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.1003894 | Genome-Wide High-Resolution Mapping of UV-Induced Mitotic Recombination Events in Saccharomyces cerevisiae | In the yeast Saccharomyces cerevisiae and most other eukaryotes, mitotic recombination is important for the repair of double-stranded DNA breaks (DSBs). Mitotic recombination between homologous chromosomes can result in loss of heterozygosity (LOH). In this study, LOH events induced by ultraviolet (UV) light are mapped throughout the genome to a resolution of about 1 kb using single-nucleotide polymorphism (SNP) microarrays. UV doses that have little effect on the viability of diploid cells stimulate crossovers more than 1000-fold in wild-type cells. In addition, UV stimulates recombination in G1-synchronized cells about 10-fold more efficiently than in G2-synchronized cells. Importantly, at high doses of UV, most conversion events reflect the repair of two sister chromatids that are broken at approximately the same position whereas at low doses, most conversion events reflect the repair of a single broken chromatid. Genome-wide mapping of about 380 unselected crossovers, break-induced replication (BIR) events, and gene conversions shows that UV-induced recombination events occur throughout the genome without pronounced hotspots, although the ribosomal RNA gene cluster has a significantly lower frequency of crossovers.
| Nearly every living organism has to cope with DNA damage caused by ultraviolet (UV) exposure from the sun. UV causes various types of DNA damage. Defects in the repair of these DNA lesions are associated with the human disease xeroderma pigmentosum, one symptom of which is predisposition to skin cancer. The DNA damage introduced by UV stimulates recombination and, in this study, we characterize the resulting recombination events at high resolution throughout the yeast genome. At high UV doses, we show that most recombination events reflect the repair of two sister chromatids broken at the same position, indicating that UV can cause double-stranded DNA breaks. At lower doses of UV, most events involve the repair of a single broken chromatid. Our mapping of events also demonstrates that certain regions of the yeast genome are relatively resistant to UV-induced recombination. Finally, we show that most UV-induced DNA lesions are repaired during the first cell cycle, and do not lead to recombination in subsequent cycles.
| Recombination occurs in both meiotic and mitotic cells. In budding yeast, there are about 100 meiotic crossovers per cell [1]. Although mitotic recombination events in S. cerevisiae are about 105-fold less frequent than meiotic exchanges [2], homologous recombination (HR) is important for the repair of double-stranded DNA breaks (DSBs) that occur spontaneously or that are induced by DNA damage. Yeast strains that lack HR grow more slowly than wild-type strains, and are sensitive to DNA damaging agents [3]. In HR events in diploid cells, the broken chromosome is repaired utilizing an intact sister chromatid or homolog as a template. Most organisms also have a pathway termed “non-homologous end-joining” (NHEJ) in which the broken ends are re-joined by a mechanism that does not require sequence homology. In diploid cells of S. cerevisiae, HR is much more important than NHEJ for repair of DNA damage [4]. We will first discuss pathways of HR, followed by a description of UV-induced DNA damage, and the recombinogenic effects of this damage.
DSBs can be repaired by a number of different HR pathways [5]. For all of these pathways, the broken DNA ends are processed by 5′ to 3′ degradation, followed by invasion of the processed chromosome end into either a sister chromatid or a homolog (Figure 1). In the synthesis-dependent strand annealing (SDSA) pathway, after strand invasion and DNA synthesis, the invading broken end is displaced and reanneals to the other broken end. The resulting product has a region of heteroduplex DNA and mismatches within the heteroduplex can be repaired to yield a gene conversion event unassociated with a crossover (Figure 1A). Alternatively, the broken ends can both engage in pairing with the intact chromosome resulting in a double Holliday junction (Figure 1B). This structure can be resolved to yield a crossover or non-crossover. As in the SDSA pathway, mismatches within the heteroduplex region can be repaired to generate a conversion event. Lastly, invasion of one broken end can result in the generation of a replication structure that duplicates sequences from the other chromosome from the point of invasion to the end of the chromosome (break-induced replication, BIR; Figure 1C).
One consequence of mitotic recombination is to cause loss of heterozygosity (LOH) for markers near the initiating lesion (gene conversions) or extending distal from the initiating lesion to the end of the chromosome (crossovers and BIR events). In Figure 2, we show the repair of DSBs in diploid mitotic cells by HR involving the homolog. In Figure 2A, we show the repair of a single broken chromatid (G2 event) using the homolog as a template. The red and black colors indicate that the two homologs have single-nucleotide polymorphisms (SNPs) that allow the detection of recombination events. Figure 2A shows a crossover between chromatids 2 and 3. If chromatids 1 and 3 segregate into one daughter cell (D1), and 2 and 4 segregate into the other (D2), a reciprocal pattern of LOH would be observed. Segregation of unrecombined chromatids 1 and 4 into one cell and the recombined chromatids 2 and 3 into the other would not lead to LOH. These two patterns of segregation are equally frequent in yeast [6]. Our previous studies [2], [7] showed that most (80%) crossovers are associated with gene conversion events (indicated by boxes in Figure 2). In Figure 2B, we show a conversion event unassociated with a crossover which produces an interstitial LOH event in one of the daughter cells. The conversion events shown in Figure 2A and 2B are termed “3∶1” events since three of the chromatids have one type of SNP and one has the other within the boxed region. A BIR event produces a region of LOH that extends to the telomere in one but not both daughter cells (Figure 2C).
The 3∶1 conversion events shown in Figures 2A and 2B are expected from the repair of a single DSB generated in S or G2 of the cell cycle. In addition, since the chromosome with the DSB acts as a recipient of information derived from the intact chromosome, these conversion events have the pattern expected if the recombinogenic DSB was on the black chromosome [4]. We observed previously, however, that over half of the mitotic conversion events had a different form from that shown in Figures 2A and 2B. In Figure 2D, we show a conversion event unassociated with a crossover in which both daughter cells have an interstitial region of LOH that is homozygous for the same SNPs; these events are called “4∶0” conversions. We interpret 4∶0 events as resulting from the repair of two broken sister chromatids in which the DSBs are located at the same positions. One simple mechanism to obtain this pattern of breakage is that the recombinogenic DSB is generated in G1, the broken chromosome is replicated, and the two resulting broken chromatids are repaired in G2 (Figure 2D). The alternative model in which the DSB is generated and repaired in G1 is ruled out because such events would not be associated with LOH for markers located distal to the conversion event [8]. If the two broken chromatids are repaired to generate conversion tracts of the same lengths, a 4∶0 event is generated. If one conversion tract is longer than the other, repair of two broken sister chromatids can also generate hybrid 3∶1/4∶0 conversion tracts [2], [7]. Our previous studies indicated that most spontaneous crossovers had conversion events consistent with a G1-initiated DSB rather than a G2-initiated DSB [9], [10], and spontaneous events resembled those induced by gamma rays in G1-synchronized yeast cells [11].
UV results in DNA lesions that are both mutagenic and recombinogenic. The primary types of lesions caused by low dosages of UV-C (∼254 nm) are pyrimidine dimers including cyclobutane dimers (CPDs) and (6-4) photoproducts (6-4 PPs) [3]. Although CPDs can be reversed in yeast by the action of photolyase, the repair of most lesions in wild-type cells likely reflects nucleotide excision repair (NER). In NER, multiple proteins act to excise a short oligonucleotide containing the damaged bases. The resulting 30-nucleotide gap is filled in by DNA polymerase delta and/or epsilon [12], and the remaining nick is sealed by Lig1p. In yeast, as in many other organisms, UV-induced lesions are more quickly repaired in transcribed genes than in non-transcribed regions [3].
Although most UV-induced lesions are removed quickly by this error-free process, a small fraction of the 30-nucleotide gaps are expanded by the action of Exo1p, resulting in large RPA-coated gaps [13], [14]. These RPA-coated regions recruit Mec1p/Ddc2p and the 9-1-1 complex, followed by subsequent recruitment of other components of the DNA damage checkpoint [15]. In addition to checkpoints triggered by the action of Exo1p, if unrepaired lesions persist into the S-phase, single-stranded regions may also be generated during the re-start of blocked replication forks. Strong activation of Mec1p by UV is observed in S-phase cells, presumably by this mechanism [16].
Although it is clear from many previous studies that UV greatly elevates the frequency of mitotic recombination in yeast [17]–[23], the recombinogenic mechanism is not well understood. There are two types of models. First, it is possible that the recombinogenic lesion is generated by NER. Consistent with this model, Galli and Schiestl (1999) [20] observed that UV of G1-synchronized cells was not recombinogenic unless the cells were allowed to replicate. They concluded that the recombinogenic lesion was likely to represent an NER-associated gap that was replicated to produce the recombination-stimulating DSB. This model predicts that the gene conversion events associated with UV-treatment of G1-synchronized cells would be exclusively 3∶1 conversion events (Figure 2A). In a preliminary study [7], however, we found that about half of the observed UV-induced conversions were 3∶1 events and about half were 4∶0 events (Figure 2D). This observation is inconsistent with the simplest form of the model proposed by Galli and Schiestl.
An alternative model is that the unexcised dimers and other DNA lesions are the recombinogenic lesion. For example, replication forks stalled at an unexcised dimer may engage in replication re-start or be broken. Although both re-start and the repair of an S-phase DSB would be expected to involve an interaction with the sister chromatid [24], some fraction of these events could involve the homolog, resulting in LOH. Kadyk and Hartwell (1993) [21] showed that UV stimulates recombination between both sister-chromatids and homologs in NER-proficient cells. In rad1/rad1 (NER-deficient) diploids, conversions, but not crossovers, were stimulated by UV in a replication-dependent manner [21]. One complication in interpreting this result is that Rad1p is involved with multiple recombination-related reactions [25]–[27] in addition to its role in NER. Regardless of this ambiguity, it is likely that unexcised dimers are recombinogenic. The summary of studies performed thus far is that some fraction of UV-induced recombination events reflects lesions resulting from NER and another fraction reflects unexcised dimers.
In the experiments described below, we examine mitotic crossovers and gene conversion events induced by UV in diploid cells. In G1-synchronized cells treated with high doses of UV, most of the events reflect the repair of two broken sister chromatids whereas at low doses, most events reflect repair of a single broken chromatid. We also show that UV induces crossovers more efficiently than BIR events. We mapped the distribution of about 100 UV-induced LOH events selected on chromosome V and about 400 unselected LOH events throughout the genome. We found that the unselected events were widely distributed throughout the genome with no very strong hotspots. The ribosomal RNA gene cluster, however, was significantly “cold” for crossovers compared to the rest of the genome.
In order to determine different types of mitotic recombination and to determine whether the conversion events are of the 3∶1 or 4∶0 configuration, we used a method of identifying recombination events that allows the recovery of both daughter cells with the recombinant chromosomes. The system used in the present study (Figure 3) is similar to that employed previously [2], [28]. Near the telomere of chromosome V, one homolog (shown in black in Figure 3A) has an insertion of SUP4-o, an ochre-suppressing tRNA gene. The diploid is also homozygous for the ade2-1 ochre mutation. Diploids homozygous for the ade2-1 mutation and zero, one or two copies of SUP4-o form colonies that are red, pink, and white, respectively [28].
In most of the experiments described below, G1-synchronized diploid cells were plated and immediately irradiated with UV. If the resulting DNA damage induces a crossover between the heterozygous SUP4-o gene and the centromere of chromosome V before the first cell division, a red/white sectored colony will be formed (Figure 3A). Since formation of a sectored colony requires a crossover, followed by the segregation pattern in which each daughter cell receives one recombined chromosome and one unrecombined chromosome (Figures 2A and 3A), only half of the crossovers induced in the first division following irradiation result in LOH. If the UV-induced DNA damage is not repaired in the first cell cycle but persists into subsequent cell cycles, a pink/white/red sectored colony could be produced (Figure 3B). As described below, most of the events induced by UV treatment in G1-synchronized cells generate a red/white sectored colony rather than a tri-colored colony. Neither gene conversion events unassociated with a crossover nor BIR events on chromosome V result in a red/white sectored colony. As will be shown below, such events can be detected as unselected events in cells that have a selected crossover on chromosome V.
The transition between heterozygous markers and homozygous markers in the sectored colony locates the position of the crossover. To detect the position of the selected crossover on chromosome V and to detect unselected LOH events throughout the genome, we used a diploid strain (PG311) derived from mating two sequence-diverged haploid strains: W303a and YJM789 [2], [7], [29]. These two strains differ by about 52,000 SNPs. We detect LOH using microarrays that examine 13,000 of these SNPs [7], allowing mapping of most events to a resolution of about 1 kb. Each SNP is represented by four 25-bp probes, two with W303a sequences (Watson and Crick) and two with YJM789 sequences. At the hybridization temperature optimized for the whole probe set, W303a genomic DNA hybridizes strongly to W303a oligonucleotides with very weak cross-hybridization to the corresponding YJM789 oligonucleotides, and vice versa for YJM789 genomic sequences. Genomic DNA is isolated from each sector of red/white sectored colonies, labeled with Cy5-tagged nucleotides, and competitively hybridized to the SNP microarray with genomic DNA from the untreated strain labeled with Cy3-tagged nucleotides. By assaying the ratio of hybridization of the differentially-tagged samples to each oligonucleotide [7], we can readily map LOH events. The transition between heterozygous and homozygous markers should be located near the site of the recombinogenic DNA lesion.
Figure 4 shows the analysis of one red/white sectored colony (59RW). In this figure, we show the normalized ratio of hybridization of genomic sequences to W303a- and YJM789-specific oligonucleotides on chromosome V with red lines and black lines, respectively; CEN5 is located near coordinate 152 kb. In the top part of Figure 4A, we depict the pattern of hybridization of genomic DNA isolated from the red sector. The ratio of hybridization is about 1 for all SNPs from coordinate 105 kb to the right telomere, indicating that SNPs in this region are heterozygous. In the red sector, SNPs centromere-distal to coordinate 105 kb on the left arm are homozygous for the W303a-derived SNPs whereas the genomic DNA from the white sector becomes homozygous at approximately the same position for YJM789-derived SNPs. In Figure 4B, the same recombination event is depicted at higher resolution; each square and diamond shows the level of hybridization to an individual YJM789-specific or a W303a-specific SNP, respectively. As shown in this figure, the red sector has a single transition between heterozygous and homozygous SNPs whereas the white sector has three transitions. The pattern of these transitions indicates that the crossover is associated with a 3∶1/4∶0 hybrid conversion tract.
Most of our experiments involve UV treatment of G1-synchronized cells with 15 J/m2; the experimental parameters used for each experiment are in Table S1. PG311 is hemizygous at the MAT locus (MATa/MATα::NAT), allowing its synchronization in G1 using the alpha pheromone [11]. The synchronized cells were plated onto solid medium and immediately irradiated at doses varying between 1 and 15 J/m2. Even at the maximum dose of UV, cell viability was 70%. No sectored colonies were observed in cells that were not treated with UV. Based on our earlier study of spontaneous crossovers in the same strain [2], the rate of crossovers in untreated cells is 1.1×10−6/division in the 120 kb interval between CEN5 and the SUP4-o marker. Relative to this rate, UV treatment stimulated sector formation by factors of 1500 (1 J/m2), 1600 (5 J/m2), 5000 (10 J/m2), and 8500 (15 J/m2). The strong stimulation of mitotic crossovers by UV is consistent with previous studies [23].
In some studies [2], [4], [28], the frequency of mitotic recombination events is higher in diploids that express both mating types than in diploids that express only one mating type. Consequently, we compared the frequency of red/white sectored colonies in G1-synchronized cultures of PG311 and PSL101 (the MATa/MATα progenitor of PG311). Because PSL101 cannot be synchronized in G1 using alpha pheromone, both strains were synchronized in G1 by growing the cells into stationary phase (Text S1). After treatment of the G1-synchronized cells with 15 J/m2 of UV, 0.4% (0.2–0.9%, 95% confidence limits) of the PG311 colonies formed red/white sectors compared to 0.6% (0.4–1%) of the PSL101 colonies. Although the confidence limits are wide, these results indicate that mating type heterozygosity does not have a large effect on the frequency of UV-induced mitotic crossovers in our system.
In addition to red/white sectored colonies, in the irradiated samples, we also observed pink/red and pink/white/red colonies. Such colonies could represent non-reciprocal recombination events (for example, BIR events), persistence of recombinogenic DNA damage beyond the first cell cycle, or an artifact (two closely-located independent cells). To exclude sectors formed artifactually, we micromanipulated individual G1-irradiated (15 J/m2 dose) single cells to specific positions on plates with solid medium, and monitored their subsequent development to form sectored or unsectored colonies. From a total of 970 isolated irradiated single cells, we observed eleven sectored colonies of the following types: seven red/white colonies, two pink/red colonies, and two pink/white/red colonies. From our SNP microarray analysis of the LOH patterns on chromosome V in these colonies (described in Text S1 and Figure S1), we found that all seven of the red/white colonies represented crossovers induced during the first cell cycle. The two pink/red sectored colonies reflected chromosome loss, resulting in a monosomic red sector and a pink sector. Only one of the pink/white/red colonies was a consequence of a UV-induced recombination event in the second division (Figure 3B, Figures S1 and S2). In summary, of the nine sectored colonies in which sectoring reflected a UV-induced crossover, eight occurred prior to the first cell division and only one occurred after the first cell division, indicating that most UV-induced DNA lesions are rapidly repaired.
We used SNP microarrays to analyze 47 sectored colonies of G1-synchronized cells treated with 5, 10 or 15 J/m2 of UV (Tables S2 and S3). 80% of the colonies were from cells treated with 15 J/m2. Nine of these colonies were derived from the single-cell experiments described above. 45 of the 47 sectored colonies examined had patterns of LOH on chromosome V consistent with a reciprocal crossover on the left arm of chromosome V. In one of the two exceptional colonies, there was a loss of one copy of chromosome V. In the other colony, there were two independent conversions that resulted in LOH events that were unassociated with a crossover. These two sectored colonies were not used in our subsequent analysis of selected events on chromosome V, although data from these colonies were used to analyze unselected recombination events.
In addition to the selected LOH events on chromosome V, we observed an average of eight unselected LOH events per sectored colony. As described below, our analysis of the 45 selected and 381 unselected events (300 gene conversion events unassociated with crossovers, 60 crossovers, and 21 BIR events) allowed us to determine several important features of the UV-induced recombination events: 1) the patterns of gene conversion in selected and unselected recombination events, 2) the lengths of gene conversion tracts associated or unassociated with crossovers, and 3) the locations of selected and unselected recombination events induced by UV. Since the frequency of selected sectored colonies in cells irradiated with 15 J/m2 was about 1%, and the selected interval on chromosome V is about 1% of the genome, we expect about one unselected crossover per irradiated cell, roughly the observed frequency (60 unselected crossovers/47 sectored colonies).
One interpretation of our observation of frequent DSCBs in G1-irradiated cells is that the repair of two very closely-spaced single-stranded DNA lesions induced by 15 J/m2 results in DSCBs in the G1-synchronized cells, whereas SCBs reflect DNA lesions on one strand. Thus, the productions of DSCBs by this mechanism would be proportional to the square of UV dosage, whereas the frequency of SCBs would be linearly proportional to the UV dosage. By this model (details to be discussed below), one might expect that a low dose of UV should have a relatively higher frequency of SCBs. Consequently, we examined the frequency and types of recombination events induced in G1-synchronized cells by 1 J/m2. As expected, the frequency of red/white sectored colonies was reduced in cells irradiated with 1 J/m2 relative to cells irradiated with 15 J/m2 (1.6×10−3/division versus 9.4×10−3/division). Ten sectored colonies were examined by whole-genome microarrays. Only four unselected events were observed. This frequency (0.4 events/sectored colony) was about twenty-fold less than that observed in samples irradiated with 15 J/m2 (8 events/sectored colony). Consequently, in the additional thirty-six sectored colonies examined, we used microarrays specific for detecting LOH on chromosome V.
The depictions of the LOH events in the 1 J/m2 irradiated samples that had the same patterns as observed for the 15 J/m2 samples are shown in Table S2; the numbers of samples with specific classes of events are shown in parentheses in this table. Patterns of LOH that were unique to the 1 J/m2 samples are shown in Table S6. The coordinates for these LOH events are shown in Table S7. The distribution of the LOH events on chromosome V for the 1 J/m2 samples was not significantly different from that observed for the 15 J/m2 samples or the spontaneous events using the same “binning” procedure and statistical test described above. The median length of conversion events associated with crossovers on chromosome V in cells irradiated with 1 J/m2 was 4.3 kb (2.3 kb–8.2 kb; 95% confidence limits) kb. In cells irradiated with 15 J/m2, the median length of conversion tracts associated with crossovers on chromosome V was 6.7 (4.2–13 kb). The distributions of tract lengths analyzed by the Mann-Whitney test showed that these distributions were not significantly different (p = 0.12).
A striking difference was observed in the distributions of events diagnostic of SCBs and DSCBs in cells irradiated with 1 and 15 J/m2 of UV. Of selected events on chromosome V in cells irradiated with 1 J/m2, we observed 5 crossovers unassociated with conversion, 31 SCB events, and 10 DSCB events. In contrast, in cells irradiated with 15 J/m2, most of the selected events on chromosome V were DSCB events (Figure 8). By the Fisher exact test, the difference in the numbers of SCB and DSCB events induced by the two different UV treatments is very significant (p<0.0001). The conclusion that G1-synchronized cells have different recombinogenic DNA lesions induced by different UV doses will be discussed further below.
Our main conclusions are: 1) UV induces high frequencies of crossovers and gene conversions in G1-synchronized yeast cells but is less recombinogenic in G2-enriched cells, 2) the LOH events induced by high doses of UV primarily involve the repair of two sister chromatids broken at the same position whereas most events induced by low doses (1 J/m2) involve repair of a single broken chromatid, 3) UV-induced LOH events are widely distributed throughout the genome, although some classes of repeated genes are significantly underrepresented, 4) most of the recombinogenic effects of UV in cells treated in G1-synchronized cells are manifested in the first cell cycle after irradiation, 5) in G1-synchronized cells, crossovers are induced about six-fold more frequently than BIR events, 6) about one-third of UV-induced conversion events are associated with crossovers, and 7) although UV effectively induces crossovers between homologs, chromosome rearrangements are not produced at detectable frequencies.
As noted in previous studies, UV very effectively induces mitotic recombination in yeast [7], [10], [20], [21], [23], [38]. In experiments involving heteroallelic recombination in synchronized cells, UV is somewhat more recombinogenic in G1-synchronized cells than in G2-synchronized cells [18], [21]; our results support these observations. Kadyk and Hartwell (1992) [24] concluded that DSBs induced by X-rays in G2-synchronized cells were repaired primarily by sister-chromatid recombination, whereas X-ray treatment of G1-synchronized cells effectively stimulated recombination between homologs. Our previous interpretation of both spontaneous and DNA damage-induced crossovers is also consistent with this conclusion [2], [7], [29]. We argue that most spontaneous crossovers between homologs are initiated by a DSB in G1 in one chromosome, and replication of the broken chromosome produces two sister chromatids broken at the same position. Since these lesions cannot be repaired by sister-chromatid recombination, they are repaired by recombination with the homolog. Although it is likely that DSBs formed in S or G2 are primarily repaired by sister-chromatid recombination, some DSBs generated in G2 are repaired by interaction with the homolog [11].
As observed in our previous studies [2], [7], [29], the mitotic conversion tracts are long compared to those observed in meiosis, and the tracts associated with crossovers are longer than the tracts unassociated with crossovers. Most of the conversion events are explicable as a consequence of repair of one broken chromatid or two sister chromatids broken at the same position by the standard HR pathways shown in Figure 1, with only conversion-type MMR and not restoration-type MMR. About 15% of the conversion events, however, are more complex, requiring “patchy” repair of mismatches within a heteroduplex (mismatches corrected by both conversion-type repair and restoration-type repair within one heteroduplex), and/or branch migration of the Holliday junction. The fraction of complex conversion tracts in the current study is similar to those observed in our previous studies [7], [29]. Although these events (described in detail in Text S1 and Figures S5, S6, S7, S8, S9, S10, S11, S12, S13, S14, S15) are explicable by modifications of the standard models shown in Figure 1, it is possible that some of these conversion events involve a substantially different mechanism such as multiple template switching events during BIR. In this context, template switching during BIR has been observed in experiments in which linear DNA fragments are transformed into yeast [39]. In addition to the complex tracts, it is possible that the very long conversion tracts reflect BIR rather than mismatch repair in a heteroduplex; 16% of the conversion events unassociated with crossovers are greater than 10 kb in length, and the longest exceeds 50 kb. Finally, it should be pointed that, although single BIR events would not be expected to generate crossovers, a model for production of a crossover by a double BIR event is shown in Figure S4 of Lee et al. (2009) [2].
A central issue is the nature of the recombinogenic DNA damage generated by UV. Based on the mechanism of NER and on the observation that unrepaired pyrimidine dimers block replication, there are two obvious potential sources of DSBs [40]. First, if a DNA molecule with an unrepaired gap resulting from NER is replicated before filling-in of the gap and ligation, the net result would be a pair of sister chromatids with a single DSB (Figure 9A). Alternatively, if a replication fork encounters an unrepaired UV-induced lesion, breakage of the fork could also result in a single broken chromatid (Figure 9B). Based on the observation that UV treatment of G1- or G2-synchronized cells was not recombinogenic unless cells were allowed to divide, Galli and Schiestl (1999) [20] suggested that cell division was required to convert DNA lesions to recombinogenic lesions, consistent with both of the possibilities described above; their assay detected only intrachromatid deletions. Kadyk and Hartwell (1993) [21] found that unrepaired UV lesions stimulate gene conversion events between homologs, but have little effect on mitotic crossovers. This conclusion may be affected by the use of the rad1 mutation to prevent dimer excision, since rad1 strains have reduced frequencies of crossovers in some assays [41].
Both of the models discussed above predict that UV-induced DNA damage in G1-synchronized cells would produce primarily gene conversion events involving a single broken chromatid (SCBs). In our study, about two-thirds of the conversion events in which cells were irradiated with 15 J/m2 reflect two broken sister chromatids, but only one-quarter of the conversions reflect two broken sister chromatids in cells irradiated with 1 J/m2 (Figure 8). Thus, there is a qualitative change in the nature of the DNA lesion with increasing UV dose. In addition, since our single-cell experiments demonstrate that UV-induced lesions are recombinogenic during the first division following treatment, DSCBs cannot be explained as reflecting the segregation of a chromosome with an unrepaired G2-associated DSB from the previous division.
We suggest that most DSCBs are a consequence of a DSB in G1. Although UV damage is generally regarded as an agent that produces DNA nicks rather than DSBs, a gel-based detection of the conversion of a circular chromosome to a linear chromosome indicated that a dose of 40 J/m2 produces 5 to 10 DSBs in G2-synchronized cells [42]. There are several related mechanisms by which NER could produce a DSB in G1 cells. First, the excision tracts resulting from removal of two closely-opposed dimers could result in very short (<6 bp) unstable duplex regions between the repair tracts, resulting in a DSB (Figure 9C). A second model is that, following the removal of two closely-opposed dimers by NER, one or both of the resulting short gaps is expanded by Exo1p (Figure 9D). A third related model is that the excision tract generated by NER is expanded into a large single-stranded gap that is cleaved by an endonuclease to yield the DSB (Figure 9E).
Based on our results and those of others, it is likely that UV produces a variety of recombinogenic lesions. In our experiments, at a low dose of UV (1 J/m2), we observed primarily SCBs, consistent with the two models shown in Figures 9A and 9B. At 15 J/m2, we observed DSCBs more frequently than SCBs. This observation supports models shown in Figure 9C and 9D that require closely-opposed lesions, and argues against the model shown in Figure 9E in which the relative fraction of DSCB and SCB events would be expected to be independent of the density of the NER tracts. It can be calculated that diploid cell irradiated with 15 J/m2 have about 7500 dimers/genome [43]; if these dimers are distributed randomly, we expect about 35 closely-opposed (separated by ≤75 bases) dimers, enough to explain the detected DSCB events. It is likely that the number of closely-spaced dimers is greater than that determined by this calculation. Lam and Reynolds (1987) [43] found that the fraction of dimers located within 15 base pairs of each other is greater than expected from a random distribution, and this fraction is somewhat independent of UV dose. These dimers may be responsible for the DSCB events detected in strains treated with the 1 J/m2 UV dose.
In summary, we suggest that low doses of UV primarily result in SCBs as a consequence of replication of a chromosome with a NER-generated DNA gap in one strand, or an unrepaired dimer resulting in breakage of one arm of the replication fork. In contrast, we suggest that high doses of UV often result in DSCB events as a consequence of a G1-generated DSB, reflecting cellular enzymes acting on closely-opposed dimers. Although this explanation seems straightforward, we cannot exclude more complex explanations of our data. For example, it is possible that the very large number of UV-induced lesions at high doses may overwhelm the DNA repair systems, resulting in changes in the use of repair pathways. In addition, we stress that our analysis based on interhomolog recombination does not yield an estimate of the relative frequencies of UV-induced recombinogenic lesions produced in G1, S, and G2, since most recombinogenic lesions produced in S and G2 are likely repaired by sister-chromatid recombination [24], a mechanism that does not lead to LOH [7].
The UV-induced recombination events were broadly distributed throughout the genome; no strong recombination hotspots were detected. The distribution of UV-induced genomic LOH events is expected to be a function of multiple factors such as: 1) the distribution of DNA damage, particularly the distribution of closely-opposed dimers, 2) the relative frequency of dimer repair by recombinogenic and non-recombinogenic pathways, and 3) the relative frequency of repair of recombinogenic DNA damage by sister-chromatid recombination, non-homologous end-joining, and recombination between homologs. The regions on each chromosome that were examined for LOH events are in Table S8.
It has been shown recently that the distribution of UV-induced pyrimidine dimers observed in vivo in yeast is primarily a function of the density of TT, TC, CT, and CC sequences in the genome ([44], Sheera Adar and Jason Lieb, personal communication). As previously discussed, in cells irradiated with 15 J/m2, the calculated frequencies of AA/TT dinucleotides among our selected (0.221) and unselected (0.218) conversion events are very close to the frequencies on the left arm of chromosome V (0.219) and the whole genome (0.217). In contrast, the frequency of these dinucleotides in the ribosomal DNA (0.19) is very significantly less (p<0.0001) than the frequency in the whole genome. Thus, the observed reduction in UV-induced recombination in the ribosomal DNA, at least in part, may be a consequence of a reduced frequency of dimer formation. Interestingly, the two motifs that are underrepresented in the unselected conversion events (tRNA and solo LTRs in Table S9) also have frequencies of AA/TT dinucleotides that are considerably less than the genomic frequency (0.124 for tRNA genes and 0.205 for the solo LTRs). Since the tRNA genes and the solo LTRs are smaller than the average conversion tract size, however, it is unlikely that the relative lack of AA/TT dinucleotides is the only factor influencing the frequency of UV-induced conversion events that include these elements. Finally, we note that the frequency of AA/TT dinucleotides in non-coding RNA genes, which are significantly over-represented in the DSCB conversion tracts (Table S9), is also higher (0.23) than the frequency for the whole genome. A more detailed discussion of the relationship between various chromosome elements and conversion tracts (Tables S9 and S10) is given in Text S1.
Although dimer formation has a simple relationship to DNA sequence, the rate of NER-mediated repair of the dimers is enhanced by transcription and reduced by chromatin silencing and other aspects of chromatin structure [45]–[47]. Our discussion of dimer repair will be limited to NER, since our experiments were done under conditions in which photoreactivation was prevented. In general, dimer repair is rapid in yeast with the majority of dimers being removed within two hours [48]. Our observation that UV treatment of G1-synchronized cells primarily results in recombination in the first cell cycle following radiation is consistent with efficient dimer repair. Nonetheless, Teng et al. (2011) [44] found genomic regions in which dimer repair was delayed. To test whether these long-lasting lesions could be more recombinogenic than lesions that were quickly repaired, we determined whether the chromosome regions containing the long-lasting lesions were over-represented in our unselected gene conversion tracts (details of the analysis in Text S1). There was not a significant enrichment of the regions with long-lasting lesions in our unselected gene conversion tracts.
Most previous studies of BIR involve transforming linear fragments of DNA or using strains in which the interacting homologous sequences are flanked by non-homologous regions [4], [49]. In contrast, our ability to distinguish BIR from crossovers is based on the recovery of both cells containing recombinant products. Among unselected LOH events examined in G1-synchronized cells irradiated with 15 J/m2, we observed 60 crossovers and 21 BIR events. By the microarray analysis, as described previously, we detect only half of crossovers. All BIR events, however, can be detected. We conclude, therefore, that crossovers are induced about six-fold more than BIR events. This conclusion is in agreement with previous observations of spontaneous recombination events [50], and events induced by the I-SceI endonuclease [51] performed by others.
About 60% of the UV-induced BIR events appear to be randomly distributed, whereas the remainder have a breakpoint located within 50 kb of the telomere. We suggest that there are two types of BIR events, the “classic” type in which one of the chromosome fragments is lost prior to second end capture, and a second type that is initiated by degradation of one of the two homologs beginning at the telomere. In two previous studies [52], [53], LOH events near the telomere were observed in strains with mutations in genes affecting telomere structure or replication (CDC13 and EST1). It was not determined in these studies whether these LOH events were crossovers or BIR events. Since the BIR events in our study were induced by UV, one interpretation of our results is that high doses of UV are associated with telomere uncapping or some other telomere defect. An alternative explanation of our observation that BIR events are enriched near the telomere is that such events are more efficiently initiated and completed than events located more internally on the chromosome, as demonstrated by Donnianni and Symington [49].
In meiosis in S. cerevisiae, roughly half of the conversion events are associated with crossovers [1], [54]. The fraction of conversions associated with crossovers varies in different studies from <5% to 50% [4]. In the unselected events induced by 15 J/m2, we observed 300 conversions without an observable crossover, and 60 crossovers. Because of the pattern of chromosome segregation, we expect only half of the crossovers will lead to LOH distal to the exchange. We calculate that there are likely 240 conversions unassociated with crossovers and 120 associated with crossovers. Thus, we conclude that about one-third of the unselected conversion events are associated with a crossover, similar to our previous conclusion based on a smaller number of events [7].
In our previous analysis of gamma ray-treated G2-synchronized diploids, we observed about two unselected chromosome aberrations per irradiated cell [31]. We found that most of these events reflected homologous recombination between Ty elements located at non-allelic positions. In our current study, although UV treatment induced about eight unselected LOH events per cell irradiated with 15 J/m2, we did not detect any large deletions, duplications, or translocations. The difference in these two studies is likely to reflect the total number of DSBs and other recombinogenic lesions generated by the two treatments. In the Argueso et al. study, the gamma ray dose (800 Gray) produces about 250 DSBs/cell. Based on the estimate that 40 J/m2 of UV results in 5–10 DSBs in G2-synchronized cells [42], we expect only about two DSBs/cell as the result of irradiating G1-synchronized cells with 15 J/m2. Since Ty elements, the main target for chromosome rearrangements, represents only a small fraction of the genome (2%), the likelihood of a UV-induced DSB within Ty elements is small, although DSBs located near Ty elements can also contribute to Ty-Ty exchanges [55]. Of the 360 unselected conversion and crossover events induced by UV, 14 included a Ty element; it is unclear whether these events initiated within or nearby the Ty. Although the lack of recombinogenic lesions within or near Ty elements may be sufficient to explain the dearth of chromosome rearrangements, other factors may also be important, since UV does not effectively stimulate recombination between non-allelic Ty elements [56], [57].
Most of our experiments were done with the diploid strain PG311, a hybrid that is heterozygous for about 52,000 SNPs [2]. The PG311 genotype is: MATa/MATα::NATade2-1/ade2-1 can1-100/can1Δ::SUP4-o ura3-1/URA3 trp1-1/TRP1 his3-11,15/HIS3 leu2-3,112/LEU2 V9229::HYG/V9229 V261553::LEU2/V261553 GAL2/gal2 RAD5/RAD5. Additional features of this strain are described in Text S1 (Supplemental Materials and Methods). We also describe the strains JSC24,JSC25 and PSL101, all of which are isogenic to PG311 except for the specified alterations. The experiments to measure recombination within the ribosomal RNA genes were done with the diploid AMC45 that is heterozygous for markers flanking the array and within the array [34]. The diploid YYy13 is a MATα::NAT derivative of AMC45. Unlike the other strains used in our analysis, AMY45 and YYy13 are not isogenic with PG311.
Standard rich growth medium (YPD) and omission media were used for these experiments [58]. We also used standard conditions for tetrad analysis, transformation, and DNA isolation.
In most of our experiments, PG311 cells were synchronized in G1 using α-factor or in G2 using nocodazole as described by Lee and Petes [11]. After two hours of treatment with these agents, the cells were plated on medium lacking arginine and irradiated with UV using a TL-2000 UV Translinker; doses varied between 1 and 15 J/m2. Following the UV treatment, the plates were covered with foil to prevent light-associated removal of dimers, and incubated for two days to allow the formation of sectored colonies. In some experiments, modifications of this protocol were employed as described in Supplemental Materials and Methods.
LOH events in PG311 and related strains were detected using SNP microarrays. For each SNP, these Agilent-constructed microarrays contain four oligonucleotides, one pair that hybridizes to the YJM789-derived SNP allele and another that hybridizes to the W303a-derived SNP allele [7]. About 13,000 SNPs distributed throughout the genome were examined. A short description of the use of SNP microarrays is in the Results section and additional details are given in St. Charles et al. (2012) [7]. In brief, genomic DNA from the experimental strain was labeled with Cy5-dUTP and control DNA from the fully heterozygous strain JSC24-2 was labeled with Cy3-dUTP. The two DNA samples were then hybridized in competition to the SNP microarrays. The microarray was examined using a GenePix scanner. By measuring the ratio of hybridization of the two differentially-labeled samples, we could determine which SNPs were heterozygous and which were homozygous.
Most of our statistical analysis involved chi-square analysis, the Fisher exact test, or the Mann-Whitney test. These tests were done using the VassarStat Website (http://vassarstats.net/) or the functions associated with Excel. To calculate 95% confidence limits on the median, we used Table B11 of Altman (1990) [59].
|
10.1371/journal.ppat.1003358 | Colocalization of Different Influenza Viral RNA Segments in the Cytoplasm before Viral Budding as Shown by Single-molecule Sensitivity FISH Analysis | The Influenza A virus genome consists of eight negative sense, single-stranded RNA segments. Although it has been established that most virus particles contain a single copy of each of the eight viral RNAs, the packaging selection mechanism remains poorly understood. Influenza viral RNAs are synthesized in the nucleus, exported into the cytoplasm and travel to the plasma membrane where viral budding and genome packaging occurs. Due to the difficulties in analyzing associated vRNPs while preserving information about their positions within the cell, it has remained unclear how and where during cellular trafficking the viral RNAs of different segments encounter each other. Using a multicolor single-molecule sensitivity fluorescence in situ hybridization (smFISH) approach, we have quantitatively monitored the colocalization of pairs of influenza viral RNAs in infected cells. We found that upon infection, the viral RNAs from the incoming particles travel together until they reach the nucleus. The viral RNAs were then detected in distinct locations in the nucleus; they are then exported individually and initially remain separated in the cytoplasm. At later time points, the different viral RNA segments gather together in the cytoplasm in a microtubule independent manner. Viral RNAs of different identities colocalize at a high frequency when they are associated with Rab11 positive vesicles, suggesting that Rab11 positive organelles may facilitate the association of different viral RNAs. Using engineered influenza viruses lacking the expression of HA or M2 protein, we showed that these viral proteins are not essential for the colocalization of two different viral RNAs in the cytoplasm. In sum, our smFISH results reveal that the viral RNAs travel together in the cytoplasm before their arrival at the plasma membrane budding sites. This newly characterized step of the genome packaging process demonstrates the precise spatiotemporal regulation of the infection cycle.
| Influenza A viruses cause one of the major respiratory infection diseases in humans. The viruses possess a genome consists of eight different RNA segments and the incorporation of all the eight RNA segments is required for the generation of an infectious virus particle. The precise process of how these eight viral RNA segments are co-packaged into progeny virus particles remains undefined due to the limitations of methodology to determine the locations of different vRNA segments in infected cells with single-molecule resolution. In this study, we established an experimental system to examine the localization of different viral RNA segments in an infected cell with high spatial precision. We found that viral RNA belonging to different segments gather together in the cytoplasm which is facilitated by cellular recycling endosomal protein Rab11. Our results supported the idea that eight different viral RNAs likely form a super-complex as they travel to the site for virion incorporation. These findings extend our knowledge on the process of influenza virus genome packaging and suggest a mechanism by which the genome assembly of different viral RNA segments is regulated.
| The Influenza A virus genome consists of eight negative-sense, single-stranded RNAs. In a virus particle or an infected cell, the viral RNAs exist in the form of viral ribonucleoprotein complexes (vRNPs) with the viral RNA (vRNA) encapsidated by the nucleoproteins (NP) and associate with the polymerase complex [1] . Since each vRNA encodes one to two essential viral proteins, the packaging of all eight vRNAs is required for the production of an infectious virus particle. Multiple pieces of evidence have shown that influenza virus selectively packages eight different vRNPs into virus particles [2], [3], [4], [5], [6]; however, when and where the selection occurs during viral infection remains unknown. The separation and assembly of different vRNA segments is difficult to determine, due to the limitation of methodology applicable to detect different vRNA segments with single-molecule sensitivity and preserve their spatial distributions in the infected cells. In this report, we established a single-molecule sensitivity fluorescence in situ hybridization (smFISH) system which detects and localizes influenza vRNAs in an infected cell, allowing the determination of the spatial relationship among different vRNPs. The assembly process of different vRNA segments during virus infection can therefore be studied with high resolution. Influenza virus is one of the rare RNA viruses which has a nuclear replication phase [1]. The vRNPs first have to be released from the virion by the disassociation of matrix protein and then they are imported into the nucleus for viral mRNA transcription and vRNA synthesis. The nuclear import of vRNPs is an active process that involves the cellular import machinery. Nuclear localization signals on the NP protein are recognized by importin α (karyopherin α) and together the vRNP and importin α form a tri-complex with importin-β that is actively transported into the nucleus through the nuclear pore complex [7]. It has been shown with biochemical analysis that the replicating vRNPs were associated with densely packed chromatin while the newly synthesized vRNPs are released into the nucleoplasm [8]. In addition, it was observed that the NP proteins were distributed to the apical face of the nucleus when the export of vRNPs was inhibited [9]. This suggested that the polarized transport of vRNPs started when they were in the nucleus. The newly assembled vRNPs are exported into the cytoplasm, where they are targeted to the plasma membrane for packaging. The export of vRNPs was also dependent on the cellular export machinery and the influenza nuclear export protein (NEP) [10], [11]. It has been demonstrated that leptomycin B treatment efficiently abrogated the export of vRNPs, showing that the transport of vRNPs out of the nucleus is a Crm1 dependent pathway [10], [12]. Even though the mechanism of vRNPs transport into and out of the nucleus has been elucidated, it is unknown whether different segments of the incoming vRNPs enter the nucleus separately or as a package and whether the newly synthesized vRNPs gather into a super-complex in the nucleus to be exported.
It has been suggested that the reassortment of influenza vRNAs happens in the cytoplasm because when the cytoplasm of two different virus infected cells was fused, the segments from the two virus strains were incorporated into progeny viruses in a random manner [13]. It is thus possible that the selection of vRNPs to be packaged happens after nuclear export. The trafficking route of the exported vRNPs to the apical plasma membrane has been shown to involve the cellular cytoskeletal system [14] and Rab11-positive recycling endosomes [15], [16], [17]. Live-cell imaging using an antibody specifically for vRNP showed that the complex moves along microtubules rapidly in both directions. Depolymerization of microtubules using nocodazole was shown to disrupt the apical targeting of the nucleoproteins, implicating the microtubule network in the polarized transport of vRNPs [16]. Rab11 is a small GTPase regulatory protein that has been shown to be localized to the endocytic recycling endosomes and plays an essential role in regulating recycling to the plasma membrane [18]. Several studies have shown that apical transport of influenza vRNPs depends on the interactions between Rab11 and the polymerase protein PB2 [17], [19]. It was unknown whether different vRNA segments traffick together with the Rab11 recycling endosomes to the plasma membrane or whether the colocalization and sorting of different vRNAs to be packaged happens at the plasma membrane.
In this present study, we applied single-molecule sensitivity fluorescence in situ hybridization assay to visualize vRNAs of different segments in influenza virus infected cells. By using this technique, the degree of colocalization between two different vRNAs can be determined with great spatial precision. By performing smFISH and colocalization analysis on virus infected cells at different time points post infection, we found that the incoming vRNPs remain associated until they are imported into the nucleus. Newly synthesized vRNPs were detected at different locations in the nucleus and the newly exported vRNPs of different identities gather together at later stages of replication when the vRNPs are loaded onto Rab11 positive vesicles. Herein, we have provided evidence that progeny vRNPs of different identities travel together in the cytoplasm. These results likely suggest that different vRNPs are selected into a pre-formed super-complex during their trafficking to the plasma membrane, where budding and genome packaging occur.
To better understand the interaction pattern between vRNAs of different identities in influenza virus infected cells, we established a single-molecule sensitivity fluorescence in situ hybridization (smFISH) system for influenza vRNAs to determine their locations in infected cells. Single-molecule sensitivity was obtained using 48 single-fluorophore-labeled short DNA oligos targeting different regions of the same vRNA. The targeted vRNAs were bound by multiple probes at the same time giving high fluorescence intensity signals, allowing them to be distinguished as diffraction-limited spots. The number of spots and the location of their centers in 3-dimensional space can then be determined using - imaging processing programs. When 48 Cy5 labeled probes against the PB2 segment were used, fluorescence spots can be observed in influenza A/Puerto Rico/8/34 (PR8) virus infected cells at 4 hours post infection (hpi) but no spots were detected in the mock infected cells (Fig. S1A). This demonstrated that the smFISH system was highly specific for vRNAs. The fact that the fluorescence spots detected by the imaging analysis program appear homogenous in size as diffraction-limited spots as well as the fact that the fluorescent particles display a unique, well-defined fluorescence intensity peak are strong indications that single molecules of vRNAs were detected using this system (Fig. S1B). To further test the specificity of the probe sets against different species of influenza vRNAs, probe sets targeting the HA vRNA encoding the head regions of the H1 subtype and the H9 subtype were designed and synthesized. The probes for the H1 HA vRNA were labeled with Cy3 fluorophores and those for the H9 HA vRNA were labeled with Cy5 fluorophores. The two differently labeled probe sets were mixed and used for in situ hybridization on cells infected with wild type PR8 virus harboring the H1 HA segment or with the recombinant PR8 cH9/1 virus which contains an HA segment coding the head region from an H9 subtype HA and a stalk from an H1 HA. In cells infected with the PR8 virus, fluorescent spots can only be observed in the Cy3 channel but not from the Cy5 channel. This demonstrated the specific binding of probes against the H1 HA vRNAs. On the other hand, fluorescent spots can only be detected in the Cy5 channel in cells infected with the recombinant PR8 cH9/1 viruses, showing the specific hybridization of the probes against the H9 HA vRNAs (Fig. S1C). This experiment demonstrated the high sequence specificity of the smFISH system and showed that the specificity could be preserved when two probe sets targeting different species of vRNAs were mixed during the hybridization process. This allowed the detection of vRNAs of different segments in the same cell at the same time.
In order to measure the degree of colocalization between vRNAs of different identities, cells infected with viruses were fixed and hybridized using two sets of probes targeting two different vRNAs, one labeled with Cy3 and the other labeled with Cy5. The center of each spot corresponding to a vRNA molecule was then located in 3-D space using the custom spot detection algorithm. Mapping of the different colored spots revealed the distances among spots and colocalization efficiency between the two vRNA segments could then be determined (Fig. 1A). A custom designed colocalization analysis tool was developed to measure the distances between color spots and their nearest neighbor spots of a different color. This allows the quantification of the number of spots being colocalized in individual cells. To validate the system, two control experiments were performed. The positive control experiment was done using two differently labeled probe sets (Cy3 and Cy5) targeting different regions of the same vRNA. Figure 1B showed the images of NA vRNAs detected by two differently labeled probe sets. The high degree of colocalization between the Cy3 and Cy5 spots in cells infected with PR8 virus at 6 hpi demonstrated the specificity and the sensitivity of our colocalization analysis. Quantitative colocalization analysis of the images showed that 80% of the Cy3 and Cy5 spots colocalized in the cytoplasm and 70% of them colocalized in the nucleus (Fig. 1D). To exclude the possibility that the high colocalization efficiency detected could be due to high density of vRNAs, we tested the colocalization efficiency between the highly expressed cellular β-actin mRNA and NA vRNA. The MDCK cells infected with PR8 virus were hybridized with Cy3 labeled probes for β -actin mRNA and Cy5 labeled probes targeting the NA vRNA. At 8 hpi, we observed NA vRNAs in the nucleus and the cytoplasm, whereas the β -actin mRNAs were mainly in the cytoplasm. The copy number of β-actin mRNA detected in the cytoplasm was 1409±283, similar to what was detected for viral RNAs at 4–8 hour post infection (Fig. 2D). When merging the two channels, a very small degree of colocalization between the two RNA species was observed (Fig. 1C). Quantification of the colocalization between the beta-actin mRNAs and NA vRNAs also showed a low percentage of colocalization, approximately 2.5% in the nucleus and 5% in the cytoplasm (Fig. 1E). These results demonstrated the specificity and sensitivity of our quantitative colocalization analysis.
To investigate the colocalization kinetics of different viral RNA species during the influenza virus life cycle, we performed smFISH and colocalization analysis on two vRNAs: PB2 and NA at different time points post infection. Figure 2A shows the typical images of the distributions of PB2 and NA vRNAs at different hour-post infections. At 2 hpi, the PB2 and NA vRNA molecules were observed mainly within the nucleus with only a few in the cytoplasm. New vRNAs were synthesized at this time point; however, the PB2 and NA vRNAs were observed in different locations in the nucleus. No colocalization between the PB2 and NA vRNAs was observed (Fig. 2B). Even though few spots were detected in the cytoplasm, the PB2 and NA vRNAs present high colocalization efficiency in this compartment. It is likely that the colocalized spots represent vRNAs of incoming virus particles which have not yet entered the nucleus. As the vRNAs were being replicated, the nucleus became packed with newly synthesized vRNAs at 4 hpi. It should be noted that the measured colocalization efficiency between PB2 and NA vRNAs increased over time in the nucleus, albeit reaching significantly lower level compared to the control. The extreme accumulation of vRNAs in the nucleus at later time points resulted in a very high spatial density of fluorescence spots. In this concentration regime, actual physical interactions between single PB2 and NA vRNAs became difficult to distinguish from crowding-induced random proximity. This effect could account for the higher colocalization fraction measured in the nucleus at later time points using our assay (Fig. 2B). The smFISH system also allowed the quantification of the copy number of vRNAs in individual cells. By doing quantification of copy number of vRNAs in the nucleus and cytoplasm compartments in each cell, it was observed that increased vRNAs can be detected in the nuclei at 2 hpi, demonstrating that the replication of vRNA started as early as 1 hour post infection. Nuclear export of newly synthesized vRNAs likely occurred after 2 hpi because higher copy number of vRNAs could be detected in the cytoplasm at 4 hpi as compared to that at 2 hpi (Fig. 2D). In fact, the newly replicated vRNAs could be seen in the cytoplasm at 3 hpi (Fig. S3A) and these vRNAs were distributed throughout the entire cytoplasm. When the colocalization between the PB2 and NA vRNAs was analyzed at 4 hpi, low colocalization efficiency was observed (Fig. 2C). This suggested that the vRNAs were exported out of the nucleus individually and vRNAs of different segments were not traveling together at the early phase post export. At later stages, large number of vRNAs was observed in the cytoplasm. Accumulation of vRNAs at the peri-nuclear regions was observed around 6–7 hpi, which is consistent with previous reports [17] (Fig. S3B). At this time point, 60% of the NA vRNAs colocalized with the PB2 vRNAs (Fig. 2C). The colocalized spots were not only detected at the peri-nuclear regions where the vRNA accumulated but also spread out in the cytoplasm. These results suggested that the PB2 and NA vRNAs started to gather and travel together between 6–7 hpi (Fig. 2&S3). As infection progressed, more vRNAs were seen in the cytoplasm and higher colocalization efficiency between the PB2 and NA vRNAs was detected. At 10 and 12 hours post infection, a significant increase of the released virus particles was observed (Fig. S4) and the colocalized vRNAs were detected mainly in the cytoplasm and near the apical surface of the cells, which likely represent the vRNAs being packaged into the budding virions. The quantification of colocalization between the PB2 and NA vRNAs in the nucleus and cytoplasm at different hours post infection is shown in Figure 2B and 2C. In order to test whether the timing of association was specific to the PB2 and NA vRNAs or a feature common to all viral segments, we measured the colocalization of five other pairs of viral RNAs. All pairs shared similar kinetics with that for the PB2 and NA vRNAs, demonstrating that the temporal pattern of association is general (Fig. S5).
Influenza virus particles enter the cells through endocytosis. Acidification of the endosomes allows the viral envelope to fuse with the endosomal membrane; influx of protons into the virus particles releases the vRNPs from the matrix protein, freeing the vRNPs into the cytosol [1]. It is unclear whether the released viral RNAs stay together or travel individually to the nucleus. In order to understand this process in more detail, MDCK cells were infected with PR8 virus at MOI = 100 and the cells were treated with ammonium chloride (NH4Cl), leptomycin B (LMB) or importazole (IPZ), to disrupt different steps of influenza vRNP nuclear transport. The colocalization efficiency of PB2 and NA vRNAs was then analyzed at 20, 40 and 60 minutes post-infection. In the mock treated cells, both PB2 and NA vRNAs could be detected in the nucleus as early as 20 minutes post infection (Fig. 3A). Replication of vRNAs started around 40 minutes post infection because the average copy number of vRNAs detected in the nucleus were comparable at 20 and 40 minutes post infection, while a significant increase was detected at 60 minutes post infection (Fig. 3D). When the colocalization of PB2 and NA vRNAs were assessed during the first hour post infection, colocalization of vRNPs in the nucleus decreased over time (from approximately 30% to 5%) while colocalization of vRNPs remained around 50% in the cytoplasm (Fig. 3B&C). This shows that vRNPs imported into the nucleus became separated, and suggests that they replicated at different positions in the nucleus. While some PB2 and NA vRNAs were not colocalized in the cytoplasm at early time points post infection, it is possible that the released vRNPs depart from each other and travel individually to the nucleus or the non-colocalized vRNPs are molecules that shuttle back into the cytoplasm after nuclear import. First, to demonstrate that the colocalized PB2 and NA vRNPs detected in the cytoplasm in the early time points post infection were predominantly vRNPs from the same virion, we retained the vRNPs in the entering virions by treating the cells with 20 mM of ammonium chloride. Ammonium chloride increases the pH value in the endosomal compartments, preventing the release of vRNPs from the matrix proteins; therefore, the vRNPs were kept in the incoming virus particles. When the cells were treated with 20 mM ammonium chloride, most of the PB2 and NA vRNAs were colocalized in the cytoplasm throughout the first hour of infection (Fig. 3A); the colocalization efficiency between these two vRNA species was also significantly higher than that in the mock treated cells (Fig. 3C). These results indicated that the uncoating and fusion events of the virions took place prior to the separation of different vRNA segments and the colocalized PB2 and NA vRNAs were the ones co-packaged in the same virus particles. In comparison with mock treated cells, lower numbers of vRNA molecules were imported into the nucleus in ammonium chloride treated cells, indicating that ammonium chloride treatment prohibited nuclear import of most vRNAs (Fig. 3D). Some vRNAs were still able to escape the inhibition and enter the nucleus. Nonetheless, the nuclear PB2 and NA vRNAs detected in the ammonium chloride treated cells exhibited decreasing levels of colocalization over time and the colocalization efficiency was significantly lower than that for the cytoplasmic vRNAs (Fig. 3B&C). These data show that vRNPs become separated once they enter the nucleus and also argue that most of the colocalized PB2 and NA vRNAs detected in the cytoplasm are from the same incoming virions instead of vRNAs from different virions infecting the same cell.
To further understand if the vRNPs released from the virions become separated upon entry to the cytosol or nucleus, infected cells were treated with importazole (IPZ) to block nuclear import of vRNPs. IPZ has been found to interfere with the interactions between importin-β and Ran-GTP, an event critical for the release of imported cargos. Treatment with IPZ, therefore, inhibits the importin-β dependent nuclear import of cargo and retains the imported cargos at the rims of the nuclei [20]. When MDCK cells infected with PR8 virus were treated with IPZ, lower copy numbers of PB2 and NA vRNPs were imported into the nucleus as compared to those in the mock treated cells (Fig. 3D). However, high colocalization efficiency between PB2 and NA was detected in both cytoplasm and nucleus at 20, 40 and 60 minutes post infection. This result suggested that the disassembly of PB2 and NA vRNPs requires nuclear import and occurred after they are released from the nuclear import machinery (Fig. 3A–C). Since there are significantly more separated PB2 and NA vRNPs observed in the mock treated cells than in cells in which the nuclear import of vRNPs is abrogated, we hypothesized that the separated cytoplasmic vRNPs are those that shuttle back into the cytoplasm after nuclear import. We then tested this hypothesis by treating the infected cells with leptomycin B (LMB) which block the export of influenza vRNPs [10]. When the cells were treated with LMB, the copy number of PB2 and NA vRNPs imported into the nucleus was similar to that in the mock infected cells (Fig. 3D) and the colocalization efficiency between the two vRNPs was also comparable in the nuclei of LMB treated and mock treated cells (Fig. 3B). These indicated that the nuclear import of vRNPs in both LMB treated and mock treated cells shared a resembling kinetics. While Crm-1 dependent nuclear export was inhibited by LMB treatment in MDCK cells (Fig. S6), higher proportion of colocalized PB2 and NA vRNAs could be observed in the cytoplasm as compared to those in mock treated cells (Fig. 3A). Quantitative analysis also showed that the colocalization efficiency between PB2 and NA vRNAs was maintained at 70% in cells treated with LMB over time, which was significantly higher than that in the control cells (Fig. 3C). These results suggested that the separated PB2 and NA vRNPs found in the cytoplasm mainly consists of the vRNPs that re-entered the cytoplasm after nuclear import. These further imply that vRNPs are exported individually into the cytoplasm during the early phase of viral infection.
Since the results of colocalization analysis of PB2 and NA vRNPs at different time points post infection indicated that vRNPs of different identities colocalized in the cytoplasm, cellular factors that may be involved in this process were further investigated. It has been reported that influenza vRNA trafficking depends on microtubules, and it has been observed that vRNAs accumulated at the microtubule-organization center (MTOC) after their export from the nucleus [17]. Thus, we first analyzed whether microtubules are involved in the colocalization of different vRNP segments by looking at the colocalization of the vRNAs with the microtubule network. To quantify the degree of colocalized vRNAs associated with microtubules, a three-color colocalization analysis was performed. The infected cells were first stained for microtubules using an antibody against tubulin followed by smFISH against the PB2 and NA vRNAs. In Figure 4A, the microtubule is represented in blue while the two vRNAs were represented in green (PB2) and red (NA). If the colocalized vRNAs were associated with microtubules, white signals should be observed. At 4 hpi, the PB2 and NA vRNAs were not colocalized so distinct green and red spots can be observed. It is of note that these vRNAs, albeit not colocalizing, were seen in positions adjacent to the microtubules, suggesting that microtubules may support the transport of individual vRNAs after they were exported from the nucleus. At 6 hpi, the PB2 and NA vRNAs started to colocalize in the cytoplasm. At this time point, white signals were observed at the peri-nuclear region demonstrating the accumulation of PB2 and NA vRNAs at the MTOC. Except for the signals observed in the MTOC regions, the colocalized PB2 and NA vRNAs in the cytoplasm were not located on the microtubule network (Fig. 4A). This suggested that the colocalization of PB2 and NA vRNAs may not happen as they travel along the microtubules. To quantify whether the association between vRNAs is modulated by their location relative to microtubules, we automatically delineated the regions of the three dimensional image stack where the microtubules were located (Fig. S7). The colocalized vRNAs that reside within the contours of these regions were considered to be microtubule associated while the others were not. By counting the number of colocalized spots inside or outside of the contours, the percentage of colocalized vRNAs associated with microtubules can be determined. This analysis showed that small proportions (4–24.5%) of colocalized vRNAs were found to be associated with microtubules, regardless of the increased colocalization efficiency between PB2 and NA vRNAs at 6 and 8 hpi, suggesting that microtubules may not play a major role in the colocalization of different vRNPs (Fig. 4B). We used the same spatial analysis to compare the colocalization efficiency of PB2 and NA vRNAs associated with the microtubules with that of vRNAs that were not associated with the microtubules. No significance difference was observed between the two groups (Fig. 4C). This further indicated that the colocalization of PB2 and NA vRNAs is independent of microtubule association.
To further confirm that an intact microtubule network is not important for PB2 and NA vRNAs to colocalize, we inhibited the polymerization of microtubules with nocodazole in PR8 virus infected cells. MDCK cells infected with PR8 were treated with nocodazole at 1.5 hpi and the cells were then fixed for hybridization at 6 and 8 hpi. Cells were stained with the Alexa488-conjugated wheat germ agglutinin (WGA) to label the plasma membrane and Golgi apparatus. At 6 hpi, large numbers of vRNAs were observed to accumulate at the juxta-nuclear positions in mock treated cells while in nocodazole treated cells, the vRNAs were spread out in the cytoplasm and accumulated toward the lateral plasma membrane (Fig. 4D). This observation was consistent with previous reports showing that microtubules are involved in the apical transport of vRNAs [16]. However, colocalization between the PB2 and NA vRNAs could still be observed in both mock treated and nocodazole treated cells (Fig. 4D). Quantification of the colocalization efficiency between the two vRNAs also showed no difference between the mock treated and nocodazole treated cells (Fig. 4E). This confirms the results from the three-color colocalization analysis that the microtubule network is involved in the destination of trafficking vRNAs but is not involved in the process of vRNA gathering. This also implies that the movement of vRNAs to the MTOC is not essential for PB2 and NA vRNAs to find each other during their journey in the cytoplasm.
It has been reported that influenza vRNAs travel to the plasma membrane in a Rab11 dependent manner [15], [16], [17], [19]. The colocalization of vRNAs in the cytoplasm at late time points post infection may occur when the vRNPs are loaded onto Rab11 positive vesicles. We therefore tested whether the transport of vRNAs with Rab11 positive organelles is critical for the colocalization of PB2 and NA vRNAs. In Figure 5A, PR8 virus infected A549 cells were hybridized with probes against the PB2 and NA vRNAs and were immunostained against Rab11 proteins. Since the antibody against Rab11 specifically detects human Rab11 molecules, two-color smFISH and immunostaining against Rab11 were performed in virus infected A549 cells. It was observed that colocalization between PB2 and NA vRNAs exhibited slower kinetics in A549 cells compared to that detected in MDCKs (Fig. S8), therefore the spatial-temporal relationships between colocalized vRNAs and Rab11 were then determined at 6, 8 and 10 hour post infection (instead of 4, 6 and 8 hour post infection in MDCK cells). At 6 hpi, the PB2 and NA vRNAs were not colocalized nor did they spatially coincide with Rab11 particles. However, at 8 and 10 hpi, PB2 and NA vRNAs displayed a high colocalization efficiency (appearance of yellow puncta in the two color FISH images) and these colocalized vRNAs were found associated with Rab11 particles, giving rise to distinct white puncta when FISH and immunostaining images were merged together. This suggests that the colocalization of PB2 and NA vRNAs likely happens when the two vRNAs are loaded onto the same Rab11-positive vesicle. To further quantify the proportion of colocalized vRNAs that associated with Rab11, a three-color colocalization analysis similar to the one described for microtubules was performed. The results show that the portion of Rab11-associated vRNPs which colocalize increases from 13.2% at 6 hpi to 44.9% at 8hpi, corresponding to the increase in the colocalization efficiency between PB2 and NA vRNAs from 31% at 6 hpi to 58% at 8 hpi (Fig. 5B). Furthermore, we found that the colocalization efficiency of the Rab11-associated PB2 and NA vRNAs was significantly higher than that of non Rab11 associated vRNAs (Fig. 5C). These results suggest that the localization of vRNAs with Rab11 enhances the degree of colocalization between PB2 and NA vRNAs.
To further demonstrate the involvement of Rab11 in the colocalization of PB2 and NA vRNAs, A549 cells were transfected with GFP tagged Rab11 in either the wild type or dominant negative forms followed by PR8 virus infection. The degree of colocalization between PB2 and NA vRNAs was compared in cells that expressed the wild type Rab11 (WT-Rab11) and cells that expressed the dominant negative Rab11 (DN-Rab11). Figure 5D shows that PB2 vRNAs colocalized with NA vRNAs in WT-Rab11 transfected cells and the colocalized vRNAs were located to Rab11 as well. On the other hand, the cells transfected with DN-Rab11 show a disperse distribution of PB2 and NA vRNAs in the cytoplasm; no colocalization was seen between the vRNAs. The DN-Rab11 proteins were also diffused in the cytoplasm, similar to what was previously reported [15]. We observed a large accumulation of vRNAs in and around the nucleus in DN-Rab11 protein expressing cells; we adjusted the contrast of the displayed images to correctly represent the vRNA particles in the cytoplasm. When a quantitative analysis was performed, the colocalization efficiency of the PB2 and NA vRNAs in the cytoplasm was significantly higher in WT-Rab11 transfected cells than in the cells expressing DN-Rab11 (Fig. 5E). These results further demonstrate the importance of Rab11 in facilitating the colocalization of different vRNAs during influenza virus infection. Together the data show that vRNAs of different segments colocalize before they reach the plasma membrane and the Rab11 positive organelles may serve as a platform for the gathering of different vRNAs as they travel to the site of assembly.
The viral proteins hemagglutinin and M2, specifically the cytoplasmic tail of M2, have been reported to be involved in the assembly of budding virions [21], [22]. We therefore tested whether HA and M2 are involved in the colocalization of vRNAs during their travel to the plasma membrane. A recombinant PR8 virus that lacks the HA ORF, PR8-HA-GFP-HA virus, was generated in MDCK cells which constitutively express the HA protein (MDCK-HA). The HA ORF of this virus is substituted with the ORF of the GFP gene. To test the role of HA protein in the colocalization of vRNAs, MDCK cells were infected with either the wild type PR8 (WT-PR8) or PR8-HA-GFP-HA viruses followed by smFISH and colocalization analysis of the PB2 and NA vRNAs at 4, 6 and 8 hpi. Since the PR8-HA-GFP-HA virus lacked the HA ORF, no HA protein could be encoded and the involvement of HA protein in this process can then be assessed. The PR8-HA-GFP-HA virus showed similar kinetics in vRNA replication and vRNA export as compared to the wild type PR8 virus in MDCK cells (data not shown). Comparable kinetics for the colocalization between PB2 and NA vRNAs were also observed for WT-PR8 and the PR8-HA-GFP-HA viruses (Fig. 6A&B). These results demonstrated that the expression of HA protein was not critical for the colocalization of different vRNAs in an infected cell.
The role of M2 protein was assessed with a similar strategy. Two recombinant viruses in the PR8 virus background were generated [23]. The PR8-WSN-M contains the full-length M segment from the WSN virus, so both M1 and M2 can be produced during infection. The PR8-WSN-ΔM2 virus lacks the M2 ORF, so it is grown in MDCK cells that expressed the M2 protein. The colocalization efficiency between the PB2 and NA vRNAs was compared between MDCK cells infected with PR8-WSN-M and PR8-WSN-ΔM2 viruses. No significant difference was observed between the PR8-WSN-M and PR8-WSN-ΔM2 virus infected cells. These results suggested that the expression of M2 protein did not play an essential role in the colocalization between PB2 and NA vRNAs (Fig. 6C&D).
In this study, we have analyzed the disassembly and the subsequent assembly of influenza vRNPs segments in virus-infected cells using smFISH. We show that vRNPs of different segments remain associated after their release from the incoming virions and they travel as a package to the nuclear membrane. Newly synthesized vRNPs of different segments do not occupy the same space in the nucleus and they are likely exported individually into the cytoplasm because colocalization of the exported vRNPs of different segments is not observed during the early stages of infection. Different viral RNPs colocalize in the cytoplasm at later stages during infection (6–8 hpi in MDCK cells) and they are often found to be associated with Rab11-recycling endosomes. These results provide evidence that vRNPs belonging to different segments follow the same trafficking route and the selection for the correct combination of the eight vRNA segments likely takes place before the vRNPs reach the cell surface (Fig. 7).
The detection of vRNPs as early as 20 minutes post infection demonstrates the high sensitivity of the smFISH methodology, which allows analysis of the spatial relationship of the vRNPs during the viral entry process. When uncoating of the incoming virus particles was blocked by increasing the pH in the endosomal compartments, the PB2 and NA vRNAs were found to colocalize with very high efficiency. These results indicate that the colocalized vRNPs originate from the same virions, in accordance to the previous report that influenza virus particles package their eight different vRNAs with high efficiency [4]. It was suggested earlier that vRNPs of different segments travel as a package before nuclear import, because when multi-nucleated skeletal myofibers were infected with influenza viruses at low MOI, viral RNAs of different segments were observed to enter the same nucleus [24]. With smFISH analysis, quantification of the level of colocalization of different vRNPs became possible. When infected cells were treated with importazole, a recently discovered compound interfering with importin-β and Ran-GTP interactions [20], the vRNPs of different segments stayed colocalized. This implied that after the vRNPs were released from the virions, they remained associated as they travel to the nucleus. In addition, it was observed that imported vRNPs of different segments remained colocalized upon IPZ treatment. Since IPZ prohibit the release of imported cargo from the cellular nuclear import machinery, it is possible that the release of vRNP complex from the import machinery is necessary for the separation of different segments or the initiation of vRNA replication. Three-dimensional reconstruction of viral RNPs inside budding virions revealed an electron dense platform at the leading tip of the vRNPs, suggesting that there are interactions among vRNPs in this region [3], [5]. Recent studies using RNA mobility shift assays of in-vitro transcribed vRNAs suggested that specific RNA/RNA interactions among vRNAs exist and that they may participate in the process of genome selection [5], [25]. However, further investigation is required to understand the detailed mechanism by which the eight vRNPs are held together during their travel in the cell.
The replication of vRNAs of cytoplasmic negative-sense RNA viruses has been found to occur in virally induced compartments, such as Negri bodies for rabies virus [26], or inclusion bodies for vesicular somatitis virus [27] and Ebola virus [28]. However, no specialized compartments have been observed for the synthesis of influenza RNAs in the nucleus. Indeed, we observed that the newly synthesized vRNAs were distributed throughout the nucleoplasm, except the nucleoli, similar to what was shown for the distribution of nucleoproteins in the nucleus [9]. The observation of separated vRNAs of different segments may indicate that the nuclear import of vRNPs drove the disassembly of the vRNP complex and each segment was distributed to a different position for transcription and replication. It is also possible that newly synthesized vRNA quickly diffused from the site of replication where different vRNA segments are in close proximity. A polymerase-mutant virus would be helpful to further elucidate the behavior of vRNAs in the nucleus. Moreover, separated newly synthesized vRNPs were detectable in the cytoplasm at the time when accumulation of vRNPs and the increased efficiency of colocalization between two different vRNPs in the nucleus were observed. This arguably suggests that the progeny vRNPs are exported individually from the nucleus and that the RNP complexes of colocalized segments only form in the cytoplasm.
The Rab11 recycling endosome system has been shown to be involved in the transport of vRNPs from the peri-nuclear regions to the apical plasma membrane [15], [16], [17]. It has also been shown that vRNPs were actively transported on microtubules [14] and an intact microtubule network was required for the targeting of Rab11 recycling endosomes to the apical surface [16], [17]. In our study, we show that microtubules are dispensable for the colocalization of different vRNPs while they are necessary for the correct membrane targeting of vRNPs. The three-color smFISH analysis shows that the interaction of vRNPs with Rab11 increases the colocalization efficiency of different vRNA segments, demonstrating that Rab11 plays a role in facilitating the colocalization between different vRNA segments. Thus, it is suggested that the interaction of vRNPs with Rab11 concentrates the vRNPs of different segments onto vesicle membranes and enhances the possibility of different vRNPs coming together. In fact, a high degree of colocalization between PB2 and NA vRNA is usually detected in cells in which accumulations of vRNPs at the peri-nuclear regions are observed, a proposed location where vRNPs are loaded onto Rab11-positive vesicles [16], [17]. We therefore propose that Rab11 recycling endosomes serve as membrane platforms which help vRNPs of different segments to find their partners. It should be noted, however, that colocalized vRNPs which were not associating with Rab11 could be observed (Fig. 6B). This indicates that there may be other factors, such as the packaging signals of vRNAs, determining the colocalization between different vRNPs (independent of Rab11 interactions). It is also possible that these colocalized vRNPs were disassociated from Rab11 as they were ready to be packaged into the virus particles. Taken together, it is likely that the selection of different vRNA segments to be co-packaged into virions takes place during vRNP trafficking to the plasma membrane.
The efficient incorporation of influenza viral genome into virions has been linked to the cytoplasmic tail of the virus transmembrane proteins: HA, NA and M2 [21], [29], [30]. Mutant influenza A virus lacking the HA and NA cytoplasmic tail showed reduced viral RNA to viral protein content; it was also suggested that this mutant virus contain more non-infectious virus particles (without full complemented genome) than the wild type viruses [29]. In addition, it has been demonstrated that truncations or mutations in the cytoplasmic tail of the M2 protein lowered the level of virion associated NP proteins and vRNAs [21], [30], [31], [32]. This effect has been linked to the interactions between the M1 protein and the M2 cytoplasmic tail, suggesting that M2 is responsible for the recruitment of M1-vRNP complexes to the virion budding sites [31], [33]. With these lines of evidence, one would postulate that the assembly of vRNPs happens at the plasma membrane. However, our study using recombinant viruses lacking the entire HA or M2 protein indicates that the colocalization between different vRNP segments is independent of HA and M2 recognition (Fig. 6), further confirm other data in this present study demonstrating that the different vRNP segments assembled prior to their incorporation into budding virions. Since it has been shown that the majority of influenza viruses package the complete set of eight vRNA segments [4] and physically associated, well-ordered eight vRNPs are observed in budding virions [3], [5], [25], it is proposed that an interlinked complex of eight vRNPs formed before them being packaged [3], [34].Even though the formation of the super-complex cannot be detected with the current smFISH experiments, the colocalization between different vRNA segments detected in the cytoplasm suggested that the process of complex formation might occur during vRNA trafficking, providing an additional hint to the current hypothesis that the selected eight vRNPs might be packaged as a complex into budding virion.
In conclusion, we have shown with smFISH and colocalization analyses that different vRNA segments in influenza virus-infected cells colocalized in the cytoplasm before they reached the plasma membrane. These findings shed light on the selection process of influenza vRNPs packaging [3], [34], however, further investigations are required to fully identify the biological principles that govern the associations between the different vRNPs.
Madin-Darby canine kidney (MDCK) epithelial cells were maintained in Modified Eagle's Medium (MEM) (Gibco, Invitrogen) supplemented with 10% fetal bovine serum (FBS). Adenocarcinomic human alveolar basal epithelial cells (A549) were grown in Dulbecco's Modified Eagle's Medium (DMEM) (Gibco, Invitrogen) supplemented with 10% FBS. Both cell lines were incubated at 37°C with 5% CO2. Influenza A/Puerto Rico/8/34 (PR8) and PR8 cH9/1 virus strains were grown in 10-day old embryonic chicken eggs as previously described [35]. Recombinant PR8-HA-GFP-HA virus was grown in HA-MDCK cells [36]; PR8-WSN-M and PR8-WSN-ΔM2 viruses were grown in M2-complementing MDCK cell lines [23] (kindly provided by Dr. Randy Albrecht). Plasmids GFP-rab11 WT and GFP-rab11 DN were purchased from Addgene (Addgene plasmid 12674 and 12678) [37]. The following antibodies were used: a rat monoclonal anti-α-tubulin antibody (Abcam), a rabbit anti-Rab11 polyclonal antibody (Invitrogen), Alexa Fluor 488-conjugated donkey anti-rabbit IgG (H+L) antibody and an Alexa Fluor 488-conjugated goat anti-rat IgG (H+L) antibody (Molecular probes).
The reverse genetics method for generating recombinant PR8-HA-GFP-HA virus was as described previously [36]. The pDZ plasmid expressing GFP ORF flanked by the HA packaging signals was constructed as previously reported [38].
Infection of MDCK cells or A549 cells with influenza viruses was performed as previously described [39]. To synchronize virus entry, the cells were first incubated on ice for 5 min and then incubated with viruses diluted in infection media (phosphate buffered saline (PBS) supplemented with 1% bovine albumin (BSA) and 1% penicillin-streptomycin) on ice for 60 min. After virus adsorption, the cells were washed and the post-infection media (DMEM supplemented with 0.3% BSA, 1% penicillin-streptomycin, and 1 µg/ml TPCK (L-1-tosylamide-2-phenylethyl chloromethyl ketone)-trypsin) pre-warmed to 37°C was added immediately to the cells. The cells were then transferred to 37°C to allow virus entry. For cells treated with drugs during infection, the compounds were present in both the infection media and post-infection media at the following concentrations: 20 mM for ammonium chloride; 40 ng/ml for leptomycin B (Sigma); 31.8 µg/ml for importazole (Sigma) and 20 µg/ml for nocodazole (Sigma).
5×104 A549 cells were transfected with 1 µg of GFP-rab11-WT or GFP-rab11-DN plasmid using lipofectamine 2000 reagent according to manufacture protocol (Invitrogen). Cover slips were coated with fibronectin (Sigma) by incubating the cover slips in 50 µg/ml of fibronectin in PBS for 45 min at RT and rinse them once with PBS. The transfected cells were plated onto the fibronectin coated coverslips and incubated at 37°C for 24 hours before virus infection.
RNA fluorescence in situ hybridization (FISH) was performed according to published protocols with some modifications [40], [41], [42]. The probes used were single-stranded DNA oligos (20 nucleotides) each labeled with one fluorophore (Cy3 or Cy5) (BioSearch, Text S1). Cells were plated onto poly-lysine coated coverslips (BD Biosciences) at a density of 5×104 cells/well and grown overnight at 37°C. At certain time points post infection, the cells were washed once with ice-cold PBSM (1×PBS, 5 mM MgCl2), followed by fixation with 4% paraformaldehyde in PBSM for 10 min at room temperature (RT). After a brief wash with ice-cold PBSM, the cells were permeabilized with 0.5% Triton X-100 in PBSM for 1 min at RT. The cells were then washed with PBSM and incubated in 2XSSC (300 mM sodium chloride, 30 mM sodium citrate) with 10% formamide for 5 min before hybridization. To detect viral RNAs, 4 µM of labeled probes in 40 µl of hybridization buffer (10% dextran sulfate, 2 mM vanadyl ribonucleoside complexes (VRC, New England BioLabs), 0.02% RNAse-free BSA, 50 µg E. coli tRNA, 2XSSC, 10% formamide) was used for each sample. Hybridization was carried out in humidified chambers maintained at 37°C for 16 hours. The samples were then washed twice with 10% formamide 2×SSC supplemented with 2 mM VRC for 30 min at 37°C. Nuclear staining using 0.5 µg/ml of DAPI was performed afterwards and the coverslips were mounted in ProLong Gold antifade mounting media (Invitrogen). The cured samples were subjected to microscopy examination or were stored at −20°C. For samples used for both immunofluorescence and FISH analyses, cells were first blocked with 1% BSA in PBSM for 1 hour at RT after fixation and permeabilization. The coverslips were then subjected to primary and secondary antibody staining in 1% BSA in PBS followed by another fixation step with 4% paraformaldehyde for 10 min. The cells were then washed once with PBSM and equilibrated with 10% formamide 2XSSC for 10 min before the in situ hybridization procedures.
Cells were placed on a Zeiss Axioplan2IE microscope equipped with a 100×, 1.4 numerical aperture (NA) oil-immersion objective (Zeiss) and a Zeiss AxioCam MRm camera. Cells were imaged using 200 nm z-dimension axis steps across a range of approximately 4 µm.
We performed subpixel localization and intensity quantification of the spots using custom-designed MatLab (Mathworks) programs which were previously described [43]. We used a custom written Matlab program for colocalization quantification. After the center positions of fluorescence spots in each channel were identified in 3-dimensional space using the spot detection program, the distances between spots of one color with their nearest neighboring spots of the other color were calculated. Colocalized events were assigned when the distance between the closest spots of different colors were within 2.5 pixels (255 nm). The value of this distance threshold was determined by comparing the distances between the nearest spots of different colors detecting the same vRNA molecule and those detecting RNA molecules that do not colocalized. The 2.5 pixels (255 nm) range was used to maximize detection of colocalization events while minimizing false positives (see Fig. S2 for further explanations). Within each image, colocalization between spots was separately quantified in various regions of interest (such as nucleus and cytoplasm) using 2D and 3D mask-images. Mask Images were generated in various ways. Individual cells were manually delineated using a plugin written for the Image J program (NIH). Nuclei were segmented using an automatic intensity threshold of a maximum intensity projection of the DAPI image stack. The microtubules and Rab11 immunofluorescence image stacks were subjected to a custom-written automated segmentation tool to generate 3D mask image stacks for colocalization analysis. Briefly, the 3D stack was first smoothed using a bandpass filter before applying an automated threshold using Otsu's Method. The result was a binary 3D image stack in which white voxels corresponded to microtubules (resp. Rab11 particles) while the rest of the 3D space was left black. Microtubules image stacks were deconvolved prior to applying the segmentation algorithm using AutoQuant X2 AutoDeblur software (Media Cybernetics). Custom written software is available upon request.
|
10.1371/journal.pcbi.1000971 | Endothelial Cell Capture of Heparin-Binding Growth Factors under Flow | Circulation is an important delivery method for both natural and synthetic molecules, but microenvironment interactions, regulated by endothelial cells and critical to the molecule's fate, are difficult to interpret using traditional approaches. In this work, we analyzed and predicted growth factor capture under flow using computer modeling and a three-dimensional experimental approach that includes pertinent circulation characteristics such as pulsatile flow, competing binding interactions, and limited bioavailability. An understanding of the controlling features of this process was desired. The experimental module consisted of a bioreactor with synthetic endothelial-lined hollow fibers under flow. The physical design of the system was incorporated into the model parameters. The heparin-binding growth factor fibroblast growth factor-2 (FGF-2) was used for both the experiments and simulations. Our computational model was composed of three parts: (1) media flow equations, (2) mass transport equations and (3) cell surface reaction equations. The model is based on the flow and reactions within a single hollow fiber and was scaled linearly by the total number of fibers for comparison with experimental results. Our model predicted, and experiments confirmed, that removal of heparan sulfate (HS) from the system would result in a dramatic loss of binding by heparin-binding proteins, but not by proteins that do not bind heparin. The model further predicted a significant loss of bound protein at flow rates only slightly higher than average capillary flow rates, corroborated experimentally, suggesting that the probability of capture in a single pass at high flow rates is extremely low. Several other key parameters were investigated with the coupling between receptors and proteoglycans shown to have a critical impact on successful capture. The combined system offers opportunities to examine circulation capture in a straightforward quantitative manner that should prove advantageous for biologicals or drug delivery investigations.
| In this work we have investigated the role of a family of cell surface molecules, proteoglycans, in blood vessel capture of proteins important to normal and diseased states under flow conditions. We developed a computer model to analyze and predict these events and, using an experimental system incorporating endothelial-lined hollow fibers as model blood vessels, tested our predictions. We found that both proteoglycans and flow exert significant influence over growth factor binding to the vessel wall. Removal of proteoglycans significantly reduced binding of these proteins; and flow rates slightly higher than that seen in capillaries had a similar effect, albeit in a different way. This knowledge will increase our understanding of interactions inside blood vessels and help to design more efficient pharmaceuticals. Also, our computer model has the potential to test the ability of existing and future drugs and biologics to successfully target blood vessels.
| The bioavailability of molecules as they circulate through the bloodstream is a crucial factor in their signaling capability. Half-life in circulation can determine the effectiveness of a drug simply by regulating the opportunities a molecule has to interact with the vessel wall. Although in vivo measurements are routinely made by researchers to monitor serum levels of molecules and to determine half-lives, interactions in the microenvironment are not easily measured or observed. While some molecules may have a long circulation life, many may have only a single opportunity to interact with the blood vessel walls before being filtered through the liver or kidneys. In addition, even molecules with a long circulation life may still face impediments to direct interaction with the endothelium. This, for example, is the case with vascular endothelial growth factor (VEGF) when bound to bevacizumab, a monoclonal antibody to VEGF [1], [2]. Bevacizumab has been shown to increase the circulating concentration of VEGF in cancer patients when compared to patients not undergoing therapy because of the increased half-life of the growth factor-antibody complex; however the complex is unable to bind to VEGF receptors [3] making delivery of the VEGF questionable. In order to better understand the vessel microenvironment and to accurately monitor drug interactions in the context of that microenvironment, better tools are needed to provide meaningful measurements that can predict the fate of molecules in circulation.
Many important measurements have and continue to be made using in vitro mammalian tissue culture methods but there are obvious limitations to the traditional two-dimensional culture approach. In circulation, the influence of flow on whether a molecule remains in the fluid phase or binds to the vessel wall can be a dominant factor. This influence cannot be ascertained in static tissue culture studies. For example, the velocity of blood in the aorta is ∼400 mm/sec while at the capillary level it is less than 1 mm/sec [4]. This reduction in velocity allows the exchange processes at the capillary level to take place more efficiently [4] and it likely also affects the activity of molecules in circulation that rely on cell surface binding in order to fulfill their roles. While direct measurement of this binding process is difficult, our model makes use of a commercial bioreactor with endothelial-lined hollow tubes operating under pulsatile flow to mimic the vascular environment architecture and to directly measure the loss of molecules as they pass through these hollow fibers. We have used a single pass method to allow better assessment of the effect of flow in either retaining molecules in the circulation or permitting their interaction with vessels. Our approach also makes use of a bolus administration, since this is a typical way in which drugs would be delivered in a clinical setting.
The binding of fibroblast growth factor-2 (FGF-2) to its cell surface receptor (FGFR) and the role of heparan sulfate proteoglycans (HSPG) in regulating the process have been of research interest for many years because of their role in angiogenesis, the growth of new blood vessels from existing vessels. Knowledge of how these processes work could aid in the development of new therapeutics to control tumor growth and assist clinically in the treatment of chronic wounds. In order to understand the mechanism of FGF-2-mediated cell proliferation, a multitude of experimental studies have been undertaken [5] and, in the past two decades, several computational models of FGF-2 binding to its receptor FGFR and HSPG have been proposed [6]–[11]. Insight can be gained through experiment-coupled modeling that could not otherwise be readily obtained. Nugent and Edelman [11] were among the earliest researchers to develop a simple model that includes three species, FGF-2, FGFR and HSPG. They measured kinetic binding rate constants experimentally and used their model to analyze the data thereby providing a foundation for investigating the complexity of FGF-2 binding. A similar approach was used by Ibrahimi et al [9] to investigate stepwise assembly of a ternary FGF-2-FGFR-HSPG complex in conjunction with their surface plasmon resonance measurements. We introduced more complexity into the FGF-2 binding model with the inclusion of heparin binding [12], receptor dimerization [8], and formation of alternative HSPG-FGFR species [13]. Recent models have moved towards including intracellular signaling [14]. With the exception of work by Filion and Popel [7], [15], which included diffusive transport, previous simulation work has been based on a static tissue culture environment that may be quite different from the dynamic in vivo environment of blood vessels.
We introduced a computational model based on a flow environment in which the competitive binding of FGF-2, FGFR, and HSPG in a pulsatile flow environment was addressed to mimic blood vessel-like hollow fibers [16], [17]. In this paper we use an updated version of that model to explore how specific parameters such as flow rate impact FGF-2 capture and receptor binding, and compare our results with experimental studies. Insights with regard to the importance of surface coupling and ligand depletion zones within the fluid phase were found. The described simulation package provides a new and valuable way to investigate growth factor capture and can be easily extended to other biologically relevant molecules and drugs.
BAECs (passage 10), cryopreserved in liquid nitrogen, were cultured in Dulbecco's modified Eagle's medium (DMEM-low glucose, phenol red-free, Invitrogen Corporation, Grand Island, NY), supplemented with penicillin (100U/mL, Invitrogen Corporation, Grand Island, NY), streptomycin (100µg/mL, Invitrogen Corporation, Grand Island, NY), glutamine (2mM, Invitrogen Corporation, Grand Island, NY), and 5% newborn calf serum (Invitrogen Corporation, Grand Island, NY). When a sufficient number of cells were grown (passage 11∼13), they were transferred to the hollow fiber cartridge.
The FiberCell polysulfone plus endothelial cartridges (C2025, FiberCell Systems Inc., Frederick, MD), also called hollow fiber bioreactors, contain 20 capillaries which are 12 cm long, 700 µm I.D., 300 µm wall, 0.1µm pore size, 53 cm2 lumen surface area (Figure 1A). They were activated with 70% ethanol (Fisher Scientific, Houston, TX), followed by multiple washes with sterile distilled water. The cartridges were then coated using 5 µg/mL fibronectin (Sigma Aldrich, St. Louis, MO) in phosphate buffered saline (PBS, Invitrogen Corporation, Grand Island, NY). BAECs (passage 11∼13) were inoculated into the cartridges (0.7–1×107 cells/cartridge) 24 hours after the coating and placed in an incubator for 4 hours (rotated 180° after 2 hours) without flow in order to promote cell attachment. The BAEC culture cartridges were then linked to the FiberCell pump system (FiberCell Systems Inc., Frederick, MD) and media circulated through the system at ∼2.6 mL/minute (5.2 mm/sec). The flow system was maintained in the incubator (37°C, 5% CO2) at all times except during the experiment periods. Cell growth and viability was monitored by measurement of the cell glucose consumption from the medium once a day with OneTouch UltraSmart blood glucose monitoring system (Lifescan, Inc., Milpitas, CA).
The flow system and cell-lined cartridges were removed from the incubator, gently washed once with warmed (37°C) PBS (60 mL), and then maintained in circulating 125 mL serum-free medium (DMEM-low glucose, phenol red-free, supplemented with 0.05% gelatin in PBS) in a sterile room-temperature tissue culture hood (Thermo Scientific, Waltham, MA). After establishing flow at the desired rate (low rate: 0.60∼0.68 mL/min (1.2–1.36 mm/sec); high rate: 1.6–1.8 mL/min (3.2–3.6 mm/sec) or 2.9–3.0 mL/min (5.8–6.0 mm/sec)) with a CellMax Quad pump (Spectrum Laboratories, Inc.) for about 2 minutes, flow was stopped to allow the growth factor of interest (FGF-2 (Sigma Aldrich, St. Louis, MO), EGF (R&D Systems Inc., Minneapolis, MN) and VEGF (R&D Systems Inc., Minneapolis, MN)) (0.11 mL) to be injected into the inlet. After the injection, the flow was resumed and the flow media collected (two drops/fraction) for the desired time period. The flow pattern was assumed to be sigmoidal based on previous studies [18], [19]. The cartridges were then gently washed with warmed PBS supplemented with 0.3M NaCl (10 mL) followed by one wash with 10 mL PBS and a wash of the whole flow system with PBS (60 mL). The system was returned to the same culture media and flow rates as described under Preparation of BAECs, allowing at least 24 hours before the next experiment. The media fractions collected during the binding experiments were stored at 4°C and analyzed with ELISA kits (R&D Systems Inc., Minneapolis, MN) within the next 24∼48 hours.
Dynamic viscosity of the test cell culture medium was measured using a DV-II++ Pro Programmable cone-plate viscometer (cone #CPE-40; Brookfield Engineering Laboratories; Boston, MA) according to the manufacturer's instructions. Viscosity measurements were made for a range (375 to 750 sec−1) of shear rates (to confirm Newtonian fluid behavior) at room (i.e., 25°C) and physiologic (i.e., 37°C) temperatures.
Heparan sulfate expression was measured in static tissue culture dishes and in the flow cartridge by heparinase treatment of cells, collection of the cleaved glycosaminoglycans, and quantitation using a dimethylene blue colorimetric assay [20], [21]. Cells in static culture contained 4.3+/−0.31×10−6 µg of heparan sulfate/cell and cells in cartridge hollow fibers contained 1.1+/−0.09×10−6 µg of heparan sulfate/cell, reflecting an ∼75% reduction in cell surface heparan sulfate under flow (0.63 mL/min (1.26 mm/sec)).
Heparinase III (0.01 unit/0.11mL, Seikagaku Corp., Japan; 0.2unit/0.11mL, Sigma Aldrich, St. Louis, MO), chondroitinase ABC (0.2 unit/0.11mL, Seikagaku Corp., Japan) and keratanase (0.33unit/0.11mL, Sigma Aldrich, St. Louis, MO) were utilized to observe their effect on growth factor flow and binding. In some experiments, the enzymes (heparinase III, chondroitinase ABC and keratanase) were mixed together as an enzymatic cocktail solution at the above concentrations. Cartridges were treated for 20 minutes at 37°C, washed with warmed PBS (10 mL), and growth factor studies performed as described above.
Non-specific binding of FGF-2 in the system was determined to be primarily due to the inlet reservoir. The reservoir chamber was removed from the cartridge, growth factors were injected into the inlet of the cartridges with a syringe, and flow was initiated. Fractions were collected as they exited the reservoir. Growth factors were measured before injection and compared to the sum of the collected fractions. The difference between the input amount and the amount collected constituted the nonspecific binding in our experiments. For FGF-2 (1.0+/−0.1 ng), the amount retained in the reservoir was 29+/−2.8% of the FGF-2 added (SD, n = 3). Additional nonspecific binding within the hollow fibers was assumed to be minimal.
The concentrations of FGF-2, EGF, and VEGF in the collected fractions were measured by ELISA. The flow rate of each experimental run was determined from the total volume collected divided by the total flow time.
To visualize the BAECs cultured in the flow system, cartridges were washed with PBS supplemented with 0.5M NaCl to extrude the endothelial cell lining from the hollow fibers and then the cell linings were fixed with 4% paraformaldehyde (Electron Microscopy Sciences, Hatfield, PA) in PBS for 10 minutes. Three washes with PBS (one minute per wash) followed and the cell linings permeabilized with PBS supplemented with 0.03% Triton and 1% BSA for 3 minutes on a shaker platform at room temperature. The cells were then treated with 10 µg/mL 4′, 6-diamidino-2-phenylindole (DAPI) (Sigma Aldrich, St. Louis, MO) in PBS supplemented with 0.03% Triton and 1% BSA for 20 minutes, followed by three PBS washes for 2 minutes each at room temperature. The cells were then visualized and photographed using a Nikon Eclipse TE 2000E fluorescent microscope (Nikon, Melville, NY) at an excitation wavelength of 350 nm (Figure 1B).
The computational model is based on the physical dimensions of the bioreactor although the system is scalable to other desired dimensions. The domain of the simulation is the hollow-fiber portion of the cartridge (Figure 1). The computational model has three coupled parts: (1) the medium flow equations; (2) the convective mass transport equations of growth factor in the flow; (3) the binding kinetics equations on the wall of the fibers [8], [16].
In order to solve the coupled equations numerically and efficiently, the following assumptions are made: (1) the walls of the hollow fibers are rigid and nonporous; (2) the flow is axisymmetric and laminar; (3) the fluid is incompressible, Newtonian and isothermal; (4) all of the hollow-fiber capillaries within the cartridge have the same dimensions, flow rate, cell densities and entrance conditions; and (5) the cells are packed tightly and distributed evenly on the wall of the hollow-fiber capillaries. Entrance effects of the flow are ignored [22], [23] and, consequently, the flow within the fibers is treated as fully developed flow in which the radial velocity is neglected. A uniform mesh is used. The kinetic pathways are shown in Figure 2 and the equations and parameter values are included in Tables 1 and 2, respectively.
In our experimental system, FGF-2 is injected into the inlet reservoir where it is assumed to quickly reach a uniform concentration. The concentration of FGF-2 in the reservoir is assumed to decrease gradually as fluid is pumped into the reservoir prior to distribution into the capillaries with each pulse cycle as:where is the volume of the reservoir, is the volume of fluid flowing into the fibers at each pulse, is the current and is the previous concentration of FGF-2 in the reservoir. , where is the amount of FGF-2 injected. The pump pulse cycle was measured experimentally and determined to be ∼36 strokes/min at a flow rate of 1.4 mm/sec.
Pulsatile flow is treated in the following manner. A pulse of fluid volume enters the pre-pump inlet reservoir (0.4 mL volume), from which a continuous flow of fluid having an axial velocity greater than or equal to zero enters the cell-lined fibers in the cartridge. The axial velocity is oscillatory but with only positive terms. Entrance effects are considered negligible [23]. The velocity of the fluid in the axial direction is determined with the following formula [17]:where qs is the average volumetric flow rate, Nf is the number of fibers inside the cartridge, R is the radius of a fiber, ω = 2π/T is the angular frequency of the pulsatile flow, and T is the pump pulse cycle.
Good agreement between the simulation and experimental results was determined based on two criteria: an amount criterion and a curve-matching criterion. The amount criterion is defined as:where Mexp is the outflow amount of protein determined experimentally, Msim is the outflow amount determined within the simulations and M is the amount of FGF-2 entering the capillary. The curve-matching criterion is calculated in the following way. The FGF-2 exit profile curve is not a continuous curve but is a series of discrete values at different time intervals. This makes use of traditional curve matching algorithms difficult. Our method aligns the initial exit times for the simulations and experiments and then calculates the distance between points on the two outflow curves using the following formula:where N is the total number of time intervals. and is the amount of FGF-2 exited at the ith time interval in experiment and simulation, respectively. The curve-matching criterion is defined as:A special program written in C/C++ that operates under Windows XP or Vista operating system has been built for solving this model and has been described previously [16], [17]. The interface allows users to easily set parameters related to the simulation such as FGF-2 injected concentration, flow rate, mesh size, time step, and total simulation time via either configuration text files or from the computer interface. The mass transport of FGF-2 within the fiber is visualized in real time during the simulation process. A Linux version of the software is also available however it lacks a user interface tool and there is no real time visualization. The binary code can be downloaded from www.cs.uky.edu/~czhanb/research.html.
In the simulations there are 800,000 cells/fiber or 16,000,000 cells/cartridge, a value which was obtained from the experimental system. The tolerance for solving the mass transport PDEs was set at 10−12. The relative tolerance for solving the kinetic ODEs was set at 10−8 and the absolute tolerance was 10−12.
All experiments were performed a minimum of three times in independent cartridges. The mean of all replicates ± standard deviation of those replicates is presented except where discrete measurements were used to more closely represent small changes in initial concentration. Significance (p<0.05) was determined using a Student t-test with a two-tail distribution and unequal variance (Excel, Microsoft).
Endothelial cells line blood vessels and are the initial entry point for access of blood-borne proteins to the underlying tissue. Our investigations focused on flow and the impact it has on endothelial cell capture of growth factors, which are important regulators of cell and tissue activity. To better approximate the microenvironment of a blood vessel, we seeded bovine aortic endothelial cells into the FiberCell cartridge system and cultured the cells under flow (Figure 1A). Cell viability was confirmed for up to 8 weeks and cell density was ∼0.3×106/cm2. The geometry is clearly more similar to in vivo than typical cell culture dishes but it was important to obtain a uniform and confluent monolayer of cells within the cartridge system to correctly perform and analyze experiments. To confirm this, cartridges were treated with a high salt wash to extrude the cell-based vessel and the cells were fixed and imaged (Figure 1B). An incision was made at one end to expose the lumen and demonstrate the continuity of the cell layer.
The average fluid velocity in human capillaries is <1 mm/sec [4]. We hypothesized that capture of regulatory growth factors from solution would be significant at these flow rates thereby facilitating growth factor activity. Using the lowest velocity setting with the standard pulsatile pump included with the Cellmax system (∼1.3 mm/sec, ∼0.65 mL/min), FGF-2 (5.0±0.4 ng) was injected into the cartridge inlet reservoir and flow was commenced. As shown in Figure 3, there is a delay in FGF-2 appearance in the outflow corresponding to the time for FGF-2 to travel through the cartridge and exit the system. The majority of FGF-2 added exited the cartridge as a large peak approximately 1 mL (or 1.5 min at this flow rate) after flow was initiated. Non-specific binding within the injection cartridge reservoir was measured directly (31+/−2.5%). Specific binding within the cell-lined hollow fibers accounted for 9+/−2.5% of total FGF-2 added to the cartridge at this concentration and ∼13% of the FGF-2 entering the cell-lined fibers, after taking into account non-specific binding (Figure 3). The results shown in Figure 3A are from three independent experiments conducted using three different cartridges illustrating the reproducibility of the system. Repeat runs conducted using the same cartridge as well as runs using radiolabeled FGF-2 instead of unlabeled FGF-2 both produced similar results (data not shown). The peak appearance time or volume in the outflow from the cartridge was insensitive to FGF-2 injection concentration in the range studied (data not shown). However, the size of the FGF-2 peak correlated with the injection concentration with the highest peak corresponding to the highest concentration of FGF-2 added (Figure 3B).
The accuracy of our measurements took into consideration specific losses that occurred with injection (i.e. tube, syringe, needle, and reservoir). Rather than averaging datasets with variable FGF-2 reservoir values, we therefore present them as discrete results. A plot of total FGF-2 retained at these discrete concentration points shows a dose responsive binding curve, reflecting the linear portion of the binding curve expected at sub-saturation ligand concentrations (Figure 3C).
Heparan sulfate proteoglycans (HSPG) are ubiquitous molecules found on virtually all cells including endothelial cells and have been shown to regulate heparin-binding growth factor binding and activity in tissue culture [6], [24]–[28]. FGF-2 is a heparin-binding molecule associated with a number of physiologic and pathologic processes [29] and, therefore, the role of HSPG in regulating FGF-2 retention under flow was examined. Although the binding affinity of FGF-2 for HSPG has been shown to be lower than the affinity for the FGF receptor, these HSPG sites can provide up to a thousand fold more binding sites for FGF-2 [6], [24] significantly impacting the cell binding “potential” for heparin-binding growth factors. Cartridges were treated with heparinase, an enzyme specific for heparin and heparan sulfate, and FGF-2 outflow quantified. After heparinase treatment, FGF (∼1 ng) was injected and pumped through the cartridge. Almost 74% of the total FGF-2 added to the system was recovered in the outflow, compared to ∼46% of the total FGF-2 recovered from the non-heparinase treated cartridge prior to subtraction of non-specific binding. The amount of FGF-2 retained in the cartridge after heparinase treatment corresponded to the measured level of non-specific binding and thus indicated no specific binding to cell-lined fibers in the absence of HSPGs (Table 3). In contrast, 25% of the FGF-2 pumped through untreated cartridges was retained after subtraction of non-specific binding. Although FGF-2 can bind to its receptor in the absence of HSPG stabilization, that binding, based on the apparent KD of the receptor for FGF-2 in the absence of heparan sulfate, the lower level of FGFR generally found, and the ligand-receptor exposure time under flow, would be expected to be at least ten-fold lower than in the presence of HSPG [24] and our data certainly support this.
To ensure that the effect with heparinase under flow was due to the specific removal of heparan sulfate and not a general effect due to enzymatic treatment of the cartridge or the enzyme incubation process, the cartridges were treated with keratanase, an enzyme having no specific known target on these cells. Keratanase, as opposed to heparinase, had no significant effect on FGF-2 retention (Table 3). Interestingly, there was a small but reproducible reduction (∼9%) after chondroitinase treatment on FGF-2 retention compared to control. Chondroitin sulfate proteoglycans are typically found on vascular surfaces but FGF-2 has not been shown to bind directly to chondroitin sulfate [30], [31]. It is not known at this time what the cause for the reduced binding is, although it has been reported that both chondroitin sulfate and dermatan sulfate under certain circumstances are able to influence FGF binding [32]–[34].
VEGF, a heparin binding protein, and EGF, which does not bind heparin, were next tested in this system. Both the initial appearance time and outflow volume for the protein as well as the general shape of the outflow peak for both VEGF and EGF were similar to FGF-2 (Figure 4). To ensure that the measured effects seen with heparinase-treatment on FGF-2 retention were due to specific responses of the growth factor to the removal of heparan sulfate and not a general response by all proteins, flow studies were done with VEGF and EGF following enzymatic treatment. EGF retention and outflow were unaffected by treatment with a cocktail of heparinase, chondroitinase, and keratanase (Table 4). Treatment with heparinase without chondroitinase or keratanase also had no effect on EGF retention or outflow (data not shown). In contrast, VEGF showed a significant decrease in specific retention between control and heparinase treated cartridges (16+/−5.8% versus −2.5+/−6.1% VEGF retained) indicating the critical role HSPG can have in heparin-binding growth factor capture under flow. The lack of a change in EGF binding or outflow profile under heparinase treatment is supportive that there are no gross changes in the cell glycocalyx that might impact the shear stress in the system.
Capture of FGF-2 by endothelial cells within the vasculature is a critical step in growth factor activity and our bioreactor is an excellent tool for investigating the capture process. However, it has limitations with regard to quantification of cellular binding behavior. The cartridges are expensive for short-term experiments and culture time and preparation can be relatively lengthy. Visualization of individual cell behavior within the culture is not feasible. In addition, the ability to predict the capture of molecules by cells under flow has value across a wide range of areas and the development of a flow-based tool for the design and testing of mechanisms related to retention is desireable. Our computer model was designed based on media flow equations and mass transport equations [35] with cell surface reaction equations to reflect the cell-growth factor interactions (see Materials and Methods-Model development). To validate the model, simulations were performed using the variables (ie FGF input concentration and flow rate) specific for an experimental series and a comparison was made. Experimental trials were run in which FGF-2 (0.92 ng) was added to the reservoir, pumped through the cartridge, and outflow collected and analyzed for FGF-2. FGF-2 in the outflow showed a characteristic peak outflow approximately 100s after flow was initiated at 0.63 mL/min (1.26 mm/sec) and 17±6.3% of the input FGF-2 was retained within the cartridge after non-specific binding was subtracted (Figure 5). Simulations performed using the same input FGF-2 value and flow rate were run and comparison was made between the simulations and experimental outflow from control (Figure 5A) or heparinase-treated (Figure 5B) cartridges. We defined good agreement based on two criteria; the amount of FGF-2 recovered and the curve similarity. Criteria one requires the relative difference in FGF-2 outlow from the experimental and simulation studies to be less than 1% while the second criteria compares the actual amounts of FGF-2 exiting from the experimental and the simulation system (see Materials and Methods). We did note that FGF-2 retention with the simulations was very dependent on the level of HSPGs with higher densities resulting in too much retention via HSPG-FGF-2 binding and subsequent FGFR coupling while lower HSPG densities resulted in too little retention (data not shown). Comparison of simulation results with our heparinase-treated data showed fine agreement with regard to our criteria when non-specific loss in the reservoir was subtracted.
Capillary flow is generally steady, and gradually becomes pulsatile at higher flow rates. We conducted simulations and in vitro experiments to compare steady and pulsatile flow at a low flow rate (0.6 mL/min, 1.2 mm/sec) to determine whether our model would predict differences between FGF-2 interactions using steady and pulsatile flow. Simulations predicted no difference in FGF-2 binding at low flow using pulsatile flow conditions versus steady flow in either the FGF binding down the cell-lined hollow fiber (Figure 6A) or in the profile of the outflow (Figure 6B). In vitro experiments were performed using a syringe pump for steady flow and the bioreactor's pulsatile flow pump (Figure 6C). FGF-2 outflow measurements indicated no overall change at 0.6 mL/min (1.2 mm/sec) suggesting that, at low rates typical of capillary flow, no significant change in FGF-2 interactions takes place.
Our experimental system does not allow easy separation between internalized FGF-2 and that bound to the cell surface or visualization of FGF-2 distribution within the cell-lined hollow fiber. Using our computer model we examined how FGF-2 would be distributed with respect to time after flow was initiated (Figure 7). At a relatively low flow rate (0.63 mL/min, 1.26 mm/sec), the FGF-2 in the reservoir had essentially all entered the hollow fibers by 150s and the peak outflow of FGF-2 was evident ∼200s after flow was initiated corresponding to the time when the bulk FGF-2 had exited the hollow fibers. Later times showed cell-bound FGF-2 either internalized or dissociated from the cell surface with little chance to reassociate. The vast majority of binding is predicted to occur near the entrance to the cell-lined hollow fibers as opposed to the middle or end of the fibers (Figure 7B). The impact of time was more pronounced in the front section also as fluid entering the hollow fiber after ∼150s was devoid of FGF-2 (<0.1% of initial FGF-2). Increasing the diffusion rate for FGF-2 in solution by increasing the diffusion coefficient by an order of magnitude is predicted to have a negligible impact on FGF-2 capture in the front of the capillary but increased significantly the FGF-2 bound down the length of the cell-lined hollow fiber. This was due to changes in the depletion zone near the cell-lined walls (Figure 8). After 44s, an FGF-2 depletion zone near the surface was evident which was reduced when the diffusive transport of FGF-2 was increased. The replenishment of FGF-2 near the wall promoted greater FGF-2 binding as complex formation is a second-order process and illustrates the importance of surface depletion in growth factor capture.
Our simulations indicate that depletion near the cell surface impacts binding and suggests that residence time in the vicinity of the cell surface is important. We therefore looked at how flow impacted cell binding of FGF-2. Simulations predict that cell binding is significantly diminished with increased flow rate (Figure 9A) although the basic result of high binding at the entrance and reduced binding down the cell-lined hollow fiber was consistent across flow rates examined (data not shown). This difference was evident regardless of the concentration of FGF-2 introduced to the system with the difference being more pronounced at higher flow rates (Figure 9B). Reduction in binding due to the loss of HSPG is less evident at higher flow rates where the specific binding was already greatly reduced. This inverse relationship between flow and cell binding is potentially important especially at these relatively low flow rates. The highest rate used in our simulations (∼3 mL/min,∼6 mm/sec) is considerably lower than average arterial flow rates (100–400 mm/sec) in larger vessels of the circulatory system [4] suggesting that, with a short half-life, retention may be relevent only in small vessels with lower velocities. Note that simulations were run to a constant time rather than volume to reduce small fluctutations in retained FGF-2 due to dissociation effects.
Experimentally, we found results that were consistent but not quantitatively exact with this model prediction (Table 5). FGF-2 retention in the hollow fibers was virtually eliminated under medium (∼1.7 mL/min, 3.4 mm/sec) and higher flow rates (3.0 mL/min, 6 mm/sec), a significant reduction compared to binding at 0.62 mL/min (1.24 mm/sec) (Table 3- control group). The simulations, in contrast, did show some level of binding even at the highest level but this likely reflects the idealized conditions used for the model system (i.e. uniform receptor and HPSG densities, free access to coupling between FGF-2 bound molecules). Heparinase treatment showed no significant further reduction in retention at the higher flow rates in agreement with the simulation results.
Simulations indicated no difference in FGF-2 binding under our pulsatile flow conditions versus steady flow (data not shown). Additional experiments were performed using a syringe pump with steady flow rather than pulsatile flow. FGF-2 outflow measurements indicated no overall change at 0.62 mL/min (1.2 mm/sec) (data not shown). Qualitatively the experimental results agreed with the simulation predictions for the overall effect of flow rate on retention although the model suggested higher retention levels for the control case and closer agreement between control and heparinase at both higher flow rates.
FGF-2 binding affinity and concentration, along with binding partner density, regulates the capture process for FGF-2 from the fluid phase. We therefore examined using our simulations how varying the affinity of FGF-2 for either HSPG (Figure 10A) or FGFR (Figure 10B) while holding all other parameters at their baseline value would impact retention. Decreasing the affinity (i.e. increasing KD) for HSPG had a dramatic effect on retention reducing it to 40% of baseline capture at the lowest value examined. The association rate constant had a greater impact than the dissociation rate constant although both followed similar trends. Somewhat surprisingly, increasing the affinity of the interaction by reducing the value of the dissociation rate constant of FGF-2 for HSPG did not alter FGF-2 binding likely due to the strong coupling present between FGFR and HSPG in the presence of FGF-2, making strict HSPG-dissociation somewhat irrelevant. For the same reason, FGF affinity for FGFR did not have a strong impact on FGF-2 capture since the vast majority of FGF-2 interacting with FGFR was via FGF-2-HSPG coupling.
Cells typically express significantly more HSPG than FGFR and we next asked how varying the cell surface densities of these binding sites would impact FGF-2 capture. In the absence of FGFR, a typical density of HSPG in our cartridge (2.5×105 #/cell) resulted in significant binding of FGF-2 in the absence of FGFR that is essentially doubled when FGFR density is 1×106 #/cell, a two-fold increase in binding sites (Figure 11A). FGFR typically are expressed at densities of approximately 1×104 #/cell thereby keeping the primary signaling receptor at a controlled level. This is predicted to result in an order of magnitude less overall FGF-2 binding than that found at typical HSPG levels but which is increased in a similar way when HSPG are present. The combination of the two surface binding sites (FGFR and HSPG) is critical. For example, when 1.0×104 FGFR are present, the retained FGF-2 is increased to ∼0.25ng from a value of ∼0.14ng without the FGFR. Looking at cell binding at the entrance of the cell-lined hollow fiber as a function of time after FGF-2 has been introduced with constant FGFR (1×104 #/cell) and variable HSPG, we found that there was a significant increase in bound FGF-2 at the higher HSPG (1×105 #/cell) when compared to the lower values and that the FGFR binding was essentially all coupled to HSPG (Figure 11B). When there are fewer HSPG, there is a lower percentage of coupled binding at least at earlier times as well as lower overall FGFR complexes.
The results with the FGF-2-HSPG affinity simulations and the density studies indicated the importance of coupling in facilitating effective FGF-2-FGFR interactions. We next looked at how varying the coupling rate constant impacted binding and internalization using simulations (Figure 12). In the absence of HSPG-FGFR coupling (kc = 0), there is a reduction in peak binding of FGF-2 and the majority of FGF-2 bound is not internalized but dissociates and exits from the system in the outflow. Even with a low level of coupling, the FGF-2 binding and internalization is dramatically increased until a peak effect is seen with kc = 0. 01 (#/cell)−1 min−1. If we looked at later times in the simulation (Figure 12B), we would find that a large fraction of the FGF-2 injected is bound during the initial pass and that this bound FGF-2 is largely internalized with little exiting the system. If coupling between HSPG and FGFR is eliminated (Figure 12C), this is not the case. In this scenario, the cells bind a smaller but still significant level of FGF-2 during the initial pass but this FGF-2 is not retained and nearly all of the FGF-2 captured ultimately exits the system in the outflow.
To further illustrate the importance of the coupling process, simulations were performed with cell-lined hollow fibers having only HSPG (2.5×105 #/cell) in the front 25% of the tube and both FGFR (1×104 #/cell) and HSPG (2.5×105 #/cell) in the back 75% of the fiber (Figure 13). The entrance area (front 25%) did not include internalization of FGF-2 by HSPG modeling an ECM-like section, however, the overall outcomes are not significantly changed when internalization is included (data not shown). HSPGs in this front section were able to capture FGF-2 but there is a significant rise in retention in the back section where both HSPG and FGFR are present. This is not simply due to the increase in binding sites due to the addition of FGFR as increasing HSPG by an equivalent level to that of the HSPG plus FGFR did not lead to the same increase in retention (data not shown). Moreover, this increase in retention is lost when the dissociation rate for FGF-2-FGFR-HSPG is reduced to that of FGF-2-HSPG and only nominally increased when the coupling rate is eliminated, reflecting the increased affinity of FGFR compared to HSPG for FGF-2 (data not shown). The effect is evident at both low and high flow rates.
Finally, we used simulations to ask whether dissociation from HSPG in an ECM-like section could lead to increased binding downstream due to slow dissociation of the growth factor and prolonged availability of the growth factor for downstream binding. When the HSPG density in the front 25% zone was increased to 5×106 HSPG/cell, a large increase in overall retention of FGF-2 in the front section was evident resulting in a decrease in FGF binding in the HSPG-FGFR section (back 75%) due to a depletion of FGF-2 in the fluid zone near the cells. This was evident at both 5 (Table 6) and 10 min (data not shown). In contrast, a low level of HSPG (5×104 or less) in the entrance section did not lead to significant binding in this zone and results in increased binding of FGF-2 in the final 75% section. FGF-2 in the fluid phase was at a higher concentration at later times after FGF-2 injection when there were more HSPG in the front section due to dissociation from the HSPGs; however, under flow conditions, this dissociated FGF-2 is not predicted to grow to a high enough concentration to meaningfully impact downstream receptor binding. This is an important difference between flow and static culture studies.
Circulation is an obligatory process for the maintenance of human life. The proper balance of solid and fluid components, flow and pressure, and chemical content are all tightly regulated to maintain homeostasis. Within these limits, however, wide fluctuations can occur. The effects of the regulatory processes that are in place to deal with these fluctuations are not well characterized. Often the overall effects can be easily measured but not the changes in the microenvironment that come together to drive these effects. Although traditional tissue culture studies have added a wealth of knowledge in such areas, they often lack the capability to emulate the in vivo environment. In the study of the effect of flow in regulating vessel wall interactions, for example, three-dimensional studies can provide valuable information. Three-dimensional studies have been used previously to measure the effects of flow on cell populations [18], [36]–[39]. We have chosen such an approach to measure the effect of flow on heparin binding protein delivery. By employing a single pass method to focus on the initial growth factor-vessel wall interaction we were able to more directly measure the effect of flow on the bioavailability of these growth factors. We measured substantial binding of all growth factors (FGF-2, VEGF, and EGF) at the lowest flow rate tested (0.61–0.66 mL/min, 1.22– 1.32 mm/sec). Had a traditional two-dimensional approach been used instead, these factors would have had few limitations on their rebinding potential since in a closed system they would not be subject to the flow that would remove them from the vessel as is typical of normal circulation. In the case of the heparin binding proteins (FGF-2 and VEGF), removal of heparan sulfate sites via enzyme digestion resulted in a significant increase in growth factor outflow (i.e. non-retention within the vessel), suggesting an important regulatory role for these proteoglycans in ligand capture. This is not necessarily surprising given the large number of binding sites these proteoglycans provide on normal cell surfaces. Certainly, it has been shown by us and others that HSPGs are important regulators of FGF-2 binding to FGF receptors in tissue culture [28], although not essential for the interaction [6], [24], [27]. Their importance with regard to capture under flow has, however, not been shown previously and suggests a critical role in the circulation.
An equally significant influence on FGF-2, VEGF, or EGF binding, regardless of heparin binding characteristics however, was the flow rate. By increasing the flow rate by less than a factor of three (∼1.8 mL/min, 3.6 mm/sec) a significant increase was seen in growth factor outflow, reflecting the absence of specific binding taking place on vessel surfaces. A higher flow rate (∼3.0 mL/min, 6 mm/sec) showed no further increase in FGF-2 outflow above that observed at the medium flow rate with both showing retention levels equivalent to that evident in the absence of heparan sulfate. This correlation of flow rate and outflow of growth factors suggests a strong regulatory effect and an environment in the bloodstream that reduces the probability of capture significantly at flow rates typically measured in arteries [4]. Although pulsatile flow is undoubtedly important in increasingly larger vessels and higher flow rates, both simulations and experiments showed that at the low flow rate typical of capillaries it had no significant effect on FGF-2 interactions when compared to steady flow.
The removal of chondroitin sulfate created a small but significant increase in FGF-2 outflow. This is interesting since a number of published findings found no significant affinity between FGF-2 and chondroitin sulfate [30], [31]. It is possible that under flow conditions subtle changes in chondroitin sulfate modifications allow for some weak interaction. Others have reported the ability of FGF-2 to bind chondroitin sulfate under certain circumstances [32]–[34]. EGF binding was, however, unaffected by treatment with a heparinase, chondroitinase and keratanase cocktail suggesting the chondroitinase effect was not universal. How this effect is manifest is currently under further study.
The minimum size of capillaries has been shown to be relatively fixed across species regardless of size [40] and is a basic assumption in the general model of allometric scaling laws proposed by West et al [41]. This suggests an optimum environment for the exchange of gases, nutrients, and the removal of waste products that is likely rooted in fundamental physical laws. In order to best make use of these environmental conditions blood flow must also be optimal. Our data demonstrate an inverse correlation between flow rate and probability of capture. Although the presence of heparan sulfate is crucial to FGF-2 capture at low flow rates, at higher flow rates the overriding regulator seems to be the flow rate itself which, based on our results, would all but preclude efficient FGF-2 binding to vessel walls in a single pass under all but the slowest flow conditions. The expectation of lower binding at increasingly higher flow rates might be somewhat expected but the relatively small increase in flow rate required to ablate binding was surprising.
Other influences, such as viscosity, and the presence of competing molecules were not addressed in this work. These are ongoing studies as we begin to add complexity to the system so as to form even more accurate models of circulation. The advantage of this method is that the conditions can be monitored and controlled much as two dimensional culture systems can be but include the three dimensional architecture and flow characteristics that are part of normal blood flow. This approach has obvious potential in the testing of both endogenous molecules and pharmaceuticals in order to provide a better perspective of molecular interactions in the microenvironment of blood vessels.
The importance of HSPGs in FGF-2 binding and signaling has been shown in many systems [6]–[11] and is a generally accepted feature for heparin-binding growth factors. Our work builds upon those studies and shows the critical importance of HSPGs in FGF-2 capture under flow (Figure 3). In this paper, we explore the impact of this critical component in detail using our computational model and show the parameters that regulate this process. In particular we show that the two-step coupling process and the accompanying decrease in dissociation are essential for effective retention of FGF-2 in a flow situation.
HSPG can mediate both the heparin-binding growth factor-receptor interaction at the cell surface and the accumulation and storage of these growth factors in the extracellular matrix [42], [43]. Removal of HSPG from the cell surface by enzymatic digestion greatly impairs FGF-2 activity in vitro and inhibits neo-vascularization in vivo [27], [28], [44]. HSPG interacts with FGFR directly [45], [46] and FGF-2 binding to cell surface HSPG can facilitate FGF-2 binding to FGFR, which in turn can result in activation of intracellular signaling cascades. Using our simple model under flow, we show in several ways that the coupling step is critical for FGF-2 retention. Elimination of coupling or decreasing the rate constant describing that interaction has a dramatic effect on both FGF-2 bound and internalized with essentially no internalization or effective binding when coupling is eliminated (Figure 12). Reducing the density of HSPG (Figure 11) or the affinity of FGF-2 for HSPG (Figure 9) significantly reduces the amount of FGF-2 bound to both the cell surface and to FGFR. In addition, simulations with only low levels of HSPG (Figures 11, 12 – entrance zone) or FGFR (data not shown) do not exhibit high retention but, when both HSPG and FGFR are present (Figure 13), the combination of both increases retention. This is evident independent of flow rate. The ability of flow to regulate the level of binding suggests how crucial the presence of HSPG is on the vessel wall, in order to increase the probability of capture of heparin-binding molecules especially given the short half-lives of some growth factors in circulation.
Under the flow condition, simulations predict that the majority of FGF-2 binding occurs at the entrance to the cell-lined hollow fiber (Figure 7). In our simulations set up to match the experimental conditions, FGF-2 enters at its highest concentration and thus is most likely to bind under those conditions. Once binding occurs, there is a depletion of FGF-2 in the fluid phase near the cell surface (Figure 8). Under flow, this zone can be replenished via diffusion as increasing the diffusion coefficient increases the concentration in this zone (Figure 8) and ultimately leads to higher binding down the cell-lined hollow fiber. We had postulated that FGF-2 bound in the entrance zone of the cell-lined hollow fiber would eventually dissociate and rebind further down the tube but this does not appear to be the case. Even when binding is extremely high at the entrance, FGF-2 that dissociated from the entrance was not in high enough concentration to impact downstream binding and was eventually washed out of the system (data not shown). In a non-flow system this would likely not be the case and exemplifies the importance of including flow in studies.
In conclusion, a simulation program previously developed by us but enhanced for our specific cell investigations of FGF-2 binding under flow [16], [17] performed well when compared to our experimental endothelial cell-lined bioreactor. Our simulations suggest that: (1) The amount of FGF-2 bound to FGFR is dominated by HSPG and the coupling rate constant, and this triad (FGFR-HSPG-FGF-2) is the key to FGF-2 capture; (2) The amount of FGF-2 bound is proportional to the diffusivity of the growth factor in solution and inversely proportional to the flow rate; (3) Flow rate and diffusivity will affect the FGF-2 outflow profile and the distribution of FGF-2 bound along the cell-lined hollow fiber wall; (4) The majority of FGF-2 binding occurs in the entrance zone of the cell-lined hollow fiber; and (5) most FGF-2 effectively bound by FGFR and HSPG will be internalized rather than dissociated. The simulation environment can provide additional information and insight into capture of FGF-2 that is not easily accessible from experimental work. We have applied the model to our in vitro bioreactor system but it has potential to be used for other growth factors as well as other cell systems where flow and capture are pivotal such as in drug and biologicals delivery testing.
|
10.1371/journal.ppat.1003173 | Viral Escape from Neutralizing Antibodies in Early Subtype A HIV-1 Infection Drives an Increase in Autologous Neutralization Breadth | Antibodies that neutralize (nAbs) genetically diverse HIV-1 strains have been recovered from a subset of HIV-1 infected subjects during chronic infection. Exact mechanisms that expand the otherwise narrow neutralization capacity observed during early infection are, however, currently undefined. Here we characterized the earliest nAb responses in a subtype A HIV-1 infected Rwandan seroconverter who later developed moderate cross-clade nAb breadth, using (i) envelope (Env) glycoproteins from the transmitted/founder virus and twenty longitudinal nAb escape variants, (ii) longitudinal autologous plasma, and (iii) autologous monoclonal antibodies (mAbs). Initially, nAbs targeted a single region of gp120, which flanked the V3 domain and involved the alpha2 helix. A single amino acid change at one of three positions in this region conferred early escape. One immunoglobulin heavy chain and two light chains recovered from autologous B cells comprised two mAbs, 19.3H-L1 and 19.3H-L3, which neutralized the founder Env along with one or three of the early escape variants carrying these mutations, respectively. Neither mAb neutralized later nAb escape or heterologous Envs. Crystal structures of the antigen-binding fragments (Fabs) revealed flat epitope contact surfaces, where minimal light chain mutation in 19.3H-L3 allowed for additional antigenic interactions. Resistance to mAb neutralization arose in later Envs through alteration of two glycans spatially adjacent to the initial escape signatures. The cross-neutralizing nAbs that ultimately developed failed to target any of the defined V3-proximal changes generated during the first year of infection in this subject. Our data demonstrate that this subject's first recognized nAb epitope elicited strain-specific mAbs, which incrementally acquired autologous breadth, and directed later B cell responses to target distinct portions of Env. This immune re-focusing could have triggered the evolution of cross-clade antibodies and suggests that exposure to a specific sequence of immune escape variants might promote broad humoral responses during HIV-1 infection.
| Since cases were first recognized in the United States in 1981, human immunodeficiency virus (HIV-1) has infected over one million Americans. Globally, this scale reaches into the tens of millions, but no effective vaccine exists. Of those infected, approximately 20–30% of patients will develop broadly neutralizing antibodies. The reasons for maturation of these potentially protective responses are presently unknown, but being able to elicit such antibodies via vaccination could curb the pandemic. Here, we defined the earliest neutralizing antibody targets and the consequent routes of viral escape in one subtype A HIV-1 infected subject who developed modest breadth. We also determined the genetic and structural characteristics of early neutralizing monoclonal antibodies circulating in this subject and found that subtle light chain alteration enhanced target contact and neutralization. Overall, our data support the idea that exposure to a specific sequence of viral variants, which have escaped from immune pressure, could program long-term potential for antibody breadth.
| Protective vaccines against viral infections generally elicit nAb responses that are comparable to those in natural infections [1]. It is, therefore, widely accepted that an optimal vaccine against HIV-1 will need to produce nAbs, but features such as the high genetic diversity and mutability of HIV-1 Env pose unique obstacles. While broad neutralization of HIV-1 will likely be difficult to achieve through immunization, renewed optimism exists because of breakthroughs in the HIV-1 vaccine and nAb research fields. In the recently concluded RV144 vaccine trial, modest protection from acquisition of infection was observed and correlated with high levels of antibodies that recognized the V1V2 hypervariable domain of Env gp120 [2]. To date, these anti-V1V2 antibodies are the only immune correlate of vaccine-mediated protection against HIV-1 in humans. In non-human primate models, Barouch et al. reported that strong vaccine-induced protection against a diverse simian immunodeficiency virus (SIV) challenge in rhesus macaques correlated with V2-binding antibody titer along with nAb titers against two neutralization-sensitive heterologous SIV Envs [3]. Taken together, these results support the concept that antibodies are important for protection against HIV-1 infection and lead to the hypothesis that even higher vaccine efficacy could be achieved if broad nAbs can be induced [4].
The latest intensified efforts to recover and characterize potent and broad mAbs from chronically infected subjects with exceptional neutralization breadth have yielded important clues regarding how these mAbs overcome Env diversity. Such cross-clade neutralizing mAbs have been shown to target conserved elements in the CD4 binding site (CD4bs) (e.g. VRC01, PGV04), V1V2-dependent and trimer-enhanced (quaternary) epitopes (e.g. PG9, PG16), the gp41 membrane proximal external region (MPER) (e.g. 4E10, CAP206, 10E8), and glycan/V3-dependent epitopes (e.g. PGT128) [5]–[11]. For each class of ‘super’ mAb, characterization of the variable domains of the immunoglobulin heavy and light chains (VH and VL, respectively), in terms of their structure, germline gene utilization, level of somatic hypermutation, and the features of their heavy chain third complementarity-determining regions (CDR H3s), has unveiled specific characteristics that facilitate extraordinary neutralizing capacity [8], [12]–[15]. Importantly, substantial nAb breadth usually requires two to three years of infection to develop and occurs in only about 20–30% of infected subjects [16], [17]. Furthermore, individuals with ‘elite’ neutralizing activity constitute only about 1% of chronically infected subjects [18]. The reasons why nAb breadth does not develop earlier or more frequently are not known but could include autoreactivity leading to clonal deletion of B cells [19], impaired affinity maturation [20], or induction of a particular Ig germline family [10], [13], [15], [21], [22]. It is also possible that early viral escape contributes to the process of increasing nAb breadth [23], [24].
A paradox of neutralization breadth is that targets known to mediate this phenomenon, such as HXB2 residue N160 in V2 (targeted by PG9, PG16) or N332 near V3 (targeted by PGT128), are well conserved and present in many transmitted/founder Envs, but broad cross-clade activity only develops in a subset of individuals. The mere presence of these targets is not, then, sufficient to elicit broadly neutralizing antibodies in early infection. Here we describe the initial nAbs in a subtype A HIV-1 infection that target the N332-proximal region of gp120 that has been previously associated with broad neutralization by mAbs recovered from a chronic subtype CRF02_AG infection [6], [8] and strain-specific nAb responses in early subtype B infection [25]. Early escape involved a single amino acid substitution in this region, which appeared to drive a modest increase in the autologous neutralization breadth of somatically related mAbs. Later escape entailed the addition and/or shifting of glycans recognized by several previously described broadly neutralizing mAbs, but these changes were not targeted by the cross-neutralizing nAbs that developed later in this subject. The combinatorial interplay among early nAb targets, viral escape pathways, and antibody somatic hypermutation could, therefore, shape the ultimate development of heterologous nAb breadth across subjects.
To examine the course and magnitude of autologous HIV-specific humoral activity in a Rwandan seroconverter, R880F, establishment and evolution of the earliest detectable nAb responses were evaluated. This subject was identified as antigen positive, antibody negative on 5Jan07 and then as antibody positive on 12Jan07. This latter date of seroconversion was designated as the 0-month time point for our analyses. Subsequent samples were chronologically coded from this originating time point forward, and each Env was given an arbitrary letter (A, B, or C) and number (1–61) designation that was preceded by the time point of isolation in months post-seroconversion. For example, Envs 0-A6 and 0-B24 were singly isolated from 0-month plasma, 2-A9 and 2-A13 from 2-month plasma, 5-A5 and 5-B52 from 5-month plasma, etc. (Table 1). These viral Envs were single-genome amplified and cloned into expression plasmids for the evaluation of Env pseudotypes. The two 0-month Envs, 0-A6 and 0-B24, had identical sequences and represented the transmitted/founder virus (Figure S1). In sum, ten envelopes from plasma at 2-months post-seroconversion, three from 5-months, five from 7-months, and two from 10-months were evaluated (Table 1).
Each Env-pseudotyped virus was assayed against autologous plasma contemporaneous to its date of isolation in the Tzm-bl neutralization assay. Plasma samples from 2-, 5-, 7-, and 10-months, but not 0-months, potently neutralized the founder Envs 0-A6 and 0-B24 (Figure 1). All longitudinal Envs were at least one log less sensitive to neutralization by contemporaneous plasma than the founder Envs and were, therefore, considered humoral escape variants (Figure 1B–E). These 0-, 2-, 5-, 7-, and 10-month Envs all succumbed to neutralization by plasma collected at 16-months post-seroconversion (Figure 1F). Hence, the induction of de novo nAbs against contemporaneous escape variants, which we and others have previously described [26]–[30], also occurred during the first year and a half of infection in R880F. In this subtype A HIV-1 infected subject, a potent nAb response was detected by 2-months following the first antibody positive time point and initiated repeated rounds of neutralization and viral escape.
To localize the earliest nAb target and elucidate consequent mechanisms of viral escape, full-length amino acid Env sequences for all 2-month nAb escape variants shown in Figure 1B were aligned and inspected for the presence of mutational hot spots. Amino acid changes concentrated in three regions of gp120 at 2-months: in C2 immediately preceding the beginning of V3, in the alpha2 helix in C3, and in V5. Figure 2 specifically diagrams these segments of gp120; Figure S1 includes the full gp160 alignment of all 22 R880F Envs. The isoleucine at position 295 (I295; HXB2 residue 293) in C2 mutated to arginine (I295R) in two Envs or threonine (I295T) in one Env (Figure 2). Additionally, glutamic acid E338 in the alpha2 helix (HXB2 residue 337) became three different residues including aspartic acid (E338D), glycine (E338G), and lysine (E338K) in six Envs (Figure 2). Of note, compared to the founder Env sequence, E338K was the sole mutation in the entire 2-A3 Env sequence (Figure S1). We concluded, then, that this single mutation directly mediated nAb escape. The aspartic acid at position 341 (D341; HXB2 residue 340), also in the alpha2 helix, changed to asparagine (D341N) in one Env (Figure 2). Finally, the glutamic acid at position 456 (E456; HXB2 residue 460) in V5 switched to lysine (E456K) in four Envs (Figure 2).
The potential escape mutations at I295, D341, and E456 were introduced into the founder Env 0-B24 by site-directed mutagenesis to determine if these alterations could individually switch the founder Env phenotype from sensitive to resistant when assayed for neutralization by 2-month plasma. In addition, amino acid changes were introduced into escape Env 2-A3 at K338 to determine whether these mutants maintained nAb resistance. The I295R and I295T substitutions in C2 independently conferred nAb escape, with I295R producing a slightly higher level of resistance that was most evident at the 1∶100 dilution of plasma (Figure 3A). For position 338, two naturally occurring substitutions (E338D from 2-A23/2-A24 and E338G from 2-B18, see Figure 2) and three artificially introduced mutations (K338I, K338Q, and K338R) independently reproduced escape Env 2-A3's wild-type level of resistance, arguing that any change at this position could provide full escape from neutralization (Figure 3B). Thus, the degree of neutralization resistance conferred by changes at I295, but not at E338, varied by the amino acid substitution. Introduction of D341N into the alpha2 helix of the founder Env 0-B24 also recapitulated the wild-type resistance level of 2-B12 (Figure 3C). Because the I295, E338, and D341 escape mutations occurred independently in the 2-month Envs, each represents a distinct lineage for escape (Figure 2). In addition, the potency of resistance was substitution-specific; I295R/T and E338D/G/K produced the highest levels of resistance, while D341N lagged somewhat behind and provided partial resistance. In contrast, the E456K mutation in V5 exerted no overt influence on neutralization phenotype when introduced into the founder Env 0-B24, despite being carried in nearly half of the 2-month escape Envs (Figure 3D). Overall, at 2-months, the viral population utilized a common amino acid substitution mechanism that diverged down three discrete escape pathways, each of which conferred nAb resistance.
Though positions 295, 338, and 341 appear disparate in the linear gp120 sequence, these residues cluster when plotted onto a 3-dimensional representation of the R880F founder Env sequence, which was modeled using all existing structures for CD4-bound HIV-1 gp120 (Figure 4). This proposed epitope emerges near the base of the V3 domain, which is well exposed on the native trimer and is also targeted by the broadly neutralizing, glycan-dependent mAb PGT128 [8]. The spatial proximity of these three residues provides evidence for a single nAb epitope during early subtype A HIV-1 infection and an explanation for why a substitution at any one of the three positions independently caused nAb resistance.
During HIV-1 infection, the antibodies circulating in patient plasma could ostensibly represent a heterogeneous pool with varying epitope specificities. Although we were able to identify a single, early nAb target in subject R880F using autologous plasma and 3-dimensional modeling, this epitope could be recognized by a polyclonal nAb response mediated by more than one B cell [31]. To illuminate the characteristics of individual monoclonal effectors, we PCR amplified and cloned antibody VH and VL genes from memory B cells present in a cryopreserved R880F peripheral blood mononuclear cell (PBMC) sample collected at 16-months post-seroconversion (Table 1). Multiple VHs and VLs were obtained, but only one VH, named 19.3H-HC, neutralized the founder Env when combined with either of two highly related VLs. Sequence analysis revealed that the R880F VH utilized IGHV3-30*02, IGHD1-7*01, and IGHJ4*02 gene segments based on matching within the SoDA database [32] and demonstrated 23.2% mutation across its framework (FWR) and complementarity-determining regions (CDR), as compared with germline at the amino acid level (Figure 5A). The VLs, named 19.3H-L1 and 19.3H-L3, were clonal relatives, both using IGLV2-14*01 and IGLJ2*01 gene segments based on matching within the SoDA database [32] and exhibiting mutation rates of 13.6% and 14.5% from the putative germline, respectively (Figure 5B). Five total amino acid differences between the 19.3H-L1 and 19.3H-L3 VLs congregated in and around CDR1: 19.3H-L3 contained two threonines (T) and one phenylalanine (F) in CDR1 that were not present in 19.3H-L1, while arginine (R) and glutamic acid (E) residues arose just downstream of CDR1 in the FWR2 region of 19.3H-L1 that were not present in 19.3H-L3 (Figure 5B). The VL CDR3 domains of 19.3H-L1 and 19.3H-L3 were identical and contained five amino acid differences from the putative germline. The two R880F mAbs produced by combination of 19.3H-HC and 19.3H-L1 or 19.3H-L3 are hereafter referenced solely by their VL designations.
Figure 5 demonstrates that both 19.3H-L1 (C) and 19.3H-L3 (D) neutralized the founder Envs 0-A6 and 0-B24, although 19.3H-L3 did so with approximately one log greater potency. In addition to neutralizing the founder Env, both mAbs neutralized the 2-month plasma escape Env 2-B12 with similar potencies. 19.3H-L3 also neutralized plasma escape variants 5-B52 and 2-B31 potently, and 2-A9 and 2-A13 to a much lesser extent. The remaining 2- and 5-month escape variants, and all 7- and 10-month escape variants were resistant to both mAbs. This result suggests that the mAbs are representative of those that circulated within the first few (2–5) months of infection; because they were isolated from memory B cells, 19.3H-L1 and 19.3H-L3 do not reflect the ability of the 16-month plasma nAbs to neutralize all longitudinal R880F Envs (Figure 1F, Table 2). To provide evidence for the specificity and authenticity of 19.3H-L1 and 19.3H-L3, the common VH, 19.3H-HC, was co-transfected with other autologous VL genes from two randomly selected R880F B cell wells. One VL utilized the same IGLV2-14*01 gene segment as 19.3H-L1 and 19.3H-L3 (Figure S2A,E); one did not (Figure S2B,E). Conversely, the 19.3H-L3 VL was paired with an autologous VH from a different R880F B cell well (Figure S2C,E). All three chimeric antibody supernatants were assayed for activity against a smaller panel of ten longitudinal R880F Envs, and no neutralizing activity was observed (Figure S2A–C), suggesting that stochastic pairing of R880F VHs and VLs does not confer neutralizing activity.
To map the specificity of mAbs 19.3H-L1 and 19.3H-L3 in finer detail, we utilized the point mutants from Figure 3, with the addition of double mutant 2-A3 K338G D341N, which was representative of escape Env 5-B52. As previously mentioned, 19.3H-L1 neutralized Env 2-B12 in addition to the founder Envs; Env 2-B12 was the only Env in the panel that shared with the founder Env all three unmutated residues at positions I295, S335, and E338 (Figure 2, Table 2). A change at any one of these positions resulted in resistance to 19.3H-L1 neutralization (Figure 5C, Table 2). 19.3H-L1 neutralized 2-B12 more potently than the founder Envs; this was directly attributed to D341N, as this single substitution introduced into Env 0-B24 (0-B24 D341N) increased the founder Env's sensitivity to that of 2-B12 (Figure 5C, Table 2). Despite sharing a common VH with 19.3H-L1, 19.3H-L3 demonstrated a distinct pattern of specificity. In contrast to 19.3H-L1, 19.3H-L3 neutralized the founder and 2-B12 Envs equivalently. In this case, then, the D341N mutation (0-B24 D341N) had very little effect on the neutralization phenotype (Figure 5D). 19.3H-L3 also neutralized Envs carrying the I295T substitution (0-B24 I295T and 2-B31) but displayed a much weaker level of neutralization capacity against Envs containing the I295R substitution (0-B24 I295R, 2-A9, and 2-A13). 19.3H-L3 neutralized Envs containing the E338G substitution when it occurred in the presence of D341N (5-B52 and 2-A3 K338G D341N) but not when E338G (or any other E338 substitution) occurred in isolation (Figure 5D, Table 2). R880F mAb 19.3H-L3, therefore, had potent neutralizing activity against two Envs (5-B52 and 2-B31) and modest activity against two Envs (2-A9 and 2-A13) that were resistant to contemporaneous plasma and to mAb 19.3H-L1. Hence, the mutational program at positions 295, 338, and 341, first witnessed at 2-months post-seroconversion to facilitate immune evasion (Figure 3), likely fueled subsequent rounds of nAb recognition, and mutations that originally evolved the virus toward an escaped phenotype here conferred sensitivity to somatically related autologous mAbs (Figure 5, Table 2).
To ascertain if 19.3H-L1 and 19.3H-L3 would compete for Env binding, three R880F gp120 monomeric proteins (the 0-A6/B24 founder Env gp120, and mutants containing I295R or E338K) were synthesized, purified, and employed in a competition ELISA assay. To first establish a baseline level of binding, the R880F mAbs were biotinylated and incubated with wild-type 0-A6/B24 gp120 protein. 19.3H-L3 demonstrated more robust binding, as compared to 19.3H-L1; the negative control mAb 6.4C (directed against a highly specific epitope in V1V2 [31]), and the broadly neutralizing mAb PGT128 [8], which shares epitope space with the R880F mAbs, both failed to bind (Figure 6A). Consistent with the neutralization data in Figure 5, neither R880F mAb could bind detectably to the I295R or E338K mutant gp120 proteins (Figure 6B–C). Wild-type 0-A6/B24 gp120 protein was then pre-incubated with 19.3H-L1, 19.3H-L3, or the negative control antibody 6.4C, washed, and incubated with either biotinylated 19.3H-L1 (Figure 6D) or 19.3H-L3 (Figure 6E) to discern if initial pre-incubation could block secondary binding. 19.3H-L1 modestly competed with itself (Figure 6D) but could not effectively compete for binding with 19.3H-L3 (Figure 6E). Conversely, 19.3H-L3 strongly competed with both itself (Figure 6E) and 19.3H-L1 (Figure 6D). Thus, 19.3H-L3 neutralizes a greater number of R880F Envs than 19.3H-L1, binds more strongly to the founder 0-A6/B24 gp120, and neutralizes the Env 0-A6/B24 pseudovirus more potently, underscoring the significance of VL alterations where antigen recognition and neutralization efficacy are concerned.
To interrogate the antigen-binding site characteristics of R880F mAbs that influenced their distinct neutralization profiles, crystal structures of the 19.3H-L1 and 19.3H-L3 Fabs were determined to the resolutions of 1.7 Å (Figure 7A) and 2.7 Å, respectively (Table S1). Although the two Fabs were crystallized in different space groups, the resultant structures were highly similar, with root mean square deviations less than 1 Å when all of the Cα atoms were superimposed (data not shown). Several structural analyses were employed, including calculations of Optical Docking Area (ODA, shown in Figure 7B, which predicted the antigen-binding sites by calculating the desolvation free energy of the surfaces), surface pockets, and electrostatic surface potentials. ODA analyses indicated that the antigen-binding sites of 19.3H-L1 and 19.3H-L3 were very flat, forming roughly rectangular shapes approximately 15 Å wide and 30 Å long on top of the six CDR loops (Figure 7C). No pockets existed in these binding surfaces, and the shared CDR H3, although it was 18 amino acids long (Kabat numbering scheme [33]), did not protrude. Such flat antigen-binding sites likely interact with epitopes formed by residues also on planar surfaces (i.e. flat-surface antigen-antibody contacts). Electrostatic surface potential analyses showed that the 19.3H-L1 and 19.3H-L3 antigen-binding sites were essentially neutral; a couple of slightly positive regions along one side of the rectangular contact area counterbalanced a slightly negative opposite region (Figure 7C, blue and red patches, respectively).
Three CDR1 residues that differed between 19.3H-L1 and 19.3H-L3 (Ser/Thr at residue 27, Gly/Thr at residue 29, and Tyr/Phe at residue 32; Kabat numbering scheme [33]; Figure 7D) did not create any substantial structural differences between the two antigen-binding sites. These changes did, however, have the potential to influence antigen-antibody interactions. The Tyr in 19.3H-L1 to Phe in 19.3H-L3 change at residue 32 likely increased the hydrophobicity at the center of the antigen-binding site, which may have augmented hydrophobic interactions with the antigen. The Gly to Thr mutation at residue 29 added a polar side chain with additional hydrogen binding possibilities. Finally, the Ser to Thr substitution at residue 27 provided a more stable side chain. As a group, these VL alterations probably enhanced the antigen-binding affinity of 19.3H-L3, explaining its increased autologous neutralization breadth.
As demonstrated in Figure 5, D341N appeared to be detrimental to the preservation of a neutralization-resistant phenotype, in the context of mAbs 19.3H-L1 and 19.3H-L3 during early infection. This mutation was, nonetheless, retained in later escape Envs. Inspection of the 7- and 10-month Env sequences containing D341N revealed that they had acquired additional substitutions, I295N (HXB2 residue 293) and/or S335N (HXB2 residue 334), absent from earlier Envs (Figure 2); each of these mutations affected a potential N-linked glycosylation site (PNGS). Accordingly, we hypothesized that these co-traveling mutations compensated for the vulnerability associated with D341N in a PNGS-dependent manner. To explore this, the I295N substitution, which created a PNGS, was introduced into two mAb-sensitive Envs: 0-A6 and 2-B12. The I295N versions of these two Envs displayed high-level resistance against mAbs 19.3H-L1 and 19.3H-L3 (Figure 8A–B, Table 2). Similarly, the S335N substitution, which also incorporated a PNGS, was inserted in three mAb-sensitive Envs: 0-A6, 2-B12, and 5-B52. The S335N versions of these three Envs also became highly resistant to 19.3H-L1 and 19.3H-L3 (Figure 8A–B, Table 2). The S335N substitution shifted a well-conserved PNGS sequon at position 333 (HXB2 residue 332; Figure 8C) that is targeted by broadly neutralizing mAbs PGT128 and 2G12 [8], [34], [35]. To determine if the observed mAb resistance was glycan-dependent, an S335Q substitution was created in Env 2-B12. Unlike S335N, which shifted the N333 sequon down two positions, S335Q destroyed the N333 sequon altogether (Figure 8C). The resulting mutant, 2-B12 S335Q, was two logs less sensitive to neutralization by mAb 19.3H-L1 than the parental Env 2-B12, but did not reach the high level of resistance achieved by 2-B12 S335N; in contrast, S335Q had only a slight effect on neutralization by mAb 19.3H-L3 (Figure 8A–B, Table 2). High-level resistance against mAbs 19.3H-L1 and 19.3H-L3, therefore, required the addition and/or shifting of PNGS sequons, but amino acid substitution S335Q also provided partial resistance that was much more effective against mAb 19.3H-L1. Together, the data strongly support a mechanism of mAb escape that was PNGS-dependent and may have introduced glycans capable of obscuring the V3-proximal space recognized by 19.3H-L1 and 19.3H-L3 (Figure 8D). Nevertheless, the two mAbs–common heavy chain notwithstanding–appear to recognize subtly distinct epitopes.
The VH, in particular the CDR H3, has generally been considered a major determinant of epitope recognition and nAb breadth. In our study, VL differences appreciably expanded the neutralization capacity of mAb 19.3H-L3 against autologous Envs. To probe whether this increase in breadth carried over to neutralization of heterologous Envs, mAbs 19.3H-L1 and 19.3H-L3 were tested against a panel of fourteen heterologous Env pseudotypes that included one A/C recombinant, four subtype A, three subtype B, and six subtype C Envs. The mAbs were unable to neutralize any of the heterologous Envs (Figure 9A–B). Thus, while mAb 19.3H-L3 possessed increased breadth against autologous Envs as compared to 19.3H-L1, this did not extend to genetically diverse Envs. Regardless of this restricted mAb cross-clade neutralization, R880F plasma collected at 16-months or 3-years post-infection did have similarly moderate breadth against heterologous Envs, which increased in potency over time (Figure 9C–D). An amino acid alignment of Envs from the heterologous breadth panel demonstrated that Envs neutralized with the greatest potency at 3-years post-seroconversion, A-Q461 and C-Z205F (IC50 values of approximately 1∶1000), contained the N335 (HXB2 residue 334) shifted glycan associated with viral escape from mAbs 19.3H-L1 and 19.3H-L3 (Figure 9E). Furthermore, Env A-Q461 also incorporated the N295 (HXB2 residue 293) substitution indicative of mAb escape. To investigate if the N295 glycan addition and/or the shifted N335 glycan in R880F Envs could have been partially responsible for the heterologous neutralization capacity that developed in this subject, several glycan knock-out mutants were created and tested with 3-year R880F plasma (Figure 10A). Within A-Q461, the N295 PNGS was eliminated either alone or in conjunction with the N335 PNGS; the N335 PNGS was also individually knocked out (Figure 10B). The positions of interest were reverted back to the amino acid present in the transmitted/founder Env 0-A6/B24. For C-Z205F, the N335 PNGS was similarly abolished (Figure 10B). Additionally, two heterologous Envs that were only modestly neutralized but that contained the highlighted glycans, C-Z109F and C-Z214M, were mutated as well. All six of the glycan knock-out mutants exactly mirrored their parental equivalents, suggesting that the particular glycans at positions 295 and 335 did not directly contribute to the breadth observed at 3-years post-infection. These data do suggest, however, that early viral escape events likely influenced how breadth developed in this subject, by expanding what was originally a narrow, regional response at the base of the V3 loop to recognize and neutralize distinct portions of Env across genetically diverse variants.
Several recent studies detail the nAb responses in early subtype B and C HIV-1 infection [24], [25], [27], [29], [31], [36]. Here we present the first such study of a subtype A infected individual, R880F, where the initial autologous nAb target was defined, along with the consequent routes of viral escape, and two mAbs from early infection were recovered. The kinetics of autologous nAb induction in R880F generally mimicked those described previously for early HIV-1 infection with subtypes A, B, and C [25]–[27], [30], [36], [37]. Reduced neutralizing activity against contemporaneous Envs at each time point indicated a well-established repeating pattern of de novo neutralization and viral escape in subject R880F. The early escape Env 2-A3 that differed by only one amino acid residue from the founder Envs, 0-A6 and 0-B24, when combined with a comprehensive panel of mutants, supports the hypothesis that the initial site of nAb recognition was a conformational target at the base of the V3 domain. Specifically, individual mutations at I295, E338, or D341 in R880F conferred escape from 2-month plasma antibodies. The region that encompasses these mutations is close to the gp120 surface area targeted by the broadly neutralizing mAb PGT128 (recovered from a CRF02_AG elite neutralizer) [6], [8], by early plasma nAbs and two mAbs recovered from a subtype B infected seroconverter [25] and by multiple autologous mAbs recovered from two subtype B infected individuals after cessation of antiretroviral treatment [38]. Thus, early nAbs across subtypes commonly target an immunogenic gp120 structure topographically situated near V3, which is well exposed on the Env trimer.
V3-adjacent regions of Env do, nevertheless, elicit strain-specific responses that are easily escaped by multiple pathways. In the study by Bar et al., nAbs in one of three subjects (CH40) targeted a putative conformational epitope composed of the same regions bordering V3 that we describe here for R880F. CH40 immune evasion in the V3 flanks was, however, preceded by escape mutations in V1; this suggests that this latter region, also immunogenic in early infection, may have been targeted first [25]. Moore et al. recently characterized 2 of 79 subtype C infected subjects who were selected because they developed heterologous plasma neutralization breadth mediated by glycan recognition at HXB2 residue N332, another V3-proximal position. In each of these individuals, the glycan motif at HXB2 residue N334 was present in the founder Env; N332 evolved later as an escape mutation and was subsequently targeted by nAbs [24]. Interestingly, in R880F, the opposite occurred: N332 (R880F residue N333) was present in the founder Env and shifted to N334 (R880F residue N335) as an escape mutation in some Envs. Furthermore, the development of heterologous breadth in R880F was not facilitated by specific recognition of N334 and, therefore, involved additional determinants and complexity. When juxtaposed, these and our studies underscore how identical mutations, when ordered differently during infection, can sometimes drive divergent phenotypic outcomes. Thus, exposure of B cells to a specific sequence of changes in Env can program the course of nAb breadth.
In our previous study of autologous nAb responses during early subtype C HIV-1 infection in subject Z205F, we reported that multiple mAbs targeted the V1V2 domain [31]. These three Z205F mAbs used somatically related IGHV3-15*01 and IGLV2-14*01 germline gene segments and recognized a series of overlapping conformational epitopes centered on residues N134 in V1 and R189 in V2. Each mAb demonstrated a distinct neutralization profile against early autologous Envs, with variable sensitivity to specific glycans. R880F mAbs similarly utilized a restricted set of IGHV3-30*02 and IGLV2-14*01 germline gene segments, but, in this case, only a single isolated VH exhibited neutralization capacity when paired with the two clonally related VLs named 19.3H-L1 and 19.3H-L3 (Figure S2). In a recent study, a single VH was recovered through phage display and conferred neutralization when paired with four somatically related variants of the same kappa VL [39]. Such VL shuffling produced mAbs with varying neutralizing activities, the most potent of which was dependent on one residue in FWR2 and one residue in CDR3. Moreover, precedent sets of clonally related mAbs that show distinct neutralization potency and/or breadth have been catalogued in HIV-1 infection [6], [7], [15], [38]–[40]. Within the context of our study, it is conceivable that only one R880F VL is authentic, while the other was generated by mutation during short-term in vitro stimulation of B cells. This caveat notwithstanding, variation between the neutralizing activities of mAbs 19.3H-L1 and 19.3H-L3 highlights a feasible mechanism for gradual acquisition of autologous breadth against highly related escape variants that was directly attributable to VL changes. Furthermore, in future studies it would be advantageous to recover a greater number of distinct antibodies, as our ability to understand breadth fully here was limited with only two highly related mAbs.
Notably, the mAbs from Z205F and R880F were predicted to utilize the same VL germline, IGLV2-14*01. This germline gene segment is also employed by the broadly neutralizing mAbs PG9 and PG16 that target a quaternary epitope involving V1V2 and V3 and is again paired with a VH3 family gene segment, IGHV3-33*05. These data suggest that VH3 and VL2 pairing is not uncommon for HIV-1 nAbs. Several instances of VH bias for anti-HIV mAbs have been demonstrated based on the epitope: anti-V3 mAbs preferentially use VH5-51 [41], [42]; anti-CD4i mAbs preferentially use VH1-69 [22]; anti-MPER mAbs in more than one instance also utilize VH1-69 [10]; and anti-CD4bs mAbs preferentially use VH1-46 and VH1-2 [13], [15]. These pairings may simply reflect common rearrangement of these germline gene segments in the human immunoglobulin repertoire or the structural features that they bind.
Defining the structural characteristics of broadly neutralizing mAbs isolated from elite neutralizers in chronic infection has been a major focus in the HIV-1 nAb field. Unlike the Bar et al. study [25], our data here supply structural information regarding HIV-specific mAbs at the opposite end of the neutralization spectrum. Indeed, we are among the first to report high-resolution crystal Fab structures from early HIV-1 infection, and to show that these mAbs likely mediate planar interactions with antigen that can be subtly altered by VL changes. Structural analyses of the 19.3H-L1 and 19.3H-L3 antigen-binding sites are consistent with the neutralization data that place their epitopes at the base of the V3 domain. As this region of gp120 lies flat, any one of the three single amino acid changes that conferred escape at 2-months could potentially disrupt the planar interactions between the 19.3H-L1 and 19.3H-L3 antigen-binding sites and their epitopes, as discussed below.
Introducing a positively charged residue with a long side chain (I295R) or a glycan (I295N) at position 295 is not compatible with the flat hydrophobic surface of the 19.3H-L1/19.3H-L3 antigen-binding site. In fact, neither mAb could bind to monomeric R880F gp120 containing the I295R mutation. In this model, the I295T substitution would be less effective at conferring neutralization escape. The long, negatively charged E338 side chain is predicted to interact with one of the positively charged surface patches (Figure 7C, blue) at the edge of the 19.3H-L1/19.3H-L3 antigen-binding site, potentially forming a salt bridge with the side chain of a positively charged residue there. The E338K mutation probably destroys this interaction and creates an electrostatic repulsion, which is also consistent with the lack of mAb binding to monomeric R880F gp120 containing the E338K mutation. Interestingly, E338D at this position does not allow 19.3H-L1 and 19.3H-L3 to neutralize the viruses, suggesting that the length of the Asp side chain is not sufficiently long to restore the possible salt bridge. These results suggest that both length and negative charge of the side chain at E338 are important for antibody binding.
The highly conserved N333 (HXB2 residue 332) PNGS at the base of V3 is located at the edge of the proposed epitope and potentially interacts with 19.3H-L1 and 19.3H-L3, as removal of this glycan (S335Q) weakens the neutralization capacities of these two antibodies, most dramatically in the case of 19.3H-L1. Moreover, the glycan shift from position 333 to 335 (S335N), toward the center of the epitope, also prevents the flat-surface antigen-antibody interaction. In combination, the structural, neutralization, and ELISA binding data indicate that mAbs 19.3H-L1 and 19.3H-L3 likely recognize overlapping epitopes that are centered on I295 and E338; however, 19.3H-L1 is more dependent on D341N and the N333 glycan motif for neutralization than 19.3H-L3. Wholly, these analyses suggest that planar motifs that lie across a flat antigen surface could mediate antibody-antigen recognition in early HIV-1 infection, prior to multiple rounds of viral escape and perhaps more extensive affinity maturation. Additionally, the specific determinants for optimal antigen recognition by each mAb, and the strengths of R880F founder Env gp120 binding, differ slightly as a result of VL variation.
In most cases, neutralization breadth in chronic infection has been attributed to the VH, with particular emphasis on the CDR H3 [8], [22], [43]–[45]. Few studies have, however, investigated the roots of neutralization breadth, as was done here. We found, somewhat unexpectedly, that in R880F, VL sequence variation influenced mAb 19.3H-L3's ability to neutralize two autologous escape variants that were not neutralized by mAb 19.3H-L1 during early infection. Significant augmentation of autologous neutralization via minor VL variation (instead of extensive CDR H3 lengthening) supports a potential mechanism for how escape variants that differ by only a few amino acids and/or glycans are neutralized. Based on this, we contend that the maintenance of VH-determined epitope specificity while light chain antigen contacts are varied could represent an important breadth-augmenting mechanism for B cells responding to highly related Env escape variants. More dramatic nAb structural adaptations such as the elongation of CDR H3 may require time for development, as longitudinal viral variants establish more complex ploys to escape.
Collectively, several factors appeared to shape the antibody maturation pathways in R880F: (i) the initial site of nAb recognition, (ii) VH and VL rearrangement, pairing, and somatic hypermutation, and (iii) repeated exposure to highly related Env escape variants. Our data are consequently consistent with the idea that neutralization breadth arises through the sequential exposure of somatically related B cells to a cascade of viral escape variants presenting altered versions of the same epitope. Additionally, and in contrast to the Moore et al. report [24], our findings demonstrate that glycans, which arose in response to the initial waves of neutralization, do not always become subsequent targets for later nAbs or promote the potential to develop heterologous breadth. Moving forward, better understanding of how initial immunoglobulin targeting affects downstream neutralization potential could positively impact HIV-1 vaccine design. Our studies suggest that the mere presence of a PNGS does not ensure its recognition by an antibody. Sequential exposure to glycans and other Env variations may be required to drive the type of specialized antibody response associated with elite neutralization. In fact, support for this type of immunization approach has been demonstrated [46]. It is, however, currently unknown exactly how to accelerate somatic hypermutation, lengthening of the CDR H3, or the acquisition of other adaptations that lead to increased breadth. We propose that a viable vaccination strategy may involve immunizing with a carefully selected series of Env immunogens that mimic defined amino acid and/or PNGS changes that occurred during the natural viral escape process and led to increased neutralization breadth, such as those described here.
Both the Emory University Institutional Review Board and the Rwanda Ethics Committee approved informed consent and human subjects protocols, and subject R880F provided written informed consent for study participation.
Longitudinal plasma and PBMC samples were obtained from ART-naïve subject R880F during enrollment in International AIDS Vaccine Initiative (IAVI) Protocol C at Projet San Francisco (PSF) in Kigali, Rwanda, as part of a multi-site study of early HIV-1 infection in adult Africans. The PSF cohort, which provides voluntary HIV-1 testing, counseling, and condom provision to cohabiting heterosexual couples, is discussed in more comprehensive detail in [47], [48]. Plasma viral load determination (reported in Table 1) was underwritten by IAVI and performed at Contract Lab Services (CLS) in South Africa using an Abbott m2000 system where typical detection ranged between 160 and 4×107 copies/ml.
Conditions for plasma viral RNA extraction and purification, cDNA synthesis, and nested single-genome PCR amplification have been described previously [49]. Subsequent full-length Env gp160 coding regions (plus Rev, Vpu, and partial Nef) were TA cloned into the CMV promoter-driven expression plasmid pcDNA3.1/V5-His-TOPO (Invitrogen) and screened for biological function as pseudoviruses following co-transfection with an Env-deficient subtype B proviral plasmid (pSG3Δenv) in 293T cells using FuGENE HD (Roche or Promega). Forty-eight hours later, supernatant was collected, clarified at 3,000 rpm for 20 min, and used to infect Tzm-bl cells. Following another 48-hour incubation, β-gal staining was performed, and wells were scored positive or negative for blue foci.
Fourteen subtype A, B, and C envelopes were used to evaluate the heterologous neutralization breadth of R880F mAbs 19.3H-L1 and 19.3H-L3 along with autologous 16-month and 3-year plasmas. One A/C recombinant and three subtype C early transmitted variants were previously cloned in our laboratory, as described in [49]: A/C-R66M is R66M 7Mar06 3A9env2; C-Z205F is Z205F 27Mar03 (“0-month”) EnvPL6.3 [29], [31]; C-Z1792M is Z1792M 18Dec07 3G7env2; and C-Z185F is Z185F 24Aug02 (“0-month”) EnvPB3.1 [29]. Ten envelopes were obtained through the AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH: from Dr. Julie Overbaugh, A-Q769-b9 is Q769ENVb9 (Cat#11545), A-Q769-d22 is Q769ENVd22 (Cat#10458), A-Q461 is Q461ENVd1 (Cat#11544), and A-Q23 is Q23ENV17 (Cat#10455) [50]–[52]; from Drs. David C. Montefiori, Feng Gao, and Ming Li, B-SS1196 is SS1196.1 (Cat#11020), and B-TRO is TRO, clone 11 (Cat#11023) [53]; from Drs. Beatrice H. Hahn, Yingying Li, and Jesus F. Salazar-Gonzalez, B-gp160opt is pConBgp160-opt (Cat#11402), C-gp160opt is pConCgp160-opt (Cat#11407), and C-Z214M is ZM214M.PL15 (Cat#11310) [54]–[56]; and from Drs. Cynthia A. Derdeyn and Eric Hunter, C-Z109F is ZM109F.PB4 (Cat#11314) [57].
Mutations were created through PCR using two overlapping primers that contained the mutated sequence per Env, in a strategy similar to that described previously [29], [31], [58], [59]. Briefly, the plasmids containing 0-A6, 0-B24, 2-A3, 2-B12, 5-B52, A-Q461, C-Z205F, C-Z109F, or C-Z214M env genes were amplified with the following sets of forward (F) and reverse (R) primer sequences, where the mutated nucleotides are underlined:
For mutants 0-A6 I295N and 2-B12 I295N: F 5′-cagcctgtgaatattacgtgtattagaactggc-3′ and R 5′-gccagttctaatacacgtaatattcacaggctg-3′
For mutant 0-B24 I295R: F 5′-gcccagcctgtgagaattacgtg-3′ and R 5′-cacgtaattctcacaggctgggc-3′
For mutant 0-B24 I295T: F 5′-cttgcccagcctgtgacaattacgtgtattag-3′ and R 5′-ctaatacacgtaattgtcacaggctgggcaag-3′
For mutants 0-A6 S335N and 2-B12 S335N: F 5′-gcatattgtaatgtcaatagaacagaatgg-3′ and R 5′-ccattctgttctattgacattacaatatgc-3′
For mutant 5-B52 S335N: F 5′-gcatattgtaatgtcaatagaacaggatgg-3′ and R 5′-ccatcctgttctattgacattacaatatgc-3′
For mutant 2-B12 S335Q: F 5′-gcatattgtaatgtccaaagaacagaatgg-3′ and R 5′-ccattctgttctttggacattacaatatgc-3′
For mutant 2-A3 K338D: F 5′-gtcagtagaacagactggaatgacactttac-3′ and R 5′-gtaaagtgtcattccagtctgttctactgac-3′
For mutant 2-A3 K338G: F 5′-gtcagtagaacaggatggaatgacactttac-3′ and R 5′-gtaaagtgtcattccatcctgttctactgac-3′
For mutant 2-A3 K338G D341N: F 5′-gtcagtagaacaggatggaatgacactttac-3′ and R 5′-gtaaagtgtcattccatcctgttctactgac-3′ followed by F 5′-gtagaacaggatggaataacactttacaacaggtag-3′ and R 5′-ctacctgttgtaaagtgttattccatcctgttctac-3′
For mutant 2-A3 K338I: F 5′-gtcagtagaacaatatggaatgacactttac-3′ and R 5′-gtaaagtgtcattccatattgttctactgac-3′
For mutant 2-A3 K338Q: F 5′-gtcagtagaacacaatggaatgacactttac-3′ and R 5′-gtaaagtgtcattccattgtgttctactgac-3′
For mutant 2-A3 K338R: F 5′-gtcagtagaacaagatggaatgacactttac-3′ and R 5′-gtaaagtgtcattccatcttgttctactgac-3′
For mutant 0-B24 D341N: F 5′-cagaatggaataacactttacaacagg-3′ and R 5′-cctgttgtaaagtgttattccattctg-3′
For mutant 0-B24 E456K: F 5′-gagatggtggtaaggatattaacag-3′ and R 5′-ctgttaatatccttaccaccatctc-3′
For mutant A-Q461 N295I: F 5′-ccaagcctgtgataattacttgtatcagacctggc-3′ and R 5′-gccaggtctgatacaagtaattatcacaggcttgg-3′
For mutant A-Q461 N335S: F 5′-gcacattgtgttgtcagtagaacagagtggaataac-3′ and R 5′-gttattccactctgttctactgacaacacaatgtgc-3′
For mutant A-Q461 N295I N335S: F 5′-ccaagcctgtgataattacttgtatcagacctggc-3′ and R 5′-gccaggtctgatacaagtaattatcacaggcttgg-3′ followed by F 5′-gcacattgtgttgtcagtagaacagagtggaataac-3′ and R 5′-gttattccactctgttctactgacaacacaatgtgc-3′
For mutant C-Z205F N335S: F 5′-caagcatattgtagcattagtaaaagtaaatggaatgac-3′ and R 5′-gtcattccatttacttttactaatgctacaatatgcttg-3′
For mutant C-Z109F N335S: F 5′-gaaaggcatattgtaaaattagtggaagtgagtggaatg-3′ and R 5′-cattccactcacttccactaattttacaatatgcctttc-3′
For mutant C-Z214M N295I: F 5′-caacttacagaagctgtaataattacgtgtatgaggccc-3′ and R 5′-gggcctcatacacgtaattattacagcttctgtaagttg-3′
The PCR cycling parameters were 1 cycle of 95°C for 2 min; 30 cycles of 95°C for 20 sec, 50°C to 60°C for 20 sec (the optimal annealing temperature was determined for each primer set), and 72°C for 2 min; and 1 cycle of 72°C for 3 min. The samples were then stored at 4°C. The 25-µl PCR mixtures contained 62.5–250 ng of each primer, 5–30 ng of the plasmid template, 0.4 mM dNTPs, and 1× reaction buffer. PfuUltra II Fusion HS DNA Polymerase (Stratagene) was used to generate the PCR amplicons, which were digested with 20 U DpnI (NEB) for 1 hr to remove contaminating template DNA and then transformed into maximum-efficiency XL10-Gold ultracompetent cells (Stratagene) so that the DNA volume did not exceed 5% of the cell volume. Transformed cells were plated onto LB-ampicillin agar plates, which generally resulted in 50 to 100 colonies. Isolated colonies were inoculated into LB-ampicillin broth for overnight shaking (225 rpm at 37°C), and plasmids were purified using a QIAprep spin miniprep kit (Qiagen). All env mutations were confirmed by nucleotide sequencing.
Sanger DNA sequencing of wild-type and mutant envelope genes and immunoglobulin genes was executed with an ABI 3730xl DNA Analyzer and BigDye Terminator v3.1 chemistry at one of two facilities: the University of Alabama at Birmingham Center for AIDS Research (P30-A127767) DNA Sequencing Shared Facility or GenScript. Nucleotide sequences were edited and assembled using Sequencher v5.0 and deposited into GenBank under accession numbers JX096639-JX096660 for wild-type env clones and JX124277-JX124282 for immunoglobulin genes. Amino acid sequences were translated and aligned using Geneious v5.0.3.
Five-fold serial dilutions of heat-inactivated R880F plasma samples, antibody-containing 293T supernatants, or purified R880F monoclonal antibodies were assayed for neutralization potential against viral pseudotypes in the Tzm-bl indicator cell line, with luciferase as the ultimate readout, as described previously [29], [37], [57], [58], [60]. In short, Tzm-bl cells were plated and cultured overnight in flat-bottomed 96-well plates. Pseudovirus (2,000 IU) in DMEM with ∼3.5% FBS (HyClone), 40 µg/ml DEAE-dextran was incubated with serial dilutions of plasma, supernatant, or antibody, and subsequently, 100 µl was added to the plated Tzm-bls for a 48 hr infection before being lysed and evaluated for luciferase activity. Data was retrieved from a BioTek Synergy HT multi-mode microplate reader with Gen 5, v1.11 software, the average background luminescence from a series of uninfected wells was subtracted from each experimental well, infectivity curves were generated using GraphPad Prism v4.0c where values from experimental wells were compared against a well containing virus only with no test reagent, and IC50 values were determined using linear regression in Microsoft Excel for Mac 2011, v14.0.2.
The subject R880F 0-B24 Env gp120 sequence was modeled using the MODELLER program [61]. The template for the homology model was a subtype A gp120 obtained by longtime all-atom molecular dynamics simulation using the CHARMM27 potential in the NAMD program [62]. This simulated gp120 was modeled using all known CD4-bound gp120 structures (Protein Data Bank [PDB] accession numbers 1G9M [63],1RZK [22], 2B4C [64], 2NY7 [65], 3JWD and 3JWO [66], and 3LMJ [43]) as templates. In all of these structures, the core of gp120 was highly similar; however, it should be noted that none of these structures is subtype A. Multiple templates were used because it has been shown that this creates high quality homology models. In addition, each template has slightly different regions of gp120 resolved. Before modeling, the templates were arranged in the trimeric state, which has been resolved using cryoelectron microscopy (PDB accession number 3DNO [67]), to ensure that the hypervariable loops did not sterically clash with the neighboring monomers. During modeling, disulfide constraints were added for the conserved cysteines present in all gp120 sequences. All sequence alignments used for modeling templates were based on sequences in the HIV-1 database (www.hiv.lanl.gov).
The subject R880F 0-B24 Env gp120 sequence mutated with N295 and N335 was modeled using the protocol described in [31]. Xleap in AmberTools kit 1.4 was used to add two glycans at residues N295 and N335. A five-mannose glycan was used in this simulation because it was found to be the most abundant glycan form in the immunodeficiency virions [68]. Amber99SB force field [69] was used for the gp120 protein, and GLYCAM06 force field [70] was used for the five-mannose glycan. All the systems were minimized using a 3-step protocol in which the protein was gradually allowed to move. These steps were: heavy atoms fixed (5,000 steps), protein backbone atoms fixed (5,000 steps), and all atoms free to move (20,000 steps). The system was gradually heated in a four-step process. The initial temperature was set to 100 K, and only hydrogen atoms were allowed to move for 25,000 fs. In the second step, the temperature was set at 300 K, and heavy atoms in the protein were harmonically constrained for the next 25,000 fs. Then the temperature was raised to 500 K, and backbone atoms were harmonically constrained for 25,000 fs. Force constants for all harmonic constraints were set to 1 kcal mol−1Å−2. Finally, the temperature was raised to 700 K, and the backbone atoms in the core of gp120 were constrained for the next 4.925 ns. The coordinates were saved once every ps, in these 5 ns. The MD simulation was performed using NAMD 2.8 [62]. The conformation at the end of the 5 ns MD simulation was used in this study.
A viably frozen PBMC sample from subject R880F was collected at 16-months post-infection and was used to recover autologous mAbs 19.3H-L1 and 19.3H-L3. The first phase of recovery was performed in the Robinson laboratory. Non-B cells were depleted using immunomagnetic beads (Miltenyi). Approximately 100,000 B cells were recovered, and for memory B cell stimulation, the cultures were incubated for 3 days in RPMI medium containing 10% FCS, Epstein-Barr virus, 2 µg/ml R848 (InVivogen), and 100 U/ml IL-2 (Dr. Maurice Gately, Hoffmann - La Roche Inc. [71]). The cells were then plated into 3,840 wells in the same medium at low cell densities (30 to 50 cells/well) in forty 96-well tissue culture plates containing irradiated macrophage or human placental fibroblast feeder cells. Starting at 12 days of culture, B cell culture supernatants were screened every 3 to 4 days for antibodies that neutralized the autologous founder pseudovirus containing Env 0-B24, or that showed ELISA reactivity with 0-B24 Env glycoproteins in previously described assays [31], [38]. Supernatant from 63 wells screened positive for inhibitory activity in the Tzm-bl assay, and 35 of these were also positive for gp120 binding activity in ELISA. The positive cultures were placed in RNAlater between days 17 to 21 after stimulation. RNA was purified from 12 lysates of these B cell cultures that had been found to be antibody positive. RNA from each well was reversed transcribed into cDNA encoding VH and lambda/kappa VL genes, which were then amplified in a nested PCR as described by Liao et al. [72]. VH and VL gene products were assembled by overlapping PCR into pairs of linear expression vectors encoding full-length human Ig heavy and light chain genes [72]. These vectors were co-transfected into wells containing 80–90% confluent 293T cells. Two days later, supernatants of transfected cultures were tested for antibody activity in the same assays used to screen B cell cultures. One well of 293T cells transfected with VH and VL genes originating from a single B cell culture designated 19.3H (plate 19, well 3H) was found to be antibody positive. The antibody (or antibodies) produced was thus named 19.3H. 293T cells expressing antibody 19.3H were serially passaged at limiting cell densities under blasticidin selection (for maintenance of the Ig vector) to obtain multiple clones. Selected 19.3H-derived clones were expanded into stable antibody producing cell lines to facilitate purification of the antibody by Protein A affinity chromatography.
To obtain VH and VL sequences that corresponded to the 19.3H antibody activity, VH and VL genes were isolated from the selected 19.3H-derived 293T cell clones using two different methods in the second phase of recovery. The first method was used in the Robinson laboratory. VH and VL genes were re-amplified from the selected 293T cell clones and inserted into expression plasmids obtained from InVivoGen: pFUSE-CHIg-hG1, containing the constant region of the human IgG1 heavy chain, and pFUSE2-CLIg-hl2 containing the constant region of human Ig lambda 2 light chain, respectively. First, the pFuse vectors were linearized by digestion with EcoRI and then subjected to PCR with primers (IgVH FWD 5′-CGAACCGGTGACGGTGTCGTGGAAC-3′ and REV 5′-ACCGGTGATCTCAGGTAGGCGCC-3′, IgVLambda FWD 5′-CCAACAAGGCCACACTGGTGTGTCTC-3′ and REV 5′-ACCGGTGATCTCAGGTAGGCGCC-3′, IgVKappa FWD 5′-GAACTGCCTCTGTTGTGTGCCTGCTG-3′ and REV 5′-ACCGGTGATCTCAGGTAGGCGCC-3′) to generate annealing sites of 15 nucleotides that were homologous with ends of the inserts [73]. Second, the SuperScript III One-Step RT-PCR System (Invitrogen) was used to amplify the Ig variable regions from 293T-cell-derived mRNA using primers designed to synthesize inserts for use with the ligation-independent In-Fusion cloning system (Clontech). The forward primer IgVH,IgVLambda,IgVKappa FWD 5′-CCTGAGATCACCGGTGCTAGCACCATGGAGACAGACACACTCC-3′ was used for both heavy and light chain inserts and contained a non-annealing tag with 15 nucleotides of homology to the upstream insertion site on the plasmid. Reverse primers for each heavy and light chain (IgVH REV 5′-CACCGTCACCGGTTCGGGGAAGTAG-3′, IgVLambda REV 5′GTGTGGCCTTGTTGGCTTGAAGCTCCTC-3′, IgVKappa REV 5′-CACAACAGAGGCAGTTCCAGATTTCAACTGCTC -3′) contained 15–20 nucleotides that overlapped with the 5′ end of the constant regions in linearized pFuse vectors. The In-Fusion reaction was performed according to manufacturer's instructions. Plasmids containing inserts were grown in JM109 competent cells, and at least five colonies were picked for subsequent nucleotide sequencing.
A second approach was performed in the Derdeyn laboratory to recover the VH and VL genes from the 19.3H-derived 293T cell clones, and from In-Fusion plasmids generated in the Robinson lab, such that all VH and VL genes would be expressed from the same plasmid vector for the neutralization studies. PCR of VH and lambda/kappa VL genes was performed essentially as described by [74], [75]. Briefly, nested PCR was performed using PfuUltra II Fusion HS DNA Polymerase (Stratagene) using the primers described. The first round amplified the leader to constant regions of the VH and VL genes, using cDNA from a 19.3H-derived 293T clonal cell line or In-Fusion plasmid DNA as a template. The second round PCR was performed to amplify the variable regions. PCR products were gel purified, digested with appropriate enzymes (AgeI and SalI for VH, AgeI and XhoI for VL, all enzymes from NEB), and cloned into the plasmid expression vectors kindly provided by Dr. Patrick Wilson (heavy - accession number FJ475055, lambda - accession number FJ517647). Plasmids were grown in One Shot TOP10 chemically competent E. coli cells (Invitrogen) and purified with a QIAprep spin miniprep kit (Qiagen). At least three separate colonies were picked and sequenced. In the end, one VH and two somatically related lambda VL genes were recovered from five 19.3H-derived 293T clonal cell lines. The VH combined with either of the VLs (but not randomly with VLs from other R880F B cell cultures) produced robust neutralizing activity against the R880F founder Envs 0-A6 and 0-B24. Further characterization of the mAbs against the larger panel of R880F Envs revealed that the VLs had distinct neutralizing capacities when combined with the 19.3H VH, but no neutralizing activity when combined randomly with R880F VHs from other B cell wells. The mAbs containing the different VLs were then designated 19.3H-L1 and 19.3H-L3.
293T cells were cultured in T-75 flasks in DMEM with 10% FBS until 80% confluency was reached. Equal amounts (6 µg) of VH- and VL-containing plasmids were mixed with FuGENE HD (Roche) at a 1∶3 ratio and used for transfection. After 24 hr, media was removed, cells were washed twice with PBS, and the media was replaced with basal media (DMEM, 1% PSG, 1% Nutridoma SP). Cells were incubated for four days at 37°C, after which the supernatant was harvested. Cell debris was removed by centrifugation at 1,500 rpm for 5 min. Approximately 50 ml culture supernatant was used for antibody purification using a Protein A/G Spin column (Pierce) according to manufacturer's instructions. Purified antibodies were concentrated using Vivaspin concentrators (GE), and protein concentrations were determined using a Nanodrop spectrophotometer (BioTek).
Four monoclonal antibodies were biotinylated with the EZ-Link Sulfo-NHS-LC-Biotinylation Kit (Thermo Scientific) for use in ELISA protocols: 19.3H-L1 and 19.3H-L3 isolated here from R880F, 6.4C isolated from Z205F [31], and PGT128 obtained through the IAVI Neutralizing Antibody Consortium (NAC) Protocol G mAb Reagent Program [6], [76]. For each mAb, 50 µg were diluted in 500 µl 1× PBS (0.1 M sodium phosphate, 0.15 M NaCl, pH 7.2) for a final protein concentration of 100 µg/ml. A 50-fold molar excess of biotin was incubated with each mAb for 1 hour at room temperature. Excess biotin was removed via Zeba Desalt Spin Column, per the manufacturer's instructions.
Reacti-Bind polystyrene 96-well plates (Thermo Scientific) were coated overnight at 4°C with 100 µl/well of 2 µg/ml R880F 0-A6/B24, R880F 0-A6/B24 I295R, or R880F 0-A6/B24 E338K purified gp120 protein (Life Technologies, GeneArt) in PBS. Note that blank control wells were coated with gp120 protein but were never subjected to mAb incubation to determine background absorbance, which averaged at 0.055, and assays were performed in duplicate. Plates were subsequently washed six times with 1× PBS-T (Thermo Scientific; 10 mM sodium phosphate, 0.15 M NaCl, 0.05% Tween-20) and blocked with 200 µl/well of 1× B3T buffer (150 mM NaCl, 50 mM Tris-HCl, 1 mM EDTA, 3.3% FBS, 2% BSA, 0.07% Tween-20) for 1 hour at 37°C in a CO2-free incubator. During this incubation step, a two-fold dilution series that spanned 11 wells was prepared in 1× B3T for each biotinylated mAb (19.3H-L1, 19.3H-L3, PGT128, or the negative control, 6.4C) to be tested for binding, beginning at a concentration of 10 µg/ml. Plates were washed six times with 1× PBS-T a second time, and 100 µl/well of serially-diluted mAb was incubated for 1 hour at 37°C. Plates were washed six times with 1× PBS-T a third time, and 100 µl/well of a 1∶10,000 dilution of high sensitivity streptavidin horseradish peroxidase (HRP) conjugate (Thermo Scientific) in 1× B3T was incubated for 1 hour at 37°C. After a final six-time wash with 1× PBS-T, 100 µl of room temperature SureBlue 3,3′,5,5′ tetramethylbenzidine (TMB) microwell peroxidase substrate solution (KPL) was added to each well and incubated for 5 minutes at room temperature. To cease colorimetric development, 100 µl/well of 2 M H2SO4 was added, and absorbance values at 450 nm were read with a BioTek Synergy HT multi-mode microplate reader. Data was retrieved with KC4 v3.4 software, and binding curves were generated using GraphPad Prism v5.0d.
The gp120 binding ELISA protocol was minimally modified to measure the competitive binding of multiple mAbs, via the following alterations: Only R880F 0-A6/B24 gp120 protein was used, and PGT128 was excluded from the competitions. The first of two 100 µl/well mAb incubation steps was performed via a three-fold dilution series that spanned 7 wells; here, each mAb to be tested for competition (19.3H-L1, 19.3H-L3, or the negative control, 6.4C) was prepared in 1× B3T, beginning at a concentration of 10 µg/ml. The second of two 100 µl/well mAb incubation steps involved addition of a constant 1 µg/ml biotinylated competitor (either 19.3H-L1 or 19.3H-L3) across all wells. Wells were washed six times with 1× PBS-T between these two 1 hour, 37°C incubations. To determine 100% binding for 1 µg/ml biotinylated 19.3H-L1, 19.3H-L3, and 6.4C, duplicate wells were incubated with 1× B3T only during the first mAb incubation step and the appropriate biotinylated competitor during the second. The average absorbance for biotinylated 6.4C alone was 0.056. Background absorbance averaged at 0.048.
Fab fragments of monoclonal antibodies 19.3H-L1 and 19.3H-L3 were crystallized using previously described methods [41], [77]–[79]. In short, Fab fragments were generated by papain digestion, purified using affinity and size exclusion chromatography, concentrated, and crystallized with the hanging drop method. Fab 19.3H-L1 was crystallized with a well solution containing 0.17 M (NH4)2SO4, 0.085 M cacodylate pH 6.5, 25.5% (w/v) polyethylene glycol (PEG) 8000, and 15% (v/v) glycerol. Fab 19.3H-L3 was crystallized with a well solution containing 28% PEG 4K, 0.17 M Li2SO4, 0.085 M Tris pH 8.5, and 15% glycerol. X-ray diffraction data were collected at beamline 23-ID-D GM/CA-CAT at the Advanced Photon Source of Argonne National Laboratory, and the data sets were processed using HKL2000 [80]. Crystal structures were solved by the molecular replacement method using MOLREP in CCP4 [81], [82]. A homologous Fab (PDB code 3NH7) was used as the starting model. The structures were refined using COOT [83] and PHENIX [84], and analyzed using ICM [85]. The Protein Data Bank (http://www.rcsb.org/pdb) accession numbers for the coordinates of the structures of Fabs 19.3H-L1 and 19.3H-L3 are 4F57 and 4F58, respectively.
|
10.1371/journal.pcbi.1004836 | Detecting Presymptomatic Infection Is Necessary to Forecast Major Epidemics in the Earliest Stages of Infectious Disease Outbreaks | We assess how presymptomatic infection affects predictability of infectious disease epidemics. We focus on whether or not a major outbreak (i.e. an epidemic that will go on to infect a large number of individuals) can be predicted reliably soon after initial cases of disease have appeared within a population. For emerging epidemics, significant time and effort is spent recording symptomatic cases. Scientific attention has often focused on improving statistical methodologies to estimate disease transmission parameters from these data. Here we show that, even if symptomatic cases are recorded perfectly, and disease spread parameters are estimated exactly, it is impossible to estimate the probability of a major outbreak without ambiguity. Our results therefore provide an upper bound on the accuracy of forecasts of major outbreaks that are constructed using data on symptomatic cases alone. Accurate prediction of whether or not an epidemic will occur requires records of symptomatic individuals to be supplemented with data concerning the true infection status of apparently uninfected individuals. To forecast likely future behavior in the earliest stages of an emerging outbreak, it is therefore vital to develop and deploy accurate diagnostic tests that can determine whether asymptomatic individuals are actually uninfected, or instead are infected but just do not yet show detectable symptoms.
| Emerging epidemics pose a significant challenge to human health worldwide. Accurate real-time forecasts of whether or not initial reports will be followed by a major outbreak are necessary for efficient deployment of control. For all infectious diseases, there is a delay between infection and the appearance of symptoms, i.e. an initial period following first infection during which infected individuals remain presymptomatic. We use mathematical modeling to evaluate the effect of presymptomatic infection on predictions of major epidemics. Our results show rigorously, for the first time, that precise estimates of the current number of infected individuals—and consequently the chance of a major outbreak in future—cannot be inferred from data on symptomatic cases alone. This is the case even if the values of epidemiological parameters, such as the average infection and death or recovery rates of individuals in the population, can be estimated accurately. Accurate prediction is in fact impossible without additional data from which the number of currently infected but as yet presymptomatic individuals can be deduced.
| A principal challenge in infectious disease epidemiology is quantifying the threat posed by disease early in emerging outbreaks [1,2]. During the earliest stages of infectious disease outbreaks, two main questions are i) will a major epidemic occur, and ii) what will the final size of the outbreak be [3]? Answering the second of these questions is impossible without understanding the answer to the first. We therefore focus on predicting whether or not reports of initial cases will be followed by a major outbreak of disease, in which a large number of individuals become infected [4–7]. Accurate real-time forecasts at the start of emerging outbreaks are essential for efficient deployment of limited resources for control [8,9]. However, the dynamics of infectious disease outbreaks are influenced by the incubation period, within which hosts are infected but do not yet show symptoms [10–12]. We use mathematical modeling to investigate how the consequent ambiguity in the number of hosts that are currently infected confounds prediction in the earliest stages of a potential major outbreak.
The basic reproductive number, R0, the average number of secondary cases caused by a single infection in a totally susceptible population, justifiably dominates any discussion of infectious disease epidemiology [4,5,13]. If R0 < 1, there will certainly not be a major outbreak [14]. When R0 is above this threshold, however, major outbreaks can but do not always occur [15,16]. In large populations, the distribution of epidemic sizes is bimodal when R0 > 1, and either the disease dies out with very few ever becoming infected, or it becomes widespread [5,7] (S1 Fig). A major outbreak can therefore naturally be defined as one where the disease becomes widespread, i.e. the total number of hosts that ever become infected lies in the part of the distribution of possible final sizes that contains the larger mode. A well-known approximation to the probability of a major outbreak in a large population can be derived from simple stochastic epidemiological models,
Prob(Major outbreak) ≈ 1−(1R0)I*,
where I* is the number that are infected at the time of estimation [6]. This estimate is widespread in the theoretical epidemiology literature particularly in the case where disease first arrives in the system and so I* = 1 [5,15,17–22]. We note that this formula has also been used in the context of the spread of the recent Ebola outbreak to Nigeria, to estimate the chance that a single undetected infected case will spark a major outbreak [23]. More sophisticated approximations to the probability of a major outbreak can be derived for models containing additional epidemiological detail, for example population structure [24,25], more refined models of individuals’ infectious periods [6], and differences in infectivity between individuals [13]. Crucially, however, the approximation above illustrates that estimates of the probability of a major outbreak require knowledge not only of the values of disease transmission parameters, but also of the total number of currently infected hosts, I*. This includes those individuals that have not yet developed symptoms. Modelers have concentrated on developing increasingly elaborate statistical machinery to estimate the parameters that constitute R0 [26–28]. Other work, most notably back-calculation [29,30], focuses on estimating the number of individuals that are currently infected, accounting for delays before symptoms emerge. What has never been examined, however, is how the lack of knowledge of precisely how many are infected in the early stages of a potential major outbreak affects predictability of whether or not a large epidemic will in fact go on to occur. In practice, epidemic forecasts for specific pathogens are typically conducted via simulation [11,31–36]. Consequently, we conduct a simulation-based study into the impact of presymptomatic infection—which the formula from the theoretical epidemiology literature suggests might disrupt forecasting—on predictions of major epidemics.
As an example of a disease for which initial cases are frequently not followed by major outbreaks, and with a significant delay between infection and emergence of symptoms, we consider Ebola virus disease. All five strains of the genus Ebolavirus cause severe acute illness, with early non-specific symptoms including asthenia and myalgia typically followed by nausea, vomiting, hemorrhagic symptoms and, in a significant proportion of cases, death [37]. There are reports of cases of Ebola in remote villages in Central and West Africa every few years [38], hypothesized to be initiated by spillover from reservoirs of infection in wild animal populations, with fruit bats most often implicated as the reservoir host [39]. Often there is no sustained human-to-human transmission, and initial cases do not lead to large outbreaks. However, since 1976 there have been twenty-five distinct reports of primary infection in humans, of which sixteen have led to epidemics causing more than twenty deaths. The largest ever Ebola outbreak started in Guinea in December 2013 and subsequently spread to and caused widespread transmission in Liberia and Sierra Leone, with additional cases in Nigeria, Mali, Senegal, Spain, USA, UK and Italy. This epidemic caused more than 11,000 fatalities before it was declared officially over by the World Health Organization on 14th January 2016, although an additional death was confirmed the following day and additional small flare-ups are still possible [40].
Modeling studies of Ebola have tended to focus on parameter estimation [41–43] and the potential effects of disease control [31,32,37,44]. Here, we instead focus on using an existing epidemiological model fitted to data from the outbreak in Uganda that killed 224 people in 2000 [45] to show how presymptomatic infection affects our ability to predict whether or not reports of initial cases will go on to cause a major outbreak. Ebola is therefore a motivating example for our investigation into how presymptomatic infection affects the predictability of infectious disease epidemics. However, since presymptomatic infection is ubiquitous, our conclusions are applicable to a wide range of pathogens.
We use simulations of stochastic compartmental epidemic models to drive our analyses. The models assume that, at any time, every member of the population belongs to a compartment describing their infection and symptom status. In a single realization of the model, whether or not an individual becomes infected is a random process. If an individual does become infected, then the model generates the time at which the individual is first infected, the time at which symptoms first appear and the time at which the individual either dies or recovers. These times are simply those at which the individual passes into the relevant compartments of the model.
We therefore produce a “dataset” for the start of an outbreak by running a simulation model. We “freeze” the outbreak at the time of the fourth death, and calculate two quantities using the model (Fig 1): the probability of a major outbreak given complete observation of presymptomatic cases (hereafter referred to as the “true” probability of a major outbreak), and the estimate of this probability that only uses data on the timings of symptoms and deaths and not the times at which individuals are initially infected. In this estimated probability, the presymptomatic cases remain hidden and the number of presymptomatic infected individuals is estimated from the data on symptoms and deaths.
For an individual outbreak, a confidence interval can be constructed around the point estimate that we consider. For the outbreak in Fig 1, the distributional estimate of the number of exposed individuals leads to a 95% confidence interval for the current number of infected individuals of [1,6], which corresponds to an extremely wide 95% confidence interval for the probability of a major outbreak of [0.24,0.78]. The point estimate corresponds to a weighted sum over the distributional estimate of the probability of a major outbreak.
Our initial analysis considers a simplified, SEIR model with exponential waiting times in each compartment. In the SEIR model, presymptomatic infecteds are confined to the uninfectious, latently infected (E) class. We relax this assumption later, and also consider the case where the waiting times follow gamma, rather than exponential distributions. Recent modeling work, focused on the recent outbreak of Ebola, often includes considerable epidemiological detail [31,32,44], although there are some exceptions [42,43,46]. Here we take advantage of a previous parameterization of the SEIR model for Ebola, noting that the SEIR model is widely-used for a number of diseases and captures what we want to investigate here—i.e. the effect of presymptomatic infection on major outbreak forecasting—in the simplest possible way.
The true and estimated probabilities of a major epidemic are calculated for many simulated outbreaks, to investigate how presymptomatic infection affects the ability to predict major epidemics early in outbreaks (Fig 2A). The probability of a major outbreak depends on the number of infected individuals at the time of estimation (S2 Fig), and hidden presymptomatic infection therefore frustrates prediction. This is even the case when there are actually no presymptomatic infected individuals in the population, since the distribution that estimates the number of presymptomatic individuals will include values other than zero.
In the SEIR model, only discrete values of the true probability of a major outbreak are possible, since the true probability is entirely controlled by the total number of infected individuals at the time of estimation. However, each individual dataset, corresponding to a separate realization of an outbreak, consists of the times that individuals become symptomatic. Slight variations in these times lead to different probability distributions of the number of presymptomatic infecteds. These differences are reflected in the estimated probability of a major outbreak. Consequently, the estimated probability for each true value is effectively a continuously varying quantity.
The key qualitative result, i.e. that the estimated and true probabilities of a major outbreak do not match, is robust to performing estimation at different stages of the start of an outbreak, and to different lengths of the incubation period and values of R0 (S3–S6 Figs). Additional uncertainty in the probability of a major outbreak occurs when the parameters for disease spread must also be estimated from the transmission data (S7 Fig). However, no matter how much parameter estimation is improved, for example using data from previous outbreaks to inform estimates, presymptomatic infection still causes significant errors in forecasting major outbreaks.
The problem of practical interest for an emerging epidemic is inferring the true probability of a major outbreak. For an individual outbreak, a (often imprecise) confidence interval can be constructed around the point estimate as we described above. However, we characterize the implications of presymptomatic infection more generally by examining many simulated outbreaks, inverting our point estimate of the probability of a major outbreak to consider the range of true probabilities that are possible for each estimated value. Similar estimated probabilities of a major outbreak can correspond to a remarkably wide range of true probabilities (Fig 2B). For example, for outbreaks in which the estimated probability is between 0.5 and 0.6, the true probability can lie between 0.23 and 0.83. We note the extreme values are themselves quite likely: in 13% of these simulated outbreaks, the true probability is in fact either 0.23 or 0.83.
Estimation of the chance of a major outbreak can be improved by the use of diagnostic tests to determine whether asymptomatic individuals are susceptible or presymptomatic infected. Since the reliability of diagnostic tests affects the extent to which forecasting is improved (Fig 3), it is not only important to develop diagnostic tests but also to ensure their continued refinement. To illustrate the general principle that diagnostic tests could be used to improve prediction, we simply choose individuals to test at random from the asymptomatic individuals in the population. With random selection, the diagnostic test must be deployed widely to reduce the error in estimates significantly, although of course careful choice of which individuals to test (e.g. via contact tracing) would reduce the need for such widespread deployment in practice.
The emergence of symptoms and the emergence of infectivity are assumed to coincide in the SEIR model. We relax this assumption by considering two other models. In the first, individuals display symptoms before becoming infectious (Fig 4A). In the second, individuals are infectious before becoming symptomatic (Fig 4B). When symptoms appear before individuals are infectious, the incubation period is reduced, so more infected individuals can be detected. As a result, predictions of major outbreaks become more accurate, although some systematic ambiguity nevertheless remains (Figs 4A and S8A). Conversely, if the incubation period is instead longer than the latent period, as is the case for many human diseases [47], it becomes more difficult to predict major outbreaks accurately (Figs 4B and S8B).
In Fig 4B, the variable heights of adjacent boxplots indicate that the distribution of infected individuals between the asymptomatic and symptomatic infectious classes affects the estimated probability of a major outbreak. For example, the heights of the second and third boxplots from the left can be explained as follows. Consider two outbreaks, each with only a single infected individual at the time that the chance of a major outbreak is being estimated. Suppose that in the first outbreak (“outbreak one”, say), the infected individual is presymptomatic, but in the second outbreak (“outbreak two”) the infected individual is symptomatic. In outbreak two, because disease is observed since the infected individual is symptomatic, the estimated probability of a major outbreak will be high compared to outbreak one. However, whilst the estimated probability of a major outbreak is higher for outbreak two, the true probability of a major outbreak is in fact higher for outbreak one. This is because, in Fig 4B, individuals can be infectious both when they are presymptomatic and when they are symptomatic. A presymptomatic individual is therefore likely to be infectious for a longer period in future than a symptomatic individual. A longer time infectious corresponds to (on average) more infections, and therefore a higher true probability of a major outbreak.
It might also naïvely be thought that prediction would be easiest in outbreaks in which many infected individuals are symptomatic. However, when a large proportion of infected individuals are symptomatic, the total number of infected individuals tends to be overestimated, causing large errors in forecasts (cf. the boxplots corresponding to a single infected individual at the time of estimation in Fig 4B).
The default assumption for compartmental models is that incubation and infectious periods are exponentially distributed. We relax this assumption, and draw periods instead from two-parameter gamma distributions to reflect the observed incubation and infectious periods for a number of diseases [49,50] (S9 Fig). Recently-infected individuals are more likely to remain infectious for a long period beyond the time of estimation than individuals that have already been infected for a long time. Consequently, outbreaks with many presymptomatic infecteds have a high true probability of a major outbreak. However, because presymptomatic individuals are unobserved, the estimated probability of a major outbreak is low in these outbreaks.
Predicting whether or not a major epidemic is likely, from the limited data typically available during the first few days of an outbreak, has received surprisingly little attention. A notable exception is the paper by Drake [51], which shows that the exact final size varies significantly between simulated outbreaks under identical conditions. He investigates how this variability scales with the contact rate between individuals and the efficacy and speed of control responses. However an incubation period is not explicitly included in the model used. Craft et al. [7] use a model of rabies in canids to show that the first four death times cannot be used to forecast major outbreaks. However, by assuming that the data consist of death times alone, the factors potentially responsible for this imprecision are confounded. Neither Drake [51] nor Craft et al. [7] quantify the error caused by presymptomatic infection. In addition to quantifying this error, our main message is that presymptomatic infection by itself is sufficient to cause error in predictions of whether or not an outbreak will be major, let alone in predicting the final size exactly. This error is particularly notable when there are no infected individuals in the population at all (i.e. the outbreak has already faded out), since the distribution that estimates the number of presymptomatic infected individuals will include values other than zero.
To focus entirely on the uncertainty caused by presymptomatic infection, we worked in an idealized setting in which symptomatic cases and deaths were recorded perfectly and in which the values of disease transmission parameters were known exactly. This allowed us to calculate the exact probability distribution of the current size of the outbreak, i.e. the total number of individuals currently infected, given that presymptomatic infection causes some infected individuals to be unobservable. This distribution drives the estimated probability of a major outbreak.
In practice, however, the distribution of possible current outbreak sizes would have to be estimated from incomplete data on symptomatic cases and deaths, without exact knowledge of parameter values and sometimes without even knowing the total population size precisely. One method for doing this is back-calculation, as originally designed by Brookmeyer and Gail for HIV-AIDS [29], which provides an estimate for the distribution of possible current outbreak sizes. Although, to the best of our knowledge, back-calculation has not been used to estimate the probability of a major outbreak, such a forecast using back-calculation as an input would necessarily be less precise than those used in our analyses here, since we have used the exact distribution of current outbreak sizes given presymptomatic infection. Indeed, by restricting our attention to the case in which there are sufficient data that the number of presymptomatic individuals is the only quantity being estimated, our results provide an upper bound on the ability of any method that seeks to predict major outbreaks from data on symptomatic cases alone. In fact, given the extensive knowledge of the epidemic assumed here, the basic formulation of back-calculation can be extended in a natural fashion to obtain the exact probability distribution of the current size of the outbreak that we use to generate our estimates for the probability of a major outbreak (S4 Text, S10 Fig).
Prediction during the recent Ebola outbreak has been criticized for overestimating the total number of cases that actually occurred [52]. Similarly, modeling studies during the 2009 H1N1 outbreak typically overestimated the total number of cases [53]. In contrast with investigations that attempt to predict the final epidemic size, we differentiated only between “minor” and “major” outbreaks. Our focus was prediction during the very early stages of an outbreak, before a major outbreak is underway, rather than forecasting the final extent of a major outbreak once the epidemic has taken off. This very initial phase of outbreaks is particularly important given the recent interest in rapid detection of disease outbreaks [54–57].
We assumed that the parameter values controlling disease spread are unchanged throughout the early stage of the outbreak, whereas in reality these parameters might vary temporally in response to changing contact networks and control interventions [58], as well as varying environmental conditions [59]. However, any such variations will only exacerbate the uncertainty that we have shown exists. Other sources of uncertainty such as under-reporting, which has posed a challenge to forecasting during the recent Ebola outbreak [60], will also decrease predictability further, although as we have shown presymptomatic infection alone is sufficient to make precise prediction impossible. A systematic investigation of the errors in forecasting caused by under-reporting in comparison to those due to other features such as presymptomatic infection or epidemiological parameter uncertainty is a possibility for a future study.
Our work shows rigorously, for the first time, that no matter how accurately disease transmission parameters are estimated, precise estimates early in outbreaks of whether a major epidemic will occur will remain unavailable without data about presymptomatic infection. This is still the case even if significant resources are devoted to recording symptomatic cases accurately. Consequently, diagnostic tests that can identify presymptomatic infecteds [61,62] are extremely important for improving forecasts of epidemic outbreaks. While our simulations consider random testing of asymptomatic individuals, in practice testing is costly [63], so it is vital that predictability is further improved in a cost-effective way by careful selection of individuals to test. This could be done by contact tracing [64] or using statistical methods to identify individuals with the highest risk of being infected [65], although of course effective and cheap diagnostics are still required. A systematic investigation into which asymptomatic individuals ought to be tested, accounting for the specificity of the tests as well as the sensitivity, would be a valuable extension to our work. A recent analysis of Ebola [66] has considered testing of individuals already exhibiting symptoms to confirm whether the patients have Ebola or a different disease with similar symptoms. That study shows that using rapid diagnostic tests in combination with slower but more accurate diagnostic tests could have significantly reduced the number of cases in Sierra Leone in the recent outbreak.
Our conclusions are robust to various characteristics of the disease, and so apply to all infectious diseases. We chose to use Ebola as a representative case study, but our results are in fact generic. In particular, our key message that presymptomatic infection drives uncertainty in whether an emerging outbreak will become major holds throughout the early stages of the outbreak (S3 and S4 Figs), as well for a number of values of the basic reproduction number of the pathogen (S6 Fig). For Ebola, there is debate as to whether the onset of symptoms and infectiousness coincide [67] or not [68]. However, symptoms and infectiousness are certainly not always concurrent: HIV is a high profile example, for which the time between infection and recognizable symptoms can take years [69], whereas individuals are infectious within months of acquiring the virus [70]. We have considered different models in which symptoms and infectiousness are not assumed to coincide (Figs 4 and S8). While we showed prediction is most reliable for diseases for which the incubation period is shorter than the latent period, even very short incubation periods can generate significant uncertainty in the number of presymptomatic infecteds, and therefore the probability of a major outbreak (S5 Fig). This means that our conclusions even hold for diseases such as influenza and norovirus, which have incubation periods of only a few days [4]. The messages we have set out are also robust to different distributions of the incubation and infectious periods, as we showed by considering models for which these periods follow gamma rather than exponential distributions (S9 Fig).
Of course, our conclusions are relevant to pathogens of agricultural and wild animals and plants, as well as humans. Xylella fastidiosa is a plant pathogen that is currently invading southern Italy, causing devastating damage to olive groves [71]. Containment and surveillance zones have been set up in an attempt to find the pathogen and subsequently mitigate spread via control interventions. Surveys in the containment zone do include some laboratory testing for presymptomatic infection, with the surveillance zone solely relying on diagnosis from visual inspection [72]. We have shown that consideration of presymptomatic infection is extremely important when forecasting the spread of pathogens, and so it is also likely to be important when planning interventions that attempt to slow or prevent spread. Studies examining the impacts of presymptomatic infection on forecasting and control of specific pathogens would represent valuable applied extensions to this publication.
At the time of writing, a point-of-care diagnostic test that can detect Ebola from blood samples has been developed and found to be accurate [73]. In light of our analysis, the continued development, deployment and improvement of this and other diagnostic tests that determine whether asymptomatic individuals are infected is of obvious public health importance, not only for Ebola but also for other infectious diseases.
We perform our analyses using stochastic compartmental models of disease spreading in a small population. Here we outline the three types of model we use: the standard SEIR model, which assumes that symptoms and infectiousness coincide; more complex models that relax this assumption; and a model that assumes that the incubation and infectious periods follow gamma, rather than exponential, distributions.
Since our concern is quantifying uncertainty caused by presymptomatic infection alone, we assume that the parameters controlling disease transmission are known, and that complete data are available from the very beginning of the epidemic for changes in the number of symptomatic infected individuals over time. These data can be used to construct the probability distribution for the number of presymptomatic infected individuals at the time of estimation (S1 Text). For the SEIR model, the data on symptomatic cases are used to estimate the probability that an asymptomatic individual is infected, which feeds into a binomial distribution to estimate the number of presymptomatic infected individuals. The approach can readily be adapted for the SEUIR and gamma-distributed incubation and infectious periods cases. In the SEAIR model case, the A class causes the complete time series of infectious individuals to be unobserved, so that the required probability cannot be calculated. Instead reversible jump Markov chain Monte Carlo (S2 Text) is used to estimate the probability distribution for the number of currently infected individuals.
To illustrate the principle that diagnostic tests can improve forecasts, the sampling of asymptomatic individuals and testing to find presymptomatic infection is modeled by choosing individuals at random out of the S or E classes without replacement. If the individual is susceptible, then infection is not detected (i.e. the test produces no false positives), whereas if the individual is presymptomatic infected, the pathogen is detected with probability pd. The results of the sample can then be integrated into the estimate of the probability distribution of the number of presymptomatic infected individuals, which therefore becomes more precise (S3 Text).
We estimate two probabilities using data from individual simulated epidemics at the time of the fourth death: the true probability of a major outbreak, and the best point estimate of this probability consistent with the transmission data. Specifically, we calculate the true probability of an outbreak by “freezing” the infection status of all individuals at the time of four deaths, simulating a very large number of outbreaks (100,000) using these data as initial conditions, and finding the proportion of simulations in which a major outbreak occurs (defined as more than 10% of the population ever becoming infected, cf. S1 Fig). Of course, this calculation is only possible since the number of presymptomatic infected individuals is known.
To calculate the estimated probability of a major outbreak, we instead imagine that the exact infection statuses of individuals that are asymptomatic (i.e. susceptible individuals and presymptomatic infected individuals) are unknown, as would be the case in practice. We use the data on symptomatic cases up to the time of the fourth death to infer the probability distribution of the number of presymptomatic infecteds. We then calculate the estimated probability of a major outbreak by running an ensemble of simulations that sample initial conditions from this distribution on each forward run.
|
10.1371/journal.pcbi.1004625 | Recombination Enhances HIV-1 Envelope Diversity by Facilitating the Survival of Latent Genomic Fragments in the Plasma Virus Population | HIV-1 is subject to immune pressure exerted by the host, giving variants that escape the immune response an advantage. Virus released from activated latent cells competes against variants that have continually evolved and adapted to host immune pressure. Nevertheless, there is increasing evidence that virus displaying a signal of latency survives in patient plasma despite having reduced fitness due to long-term immune memory. We investigated the survival of virus with latent envelope genomic fragments by simulating within-host HIV-1 sequence evolution and the cycling of viral lineages in and out of the latent reservoir. Our model incorporates a detailed mutation process including nucleotide substitution, recombination, latent reservoir dynamics, diversifying selection pressure driven by the immune response, and purifying selection pressure asserted by deleterious mutations. We evaluated the ability of our model to capture sequence evolution in vivo by comparing our simulated sequences to HIV-1 envelope sequence data from 16 HIV-infected untreated patients. Empirical sequence divergence and diversity measures were qualitatively and quantitatively similar to those of our simulated HIV-1 populations, suggesting that our model invokes realistic trends of HIV-1 genetic evolution. Moreover, reconstructed phylogenies of simulated and patient HIV-1 populations showed similar topological structures. Our simulation results suggest that recombination is a key mechanism facilitating the persistence of virus with latent envelope genomic fragments in the productively infected cell population. Recombination increased the survival probability of latent virus forms approximately 13-fold. Prevalence of virus with latent fragments in productively infected cells was observed in only 2% of simulations when we ignored recombination, while the proportion increased to 27% of simulations when we allowed recombination. We also found that the selection pressures exerted by different fitness landscapes influenced the shape of phylogenies, diversity trends, and survival of virus with latent genomic fragments. Our model predicts that the persistence of latent genomic fragments from multiple different ancestral origins increases sequence diversity in plasma for reasonable fitness landscapes.
| Increasing evidence suggests that HIV-1 released from activated latent cells survives in productively infected cells in patient plasma despite competition against better adapted virus variants that have evolved in response to the host immune pressure. Long-term survival requires that latent virus forms adapt to the host immune response so that they are not outcompeted. We simulated the dynamics of HIV-1 envelope sequence evolution in response to host immune pressure to investigate how virus from activated latent cells can survive despite having reduced fitness compared to the more evolved virus variants in patient plasma. The evolutionary trends of our simulated virus populations followed closely those observed in HIV-1 sequence data from 16 patients. Our simulation results suggest that recombination facilitates the survival of genomic fragments originating from virus activated from latent cells. Our model further predicts that sequence diversity increases with the number of latent genomic fragments from different origins that persist in plasma.
| Patients infected with HIV-1 require lifelong highly active antiretroviral therapy (HAART) to suppress infection. Treatment cessation typically leads to HIV viral rebound to pre-therapy levels; the resurgence is thought to be associated with activation of long-lived, latently HIV-infected cells. A cure for HIV therefore requires either clearance of all cells harboring latent virus, or prevention of virus release from the reservoirs after discontinuation of treatment. Increasing evidence suggests that latency plays an integral role throughout the life cycle of the virus. We recently observed that the majority of HIV-1 plasma sequences in two untreated chronically infected patients had accumulated significantly less mutations than expected, suggesting a period of latency during which no replication occurred in the history of these lineages [1]. Moreover, viral variants with reduced evolution consistent with periods of latency are frequently involved in transmission events [2–4]. While recent advances have shed light on the mechanisms leading to the establishment and maintenance of latent reservoirs [5, 6], the prevalence of viral sequences displaying a signal of latency in the replicating population remains enigmatic, especially in the absence of antiviral treatment.
HIV-1 is subject to selection pressure exerted by the immune system; strains that can avoid the immune response have an advantage within a host. The neutralizing antibody response in patient sera is much stronger against virus circulating in infection earlier than contemporaneous variants, with immunological memory persisting for years [7–9]. Virus from activated latent cells is therefore less fit due to long-term immunological memory than variants that have continually evolved in response to the host immune pressure. Yet, we detect a signal of latency in the replicating population, suggesting that virus from activated latent cells persists despite competing against better-adapted contemporaneous variants.
In order to survive, activated latent forms must either have a replicative advantage outweighing immune selection, or quickly acquire escape mutations to evade the immune response. While mutations associated with immune escape decrease the replicative capacity of the virus between 0–24% [10], compensatory mutations can completely restore fitness in some variants [11]. It is thus unknown whether virus from activated latent cells is able to compete against better adapted contemporaneous virus due to higher replicative fitness. To persist in plasma after activation, the virus must adapt to immune pressure before being outcompeted. The virus may adapt by sequentially accumulating escape mutations at recognized epitopes, or by recombining with contemporaneous virus and simultaneously gaining multiple epitopes not recognized by the immune system.
Recombination provides a mechanism to preserve genomic fragments originating from activated latent cells within contemporaneous virus backbones adapted to the immune response. Further recombination events may propagate latent genomic fragments through the replicating plasma population, enhancing the adaptability potential of the virus. In this paper, we investigate the persistence of latent envelope genomic fragments in the replicating plasma population by developing a detailed model of HIV-1 sequence evolution and the cycling of lineages in and out of the latent reservoir.
There has been an increasing focus on modeling the dynamics of the latent reservoir, particularly in the context of delineating the effect of latency on viral blips during treatment, and the re-emergence of infection after treatment failure/interruption [12–16]. A common feature of these models is the cycling of infected cells through a latent compartment, in which they may divide or die before becoming activated and re-joining the productively infected cell population. Our agent-based model integrates the dynamics of the latent reservoir proposed in [12–14], with a sequence evolution framework first introduced in [17, 18], which follows how individual viruses mutate and recombine.
Vijay et al. developed an HIV sequence evolution model of infected patients that incorporates mutation, infection of cells by multiple virions, recombination, fitness selection, and epistatic interactions between multiple loci [17]. Their model, which describes quantitatively the evolution of HIV-1 diversity and divergence in patients, has been applied to a wide variety of questions, including the effect of recombination on sequence diversification [17], the effective population size of HIV-1 [18], the genetic structure of HIV-1 quasispecies and its error threshold [19], and the fraction of progeny viruses that must incorporate a drug treatment target for suppression of productive infection [20].
We extend the sequence evolution model by incorporating a latent compartment in which no evolution takes place, and by explicitly simulating how the virus population interacts with and adapts to the immune response. To evaluate whether our model captures sequence evolution in vivo, we compare our simulated sequences to sequences from 16 untreated or unsuccessfully treated patients followed longitudinally from seroconversion [9, 21, 22] in terms of sequence divergence (i.e. average distance of a population of sequences from a founder sequence), diversity (i.e. average pairwise distance between all sequences in a population), and phylogenetic tree shape measures.
We developed an agent-based model to simulate within-host HIV-1 sequence evolution after primary HIV-1 infection (PHI), tracking how long each virus lineage has spent in the latent reservoir throughout its history (Fig 1). The model incorporates a detailed mutation process including recombination, latent reservoir dynamics, diversifying selection pressure driven by the immune response, and purifying selection pressure asserted by deleterious mutations. During each generation, a population of virus infects a population of uninfected cells, with infecting viruses chosen according to their relative fitness. Genetic variability is introduced through recombination and substitution during reverse transcription of viral RNAs to proviral DNA in infected cells. Infected cells have a nonzero probability of moving into the latent reservoir, while latent cells have a nonzero probability of dying, proliferating, or becoming activated. Infected cells and activated latent cells produce progeny virus, forming the population of virus infecting the next generation of cells.
The relative fitness of each virus depends on whether it has acquired mutations at invariant sites, which are considered deleterious, and how well it is recognized by the immune system. All virus sequences have multiple epitopes at pre-defined sites, which are presented to the immune system. If an epitope variant reaches a sufficient frequency in the plasma, it elicits an immune response, which remains in memory for the duration of the simulation. The breadth of the immune response is updated each generation by adding any newly recognized epitope variants to immunological memory.
While we simulate infections established by a single founder strain, the model can also accommodate multiple founder strains, which are observed in 20% of heterosexual cases [23]. The simulations were implemented using a computer program written in R [24], and the phylogenies were created using the R package ape (Analyses of Phylogenetics and Evolution) [25]. The algorithm is outlined in the supplement (S1 Text).
We initialized the simulation by generating a set of NV identical virions, where each virion consists of a string of L nucleotides (A, U, G, C). While HIV contains two RNA molecules, we considered only a single RNA molecule for each virus to simplify computations. Because recombination between two identical parental strains does not lead to new genetic variants, restricting our simulations to one RNA molecule per virus should have a negligible effect on our simulation results. We chose as the starting sequence the consensus of the first time point sequences, sampled close to seroconversion, of Patient A from the Amsterdam cohort [9]. The sequence is from env, and consists of approximately 700 nucleotides.
At each generation, NC viruses are chosen to infect NC uninfected cells such that each cell is infected with a single virus. Multiple infection of cells is described below. The probability of selecting a given virus is proportional to its relative fitness, which is a function of the strength of the immune response against it, and the number of deleterious mutations it has acquired. Fitness selection is described in detail below. Upon infection each cell has a small probability η of becoming latent and moving to the latent reservoir. The virus in the infected cells that remain productive undergo reverse transcription, where the viral RNA is copied into proviral DNA. This process includes both recombination and substitution.
To simplify computations, we only considered dual infections that lead to recombination. Based on estimates of Josefsson et al. [26], we assumed that 10% of cells are infected with a second virion, ignoring multipe infection with three or more virions. On average half of the progeny virus produced by a cell infected with two virions acquires one RNA molecule from each parent. To simplify computations, we allowed at most one cross-over event per sequence. This simplification is justified because while there are on average 2.8 cross-overs per each HIV-1 genome of 10,000 nucleotides every life cycle [27], our simulated env sequences were over 10 times shorter.
For a template switching rate of 3 × 10−4 per site per generation [27, 28], the expected number of cells accommodating a recombination event between two different parental strains is NR = 0.1 × 0.5 × 3 × 10−4 × NC L. We incorporated dual infection into the model by sampling NR additional viruses from the virus population with probability proportional to their relative fitness and adding them to randomly chosen infected cells. The dually infected cells then undergo recombination.
We assumed that cross-over events are distributed uniformly across the genome, selecting the starting strand and the cross-over position ∈ {2,L − 1} randomly. The sequence of the starting strand is copied until the cross-over position, the strand is then switched, and the second strand is copied to the end of the sequence. The recombinant sequence replaces the two parent strands, so that all infected cells contain a single viral sequence. The sequence is then mutated according to a general-time-reversible substitution model [29]. We used the nucleotide-specific substitution rates estimated from the sequences of Patient A via maximum likelihood (PAUP*, [30]), which were representative of the rates found for all 16 patients. The proviral DNA of each infected cell is transcribed to viral RNAs, which are released as new virions. Each infected cell produces P identical virions, which form the current population of virus from which viruses that infect the next generation of cells are selected.
As the latent reservoir is established early in infection [31, 32], we initialized the reservoir by infecting each of the NL cells with the same founder virus used to infect the replicating population. During effective antiretroviral treatment, the reservoir’s half-life has been estimated to range from 4 to 44 months [33–36]. Assuming that the size of the latent reservoir, NL, remains constant during untreated chronic infection, we set the probability of an infected cell becoming latent equal to the decay rate of the latent reservoir. Each generation, we randomly select which (if any) infected cells become latent.
For simplicity, we assumed homogeneous activation and death rates, randomly selecting which cells in the latent compartment die or become activated each generation. Activated latent cells join the population of productively infected cells and produce progeny virus. We assumed that the latent reservoir is maintained through homeostatic proliferation [12, 13, 37], which we simulated by randomly duplicating cells to replace those lost to death and activation to keep the size of the latent pool constant. Specifically, if the size of latent reservoir NL after the addition of any newly latent cells and the removal of any activated or dead cells is less than the target size N L *, we randomly duplicate N L * - N L of the remaining cells in the latent compartment. No proliferation takes place if the size of the latent reservoir exceeds N L *.
The model tracks how many generations each cell spends in the latent reservoir before being activated, with progeny viruses retaining the information for the duration of the lineage. Because viruses that have been latent for different amounts of time can recombine, we track latency separately at each site in the genome. If a virus that has never been latent recombines with a virus that has been latent for n generations, we consider the recombinant progeny a latent virus form, with the portion of its genome acquired from the non-latent parent recorded as having been latent for zero generations, and the portion acquired from the latent parent as having been latent n generations. At the end of the simulation, we know the position and age of every latent genomic fragment for each virus sequence. We are interested in the persistence of sequences that contain at least one latent genomic fragment consisting of at least one site originally from a latent ancestor.
The immune response against antigens with recognized epitopes consists of two major arms; neutralization of a pathogen via antibodies, and killing of infected cells by cell-mediated immune responses. Because we simulated the viral envelope, we focused primarily on the antibody response, assuming that 1) variants with a recognized epitope are less fit than those able to evade the immune response, and 2) once an epitope is recognized, it is retained in memory for all time. The immune response imposes diversifying selection pressure on the virus population, resulting in sequential escape at each epitope. A virus variant not recognized by the immune system has a fitness advantage, and starts to take over the virus population, eventually eliciting an immune response, and thus selection for new escape variants.
We chose the location of NE epitopes randomly in the envelope fragment that we are simulating. Epitopes are approximately 30 nucleotides in length, and may be non-contiguous and overlapping [38]. As synonymous mutations at the nucleotide level have no effect at the amino acid level, we ignored third-codon positions, and assumed that the epitopes are 20 nucleotides long and non-overlapping.
Serum from HIV-1 infected individuals does not neutralize contemporaneous virus, but rather virus that dominated the population 3 to 6 months earlier [7]. To model this behavior, we imposed an immune response against a new escape variant not when it first appeared but after it had risen in frequency in the plasma, and increased the strength of the subsequent response gradually over time.
We began the immune response against the virus after 30 generations. Every generation thereafter, we determined which epitope variants had reached a sufficient frequency, f, in the plasma, and initiated an immune response against them. The epitopes were stored in immunological memory for the duration of the simulation, and all virus variants containing them had a fitness disadvantage. For simplicity, we assumed that to be recognized the nucleotide sequence of a virus epitope must match perfectly with the sequence of a stored epitope. Note that virus may be recognized at multiple epitopes simultaneously.
Because antibody responses mature over time, we assumed that the fitness cost to the virus imposed by the immune response to epitope i of variant j increases linearly until it reaches some maximum value c i *. Thus, the fitness cost of epitope i of variant j is given by
c i j ( t ) = m i n { c i * , c i * ( t - t i j 0 ) / d } δ i j , t ≥ t i j 0 ,
where t i j 0 is the time when the immune response against epitope i of variant j is introduced, d is the time it takes to reach full potency, and δij is the Dirac delta function, where δij = 1 when epitope i of variant j is recognized, and δij = 0 when it is not recognized.
The strength of the immune response against a particular epitope depends in part on how accessible it is to antibodies/killer T-cells. While escape mutations appear shortly after seroconversion in HIV-1/SIV infection at some epitopes [10, 39], it may take years to see evidence of selection at others [40]. We initialized the maximum fitness cost of each epitope in the beginning of the simulation by drawing it from a uniform distribution, ci*∼U [0, cmax]. We assumed that neutralization via antibodies is the primary driver of selection in chronic infection, and that virus is saturated with antibodies. In this scenario, the fitness loss of a virus variant upon antibody recognition is driven by the most potent antibody that binds it, c j i m m = m a x ( c i j δ i j ). The distribution of fitness costs associated with the epitopes defines the fitness landscape, which influences the evolutionary trajectory of the virus population.
Purifying selection conserves sites in the HIV-1 genome where mutations would be deleterious to the virus. We assume that mutations at invariant sites (where no variability is observed across sequences from different time points) are inherently deleterious, and incur a multiplicative fitness cost, c j i n v = 1 - ( 1 - ψ ) k, where ψ is the reduction in fitness per mutation, and k is the number of mutations. At the beginning of each simulation, we randomly distribute Lpinv invariant sites across the sequence, where pinv is the proportion of invariant sites. Note that the positions of the invariant sites vary between simulations, and may occur in both non-epitope and epitope regions. Following Ganusov et al. [41], we define the relative fitness of variant j as f j = ( 1 - c j i m m ) ( 1 - c j i n v ) = ( 1 - m a x ( c i j δ i j ) ) ( 1 - ψ ) k.
We considered a pool of NC = 15000 infected cells, which is in line with the mean effective population size of HIV-1 in chronic infection [18]. HIV-1 has a large burst size of approximately 50,000 [42], but only a small fraction of one in 1000 to 10,000 virions appear to be infectious [43–45]. Each productively infected cell therefore produces between 5 and 50 infectious virions. For consistency with the sequence evolution models developed by Vijay et al. and Balagam et al., we set the number of progeny virions produced by each productively infected cell to P = 5 [17, 18]. Following again Balagam et al., we set the generation time to 1.2 days [18], and the substitution rate to the mean of the mutation rates estimated by Mansky et al., μ = 3.5 × 10 − 5 [46]. We set the proportion of invariant sites to 50%, corresponding to the mean proportion of invariant sites estimated from the HIV-1 sequence alignments of patients described in Bunnik et al. and Karlsson et al. [9, 22]. We ran the simulations for 3000 generations, corresponding to 10 years.
To maintain the ratio of productively infected to latent cells predicted by the homeostatic proliferation model of latency introduced by Kim et al. and Rong et al., we set the size of the latent reservoir to NL = 100 [12, 13]. Assuming a total body load of 2 × 108 productively infected cells [47] and between 2.2 × 105 to 1.6 × 107 latent cells with replication competent DNA [48], there are between 0.0011 to 0.08 latent cells for every productively infected cell. In our simulations with 15,000 productively infected cells, this corresponds to a latent reservoir size NL of 16 to 1200 cells; our chosen latent reservoir size is close to the geometric mean of this range.
We chose a conservative estimate of 44 months for the half life of the latent reservoir [33, 36], which we used to calibrate the probability η of an infected cell becoming latent upon infection such that the size of the latent reservoir was maintained. Archin et al. recently estimated that the composite parameter βη is on the order of 10−14 [15], where β is the mass-action infection rate constant in ml−1 day−1. The latter has been estimated to be approximately β = 1.5 × 10−8 [49, 50]. Our value of η = 3.5 × 10−6 is therefore in line with these estimates. We set the death rate of latent cells to dL = 0.004 per generation, corresponding to the estimated death rate of memory cells, which make up the bulk of the latent reservoir [13, 14]. Following Rong et al., we varied the activation rate from aL ∈ (0.001,0.01) per generation so that between 0.1 and 1 latent cells were activated each generation [13, 14].
Barr et al. estimated that the minimum efficacy of three early neutralizing antibodies at blocking de novo infections ranged from 19.6% to 35.2% [10]. To account for the observation of Richman et al. [7] that some antibodies isolated from patient serum show no neutralization activity, we allowed the maximum fitness cost, c i *, due to recognition of a viral epitope to range from 0 to 0.40, with a mean of 0.2.
The remaining parameters associated with the immune response are not well known. For our default parameter setting, we assumed that there are ten epitopes in the approximately 700 nucleotide region of envelope that we simulate. In subsequent sensitivity analyses, we varied the number of epitopes between 0 and 15. While it is not known how many epitopes there are on the envelope, analysis of escape mutations to SIV in rhesus monkeys suggests that there are several [51]. We assumed that an antibody is created against an epitope variant once it reaches 1% frequency in the plasma. In sensitivity analyses, we varied the antigen frequency required for stimulating a new antibody response between 0.1% and 10%. Because patient serum does not neutralize contemporaneous virus but rather virus that circulated in the plasma at least 3 months earlier [7], we assumed that it takes 90 generations for a newly introduced immune response to reach its full potency (i.e. impose maximal fitness cost to the virus variant it recognizes, with fitness cost increasing linearly from zero over the 90 generations). The default values of the model parameters are summarized in Table 1.
To investigate the robustness of our results, we also considered alternative parameter values proposed in the literature. Recent estimates of the basic reproductive number of HIV-1 during primary infection suggest that an infectious cell generates at least eight new infected cells at the start of infection when target cells are not limiting. Thus, each infected cell produces at least P = 8 infectious progeny virions [52]. Following Pearson et al., we set the number of progeny virions to P = 10 to account for viral clearance [53]. We set the generation time of HIV-1 in vivo to two days, estimated from the decay dynamics of productively infected cells [54], and the substitution rate to the reverse transcriptase nucleotide substitution rate of approximately 2.2 × 10−5 [46, 55]. We further assumed that 10% of sites in the envelope region of interest are invariant, corresponding to the minimum proportion of invariant sites in the patient data. Because the mechanism of recombination is central to our investigation, we also varied the number of recombination events per generation between 0.1NR, and 10NR.
Estimating sequence divergence and diversity allows us to quantify viral evolution. Divergence is a measure of how far viral genomes have evolved from the founder strain, whereas diversity is a measure of the genomic variation in the viral population at any given time. Every 30 generations, we calculated the divergence and diversity of the HIV-1 sequences in the entire populations of productively infected and latent cells, recording the mean, median, and 5% and 95% quantiles. Furthermore, we randomly sampled 100 productively infected cells and 100 cells from the latent reservoir every 300 generations, storing the HIV-1 sequences for later phylogenetic analysis. Using d(i, j) to denote the mean number of differences per position between sequences i and j, divergence and diversity are defined as follows for a collection of k sequences:
d i v e r g e n c e = 1 k ∑ i = 1 k d i , f o u n d e r d i v e r s i t y = ∑ i = 1 k - 1 ∑ j = 1 + 1 k d ( i , j ) 1 2 k ( k - 1 ) .
Stochastic effects dominate at small population sizes—the smaller the population, the larger the probability that stochastic fluctuations lead to extinction. We therefore investigated how the size of the simulated population affects the ability of latent genomic fragments to survive in patient plasma, assuming that recombination is necessary for survival. To this end we determined the expected number of dual infections, a precursor for recombination, involving at least one latent form when the sizes of the replicating and latent populations were increased ten-fold. We ran 100 simulations, tracking the progeny of the activated latent cells for ten generations, introduced into a replicating population of size nr at generation zero. The relative fitness of each latent virus was set to 0.5, while the relative fitness of each non-latent virus was drawn from a uniform distribution between 0.5 and 1. As in our full model, each generation of the simulation consisted of every infected cell producing five progeny virions, and downsampling the virus population to infect the next generation of nr uninfected cells. The expected number of dual infections involving at least one latent virus was then estimated for each simulation run.
Simulation results were compared to HIV-1 DNA sequence data from 16 untreated or unsuccessfully treated patients [9, 21, 22]. All patients were infected with HIV-1 subtype B. The patients were followed longitudinally from seroconversion with 2–22 sequences sampled at intervals of 1–67 months (mean 13 months). The sequences were derived from the envelope, and were between 532–948 nucleotides long (mean 683 nucleotides). For each patient, we calculated the mean divergence and diversity at each time point the same way as for the simulated data. Note that in [21], sequence divergence and diversity were estimated from mean pair-wise distances determined using either a two-parameter Kimura model or a general time-reversible model with site-to-site variation in substitution rates, while here only simple pairwise Hamming distances were used. We considered the unsuccessfully treated patients as effectively untreated, because their viremia was not suppressed, and their divergence and diversity trends were similar to those of the untreated patients. We also estimated the proportion of invariant sites across all time points for each patient using PhyML [56]. The mean proportion of invariant sites across all patients was 39%. The lowest proportion of invariant sites was 13% for Patient 9 in the Shankarappa cohort, while the mean was greater than 50% for both the Amsterdam and Karlsson patients.
To be sure that our model invokes realistic trends of HIV-1 genetic evolution, we compared our simulated patient HIV-1 populations generated by simulations incorporating recombination to HIV-1 DNA sequence data from 16 untreated or unsuccessfully treated patients [9, 21, 22]. Reassuringly, the empirical HIV-1 diversity and divergence trends over time were qualitatively and quantitatively similar to our simulated patient HIV-1 populations (Fig 2). Divergence increased linearly over time to a mean of 0.06 substitutions per site at 10 years post-primary HIV-1 infection (post-PHI), while diversity initially increased rapidly, and then saturated to approximately 0.05 substitutions per site at 10 years post-PHI. As seen in S1 Table in the supplement, there was no significant trend between the activation rate and mean sequence diversity (p = 0.67, cor = 0.15; Pearson’s product-moment correlation). Increasing the activation rate slightly decreased sequence divergence at the end of the simulations (p = 0.04, cor = −0.65; Pearson’s product-moment correlation). Unsurprisingly, both mean sequence divergence and diversity increased in the latent reservoir at a much slower initial rate than in productively infected cells (Fig 3). After approximately 6 years post-PHI, divergence in the latent reservoir proceeded at nearly the same rate as in productively infected cells. Because diversity in the latent reservoir grew linearly but started to saturate in plasma, by 10 years post-PHI diversity in the latent reservoir almost reached that in productively infected cells.
Reconstructed phylogenies of the simulated patient HIV-1 populations, with 20 sequences sampled every 12 months, showed similar topological structures as patient HIV-1 populations. While the serially sampled HIV-1 phylogenies ranged from star- to ladder-like structures, they generally displayed a clear time trend (Fig 4). Both in the simulated and clinical data, the fraction of surviving phylogenetic lineages between samplings ranged from 0.1 to 1.0 (S1 Fig). Low survival of lineages corresponds to a ladder-like phylogeny while high survival corresponds to a star-like phylogeny.
The tree shapes were largely explained by the fitness landscape, where convex profiles lead to more ladder-like trees, and concave profiles lead to bushier, star-like, trees. Intermediate fitness landscapes (nearly linear profiles) generated trees more typically observed in HIV-1 infected patients. Importantly, these intermediate landscapes also showed typical patient diversity trends (Fig 4). Interestingly, in intermediate and concave profiles we sometimes observed dramatic selective sweeps where the diversity dropped drastically and then recovered. The fitness landscape also influenced the number of latent origins (latent virus of different ages reaching at least 1% frequency in plasma virus population). The mean (s.d.) number of different latent origins was 0.16 (0.42) for convex landscapes, 0.42 (0.76) for nearly linear landscapes, and 7.6 (7.5) for concave ones. Thus, our simulations suggest that the fitness landscape influences general tree shape, diversity trends, and the survival of latent forms.
The shape of the landscape is defined by the maximum fitness costs exerted by immune responses against different viral epitopes. The concave landscapes have a higher number of different immune responses exerting similar high fitness costs on the virus than convex or nearly linear landscapes. Escaping the immune response exerting the maximum fitness cost only marginally increases fitness if there are simultaneous immune responses against other epitopes that also impose high fitness costs. Therefore, a more concave fitness landscape exerts a stronger selection pressure on the virus, reducing both the mean and variance of the relative fitness in the virus population. The survival of less adapted latent forms in plasma therefore increases, while the reduction in fitness differences between virus variants in productively infected cells increases diversity.
Fig 5 shows that the more genomic fragments with different latent origins persist in the plasma, the higher the sequence diversity (panel A). Compared to simulations with no surviving latent origins, sequence diversity became significantly higher after 9.5 years post-PHI in simulations where there were between 1 and 5 latent origins (p < 0.05, Wilcoxon rank-sum test), while in simulations where there were more than 5 latent origins, sequence diversity became significantly higher much earlier, after approximately 5 years post-PHI (p < 0.05, Wilcoxon rank-sum test). At 10 years post-PHI, mean sequence diversity was 0.038 substitutions/site when no latent genomic fragments survived, increasing by 16%, when there were between 1 and 5 latent origins, and by 95% when there were more than 15 latent origins. The propagation of genomic fragments that have accumulated fewer mutations than contemporaneous variants should reduce average sequence divergence in the plasma. As expected, when the number of surviving genomic fragments of different latent origins increased, mean sequence divergence decreased (Fig 5, panel B).
Our model predicts that multiple introductions of latent forms into plasma are necessary for increased diversity; simulations with only one surviving latent origin did not have significantly higher diversity than those with no latent forms in plasma. Because sequence diversity increased almost linearly for concave fitness landscapes, the introduction of older virus forms into plasma did not further increase diversity (S2 Fig). Furthermore, because the survival of multiple latent origins was rare for convex fitness landscapes, it was not possible to determine whether latency could have had an effect on diversity under such conditions. However, the phylogenies and diversity trends generated by convex or concave fitness landscapes did not resemble those observed in typical patients.
Fig 6 shows the effect of recombination on the proportion of virus with latent genomic fragments in productively infected cells in 2000 simulations of untreated patients. Our simulation results suggest that recombination facilitates the survival of latent genomic fragments. In simulations with recombination, 27% of the HIV-1 replicating plasma populations reached a mean of ≥ 10% virus with latent genomic fragments between 5–10 years post-PHI, while only 2.0% reached ≥ 10% latent forms without recombination. Thus, recombination increased the survival of latent genomic fragments in productively infected cells approximately 13-fold.
The proportion of virus with latent genomic fragments was higher when we allowed for recombination than when we ignored it. This distinction is clear from even the first sample time, after 30 generations (p < 0.05, Wilcoxon rank-sum test). Furthermore, the rate of increase of virus with latent genomic fragments over time in the replicating population was approximately 9 times faster when recombination was involved Fig 6. In most simulations without recombination, virus with latent fragments immediately disappeared from productively infected cells upon introduction, while in some simulations, sharp peaks were observed where the proportion of latent forms increased rapidly to a high fraction but then fell quickly and disappeared completely. In simulations with recombination, the proportion of virus with latent genomic fragments often increased gradually, and stabilized both at intermediate values and the extremes.
Most recombinant HIV-1 that include a fragment from a latent virus underwent relatively few recombination events; 96% had fewer than 5 latent fragments (Fig 7A). However, only 0.7% of sequences with latent sites at 10 years post-PHI had not acquired genomic fragments from non-latent contemporaneous virus through recombination. The most common latent recombinant observed in our simulations had one latent fragment, with a mean (s.d.) of 85 (123) sites (Fig 7A). When such a recombinant proliferates further, additional recombination with contemporaneous virus removes sites from the latent fragment. Overall, the mean (s.d.) number of latent sites at 10 years post-PHI was 99 (121), while the mean (s.d.) fragment length was 66 sites (Fig 7B).
When the number of latent fragments in a viral sequence increased, the fragment length decreased (Fig 7B). Interestingly, viruses that had more latent fragments also had more latent origins (Fig 7C). Hence, the most common forms of latent recombinants had few breakpoints, and multiple breakpoints frequently involved latent virus with different origins in time. Increasing the recombination rate increased the survival of virus with latent genomic fragments (S2 Table). As expected, the number of latent fragments per sequence also increased, while the fragment length decreased.
Higher proportions of surviving lineages through time imply higher numbers of virus with genomic fragments of different latent origins that have long branches reaching far back into the phylogeny (Fig 8). When the prevalence of latent genomic fragments in productively infected cells was high, they were typically of different latency ages, i.e., deposited and activated at different times (S3 Fig). These observations are supported by the noted increase in both the proportion of virus with latent genomic fragments and the observed number of different latent origins in productively infected cells when the activation rate was increased (S4 Fig).
When a person becomes infected with HIV-1 the initial spike in viral load may deposit many similar copies of the initial virus in the reservoir. We modeled this by initially filling the reservoir with the infecting strain. Thereafter, a relatively slow deposit rate and proliferation rate refill and maintain a diversifying latent virus population. Interestingly, at 10 years post-PHI, our simulations predict that about 27% of the reservoir still consists of virus deposited in the first year of infection (Fig 9A). Conversely, the latent genomic fragments in the plasma population originated mostly from more recently deposited and activated virus (Fig 9B). Overall, the age structure among latent genomic HIV-1 fragments in the plasma of an untreated patient followed that of the latent reservoir with the exception of the initially deposited virus, which is strongly selected against upon activation by the refined immune response.
The evolutionary trends of the virus populations simulated under the alternative model parameters (P = 10, μ = 2.2 × 10−5, inv = 0.1, τ = 2 days) were qualitatively and quantitatively similar to those seen previously under the default parameters. The mean sequence divergence and diversity of 100 simulations with the alternative model parameters are shown in (S5 Fig). We found that the prevalence of virus with latent genomic fragments in productively infected cells was much higher in simulations with recombination than without (S6 Fig). Our simulation results are therefore robust to the choice of parameters defining the infection and replication dynamics.
The sensitivity of the model to the immune system parameters, i.e., the antigen frequency necessary to elicit a new immune response, and the number of epitopes accessible to the immune system, are described in (S3 and S4 Tables), respectively. Fifty simulation runs were performed for each parameter value. When the antigen frequency required for stimulating a new immune response was increased from 0.1% to 10%, the survival of virus with latent fragments and sequence diversity decreased (p = 0.025, cor = -0.67; p = 0.013, cor = -0.71, respectively; Pearson’s product-moment correlation), while sequence divergence increased (p = 0.003, cor = 0.79, Pearson’s product-moment correlation).
When the immune response to a new escape mutant is delayed, the frequency of virus variants carrying the advantageous escape mutation increases. The increase in the number of virus variants that have evolved away from the founder sequence increases divergence but the simultaneous convergence towards the new escape variant reduces diversity. Because the mean relative fitness of virus in productively infected cells increases, the survival probability of virus from activated latent cells decreases. The survival of virus with latent genomic fragments was low in all simulations without recombination.
On the other hand, when the number of epitopes was increased, sequence diversity at 10 years post-PHI initially decreased from one epitope to 4 epitopes and then increased from 4 to 15 epitopes (S4 Table). Sequence divergence initially increased from one to 4 epitopes, and then slightly decreased as the number of epitopes was increased. The survival of virus with latent genomic fragments was more stochastic, but generally displayed a similar trend. Virus with latent genomic fragments rarely survived in simulations without recombination. Therefore, our fundamental model result suggesting that recombination is an essential mechanism facilitating the survival of latent genomic fragments is robust to large perturbations in the largely unknown parameters describing the immune response.
When the number of epitopes was zero, no immune pressure was exerted on the virus. This resulted in high diversity (0.081 subst/site) because all virus variants without mutations at invariant sites had equal probability of surviving, and low divergence (0.047 subst/site), because rapid accumulation of escape mutations was not selected for. Virus with latent genomic fragments survived in all simulations, because earlier virus forms were not at a fitness disadvantage compared to more recent variants.
When the number of epitopes was increased from 1 to 4, it became increasingly more difficult but not impossible for virus to escape the immune response. When two epitopes had high fitness costs associated with them, it took longer for any virus variant in the population to simultaneously escape both responses, and the significant fitness gain and proliferation due to escape resulted in decreased diversity (S7 Fig). Such genetic sweeps were more dramatic when three epitopes had high fitness costs. When four epitopes had high fitness costs, escape at all epitopes to gain fitness was nearly impossible, and happened in only a small fraction of the simulations. Therefore, more concave fitness landscapes with four or more epitopes with high fitness costs imposed selection pressure resulting in high diversity (S7 Fig). The probability that a randomly drawn fitness landscapes had 2 or 3 epitopes with high fitness costs was highest when the total number of epitopes was 4 or 5. When the number of epitopes was increased from 6 to 15, the probability that each fitness landscape had 4 or more epitopes of high fitness costs also increased.
Surviving in the replicating population long enough to recombine with a replicating virus forms the bottleneck that largely determines the long-term fate of a latent lineage. When one latent form was introduced into a population of 15,000 cells at generation t = t0, corresponding to the expected number of cells activated each generation from the latent reservoir of 100 cells at the highest activation rate α = 0.01, the expected number of dual infections involving at least one latent virus was 0.1 over 10 generations. However, when 10 latent forms were introduced to a population of 150,000 cells, i.e., at the same proportion but in a ten-fold population, the total expected number of dual infections was 0.94. Thus, when the simulated population size was increased ten-fold, the probability that at least one latent form ended up in a dually infected cell before disappearing from the replicating population increased nearly ten-fold. The long-term behavior in simulations where virus with latent genomic fragments proliferates is expected to be robust to stochastic effects due to the increased number of such viruses, however, the variance observed in the survival of latent virus forms between simulations is overestimated in a smaller population due to early stochastic extinction. The proportion of simulations with persistence of virus with latent genomic fragments is therefore likely an underestimate.
We developed a within-host HIV-1 evolution model that includes point mutation, recombination, immune (positive) selection, negative selection, and latency. This model quantitatively captures previously observed viral sequence divergence and diversity trends and mechanistically explains these patterns, allowing us to model realistic within-host natural evolution of HIV-1. We used this model to investigate when and how latent forms of HIV-1 can survive in productively infected cells despite not being well adapted to the immune response, therefore having reduced relative fitness. Our simulation results suggest that recombination is a key mechanism facilitating the survival of virus forms with latent genomic fragments. We have previously shown that the majority of phylogenetic lineages in HIV-1 populations of untreated patients display a statistically defined signal of latency [1]. The results in this paper suggest that such lineages likely survived because of recombination. Our model further predicts that the survival of latent genomic fragments in plasma depends on the fitness landscape induced by the immune response.
By comparing simulations where the surviving genomic fragments originated from different numbers of latent ancestors, we found that latency reduced mean sequence divergence, but increased mean sequence diversity for reasonable fitness landscapes. This effect may explain how HIV-1 can keep a high adaptation potential without mutating too far away from the infecting strain. High adaptation potential is useful for escaping immune and antiviral drug pressures, while less divergent sequences closer to the infecting form are arguably more fit to infect new hosts [2].
The dynamics of our simulated immune response depend on the number of epitopes and the antigen frequency triggering a new immune response, neither of which are well known. To test the sensitivity of our model, we varied these parameters to the extremes where measures of sequence divergence and diversity became unrealistic. Importantly, the trend of much higher survival of virus with latent genomic fragments in simulations with recombination than in simulations without recombination was robust for the full range of immune system parameters. This suggests that our main result is not sensitive to the exact details of how the immune response was implemented in our model.
While in vivo HIV-1 viral populations are typically very large, computational considerations limited us to follow instead the relatively small population of 15,000 productively infected cells, which is in line with recent estimates of the effective population size in real HIV-1 populations [18]. The variability in the survival of virus with latent genomic fragments observed in our simulations is likely a result of stochastic fluctuations when latent virus first enters the replicating population. The proportion of simulations in which latent virus forms survive is therefore likely to be an underestimate. However, we expect the mean dynamics of virus with latent genomic fragments to be robust to scaling issues after the initial bottleneck.
Previous studies have suggested that escape mutations have a fitness cost [10], and thus virus closer to the infecting strain may have higher replicative fitness due to having accumulated fewer escape mutations to evade immune surveillance. One of the limitations of our model is that we do not consider fitness costs associated with immune escape. However, compensatory mutations may arise in other parts of the genome, and at least partially restore fitness [11]. Furthermore, mutations can hitchhike and eventually affect population fitness. We do not attempt to incorporate epistatic effects here but rather investigate the theoretical mechanisms that may allow the persistence of latent genomic fragments in the plasma population despite virus activated from latent cells having a fitness disadvantage. Assuming that latent virus does not have a replicative advantage may lead us to underestimate the survival of latent genomic fragments in the plasma virus population.
Furthermore, if activated virus from latent cells has similar fitness as contemporaneous variants, recombination may not be as important for survival. However, recombination does not merely facilitate survival, but rather allows latent genomic fragments to propagate through the plasma virus population. In the most extreme case where no selection pressure was exerted by the immune response (0 epitopes, S4 Table), approximately 60% of viral sequences contained latent genomic fragments during the last five years of infection in simulations with recombination, while only 6% of sequences were latent in simulations without recombination. While continual recombination with virus circulating in the plasma reduces the size of the genomic fragment inherited from a latent ancestor, even a few latent sites of high replicative fitness may allow the virus to better adapt to different evolutionary pressures. Latency may therefore expand the adaptive potential of HIV-1 besides enabling the virus to hide from immune surveillance and antiretroviral treatment.
Because the patient sequences we chose to study were in envelope, we focused on antibody responses and escape from them. While we did not explicitly account for CTL responses, the epitopes in our model are approximately the same length as CTL epitopes, so our model could be interpreted as one in which CTL responses instead of or in addition to antibodies are providing immune pressure. However, CTL responses exert selection pressure on the whole proteome. Since we only simulated sequence evolution in env and ignored the interplay between different genes, our model predictions may not be generalizable to the whole virus. Our future directions include investigating further the balance between replicative fitness and immune escape on the level of the whole virus genome, and adapting our model to simulate HIV-1 evolution and latency under different treatment scenarios.
|
10.1371/journal.pntd.0004753 | Antibody Secreting Cell Responses following Vaccination with Bivalent Oral Cholera Vaccine among Haitian Adults | The bivalent whole-cell (BivWC) oral cholera vaccine (Shanchol) is effective in preventing cholera. However, evaluations of immune responses following vaccination with BivWC have been limited. To determine whether BivWC induces significant mucosal immune responses, we measured V. cholerae O1 antigen-specific antibody secreting cell (ASC) responses following vaccination.
We enrolled 24 Haitian adults in this study, and administered doses of oral BivWC vaccine 14 days apart (day 0 and day 14). We drew blood at baseline, and 7 days following each vaccine dose (day 7 and 21). Peripheral blood mononuclear cells (PBMCs) were isolated, and ASCs were enumerated using an ELISPOT assay. Significant increases in Ogawa (6.9 cells per million PBMCs) and Inaba (9.5 cells per million PBMCs) OSP-specific IgA ASCs were detected 7 days following the first dose (P < 0.001), but not the second dose. The magnitude of V. cholerae-specific ASC responses did not appear to be associated with recent exposure to cholera. ASC responses measured against the whole lipolysaccharide (LPS) antigen and the OSP moiety of LPS were equivalent, suggesting that all or nearly all of the LPS response targets the OSP moiety.
Immunization with the BivWC oral cholera vaccine induced ASC responses among a cohort of healthy adults in Haiti after a single dose. The second dose of vaccine resulted in minimal ASC responses over baseline, suggesting that the current dosing schedule may not be optimal for boosting mucosal immune responses to V. cholerae antigens for adults in a cholera-endemic area.
| The bivalent whole-cell (BivWC) oral cholera vaccine (Shanchol) is effective in preventing cholera. Despite its increasing use as part of comprehensive cholera prevention and control efforts, evaluations of immune responses following vaccination with BivWC have been limited. In this study, we measured the development of cholera-specific antibody secreting cells, markers of mucosal immunity, following vaccination with BivWC among a population of adults in Haiti, where cholera is now endemic. BivWC induced development of robust immune responses following the first dose of vaccine, but similar ASC responses were not detected following the second dose, suggesting that the currently recommended 14-day interval between doses may not be optimal for boosting mucosal immune responses among adults in cholera endemic regions. These findings suggest that additional evaluation of the optimal dosing schedule for oral cholera vaccines is warranted with the goal of improving long-term immunity.
| Cholera, a diarrheal disease with epidemic potential caused by the bacterium Vibrio cholerae, remains a significant cause of morbidity and mortality, accounting for approximately 1.4 to 4.3 million cases of acute diarrhea and 28,000 to 142,000 deaths annually.[1] V. cholerae serogroup O1 is the current predominant cause of cholera, while serogroup O139 was a major cause of cholera in the 1990s and early 2000s.[2] The O1 serogroup is further classified into Ogawa and Inaba serotypes.[3] These differ by the presence of a 2-O-methyl group present in the terminal carbohydrate of the O-specific polysaccharide (OSP) in the lipopolysaccharide (LPS) of the Ogawa serotype.[4–6] Protection against cholera is serogroup specific, and serogroup specificity is defined by the OSP moiety of LPS.[7–9]
Recent years have seen the frequency and severity of cholera outbreaks increase worldwide.[10] For example, the cholera epidemic that began in Haiti in October 2010 has accounted for 750,752 reported cases of acute diarrhea and 9,031 deaths,[11] making it the largest outbreak of cholera in recent history. While the development of robust water and sanitation infrastructure is essential in regions affected by cholera, vaccination offers another important and complementary tool for cholera prevention and control.[12]
There are currently three World Health Organization (WHO) pre-qualified, commercially available oral cholera vaccines (OCV): Dukoral (Crucell, Sweden), which consists of whole-cell killed V. cholerae O1 Inaba and Ogawa and recombinant cholera toxin B subunit (WC-rBS); and Shanchol (Shantha Biotechnics, India) and Euvichol (EuBiologics, Korea), which are similar bivalent whole-cell (BivWC) vaccines containing multiple biotypes of V. cholerae O1 and O139 which lack the cholera toxin B subunit (CtxB) found in Dukoral. Following Shanchol’s prequalification by the WHO in 2011, multiple programs in diverse settings including Haiti have demonstrated the feasibility and efficacy of integrating vaccination with BivWC into comprehensive cholera control strategies.[13,14] In 2013, the WHO created an OCV stockpile to respond to cholera outbreaks worldwide.[15]
Despite the rapidly growing use of oral cholera vaccination with BivWC, there remain significant gaps in our understanding of immune responses following vaccination with BivWC. Since vaccine immunogenicity is a surrogate for protection, such studies are especially important in settings where large scale evaluations of direct effectiveness are not practical, such as in comparing vaccine formulations, dosing schedules, and in estimating vaccine efficacy across specific populations. Trials to assess safety and immunogenicity have demonstrated development of significant vibriocidal antibody and OSP-specific plasma immunoglobulin A (IgA) responses among adults and children following vaccination[16–20], including comparable vibriocidal antibody responses following a 14-day versus a 28-day dosing interval in a cholera-endemic area of Kolkata, India.[18] Previously, our group demonstrated significant vibriocidal responses and OSP-specific IgA responses following vaccination with BivWC among adults, children, and HIV-infected adults in Haiti.[21,22] However, compared to studies characterizing immune responses following natural cholera infection[23–28] and vaccination with WC-rBS[29–33], studies evaluating immune responses following vaccination with BivWC are limited.
Among evaluations of immunogenicity, the detection of antibody secreting cells (ASCs) in blood is a standard measure of the mucosal immune response. Following stimulation in the gastrointestinal tract by live V. cholerae or by OCVs, these cells transiently migrate into the systemic circulation where they can be detected by standard Enzyme-Linked ImmunoSpot (ELISPOT) assays.[34] This migration peaks approximately at day 7 following primary infection or vaccination. Many of these circulating ASCs express gut-homing markers and return to mucosal tissue.[35] Detection of LPS-specific ASCs in the peripheral blood correlates with the development of V. cholerae antigen-specific antibodies in the small intestinal lamina propria up to 6 months following infection.[36] As such, ASCs provide a critical and early window into subsequent immunologic memory at the mucosal surface.
To determine whether vaccination with BivWC induces significant mucosal immune responses, we measured levels of ASC responses against V. cholerae LPS and OSP at multiple time points following vaccination with BivWC among healthy participants in Haiti.
Healthy adults, ages 18–60 years old attending the Saint Nicolas Hospital outpatient clinic in Saint Marc, Haiti were invited to participate in May 2015, and were deemed eligible if they lived in the catchment area. Individuals with an active gastrointestinal disorder, pregnancy (as determined by urine pregnancy test), those that had previously received OCV, and those unable to give informed consent were excluded from the study. The BivWC vaccine was stored at 2°C–8°C prior to administration. Two doses of vaccine were given 14 days apart, in accordance with the package insert. Venous blood from each participant was obtained prior to immunization and 7 days after each dose of vaccine (on days 0, 7, and 21). A stool specimen was collected on day 7 to test for enteric pathogens, and a questionnaire on social and nutritional factors was conducted on day 14. The participants enrolled in this study were part of a larger ongoing study to measure long-term immune responses among adult cholera vaccine recipients.
Written informed consent was obtained from all potential adult participants prior to enrolling in the study. The study protocol was reviewed and approved by the institutional review board of Partners HealthCare in Boston, Massachusetts, and the Haitian National Bioethics Committee in Port-au-Prince, Haiti.
We collected venous blood from each participant in Vacutainer blood collection tubes (Becton Dickinson), containing sodium heparin anticoagulant. Samples were obtained at day 0 (before vaccination), day 7 (7 days after the first dose of vaccine), and day 21 (1 week after the second dose of vaccine). Following sample collection, blood was diluted in PBS (pH 7.4) and processed for separation of plasma and peripheral blood mononuclear cells (PBMCs). PBMCs were prepared by differential centrifugation in Leucosep tubes (Greiner Bio-One Ltd) filled with Ficoll-Isopaque (Pharmacia, Piscataway, NJ), and after two subsequent PBS washes, resuspended at a concentration of 5 x 106 cells/mL in RPMI complete medium (Gibco, Carlsbad, CA) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (HyClone, Logan, UT) and 1% Penicillin-Streptamycin (Sigma-Aldrich). These cells were used immediately as described below for measuring ASC responses. Plasma was shipped in a liquid nitrogen dry vapor unit and stored at -80°C prior to analysis in Boston.
Vibriocidal antibody responses in plasma were measured using a standard protocol described previously.[24] Briefly, V. cholerae O1 El Tor Ogawa (X25049) and Inaba (T19479) were used as the target organism. The vibriocidal titer was defined as the reciprocal of the highest plasma dilutions resulting in a greater than 50% optical density reduction associated with V. cholerae O1 growth when compared to control wells without plasma. An increase of titer by 4-fold was considered to represent seroconversion, by convention. Control plasma, a pool of day 7 plasma from culture-confirmed V. cholerae infected individuals in Dhaka, Bangladesh, was used to monitor intra-assay variability between plates.
OSP was purified and conjugated to bovine serum albumin (BSA) as previously described;[37] source strains of OSP Inaba and Ogawa were V. cholerae O1 El Tor PIC018 and PIC058, respectively.[38] Anti-OSP IgA responses in plasma were measured using a previously described ELISA protocol.[24] In brief, 96-well polystyrene plates (Nunc, low affinity plates) were coated with V. cholerae O1 Ogawa OSP conjugated to BSA, or Inaba OSP conjugated to BSA (1 μg/mL), dissolved in 50mM carbonate buffer (pH 9.6). We added 100 μL of plasma diluted 1:25 in 0.1% BSA in PBS-0.05% Tween. Plates were incubated at 37°C for 60 minutes while shaking (100 rpm) and then washed. We detected OSP-specific antibodies using 100 μL per well of horseradish peroxidase-conjugated goat anti-human IgA at a dilution of 1:1,000 (Jackson Immunoresearch, West Grove, PA) in 0.1% BSA in PBS-0.05% Tween. The plates were incubated at 37°C for 1 hour while shaking (100rpm), washed three times with PBS-0.05% Tween, and developed with ABTS/H2O2 (Sigma-Aldrich, St. Louis, MO) in 0.1M citrate-phosphate buffer (pH 4.5) and 0.03% hydrogen peroxide. Measurements were made using a kinetic reading at 405 nm wavelength using the SoftMax Pro software.
ASC responses were measured by enzyme-linked immunosorbent spot (ELISPOT) as described previously.[34] To measure the total number of circulating ASCs, nitrocellulose-bottom plates (Millipore, Bedford, MA) were coated with 100μl of affinity-purified goat anti-human IgG F(ab)2 (Jackson Immunology Research, West Grove, PA) at a concentration of 5 μg/mL in PBS (pH 7.4). Nitrocellulose plates were also coated with 100μl of the following antigens, to detect V. cholerae-specific antigen reponses: V. cholerae O1 Ogawa and Inaba OSP:BSA (10 μg/mL), Ogawa and Inaba-LPS (25 μg/mL), CtxB (2.5 μg/mL;Sigma-Aldrich), and keyhole limpet hemocyanin (KLH, Pierce Biotechnology, Rockford, IL, 2.5 μg/mL). While KLH is a standard control antigen, CtxB was also used to control for ASC responses from possible exposure to natural V. cholerae O1 infection. The BivWC vaccine lacks recombinant CtxB, therefore should not elicit a detectable CtxB-specific response.
Plates were incubated with antigen overnight at 4°C, and then blocked for 1 hour at 37°C with 200 μL of 2% milk-RPMI. A total of 1 x 105 PBMCs were added to the total Ig-coated wells and serially diluted two times by a factor of 10. A total of 5 x 105 PBMCs were added to the OSP, LPS, CtxB and KLH-coated wells. Plates were then incubated at 37°C for 3 hours and then washed. To detect IgG and IgA ASCs, plates were incubated overnight at 4°C with horseradish peroxidase-conjugated mouse anti-human IgA (Hybridoma Reagent Laboratory, Baltimore, MD) and alkaline phosphatase-conjugated IgG (Southern Biotech, Birmingham, AL), each diluted 1:500 in sterile filtered PBS-0.1%-BSA-tween-0.05%. Following overnight incubation at 4°C, plates were developed with 5-bromo-4-chloro-3-indolylphosphate-nitroblue tetrazolium (BCIP/NBT, Sigma-Aldrich) and 3-amino-9-ethylcarbazole (AEC pre-mix solution, Sigma-Aldrich). Two individuals using a stereomicroscope independently quantified ASC numbers. The number of antigen-specific IgG and IgA ASCs were expressed per 106 total PBMCs and as a percentage of total IgG or IgA expressing cells.
Statistical analyses were performed using STATA Version 14 (StataCorp, LP, College Station, TX) and Graph-Pad Prism (Graph Pad Software, Inc., La Jolla, CA). Vibriocidal and antigen-specific plasma antibody responses were expressed as geometric mean titers (GMT) with 95% confidence intervals. We used the Wilcoxon signed-rank test to compare within-person antigen-specific ASC responses, vibriocidal antibody responses, and antigen-specific plasma antibody responses across different time points.
We used Spearman’s correlation to quantify associations between antigen-specific ASC, vibriocidal and antigen-specific antibody responses. Comparisons between groups previously exposed and unexposed to V. cholerae were performed using the Mann Whitney-U test. We tested for equivalence between anti-LPS and anti-OSP ASC responses using two-sided 90% confidence intervals and an equivalence bound of +/- 5 ASCs per 106 PBMCs difference. Confidence intervals within this bound were considered equivalent. All P-values were two-tailed, with a P ≤ 0.05 defined as the threshold for statistical significance.
Our sample size was based on the number of participants needed to detect a difference between the baseline and post-vaccination frequency of LPS-antigen specific ASCs. Extrapolating from other studies,[32,34,39] we estimated that 19 participants would be needed to detect an increase in both IgG and IgA ASC responses, with at least 90% power following vaccination.
Table 1 describes the demographic characteristics of the 24 individuals that enrolled in the study. One participant dropped out following the first blood draw on day 0, and another dropped out following the second blood draw on day 7. A third participant was excluded from the analysis of ASC responses because of excessive red blood cell contamination in PBMCs at all time points. Thus, 22 participants were included in the analysis of responses following first dose of vaccine; Twenty-one individuals were included in the analysis following the second dose of vaccine. Baseline vibriocidal titers among the study population were consistent with widespread exposure to cholera in a substantial portion of the cohort. 8/22 (36%) participants had a vibriocidal ≥ 80 for the Ogawa serotype, and the same proportion of participants (36%) had a vibriocidal ≥ 80 for the Inaba serotype. A total of 11/22 (50%) had a vibriocidal ≥ 80 for either the Ogawa or Inaba serotype.
A summary of immune responses to BivWC at each time point is listed in Table 2. Vibriocidal antibody responses are also depicted in Fig 1. Robust vibriocidal antibody responses to Ogawa and Inaba were detected by day 7 after a single dose of vaccine (p < 0.0001) with a geometric mean fold rise (GMF) of 7.3 to V. cholerae O1 Ogawa and 9.1 to Inaba. Vibriocidal titers at day 21 were slightly lower compared to day 7 responses, although these differences were not significant. However, vibriciocidal GMT at day 21 remained significantly increased compared to baseline values for Ogawa (p < 0.01) and Inaba (p < 0.001). Vibriocidal antibody seroconversion rates are depicted in Table 2. The majority of participants seroconverted following the first dose of BivWC; Ogawa (64%) and Inaba (73%). The seroconversion rate increased to 76% and 81% for Ogawa and Inaba, respectively, following two doses of vaccine. In addition to vibriocidal antibody responses, we observed significant OSP-specific plasma IgA responses to both Ogawa and Inaba serotypes at days 7 and 21 (Table 2).
We did not observe an overall increase in the number of total IgA and IgG antibody secreting cells either in total or relative to the total number of peripheral blood mononuclear cells following vaccination with BivWC (Table 2). Among antigen-specific responses, there was no response to either of the control antigens, CtxB or KLH, at any time point. Some artefactual spots (typically <3) are typical of ELISPOT assays. We observed a single KLH spot in only three of the 130 ELIPSPOT samples analyzed (2.3%), and no more than a single KLH spot was observed in any of the samples. The low number of KLH spots suggests that there was negligible artefactual background in the other antigen-specific ASC measurements. Similarly, in four of the 130 samples (3.1%) analyzed as part of the study, we observed a single CtxB spot, and again, in no samples, were more than a single CtxB spot observed. As such, we defined the presence of ≥ 2 ASCs per 106 PBMCs as a positive ASC response. The lack of CtxB responses suggests that responses detected to the OSP or LPS antigens were not likely the result of intercurrent infection with or exposure to V. cholerae, since natural infection is associated with significant responses to the CtxB antigen.
OSP-specific ASC responses are shown in Fig 2. Notably, OSP-specific IgA ASC responses to Ogawa and Inaba peaked on day 7 following the first dose of vaccine, and were significantly increased from baseline (P < 0.001). Notably, day 7 OSP-specific IgA antibody responses detected in plasma were strongly correlated with day 7 OSP-specific IgA ASC responses to Ogawa (Spearman r = 0.6877, P = 0.0004) and Inaba (Spearman r = 0.6077, P = 0.0027)
An analysis of ASC response based on stratification by a pre-vaccination vibriocidal titer of ≥ 80, shows no significant difference in mean ASC responses. OSP-specific IgA responses at day 7 to the Ogawa serotype were 5.2 ASCs per million PBMCs and 7.9 ASCs per million PBMCs among those exposed and unexposed (P = 0.8894), respectively, whereas OSP-specific IgA responses to the Inaba serotype were 10.4 ASCs per million PBMCs and 9 ASCs per million PBMCs (P = 0.285), respectively. Additionally, no significant differences were seen when comparing OSP-specific IgA ASC responses at day 7 to the Ogawa and Inaba serotype between individuals with previous exposure to either Ogawa or Inaba (P = 0.6905 for Inaba OSP response, P = 0.7889 for Ogawa response). However, our sample size is limited and not sufficiently powered to definitively assess whether prior exposure in this setting impacts mucosal immune responses.
OSP-specific IgA ASC responses after the second dose of the vaccine decreased nearly to baseline levels by day 21. OSP-specific IgG antibody responses were proportionally lower than IgA responses, although OSP-specific IgG ASC responses nonetheless demonstrated a statistically significant increase from baseline for the Ogawa (p < 0.001), but not the Inaba serotype (p = 0.099) (Fig 3).
We noted the OSP and LPS responses were not only highly correlated, but appeared nearly equivalent. (Fig 4) Testing for equivalence of IgA ASC responses to LPS and OSP demonstrated a mean difference of 0.70 ASCs per million PBMCs (90% CI: -0.799–2.21) and 0.73 ASCs per 106 PBMCs (90% CI: -3.86–2.41) for Ogawa and Inaba, respectively. The confidence intervals for these mean cell differences lie within the predetermined equivalence range of +/- 5 ASCs per 106 PBMCs, suggesting equivalence or near equivalence between anti-LPS and anti-OSP ASC responses at a significance level of 0.05.
To our knowledge, this study is the first report of ASC responses to the BivWC oral cholera vaccine. We demonstrate that when administered according to the recommended two dose schedule in adults living in Haiti, BivWC induces a robust mucosal response to the V. cholerae O1 antigen following the first vaccine dose. However, a second dose of the vaccine administered 14 days later did not result in a robust mucosal response to the V. cholerae O1 antigen.
V. cholerae is a non-invasive pathogen, and it is thought that immunity against cholera is most likely mediated by V. cholerae O1-antigen specific IgA produced at the mucosal surface. For this reason, IgA-secreting ASCs, which transiently appear in the circulating following vaccination and ultimately differentiate into mucosal plasma cells provide an important window into mucosal immunity. In this context, our finding that a second dose of the OCV administered 14 days after the first dose does not result in significant ASC responses in Haitian adults raises important questions about the appropriate dosing interval for the BivWC vaccine in cholera-endemic areas. While the 14-day interval may allow for the delivery of two vaccine doses to people during an outbreak as quickly as possible, it is not clear whether this is a sufficient time period to generate sufficient immunologic memory in order to achieve a prime-boost effect with the second vaccine dose. Given that our findings mirror previous studies conducted in cholera-endemic areas of South Asia following vaccination with the WC-rBS vaccine,[32,34,40] we suspect that the current 14-day dosing schedule is too short for the generation of memory B cells that can respond to secondary antigenic stimulation upon re-exposure to V. cholerae O1 OSP.
To address this question regarding the timing of the second vaccine dose definitively would require additional studies. Specifically, evaluating the effect of re-vaccination with oral cholera vaccine at different intervals and determining whether repeated vaccination results in increasing levels of class switching and affinity maturation among the resulting V. cholerae-specific ASCs, would determine whether bona fide functional and long lasting B cell memory can be boosted following repeated doses of oral cholera vaccination administered at longer intervals. These results could then provide mechanistic context for larger vaccine effectiveness studies comparing both single and multi-dose OCV regimens.
Another notable finding in this report was derived from measuring ASC responses separately to both V. cholerae O1 LPS and the O-antigen-specific polysaccharide (OSP) moiety of LPS. These responses were not only strongly correlated, but also nearly equivalent, suggesting that immune responses primarily target the OSP portion of the antigen, rather than the O-antigen core, lipid A, or other component of the LPS preparation. These findings have implications both for subsequent evaluations of immune responses following cholera infection or vaccination, and the development of more immunogenic vaccines for cholera.
Another finding in our study, was the significant correlation we observed between day 7 antigen-specific plasma IgA responses and day 7 antigen-specific ASC responses. This correlation has been previously described following infection with V. cholerae or vaccination, [8,34] and suggests that plasma IgA responses on day 7 following vaccination does indeed serve as a potential proxy for the mucosal ASC response in some circumstances.
Our study has a number of limitations. Because cholera is now endemic in Haiti, our cohort had high pre-existing levels of vibriocidal antibodies before vaccination with 50% of participants having a vibriocidal ≥ 80 for either the Ogawa or Inaba. These findings are in line with a serosurvey conducted in Haiti in 2011 demonstrating that 64% of individuals had a vibriocidal antibody titer ≥ 80 for either V. cholerae Inaba or Ogawa, suggesting high levels of prior infection.[41] This is important because ASC responses to vaccination are likely to differ between populations based on the level of previous antigen exposure. For example, in contrast to our results in Haiti where evidence of cholera exposure was common, a previous study conducted among Swedish volunteers demonstrated significant CT-specific ASC responses to a WC-rBS vaccine following the second dose of vaccine as part of a similar vaccination schedule spaced 14 days apart, suggesting differences between responses in endemic and non-endemic settings.[39] Levels of previous exposure also effect the kinetics of the ASC response, and ASC responses peak earlier among primed populations.[42] Because we only measured ASC responses at a single time point after each vaccination (7-days) we could not evaluate the kinetic aspects of the response. While this is a limitation of the study, we do not think it changes the major findings, because based on previous studies, it is very unlikely that such priming would shift the kinetics of ASC responses to the extent they would become undetectable by day 7.[42] Lastly, while our results should be generalizable to adults living in other cholera endemic areas, additional studies are also needed in children and other subpopulations in cholera-affected regions, who may differ in their mucosal response to the BivWC vaccine.
In conclusion, vaccination with the BivWC oral cholera vaccine induces a significant mucosal ASC response targeting the OSP portion of V. cholerae LPS in Haitian adults, while a second dose of vaccine does not appear to stimulate antibody secreting cell responses in this population. These findings are similar to responses with the WC-rBS oral cholera vaccine in other cholera-endemic regions. More detailed evaluations of whether alternate vaccine dosing intervals can stimulate more robust mucosal immune responses, and contribute to improved duration of memory B cell targeting the V. cholerae O-antigen should be a priority for future studies.
|
10.1371/journal.pbio.1001732 | A Downy Mildew Effector Attenuates Salicylic Acid–Triggered Immunity in Arabidopsis by Interacting with the Host Mediator Complex | Plants are continually exposed to pathogen attack but usually remain healthy because they can activate defences upon perception of microbes. However, pathogens have evolved to overcome plant immunity by delivering effectors into the plant cell to attenuate defence, resulting in disease. Recent studies suggest that some effectors may manipulate host transcription, but the specific mechanisms by which such effectors promote susceptibility remain unclear. We study the oomycete downy mildew pathogen of Arabidopsis, Hyaloperonospora arabidopsidis (Hpa), and show here that the nuclear-localized effector HaRxL44 interacts with Mediator subunit 19a (MED19a), resulting in the degradation of MED19a in a proteasome-dependent manner. The Mediator complex of ∼25 proteins is broadly conserved in eukaryotes and mediates the interaction between transcriptional regulators and RNA polymerase II. We found MED19a to be a positive regulator of immunity against Hpa. Expression profiling experiments reveal transcriptional changes resembling jasmonic acid/ethylene (JA/ET) signalling in the presence of HaRxL44, and also 3 d after infection with Hpa. Elevated JA/ET signalling is associated with a decrease in salicylic acid (SA)–triggered immunity (SATI) in Arabidopsis plants expressing HaRxL44 and in med19a loss-of-function mutants, whereas SATI is elevated in plants overexpressing MED19a. Using a PR1::GUS reporter, we discovered that Hpa suppresses PR1 expression specifically in cells containing haustoria, into which RxLR effectors are delivered, but not in nonhaustoriated adjacent cells, which show high PR1::GUS expression levels. Thus, HaRxL44 interferes with Mediator function by degrading MED19, shifting the balance of defence transcription from SA-responsive defence to JA/ET-signalling, and enhancing susceptibility to biotrophs by attenuating SA-dependent gene expression.
| The highly conserved Mediator complex plays an essential role in transcriptional regulation by providing a molecular bridge between transcription factors and RNA polymerase II. Recent studies in Arabidopsis have revealed that it also performs an essential role in plant defence. However, it remains unknown how pathogens manipulate Mediator function in order to increase a plant's susceptibility to infection. In this article, we show that a secreted effector, HaRxL44, from the Arabidopsis downy mildew pathogen Hyaloperonospora arabidopsidis (Hpa), interacts with and degrades the Mediator subunit MED19a, resulting in the alteration of plant defence gene transcription. This effector-mediated interference with host transcriptional regulation perturbs the balance between jasmonic acid/ethylene (JA/ET) and salicylic acid (SA)–dependent defence. HaRxL44 interaction with MED19a results in reduced SA-regulated gene expression, indicating that this pathogen effector modulates host transcription to promote virulence. The resulting alteration in defence transcription patterns compromises the plant's ability to defend itself against pathogens, such as Hpa, that establish long-term parasitic interactions with living host cells via haustoria (a pathogen structure that creates an expanded host/parasite interface to extract nutrients) but not against necrotrophic pathogens that kill host cells. HaRxL44 is unlikely to be the sole effector that accomplishes this shift in hormonal balance, and other nuclear HaRxL proteins were reported by other researchers to interact with Mediator components, as well as with other regulators of the JA/ET signalling pathway. Functional analyses of these effectors should facilitate the discovery of new components of the plant immune system. These data show that pathogens can target fundamental mechanisms of host regulation in order to tip the balance of signalling pathways to suppress defence and favour parasitism.
| Plants and microbial pathogens co-evolve; pathogens are selected to evade host defence, and plants are selected to detect and resist pathogens [1],[2]. Resistance mechanisms include not only pattern-triggered immunity (PTI) and effector-triggered immunity (ETI) [1], but also local and systemic plant defence responses that are controlled through distinct, but partially interconnected pathways involving the hormones salicylic acid (SA) and jasmonic acid (JA)/ethylene (ET) [3]. Adapted pathogens have a substantial repertoire of effectors that can suppress PTI by various mechanisms [4] but only one effector has been shown to interfere with SA-triggered immunity (SATI) [5]. An important role in plant defence has been attributed to nuclear processes, since there are many reports that nuclear localisation of pathogen effectors, R proteins, and key host components, including transcription factors and regulators, is essential for plant immunity [6]. This observation suggests that effectors may manipulate host transcription or other nuclear regulatory components for the benefit of the pathogen.
Although filamentous phytopathogens such as fungal rusts and powdery mildews and oomycete downy mildews and white rusts are more damaging to agriculture than bacteria, their effector functions are more poorly understood. Fungal and oomycete effectors are secreted, and then taken up by the host cell via a poorly understood mechanism that for many oomycetes involves the N-terminal RxLR motif [7],[8]. Sequencing of several oomycete genomes including the model organism Arabidopsis downy mildew Hyaloperonospora arabidopsidis (Hpa) has allowed prediction of a repertoire of effector candidate genes that share N-terminal sequence motifs with known effectors [9],[10]. To establish an inventory of the Hpa RXLR effectors (HaRxLs), the draft genome of Hpa Emoy2 was scanned and HaRxL effector candidates were cloned. Because transformation of biotrophic pathogens such as Hpa is difficult, we developed heterologous systems to assess HaRxL functions [11],[12]. We first deployed a Pseudomonas syringae pv. tomato (Pst) type three secretion (T3S)–based delivery system (EDV) to look for HaRxLs that enhance Pst virulence and/or that suppress host defence outputs such as callose deposition, in order to prioritize effectors for follow-up studies [13],[14]. We next screened for the subcellular localisation of the HaRxL collection and identified 15 HaRxL effectors that localise to the plant cell nucleus when stably expressed in Arabidopsis [15],[16] and interact in yeast with nuclear plant proteins implicated in transcription [16],[17]. In particular, in yeast 2-hybrid (Y2H) assays, HaRxL44 interacts with MED19a, a subunit of the Arabidopsis Mediator complex [18]. Six other Hpa effectors interact with host Mediator or regulators of Mediator ([18]; Figure S1).
Mediator is a conserved multisubunit complex that acts as a molecular bridge between transcriptional regulators at gene enhancer sequences and the activation of transcription by RNA polymerase II at the transcription start site [18],[19]. Eight of 10 essential Mediator genes conserved between S. pombe and S. cerevisiae (including MED19) also have a metazoan homologue, indicating that a Mediator core has been conserved throughout evolution and is present in all eukaryotic cells [20]. Mediator is a large complex (>25 components), but different subunits are implicated in integration of specific external stimuli [21],[22]. Mediator has numerous functions in addition to interacting directly with RNA polymerase II as it can interact with and coordinate the action of many other co-activators and co-repressors, including those acting on chromatin [20]. These interactions ultimately allow the Mediator complex to deliver outputs ranging from the maximal activation of genes, through the modulation of basal transcription, to long-term epigenetic silencing [20]. Despite the importance of Mediator, this complex has been little studied, due to the lethality of mutants in most multicellular organisms. However, null mutations of Mediator subunit genes are often not lethal in plants, making these organisms a valuable model for studying the Mediator complex. In Arabidopsis, several Mediator subunits have been shown to have a specific function in the activation of signalling pathways during plant development and in response to abiotic stress. MED12/CRP (CRYPTIC PRECOCIOUS) and MED13/MAB2 (MACCHI-BOU2) are required for early embryo patterning, and also regulate flowering and cotyledon organogenesis, respectively [23],[24]. MED14/SWP (STRUWWELPETER) is a key regulator of cell proliferation [25]. MED16/SFR6 (SENSITIVE TO FREEZING6) integrates cellular and environmental cues into the circadian clock [26]–[28] and is required for cold acclimation. MED17, MED18, and MED20a play an important role in the production of small and long noncoding RNAs [29]. MED25/PFT1 (PHYTOCHROME AND FLOWERING TIME1) was first identified as a key regulator of flowering [30] and later found to regulate final organ size and light signalling [31],[32]. MED33a/RFR1 (REF4-RELATED1) and MED33b/REF4 (REDUCED EPIDERMAL FLUORESCENCE4) are required for phenylpropanoid homeostasis [33].
Mediator was recently shown to play a role in plant immunity and pest resistance. It was initially shown to be important for the activation of JA/ET-dependent defences against necrotrophic pathogens, via MED21 and MED25 [34],[35]. Other studies reveal a role for Mediator in the activation of SATI [36]. The Mediator subunits MED14, MED15, and MED16 have all been reported to be required for the biological induction of systemic acquired resistance (SAR) [37]–[39], suggesting that the Mediator may function in SAR activation. Both MED14 and MED15 appear to function downstream of NPR1 and do not affect the nuclear localisation or stability of NPR1 [37],[39], whereas MED16 makes a positive contribution to the accumulation of NPR1 protein [39]. The Mediator complex thus appears to be a “hub” for the plant immune system, but little is known about how the pathogen manipulates its function to promote disease.
We report here the functional analysis of a nuclear downy mildew effector, HaRxL44, which interacts with Mediator subunit 19a (MED19a), and causes its degradation via proteasome-mediated degradation of this subunit. Expression profiling revealed an induction of JA/ET signalling in the presence of HaRxL44, mimicking that observed after 3 d of compatible interaction. This increase in JA/ET signalling was associated with low levels of SATI in both Arabidopsis plants expressing HaRxL44 and in med19a knock-out mutants, whereas high levels of SATI were observed in plants overexpressing MED19a. Using the PR1::GUS reporter, we confirmed that Hpa abolishes PR1 expression specifically in cells containing haustoria. Thus, HaRxL44 affects via MED19a the balance between JA/ET and SA signalling and thus enhances biotroph susceptibility.
In a previous functional screen for Hpa virulence factors, we identified HaRxL44 (Figure 1A) as an enhancer of bacterial virulence in Arabidopsis [13]. The amino-acid sequence of HaRxL44 displays similarity to two predicted RXLR effectors from Phytophthora infestans, PITG-04266 and PITG_07586, and avh109 from P. sojae (Figure S2A). As observed for its homolog PITG_07586 from the “plastic secretome” of P. infestans [40], HaRxL44 is found in a region of the Hpa genome enriched in retrotransposons (Figure S2B) and is conserved between Hpa races (Figure S2C). We confirmed the effect of HaRxL44 on virulence (Figure 1B) by generating transgenic lines of Arabidopsis expressing HaRxL44 under the control of various promoters (Figure S3). Subcellular localization of GFP-HaRxL44 in a stably transformed Arabidopsis line (Figure 1C) confirmed its nuclear localization during Hpa infection, during which the nucleus is found closely associated with Hpa haustoria [15].
In an extensive Y2H screen [17], HaRxL44 was found to interact with several nuclear proteins, including MED19a (Figure S3A, S3B). We assessed the functional role of Mediator in immunity to Hpa, by studying the contribution of MED19a during Hpa infection. We first isolated med19a loss-of-function alleles (Figure 2A, 2B) and found that med19a mutant plants had a wild-type (WT) phenotype, with the exception of abnormally shaped siliques (Figure 2C). In parallel, we generated Col-0 Arabidopsis transgenic lines overexpressing a construct encoding MED19a fused to a GFP tag (OE MED19a; Figure 2D). Homozygous med19a-1 and med19a-2 mutants expressing GFP-MED19a were produced in order to check for complementation. We tested by Western blot the expression of GFP-MED19a in the mutant background, and selected lines with lower expression levels than observed for OE MED19a lines (C1, C2; Figure S4). In these selected lines, GFP-MED19a rescued the phenotype observed during plant development (Figure 2C).
We confirmed that the fusion protein was functional, by checking that GFP-MED19a interacted with the Mediator complex. Immunoprecipitation of the GFP-MED19a protein in Arabidopsis led to the detection of both MED6 and MED7 in pull-down assays with native antibodies (Figure 2E).
We then analysed the subcellular localisation of MED19a in vivo in Arabidopsis by confocal microscopy. Live-cell imaging showed that GFP-MED19a and HaRxL44 were present in the same compartments: the nucleoplasm and nucleolus of the plant cell (Figure 2F). Western-blot analysis of two independent transgenic lines producing GFP-MED19a (Figure 2G) demonstrated the presence of a GFP-MED19a protein of the expected size (50 kDa), together with additional signals at higher molecular weights (60 kDa and 70 kDa), suggesting that MED19a is modified post-translationally in planta.
We then challenged the transgenic lines with Hpa and monitored pathogen growth after six days. Both the med19a-1 and med19a-2 mutants were more susceptible to Hpa than wild type, similar to a med15 mutant, which has impaired SATI (med15 [37]; Figure 2H). Complemented lines displayed the same level of susceptibility as wild type plants (Figure 2H), confirming the functionality of the fusion protein. By contrast, transgenic lines overproducing MED19a were more resistant to Hpa than the WT or Arabidopsis lines expressing GFP alone. Therefore, GFP-MED19a can associate with other Mediator subunits and complements med19a loss of function alleles, which suggests that the fusion protein is functional. Thus, the Mediator subunit MED19a is a positive regulator of plant immunity to Hpa.
We monitored the subcellular location of RFP-MED19a and GFP-HaRxL44 using confocal microscopy. Both proteins localise to the nucleoplasm and nucleolus, whereas Bimolecular Fluorescence Complementation (BiFC) signals resulting from the co-expression of YFPc-MED19a and YFPn-HaRxL44 constructs are restricted to the nucleolus, following transient expression in N. benthamiana. No BiFC signal was detected in the nucleoplasm, the site of Mediator function (blue arrow, Figure 3A). The destabilisation of RFP-MED19a in the presence of GFP-HaRxL44 was quantifiable by both Western blotting (blue arrow, Figure 3B, Figure S5A) and confocal microscopy (blue arrow, Figure 3C). Furthermore, no decrease in the amount of GFP-MED19a was observed in coexpression experiments with RFP-24 and RFP-45 constructs, which encode other nuclear HaRxLs (Figure 3C, Figure S5A), suggesting that MED19a is specifically targeted by the HaRxL44 effector. As MED19a transcript levels were not affected in HaRxL44 lines (Figure S5B), we conclude that HaRxL44 destabilizes MED19a at the protein level. Taken together these results show that MED19a, which is found in both nucleoplasm and nucleolus, disappears in the nucleoplasm in the presence of HaRxL44, and perhaps persists in the nucleolus because of low proteasome activity in the nucleolus. Since Mediator is known to function in the nucleoplasm, this HaRxL44-mediated degradation of MED19 likely affects Mediator activity.
In the Y2H screen [17], HaRxL44 was found to interact with two E3 ligases (Figures 4A and S3), BOTRYTIS SUSCEPTIBLE 1 (BOI; AT4G19700) and MED25-BINDING RING-H2 PROTEIN-like (MBR1-like; AT1G17970). We investigated whether these E3 ligases are present in the same plant cell compartment as HaRxL44 and MED19a. We investigated the subcellular distribution of these two E3 ligases, by transiently expressing GFP-tagged versions of BOI and MBR1-like in N. benthamiana (Figure 4B). GFP-BOI localises to the nucleoplasm and accumulates in foci, in four to five large bodies. Furthermore, no GFP-BOI signal was detected in the plant cell nucleolus (Figure 4B). GFP-MBR1–like was also localised to the plant cell nucleus (Figure 4B), in a pattern similar to that observed for proteins involved in RNA splicing [41]. GFP-MBR1–like accumulated in large amounts in the plant cell nucleolus and had a punctate distribution in the nucleoplasm (Figure 4B). In order to test whether one of the two E3-ligases interacting with HaRxL44 in Y2H might be responsible for MED19a degradation, we tested the phenotype of BOI and MBR1-like loss-of-function mutants during Hpa infection. Surprisingly, both the boi RNAi line and the mbr1-like T-DNA KO line were more susceptible to Hpa (Figure S5C). However, such loss-of-function experiments are difficult to interpret because BOI and MBR1-like might also affect other components of the plant immune system, leading to an increase in plant susceptibility.
As HaRxL44 interacts in Y2H analysis with E3 ligases located in the plant cell nucleus (Figures 4 and S3), we hypothesised that HaRxL44 acts as an adaptor protein for E3 ligases, mediating the degradation of MED19a. Indeed, we showed that inhibition of the proteasome by the addition of 100 µM MG132 for 4 h prevented HaRxL44-induced degradation of GFP-MED19a (Figure 4C). The addition of 100 µM MG132 during protein extraction prevented the degradation of GFP-MED19a in the presence of HA-HaRxL44 and made it possible to confirm the interaction of these proteins in planta, by co-immunoprecipitation (red arrow, Figure 4D). We next tested if blocking the proteasome would allow the detection of the interaction between HaRxL44 and MED19a in the nucleoplasm. We showed that addition of MG132 1 h before observation with confocal microscopy allowed the detection of the interaction between YFPc-MED19a and YFPn-HaRxL44 in the nucleoplasm by BiFC (Figure 3A). Thus, HaRxL44 interacts with MED19a, a positive regulator of plant immunity to Hpa, leading to its destabilisation in a proteasome-dependent manner.
In order to check if the interaction between HaRxL44 with MED19a is important for its degradation, we generated a series of HaRxL44 mutants by NAAIRS-scanning mutagenesis [42]. We obtained one mutant, HaRxL44M, mutated in the nucleolus-localization signal (Figure S6), which no longer interacts with MED19a by Co-IP when transiently expressed in N. benthamiana (Figure 5A). In contrast with HaRxL44, which is visible in the nucleoplasm and the nucleolus (Figure 5B), HaRxL44M presents a nuclear-cytoplasmic localisation (Figure 5B). Using both cell biology (Figure 5B, 5C) and biochemistry (Figure 5D) we showed that HaRxL44M no longer degrades MED19a when transiently co-expressed in planta. Thus, the interaction between HaRxL44 and MED19a is important for proteasome-dependent MED19a degradation.
First, we verified the degradation of MED19a in the presence of HaRxL44 in Arabidopsis, by generating a transgenic line expressing both GFP-MED19a and 3HA-Strep2-HaRxL44 (or 3HA-Strep2-GUS as control). We showed that, as we observed in N. benthamiana, MED19a is degraded in the presence of HaRxL44 in Arabidopsis and the addition of MG132 blocks the effect of HaRxL44 on MED19a stability (Figure S5D). We then investigated whether the presence of HaRxL44 affects the interaction between MED19a and the Mediator complex. We found that, even in the presence of 3HA-HaRxL44, MED6 co-immunoprecipitates with GFP-MED19a in Arabidopsis (Figure S5E), suggesting that MED6 and GFP-MED19a also associate in the nucleolus. However, overproduction of MED19a and HaRxL44 in Arabidopsis may affect the stoichiometry or nuclear/nucleolar distribution of interactions between MED19a and the Mediator complex, or Mediator subcomplexes, obscuring potential effects of HaRxL44 on the integration or stability of MED19a subunits in the Mediator complex.
As MED19a is part of a major transcriptional regulatory complex, we then investigated whether and how HaRxL44 expression affects transcription. Illumina RNA-sequencing revealed a positive correlation between the genes differentially up-regulated in HaRxL44-lines and by methyl JA (MeJA) treatment [43] (Hypergeometric probability <0.001; Figure 6, Tables S1 and S2). No correlation was observed for down-regulated genes in HaRxL44–line 1 (Hypergeometric probability = 0.98). This result can be explained by the lower number of genes differentially expressed in HaRxL44-line 1 compared to HaRxL44–line 2. However, the average fold change in HaRxL44–line 1 is still correlated to what is observed in HaRxL44–line 2 (Figure 6, Table S2). We confirmed, by QRT-PCR, that JA/ET marker genes (PDF1.2, JAZ1, and JAR1) were induced in HaRxL44-lines and in med19a mutants, with respect to WT levels (Figures 7A, 7B and S7A, S7B). Two of the five JA-responsive genes from the JA biosynthesis pathway [44], OPR3 (AT2G06050) and LOX2 (AT3G45140), were up-regulated (Figure S7C, Table S2), suggesting that HaRxL44 may induce JA/ET signalling.
We then checked whether the induction of JA/ET-responsive genes in the presence of HaRxL44 was biologically significant. We conducted gene expression profiling over a time course of Hpa infection in Arabidopsis and found that PDF1.2 induction is observed 3 d after infection (DAI), when HaRxL44 transcription was induced (Figure 7C, 7D). Furthermore, the induction of JA/ET-responsive genes in HaRxL44 transgenic lines was similar to the induction observed during early stages of Hpa infection in susceptible accessions of Arabidopsis (Figure 6, Tables S1 and S2). Thus, JA/ET signalling is induced in the presence of HaRxL44, the absence of MED19a, and 3 d after Hpa infection.
In Pst, the phytotoxin coronatine (COR) acts as an analogue of JA and contributes to bacterial invasion [45]. COR biosynthetic (COR−) mutants of Pst strain DC3000 exhibit reduced virulence on Arabidopsis when surface-inoculated [45]. In order to test if HaRxL44 was able to complement PstCOR− strain, we delivered HaRxL44 in planta by using EDV system (EDV-HaRxL44, [13]). When spray-inoculated in Arabidopsis Col-0 plants, PstCOR− growth was reduced by two logs (cfu/cm2) compared to Pst (Figure 7E). PstCOR− EDV-HaRxL44 growth was increased by one log compared to PstCOR− (**p value<0.01, Figure 7E). These results indicate that HaRxL44 is able to complement the deficiency of COR production in PstCOR− and supports a key role for this Hpa effector in the activation of the JA/ET pathway.
Because JA/ET-induced defence is effective against necrotrophs [3], we next challenged the transgenic lines expressing HaRxL44 with Botrytis cinerea (Figure 7F). As control, we used loss-of-function alleles of HISTONE MONOUBIQUITINATION1 (HUB1) shown to increase susceptibility to B. cinerea, and HUB1-OE lines that confer resistance to B. cinerea [34]. We observed that B. cinerea grew less well in HaRxL44-OE lines than in the WT, as also observed for HUB1-OE lines (*p value<0.01; Figure 7F, Figure S7D to S7F). Altogether, these results suggest that JA/ET-dependent defence is promoted in lines that express HaRxL44.
The Mediator complex is known to be important for JA/ET signalling [46]. In particular, MED25 and MED21 are key components of Mediator that regulate JA/ET-induced gene expression [34],[35]. We then tested if the med21 and med25 loss-of-function mutants are altered in Hpa growth. We observed that in both med21 RNAi line and med25 knock out (KO) mutants, Hpa growth was reduced compared with WT (Figure 2H). Thus, JA/ET-responsive gene transcriptional activation via Mediator is important for Hpa virulence.
As the activation of the JA/ET defence pathway can antagonise SATI [3], we next assessed whether HaRxL44 suppresses SATI. We first observed, by QRT-PCR, that SA marker genes (PR1, LURP1, WRKY70, PR2, PR5 genes) are down-regulated in HaRxL44 transgenic lines (Figures 8A–C and S8A, S8B). We then assessed PR1 induction after elicitation. In HaRxL44 transgenic lines, basal PR1 transcript levels are lower than those in the WT, resulting in a reduction of PR1 induction levels 8 h after SA treatment (Figure 8D). Similar results were observed in med19a mutants (Figure 8E), whereas MED19a OE led to stronger PR1 induction (from 5 to 15 times higher level of PR1 expression in MED19 OE lines compared to control plants; Figure 8F). We then investigated whether Hpa suppresses SATI. Expression profiling in Col-0 plant infected with Hpa Waco9 revealed a 40-fold change in PR1 gene induction 3 DAI (Figure 9A). We then investigated the cell-specific expression pattern of PR1, by infecting PR1::GUS lines with Hpa. PR1::GUS staining was restricted to the plant vascular tissues in contact with Hpa 3 DAI (Figure S8C, S8D), whereas strong GUS staining was observed throughout the entire leaf 6 DAI (Figure 9B). An analysis of PR1 expression patterns at the cellular level showed that PR1::GUS staining was absent from Hpa-infected cells, whereas PR1::GUS staining was observed only in the cell layer surrounding the mesophyll cells into which haustoria had penetrated (Figure 9B). Thus, Hpa suppresses SATI specifically in the haustoria-containing mesophyll cells to which the effector proteins are delivered. As expected, the amount of PR1 mRNA generated in response to Hpa was lower in the absence of med19a, as shown by QRT-PCR (Figure 9C).
We next tested whether MED19a is degraded upon infection by Hpa. In GFP-MED19a lines, we tried to image signal in an infected mesophyll cell and compare this to the signal level to the signal in the neighbouring cells. However, measurement of fluorescence by confocal microscopy in deep tissues was too difficult to allow us to obtain reliable results. Therefore, we used the med19a mutant lines complemented with GFP-MED19a in order to check by Western blot analysis the GFP-MED19a protein level in Hpa-infected tissues compared to uninfected tissues. GFP-MED19a signal in infected tissues was reduced compared to uninfected tissues (Figure 9D), confirming that this positive regulator of plant immunity against Hpa is degraded after infection. We suggest that the destabilisation of MED19a by HaRxL44 results in transcriptional reprogramming, leading to changes in the balance between the JA/ET and SA pathways, promoting biotrophy.
We report here the functional analysis of an Hpa nonpolymorphic effector, HaRxL44. We verified Y2H data suggesting that HaRxL44 interacts with MED19a. We found MED19 to be a positive regulator of plant immunity against Hpa, leading to proteasome-dependent degradation of MED19a. Expression profiling reveals that JA/ET signalling is elevated in the presence of HaRxL44, in med19a knock-out mutants, and 3 d after Hpa infection. Strong JA/ET signalling is associated with weak SATI in both Arabidopsis plants expressing HaRxL44 and in med19a KO mutants, whereas strong SATI is observed in plants overexpressing MED19a. We confirmed that Hpa represses PR1 expression specifically in the cells containing haustoria, into which RxLR effectors are delivered. Thus, HaRxL44 hijacks nondefensive aspects of the JA/ET signalling pathway, at the transcriptional level, via MED19a, resulting in reduced capacity to defend against biotrophs. A translocated chorismate mutase from Ustilago maydis was reported to be able to lower SATI by acting on SA biosynthesis [5]. In contrast, we report here a new mechanism of SATI suppression by means of a biotrophic oomycete effector that alters SA-dependent transcription by promoting degradation of MED19a, a transcriptional component involved in SA/JA crosstalk.
MED19/Rox3 was originally identified in a search for mutants increasing aerobic expression of the CYC7 gene in yeast [47]. The nuclear localisation of this protein and the nonviability of null mutants suggest that the MED19/Rox3 protein is a general regulatory factor [47]. The purification of Mediator from a strain lacking the MED19 subunit [48] led to the demonstration that MED19/Rox3 regulated intermodule interactions in the S. cerevisiae Mediator complex. In Arabidopsis, MED19 is encoded by two genes, MED19a and MED19b. Only MED19a has been reported to be involved in Mediator complex formation in Arabidopsis [18]. HaRxL44 interacts with both MED19a and MED19b in Y2H screen [17]. We therefore tried to amplify both genes, but were unable to amplify the MED19b gene from cDNA or genomic DNA. Furthermore, no T-DNA insertion into the MED19b gene is available, limiting analyses of the function of this gene in response to Hpa. In this study, we focused on the role of MED19a during Hpa infection. However, it should be borne in mind that the phenotype observed for med19a KO mutants may be only partial, because MED19a and MED19b could have redundant functions. In Arabidopsis, there are other Mediator subunits encoded by several genes, such as MED10, MED20, MED22, and MED33. Transcript profiling with med20a and the RNA polymerase II subunit RPB2 mutant nrpb2-3 revealed a high degree of overlap in the lists of genes displaying down-regulation in the two mutants [29]. This suggests that even a single mutation in one of several paralogs encoding an Arabidopsis Mediator subunit can lead to a quantifiable phenotype.
We first confirmed that MED19a was part of the Mediator complex, by demonstrating its interaction with MED6 and MED7 in planta. We then investigated the subcellular distribution of MED19a, which was found to be localised to the plant cell nucleoplasm, as reported for MED16 [26]. MED19a was also localised to the plant nucleolus. This is surprising, because Mediator is thought to associate with RNA polymerase II in the nucleoplasm. It has been suggested that Mediator regulates the action of other plant RNA polymerases [49]. The similarities between RNA polymerases II, IV, and V raise the possibility that Mediator may associate with another polymerase, either polymerase IV or polymerase V [49]. MED19a may even associate with the nucleolar RNA polymerase I or III. Indeed, Mediator subunits have been shown to interact with RNA polymerase I subunits in Y2H assays ([50], Figure S1). However, as the evidence concerning the possible role(s) of Mediator in directing the activity of other RNA polymerases remains inconclusive, we decided to focus on the role of MED19a in the regulation of transcription by RNA polymerase II. A proteomic analysis of human nucleoli revealed the presence of a large number of proteins with no known nucleolar function [51]. Nucleolar protein composition is not static and may undergo significant modification in response to the metabolic state of the cell [52]. The regulation of protein activity by nucleolar sequestration has been reported before [53],[54]. Indeed, this phenomenon has already been reported for human MED1 [55]. MED1 is phosphorylated by MAPK1 or MAPK3 during the G2/M phase, enhancing protein stability and promoting the entry of this molecule into the nucleolus [55]. We can speculate that MED19a is sequestered in the nucleolus to remove it from the functional pool of MED19a in the nucleoplasm.
We then investigated whether the presence of HaRxL44 affected the interaction between MED19a and the Mediator complex. We showed that even in the presence of HaRxL44, MED19a associated with MED6 in Arabidopsis. However, we cannot exclude the possibility that the overproduction of MED19a and HaRxL44 in Arabidopsis affects the stoichiometry between MED19a and the Mediator complex, obscuring potential effects of HaRxL44 on the integration of MED19a subunits into the Mediator complex.
We show here that HaRxL44 interferes with Mediator function by promoting the proteasome-dependent degradation of MED19a. Effectors from plant pathogens have been reported to suppress various layers of plant defence by controlling the ubiquitination and degradation of proteins important for plant immunity via the proteasome.
AvrPtoB is a well-studied Pseudomonas syringae effector that mimics a plant E3 ligase [56] and facilitates the degradation of key components of PAMP-triggered immunity [57]–[60]. The Xanthomonas effector XopL has been shown to display E3 ubiquitin ligase activity in vitro and in planta, to induce plant cell death, and to suppress plant immunity [61]. The structural fold of the E3 ubiquitin ligase domain in XopL is unique, and the lack of cysteine residues in the XL-box suggests a noncatalytic mechanism for XopL-mediated ubiquitination [61]. The P. syringae effector HopM1 mediates the degradation, by the proteasome, of AtMIN7, a plant protein involved in the vesicular trafficking of defence components [62],[63]. Unlike AvrPtoB and XopL, HopM1 has no E3 ligase activity, suggesting that this effector acts as an adaptor protein, connecting AtMIN7 and the proteasome [62]. Several ubiquitin proteins have been identified in the Meloidogyne incognita secretome, and a ubiquitin extension protein secreted from the dorsal pharyngeal gland of Heterodera schachtii has also been detected [64],[65]. The Magnaporthe oryzae effector AvrPiz-t was recently reported to interact with a RING E3 ubiquitin ligase, APIP6, abolishing its ubiquitin ligase activity [66]. In addition, the P. infestans RXLR effector AVR3a has been shown to target and stabilise the nucleolar E3 ligase CMPG1, which is required for the programmed cell death triggered by the elicitin INF1 [67],[68]. However, the targets for the ubiquitination of these E3 ligases have yet to be determined.
We show here that HaRxL44 interacts with MED19a, destabilising this Mediator subunit in a proteasome-dependent manner. As HaRxL44 displays no sequence similarity to plant E3 ligases, we suggest that, like HopM1, HaRxL44 acts as an adaptor, presenting MED19a to the proteasome or to an E3 ligase. However, the mechanism by which HaRxL44 induces the degradation of MED19a remains unclear. Y2H screens have shown that HaRxL44 interacts with two E3 ligases: BOI and MBR1-like [17]. BOI is encoded by a gene from a multigene family with four known members, including BOI-RELATED GENE [69]. BOI was identified in a screen for proteins interacting with BOTRYTIS SUSCEPTIBLE 1 (BOS1), which encodes an R2R3 MYB transcription factor involved in restricting necrotroph-induced necrosis [70]. BOI is an important player in plant immunity to necrotrophic pathogens [71]. BOI ubiquitinates BOS1, leading to its rapid degradation by the proteasome [71]. In addition to its role in restricting necrosis, BOI may integrate plant responses to diverse signals [72]. Indeed, Park et al. (2013) recently showed that BOI and DELLA proteins inhibit GA responses by interacting with each other, binding to the same promoters of GA-responsive genes, and repressing these genes. In the Y2H screen carried out by Mukhtar et al. (2011) [17], BOI was found to interact with four nuclear effectors from Hpa: HaRxL44, HaRxL10, ATR1, and ATR13. Thus, Hpa effectors may act on BOI function, to render the plant more susceptible to biotrophic pathogens.
It is not clear whether the HaRxL44-mediated degradation of MED19a by the proteasome has a positive or negative impact on transcription. It is well known that one major way of regulating transcription is to couple the activity of transcription factors to their destruction by the proteasome [73]. This “transcription-coupled destruction” mechanism of activator action [74] must serve a functional purpose, such that, if blocked, repeated rounds of transcriptional activation cannot occur [73]. This “unstable when active” phenomenon is seen with many transcriptional regulators, including the Mediator subunit MED25 [75]. In Arabidopsis, MED25 is a highly unstable protein, degraded by the proteasome both in vitro and in vivo [75]. A blockade of proteasome activity prevents MED25 from inducing flowering [75]. Two E3 ubiquitin ligases, MBR1 and MBR2, have been shown to polyubiquitinate MED25 in planta, supporting the “transcription-coupled destruction” model for the regulation of MED25. MBR1 and MBR2 are part of a small cluster of E3 ligases in Arabidopsis [76]. HaRxL44 has been shown to interact with MBR1-Like in Y2H screens [17]. Thus, HaRxL44 may recruit different E3-ligases, to promote the destruction of MED19a, thereby promoting Hpa growth.
In metazoans, Mediator complex subunits are degraded upon cell differentiation [77]–[79]. This observation is consistent with the notion that subcomplexes of Mediator may display cell type–specific activity [78]. The degradation of some subunits helps to turn off the expression of a large portion of genes, whereas the retention of other subunits is required for the expression of a smaller, highly specific subset of genes [78],[80]. Based on our results, we hypothesise that HaRxL44 targets MED19a for degradation, to block the transcription of genes important for plant immunity (i.e., genes important for SA-dependent defence), whereas MED19a degradation allows the transcription of a small number of genes beneficial for Hpa, including JA/ET-induced genes.
We showed that JA/ET signalling is induced in the presence of HaRxL44 (or the absence of MED19a). Expression profiling using Illumina RNA sequencing revealed a positive correlation between the genes differentially up-regulated in HaRxL44-lines and by MeJA treatment [43]. No correlation was observed for down-regulated genes in HaRxL44–line 1, but this result can be explained by the low number of genes differentially expressed in HaRxL44–line 1 compared to HaRxL44–line 2. However, the average fold change in HaRxL44–line 1 is still correlated to what is observed in HaRxL44–line 2. This result is consistent with the quantitatively different phenotypes observed in these transgenic lines, such as susceptibility to Hpa (Figure 1B, [13]), induction of PDF1.2 (Figure 7A), and suppression of SA-responsive genes (Figure 8A-D). Thus, we believe that HaRxL44 affects JA/ET-regulated gene expression. Indeed, HaRxL44-expressing plants showing activation of JA/ET-responsive genes are more resistant to the necrotrophic pathogen B. cinerea for which JA/ET-dependent defence is required. Conversely, HaRxL44 expression (or the absence of MED19a) resulted in a loss of PR1-induction and higher rates of biotrophic pathogen growth. These results suggest that HaRxL44 affects the hormonal balance between JA/ET and SA, promoting biotrophy, by acting on the transcriptional machinery of the plant. Hpa infection also led to the expression of JA/ET-responsive genes, confirming the biological significance of the results obtained in the functional analysis of HaRxL44.
JA/ET and SA-dependent defences are known to be antagonistic. Arabidopsis mutants with impaired SA accumulation, such as eds4, eds5, and pad4, display high levels of PDF1.2 expression in response to inducers of JA/ET-dependent gene expression [81],[82]. Convincing evidence for such an antagonistic effect has also been reported for NON-EXPRESSOR OF PATHOGENESIS-RELATED GENES1 (NPR1) [83]. NPR1 is the key regulator of SAR, an important reaction in defence against pathogens. The Arabidopsis npr1 mutant displays high levels of JA/ET-responsive gene transcript accumulation and of JA and ET accumulation in response to P. syringae infection, suggesting that NPR1 is involved in the SA-mediated suppression of JA/ET signalling [83]. Reciprocally, an mpk4 mutant has been shown to display constitutive SA-dependent gene expression and higher SA levels and enhanced resistance to biotrophic pathogens [84]. MPK4 up-regulates JA/ET-responsive genes and simultaneously suppresses SAR, placing MPK4 at the heart of the antagonistic interaction between JA/ET and SA [84],[85].
The role of the Mediator complex in JA/ET and SA-responsive gene expression has recently been investigated. MED25, MED21, and MED8 have been shown to be important for the activation of JA/ET-induced gene transcription [34],[35]. MED25 plays a major role in the JA-responsive gene transcription pathway, through its interaction with the transcription factor MYC2, which plays a key role in the activation of JA-induced gene expression [86]–[88]. MED25 regulates JA-dependent defence responses, conferring resistance to necrotrophic pathogens, and a med25 mutant has been shown to be more susceptible than the WT to the hemibiotroph Fusarium oxysporum [89]. The effect of a med8 mutation on the JA/ET-induced expression of PDF1.2 is readily detectable only in med8 med25 double mutants [35]. MED21 RNA interference lines are susceptible to both B. cinerea and A. brassicicola [34]. MED21 has been shown to interact with a RING E3 ligase, HISTONE MONOUBIQUITINATION1 (HUB1), increasing resistance to necrotrophs [34].
MED14, MED15, and MED16 were recently reported to up-regulate SAR in Arabidopsis [37]–[39],[90]. We show here that mutations of the gene encoding MED19a increase the basal level of JA/ET-responsive gene transcription and decrease the responsiveness of PR1 gene expression to SA. The abolition of PR1 expression or the absence of MED19a (or the presence of HaRxL44) was associated with faster growth of Hpa in med19a KO mutants. Thus, HaRxL44 targets a positive regulator of plant immunity to biotroph pathogens, thereby interfering with hormonal balance and promoting biotrophy.
When the first results from expression profiling host gene expression became available, a paradox emerged [91]. Even susceptible plants, in which Hpa is presumably suppressing host defences, show strong activation of a set of plant genes induced by SATI during SAR. Why does this defence activation not preclude pathogen infection? Our cell biology analysis reported here resolves this paradox. We show that, during Hpa infection, the pathogen blocks PR1 induction in cells with haustoria, suggesting that the HaRxL effectors act at the transcriptional level, blocking PR1 expression (and presumably other genes of the SATI regulon), to promote virulence. Further analysis requires methods, currently under development, to expression profile specifically from infected cells. HaRxL44 is unlikely to be the sole effector that accomplishes this shift in hormonal balance that promotes biotrophy. Indeed, other nuclear-HaRxLs have been shown to interact with the Mediator complex as well as with other regulators of JA/ET pathway, like JAZ proteins [17]. Functional analyses of these effectors should facilitate the discovery of new components of nuclear immunity and the engineering of improvements to plant defences, to strengthen disease resistance in crops.
To generate HaRxL44 constructs, primers were designed from the Hpa Emoy2 genome version 8.3. HaRxL44 was amplified from the signal peptide cleavage site (ΔSP-HaRxL44) until the stop codon using genomic DNA extracted from Hpa Emoy2 conidiospores, proof reading polymerase (Accuprime Pfx, Invitrogen), and standard PCR conditions. The HA tag sequence was added to the Fw primer (CACCATGTATCCGTACGACGTACCAGACTAC GCAATTGAAGTTGTCCCC) in order to create an HA-HaRxL44–tagged version. The PCR fragment was inserted into the pENTR-D-TOPO and then in the plant expression vectors pK7WGF2, dP2 [92], and pBAV150 using Gateway Technology (Invitrogen). The constructs were sequenced by The Genome Analysis Centre (Norwich, UK) and transformed into Agrobacterium tumefaciens strains GV3101 and GV3103.
For the prediction of HaRxL44 nucleolar localisation signal, NoD [93] was used http://www.compbio.dundee.ac.uk/www-nod/index.jsp. HaRxL44M NAAIRS mutant was generated by overlapping PCR using the primers Fw AATGCTGCTATA CGATCGAAACACAAGAGG and Rev CGATCGTATAGCAGCATTCTTGTGCCAGCC.
MED19a (AT5G12230) was amplified from Arabidopsis Col-0 cDNA obtained from flowers using the primers: MED19a F1-CACCATGGAGCCTGAACGTTTAAA and MED19a R1-TTAGCCAGCAACCCTTATTGCACC. BOI was amplified from Arabidopsis Col-0 genomic DNA using the primers F2-CACCATGGCTGTTCAAGCTCATC ACATGAACATTTTC and R2-TCAAGAAGACATGTTAACATGCACACTAGCGTTCA TGACCATATCGC and MBR1-like (At1G17970) using the primers F3-CACCATGTCTTCTACAACAATCGGCGAGCACATCAG and R3-TTAAGGCTTGCC ATATGCTGCCTTCTTACAGACCG. The PCR fragment was inserted into the pENTR-D-TOPO and then in the plant expression vectors pK7WGF2 and pH7WGR2 using Gateway Technology (Invitrogen). The constructs were sequenced by The Genome Analysis Centre (Norwich, UK) and transformed into A. tumefaciens strain GV3103.
To isolate homozygous med19a-1/med19a-1 and med19a-2/med19a-2 plants, we could not analyse the segregation of the kanamycin marker carried by the T-DNA on progenies because of the loss of kanamycin resistance in these SALK lines (SALK_037435.47.85x and SALK_034955.56.00x). For mbr1-like mutant, we use a homozygous line from the SALK named SALK_025248.37.45.x. T-DNA insertions were checked by PCR genotyping using T-DNA left border and gene-specific primers designed by the Salk Institute Genomic Analysis Laboratory (SIGnAL) (http://signal.salk.edu/tdnaprimers.2.html) using default conditions. Homozygote lines were identified.
For protein extraction, frozen plant tissues were ground and mixed with an equal volume of cold protein isolation buffer [20 mM Tris-HCl (pH 7.5), 1 mM EDTA (pH 8.0), 5 mM DTT, 150 mM NaCl, 0.1% SDS, 10% glycerol, 1× Protease Inhibitor Cocktail (Sigma)]. The mixture was spun down, and the supernatant was transferred to a new tube and boiled in 5× SDS loading buffer [300 mm Tris-HCl (pH 6.8), 8.7% SDS, 5% β-mercaptoethanol, 30% glycerol and 0.12 mg/ml bromophenol blue].
For co-immunoprecipitation experiment, frozen leaf samples were ground in liquid nitrogen. The resulting powder was transferred into prechilled SM-24 20 mL centrifuge tubes containing chilled extraction buffer (4–10 mL) [1 M Tris HCl pH 7.5, 5 M NaCl, 0.5 M EDTA, 20% glycerol, 10 mM DTT, 1× Protease inhibitor (Sigma), 20% Triton X-100, 2% PVPP]. Tubes were vortexed and equilibrated before centrifugation 20 min at 20,000 rpm at 5°C. After centrifugation, supernatants were filtered to remove plant debris (Biorad Poly-Prep Chromatography columns). Proteins were quantified by Bradford assay. Three micrograms of total protein extracts were used for co-immunoprecipitation in protein Lo-Bind safe-lock tubes (Eppendorf) in which 25 µL of slurry solution of GFP beads (Chromotek) were added. Tubes were incubated on a rolling wheel for 2 to 4 h at 5°C. After incubation beads were washed with extraction buffer without PVPP by repeated low-speed centrifugations (up to four washes). Beads were resuspended in 5× SDS loading buffer prior to flash-freezing in liquid nitrogen.
Proteins were separated by SDS-PAGE, electro-blotted onto PVDF membrane (Biorad), and probed with horseradish peroxidase-conjugated anti-RFP (Abcam) or anti-GFP (Roche) antibody. MED6 and MED7 primary antibodies (from Bjorklund's lab) were used at 1∶1,000. Bands were visualized by chemiluminescence using Pico/Femto (Thermo Scientific).
For Hpa infection, 10-d-old plants were spray-inoculated to saturation with a spore suspension of 5.104 spores/ml. Plants were kept in a growth cabinet at 16°C for 3 d with a 16 h photoperiod. To evaluate conidiospore production, 10 pools of 2 plants were harvested in 1 ml of water for each line. After vortexing, the amount of liberated spores was determined with a haemocytometer as described by [94]. Statistical analyses have been performed from three independent experiments, using ANOVA.
For B. cinerea infection, spores from the fungus strain B05.10 were obtained from Dr. Henk-Jan Schoonbeek (John Innes Centre, Norwich, UK). Inoculation of Arabidopsis with B. cinerea spores was performed as described previously [95]. Briefly, 5-wk-old plants were inoculated with a suspension of 2.5×105 spores/mL in quarter-strength potato dextrose broth (6 g/L). Five-microliter droplets of spore suspension were deposited on six leaves per plant, with eight to 12 plants per experiment, and lesion diameters were measured at 3 d postinfection.
Pst infection was performed as described by [96]. Briefly, Arabidopsis plants were sprayed with bacterial suspensions carrying the EDV construct generated by Fabro et al. (2011) [13] (supplemented with 0.05% Silwet L-77). Plants were then covered with a transparent lid for 48 h. Infected leaf samples were collected at 4 DAI, ground in sterile 10 mM MgCl2, serially diluted, and spotted on NYG or low-salt LB (Luria-Bertani) agar medium containing appropriate antibiotics. Numbers of colonies were counted after 2 d of incubation at 28°C.
For SATI assay, 5-wk-old Arabidopsis plant were used. Leave disks were equilibrated in water in the dark overnight, and the solution was changed for 200 µM SA (Sigma) in the morning. After 8 h of incubation with SA or mock, leaf disks were quickly dried and flash-frozen in liquid nitrogen. About 20 leaf disks per condition were used for RNA extraction.
For transient assay analysis, A. tumefaciens strains GV3101 and GV3103 were used to deliver respective transgenes in N. benthamiana leaves, using methods previously described [97]. Protein stability was assessed using Western blot, as described by [98]. For stable expression in planta of selected candidates, Arabidopsis WT (Col-0) plants were transformed using the dipping method [99]. Briefly, flowering Arabidopsis plants were dipped with A. tumefaciens carrying a plasmid of interest, and the seeds were harvested to select the T1 transformants on selective GM media. T1 plants were checked for expression of the construct of interest either by fluorescence microscopy and/or by Western blot analysis. T2 seeds were sown on selective GM media, and the proportion of resistant versus susceptible plants was counted in order to identify lines with single T-DNA insertion. Transformed plants were transferred to soil and seeds collected. For each construct, three independent transformed plants were analyzed. T3 homozygotes plants were used for in vivo confocal microscopy and pathotests.
Frozen plant tissues were ground to a fine powder in liquid nitrogen using a precooled pestle and mortar. The powder was immediately transferred to a 1.5 ml tube and rapidly frozen in liquid nitrogen. Batches of 12 samples were thawed on ice, and 1 ml Tri-Reagent (Sigma) was added to the tubes and incubated at room temperature for 10 min. The solution was centrifuged for 20 min at 12,000× g, and the supernatant was transferred to a clean tube containing an equal volume of isopropanol. The tube was incubated overnight at −20°C and centrifuged for 10 min at 12,000× g, 4°C. Pellets were washed with 70% ethanol, air dried, and resuspended in RNase-free water. The yield and integrity of the RNAs were assessed by measuring the optical density at 260 nm and 280 nm Micro-Volume UV-Vis Spectrophotometer for Nucleic Acid and Protein Quantitation (Nanodrop, Thermo Scientific, UK) and agarose gel.
Five micrograms of total RNAs were used for generating cDNAs in a 20 µl volume reaction according to Invitrogen Superscript II Reverse Transcriptase protocol. The obtained cDNAs were diluted five times, and 5 µl were used for 10 µl qPCR reaction, and 10 µl were used for 20 µl PCR reaction.
qPCR was performed in 20 µl final volume using 10 µl SYBR Green mix (Sigma), 10 µl diluted cDNAs, and primers. qPCR was run on the CFX96 Real-Time System C1000 thermal cycler (Biorad) using the following program: (1) 95°C, 4 min; (2) [95°C, 10 s, then 62°C, 15 s, then 72°C, 30 s]×40, 72°C, 10 min followed by a temperature gradient from 65°C to 95°C, and then 72°C, 10 min. The relative expression values were determined using EF1α (At5g60390) as reference gene and the comparative cycle threshold method (2−ΔΔCt). Primers were designed using Primer3 with the default settings.
For RNA sequencing, total RNAs were extracted using TRI reagent (Sigma) and 1-bromo-3-chloropropane (Sigma) according to the procedure of the manufacturer. RNAs were precipitated with half volume of isopropanol and half volume of high salt precipitation buffer (0.8 M sodium citrate and 1.2 M sodium chloride). RNA samples were treated with DNaseI (Roche) and purified by RNeasy Mini Kit (Qiagen) according to the procedure of the manufacturers.
RNA sequencing was performed as described by [100]. Briefly, total RNAs (3 µg) were used to generate first strand cDNAs using an oligo(dT) primer comprising P7 sequence of Illumina flow cells. Double-strand cDNAs were synthesised as described previously [101]. Purified cDNAs were subjected to Covaris shearing (parameters: intensity, 5; duty cycle, 20%; cycles/burst, 200; duration, 90 s). End repairing and A-tailing of sheared cDNAs were carried out as described by Illumina. Y-shaped adapters were ligated to A-tailed DNA and subjected to size selection on agarose gel. The gel-extracted library was PCR enriched and quantified using qPCR with previously sequenced similar size range Illumina libraries. The libraries were sequenced on Illumina Genome Analyzer II.
Illumina libraries were quality-filtered using FASTX Toolkit 0.0.13 with parameters −q20 and −p50 (http://hannonlab.cshl.edu/fastx_toolkit/index.html). Reads containing “N” were discarded, and read qualities were converted from Illumina fastq to Sanger fastq format. The libraries were separated using perfect match to the barcode. The sub-library was artefact-filtered using FASTX-toolkit. Quality-filtered libraries were aligned to the Arabidopsis Col-0 genome sequence (TAIR10) using Bowtie version 0.12.8 [102] and reads with up to 10 reportable alignments were selected. Unaligned reads from previous steps were used to align to transcript sequences of Arabidopsis Col-0 (ftp://ftp.Arabidopsis.org/home/tair/Sequences/blast_datasets/TAIR10_blastsets/TAIR10_cdna_20101214_updated) using Bowtie version 0.12.8. Linking of each sequenced read (Tag) to gene was carried out using the following considerations: reads aligning to each gene limits were assigned to that gene; reads aligning to genes with overlapping gene limits were split equally between them; and reads aligning to more than 10 genes were discarded. Differential expression analysis was performed using the R statistical language version 2.11.1 with the Bioconductor [103] package, edgeR version 1.6.15 [104] with the exact negative binomial test using tagwise dispersions.
For co-localisation assays in N. benthamiana, cut leaf patches were mounted in water and analysed on a Leica DM6000B/TCS SP5 confocal microscope (Leica Microsystems) with the following excitation wavelengths: GFP, 488 nm; YFP, 488 nm; RFP, 561 nm. For in vivo localisation in Arabidopsis, 10-d-old Hpa-infected seedlings were mounted in water and analysed on a Leica DM6000B/TCS SP5 confocal microscope (Leica Microsystems) with the following excitation wavelengths: CFP, 458 nm; GFP, 488 nm; RFP, 561 nm.
GUS activity was assayed histochemically with 5-bromo-4-chloro-3-indolyl-β-d-glucuronic acid (1 mg/ml) in a buffer containing 100 mM Sodium Phosphate pH 7, 0.5 mM Potassium Ferrocyanide, 0.5 mM Potassium Ferricyanide, 10 mM EDTA, 0.1% Triton. Arabidopsis leaves were vacuum-infiltrated with staining solution and then incubated overnight at 37°C in the dark. Destaining was performed in 100% ethanol followed by incubation in chloral hydrate solution. Sections were observed with a Zeiss Axioplan 2 microscope (Jena, Germany).
Aniline blue staining was used to stain callose structures in plant tissues [105], which appeared after infection, like ring or encasements of Hpa haustoria, or like dots after Pseudomonas infection or PAMP treatment. Samples (either Hpa-infected seedlings or leaf disks punctured from PAMP/Pseudomonas-infiltrated leaves) were cleared in 100% methanol, washed in water, and then stained with aniline blue (0.05% w/v in 50 mM phosphate buffer pH 8) overnight. Samples were observed with a Leica DM6000B/TCS SP5 confocal microscope (Leica Microsystems).
|
10.1371/journal.pgen.1007608 | A novel gene-diet pair modulates C. elegans aging | Diet profoundly affects metabolism and incidences of age-related diseases. Animals adapt their physiology to different food-types, modulating complex life-history traits like aging. The molecular mechanisms linking adaptive capacity to diet with aging are less known. We identify FLR-4 kinase as a novel modulator of aging in C. elegans, depending on bacterial diet. FLR-4 functions to prevent differential activation of the p38MAPK pathway in response to diverse food-types, thereby maintaining normal life span. In a kinase-dead flr-4 mutant, E. coli HT115 (K12 strain), but not the standard diet OP50 (B strain), is able to activate p38MAPK, elevate expression of cytoprotective genes through the nuclear hormone receptor NHR-8 and enhance life span. Interestingly, flr-4 and dietary restriction utilize similar pathways for longevity assurance, suggesting cross-talks between cellular modules that respond to diet quality and quantity. Together, our study discovers a new C. elegans gene-diet pair that controls the plasticity of aging.
| For animals living in the wild, being able to utilize a wide range of diet is evolutionarily advantageous as they can survive even when their optimal diet is depleted. Since diet is known to influence the rate of aging, animals seem to have evolved intricate mechanisms to maintain homeostasis and normal life span, but the molecular mechanisms are less understood. Using a small nematode, C. elegans as a model, we show that the adaptive capacity to different diet is maintained by a kinase gene. When this gene is mutated, worms start living longer on one strain of bacterial diet but not on the other. We identify the molecular cascade required for this food-type-dependent longevity. We show that this cascade of events significantly overlaps with the pathway that determine food quantity-dependent life span enhancement. Our study thus elucidates a part of the molecular monitoring system that regulates longevity dependent on the available quality and quantity of diet.
| Animals dwell in a complex ecosystem where they interact with a host of other organisms; some of them may alter their life history traits. For example, the nematode Caenorhabditis elegans is found in decaying moist vegetation that is co-inhabited by different types of bacteria that they feed on. So, the worms are often exposed to various pathogenic bacteria that they either avoid or use a conserved innate immunity pathway to counter. The worms are also presented with a range of bacteria of different nutritional values that they choose between [1–3]. They encounter E. coli¸ Bacillus and Comamonas etc. that are known to modulate development, reproduction, fat storage and life span [2, 4–7]. In the laboratory, worms are mostly maintained on E. coli OP50 but are often exposed to the RNAseIII-deficient HT115 during RNAi experiments. The two strains differ considerably, particularly in terms of carbohydrate content, with the OP50 strain considered as a low quality diet that induces less satiety and promotes higher fat storage [3, 6–9]. Since the rate of aging is greatly influenced by dietary composition, worms seem to have evolved intricate adaptive strategies to maintain normal aging [10, 11]. As a result, although the two diets differentially affect metabolism, the worms are able to ensure relatively normal life span when fed either bacteria [7]. How C. elegans sense different diet to alter metabolism and life history traits, including complex traits like aging, is an emerging area of research. These studies are being facilitated by the discovery of gene-diet pairs where the function of a gene becomes discernible only on a particular diet [12].
Previous work has shown that sensory neurons may process signals that differentiate between different food-types and regulate life span in flies and worms [10, 13, 14]. In C. elegans, the neuromedin U receptor-like gene, nmur-1 mutant as well as osm-3 (kinesin motor protein required for cilia formation) mutant has extended life span on OP50 but not on HT115 [10, 14]. These life spans were found to be dependent on the FOXO transcription factor, DAF-16 [14]. On the other hand, the proline metabolism gene alh-6 works in the muscle to preserve mitochondrial structure and functional homeostasis in response to OP50 [11]. This requires a functional NMUR-1 receptor signalling [11]. Interestingly and intuitively, it would appear that the intestine may also contribute to this phenomenon as it gets to sample different food that the worms ingest. However, the role of intestine in food-type-dependent life span regulation is not as well-characterized.
Here we show that a serine-threonine kinase gene, flr-4 regulates food-type-dependent life span by functioning both in the neurons and the intestine. We find that when flr-4 is knocked down by RNAi or its function disrupted by a P223S missense mutation in its kinase domain, life span is dramatically increased. The life span of the kinase-dead mutant flr-4(n2259) is increased only when the mutant is fed HT115 and not OP50. We show that knocking down flr-4 leads to increased cytoprotective xenobiotic detoxification pathway (XDP) gene expression, through the nuclear hormone receptor NHR-8, that plays a causal role in its increased life span. Interestingly, this elevated gene expression as well as the increased life span is dependent on the conserved p38 MAPK signalling. In flr-4(n2259), OP50 is unable to strongly activate the p38 MAPK while HT115 leads to increased phosphorylation of the MAPK. Consequently, in the mutant, HT115 is able to increase the levels of the XDP genes while OP50 does not. Finally, we demonstrate that FLR-4 uses a pathway similar to DR to ensure longevity, dependent on FOXA/PHA-4 but independent of FOXO/DAF-16. Together, our study establishes flr-4 as a new longevity gene that controls adaptive capacity of C. elegans towards bacterial diet by preventing differential activation of XDP genes through p38MAPK pathway, dependent on food-type.
Flr-4 was originally identified in a screen for genes that regulate fluoride resistance and the mutants exhibit temperature-sensitive defecation defects [15]. However, at 20 oC they do not have defects in defecation [15]. We initially became interested in the flr-4 gene as it has 26% homology and 40% identity to drl-1 [16], a gene we have recently characterized to be involved in Dietary Restriction (DR). Knocking down flr-4 using a cDNA RNAi construct increased life span dramatically (average life span increase 40–60%, Fig 1A, S1 and S2 Tables). Similar life span extension was observed in absence of FUDR, a DNA synthesis inhibitor used to arrest confounding effects of progeny population during life span analysis (S1A Fig). The increased life span was also associated with better health as evident from lower lipofuscin pigment accumulation (S2A Fig), lesser muscular atrophy with age (S2B Fig) and consequently, better motility (S2C Fig). These worms were smaller in size (S2D Fig) but did not have any major defects in developmental rates (S3 Fig). However, the increase in life span was not associated with enhanced tolerance towards heat stress (S4B Fig); UV stress tolerance was only increased 13–15% compared to 40–60% increase in life span (S4A Fig). The increased life span of flr-4 knock down was also not dependent on the heat shock transcription factor hsf-1 (S4C and S4D Fig). Thus, flr-4 seems to decouple longevity from stress resistance and is a novel longevity modulator.
Next, we asked whether an flr-4 mutant has attributes similar to the flr-4 RNAi. We used the flr-4(n2259) allele that has a P223S missense mutation in the activation loop of the protein kinase domain and shows increased fluoride resistance [15]. This allele has similar developmental rates as wild-type (S5 Fig) and shows no dauer arrest [15]. We found that flr-4(n2259) increased life span that was not further increased when grown on flr-4 RNAi (Fig 1B, S1 and S2 Tables). The life span of the mutant was also not affected by the presence of FUDR (S1B Fig). This suggests that flr-4(n2259) behaves as a null allele to regulate life span.
C. elegans is known to respond differentially to bacterial diet to modulate life history traits, including life span [5–7, 14, 17]. Interestingly, we found that the life span extension in flr-4(n2259) is dependent on the food-type. When the mutant was grown on the E. coli HT115 (a K12 strain), life span was dramatically extended (Fig 1C, S1 and S2 Tables). In contrast, when grown on E. coli OP50 (a B strain), no extension of life span was observed (Fig 1C). We checked for differences in pumping under these conditions and found no change (S2E Fig). Also, ingestion of RFP-labelled beads was similar in WT and flr-4(n2259) (S2F Fig), suggesting that the differences in life span are not due to altered feeding behaviour. Furthermore, the developmental rate was similar in both the strains when grown on the two E. coli strains (S5 Fig). Remarkably, knocking down flr-4 using an OP50-based RNAi system [18] increased life span similar to that of the HT115-based system (Fig 1D), suggesting that the food-type-dependent life span extension may be attributed to the kinase function of FLR-4. This conclusion is further supported by the observation that rescuing flr-4(n2259) with a kinase-proficient wild-type transgene suppresses the increased life span on HT115 (S16 Fig). Together, FLR-4 kinase suppresses pro-longevity cues in a food-type dependent manner.
Next, we asked where FLR-4 functions to regulate longevity. We constructed a flr-4p::gfp transgenic line and found that the flr-4 promoter drove expression of gfp in the intestine and a few neurons (Fig 2A). We used a tissue-specific RNAi system to determine the tissue where flr-4 functions. We found that flr-4 knockdown specifically in the intestine or the neurons was sufficient for life span extension; no extension in life span was observed when the gene is solely knocked down in muscle or hypodermis (Fig 2B–2E, S1 and S2 Tables). Together, flr-4 functions in the intestine and neurons to negatively regulate life span of the worms.
In C. elegans, longevity genes need to be knocked down at temporally distinct points in development to increase life span. For example, the mitochondrial electron transport gene cco-1 or the MEKK-3-like kinase drl-1 needs to be knocked down early in development to increase life span while insulin-like signalling pathway functions in adulthood to exhibit the beneficial longevity effects [16, 19, 20]. We initiated flr-4 RNAi at different stages of development of the worms and found that knocking down at L1 or L2 produced maximum life span extension; knocking down at L3 or later had diminished or no effect (S6A–S6E Fig). Together, flr-4 functions in the intestine and neurons, during larval development to regulate adult life span.
Since FLR-4 is required to suppress the effect of food-type on longevity, we asked what signalling cascade may be mediating this effect. The p38 MAPK pathway is a central signalling mediator required for mounting the innate immune response when worms are challenged with pathogens [21–25]. The worm p38 homolog PMK-1 is activated by its upstream MAPKKK NSY-1 and MAPKK SEK-1 [26, 27]. The TIR domain adaptor protein TIR-1, an ortholog of human SARM [28, 29] and UNC-43, a Ca2+/calmodulin-dependent protein kinase II (CaMKII) [30] lie further upstream of these serine-threonine kinases. While TIR-1 works upstream of the p38 MAPK pathway to regulate innate immunity genes [28, 29], both TIR-1 and UNC-43 functions in the neurons to control neuronal cell fate and asymmetric patterns of odorant receptor expression [30, 31]. We grew wild-type, pmk-1(km25), sek-1(km4), nsy-1(ag3), nsy-1(ok593), tir-1(tm3036) and unc-43(e403) on control or flr-4 RNAi and performed life span analysis. We found that increased life span on flr-4 RNAi is suppressed when any of the kinases in the p38 MAPK pathway or tir-1 or unc-43 is mutated (Fig 3A, S1 and S2 Tables). Interestingly, pmk-3 is not required for the life span extension, showing specificity of the process (S1 and S2 Tables). Similar suppression of life span was observed when flr-4(n2259) was grown on sek-1 RNAi (S7A Fig). We also created a flr-4(n2259):sek-1(km4) double mutant and found that the life span was suppressed (S7B Fig). Further, in order to determine biochemically whether knocking down flr-4 activates the p38 MAPK, we performed western blot analysis using a phospo-PMK-1-specific antibody (Figs 3B and S14). We found that in WT worms when flr-4 is knocked down, PMK-1 phosphorylation increases in a sek-1-dependent manner. Further, TIR-1 or UNC-43 seems to be working in the same linear pathway as flr-4; in the tir-1(3036) and unc-43(e408), flr-4 knockdown failed to increase phosphorylation of PMK-1 (S7C Fig). Together, the FLR-4 longevity signals are mediated by the p38 MAPK pathway.
In order to understand how flr-4 knockdown increases life span, we performed transcriptomic analysis of wild-type worms grown on control or flr-4 RNAi. We found that 1957 genes were upregulated (> 2 folds, P ≤ 0.05) while 538 genes were down-regulated. We determined the biological functions of these genes using DAVID [32] and show that they are significantly enriched for genes involved in the Xenobiotic Detoxification Pathway (XDP) (Figs 4A and S8A). All these cytoprotective XDP genes are expressed in the intestine of the worms (www.wormbase.org). Using quantitative reverse transcriptase PCR (qRT-PCR), we verified 13 genes that were upregulated in our RNA-seq experiment (Figs 4B and S8B). We also used a transgenic worm expressing gfp driven by the cyp-35B1 promoter and show that the expression of GFP is enhanced in the hind-gut region, when the worms were grown on flr-4 RNAi (S8C Fig).
Next, in order to determine whether p38 MAPK pathway has any role in the regulation of these cytoprotective genes, we performed qRT-PCR analysis of these genes after knocking down flr-4 in sek-1(km4). Interestingly, majority of the XDP genes that were upregulated in the wild-type failed to do so in sek-1(km4) (Figs 4C and S8D). Since the levels of some of the genes do not fall back to basal level, it is possible that other signalling pathways or transcriptional regulators are also involved. These experiments suggested that p38 MAPK pathway regulates XDP genes downstream of flr-4.
In order to determine whether XDP genes are indeed required for life span extension brought about by flr-4 knockdown, we used a mutant of nhr-8, a transcription factor required for XDP gene expression [16, 33]. First, we knocked down flr-4 by RNAi in nhr-8(ok186) and found that it failed to increase life span to the same extent as in wild-type (Fig 4D). We also knocked down nhr-8 using RNAi in flr-4(n2259), and found that the life span of the mutant is significantly suppressed (S8E Fig). Finally, we show that in nhr-8(ok186), the XDP genes fail to upregulate to the same extent as in WT, when flr-4 is knocked down (S8F Fig). Thus, increased expression of XDP genes is required for life span extension by flr-4 knockdown.
Next, we asked whether NHR-8 functions downstream of p38 MAPK pathway to regulate XDP genes. For this, we decoupled p38 MAPK from flr-4 and activated it by knocking down the phosphatase VHP-1 using RNAi [34, 35]. Knocking down vhp-1 led to upregulation of cyp-35B1, as measured by increased GFP expression in the cyp-35B1p::gfp transgenic line. This enhancement was suppressed in the cyp-35B1p::gfp;nhr-8(ok1853) worms, showing that NHR-8 functions downstream of p38MAPK (Fig 4E). QRT-PCR analysis showed that three other XDP genes that are upregulated on vhp-1 knockdown are dependent of NHR-8 (Fig 4F). Interestingly, knocking down vhp-1 was not sufficient to increase life span of WT (S8G Fig), suggesting that additional downstream events may have to be coactivated in order to get life span benefits similar to flr-4 knockdown.
Since flr-4(n2259) shows differential response to OP50 and HT115 to extend life span, we suspected that this may be due to the ability of a diet to upregulate a specific set of genes. So, we performed transcriptomic analysis of wild-type and the mutant on the two bacterial diets. We found that the XDP genes are upregulated only when flr-4(n2259) was grown on HT115 but not when grown on OP50 (Fig 5A). We verified several of these genes using qRT-PCR and found them to be upregulated only on HT115 (Figs 5B and S9A). Additionally, we used the flr-4(n2259);cyp-35B1p::gfp strain to show that the expression of GFP is induced only when the worms are grown on HT115 (Fig 5C). Together, this shows that flr-4 mutant worms mount a specific p38-dependent transcriptional response when fed HT115 that provide cytoprotective benefits leading to enhanced life span.
Next, we asked whether only HT115 can differentially activate the p38 MAPK pathway. For this, we performed western blot analysis with WT and flr-4(n2259) grown on HT115 or OP50. Interestingly, we found that flr-4(n2259) grown on HT115 showed enhanced phosphorylation of PMK-1 compared to wild-type (Figs 5D, S9B and S15). However, the levels of phosphorylation were unchanged in WT maintained on the two bacteria. Thus, FLR-4 prevents differential activation of p38 MAPK dependent on diet, maintaining adaptive capacity in C. elegans.
Since flr-4 mutant worms responded differentially to diet, we evaluated the interaction of the gene with two nutrient sensing pathways. First, using RNAi we knocked down flr-4 in the IIS pathway mutant daf-2(e1370) and found that the life span of the mutant is further extended, suggesting independent mechanisms (Fig 6A, S1 and S2 Tables). On the other hand, the life span of eat-2(ad1116) is not further extended; in fact, the life span was suppressed by 10–14% (Fig 6B). However, flr-4 knockdown does not affect pharyngeal pumping of eat-2(ad1116), similar to wild-type (S13 Fig), showing that the lack of additive effect on life span with eat-2(ad1116) is not mechanical. In case of another genetic mimic of DR [16], the extended life span of drl-1 RNAi worms was also not extended further by flr-4 mutation (S10A Fig). This suggested that flr-4 uses cellular signalling pathways utilized by DR, but not the IIS pathway to ensure longevity. This was further supported by the fact that life span of flr-4 RNAi, as in eat-2 mutants and on drl-1 knockdown, was dependent only on the FOXA transcription factor PHA-4, and not the FOXO factor DAF-16 that is required by IIS pathway mutants (Figs 6C, 6D and S10B). In the pha-4(zu225), life span extension by flr-4 knockdown was completely abrogated. Interestingly, the life span of the flr-4(n2259) or flr-4 RNAi worms was independent of the NRF2 ortholog, SKN-1, a common output of insulin-like signalling and DR [36, 37](S10C and S10D Fig); skn-1 abrogation by mutation or RNAi affects the life span of WT and flr-4 knockdown worms to similar extent. Further, similar to DR [16, 38], the flr-4 RNAi did not further extend the already long life span of germline defective mutants (S11 Fig). Like many long lived mutants, flr-4 mutants have delayed reproductive span and lower brood size compared to WT, mainly on HT115 (S12 Fig). This may be due to more resource allocation towards somatic maintenance during DR [39] and may be caused by lower germ cell proliferation as seen in case of insulin-IGF-1 signalling pathway mutants [40]. The fact that FLR-4 utilizes the DR machinery for longevity assurance is also consistent with its role in ensuring adaptive capacity to diet.
A complex interaction between genes and diet determines the rate of aging and predisposes an individual to age-related diseases. The term gene-diet pair is used when the consequences of mutating a gene is visible only on a specific diet [12]. Surprisingly, only a few gene-diet pair have been identified that regulates aging, mainly through studies in C. elegans [12]. In this report, we identify a novel gene-diet pair and show that the adaptive capacity to different food-type is modulated by the protein kinase FLR-4. This protein prevents differential activation of the p38 MAPK pathway dependent on the food-type and consequently, the expression of cytoprotective genes by transcription factor NHR-8 (Fig 6E). Interestingly, this pathway overlaps with the DR pathway, suggesting a cross-talk between cellular signalling that senses food quality and quantity to regulate life history traits like aging.
FLR-4 is a serine-threonine protein kinase similar to mammalian Cyclin-dependent protein kinase 3 (31% identity, 50% similarity, E value 8e-22). It was initially identified in a screen for genes involved in fluoride tolerance [41], but was subsequently shown to have defects in ultradian rhythm in the intestine that controls defecation [15]. In this study, we elucidate a novel function for FLR-4 in the intestine and neurons that is independent of its role in defection. The temperature-sensitive kinase dead mutant flr-4(n2259) has normal defecation cycle at 20 oC [15], a temperature at which most of our assays were performed. In view of the central role that this kinase plays in controlling multiple important phenotypes, future research needs to be directed towards finding its immediate cellular targets. This is particularly important as we found that the food-type-dependence is specific to the kinase dead mutant; RNAi knockdown using an OP50-based system also increased life span. This suggests that the kinase-dead mutant may not be able to phosphorylate a substrate(s) that is required to maintain life span homeostasis on the different bacterial diets.
Considering the importance of gene-diet pairs in aging and disease, our understanding of the mechanisms of adaptive capacity to food-type is still in its infancy. Previous studies have identified a few genes that play a role in this process. Notably among them are the RICTOR ortholog rict-1 [6], neuromedin U receptor ortholog nmur-1 [14] and mitochondrial 1-pyrroline-5-carboxylate dehydrogenase (P5CDH) alh-6 [11]. The rict-1 mutants have phenotypes similar to flr-4 mutant worms such that they have shorter life span on OP50 while exhibiting life span extension on HT115 [6]. However, in contrast to the flr-4, rict-1 regulates feeding behaviour when an animal encounters diets of different qualities. The rict-1 mutants have different pumping rates and exhibit avoidance behaviour on food of diverse quality [6, 12]. The alh-6 mutants, on the other hand, show accelerated aging when fed OP50 while retaining normal rates of aging on HT115 [11]. Interestingly, the nmur-1 mutants have longer life span on OP50 but not on HT115 [14]. Thus, these gene-diet pairs seem to control diverse aspects of an animal’s response to different food. In future, it will be interesting to study the interaction of these genes with flr-4, considering the fact that opposing phenotypes controlled by these genes may indicate homeostatic control of life span in response to different diet.
As the above mutants differ in their response to different diet, they may also activate diverse signalling cascades. This is apparent from the fact that rict-1 and nmur-1 interact differentially with the downstream components of the insulin signalling pathway [6, 14, 42]. While the life span extension of rict-1 knockdown is dependent on NRF2 ortholog skn-1, nmur-1 mutants require the FOXO transcription factor DAF-16. We found that the flr-4 mutants require the FOXA transcription factor PHA-4 for life span extension and is independent of DAF-16; its transcriptional response may thus be different from other gene-diet pairs. Since both alh-6 and flr-4 may signal through a pathway used by the eat-2 model of DR, it will be interesting to study the transcription factor requirements of the former. Comparative gene expression profiles of these mutants on different diet will help us understand the complex gene expression modalities controlled by these gene-diet pairs.
Gustatory and olfactory neurons that perceive chemical signals have previously been shown to affect life span [10, 14, 43]. Here we show that flr-4 knockdown in the neurons can also increase life span. In fact, the life span extension by flr-4 RNAi requires the CAMKII ortholog UNC-43 and SARM ortholog TIR-1 that is known to act in the neurons to determine cell fate. On the other hand, intestine-specific knockdown of flr-4 also increases life span. Although it appears intuitive to suggest that the gut may play an important role in sensing different diet that is ingested, the role of the intestine in food-type-dependent life span extension is less known. Interestingly, TORC2 that also regulates food-type-dependent life span, requires SKN-1/NRF2 in the intestine to regulate life span [42]. However, FLR-4 life span is independent on skn-1, indicating to extensive insulation as well as cross-talks among these pathways. Future research needs to be directed to understand the partitioning of the p38 MAPK pathway in the neurons and intestine as well as their cross-talk to regulate flr-4-mediated life span.
Animals in the wild, unlike those in laboratories, are exposed to a wide variety of food that they have adapted to. Being able to utilize a wide range of diet is evolutionarily advantageous as the animals can survive when their optimal diet is depleted. Giant panda that depend mainly on bamboo for nutrition is facing extinction due to loss of habitat (wwf.panda.org)[44]. Since diet influences the rate of aging, the animals have evolved intricate mechanisms to maintain homeostasis. In addition to the quality of diet, the quantity of food regulates the plasticity of aging. As a result, DR is able to delay aging and increase life span across the animal kingdom [45, 46]. In our study as well as in that of alh-6 [11], we observe genetic overlap with DR, suggesting that organisms have evolved cellular modules that evaluate both quality and quantity of diet to regulate life span.
We show that FLR-4 signals through the p38 MAPK pathway to regulate the expression of cytoprotective genes dependent on diet. In C. elegans, this pathway has been extensively characterized for its role in mediating innate immune response towards pathogenic bacteria as well as in mounting an oxidative stress response [21, 23, 47–49]. On the other hand, the XDP genes have been shown to provide cytoprotective effects leading to enhanced life span in multiple models of longevity [16, 50–52]. The fact that FLR-4 would signal differences of diet through the p38 MAPK seems quite intuitive for an organism that feeds on bacteria and uses this same pathway to differentially activate immune genes on encountering pathogens. But in case of flr-4 knock down, immune effector genes are not upregulated, showing specificity of this module. However, in mammals, the p38 MAPK has important role in regulating metabolism in liver and adipocytes during fasting, mediated by glucagon and insulin [53]. It will be interesting to study the role of p38 MAPK pathway in gene-diet interaction networks in mammals.
How flr-4 mutants sense the differences in bacterial food remains to be answered. One possibility is that the mutants become sensitive to the presence or absence of a metabolite secreted by the bacteria and mount the specific response, whereas the WT worms are able to maintain homeostasis. Detailed metabolomics study will be able to reveal the exact nature of the molecule. One interesting observation is that genes that are upregulated in flr-4(n2259), grown specifically on HT115, are enriched in amino acid metabolism. Previous studies have shown that methionine metabolism greatly influences life span, metabolism, and stress resistance [54–56]. Vitamin B12 acts an important cofactor in methionine as well as propionic acid metabolism, maintaining optimal level of Homocysteine and propionic acid, thereby preventing toxicity [54]. It is possible that the molecule may be vitamin B12, as seen in case of Comamonas [17, 57, 58]. In line with this idea, we found that the levels of the metabolic sensor acdh-1 is much suppressed in HT115-fed flr-4(n2259) in our RNA-seq data, similar to the effect of vitamin B12 treatment. However, flr-4 mutants do not have any significant defect in development or fat storage. In future, why flr-4 mutants become sensitive to a metabolite or whether the two bacteria differ in production of soluble metabolites needs to be addressed. We also need to understand why the RNAi knockdown of the gene do not induce food-type-dependent life span response.
Together, our study discovers a new gene-diet pair that controls the plasticity of aging in C. elegans and reveals a complete signal transduction cascade involved in this process.
C. elegans strains used in this study were obtained from the Caenorhabditis Genetics Center and maintained on NGM agar plates at 20°C, unless otherwise stated, on Escherichia coli OP50 lawns. All RNAi experiments were initiated with synchronized L1 worms. Strains used in the study are: N2 Bristol as wild-type, flr-4(n2259)X, rde-1(ne219) V, rde-1(ne219) V;kzIs9, rde-1(ne219) V;kzIs20, rde-1(ne213) V;kbIs7, sid-1(pk3321) V, sid-1(pk3321) V;uIs69 V, ccIs4251 [pSAK2 (myo-3::NGFP-LacZ)], tir-1(tm3036)III, unc-43(e408)IV, nsy-1(ag3)II, nsy-1(ok593)II, sek-1(km4)X, pmk-1(km25)IV, pmk-3(ok169)IV, flr-4(n2259)X; sek-1(km4)X, nhr-8(ok186) IV, daf-2(e1370)III, bvIs5 [cyp-35B1p::GFP + gcy-7p::GFP] referred to as cyp-35B1p::gfp in this manuscript, nhr-8(ok186) IV;bvIs5, flr-4(n2259)X;bvIs5, rrf-3(pk1426)II;eat-2(ad1116)II, rrf-3(pk1426)II, daf-16(mgdf50)I, smg-1(cc546), smg-1(cc546)I;pha-4(zu225)V, skn-1(zu169) IV/nT1[unc-?(n754) let-?](IV;V), glp-1(e2141)III, gld-1(op236)I, glp-4(bn2ts)I, hsf-1(sy441).
Gravid adult worms, initially grown on E. coli OP50, were bleached and the eggs were L1 synchronized in M9 buffer for 16 hours before placing them on the respective RNAi plates (say ‘X’ gene RNAi). Once worms reached L4 stage, they were transferred to intermediate RNAi plates (seeded with the same ‘X’ gene RNAi) for 12 hours. After that, the worms were picked onto fresh ‘X’ gene RNAi plates overlaid with 5-fluorodeoxyuridine (FUDR, final concentration 0.1 mg/ml of media). For life span analysis on plates without FUDR, worms were transferred to fresh plates on alternate days till the end of the reproductive span. Life span scoring was initiated at day 7 of adulthood and continued every alternate day. For statistical analyses of survival, OASIS software (http://sbi.postech.ac.kr/oasis) was used and P-values were calculated by using a log rank (Mantel-Cox method) test.
For temporal requirement experiments, L1 synchronized worms were placed on control RNAi plates. Worms from the plates were transferred to flr-4 RNAi plates at L2, L3, L4 or YA stages. FUDR was overlaid on the plates 12 hrs after the worms reached L4.
For life span on different bacterial feed, L1 synchronized worms were placed on HT115 and OP50-seeded plates and the lifespan was initiated as mentioned above.
All life span analysis referred in the main text is provided in the S1 Table. Two independent biological replicates are provided in S2 Table.
A total of 20 worms each from control or flr-4 RNAi plates were transferred to an unseeded plate on day 2, 5 and 10 of adulthood. Each worm was gently prodded on the tail with a platinum wire and total number of body bends per 30 seconds was counted. A body bend was scored every time the area behind the pharynx reached a maximum bend in the opposite direction from the last bend counted.
The ccIs4251 [pSAK2 (myo-3::NGFP-LacZ)] worms were grown on control or flr-4 RNAi plates. On day 2, 5 and 10 of adulthood, the worms were paralyzed on 2% agarose pads in the presence of 20 mM sodium azide. Photographs of the worms were captured at 630X magnification using an AxioImager M2 microscope (Carl Zeiss, Germany) fitted with Axiocam MRm [Excitation 488nm and Absorbance at 520nm]. For each RNAi, at least 10 nuclei of 10 worms each were photographed. Morphology of each muscle nuclei was scored as ‘intact’, ‘moderately damaged’ or ‘severely damaged’. A nucleus was scored as ‘intact’ if it had intact membrane with no degradation, ‘moderately damaged’ when the nuclear membrane appeared to disintegrate but the nucleoplasm displayed no or very little dark patches and ‘severely damaged’ when the nuclei had increased nucleolar size, dark patches in the nucleoplasm, distorted appearance and membrane disintegration.
To determine lipofuscin autofluorescence, 20 worms each grown on control or flr-4 RNAi were anesthetized in 20mM Sodium Azide and mounted on 2% agarose pads on day 1, 5 and 10 of adulthood. The worms were visualized under microscope using FITC filter and images were captured using a constant exposure time (1.2 sec).
Twenty five L4 worms, grown on respective bacterial feed, were picked and placed onto NGM plates seeded with 250:1 (vol:vol) of bacteria and Fluoresbrites Multifluorescent microspheres/RFP beads (0.2 μm diameter, Polyscience Inc., USA). After 10 minutes, worms were collected and washed twice with 1X M9 buffer to remove any bead attached to the body surface. Worms were finally re-suspended in 30 μl of 1X M9 buffer and transferred to a freshly prepared 2% agarose pad slides. The images of worms were taken using AxioImager M2 microscope (Carl Zeiss, Germany). Quantification was performed using NIH ImageJ software.
An one minute video of Day 1 adult worms was taken using Axiocam MRm camera attached to M205FA microscope (Leica, Germany). The video was slowed down and pharyngeal pumping was counted for that 10 second period.
Worms were imaged one day after they reached L4 using Axiocam MRm camera attached to an AxioImager M2 microscope (Carl Zeiss, Germany). Area of the worms was quantified using NIH ImageJ software.
flr-4 RNAi: The full length cDNA sequence of flr-4 was amplified using primers listed in S3 Table and cloned into pL4440 RNAi vector.
A transcriptional fusion of the flr-4 promoter and a green fluorescent protein (GFP) gene was constructed in pPD95.75. The 3.5 kb promoter region upstream of start codon of F09B12.6 was amplified using primers listed in S3 Table and HiFidelity PCR system (Kapa Biosystems, USA) and cloned into pPD95.75 using BamHI and KpnI restriction sites. The recombinant plasmid was injected at a concentration of 5ng/μl into the syncytial gonad of wild-type worms along with 100ng/μl pRF4 (rol-6) co-injection marker using a Microinjection setup consisting of Nikon TiS inverted microscope fitted with Eppendorf Femtojet Express and Transferman NK2. Transformants were selected based on the rolling phenotype as well as the presence of GFP expression. Fluorescence images of transgenic worms were captured under AxioImager M2 microscope (Carl Zeiss, Germany) fitted with Axiocam MRm at 40X magnification [Excitation 488nm and Absorbance at 520nm].
The full length cDNA sequence of flr-4 was amplified using primers listed in S3 Table. The gfp sequence of Pflr-4::gfp plasmid was then excised using KpnI and EcoRI and replaced with the amplified flr-4 cDNA sequence, generating the Pflr-4::flr-4 cDNA construct. The Pflr-4::flr-4 cDNA construct and pRF4 were co-injected in the germline of flr-4(n2259) (concentrations: 150 ng/μL pRF4 and 5 ng/μL Pflr-4::flr-4 cDNA). Wild-type and flr-4(n2259) roller lines were generated by injecting 150 ng/μL pRF4. Lines were maintained by picking rollers.
Synchronized L1 worms grown on OP50 or RNAi plates were collected at Day 1 of adulthood in M9 buffer and washed thrice using M9 buffer. Then, Trizol was added to about 4 times the volume of the worm pellet and the worms lysed using two freeze thaw cycles, followed by vigorous vortexing. RNA was purified by phenol:chloroform:isoamylalcohol extraction and isopropanol precipitation. For quantitative Reverse Transcriptase PCR (qRT-PCR) experiments, the concentrations of the RNA were determined using NanoDrop 2000 (Thermo Scientific, USA) and the quality of the ribosomal 28 S and 18 S on denaturing agarose gel was used for evaluation of RNA integrity. For transcriptomic analysis, the quality was evaluated using Bioanalyzer (Agilent, USA) and only RNA with RIN number above 9 was used for RNA-seq.
About 2.5 μg of RNA was converted to cDNA using Superscript III Reverse Transcriptase enzyme and poly-T primers (Invitrogen, USA). QRT-PCR analysis was performed using the DyNAmo Flash SYBR Green mastermix (Thermo Scientific, USA) and Realplex PCR system (Eppendorf, USA) to determine the relative gene expression levels. Statistical analysis was performed using GraphPad 7.0. All the primers used are listed in S3 Table.
RNA-Sequencing (RNA-seq) libraries of WT grown on Control RNAi or flr-4 RNAi, and WT or flr-4(n2259) grown on HT115 or OP50 at Day 1 adulthood were prepared as recommended by the Illumina TruSeq RNA Sample Preparation kit using Low-Throughput (LT) Protocol (Illumina, Inc., USA). Sequencing of libraries was performed using Illumina GAIIX for 78 cycles including 6 additional cycles for index read. Sequence reads were aligned using CLC Genomics Workbench 6.5.1 with default setting against C. elegans genome assembly (WS231). Unpaired group comparisons, based on RPKM (Reads per Kilobase per Million mapped reads), were chosen as expression values for comparing the samples. A fold change ±2.0 and P value ≤0.05 (Kal's Z test) were used to filter the differentially expressed genes. GO‐term enrichment analysis was performed using the DAVID Bioinformatics Database [32]. The sequencing data is available as BioProject ID: PRJNA362992.
Synchronized L1 worms, grown on OP50 or HT115-seeded plates, were collected at Day 1 of adulthood in 1xM9 buffer and washed thrice using the same buffer. The pellet was freeze-thawed 3 times in a protein extraction buffer (20 mM Hepes buffer pH 7.9, 25% glycerol, 0.42 mM NaCl, 1.5 mM MgCl2 hexahydrate, 0.2 mM EDTA dihydrate, 0.5 mM DTT) in presence of a protease inhibitor cocktail (Sigma, USA), sonicated in a waterbath-based sonicator (Diagenode, USA) and centrifuged at 10,000 rpm for 10 mins. The protein concentration in the supernatant was estimated by using Bradford reagent (BioRad, USA).
About 30 μg of protein was separated on a 12% SDS-PAGE and transferred to Nitrocellulose membrane. The membranes were blocked for 1 hour in 5% non-fat milk and 5% BSA dissolved in 1X TBST (TBS with 0.1% Tween 20) and probed with anti-PMK-1 antibody (1:2,000 dilution in blocking buffer; Cell Signaling Technology, USA) or anti-phospho-PMK-1 antibody (1:2,000 dilution in blocking buffer; Cell Signaling Technology, USA), incubated overnight at 4°C. Next day, the membranes were washed thrice with 1X TBST and further incubated with 1:10,000-diluted secondary antibody (anti-rabbit conjugated to HRP, Cell Signaling Technology, USA) for 1 hr at room temperature. The blots were then washed 4–5 times with 1X TBST, each wash lasting 10 min. The blots were developed using enhanced chemiluminiscent substrate (Millipore, USA).
For the quantification of PMK-1 activity, the band intensities of pPMK-1 and total PMK-1 were quantified using ImageJ software (National Institutes of Health, Bethesda, MD; http://rsb.info.nih.gov/ij/) and divided with the intensity of the beta-actin bands. The value thus acquired for pPMK-1 was then divided by that of total PMK-1 and represented as percentage. The immunoblots of four independent experiments were quantified.
Well-fed young adult worms from control RNAi-seeded plates were collected and washed thrice in M9 buffer. The worm pellet was then divided into two halves; to one half 120 μl of 1X M9 buffer was added while to the other half 120 μl of 1X M9 buffer containing 20mM sodium arsenite was added. After incubation at 20°C for 20 minutes, the worms were washed thrice with M9 buffer. The worm pellet was then processed for protein isolation and western blotting using the above-mentioned method.
Worms were grown on respective RNAi from L1 onwards. For each RNAi, four 60 mm unseeded NGM plates with approximately 25 L4 worms per plate were irradiated using a 254 nm UV bulb at 10Jm-2min-1 in a CL-1000 UV Crosslinker (Ultra-Violet Products Limited), followed by transfer to the respective RNAi-seeded NGM. All UV-resistance assays were performed at 20 oC. Survival to stress was scored every 24 hrs post UV exposure.
The worms were grown on RNAi plate as above. For each RNAi, three 60 mm NGM plates with approximately 40 L4 worms per plate were incubated at 35 oC. Animal survival was scored every 60 min.
Wild-type or flr-4(n2259) mutant worms were grown on two different E. coli feed, OP50 or HT115 till late L4 stage. Five worms were picked onto fresh plates (OP50 or HT115 seeded) and allowed to lay eggs for 24 hours. Three such plates were used for each assay so that ‘n’ was 15 per experiment. The worms were then transferred to fresh plates every day and the eggs/L1s on previous day’s plate were counted. Worms that crawled off the plates or ruptured before the fertile period ended were discarded. Eggs that produced viable progeny were considered as total L1s and the un-hatched eggs were considered as dead eggs. Pool of total L1s and dead eggs are defined as brood size. Data is presented as brood size ± SEM. For calculating reproductive span, total number of L1s is expressed per worm per day. Data is shown as viable progenies plotted against number of days, with SEM at each time point.
Gravid adult worms, initially grown on E. coli OP50, were bleached and the eggs were L1 synchronized in M9 buffer for 16 hours before placing them on seeded NGM plates. The worms were then scored every 12 hours till 60th hour for their development stage.
The study was performed with approval from the Institutional Biosafety committee. Only invertebrate nematodes were used for the study.
|
10.1371/journal.pgen.1007488 | Dynamic expression patterns of Irx3 and Irx5 during germline nest breakdown and primordial follicle formation promote follicle survival in mouse ovaries | Women and other mammalian females are born with a finite supply of oocytes that determine their reproductive lifespan. During fetal development, individual oocytes are enclosed by a protective layer of granulosa cells to form primordial follicles that will grow, mature, and eventually release the oocyte for potential fertilization. Despite the knowledge that follicles are dysfunctional and will die without granulosa cell-oocyte interactions, the mechanisms by which these cells establish communication is unknown. We previously identified that two members of the Iroquois homeobox transcription factor gene family, Irx3 and Irx5, are expressed within developing ovaries but not testes. Deletion of both factors (Irx3-Irx5EGFP/Irx3-Irx5EGFP) disrupted granulosa cell-oocyte contact during early follicle development leading to oocyte death. Thus, we hypothesized that Irx3 and Irx5 are required to develop cell-cell communication networks to maintain follicle integrity and female fertility. A series of Irx3 and Irx5 mutant mouse models were generated to assess roles for each factor. While both Irx3 and Irx5 single mutant females were subfertile, their breeding outcomes and ovary histology indicated distinct causes. Careful analysis of Irx3- and Irx5-reporter mice linked the cause of this disparity to dynamic spatio-temporal changes in their expression patterns. Both factors marked the progenitor pre-granulosa cell population in fetal ovaries. At the critical phase of germline nest breakdown and primordial follicle formation however, Irx3 and Irx5 transitioned to oocyte- and granulosa cell-specific expression respectively. Further investigation into the cause of follicle death in Irx3-Irx5EGFP/Irx3-Irx5EGFP ovaries uncovered specific defects in both granulosa cells and oocytes. Granulosa cell defects included poor contributions to basement membrane deposition and mis-localization of gap junction proteins. Granulosa cells and oocytes both presented fewer cell projections resulting in compromised cell-cell communication. Altogether, we conclude that Irx3 and Irx5 first work together to define the pregranulosa cell population of germline nests. During primordial follicle formation, they transition to oocyte- and granulosa cell-specific expression patterns where they cooperate in neighboring cells to build the foundation for follicle integrity. This foundation is left as their legacy of the essential oocyte-granulosa cell communication network that ensures and ultimately optimizes the integrity of the ovarian reserve and therefore, the female reproductive lifespan.
| Fertility is a concern in women’s health, especially when the reported average age of mothers is rising (2016, CDC DB232). Of concern in aging mothers is ovarian follicle health, which requires active communication between its cell components that include a single egg and its surrounding granulosa support cells. The process used to establish cell-cell communication within follicles is unknown but begins during fetal development. We discovered that two related genes, Irx3 and Irx5, direct network construction. If Irx3 and Irx5 are eliminated in mice, oocyte-granulosa cell contact is lost, and all follicles die within short order. Analysis of additional Irx3 and Irx5 mutant mice highlighted that each factor is important for establishing follicle integrity. During fetal development, they work together in the same cell to establish the identity of future granulosa support cells. Then, as the follicles form, they continue to work cooperatively, but in different cells. While Irx5 expression remains in granulosa cells, Irx3 expression transitions to the oocyte and together, they promote cell patterns and interacting zones to synchronize the follicle as an interactive unit. Although the work of Irx3 and Irx5 is finished in the perinatal ovary, their impact lasts a lifetime as it establishes long-term follicle integrity and therefore, fertility.
| Mammalian neonatal ovaries are endowed with a finite and non-replenishable reserve of oocytes that will ultimately define the length of the female’s entire reproductive lifespan. The oocytes that define the ovarian reserve are maintained within primordial follicles, which are comprised of a single oocyte surrounded by a protective layer of somatic cells called granulosa cells. Once females reach reproductive age, some primordial follicles are recruited at regular intervals for maturation and potential ovulation or programmed atresia, but many of them will remain dormant for years. The integrity of those follicles depends on appropriate communication between oocytes and granulosa cells to ensure survival. Early depletion of primordial follicles and other follicle loss result in conditions such as premature ovarian insufficiency (POI), which poses great risk to a woman’s fertility and overall systemic health, but the underlying causes remain largely unknown [1, 2].
After sex determination in female (XX) gonads, primordial germ cells proliferate to form germline nests [3–5]. Just before birth in mice, germline nests break down to form primordial follicles through processes that include pre-granulosa cells extending cytoplasmic projections between oocytes and selective oocyte apoptosis [6–8]. Factors identified that promote this process include FOXL2, NOBOX, Notch and steroid hormone signaling pathways [9–13]. Later, when primordial follicles mature to the primary follicle stage, oocyte-derived signaling molecules including BMP15 and GDF9 begin to direct communication with surrounding granulosa cells to promote follicle growth and maturation [13–18]. As follicles mature, granulosa cells respond to autocrine, paracrine, and hormonal signals via IGF, KIT ligand, and gonadotropins, among others, to coordinate follicle development and eventually ovulation [15, 19–22]. The communication between oocyte and surrounding granulosa cells is key to each of these events and intercellular crosstalk begins as early as the germline nest and then oversees progression through primordial follicle formation, follicle maturation, and ovulation. The events that establish the initial interactions between pre-granulosa cells and oocyte within primordial and primary follicles, however, are still unknown.
Previously, we discovered that deletion of the Fused toes (Ft) locus, which includes Irx3 and Irx5, caused POI with abnormal primordial follicle formation that failed to progress beyond the primary follicle stage [23]. IRX3 and IRX5, like all members of the Iroquois homeobox gene family, are characterized by an 11-amino acid Iro motif and a highly conserved DNA binding homeodomain that includes three extra amino acids, thereby distinguishing them as TALE (three amino acid loop extension) homeodomain transcription factors. Iroquois factors are known for their role in patterning and embryogenesis and are conserved from worms to vertebrates [24]. Among the six members of the Iroquois homeobox gene family, Irx3, Irx5, and Irx6 comprise the IrxB cluster that resides on chromosomes 8 and 16 in mouse and human, respectively [25, 26]. Irx3 and Irx5 exhibit considerable transcript and protein homology and their expression patterns overlap in several developing tissues including the central nervous system, heart, gastrointestinal tract, skin, mammary gland, and limb [24–27]. Recently, our laboratory showed that both Irx3 and Irx5 were detected within the gonad in a sexually dimorphic pattern with enriched expression in developing ovaries, while Irx6 was not detected [23, 27–31]. The timing of the peak of Irx3 and Irx5 expression corresponded to germline nest breakdown and primordial follicle formation, suggesting an important role in ovarian development [23]. This premise was supported by our findings that, like the Ft mutant mouse ovaries, targeted deletion of both Irx3 and Irx5 caused disruption of granulosa cell-oocyte contacts during early follicle development leading to oocyte death [3]. We therefore hypothesized that each Iroquois factor (IRX3 and IRX5) is critical to maintain follicle integrity and promote fertility. Here we report that deletion of either Irx3 or Irx5 impairs female fertility, but with distinct outcomes from each strain suggesting specific roles for each gene. This was supported by their dynamic expression profiles that are initially shared in pre-granulosa cells of germline nests and then subsequently diverge to granulosa cell- or oocyte-specific as nests breakdown into primordial follicles. Investigations into potential causes of follicle death in Irx3 and Irx5 double knockout ovaries uncovered defective cell biology in both granulosa cells and oocytes including abnormal deposition of the basement membrane, mis-localized gap junction proteins, and diminished extension processes emanating from both cell types. Taken together, we conclude that Irx3 and Irx5 work together in the same cells during fetal development, and then cooperate in neighboring cells to synchronize the formation of oocyte-granulosa cell interactions and promote granulosa cell function to establish the foundation required for follicle integrity and optimal female fertility.
Previously, we reported that follicles from Fused Toes (Ft) mutant mouse ovaries failed to mature as a result of defective granulosa cell-oocyte interactions. Of the six genes eliminated in the Ft mutation, only Irx3 and Irx5 exhibited an ovary-enriched expression profile suggesting that they played an integral role in follicle formation and maturation [23]. To test their roles in ovary development independent of the Ft mutation, we analyzed a series of Irx3 and Irx5 mutant mouse models.
Similar to the Ft mutation, the Irx3-Irx5EGFP/Irx3-Irx5EGFP double knockout (Irx3/5 DKO, Fig 1A) is lethal in mice by approximately embryonic day 13.5–14.5 (E13.5–14.5) [32]. Ovaries from E12.5 wild type and Irx3/5 DKO female embryos were transplanted under the kidney capsule (KCT) of ovariectomized nude mouse hosts for 10 days, 2 or 3 weeks to evaluate ovary development beyond the lethal stage. Ten days after surgery, both wild type control and mutant grafts contained follicles, suggesting that Irx3 and Irx5 are not necessary for the initiation of follicle formation (Fig 1B and 1C). Wild type primary follicles exhibited granulosa cells with their characteristic cuboidal shape surrounding and intimately associated with the oocyte in transplant grafts after 10 days and three weeks KCT (Fig 1B and 1D). In contrast, granulosa cells in primary follicles of Irx3/5 DKO grafts were rounded in shape and unevenly distributed around an oocyte with disrupted cell-cell contacts (Fig 1C and 1E). Closer examination by transmission electron microscopy (TEM) verified significant gaps in granulosa cell-oocyte contacts in Irx3/5 DKO follicles compared to the wild type control (Fig 1F and 1G)[3]. The relative incidence of abnormal follicle morphology was quantified. Abnormal follicles were defined as those that exhibited at least one of the following characteristics: mis-shaped granulosa cells (rounded, not cuboidal), non-uniform or wisp-like oocyte cytoplasm, presence of apoptotic bodies, asymmetric accumulation of granulosa cells relative to a central oocyte, and increased distances between granulosa cells and/or between granulosa cells and oocytes (S1 Fig). Compared to wild type (n = 44 follicles) grafts, Irx3/5 DKO (n = 52 follicles) grafts harbored substantially higher percentages of abnormal primordial (14% vs 0%), primary (80% vs 17%) and secondary/pre-antral (90% vs 59%) follicles.
To facilitate evaluation of ovary development without embryonic lethality, we generated Irx3-Irx5EGFP/Irx3floxIrx5EGFP mice (herein referred to as Irx3/5 hypomorph) that maintain a single functional Irx3 allele (flox; Irx3-/flox) on an Irx5-null background (Irx5EGFP/EGFP) (Fig 2A) [33]. Some of the Irx3/5 hypomorph mice were viable, but they were too small and feeble to use in a standard breeding study (8-week-old body weights: Irx3/5 hypomorph 15.14 +/- 1.11g; littermate controls 28.84 +/- 2.55g, p = 0.0003). Instead, we evaluated ovary function using superovulation and in vitro fertilization (IVF). By 8 weeks of age, the hypomorph mice were of sufficient size [34] to assess their fertility using this protocol. After superovulation, fragmented eggs were counted and then discarded and not included in IVF, but there was a higher fragmentation rate in Irx3/5 hypomorph oocytes indicating poor quality of ovulated eggs (normalized fragmentation rate of 8.0 +/- 4.2 vs 1.0 +/- 0.1 for control oocytes, p = 0.15; S2 Fig). In addition, compared to controls, Irx3/5 hypomorph females ovulated significantly fewer oocytes and therefore presented less 2-cell embryos (p < 0.05, Fig 2B, S2 Fig). The hypomorph mice also exhibited a trend of lower efficiency of progression to 2-cell embryos compared to the controls (p = 0.18, Fig 2B).
Fewer ovulation events in Irx3/5 hypomorph mice was supported by the presence of fewer corpora lutea (CL) than control ovaries (Fig 2C). Quantification confirmed statistically fewer CLs in addition to reduced numbers of large atretic follicles in Irx3/5 hypomorph compared to control ovaries (Fig 2D). No statistical difference was detected between groups for other follicle stages at this 8-week old time point (Fig 2D). Irx3/5 hypomorph ovaries were smaller than controls (Fig 2C). This is explained by the substantial volume that CLs contribute to the size of control ovaries along with the disparity in body size. Altogether, these results suggest that growing follicles are unresponsive to ovulatory signals in Irx3/5 hypomorph mice.
Next, we examined the individual effects of Irx3 and Irx5 using Irx3LacZ/LacZ and Irx5EGFP/EGFP mice, respectively (Fig 3A and 3B). Each strain was previously documented to be robust and able to produce live offspring [33, 35, 36]. To assess their fertility, 6-week-old females from each strain were placed in a breeding study for six months. Irx3LacZ/LacZ mice gave birth to fewer pups than their littermate controls throughout the breeding study (Fig 3C). Irx5EGFP/EGFP mice also produced significantly fewer pups, but not until 120 days (about 4 months) into the breeding study (Fig 3D). Quantification of structures within ovaries after the conclusion of the breeding study supported that fertility defects were different between the two mutant strains. Irx3LacZ/LacZ ovaries harbored significantly fewer total follicles caused by reduced contributions from young stage follicles and a trend of an increased number of atretic follicles, most of which were small atretic follicles (Fig 3E). Histological quantification of Irx5EGFP/EGFP ovaries did not reveal any differences from the controls, except for a trend of relatively fewer secondary follicles and atretic follicles, both small and large (Fig 3F). Together, these results suggest that while Irx3 and Irx5 each contribute to fertility, their roles are potentially different. Further, when comparing the results between Irx5EGFP/EGFP, Irx3LacZ/LacZ and the Irx3/5 hypomorph (Irx5EGFP/EGFP plus loss of one Irx3 allele) strains, we conclude that defects associated with Irx3 occur earlier and cause a more severe phenotype than the loss of Irx5.
Because fertility studies suggested unique roles for Irx3 and Irx5, we hypothesized that their expression profiles would be distinct. Previously, we reported that Irx3 and Irx5 initiate female-specific expression in developing gonads starting at E12.5. Both transcripts increase and then peak in ovaries around the time of birth (postnatal day 0, P0) followed by a rapid decline (S3 Fig) [23, 28, 29]. Because no reliable antibodies are currently available for IRX3 or IRX5, we used heterozygous Irx3LacZ/+ and Irx5EGFP/+ reporter mice [32, 37] to mark Irx3 and Irx5 expression, respectively, in specific cell types in the developing ovary over time. It is important to note that while each reporter reflects cellular localization over time, they do not reflect the endogenous subcellular localization of IRX3 or IRX5.
During fetal stages, cytoplasmic β-galactosidase staining, marking Irx3 expression, was detected in GATA4-positive somatic cells but not TRA98- or VASA-positive germ cells of germline nests (S4 Fig, Fig 4A and 4B). While GATA4 marks all somatic cells, Irx3-LacZ-positive cells were associated only with somatic cells that surrounded germline nests (S4 Fig). As germline nests broke down and primordial follicles formed at birth, Irx3-LacZ expression was also detected in VASA-positive oocytes while being maintained in somatic cells (Fig 4C and 4D). Shortly after birth, strong Irx3-LacZ expression remained in oocytes of primary follicles, but its expression diminished and was eventually eliminated from granulosa cells (Fig 4E and 4F). Oocyte-enriched Irx3-LacZ expression was maintained throughout postnatal development and into adulthood (Fig 4O and 4P). Other members of the TALE homeodomain transcription factor family have been localized to both nuclear and cytoplasm [38, 39]. To evaluate subcellular localization of IRX3, we used antiserum against IRX3 that was previously made by our laboratory [28]. Results showed that during developmental stages, IRX3 was expressed within the nucleus of somatic cells that surrounded germline nests (Fig 4M). Just before germline nest breakdown, IRX3 expression was maintained within somatic cell nuclei, but was also detected in both cytoplasm and nuclei of germ cells within the germline nest and in the cytoplasm of ovarian surface epithelial cells (Fig 4N). Early-stage primordial follicles retained IRX3 expression in nuclei of pre-granulosa cells and in both nuclei and cytoplasm of germ cells. Finally, IRX3 was no longer detected in pre-granulosa cells, but was maintained in germ cells of more mature primordial follicles within the ovary medulla (Fig 4N). Unfortunately, this antiserum is no longer available. These data support the timeline of Irx3 expression delineated by the reporter mouse strain and provide important new information. A summary of the subcellular localization of IRX3 over time in developing ovaries is presented in S1 Table.
Irx5 is separated from Irx3 by 550 kb on mouse chromosome 8 and shares a similar transcript expression profile with Irx3 in several developing tissues [24, 25, 27], but cell specific localization varies depending on developmental stage and tissue type [32, 40]. Similar to Irx3-LacZ, Irx5-EGFP expression was absent in TRA98- and VASA-positive germ cells and was mostly restricted to GATA4-positive cells surrounding germline nests during fetal development (Fig 4G and 4H, S4 Fig). Further, during early postnatal days it was also detected (weakly) in oocytes of primordial and some primary follicles while being maintained in granulosa cells (Fig 4I and 4J). In contrast to Irx3-LacZ, oocyte expression of Irx5-EGFP diminished in later staged follicles, but its presence in GATA4-positive granulosa cells remained robust until at least P7 (Fig 4K and 4L). By P21, Irx5-EGFP was no longer detected in any cell type (Fig 4Q). There is no antibody available to evaluate subcellular localization of IRX5.
To validate Irx3 and Irx5 co-expression, we generated Irx3LacZIrx5+/Irx3floxIrx5EGFP dual-reporter mice. In support of the previous results, double immunofluorescence with β-galactosidase and EGFP antibodies indicated that both Irx3 and Irx5 were expressed in somatic cells outlining germline nests at E16.5 (Fig 4R, 4R’, 4U and 4V). Upon germline nest breakdown and primordial follicle formation at birth (P0), Irx3 and Irx5 were co-expressed in pre-granulosa cells at the ovarian cortex with decreased signal intensity for both in the medulla (Fig 4S, 4S’, 4W and 4X). By P7, their expression patterns were distinct with Irx3 and Irx5 confined to oocytes and granulosa cells, respectively (Fig 4T, 4T’, 4Y and 4Z). In addition, both Irx3 and Irx5 were detected in the ovarian surface epithelial cell layer at all three time points (Fig 4R–4Z).
At least three different populations of somatic cells have been reported in the developing ovary [41]. To identify the specific somatic cell populations for Irx3 and Irx5 expression, we evaluated whether their reporters co-localized with known ovarian somatic cell markers including the vasculature-associated somatic cell marker, COUP-TFII (NR2F2), and the pre-granulosa cell marker, FOXL2 [14, 41]. Results from ovaries at E15.5 and P0 showed that both Irx3 and Irx5 were not expressed in NR2F2-positive cells (Fig 5A–5D). Co-expression with FOXL2-positive cells was more nuanced. FOXL2 was detected in Irx3- and Irx5-positive somatic cells in the medulla at E15.5 but not in the cortex (Fig 5E and 5F). By P0, all pre-granulosa cells expressed FOXL2 (Fig 5G and 5H). While Irx3 was detected in all FOXL2-positive cells though weaker in the medulla (Fig 5G), Irx5 expression co-localized with FOXL2-positive cells only within the cortex and was absent in the medulla (Fig 5H). As described above (Fig 4R–4Z), Irx3 and Irx5 were expressed in the ovarian surface epithelium, which was negative for FOXL2 and NR2F2 at both E15.5 and P0 (Fig 5A–5H).
Thus far, we have demonstrated that Irx3 and Irx5 both contribute to female fertility and that ovarian function is increasingly impaired as alleles of Irx3 are eliminated in the context of the Irx5-null mouse. The most severe phenotype is presented in Irx3/5 DKO ovarian grafts with follicles at varying stages of cell dysfunction and death (Fig 1) [3]. After having determined that Irx3 and Irx5 are expressed in both cell types that make up a follicle, we next set out to identify potential mechanisms to explain defects in mutant ovaries.
Potential causes for granulosa cell dysfunction include loss of cell identity or defective cell polarity/orientation within the follicle structure. Immunofluorescence results showed that Irx3/5 DKO granulosa cells maintained their identity as expression of the granulosa cell markers, anti-Müllerian Hormone (AMH) and FOXL2, was not different from controls, and no upregulation of SOX9 expression was detected by 2 weeks post KCT surgery (S5 Fig). Instead, immunofluorescence and TEM approaches suggested a loss of granulosa cell polarity or orientation (Fig 6). Granulosa cells produce and guide deposition of the basement membrane that surrounds each follicle, which in turn influences their proliferation and follicle growth [42–44]. Evaluation of laminin, a basement membrane marker, demonstrated that a continuous basement membrane surrounded follicles in mutant and control tissues; however, the intensity of its expression was reduced in Irx3/5 DKO follicles, especially as they matured (compare Fig 6A–6C and 6D–6F, quantified in Fig 6G). Laminin expression was detected in primary and early secondary follicles but diminished in later staged follicles (compare Fig 6D, 6E and 6F). This finding was supported by results that showed no difference in laminin transcript levels at early developmental stages of E13.5 or 7d KCT (S6 Fig). Closer examination by TEM showed that in contrast to the smooth, organized basement membrane adjacent to wild type granulosa cells (Fig 6H), the structure in Irx3/5 DKO tissues presented with a variety of phenotypes. We observed increased incidences of abnormal membrane deposition including double layers (Fig 6I), abnormal production with looping (Fig 6J), and diffuse, permeable deposition (example, Fig 6K). The incidences of these abnormalities are outlined in S2 Table.
Granulosa cells must communicate with each other, primarily via connexin 43 (Gap junction protein 1, GJA1), to integrate signals that promote their proliferation, differentiation, and survival [45, 46]. In addition, Irx3 is known to repress Gja1 transcription in neonatal ventricular myocytes [37]. Although there was no difference in transcript accumulation for Gja1 between wild type and mutant ovaries at earlier stages including E13.5 or 7d KCT (S6 Fig), our immunofluorescence analysis uncovered abnormal protein localization patterns that varied depending on follicle stage. In early stage follicles (transitioning primordial and early primary follicles), GJA1 was detected as expected between granulosa cells, but also at the granulosa cell—oocyte interface in mutant tissues (S7 Fig). Later, starting at the late primary follicle stage, oocytes synthesize the zona pellucida, a specialized extracellular matrix that plays critical roles in oocyte growth, fertilization, and early embryonic development. Simultaneously, granulosa cells develop transzonal processes that are designed to traverse the zona pellucida to extensively interact with each other and the oocyte [45, 47]. Robust GJA1 expression was detected within the zona pellucida of late primary and more mature follicles in control tissues (Fig 6A–6C and 6A’–6C’). In contrast, GJA1 signal decreased in intensity in this region as follicles matured in Irx3/5 DKO ovaries indicating gradually diminishing interactions between transzonal processes over time (Fig 6D–6F and 6D’–6F’). In support of GJA1 data, TEM images validated the decrease in transzonal projections from different granulosa cells. While frequent interactions were observed among the trans-zona in control follicles (Fig 6L–6N), few were observed in Irx3/5 DKO follicles (Fig 6O–6Q). In other areas of Irx3/5 DKO antral follicles, populations of granulosa cells exhibited abnormal accumulation of GJA1, including in cell membranes at the interface with the basement membrane (compare Fig 6B’ to 6E’ and 6F).
Finally, TEM uncovered a generalized paucity of interactions in the zona pellucida between the oocyte and granulosa cells (Fig 6L–6Q). Besides distinctly fewer transzonal processes extending from granulosa cells, the organization of structures originating from the oocyte was also different in Irx3/5 DKO follicles. Widened and inconsistent distances separated microvilli extensions from Irx3/5 DKO oocytes compared to controls and there were fewer microvilli in many cases (Fig 6M, 6N, 6P and 6Q). Taken together, both cell types contribute fewer interacting processes that result in fewer opportunities for cell-cell communication within the interacting zone of Irx3/5 DKO follicles. To test whether the disrupted communication was the cause or the result of follicle death, we tested for apoptosis using immunofluorescence with antibodies against cleaved caspase 3 and laminin on wild type (n = 6) and Irx3/5 DKO (n = 5) grafts evaluating a range of 2–16 follicles per slide. Cleaved caspase 3-positive cells were rare in both wild type and Irx3/5 DKO follicles indicating that these disruptions occur before follicle death (S8 Fig). Taken together, we conclude that Irx3 and Irx5 function in oocytes and granulosa cells to promote directional intercellular interactions along with a functional follicle-ECM niche to ensure follicle survival.
It has long been established that somatic cell-oocyte communication is critical for follicle integrity and promotes healthy ovarian development and function. During ovarian development, factors are expressed in specific spatio-temporal fashion to not only determine oocyte and somatic cell identity, but also coordinate cell-cell interactions that ensure proper breakdown of germline nests and establishment of the primordial follicle pool for future recruitment [7–13, 17, 48]. The mechanisms by which cell-cell communication networks are established between pre-granulosa cells and oocytes within new primordial follicles, however, remain unknown. We previously showed that Irx3 and Irx5 transcripts peaked in the ovary around birth and that their expression was required to maintain oocyte-granulosa cell contacts within early-staged follicles [3, 23]. The objective for this study was to investigate roles for Irx3 and Irx5 in ovarian development and function. We demonstrated that female fertility was significantly impaired in both Irx3LacZ/LacZ and Irx5EGFP/EGFP single mutant mice. Their subfertility phenotypes were different, suggesting their distinct roles. In support, careful analysis of Irx3 and Irx5 expression profiles highlighted primordial follicle formation as a critical time point when their expression patterns transitioned from shared to unique cell localization. Evaluation of Irx3/5 DKO follicles underlined the importance of their individual expression patterns. Most follicles exhibited abnormal morphology within early postnatal stage Irx3/5 DKO ovarian tissue. Evidence points to defective granulosa cell function and granulosa cell-oocyte communication structures as a cause of follicle death. Based on these data, we conclude that Irx3 and Irx5 direct construction of the communication infrastructure within and between pre-granulosa cells and the oocyte as nascent primordial follicles form.
It has been proposed that two distinct populations of primordial follicles exist in the postnatal ovary that are activated in two separate waves to play different roles in ovarian development and fertility [49, 50]. Foxl2 and Lgr5 are reported to mark pre-granulosa cell populations that distinguish the two pools of primordial follicles [41, 48, 51]. We propose that Irx3 and Irx5 are present in both populations. During embryonic development (E15.5), Irx3 and Irx5 are expressed in the ovarian surface epithelium and in somatic cells that outline germline nests, whereas FOXL2 expression is restricted to nests located in the ovarian medulla. These findings support previous reports that FOXL2 marked early somatic cells within the medulla that give rise to granulosa cells within primordial follicles that are activated during the first wave, immediately after birth [48, 50]. Mork and colleagues suggested a second wave of primordial follicles that established the lifetime ovarian reserve. These follicles incorporated pre-granulosa cells that originated from FOXL2-negative ovarian surface epithelial cells, which would eventually mature into FOXL2-positive granulosa cells [48]. These ovarian surface epithelial cells were found to be Lgr5-positive, and they matured into pre-granulosa cells of primordial follicles that populated the cortex to define the ovarian reserve at birth [41, 52, 53]. Our results show that Irx3 and Irx5 are expressed in ovarian surface epithelium (Lgr5-positive) and both FOXL2-negative (presumably Lgr5-positive) and FOXL2-positive pre-granulosa cells, suggesting that Irx3 and Irx5 mark cells destined to be granulosa cells within follicles assembled for both populations of primordial follicles.
Surprisingly, Irx3 and Irx5 were found to be expressed briefly in oocytes while they were still in pre-granulosa cells during the perinatal stage when germline nests break down and primordial follicles form. As follicles develop beyond the primordial and primary stages, Irx3 expression was detected only in oocytes while Irx5 expression was restricted to granulosa cells (summarized in the model in Fig 7A). Notably, evaluation with IRX3 antiserum indicated that IRX3 was expressed in the nucleus and cytoplasm of the oocyte whereas pre-granulosa cell expression was confined to the cell nucleus (S1 Table). The divergence of Irx3 and Irx5 expression patterns occurs as follicles transition from primordial to primary follicles, which represents an event of follicle development of which little is known. The shared function of Iroquois family factors, including IRX3 and IRX5, is to promote patterning during development [25, 27, 33, 54]. Other members of the TALE homeodomain transcription factor family are known to shuttle between the nucleus and cytoplasm in certain cells at specific time points during development. Notably, when these factors were detected in the cytoplasm, they co-localized with cytoskeletal proteins (actin, non-muscle myosin) [38, 39]. The activity of these factors in the cytoplasm is still under investigation, but there have been links to their facilitating responses of extracellular signals. That Irx3 and Irx5 are expressed in two distinct but neighboring cell types suggest their cooperative role in orchestrating cell-cell interactions. Indeed, studies in heart development revealed that Irx3 and Irx5 performed both supportive and antagonistic functions during heart development and in postnatal cardiac electrophysiology induction, respectively [32]. When Irx3 and Irx5 were not co-expressed, they worked together via different but spatially connected cells to promote electric propagation [32]. In support of studies within other developing organs, we find that Irx3 and Irx5 impart spatial expression patterns that transitions at the critical stage of primordial follicle formation.
Irx3LacZ/LacZ and Irx5EGFP/EGFP mutant mice have been used to investigate functions of other systems and are healthy and fertile [33, 35, 37, 55]. Six-month breeding studies, however, revealed that both strains accumulated ≤ 50% pups compared to their littermate controls, but in different patterns. Irx3LacZ/LacZ females consistently produced fewer pups throughout the entire study. At 8 months of age, Irx3LacZ/LacZ ovaries harbored fewer early stage follicles along with a trend for increased atretic follicles compared to controls suggesting that fewer oocytes were available for recruitment. At this rate, it is likely that Irx3LacZ/LacZ mutants would deplete their ovarian reserve much earlier than the controls, as in POI cases. Furthermore, we demonstrated that Irx3 was present in the oocyte beyond the primordial follicle stage. These data, along with increased oocyte fragmentation detected in Irx3/5 hypomorph mice suggest that Irx3 impacts oocyte quality and therefore, fertilization success. These outcomes, together with its transitioning expression profile, suggest that Irx3 is important in both oocytes and granulosa cells.
Irx5EGFP/EGFP mutant females accumulated the same number of offspring as controls for the first three months but failed to increase their numbers thereafter. Thus, it was somewhat surprising to find no obvious difference in follicle numbers between groups. Likewise, Irx3/5 hypomorph females that lack one Irx3 and both Irx5 alleles ovulated significantly fewer eggs, but ovary histology did not differ from the controls other than an expected reduction in the number of CLs. Successful ovulation requires response and coordination of granulosa cells [22]. These fertility defects, along with the knowledge that all Irx5 expression is eliminated before puberty suggest that during the perinatal stage, Irx5 prepares granulosa cells for future responsiveness to external signals required for maturation and ovulation. This process is substantially more successful when Irx5-positive granulosa cells can communicate with oocytes harboring their full complement of Irx3 expression. It is important to note however, that there remains a possibility that decreased fecundity may also be caused by implantation defects. This is currently under investigation. Altogether, these results are similar to that of Bmp15-/- ovaries that exhibited impaired ovulatory function with generally normal morphology except for evidence of trapped oocytes [56, 57]. Both mutant mouse models support the need for the early onset of communication between cell types in the growing follicle.
Complete loss of both Irx3 and Irx5 has devastating effects on follicle health much earlier than the other Irx mutant mouse models. We found that Irx3/5 DKO follicles deposited basement membrane sections with abnormal morphology and that it appeared to diminish as follicles matured. Previous in vitro studies have demonstrated that extracellular matrix (ECM) proteins are critical for follicle survival [58]. As such, ultrastructure images provided evidence that some follicles, though generally healthy looking, contained less competent oocytes when they were surrounded by looped or multiple layers of basement membrane deposits and the oocytes progressed through atresia when basement membrane became fragmented [59]. Laminin is a major ECM component of the basement membrane surrounding follicles that is produced by granulosa cells, among other cells, suggesting that they contribute towards follicle basement membrane organization [60–62]. Thus, the presence of insufficient follicular basement membranes in Irx3/5 DKO follicles suggest that Irx3 and Irx5 are required for granulosa cell function. Moreover, it has been proposed that oocytes regulate basement membrane deposition via oocyte-specific glycoprotein signals through theca cells [63]. Therefore, we should include the possibility that the oocyte contributes to defects observed in Irx3/5 DKO follicles, especially given the presence of Irx3 in oocytes.
More alarming than basement membrane insufficiency was the disarray of cell-cell communication at multiple levels. Irx3/5 DKO follicles exhibited mis-localized and abnormal accumulation of GJA1 providing further support that granulosa cells failed to establish functional communication ports. Polarized junction protein expression is an indication of normal granulosa cell function and is required to establish appropriate cell-cell communication [45, 64]. In addition, in Irx3/5 DKO follicles, GJA1 expression was nearly absent in the region of the zona pellucida of pre-ovulatory-stage follicles suggesting few granulosa cell-granulosa cell interactions near the oocyte. This was supported by TEM images that validated the lack of granulosa cell-derived transzonal processes, but also highlighted abnormal patterns of microvilli extensions from the oocyte. Together, these data highlight roles in both granulosa cell and oocyte for Irx3 and Irx5 in maintaining follicle integrity.
Based on the data reported here, we propose a model to illustrate our conclusion that Irx3 and Irx5 work together, in the same and then in neighboring cells (pre-granulosa cells and oocyte) over time to determine the pattern, polarity, and therefore, the foundation by which communication networks are established within new follicles. This model is remarkably similar to that proposed by Gaborit et al. in which Irx3 and Irx5 have complex interactions in the developing heart. During heart development, they exhibit redundant activity as they colocalize in the endocardium. Thereafter, in the postnatal heart, their expressions transition to neighboring cells where they modulate the expression of intercellular channels to maintain appropriate conductance [32]. Fig 7A models the expression profiles for each factor over time with the transition into cell-specific expression patterns that coincides with the establishment of the ovarian reserve in the form of primordial follicles. We propose that cytoplasmic IRX3 or IRX5 interacts with cell cytoskeleton to direct polarity and organize cell extensions. Further, these cells are simultaneously producing and organizing ECM components to establish the niche for the new follicle. These parameters are established in the perinatal ovary, but their effects continue to impact follicular health throughout the reproductive lifespan of the ovary. Fig 7B highlights the impact of the Irx3/5 DKO mutation. Although follicles have formed with a single oocyte surrounded by granulosa cells, the cell-cell and cell-ECM interactions have not been established for lasting quality. Instead, cell-cell interactions become compromised and eventually, the oocyte and granulosa cells lose contact and die.
The biggest challenge for our study was the embryonic lethality of Irx3 and Irx5 double mutation. With the generation of Irx3/5 hypomorph and Irx3 and Irx5 single mutant females, we gained insight into factor-specific roles. The reporter lines were critical to highlight the potential for time- and cell-specific effects. Thus, the results of this current study provided a map for future investigations of Irx3 and Irx5 functions in the ovary using cell specific mutations. Furthermore, the mis-expression of GJA1 and abnormal basement membrane in Irx3/5 DKO follicles render it necessary to investigate the interactions of Irx3 and Irx5 with their downstream targets and other components of cellular architecture. Here, we show convincing evidence that Irx3 and Irx5 are critical factors for the developing ovary while they are not expressed in the developing testis. In the ovary, they work together to mark future granulosa cells during the fetal stage, and then function within both the oocyte and its protective somatic cell layer to coordinate formation of the nascent primordial follicles. Their legacy is manifest in functional oocyte-granulosa cell interactions that ensure healthy follicle maturation and competent oocytes.
Adult animals were euthanized by CO2 asphyxiation followed by cervical dislocation. Embryonic pups were euthanized by decapitation with a razor blade. Animal housing and all procedures described were reviewed and approved by the Institutional Animal Care and Use Committee at the University of Wisconsin—Madison and were performed in accordance with National Institute of Health Guiding Principles for the Care and Use of Laboratory Animals.
Mouse strains included CD1 outbred mice (Crl:CD1(ICR), Charles River, MA), nude mice (Crl:NU-Foxn1nu, Charles River, MA), Irx3LacZ [37], Irx3floxIrx5EGFP (referenced as Irx5EGFP here) [32], and Irx3-Irx5EGFP [32], all of which were maintained on a CD1 genetic background. Genotyping for Irx3LacZ, Irx5EGFP and Irx3-Irx5EGFP was carried out as previously reported [32, 37]. Timed mating was identified by the presence of a vaginal plug, which was designated as embryonic day 0.5 (E0.5). Ovaries were collected at the indicated time points for further analysis.
Gonads were harvested from embryonic day (E) E12.5 embryos resulting from the breeding of Irx3-Irx5EGFP/Irx3+Irx5+ male and female mice. Post-genotyping, Irx3+Irx5+/Irx3+Irx5+ (wild-type, WT) and Irx3-Irx5EGFP/Irx3-Irx5EGFP (Irx3/5 DKO) ovaries were transplanted under the kidney capsule of an ovariectomized nude mouse of at least eight-week-old as previously described [23]. The nude mice were ovariectomized at least 2 weeks prior to KCT surgeries. Grafted ovaries were recovered 10 days, 2 weeks or 3 weeks post transplantation.
Irx3-Irx5EGFP/Irx3FloxIrx5EGFP (Irx3/5 hypomorph) pups often died shortly after birth. Those that survived were the same size as littermate controls at first, but growth was significantly retarded; therefore, they were maintained with their dam to encourage survival with continued access to milk. Once pups reached sufficient size to respond to hormone induction (approximately 8 weeks of age)[34], they and their littermate (age matched) controls were subjected to a superovulation protocol including intraperitoneal (IP) injections with 5 IU each of pregnant mare’s serum gonadotropin (PMSG) and human chorionic gonadotropin (hCG) 64 hours and 16 hours, respectively, prior to oocyte harvest. Oocytes were extracted at the Transgenic Animal Facility at UW-Madison Biotechnology Center and in vitro fertilized with wild type sperm. The number of ovulated oocytes, fragmented oocytes, oocytes used for IVF and 2-cell embryos post fertilization were recorded. Due to variability across superovulation and IVF experiments, each of the numbers were normalized relative to the control data within the same experiment.
Irx3LacZ/LacZ, Irx5EGFP/EGFP females and their respective littermate control females were set up with wild type males to breed for 6 months. WT males were replaced and rotated at least once. The litter sizes and birth dates were closely monitored and recorded throughout the breeding period.
Ovaries were harvested from each experimental paradigm, fixed in 4% PFA overnight and then embedded in paraffin. Paraffin blocks were sectioned at 8 μm thickness and then stained with hematoxylin and eosin (H&E) for histological analysis.
Ovaries were sectioned completely through and every tenth section was used to quantify the numbers for each structure. Corpora lutea (CLs) and large follicles were counted in each quantified section for each ovary; therefore, these structures were counted multiple times resulting in the high numbers/ovary. All counting was completed by investigators blinded to ovary genotype.
Mouse ovaries were harvested at E15.5 and P0, P3, P7, P21 and 8 months, fixed in 4% paraformaldehyde (PFA) in phosphate buffer saline (PBS) at 4°C overnight, and then washed in PBS. Samples were dehydrated through an ethanol gradient, cleared in xylene and then embedded in paraffin. Ovary grafts used for GJA1 and laminin staining were snap-frozen in Tissue-Plus O.C.T. Compound embedding medium immediately after dissection and were fixed in acetone for 10 minutes after sectioning prior to storage. Primary antibodies were applied to 8 μm paraffin tissue sections and 8 μm frozen sections and then incubated at 4°C overnight (Table 1). Secondary antibodies (Table 2) were then incubated at room temperature for 1 hour. A 10X DAPI (4’,6-diamidino-2-phenylindole) in PBS solution (1:500) was used as a nuclear counterstain. Images were collected on a Leica SP8 confocal microscope and a Keyence BZ-X700 microscope at the School of Veterinary Medicine, University of Wisconsin—Madison. Other images were acquired using a Nikon C1 confocal microscope at the Monash Micro Imaging Facility and on a Zeiss LSM800 confocal microscope at the Biological Optical Microscopy Platform (BOMP) at the Department of Anatomy and Neuroscience, the University of Melbourne. Images were processed with ImageJ or Adobe Photoshop.
Ovary grafts collected at 2 weeks post KCT surgery were processed for transmission electron microscopy by the University of Wisconsin Electron Microscopy Service. Briefly, samples were immersion fixed for 2 hours in 2.5% glutaraldehyde, 2% PFA buffered in 0.1M sodium phosphate buffer (PB) at room temperature (RT). After rinsing, samples were post-fixed in 1% osmium tetroxide, 1% potassium ferrocyanide in 0.1M PB for 1 hour at RT, rinsed, and then stained in saturated aqueous uranyl acetate for 2 hours at RT. Dehydration was performed at RT in a graded ethanol series and then transitioned in propylene oxide (PO). Fully dehydrated samples were then infiltrated in increasing concentrations of PolyBed 812 (Polysciences Inc. Warrington, PA) and PO mixtures. Embedding and polymerization took place in fresh PolyBed 812 for 48 hours at 60°C. Semi-thin sections (1 mm) were first stained with methylene blue/Azure II for light microscopic inspection. The samples were then sectioned on a Leica EM UC6 ultramicrotome at 90nm. The sections were collected on Cu, 300 mesh thin-bar (EMS Hatfield, PA), and post-stained in uranyl acetate and lead citrate. The sectioned samples were viewed at 80kV on a Philips CM120 transmission electron microscope, equipped with MegaView III camera (Olympus Soft Imaging System Lakewood, CO).
Statistical evaluation of superovulation, IVF, breeding study, follicle quantification, and laminin intensity quantification results between groups were carried out using a two-tailed t-test assuming unequal variances. Results were considered statistically different if p-values were ≤ 0.05. Results of p < 0.2 are also reported. Gene expression value analysis was conducted as described in the figure legend.
|
10.1371/journal.ppat.1002581 | Additive Function of Vibrio vulnificus MARTXVv and VvhA Cytolysins Promotes Rapid Growth and Epithelial Tissue Necrosis During Intestinal Infection | Vibrio vulnificus is a pathogen that causes both severe necrotizing wound infections and life-threatening food-borne infections. Food-borne infection is particularly lethal as the infection can progress rapidly to primary septicemia resulting in death from septic shock and multiorgan failure. In this study, we use both bioluminescence whole animal imaging and V. vulnificus bacterial colonization of orally infected mice to demonstrate that the secreted multifunctional-autoprocessing RTX toxin (MARTXVv) and the cytolysin/hemolysin VvhA of clinical isolate CMCP6 have an important function in the gut to promote early in vivo growth and dissemination of this pathogen from the small intestine to other organs. Using histopathology, we find that both cytotoxins can cause villi disruption, epithelial necrosis, and inflammation in the mouse small intestine. A double mutant deleted of genes for both cytotoxins was essentially avirulent, did not cause intestinal epithelial tissue damage, and was cleared from infected mice by 36 hours by an effective immune response. Therefore, MARTXVv and VvhA seem to play an additive role for pathogenesis of CMCP6 causing intestinal tissue damage and inflammation that then promotes dissemination of the infecting bacteria to the bloodstream and other organs. In the absence of these two secreted factors, we propose that this bacterium is unable to cause intestinal infection in humans.
| Vibrio vulnificus causes disease both by infection of wounds from seawater and by consumption of contaminated foods, especially oysters. Wound infection results in necrotizing fasciitis and edema in extremities with mortality of ∼25% as the incidence of septicemia is low. Contaminated food consumption by contrast can lead to highly invasive infections that progress rapidly from an intestinal infection to primary septicemia. Case-fatality rates are ≥50%, with rates as high as 100% in individuals who receive no antibiotic therapy. The aim of this study is to elucidate virulence mechanisms of food-borne infection of the most highly virulent strains of V. vulnificus. We developed a novel intragastric infection model for a highly virulent clinical isolate from Korea in which we can observe the bacterial load in live mice and applied this to study of wild type and strains genetically altered to delete genes for two secreted cytotoxins. Using this model, we show that both the multifunctional-autoprocessing RTX toxin (MARTXVv) and the cytolysin VvhA contribute to rapid in vivo growth of bacteria and that the presence of these factors directly correlates with mouse mortality. These exotoxins are then directly linked to intestinal damage and inflammation.
| Vibrio vulnificus is a motile, Gram-negative, opportunistic human pathogen capable of causing severe to life-threatening infection in individuals with predisposing conditions, including liver damage, hereditary hemochromatosis and compromised immune systems [1]–[3]. Infection can result from consumption of contaminated seafood or from exposing an open wound to water harboring the pathogen. Wound infection can progress to edema, cellulitis, ecchymoses and necrotizing fasciitis at the site of infection [4], [5]. The mortality of wound infection is about 25% because primary septicemia does not frequently occur [5], [6]. By contrast, V. vulnificus food-borne infection rapidly progresses to primary septicemia with symptoms that include high fever, chills, decreased blood pressure and septic shock [5], [7]–[9]. These infections result in a much higher mortality rate (≥50%) with rates as high as 100% in the absence of antibiotic therapy [5], [6], [10]. Hence, a critical aspect of V. vulnificus pathogenesis is its ability to infect a host via the gastrointestinal tract and then rapidly spread from the small intestine to the blood stream.
Although several secreted virulence factors of V. vulnificus have been identified [3], [11], only two have been previously associated with increased death during intestinal infection: the secreted cytolytic/hemolysin pore-forming toxin encoded by vvhA [12] and the multifunctional autoprocessing RTX (MARTXVv) toxin encoded by gene rtxA1 [13]–[15]. In vitro, both of these toxins are cytolysins associated with lysis of a variety of cell types including erythrocytes, epithelial cells and macrophages, albeit by different molecular mechanisms [12], [16]–[22].
The role of these toxins in vivo during infection has been less well-characterized. When injected directly to the bloodstream, purified VvhA is lethal at sub-µg levels and causes hypotension and tachycardia, along with skin and pulmonary damage [12], [23]. However, deletion of vvhA from V. vulnificus had either a slight or no defect in virulence when delivered intraperitoneally (i.p.) and no defect when delivered intradermally (i.d.) [23]. When delivered intragastically (i.g.) to neutropenic mice, loss of vvhA resulted in a detectable, albeit modest, 4–5 fold increase in median lethal dose (LD50) [12], [23].
In comparison to VvhA, MARTXVv has been shown to have a significantly greater contribution to mouse lethality. A mutant in rtxA1 has a 100- to 500-fold increase in LD50 compared to wild-type when inoculated i.p. [14], [17], [18] and a 13-fold increase when inoculated subcutaneously (s.c.) [19]. A deletion of rtxA1 caused a 180 to 2600-fold increase in LD50 in an i.g. infection model with the contribution of the gene deletion to virulence varying depending on the specific toxin variant that is expressed [15]. Comparison across different studies suggest that the MARTXVv toxin is the most significant virulence factor of V. vulnificus and both MARTXVv and VvhA exert a greater effect on i.g. and septicemic infection compared to i.p., s.c. or i.d. infection.
In this study, we sought to understand how cytotoxins MARTXVv and VvhA contribute to food-borne infection by highly virulent V. vulnificus strains that produce a particularly potent variant of the MARTXVv toxin [15]. We used bioluminescence imaging (BLI) and measurement of bacterial colonization to monitor early events in growth and dissemination of V. vulnificus strain CMCP6 in mice after i.g. infection. This study shows that both MARTXVv and VvhA from CMCP6 contribute to the onset of colonization and to significant bacterial growth sooner after inoculation. These data are consistent with a role of both toxins in disabling innate immune cells in the small intestine allowing for more rapid growth. However, the effect of the toxins is not limited to innate cells as these toxins are also here shown to directly cause epithelial tissue damage. The combination of rapid growth and tissue damage is essential for the dissemination to the bloodstream earlier during the infection cycle and this rapid dissemination is the leading factor promoting death.
V. vulnificus CMCP6 can cause lethal infection of adult mice inoculated i.g. [15]. To more directly measure how disease progresses during early infection, we transformed V. vulnificus CMCP6 with plasmid pHGJ1 that expresses the Photorhabdus luminescens lux genes from the constitutively active Vibrio cholerae ompC promoter (see material and methods). The LD50 for the resulting strain CMCP6lux (HG0905) was 3.1×105 CFU (Table S1), about 13-fold higher than the LD50 of 2.4×104 CFU previously determined for the parent strain CMCP6 in this mouse infection model [15]. The presence of the plasmid also caused an in vitro defect in growth in antibiotic-free culture media (Figure S1). The difference in both in vitro growth and in vivo virulence between the parent strain CMCP6 and CMCP6lux (HG0905) is likely due to spontaneous bacterial death upon plasmid loss. To maintain the lux reporter without antibiotic selection, the lux plasmid (pHGJ1) carries the hok/sok plasmid addiction system [24]. Bacteria that lose the plasmid during cell division will die upon dilution of the less stable antitoxin. The advantage of this system is that only bacteria that produce luciferase survive and thus there is no contribution to infection from bacteria that are not lux+. In addition, the expression of the luciferase genes also probably contribute to the reduced virulence of HG0905 compared to CMCP6 since a CMCP6 that carries a plasmid deleted of the lux genes (HG0909) was slightly more virulent than HG0905 (Figure S1).
In this paper, we compared lux+ strains derived from parent strain CMCP6 and we confirmed that there was no in vitro growth difference in mutant strains CMCP6luxΔrtxA1 (HG0906), CMCP6luxΔvvhBA (HG0907) and CMCP6luxΔrtxA1vvhBA (HG0908) containing plasmid pHGJ1 in antibiotic-free culture media (Figure 1G) compared to the isogenic parent CMCPlux (HG0905). Further, we confirmed by lactate dehydrogenase (LDH) release assays that the mutants carrying pHGJ1 have defects in HeLa cell lysis consistent with previous findings [14], [18]. Specifically, the ΔrtxA1 mutation reduced rapid HeLa cell lysis while the ΔvvhBA mutation eliminated slow cell lysis. The double mutant did not lyse cells (Figure S2).
The effect on virulence of the lux+ plasmids was less evident in the mutant strains since these strains likely are not growing in vivo and thus less likely to lose the plasmid during rapid replication. Thus, the LD50 for CMCP6luxΔrtxA1 (HG0906) matched our previously determined LD50 of 8.0×107 CFU for plasmid-free CMCP6ΔrtxA1 and thus the deletion of rtxA1 exhibited only a 260-fold effect on virulence due to the reduced virulence of the isogenic wild type. Deletion of vvhA caused a 61-fold decrease in virulence with an LD50 of 1.9×107. This result was surprising since previous studies of i.g. infection with a ΔvvhA mutant in isolate YJ016 revealed only a modest 5-fold virulence defect [12]. However, the previous study was conducted in highly susceptible iron-overloaded, neutropenic mice, which may have masked the importance of this factor for intestinal infection. Consistent with a role of both factors in infection by CMCP6, the CMCP6luxΔrtxA1ΔvvhBA double mutant (HG0908) was essentially avirulent with an LD50>109 (Table S1).
To monitor how rapidly V. vulnificus bacteria expand in vivo, bioluminescence from mice infected with CMCP6lux (HG0905) was observed and quantified at defined intervals using an IVIS 100 bioluminescence imager (Xenogen Corp.). As previously described by others [25], use of anesthesia during orogastric inoculation can result in accidental lung infection due to contamination of the larynx during infection. In pilot studies, we similarly found that some mice developed lung infection. These mice usually died rapidly, often by 6 hours after infection. In our study, 3 mice infected with HG0905 developed infection of the lung detectable by IVIS imaging by 4 hr. These mice were euthanized and removed from analysis. All other mice did not show detectable lung infection by IVIS imaging.
At a dose of 1×106 CFU, 100% of mice with an intestinal infection died between 12 and 22.5 hr post-inoculation (Figure 1B). Prior to death, all of the CMCP6lux (HG0905) inoculated mice (6/6) showed detectable levels of photon flux. Two of the mice reached our preset detection limit by 4 hr, 3 mice by 8 hr, and all mice had detectable levels by 12 hr (Figure 1A and B). Of note, all mice showed a steady rise (mean slope between onset and peak (m) equals 0.174; Figure 1F) in light emission until the animal was sacrificed for severe morbidity between 12 and 22.5 hr demonstrating constant replication of the wild-type bacteria. However, even though all mice reached at least 7.3 RLU (Relative Luminescence Unit on a logarithmic scale; Luminescence Unit represents the photons s−1 cm−2 sr−1) before death, attainment of this level did not predict eminent death as three mice survived for 4–22.5 hr after crossing this threshold.
To demonstrate that photon flux is representative of changes in intestinal colonization, in a separate experiment, 5 mice inoculated with 106 CFU were euthanized at both 8 and 12 hr. In accordance with flux readout in the previous experiment (Figure 1A and B), there was variability in recovered CFU from the small intestine at 8 hr ranging from 104–108 CFU or −2 to +3 log unit change from the inoculation dose (Figure 2A). By 12 hr, all mice were colonized above the infection dose representing 2–5 log units growth (Figure 2B). In addition, there was dissemination of the bacteria to the liver and spleen by 8 hr, indicating progression to septicemia in all mice during the earliest stages of infection (Figure 2C and D). Overall, V. vulnificus CMCP6lux (HG0905) was shown to expand in vivo and this rapid growth occurred coincident with dissemination to other tissues shortly after inoculation.
To determine if MARTXVv is the factor that promotes the rapid growth seen in mice after i.g. infection, we monitored the effect of deletion of the rtxA1 gene from CMCP6lux on disease progression. The most apparent phenotype of the resulting strain HG0906 compared to CMCP6lux (HG0905) was a delay in the time required to the BLI detection limit (Figure 1A). One of 9 mice was sacrificed due to lung infection and one mouse was not infected. Among the 7 infected mice, light production was detectable in only one (14%) by 4 hr and 3/7 (43%) of mice by 8 hr (Figure 1C). Four mice (50%) showed delayed onset with detectable light emission only after 8–12 hr. After onset, the average rate of growth in all infected mice was similar to wild-type (m = 171, Figure 1F). However, unlike CMCP6lux (HG0905) infected mice, 3 of 7 mice ultimately survived to 36 hr despite the in vivo bacterial load. In addition, several of the mice succumbed only late in the experiment indicating, as suggested by LD50 studies (Table S1), that more mice might have survived except for the stress imposed by repeated anaesthetic regimen necessary for imaging. Thus, the major effect of loss of the MARTXVv toxin was delayed onset of bacterial growth to detectable levels. Further, among the mice that attained high bacterial loads, half failed to progress to death and the mice cleared the infections. As further evidence for delayed growth, there was a trend toward reduced colonization of the small intestine by the CMCP6luxΔrtxA1 mutant (HG0906) at 8 hr that reached statistical significance by 12 hr (Figure 2A and B). In addition, by 8 hr, there was significantly reduced dissemination of bacteria to the liver and spleen (Figure 2C and D). Overall, our results suggest that loss of rtxA1 results in an inability to consistently establish an infection that can progress to other organs shortly after ingestion of bacteria and thus the infections are delayed and less severe at least 50% of the time.
To test if VvhA also contributes to infection, we next tested a CMCP6luxΔvvhBA mutant (HG0907). Results were intermediate between CMCP6lux and the isogenic ΔrtxA1 mutant with 7 of 9 mice showing increased light emission beginning 4–12 hr after inoculation rising to values greater than 7.3 RLU. Similar to both CMCP6lux and the isogenic ΔrtxA1 mutant, the 7 mice successfully infected with CMCP6luxΔvvhBA mutant showed a similar rate of increasing light emission with other strains indicating it does grow in vivo (m = 0.142; Figure 1F). 1 of these 7 mice reversed course and began to clear the infection while the other 6 succumbed to infection by 22.5 hr. When assessed for colonization, there was a consistent trend for reduced colonization of CMCP6luxΔvvhBA mutant (HG0907) in the intestine at 8 hr and 12 hr (Figure 2A and B) and reduced dissemination during early infection to the liver and spleen but these values did not achieve statistical significance (Figure 2C and D). Thus, mice infected with the ΔvvhBA mutant showed detectable decreases in numerous parameters of infection including delayed and reduced death but this cytolysin does not exert the same impact on progression of CMCP6 infection as MARTXVv.
Despite its minimal effect when the rtxA1 is intact, expression of VvhBA by CMCP6lux does account for the residual virulence of the CMCP6luxΔrtxA1 mutant. A CMCP6lux double mutant eliminating both vvhBA and rtxA1 (HG0908) was nonlethal in mice at 1×106 CFU (Figure 1E) except for one mouse sacrificed due to lung infection. Three of the mice were overall defective for bacterial growth and did not achieve 7.3 RLU at any time point and 1 was not infected at all. In 2/8 (25%) mice there was long 12 hour delay to detectable light production (Figure 1E). When bacteria did expand in vivo, the mean slope from onset of detection to peak (m = 0.164) was similar with those of wild type (Figure 1F). However, in all cases where mice did achieve high bacteria loads, the emission of light reversed from a peak between 15 and 22.5 hr post-inoculation and all mice ultimately cleared the infection.
When tested for colonization, CFU recovered from the small intestine were significantly reduced at both 8 and 12 hr (Figure 2A and B) and the bacteria did not disseminate to the liver and spleen by 8 hr (Figure 2C and D). Overall, these data indicate that in V. vulnificus strain CMCP6, MARTXVv in conjunction with a secondary additive contribution from VvhA is essential during the early stages of infection to promote initiation of the infection and dissemination to the bloodstream.
Lack of bacterial growth during in vivo infection due to loss of secreted factors can often be restored by co-infection wherein mutant bacteria benefit from alteration to the host environment by the co-infecting strain. These data can reveal that mutant bacteria are not defective in their ability to replicate in vivo per se, but lack the capacity to modify the host environment to promote their growth. To test if HG0908 could be restored for in vivo growth by co-infection, we transferred a plasmid from which the luxCDABE operon was deleted (pHGJ2) into CMCP6 and the double cytolysin mutant strains and competed strains 1∶1 with the lux+ double mutant (HG0908). Thereby, if HG0908 is rescued by co-infection, total flux during co-infection should increase since all light signal would originate from HG0908 and not the cytolysin producing co-infecting strain.
When 5×105 CFU of lux− wild-type (HG0909) and 5×105 CFU of lux+ double mutant (HG0908) were co-inoculated, median light production produced by the double mutant was 6.9 RLU at 8 hr post-infection and reached 7.6 RLU after 12 hr infection. By contrast, mice infected with 5×105 CFU of lux+ double mutant (HG0908) and 5×105 CFU lux− double mutant (HG0910) was 5.7 RLU at 8 hr post-infection and 5.9 RLU after 12 hr (Figure 2E and F).
Thus, the in vivo growth defect of the double cytolysin mutant HG0908 can be restored by co-infection with a cytolysin producing strain indicating that HG0908 is not defective in its ability to replicate in vivo but in its ability to modify the host environment.
We have demonstrated that MARTXVv and VvhA of V. vulnificus CMCP6 are required for earlier onset of in vivo growth after i.g. inoculation. This result is consistent with a recent study conducted by s.c. infection to model wound-induced infection using strain YJ016. The s.c. study of YJ016 suggested the requirement for MARTXVv during infection is primarily for protection from phagocytes to promote growth [19]. This conclusion seems to conflict with evidence that cytotoxins of CMCP6, YJ016, and other V. vulnificus strains are linked to lysis of both epithelial cells and macrophages in vitro [14], [16]–[19], [21]. To reveal whether the cytotoxins have an additional role beyond promoting rapid growth during intestinal infection, mice were inoculated with a lethal dose of CMCP6lux (HG0905) and the terminal ileal tissue was collected after 8 hr infection for various histopathological staining. Severe disruption of the intestinal barrier occurred in the ileum infected with HG0905 (Figure 3A–C) with many broken villi and barrier disruptions, consistent with pathology reported in earlier studies in neutropenic mice using strain YJ016 [12]. Excessive amounts of epithelial cell debris and heavy cellular infiltration of lamina propria were observed in the lumen and mucosa of the ileum from mice infected with the wild type (Figure 3A–C). Staining with anti-CD45 showed extensive influx of monocytes and other immune cells to the tissue and the lumen (Figure 3D). Within the destroyed tissue, F4/80 positive macrophages are present and proinflammatory cytokine IL-1β is secreted and found distributed in the ruptured tissue (Figure 4A). The lumen is filled with epithelial debris (stained positive with β-catenin), lysed macrophages, and IL-1β presumably release from necrotic macrophages (Figure 3C and 4A).
By contrast, mice infected with 1×106 CFU of either the CMCP6luxΔrtxA1 mutant (HG0906) or the CMCP6luxΔvvhBA mutant (HG0907) showed no destruction of the villi architecture and infiltration of the lamina propria except only slight swelling (Figure 3E and F). Mice infected with double mutant HG0908 showed no pathology distinct from PBS mock control (Figure. 3G and H). However, the absence of tissue damage cannot be conclusively linked to the toxins by this approach because the bacterial load of wild type in the ileum 8 hours after infection of 106 CFU would be much higher than that of single and double mutant strains due to affects of the loss of cytotoxins on bacterial growth (Figure 2A and B).
Therefore, we infected mice with increasing CFU so that the bacterial load in ileum at the point of euthanasia would be equalized. In mice infected with a CMCP6luxΔrtxA1ΔvvhBA double mutant (HG0908) at a dose of 5×107 or 1×109 CFU (n = 6), there was no tissue damage in the ileum of any of the mice (Figure 5A) and the epithelial lining appears similar to the PBS mock infected control group (Figure 3H and I). Notably, even at the dose high enough to kill 1/6 mice in 8 hour, no tissue damage occurred. This finding is significant because it indicates that no other secreted protease or toxin produced by strain CMCP6 is sufficient to cause visible tissue damage in the absence of MARTXVv or VvhA, even when a concentration of bacteria in the lumen exceeded that normally found for wild type by 8 hr post inoculation.
Staining shows macrophages are present in the lamina propria at low dose infection of double mutant and are secreting only low amounts of IL-1β consistent with the low levels of colonization at 8 hours post inoculation (Figure 4A). By contrast, at a high dose, macrophages are less apparent in the lamina propria and may have moved to the lumen, where high density staining of IL-1β is seen (Figure 4). This focusing of a proinflammatory immune response to the lumen is an effective response since infected mice are surviving a dose that would kill 100% within 8 hr if infected with CMCP6lux (HG0905).
To determine whether MARTXVv and/or VvhA is directly responsible for the tissue damage caused by CMCP6lux, similar increasing dose infections were performed with the single toxin deletion strains. When 1×109 CFU of either the isogenic ΔrtxA1 mutant (n = 5) or the ΔvvhBA mutant (n = 5) was inoculated, 40% or 80% of the mice died within 8 hr post-infection, respectively (Figure 5B and C). The difference in survival compared to the double mutant HG0908 shows that the toxins are able to function independently while the difference between HG0906 and HG0907 further exemplifies the relative import of MARTXVv over VvhA for overall survival of CMCP6lux.
At the intermediate infection dose of 5×107 CFU, the median number of each single mutant recovered from the small intestines were 8.1 and 8.2 Log CFU/organ, respectively (Figure 5B and C); not significantly different from the 7.9 Log CFU/organ recovered from CMCP6lux (HG0905) inoculated at only 1×106 CFU. Note that these median values were calculated from surviving, colonized mice that were sacrificed for histopathology and do not include mice that rapidly succumbed to infection or noncolonized mice with recovered CFU below the detection limit. In the CMCP6luxΔrtxA1 mutant infected mice, only a small portion of the villi showed necrotic epithelial cells and hypercellularity in the lamina propria (Figure 5B) indicating, in the absence of MARTXVv, VvhA induces mild tissue damage accounting for the modest dissemination of this mutant to the liver (Figure 2C). By contrast, samples of intestines from mice infected by the MARTXVv+ strain ΔvvhBA mutant showed sloughed villi and an infiltration of the lamina propria into the lumen indicating MARTXVv induces tissue damage that is more severe than associated with VvhA (Figure 5C). This finding is consistent with the significant ability of this mutant to disseminate to the liver (Figure 2C). Importantly, we did not observe the severe tissue destruction similar to that found in the CMCP6lux (HG0905) infected group in any of these mice suggesting that both cytotoxins target intestinal epithelial cells and both cause tissue damage that is additive or possibly even synergistic.
Successful and rapid in vivo growth of V. vulnificus is generally regarded as an essential step in its pathogenesis [11], [26]. We developed a BLI system using the highly virulent V. vulnificus strain CMCP6 to directly observe how rapid in vivo growth during early infection can influence the outcome of infection. Wild-type V. vulnificus CMCP6lux infection expanded quickly in mice very early after intestinal infection and all the animals progressed to lethality. By contrast, deletion of one or both of the rtxA1 and vvhA genes led to bacteria with a in vivo growth delay leading to reduced CFU in animals by 8–12 hr post infection in the small intestine and other organs, although there was no in vitro growth defect (Figure 1). However, while some mice infected with strains missing just one toxin showed little or no growth, many mice still died from infection and others that survived infection emitted a high level of light up to 15 hr (Figure 1A–E). In these mice with detectable light, the slope of light emission representing the growth rate was similar to wild type from first detection to peak infection. Furthermore, co-infection studies revealed the mutant that produces no cytotoxins has the capacity to grow in vivo, but does not have the capacity to alter the host environment to promote its own growth. These findings suggest that the cytolysins cause another phenomenon beyond simply manipulating bacterial load in the animal and that this event is important to cause lethal infection.
Previous studies indicate that dissemination of infection to the liver is a major predictor of mouse mortality after wound infection [9], consistent with clinical reports indicating that hepatic hemorrhage is a frequent cause of death of patients after both wound and food-borne V. vulnificus infection [27]–[29]. MARTXVv and VvhBA from strain CMCP6 are here shown to not only be associated with enhanced in vivo growth, but also with necrosis of tissue in the small intestine and translocation of V. vulnificus from the small intestine to the liver (Figure 2C). Notably, the small intestine has already been recognized as the site of the most severe tissue necrosis in human autopsy of V. vulnificus patients [27].
Although both single toxin gene mutants induced from moderate to severe necrosis and dissemination, the double mutant was completely restricted to the intestine and no damage was evident (Figure 3). Close examination of Figure 1A suggests that light signal from the double mutant occurred predominantly in the lower abdomen compared to wild type light emission from the mid-abdomen, an observation consistent with the ability of the double mutant to grow in vivo during transit through the upper and lower bowel, accounting for the increased light signal, but the infection never progressed to the liver.
We next sought to understand if the role of the MARTXVv in dissemination of CMCP6 during gut infection was to increase the growth rate within the small intestine during the first few hours to create a larger pool of bacteria to express VvhA, proteases or other cytolysins to promote dissemination as proposed by Lo et al. [19] for wound related infections or, if the role of the toxins is to utilize the cytolytic activity to directly lyse intestinal epithelial cells to create a pathway through which the bacteria could disseminate as proposed by Kim et al. [14]. Our study found that, for strain CMCP6, both MARTXVv and VvhA function additively to cause intestinal tissue necrosis. We also found that in vivo growth does occur in the absence of toxins but is restricted to the intestine, possibly to the colon as recovered CFU from the small intestine at 12 hr was decreased compared to wild type despite strong light emission in some animals. Indeed, studies examining wound infection with strain YJ016 also show less dermal tissue damage upon deletion of rtxA1, but the effect was negated as secondary to the effect on decreased bacterial load [19]. Using identical inocula, we came to the same conclusions. It was only after we used increased inocula such that bacterial load at the time of euthanasia was equivalent that the role of toxins on tissue necrosis became evident. Thus, it is possible that MARTXVv and VvhA will be shown to also be involved in tissue necrosis during wound infection, at least for strains CMCP6 and YJ016. However, lethal dose studies have shown that MARTXVv and VvhA are in general less important to wound induced lethality than i.g. infection [12], [15], [18], [19], [23], supporting the conclusion by Lo et al. [19] that alternate factors have a significant role during wound infection.
While our data support that the cytotoxins target the intestinal epithelium, our results do not negate that the toxins have a significant role in innate immune defense as well. Both in vitro and in vivo, V. vulnificus is also known to cause killing of phagocytic cells [19], [21]. In vitro, both toxins are known to induce NLRP3 dependent caspase-1 activation resulting in necrosis of macrophages [21]. The absence of phagocytes in hepatic tissue has been previously noted as a factor that can contribute to patient mortality [30]. However, our study reveals that V. vulnificus is inducing massive inflammation, leading to recruitment of monocytes, neutrophils and F4/80-positive macrophages. These results are consistent with an increase of the proinflammatory cytokines TNF-α, IL-6 and IL-1ß that were detected in the sera of V. vulnificus septicemic patients [31]. In our study, the increase of IL-1ß secretion in mice inoculated with a lower concentration of CMCP6lux suggest that in vivo, the action of toxins against macrophages induces pyroptosis, just as it does in vitro [21].
A question that remains is whether the toxins are simultaneously promoting inflammation while attempting to keep it at bay by killing the recruited cells. A recent study has revealed that some gut pathogens specifically induce inflammation as a mechanism to promote rapid growth. Salmonella is known to induce pyroptosis leading to inflammation [32]. This inflammation then allows Salmonella to use tetrathionate respiration in the anaerobic environment of the gut, which promotes bacterial replication and transmission [33], [34]. In the present paper, the tissue damage of villi in small intestine was clearly apparent (Figure 3 and 5) and the inflammation as early as 8 hr post infection is severe (Figure 3D). Thus, while killing of phagocytes is one mechanism that would promote rapid in vivo growth, particularly in the bloodstream, it is possible that inflammation itself may promote growth in the intestine. The ttr genes that encode the tetrathionate respiration system necessary for Salmonella to grow in response to inflammation [34] are present in V. vulnificus YJ016 on what appears to be a pathogenicity island [33], [35]. However, the other sequenced V. vulnificus strains [36], including CMCP6 used in this study, do not seem to have acquired this island. If the fact that this pathogen induces the host inflammation to promote their own outgrowth is a general consequence, our results suggest that a novel mechanism unrelated, (or in the case of ttr+ YJ016 in addition to tetrathionate) is required, depending on the strain isolate.
A final important finding of our work is the evidence that MARTXVv and VvhBA are directly linked to death of the host. One mechanism that accounts for the linkage to cell death is that the cytotoxins promote movement of the bacteria to bloodstream leading to primary septicemia and septic shock. High bacterial load in the bloodstream and high serum TNFα concentrations have been directly linked to death of patients [37], [38]. In addition to septic shock, the toxins might also contribute to multiorgan failure. This can include necrosis of lung tissue and liver, a common finding in autopsy patients [28], [39]. They could also cause progression of the infections out of the bloodstream into the muscle tissue to cause necrotizing fasciitis, another complication of V. vulnificus infections.
Overall, the present study demonstrates that, for V. vulnificus isolate CMCP6, MARTXVv, along with VvhBA, performs an essential role during food-borne V. vulnificus infection after consumption. These toxins have multiple roles including promotion of rapid in vivo growth, destruction of epithelial tissue, causing inflammation through induction of pyroptosis, and possibly causing patient death through tissue destruction in peripheral organs.
Similar studies performed in other V. vulnificus clinical isolates will be necessary to determine if findings here with the highly virulent strain CMCP6 will be broadly applicable to all V. vulnificus clinical isolates. In particular, we recently found that the predominance of US clinical isolates from patients with primary septicemia (represented by strain MO6-24/O) carry a variant of the MARTXVv toxin that arose by a recombination in the rtxA1 gene with the rtxA gene from Vibrio anguillarum. This recombination likely accounts for an overall 10-fold reduced virulence of MO6-24/O in this animal i.g. infection model compared to CMCP6 [15]. Notably, the loss of a domain of unknown function from the MO6-24/O-type MARTXVv variant did not affect the function of MARTXVv as a cytolysin in vitro [17] suggesting it would likely retain the ability to induce necrotic tissue damage during intestinal infection.
However, it is possible that the reduced potency of the toxin variant could impact the relative contribution of the toxin to in vivo growth and tissue damage such that, perhaps, MARTXVv and VvhA could be found to have a more equal contribution to intestinal infection by MO6-24/O and similar strains, although we predict that both toxins would continue to have an additive contribution to virulence. Alternatively, the kinetics of infection could be altered such that critical levels of colonization and/or damage will occur later in the infection cycle.
In addition to MO6-24/O and CMCP6-type MARTXVv variants, other variants of the toxin have been described that have undergone more significant changes by horizontal gene transfer and homologous recombination, including the acquisition of the ability to covalently crosslink actin in epithelial cells [15], [40]. One might predict that these rare variants will have an even more distinct infection profiles when compared to CMCP6. In any event, results in this study have established that both MARTXVv and VvhA contribute to virulence and provide a baseline for determining if other isolates have similar patterns of disease progression or whether V. vulnificus infection develops differently dependent upon the variant of MARTXVv it expresses.
Finally, while it has been shown here that MARTXVv and VvhA are critical to infection, these are very likely not the only important virulence factors necessary for intestinal infection. Most notably, V. vulnificus Biotype 1 can be separated into two distinct evolutionary lineages: a clinical lineage and environmental lineage [41]. We have recently shown that bacteria from both lineages carry genes for both cytolysins [15], even though strains from the environmental lineage rarely cause clinical infection and are less virulent in mice [9]. Thus, there must be additional V. vulnificus factors to define host selection that have yet to be identified and characterized. These would then work in concert with the cytotoxins perhaps by improving growth in the human intestine during infection or by facilitating colonization of the small intestine by interacting with a human epithelial receptor. Regardless of the nature of this other virulence gene, the importance of the cytotoxins cannot be negated since our work demonstrates that the additive destruction of these toxins is essential to disease progression.
This study was carried out in strict accordance with the recommendations in the United States Public Health Service (USPHS) regulations and applicable federal and local laws. The protocol (Protocol No. 2009-1016) was approved by the Northwestern University Institutional Animal Care and Use Committee (IACUC) as detailed in methods. All surgery was performed under ketamine-xylazine and isoflurane anesthesia, and all efforts were made to minimize suffering.
The strains and plasmids used in this study are listed in Table 1. Escherichia coli strains used for DNA replication or conjugational transfer of plasmids and Vibrio vulnificus strains were grown in Luria-Bertani (LB). When appropriate, antibiotics were added to media at the following concentrations: kanamycin (50 µg/ml), rifampicin (50 µg/ml) and chloramphenicol (5 µg/ml). Bacterial growth in LB was monitored using a Beckman DU530 Spectrophotometer.
To inactivate rtxA1, vvhBA and rtxA1vvhBA, overlapping PCR was applied for the construction of rtxA1 (HG0901), vvhBA (HG0902) and rtxA1vvhBA (HG0903) deletion mutants [42] (Table 1). The 9635 bp deleted rtxA1 and the 793 bp deleted vvhBA open reading frame (ORF) were ligated with SalI-SacI and XbaI-SacI digested pDS132 [43] forming pHGJ3 and pHGJ4. To generate the ΔrtxA1 and ΔvvhBA mutants by homologous recombination, E. coli SM10λpir and S17λpir (containing pHGJ3 and pHGJ4) were used as a conjugal donor to V. vulnificus CMCP6 with spontaneous rifampicin resistance. The ΔrtxA1vvhBA double mutant was also generated through conjugation of pHGJ3 to HG0902 (Table 1). The conjugation and isolation of the transconjugants were conducted using sucrose counter selection previously described [42].
A pCM17 containing luxCDABE and hok/sok plasmid [24] was used for generation of bioluminescent V. vulnificus strains. To create the conjugatable plasmid, 251 bp of oriT DNA from pGP704 was inserted into NheI-SalI digested pCM17 to create pHGJ1. pHGJ1 was then digested with HindIII followed by religation to inactivate the luciferase genes and create pHGJ2 (Table 1). CMCP6lux (HG0905) and isogenic rtxA1 (HG0906), vvhBA (HG0907) and rtxA1vvhBA (HG0908) mutants were generated by conjugal transfer of pHGJ1 and HG0909 and HG0910 were generated by conjugal transfer of pHGJ2 (Table 1).
The roles of the V. vulnificus CMCP6 MARTXVv and VvhA in pathogenesis were examined using a mouse model. Female C57BL/6 mice (5–6 weeks old, Harlan, Indianapolis, IN) were individually anesthetized with an i.p. injection of 100 µl of PBS solution containing 10 µg/ml ketamine and 2 µg/ml xylazine per mouse, i.g. inoculated with 50 µl of 1×106 CFU of the indicated V. vulnificus strains. Images were acquired using an IVIS 100 (Xenogen Corporation, Alameda, CA). During image acquisition, mice were anesthetized using ketamine-xylazine cocktail at each image cycle. All images were acquired at a preset exposure of 20 sec with medium binning and f/stop = 1 so images could be compared over time. Photons per second emitted by each mouse were quantified and analyzed by defining regions of interest (ROI), using the Living Image 1.0 software. Severely moribund mice unlikely to survive to the next imaging cycle were euthanized after imaging and counted as non-survivors.
To observe the tissue damage in mice small intestine, infected mice were sacrificed at specific time points and 1 cm of ileum immediately adjacent to the cecal-ileal junction was fixed by 10% neutral phosphate buffered formaldehyde solution (Sigma, St. Louis, MO) for 16 hr. Histopathology was performed at the Northwestern University Pathology Core Facility. The ileum was embedded in paraffin, and stained with hematoxylin and eosin (H&E).
A single immunohistochemical staining procedure was performed to characterize the necrotized cells and to detect the cytokines secretion. Briefly, tissue sections were placed in a 60°C oven overnight for tissue to adhere. The sections were dewaxed in xylene, rehydrated through graded alcohols to water. Antigen retrieval was done by placing the slides in citrate buffer and pressure cooked up to 125°C for 30 sec and gradually reduced to 90°C over 40 min. Slides were then cooled down at room temperature for 20 min and placed in DAKO butter (DAKO, Carpineteria, CA). Then immunostaining for ß–catenin, CD45, F4/80, and IL-1ß was performed on a DAKO Autostainer Plus using a DAKO Envision system (DAKO). Sections were first quenched with hydrogen peroxide (H2O2) for 10 min and incubated with primary antibodies for 60 min. Primary rabbit polyclonal antibodies to ß-catenin, CD45 and IL-1ß (Abcam, Cambridge, UK), at 1∶50 dilution, and rat monoclonal antibodies to F4/80 (Abcam), at a 1∶100 dilution, were used. Secondary antibodies were applied at a dilution of 1∶200 for 30 min, followed by incubation with polymer link streptavidin-horseradish peroxidase (HRP) reagent and 3,3′-diaminobenzidine (DAB; DAKO). The slides were counter-stained with blue Mayer's Hematoxylin and primary antibodies were omitted in negative controls. Then the stained slides were photographed using a Zeiss Axioskop (MicroImaging, Thornwood, NY) microscope with Nuance spectral camera (CRI, Woburn, Mass).
Mouse colonization assays were performed essentially as described earlier for Vibrio cholerae infection [44]. Briefly, five C57BL/6 female mice (5–6 weeks old, Harlan, Indianapolis, IN) per each group were euthanized by cervical dislocation under anesthesia at 8 or 12 hr after inoculation of the indicated V. vulnificus strains or PBS. After terminal ileum dissection for histology, the liver, spleen and remaining small intestine were dissected. Then it was homogenized in 5 ml (small intestine and liver) or 3 ml (spleen) of PBS and serially diluted for plate counts of recovered colony forming units (CFU) on LB plate containing rifampicin. Mice for which fewer than 10 colonies were recovered from 50 µl of the homogenated extract were plotted below the detection limit line. Recovery of bacteria is reported as a Colonization Index (Col. Ind.) calculated as CFUrecovered/CFUinoculated or CFU/organ at logarithmic scale. The remaining homogenates of small intestine was centrifuged at 13400×g for 5 min at 4°C and the supernatant were kept at −80°C. The supernatants were thawed on ice immediately prior to assay. IL-1ß levels in small intestines were determined from homogenated extracts by ELISA (Enzyme-linked immunosorbent assay, BioLegend, San Diego, CA) kits according to manufacturer's instruction.
Virulence of lux+ V. vulnificus strains was determined in a morbidity assay as previously described [15] and the LD50 for each strain was calculated by the method of Reed and Muench [45].
To examine the cytotoxicity of lux+ V. vulnificus strains, the HeLa cells were grown in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum and seeded in 12 well culture plates to a density of 8.5×105 cells per well. After growing overnight at 37°C in 5% CO2, the monolayer of HeLa cells were infected with lux+ V. vulnificus strains at a multiplicity of infection of 25 and the cytotoxicity was then determined by measuring the activity of LDH in the supernatant at 1 to 5 hr post-infection using a CytoTox 96 Non-Radioactive Cytotoxicity Assay Kit (Promega, Madison, WI) according to manufacturer's instructions.
All data were graphed and analyzed using GraphPad Prism 4 for MacIntosh Software (San Diego, CA). Statistical significance for LDH assays, growth curve, and ELISA assays was determined in pairwise comparisons using a student t-test. A Mann-Whitney non-parametric t-test comparing means was used for mouse colonization studies. Significance of survival curves was determined using the log-rank test.
|
10.1371/journal.pntd.0003918 | Haemophilus ducreyi Cutaneous Ulcer Strains Are Nearly Identical to Class I Genital Ulcer Strains | Although cutaneous ulcers (CU) in the tropics is frequently attributed to Treponema pallidum subspecies pertenue, the causative agent of yaws, Haemophilus ducreyi has emerged as a major cause of CU in yaws-endemic regions of the South Pacific islands and Africa. H. ducreyi is generally susceptible to macrolides, but CU strains persist after mass drug administration of azithromycin for yaws or trachoma. H. ducreyi also causes genital ulcers (GU) and was thought to be exclusively transmitted by microabrasions that occur during sex. In human volunteers, the GU strain 35000HP does not infect intact skin; wounds are required to initiate infection. These data led to several questions: Are CU strains a new variant of H. ducreyi or did they evolve from GU strains? Do CU strains contain additional genes that could allow them to infect intact skin? Are CU strains susceptible to azithromycin?
To address these questions, we performed whole-genome sequencing and antibiotic susceptibility testing of 5 CU strains obtained from Samoa and Vanuatu and 9 archived class I and class II GU strains. Except for single nucleotide polymorphisms, the CU strains were genetically almost identical to the class I strain 35000HP and had no additional genetic content. Phylogenetic analysis showed that class I and class II strains formed two separate clusters and CU strains evolved from class I strains. Class I strains diverged from class II strains ~1.95 million years ago (mya) and CU strains diverged from the class I strain 35000HP ~0.18 mya. CU and GU strains evolved under similar selection pressures. Like 35000HP, the CU strains were highly susceptible to antibiotics, including azithromycin.
These data suggest that CU strains are derivatives of class I strains that were not recognized until recently. These findings require confirmation by analysis of CU strains from other regions.
| Cutaneous ulcers (CU) in children living in equatorial Africa and the South Pacific islands have long been attributed to yaws, which is caused by Treponema pallidum subsp. pertenue. However, PCR-based cross sectional surveys done in yaws-endemic regions show that Haemophilus ducreyi is the leading cause of CU in these regions. H. ducreyi classically causes the genital ulcer (GU) disease chancroid and was once thought to be exclusively sexually transmitted. We show that CU strains obtained from Samoa and Vanuatu are genetically nearly identical to class 1 GU strains and contain no additional genetic content. The CU strains are highly susceptible to antibiotics, including azithromycin. The data suggest an urgent need to obtain and analyze CU isolates from Africa and other countries in the South Pacific and to search for environmental sources of the organism.
| Haemophilus ducreyi classically causes chancroid, a sexually transmitted disease that presents as painful genital ulcers (GU), which are often accompanied by infected regional lymph nodes. Although the current global prevalence of chancroid is undefined due to syndromic management of genital ulcer disease and lack of surveillance programs, the worldwide prevalence of chancroid has declined over the last decade [1]. In addition to causing its own morbidity, chancroid facilitates the acquisition and transmission of the human immunodeficiency virus type 1 [1].
In addition to causing chancroid, H. ducreyi has been isolated from or its DNA has been detected in chronic cutaneous ulcers (CU) in yaws-endemic regions in the South Pacific islands and equatorial Africa [2–7]. Yaws is a chronic infection of skin, bone, and cartilage that occurs mainly in poor communities in tropical areas of Africa, Asia, and Latin America; yaws is caused by Treponema pallidum subspecies pertenue, which is closely related to T. pallidum subsp. pallidum, the cause of venereal syphilis. A prospective cohort study by Mitjà and colleagues in yaws-endemic villages of Papua New Guinea showed that H. ducreyi is a major cause of chronic CU in children younger than 15 years old [6]. In that study, nearly 60% of patients with ulcers had detectable lesional H. ducreyi DNA, while only 34% were positive for lesional T. pallidum subsp. pertenue DNA. Approximately 2% of the total population and more than 7% of the children aged 5–15 years had ulcers positive for H. ducreyi as detected by PCR. Similar findings were reported from yaws-endemic communities in the Solomon Islands [8].
Mass drug administration (MDA) of oral azithromycin (AZT) for yaws in Papua New Guinea with a population coverage rate of 84% reduced the prevalence of CU by 90% [9]. Although MDA significantly reduced the proportion of ulcers with T. pallidum subsp. pertenue DNA, the proportion of ulcers containing H. ducreyi DNA was not affected [9]. The presence of H. ducreyi-positive CU was also reported from districts of Ghana that had received several rounds of MDA of AZT for trachoma [7]. These data raise the possibility that CU strains may be resistant to AZT, exist in an environmental reservoir, or are so infectious that MDA at the above coverage rate fails to eradicate H. ducreyi.
Multilocus sequence analysis is frequently used to determine the genetic relatedness of bacterial strains. Based on analysis of 11 H. ducreyi genes, GU strains form two genetically distinct classes, designated class I and class II, which diverged from each other approximately five million years ago (mya) and may represent distinct species [10]. A similar analysis including four CU strains suggests that they are a subset of class I GU strains [11]. However, this analysis was limited by the fact that it was based on only three informative loci.
To obtain additional insights into the evolutionary relationship of CU and GU strains, here we performed whole-genome sequencing of CU strains isolated from patients infected in Samoa and Vanuatu and archived class I and class II GU strains. Due to the persistence of CU strains after MDA of AZT, we also determined the in vitro susceptibilities of CU and GU strains to antimicrobials used for the treatment of chancroid.
The 5 CU strains used in this study were the only strains available at the time the study was initiated (Table 1); their associated clinical features are listed in S1 Table. The class I and class II strains used in this study were chosen because these strains had been previously analyzed by multilocus sequencing (Table 1) [10]. 35000HP, whose genome has been sequenced (GenBank accession no. NC_002940.2), was used as the reference strain in this study; 35000HP was isolated from a volunteer who was experimentally infected on the arm with strain 35000 and has been extensively characterized in human inoculation experiments [12, 13]. The H. ducreyi strains were grown on Columbia agar plates or in Columbia broth supplemented with 1% bovine hemoglobin (Sigma-Aldrich), 1% IsoVitaleX, and 5% fetal bovine serum (Hyclone) at 33°C with 5% CO2.
Genomic DNA was extracted from H. ducreyi strains using the DNeasy Blood & Tissue kit (Qiagen) and quantified using the Quant-It High Sensitivity dsDNA Assay kit (Life Technologies).
The sequencing libraries were prepared using the NexteraXT DNA Library Preparation kit (Illumina, Inc.) following the manufacturer’s instructions. Samples were multiplexed using the NexteraXT Dual Index Primer kit. Equimolar concentrations of indexed libraries were combined into a single pool and were sequenced at the Tufts University Genomics Core Facility. Paired-end 250-bp sequencing was performed on the Illumina MiSeq platform using the MiSeq V2 500 cycles chemistry. The de novo assembly was performed using Edena, with a customized bash script that optimizes the assembly process by optimizing three key Edena parameters [14]. The assembled contigs were annotated using the RAST online annotation tool [15].
A flow chart of comparative genome analysis of CU and GU strains is depicted in S1 Fig. For all comparative genome analyses in this study, the genome sequence of 35000HP was used as the reference. The de novo assembled contigs were ordered into Locally Collinear Blocks (LCBs) by Mauve Contig Mover (MCM) [16]. The breakpoints between LCBs were resolved by using BLAST analysis of the unaligned contigs produced by MCM, the breakpoint regions in the de novo assembled contigs, and by alignment of raw reads against 35000HP. After resolving the breakpoints, the ordered contigs were concatenated into draft genomes using Emboss 6.3.1. Pairwise genome conservation distances, which represent both gene content and sequence similarity, were estimated from draft genomes using ProgressiveMauve and plotted as heat map using CIMminer [17, 18].
The draft genome sequences for the 14 H. ducreyi strains NZS1, NZS2, NZS3, NZS4, 82–029362, 6644, HD183, HMC46, HMC56, NZV1, 33921, CIP542, DMC64, and DMC111 were deposited in GenBank under the accession numbers CP011218, CP011219, CP011220, CP011221, CP011222, CP011223, CP011224, CP011225, CP011226, CP011227, CP011228, CP011229, CP011230, and CP011231, respectively.
Genome rearrangements were identified from multiple alignments of the draft genomes generated by ProgressiveMauve, BLAST Ring Image Generator, and nucleotide BLAST and from the assembly of raw reads against 35000HP by SeqMan NGen [17, 19, 20]. Because reference-based alignment can miss additional genes that might be absent in the 35000HP genome, the de novo assembled contigs that did not align to the 35000HP genome by ProgressiveMauve were aligned against other microbial genomes using translated nucleotide BLAST.
SNPs and small insertions and deletions (indels, <10 bp) were detected using DNASTAR Lasergene (DNASTAR, Inc., Madison, WI). Briefly, the sequenced reads were assembled by SeqMan NGen against 35000HP. SNPs and indels were discovered by Seqman Pro using default parameters except that a minimum frequency of 90% reads and a minimum coverage of 50 reads were used for the analysis. SNPs were grouped as non-coding, synonymous, or nonsynonymous. Nonsynonymous SNPs were further categorized as substitutions, no-start, no-stop, nonsense, or frameshifts. All SNPs in the genomes of the CU strains were manually verified for accuracy.
Diversity analyses of whole-genome nucleotide sequences and translated concatenated coding sequences were performed using Mega 6.0 [21]. The reliability of the diversity analyses was tested using 1000 bootstrap replicates.
Recombination analysis was performed using the Phi test implemented in PhiPack and the likelihood ratio test implemented in TOPALi v2 [22, 23]. For both tests, a threshold P < 0.05 was used to define a recombination event.
Phylogenetic analyses were performed using Mega 6.0 and Realphy [21, 24]. Briefly, whole-genome alignments were imported into Mega 6.0 and subjected to model testing to identify the best-fit models of nucleotide substitution. Model testing identified Hasegawa-Kishino-Yano plus invariant sites plus gamma-distributed model as the best-fit nucleotide substitution model for our data. Using the best-fit model, phylogenetic analyses were performed with both whole-genome alignments and alignments of translated amino acid sequences from concatenated protein-coding regions using different methods of phylogeny reconstruction, including Maximum Likelihood, Maximum Parsimony, Minimum Evolution, and Neighbor Joining with different gap treatment approaches. We also inferred phylogenies using Realphy, which generates phylogenetic trees by merging alignments obtained by mapping to multiple reference genomes. A rooted Maximum Likelihood tree was reconstructed by including other Pasteurellaceae members (Actinobacillus pleuropneumoniae, Mannheimia haemolytica, Pasteurella multocida, Aggregatibacter actinomycetemcomitans, and Haemophilus influenzae) as outgroups. The reliability of all the trees generated was verified by 1000 bootstrap replicates.
The times to the most recent common ancestor (MRCA) were estimated by Bayesian molecular clock method using Beast v1.8.1 [25]. Hasegawa-Kishino-Yano plus invariant sites plus gamma-distributed model and a relaxed clock model were used to account for variation in substitution rates. The results from the Beast analysis were visualized using Tracer v1.6. A best-fit tree was identified from the tree data generated by Beast using TreeAnnotator and visualized using FigTree v1.4.2. As described previously, we used a substitution rate of 4.5 × 10−9 per site per year to calibrate the tree [26].
Selection analyses were performed using Mega 6.0 and Hyphy 2.1 [21, 27]. Briefly, protein-coding regions were extracted from the annotated genomes, ordered against 35000HP using MCM, concatenated using Emboss 6.3.1, and aligned using ProgressiveMauve [16, 17, 28]. The alignments were manually edited for accuracy to obtain a codon-delimited alignment, which was used for all the selection analyses. Rates of nonsynonymous (dN) and synonymous (dS) substitutions are widely used as a sensitive measure of selection occurring in a protein with dN = dS, dN > dS, and dN < dS indicating neutral, positive, and negative selection, respectively. Alignment-wide evidence for selection was tested using the codon-based Z test. For the codon-based Z test, we first calculated dN and dS and their variances using 1000 bootstrap replicates. We then used this information to test the null hypothesis of neutrality (dN = dS) versus alternative hypothesis of positive (dN > dS) or negative (dN < dS) selection using a Z-test. A branch-site random effects likelihood test was used to test whether any of the branches in the tree are evolving under positive selection. A branchTestDNDS test was performed to test whether a prespecified branch of the tree is evolving under different selection strength than the rest of the tree [27]. Individual sites under positive or negative selection were identified using the single likelihood ancestral counting and fixed effects likelihood methods [27].
The draft genomes were interrogated for the presence of genes that are required for the virulence of strain 35000HP in the human inoculation experiments, using nucleotide BLAST [13]. For identifying sequence variation, the nucleotide sequences of virulence genes were translated into amino acids and the translated sequences were aligned using Clustal Omega [29].
The draft genomes were searched for the presence of known antimicrobial resistance genes using ResFinder with default parameters [30]. AST was performed using the agar dilution method as described previously with some modifications [31–33]. Briefly, H. ducreyi strains were grown on Columbia agar (Difco) containing 1% hemoglobin (BBL), 0.2% activated charcoal (Sigma-Aldrich), 5% fetal bovine serum (Atlanta Biologicals), and 1% IsoVitaleX (BBL) for 48 h at 33°C under microaerophilic conditions. The colonies were suspended into Mueller-Hinton (BBL) broth containing 1% IsoVitaleX and 0.002% Tween-80 (Sigma-Aldrich), passed through a 22-gauge needle and left at room temperature for 15 min. The optical density of the culture was adjusted to that of a 0.5 McFarland standard using a Spectronic 20 Plus spectrophotometer (Milton Roy). AST was performed on Mueller-Hinton II medium (BBL) containing 33% lysed horse blood (Remel), 5% fetal bovine serum, and 1% IsoVitaleX. The following antibiotics were tested: amoxicillin (AMX), amoxicillin/clavulanic acid (AMC; 2:1), azithromycin (AZT), ciprofloxacin (CIP), ceftriaxone (CRO), doxycycline (DOX), erythromycin (ERY), and penicillin (PEN; all from Sigma-Aldrich). The H. ducreyi strains CIP542, 35000HP, and the H. influenzae strain 49247 were used as controls. A 104/ml suspension of each strain was delivered onto each plate with a Steer’s Replicator (CMI-Promex, Inc.), and the plates were dried for 15 min at room temperature. The minimal inhibitory concentrations were recorded after incubating the plates for 48 h at 33°C under microaerophilic conditions. The presence of three or fewer colonies was recorded as no growth.
Whole-genome sequencing generated between 0.5 and 4 million reads for each of the 14 strains (Table 1). The estimated genome sizes ranged from 1.52 Mb to 1.74 Mb, with an average GC content of 37.8% to 38.6% (Table 1). The total number of contigs for each strain ranged from 40 to 129 (Table 1). The estimated average genome coverage ranged from 92 to 596 fold (Table 1). Contig ordering generated 2 to 6 LCBs for the CU strains, 5–15 LCBs for the class I strains, and 12–16 LCBs for the class II strains. Inspection of the LCBs revealed that the majority of the putative breakpoints between LCBs occurred in genes that share high homology with other genes in the H. ducreyi genome such as lspA1 and lspA2, genes encoding rRNAs, and bacteriophage-related genes and that the majority of the breakpoints did not contain any rearrangements.
Analysis of pairwise genome conservation distance of the draft genomes showed that CU strains form a subcluster within class I strains and that class II strains form a separate cluster from CU and class I strains (Fig 1).
Compared to 35000HP, all the CU strains consistently contained ~20-kb deletion (HD1528 to HD1565) in a bacteriophage locus that is homologous to Pseudomonas aeruginosa bacteriophage B3 and five small deletions that ranged in size from 30–767 bp (Fig 2 and S2 Table). The class I strain HMC56 contained a 50-kb deletion (HD0897 to tRNA-Lys-1) in a region homologous to the H. influenzae ICEHin1056 integrative conjugative element (Fig 2 and S3 Table). All the class II strains contained 3 major deletions of 37 kb (HD0087 to HD0161), 35 kb (HD0478 to HD0495) and 50 kb (HD0897 to tRNA-Lys-1), which are homologous to Escherichia coli bacteriophage D108, Haemophilus bacteriophage SuMu, and H. influenzae ICEHin1056, respectively (Fig 2 and S4 Table). The class II strains also contained several deletions (between HD1528 and HD1618) in a region that is homologous to P. aeruginosa bacteriophage B3 (Fig 2 and S4 Table). All the GU strains also contained several other small deletions as listed in S3 and S4 Tables.
Compared to 35000HP, we did not find any inversions in CU strains with the exception of NZV1, which contained an inversion of ~428 kb that spanned from HD0054 (tuf) to HD0659 (S2 Fig). Among the class I strains, HMC56 contained an inversion of ~300 kb that spanned from glpA (HD1157) to lspA1 (HD1505) (S2 Fig). HD183 contained a ~161 kb inversion that spanned from hhdA (HD1327) to lspA1 (HD1505) (S2 Fig). All the class II strains contained an inversion of ~17 kb that spanned from HD1532 to HD1565 (S2 Fig). However, BLAST analysis of the inversion breakpoints showed no major changes in their genetic content.
Compared to 35000HP, the CU strains, the class I strain 82–029362, and the class II strain CIP542 contained no additional genes in their genomes. All the remaining class I and class II strains contained several additional genes as listed in S5 Table.
To get a deeper understanding of the relationship of CU strains to GU strains, we next performed whole-genome SNP analysis using 35000HP as a reference. CU strains differed from 35000HP by ~400 SNPs (Table 2). The class I strain HD183 differed from 35000HP by ~160 SNPs, while all other class I strains differed by ~2,000 SNPs (Table 2). The class II strains differed from 35000HP by ~30,000 SNPs (Table 2).
Analysis of within lineage genetic diversity showed that CU strains had the least nucleotide and amino acid divergence followed by class I and class II strains (Table 3). Analysis of interlineage diversity showed that there was little divergence between CU and class I strains; however, a greater amount of divergence was observed between CU and class II strains (Table 3). Interlineage diversity analysis showed that there was high divergence between class I and class II strains (Table 3).
While the Phi test showed no evidence of recombination (P = 0.44), the likelihood ratio test identified five putative recombination events in CU and GU strains (S6 Table). Removal of recombination regions had no major effect on the overall topology of the phylogenetic tree described in the following section, except for minor differences in bootstrap values and positioning of individual species within class clades (S3 Fig).
In general, all methods showed that class I and class II strains formed two separate phylogenetic clusters and that CU strains formed a subcluster within the class I clade with minor differences in bootstrap values and positioning of individual species within class clades. A rooted tree generated by the Maximum Likelihood method and Pasteurellaceae members as outgroups was used as the final tree (Fig 3).
To determine the approximate time to the MRCA of the CU strains, we performed a molecular clock analysis using the Bayesian method and the mutation rates proposed by Ochman et al. for calibration [25, 26]. The divergence time of the CU strains from the MRCA of the class I strains 35000HP and HD183 was estimated as 180,000 years ago (Fig 4). The divergence time of the CU strains, 35000HP, and HD183 from the MRCA of other class I strains was estimated as 450,000 years ago (Fig 4). The divergence time of class I strains from the MRCA of class II strains was estimated as 1.95 mya (Fig 4). Molecular clock analysis also showed that the CU strains began to diversify from each other around 27,000 years ago (Fig 4). Thus, CU strains appear to have recently diverged from class I GU strains.
Pairwise analysis of rates of nonsynonymous (dN) and synonymous (dS) substitutions and their variances showed that the Z-test rejected the null hypothesis of neutrality (dN = dS) in favor of the alternative hypothesis of negative selection (dN < dS) (Table 4). The dN-dS value averaging over all sequence pairs was -75.55 (P = 0.0000000001). Utilizing the rates of nonsynonymous (dN) and synonymous (dS) substitutions, we also calculated the overall mean and pairwise mean dN/dS ratios; the overall mean dN/dS ratio for all genomes was 0.31 and the pairwise mean dN/dS ratios for most comparisons were less than 1 (Table 4). The pairwise mean dN/dS ratio between the CU and GU lineages was 0.35, between CU and class I lineages was 0.38, and between CU and class II lineages was 0.33. Consistent with these analyses, the single likelihood ancestral counting and the fixed effects likelihood analyses identified 141 and 132 negatively selected sites, respectively.
To determine whether CU strains evolved under different selection strength than GU strains, we performed a TestBranchDNDS analysis. This analysis showed that the strength of selection in CU strains was not significantly different than in GU strains (likelihood ratio difference = 3.8; P = 0.58).
We determined whether the genomes of CU strains contained the genes that are required for the virulence of strain 35000HP in the human challenge model of infection and whether there were variations in these virulence determinants compared to GU strains [13]. BLAST analysis showed that all the CU and GU strains contained all of the genes known to be required for virulence in the human challenge model (S7 Table). Alignment of amino acid sequences of the virulence determinants showed that the DsrA, LspA1, and LspA2 proteins of the CU strains differed by at least 1 amino acid from class I strains (S7 Table).
To determine whether CU strains were resistant to clinically relevant antimicrobials, we performed AST using the agar dilution method. The CU strains from Samoa and Vanuatu were AZT susceptible, and had similar susceptibility patterns as the type strains 35000HP and CIP542 (Table 5). With the exception of 82–029362 and 35000HP, all the class I strains were resistant to penicillin (MIC, >256 μg/ml), amoxicillin (MIC, 64–256 μg/ml), and doxycycline (MIC, 8–16 μg/ml) (Table 5). With the exception of CIP542, all class II strains were resistant to amoxicillin and penicillin (MIC, 128–256 μg/ml) (Table 5). The class II strains 33921 and DMC64 were also resistant to doxycycline (MIC, 8–16 μg/ml) (Table 5). All the strains were susceptible to ciprofloxacin, azithromycin, erythromycin and ceftriaxone (Table 5).
Consistent with their susceptibility to clinically relevant antimicrobials, the CU strains contained no horizontally acquired genes encoding antimicrobial resistance determinants in their genomes. Consistent with their resistance to penicillin/amoxicillin and doxycycline, the genomes of GU strains contained genes that confer resistance to penicillin/amoxicillin (blaTEM-1B) and doxycycline [tet(B), tet(32) or tet(M)] (Table 5).
H. ducreyi was previously thought to exclusively cause the sexually transmitted disease chancroid but has emerged as a major cause of the nonsexually transmitted CU in children in yaws-endemic regions of South Pacific islands and equatorial Africa. Here, we performed whole-genome sequencing of a limited number of CU strains and compared them to class I and class II GU strains. Comparative genome analyses showed that the CU strains are remarkably similar to class I strains. Phylogenetic analyses showed that the CU strains evolved from class I GU strains.
Analysis of genome conservation of CU and GU strains showed that CU strains had 98–99% similarity to each other, 94–98% similarity to class I strains, and 81–92% similarity to class II strains. Kunin et al., estimated genome conservation within different bacterial taxonomic ranks and found that strains within most bacterial species have a genome conservation of approximately 87% (range, 73–101%) [34]. Thus, the H. ducreyi genome conservation values are well within the range of those of other bacterial species. Genome conservation analysis also showed that CU strains form a subcluster within class I GU strains and that class II strains form a distinct cluster from class I and CU strains. These findings are in good agreement with the results of the whole-genome phylogenetic analysis as well as with previous multilocus sequence-based phylogenetic analysis [11].
Consistent with the genome conservation data, analysis of whole-genome genetic diversity also showed that there was smallest amount of genetic diversity within the CU strains (dnucleotide = 0.000013), little genetic diversity between CU and class I strains (dnucleotide = 0.00012), and a greater amount of genetic diversity between CU and class II strains (dnucleotide = 0.0098) and class I and class II strains (dnucleotide = 0.01). Cejcova et al., reported a whole-genome nucleotide diversity of 0.00033 for strains within T. pallidum subsp. pallidum and of 0.00032 for strains within T. pallidum subsp. pertenue [35]. Thus, the H. ducreyi genetic diversity values for CU and Class I strains are similar to those of the two Treponema species that inhabit similar ecological niches as H. ducreyi, while the diversity values between CU and class II strains and class I and class II strains are higher than those of the two Treponema species.
Our study estimated that class I strains diverged from class II strains 1.95 mya and that CU strains diverged from class I strains 0.18 mya. Previous studies estimated that class I strains diverged from class II strains 5 mya and that CU strains diverged from class I strains 0.355 mya [10, 11]. In our study, divergence times were estimated using entire genomes. The previous studies used only 11 H. ducreyi loci, all of which were selected to contain variant alleles to allow for epidemiological typing. Since a large proportion of genes in the genome do not contain variant alleles, averaging the variance over the entire genome would result in relatively lower divergence times than those estimated in previous studies.
Although CU strains lacked additional genetic material compared to 35000HP, CU strains differed from 35000HP by ~400 SNPs. Nearly 40% of these SNPs were nonsynonymous and 25% were located in noncoding regions of the genome. Previous studies have shown that SNPs can have a profound impact on global gene expression in bacteria [36]. Whether SNPs in the CU strains would result in a different global gene expression pattern than 35000HP requires additional investigation.
A large number of SNPs in the CU strains were located in 21 genes that individually or in combination are required for H. ducreyi infection in human volunteers. CU strains differed from class I strains by at least one amino acid in 3 virulence determinants, specifically DsrA, LspA1, and LspA2. DsrA is a surface protein and LspA1 and LspA2 are secreted proteins; all three are required for evasion of immune defenses [37, 38]. Thus, the variations of these proteins in CU strains are likely an effect of host immune pressure. Consistent with the fact that class I strains differ from class II strains in several of the known virulence determinants and our finding that CU strains formed a subcluster under class I strains, CU strains differed from class II strains by at least one amino acid in 19 of the 21 virulence determinants [10]. In agreement with a previous study, compared to class I strains, the nucleotide sequences of DsrA and NcaA of class II strains also contained several short rearrangements including deletions and insertions [10].
Analysis of rates of nonsynonymous substitutions (dN) and synonymous substitutions (dS) in the CU and GU genomes showed that synonymous substitutions were found at a higher rate than nonsynonymous substitutions with an overall mean dN/dS ratio for all strains of 0.31. Similarly, the pairwise mean dN/dS ratio between CU and GU lineages was 0.35. These data suggest that CU and GU strains evolve under negative selection. Other sexually transmitted bacterial pathogens such as Neisseria gonorrhoeae and Chlamydia trachomatis also evolve under negative selection, with overall mean dN/dS ratios of 0.3184 and 0.4021, respectively [39, 40]. These findings are in agreement with the neutral theory of molecular evolution, which postulates that selective fixation of neutral mutations by genetic drift is the major determinant behind species divergence [41]. Our data also showed that both CU and GU strains evolve under similar selection strength, which may be due to the similar immunological pressures that these strains encounter in their respective ecological niches of human skin versus mucosal surfaces and human skin.
Using SNPs, molecular dating analysis indicates that the CU strains began to diversify from each other ~27,000 years ago. The CU clade is characterized by several shared, derived deletions of defined lengths (synapomorphies), which were most likely inherited from the common ancestor of modern CU strains. Given that these deletions were absent in all the class I GU strains including 35000HP and HD183, we speculate that the Samoan/Vanuatu CU lineage may have existed for at least 27,000 years.
Mitjà and colleagues hypothesized that syndromic management of genital ulcers in the South Pacific may have forced H. ducreyi into a new niche of cutaneous ulcers in children [6]. Syndromic management of GU in the South Pacific was introduced in 2002, while CU due to H. ducreyi was first reported in 1989 [6]. The fact that the CU strains diverged from GU strains ~180,000 years ago and from each other ~27,000 years ago supports the idea that cutaneous infection with H. ducreyi preceded syndromic management of GU. A possible explanation why H. ducreyi was not recognized as a cause of CU previously is that CU in the South Pacific has traditionally been empirically treated with penicillin [42]. As CU strains are susceptible to penicillin, CU due to H. ducreyi would have responded to empirical treatment. The current World Health Organization case definition of yaws includes a patient with a chronic atraumatic skin ulcer and seropositivity for T. pallidum subsp. pertenue. In the cross sectional survey in Papua New Guinea, a reasonable proportion of children with detectable H. ducreyi DNA in ulcers were also seropositive for T. pallidum subsp. pertenue [6] and therefore would be classified as having yaws. This could account for the lack of earlier recognition of H. ducreyi as a source of CU.
Although penicillin had been the cornerstone of yaws eradication efforts for the last several decades, MDA of AZT is the mainstay of the World Health Organization’s new program for the eradication of yaws [42]. MDA was given to 84% of the villagers who were studied in Papua New Guinea [9]. At 12-months follow-up, MDA reduced the prevalence of CU by 90% [9]. In those who had ulcers at follow-up, there was a significant reduction in the proportion of ulcers with T. pallidum subsp. pertenue DNA [9]. However, the proportion of ulcers containing H. ducreyi DNA was unchanged relative to the baseline level of 60% [9]. The CU strains from Samoa and Vanuatu were as susceptible to AZT as 35000HP. Whether CU strains from Papua New Guinea are susceptible to AZT is not known. If they are susceptible, their persistence after MDA suggests that CU strains may have a higher level of infectivity than T. pallidum subsp. pertenue or may be present in an environmental reservoir.
Inoculation of the upper arm of human volunteers with the GU strain 35000HP produces an infection that is clinically and histopathologically nearly identical to natural chancroid [13, 43, 44]. Evolutionary analyses showed that CU strains are closely related to 35000HP. Similar to 35000HP, CU strains are capable of infecting nongenital skin. Our data showed that CU strains evolve under selection strength similar to that of GU strains. Due to lack of biopsy specimens, we do not know whether the histopathology of a CU lesion is similar to that of an experimental lesion caused by 35000HP or natural chancroid. Nevertheless, these data suggest that H. ducreyi likely encounters similar host pressures in the genital and nongenital skin.
Placement of 106 CFU of 35000HP on intact skin does not cause disease in human volunteers; but as few as one bacterium delivered by a puncture wound causes infection [13]. These data raises the possibility that either wounds are required for CU strains to initiate infection or that CU strains possess additional genes that allow them to penetrate intact skin. Our data showed that the CU strains did not contain additional genetic elements, suggesting that CU strains likely use wounds to initiate infection. In Papua New Guinea, up to 7% of children have CU with detectable H. ducreyi DNA [6]; it is difficult to imagine that wound to wound transmission is responsible for this astoundingly high prevalence. In the Papua New Guinea study, many children infected with H. ducreyi were seropositive for T. pallidum subsp. pertenue and some ulcers contained both H. ducreyi and T. pallidum subsp. pertenue DNA [6]. Thus, T. pallidum subsp. pertenue may serve as an instigating pathogen while H. ducreyi superinfects yaws lesions. Photographs of typical CU lesions show that flies frequently land on ulcers [6]. Thus, it is possible that CU strains are transmitted from person to person by direct contact of wounds with infected lesions, or by vectors such as flies.
In a randomized controlled clinical trial, treatment with 1 gram of AZT prevented experimental infection of adult volunteers with 35000HP for nearly two months [45]. Given that a 2-gram dose of AZT is being used to eradicate yaws, MDA may provide treatment and prophylaxis against CU strains for a similar period of time. These data also suggest that repetition of MDA on a bimonthly basis and/or higher coverage rates may contribute to successful eradication of CU strains from yaws-endemic areas.
By PCR-based testing, 2% of commercial sex workers in a chancroid endemic region are asymptomatically colonized in the cervico-vaginal tract with H. ducreyi [46]. Whether CU strains asymptomatically colonize the skin of humans living in the tropics is unknown, but colonization would provide a source of bacteria that could enter wounds. As AZT is concentrated intracellularly especially in fibroblasts [45], colonization of the skin surface could allow CU strains to escape AZT treatment.
Our study has several limitations. We only reported draft genomes, and the genetic variation among the strains was not confirmed by PCR and sequencing. Our study involved a small number of GU strains, with limited clinical and epidemiological data. Our analysis only included CU strains that were acquired in Samoa and Vanuatu; our findings should not be extrapolated to CU strains from other regions. All strains used in this study were obtained following culture and storage; their sequences could have been affected by these factors over time. Finally, the CU strains were not compared to contemporaneous GU strains from the same or other regions; to our knowledge and due to syndromic management, few such GU strains exist.
This was the first study using comparative genomics to examine a small number of cultured H. ducreyi strains isolated from CU and GU. Our findings show that CU strains are derivatives of class I GU strains whose lineage may be 27,000 years old. Further studies are needed to determine the phylogeny of CU strains from other endemic areas, such as Papua New Guinea, Ghana, and the Solomon Islands, and to examine strains that persist after MDA of azithromycin. Flies and nonhuman primates are thought to serve as reservoirs for T. pallidum subsp. pertenue [6, 47]; it would be interesting to determine whether they serve as reservoirs for CU strains or whether humans who reside in endemic areas are colonized with H. ducreyi.
|
10.1371/journal.pntd.0002795 | Sulphonylurea Usage in Melioidosis Is Associated with Severe Disease and Suppressed Immune Response | Melioidosis is a problem in the developing tropical regions of Southeast Asia and Northern Australia where the the Gram negative saprophytic bacillus Burkholderia pseudomallei is endemic with the risk of fulminant septicaemia. While diabetes mellitus is a well-established risk factor for melioidiosis, little is known if specific hypoglycemic agents may differentially influence the susceptibility and clinical course of infection with B. pseudomallei (Bp).
In this cohort study, patients with pre-existing diabetes and melioidosis were retrospectively studied. Outcome measures: mortality, length of stay and development of complications (namely hypotension, intubation, renal failure and septicaemia) were studied in relation to prior diabetic treatment regimen. Peripheral blood mononuclear cells (PBMC) from diabetic patients and healthy PBMC primed with metformin, glyburide and insulin were stimulated with purified Bp antigens in vitro. Immune response and specific immune pathway mediators were studied to relate to the clinical findings mechanistically. Of 74 subjects, 44 (57.9%) had sulphonylurea-containing diabetic regimens. Patient receiving sulphonylureas had more severe septic complications (47.7% versus 16.7% p = 0.006), in particular, hypotension requiring intropes (p = 0.005). There was also a trend towards increased mortality in sulphonylurea-users (15.9% versus 3.3% p = 0.08). In-vitro, glyburide suppressed inflammatory cytokine production in a dose-dependent manner. An effect of the drug was the induction of IL-1R-associated kinase-M at the level of mRNA transcription.
Sulphonylurea treatment results in suppression of host inflammatory response and may put patients at higher risk for adverse outcomes in melioidosis.
| Melioidosis is a problem in the developing tropical regions of Southeast Asia and Northern Australia where the Gram negative bacillus Burkholderia pseudomallei can cause life-threatening infection. Diabetes mellitus is a recognised risk factor for melioidiosis; however little is known if commonly used anti-diabetic drugs may affect the clinical course of the disease. In this study, we found that patients who were receiving sulphonylureas for diabetic treatment had more severe septic complications requiring intensive care as well as increased risk of deaths. This may be attributable to the capacity of sulphonylureas in modulating the host immune response. We highlight caution in the prescription of this class of drug, which is popular due to its low cost and easy availability, especially in melioidosis-endemic tropical regions.
| Melioidosis is an environmental saprophyte endemic in tropical regions. The infection is caused by the Gram negative saprophytic bacillus Burkholderia pseudomallei found commonly in environmental soil and surface water [1]. Clinical presentations of melioidosis include bacteremia, abscesses in any organ systems, pneumonia or soft tissue infection. The clinical course can be mild and chronic through to fulminant leading to septic shock and death [2].
Diabetes mellitus is a well-established risk factor for susceptibility to melioidosis [3], [4]. Approximately 50% of patients presenting with melioidosis are diabetics, the majority of whom are already receiving anti-diabetic therapy. However little is known if specific hypoglycemic agents, by themselves, may differentially influence the susceptibility and clinical course of infection with B. pseudomallei. This is important in light of recent knowledge that sulphonylureas (glyburide) and the biguanides (metformin) may possess modulating capabilities on host immune response [5], [6]. It is known that the activation of host immune defence and the elaboration of inflammatory cytokines against the pathogen is requisite [7], [8] but opinion on the optimal cytokine milieu for facilitating an effective host response during melioidosis remains unclear [9], [10].
Hence we aim to study whether commonly prescribed anti-diabetic medications, in particular the sulphonylureas by virtue of their ascribed immunomodulating potential, may influence outcomes in melioidosis. Consequently, we also will attempt to underline the mechanisms by which the specified drug may exert their effects on host immune response against B. pseudomallei.
The study was conducted at two of Singapore's largest hospitals, the National University Hospital and the Tan Tock Seng Hospital, each with more than1000-beds capacity. We identified patients diagnosed with melioidosis between January 2001 to December 2010 from the integrated hospital electronic medical record systems. A total of 116 such patients were identified, of which 15 subjects (12.9%) did not have diabetes by history and laboratory investigation, while 27 subjects (23.3%) were newly diagnosed diabetics upon presentation with melioidosis. The remaining 74 patients (63.8%) had pre-existing type II diabetes and receiving medication prior to presenting with melioidosis and this was the study cohort being analysed.
All patients received the recommended melioidosis treatment regimen consisting of an initial intensive therapy consisting of either ceftazidime, meropenem or imipenem intravenously for at least 2 weeks, followed by a prolonged oral eradication phase consisting of a combination of 2 oral agents from the 3 antibiotics: trimethoprim-sulfamethoxazole (TMP-SMX), amoxicillin-clavulanate, doxycycline for at least 10 weeks [11]. Trimethoprim-sulfamethoxazole was the backbone of the 2-drug eradication regimen unless there were drug tolerance issues with individual patients. The patients were followed up for at least 12 weeks. Records of pre-existing diabetic medications were obtained from clinical notes and/or verified from computerised outpatient pharmacy archives. Most of the glycosylated haemoglobin (HbA1c) levels were measured upon admission to hospital with melioidosis. When this was not available, the last known HbA1c level (up to 3 months prior presentation) was traced through electronic outpatient records. The primary outcome measure was mortality. The secondary outcome measures included length of hospital stay and measures of disease severity, namely: (i) hypotension requiring the infusion of inotropes (ii) respiratory distress requiring mechanical ventilation (iii) renal impairment with need for renal replacement therapy (iv) septicaemia as per defined by the Surviving Sepsis Campaign Guidelines [12].
The Toll like receptor (TLR) 4 ligand lipopolysaccharide (LPS-EK Ultrapure, Escherichia coli serotype K-12) was purchased from Invivogen (San Diego, CA). Metformin (a biguanide), glyburide (a sulphonylurea) and human recombinant insulin were purchased from Sigma-Aldrich (Singapore) and re-suspended in DMSO. The antibodies ERK 1&2 (H-72) and p-ERK (E-4) were from Santa Cruz Biotechnology, Inc (Dallas, TX).
Overnight culture of B. pseudomallei, Bp22, was inoculated into LB broth (Difco Laboratories, Detroit, Michigan) and incubated for 4 h at 37°C until an OD600 reading of 0.8–1.0. The bacteria was pelleted at 4000 g for 10 min and washed twice with 1X phosphate buffer saline (PBS), prior to reconstitution in 1X PBS containing 0.2 µg/ml leupeptin (Sigma-Aldrich, St Louis, MO), 0.2 µg/ml pepstatin A (Sigma-Aldrich, St Louis, MO) and 2.5 Kunitz units of DNase I (Sigma-Aldrich, St Louis, MO). To extract bacterial proteins, the suspension was loaded into lysing Matrix B tubes (MP Biochemicals, Solon, OH), and three rounds of homogenization was performed using Fastprep instrument (MP Biochemicals, Solon, OH) at a speed of 6 m/s for 30 s with a pause of 10–15 min in between for cooling. The bacterial lysate was centrifuged at 13000 g for 2 min, and passed through 0.22 µm filter unit (Merck Millipore, Billerica, MA) to remove any live bacteria. The supernatant was further purified and concentrated using Amicon Ultra-15 centrifugal units with 10 kDa nominal molecular weight limit (NMWL) membrane (Merck Millipore, Billerica, MA). The final purified Bp antigen were recovered in 1XPBS, quantified by BCA protein assay kit (Pierce Biotechnology, Rockford, IL) and used for further stimulation experiments.
Separation and stimulation of peripheral blood mononuclear cells (PBMC) from healthy volunteers and diabetic patients was performed as previously described [13]. Blood was drawn into sodium heparin tubes (BD Vacutainer Franklin Lakes, NJ) after informed consent. PBMC were isolated by density centrifugation on Ficoll-Hypaque (Pharmacia Biotech, Uppsala, Sweden). Cells were washed, counted and adjusted to 5×106 cells/mL in culture medium (RPMI 1640 DM supplemented with gentamicin, L-glutamine, and sodium pyruvate). Stimulation assays were performed in 96-well round-bottom plates using 100 µl of PBMC with the various stimuli and drugs to a total volume of 200 µl per well. PBMC were pre-incubated with metformin, glyburide, insulin or control media for 1 hour after which purified Bp antigen was added. After 24 h or 48 h of incubation in humidified atmosphere (5% CO2) at 37°C, the supernatants were collected and stored at −20°C until further assay.
Interleukin(IL)-6, IL-10, IL-1β, tumour necrosis factor-alpha (TNF-α) and interferon- gamma (IFN-γ) were measured by commercial ELISA kit according to the instructions of the manufacturer (eBioscience, San Diego, CA). Detection limits were 20 pg/ml (IL-1b, TNF-α and IFN-γ) and 10 pg/ml (for IL-6 and IL-10) respectively.
PBMCs from healthy volunteers were treated with glyburide, metformin, insulin and the carrier medium (DMSO) for 1 h, followed by stimulation with purified Bp antigen for 30 mins. Whole-cell lysates were prepared in RIPA buffer and protein content was determined by the Bradford assay (Bio-Rad). 25 µg of lysates were electrophoresed on 10% polyacrylamide gels and transferred to polyvinylidene difluoride membranes. After blocking in 5% non-fat dry milk, the membranes were incubated with p-ERK (sc-7383), p-JNK (#4668) or p-p38 (#9215) primary antibodies. Horseradish peroxidase-conjugated secondary antibody was used and immunoreactive proteins were visualized with chemiluminescence detection substrate (Pierce). The membranes were stripped and re-probed for total ERK (sc-292838), JNK (#9252) and p38 (#9215). Western blot was also performed for MyD88 (sc-11356). Beta-actin (sc-47778) was used as the loading control. Antibodies from Santa Cruz Biotechnology or Cell Signaling were used.
PBMC were incubated with glyburide, metformin, insulin and the carrier medium (DMSO) for 2 h prior to stimulation with purified Bp antigen for 30 minutes. Following stimulation, cells were washed twice with 1XPBS supplemented with 0.2% (w/v) BSA and stained with anti-CD14-V450 antibody (BD Biosciences, San Jose, CA). Cells were then fixed with BD Cytofix (BD Biosciences, San Jose, CA) and permeabilized using Perm III buffer (BD Biosciences, San Jose, CA) for 30 minutes. The permeabilized cells were stained with mouse anti- p-ERK1/2 antibody (BD Biosciences, San Jose, CA), mouse anti- phospho-JNK antibody (Cell signaling Technology, Danvers, MA) and goat anti-IRAK-M antibody (Santa Cruz Biotechnology, Dallas, TX). For cells stained with unlabelled goat anti-IRAK-M antibody, secondary labeling with donkey anti-goat-FITC antibody (Santa Cruz Biotechnology, Dallas, TX) was carried out. Stained cells were then analyzed on FACSCanto flow cytometer (BD Biosciences, San Jose, CA) and data analysis was performed using Flow Jo software (version 7.6.5) (Tree Star, Ashland, OR).
The expressions of interleukin-1 receptor-associated kinase-1(IRAK-1) and interleukin-1 receptor-associated kinase-M (IRAK-M) in PBMC primed with glyburide, metformin, insulin and control medium were determined. RNA was extracted from 107 PBMC by using 1 ml TRIzol reagent (Sigma, St. Louis, MO). Subsequently, 200 µl chloroform and 500 µl 2-propanol were used to separate the RNA from DNA and proteins. After washing with 75% ethanol, the dry RNA was dissolved in 50 µl of diethylpyrocarbonate (DEPC) water. To obtain cDNA, we reverse-transcribed 1 µg DNase-treated total RNA with oligo(dT) primers (0.01 µg/ml) in a reverse transcription-PCR mixture with a total volume of 20 µl. Quantitative PCR was performed using the Bio-Rad iCycler and SYBR Green. The following primers were used (5′-3′): GTACATCAAGACGGGAAGGC (forward) and AGTGTGCTCTGGGTGCTTCT (reverse) for IRAK-1, GTACATCAGACAGGGGAAACTTT (forward) and GACATGAATCCAGGCCTCTC (reverse) for IRAK-M, and ATGAGTATGCCTGCCGTGTG (forward) and CCAAATGCGGCATCTTCAAC (reverse) for β2 microglobulin (B2M) [9]. Quantification of the PCR signals for each sample was performed by comparing the cycle threshold values, in duplicate, for the gene of interest with the cycle threshold values for the B2M as housekeeping gene. All primers were validated according to the protocol. Mean relative mRNA expression was calculated using Pfaffl method. Values are expressed as a ratio of fold increase to mRNA levels of unprimed cells.
The SPSS version 20.0 statistical software package was used to perform the calculations. In the cohort study, the differences between patient groups were analyzed using the chi-squared test or Fisher exact probability where appropriate, for categorical variables and the t-test for continuous variables. Adjustments for the effect of HbA1c and diabetic drug usage on outcome measures were performed using binary logistic regression. For the in-vitro experiments, results were pooled from at least 3 sets of experiments and analyzed, unless otherwise specified. The cytokine data was presented as mean±standard error of the means (SEM). The differences in cytokine production were tested using Mann-Whitney U test. The level of significance was set at p<0.05.
Ethics approval for the conduct of the above study had been attained through the Domain Specific Review Boards, National Healthcare Group, Singapore (no. 2012/00596 and no. 2012/00949). Adult study subjects had provided written informed consent to participate in the study. There were no study subjects under the age of 21 years old.
Of the 74 subjects with pre-existing diabetes on treatment and melioidosis, 44 (57.9%) were taking a sulphonylurea group drug either alone or in combination with other diabetic medications. The demographics and outcomes of this group of patients were studied against patients who did not receive sulphonylureas as part of their diabetic drug regimen. As described in Table 1, the background characteristics between the 2 groups of patients including immune status and renal function were similar. Glycaemic control as judged by their most recentHbA1c sugar control was similar between both groups (9.00 versus 9.59, p = 0.353).
However, patient receiving sulphonylurea-containing regimens had higher incidence of secondary end points indicating severity (47.7% versus 16.7%, p = 0.006). In particular, melioidosis patients on sulphonylurea were more likely to be hypotensive requiring inotropic support (29.5% versus 3.3%, p = 0.005). The other analysed criteria showed increased tendencies for mechanical ventilation and acute kidney injury requiring renal replacement therapy. Similarly there was a trend towards increased overall mortality in patients on sulphonylurea (15.9% versus 3.3%, p = 0.080). After adjustment for HbA1c, sulphonylurea usage remained a significant predictor of development of severity manifestations (adjusted odds ratio AOR 4.89, 95% confidence interval C.I. 1.37–17.5, p = 0.015). Mortality was not significantly linked to prior sulphonylurea exposure (AOR 4.48 95% C.I. 0.49–41.4, p = 0.186) even after HbA1c correction.
Of note, concurrent insulin therapy was prescribed more in diabetic regimens not containing sulphonylureas, and the converse for α-glucosidase inhibitor. Melioidosis patients who had insulin as part of their diabetic therapy had a lower incidence of severe manifestations during infection (19.0% versus 41.5%, p = 0.104). The use of metformin did not influence the clinical outcomes (data not shown).
To specifically elucidate the influence of the respective diabetic drugs on the immune response, PBMC from healthy volunteers were primed with metformin, glyburide and insulin and then stimulated with purified Bp antigen. Over a drug concentration range of 0.001 mg/mL to 0.01 mg/mL, glyburide clearly diminished the production of TNF-α, IL-1β and IL-10 in a dose-dependent manner as compared to Bp control (in carrier medium) (Figure 1A–D). Conversely, insulin had a tendency to accentuate IL-1β response but this is only most evident at the dose of 0.01 mg/mL (Figure 1B). Low dose metformin had a limited effect in enhancing TNF-α, IL-1β and IFN-γ production (Figures 1A–C).
To further validate the above findings in the patient cohort, PBMC from diabetic patients were stimulated with purified Bp antigen. We found that patients who had sulphonylurea in their treatment regimen for diabetes had a significantly weaker IL-1β response (p = 0.04) as compared to patients who were on non-sulphonylurea containing regimens. The attenuative effects of sulphonylurea on the other cytokines in the patients were suggestive but not significant (Figure 1E).
Further to the above findings, we found that the attenuation of cytokine production by glyburide was effected upstream at the level of transcription. In the presence of glyburide, the IL-1β and TNF-α mRNA transcription was significantly reduced. In comparison, metformin and insulin induced minimal transcriptional changes (Figure 2).
Because the suppressive effects induced by glyburide was generally seen with IL-1β, TNF-α and IL-10 and possibly not localized to a specific arm of the immune signaling pathway, we decided it was logical to study first, the major upstream adaptor molecule MyD88 and the mitogen-activated protein (MAP) kinases JNK, ERK and p38 in the canonical innate signaling pathways. However we found that the expression of MyD88, JNK, ERK and p38 were not affected glyburide-treated cells (Figures 3A–D).
Recently it has been proposed that elevated expression of IL-1R-associated kinase-M (IRAK-M), a regulator of the inflammatory signaling pathway, was associated with adverse outcome in septic melioidosis [9]. The role of IRAK-M in the context of diabetes treatment and melioidosis, however, is not known to date. We saw increased expression of IRAK-M mRNA in glyburide-treated cells. The interleukin-1 receptor-associated kinase 1 (IRAK-1) mRNA expression was not affected (Figure 4A). The effects of insulin and metformin on IRAK-M and IRAK-1 were limited. This was further confirmed on flow cytometery showing increased IRAK-M expression in monocytes exposed to glyburide (Figure 4B). This up-regulation of IRAK-M expression reasonably accounts for the immune suppressive effects induced by glyburide, and the association with disease severity observed with sulphonylurea use in melioidosis.
During an infection the capacity of the host to mount an appropriate immune and inflammatory response against the pathogen is pivotal. In the context of recent reports that diabetic medications like glyburide and metformin may possess inherent immune modulating capabilities, at least in the in-vitro setting [5], [6], we felt that it was important to study this further in the clinical context of melioidosis whereby diabetics are specifically predisposed to.
In this study, we found that the use of a sulphonylurea-containing diabetes treatment regimen is linked to a more severe clinical course especially hypotension during melioidosis. This finding has implications for the large number of diabetics living in melioidosis endemic regions, particularly Singapore, Thailand and Northern Australia. The low cost of sulphonylureas, its availability and its relatively wide therapeutic window contributes to the popularity of the drug. While glycaemic control between the 2 treatment groups might have been a potential factor contributing to the difference in outcomes, we have shown that HbA1c was not a significant confounder of the results.
We showed that glyburide had the capacity to suppress cytokine production. These effects were seen not only in-vitro in cells primed with the drug, but also similarly demonstrated in diabetic patients who were taking sulphonylureas. The importance of this latter finding is to be highlighted in that the differences in cytokine trends could still be elicited despite the numerous of confounders like multiple co-morbidities and polypharmacy in the clinical cohort. The attenuative effect on cytokine production was exerted at the level of mRNA transcription and one of the effects of glyburide was through the enhancement of IRAK-M which assumes a regulatory or inhibitory role in proinflammatory cytokine production in the TLR/IL-1R signaling pathway though inhibition of nuclear transcription factor NF-kappaB [14]. Conversely, IRAK-1 which up-regulates NF-kappaB, is suppressed. These findings on the mechanistic action of glyburide are novel as to date; glyburide is reported to act through the NALP3 inflammasome complex [5].
The response to infection by B. pseudomallei requires the host defence to first mount a prompt and effective innate immune response [7]. Results from our studies indicate that this capacity to mount an inflammatory cytokine milieu is compromised in the background presence of sulphonylureas in the body. This hypothesis is supported by the recent finding that increased IRAK-M with consequent immunosuppresion was associated with poor outcome in melioidosis, though the specific aspect of diabetes and drugs was not looked at in that study [9].
The findings of our study is in contrast to that by another group [10], who had reported reduced mortality, hypotension and respiratory failure with glyburide usage in melioidosis in Thailand. The group had hypothesized that the inhibition of the inflammasome and the neutrophil-mediated inflammatory process by glyburide might be beneficial in limiting sepsis-induced tissue injury and organ damage. Conversely, it is imperative that the host innate immunity be able to mount a robust inflammatory response against the pathogen during the initial phase of infection. The presence of sulphonylurea in the body compromises the immune response as we have demonstrated. Furthermore, such modulation of the IRAK complex as induced by sulphonylurea leads to susceptibility to infections as can be seen in patients with these genetic deficiencies [15]. In our smaller cohort of patients in Singapore, we were able to document diabetic drug history from our electronic records as well as the prescribed appropriate initial intensive therapy for at least 2 weeks followed by oral eradication therapy of at least 10 weeks as per recommended for melioidosis [16]. While HbA1c was not captured in the Thai cohort, we had sought to obtain the last known HbA1c of our patients and had factored glycaemic control into our analysis to show that sulphonylurea usage was an independent predictor of developing severe disease in melioidosis.
In our study, the diabetic patients who were receiving insulin tended not to be prescribed concurrent sulphonylurea. This was expected as per prescribing practice in the management of diabetes. Our data suggests that patients on insulin might have a lower risk of severe septic manifestations of melioidosis. However we are cautious in not over-interpreting these results at this point as functionally, the immunomodulatory effects of insulin are not as clearly evident as compared to the suppression of cytokines induced by sulphonylurea. To validate these findings, the mechanistic effects of insulin need to be elucidated further.
Diabetes mellitus remains the single most important risk factor for development of melioidosis. We find here that a sulphonylurea-containing diabetes treatment regimen suppresses the host inflammatory response and puts patients at higher risk for adverse outcomes. The implication of this finding may extend beyond melioidosis to other Gram negative septicaemia. Against the background of the popularity of sulphonylurea use in diabetes, this study highlights caution in the prescription of this class of drug especially in melioidosis-endemic regions.
|
10.1371/journal.pbio.1001314 | Essential Role for miR-196a in Brown Adipogenesis of White Fat Progenitor Cells | The recent discovery of functional brown adipocytes in adult humans illuminates the potential of these cells in the treatment of obesity and its associated diseases. In rodents, brown adipocyte-like cells are known to be recruited in white adipose tissue (WAT) by cold exposure or β-adrenergic stimulation, but the molecular machinery underlying this phenomenon is not fully understood. Here, we show that inducible brown adipogenesis is mediated by the microRNA miR-196a. We found that miR-196a suppresses the expression of the white-fat gene Hoxc8 post-transcriptionally during the brown adipogenesis of white fat progenitor cells. In mice, miR-196a is induced in the WAT-progenitor cells after cold exposure or β-adrenergic stimulation. The fat-specific forced expression of miR-196a in mice induces the recruitment of brown adipocyte-like cells in WAT. The miR-196a transgenic mice exhibit enhanced energy expenditure and resistance to obesity, indicating the induced brown adipocyte-like cells are metabolically functional. Mechanistically, Hoxc8 targets and represses C/EBPβ, a master switch of brown-fat gene program, in cooperation with histone deacetylase 3 (HDAC3) through the C/EBPβ 3′ regulatory sequence. Thus, miR-196a induces functional brown adipocytes in WAT through the suppression of Hoxc8, which functions as a gatekeeper of the inducible brown adipogenesis. The miR-196a-Hoxc8-C/EBPβ signaling pathway may be a therapeutic target for inducing brown adipogenesis to combat obesity and type 2 diabetes.
| Obesity is caused by the accumulation of surplus energy in a fatty tissue called white adipose tissue (WAT) and can lead to important health problems such as diabetes. Mammals additionally possess brown adipose tissue (BAT), which serves to generate body heat to stabilize body temperature under exposure to cold, and is abundant in hibernating animals and human neonates. In performing its function BAT consumes energy, thereby reducing WAT fat accumulation. Recent studies have shown that exposure to a cold environment stimulates the partial conversion of WAT to BAT in mice, and given that human adults have a limited amount of BAT, such a conversion has the potential to afford a novel method of obesity control. Here, we analyze the molecular mechanism of this conversion using genetically manipulated mice and cells isolated from human adipose tissue. We find that the expression levels of a microRNA, miR-196a, positively correlate with the conversion of WAT to BAT under cold exposure conditions. We show that forced expression of miR-196a in mouse adipose tissue increases BAT content and energy expenditure, thereby rendering the animals resistant to obesity and diabetes. Mechanistically, we observe that miR-196a acts by inhibiting the expression of the homeotic gene Hoxc8, a repressor of brown adipogenesis. These findings introduce the therapeutic possibility of using microRNAs to control obesity and its associated diseases in humans.
| Brown adipose tissue (BAT) combusts excess energy through mitochondrial energy uncoupling mediated by Uncoupling protein-1 (Ucp1, also known as thermogenin) in nonshivering thermogenesis [1]. Recent discoveries of metabolically active BAT in adult humans [2]–[6] have highlighted BAT as a new therapeutic target for treating obesity and its associated diseases, such as type 2 diabetes mellitus [7]. The activity of BAT is inversely correlated with body mass index in humans [3]–[4], implying a significant role for BAT in the development of obesity. Importantly, the brown adipocyte-like cells in white adipose tissue (WAT) can be generated by cold exposure or β3-adrenergic stimulation in rodents [8]–[9], and the activity of BAT can be increased by cold exposure or β3-adrenergic stimulation in humans [2]. The molecular mechanisms underlying this inducible brown adipogenesis have not been fully elucidated.
The expression patterns of the Hox family of homeobox genes (Hox genes) are characteristically distinct between BAT and WAT [10]–[12], which implies a significant role of Hox genes in the determination of two fat types. But its significance has not been fully understood. Hox genes are representative of developmental genes and confer an anteroposterior positional identity during embryogenesis. Several Hox genes have roles in differentiation systems, such as hematopoiesis [13], myogenesis [14], and cardiogenesis [15], but relatively less is known about their roles in adipogenesis. Among the differentially expressed Hox genes, Hoxc8 is more highly expressed in WAT than in BAT and is categorized as a white-fat gene [11],[16]. These observations imply that Hoxc8 may have an unknown role in the determination of the two fat types.
microRNAs (miRNAs) are important regulators of the gene networks underlying diverse biological phenomena [17]. miRNAs are small, non-coding RNAs that base pair with specific mRNAs and suppress gene expression post-transcriptionally [18]. miRNAs constitute an essential regulatory layer at the level of the transcriptional network [19]. Through their regulatory capacity, miRNAs affect the output of signaling networks by fine-tuning or switching output levels [19] and promote or redirect dynamic flow in genetic circuits and affect differentiation [20]. The roles of miRNAs in the inducible brown adipogenesis in WAT are not well understood.
We here show that single miRNA miR-196a is capable of recruiting the metabolically functional brown adipocytes in WAT in mice. The miR-196a expression is induced in the WAT-progenitor cells in mice exposed to cold or β3-adrenergic stimulation. The induction of miR-196a is required for the brown fat gene expression and is sufficient to generate the metabolically functional brown adipocyte-like cells in WAT in mice. The target gene of miR-196a is white-fat gene Hoxc8, which directly represses the expression of C/EBPβ, a master regulator of brown adipogenesis.
Recent reports have shown that the WAT-derived progenitor cells undergo brown adipogenesis in vitro in both mice [16],[21] and humans [16],[22]. Consistently, the human fat progenitor cells derived from flank subcutaneous WAT (hereafter, WAT-progenitor cells) exhibited increased brown-fat gene expression after differentiation (Figure S1A and S1B). HOXC8 is categorized as a white-fat gene [16] and RNA-seq analysis revealed that HOXC8 was most highly expressed among the clustered HOX genes in the human WAT-progenitor cells (Figure 1A and 1B). We noticed that HOXC8 was down-regulated in the differentiated adipocytes (Figure 2A and 2B). Contrarily, the expression of HOXC6 did not change after differentiation (Figure S1C) and was not particularly high in WAT (Figure S1D), though HOXC6 is located adjacent to HOXC8 in HOXC cluster and was the second most highly expressed gene (Figure 1A and 1B). These results implied the existence of specific regulatory machinery for HOXC8 expression. Down-regulation of HOXC8 was observed at the protein level (Figure 2C) but not at the mRNA level (Figure 2D). These results implied that HOXC8 might be regulated post-transcriptionally. Transduction of HOXC8 in the human WAT-progenitor cells suppressed the brown-fat genes including C/EBPβ [23], UCP1 [24], and ADIPSIN (also known as CFD) (Figure 2E) [23]. In contrast, HOXC8 did not suppress the white-fat genes including leptin [11], CD24 [25], HMGA2 [26], and ADIPOQ (also called adiponectin) (Figure 2E). These results suggested that HOXC8 might regulate the brown-fat genes and that HOXC8 might be an important regulator for brown adipogenesis of the WAT-progenitor cells.
To extend our findings in vitro to in vivo, we proceeded to a mouse model of brown adipogenesis. In mice, the Hoxc8 expression was higher in WAT than BAT and other tissues (Figure S2). Stromal vascular fraction (SVF) of fat depots contains fat progenitor cells (hereafter, SVF cells). The Hoxc8 expression was suppressed after the SVF cells were induced to undergo brown adipogenesis (Figure 3A and 3B) and expressed Ucp1 (Figure 3C), Pgc-1α, and C/EBPβ (Figure 3D). In mice, brown adipogenesis can be induced in WAT by administering a β3-adrenergic agonist, CL-316,243, or by exposing mice to cold environment. After administration of CL-316,243, the expression of Hoxc8 was down-regulated prominently in inguinal WAT (ingWAT) (Figure 3E). The down-regulation of Hoxc8 was relatively modest in epididymal WAT (epiWAT) and interscapular BAT (iBAT) than in ingWAT (Figure 3E). To delineate the Hoxc8 expression changes during white and brown adipogenesis, the Hoxc8 expression levels were compared between the progenitor cell fraction (SVF) and tissue fraction mainly composed of mature adipocytes. As a result, the Hoxc8 expression is slightly increased in saline-treated WAT than in SVF and is down-regulated in CL-316,243-treated fat that underwent brown adipogenesis, indicating that Hoxc8 is down-regulated specifically during brown adipogenesis, but not during white adipogenesis (Figure 3F). Thus, the down-regulation of Hoxc8 is observed during brown adipogenesis both in vitro and in vivo.
We next sought to identify the mechanism underlying the down-regulation of Hoxc8 during brown adipogenesis. Post-transcriptional regulation of Hoxc8 was suggested by the in vitro experiments. Characteristically, a number of Hox genes are regulated by miRNAs [14],[27]–[29] and the Hoxc8 expression can be down-regulated by evolutionally conserved miR-196a via translational inhibition during vertebrate development [28]. There are two genes encoding miR-196a (miR-196a-1 and miR-196a-2) located within the Hox gene clusters [28]. Based on the hypothesis that Hoxc8 might be regulated by miR-196a, we investigated the miR-196a expression during the brown adipogenesis in mice. We found that the miR-196a expression was induced in WAT depots of mice exposed to cold environment or β3-adrenergic stimulations (Figure 4A). More specifically, miR-196a was more highly induced in the SVF cells (Figure 4B) than in mature adipocytes (Figure S3). Thus, miR-196a expression is induced in the SVF cells in mice exposed to β3-adrenergic stimulation or cold exposure. The in situ hybridization analysis of miR-196a showed the induction of miR-196a in WAT after CL-316,243 administration (Figure 4C). Based on the finding that the miR-196a expression is induced during the brown adipogenesis in WAT in mice, we next investigated whether the miR-196a induction is required for the induction of brown adipogenesis and Hoxc8 suppression. In vitro, the miR-196a expression is induced during the differentiation of WAT-progenitor cells derived from both mice (Figure 4D) and humans (Figures S4A). More detailed analyses showed that miR-196a was induced by forskolin, an adenylyl cyclase activator, implying the significant role of cyclic AMP pathway to regulate miR-196a expression (Figure S4B). To address the necessity of miR-196a in the brown adipogenesis, antisense oligonucleotide (ASO) against mR-196a was transfected to the mouse SVF cells. The miR-196a expression was suppressed in the transfected cells (Figure 4E) and the Hoxc8 expression was recovered in the transfected adipocytes (Figure 4F), indicating that Hoxc8 suppression was mediated by miR-196a. The ASO against miR-196a suppressed the expression of Ucp1 (Figure 4G and 4H) and other brown-fat genes (Figure 4H), but not the leptin expression, indicating that miR-196a is necessary for the brown fat gene expression. Thus, the upregulation of miR-196a is required for the induction of brown fat gene expression during the differentiation of WAT-progenitor cells.
We next sought whether the findings above are possible to be generalized to the conventional brown adipogenesis, which occurs in the iBAT. The miR-196a expression level was significantly lower in iBAT than WAT (Figure S4C) and was not altered during the differentiation of the iBAT-SVF cells (Figure S4D), suggesting that miR-196a might not be involved in conventional brown adipogenesis in iBAT. Furthermore, endogenous expression of Hoxc8 was not detected in iBAT-SVF cells (Figure S5). Taken together, miR-196a is upregulated in the WAT-progenitor cells during the inducible brown adipogenesis in mice and is required for the induction of brown fat gene expression.
We next asked whether Hoxc8 was an essential target of miR-196a for the induction of brown-fat genes. We cloned the wild-type (Hoxc8-wt3′UTR) and miR-196a-binding site-deleted (Hoxc8-ΔmiR-196-BS) Hoxc8-3′UTR into a pCX4 retroviral vector and transduced these constructs into human WAT-progenitor cells (Figure S6A). The exogenous expression levels were comparable among the constructs (Figure S6A). After the adipogenic induction, the protein expression of Hoxc8 was suppressed in the Hoxc8-wt3′UTR-transduced cells than in Hoxc8-ΔmiR-196-BS- or Hoxc8-transduced cells (Figure S6B), suggesting that the suppression of Hoxc8 was dependent on the miR-196a-binding site in the Hoxc8 3′UTR. The brown fat gene expression was specifically high in the Hoxc8-wt3′UTR-tranduced cells (Figure S6C), indicating that the induction of brown-fat genes was regulated in a manner dependent on the miR-196a-binding site of Hoxc8 mRNA. These results suggest that miR-196a regulates brown-fat genes through suppression of Hoxc8. To further corroborate that Hoxc8 suppression is an important step, Hoxc8 was knocked down using Hoxc8 shRNA (Figure S7). As a result, the brown-fat genes including C/EBPβ and Ucp1 were induced (Figure S7A and S7B), indicating that the suppression of Hoxc8 is a critical step for the induction of brown-fat genes.
Based on the finding that miR-196a is required for the inducible brown adipogenesis, we next addressed whether miR-196a is capable of inducing brown adipogenesis in mice. We created transgenic mice in which miR-196a and EGFP were expressed under the control of the aP2 promoter/enhancer, which is exclusively active in adipose tissues [30]. The transgenic mice (hereafter, the miR-196a mice) were born in a Mendelian ratio and were viable. The SVF cells isolated from the miR-196a mice were EGFP-negative immediately upon isolation, but they became EGFP-positive while they were kept in culture (Figure S8A) and expressed miR-196a (Figure S8B), resulting in Hoxc8 suppression (Figure S8C and S8D). After differentiation induction, the cells expressed more intense EGFP and underwent adipogenesis. The aP2 promoter activity was observed in the fibroblast-like cells in ingWAT depots (Figure S8E), which might represent the fat progenitor cells undergoing adipogenesis. The SVF cells isolated from the miR-196a mice expressed brown-fat genes more highly than the cells from wild-type (WT) mice after differentiation in vitro (Figure S8F), indicating that miR-196a promotes brown adipocyte differentiation of the WAT-progenitor cells. To ask whether the miR-196a function is cell-autonomous, the human WAT-progenitor cells were transduced with lentivirus expressing miR-196a. As a result, miR-196a enhanced the brown fat gene expression during differentiation, indicating the cell-autonomous function of miR-196a (Figure S9).
In vivo, the gene-expression analysis revealed an induction of brown-fat genes, including C/EBPβ, Prdm16, and Ucp1 in ingWAT (Figure 5A), and the histological analysis revealed clusters of multilocular cells with Ucp1 expression (Figure 5B). It is known that different WAT depots respond to brown fat-inducing stimulations to different extents [31], and we therefore addressed the responses to the miR-196a expression in different fat depots. The miR-196a expression levels were comparable among the different fat depots in the miR-196a mice (Figures 5C and S10). The induction of C/EBPβ, Ucp1, and Pgc-1α was more prominent in the ingWAT than in the epiWAT (Figure 5D and 5E) and was further augmented after CL-316,243 treatment (Figure 5D and 5E). In the iBAT, no appreciable influence of miR-196a was observed (Figure 5D and 5E). Thus, miR-196a induces the brown adipocyte-like cells with characteristic appearance and gene expression profile of brown adipocytes in WAT.
Based on the finding that miR-196a is capable of inducing the brown adipocyte-like cells, we next addressed whether they were metabolically functional. The miR-196a mice showed a tendency to be leaner than WT mice (Figure 6B), and even when fed a high-fat diet, they exhibited resistance to obesity (Figure 6A and 6B), despite the fact that their food intake tended to be increased compared with that of the WT littermates (Figure 6C). The weight reduction was attributable to a reduced fat accumulation (Figure S11). To interrogate the mechanism behind the obesity resistance of the miR-196a mice, indirect calorimetry was performed. We used mice with similar body weight under a normal diet. As a result, the oxygen consumption (Figure 6D) and the energy expenditure (Figure 6E and Table S1) were enhanced during both the light and dark phases in the miR-196a mice compared to the WT mice, indicating the accelerated energy metabolism. The difference of the oxygen consumption and the energy expenditure was even enlarged when the mice were fed a high-fat diet (Figure S12). The core body temperature was higher in the miR-196a mice than in the WT mice (Figure 6F). These findings suggest that miR-196a boosted the cellular energy combustion through the induction of brown adipocyte-like cells. We next analyzed impacts of miR-196a on glucose metabolism in the miR-196a mice. In the glucose tolerance tests, the miR-196a mice showed lower blood glucose (Figure 6G) and insulin levels (Figure 6H). After insulin administration, they exhibited more pronounced declines in their blood glucose levels (Figure 6I). These results imply that miR-196a prevented the mice from developing insulin resistance, the premorbid condition of type 2 diabetes. Taken together, these findings suggest that the miR-196a-induced brown adipocyte-like cells are metabolically functional and have favorable impacts on glucose metabolism in mice.
The concept that miR-196a induces brown adipogenesis through the suppression of Hoxc8, which might function as a gatekeeper of brown adipogenesis in WAT, facilitated us to investigate the target gene of Hoxc8 transcription factor. The chromatin immunoprecipitation (ChIP) assays among the candidate genes revealed a significant enrichment of Hoxc8 in the C/EBPβ locus in the mouse genome (Figure 7A). C/EBPβ is a crucial regulator of brown adipogenesis, which is highly expressed in BAT compared to WAT [23]. The enrichment was found in the 3′ region, which harbors high interspecies conservation (Figure 7B, “4”). In human WAT-progenitor cells, too, the enrichment of HOXC8 was observed in the C/EBPβ 3′ region (Figure 7C and 7D). The enrichment of HOXC8 was also observed in the promoter of osteopontin (OPN) gene used as a positive control (Figure 7C) [32]. To ask whether the binding of Hoxc8 in the 3′ of C/EBPβ has a regulatory role, we performed the reporter assay by replacing the C/EBPβ coding region with luciferase gene. Indeed, the C/EBPβ 3′ sequence induced luciferase activity, which was further augmented by adipogenic stimulation (Figure 7E). This luciferase expression was suppressed by concomitant transfection of Hoxc8 but not by that of Hoxc8 with a mutated homeodomain (HDm) lacking DNA-binding capacity (Figure 7F) [33]. These results implied that Hoxc8 regulates the C/EBPβ expression via the C/EBPβ 3′ regulatory sequence. Furthermore, the suppressive effect of Hoxc8 was abolished by trichostatin A, a histone deacetylase (HDAC) inhibitor, indicating that the suppressive effect involves histone deacetylation (Figure 7G). In this regard, Hoxc8 interacted with HDAC3 (Figure 7H) [34]–[35], but not with HDAC1 or HDAC2. The interaction was independent of the DNA binding capacity of Hoxc8 (Figure 7I). To further corroborate that HDAC3 cooperates with Hoxc8, HDAC3 was suppressed using siRNA (Figure 7J), resulting in partial elimination of the suppressive effects of Hoxc8 (Figure 7K). To demonstrate that C/EBPβ is an essential target of Hoxc8, C/EBPβ was transfected into the human WAT-progenitor cells that stably expressed human HOXC8, resulting in restoration of the brown-fat gene expression that had been suppressed by HOXC8 (Figure 7L). Thus, Hoxc8 targets and represses C/EBPβ in an HDAC3-dependent manner.
In summary, during the brown adipogenesis induced by cold exposure or β3-adrenergic stimulations, miR-196a is induced in WAT-progenitor cells and suppresses Hoxc8, which targets C/EBPβ, an essential regulator of brown adipogenesis. The miR-196a expression is required for the brown-fat gene expression and sufficient to induce metabolically functional brown adipocyte-like cells in WAT in mice. Our findings imply the therapeutic potential of targeting the miR-196a-Hoxc8-C/EBPβ signaling pathway that induces metabolically functional brown adipocytes in WAT to treat obesity and its associated diseases.
Recent discoveries of metabolically active BAT in adult humans have highlighted BAT as a therapeutic target for treating obesity and its associated diseases. The brown adipocyte-like cells in WAT can be generated by cold exposure or β-adrenergic stimulation in rodents, but the molecular mechanisms underlying these phenomena have not been fully elucidated. In this work, we elucidated that miR-196a induces functional brown adipocytes in WAT in mice. miR-196a is upregulated in WAT-progenitor cells during brown adipogenesis induced by cold or β-adrenergic stimulations. miR-196a is required for the brown fat gene expression and is sufficient to induce metabolically functional brown adipocyte-like cells in mice. The target gene of miR-196a is Hoxc8, which is categorized as a white-fat gene with a previously undermined role in adipogenesis. Hoxc8 directly targets and represses C/EBPβ, a master switch of brown adipogenesis. Thus, the miR-196a-Hoxc8-C/EBPβ pathway underlies the brown adipogenesis in WAT (Figure 8) and might be a therapeutic target for the treatment of obesity and type 2 diabetes.
Elucidation of the molecular mechanism regulating the brown adipogenesis in WAT is important from both a biological and clinical viewpoint. Recent studies uncovered the existence of WAT-progenitor cells that harbor a potential to differentiate to brown adipocytes [16],[21]–[22],[36]. The molecular mechanism behind the inducible brown adipogenesis in WAT is relatively unknown, but recent studies elucidated the importance of cyclooxygenase-2 [36]–[37] and Prdm16 [38]. C/EBPβ is an essential regulator of brown fat gene program [23],[39]–[41], but whether C/EBPβ has a significant role in the inducible brown adipogenesis was not fully understood. We found that miR-196a suppresses Hoxc8, thereby derepressing C/EBPβ, which leads to the activation of the brown fat gene program. Our findings imply the relevance of C/EBPβ not only in the conventional brown adipogenesis but also in the inducible brown adipogenesis in WAT.
The cellular origin of the inducible brown adipocyte-like cells in WAT is an important question. Transdifferentiation is a significant mechanism that has been reported to contribute to brown adipocyte recruitment in WAT [42]–[43]. Because the increase in Ucp1 mRNA is detectable within a few hours after cold stimulation [1],[31], and in vitro SVF cell differentiation is a longer process, transdifferentiation might have a significant role in the rapid response to stimulation. The important questions include the relative contribution of transdifferentiation and the progenitor cell-mediated mechanism in brown adipocyte recruitment throughout the different phases upon exposure to a cold environment and physiological energy regulation.
miRNAs regulate the gene networks underlying various physiological and pathological phenomena and might be therapeutic targets [18]–[19],[44]–[46]. miR-196a has been implicated in the in vitro osteoblast differentiation of human fat progenitor cells, where miR-196a suppresses Hoxc8 [47], but the in vivo relevance remains unknown. We elucidated that miR-196a is induced in the WAT-progenitor cells after the induction of brown adipogenesis, is required for the induction of brown fat gene expression, and is sufficient to induce the metabolically functional brown adipocyte-like cells in WAT.
Our observations indicate that miR-196a has only a modest, if any, effect on iBAT. The endogenous expression of Hoxc8 and miR-196a was much lower in iBAT than in ingWAT and epiWAT. The forced expression of miR-196a in mice did not yield appreciable effects in iBAT. Treatment of mice with β3-adrenergic receptor agonists usually leads to a much more moderate induction of Ucp1 expression in iBAT than in WAT depots. Although the primary cultures of brown adipocytes from iBAT are highly sensitive to β3-adrenergic activation [1], a moderate but significant induction of Ucp1 was reported in iBAT in response to β3-adrenoreceptor agonists in vivo [48]. A relatively modest response from iBAT to the β3-adrenergic receptor agonist compared with subcutaneous and visceral WAT has also been reported in other studies [16],[43],[49]. These results imply that distinct machinery regulates brown adipocyte recruitment in iBAT, which was previously suggested by Petrovic et al. [21].
A number of miRNAs function as a molecular switch [46],[50]–[53], and further elucidating how the miRNAs influence the physiological output will enable better understanding and clinical use of miRNAs.
The significance of the distinct expression patterns of Hox genes between BAT and WAT has been unknown [10]–[12]. We here demonstrate that Hoxc8 functions as an important determinant of white fat lineage and negatively regulates the induction of brown adipogenesis in WAT-progenitor cells by repressing C/EBPβ, which is a master switch of brown adipogenesis [39]–[41]. Mechanistically, Hoxc8 directly represses the C/EBPβ expression through the 3′ regulatory sequence. Similar conserved non-coding regulatory elements have been reported for the Foxp3 gene [54], and previous studies suggested that the majority of transcription factors bind to sites other than the promoter [20],[55]. Hoxc8 recruits HDAC3, which is implicated in the regulation of metabolic genes [34],[35]. Since the HDAC proteins lack DNA-binding activity, they are recruited to target genes via association with transcriptional factors [56]. Our findings imply the possible therapeutic efficacy of HDAC inhibitors for obesity through inducing brown adipogenesis, but further study is required to address the possibility.
The induction of brown adipogenesis in WAT has great therapeutic potential. Our findings suggest that the miR-196a-Hoxc8-C/EBPβ pathway may constitute a promising strategy for addressing the social and health problems caused by obesity and its associated diseases.
Mice were handled in accordance with protocols approved by the Ethics Committee for Animal Experiments of the Osaka University Graduate School of Medicine.
The coding sequence of human Hoxc8 (Gene ID: 3224) was cloned into pCX4-puro [57] and pCAGIP vector [58]. The pCX4-Hoxc8 retroviral vector was used to generate human WAT-progenitor cells stably expressing Hoxc8. Human C/EBPβ was cloned into the pCAGIP vector. The homeodomain mutant (I195A/Q198A/N199A/M202A) [33] of Hoxc8 (HDm) was created by site-directed mutagenesis. For lentivirus-mediated shRNA expression, pLenti6-miR-196a, -shHoxc8, and -shLacZ were generated from pcDNA6.2 constructs by Gateway reactions. Lentivirus was generated by cotransfection of the pLenti6 construct with packaging plasmids into 293FT cells according to the manufacturer's instruction (Invitrogen). For Hoxc8 3′UTR analysis, human Hoxc8 3′UTR sequence was cloned and inserted to the 3′ end of Hoxc8 cDNA. The miR-196a binding site (CCCAACAACTGAGACTGCCTA) was deleted to generate Hoxc8-ΔmiR-196a-BS.
Total RNA was isolated using the RNeasy Lipid Tissue Mini Kit (QIAGEN, CA). Reverse transcription and quantitative PCR were performed as previously described [59]. For microRNA quantification, total RNA was isolated using a mirVana miRNA isolation kit (Applied Biosystems). Reverse transcription and quantitative PCR were performed according to the manufacturer's instructions. A list of probes is provided in Text S1.
RNA from human white fat (WAT) progenitor cells was extracted with RNeasy (QIAGEN) following the manufacturer's instructions. 12.5 µg of total RNA were subjected to two rounds of oligo-dT purification using Ambion MicroPoly(A) Purist Kit (Ambion). 50 ng of the fragmented poly(A) RNA by using RNaseIII were ligated to SOLiD Adaptor Mix and were reverse-transcribed by using SOLiD Total RNA-Seq Kit (Life Technologies). First-strand cDNA from 100 bp to 150 bp was selected by using Agencourt AMPure XP reagent (Beckman Coulter Genomics) and was amplified by SOLiD 5′ PCR primer and barcoded SOLiD 3′ PCR primers (Life Technologies). Sequencing libraries were prepared according to Life Technologies' protocol. RNA-seq libraries were sequenced with SOLiD 4. Mapping of resulting reads was performed by Bioscope (Life Technologies), and analysis of mapped reads (31,825,850 reads in hADSC_1 and 42,009,231 reads in hADSC_2) was performed by Cufflinks [60].
Human WAT-progenitor cells were isolated from human flank subcutaneous fat lipoaspirate (Lonza, Switzerland) and maintained in mesenchymal stem cell growth medium (Lonza). For adipogenesis, 2-d post-confluent cells were treated with an induction medium containing 0.5 mM IBMX, 10 µg/ml insulin, and 1 µM dexamethasone (MDI). The induction medium was changed every 2 d. Forskolin (40 µM, Sigma-Aldrich) was added to the medium as noted. Antisense oligonucleotide against miR-196a (Anti-miR miRNA inhibitor, AM10068, Ambion) was transfected according to the manufacturer's instruction. The fat progenitor cells were isolated from inguinal white adipose tissue (WAT) or interscapular BAT (iBAT) of C57Bl/6 mice using a standard method [61]. Adipogenic induction was performed by treating the cells with the induction medium for 2 d.
Western blotting was performed with antibodies against Hoxc8 (1∶1,000, ab86236, abcam), C/EBPβ (1∶200, sc-150, Santa Cruz Biotechnology, CA), UCP1 (1∶1,000, U6382, Sigma-Aldrich), PGC-1α (1∶1,000, ab54481, abcam), β-actin (1∶5,000, AC-15, Sigma-Aldrich), and GAPDH (1∶5,000, ab8245, abcam). The secondary antibodies (GE Healthcare) were used at a 1∶1,000 dilution ratio. Immunoreactive bands were detected with Chemi-LumiOne L (Nacalai Tesque) or ECL plus (GE Healthcare). Densitometry was performed with the ImageJ software (NIH; http://rsb.info.nig.gov/ij/).
Immunocytochemistry was performed using antibodies against Hoxc8 (1∶200, MMS-286R, Covance), Hoxc6 (1∶200, ab41587, Abcam), Pgc-1α (1∶300, ab54481, Abcam), or UCP1 (1∶500, ab10983, Abcam) as previously described [59]. The primary antibodies were detected using anti-mouse-Alexa Fluor 546, anti-mouse-Alexa Fluor 488, or anti-rabbit-Alexa Fluor 546 (1∶1,000, Invitrogen). Cells were counterstained with CellTracker Green Bodipy (Invitrogen), Bodipy 493/503 (D3922, Invitrogen), and 4′-6-diamidino-2-phenylindole (DAPI, Invitrogen).
These experiments were approved by the Ethics Committee for Animal Experiments of the Osaka University Graduate School of Medicine. Male outbred C57Bl/6 mice were used. For acute cold-exposure studies, 3- to 4-mo-old male mice were housed at 4°C for 5 h. For β3-adrenaline receptor stimulation, CL-316,243 (Sigma), at 0.5 mg/kg, was injected intraperitoneally once daily for 7 d. Transgenic mice with fat-specific forced expression of miR-196a were generated using a transgene encoding miR-196a driven by the enhancer/promoter of the aP2 gene [30], and littermates were used as the wild-type controls.
Inguinal fat sections were fixed in 10% buffered formalin and stained with hematoxylin-eosin. For immunohistochemistry, paraffin-embedded sections were incubated with antibodies against UCP1 (1∶1,000, ab10983, Abcam) followed by detection using ABC Vectastain-Elite kit (Vector Labs). Nuclei were counterstained with modified Mayer's hematoxylin (Diagnostic BioSystems).
Inguinal WAT depots of mice were dissected after perfusion and fixation with Tissue Fixative (Genostaff), embedded in paraffin, and sectioned at 6 µm. The sections were de-waxed with xylene and rehydrated. The sections were fixed with 4% paraformaldehyde (PFA) for 15 min, treated with 8 µg/ml proteinase K for 30 min at 37°C, re-fixed with 4% PFA, and placed in 0.2 N HCl for 10 min. The sections were acetylated with 0.1 M tri-ethanolamine-HCl, pH 8.0, and 0.25% acetic anhydride for 10 min. After being washed with PBS, the sections were treated with PBS at 80°C for 5 min. The sections were hybridized with 3′-digoxygenated probes (18 pmol/ml, miR-196a-AS-LNA1: cCcaAcaAcaTgaAacTacCta, Control (Ctrl)-LNA1: cGacTacAcaAatCagCgaTtt, capitals denote LNA) in Probe Diluent-1 (Genostaff) at 50°C for 16 h and washed in 5× HybriWash (Genostaff) at 50°C for 20 min, 50% formamide in 2× HybriWash at 50°C for 20 min, twice in 2× HybriWash at 50°C for 20 min, and twice in 0.2× HybriWash at 50°C for 20 min. The sections were treated with 0.5% blocking reagent (Roche) in TBST for 30 min and incubated with anti-DIG AP conjugate (1∶1,000, Roche) for 2 h at RT. The sections were washed twice with TBST and incubated in a solution with a composition of 1,000 mM NaCl, 50 mM MgCl2, 0.1% Tween-20, 100 mM Tris-HCl, pH 9.5. Coloring reactions were performed with NBT/BCIP solution (Sigma) overnight followed by counterstaining with Kernechtrot stain solution (Mutoh).
Mice were given a standard diet or a high-fat diet (20.4% protein, 33.2% fat, 46.4% carbohydrates by calories; MF+; Oriental Yeast Co., Japan). Metabolic measurements were performed on 3- to 4-mo-old mice with similar body weight that were given a standard diet. Food intake and body weight were measured daily and weekly, respectively. For glucose tolerance tests, the mice were deprived of food for 16 h and were injected intraperitoneally with glucose (2 g/kg). For insulin tolerance tests, the mice were allowed ad libitum access to food followed by an intraperitoneal injection of human insulin (0.75 U/kg, Eli Lilly). The plasma concentration of glucose was measured with a Glucometer (Sanwa Kagaku Kenkyusho, Japan), and insulin was measured with an ELISA (Morinaga Institute of Biological Science, Japan). Indirect calorimetry was performed under 12 h light and dark cycles beginning at 8:00 a.m. and 8:00 p.m., respectively. After 1 d of acclimation, V˙O2 and V˙CO2 were recorded every 3 min over 3 d using the Metabolism Measurement System (MK-5000, Muromachi Kikai, Japan). Energy expenditure (EE) was calculated using the equation of Weir: EE (kcal/kg/h) = (3.815×V˙O2)+(1.232×V˙CO2). For body temperature measurement, mice were housed singly and unrestrained and had free access to food and water. Body temperature was measured using a rectal probe (Perimed, Sweden).
Chromatin immunoprecipitation was performed as previously described [62] with 3T3-L1 preadipocytes expressing Flag-tagged human Hoxc8. Primer sequences are listed in Text S1.
The C/EBPβ3′-luciferase constructs (C/EBPβ-Luc) were generated by cloning the 3′ sequence of the human C/EBPβ gene (+1,021 to +1,837) into the downstream of luciferase gene in pGL3 promoter plasmid (Promega). Dual luciferase assays were performed as previously described [62] with 3T3-L1 preadipocytes. Trichostatin A (330 nM, Sigma-Aldrich) was added 4 h after transfection as indicated. Mission siRNA (Sigma) for HDAC3 (sense: 5′GUAUCCUGGAGCUGCUUAATT, antisense: 5′UUAAGCAGCUCCAGGAUACTT) was transfected using Neon transfection system (Invitrogen).
Nuclear extracts were prepared as previously described [62] from 3T3-L1 preadipocytes transfected with Flag-Hoxc8, pretreated with Protein G Sepharose beads (Amersham Bioscience), and incubated with anti-Flag M2 Affinity Gel (A2220, Sigma-Aldrich) or control mouse IgG AC (Santa Cruz) overnight at 4°C. The beads were washed 3 times with nuclear isolation buffer containing 500 mM NaCl and 0.15% NP-40. Purified proteins were subjected to immunoblotting using antibodies against HDAC1 (3∶1,000, Millipore), HDAC2 (1∶2,000, H3159, Sigma), and HDAC3 (1∶500, ab16047, Abcam).
The statistical analysis was performed with StatView 5.0 software, JMP8 (SAS Institute, NC) and SPSS (IBM). All results are expressed as mean ± SEM. The data were compared using ANOVA, followed by Dunnett's test for pairwise comparisons against controls and by Tukey's test for multiple comparisons. For the analysis of energy expenditure, a one-way analysis of covariance (ANCOVA) was conducted. The body weight was used as the covariate. Statistical significance was defined as p<0.05.
The RNA-seq data have been submitted to the NCBI Sequence Read Archive (SRA). The accession number is SRA048274.1.
|
10.1371/journal.pbio.2002117 | Hierarchical patterning modes orchestrate hair follicle morphogenesis | Two theories address the origin of repeating patterns, such as hair follicles, limb digits, and intestinal villi, during development. The Turing reaction–diffusion system posits that interacting diffusible signals produced by static cells first define a prepattern that then induces cell rearrangements to produce an anatomical structure. The second theory, that of mesenchymal self-organisation, proposes that mobile cells can form periodic patterns of cell aggregates directly, without reference to any prepattern. Early hair follicle development is characterised by the rapid appearance of periodic arrangements of altered gene expression in the epidermis and prominent clustering of the adjacent dermal mesenchymal cells. We assess the contributions and interplay between reaction–diffusion and mesenchymal self-organisation processes in hair follicle patterning, identifying a network of fibroblast growth factor (FGF), wingless-related integration site (WNT), and bone morphogenetic protein (BMP) signalling interactions capable of spontaneously producing a periodic pattern. Using time-lapse imaging, we find that mesenchymal cell condensation at hair follicles is locally directed by an epidermal prepattern. However, imposing this prepattern’s condition of high FGF and low BMP activity across the entire skin reveals a latent dermal capacity to undergo spatially patterned self-organisation in the absence of epithelial direction. This mesenchymal self-organisation relies on restricted transforming growth factor (TGF) β signalling, which serves to drive chemotactic mesenchymal patterning when reaction–diffusion patterning is suppressed, but, in normal conditions, facilitates cell movement to locally prepatterned sources of FGF. This work illustrates a hierarchy of periodic patterning modes operating in organogenesis.
| Repeating anatomical units, forming a pattern, are present in different parts of the body. This is particularly obvious in the skin, where the many hair follicles become regularly arranged and fixed in place in the embryo as a pattern of evenly spaced spots. The basis for the formation of such repeating patterns has attracted a number of theoretical explanations. One influential theory is that cells first issue chemical signals to one another to produce a map of the pattern, which they then follow to assemble a structure, like a hair follicle. Alternative theories instead suggest that cells cluster together directly, without reference to a pre-existing map, to drive pattern formation. In this study, we find that a set of chemical signals known to be important for early hair follicle formation form a network capable of producing a patterned template to which cells then respond, moving according to its instructions. This supports the signal-based theory for pattern formation. Strikingly, though, by imposing the conditions of the incipient hair follicle evenly across the entire skin, we reveal that skin cells can also form patterns directly by coalescing, without requiring instructions from a pre-existing pattern. However, this ability of cells to pattern directly is normally subservient to their following of a quickly forming pattern template defined by chemical signals. This work highlights that different ways of making biological patterns can coexist in the same embryonic organ but that one is normally subordinated to the other.
| Diverse structures, such as the skeletal elements of the limb, rugae of the palate, cartilaginous rings of the trachea, intestinal villi, and feathers, scales, or hair follicles, develop in a periodically patterned manner. Although many specific models to explain the spontaneous emergence of such repeating patterns in embryonic tissues have been proposed [1], these can be grouped into 2 general classes (Fig 1A). The first class, based on the Turing reaction–diffusion system, relies on the operation of 2 opposing signalling processes: an activator, which is self-enhancing and has a limited spatial range, coupled with the production of an inhibitor with a greater spatial range. These activating and inhibiting processes were originally presented as a pair of chemical signals [2], though more complex networks can produce similar patterns [3–7]. These activating and inhibitory processes induce a spatially patterned change in cell state, typically reflected in altered gene expression. This arrangement of cell states, termed a prepattern, is then used as a template to produce anatomical structures by locally influencing cell aggregation, growth, or survival (Fig 1A). The second class of models focuses on the potential of mobile mesenchymal cells to generate periodic foci of high cell density directly through chemotaxis, adhesion, or mechanical deformation of their environment [8–13]. Such mesenchymal patterning does not require opposing intercellular signals, as the self-activation phenomenon arises from local cell clustering, while the inhibitory effect is achieved by widespread cell depletion as these are drawn away to nascent clusters (Fig 1A). Thus, these 2 theories differ in the entity that is moving: in the Turing reaction–diffusion system, cell movement is limited and chemical signals diffuse, while in mesenchymal self-organisation systems, moving cells are themselves agents of pattern formation. Both types of model produce similar patterns in computational simulation [4, 14–16] as they both ultimately rely on the principle of local self-activation and long-range inhibition [17]. Thus, experimental investigation is required to define the contribution made by each mechanism during organ development.
In the mouse embryo, a periodic pattern of hair follicles first arises at late embryonic day 13 (E13) [18, 19]. At this stage, the skin is composed of an epidermal sheet overlying an extracellular matrix containing dispersed mesenchymal cells. Sites of hair follicle initiation first become identifiable as groups of cells expressing specific marker genes and by cellular reorganisation (Fig 1B). The latter involves the closer packing of epidermal cells to form a placode [20] and the clustering of mesenchymal cells to form a dermal condensate directly underneath. Modulation of hair follicle size, shape, and spacing during skin growth and after experimental perturbation has provided the strongest evidence that a local self-activation and long-range inhibition system ultimately determines the hair follicle arrangement. Such a process can explain why new follicles are inserted between existing ones as the skin expands [21] and why the follicles will align along the edge of a cut made prior to pattern formation [4, 22]. This mechanism also explains how the hair follicle pattern can transition from an array of spots to one of stripes in a labyrinthine pattern [4]. As both cell–cell signalling and mesenchymal cell aggregation are prominent features of early hair follicle formation, it is plausible that either process, or a combination of both, is responsible for defining the hair follicle array.
Classical tissue recombination experiments indicated that the dermal mesenchyme is the compartment in which pattern generation occurs, this spatial information subsequently being conveyed to the epidermis through inductive signalling [23, 24]. This sparked a search for the molecular identity of the “first dermal message” thought to induce the epidermal placode pattern according to a dermal prepattern. Several intercellular signalling pathways have since been found to be critical for early hair follicle development, though none have the exact characteristics of the hypothesised first dermal message. Nascent follicle primordia are foci of high wingless-related integration site (WNT) activity [18, 25, 26] and display elevated production of fibroblast growth factors (FGFs) [27], bone morphogenetic proteins (BMPs), and BMP inhibitors [22, 28, 29]. Consistent with a driving role for the dermis in hair follicle patterning, this process does not initiate in mice upon abolition of dermal WNT/β-catenin activity [30]. However, activity of WNT/β-catenin in the epidermis is also essential for hair follicle patterning, as no focal expression of any molecular markers or cellular rearrangements that accompany normal hair follicle formation occur when this is ablated [25]. Furthermore, forced activation of epidermal β-catenin is sufficient to drive placode identity across the entire epidermis [31, 32]. Mutation of epithelial FGF receptor 2 leads to loss of cell rearrangements and placode markers [33], while mutation of Fgf20 allows formation of epidermal placodes and expression of a nearly full suite of epidermal placode markers without any sign of an accompanying dermal condensate nor of patterned dermal gene expression [27]. However, administration of FGF7 to skin inhibits hair follicle formation [34], suggesting both positive and negative roles for FGF signalling in this process. The BMP family act as inhibitors of hair follicle formation, based on their effects when applied to cultured skin [21, 22, 29], an increased primary follicle density when epithelial BMP receptor is deleted [35], and the suppression of follicle formation when the BMP inhibitor NOGGIN is ablated [28].
Once the spatial pattern has been defined, the cells selected to become a follicle activate expression of other genes to progress their development into construction of the mature organ. These later-acting genes are typically expressed in the ‘de novo’ mode, indicating their activation only after assumption of the new cell fate, as opposed to the ‘restrictive’ mode of expression, which is characteristic of genes involved in the patterning process itself [36]. An example is Shh, which is expressed only once hair follicle cell fate has been established [37], and, agreeing with its role being only subsequent to definition of the follicle pattern, mutations in this gene do not impair hair follicle fate acquisition but rather arrest the follicles’ development due to lack of growth [38].
Simple signalling interactions have been proposed as the basis of hair follicle periodic patterning, though none represent a complete system capable of de novo pattern formation. WNT signalling coupled with induced expression of Dickkopf (DKK) family members was proposed to contribute to a reaction-diffusion system, though no WNT activator positive feedback loop was identified [39]. An activator/inhibitor relationship between ectodysplasin A receptor (EDAR) and the BMP pathway was suggested to contribute specifically to primary hair patterning [5, 22]. This system, however, stabilises a labile prepattern of focal WNT/β-catenin activity [22, 25, 40, 41], rather than acting to create order in naïve tissue. Thus, these proposed models do not integrate all pathways implicated in patterning and do not report a set of interactions sufficient to act as a periodicity generator, raising questions as to whether a simple reaction–diffusion system underlies hair follicle patterning or whether this process may have inputs other than intercellular signalling that contribute to symmetry breaking.
Here, we identify a set of interactions between the BMP, FGF, and WNT pathways capable of breaking symmetry to produce a prepattern that guides local dermal cell condensation. Strikingly, by modulating components of this system to mimic the microenvironment of the hair follicle primordium, we identify a transforming growth factor (TGF) β-driven patterning system in dermal mesenchyme that is independent of epidermal instruction. Whereas TGFβ signalling is capable of driving mesenchymal self-aggregation when reaction–diffusion signalling is suppressed, in normal development, this pathway potentiates cell accumulation at sites of focal FGF production, thereby assuring timely dermal condensate formation according to the epidermal template. Thus, both reaction–diffusion signalling and mesenchymal self-organisation potentials reside in the developing skin, but the former dominates and directs the latter, highlighting the hierarchical nature of patterning mechanisms.
To determine the functions and relationships between cell signalling pathways and whether their interactions are sufficient to constitute a complete pattern forming network, we focussed on the BMP, FGF, and WNT/β-catenin pathways [25–34]. We assessed the effects of modulating each pathway in E13.5 cultured skin on both epidermal placode, defined by expression of Dkk4 [41, 42], and using the TCF/Lef::H2B-green fluorescent protein (GFP) reporter line [43] to visualise dermal cell arrangement and condensation (Fig 1C and S1C Fig). Treatment with BMP4 inhibited placode and dermal condensate formation, while the BMP receptor inhibitor LDN193189 increased placode and condensate size. We focused primarily on the FGF9/16/20 subfamily signalling as FGF family representatives due to their demonstrated involvement in hair follicle formation [27, 44]. Treatment with recombinant FGF9 suppressed placode and condensate appearance (Fig 1C), as did FGF7, a member of a different subfamily (S2 Fig). Blocking FGF signalling with SU5402 inhibited both Dkk4 expression and dermal condensate appearance, consistent with FGF signalling being both necessary for, and inhibitory to, hair follicle development (Fig 1C). Treatment of skin explants with the GSK3β-inhibitor CHIR99021 to stimulate β-catenin activity led to the enlargement of hair follicle primordia, transitioning from discrete spots to a merged labyrinthine pattern. Inhibition of WNT/β–catenin signalling reduced placode density, accompanied by weaker expression of Dkk4 (Fig 1C). In each condition, the effects on placode pattern were matched by corresponding changes in dermal condensate pattern. These pattern transitions are in agreement with WNT signalling being activatory, BMPs functioning as inhibitors, and FGFs playing a more complex role, being both required for and inhibitory to hair follicle initiation.
Having defined conditions to stimulate and repress each of these 3 signalling pathways, we set out to delineate their transcriptional regulatory interactions and assess whether these could comprise a functional pattern forming network. Turnover of activator and inhibitor molecules is required to produce a reaction–diffusion pattern [2, 4, 17], and simulations show that activator and inhibitor half-lives have a major influence on the timing of pattern emergence [4]. As mouse skin takes approximately 10 h to form a pattern from a naïve state [22], in a simple reaction–diffusion network, this constrains the half-lives of the oscillating components of this system to being under 90 minutes [4]. We used this temporal constraint as a filter to identify transcripts from these 3 pathways that could play a driving role in pattern formation.
We performed a transcriptome-wide screen of mRNA half-lives in E13.5 mouse dermis and epidermis, subsequently applying a data filtering criterion based on half-life (t1/2 ≤ 90 minutes) and focusing upon extracellular molecules (ligands, receptors, and extracellular signal-attenuating molecules) of the BMP, FGF, and WNT pathways (S3 Fig, S2 and S3 Tables, S1A Appendix for detail). The focus on ligand, receptor, and extracellular antagonist molecules is motivated by their critical role in reaction–diffusion processes by conveying information between cells [2, 17, 45]. This yielded a candidate list of mRNA species that fit within the 3 signalling pathways of interest, have half-lives of less than 90 minutes, encode extracellular proteins, and are expressed in embryonic skin (S1 Table). This set contains many genes already implicated in skin patterning, including Bmp2 and Bmp4 [21, 22, 29], the BMP inhibitor Nog [28, 46], the WNT inhibitors Dkk1 and Dkk4 [26, 39, 41, 47], and Fgf7 and Fgf20 [27, 34]. Transcripts encoding proteins in the EDAR and TGFβ pathways had longer half-lives, consistent with their acting later to stabilise emerging patterns [22] and promote morphogenesis [48].
To identify regulatory relationships between the BMP, FGF, and WNT pathways, we stimulated and repressed each pathway in unpatterned E13.5 skin explants for 6 h and assessed the resulting change in abundance of each candidate transcript (Fig 1D). This approach identified a number of previously reported regulatory interactions, including upregulation of Fgf20 and Dkk4 by WNT activity [27, 41, 42] and stimulation of Bambi and Nog by BMP signalling [49, 50].
After excluding candidates unresponsive to BMP, FGF, or WNT signal modulation, 15 genes were found to be regulated within this period (Bmp2, Bmp4, Ctgf, Dkk1, Dkk4, Wnt9a, Ror2, Fgf7, Fgf10, Fgf18, Fgf20, Rgma, Fzd10, Bambi, and Nog). From these data, we derived a transcriptional network describing the interactions between the candidates (Fig 1E). In this network, WNT/β–catenin signalling stimulates expression of patterning genes and hair follicle primordium markers, while BMP has opposing effects on most of these. FGF represses expression of all epithelial WNT/β–catenin activated targets (Dkk4, Bmp2, and Fgf20), which are indicators of placode fate, but does not alter expression of any dermally expressed genes in the network, even though we detect the classical FGF target Etv5 being upregulated in both tissue layers (S4A Fig). The inhibition of hair follicle development following widespread application of FGF to skin cultures (Fig 1C) can thus be explained by its suppression of epithelial β-catenin activity, consistent with the increased epidermal β-catenin activity reported in Fgf20 null embryonic skin [27] and reduction of active β-catenin protein in FGF-stimulated epidermis (S4B and S4C Fig).
In this network, some features of classical Turing activator–inhibitor dynamics are present, including activator stimulation of inhibitor production, but we did not detect direct positive feedback for any pathway. Rather, each pathway displays prominent self-inhibition. Thus, in its number of components and its structure, this network does not conform to the classical topology of a reaction–diffusion system. Therefore, we took a mathematical approach (S1B Appendix) to assess whether the structure of this network is capable of producing a periodic pattern.
We grouped the candidates into 5 species: BMP (Bmp2, Bmp4, Rgma), WNT (Wnt9a, Ror2, Fzd10), WNT inhibitor (Dkk1 and Dkk4), BMP inhibitor (Bambi and Nog), and FGF (Fgf20) (see S1A Appendix). We then defined a matrix in which the interactions between these species were represented by either a + sign (stimulation) or a − sign (inhibition) (S3F Fig). From the matrix of interactions, we performed linear stability analysis [2], a method that assesses whether small perturbations in a system will decrease or increase over time. If they decay, patterning cannot occur, and if they grow, there is a prospect of pattern formation, as a result of diffusion in the case of the Turing mechanism. This analysis found that the conditions of Turing instability are met when all 5 species are considered to be diffusible (Fig 1F), which is expected as each grouping contains at least 1 secreted factor. The structure of this BMP, FGF, and WNT network can, therefore, generate a stable periodic pattern (see S1C Appendix for detail) through rapid regulatory interactions. Intuitively, the main drivers in this system lie in WNT stimulating expression of a suite of hair follicle primordium-specific genes, while BMP broadly inhibits their expression and FGF acts as an inhibitor selectively in the epidermis. While direct positive feedback is not apparent in this system, the sequence through which WNT upregulates FGF, FGF downregulates BMP, and BMP downregulates WNT could have an overall positive effect that would allow WNT to undergo indirect self-upregulation.
Having defined the reciprocal relationships between BMP, FGF, and WNT signalling, including their organisation into a network capable of periodic patterning, we investigated the origin of the dermal condensate and the influences of these pathways on its formation. We performed live cell imaging of TCF/Lef::H2B-GFP E13.5 dorsal skin explants using confocal microscopy (Fig 2A, S1 and S2 Videos) to track dermal cells during condensate formation. This revealed undirected movement of dermal cells up to the timepoint of condensate formation, when local directed cell movement produces the structure. Condensate-entering cells were observed to do so individually, with no sign of collective migration (S1 Video). We found that the dermal cells which ultimately make up the condensate are those present in its immediate vicinity, with a simple relationship between initial cell location and probability of incorporation into the condensate (Fig 2B).
We performed an analysis of the direction of movement of individually tracked cells by dividing each cell track into 6-h windows and determining the Euclidean angle of movement with respect to the future condensate centre for each window, as well as the Euclidean and accumulated distance travelled over this period. To determine whether individual cell movement was directed, we compared the distribution of Euclidean angles and distances for 2 classes of cells: (i) those cells initially outside the condensate area which subsequently entered it (‘condensate’) and (ii) those cells which remained outside the condensate area throughout (‘intercondensate’) (Fig 2C). Intercondensate Euclidean angles (n = 893) in these windows did not deviate from a uniform distribution (Kolmogorov–Smirnov test D = 0.05, p > 0.05) while condensate-bound track angles (n = 45) were consistently lower (Kolmogorov–Smirnov test D = 0.44, p < 0.001). The median Euclidean distance travelled in these 6-h windows was also greater for condensate-bound cells (Mann–Whitney U test p < 0.0001) than intercondensate cells (Fig 2C). Condensate-entering cells therefore exhibit directed movement towards sites of follicle initiation.
To investigate the timing of the distinct behaviours of cells entering condensates, we quantified cell behaviour by comparing the three 6-h windows prior to follicle entry (or the end of the track for intercondensate cells). We compared the Euclidean angle and level of persistence (calculated as the Euclidean distance/accumulated distance) of movement by the 2 cell classes (intercondensate and condensate) at 0–6, 6–12, and 12–18 h prior to follicle entry (Fig 2D). The median Euclidean angle relative to the nearest prospective condensate centre differed significantly between intercondensate and condensate-bound cells in the 0–6 h window (Kruskal–Wallis test p < 0.0001, Mann–Whitney U test with Bonferroni’s correction p < 0.01). Consistent with this behaviour, the median persistence for condensate-bound cells also differed from intercondensate cells in the 0–6 h window (Kruskal–Wallis test p < 0.01, Mann–Whitney U test with Bonferroni’s correction p < 0.01). These results show that the directed cell movement in the condensate-entering cell population occurs locally and only in the hours immediately prior to condensate appearance. Prior to this time, the behaviour of the entire mesenchymal cell population is the same, indicating that condensate formation involves selection of cells from an equivalent population based simply on their location.
Using the TCF/Lef::H2B-GFP line, we set out to determine the relative order of placode specification and cell condensation. We fixed cultured dorsal skin cultures at intermediate stages of development and compared the timing of appearance of Dkk4 expression with that of cell condensates in individual samples of skin. We found that patterning is first detectable in the epidermis as spatially organised focal Dkk4 expression with a lack of corresponding dermal cell clustering (Fig 2E). As patterning continues, dermal condensates become apparent, and the foci of Dkk4 expression resolve. However, at this stage, not all epidermal placodes have corresponding dermal condensates, and there are regions of skin where the epidermal pattern is present in the absence of dermal organisation. As the process completes, each Dkk4-positive placode is underlain by a dermal condensate (Fig 2E). These results show that an epidermal gene expression prepattern precedes the formation of dermal condensates.
Based on these findings, we hypothesised that local signalling from the epithelial placode might coordinate the condensation of the underlying dermal cells. We considered FGF a good candidate for the local attractant signal as this induces dermal condensations during feather development [51, 52] and because knockout mice lacking Fgf20, a gene selectively expressed in epidermal placodes, do not form dermal condensates [27]. To investigate whether a local FGF source could attract dermal cells, we cultured E12.75 TCF/Lef::H2B-GFP dorsal skin explants with beads soaked in recombinant FGF9 or control protein bovine serum albumin (BSA) beads. FGF9 beads induce cell accumulation in their vicinity, unlike BSA beads (Fig 2F and S3 Video). The area surrounding the FGF9 bead-induced condensate contains fewer GFP+ve nuclei, indicative of cell depletion from this area, and exhibits lack of hair follicle primordia (Fig 2F). Together, these observations show that local sources of FGF stimulate dermal condensate formation, demonstrating a positive role for FGF in hair follicle formation in the mesenchyme, in contrast to its effect on the epidermis.
Having established that FGF attracts mesenchymal cells to form a condensate and knowing that reception of WNT signals by mesenchymal cells is also required for this process [30, 53], we set out to define the effects of BMP signalling on dermal cell organisation. TCF/Lef::H2B-GFP skin explants treated with the BMP inhibitor LDN193189 have enlarged dermal condensates and modestly increased placode size (Fig 3A and 3C). We found that inhibition of BMP signalling significantly reduced the density of nuclei in the area between condensates (Fig 3A and 3D), indicating that increased dermal condensate size results from greater recruitment of dermal cells to produce enlarged follicle primordia. Thus, active BMP signalling normally limits cell recruitment to incipient condensates, and when BMP is impaired condensates continue to expand through cell recruitment, depleting cells from the intercondensate region.
To further define the influence of BMP on dermal cell arrangement, we stimulated this pathway in skin cultures containing existing dermal condensates. BMP treatment led to erosion of condensates, with a profound reduction in their size (Fig 3E and 3F). This demonstrates that BMP stimulation destabilises and depletes cells from mesenchymal aggregates when present in excess. Although dermal condensates are sites of high BMP4 production, BMP activity within the hair follicle rudiments is normally restrained through expression of the inhibitors NOGGIN and CTGF in the condensate and placode, respectively [22, 28, 29]. This, together with selective Fgf20 expression in the placode [27], produces a niche microenvironment of low BMP and high FGF activity at sites of follicle formation.
These results indicate roles for FGF and BMP signalling in influencing mesenchymal cell aggregation. We tested whether these influences could be modulated to trigger mesenchymal condensation by mimicking the hair follicle primordium microenvironment—that is, high-FGF and low-BMP signalling (hereafter, FGFHiBMPLo)—across the entire skin. Strikingly, imposing these conditions using FGF9 together with LDN193189 caused dermal condensates to arise in a periodically spaced manner without corresponding expression of the epidermal placode marker Dkk4 (Fig 4A). Histological sections from these samples showed the presence of large dermal condensates at the dermal–epidermal junction but an absence of epidermal placodes (Fig 4A). Neural cell adhesion molecule (NCAM) staining reveals the continued dermal identity of cells in the condensates thus induced, together with the absence of the distinct placodal NCAM expression that indicates an epidermal contribution to patterning (Fig 4B). As observed for Dkk4, other markers of epidermal placodes (Shh, Bmp2 and Edar) [18, 54] are not expressed in FGF9- and LDN193189-treated skins, while dermal condensate markers Bmp4 and Sox2 [54, 55] do display patterned expression, demonstrating the dermal condensate identity of the mesenchymal aggregates (Fig 4C). In FGFHiBMPLo conditions, epidermal TCF/Lef::H2B-GFP reporter signal is not detectable, demonstrating that FGF suppression of epidermal β-catenin signalling is not alleviated by simultaneous inhibition of BMP signalling. The dermal condensates that form under FGFHiBMPLo conditions have increased overall condensate size and markedly low cell density between the broad condensates (Fig 4D). The ability to pattern mesenchymal condensates in an FGFHiBMPLo environment is not restricted to a specific developmental stage (S5 Fig), and these structures, once induced to form, are autonomously stable upon restoration of normal FGF and BMP function (S6 Fig).
To determine the relative rates of dermal condensate patterning, we cultured skin explants under normal or FGFHiBMPLo conditions and imaged the samples at different time points. Pattern formation under normal conditions is more rapid than in FGFHiBMPLo conditions, with the control pattern appearing and stabilising quickly, while the FGFHiBMPLo pattern is slower to appear (S7 Fig).
Time-lapse imaging of the formation of mesenchymal condensates under FGFHiBMPLo conditions reveals the gradual emergence of these focal aggregates across the entire skin (Fig 5A, S4 and S5 Videos). To investigate the cell behaviours underlying mesenchymal self-organisation, we performed live cell imaging of TCF/Lef::H2B-GFP E13.5 dorsal skin explants using confocal microscopy (Fig 5B–5D). We used the 6-h window analysis as before to determine the Euclidean angle of movement with respect to the future condensate centre for each window, as well as the Euclidean and accumulated distance travelled over this period. In FGFHiBMPLo conditions, the intercondensate Euclidean angles (n = 612) deviate from a uniform distribution (Kolmogorov–Smirnov test D = 0.09, p < 0.01), while condensate-bound track angles (n = 91) were consistently altered from both a uniform distribution (Kolmogorov–Smirnov test D = 0.32, p < 0.001) and the distribution of intercondensate angles (Kolmogorov–Smirnov test D = 0.25, p < 0.001). As observed under normal (control) conditions (Fig 2C), the median Euclidean distance travelled in these 6-h windows was greater for condensate-bound cells (Mann–Whitney U test p < 0.001) than intercondensate cells (Fig 5B).
The behaviour of these cells entering condensates is remarkably similar to that observed in conditions of normal patterning (Fig 2C and 2D), with cells showing directed movement to condensates (Fig 5B–5D). The median Euclidean angle of movement differed significantly between intercondensate and condensate-bound cells in the 0–6-h window (Kruskal–Wallis test p < 0.001, Mann–Whitney U test with Bonferroni’s correction test p < 0.001) (Fig 5C). Consistent with this behaviour, the median persistence for condensate-bound cells also differed from intercondensate cells in the 0–6-h window (Kruskal–Wallis test p < 0.001, Mann–Whitney U test with Bonferroni’s correction p < 0.001) (Fig 5D), further demonstrating directed movement of condensate-entering cells.
To further investigate cell behaviour between normal (control) and FGFHiBMPLo conditions, we compared summary statistics for individual cell tracks (Fig 5E). As expected, the median accumulated velocity of condensate-entering cells in normal and FGFHiBMPLo conditions (Fig 5E) was increased when compared to intercondensate cells under control conditions (Kruskal–Wallis test p < 0.0001, Mann–Whitney U test with Bonferroni’s correction p < 0.001 and p < 0.0001, respectively). However, surprisingly, the median accumulated velocity of intercondensate cells in the FGFHiBMPLo conditions was also significantly higher than intercondensate cells in control conditions (p < 0.0001) (Fig 5E). This was not reflected by a change in the median Euclidean velocity nor the persistence of intercondensate cells under FGFHiBMPLo conditions (Kruskal–Wallis test p < 0.0001, Mann–Whitney U test with Bonferroni’s correction p > 0.05 for both cases), suggesting that the increase in cell movement was due to a chemokinetic effect of FGFHiBMPLo conditions (Fig 5E). Taken together, these results show that slower pattern emergence under FGFHiBMPLo conditions is not a result of sluggish cell movement but a reflection of the dynamics of the pattern-forming process itself. To gain an overall view of cell displacement across the field during the course of cellular condensation, we used particle image velocimetry (see S1D Appendix) to analyse the time-lapse videos (S1 and S5 Videos). This approach delineates the average paths of cells (Fig 5F, see S1D Appendix) revealing that a major distinction between unperturbed and mesenchyme-only patterning is the large zone of attraction which extends to collect cells in the latter condition.
As the underlying individual cell behaviour driving condensate formation is the same in control and FGFHiBMPLo conditions, we asked whether these are fundamentally distinct patterning mechanisms or whether they might be differently regulated outputs of a single underlying mechanism. Reaction–diffusion- and cell aggregation-based patterning mechanisms behave differently at tissue boundaries [4, 22, 56, 57]. Simulations of reaction–diffusion systems display an edge affinity in the arrangement of their cell clusters [4]. In contrast, simulations of cell aggregation-based systems display cell clusters that form at a distance from the cut edge (Fig 5G, see S1E Appendix). We manipulated the tissue boundaries under both control and FGFHiBMPLo conditions by introducing a cut edge into the skin explants and found, consistent with the mesenchyme-only patterning arising from a different mechanism, edge effects to be different for normal versus mesenchyme-only patterning. The normal pattern aligns close to the cut edge of the tissue, as previously reported [4, 22], while the latter respects the shape of the boundary but maintains a large distance from the edge to the nearest pattern foci (Fig 5H). Intuitively, these behaviours can be thought of as recognising an advantage for cells adjacent to an edge in a reaction–diffusion system, as they lack competition on 1 side and are able to dilute the inhibitor that they produce into the culture medium. Conversely, in a patterning system based on cell aggregation, the cells close to the edge are disadvantaged, as they have reduced numbers of neighbours with which to nucleate clustering.
As the mesenchymal patterning mechanism is fundamentally distinct from the normal condition, we sought a mechanistic basis for this process. We treated skin cultures with modulators of pathways previously implicated in dermal cell condensation to identify signalling pathways with selective effects on mesenchyme-only patterning. We found that mesenchyme, in both control and FGFHiBMPLo conditions, patterned robustly, despite the presence of inhibitors or activators of the CXCL/CXCR, Notch, or platelet-derived growth factor (PDGF) pathways (S8 Fig).
Alteration of TGFβ signalling, however, profoundly suppressed mesenchyme-only patterning while having relatively modest effects on normal patterning. The TGFβ receptor type I and II inhibitor LY2109761 (Fig 6A) slowed the assembly of normal dermal condensates, while placode patterns became expanded and threaded through the epidermis (Fig 6B and S9 Fig), matching the delayed hair follicle phenotype reported for the Tgfb2 mutant mouse [48]. Augmenting signalling by administration of recombinant TGFβ2 permitted a normal array of placodes and condensates to arise, with condensates appearing more prominent than in control conditions. However, either suppression or widespread stimulation of TGFβ2 signalling abolished the ability of mesenchyme to pattern autonomously in FGFHiBMPLo conditions (Fig 6B). Thus, active TGFβ signalling is required for mesenchyme-only patterning but must be restricted for this process to yield a spatially organised array of condensates.
Tgfb2 is broadly expressed by mesenchymal cells [54, 58] during primary hair follicle formation and in cell clusters at sites of dermal condensation (Fig 6C). This expression is consistent with widespread SMAD2 phosphorylation in early dermal fibroblasts, becoming focussed in incipient dermal condensates (Fig 6D) [59]. To test whether TGFβ2 can attract mouse dermal mesenchymal cells, as previously reported for chicken mesenchyme [60], we applied TGFβ2-coated beads to skin and detected a strong accumulation of cells around the bead (Fig 6E). Thus, TGFβ2 represents a widely expressed attractant serving to draw mesenchymal cells together.
Having defined restricted TGFβ as a required self-attractive factor in mesenchyme-only patterning, we sought its role in an interplay between normal patterning and mesenchymal self-organisation. In addition to its function as a direct mesenchymal chemoattractant (Fig 6E) [61], TGFβ signalling also influences cell migration and cell–matrix interactions through modulation of gene expression to create an environment conducive to cell migration [62, 63]. We set out to identify in developing skin whether TGFβ regulates the expression of genes known to be associated with cell migration, adhesion, and matrix composition in other tissues [64–68]. We compared responses to TGFβ in both E13.5 and E13.75 skin with those elicited by similar FGF and BMP treatments, 2 signalling pathways we previously found to promote or antagonise dermal cell aggregation, respectively (Figs 2F and 3). We identified TGFβ regulation of genes encoding the cell migration modulators transforming growth factor beta induced (Tgfbi) and thrombospondin family members (Thbs2 and Thbs4), as well as the matrix components Fibronectin (Fn1), Syndecan-1 (Sdc1), and Tenascin-C (Tnc) (Fig 7A and 7B), suggesting that TGFβ alteration of cell–matrix interaction may contribute to its aggregative effect. The TGFβ gene regulatory effects are broadly distinct from FGF-elicited responses, despite both signals sufficing to stimulate dermal cell aggregation at their local sources, while BMP suppresses Fn1 and Sdc1 expression and potentially downmodulates TGFβ signalling through induction of inhibitory Smad7 and suppression of the TGFβ type III receptor (Tgfbr3), a known enhancer of TGFβ2 signalling [69].
These gene expression changes related to cell–matrix interactions and movement suggested that TGFβ may play a general role in promoting cell migration. To assess whether TGFβ could promote cell attraction to placodes, as indicated by the prominent condensates it induces during patterning (Fig 6B), we tested the effect of generalised TGFβ availability on directed movement to FGF sources, where loaded beads replicate the local source of FGF produced in a placode. We found that the extent of mesenchymal cell attraction to FGF-coated beads was strongly increased by the presence of TGFβ2 in the culture medium. Conversely, cell attraction to FGF beads was greatly diminished by TGFβ signal suppression (Fig 7C and 7E). We detected no reciprocal effect of ubiquitously available FGF on cell recruitment to local TGFβ2 (Fig 7D and 7F). Thus, TGFβ generates an environment conducive to efficient cell recruitment at FGF sources, explaining the phenomenon of delayed dermal condensate formation when this signal is suppressed.
Embryonic pattern formation proceeds rapidly; periodic arrangements of limb digits arise within approximately 16 h [70] and hair follicles in approximately 10 h [22]. By focussing on short-lived mRNAs, we have identified a network of interactions between the BMP, FGF, and WNT pathways capable of breaking symmetry to produce a periodic pattern. In this system, WNT signalling acts to stimulate expression of genes denoting placode fate (Dkk4, Bmp2, Fgf20), while BMP and FGF signalling inhibit their expression. We note that the half-lives of the proteins encoded by components of the network would also need to be relatively short to achieve rapid pattern formation. The many mRNAs with long half-lives that become preferentially expressed in the placodes, including Ctnnb1, Wnt10a, Wnt10b, and Edar, may act to stabilise the rapidly emerging pattern and promote hair follicle growth, as previously shown for EDAR function [22, 25, 40].
Following definition of the epidermal placode arrangement, dermal cells are locally recruited to assemble a condensate. Previously, a mesenchymal cell-sorting mechanism for hair follicle patterning was proposed based on genetic correlations between hair follicle number and size [71], conceptually similar to the well-studied and diverse colour patterns on fish skin arising from sorting of distinct mesenchymal lineages [15]. Though there is abundant heterogeneity of dermal mesenchymal cells that could be exploited to achieve such a mechanism [72], our quantitative analyses do not support the operation of such a system and instead show that the cells ultimately forming the condensate are those located in its immediate proximity (Fig 2). The local movement of the dermal mesenchymal cells is thus similar to the short-range movements that act to construct the epithelial placode [20].
In addition to their contribution to pattern formation through rapid gene regulatory interactions, the BMP and FGF pathways also influence mesenchymal cell behaviour. FGF, normally produced locally at the placode, directs mesenchymal clustering, while BMP suppresses aggregation. By mimicking the hair follicle microenvironment of high-FGF and low-BMP signalling, we identify a latent potential of dermal mesenchyme to self-organise, yielding a stable periodic pattern of cell aggregates in the absence of signalling directives from the epidermis. This demonstrates that dermal condensates do not require spatially restricted signals to form. Rather, the widespread BMP activity in the skin prior to hair follicle patterning, together with the slow rate of dermal self-organisation, subordinates the mesenchyme’s patterning potential to reaction–diffusion patterning in the epidermis. Our identification of a mesenchyme-only patterning condition is, in structural terms, reciprocal to that arising from Fgf20 deficiency, which results in epidermal placode formation without accompanying mesenchymal condensates [27]. Thus, epidermal and dermal patterning are separable self-organising systems, coupled through epidermal FGF-guided mesenchymal cell attraction with TGFβ potentiation of this accumulation.
Mechanistically, however, these epidermis-only and mesenchyme-only patterns arise through very different mechanisms. Epidermis-only patterning in the Fgf20 mutant is achieved through focally restricted signals, agreeing with the majority of patterning signals being produced in this tissue layer (see Fig 1D and S1 Table), while it is motile cells that determine the dermis-only pattern. The expansion of placode identity upon suppression of TGFβ signalling (Fig 6B), similar to that of the unresolved prepattern (Fig 2E), may be a secondary result of delayed condensate formation together with its accompanying inhibitors (BMP4 and DKK1) failing to define the placode edges.
Our findings of dermal condensate formation in the absence of an epidermal prepattern are consistent with modes of tissue patterning in which moving mesenchymal cells themselves are agents of pattern formation. These models do not require interacting diffusible signals and such mechanisms integrate tissue morphogenesis and patterning as part of a single process, rather than occurring in sequence, as in the Turing system [4, 8, 9, 73]. An aggregative potential of dermal papilla cells, the product of the embryonic dermal condensate, has been noted [74] and may not be unique to the skin, as spontaneous condensation of mesenchymal stem cells from a range of organs has been identified, given suitable culture substrates in vitro [75]. Beyond the formation of simple condensates, spatially organised aggregates of embryonic limb bud mesenchymal cells arise spontaneously in culture [76], with a suggested role for TGFβ2 driving this process for mesenchymal cells cultured in isolation from epithelium [77]. These observations may suggest a broad role for TGFβ in driving mesenchymal morphogenesis, whether spatial organisation is guided by a prepattern or occurs de novo through cell movement.
Together, these results support a model for hair follicle positioning and construction in which a reaction–diffusion system, operating largely in the epidermis [25, 27, 39], limits the locations at which mesenchymal organisation can occur by providing both local FGF direction and relief from BMP activity (Fig 8). Dermal cells respond to these sources of FGF, with widespread TGFβ2 generating an environment promoting entry into condensates according to the FGF gradients. As it forms, the condensate produces BMP4, thereby restricting further condensate and placode expansion. Ultimately, the signals operating in reaction-diffusion patterning also serve to modulate mesenchymal behaviour to direct, inhibit, or stimulate mesenchymal condensation, such that, under FGFHiBMPLo conditions, the components of the normal patterning network are recast to achieve mesenchyme-only patterning. Thus, we conclude that the dermal mesenchyme does possess pattern-generating ability but that this is preceded by a rapid, primarily epidermal, pre-patterning system that acts to specify the restricted locations at which mesenchymal organisation is permitted.
The rationale for the existence of 2 distinct routes to achieve periodic patterning during hair follicle formation may lie in a shifting balance between these 2 potentials in the waves of hair follicles that form at different developmental stages. Thus, it may be that the pattern of secondary or tertiary hair follicles is defined by mesenchymal cell behaviour to a greater extent than the primary follicles focussed upon here. Alternatively, the existence of 2 periodicity generators may be a remnant of the evolutionary history of skin development, as it has been suggested that turnover of patterning mechanisms can occur during the course of evolution [78, 79] and that direct cell-driven patterning systems tend to become captured by signal-based pre-patterning systems [80]. Our work demonstrates experimentally such a superimposition of different patterning mechanisms, with 1 potential for self-organisation being suppressed and subjected to instruction by another.
All animal work was conducted under approval of the Animal Welfare and Ethical Review Body (AWERB) at The Roslin Institute, University of Edinburgh, and by the United Kingdom Home Office in accordance with the Animals (Scientific Procedures) Act 1986. Euthanasia was carried out according to Schedule 1 of the Animals (Scientific Procedures) Act 1986.
Mice were on the FVB/N background, apart from those employed in time-lapse and static imaging of cell rearrangement, which were offspring of a cross between male hemizygous TCF/Lef::H2B-GFP mice [43] on a mixed C3H/C57 genetic background and female FVB/N mice. Noon on the day of discovery of the vaginal plug was designated day 0.5 of development. Skins were examined for pre-existing signs of hair morphogenesis and not used if these were detected for experiments designated to start prior to follicle morphogenesis. Embryos were harvested in high-glucose Dulbecco’s modified Eagle’s medium (DMEM) (Sigma) supplemented with 1% penicillin/streptomycin (Gibco, Life Technologies) and kept on ice for short periods.
For the qRT-PCR experiments used in the network derivation, dorsal skin explants were halved along the midline to generate treated and control halves from each embryo. Skin halves were placed onto a cellulose filter (Millipore, pore size 0.45 μm). The tissue was then submerged in DMEM supplemented with 2% FBS (complete DMEM) in a centre-well dish (Falcon) and incubated at 37°C and 5% CO2 for 6 h. For the TGFβ2, FGF9, and BMP4 experiments, whole dissected skins were cultured in complete DMEM containing the required recombinant protein for either 8 h or 24 h.
For longer-term culture, whole skin explants were dissected and mounted on filters as described above, submerged in complete DMEM, and supported by a metal grid in a centre-well dish. E13.5 explants were cultured for 27 h unless stated otherwise. For experiments extending beyond 27 h, medium was replaced every 24 h.
Epidermal–dermal separations for RNA-seq and qRT-PCR experiments were performed by incubating skin samples with 2 mg/mL Dispase II (Gibco) at 37°C for 10 minutes. For imaging, epidermal–dermal separations were performed by incubating skin in 10 mM EDTA/PBS for 25 minutes at 37°C followed by 20 minutes in PBS at RT prior to separation with fine forceps. For counterstaining, samples were fixed in 4% PFA, washed several times in PBS/0.1% Tween 20 (PBST), treated with 20 μg/ml proteinase K for 3 minutes, washed in PBT, incubated with 100 μg/ml RNase for 20 minutes, washed again with PBT and then stained with 1/2000 diluted propidium iodide (PI) solution (Life Technologies) for 5 minutes. Following staining, samples were washed in PBT and mounted in Prolong Gold (Life Technologies).
Recombinant FGF7 (mouse), FGF9 (mouse and human), BMP4 (mouse), and TGFβ2 (human) were from R&D Systems. LDN193189 (Stemgent), CHIR99021 (Axon biochem), IWR-1 (Tocris), DAPT (Bio-Techne), LY2109761 (Cambridge Biosciences), Pertussis toxin (Tocris), Imatinib Mesylate (Sigma Aldrich), and SU5402 (Sigma Aldrich) were reconstituted according to the manufacturer’s recommendations. We did not detect any significant effect from treatment with recombinant FGF20 protein obtained from 2 commercial suppliers on either placode pattern formation, skin development, or on expression of the direct FGF pathway target gene Etv5.
Total RNA was isolated from skin explants using Tri Reagent (Sigma) and treated with RQ1 DNAse (Promega) to remove contaminating genomic DNA. cDNA was synthesised from total RNA using random primers and Superscript III reverse transcriptase (Roche) in a 20-μl reaction. Reactions were diluted 20-fold and had 3 μl used as a template for each qRT-PCR. Each reaction was performed in a 20-μl volume using Universal SYBR Green Master Mix (Roche) containing Rox reference dye. Reactions were performed in triplicate, with at least 3 biological replicates used to determine each data point. Relative expression levels were determined from cDNA dilution standard curves and normalised to Tbp (or Capzb for half-life determination). Oligonucleotide sequences used are given in S1 Supporting Methods.
See S1 Supporting Methods for Actinomycin D dose determination and treatment, sample quality control and processing, RNA-sequencing, analysis, and qRT-PCR validation.
Skin explants were fixed overnight in 4% PFA at 4°C. Samples were dehydrated into 100% methanol, bleached in 6% H2O2, then rehydrated and treated with 20 μg/mL proteinase K. After postfixing in 4% PFA containing 0.2% glutaraldehyde for 20 minutes, skin explants were hybridised with probe at 60°C overnight in 50% formamide, 5 X saline sodium citrate (SSC), 1% SDS, 50 μg/mL heparin, and 50 μg/mL yeast RNA in diethyl pyrocarbonate (DEPC)-treated H2O. Samples were washed to remove unbound probe and signal detected using an alkaline phosphatase sheep antidigoxigenin antibody (Roche, 1:1,000 dilution) and 5-bromo-4-chloro-3'-indolylphosphate/nitro-blue-tetrazolium (BCIP/NBT) colour reaction (Sigma).
For immunohistochemistry, samples were fixed in 4% PFA, embedded in 0.12 M sodium phosphate buffer/7.5% gelatin/15% sucrose, and cryosectioned. Sections were rehydrated in PBS at 37°C for 30 minutes, briefly washed with tris-buffered saline containing 0.01% Tween 20 (TBST), then incubated in blocking buffer (TBST containing 5% heat-treated sheep serum and 1% BSA) for 1 h at RT before overnight incubation at 4°C with primary rabbit antibodies (1:200 mouse antiactive beta catenin (Millipore #05–665) or rabbit antiphosphorylated SMAD2 (pSMAD2) (Cell Signalling Technologies #3108) diluted in blocking buffer. Samples were washed in TBST, then incubated with fluorescent secondary antibodies (1:500 Life Technologies) in blocking buffer for 1 h at RT. Samples were washed with TBST, counterstained with DAPI (Sigma) and mounted in Prolong Gold (Life Technologies). Sections were imaged using a Zeiss LSM710 confocal microscope (Carl Zeiss).
For immunoblotting, protein was extracted from skin samples using radioimmunoprecipitation assay (RIPA) lysis buffer (Santa Cruz Technologies) and a handheld homogeniser. Protein samples were diluted in NuPage LDS sample buffer (Life Technologies), separated by gel electrophoresis on 4%–12% Bis-Tris NuPage precast gels under denaturing conditions and transferred to nitrocellulose membrane (Amersham). Membranes were blocked in 5% milk/TBST for 1 h at RT, followed by overnight incubation at 4°C in 5% BSA/TBST containing primary antibody (1:1,000 mouse anti-active β-catenin [Millipore #05–665], 1:3000 rabbit anti-β-catenin [BD biosciences #610153], 1:1,000 rabbit anti-phospho SMAD 2/3 [Cell Signalling Technologies #8828], 1:1,000 rabbit anti-SMAD2 [Cell Signalling Technologies #5339], or 1:3,300 mouse anti-γ tubulin [Sigma Aldrich #T6557]). Following primary antibody incubation, membranes were washed in TBST and incubated for 1 h at RT in 5% milk/TBST containing horseradish peroxidase (HRP)-conjugated species-specific secondary antibodies (Dako). Membranes were washed several times with TBST before detection with the Novex ECL chemiluminescent substrate reagent kit (Life Technologies) and developed on ECL Film (Amersham).
Affi-Gel Blue Gel beads (Bio-Rad) were washed twice in PBS and incubated in either 100 μg/ml recombinant human FGF9, 100 μg/ml recombinant human TGFβ2, or 100 μg/ml BSA diluted in PBS for at least 2 h at RT or overnight at 4°C. Beads were placed onto a nitrocellulose Millipore filter (pore size 0.45 μm), and dissected E12.5 TCF/Lef::H2B-GFP dorsal skin explants were manoeuvred and placed on top of the beads, dermis side down. Skins were then imaged, as described in S1 Supporting Methods, over a period of 48–72 h.
E13.5 TCF/Lef::H2B-GFP dorsal skin explants were dissected and imaged as described previously using a custom imaging chamber [81] with the following modifications. Instead of using a central imaging clip, the entire base of the chamber was filled with 1% (weight/volume) agarose in PBS. Dissected embryonic skin was mounted dermis-side down onto a black nitrocellulose filter membrane (Millipore) with a 45 μm pore size. The membrane was subsequently placed onto the agarose block, and a lummox membrane (Greiner) was clamped across it with an o-ring such that the skin was sandwiched between the 2 membranes. The imaging chamber was filled with DMEM without phenol red containing 4,500 mg/L glucose, 2% FBS, 1% penicillin/streptomycin, and 0.584 g/l L-glutamine. Images were captured with a 20X objective using a Nikon A1R inverted confocal microscope in a heated chamber supplied with 5% CO2 in air. Bead and FGF9/LDN193189 combination experiments were performed using a Zeiss Live Cell Observer and are described in S1 Supporting Methods.
All image analysis tasks were performed using custom written macros for the Fiji [82] distribution of ImageJ, an open-source image analysis package based on NIH Image [83]. The source code is available through GitHub (https://github.com/richiemort79/cell_patterning). In order to track cell behaviour in an unbiased manner, maximum intensity Z-projections of time-lapse sequences were drift-corrected and cropped to include the area of the final condensate (of varying size) with approximately 100 μm of space surrounding each. A window of at least 1,500 minutes that incorporated condensate formation was considered. At least 50 cells per condensate from at least 4 independent skins were then selected at random prior to condensate formation and tracked manually until they either entered a condensate or the video ended.
A cell was deemed to acquire a ‘condensate identity’ if, at the end of the tracking period, it was incorporated within this structure; for analysis purposes, tracking was halted on follicle entry. Cells that did not enter the condensate were termed ‘intercondensate.’ Visual analysis of these cells up to a later time point was also performed to ensure that these cells did not become part of the condensate at a later stage.
|
10.1371/journal.pbio.1000408 | A Genome-Scale DNA Repair RNAi Screen Identifies SPG48 as a Novel Gene Associated with Hereditary Spastic Paraplegia | DNA repair is essential to maintain genome integrity, and genes with roles in DNA repair are frequently mutated in a variety of human diseases. Repair via homologous recombination typically restores the original DNA sequence without introducing mutations, and a number of genes that are required for homologous recombination DNA double-strand break repair (HR-DSBR) have been identified. However, a systematic analysis of this important DNA repair pathway in mammalian cells has not been reported. Here, we describe a genome-scale endoribonuclease-prepared short interfering RNA (esiRNA) screen for genes involved in DNA double strand break repair. We report 61 genes that influenced the frequency of HR-DSBR and characterize in detail one of the genes that decreased the frequency of HR-DSBR. We show that the gene KIAA0415 encodes a putative helicase that interacts with SPG11 and SPG15, two proteins mutated in hereditary spastic paraplegia (HSP). We identify mutations in HSP patients, discovering KIAA0415/SPG48 as a novel HSP-associated gene, and show that a KIAA0415/SPG48 mutant cell line is more sensitive to DNA damaging drugs. We present the first genome-scale survey of HR-DSBR in mammalian cells providing a dataset that should accelerate the discovery of novel genes with roles in DNA repair and associated medical conditions. The discovery that proteins forming a novel protein complex are required for efficient HR-DSBR and are mutated in patients suffering from HSP suggests a link between HSP and DNA repair.
| All cells in our bodies have to cope with numerous lesions to their DNA. Cells use a battery of genes to repair DNA and maintain genome integrity. Given the importance of an intact genome, it is not surprising that genes with roles in DNA repair are mutated in many human diseases. Here, we present the results of a genome-scale DNA repair screen in human cells and discover 61 genes that have a potential role in this process. We studied in detail a previously uncharacterized gene (KIAA0415/SPG48) and demonstrated its importance for efficient DNA double strand break repair. Further analyses revealed mutations in the SPG48 gene in some patients with hereditary spastic paraplegia (HSP). We showed that SPG48 physically interacts with other HSP proteins and that patient cells are sensitive to DNA damaging drugs. Our data suggest a link between HSP and DNA repair and we propose that HSP patients should be screened for KIAA0415/SPG48 mutations in the future.
| Mutations in DNA repair genes are associated with different diseases and disorders including cancer [1], accelerated aging [2], and neuronal degeneration [3]. Neurons appear to be particularly vulnerable to mutations in DNA repair genes, possibly due to the lack of proliferation and high oxidative stress within these cells. As a consequence, several neurological diseases have been linked to defects in DNA repair such as Ataxia-telangiectasia [4], Ataxia-telangiectasia-like disorder [5], Seckel syndrome [6], Nijmegen breakage syndrome [7], and Charcot-Marie-Tooth syndrome [8].
A particularly dangerous DNA lesion for a cell is a double strand break (DSB), in which two strands of the DNA are broken in close proximity to one another [9],[10]. DSBs are repaired mainly via two parallel pathways: homologous recombination and nonhomologous end joining (NHEJ). Repair via homologous recombination typically restores the genetic information, whereas repair via NHEJ often leads to mutations [10],[11].
Recently, several RNAi screens have addressed different aspects of mammalian DNA repair, such as increased sensitivity towards PARP inhibition [12], increased sensitivity towards cisplatin [13], accumulation of 53BP1 foci [14],[15], or altered phosphorylation of the histone variant H2AX [8]. These screens have greatly enhanced our understanding of human DNA repair processes and delivered a number of novel genes implicated in various aspects of DNA repair. Here, we report a genome-scale RNAi screen for genes implicated in homologous recombination-mediated DSB repair, uncovering a variety of known and so far uncharacterized genes implicated in this process. In this work, we mine this screen employing a structural bioinformatics approach and identify KIAA0415/SPG48 as a putative helicase that is associated with hereditary spastic paraplegia (HSP).
For a comprehensive search of genes associated with DNA DSB repair, we performed a genome-scale RNAi screen, utilizing an endoribonuclease-prepared short interfering RNA (esiRNA) library [16] and employing the well-established DR-GFP assay [17]. First, a stable HeLa cell line with two non-functional GFP alleles was generated, in which GFP expression is efficiently activated only after HR-DSBR (Figure 1A). We then tested the robustness of the assay by co-transfection of these cells with the I-SceI expression plasmid and an esiRNA targeting Rad51, which is an essential factor for the early stages of homologous pairing and strand exchange [18]. Depletion of Rad51 resulted in a marked reduction of GFP positive cells, and comparisons to negative control transfected cells suggested a high dynamic range for candidate factors influencing HR-DSBR (Figure 1B and histograms Figure 1C).
The RNAi screen was carried out in duplicate in 384-well plates by co-transfection of an I-SceI encoding plasmid with the individual esiRNAs targeting over 16,000 human genes [16]. The percentage of GFP positive cells was determined by high throughput FACS, providing a sensitive readout for esiRNAs influencing the frequency of HR-DSBR (Figure 1C). Knockdown of 228 and 141 transcripts significantly decreased or increased the percentage of GFP positive cells, respectively (Figure 1D, Table S1). Among the strongest knockdowns affecting HR-DSBR were genes with well-characterized roles in DNA repair such as Rad51, BRCA1, and SHFM1. Gene ontology enrichment analysis of the candidates revealed a 5-fold enrichment for genes reported to be implicated in DNA repair (Figure 1E), confirming that the screen was efficient.
To validate the candidate hits we examined their expression in HeLa cells and resynthesized all esiRNAs for the genes that were expressed. We also generated a second, independent, and non-overlapping esiRNA for these genes and tested all esiRNAs again in the DR-GFP assay in multiple replicates. Using stringent selection criteria (see Online Methods), 45 genes decreased the frequency of homologous recombination, while 17 genes increased it with two independent silencing triggers (Table 1). To further narrow down the list of these 62 candidates, we tested the esiRNAs for their impact on intracellular GFP levels. EsiRNAs that influence GFP levels, for example by targeting a transcriptional activator for GFP expression, could score in the DR-GFP assay and contaminate the hit list. We therefore transfected the esiRNAs into GFP expressing HeLa cells and assayed GFP levels by FACS. EsiRNAs targeting MKNK2 reduced GFP levels in these cells. Therefore, this gene was excluded from further analysis, reducing the final hit list to 61 genes (Table 1). The effectiveness of this stringent validation was monitored again by gene ontology enrichment analysis, with an enrichment of now 20-fold for genes annotated in the category DNA repair (Figure 1E).
Silencing of 17 genes significantly increased the number of GFP positive cells in the DR-GFP assay. Hence, the knockdown of these genes promoted HR-DSBR, which might be of interest for several biological applications such as increasing the targeting efficiency of genes by homologous recombination [19]. Different reasons might account for the increased number of GFP positive cells observed. One possibility is that the knockdown led to an inhibition of the NHEJ pathway, thereby shifting the ratio of the two possible pathways toward repair via HR. Support for this reasoning comes from experiments in yeast and flies, where the knockout of DNA ligase IV, a gene that is required for NHEJ [20], significantly increased gene targeting by homologous recombination [21],[22]. Interestingly, the knockdown of human Lig4 resulted in a striking increase in GFP positive cells in the DR-GFP assay (Table 1), suggesting that inhibition of the NHEJ pathway can increase the frequency of HR-DSBR also in mammalian cells. This idea is further supported by inspection of other known NHEJ proteins, including XRCC4, XRCC5, XRCC6, PRKDC, and DCLRE1C [23],[24]. Knockdown of all of these proteins increased the frequency of homologous recombination in the DR-GFP assay (Table S1). Hence, we speculate that other genes that increased the number of GFP positive cells might be implicated in the NHEJ pathway and that knockdown of these genes could enhance gene targeting by homologous recombination in mammalian cells.
The list of genes that decreased the frequency of HR-DSBR was enriched for proteins with well-defined roles in HR-DSBR, such as Rad51 and BRCA1. In addition, genes, such as E2F1, that more indirectly influence HR-DSBR were also identified in the screen. E2F1 is involved in cell cycle and apoptosis regulation after DNA damage [25] and has recently been implicated in transcriptional regulation of Rad51 and BRCA1 [26], possibly explaining why the knockdown of E2F1 scored in our screen. Interestingly, the assay also uncovered a number of genes that have roles in DNA repair processes other than HR-DSBR, such as XPC, which has a role in nucleotide excision repair (NER) [27], and the base excision repair (BER) DNA helicase RECQL4 [28]. However, a polymorphism in the XPC gene has recently been shown to correlate with bleomycin-induced chromosomal aberrations [29], and RECQL4 has been reported to coincide with foci formed by Rad51 after induction of DSBs [30], suggesting possible links between the different DNA repair pathways. Finally, the gene list is enriched for proteasome subunits, including PSMD4, PSMD1, PSMD14, and SHFM1. Treatment with proteasome inhibitors has been shown to specifically suppress HR-DSBR possibly because of the lack of proteasome-mediated degradation of chromatin bound proteins blocking the access to the lesion [31],[32]. Moreover, SHFM1 has been shown to be required for Rad51 foci formation upon DNA damage [33], implicating a more direct role of this proteasome subunit in HR-DSBR and possibly providing an explanation why SHFM1 was one of the strongest hits in our screen. Based on these results we were encouraged to investigate further the knockdowns that decreased the number of GFP positive cells in the DR-GFP assay.
To characterize in detail the 44 knockdowns that decreased the frequency of HR-DSBR, we performed several additional assays. First, we tested the influence on cell viability of these esiRNAs in HeLa cells. Thirteen esiRNAs considerably decreased cell numbers and were excluded from further analyses (Table 1). Second, we performed mitomycin C (MMC), cisplatin, and ionizing radiation (IR) sensitivity assays. MMC predominantly causes interstrand cross-links, which result, among other things, in DSBs due to a block of replication forks [34]. Cisplatin damages DNA in a different way and generates predominantly intrastrand cross-links [35], whereas IR gives rise to a variety of DNA lesions [36]. Cells with impaired DNA repair pathways might be more sensitive to these treatments, which should manifest in reduced cell viability. Twenty-four hours post-transfection of the esiRNAs, the cells were treated for 1 h with MMC, cisplatin, or exposed to IR and cells were counted after an additional incubation for 48 h. A number of knockdowns increased the sensitivity towards one or more treatments, substantiating a role of these genes in DNA repair, with some of the knockdowns showing an effect for one, but not the other treatment (Figure 2A, Table 1). For instance, the knockdown of RBBP8 (also known as CtIP), which promotes DNA end resection [37], did not cause increased sensitivity towards cisplatin. However, substantially less cells were counted after MMC treatment, indicating that RBBP8 depletion primarily sensitized the cells against this drug. Third, we employed a gamma-H2AX removal assay. The histone H2AX is phosphorylated on serine 139 predominantly by ATM/ATR [38],[39] at sites of DSBs until the lesion is repaired. After successful DNA repair this phosphorylation is reverted by the phosphatase PP2A [40]. Several knockdowns resulted in extended time before gamma-H2AX was removed from irradiated cells (Figure 2B, Table 1), suggesting a delay in DSBR, and potentially explaining the observed reduction of GFP positive cells in the DR-GFP assay. Surprisingly, a few knockdowns showed overall reduced numbers of gamma-H2AX positive cells, or accelerated removal of gamma-H2AX after irradiation. For example, depletion of ARHGEF1 resulted in a reduced number of gamma-H2AX positive cells 1 h after irradiation (Figure 2B). Potentially, this Rho guanine nucleotide exchange factor [41] is required for efficient recruitment of H2AX phosphorylation factors, which ultimately translates into less efficient HR-DSBR. In contrast, the knockdown of FIZ1, a Flt3 interacting zinc finger protein [42], resulted in similar numbers of gamma-H2AX positive cells 1 h after irradiation in comparison to the control transfected cells. However, gamma-H2AX was more rapidly removed in these cells (Figure 2B), potentially compromising effective DSBR. Taken together, these results validate the effectiveness of our screen and serve as an initial classification of molecular pathways for a number of genes that can be explored in future studies.
For this work, we mined the screen by performing bioinformatics analyses on the uncharacterized sequences in an attempt to reveal possible molecular functions. KIAA0415 emerged as particularly notable. By applying threading techniques (see Online Methods for details), we identified potential structural homologies of KIAA0415 with proteins belonging to the fold family “P-loop containing nucleoside triphosphate hydrolases” (SCOP c.37; Table S2). This fold family contains the so-called “helicase C domain” (PF00271) formed by a tandem repeat of two RecA-like domains (Tandem AAA-ATPase superfamily). Top scoring sequence-to-structure alignments were obtained with the KIAA0415 sequence and the structure of the helicase C domains of SF2 helicases that are involved in DNA repair such as UvrB, Hel308, RecG, and TRCF (Figure 3). Visual inspection of the generated 3D model (see Online Methods) confirmed the existence of potential SF2 helicase motifs in KIAA0415 (Figure S1). Molecular dynamics simulations were used to refine the KIAA0415 model and corroborated its stability and its putative ADP and Mg2+ recognition (Video S1, Online Methods). These results further support the prediction of a helicase-like domain within KIAA0415 and substantiate the conservation in 3D of residues important for its function as a putative SF2 helicase.
Based on these results, we decided to further elucidate possible molecular functions of KIAA0415. We first tested the potency of the employed KIAA0415 esiRNAs in more detail. Both esiRNAs efficiently depleted KIAA0415 mRNA transcripts (Figure 4A) and protein (Figure 4B). We then repeated the DR-GFP assay in the HeLa reporter cell line and found 3.4 (esiRNA1) and 4.3 (esiRNA2) fold decrease in GFP positive cells in comparison to controls, suggesting reduced frequencies of homologous recombination (Figure 4C). We examined the expression levels of I-SceI after the knockdowns to rule out the possibility that I-SceI-generated DSBs are compromised (Figure S2). To exclude a possible cell-type specific effect, we also tested the knockdowns in a different cell line. U2OS cells carrying a single insertion site of the DR-GFP construct showed a similar reduction of GFP positive cells upon KIAA0415 knockdown (Figure 4D), indicating that this effect was not cell line specific. Finally, we excluded possible off-target effects by performing cross-species RNAi rescue experiments [43]. Stable expression of mouse KIAA0415 in the human DR-GFP cell line rendered this cell line resistant to the human esiRNAs, authenticating a role of KIAA0415 in HR-DSBR (Figure 4E). In summary, these results suggest that KIAA0415 is a novel putative SF2 helicase required for efficient HR-DSBR.
To further characterize KIAA0415, we tagged the gene on a bacterial artificial chromosome (BAC) applying the TransgeneOmics approach [44]. This method allows expression of tagged proteins from its native promoter in its genomic context, and therefore, the protein is expressed near physiological levels [44],[45]. C- and N-terminally tagged KIAA0415 was successfully cloned and expressed in HeLa cells. The fusion protein showed disperse, cytoplasmic, and nuclear localization, which did not change considerably upon induction of DNA damage (unpublished data). Immunoblotting of cell extracts revealed two major protein bands, possibly reflecting two KIAA0415 isoforms (Figure 4B). Cell fractionations showed that the shorter isoform was predominantly nuclear, whereas the longer form was found mostly in the cytoplasm (Figure 4F). Immunoprecipitation experiments followed by spectrometric identification of co-isolated proteins revealed interactions of KIAA0415-LAP with SPG11, SPG15, C20orf29, and DKFZp761E198 (Figure 5A,B and Table S3). In order to validate these interactions we generated cell lines expressing C-terminally tagged SPG11, SPG15, and DKFZp761E198 again using the TransgeneOmics approach. Reciprocal immunoprecipitation experiments followed by mass spectrometry analyses of in-gel and in-solution digests confirmed the existence of a protein complex, which consists of at least five core proteins: KIAA0415, SPG11, SPG15, C20orf29, and DKFZp761E198 (Figure 5B and Table S3). In order to test whether protein interaction partners of KIAA0415 would also affect HR-DSBR, we tested esiRNAs targeting these genes in the DR-GFP assay. Interestingly, significant reduction of GFP positive cells were observed upon silencing of C20orf29 and SPG15 with two independent esiRNAs (Figure 6), suggesting that these proteins are also required for efficient HR-DSBR. Knockdown of SPG11 and DKFZp761E198, however, did no have an effect on the percentage of GFP positive cells. Together, these experiments reveal a novel protein complex, which at least in part is required for efficient HR-DSBR.
The KIAA0415 interaction partners SPG11 and SPG15, also known as spatacsin and spastizin, are encoded by two genes that have been associated with hereditary spastic paraplegia with thin corpus callosum (HSP-TCC) [46],[47]. HSP-TCC is a subset of hereditary spastic paraplegia (HSP), which are inherited neurological disorders caused by the degeneration of the cortico-spinal tracts leading to lower-limb spasticity. HSP is a highly heterogeneous condition with at least 46 loci identified so far [48]. A potential interaction of SPG11 and SPG15 has been suggested on the basis of similar neurological symptoms [49], however a physical interaction of SPG11 and SPG15 has not been reported thus far. Because of the physical interaction of KIAA0415 with these two proteins encoded by genes associated with HSP, we decided to investigate if any unexplained HSP cases could be linked to mutations in KIAA0415. Direct sequencing of KIAA0415 in 166 unrelated HSP patients, including 38 and 64 cases with a recessive or dominant inheritance pattern and 64 sporadic cases (see Online Methods), identified 7 known and 15 new variants, respectively. Most of these variants were not considered causative, because they did not affect protein sequence, were not predicted to alter correct splicing, or were also found frequently in control samples (Table S4). However, one of these identified variants led to a premature stop codon at position 527 (c.1413_1426del14/p.L471LfsX56, Table S4) and was absent in 158 Caucasian and 84 North-African control chromosomes. The mutation was heterozygous and no other mutation or variant was found in the coding sequence of KIAA0415 or in its regulatory regions in this apparently sporadic patient (FSP-70-1). No other subjects from the family were available for sampling and no copy number variations were detected on chromosome 7 in the affected patient (unpublished data), but small heterozygous rearrangements or mutations in uncovered regions (unknown exons or introns) might have escaped detection.
More interestingly, we also found a homozygous mutation in two French siblings (FSP-083), which was not detected in 156 Caucasian and 242 North-African control chromosomes. In these patients, a complex indel in exon 2 (c.[80_83del4;79_84ins22], Figure 7C) generates a frameshift and a stop codon following amino-acid 29 (p.R27LfsX3, Figure 7A). Interestingly, the insertion is an imperfect quadruplication of the sequence CTGTAA(A), suggesting DNA polymerase slippage during DNA synthesis as the mechanism for introduction of the mutation. Both affected patients presented with progressive spastic paraplegia associated with urinary incontinence since age 50 and 49, respectively. Cerebral MRI was normal but spinal hyperintensities at C3-C4 and C7 were observed in one. Both parents died at the age of 72 and 77, respectively, of non neurological causes. They originated from two neighbouring villages, but there was no known consanguinity. However, the analysis of three close microsatellite markers (D7S531, D7S517, and D7S1492) and the loss of heterozygosity (LOH) search using CYTO_12 microarrays (unpublished data) confirmed that the region is homozygous in both affected patients (Figure 7B).
To further substantiate a role of KIAA0415 in DNA repair, we compared drug sensitivity in lymphoblast cell lines established from a patient carrying the KIAA0415 mutation (FSP-083-4) and a patient carrying a mutation in SPG15 (FSP-708-22 [50]) to control lymphoblast cell lines. Strikingly, the KIAA0415 mutant cells were significantly (p<0.05) more sensitive to MMC and bleomycin treatments compared to any of the control cell lines (Figure 8). In addition, also the SPG15 cell line showed a mild sensitivity to these drugs, phenocopying the results observed in HeLa and U2OS cells.
Taken together, these experiments identify KIAA0415 as a novel gene, which is mutated in patients with HSP, and implicate a link between HSP and DNA repair.
Using a well-characterized esiRNA library [16] we performed a genome-scale RNAi screen and identified 61 genes that reproducibly decreased or increased the frequency of DNA repair in an assay for homologous recombination [17]. Secondary assays for processes relevant to DNA repair corroborated many of the initial findings. Hence, we provide a dataset that should accelerate the discovery of novel genes with roles in DNA repair and associated medical conditions. Eighteen out of the 61 candidate genes have been described in other large-scale mammalian DNA repair studies [8],[13],[15],[51], demonstrating the effectiveness of our screen, but also highlighting that the use of different assays can uncover novel players. Hence, we predict that the development of alternative DNA repair assays for RNAi screens will reveal additional genes implicated in DNA repair. For our screen we co-transfected the “DNA damaging reagent,” I-SceI, together with the esiRNA silencing triggers. Hence, proteins with long half-lives may have been missed in this screen. Assays in which the DSB is introduced some time after the cells were transfected with the silencing triggers could uncover additional genes playing a role during DNA repair.
To prioritize the molecular investigation of the uncharacterized proteins identified in the screen, we employed a structural bioinformatics approach. Based on the prediction that KIAA0415 represents a novel putative helicase we investigated this gene in more detail. Tagging of the gene using the TransgeneOmics approach revealed nuclear as well as cytoplasmic localization and physical interaction with at least four proteins. Investigations of the interaction partners showed that at least two of these proteins are also required for efficient HR-DSBR. Possibly, these proteins form a complex that is required for efficient HR-DSBR. Consequently, the complex would lose its activity when one of the three proteins is depleted.
Two of the interaction partners of KIAA0415 are encoded by genes that are associated with spastic paraplegia. This result prompted us to examine whether KIAA0415 mutations can explain spasticity in patient samples not linked with mutations in any of the known spastic paraplegia genes. We report a homozygous mutation in KIAA0415, responsible for the spastic paraplegia observed in two siblings. Hence, we identify KIAA0415 as a novel spastic paraplegia associated gene. Based on this finding, we propose to rename KIAA0415 to SPG48 according to the HUGO nomenclature. The fact that three proteins that form a protein complex result in similar phenotypic consequences argues that the whole complex is exerting an important function, which is disturbed when one of the proteins is missing or non-functional. It will therefore be interesting to investigate the remaining interaction partners, C20orf29 and DKFZp761E198, for possible mutations in HSP patients, even though they do not map to known HSP loci [49]. Although only demonstrated for one case, cell lines derived from a patient carrying a SPG48 mutation were more sensitive to DNA damaging drugs than control cells, corroborating a role of SPG48 in DNA repair. Unfortunately, material from other patients with SPG48 mutations was not available. However, we propose that in the future HSP patients be screened for mutations in SPG48 and that cells from these individuals be checked for DNA repair defects.
Genes mutated in HSP have been associated with several biological functions, including intracellular transport, axonal pathfinding, mitochondrial functions, cholesterol metabolism, myelin formation/stability, and chaperonin activity [48]. Based on our findings, we propose that HSP might also be a result of impaired DNA repair, adding HSP to the growing list of neurodegenerative diseases caused by DNA repair deficiencies [4],[5],[7],[8]. In agreement with this hypothesis, SPG11 has recently been reported to be phosphorylated upon DNA damage by ATM or ATR [51]. Whether SPG48 (and its associated proteins) is a direct component of the HR-DSBR pathway or more indirectly linked to DNA repair remains to be established. Biochemical analysis of the putative helicase domain of SPG48 appears to be an attractive entry point into gaining mechanistic insights into the DNA repair function(s) of SPG48.
The technological advances in RNAi screening have increased the speed at which phenotypic data can be obtained. However, interpretation of the resulting genotype-phenotype relationships remains challenging, and approaches that help to decipher the screening data are highly desirable. Approaches that analyze phenotypic data from unrelated RNAi screens and that combine phenotypic- with localization- and proteomic data [52],[53] have been used successfully to bootstrap phenotype-to-function analyses. Here, we explored the possibility of combining RNAi screening data with structural bioinformatics approaches. The obtained results demonstrate that this combination generates valuable information, which helps to prioritize the follow-up studies of uncharacterized candidate genes. We envision that the design of an automatic pipeline to analyze possible structural and functional features beyond protein sequence similarities will further accelerate the characterization of genes identified in RNAi screens. In the future, it will be important to combine the different “omics” and bioinformatics approaches to understand DNA repair at a systems level and to further accelerate the discovery of genes relevant to human pathology.
Ten µg of the DR-GFP construct [17] were transfected into 2.5×106 HeLa cells using 12 µl Enhancer (Qiagen) and 14 µl Effectene (Qiagen) according to the manufacturer's protocol. Stable cell lines were selected with 3 µg/ml puromycin (Sigma-Aldrich) and single clones were obtained by FACS sorting on a FACSAria (BD Biosciences). Colonies derived from individual clones were expanded and tested for their behaviour after transfection with a plasmid encoding the I-SceI endonuclease. A cell line with virtually no GFP positive cells before I-SceI treatment and high number of GFP positive cells after I-SceI treatment was chosen for the screen.
Cells were grown on glass coverslips and fixed with 3% paraformaldehyde (PFA) as described previously [44]. Immunofluorescence stainings were performed with a primary mouse anti-GFP antibody (Roche Diagnostics, 1∶4,000 dilution) and a secondary donkey anti-mouse antibody conjugated to Alexa488 (Molecular Probes, 1∶500 dilution). Genomic DNA was counterstained with ProLong Gold antifade reagent containing DAPI (Invitrogen). Images were acquired on an Axioplan II Microscope (Zeiss) operated through MetaMorph (Molecular Devices).
Western blot analysis was performed as described previously [52]. In this study the following primary antibodies were used: mouse anti-GFP (Roche Diagnostics, 1∶4,000 dilution), mouse anti-DM1alpha tubulin (MPI-CBG Antibody Facility, 1∶50,000 dilution), and rabbit anti-Histone H3 (Abcam 1∶25,000 dilution).
The esiRNA library employed has been described elsewhere [16],[54]. For the screen the I-SceI expression plasmid [17] was co-transfected with individual esiRNAs in an arrayed fashion. Briefly, 50 ng of each esiRNA in 5 µl TE Buffer was pipetted into 384-well tissue culture plates (BD Biosciences) and stored at −20°C. Each plate contained four esiRNAs against Rad51 as positive control (at positions C3, C21, M5, M18) and 12 esiRNAs targeting renilla luciferase (Rluc) as negative control (at positions C4, D3, D4, C22, D21, D22, M6, N5, N6, M19, N18, N19 as shown in Figure 1C). Using a multi-well dispenser (WellMate, Thermo Scientific) a mixture of the I-SceI plasmid (12.75 ng/well) and the Enhancer (0.142 µl/well) in 5 µl/well EC Buffer (Qiagen) was dispensed and briefly spun in a Heraeus Multifuge 4KR (Thermo Electron Corporation). After incubation for 5 min, Effectene (0.12 µl/well) diluted in 5 µl/well EC Buffer was added to each well and plates were briefly spun again. The transfection mixture was incubated for 5 min to allow complex formation. In the meantime HeLa cells carrying the DR-GFP reporter construct were harvested, counted, and diluted to a final concentration of 40 cells/µl in DMEM (Invitrogen) containing 12.5% Fetal Bovine Serum (Invitrogen). Fifty µl of the cell suspension was added to each well using a multi-well dispenser (Wellmate, Thermo Scientific). In order to prevent evaporation, plates were sealed with breathable plate sealing foils (Corning) and incubated in a tissue culture incubator at 37°C in 5% CO2. The medium was replaced 24 h post-transfection. After another 72 h cells were washed with PBS and detached by adding 15 µl/well trypsin/EDTA (Invitrogen). After 25 min cells were fixed by addition of 15 µl/well 3% PFA and stored no longer than 48 h at 4°C. Cells were assayed with a FACSCalibur (BD Biosciences) equipped with a High Throughput Sampler (BD Biosciences). Data were acquired and analyzed using CellQuest Pro (BD Biosciences).
Z-scores were calculated for the percentages of GFP positive cells using the following equation: z = (x−μ) σ−1 with: x − percentage of GFP positive cells; μ − mean percentage of GFP positive cells; σ − standard deviation of the number of GFP positive cells. In the primary screen mean and standard deviations were calculated separately for each plate over all samples on the plate excluding controls. Z-scores were calculated for each esiRNA and averaged for duplicates. The transfection of esiRNA targeting Rad51 was used as positive control and as reference for the assay performance. esiRNAs for which the average z-score was below −2 or over 2 were considered as primary hits (Table S1).
In further validation experiments, the z-scores were calculated based on the mean and standard deviation of the negative control (Rluc transfection). EsiRNAs for which the average z-score for 4 replicates were below −2 or over 2 for one esiRNA and below −1.5 or over 1.5 for a second esiRNA were classified as validated hits. Primer sequences for utilized esiRNAs are presented in Table S5.
Gene enrichment analysis was performed using the Panther Analysis Tools (http://www.pantherdb.org/tools/).
Fifteen ng of each esiRNA diluted in 5 µl Opti-MEM (Invitrogen) was pipetted in 384-well tissue culture plates (Greiner). 0.2 µl Oligofectamine (Invitrogen) was diluted with 4.8 µl Opti-MEM, incubated for 5 min and pipetted to each well of the plate. The mixtures were incubated for 20 min to allow complex formation and 1,000 cells in 40 µl medium were added to each well. Twenty-four hour post-transfection cisplatin (100 ng/ml) or MMC (100 ng/ml) were added for 1 h or cells were exposed to 10 Gy IR. Cells were washed carefully with PBS and new medium was added. After additional 48 h cells were fixed with −20°C cold methanol for 20 min, washed twice with PBS, and blocked with Blocking Buffer (0.2% Gelatin from cold water fish skin (Sigma-Aldrich Chemie) in PBS) for 5 min. Cell nuclei were stained with DAPI (1 µg/ml) and cells were preserved with 0.02% sodium azide in PBS. Images were acquired on an Olympus IX81 microscope (Olympus) and cell numbers were determined using the Scan∧R Analysis software (Olympus). Every knockdown was repeated 3 times. Cell numbers with and without DNA damaging agents were compared to Rluc transfections.
HeLa cells were treated with 10 Gy IR 48 h post esiRNA transfection and fixed 1 h or 6 h later. Cells were stained with a phospho-H2AX antibody (clone JBW301, Upstate Biotechnology, 1∶600 dilution) and with donkey anti-mouse TxRed conjugated antibody (Molecular Probes, 1∶400 dilution). DNA was stained with DAPI (1 µg/ml). Cells were preserved with 0.02% sodium azide in PBS and images were acquired on an Olympus IX81 microscope and analyzed by Scan∧R Analysis software (Olympus). Every knockdown was repeated 3 times. Percentages of gammaH2AX positive cells were compared to Rluc transfections. p values were calculated by Student's t test.
Sequence-based analysis (Blast) failed to identify any statistically significant sequence homology between KIAA0415 and any previously characterized protein. Fold recognition techniques were applied to search for potential structural homologies of KIAA0415 with known protein structures. The threading algorithm ProHit (ProCeryon Biosciences) was used to search for structural resemblance of the uncharacterized KIAA0415 sequence with protein structures of the Brookhaven Protein Databank (PDB). Threading calculations were performed with parameters and scoring functions as previously published [55]. A fold library consisting of 19.961 protein chains representing the PDB at 95% sequence identity was used. Three-dimensional (3D) models for KIAA0415 were generated by threading its sequence through each fold of the fold library. Inspection of fold coverage, gaps position and content in the sequence-to-structure alignments obtained, together with the analysis of the secondary structure prediction obtained for KIAA0415 by PredictProtein (http://www.predictprotein.org/) were used to discard possible false positives in top scoring folds. A three-dimensional model of KIAA0415 was built based on the threading alignments obtained with high confidence predicted folds and four template structures (PDBId: 2d7d, 2p6r, 1gm5, and 2eyq) by using Modeler in Discovery Studio (Accelrys v1.7). Manual docking of ADP and Mg2+ onto the resulting KIAA0415 3D model was done based on the X-ray structures of 2d7d and 1gm5. Refinement of the obtained complex was done with AMBER 10 [56]. A first step of energy-minimization by 1,000 cycles of steepest descent and 500 cycles of conjugate gradient with harmonic force restraints on protein atoms was followed by 3,000 cycles of steepest descent and 3,000 cycles of conjugate gradient without constraints. The system was then heated from 0 to 300K for 10 ps. An equilibration step of 30 ps at 300K was followed by a 10 ns MD productive run. The ff03 force field, periodic boundary conditions at constant pressure with Langevin temperature coupling and Berendsen pressure coupling, TIP3P explicit solvent, counterions, 8 Å cut-off for non-bonded interactions, and the SHAKE algorithm for hydrogens were used.
BAC recombineering and the generation of BAC-transgenic cell lines was performed as described previously [44],[57]. A list of all BACs and primers used in this study is provided in Table S6.
A goat anti-GFP antibody (MPI-CBG Antibody Facility) immobilized on G-protein sepharose (GE Healthcare) or GFP-Trap (Chromotek) were used for immunoprecipitation [44],[52]. Glycine eluated KIAA0415-LAP and SPG11-LAP complexes were analyzed on silver stained SDS PAGE. Excised slices were in-gel digested and analyzed by nanoLC-MS/MS on a LTQ (Thermo Fisher Scientific) as previously reported [58],[59]. Glycine eluates from KIAA0415-LAP, KIAA0415-NFLAP, SPG11-LAP, and DKFZp761E198-LAP immunopurifications were used for in-solution digestion and analyzed by shotgun-LC-MS/MS on a LTQ Orbitrap (Thermo Fisher Scientific) [60]. Proteins identified in more than 15% of 193 independent immunoprecipitations performed in ongoing collaborations projects from unrelated baits were considered common backgrounds and further excluded.
Cell fractionation was performed with the commercially available ProteoExtract kit (Novagene, Merck Biosciences) according to the manufacturer's protocol.
We selected 166 unrelated index cases with spastic paraplegia diagnosed according to the Harding's criteria [61]; 109 had a pure form of the disease and 57 had a complex form partially overlapping with the SPG11 typical phenotype. They included 64 index patients from families with dominant inheritance (mean age at onset: 27.0±16.6 y), 38 index patients with inheritance compatible with an autosomal recessive trait (mean age at onset: 25.6±19.9 y), and 64 patients with no family history of the disease (mean age at onset: 31.2±16.9 y). Most patients were French (n = 137) while the remaining patients originated from other countries in Europe (n = 16), North-Africa (n = 8), or elsewhere (n = 5).
This study was approved by the local Bioethics committee (approval No. 03-12-07 of the Comité Consultatif pour la Protection des Personnes et la Recherche Biomédicale Paris-Necker to Drs A. Durr and A. Brice). Informed and written consents were signed by all participating members of the families before blood samples were collected for DNA extraction. All clinical evaluations were performed according to a protocol established by the European and Mediterranean network for spinocerebellar degenerations (SPATAX, coordinator: Dr. A. Durr) that included: a full medical history and examination, estimation of the age at onset by the patient, observation of additional neurological signs, electroneuromyographic (ENMG) studies, and brain MRI, when possible. Disability was assessed on a 7-point scale as previously described [62],[63].
Mutations in SPAST, SPG3, SPG6, and SPG42 were previously excluded in most of the index patients with dominant transmission by direct sequencing and multiplex ligation-dependent probe amplification in the case of SPAST and SPG3 [64] and unpublished data. Among autosomal recessive and sporadic patients, mutations in the CYP7B1/SPG5 gene were excluded in most patients [63] while SPG11 and SPG15 mutations have been excluded in all complex autosomal recessive forms [50].
All coding exons of the gene KIAA0415 (Ensembl gene ID: ENSG00000164917) and its splice junctions were amplified by PCR on a Thermocycler 9700 (Applied Biosystems, Foster City, CA) using specific primers (see Table S7). 3.1 Kb on the 3′ and 1.5 Kb on the 5′-UTRs were also sequenced in patients with an autosomal recessive transmission carrying a single heterozygote variant. The amplicons were sequenced in both directions using the BIGDYE V3 chemistry in an ABI3730 automated sequencer (Applied Biosystems) as recommended by the supplier. The seqscape v2.6 (Applied Biosystems) software was used to highlight nucleotide variations in comparison to the normal consensus sequence of both genes. In family FSP70, the mutation was confirmed after subcloning of PCR products into the pcDNA3.1/V5-His TOPO TA vector using TOP10 bacteria according to the manufacturer's recommendations (Invitrogen) and direct sequencing of at least 5 independent clones of both alleles.
After identification of a variant, reamplification and resequencing was systematically performed. Segregation of the mutations/polymorphisms with the disease was verified by direct sequencing in additional family members whose DNA samples were available. In addition, 79 and 121 unrelated healthy Caucasian and North-African subjects were screened to evaluate the frequency of new nucleotide changes. In order to estimate evolutionary conservation, gene sequences of different species were downloaded from the Ensembl genome browser (www.ensembl.org) and aligned using the ClustalW algorithm (http://www.ebi.ac.uk/Tools/clustalw2/index.html). All variants were systematically tested for their effect on splicing at: http://rulai.cshl.edu/cgi-bin/tools/ESE3/esefinder.cgi, http://rulai.cshl.edu/new_alt_exon_db2/HTML/score.html, http://www.fruitfly.org/seq_tools/splice.html. Predicted effects of missense changes were tested using SIFT and POLYPHEN at http://sift.jcvi.org/www/SIFT_seq_submit2.html and http://genetics.bwh.harvard.edu/pph/.
Cell lines were obtained from patients by infection with Epstein-Barr-Virus (Table S8). Lymphoblast were cultured in RPMI medium supplemented with 1% Pen/Strep, 2 mM L-Glutamine, 10 mM Hepes, 1% Fungizone, and 20% FCS. 200.000 cells were plated in 6-well plates and cultured without or with 10 ng/ml MMC or exposed to 10 ug/ml bleomycin for 1 h. Growth of the cells was monitored daily by counting the trypan blue negative cells using a Countess Automated Cell Counter (Invitrogen). Four days after incubation 100.000 cells were stained with the FITC Annexin V Appoptosis Kit II (BD Biosciences) followed by FACS (BD Biosciences) analyses following the manufacturer's protocol. Experiments were performed two times in duplicates.
|
10.1371/journal.ppat.1002956 | Post-Transcriptional Regulation of the Sef1 Transcription Factor Controls the Virulence of Candida albicans in Its Mammalian Host | The yeast Candida albicans transitions between distinct lifestyles as a normal component of the human gastrointestinal microbiome and the most common agent of disseminated fungal disease. We previously identified Sef1 as a novel Cys6Zn2 DNA binding protein that plays an essential role in C. albicans virulence by activating the transcription of iron uptake genes in iron-poor environments such as the host bloodstream and internal organs. Conversely, in the iron-replete gastrointestinal tract, persistence as a commensal requires the transcriptional repressor Sfu1, which represses SEF1 and genes for iron uptake. Here, we describe an unexpected, transcription-independent role for Sfu1 in the direct inhibition of Sef1 function through protein complex formation and localization in the cytoplasm, where Sef1 is destabilized. Under iron-limiting conditions, Sef1 forms an alternative complex with the putative kinase, Ssn3, resulting in its phosphorylation, nuclear localization, and transcriptional activity. Analysis of sfu1 and ssn3 mutants in a mammalian model of disseminated candidiasis indicates that these post-transcriptional regulatory mechanisms serve as a means for precise titration of C. albicans virulence.
| Candida albicans is a fungus that resides on the skin and in the gastrointestinal tract of humans and other mammals. However, this commensal organism is also capable of proliferating and causing disease in people who have received antibiotics, who are immunocompromised, or who have suffered injury to epithelial layers. We previously identified a novel transcription factor called Sef1 that promotes C. albicans virulence by activating the expression of iron uptake genes in iron-poor environments, such as the host bloodstream. However, in iron-replete environments such as the gastrointestinal niche, the SEF1 gene is repressed by a second transcription factor called Sfu1. Here, we report our discovery of a series of post-transcriptional regulatory events that determine the intracellular localization, stability, and activity of Sef1 protein. Mutants that disrupt these post-transcriptional events alter C. albicans virulence in a mammalian model of disseminated infection. The existence of multiple levels of regulation speaks to the importance of Sef1 in C. albicans virulence and suggests that close titration of Sef1 activity is important for adaptation to distinct microenvironments within the mammalian host.
| Candida albicans is a ubiquitous component of the mammalian microbiome [1] as well as the most common fungal pathogen of humans [2], [3], [4], [5]. As this organism transits between its commensal niches (mammalian skin and gastrointestinal tract) and those of virulence (bloodstream and internal organs), it experiences profound shifts in the levels of nutrients, the physical environment, and immune surveillance. We previously demonstrated that a novel C. albicans transcriptional regulatory circuit is required for survival in at least two distinct habitats, the host bloodstream and gastrointestinal tract [6], where levels of bioavailable iron differ by more than 20 orders of magnitude [7], [8]. In the bloodstream, where iron is tightly sequestered by host transferrin [7], C. albicans defends against iron deficiency through expression of Sef1, a Cys6Zn2 transcriptional activator of iron uptake genes and an indirect suppressor of the gene for Sfu1 [6]. In the gastrointestinal tract, where iron is abundant thanks to diet and sloughed cells [8], [9], C. albicans defends against iron toxicity through the expression of Sfu1 [6], a GATA family transcriptional repressor that inhibits both SEF1 and genes for iron uptake [6], [10]. Remarkably, the opposing roles of Sef1 and Sfu1 in iron homeostasis extend to differing relationships with the host, with Sef1 promoting virulence and Sfu1 promoting commensalism in animal models [6]. However, the details of how these transcriptional regulators are themselves regulated by iron remain to be elucidated.
Sfu1 is broadly conserved among ascomycetes, and orthologs from multiple species have been shown to play a negative role in iron homeostasis through repression of iron uptake genes [10], [11], [12], [13], [14], [15]. The best-characterized ortholog is Fep1 from Schizosaccharomyces pombe that, like Sfu1, is subject to repression at the transcriptional level when environmental iron is limiting [16], [17]. In this species, protein activity is also regulated by iron, since only iron-bound Fep1 can associate with DNA [18]. By contrast, orthologs of Sef1 have not been extensively characterized, in part because the genomes of only a handful of species in the Saccharomyces and Candida lineages encode this protein [6]. Moreover, C. albicans Sef1 appears to function differently from its S. cerevisiae ortholog, since iron homeostasis in the latter species is controlled by Aft family proteins [19], [20], [21] and is not dependent on Sef1 [6].
Here we describe studies that reveal an unexpected, transcription-independent role of C. albicans Sfu1 in inhibiting Sef1 function, as well as a role for a predicted protein kinase, Ssn3, in Sef1 activation. Specifically, we find that, under iron-replete conditions, Sfu1 physically associates with Sef1 and sequesters it in the cytoplasm, where it is destabilized. In contrast, under iron-depleted conditions, Sef1 forms an alternative complex with Ssn3, resulting in Sef1 phosphorylation, nuclear localization, and the transcriptional activation of iron uptake genes. These post-transcriptional regulatory events are of direct consequence to C. albicans virulence, since either overexpression of SFU1 or deletion of SSN3 results in attenuated virulence in a mammalian model. We hypothesize that these multiple, opposing mechanisms for Sef1 regulation, including a surprising protein-protein interaction with its own transcriptional inhibitor, enable this obligate mammalian parasite to fine-tune its interactions with the host on a spectrum from commensalism to virulence.
Given the important role of Sef1 in promoting C. albicans virulence [6], we speculated that it would be a prime target for regulation. We and others had previously shown that, under iron-replete conditions, transcription of SEF1 is repressed by Sfu1 [6], [10], the C. albicans structural and functional ortholog of S. pombe Fep1 [22]. To determine whether additional regulators contribute to SEF1 gene expression, we used RT-qPCR to compare SEF1 transcript levels in a wild-type strain vs. an isogenic strain lacking the SFU1 gene. The result was that deletion of SFU1 was sufficient to fully derepress SEF1, independent of the iron content of the growth medium (Figure 1a, compare the level of SEF1 in wild-type cells grown in iron-depleted medium [bar 2, derepressing condition] to that in the sfu1ΔΔ strain, grown in either iron-replete [bar 3] or iron-depleted medium [bar 4]); numerical values and statistical analysis are provided in Table S1. These results suggested that iron-dependent transcriptional repression by Sfu1 is sufficient to account for SEF1 transcript levels in wild-type cells.
To determine whether forced overexpression of SFU1 could further suppress SEF1 gene expression, we created a strain in which the endogenous promoter of SFU1 was replaced with the strong, constitutively active TDH3 promoter (SFU1OE); increased levels of SFU1 RNA and protein were confirmed by RT-qPCR and immunoblot analysis, respectively (Figure S1a and S1b). Overexpressed Sfu1 did not substantially diminish the level of SEF1 mRNA under iron-replete or iron-depleted conditions (Figure 1a, compare bar 1 with bar 5 and bar 2 with bar 6). The failure of overexpressed SFU1 to inhibit the transcription of SEF1 under iron-depleted conditions suggested that C. albicans Sfu1 might, like its S. pombe ortholog [18], require iron as a cofactor for binding to DNA and transcriptional repression.
Although SEF1 mRNA levels were normal in the SFU1-overexpression strain (Figure 1a), this strain demonstrated hypersensitivity to treatment with the iron chelator, bathophenanthroline disulfonic acid (BPS; Figure S2), suggestive of a potential defect in iron acquisition. Addition of FeCl3 to the BPS-treated medium was sufficient to reverse the growth defect (Figure S2), confirming the specificity of the iron-chelation phenotype. To determine whether Sef1 protein levels were affected in the SFU1OE strain, we utilized an epitope-tagged version of Sef1 in which 13 copies of the Myc epitope were fused in-frame at the C-terminus; this fusion protein is fully functional [6]. Surprisingly, the steady state level of Sef1-Myc was substantially reduced, particularly under iron-depleted conditions (Figure S3). The observations that overexpression of SFU1 does not affect SEF1 mRNA levels but strongly decreases Sef1 protein levels raised the possibility that Sfu1 may have a second function in the post-transcriptional regulation of Sef1.
To determine whether Sef1 localization is regulated, we used indirect immunofluorescence to visualize Sef1-Myc in wild-type cells exposed to varying concentrations of iron. Under iron-replete conditions, Sef1-Myc was localized primarily in the cytoplasm (Figure 1b, WT strain, H; note the absence of green Sef1-Myc signal in the FITC channel in areas that correspond to red DNA signal in the DAPI channel; a negative control showing minimal staining of an isogenic strain that lacks the Myc epitope is shown in Figure S4a). Under iron-depleted conditions, however, Sef1-Myc was primarily nuclear, with prominent areas of yellow overlap when the FITC and DAPI channels were merged. Notably, examination of Sef1-Myc in an sfu1ΔΔ mutant revealed constitutive nuclear localization, even under iron-replete conditions (sfu1ΔΔ strain, Figure 1b). Conversely, overexpression of SFU1 resulted in substantial cytoplasmic localization of Sef1-Myc even under iron-depleted conditions in which it is usually nuclear (SFU1OE strain, Figure 1b). By comparison, an Sfu1-Myc fusion protein was found to be distributed between the nucleus and cytoplasm in wild-type cells propagated under iron-replete conditions and primarily cytoplasmic under iron-limiting conditions (Figure S4b).
These results established that Sef1 localization varies as a function of iron, that Sfu1 promotes Sef1 localization in the cytoplasm, and that the protein localizing activity of Sfu1—unlike its transcriptional repression activity (Figure 1a)—does not inherently require iron.
Immunoblot analysis of Sef1-Myc recovered from wild-type cells grown under iron-replete vs. iron-depleted conditions demonstrated an inverse relationship between Sef1 protein abundance and iron levels (Figure 2a, lanes 1 and 2), which was expected based on the known, iron-dependent inhibitory activity of Sfu1 on SEF1 gene expression. An unexpected finding was that the electrophoretic mobility of Sef1 also varied in an iron-dependent fashion. This subtle but reproducible decrease in Sef1 mobility under iron-depleted conditions was observed not only in wild-type cells, but also in an sfu1ΔΔ deletion mutant (lanes 3 and 4), arguing against a role for Sfu1 in this process.
We hypothesized that the lower mobility form of Sef1 might result from covalent phosphorylation. To test this hypothesis, we used a tandem affinity purification strategy to recover TAP-tagged Sef1 from C. albicans grown under iron-replete or iron-depleted conditions. Purified TAP-tagged Sef1 exhibited an iron-dependent mobility shift similar to that of Sef1-Myc, with protein from the iron-depleted cells running with slightly lower mobility (Figure 2b, compare lanes 1 and 3). Treatment of the purified proteins with lambda phosphatase, a broad specificity enzyme with activity on phospho-serine, phospho-threonine, and phospho-tyrosine residues, resulted in conversion of the lower mobility form of Sef1-TAP to the higher mobility form (Figure 2b, compare lane 4 to lanes 1 and 2), in support of our hypothesis.
To identify the kinase responsible for low-iron-dependent phosphorylation of Sef1, we tested the 31 available homozygous knockout mutants affecting predicted kinases for sensitivity to BPS. Our reasoning was that, if phosphorylation of Sef1 is required for full induction of iron uptake genes, then a mutant lacking the relevant kinase might be hypersensitive to iron depletion, that is, phenotypically similar to sef1ΔΔ itself [6], [23]. Our screen identified the ssn3ΔΔ mutant as being hypersensitive to iron depletion (Figure S2). Further, an immunoblot of Sef1-Myc recovered from the ssn3ΔΔ strain revealed persistence of the higher mobility form under iron-depleted conditions (Figure 2c), consistent with a role for Ssn3 in phosphorylation of Sef1. The identical result was obtained when Sef1-Myc was examined in a strain encoding a predicted kinase-dead allele of Ssn3 (Ssn3D325A, Figure S5a).
Although C. albicans Ssn3 has not yet been characterized, its S. cerevisiae ortholog is a cyclin-dependent kinase with two known functions: first, it is a component of the Mediator complex with inhibitory activity on RNA polymerase II [24]; second, it phosphorylates a number of specific transcription factors to regulate their activity, nuclear-cytoplasmic localization, and/or stability [25], [26], [27]. To determine whether C. albicans Ssn3 influences the localization of Sef1, we performed indirect immunofluorescence on Myc-tagged Sef1 in the ssn3ΔΔ mutant. As shown in Figure 2d, deletion of SSN3 resulted in constitutive cytoplasmic localization of Sef1-Myc under both iron-replete and iron-depleted conditions; similar mislocalization was observed in a strain containing Ssn3D325A (Figure S5b). Unlike the case with SFU1, however, overexpression of SSN3 via the TDH3 promoter (SSN3OE) had no obvious effect on Sef1-Myc localization (Figure S5b), perhaps indicating that the nuclear localization activity of Ssn3 is restricted to low iron conditions.
The preceding results were suggestive of a model in which Sfu1 and Ssn3 have opposite and competing roles in Sef1 localization, with Sfu1 promoting cytoplasmic localization and Ssn3 promoting nuclear localization. To test this model, we utilized the SFU1-overexpression strain that mislocalizes Sef1-Myc to the cytoplasm under iron-depleted conditions (Figure 1b). We predicted that, if Ssn3 competes with Sfu1 for localization of Sef1, then overexpression of SSN3 might rescue this Sef1 mislocalization phenotype. Indeed, a strain in which both genes are driven by the strong TDH3 promoter exhibits substantial restoration of nuclear Sef1-Myc under iron-depleted conditions, with normal cytoplasmic localization under iron-replete conditions (Figure 3a). These results indicate that Sfu1 and Ssn3 exert opposing roles on Sef1 localization, but only under iron-depleted conditions (when Sef1 is phosphorylated).
To determine whether Sef1 physically associates with Sfu1 and/or Ssn3, we created a series of double epitope-tagged strains, each containing a Myc-tagged version of one of the three potentially interacting proteins and a TAP-tagged version of another; the TAP epitope consists of a calmodulin binding domain fused to a TEV cleavage site and a Protein A domain (Figure S6a; [28], [29]). Co-immunoprecipitation experiments were performed using whole cell extracts prepared from cells grown under iron-replete or iron-depleted conditions. Extracts were incubated with IgG-sepharose, which binds to the Protein A component of the TAP epitope, followed by extensive washing of the immunoprecipitated complexes and protein electrophoresis under denaturing conditions (SDS-PAGE; see Figure S6b for a schematic of the protocol). Finally, immunoblots were probed with anti-Myc antibodies to determine the presence or absence of a Myc-tagged putative binding partner. Specificity of IgG-sepharose for the TAP tag was confirmed by performing experiments with strains containing Myc-tagged fusion proteins and an unfused TAP tag (Figure S6c), and specificity of the anti-Myc antibodies for the Myc epitope was confirmed using cells containing only the TAP-tagged fusion proteins (Figure 3b).
Shown in Figure 3b are the results with Sfu1-Myc and Sef1-TAP. Sfu1-Myc was efficiently co-immunoprecipitated with Sef1-TAP when cells were propagated in iron-replete medium (lane 3, IP), but not when iron-starved cells were used (lane 4, IP). On the other hand, when the epitope tags were reversed, co-immunoprecipitated Sef1-Myc was poorly visualized using extracts of iron-replete cells (Figure 3c, lane 3) but was easily seen using iron-starved cells (Figure 3c, lane 4). Together, these results suggest that Sef1 and Sfu1 interact physically in a manner that is independent of iron levels, whereas the sensitivity of our biochemical assay is a function of the relative abundance of the Myc-tagged protein in the extract.
Co-immunoprecipitation experiments combining either Sef1-Myc or Sfu1-Myc with Ssn3-TAP revealed a robust interaction between Ssn3 and Sef1, but no detectable interaction between Ssn3 and Sfu1 (Figure 3d). That is, Sef1-Myc was efficiently co-immunoprecipitated with Ssn3-TAP from an extract of iron-depleted cells (Figure 3d, lane 4, IP), which express relatively high amounts of Sef1-Myc protein (Figure 3d, lane 4, input), whereas Sfu1-Myc was not co-immunoprecipitated under any condition (Figure 3d, lanes 5 and 6, respectively). When the epitope tags were reversed, Ssn3-Myc was efficiently co-immunoprecipitated with Sef1-TAP using either iron-replete (Figure 3e, lane 3) or iron-depleted (Figure 3e, lane 4) cells; note that Ssn3-Myc is relatively abundant under both conditions. These results suggest that Sef1 physically associates with Ssn3 as well as Sfu1, but these appear to represent alternative complexes since Ssn3 and Sfu1 do not associate with each other.
To learn whether the stability of Sef1 varies with its intracellular localization, we determined the half-life of Myc-tagged Sef1 in wild-type C. albicans and in mutants in which Sef1 is stably localized in either the nucleus or the cytoplasm. Under iron-replete conditions, Sef1 is predominantly cytoplasmic in wild-type C. albicans but is mislocalized to the nucleus in sfu1ΔΔ (Figure 1b). To obtain sufficient Sef1 protein for the analysis and to uncouple the role of Sfu1 in Sef1 localization from its effects on SEF1 transcription, we replaced the endogenous SEF1 promoter with a constitutively active TDH3 promoter in both wild-type and sfu1ΔΔ strains; overexpressed Sef1-Myc exhibited the same pattern of iron-dependent nuclear vs. cytoplasmic localization as Sef1-Myc expressed from its endogenous promoter (Figure S7). The strains were propagated to mid log phase in iron-replete medium, followed by addition of cycloheximide to block further translation, and serial sampling for measurements Sef1-Myc abundance. Shown in Figure 4a is a quantitative immunoblot of Sef1-Myc and alpha tubulin, which was used as an internal control for protein loading. Under these iron-replete conditions, the calculated half-life of cytoplasmic Sef1-Myc was ∼80 minutes (wild type, R2 = 0.94) and that of nuclear Sef1-Myc was ∼160 minutes (sfu1ΔΔ, R2 = 0.92). Next, we examined Sef1-Myc stability under iron-depleted conditions, in which the protein is predominantly nuclear in wild-type cells (Figure 1b) but mislocalized to the cytoplasm in the ssn3ΔΔ mutant (Figure 2d). Wild-type and ssn3ΔΔ strains expressing SEF1-MYC from the endogenous SEF1 promoter were propagated in iron-depleted medium to mid-log phase, then treated with cycloheximide and visualized as above (Figure 4b). Under these iron-depleted conditions, the calculated half-life of nuclear Sef1-Myc (∼150 minutes in wild type; R2 = 0.98) was once again more stable than that of cytoplasmic Sef1-Myc (∼40 minutes in ssn3ΔΔ; R2 = 0.96). The most parsimonious explanation for these results is that Sef1 is degraded more rapidly in the cytoplasm than in the nucleus; however, we cannot exclude the possibility that Ssn3 and Sfu1 exert independent effects on Sef1 degradation that are unrelated to its intracellular localization.
Our current model of Sef1 regulation, which integrates these results with previously published findings [6], [10], [22], is depicted in Figure 4c. According to the model, Sef1 is subject to two distinct forms of Sfu1-mediated repression when environmental iron is replete: 1) transcriptional repression of the SEF1 gene, through direct binding and repression of transcriptional initiation; and 2) post-translational inhibition of Sef1 protein, through direct binding and retention in the cytoplasm, where Sef1 is more rapidly degraded. Alternatively, under iron-limiting conditions, when Sfu1 protein is depleted, Sef1 is bound by Ssn3, phosphorylated, and localized in the nucleus, where it activates expression of iron uptake genes. Our recent observation that Sef1-Myc is constitutively cytoplasmic in an sfu1ΔΔ/ssn3ΔΔ double mutant strain (Figure S8) suggests that Ssn3 may play actively promote the nuclear localization of Sef1, beyond merely extricating Sef1 from Sfu1.
We previously demonstrated that SEF1 gene expression is induced in the iron-limiting environment of the host bloodstream and that SEF1 is required for virulence in a murine model of bloodstream candidiasis [6]. Conversely, we showed that SFU1 is not required for virulence but rather that the sfu1ΔΔ mutant exhibits increased competitive fitness relative to wild-type C. albicans, presumably because of an enhanced ability to take up extracellular iron [6]. Our current results suggest that the negative effect of Sfu1 on C. albicans virulence likely results from mislocalization of Sef1 to the cytoplasm rather than from repression of SEF1 gene expression, since only the former activity is observed under conditions of iron depletion (compare Figure 1a and Figure 1b). We tested this hypothesis by examining the virulence of mutants with moderate (SFU1OE, Figure 1b, low iron condition) to severe (ssn3ΔΔ, Figure 2d, low iron condition) defects in Sef1 nuclear localization. As shown in Figure 4d and 4e, both mutants were significantly attenuated in the murine bloodstream infection model, such that mice infected with either mutant survived longer than mice infected with wild type. Note also that the strength of the virulence defects paralleled the strength of the Sef1 mislocalization defects of the two mutants, with those of ssn3ΔΔ being worse, although contributions from additional misregulated targets of Ssn3 cannot be excluded.
Human pathogenic microorganisms encounter a dearth of iron in the host bloodstream and internal organs [30], [31], [32], and specialized systems for iron acquisition have been demonstrated to be essential for the virulence of numerous bacterial, fungal, and parasitic pathogens, e.g. [33], [34], [35], [36], [37], [38], [39], [40]. Meanwhile, commensal-pathogens such as C. albicans face the additional challenge of potential iron toxicity (from free radicals generated by the Fenton reaction [41]) in niches such as the gastrointestinal tract, where iron is relatively abundant [8], [9]. We previously showed that the C. albicans transcription factors Sef1 and Sfu1 are key components of an iron homeostasis regulatory circuit that permits adaptation to these widely divergent host niches [6]. Sef1 protects against iron deficiency in the bloodstream through the induction of iron uptake genes and repression of SFU1, whereas Sfu1 protects against iron toxicity in the gut through repression of iron uptake genes as well as SEF1. Still unanswered are the questions of how the activities of Sef1 and Sfu1 are themselves tied to iron levels and whether additional regulatory inputs are involved. Here, we define a system for post-transcriptional, iron-dependent regulation of Sef1 protein that precisely controls the virulence of this obligate commensal-pathogen.
Sef1 plays a central role in C. albicans pathogenesis through promoting the expression of virulence factors as well as iron uptake genes, whereas Sfu1 is essential for commensalism [6]. Given its role in virulence and, perhaps, in the choice between commensal and virulent lifestyles, we hypothesized that Sef1 would be a prime target for regulation beyond transcriptional repression by Sfu1. Indeed, our analysis of Myc-tagged Sef1 in wild-type C. albicans has revealed multiple levels of iron-dependent regulation, including nuclear vs. cytoplasmic localization, phosphorylation, and differential protein stability. In wild-type cells, Sef1 protein is nuclear, phosphorylated, stable, and competent for transcriptional activation only under iron-depleted conditions such as those encountered in the bloodstream.
Our analysis of Sef1 in C. albicans mutants has shed further light on the mechanisms of Sef1 regulation. Surprisingly, in the sfu1ΔΔ mutant, Sef1-Myc is constitutively nuclear, whereas in an SFU1-overexpression strain it is predominantly cytoplasmic. These results clearly suggested a role for Sfu1 in the cytoplasmic localization of Sef1. Our screen of C. albicans mutants affecting predicted kinases exposed a role for Ssn3 in promoting cellular resistance to iron depletion as well as phosphorylation of Sef1. Co-immunoprecipitation experiments indicating that Ssn3 forms a physical complex with Sef1 supported a direct role for Ssn3 in Sef1 phosphorylation. Our finding that Sef1-Myc is constitutively cytoplasmic in the ssn3ΔΔ mutant suggested that Ssn3 might oppose Sfu1 by promoting the nuclear localization of Sef1. This hypothesis was validated by the ability of overexpressed SSN3 to overcome the cytoplasmic Sef1-mislocalization phenotype (under low iron conditions) of an SFU1-overexpression strain. Finally, our observations that Sfu1 and Ssn3 were both detectable in complexes with Sef1, but that neither could be found associated with the other, suggested that the functional antagonism between Sfu1 and Ssn3 occurs in part through competitive binding to Sef1 protein. Meanwhile, the observation that Sef1 is constitutively cytoplasmic in an sfu1ΔΔ/ssn3ΔΔ double mutant argues that Ssn3 plays at least one additional role in Sef1 nuclear localization.
These studies led to a revised model of Sef1 regulation (Figure 4c). According to the model, under iron-replete conditions, Sfu1 utilizes two distinct mechanisms to inhibit the function of Sef1: 1) Transcriptional repression, via direct binding to the SEF1 promoter, and 2) Post-transcriptional repression, via binding to Sef1 protein and forced localization in the cytoplasm, where Sef1 is unstable and unable to participate in transcription. To our knowledge, this would be the first example of a regulatory factor that regulates it target by both transcriptional and post-transcriptional mechanisms. Under iron-limiting conditions, Sfu1 protein is depleted, and Sef1 associates with the predicted protein kinase, Ssn3. Ssn3 most likely phosphorylates Sef1 directly, and either the complex or free Sef1 is transported to the nucleus, where Sef1 functions as a transcriptional activator. A key goal of future studies will be to understand how iron regulates these newly described activities of Sfu1 and Ssn3.
The findings that Ssn3 and Sfu1 post-transcriptionally regulate Sef1, an important virulence factor, raised the question of whether these regulatory events impact C. albicans virulence. Previously, we observed that deletion of SFU1 leads to hypervirulence in the murine bloodstream infection model, with the sfu1ΔΔ mutant significantly better at colonizing host kidneys than wild-type C. albicans [6]. We attributed this enhanced fitness to derepression of SEF1 and iron uptake genes in the mutant, resulting in an increased capacity for iron acquisition. In light of our current results showing that Sfu1 requires iron for transcriptional repression activity, a more likely explanation for the fitness advantage of sfu1ΔΔ is that Sef1 is constitutively nuclear (and therefore transcriptionally active) in this strain, whereas in wild type some fraction of Sef1 is retained in the cytoplasm and degraded. Our current observations with SFU1OE and ssn3ΔΔ mutants dovetail with these findings by showing the converse, i.e. that mutants with incremental defects in the nuclear localization of Sef1 have proportional defects in virulence. Together, these results strongly support the hypothesis that C. albicans iron acquisition (and therefore virulence) can be modulated up or down, respectively, through the activities of Ssn3 or Sfu1 on Sef1 localization and stability. We hypothesize that the evolution of such fine-tuned regulation of a potent transcription factor is particularly advantageous to an obligate commensal-pathogen, such as C. albicans, which must continuously adapt to differing iron concentrations among the various microenvironments of its mammalian host, while avoiding excessive expression of pathogenicity genes during its usual role as a commensal.
All procedures involving animals were approved by the Institutional Animal Care and Use Committee at the University of California San Francisco and were carried out according to the National Institutes of Health (NIH) guidelines for the ethical treatment of animals.
C. albicans strains were routinely propagated in YPD, also referred to as “iron-replete” medium. “Iron-depleted” medium is YPD supplemented with one of the specific iron chelators, bathophenanthroline disulfonic acid (BPS) or 2,2′-dipyridyl (DIP), as previously described [6].
All C. albicans strains used in this study are described in Table S2, primers are listed in Table S3, and plasmids are listed in Table S4. Construction of C. albicans knockout mutants, complemented (gene addback) strains, and strains containing Myc-tagged fusion proteins was performed as previously described [6], [42], [43], [44].
For introduction of TAP epitopes at the C-terminus of Sef1, Sfu1, and Ssn3, a series of plasmids was constructed using PCR and homologous recombination in S. cerevisiae [45]. The vector was pRS316 [46], and the insert consisted of (5′ to 3′): a PmeI restriction site; 350–450 bp of target ORF sequence up to, but not including, the stop codon; the TAP tag coding sequence [29]; a SAT1 (dominant selectable marker)-flipper cassette [47]; 350–450 bp of sequence downstream of the target ORF; and a second PmeI restriction site. Plasmids were called pSN150 (Sef1-TAP), pSN228 (Sfu1-TAP), and pSN219 (Ssn3-TAP). PmeI-digested plasmids were transformed into wild-type C. albicans reference strain SN250 [42], and nourseothricin-resistant C. albicans transformants were screened by colony PCR to verify the correct orientation of the C-terminal TAP tag and SAT1-flipper cassette. Strains expressing both Myc- and TAP-tagged fusion proteins were constructed by transforming strains already expressing the Myc-tagged protein with the appropriate PmeI-digested TAP-tag integration fragment, as described above.
Overexpression strains for SEF1- and SSN3 were created by replacing portions of the endogenous promoters with the highly active TDH3 promoter. PCR and homologous recombination in S. cerevisiae [45] were used to create plasmids containing (5′ to 3′): a PmeI restriction site; 350–450 bp of sequence homology ending ∼500 bp upstream of the target ORF; the SAT1 gene (dominant selectable marker); the TDH3 promoter; 350–450 bp of sequence homology beginning with the start codon of the target ORF; and a second PmeI site. The vector was pRS316 [46], the source of NAT1-TDH3 promoter was pCJN542 [48], and the resulting plasmids were named pSN147 (SEF1OE) and pSN229 (SSN3OE). Correct integration of the inserts in nourseothricin-resistant transformants was verified by colony PCR, and overexpression of SEF1 and SSN3 was confirmed by RT-qPCR.
The SFU1 overexpression strain (SN742) was created using an analogous method. pSN141 was engineered to contain (5′ to 3′): a PmeI site; 350–450 bp of sequence upstream of the C. albicans LEU2 ORF; the C. dubliniensis ARG4 gene (selectable marker); the TDH3 promoter; the SFU1 ORF; 350–450 bp sequence downstream of the LEU2 ORF; and a second PmeI restriction site. After digestion with PmeI, the plasmid was transformed into SN515 (sfu1ΔΔ). Correct integration of the insert in Arg+ transformants was verified by colony PCR, and overexpression of SFU1 was confirmed by RT-qPCR.
The Ssn3D325A kinase-dead mutant (SN977) was created in a similar manner to that of the SFU1OE strain. First, PCR and primers SNO1394 through SNO1397 (Table S3) were used to create a D325A-encoding variant of the SSN3 ORF. Next, plasmid pSN239 was engineered to contain (5′ to 3′): a PmeI site; 350–450 bp of sequence upstream of the C. albicans LEU2 ORF; C. dubliniensis ARG4 (selectable marker); the TDH3 promoter; the mutant SSN3 ORF; 350–450 bp sequence downstream of the LEU2 ORF; and a second PmeI restriction site. A Myc-tagged version of Ssn3D325A (SN987) was created using a plasmid (pSN273) that contains (5′ to 3′): a PmeI site; 350–450 bp of SSN3 ORF sequence up to, but not including, the stop codon; sequence encoding 13×Myc; a SAT1-flipper cassette [47]; 272 bp of sequence downstream of the SSN3 ORF; 350–450 bp of sequence downstream of the LEU2 ORF; and a second PmeI restriction site. PmeI-digested plasmid was transformed into SN977, and correct integration in nourseothricin-resistant transformants was verified by colony PCR. Sequences of all PCR products were verified by DNA sequencing.
C. albicans was grown at 30°C for 5–6 hours in “iron-replete” (YPD) or “iron-depleted” medium (YPD supplemented with 500 µM BPS) to OD600 = 0.8–1.0. Cell fixation, cell wall digestion, and antibody hybridization were performed as previously described [49] except that the 9E10 anti-c-Myc antibody (Covance Research) was used at a 1∶300 dilution and detected with a 1∶400 dilution of Cy2-conjugated secondary antibody (Jackson ImmunoResearch, 715-225-151). Images were acquired under 100× oil objective using a cooled CCD camera (Cooke Sensicam) mounted on an inverted microscope (Zeiss Axioplan 200 M; Carl Zeiss MicroImaging) or a Nikon Eclipse TE2000-E fluorescence microscope. All images were processed with ImageJ software (National Institutes of Health).
C. albicans protein extracts were prepared under denaturing condition using a protocol adapted from a previously described method [50]. Lysates corresponding to 1 OD600 of cells were analyzed by SDS-PAGE and immunoblotted with either anti-c-Myc (9E10, Covance Research) for Myc-tagged proteins or anti-peroxidase soluble complex antibody (Sigma, P2416) for TAP-tagged proteins. Immunoblots were also probed with anti-alpha tubulin antibody (Novus Biologicals, NB100-1639) as a loading control.
C. albicans strains were grown on YPD medium (“iron-replete”) or YPD medium supplemented with the specific iron chelator 2,2′-dipyridyl (DIP) at a final concentration of 0.5 mM (“iron-depleted”). A sample of 1–1.5 OD600 cells was taken immediately (zero time point) before addition of cycloheximide to a final concentration of 2 mg/ml. At the indicated times, 1 OD value of cells was collected and harvested for protein preparations and immunoblotting. Semiquantitative detection of protein levels was performed using the LiCor Odyssey Infrared Imager (Lincoln, NE). Integrated fluorescence intensities of individual bands were measured and background subtracted using the Odyssey Application software. The signal from Sef1-Myc bands was normalized to that of alpha tubulin. Calculations of half-life were performed as previously described [51].
Cells expressing TAP-tagged Sef1, Sfu1 or Ssn3 were grown in YPD medium to OD600 = ∼0.3–0.35 and centrifuged for 5 min at 3,000 rpm. Cell pellets were resuspended in “iron-replete” medium (pre-warmed YPD) or “low iron” medium (pre-warmed YPD supplemented with DIP), and grown for an additional 4 hours in the dark. Cells were collected by centrifugation, washed three times with ice-old water, and resuspended in 1 ml of lysis buffer (20 mM Tris, pH 7.4, 100 mM KCl, 5 mM MgCl2, 20% glycerol) with protease and phosphatase inhibitors (Roche). Cells were lysed using a Bead Beater and one-third volume of glass beads. Cell lysates were centrifuged for 2×20 minutes at 14,000 rpm at 4°C. Protein concentration of the supernatants was measured by the Bradford assay. 3 mg of proteins was used for immunoprecipitation with 50 µl of immunoglobulin G-Sepharose resin (IgG Sepharose 6 Fast Flow, GE Healthcare). After 24 h of protein binding with rotation at 4°C, the resin was washed 4 times with lysis buffer and 2 times with tobacco etch virus (TEV) protease cleavage buffer (10 mM Tris-HCl, pH 8, 150 mM NaCl, 0.5 mM EDTA, 0.1% Tween-20). TEV protease (100 U) cleavage was performed in 1 ml buffer at 4°C overnight. The TEV eluate was collected and proteins were recovered by TCA (trichloroacetic acid) precipitation.
Total RNA was prepared using a hot-phenol method [52] and treated with DNaseI using the Turbo DNA-free kit (Ambion). Ten micrograms of RNA was used in standard RT reactions using oligo [(dT)20-N] primers. cDNAs were quantified by qPCR with the primers listed in Table S3 and normalized against ACT1.
As previously described [6], groups of 10 female (8- to 10-week-old) BALB/c mice (Charles Rivers) were injected by tail vein with 5×105 CFUs of wild type (SN425), SFU1OE (SN742), ssn3ΔΔ (SN982), or ssn3ΔΔ/SSN3 (SN978). Mice were monitored twice daily and euthanized when morbidity criteria were met (weight loss >15%, hunched posture, inactivity).
|
10.1371/journal.pntd.0003610 | Mass Drug Administration for Trachoma: How Long Is Not Long Enough? | Blinding trachoma is targeted for elimination by 2020 using the SAFE strategy (Surgery, Antibiotics, Facial cleanliness, and Environmental improvements). Annual mass drug administration (MDA) with azithromycin is a cornerstone of this strategy. If baseline prevalence of clinical signs of trachomatous inflammation – follicular among 1-9 year-olds (TF1-9) is ≥10% but <30%, the World Health Organization guidelines are for at least 3 annual MDAs; if ≥30%, 5. We assessed the likelihood of achieving the global elimination target of TF1-9 <5% at 3 and 5 year evaluations using program reports.
We used the International Trachoma Initiative’s prevalence and treatment database. Of 283 cross-sectional survey pairs with baseline and follow-up data, MDA was conducted in 170 districts. Linear and logistic regression modeling was applied to these to investigate the effect of MDA on baseline prevalence. Reduction to <5% was less likely, though not impossible, at higher baseline TF1-9 prevalences. Increased number of annual MDAs, as well as no skipped MDAs, were significant predictors of reduced TF1-9 at follow-up. The probability of achieving the <5% target was <50% for areas with ≥30% TF1-9 prevalence at baseline, even with 7 or more continuous annual MDAs.
Number of annual MDAs alone appears insufficient to predict program progress; more information on the effects of baseline prevalence, coverage, and underlying environmental and hygienic conditions is needed. Programs should not skip MDAs, and at prevalences >30%, 7 or more annual MDAs may be required to achieve the target. There are five years left before the 2020 deadline to eliminate blinding trachoma. Low endemic settings are poised to succeed in their elimination goals. However, newly-identified high prevalence districts warrant immediate inclusion in the global program. Intensified application of the SAFE strategy is needed in order to guarantee blinding trachoma elimination by 2020.
| Trachoma, the world’s leading infectious cause of blindness, is scheduled for elimination by 2020. Reaching this elimination target depends on successful implementation of the SAFE strategy (Surgery, Antibiotics, Facial cleanliness, and Environmental improvements). Annual mass antibiotic distributions are key to breaking the cycle of transmission in a community. However, it is not clear how many annual mass treatments need to be carried out in order to achieve elimination. Our study analyzes the effect of mass antibiotic distribution on different baseline prevalence levels of trachoma, in order to assess factors that affect the success of reaching elimination goals. We find that the prevailing belief, which suggests that 3 annual mass treatments can achieve local elimination of trachoma at prevalences between 10–30%, and 5 annual mass treatments for districts above this benchmark, is probably incorrect. In fact, much longer intervals may be required with “business as usual” programmatic strategies, which often include skipped years of treatment. Districts with high prevalence levels may require more intense treatment strategies to eliminate trachoma. Intensified recommendations must be implemented without delay in order to reach the 2020 elimination deadline.
| Trachoma remains the world’s leading infectious cause of blindness, although it has disappeared from much of the developed world due to advances in hygiene and sanitation. The World Health Organization (WHO) has classified it amongst the neglected tropical diseases (NTDs), as where it remains, it is concentrated among the world’s poorest populations. These communities live “at the end of the road,” beyond the reach of development infrastructure, and lack access to the basic sanitation measures that prevent disease transmission. Currently, WHO estimates that 232 million people live in endemic areas, 21.4 million have active trachoma, and 7.3 million suffer from trachomatous trichiasis (TT) and are at immediate risk of becoming blind [1–3]. However, through implementation of the SAFE strategy (Surgery, Antibiotics, Facial cleanliness, and Environmental improvements), we hope to reduce active disease, defined as trachomatous inflammation—follicular among children aged 1–9 (TF1–9) [4] to below 5% prevalence in every endemic district by 2020. As over 100 repeated infections are required to cause the scarring that leads to blindness [5], this will ensure that no one accrues sufficient infections to progress to the disease’s blinding end stages, thus accomplishing elimination of blinding trachoma.
In order to achieve sustainable elimination, effective implementation of each component of the SAFE strategy is essential. Treatment with Zithromax (azithromycin) successfully clears individual infections [6,7], but many factors affect the impact of mass drug administration (MDA) at the population level, such as MDA coverage [8,9] and concurrent implementation of environmental improvements and hygiene education [10,11]. Current recommendations from WHO are to perform at least three annual MDAs prior to an impact survey when baseline TF1–9 prevalence is 10–29%, and at least five MDAs before an impact survey when baseline TF1–9 prevalence is ≥30% [12]. These benchmarks were instituted in 2010 as an update to the original guidelines from 2006 [13], which proved insufficient for some high endemic areas.
Many perceive these benchmarks to suggest that a certain number of years of treatment “guarantee” elimination, but this may be incorrect. Even in relatively low-endemic regions, elimination may take more than three annual MDAs [14,15]. Three treatment rounds were also not sufficient for sustained elimination at roughly 30% baseline TF1–9 prevalence [16]. Modeling suggests that where TF1–9 prevalence is ≥50%, five years of annual treatment is likely not enough [17,18]. Indeed, 7–10 MDAs may be necessary [9].
Given the increase in available research and programmatic data, these recommendations can be assessed and refined to allow trachoma control programs to appropriately plan and budget for elimination. In this study, we used a global dataset of baseline and impact surveys to assess the evidence base for the effect of MDA on trachoma prevalence, with the goal of determining whether improved recommendations can be developed in order to improve programmatic efficiency and ensure continuous progress towards elimination.
In order to effectively coordinate the Zithromax donation on behalf of Pfizer, the International Trachoma Initiative (ITI) maintains a comprehensive database of trachoma prevalence and Zithromax treatments performed around the world. This database allows ITI to effectively allocate drugs, and conduct forecasting and planning of programmatic scale-up [19,20]. Data sources include published literature reports and annual applications for Zithromax submitted to ITI, personal communication with national program staff and researchers, and targeted review of other sources. This study includes database updates through February 2014.
Each observation in the database includes the following information, if available: active trachoma prevalence and the clinical sign used as an active indicator (TF or TF/TI), trachomatous trichiasis (TT) prevalence, age range of individuals surveyed for TF and TT, survey location, survey year, survey design and sampling methodology, and data source. Where multiple surveys were conducted at a given location, they were coded to indicate if they preceded or followed treatment. Where treatment was conducted, some entries include estimates of district population, reported antibiotic distribution in doses, and coverage (estimated as doses distributed divided by total population).
There is substantial variation between some of the surveys represented in the database. For example, the indicator used for active trachoma is a measure of circulating disease in a community. Though the WHO standard is to measure trachomatous inflammation—follicular (TF) among children aged 1–9 years (TF1–9), some surveys assessed TF among school-aged children or children under 6 years old. All surveys included in the database used the simplified clinical grading system for trachoma [4], but some measured TF as an indicator for active trachoma and others used a combination of TF and TI (trachomatous inflammation, intense).
While cross-sectional population-based prevalence surveys (PBPS) are considered the gold standard for assessing trachoma prevalence at a given location [19,21], data from trachoma rapid assessments (TRAs) and acceptance sampling trachoma rapid assessments (ASTRA) were reported from some locations. The trachoma community experimented over several years with alternative methods for providing evidence to start programmatic implementation, however, neither have been routinely adopted [21]. TRAs are designed to provide biased prevalence estimates, as they prioritize finding trachoma where it exists [22,23]. In most cases, these TRAs were used to determine areas where a PBPS should be implemented. Prevalence surveys are intended to take place using the district as the implementation unit (where district is defined as an administrative unit of 100,000–250,000 people), but are sometimes performed at a larger geographic area, such as the zonal level, if trachoma is expected to be hyperendemic [12]. Sub-district analyses are also required if TF1–9 prevalence is below 10% at district level [12].
We assessed the factors affecting change in prevalence over time in pairs of surveys collected at the same location. The database initially contained 2365 surveys. These represented 29 countries and were performed between the years 1992–2013. We censored 156 TRAs and 46 ASTRAs. Of the 2157 remaining surveys, 353 represented follow-up after treatment, 1318 represented baseline that preceded treatment, and the remaining 486 represented surveys that did not prompt treatment. All 1671 surveys that preceded or followed treatment were assigned unique IDs by location and matched. Matches were parsed into pairs corresponding to two prevalence surveys in the same location and ordered chronologically. Matched pairs were merged with data on treatment and coverage that used the same unique IDs by location. In areas where follow-up assessment was conducted at a smaller implementation level than the baseline survey (e.g. district surveys following a zonal survey), the follow-up data was averaged across the original unit of implementation to allow comparison.
We investigated adjustment factors where active disease prevalence was not measured as TF1–9. In settings with TF prevalence exceeding 20%, the age-prevalence peak may shift such that younger individuals are more likely to have a greater share of disease burden [5,24–26]. However, data from the PRET trial showed a very high level of correlation between active disease among children 0–5 and 1–9 years old [27,28]. Thus, we did not apply a scaling factor where TF prevalence was assessed among children under six. As the only surveys in the dataset that sampled children aged 6–15 were conducted in Vietnam, where school attendance is high and prevalence peaks among school-aged children [29], no adjustment was applied. If TF/TI was used as an active indicator rather than TF alone, it was adjusted by a factor of 0.87. This was calculated as an average of the relative difference between TF and TF/TI prevalences in published studies [30–33]. Finally, among surveys for which a year range was specified, the survey year was coded as the median of that range or the most recent year of a two-year range.
Pairs were identified as representing MDA if any treatment was recorded between the survey dates, or if ITI coding indicated that MDA had taken place. All other pairs were considered to represent “background” prevalence change. Variables were created representing annual MDAs between treatment (number of MDAs that took place between baseline and follow-up surveys), number of years between surveys, number of years before treatment (years between baseline survey and first MDA), number of years since treatment (years between first MDA and follow-up survey), total annual MDAs (number of MDAs before the follow-up survey, regardless of whether they took place after the baseline survey), skips between (“treatment holiday,” or skipped years between annual MDAs), and total skipped years (any years without treatment before the follow-up survey and after the beginning of treatment). See Fig. 1 for a representation of this coding scheme.
As all temporal information in the database is based on calendar years, discrimination between time intervals smaller than a year was not possible. Thus, a given “year” could be as short as 12 months or as long as 23 (e.g., if a baseline survey took place at the beginning of one calendar year and an MDA took place at the end of the next calendar year). Coding proceeded on the assumption that baseline surveys would be followed by treatment, while impact surveys followed treatment. Instances of anomalous code were manually inspected and cleaned. The final dataset had 170 pairs of surveys corresponding to baseline and follow-up after MDA, and 112 pairs that did not correspond to MDA. All of these represented population-based prevalence surveys.
In order to perform ordinal logistic regression modeling (described below), we created a categorized ordinal variable for TF1–9. TF1–9 categories were specified based on the thresholds that define current WHO recommendations for treatment [12]. An additional category, in which prevalence exceeded 50%, was added to represent hyperendemic settings where trachoma is entrenched (see Fig. 2). These thresholds correlate with number of rounds MDA applied, and often years between surveys, and thus categorize the data into similar groups.
Coverage data, applicable only to the treatment dataset (since the background dataset did not by definition involve MDA), was only reported in 2010–2012. Therefore, coverage data was available for the end of the treatment cycle for only those survey pairs whose treatment interval included at least one of these years: this was true of just 52 (approximately 31%) of the survey pairs in the treatment dataset. We therefore omitted this variable from modeling.
The final dataset contained 282 pairs of surveys, which were conducted between 1996–2013. We used SAS 9.4 (SAS Institute, Cary, NC, USA) to produce descriptive statistics of the dataset (Table 1). Generalized linear models were fit to the “background” dataset, which represented change in prevalence in the absence of MDA, and the “treatment” dataset, which represented MDA’s effect on prevalence. The outcome variable for each was defined as TF1–9 prevalence at follow-up. Stepwise selection and backwards elimination strategies, with entry and stay criteria of α = 0.10, respectively, were used for model building, with all possible variables included at the outset. Aikake Information Criterion (AIC) was used to compare models. The assumption of linearity was confirmed using an overall F test, as well as by plotting the residuals of the explanatory variables. Univariate and multivariate logistic regression models were fitted to banded TF1–9 prevalence at follow-up (see Fig. 2 for categories) to demonstrate the odds of reduction to lower categories of follow-up TF1–9 prevalence. Stepwise selection and backwards elimination were again used to determine final model candidates. Maximum likelihood was used to estimate the coefficients for model predictors [34]. Collinearity was assessed for linear modeling using variance inflation factors, and for logistic modeling using condition indices and variable decomposition factors, calculated with a SAS macro [35]. Given a condition index of ≥30, we investigated variables associated with decomposition factors ≥0.5 [34].
In the treatment dataset, 28 observations coded as representing MDA but missing data on treatment were dropped from the linear and logistic models due to missing predictor values. Pairs dropped included data from Ghana, Nigeria, Tanzania, The Gambia, and Vietnam.
Using several selection strategies in generalized linear modeling, we included the following variables in the final model for the treatment dataset: baseline TF1–9 prevalence (0.13, 95% CI: -0.17, 0.43), rounds of MDA (-2.59, 95% CI: -4.47, -0.71), years since treatment began (1.80, 95% CI: 0.67, 2.93), years before treatment began (-0.94, 95% CI: -1.79, -0.17), and the interaction between rounds of MDA and baseline TF1–9 prevalence (0.062, 95% CI: 0.003, 0.12). These were significant at the 0.05 level, with the exception of baseline prevalence, which also exhibited collinearity with the interaction term but had to be retained for a hierarchically well-formulated model. The final multivariate model, specified below, had an r2 value of 0.40:
TFPr2 = 3.22 + 0.13 * TFPr1–2.59 * Rounds MDA + 1.80 * Years Since Treatment Start - 0.94 * Years Before Treatment + 0.062 * (TFPr1 * Rounds MDA)
In contrast, the best model fit to the background dataset (without MDA) accounted for only about 8% of the variation in the data, demonstrating that these model parameters do not do a good job of accounting for TF1–9 prevalence change in the absence of treatment.
Univariate ordinal logistic regression performed on the treatment dataset (Table 2) demonstrated that increased baseline TF1–9 prevalence was significantly associated with reduced likelihood of achieving lower categories of follow-up TF1–9 prevalence. Years since treatment began and total skipped years since treatment began were also significant. Increased number of annual MDAs and years skipped between annual MDAs also showed a non-significant trend towards association with reduced likelihood of reduction.
A multivariate ordinal regression model fitted to the treatment dataset was used to model the odds of reduction to a lower category of follow-up TF1–9 prevalence. The proportional odds assumption was satisfied for this model. An increase in the following was associated with significantly lower odds of TF1–9 prevalence reduction (see Fig. 3): increased baseline TF1–9 prevalence (OR = 0.92, 95% CI 0.89–0.94), and years since treatment began (0.77, 95% CI = 0.61–0.97). However, an increase in annual MDAs (OR 1.56, 95% CI 1.16, 2.10) and years before treatment (OR 1.30, 95% CI = 1.08, 1.57) were associated with significantly increased odds of TF1–9 prevalence reduction. Censoring of the “super-district” observations, which used mean follow-up TF1–9 prevalence to account for baselines measured at the zonal level, did not have a significant effect on these ORs.
The unmodeled data demonstrated a general trend of reduction from baseline to follow-up in the treatment dataset, though this was more pronounced at lower baseline prevalence levels (see Fig. 4). Correspondingly, there was a significantly greater probability of reduction to a lower prevalence category at lower TF1–9 prevalence levels in the multivariate logistic model (see Fig. 5). While the model predicted a 75% probability of reduction to below 10% given 3 treatment annual MDAs at 20% baseline TF1–9 prevalence, the probability of reduction to below 10% given a 30% baseline TF1–9 prevalence was 56%. At higher baseline endemicities, the point estimate for probabilities became lower, and the error increased. So while a 56% probability of reduction was predicted for a baseline TF1–9 prevalence of 30% given 3 annual MDAs, this was not statistically significant. As number of MDAs increased, the confidence interval narrowed, such that a 64% chance of reduction from 30% baseline was predicted for 5 treatment rounds. Even if the number of MDAs was increased to 10 for an area at 50% endemicity, the probability of reduction (estimated at 42%) was non-significant.
Although various simple measures of skipped years were not significant in the multivariate model, an increase in years since treatment began was significantly associated with reduced odds of prevalence reduction, such that adding a year to the treatment cycle (without a corresponding increase in treatment rounds) led to about a 5% reduction in the probability of success achieving reduction below 10%. The model also predicts increasing success with a waiting period before implementing treatment.
In this study, using data collected in a programmatic context over ITI’s 15-year history, we have demonstrated that the context in which mass drug administration for trachoma is conducted may be as important as the number of annual rounds implemented. Hyperendemic districts (baseline TF1–9 prevalence >50%) should implement at least seven MDAs before considering an impact survey, while relatively low-endemic districts (<20% baseline TF1–9 prevalence) likely could resurvey after three annual MDAs. However, our models are built using data that represents the imperfect world in which trachoma control programs have operated, with skipped treatment years and little data on antibiotic coverage and improvements in hygiene and sanitation. The context in which MDA is implemented is also crucial, and is likely key to successful elimination of trachoma.
Some of the principles demonstrated by our models regarding treatment context are well recognized. Trachoma tends to decline slowly on its own, probably due to the effects of gradual development and improvements in hygiene and sanitation [36,37]. This is likely represented by the variable for years before treatment, which predicts that in the absence of treatment (or before treatment), there is a modest decrease in prevalence at follow-up. Furthermore, trachoma is more likely to reemerge after treatment in higher prevalence settings [8,18,38,39], while in lower prevalence settings it disappears after treatment [40,41]. The variable for baseline TF1–9 prevalence demonstrates that the effect of MDA varies at different endemicities. We had limited ability to investigate interactions between variables due to insufficient power and a small number of potential variables. As such, although the interaction term in the linear models shows that a higher baseline prevalence is less responsive to treatment, neither this term nor a potentially interesting interaction between baseline prevalence and skipped years could be included in the logistic models due to unacceptable levels of multicollinearity. However, in all the models, skipped years, or additional years since treatment began made reduction less likely. We see this effect despite the fact that a single “year” in our data may represent anywhere from 12 to 23 months, given that reporting is agnostic to timing of surveys and treatment during the calendar year.
We assessed the combined effects of these variables by generating predictions for various treatment schemes. The multivariate logistic model predicts that increasing the number of annual MDAs leads to a higher probability of TF1–9 prevalence reduction. No matter how many continuous MDAs are conducted, achievement of the elimination target levels becomes less likely as baseline prevalence increases. Of the ten districts in the treatment dataset with baseline TF1–9 prevalence >50%, none showed reduction to below 5%, and only one achieved reduction to below 10%, despite the application of up to seven annual MDAs (see Fig. 4). This limits the capacity of the model to predict successful reduction in hyperendemic conditions. Even at TF1–9 prevalences between 30–50%, only about half of the districts achieved reduction below 10%.
The model suggests, therefore, that low endemic districts (<20%) are likely to achieve reduction to below 10% after three annual MDAs, and should be resurveyed at that time. However, at 30% baseline TF1–9 prevalence, the model predicts a 56% chance of reduction to below 10%. This probability dwindles as baseline TF1–9 prevalence increases. From the limited available evidence, even 7 annual MDAs were insufficient in hyperendemic districts (>50% TF1–9 prevalence) to make a meaningful public health difference. In such programmatic contexts, over 7 annual MDAs may be necessary to achieve the target. These findings are supported by other studies: in a programmatic context in Mali, three annual rounds of MDA were not sufficient at baseline prevalences of close to 30% [16], while seven to ten years of annual treatment were also suggested by a research study in a hyperendemic setting in Tanzania [9].
Once again, our models do not represent the effect of MDA conducted in controlled conditions. It is likely that many of the districts in our dataset did not achieve their prevalence reduction goals due to inconsistent application of the SAFE strategy. For example, most of the high endemic districts experienced discontinuous treatment. As described, skipped treatment years significantly decrease the probability of TF1–9 prevalence reduction. Our models also omit data on other factors known to influence the effect of MDA, such as treatment coverage [8]. Coverage data was available in such a small subset of surveys that it could not be included in our models; less than half of the districts surveyed in 2010–12 reported any kind of MDA coverage measures to ITI. However, even if more programs provided these estimates, the quality of coverage data currently collected by trachoma control programs is known to vary greatly [42].
We also lack measures of hygiene and environmental factors, the F and E components of the SAFE strategy. Reduction in trachoma has been associated with clean faces and hygiene indicators [43], latrine provision [24,44], and insecticide spraying to control flies, which can act as trachoma vectors where they are prevalent [45,46]. Direct causative evidence is lacking to guide the development of metrics that could be used by control programs. Nonetheless, the endemic equilibrium that leads to reemergence of trachoma is likely dependent on environmental factors [5,17,39]. If the setting in which antibiotic treatment is applied is unchanged, “elimination” will be transient at best.
Despite these omissions, our results are valuable precisely because they represent the effect of MDA as it is conducted by trachoma control programs. Although low endemic districts are likely to succeed in their elimination goals under the current WHO recommendations, we must consider carefully how to support the remaining districts with baseline TF1–9 prevalence over 30%. With just under five years left before the 2020 elimination goal, those districts must plan for intensified treatment programs. They may consider alternatives such as targeted treatment [47] or biannual treatment [8,48]. There may be substantial cost savings associated with proposed integration of efforts to survey and distribute treatment with programs for other NTDs [49–51]. Most importantly, we must recognize that in the imperfect context in which programs on the ground operate, adding more annual MDAs without regard to coverage, programmatic continuity, and underlying environmental context will not guarantee trachoma elimination.
In order to continue our progress towards trachoma elimination, we must emphasize the WHO recommendations that call for programmatic continuity, which should be attainable even in countries where program implementation is difficult, given increased donor support. We must also emphasize the importance of antibiotic coverage, hygiene education, and sanitation improvements. This should start at the level of the data we collect. We cannot track progress, measure success, or even understand what success looks like for variables we do not measure.
Trachoma serves as an object lesson that antibiotic interventions, such as azithromycin mass treatment, can only go so far in the context of poor development. With increasing rounds of MDA, we may eventually reduce TF1–9 prevalence to below 5%, even in the most high-endemic districts remaining. Our data suggests that such districts ought to prepare for extended MDA timelines. However, we should not rely on antibiotics alone to achieve trachoma elimination. The most effective and efficient solution is likely to implement all aspects of the SAFE strategy, which recognizes that though high-coverage, continuous MDAs are essential, clean water and good hygiene may be as important. For programs seeking real and sustainable elimination, it may be that no amount of time is long enough to achieve trachoma elimination without lasting change of the environment in which it persists.
|
10.1371/journal.pgen.1000496 | Consequences of Lineage-Specific Gene Loss on Functional Evolution of Surviving Paralogs: ALDH1A and Retinoic Acid Signaling in Vertebrate Genomes | Genome duplications increase genetic diversity and may facilitate the evolution of gene subfunctions. Little attention, however, has focused on the evolutionary impact of lineage-specific gene loss. Here, we show that identifying lineage-specific gene loss after genome duplication is important for understanding the evolution of gene subfunctions in surviving paralogs and for improving functional connectivity among human and model organism genomes. We examine the general principles of gene loss following duplication, coupled with expression analysis of the retinaldehyde dehydrogenase Aldh1a gene family during retinoic acid signaling in eye development as a case study. Humans have three ALDH1A genes, but teleosts have just one or two. We used comparative genomics and conserved syntenies to identify loss of ohnologs (paralogs derived from genome duplication) and to clarify uncertain phylogenies. Analysis showed that Aldh1a1 and Aldh1a2 form a clade that is sister to Aldh1a3-related genes. Genome comparisons showed secondarily loss of aldh1a1 in teleosts, revealing that Aldh1a1 is not a tetrapod innovation and that aldh1a3 was recently lost in medaka, making it the first known vertebrate with a single aldh1a gene. Interestingly, results revealed asymmetric distribution of surviving ohnologs between co-orthologous teleost chromosome segments, suggesting that local genome architecture can influence ohnolog survival. We propose a model that reconstructs the chromosomal history of the Aldh1a family in the ancestral vertebrate genome, coupled with the evolution of gene functions in surviving Aldh1a ohnologs after R1, R2, and R3 genome duplications. Results provide evidence for early subfunctionalization and late subfunction-partitioning and suggest a mechanistic model based on altered regulation leading to heterochronic gene expression to explain the acquisition or modification of subfunctions by surviving ohnologs that preserve unaltered ancestral developmental programs in the face of gene loss.
| Gene duplication may facilitate the acquisition of genetic diversity. Little is known, however, about the impact of gene loss on the functions of surviving genes. When a gene is lost, can other closely related genes evolve to perform the functions of the lost gene? Answering this question can be difficult because the proof for gene loss is based on negative evidence and thus can easily pass unnoticed. Here, we illustrate how the comparison of genomic neighborhoods in different species can help reconstruct the chromosomal history of a gene family and provide robust evidence for gene loss, even without an appropriate early-diverging comparator group. Identifying gene loss is important because it helps distinguish between gene gain as a lineage-specific innovation and gene loss as a lineage-specific simplification. As a case study, we investigated the expression of the Aldh1a family, which is crucial for retinoic acid signaling in development of eyes, limbs, the brain, and in cancer. Results showed that gene loss is indeed associated with the evolution of functional change in surviving gene family members. Our results highlight the relevance of comparative genomics for identifying gene loss and improving the functional connectivity among human and model organism genomes.
| Understanding the evolution of gene functions during vertebrate evolution is important for the proper interpretation of comparative analyses, especially when using model organisms to understand human gene functions. Gene duplication has been proposed to facilitate the evolution of gene functions [1], and the mechanisms of neofunctionalization and subfunctionalization may play a role [1]–[3] (reviewed in [4]). Human gene families show the signatures of two rounds of whole genome duplication (R1 and R2) that occurred during early vertebrate evolution [1], [5]–[14] (but see [15]). Mutations in gene copies that arose in these R1 and R2 events often cause related diseases (for example, osteogenesis imperfecta (COL1A1) and spondyloepiphyseal dysplasia (COL2A1), bullous erythroderma ichthyosiformis (KRT1) and epidermolysis bullosa (KRT5), and syndactyly type II (HOXD13) and hand-foot-uterus syndrome (HOXA13)). Comparative analysis shows that fish genomes have two co-orthologs for many human genes as a result of a third round of genome duplication (R3) that occurred at the base of the teleost radiation [16]–[28]. Early on, S. Ohno [1] recognized the relevance of increased genetic diversity after genome duplication, and in his honor, gene duplicates originated by genome duplication are called “ohnologs” [29]. This term is useful because of the special properties that ohnologs possess at their birth compared to duplications that arise by other mechanisms such as unequal crossing-over, tandem gene duplication, or retrotransposition.
While many studies focus on how gene duplications can facilitate the acquisition of evolutionary innovations during vertebrate evolution, less attention has been focused on the evolutionary impact of lineage-specific gene losses. Differential ohnolog loss is important because it decreases genetic diversity within a species but increases genetic diversity between species. Loss of one copy of a pair of fully redundant gene duplicates should not usually have significant impact, but duplicate loss after functional divergence can have evolutionary consequences. Reciprocal paralog loss in different lineages can affect a species' biology, decrease evolvability, and diminish adaptability to changing environments [30]–[34]. In other cases, gene loss can be adaptive, and thus relevant for a species' evolution, perhaps even for human origins [35]. Furthermore, reciprocal loss of even fully redundant gene duplicates in two populations may contribute to speciation [33],[36].
Global estimations of gene loss in fully sequenced vertebrate genomes have been inferred by massive phylogenetic reconstructions of gene families [37]–[39]. Large-scale analyses, however, are sensitive to uncertainties of phylogenetic analysis, for example, asymmetric rates of evolution among paralogs can affect tree topologies and generate gene phylogenies that are not congruent with the species phylogenies of which they are a part [40],[41]. Furthermore, published genome-wide studies have not addressed gene function. In principle, gene functions that are associated exclusively with a certain gene may disappear if the gene is lost. It is possible, however, that exclusive gene functions might not disappear in situations in which a surviving paralog might acquire or maintain the expression domain of the lost paralog, and thereby the ancestral developmental or physiological program can remain unaltered [42],[43]. Because the evidence for gene loss is negative and can pass unnoticed and is subject to uncertainties in the completion or assembly of sequenced genomes and in copy number polymorphisms [44],[45], the impact of gene loss in the evolution of function of surviving paralogs is under-investigated. Identification of gene loss is especially important to avoid misinterpretations when human gene functions are inferred from the study of model organisms that might have suffered lineage-specific paralog loss, so that the model has no true ortholog of the phylogenetically most closely related human gene, or vice versa.
To evaluate the evolutionary relevance of gene loss on the functions of surviving paralogs, it is first important to understand gene phylogeny. For genes lost following large-scale genome duplications, conserved syntenies can identify duplicated genomic regions and provide evidence for gene loss, often even in situations lacking a proper outgroup [46]. Genes lost after genome duplication events have been called “ohnologs gone missing” (ogm), and their identification is important to properly distinguish orthologs from other types of paralogs [32],[47],[48]. We propose that identification of gene loss by automated comparative genomic analysis of conserved syntenies can: (1) help resolve uncertain gene phylogenies; (2) help discriminate cases of evolutionary innovations from evolutionary simplifications; (3) facilitate understanding of the diversification of gene functions among species; and, importantly, (4) improve functional connectivity of human and model organism genomes.
To explore the roles of gene loss in a functional context, work reported here focuses on the vertebrate Aldh1a retinaldehyde dehydrogenase gene family (formerly known as Raldh) as a case study. Understanding the evolution of Aldh1a genes is important because this family encodes enzymes responsible for the synthesis of retinoic acid (RA), the active derivative of vitamin A (retinol). In humans, as in other vertebrates, RA plays important roles during embryogenesis, for example, in axial patterning, limb development, and differentiation of eyes and nervous system, as well as functioning in adult organ homeostasis (recently reviewed in [49],[50]). Alterations of RA metabolism can lead to human pathologies including breast and prostate cancers, osteoporosis, rheumatoid arthritis, dermatologic diseases, developmental anomalies and premature births.
The evolutionary origin of the Aldh1a family probably predates the origin of stem bilaterians [51],[52], but the ability of the Aldh1a enzyme of basally diverging bilaterians to synthesize RA remains unknown. Aldh1a likely arose by duplication of an ancestral gene related to the Aldh2 gene family, which encodes a mitochondrial Aldh that plays a major role in acetaldehyde oxidation and is broadly represented in most extant organisms from bacteria to humans [53]. Humans and many other vertebrates have three genes that encode Aldh1a family enzymes: ALDH1A1, ALDH1A2 and ALDH1A3 [54]. Studies of model organisms such as mouse, chicken, frog and zebrafish have provided insights into the roles of each Aldh1a gene in the synthesis of RA (reviewed in [49], [50], [55]–[58]). Variation in Aldh1a gene number in different animal lineages has been hypothesized to be relevant to animal evolution due to potential effects of RA metabolism on the mechanisms of development [59]–[61]; reviewed in [62].
Rodents have a fourth Aldh1a paralog that is mostly expressed in kidney (termed, Aldh1a4 in rat [63], and its ortholog Aldh1a7 in mouse [64]); these genes originated by a tandem gene duplication in the rodent lineage after it diverged from the human lineage. Experiments using a heterologous Xenopus system to express mouse Aldh1a7 suggested that Aldh1a7 might not be involved in RA synthesis [64]. In contrast to rodents with four Aldh1 genes, most teleost fish have just two, aldh1a2 and aldh1a3, but they lack aldh1a1 [59],[65]. Phylogenetic relationships of vertebrate Aldh1a1 genes are still controversial, and whether Aldh1a1 is a tetrapod innovation or its absence from teleosts is due to gene loss is still unknown. Furthermore, the functional consequences of these gene copy number variations have not yet been investigated.
Here, we show how comprehensive comparative genomic analyses of syntenic conservation provides a framework necessary for the examination of the general mechanisms by which lineage-specific gene loss can impact the functions of surviving paralogs. This work reveals multiple losses of Aldh1a ohnologs and proposes an evolutionary genomic model that reconstructs the history of Aldh1a-related vertebrate chromosomes and the evolution of Aldh1a gene functions during and subsequent to the R1, R2, and R3 genome duplications. Results show that acquisition or modification of expression domains by surviving paralogs may lead to lineage-specific innovations that preserve unaltered ancestral developmental programs in the face of gene loss. This work highlights the importance of comparative genomics for understanding the historical basis of gene loss, and to improve functional connectivity between model organism and human genomes.
To understand the history of gene gain and loss in the Aldh1a family, it is important to first understand the phylogeny of family members. Unfortunately, evolutionary relationships among vertebrate Aldh1a paralogs are currently unclear. In one analysis, the three vertebrate Aldh1a clades collapsed to an unresolved trichotomy [59], and in another, Aldh1a2 and Aldh1a3 appeared as sister groups (Aldh1a1, (Aldh1a2, Aldh1a3)), supported by low bootstrap values [65]. These problems may stem from sequence similarities among the Aldh1a1, Aldh1a2 and Aldh1a3 proteins and the use of the evolutionarily distant mitochondrial Aldh2 family to root the tree. To overcome this uncertainty, we turned to a chordate outgroup, the cephalochordate amphioxus, whose lineage diverged from that of the vertebrates before the R1 and R2 events [66],[67]. Amphioxus has both Aldh1a and Aldh2 gene families [59], and hence its Aldh1a genes are much more closely related to vertebrate Aldh1a1 genes than is the Aldh2 gene family. We found that several different phylogenetic methodologies, including Bayesian inference, Maximum-likelihood, gamma-corrected Neighbor-Joining and Maximum-Parsimony all agreed on the same tree topology ((Aldh1a1, Aldh1a2), Aldh1a3)), with Aldh1a1 and Aldh1a2 as sister groups (Figures 1 and S1). This phylogeny differs from both published results: the trichotomy result and the view of Aldh1a2 and Aldh1a3 as sister clades [59],[65]. Our results still provided only a moderately high probability of 0.76 supporting the Aldh1a1/2 clade under the Bayesian phylogenetic inference (Figure 1); thus, phylogenetic analysis alone is insufficient to definitively resolve Aldh1a relationships. To further test historical relationships among Aldh1a paralogs, we examined a data set independent of Aldh1a gene sequence by conducting comparative genomic analyses of the entire genomic neighborhoods (GN) surrounding Aldh1a genes in the genomes of humans and other vertebrates.
The results of our phylogenetic analysis ((Aldh1a1, Aldh1a2), Aldh1a3) (Figure 1) implies that the duplication event that gave rise to Aldh1a1 and Aldh1a2 was more recent than the duplication event that gave rise to Aldh1a3 and the ancestral Aldh1a1/2 gene. If the duplication events that produced the Aldh1a family involved whole genomes or large chromosomal segments, then the phylogenic hypothesis of relationships (Figure 1) predicts more syntenic conservation between the genomic neighborhoods (GN) surrounding Aldh1a1 and Aldh1a2 than between the genomic neighborhood of Aldh1a3 and either Aldh1a1 or Aldh1a2. To test this hypothesis, we conducted a comparative genomic analysis of conserved synteny among the genomic neighborhoods of each ALDH1A paralog in the human genome.
The three human ALDH1A genes are located on two chromosomes: ALDH1A1 is on Hsa9 (human chromosome 9), while ALDH1A2 and ALDH1A3 are on Hsa15 separated by 43 megabases (Mb). We first made a composite dotplot to represent the genome-wide distribution of the paralogs of all genes within a 10 Mb-window surrounding each human ALDH1A gene throughout the 23 human chromosomes (y-axis) (we refer to this set of genes as ALDH1A-neighbor paralogs (red, blue and green crosses in Figure 2A)). Table S1 lists gene names, reference numbers, genomic positions and outgroup (i.e. Branchiostoma floridae and Ciona intestinalis) gene information used to construct each paralogy group in the dotplot. This plot showed that while some ALDH1A-neighbor paralogs appeared randomly scattered throughout the genome, some chromosomal regions contained a concentration of ALDH1A-neighbor paralogs (yellow and pink boxes in Figure 2A). These chromosome regions with syntenic conservation to ALDH1A-neighbor paralogs likely represent chromosome fragments that were duplicated during the whole genome duplication events R1 and R2 and are historically related to the expansion of the Aldh1a family. The presence of ALDH1A-neighbor genes conserved among ALDH1A genomic neighborhoods (pink-shaded dotted boxes) suggests that the ALDH1A family expanded by large-scale genome duplications rather than by local tandem gene duplications. The dotplot analysis also identified four genomic regions that share syntenic conservation with ALDH1A genomic neighborhoods, but do not contain ALDH1A genes (yellow-shaded boxes on Hsa1, Hsa5, Hsa9 and Hsa19). The paralogs of each gene contained in these four yellow boxes were also included in the dotplot (Figure 2A: golden, black, pink and brown crosses). In principle, the existence of the yellow-boxed regions that lack ALDH1A paralogs but show syntenic conservation with the ALDH1A genomic neighborhood could be explained by genome duplications followed by a loss of the ALDH1A paralog (i.e. ALDH1A ohnologs gone missing), or alternatively by the translocation of a portion of the genomic neighborhood away from the ALDH1A gene itself.
When gene functions are compared among different organisms, it is important to distinguish whether the compared genes are orthologs or paralogs. In some cases, reciprocal loss of paralogs in different organisms can lead to the misinterpretation of paralogs as orthologs. Intriguingly, while in most vertebrates Aldh1a2 and Aldh1a3 are on the same chromosome separated by an intervening region of few tens of megabases, in rodents Aldh1a2 and Aldh1a3 are on different chromosomes [63],[64]. In rats, for instance, Aldh1a3 is on the same chromosome as Aldh1a1 rather than being on the same chromosome as Aldh1a2 as in human. This arrangement would be expected if the rat Aldh1a3 gene were a paralog rather than an ortholog of human ALDH1A3. Phylogenetic analysis provided strong support for the conclusion that all vertebrate Aldh1a3 genes are orthologs [59],[65],[69], but evidence for an ALDH1A3 ohnolog gone missing from the human genome raises the possibility of reciprocal paralog loss that would have caused human and rodent Aldh1a3 genes to be paralogs rather than orthologs.
To see whether the mouse Aldh1a3 gene is orthologous to human ALDH1A3 or to ALDH1A3-ogm, we first constructed a dotplot that displayed the distribution of the mouse orthologs of human genes within 10 Mb of ALDH1A1, ALDH1A2, ALDH1A3 and ALDH1A3-ogm (Figure 3A and Table S2). The dotplot revealed that most mouse orthologs of the human ALDH1A-neighbor genes tightly clustered on four mouse chromosomes (Mmu7, Mmu9, Mmu13, and Mmu19) (Figure 3A). Next, we compared these four mouse chromosomes to their orthologons on human chromosomes Hsa5, Hsa9 and Hsa15 in a circleplot (Figure 3B). These analyses identified four clusters of orthology in the Synteny database [48] that unequivocally related mouse orthologs of human ALDH1A1, ALDH1A2, ALDH1A3, and ALDH1A3-ogm genome neighborhoods (Figure S2). The identification of a genomic region on Mmu13 that lacks any Aldh1a gene but that nevertheless possesses orthologous syntenic conservation to ALDH1A3-ogm genomic neighborhood on Hsa5 (golden bundle in Figure 3B) provides strong evidence that the loss of Aldh1a3-ogm predated the split between the lineages leading to humans and rodents, and discards the hypothesis of reciprocal paralog loss. These results conclusively rule out the hypothesis that the rodent Aldh1a3 is an ortholog of human ALDH1A3-ogm, and independently supports orthologous relationships between human and mouse Aldh1a3 genes inferred by phylogenetic methods (Figure 1).
Because the number of Aldh1a paralogs detected in genome databases is lower in teleosts than in tetrapods [59],[65], we performed a comparative genomic analysis of conserved synteny between Aldh1a genomic neighborhoods in the genomes of three teleosts and human to learn the historical basis of different numbers of gene family members (Figure 4).
This work illustrates how comparative analysis of whole genomes is important for functional connectivities between humans and model organisms. Analysis of conserved syntenies related to individual gene families helps identify lineage-specific gene gains and losses that can translate to evolving developmental mechanisms. Using the evolution of the Aldh1a family as a case study, we sought to probe the general mechanisms underlying the impact of gene loss on the functional fate of surviving paralogs after genome duplications while preserving unaltered ancestral developmental programs.
Retinoic acid plays important morphogenetic roles in chordate embryonic development. The recent identification of components of the RA genetic machinery in non-chordate deuterostomes and in protostomes opens the possibility that expansion and reduction in RA-related gene families could have played a role in the developmental diversification of bilaterians [51],[52]. The Aldh1a gene family, which encodes enzymes that synthesize RA, has expanded independently several times during the evolution of the three chordate subphyla, the Cephalochordata, Urochordata and Vertebrata [59]. Within vertebrates, the expansion of the Aldh1a family generated three main paralogs - Aldh1a1, Aldh1a2 and Aldh1a3 - but the phylogenetic relationships and origins of these genes remained uncertain [59],[65].
To identify gene gains and losses, one must first reconstruct the evolutionary genomic history of a gene family. We undertook a combination of phylogenetic and comparative genomic analyses of conserved syntenies that clarified the evolutionary history of the Aldh1a family. Phylogenetic results showed that Aldh1a1 and Aldh1a2 form sister clades and Aldh1a3 occupies a basal position in the phylogenetic tree rooted on cephalochordate Aldh1a genes (Figure 1). This analysis breaks the trichotomy observed in one previous analysis [59] and is opposite to the topology rooted on the far more distant Aldh2 gene family in another analysis [65].
Further support for the new understanding of Aldh1a family member relationships ((Aldh1a1, Aldh1a2) Aldh1a3) comes from comparative genomic analyses of conserved syntenies in the genomic neighborhoods of Aldh1a paralogs in human, mouse, zebrafish, stickleback and medaka, which showed extensive conservation of syntenies between Aldh1a1 and Aldh1a2 genetic neighborhoods (Figure 2B). The congruency of the inferred historical relationships that arise from the new phylogeny and conserved syntenies, which are independent datasets, forces the conclusion that Aldh1a1 and Aldh1a2 are sisters and both are cousins to the Aldh1a3 gene.
Based on results obtained from the analysis of synteny conservation of the Aldh1a1 genomic neighborhoods across human and model organism genomes, we infer an evolutionary model that reconstructs the genomic history of the Aldh1a family, and integrates previous work by Nakatani et al., (2007) [26] that had reconstructed the re-organization of the ancestral chromosomes (named A to J) of the last common ancestor of vertebrates through R1, R2 and R3 genome duplications (Figure 7). Because Aldh1a2 and Aldh1a3 are syntenic (on the same chromosome) in human, zebrafish, and stickleback genomes, we conclude that this was the state in their last common ancestor (Figure 7 step 1). According to Nakatani's reconstruction, Hsa15 mostly derives from the post-R2 ancestral chromosome “A4”, which allows us to infer that Aldh1a2 and Aldh1a3 were syntenic in the ancestral chromosome A4 (Figure 7 step 1). After our comparative analysis of synteny conservation between human and mouse, which ruled out the possibility of reciprocal Aldh1a3 paralog losses (Figure 3) and showed that Aldh1a3 genes are actual orthologs (Figure 1), we conclude that the Aldh1a3-ogm was already absent in the last common ancestor of tetrapods and teleosts (Figure 7 step 1). If Aldh1a2 and Aldh1a3 were syntenic in the ancestral state, we reason that a chromosomal translocation might have occurred during the evolution of the rodent lineage to separate them into different chromosomes (e.g Mmu9 and Mmu7 in Figure 7 step 2). Because the fourth Aldh1a paralog of rodents (i.e. Aldh1a7 in mouse) is adjacent and oppositely oriented to Aldh1a1, separated only by 0.5 Mb with no intervening genes, we conclude that the fourth Aldh1a rodent paralog originated by a rodent-specific tandem gene duplication associated with a local inversion (Figure 7 step 2) that was probably followed by subsequent amino acid sequence changes that destroyed its ability to synthesize RA [64].
In contrast to tetrapods, teleosts lack an Aldh1a1 ortholog, and whether this is due to a gene loss in teleosts, or a gene gain by tetrapods was previously unknown. Our whole-genome comparisons of conserved synteny answer this question by identifying genomic neighborhoods orthologous to the human ALDH1A1 genomic neighborhood in zebrafish, stickleback and medaka (Figure 5). This finding is consistent with the new Aldh1a phylogeny (Figure 1) and provides strong evidence supporting the conclusion that Aldh1a1 was present in the last common ancestor before the tetrapod and teleost lineages split (Figure 7 step 1). Thus, we conclude that the absence of Aldh1a1 in teleosts is due to gene loss, probably in stem teleosts or ealier in stem actinopterygians (Figure 7 step 3), and discards the hypothesis that Aldh1a1 is a tetrapod innovation. This finding illustrates the power of comparative genomics to discern cases of gene losses from cases of gene gains, even in situations in which no proper outgroup is available.
In human and mouse, Aldh1a1 and Aldh1a3-ogm genomic neighborhoods are not syntenic (e.g. on Hsa9 and Hsa5, respectively). Interestingly, however, in zebrafish, stickleback and medaka, genomic neighborhoods orthologous to those of human ALDH1A1 and ALDH1A3-ogm are syntenic (e.g., on Dre5, GacXIII and Ola9, see Figure 4A–C and right panels of Figure 7). Thus, just as Aldh1a2 and Aldh1a3 were syntenic after R2, it is likely that Aldh1a1 and Aldh1a3-ogm were also syntenic before the tetrapod-teleost split (Figure 7 step 1). This reasoning lead us to conclude that there might be a chromosomal translocation event that separated the Aldh1a1 and Aldh1a3-ogm genomic neighborhoods to two different chromosomes during the evolution of the tetrapod lineage (Figure 7 step 4). This predicted translocation event is supported by the reconstruction of ancestral chromosomes made by Nakatani et al. (2007), [26], in which a post-R2 ancestral chromosome named “A0” split into two main pieces that are today on Hsa5 and Hsa9. Thus, we conclude that Aldh1a1 and Aldh1a3-ogm were syntenic in the ancestral chromosome A0 (Figure 7 step 1), which broke apart in tetrapods (Figure 7 step 4) but remained intact in the teleost lineage (Figure 7 step 3). Consistent with our finding that Hsa1 and Hsa9 are related to the ALDH1A genomic neighborhoods in the human genome despite their lack of ALDH1A genes (Figure 2A yellow boxes), Nakatani's reconstruction also shows that most of Hsa1 and Hsa19 derive from “A2–A5” and “A1–A3”, respectively, which are the other post-R2 homeologs derived from the ancestral chromosome “A” present in the genome of the last common pre-R1 vertebrate ancestor. Therefore, we conclude that the conserved synteny related to ALDH1A we detected in Hsa1, Hsa5, Hsa9, Hsa15 and Hsa19 originated by R1 and R2 from the ancestral chromosome “A” in the genome of the last common ancestor of vertebrates.
In Figure 7, we propose two hypotheses to explain how a single gene located in pre-duplication chromosome “A” generated Aldh1a2 and Aldh1a3 in ancestral chromosome “A4”, and Aldh1a1 and Aldh1a3-ogm in ancestral chromosome “A0” inferred after R2 (Figure 7 step 1). The first hypothesis suggests Aldh1a duplication before R1 (pink box in Figure 7), and the second hypothesis invokes a translocation either before or after R2 (tan box in Figure 7) (see legend in Figure 7 for details). Independently of the order of events, however, both scenarios agree that the first duplication generated Aldh1a1/2 and Aldh3/3-ogm ancestral genes from the precursor Aldh1a1/2/3/3-ogm gene in the ancestral chromosome “A”. Because the available genomic databases of basally divergent vertebrates such as cartilaginous fishes (e.g. dogfish, little skate or elephant shark), or from basally divergent craniates (e.g. lampreys or hagfish), are still too fragmented to perform a comprehensive analysis of conserved synteny, testing the hypothetical “pre-R1 duplication” or “translocation” scenarios must be delayed.
Supporting the postulated R3 teleost-specific genome duplication, our analysis of conserved synteny between the ALDH1A genomic neighborhoods and teleost genomes (Figure 4) revealed that fish orthologs of human ALDH1A neighbors are mostly confined to two main chromosomes in each fish species, and no extra R3-generated aldh1a ohnologs have been preserved in duplicated copies (Figure 7 step 5). Analysis of conserved synteny (Figure 4) supports the conclusion that each preserved duplicated Aldh1a gene is an actual ortholog of its partners within teleosts, and no evidence supports the complementary loss of aldh1a paralogs after R3 in different teleost lineages.
The distribution of conserved co-orthologs in teleost paralogons, however, was asymmetric. In each of four genomic regions for three teleost species, one homeolog (the primary chromosome) conserved substantially more genes in the observed region than the other chromosome (the secondary chromosome) (Figure 4D). This asymmetric distribution of syntenic gene conservation appears to be a common characteristic for R3-generated genomic neighborhoods, in agreement with previous observations of the analysis of the hox and parahox genomic neighborhoods in teleosts [17], [89]–[91] and the analysis of syntenic blocs formed following tetraploidy in Arabidopsis [92]. Evolutionary sequence divergence among paralogs also often display asymmetry, with one paralog evolving at a rate similar to its tetrapod ortholog and the other paralog evolving at an accelerated rate, suggesting neofunctionalization [39], [93]–[95]. During the analysis of the hox cluster it was noted that the fastest evolving hox genes belong to clusters that tend to lose their hox genes faster [89],[96]. Furthermore, the asymmetric distribution of synteny conservation between parahox cluster paralogons in teleosts, was accompanied by asymmetric accumulation of introns and repetitive DNA elements in type III RTK genes, and asymmetric conservation of potential regulatory elements [91]. Thus, our observation of asymmetric chromosomal distribution of surviving co-orthologs in the aldh1a genomic neighborhoods extend previous observations in the hox and parahox genomic regions, to genomic neighborhoods with a great variety of gene types, suggesting that the probability of duplicate gene preservation depends not only on inherent evolutionary forces depending on gene function (i.e. subfunctionalization and neofunctionalization), but also on properties pertaining to the architecture of the local genomic neighborhood. The R3 Aldh1a-ogm genes appear to represent cases in which, once gene organization had become altered in one of the duplicated regions, constraints that preserve genes became more relaxed, and therefore the chances of additional gene losses and further chromosomal rearrangements in the secondary chromosome were increased.
At least two possible mechanisms could explain asymmetric co-ortholog retention: first, enhancers shared or embedded in genes at distant sites [70],[91], or second, epigenetic regulatory mechanisms based on chromatin architecture [71],[72]. In principle, shared or distant enhancers or epigenetic regulatory signals must be retained in one homeolog, thus facilitating neighborhood gene retention, but can be lost from the other, allowing more gene loss and more rapid gene evolution due to greater relaxation of evolutionary constraints.
In addition to the loss of aldh1a1 in stem teleosts (Figure 7 step 3), our genomic surveys revealed that aldh1a3 is absent from the genomic database of medaka fish (Figure 7 step 6). Identification of a genomic neighborhood in medaka that shows conserved orthologous synteny with the stickleback and human Aldh1a3 genomic neighborhoods (Figure 6) provides evidence that aldh1a3 was lost in the medaka lineage after it diverged from the stickleback lineage (Figure 7 step 6). This finding illustrates again the power of comparative analysis of conserved synteny to provide evidence of gene loss. The finding of an apparent LTR-retrotransposon in the orthologous position occupied by aldh1a3 in stickleback and human suggests that the insertion of this mobile element may have caused the loss of aldh1a3 in medaka. Genomic data from medaka species phylogenetically close to Oryzias latipes is not yet available to more narrowly define the timing of this insertion event.
The finding of the loss of aldh1a3 in medaka makes this organism the first known vertebrate with a single surviving Aldh1a paralog (i.e. aldh1a2), and made us wonder about the functional implications of gene loss. As a measure of gene function, consider expression patterns of Aldh1a genes. In the developing retina of mouse, frog, zebrafish and medaka, Aldh1a genes are expressed in a dorsal sector and in a ventral sector at the completion of optic cup invagination (about E11.5 in mouse, stage 35 in frog, and 1.5 days post fertilization in zebrafish and medaka; Figure 8A). Different vertebrates express different Aldh1a genes in different dorso-ventral sectors of the eye. The right column of Figure 8 summarizes the main expression patterns of the Aldh1a family in the retina of different animals (Aldh1a1 in red, Aldh1a2 in blue, and Aldh1a3 in green). Aldh1a paralogs expressed in the dorsal sector of the retina include Aldh1a1 (but not Aldh1a2) in mouse; both Aldh1a1 and Aldh1a2 in frogs and birds (e.g. chicken and quail, not included in Figure 8A for simplicity); and Aldh1a2 (but not Aldh1a1) in teleosts (e.g. zebrafish and medaka). The main Aldh1a paralog expressed in the ventral sector of the retina is Aldh1a3 both in tetrapods (e.g. mouse, frog and birds) and in at least one teleost (e.g. zebrafish). In contrast, in medaka, which lacks an aldh1a3 paralog, we found strong expression of aldh1a2 ventrally (Figure 6). Dotted regions depict weak expression of Aldh1a genes in a small part of each dorso-ventral sector or from earlier developmental stages prior to the complete invagination of the optic cup in Figure 8A.
The rules of ancestral reconstruction imply that the retina of the last common vertebrate ancestor probably had a dorsal and a ventral sector, and the original Aldh1a1/2/3/3-ogm gene prior to the expansion of the Aldh1a family gene was likely expressed in both dorsal and ventral sectors (Figure 8A step 1). According to the evolutionary model proposed in Figure 7, the first expansion of the Aldh1a family occurred before R2 and generated Aldh1a1/2 and Aldh1a3/3-ogm. Because Aldh1a3 is the major paralog in the ventral sector of the retina in extant tetrapods and teleosts, and because Aldh1a1 or Aldh1a2 are the major paralogs in the dorsal sector of the retina in both tetrapods and teleosts, we infer that after the first duplication prior to R2, Aldh1a1/2 inherited the subfunction leading to expression in the dorsal sector of the retina, and Aldh1a3/3-ogm inherited the subfunction causing expression in the ventral sector (Figure 8A step 2). It is probable that this subfunctionalization event contributed to the preservation of both paralogs as expected under the duplication-degeneration-complementation (DDC) model, in which the summation of the subfunctions that were split between gene duplicates equals the ancestral function prior the duplication event [3]. After R2, Aldh1a3-ogm was lost and Aldh1a3 became the main ventral source of RA in the retina. On the other hand, both Aldh1a1 and Aldh1a2 retained expression in the dorsal sector because it is preserved in frog, chicken and quail, but not in mouse. Thus we conclude that the absence of Aldh1a2 dorsal expression in mouse retina is due to a loss of an ancestral expression domain, which can be interpreted as an evolutionary innovation due to late subfunction partitioning [3], in which a function that was originally possessed by both Aldh1a1 and Aldh1a2 became partitioned exclusively to Aldh1a1 (Figure 8A step 3). Analysis of the ALDH1A2 expression pattern in the human retina will help elucidate whether the loss of the Aldh1a2 dorsal expression domain occurred before the split of lineages leading to human and rodents, or if it is a feature that has been acquired specifically in the rodent lineage.
An important question is how gene loss can impact the evolution of gene regulation and gene function in surviving paralogs. After the loss of Aldh1a1 in teleosts, Aldh1a2 became the only source of RA in the dorsal retina, taking full responsibility for subfunctions originally shared with Aldh1a1. Natural selection would have gradually increased the strength of the ancestral dorsal domain of Aldh1a2 (Figure 8A step 4). Medaka lacks both aldh1a1 and aldh1a3 orthologs, and the only surviving Aldh1a gene is aldh1a2, which is expressed in both the dorsal and ventral domains of the retina (Figure 8A step 5). The fact that in zebrafish and mouse, Aldh1a2 is expressed early in the ventral retina prior to the closure of the optic cup and becomes progressively down-regulated until the completion of optic cup invagination (arrow in Figure 6B) [88], suggests that early expression followed by down-regulation of Aldh1a2 is an ancestral feature and that medaka evolved an innovative heterochronic mechanism to avoid the ventral down-regulation of aldh1a2 and to increase its ventral expression at later stages. Thus, it is likely that the dorsal and ventral paracrine sources of RA that have been suggested to regulate the development of perioptic mesenchimal derivative structures [56] is an ancestral feature that might be still preserved in teleosts. Comparative and functional analysis of the regulation of aldh1a paralogs during the development of the eye and other tissues in medaka, zebrafish and in other fishes, particularly outgroups, will be required to test this hypothesis.
The evolution of functions among Aldh1a paralogs illustrates what may be a general phenomenon associated with evolution after genome duplication: gene loss without altering developmental programs due to the preservation of functions in surviving paralogs. In our case study, the unaltered ancestral program provides both a dorsal and ventral supply of Aldh1a enzyme and hence dorsal and ventral sources of RA during retinal development. Comparative analysis shows that different paralogs can perform equivalent functions in different species. For instance, the ventral sector of the retina expresses aldh1a2 in medaka and aldh1a3 in zebrafish; and the dorsal sector of the retina expresses Aldh1a1 in mouse and aldh1a2 in zebrafish. Similar cases of what has been called function shuffling have been described for Hox genes [42]; Bmp genes [97], and Twist genes [43]. Gitelman (2007) proposed the term synfunctionalization to describe the process by which a paralog acquires the expression pattern of another paralog by gaining new regulatory elements, and thereby allowing losses of genes without changing the ancestral developmental program [43]. The acquisition of enhanced ventral expression by aldh1a2 in the face of aldh1a3 loss in medaka suggests several possible mechanisms for the apparent shuffling of functions between aldh1a3 and aldh1a2 that do not require the evolutionary gain of new regulatory elements (Figure 8B). Based on our findings, we propose a general mechanistic model to explain the loss of a paralog without altering the ancestral developmental program. After gene duplication from an ancestral gene a/b (Figure 8B Step 1), paralog b (e.g. aldh1a3) could lose the dorsal subfunction without penalty (Step 2) because it is covered by paralog a (e.g. aldh1a2). Next, mutations in negative regulatory elements or in upstream negative regulators that normally down-regulate paralog a expression in later developmental stages (e.g., after retina cup invagination) would facilitate an innovative heterochronic paralog a expression (Step 3). Finally, natural selection or genetic drift could act on natural variation that positively strengthens paralog a expression in the ventral domain (Step 3), while allowing relaxed selection for paralog b expression (Step 4), thereby facilitating the loss of paralog b (Step 5) without loss of the ancestral developmental program (Step 6).
Overall, our results illustrate how comparative genomic analyses of conserved synteny, coupled with reconstruction of ancestral chromosomes, can provide a phylogenetic framework necessary for the identification of lineage-specific gene losses. Our analysis provides evidence for early subfunctionalization and late subfunction-partitioning, and for the acquisition or modification of subfunctions by surviving paralogs that preserve unaltered ancestral developmental programs in the face of gene loss. Understanding the evolution of gene functions is fundamental for the proper interpretation of comparative analyses, especially when using model organisms to understand human gene functions. In the case of the Aldh1a family, although RA is important in human disease, we still know little about the spatio-temporal dynamics of the expression domains and functions of ALDH1A1, ALDH1A2 and ALDH1A3 genes during human development and adult organ homeostasis, other than RT-PCR studies [98], which do not provide enough resolution at the single cell level to understand how the sources of RA regulate physiological action. The evolutionary framework defined here provides information essential for the functional connectivity of human and model organism genomes, not only for RA signaling in eye development, but for the many organs in which RA plays important functions, including axis and limb development and cancer biology.
All animals were handled in strict accordance with good animal practice as defined by the relevant animal welfare bodies, and all animal work was approved by the University of Oregon Institutional Animal Care and Use Committee (A-3009-01, IACUC protocol #08-13).
Alignments of ALDH1A proteins from vertebrates and cephalochordates were generated with clustalX [99] and corrected by eye. Only the unambiguous part of the alignment was considered for phylogenetic tree reconstructions (Figure S1 provides sequence alignments). The ProtTest tool was used to choose the best-fit protein evolutionary model [100], resulting in the LG+I+G [101] and the JTT+I+G [102] as the top two selected, with a relatively low value of deltaAIC = 92.92 (AIC = 18797.45 and 18890.37, respectively). Because different phylogenetic methods have different limitations [103], we compared results from four phylogenetic approaches: i) Bayesian phylogenetic inferences were calculated with MrBayes [104], using the JTT model as well as a gamma distribution for rate variation (divided into four categories) and a proportion of invariant sites. We ran two chains for 5 million generations, sampling every 100 iterations with a 25% burn-in. ii) Maximum-likelihood (ML) analysis was conducted using PHYML [105], with an LG+I+G and JTT+I+G model. The alpha parameter of the gamma distribution (1.41) and the proportion of invariable sites (0.19) were estimated from the sample, considering four categories of substitution rates. The topology, branch lengths, and rate parameters of the tree were required to be optimized. iii) Maximum-parsimony (MP) analysis (MEGA package, [106] used the close-neighbor-interchange approach with one level of search, and added 10 replicas of random trees, and 100 replications to calculate the bootstrap value that supports each node of the tree. iv) Neighbor-joining phylogenetic (NJ) tree (MEGA package, [106] was inferred taking into account among-site rate heterogeneity with four gamma-distributed categories. This approach has been previously shown to provide equivalent results to those obtained by ML under conditions of low sequence divergence, with the advantage of a low computing-time cost [107]. The alpha parameter 1.41 was estimated from the sample using PHYML under a JTT substitution model. Concordance of trees from each of the different methods, bootstrap proportions and posterior probability estimates were used to examine the robustness of nodes. Aldh1a1/2/3 proteins predicted from gene sequence data from the cephalochordate Branchiostoma floridae were used to root the phylogenetic tree of the vertebrate Aldh1a family. Tunicate Aldh1a1/2/3 proteins were not included to avoid possible artifacts arising from long branches shown previously for Aldh genes [59].
The automatic tools developed by Catchen et al. [48] to detect synteny conservation allowed us to perform comprehensive genomic comparisons between the human genome and other fully or partially assembled genomes from a wide variety of model organisms. These automatic tools use a reciprocal best hit BLAST analysis pipeline to define groups of paralogy between a primary genome and an outgroup genome. For instance, when the human genome is compared with outgroup genomes that diverged prior the two rounds of genome duplication R1 and R2 (i.e. the urochordate Ciona intestinallis or the cephalochordate Branchiostoma floridae in Figure 2A), each human gene will belong to a group of paralogy that is anchored to a gene from the outgroup genome. Use of multiple outgroup genomes and merging clusters anchored by outgroup paralogs help to minimize errors derived from the automatic reciprocal best hit BLAST pipeline due to the effect of losses, duplications or sequence divergence of outgroup genes (for details on best hit BLAST pipeline analysis, see [48]).
Dotplots graphically represent the distribution of paralogous genes (crosses) within the primary genome (e.g. Figure 2A), or the distribution of orthologous genes between the primary and outgroup genomes (e.g. Figure 3A), using the results generated with the automatic BLAST analysis pipeline. In the case of an orthology dotplot, genes belonging to a selected chromosome in the outgroup are displayed along the x-axis of the plot in the order they appear in that genome. Orthologs of those genes are displayed on their respective chromosomes in the primary genome directly above the location of the gene on the selected chromosome in the outgroup, not in their order in the second genome. Scaled dotplots represent a variant in which the paralogs (or orthologs) of genes on the selected chromosome are displayed according to their natural chromosomal positions in the genome (e.g. Figure 2A). For instance, given an orthologous dotplot with Danio rerio as the primary genome and human as the outgroup (Figure 4A), each two paralog genes originated by R3 in Danio will be aligned above their human ortholog on the x-axis. Composite dotplots overlap multiple dotplots from the analyses of various regions of interest (crosses labeled with different colors) and different outgroup genomes (e.g. Figure 2A). Circleplots represent user-selected chromosomes as arcs along the circumference of a circle. The origins of lines connecting positions along the arcs represent pairs of paralogous genes within the same species (Figure 2D–F) or orthologous genes between two different species (Figure 3B). Gene loci that are close to each other may appear overlapped as single crosses in the dotplot or a single connecting line in circle-plots due to the selected graph resolution.
Clusters in the Synteny Database were created by coupling results from the reciprocal best hit BLAST pipeline with the use of a sliding window analysis that links chromosome segments with conserved synteny (for details see [48]). Clusters that link chromosomal segments within the same species represent paralogous syntenic conservation (e.g. Figure 2B–C), and clusters that link chromosomal segments between different species represent orthologous syntenic conservation (e.g. Figure 4E–G). The Synteny Database provides clusters produced using several different sliding window sizes measured in terms of contiguous gene number. Smaller window sizes identify tightly-conserved syntenic regions in which gene order and orientation are well preserved while larger window sizes can accommodate chromosomal rearrangements (inversions, transpositions, translocations, and small duplications). The Synteny Database is especially useful to provide evidence of ohnologs gone missing (ogm) by uncovering the putative chromosomal region that still preserves paralogous syntenic conservation, but lacks a certain ohnolog of interest (e.g. Figures 5 and 6).
Full coding sequence of aldh1a2 cDNA from Medaka Oryzias latipes (Cab strain) and the aldh1a3 cDNA from zebrafish Danio rerio were cloned after being amplified from cDNA by PCR with specific primers designed from genomic scaffold sequence data (medaka: 200506-scaffold21 and zebrafish: Zv5Scaffold1492 and NA2068) (Ola1a2F: 5′ATGACTTCCAGTAAGATCGAGATCCC3′ and Ola1a2R: 5′CATTAACGTTTCATCCATTACTGTCC3′; Dre1a3F: 5′GTCCACACAATAATCTACTCTACAGC3′; Dre1a3R 5′CATATGTTTGCGCTTAGCTGCCATG3′). Full length cDNA sequences were submitted to GenBank (medaka aldh1a2: FJ516380, and zebrafish aldh1a3: DQ300198). A zebrafish adh1a2 clone [86], a clone containing a zebrafish aldh1a3 800 nt-fragment from exon 7 to exon 13 (cloning primers: 5′GGAGCTGCGATCGCTGGTCACATG3′ and 5′CTGAGTTTGATAGTGATGGCTTTGAC3′), and a clone containing a medaka aldh1a2 527-nt fragment from exon 12 (cloning primers: 5′GGAGGATACAAAATGTCTGGGAATGG3′) to the 3′UTR (5′CATTAACGTTTCATCCATTACTGTCC3′) were used to synthesize riboprobes for whole-mount in situ hybridization using standard procedures [108],[109], with slight variations: NBT and BCIP were used instead of the BM purple.
|
10.1371/journal.pcbi.1003319 | Cell-Based Multi-Parametric Model of Cleft Progression during Submandibular Salivary Gland Branching Morphogenesis | Cleft formation during submandibular salivary gland branching morphogenesis is the critical step initiating the growth and development of the complex adult organ. Previous experimental studies indicated requirements for several epithelial cellular processes, such as proliferation, migration, cell-cell adhesion, cell-extracellular matrix (matrix) adhesion, and cellular contraction in cleft formation; however, the relative contribution of each of these processes is not fully understood since it is not possible to experimentally manipulate each factor independently. We present here a comprehensive analysis of several cellular parameters regulating cleft progression during branching morphogenesis in the epithelial tissue of an early embryonic salivary gland at a local scale using an on lattice Monte-Carlo simulation model, the Glazier-Graner-Hogeweg model. We utilized measurements from time-lapse images of mouse submandibular gland organ explants to construct a temporally and spatially relevant cell-based 2D model. Our model simulates the effect of cellular proliferation, actomyosin contractility, cell-cell and cell-matrix adhesions on cleft progression, and it was used to test specific hypotheses regarding the function of these parameters in branching morphogenesis. We use innovative features capturing several aspects of cleft morphology and quantitatively analyze clefts formed during functional modification of the cellular parameters. Our simulations predict that a low epithelial mitosis rate and moderate level of actomyosin contractility in the cleft cells promote cleft progression. Raising or lowering levels of contractility and mitosis rate resulted in non-progressive clefts. We also show that lowered cell-cell adhesion in the cleft region and increased cleft cell-matrix adhesions are required for cleft progression. Using a classifier-based analysis, the relative importance of these four contributing cellular factors for effective cleft progression was determined as follows: cleft cell contractility, cleft region cell-cell adhesion strength, epithelial cell mitosis rate, and cell-matrix adhesion strength.
| Branching morphogenesis is a complex and dynamic embryonic process that creates the structure of many adult organs, including the salivary gland. During this process, many cellular changes occur in the epithelial cells, including changes in cell-cell adhesions, cell-extracellular matrix (matrix) adhesions, cell proliferation, and cellular contraction, resulting in formation of clefts in the epithelial cells of the organ. A comprehensive understanding of the relative contributions of these cellular processes has crucial therapeutic implications for organ regeneration and functional restoration of organ structure in diseased salivary glands. Here, we have developed a cell-based model of cleft progression and simulated cleft progression under conditions of altered cell-cell adhesions, cellular contractility, cell-matrix adhesion and cell proliferation to identify the optimum cellular conditions that cause clefts to progress. The model predicts that cleft progression requires a moderate level of cleft cell contractility, a low epithelial proliferation rate, reduced cell-cell adhesion strength in the cleft and high cell-matrix adhesion strength also in the cleft region. The results of our classification analysis demonstrate that cellular contractility in the cleft cells has a significant effect on cleft progression, followed by cell-cell adhesion strength, rate of cell proliferation, and strength of cell-matrix adhesion energies.
| Branching morphogenesis is a specific type of tissue morphogenesis that is a crucial developmental process occurring in several organs, such as the mammary glands, lungs, kidney, and salivary glands to maximize epithelial surface area for secretion or absorption of fluids and gases [1]. The process of branching morphogenesis is complex and dynamic, requiring reciprocal interactions between the epithelium and the mesenchymal cell types [2], [3]. Since many organs develop by branching morphogenesis, one strategy for a regenerative medicine-based restoration of diseased or damaged branched organs would be to reactivate the cellular and molecular mechanisms that produce these organs during development. Deciphering the coordinated mechanisms driving branching morphogenesis is therefore relevant to the basic understanding of development and may be applicable to future regenerative medicine strategies.
Submandibular salivary gland (SMG) is one of the best-characterized organ systems for the study of branching morphogenesis [4] since the embryonic organs can be grown ex vivo and manipulated genetically [5] or pharmacologically [6]–[9] and monitored using time-lapse imaging [10], [11]. The gland starts to develop at embryonic day 11 (E11) when the epithelium protrudes into the neural crest-derived mesenchyme. At E12, clefts, or indentations, initiate in the surface of the primary epithelial bud, which progress inward towards the interior of the epithelium, subdividing the primary bud into multiple buds by E13. Cleft progression is associated with proliferation of the epithelial cells causing tissue outgrowth [2]. In successive days, embryonic development continues into postnatal development with continued cleft formation and bud outgrowth together with duct formation, thereby forming a highly arborized adult structure. Cellular differentiation begins at E15, concomitant with continued branching to create functional cell types, leading to saliva secretion [3]. Since the salivary glandular structure is presumably important to facilitate its function, the question of how this ramified epithelial structure is established has been the subject of many biological studies and some recent computational modeling studies.
Analysis of the physics of complex systems has demonstrated that collective behaviors arising from ensembles of a large number of interacting components cannot be interpreted from behavioral analysis of individual components [12]. Thus, several researchers have utilized various systems biology and computational modeling approaches as tools to try and understand salivary gland morphogenesis [13]. Starting at the organ level, Lubkin's group developed a 2D model for cleft formation during early salivary gland branching morphogenesis. In this work, the epithelium and mesenchyme were both modeled as immiscible Stokes fluids, separated by an interface representing the basal lamina. Using a 2D model, they predicted that mesenchymal viscosity drives a clefting force that affects the time required for branching and that the ratio of viscosities of the epithelium to mesenchyme affects the shape of clefts [14]. In subsequent work, they developed a more complex 3D model that incorporated the mesenchyme-generated traction forces. This model predicted that these mesenchymal traction forces were sufficient to drive cleft formation [15]. Although these computational models were the first attempt in modeling complex tissue-driven forces and were able to successfully generate clefts, the cleft shape did not mimic the actual shape observed in the developing salivary glands. Additionally, the 3D model could not explain how branching morphogenesis can occur in the absence of mesenchymal cells when epithelial rudiments are grown in an artificial basement membrane together with growth factors [10], [16]–[19]. The fact that branching morphogenesis can occur without mesenchymal cells indicates that a cell-based model system that focuses on epithelial cellular processes may have utility in modeling the process of cleft formation.
Previous experimental research using ex vivo embryonic organ explants and transgenic mouse models has made possible the identification of many molecules and cellular processes required for cleft formation in the submandibular salivary gland; however an integrated model for cleft formation does not exist. Using a cell-based modeling environment we set out to incorporate as much of the experimental data as possible into a computational model. Early work indicated that actin microfilaments are required for forming clefts [20], [21]. Since actin is known to regulate cell shape, a simple model for cleft formation was proposed where localized actin contraction at the basal cell surfaces alternating with contraction at the apical surfaces in the outer monolayer of epithelial cells bends this peripheral cell layer to generate clefts. However, subsequent electron microscopy studies did not detect basal actin bundles [11]. According to recent experimental work, cleft formation can be subdivided into four fundamental steps: initiation, stabilization, progression and termination. While the events leading to cleft initiation remain unclear, recent studies indicate that cleft stabilization requires formation of cell-ECM adhesions containing active focal adhesion kinase (FAK) [7]. Initiated clefts can only progress when they have been stabilized by an inside-out integrin signaling that promotes activation of focal-adhesion protein complexes that can overcome a presumed mechanochemical barrier to progression. Cleft progression was shown to require Rho kinase I (ROCK I)-stimulated non-muscle (NM) myosin II/-mediated actomyosin contractility for basal fibronectin (FN) assembly in the cleft region and associated cell proliferation, at least part of which is stimulated by FN [6]. FN assembly induced epithelial cell proliferation, which had a major impact on cleft progression and bud outgrowth but not on cleft initiation. Explants treated with hydroxyurea, a known pharmacological S-phase inhibitor, showed a reduction of progressive clefts with no effect on number of initiated clefts as compared to vehicle control glands [6]. With time-lapse imaging studies, Kadoya and Yamashina [11] showed that clefts progress with a very subtle replacement of cell-cell adhesions with cell-ECM adhesions with very little space between the cells on each side of the cleft. They proposed that local folding of the plasma membrane near the base of the cleft produces a “shelf” containing an accumulation of actin filaments. The shelf was proposed to be the contact point between the epithelium and matrix, and the cleft progressed in the groove between the shelf and the cleft cell walls, through retraction of the groove [11]. Cleft formation was also found to be accompanied by accumulation of FN in the cleft bases and concomitant loss of adjacent E-cadherin based cell-cell junctions [5]. This conversion of cell-cell adhesions to cell-matrix adhesions was found to be regulated transcriptionally through increases in BTB (POZ) domain containing 7 (Btbd7) to activate a local epithelial-to-mesenchymal transition (EMT) found near the base of the cleft [22]. Btbd7 is assumed to assist in separating the adjacent epithelial cells, while assembled FN keeps accumulating at the newly separated cleft base cells, promoting continuous cleft progression [23].These experimental studies point to a coordinated requirement for cell proliferation, actomyosin contractility, cell-cell adhesions and cell-matrix adhesions in cleft progression.
To develop a relevant cellular level model of morphodynamic pattern formation in developing salivary glands, we used a modeling environment that specifically attempts to simulate several cellular events including mitosis or cell proliferation, actomyosin contraction, cellular organization with cell-cell interactions and cell-matrix interactions and allows independent computational manipulation of each parameter within specific cell populations. The Glazier-Graner-Hogeweg (GGH) model [24], [25] was originally developed to model cellular rearrangements as a function of inter-cellular surface energy, cell membrane fluctuations and energy between cells and their external environment [26]. The GGH model has been utilized to recapitulate cellular events during pattern formation and morphogenetic movements in several organisms and organ systems [27]–[31]. The GGH model represents each cell as an aggregation of lattice points, or pixels, in a 2D space. Each cell is assigned an energy signature denoting the probability of the cell to grow, move, adhere, and organize into different patterns. GGH thus enables cell-centered modeling to simulate changes in collective ensembles of cells within tissues to facilitate the testing of how specific cell behaviors affect a larger morphological process.
In this study, we construct a GGH model of salivary gland cleft progression using CompuCell3D (CC3D), an open-source implementation of the GGH model. We developed both a single cleft and a whole epithelial tissue model, which include GGH-based representations of cellular adhesions, cellular contractility, cell-matrix adhesions and cell proliferation within the epithelial cells that are surrounded by a simplified mesenchymal compartment. The whole tissue model demonstrated a mutual dependence of cleft progression on neighboring clefts, and the single cleft model was used to investigate the contribution of the cellular parameters to individual cleft progression. We used morphometric quantification of cleft depths from time-lapse images of ex-vivo cultured glands to create a temporally and spatially accurate model. The clefts obtained during the simulations were assessed for quality using three morphometric features – cleft depth, cleft spanning angle, and cleft tilt angle. Comparisons with ex-vivo cultured glands were generated from image data that was measured using the same features. Using the single cleft model we have been able to make the following predictions regarding the contributions of cellular parameters to branching morphogenesis: (i) cleft progression requires an intermediate level of actomyosin contractility in the cleft region, and lower contractility is more detrimental to cleft progression than higher levels of contractility, (ii) proliferation rates and location of the proliferating cells affect cleft progression such that very low proliferation rates are required and an equal number or majority of the proliferating cells should be in the outer columnar epithelial layer rather than in the inner cells, (iii) low levels of cell-cell adhesion in the cleft promote progressing clefts, and (iv) cell-matrix adhesions do not have as significant an effect on cleft progression as do cell-cell adhesions. Since it is difficult to make assessments of the relative importance of cellular factors to branching morphogenesis using experimental methods, we used ex-vivo data-sets to formulate three classes of cleft progression and used classifiers to identify the most important factors during cleft progression. Our results show that epithelial cell contractility in the cleft cells is the most influential factor during cleft progression, closely followed by mitosis rate and cell contractility.
This study involves CD1 mice and was approved by the University at Albany, SUNY IACUC under protocol numbers 09-013 and 12-013.
The GGH model is built on the energy minimization-based Ising model, using imposed fluctuations via a Monte Carlo approach [24]. The simulation space is divided into a lattice, which may be two- or three-dimensional, and cells are represented by groups of adjacent lattice points; each lattice point has an associated energy value that is assigned based on its interactions with other lattice points. Energy is also assigned to cells based on cell-cell interactions, and the sum of energies across all lattice points and cells in the simulation space is the effective energy. The energy assignment of a lattice point is based on functions representing biological behaviors or constraints, and the effective energy of the simulation can be written as a Hamiltonian equation, where each term represents the sum contribution of a particular energy function. The model is based on the assumption that the most favorable state is the lowest energy state.
To develop the 2D GGH single cleft and tissue models of cleft progression, we used the following terms:
Contact energy represents differential adhesion between model cells of different types by assigning an energy penalty to adjacent lattice points belonging to different cells. Each possible pair of cell types (τa,τb) is enumerated and assigned an energy penalty J(τa,τb), including same-type pairs. Cell types that adhere to each other are assigned a lower energy penalty; the cell type τ of a particular lattice point i is given by τσ(i), where σ is the cell ID. The contact energy penalty assigned to a pair of lattice points (i,j) is therefore given as J(τσ(i),τσ(j)). To prevent lattice points within the same cell from being assigned a contact energy penalty, this is multiplied by (1−δσ(i),σ(j)), where δ is the Kronecker delta. The term for contact energy in the Hamiltonian equation across all pairs of lattice points (i,j) is therefore given as:(1)for all neighboring lattice sites i, j.
Area (a) represents cell volume in two dimensions. It is a cell-based energy function that penalizes cells for deviating from a target size, simulating the biological tendency for cells to grow to and maintain a certain size. It has two constants, a target area A, and a strength factor λ. The term is thus:(2)for each cell σ and cell type τ
Perimeter (p) is a representation of surface area in two dimensions. Like area, it is a cell-based energy function, and it imposes an additional constraint on cell size based on the amount of plasma membrane available to a cell. It also uses two constants, a target perimeter P, and a strength factor λ. The energy term is given as:(3)for each cell σ and cell type τ
Focal point plasticity is a cell-based energy term that assigns an energy penalty for linked cells that deviate from a target length L, based on the distance between the cell centroids (l). Although it was developed to simulate actomyosin-dependent contractility, it is used in our model to simulate the effects of actomyosin contractility-dependent FN assembly. Since we are unable to represent the FN wedge as a physical structure, we reproduce its cleft-forming effects by exerting a separating effect on opposing cells of the cleft wall through FPP. Within the cleft, the target distances between opposing cells are assigned based on depth, and represent the shape constraint imposed by the FN structure. The λ value modulates the effect of focal point plasticity, and corresponds to the amount of actomyosin contractility present in the simulation. The energy term is:(4)for linked cells σ and σ', and cell type τ
The target distances that produce the characteristic shape of the cleft are assigned based on an inverse relationship with the depth; cells near the bottom of a cleft are assigned shorter target distances than the cells at the top of the cleft. This relationship was determined through examination of images of progressed clefts from time-lapse images of embryonic day 12 (E12) organ explants. Additionally, we used a simplified two-cell model to investigate the effects of FPP relative to Cell-Matrix (CM) contact energy, λ, and target distance, for constant cell-cell contact energy (CC) value = 10 to determine the values of λ to use in the model (Figure S4).
The full Hamiltonian equation for our simulation is thus given as the sum of these four equations:(5)
Energy minimization is carried out by choosing pairs of adjacent lattice points from different cells, and an attempt is made to copy the cell ID from the first point to the second. This copy attempt grows one cell, either by forcing another cell to shrink, or expanding into the medium. The effective energy is calculated before and after the change, and if the new energy is lower, the change is made permanent. However, if the resulting energy is higher, the change is only retained with some probability using a Boltzmann acceptance function, e−ΔH/T. In the context of the GGH simulation, T is a constant that controls the intrinsic motility of the cell, corresponding to the amplitude of cytoskeletally derived membrane fluctuations. Using T, we have allowed a certain amount of cell motility. Allowing some amount of energy-raising lattice-copy events is important as it prevents the model from stalling at local energy minima. A single step in the GGH model actually consists of N lattice copy attempts, where N is the total number of lattice sites in the simulation space. These attempts are carried out through a Monte Carlo simulation using modified Metropolis dynamics, designated as Monte Carlo steps (MCS) [24].
Cell proliferation in the GGH model is accomplished by dividing an existing cell into two equally sized new cells. To simulate mitotic cells, a subset of cells is instructed to grow to twice their original size and divide every 100 Monte Carlo steps (MCS), mimicking the growth and mitosis of biological cells.
Simplification: Although the GGH model is able to mimic parameters such as growth factor absorption kinetics, we have omitted these from this initial study to reduce complexity and focus on the cellular behaviors. Similarly, we have simplified the basement membrane and mesenchymal compartment, which contains nerves and blood vessels [2], [32] in addition to mesenchymal fibroblasts; surrounding the epithelium into a single compartment we call “matrix” and that is often called “medium” in GGH models. The matrix compartment is essentially represented here as a single special GGH cell that is not subjected to area and perimeter constraints. We have not included apoptosis in our model since there is currently no biological data to suggest that apoptosis is important in cleft progression.
Embryos from timed-pregnant female mice (strain CD-1, Charles River Laboratories) at embryonic day 12 (E12) (with day of plug discovery designated as E0), were used to obtain submandibular salivary gland rudiments (SMGs) following protocols approved by the University at Albany, SUNY IACUC committee (protocols 09–013 and 12-013), as reported previously [6], [7], [33], [34]. E12 SMGs that contain 1 primary bud were micro-dissected from mandible slices and cultured, as described previously. For culturing ex-vivo organs, 13 mm, 0.1 µm pore size Nucleopore Track-Etch membrane filters (Whatman) were used. The SMGs were floated on top of the filters that sit on 200 µL of 1∶1 DMEM/Ham's F12 Medium (F12) lacking phenol red (Invitrogen) in glass-bottomed 50 mm microwell dishes (MatTek Corporation). The medium was supplemented with 50 µg/mL transferrin, 150 µg/mL L-ascorbic acid, 100 U/mL penicillin, and 100 µg/mL streptomycin, to make complete DMEM/F12 medium. Brightfield images were acquired on a Nikon Eclipse TS100 microscope equipped with a Canon EOS 450D digital camera at 4X (Plan 4X/0.10 NA) magnification.
Whole-mount immunocytochemistry was performed as previously described [6], [7], [33], [34]. E12 SMGs were fixed in 4% paraformaldehye (PFA) in 1X phosphate buffered saline (1XPBS) containing 5% (w/v) sucrose for 20 min at room temperature. SYBR Green I (1∶10000, Invitrogen) was used to detect nuclei and proliferating cells were detected using phospho-Histone H3 (pHH3) antibody (1∶100, Cell Signaling Technology). Epithelium was detected using an antibody recognizing E-cadherin (1∶250, BD Biosciences), F-actin was detected using Alexa Fluor 546 Phalloidin (Invitrogen, 1∶350), and mesenchyme was detected using an antibody recognizing PDGF receptor (R)-β (1∶100, Epitomics). Appropriate cyanine dye-conjugated AffiniPureF (ab′)2 fragments were used as secondary antibodies (Jackson ImmunoResearch Laboratories, 1∶100). SMGs were imaged on a Zeiss LSM510 confocal microscope at 20X (Plan Apo/0.75 NA), or 63X (Plan Apo/1.4 NA) magnification.
E12 SMG organ explants were treated with 200 µl of Hank's balanced salt solution (HBSS lacking Ca2+ or Mg2+, Life Technologies) containing 0.4% (v/v) dispase (Life Technologies) for 25 min at 37°C, and the mesenchyme was physically removed by microdissection, as described in [10]. The epithelial rudiment was cultured in a final concentration of 6 mg/mL Matrigel (BD Biosciences) diluted in DMEM/F12 containing 20 ng/mL EGF and 200 ng/mL FGF7 (R&D Systems). The gland was imaged using time-lapse microscopy with a 20X objective lens using a Zeiss 510 Meta Confocal microscope. 120 images were captured as 5 µm sections at 10 minute intervals for a 20 hour time period using the MultiTime macro. The 543 nm laser was used to capture a near-DIC image. Images were captured at a 512×512 pixel resolution using a scan speed of 9 in line averaging mode. A total of 30 glands were imaged for 20 hours in three separate sets and 40 clefts were measured using image analysis software ImageJ [35]. The first frame and the last frame (after 20 hours) were used to measure the depth in pixels for each cleft and according to the scale, the distances were converted to micrometers (µm).
To enhance the contrast of the grey-scale pHH3 images and the SYBR green images, we applied the contrast-limited adaptive histogram equalization algorithm (CLAHE) [36] to the image. The CLAHE algorithm considers the image as a collection of smaller regions and applies histogram equalization on these regions. The objective of histogram equalization is to transform the image so that the intensity histogram of the output image approximately matches a specified histogram; in our case we use a curved histogram. The CLAHE algorithm evens out the distribution of used grey values and thus makes hidden features of the image more visible. Noisy regions of the images are removed by considering regions of intensity greater than a pre-determined threshold. For the E-cadherin marker images, we applied a Gaussian smoothing followed by the CLAHE algorithm, and then removed noisy regions based on a predetermined threshold. Binary masks were created for the SYBR green and pHH3 histogram equalized images by applying an OR operation on the histogram equalized image and the E-cadherin marker. The total area of the connected components in both images was calculated, and the ratio yielded the percentage of SYBR green-positive cells (total cells) that are in mitosis, or M phase, of the cell cycle.
Four values for mitosis rate (MR), six values for contractility (FPP λ), five values for cleft region adhesion (CC), and five values for cleft-matrix adhesion (CM) were chosen from the hypothesis driven individual analyses, and 40 simulations were run for each of the 600 possible combinations. Cleft simulations were classified as failed (less than 17.8 µm), non-progressive (17.8 to 30.5 µm), progressive (30.5 to 40.7 µm), and super-progressive (greater than 40.7 µm) based on minimum, first quartile, and third quartile depths of ex-vivo cleft measurements. Parameter combinations were assigned an overall class based on the cleft depths attained in a majority class within the 40 runs; in the case of a tie, the median depth was used to classify the parameter combination. This resulted in 275 failed, 188 non-progressives, 85 progressive, and 52 super-progressive results. To determine the importance of each GGH parameter in cleft progression, we formulated the problem as a supervised learning feature selection task, with each combination as a data point and the parameter values as features. Samples were created by drawing 50 random points from each class. For each of the 15 possible combinations of the four features, a 10-fold cross-validation using a radial basis kernel support vector machine (SVM) was performed on the sample, reporting the training and testing accuracies [37]. A greater decrease in classification accuracy corresponds to a more important feature. Additionally, analysis of the parameters resulting in progressive clefts was performed to confirm the importance of each parameter; parameters that were essential to progressive clefts were expected to be distributed around a particular value with low variance.
We chose to start our model at E12, when the mouse SMG undergoes the first round of branching morphogenesis. At E12, the gland is a single epithelial mass, or bud, atop a stalk, surrounded by a condensed mesenchyme (Figure 1A, 1B). Clefts initiate as indentations in the epithelium, which progressively furrow interiorly. Since cleft initiation and cleft progression are biochemically independent steps [6] and little biological information is available regarding mechanisms of cleft initiation, we chose to pre-specify an individual initiated cleft in the model and simulate only the stage of cleft stabilization and cleft progression (Figure 1E). At E12, the epithelium expresses E-cadherin (Figure 1C,1D) but later stage differentiation marker proteins are not yet expressed [38], [39]. We therefore assumed that the cell-cell adhesions present are E-cadherin-containing adherens junctions with an absence of tight junctions, as previously reported [38], [39]. The epithelium is surrounded by mesenchyme that expresses PDGFR-β, which can be used to distinguish the latter from the former (Figure 1C, 1D). Closely associated with the epithelial cells is the basement membrane, a specialized extracellular matrix (ECM) that forms a boundary between the epithelial and mesenchymal tissue compartments [5], [40]. Since we are focusing on epithelial cell parameters that control cleft progression, we modeled the basement membrane and the entire mesenchyme compartment as a simplified single cell, designated as “matrix,” which lacks area and perimeter constraints.
At E12, there are two structurally distinct epithelial precursor cell populations [34], [38]. The outer columnar cells (OCCs) that contact the basement membrane surround a cluster of less organized inner polymorphic cells (IPCs) (Figure 2A), and this cell arrangement is maintained during 24 hours of ex-vivo culture (Figure 2B). The 6×6 pixel square cells were arranged in a homogenous grid, a simplification that approximates the initial cell distribution with OCCs labeled in dark green and IPCs in light green (Figures 3A, 3B). To calibrate the model with image data, we performed time-lapse imaging of multiple E12 mesenchyme-free SMG organ explants for 20 hours and measured the length of the resulting clefts (Figures 3C, 3D, Video S1). Clefts achieved an average depth of 36.2 µm and a median depth of 35 µm. Based on the cleft depths obtained from the time-lapse analysis, we defined normal cleft depth in the CC3D model as 36 pixels, using 6 cells per cleft, shown in light and deep blue (Figures 3A, 3B). To distinguish OCCs from IPCs, we use a baseline perimeter equivalent to the perimeter of a square for the initial cell area. Relative to this baseline, we allow a marginal increase in the target perimeter for IPCs, which encourages them to take on more irregular shapes, whereas OCCs were confined to a smaller perimeter, encouraging them to maintain a more ordered columnar shape as they do in-vivo.
Cells exhibit differential adhesion that can drive complex tissue-level behavior [30]. The IPCs demonstrated a slightly more diffuse distribution of the adherens junction protein E-cadherin than the OCCs, suggestive of reduced adherence of the IPCs to each other [38]. To represent cell-cell adhesions in the GGH model, we start with a baseline contact energy penalty; increasing or decreasing the penalty simulates lower and higher adhesion, respectively, as explained by Eq. 1. Relative to this baseline, we designated increased cell-cell contact energy between IPCs to represent decreased adhesion properties and decreased contact energy between OCCs, simulating a possible increased adhesion that may help OCCs maintain their regular shape. During cleft progression, contact energy between the OCCs representing the cleft walls is directed to increase relative to the baseline, while contact energy between cleft cells and the matrix is decreased. This decrease in contact energy allows cell-matrix contacts to be established between the cleft cells.
The basement membrane is a dynamic structure that plays a critical role in branching morphogenesis, and cell-matrix adhesions are known to change dynamically during branching morphogenesis [5], [6], [8], [10]. In the GGH model, we represent basement membrane through the contact energy settings between the OCCs and the matrix, which is represented as a single homogenous cell not subject to area and perimeter constraints. This contact energy is designated in our model as the “cell-matrix” contact energy and behaves as defined by Eq. 1.
The actin cytoskeleton has long been known to be required for branching morphogenesis and was specifically shown to be required to maintain initiated clefts [20], [21]. In salivary gland epithelial cells, the actin cytoskeleton is organized primarily into cortical actin filaments at the cell perimeter (Figure 2C, 2D) in an E12 organ explant grown ex vivo for 0 or 24 hours. Our subsequent work indicated that actin and non-muscle (NM) myosin II–mediated contraction are required to regulate cleft progression [6]. The current model for cleft progression assumes that actomyosin contraction is required for assembling fibronectin through integrin activation [5], [6], [7], which then stimulates local EMT through upregulation of Btbd7 and Slug and reduction of E-cadherin levels [22]. Since EMT is one of the chief factors promoting cleft progression, we utilized variable cell-cell and cell-matrix contact energies to facilitate cleft progression. Without any other energy factors affecting cleft progression, the resultant clefts were poorly formed (Video S2).
During early cleft formation, the cleft evolves as a thin opening between OCC cells, possibly primarily aided by random cell movements [5], [11] and possibly from a hypothesized force generated by FN assembly [10] pushing assembled basement membrane into the cleft opening. FN assembly, dependent on strength of actin contractility for integrin activation, might cause the two cleft-forming epithelial cell layers to separate. FN assembly also stimulates proliferation [6], presumably causing an outward force that emanates from inside the bud to counteract an inward cleft movement force produced by FN.
Since our model lacks specific structural representation of basement membrane assembly dynamics, we could not simulate the FN generated “cleft forming force” which was hypothesized to be the primary cause for progressive clefts [10]. Therefore, we attempted to simulate the effect of this FN-actomyosin dependent “cleft forming” force through an energy function called focal point plasticity (FPP). This function establishes links between selected cells and regulates the distance between them, assigning an energy penalty for deviating from a target distance. As noted in Eq. 4, the penalty varies based on the target distance, and the λ term. To replicate the wedge-shaped cells in the cleft, we paired opposite cells on each side of the cleft, and set decreasing target distances for pairs deeper within the cleft. These target distances were determined by examining cleft depths from ex-vivo time-lapse images and measuring cleft width as a function of depth (Figure S1). We found that a target distance inversely proportional to the cleft depth approximated the observed shape. Modulating the λ term adjusts the strength of this cleft-opening/maintaining effect. Due to the fundamental role of actomyosin contractility in FN assembly, it can be viewed as modulating contractility levels within the cleft cells.
In case of cleft progression, the exact roles for actin contractility in force generation during progression is unknown and although phosphorylated NM-myosin II was detected in the OCCs [6], it is not known if OCCs contract by pulling on each other through the actomyosin bundles. So, we utilized FPPλ to assign lateral FPP links in the OCC layer between adjacent cells and additional vertical links and lateral links between cleft cells (Figures 3A, 3B). These lateral links in the OCCs helped control the shape of the boundary cells along with maintaining a constant epithelial boundary. We then utilized the lateral FPP links in the cleft cells to simulate the effect of this actomyosin dependent FN “wedge.” The varying target distances in the cleft region are manipulated dynamically to simulate the effects of a “clefting force” generated by continuous actomyosin-mediated FN assembly between the cleft cells as the cleft progresses inward.
Previous work shows cell proliferation to be dispensable for cleft initiation [41], but to be required for cleft progression [6]. Although cytoskeletal contraction can induce cell proliferation [42], [43], in the CC3D environment, cell proliferation can be regulated separately from cell contractility. In the model we designated not only the percentages of mitotic cells but also their location within each epithelial cell subtype.
We ran initial simulations for an extended number of MCS steps to determine the range of MCS steps corresponding to the time frame encompassing cleft initiation through progression (Figure 1E). A termination value of 1500 MCS steps was selected, equating to a temporal conversion of 1 MCS≈48 seconds (Figure 3E, 3F). Figures 3A and 3B show the model at time 0 hrs (0 MCS) and time 20 hrs (1500 MCS), respectively.
Within the CC3D environment, we established a set of base values for the five primary epithelial parameters included in this computational model under which cleft progression could occur (Table 1). To conduct a parametric search, we fixed the temperature (T) at 10. Due to its central role in the energy minimization step, modification of T impacts every other energy-based parameter. We vary T and select a fixed value that permits cells to fluctuate fluidly without becoming fragmented [27], consistent with previous observations that epithelial cells undergo dynamic movements during branching morphogenesis [10], [23]. This simulates a basal level of cell migration in both OCC and IPC epithelial cells. Interestingly, the random cell movement observed produces some exchange of cells between the OCC and IPC layer. With T at 10, we conducted a parametric search on these parameters: focal point plasticity (FFP λ), mitosis rate, mitosis location, cell-cell contact energy, and cell-matrix contact energies.
To yield a final cleft depth of 36 pixels in 1500 MCS (Video S3), we fixed these base values for the five parameters: Mitosis rate was set to 1% (per 100 MCS steps), evenly divided within OCCs and IPCs; FPP λ values in the OCCs and cleft cells was set at 10; cell-cell contact energy was set to 10 for cleft cells and 5 for all other cells; and cell-matrix adhesion in cleft cells was set to 3. Under these parameters, our model achieved an average cleft depth of 34.1 pixels, thereby yielding a spatial conversion of 1 µm = 1.06 pixels. Each simulation was run 100 times to ensure the consistency of the results given the stochastic nature of the GGH model. Figure 3F shows an example of the temporal evolution of cleft depths, achieving a 34.1 µm depth in 1500 MCS. With T value fixed at 10, we tracked 1725 individual cells in the base case simulation for 10 runs. The average net displacement was found to be 7.3 µm and the total path length was 94.6 µm (Figure S2). Thus the cell velocity was calculated to be 4.7 µm/hour and the meandering index to be 0.08 in the base model.
For quantitative and consistent methods to measure the quality of simulated clefts by comparison with equivalent measurements from organ explants, we developed descriptive cleft measurement indices – cleft depth, spanning angle, and tilt angle. First, the cleft center was located at the epithelial-mesenchymal boundary by examining the angle formed by each boundary point and its 8-distance neighbor on either side. As the deepest point of the cleft, the cleft center should have the lowest such angle value. The extrema are identified by using the mean-squared error (MSE) of the best-fit line for the boundary on each side of the cleft center; for each side, we progressively include points from the boundary until the MSE exceeds a predetermined threshold. The cleft center and extrema are shown in the example image in Figure 4A, 4D. Cleft depth is calculated as the distance from the cleft center to the midpoint of the line segment joining the two extrema (Figure 4B, 4E). Spanning angle is calculated as the angle formed by the line segments joining the cleft center to each extrema (Figure 4C, 4F). Clefts measuring less than 5 pixels in depth or exceeding 160° in spanning angle were discarded. The tilt angle is a measure of the perpendicularity of a cleft to the bud surface. It is calculated as the smaller of the complementary angles formed by the line segment between the extrema, and the line segment from the cleft center to the midpoint of the line segment joining the two extrema, as shown in Figure S3. Clefts with a tilt angle of less than 45° were labeled as “failed clefts”. The cleft categorization criteria were based on measured properties of clefts from multiple time-lapse images of organ explants.
In a developing salivary gland, multiple clefts form on the surface of the primary bud during branching morphogenesis, and they do not all form simultaneously. To determine if the progression of one cleft has an effect on adjacent clefts, we constructed a GGH-based salivary gland organ model consisting of a single bud with three equally-spaced clefts that progress simultaneously (Figure 5A) We ran 70 independent simulations with the same base case parameters that were used for the single cleft model, each for 1500 MCS (Figures 5A, 5B, Video S4). Quantitative analyses show that each individual cleft is comparable to those produced by the single local cleft model, albeit the average final cleft depth is slightly lower at (Figure 5C) 29.7 µm rather than 34.1 µm for the single cleft model. Correspondingly, marginally higher spanning angle values were observed compared to the base case. This result interestingly predicts that the behavior of clefts is somewhat dependent upon adjacent clefts. However, to focus on the cellular parameters necessary for progression of a single cleft, we used the single cleft model in all subsequent studies.
Having built a cellular model replicating cleft progression, we ran simulations comprising combinatorial variations of two parameters to simulate a specific biological state. As cleft progression requires ROCKI signaling, which stimulates both actomyosin contractility and proliferation [6], we simulated these cellular conditions by reducing the lateral FPP λ values in the cleft cells from 10 to 1 to simulate reduced cellular contractility and correspondingly lowered the mitosis rates in all cells from 1% to 0.5%. We performed 100 simulations and quantified cleft quality using the three cleft measurement indices. As shown in Figure 6, the cleft depths were reduced by 40.8% in simulations and 94.7% in ex-vivo studies using 10 µM Y27632 treatments for 24 hours (Figure 6, Video S5). We also simulated the effects of blebbistatin, a pharmacological inhibitor that prevents high affinity interactions between actin and myosin to inhibit cleft progression but does not affect cell proliferation [6] using an FPP λ value of 1 in the cleft cells but without changing cell proliferation rate. We observed similar trends in the reduction of cleft depths: a 48% reduction with 20 hours of in silico simulation in comparison to 88.9% reduction in organ explants treated for 24 hours with 25 µM blebbistatin (Figure 6, Video S6). Interestingly, the computational model agrees qualitatively with the experimental data that cell contractility and mitosis affect cleft progression.
Use of the single cleft model to test hypotheses regarding the mechanisms of cleft progression-
Cell proliferation has long been understood to occur during branching morphogenesis. An early study indicated that cell proliferation was not required for salivary gland cleft formation [41], but later work demonstrated that although cleft initiation does not require cell proliferation, the biochemically independent step of cleft progression does require cell proliferation [6]. However, it has not been possible to experimentally increase the cell proliferation rate without affecting other cellular parameters. To test the sub-hypothesis that high cell proliferation rates are required for cleft progression, we performed simulations in which we tested increasing amounts of cell proliferation by varying the value for the GGH parameter, mitosis rate (MR). For this in silico experiment, we chose to assign the dividing cells equally in the OCC and IPC epithelial cells. We ran 100 simulations for 5 different values of MR, from 0.5% to 5%. Surprisingly, we found that high MR levels were inhibitory for cleft progression and that the best conditions for promotion of cleft progression were at a MR of 1%, where the cleft depth was the highest and the spanning angle was the lowest (Figures 7A, 7B). To experimentally validate the prediction of the single cleft model that 1% cell proliferation is ideal for cleft progression, we grew organ explants for 24 hours, and fixed a subset of tissues for immunocytochemistry with pHH3 to detect cells in M phase and staining with SYBR green to detect total nuclei at 2, 8, 12, and 24 hrs. Mitotic cells were detected in both the OCC and IPC layers, and the percentages of dividing cells were calculated from single confocal images for each tissue compartment (Figure 2E, 2F). Although the mitosis rate varied over the time period of the assay, the average mitosis rate was calculated to be 0.99% in the epithelial region (Figure 7C), as predicted by the single cleft model.
It has not been experimentally possible to manipulate cell proliferation rates in specific regions of the gland; therefore, it is not known if there is a regional preference for cell proliferation within the epithelium during cleft progression. Using the single cleft model, it is possible to test the sub-hypothesis that the OCCs proliferate preferentially over the IPCs. We performed simulations where we both varied the epithelial location of the proliferating cells and also varied MR. MR was set at 0.5%, 1%, 2%, 3%, or 5%, with the location for mitotic cells designated as 25%, 50%, or 75% in the OCC population (Figure 7). When 25% of the proliferating cells were located in the OCCs, we found that increasing the mitosis rate from 0.5% to 1% caused a minor decrease in cleft depth (32 µm to 31.2 µm) (Figure 7A). Further increases in MR to 2%, 3% and 5% decreased cleft depths to 25.7 µm, 15.2 µm, and 8.7 µm, respectively. Interestingly, when 50% or 75% of the mitotic cells were located in the OCC region, by increasing MR from 0.5% to 1%, a slight increase in cleft depth from 32.3 µm to 34.1 µm was observed; but further increases in rates to 2%, 3%, and 5% caused progressive decrease in depths, irrespective of the location of the dividing cells. The trends in spanning angle are shown in Figure 7B. Thus, 1% MR with 50–75% of the cells located in the OCC region was found to be the optimal condition for cleft progression.
In previous work, we demonstrated that ROCKI is required for cleft progression through modulation of actomyosin contractility [6]. ROCKI was required for phosphorylation of NM myosin II to stimulate contractility and down-regulation of cellular contractility with blebbistatin similarly reduced cleft progression. Pharmacological inhibitors cause a global reduction of actomyosin contractility, making it impossible to assess the effect of cellular contractility specifically in the cleft cells. Actin-based contractility is responsible for dynamic cell movements and FN assembly by the cleft cells, and so we modulated the strength of the lateral FPPλ links in the cleft cells to test the hypothesis that actomyosin contractility in the cleft cells is required for cleft progression. In the cleft region, we assigned vertical FPP links between cleft cells and also lateral links between the adjacent cleft cells on either side of the cleft. Also, each pair of cleft cells was assigned a different target distance, with a lower target distance being set for cells deep within the cleft. During the course of simulation, each pair of cleft cells strives to acquire the set target distance, and after 1500 MCS, under unaltered cell and matrix contact energy settings, FPP energy is solely dictated by λ values since the distance deviation for each pair of cleft cells remain almost constant for varying λ values. Thus, by altering the lateral FPPλ values in the cleft cells, we aimed to vary actomyosin-based cellular contractility specifically in these cells to assess its effect on cleft progression.
To test the hypothesis that contractility in cleft cells is required for cleft progression, we varied the lateral FPP λ values that hold the cleft cells together, between 1 and 30. A cleft depth of 34.1 µm and a spanning angle of 46.0° was found for FPP λ = 10, whereas lowering FPP λ values to 5 or 1 caused the average cleft depth to decrease to 29.0 and 19.5 µm respectively (Figure 8A) with associated increases in spanning angle (Figure 8B). This manipulation mimicked the effect of decreasing actomyosin contractility, as performed experimentally using blebbistatin. Interestingly, when we used higher values of FPP λ such as 15, 20, and 30, progressively lower cleft depths (33.5, 31.6, and 26.0 µm) and higher spanning angles (47.3, 49.3, and 71.5°) were observed, which has not been experimentally tested. When no FPP links were used, shallow clefts were formed with an average cleft depth of 12.6+/−3.94 µm and spanning angle of 99.9° (Video S2). The single cleft model thus predicts that actomyosin contractility in the cleft is essential for cleft progression and that a moderate level of this cellular contractility favors cleft progression, with low contractility being more detrimental to cleft progression than high contractility.
It was previously reported that loss of E-cadherin-containing cleft-cell adhesions and gain of fibronectin-driven cell-matrix adhesions within the cleft region are required for cleft progression, and further that an epithelial-mesenchymal transition occurs in a subset of cells at the base of the cleft to facilitate cleft progression [5], [22]. Although experimental manipulations have been performed to examine the effect of decreasing E-cadherin-based cleft-cell adhesions [10], [38] and to increase or decrease cell-matrix adhesions [5], [6] these manipulations have been performed with whole organ explants or with whole epithelial rudiments grown in an artificial matrix and the requirement for these changes specifically within the cleft region has not been possible to address. With the single cleft model, it is possible to manipulate cleft-cell and cell-matrix adhesion strengths within a subset of cells in the cleft region by running simulations at multiple values.
To recapitulate a progressive EMT occurring in the progressing cleft, we assigned a low cell-matrix adhesion value of 3 and a higher cell-cell adhesion value of 10 in the cleft cells forming the cleft wall, whereas all other cell-cell contact energies remained at 5. These values are assigned at the onset of simulation, and in each pair of cleft cells, cell-cell adhesions are replaced by epithelial-matrix adhesions during the temporal progress of cleft deepening. In order to preserve the epithelial organization of the OCC and IPC relative to the matrix, we assign a lower energy penalty between OCCs and a higher penalty for IPC-matrix. When cleft-cell adhesion was decreased (raised contact energy), cleft depths increased beyond 34.1 µm to 38.4 (CC value = 15) and 40.7 µm (CC value = 20) (Figure 8C) with corresponding spanning angle measurements that decreased from a base value of 46.0° to 39.3° and 35.6° (Figure 8D), respectively, following 100 simulations. Interestingly, increasing cleft-cell adhesions (lowering contact energy values to 5 and 1) caused shallower clefts with depths of 24.7 µm and 19.2 µm and increased spanning angles to 76.2° (C-C value = 5) and 92.4° (C-C value = 1). Thus, the single cleft model predicts that low cell-cell adhesion strengths within the cleft are most beneficial for cleft progression.
It has also not been possible to determine experimentally whether it is the loss in cleft-cell adhesions or the increase in cell-matrix adhesions that occurs in progressive clefts that has the most significant effect on cleft progression. In the single cleft model, we varied the cleft-matrix contact energy from 1 to 5 and ran 100 simulations with each value. Unexpectedly, variations in the cell-matrix contact energy values had minimal effects on cleft progression. Higher cleft cell-matrix contact energy values (lowered adhesion) yielded slightly shallower clefts (30.8 µm for CM = 4 and 27.3 µm for CM = 5) than the base value of 3 (Figure 8E), with corresponding changes in spanning angles (Figure 8F). Lower cleft cell-matrix contact energies (increased adhesion) yielded marginally deeper clefts (36.1 µm for value = 2 and 35.1 µm for CM = 1), with corresponding changes in spanning angles, than obtained with the base value of 3. Thus, the single cleft model predicts that for efficient cleft progression in the cleft region, a low cell-cell adhesion value is required more so than specific levels of cell-matrix adhesion; however, higher cell-matrix adhesion levels are somewhat beneficial for cleft progression.
Although multiple studies have been performed to assess the importance of individual cellular factors in the process of branching morphogenesis, it is not possible to rank the importance of these cellular factors using experimental methods alone. Using the single cleft simulation model, we varied each of the four parameters independently for a total of 600 parameter combinations. Each parameter combination was simulated 40 times, and classified into one of four categories based on the majority result. These cleft classes were designated based on the distribution of measurements derived from time-lapse data, and labeled “failed,” “non-progressive,” “progressive,” and “super-progressive.” Failed (F) clefts did not stabilize and regressed back to the epithelial boundary and non-progressive (NP) clefts stabilized but failed to progress. Progressive (P) clefts fall within the normal size range of clefts measured from time-lapse data, whereas super-progressive (SP) clefts exceed the average size. The depths that each class corresponds to are shown in Figure 9A. Out of the 600 parameter combinations, we obtained 275 F, 188 NP, 85 P, and 52 SP combinations in each of the cleft classes.
The number of stabilized versus progressive clefts, for each parameter variation have been outlined in Figure S5 and in Tables S1 and S2. The proportion of F, NP, and P clefts defined the limits of our parametric search for each hypothesis. For instance, the range of FPP λ values was chosen after assessing the number of progressive clefts (P) obtained from each variation. Figure S6B shows that at λ value 0.5, no P was obtained. Hence the lower limit for FPP λ variation was set to 1. The number of P decreased with increasing contractility and a higher value of 30 was set as the upper limit of the range. Also from Table S1, the number of F and NP increased highly at λ value 30, and beyond 30 there were almost no P, with mostly destabilized failed clefts.
To assess the relative importance of each GGH parameter, we measured how accurately a classifier could predict the cleft class of a parameter configuration, with the expectation that parameters that have a high importance in cleft progression should also serve as good predictors of cleft class (simulation outcome). Conversely, when removed as a feature, the absence of such a parameter should have a strong negative impact on the classification accuracy. Using a radial basis kernel SVM classifier, we were able to achieve 75.6% accuracy when the classifier was provided with all four GGH parameters as features. We then attempted classification using the remaining 14 possible combinations of three, two, or one parameter. The cross-validated training and testing accuracies and decrease in accuracy relative to the full parameter set are reported in Table 2. The testing accuracy for combinations where a single parameter was removed is also shown in Figure 9B. Individual removal of the three parameters MR, FPP λ, and CC resulted in similar drops in classification accuracy of 15.3%, 18.5%, and 15.6%, respectively. The removal of cellular contractility (FPP λ) had a marginally higher impact on the classification accuracy than cell-cell adhesion (CC) and mitosis rate (MR). In contrast, omitting CM resulted in a classification accuracy of 75.0%, which is only a 0.6% decrease from the 75.6% accuracy level obtained with all four features, suggesting that the contribution of cell-matrix adhesions are the least significant contributor to cleft progression in our model. Thus, our analysis that considers the drop in classification accuracy as the metric for importance suggests that FPP λ is the most significant contributor to cleft progression in our model, followed closely by CC and MR.
After classifying each parameter set into the four cleft types, we questioned which specific parameter values are important for achieving clefts that fall into the data-driven normal cleft range, described as “progressive” clefts (30.5–40.7 µm ∼85 out of the 600 simulations). The distributions of FPP, CC, MR and CM values of all clefts falling into the progressive class are shown in Figure 9C. The results indicate that the conditions to form “progressive” clefts vary slightly from the conditions required to form clefts, in general: (i) the optimal FPP λ value was 5, rather than 10 (base case of 34.1 µm cleft depth), indicating that a slightly lower value of cleft cell contractility is sufficient for progressive clefts. (ii) CC values peaked at 15 and 20, showing that a lower level of epithelial adhesion favors cleft progression in the “progressive class” than in the base case category where the deepest clefts peaked at a value of 10. (iii) CM adhesion values did not peak particularly at any value, but a lower value (∼1) promoted cleft progression, whereas a value of 3 was optimal for the base case. The optimal MR was similar to that predicted by the base case cleft categorization. This analysis also indicated that CM had the lowest impact on cleft progression.
We describe here the first cell-based model of salivary gland branching morphogenesis, which is able to recapitulate many crucial epithelial cell behaviors and make predictions regarding the manipulation of these behaviors on the outcome of the tissue structure, thus spanning two biological scales. Since organ formation is a complex process that encompasses several conserved molecular, cellular, and genetic mechanisms that cooperatively aid in the formation of tissue structure, it is difficult with experimental manipulations alone to identify critical factors contributing to the overall morphogenetic process. Using the single cleft GGH-based model, we were able to assess the relative quantitative importance of various cellular parameters in the process of cleft progression and found that cleft cell contractility was comparatively the most significant cellular contributor to cleft progression, followed closely by cell-cell adhesion and mitosis rate, with cell-matrix adhesions showing less significant contributions to cleft progression. It is particularly significant that actomyosin contraction, the biological effects of which are closely mimicked by the GGH term, focal point plasticity (FPP λ), was the most crucial contributor to cleft progression in this model, thus supporting our prior experimental results indicating that actomyosin contractility is essential for cleft progression [6].
We mimicked the effects of actin contractility by establishing FPP links in the cleft cells both laterally and vertically. Cortical actin microfilaments run along the cell perimeter in SMG epithelial cells, and together with myosin provide tensile forces in and between the cells [42], [43]. In the GGH model, FPP links establish a similar kind of restraint to inter-cell dynamics. The λ value dictates the strength of these connections. Thus, upon varying FPP λ in the cleft cells, there was a biphasic response of cleft depth to strength of contractility. Previously, cleft progression was studied experimentally only with reduced actomyosin contractility, which resulted in non-progressive clefts [6]. The effect of increased contractility on submandibular gland branching morphogenesis has not yet been studied, but in lung morphogenesis, a general Rho activator caused a biphasic response to branching [44]. Although the highest dose of the Rho activator increased cellular contractility, it decreased the number of buds, in support of the idea that the effect of contractility on cleft progression is biphasic.
Modulation of the distribution of proliferating cells is not easily addressed experimentally. Upon varying proliferation rates and locations, our in silico results indicate that a low rate of cell division is conducive for cleft progression and the dividing epithelial cells should either be equally divided between the OCCs and the IPCs or located primarily in the outer cells. Increasing cell proliferation levels, irrespective of the location of the dividing cells, caused a decrease in cleft depth. These results apparently conflict with experimental studies demonstrating that growth factors such as FGF, EGF, and PDGF [8], [9], [16]–[19], [40], [45]–[47] that promote branching morphogenesis of organ explants by increasing proliferation through various complex regulatory networks. However, most of these studies did not specifically examine cleft progression per se, but found that proliferation is generally important for increased bud formation, ductal outgrowth, and regulating expansion/maintenance of progenitor cell populations. Our model does not address ductal outgrowth or specific progenitor cell populations, so modeling of these more complex events requires the addition of more complexity to our model [48], [49].
We developed the salivary gland cleft model to be as realistic as possible given the current limitations of the GGH modeling environment. The model was built with 6×6 pixel cells that were structured on a regular lattice, organized as two epithelial cell layers - the outer and the inner cells along with pre-ordained cleft cells in the outer layer. An oversimplification of the model is that the cell shapes are not accurately represented, as the shapes of epithelial cells are known to be irregular during early development [10], [22]. Since cell placement and cell shape changes are essential components of multiple developmental processes [33], [34], [50], it will be informative in future studies to utilize cell-graphs as a quantitative tool to define the accurate placement of cells into a GGH model, thus allowing us to accurately model the cell shape changes in correlation with experimental cellular events. The previously reported dynamic, irregular shape of cells in developing salivary glands likely relates to their movement during early development [10]. This movement may be facilitated by weak E-cadherin-mediated cell adhesions. In our local cleft model, fixing temperature (T) at 10 provided us with lower velocity, displacement and meandering index values than those previously calculated [10]. Increasing T in the GGH model allows cells to change their shape more freely; however, it does not simulate the extensive cell movements previously observed in embryonic glands using time-lapse imaging [10]. Future improvements to the model will include more accurate representation of cell shapes and more accurate modeling of the kinetics of cell motility in the epithelium.
Although loss of cleft cell-cell adhesion is closely associated with progression of cell-matrix adhesions, our feature selection results indicate that cell-matrix adhesions are not important for cleft progression, which may result from inadequate modeling of the basement membrane properties. Several biological studies have demonstrated a role for basement membrane proteins including fibronectin [5], [6], [10], collagens [45], [47], laminin α5 [9], and perlecan [19] in branching morphogenesis. Previous studies have observed collagen III to be accumulated in the narrow cleft base region [47]; thus, a model was proposed where interstitial collagen secreted by mesenchymal cells was proposed to initiate clefts that were stabilized through GAGs, resulting in accelerated proliferation. Structural representation of basement membrane components in future models will make it possible to model cleft initiation, which was not addressed in this study. Further research to combine the epithelial cellular factors along with assembly of secreted matrix proteins, needs to be conducted to so that cleft initiation, stabilization, and progression can be studied in a synchronized fashion.
The mesenchyme is complex and also contains both developing nerves and blood vessels along with fibroblastic mesenchymal and progenitor cells within an elaborate extracellular matrix. Components of the mesenchyme are important for morphogenesis as mesenchymal-epithelial interactions are required [2], [3], [32], [51]. Our recent study using cell-graphs uncovered a previously undetected rearrangement of mesenchymal cells in ROCK inhibitor-treated glands relative to untreated controls, suggesting that mesenchymal rearrangements impact branching morphogenesis [33]. Recent work has investigated the dynamics of epithelial cell progenitor populations in developing salivary glands [32], [49]. Since the GGH model is capable of modeling reciprocal interactions between multiple cellular subtypes in pattern formation and disease progression [52], [53], [54], it will be informative in future work to use GGH modeling to evaluate contributions of progenitor cell populations and epithelial cell subtypes to morphogenesis. Thus, the modeling of specific epithelial and mesenchymal cell subtypes into future modeling work will make it possible to more accurately assess the cellular mechanisms driving branching morphogenesis.
Thus, this study provides a realistic model of one of the significant events in salivary gland organ development – cleft progression. Various cellular factors that affect this morphodynamic pattern formation have been explored in detail, and biological validation has been provided wherever possible. Although manipulations of genes and protein functions using organ explants has provided insight into the molecular mechanisms driving branching morphogenesis, there are many experimental manipulations that cannot be performed with either ex-vivo explants or in-vivo organisms due to technical impossibilities or to limited resources. In silico analysis of multifactorial developmental events circumvent this disadvantage and provide us with crucial molecular clues that can be investigated using experimental biology, thus improving our understanding of the complex process of organogenesis.
|
10.1371/journal.pntd.0000849 | Biochemical Properties of a Novel Cysteine Protease of Plasmodium vivax, Vivapain-4 | Multiple cysteine proteases of malaria parasites are required for maintenance of parasite metabolic homeostasis and egress from the host erythrocyte. In Plasmodium falciparum these proteases appear to mediate the processing of hemoglobin and aspartic proteases (plasmepsins) in the acidic food vacuole and the hydrolysis of erythrocyte structural proteins at neutral pH. Two cysteine proteases, vivapain (VX)-2 and VX-3 have been characterized in P. vivax, but comprehensive studies of P. vivax cysteine proteases remain elusive.
We characterized a novel cysteine protease of P. vivax, VX-4, of which orthologs appears to have evolved differentially in primate plasmodia with strong cladistic affinity toward those of rodent Plasmodium. Recombinant VX-4 demonstrated dual substrate specificity depending on the surrounding micro-environmental pH. Its hydrolyzing activity against benzyloxycarbonyl-Leu-Arg-4-methyl-coumaryl-7-amide (Z-Leu-Arg-MCA) and Z-Phe-Arg-MCA was highest at acidic pH (5.5), whereas that against Z-Arg-Arg-MCA was maximal at neutral pH (6.5–7.5). VX-4 preferred positively charged amino acids and Gln at the P1 position, with less strict specificity at P3 and P4. P2 preferences depended on pH (Leu at pH 5.5 and Arg at pH 7.5). Three amino acids that delineate the S2 pocket were substituted in VX-4 compared to VX-2 and VX-3 (Ala90, Gly157 and Glu180). Replacement of Glu180 abolished activity against Z-Arg-Arg-MCA at neutral pH, indicating the importance of this amino acid in the pH-dependent substrate preference. VX-4 was localized in the food vacuoles and cytoplasm of the erythrocytic stage of P. vivax. VX-4 showed maximal activity against actin at neutral pH, and that against P. vivax plasmepsin 4 and hemoglobin was detected at neutral/acidic and acidic pH, respectively.
VX-4 demonstrates pH-dependent substrate switching, which might offer an efficient mechanism for the specific cleavage of different substrates in different intracellular environments. VX-4 might function as a hemoglobinase in the acidic parasite food vacuole, a maturase of P. vivax plasmepsin 4 at neutral or acidic pH, and a cytoskeleton-degrading protease in the neutral erythrocyte cytosol.
| Plasmodium vivax affects hundreds of millions each year and results in severe morbidity and mortality. Plasmodial cysteine proteases (CPs) play crucial roles during the progression of malaria since inhibition of these molecules impairs parasite growth. These CPs might be targeted for new antimalarial drugs. We characterized a novel P. vivax CP, vivapain-4 (VX-4), which appeared to evolve differentially among primate Plasmodium species. VX-4 showed highly unique substrate preference depending on surrounding micro-environmental pH. It effectively hydrolyzed benzyloxycarbonyl-Leu-Arg-4-methyl-coumaryl-7-amide (Z-Leu-Arg-MCA) and Z-Phe-Arg-MCA at acidic pH and Z-Arg-Arg-MCA at neutral pH. Three amino acids (Ala90, Gly157 and Glu180) that delineate the S2 pocket were found to be substituted in VX-4. Alteration of Glu180 abolished hydrolytic activity against Z-Arg-Arg-MCA at neutral pH, indicating Glu180 is intimately involved in the pH-dependent substrate preference. VX-4 hydrolyzed actin at neutral pH and hemoglobin at acidic pH, and participated in plasmepsin 4 activation at neutral/acidic pH. VX-4 was localized in the food vacuoles and cytoplasm of the erythrocytic stage of P. vivax. The differential substrate preferences depending on pH suggested a highly efficient mechanism to enlarge biological implications of VX-4, including hemoglobin degradation, maturation of plasmepsin, and remodeling of the parasite architecture during growth and development of P. vivax.
| Plasmodium vivax, one of the most predominant human malaria species worldwide, causes hundreds of millions of illnesses each year, and can result in severe morbidity and mortality, especially in children [1], [2]. Emergence and spread of multidrug resistant vivax malaria is an increasing problem, which is associated with fatal disease [3]–[5].
Cysteine proteases of malaria parasites are intimately involved in a variety of physiological processes essential for the parasite's survival. The potential roles of the cysteine proteases of P. falciparum such as falcipain-2 (FP-2), FP-2B (synonym of FP-2′) and FP-3 in hemoglobin degradation in the acidic parasite food vacuole [6], [7], processing of food vacuole plasmepsins to active proteases [8], and erythrocyte rupture via cleavage of cytoskeletal proteins followed by merozoite release [9] have been well characterized. Genes encoding three closely related FPs cluster on chromosome 11 within a 12-kb stretch, called the cysteine protease island [10]. FP-2 and FP-2B share similar primary structure and enzymatic properties [11]. Knockout of the FP-2 gene leads to a transient block in hemoglobin hydrolysis, but parasites compensate for the loss through production of FP-2B and/or FP-3, and they multiply at the same rate as wild type parasites. In contrast, FP-3 appears to have an essential role, as knockout of the FP-3 gene is lethal [12]. Another papain-like cysteine protease FP-1, of which gene locates on chromosome 14, appears to be active in early invasive merozoites and in oocyst production in mosquitoes [13], [14]. Plasmodium cysteine proteases appear to display non-overlapping roles with differentiated biochemical properties and expression patterns, as well as redundancy and complementation [15].
Two cysteine proteases, vivapain (VX)-2 and VX-3, which are encoded on chromosome 9, have been identified in P. vivax [16], [17]. VX-2 and VX-3 share a number of biochemical properties with FP-2 and FP-3, including acidic pH optima, requirement for reducing conditions for maximal enzyme activity, and preference toward peptide substrates with positively charged residues at the P1 position and Leu at P2 [17]. Structural modeling of VX-2 and VX-3 has also revealed a topology similar to those of FP-2 and FP-3; however, some substantial differences are detected in the predicted sizes of the binding pockets and residues involved in substrate binding [18]. A gene (XM_001612308) encoding a protein with significant similarity to FP-1 has recently become available in the nucleotide sequence of P. vivax (PVX_195290, PVX_239290, and PVX_240290); its biochemical properties and biological activity remain unclear.
Interests in specific inhibitors impeding plasmodial cysteine proteases have focused on their chemotherapeutic applicability; effective inhibitors impair normal parasite growth in vitro [19]. In addition, rupture of the erythrocyte membrane by mature parasites is inhibited by broad-spectrum inhibitors of serine and cysteine proteases [20], [21]. Identification and further characterization of P. vivax cysteine proteases will be helpful to investigate their biological roles and to characterize targets for antimalarial drugs. However, comprehensive studies of P. vivax cysteine proteases have been hindered by an inability to culture P. vivax.
In the present study, we describe the biochemical properties of a novel cysteine protease of P. vivax, designated vivapain-4 (VX-4), which displays unusual pH-dependent substrate specificity. Molecular modeling and subsequent mutation analysis demonstrated that Glu180 is involved in the pH-dependent substrate specificity of VX-4. The protease effectively hydrolyzed hemoglobin at acidic pH, actin at neutral pH, and plasmepsin 4 at neutral and acidic pHs, supporting its role in the maintenance of metabolic homeostasis and architectural remodeling of the parasite during growth and development.
All animals used in this study were housed in accordance with guidelines from the Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC). All protocols were approved by the Institutional Review Board and conducted in the Laboratory Animal Research Center of Sungkyunkwan University.
Genes putatively coding for cysteine proteases were identified from primate and rodent Plasmodium sequences deposited in PlasmoDB (http://plasmodb.org) and GenBank (http://www.ncbi.nlm.nih.gov/) through BLAST searches. The amino acid (AA) sequences of cysteine proteases of P. falciparum (FP-1, FP-2, FP-2B, and FP-3), P. vivax (VX-2 and VX-3), P. yoelii (yoelipain [YP]-1 and YP-2), and P. berghei (bergheipain [BP]-1 and BP-2) were used in multiple queries, with a threshold at 0.001 (E-value cut-off). After excluding redundancies, the AA sequences were aligned with ClustalX and optimized with GeneDoc. The alignment was used as an input in the construction of neighbor joining and maximum likelihood trees using PHYLIP (ver. 3.6b) and TREE_PUZZLE (ver. 5.2). The standard error in each of the connecting nodes was estimated by bootstrapping of 1000 replicates. Two novel cysteine proteases isolated from P. vivax were annotated as P. vivax cysteine protease 1 (VX-1; XP_001615807) and 4 (VX-4; XP_001615272), according to their clustering patterns in the trees.
The open reading frame (ORF) of VX-4 was amplified with forward (5′- ATGGAATATCACATGGAGTACTCGAAC-3′) and reverse (5′-CTAGTCAAGCAGGGGGACGTACGCCTC-3′) primers using Ex Taq DNA polymerase (Takara) and P. vivax genomic DNA (100 ng) isolated from a Korean patient (a generous gift from Dr. JS Yeom). The product was gel-purified, ligated into the pCR2.1 vector (Invitrogen) and transformed into competent E. coli Top10 cells (Invitrogen). The nucleotide sequence was determined with an ABI PRISM 377 DNA sequencer (Applied Biosystems). The DNA fragment harboring the mature region and a portion of the prodomain from AA position 182 were amplified using 2 primers; 5′-GAGCTCGAGATGCAGCAGAGGTACCT-3′ (contains a 5′ Sac I site) and 5′-CTGCAGCTAATCCACGAGCGCAACGA-3′ (contains a 5′ Pst I site). The PCR product was ligated and transformed as described above, and ligated into the pQE-30 expression vector (Qiagen). The plasmid was transformed into competent E. coli M15 (pREP4) cells (Qiagen), grown overnight in LB medium and induced with 1 mM isopropyl-1-thio-β-D-galactopyranoside for 3 h at 37°C. The bacterial cells were suspended in lysis buffer and then centrifuged. rVX-4 was purified from the supernatant by nickel-nitrilotriacetic acid (Ni-NTA, Qiagen) chromatography, following the manufacturer's instruction. Optimal refolding conditions for rVX-4 were determined with 100 different buffer combinations in a microplate format [22]. For large-scale refolding, purified rVX-4 (100 mg) was diluted 100-fold in optimized refolding buffer (250 mM L-arginine, 1 mM ethylenediaminetetraacetic acid [EDTA], 5 mM reduced glutathione [GSH], 1 mM oxidized glutathione [GSSG], and 100 mM Tris-HCl, pH 8.0), and incubated overnight at 4°C. To allow processing to the active enzyme, the pH was adjusted to 5.5 in the presence of 10 mM dithiothreitol (DTT), the sample was incubated at 37°C for 2 h, and the pH was then readjusted to 6.5. The protein was concentrated with a Centriprep concentrator (cut-off: 10 kDa, Millipore).
The fully processed rVX-4 was separated by 12% SDS-PAGE. The protein was transferred to a polyvinylidene difluoride (PVDF) membrane (Millipore) and stained with Coomassie blue. The band was excised and subjected to protein sequencing on an ABI model 477A protein sequencer and an ABI model 120A PTH analyzer (Applied Biosystems) at the Korea Basic Science Institute (Daejeon, Korea).
Six-wk-old, specific pathogen free (SPF) BALB/c female mice were subcutaneously immunized 3 times with the purified rVX-4 (30 µg per each mouse per each time) in Freund's adjuvants (Sigma-Aldrich) at 2-wk intervals. One week after the final inoculation, 10 µg protein were injected via tail vein. One week later, the blood was collected by heart puncture, after which the antiserum was prepared. BALB/c mouse (6-wk-old) serum obtained from SPF strain was used as a normal control.
Cysteine protease activity was ascertained by the hydrolysis of benzyloxycarbonyl-L-leucyl-L-arginine 4-methyl-coumaryl-7-amide (Z-LR-MCA) (Peptide International). Enzyme (30 µl; 200 nM) was added to 100 mM sodium acetate (220 µl, pH 5.5) containing 5 µM Z-LR-MCA and 10 mM DTT. The release of fluorescence was assessed at excitation and emission wavelengths of 355 nm and 460 nm with a SpectraMAX Gemini fluorometer (Molecular Devices). For activity gel electrophoresis, refolded rVX-4 was mixed with SDS-PAGE sample buffer lacking 2-mercaptoethanol and subjected to 12% SDS-PAGE co-polymerized with 0.1% gelatin. The gel was washed with 2% Triton X-100 (30 min), incubated overnight with 100 mM sodium acetate (pH 5.5) containing 10 mM DTT at 37°C and stained with Coomassie Blue. For kinetic analysis, the rVX-4 (25 nM) was incubated with varying concentrations of peptide substrates at pH 5.5, 6.5 and 7.5 in appropriate buffers, each supplemented with 10 mM DTT. The release of MCA was monitored over 10 min at room temperature as described above. Activities were compared as fluorescence over time. The kinetic constants Km and Vmax were determined using PRISM (GraphPad Software). The optimal pH was assessed in 100 mM sodium acetate (pH 4.5–5.5), 100 mM sodium phosphate (pH 6.0–6.5) and 100 mM Tris-HCl (pH 7.0–8.5). The enzyme (50 nM) was added to each buffer supplemented with 10 mM DTT and 5 µM Z-L-phenyl-L-arginine 4-methyl-coumaryl-7-amide (Z-FR-MCA), Z-leucyl-L-arginine-MCA (Z-LR-MCA), or Z-L-arginyl-L-arginine 4-methyl-coumaryl-7-amide (Z-RR-MCA). The appropriate buffers were separately employed as controls at each pH. Enzyme activity was measured as described above. The effects of reducing agents were examined under various concentrations of GSH, and pH stability was examined at pH 5.0 and 8.0 by incubating rVX-4 at 37°C in the appropriate buffer. Active site titration was done using a specific inhibitor, trans-epoxysuccinyl-L-leuciloamido-(4-guanidino) butane (E-64).
Two synthetic combinatorial libraries were used to determine the substrate specificities of the S1–S4 subsite of rVX-4. To determine P1 specificity, a P1 diverse library consisting of 20 sublibraries was employed. In each sublibrary, the P1 position contained one native AA, and the P2, P3, and P4 positions were randomized with equimolar mixtures of AAs for 6859 tetrapeptide substrates sequenced per sublibrary (in each case, cysteine was omitted and methionine was replaced by norleucine). A total of 20 aliquots (5×10−9 M) of each sublibrary were dispensed into wells of a 96-well microfluor-1 U-bottom plate (Dynex) at a final concentration of 7.3 nM. To determine P2, P3, and P4 specificity, a complete diverse library was used in which the P2, P3, or P4 position was spatially addressed with 20 AAs (norleucine was substituted for cysteine) and the remaining 3 positions were randomized. Aliquots (2.5×10−8 M) from each sublibrary were added to 60 wells of a 96-well microfluor-1 U-bottom plate. Each well contained 8000 compounds (final concentration of 30 nM). Hydrolytic reactions were initiated by the addition of rVX-4 (10 nM) and monitored fluorometrically as described above. Assays were performed at 37°C in 100 mM sodium acetate (pH 5.5), 100 mM sodium phosphate (pH 6.5), or 100 mM Tris-HCl (pH 7.5), in each case with 100 mM NaCl, 10 mM DTT, 1 mM EDTA, 0.01% Brij-35 and 1% dimethylsulfoxide (DMSO).
To observe possible roles of VX-4 in the processing of plasmepsin (PM), we cloned P. vivax plasmepsin (PvPM) 4 (XM_001616821) employing P. vivax genomic DNA obtained from the Korean patient as previously described [23]. Recombinant PvPM4 expressed in E. coli cells was purified by Ni-NTA chromatography (Qiagen) and refolded as described above. rVX-4 (50 nM) was incubated with PvPM4 (20 µg) in 100 mM sodium acetate (pH 5.0–5.5), 100 mM sodium phosphate (pH 6.0–6.5), or 100 mM Tris-HCl (pH 7.0–7.5) supplemented with 10 mM DTT for 3 h. The experiments were also performed in the presence of E-64 (1 µM) and/or pepstatin A (10 µM). Hemoglobinase activity of rVX-4 (30 nM), as well as those of rVX-2 and rVX-3 expressed as previously described [17], was assessed using human hemoglobin (Sigma-Aldrich) in different pHs (5.0–7.5) in the presence of 1 mM GSH at 37°C. Erythrocyte ghosts purified from fresh human blood by hypotonic lysis were incubated with rVX-4 (200 nM) at pH 7.0 or 7.5 at 37°C for 3 h, after which reaction products were analyzed by SDS-PAGE. For immunoblotting, the electrophoretically resolved proteins were transferred to PVDF membranes (Millipore) followed by blocking with 0.05% Tween 20 in phosphate buffered saline (PBST) containing 2% bovine serum albumin. The membrane was incubated with appropriate antibodies including anti-human spectrin (Sigma-Aldrich, 1∶500 dilutions), anti-human band 3 (Sigma-Aldrich, 1∶3000 dilutions), or anti-human actin (Sigma-Aldrich, 1∶1000 dilutions). Blots were subsequently incubated with horseradish peroxidase-conjugated host specific antibodies. The immunoreactive bands were visualized using 4-chloro-1-naphthol (4C1N; Sigma-Aldrich) supplemented with 3% hydrogen peroxide.
Computational analyses were accomplished in a Silicon Graphics Octane 2 workstation, equipped with two parallel R12000 processors (SGI). Homology modeling was orchestrated within the SYBYL 6.9 COMPOSER module (Tripos Associates, MO). Energy minimization and molecular dynamic studies were performed with the DISCOVER module of InsightII 2000 (Accelrys). The geometrical and local environmental consistency of the model was assessed within the PROSTAT and InsightII 2000 Profiles-3D modules, together with the SYBYL 6.9 Matchmaker module. Structural models of FP-2, FP-3, VX-2, VX-3 and VX-4 mature domains were prepared on the basis of their sequence homology with several cysteine proteases using an analogous approach [18]. More than 35% sequence identity was observed between the protein homologs and the target AA sequence. The homologs used in this analysis included human cathepsins K (1ATK), V (1FH0) and S (1MS6); cruzain (1AIM), a cysteine protease from Ginger rhizome (1CQD) and actinidin (1AEC). Terms in parentheses refer to the Protein DataBank accession numbers for the corresponding crystal structures.
Site-directed mutagenesis was performed using a QuickChange II Site-Directed Mutagenesis Kit (Stratagene). A pair of complementary primers with 39 bases was designed and a mutation to replace Ala90 to Ile (A90I), Gly154 to Ser (G154S) or Glu180 to Ala (E180A) was placed in the middle of the primers. Parental DNA inserted in pQE-30 was amplified using Pfu Ultra HF DNA polymerase with these primers for 16 cycles in a DNA thermal cycler (Perkin-Elmer). After digestion of the parental DNA with Dpn I, the amplified DNA with nucleotide substitution was incorporated and transformed into E. coli XL1-Blue (Stratagene). The mutations were verified by DNA sequencing. Double and triple point mutagenesis of A90I, G154S, and E180A were also done as described above. Each mutant plasmid was transformed into competent E. coli M15 (pREP4) cells (Qiagen). Each recombinant protein was individually expressed, purified and refolded as described above.
Thin blood smears (2 µl) were prepared from EDTA-containing venipuncture blood immediately after sampling from patients infected with P. vivax (gift from Dr. JS Yeom). A part of the slides were stained with 3% Giemsa, rinsed and air dried. The unstained thin films were treated with 3% H2O2 for 5 min and incubated with 1% bovine serum albumin. The films were incubated with mouse anti-rVX-4 antibody (1∶500 dilutions in PBS). The reactions were visualized with an avidin-biotin complex (DAKO) and examined under a light microscope (Axiophot, Carl Zeiss).
By data-mining of the P. vivax genome (TIGR, Release 2.0), we identified two genes putatively coding for novel cysteine proteases, in addition to the previously identified genes encoding VX-2 (PlasmoDB code PVX_091415) and VX-3 (PVX_091410). We designated these genes as VX-1 (PVX_195290) and VX-4 (PVX_091405). The other primate Plasmodium genomes examined, such as P. falciparum, P. reichenowi and P. knowlesi, also harbored four closely related cysteine protease genes. Conversely, avian and rodent malaria parasites including P. gallinaceum, P. yoelii, and P. berghei possessed only two paralogous genes (Figure 1). The deduced AA sequence of VX-4 (TC5625, 484 AAs) revealed considerable degrees of identity to that of VX-2 (TC5622, 59%) and VX-3 (TC5618, 48%), while that of VX-1 (TC5613, 583 AAs) was highly related to the FP-1-like proteases of P. falciparum, P. knowlesi, P. ovale and P. fragile (37–77% identity). The greater length of VX-1 might be attributable to an N-terminal extension [24]. Physiological implications and specific domain(s)/signature(s) of VX-1 remain largely elusive. The primary structure of VX-4 tightly conserved the AA residues lining the catalytic site (Gln, Cys, His, Asn and Trp) that are essential for the stabilization of a thiolate-imidazolium ion pair and/or the transition state of the catalytic site (AA positions highlighted in blue in Supplementary Figure S1). The regulatory motifs of the plasmodial cysteine proteases such as a bipartite trafficking domain, inhibitor domain with ERFNIN signature and hemoglobin-binding FP2 arm were also clearly identified in each of the corresponding regions (Supplementary Figure S1) [25], [26]. The eight Cys residues, which are involved in the maintenance of structural geometry, were well conserved in these proteins, whereas the last Cys was replaced by Asn in VX-4 and KP-4 (arrowheads in Supplementary Figure S1). Given the fact that a disulfide bridge between the seventh and eighth Cys residues is intimately engaged in the stabilization of the S2 and S1′ sites of FP-2 [27], the more flexible binding pocket of VX-4 might allow broader accessibility of proteolytic substrates. In addition, several AA substitutions found in critical domains of VX-4 suggest a distinctive physiological role for this protease (Supplementary Figure S1). These collective data demonstrate that VX-4 is a distinct cysteine protease that shares significant identity with, but clearly differs from previously characterized P. vivax cysteine proteases.
A neighbor-joining tree of VX-1 and VX-4 homologs, which were retrieved from PlasmoDB and GenBank, was constructed employing the AA sequences of mature domains (Figure 1). The Plasmodium proteases were largely separated into two distinct clusters consistent with their predicted biological roles: FP-1 clade, of which members are implicated in host cell invasion [13] and oocyst production [14], and FP-2 clade, the majority of which play central roles in hemoglobin degradation [6], [11], [28]. An overall topology similar to that of the neighbor-joining tree was observed in a quartet maximum likelihood tree (TREE_PUZZLE program; data not shown) and the major branching nodes were supported by significant bootstrapping or quartet values. The falcipain homolog genes appeared to have duplicated from a common ancestor before diverging into each of the avian and mammalian parasite lineages. The FP-1 family proteins seemed to have diverged along with their specific donor organisms without any provocative genetic event. Meanwhile, members of FP-2 clade might have more complicated evolutionary pathways, including either multiplication(s) in primate malaria or deletion(s) in rodent malaria. The genes orthologous to VX-2 and VX-3 may have been deleted in the rodent parasites, considering the polytomic relationships among the P. vivax and P. knowlesi paralogs and the tight clustering of VX-4/KP-4 with rodent malarial proteins. This suggestion is further supported by the fact that P. falciparum and P. reichenowi, which comprise a basal clade in mammalian Plasmodium lineages [29], [30], contain three paralogous genes. The three paralogous genes occupying distinct but highly linked genomic loci (cysteine protease island) may have undergone a kind of convergent evolution events in these basal malaria genomes.
Adding to increased genic dosage, the degree of sequence divergence was prominent among the primate FP-2 clade members (0.812±0.078), compared to related rodent proteins (0.271±0.034). The members of primate (0.266±0.035) and rodent (0.377±0.056) FP-1 clade displayed values similar to that of the rodent FP-2-like proteins (Supplementary Table S1). Alteration in gene copy number provides a simple way to change expression levels or to enlarge protein pools with non-overlapping functions. Biochemical studies have demonstrated that the primate malaria proteins belonging to the FP-2 clade exhibit similar enzymatic properties; however, those of P. vinckei (VP-2) and P. berghei (BP-2) demonstrated quite dissimilar features, particularly in terms of their substrate preference and inhibitor specificity [31],[32]. Therefore, the large divergence among the primate FP-2 proteins and tight clustering of VX-4 and KP-4 with rodent Plasmodium proteins (bootstrapping value 76) further suggest biological roles of VX-4 that are distinct from those previously described for VX-2 and VX-3 [16], [17].
The full-length VX-4 gene amplified from a Korean P. vivax patient's blood displayed nucleotide sequence identical to that of the reference Sal I strain (nucleotide sequence data is available in the GenBank under the accession no. AY584068). A rVX-4 protein comprising a portion of the prodomain and entire mature domain was expressed in E. coli (Figure 2A). Purified rVX-4 was refolded followed by maturation under reducing and mild acidic (pH 5.5) conditions. The fully processed 28 kDa protein (left panel, Figure 2B) exhibited protease activity by gelatin-gel electrophoresis (right panel, Figure 2B), which was completely inhibited by the cysteine protease inhibitor E-64 (data not shown). The N-terminal sequence of fully processed rVX-4 was NSPYV (Supplementary Figure S1).
rVX-4 hydrolyzed synthetic dipeptidyl substrates with hydrophobic AA residues at their P2 site such as Z-LR-MCA and Z-FR-MCA under acidic conditions. Activity was highest at pH 5.5. The pH-optimum was substantially different with a substrate containing a basic AA at P2 (Z-RR-MCA) (Figure 2C) with maximal activity at pH 6.5, and activity seen above pH 8. These results suggest either that electrostatic conditions near the S2 site are highly dependent on the surrounding pH or that the geometry of the catalytic site can be changed in a pH-dependent manner. rVX-4 was relatively stable after incubation at acidic and neutral pH, while it was highly unstable under alkaline conditions (pH 8.0 and 8.5) against Z-LR-MCA (Figure 2D), which was similar to results observed for FP-2 and FP-3 [28], [32]. These data suggest that decreased hydrolyzing activity of rVX-4 against Z-LR-MCA and Z-FR-MCA at alkaline conditions might be due to an irreversible change of the protein conformation. The requirement for increased concentrations of E-64 to inhibit rVX-4 at higher pH also supports altered structural geometry as the explanation for altered substrate preference (Figure 2E).
Steady-state kinetic analyses confirmed varied substrate utilization depending on pH (Table 1). rVX-4 showed a similar catalytic efficiency against three peptide substrates at pH 5.5. However, at pH 7.5 kcat/Km against Z-RR-MCA increased 2.6-fold whereas that against Z-LR-MCA decreased 5.5-fold and Z-FR-MCA was not hydrolyzed. The rVX-2 and rVX-3 exhibited much higher kcat/Km values than that of rVX-4 toward Z-LR-MCA at the pH conditions selected, although the optimal pH for rVX-2 was 6.5, rather than 5.5. Interestingly, rVX-2 and rVX-3 could not hydrolyze Z-FR-MCA or Z-RR-MCA. Phe has a large aromatic R group, and it might not fit into the S2 pocket of rVX-2 and rVX-3, which are stabilized by the disulfide bond between the seventh and eighth Cys residues (Supplementary Figure S1).
Substrate preferences were further evaluated using a combinatorial tetrapeptide library (Figure 3). AA utilization patterns of vivapains at the P1, P3 and P4 positions were comparable to those of FP-2 and FP-3 [15]. pH changes did not lead to significant alterations in the substrate preference of VX-4 at these sites, except that for tetrapeptides with Arg at P3; VX-4 was active at pH 5.5, with substantial decreases as pH increased. The most striking specificity was observed with substrates with different AAs at P2. The vivapains preferred hydrophobic AAs such as Leu and Val under acidic (VX-3 and VX-4, pH 5.5) and neutral (VX-2, pH 6.5) conditions. VX-4 also exhibited significant activity against P2 Met and His under acidic conditions. However, relative activity against these hydrophobic AA residues was significantly decreased, while that against Arg increased at pH 6.5 and 7.5, consistent with results obtained using synthetic peptide substrates (Table 1).
Homology modeling of VX-4 demonstrated an overall topology similar to those of FP-2, FP-3, VX-2 and VX-3 with the average pairwise RMSD of 0.98 for the Cα atoms (data not shown). However, a number of substitutions are recognized between VX-4 and the other VXs, including three prominent AA residues delineating the S2 pocket (Ala90, Gly154 and Glu180; numbering from the mature domain of VX-4) (Figure 4A; see also box in Figure 1 and Supplementary Figure S1). The substrate preferences of VX-4 were found to depend on AA residues occupying P2 site and thus, the diagnostic AA substitution might be relevant to the differential biochemistry of VX-4 compared to those of VX-2 and VX-3. Seven mutant forms of VX-4, in which these three AA residues were substituted by single, double, or triple site-directed mutagenesis (A90I, G154S, E180A, A90I/G154S, A90I/E180A, G154S/E180A, and A90I/G154S/E180A), were expressed in E. coli, and their proteolytic activities were examined. All of the refolded recombinant proteins showed hydrolytic activity against gelatin (lower panel, Figure 4B). In assays against peptide substrates, all of the mutants harboring A90I and G154S by single or multiple substitutions maintained the pH-dependent substrate specificity of wild-type VX-4. Conversely, those containing E180A lost activity against Z-RR-MCA at pH 7.5 (Figure 4C). These results demonstrate that Glu180 plays a key role in the pH-mediated switching of substrate specificity of VX-4.
The impact of a single AA substitution at a critical position has been shown in a Leishmania major cathepsin B-like protease, in which a Gly residue at the putative S2 pocket provided no detectable proteolytic activity against Z-RR-AMC, while its replacement with Glu restored activity [33]. A similar result was also observed for papain, which exhibited a preference for Phe over Arg at the P2 position, but exhibited cathepsin B-like specificity when the S2 subsite was altered [34]. A cathepsin B-like cysteine protease of Giardia lamblia that harbors a Glu residue at the S2 pocket was active against both Z-FR-MCA and Z-LR-MCA [35]. The crystallographic structure of cruzain, an essential cysteine protease of Trypanosoma cruzi, demonstrated that the side chain of Glu205 might vary positions and interact with different substrates according to pH and availability of an electrostatically appropriate partner in the S2 pocket [36]. Therefore, the S2 subsite might be intimately involved in the determination of ligand specificity. Although most of the residues delineating the S2 pocket are hydrophobic, a polar residue is present at the pocket's hollow end in some cysteine proteases [37] including VX-4 (Supplementary Figure S1; see also above). The S2 pocket of these enzymes might retain a negative charge at physiologic pH, allowing the capability to bind the polar guanadino group of Arg at the P2 position.
Comparative analysis revealed that two motifs, the FP2 nose and FP2 arm, specific to the hemoglobin-degrading falcipain homologs, were conserved in VX-4 (Supplementary Figure S1). The FP2 nose interacts with the protease core via a highly conserved KEA motif to provide proper folding of the mature protein, while the FP2 arm mediates interaction between the enzyme and hemoglobin [26], [38], [39]. We recognized some differences in the FP2 arm motif of VX-4, in which residues Phe192, Ser194 and Ala198 (numbered from the mature sequence of VX-4) showed different degrees of hydropathy compared to those of other VXs. In addition, Ala198 of VX-4 offered a unique hydrophobic polymorphism, which in structural modeling contributed considerable change in the arm structure (data not shown). These observations suggested that VX-4 may act principally on substrates other than hemoglobin.
We assessed whether VX-4 plays a role in PM processing since a recent study has revealed that FPs function as maturases for PMs within the food vacuole of P. falciparum [40]. Plasmodium species infecting mammals harbored genes for seven PMs (PM4-PM10), of which PM4 orthologs were found in the food vacuole [23]. P. falciparum genome encoded additional food vacuole-related proteins, PM1, PM2, and histo-aspartic protease (HAP), although genes orthologous to these proteins genes were not detected in non-falciparum species. [23], [41], [42]. These results suggest that PvPM4 is the major, if not all, plasmepsin targeted into the food vacuole of P. vivax. We examined possible roles for VX-4 during maturation of recombinant PvPM4 (rPvPM4), which was expressed in E. coli. As shown in Figure 5A, autocatalytic processing of rPvPM4 occurred at acidic pH and, to a less extent, at neutral pH (6.5–7.0). This processing was completely blocked by the aspartic protease inhibitor pepstatin A. This cleavage was significantly accelerated in the presence of VX-4 in a dose- and time-dependent manner (data not shown). In the presence of pepstatin A to block autocatalysis, rVX-4 effectively cleaved rPvPM4 at pH 5.0–7.0, and this process was specifically and significantly inhibited by E-64. These results suggest that VX-4 is a key molecule regulating PvPM4 maturation. Processing may occur during trafficking of the enzymes from endoplasmic reticulum (ER)-derived transport vesicles or the parasitophorous vacuolar space (PVS), where pH is neutral, or in the acidic food vacuole (pH 5.4–5.5) [43], [44]. VX-2/VX-3 might also participate in the processing in the food vacuole.
The major hemoglobinases of P. falciparum are targeted into the food vacuole through ER-derived vesicles, but it is unclear whether the ER-derived, protease-containing vesicles fuse with hemoglobin-containing transport vesicles derived from cytosomes, or if they directly contact the food vacuole [43], [45]. The bipartite signals, composed of cytoplasmic, transmembrane and lumenal motifs, were found to be required for trafficking of FP-2 and FP-3 to the food vacuole, and they are conserved in VX-2, VX-3, and VX-4 (Supplementary Figure S1) [39], [46]. The hemoglobinase activity of VX-4 was compared to that of VX-2 and VX-3. pH-dependent and time-lapse analyses demonstrated that the hemoglobinolytic activity of VX-4 was relatively weak. Maximal hemoglobin degrading activity of VX-2, VX-3 and VX-4 was observed between pH 6.0–6.5, 5.0–6.0, and 5.5–6.0, respectively (Figure 5B). Considering their peak activities at different pHs, the action points of different VXs may be temporally segregated during hemoglobin degradation. However, it is unclear whether the biochemical differences between VX-2, VX-3, and VX-4 are most important to foster cooperative action against hemoglobin or to provide activities against different substrates over the course of erythrocytic infection by P. vivax.
To consider other potential substrates for VX-4, we examined hydrolytic activity against erythrocyte cytoskeletal proteins (Figures 5C and D). VX-4 cleaved the majority of erythrocytic ghost proteins under acidic conditions (pH 5.0–6.0), whereas some activities were negligible at neutral pH (6.5–7.5). However, VX-2, VX-3, and VX-4 all degraded band-3 (anion exchanger 1, AE1) and actin at neutral pH. The proteolytic activities of VX-4 against erythrocyte actin and band-3 suggest an additional role for the protease in remodeling of erythrocyte cytoskeleton during the process of egress of merozoites from erythrocytes at the conclusion of the parasite erythrocytic cycle. Alternatively, actin degradation may be directly related to hemoglobin transport into the food vacuole, as a recent study showed that actin filament turnover in P. falciparum might be essential for both cytostome formation and hemoglobin translocation [44].
The major hemoglobinases of P. falciparum (FP-2, FP-3 and plasmepsins) are targeted into a food vacuole through the ER-derived vesicles. However, it could not be clearly concluded whether the ER-derived, protease-containing vesicles are fuse with hemoglobin-containing transport vesicles, which were pinched off from cytosomes, or they directly contact with the food vacuole [43], [45]. However, investigations have been highly limited with the P. vivax proteins, not only due to low parasitemia in the patients' blood, but also due to failure of experimental maintenance of the parasite. The N-terminal regions of vavapains conserved the characteristic bipartite signal for trafficking to the food vacuole, which included cytoplasmic, transmembrane and lumenal motifs [26] (Supplementary Fig. S1). Alternatively, the proteolytic activities of VX-4 against erythrocytic actin and band-3 might suggest its additional role(s) in the remodeling of erythrocytic cytoskeleton, although the low activity with spectrin, which is one of the major cytoskeletal proteins in erythrocyte, makes it unclear (Figure 5D).
We prepared a mouse antiserum specific to rVX-4, which showed negligible cross-reactions with rVX-2 and rVX-3 as well as erythrocyte proteins (Figure 6A). We assessed the spatiotemporal expression pattern of VX-4. As shown in Figure 6B, VX-4 was shown to be expressed through all of the intraerythrocytic stages of P. vivax, from ring to schizont/gametocyte stages. VX-4 localization appeared to be largely limited to the food vacuoles with dark homozoin pigment, while the protein was also labeled diffusely in the parasite cytoplasm. P. falciparum FP-3 seemed to have a biological implication(s), which is pivotal to the parasite's survival, in addition to the hemoglobin degradation [46] and showed a distribution pattern similar to that of VX-4 [47]. Given the fact that VX-4 has hydrolytic activity against cytoskeletal proteins, the cytoplasmic distributions of VX-4 and FP-3 might suggest their cytosolic roles such as cytoskeletal remodeling and hemoglobin transportation, which is pivotal for the maintenance of intraerythrocytic stage of the parasites.
The substrate specificity of proteases depends largely on interactions between a substrate and the enzyme active site. The binding efficiency is greatly affected by the physicochemical micromilieu. The reaction pH may confer substrate preference, as has been seen with cruzain [36]. The pH-dependent substrate switching of VX-4 might be relevant to its multiple biological roles; the protein might function as a maturase of PvPM4 in the plasma membrane or cytosomes at neutral pH, while it participates in the degradation of hemoglobin in the acidic food vacuole. VX-4 might also be involved in cytoskeletal remodeling for the invagination of parasite plasma membrane to form cytostomes and/or the hydrolysis of host proteins to facilitate parasite egress from the erythrocyte. VX-4 thus may be a multifunctional enzyme, performing pivotal functions to ensure parasite survival during the complex life cycle of P. vivax. Given the multifunctional activities of VX-4, which are critical for the survival and/or metabolic homeostasis of the parasite, the enzyme might be an attractive target for the development of new antimalarial chemotherapeutics. Work toward further identification of natural substrates and distinct protease functions are currently underway to facilitate a more comprehensive understanding of the biological significance of this enzyme.
|
10.1371/journal.pgen.1006953 | Co-option of the bZIP transcription factor Vrille as the activator of Doublesex1 in environmental sex determination of the crustacean Daphnia magna | Divergence of upstream regulatory pathways of the transcription factor Doublesex (Dsx) serves as a basis for evolution of sex-determining mechanisms in animals. However, little is known about the regulation of Dsx in environmental sex determination. In the crustacean Daphnia magna, environmental sex determination is implemented by male-specific expression of the Dsx ortholog, Dsx1. Transcriptional regulation of Dsx1 comprises at least three phases during embryogenesis: non-sex-specific initiation, male-specific up-regulation, and its maintenance. Herein, we demonstrate that the male-specific up-regulation is controlled by the bZIP transcription factor, Vrille (Vri), an ortholog of the circadian clock genes—Drosophila Vri and mammalian E4BP4/NFIL3. Sequence analysis of the Dsx1 promoter/enhancer revealed a conserved element among two Daphnia species (D. magna and D. pulex), which contains a potential enhancer harboring a consensus Vri binding site overlapped with a consensus Dsx binding site. Besides non-sex-specific expression of Vri in late embryos, we found male-specific expression in early gastrula before the Dsx1 up-regulation phase begins. Knockdown of Vri in male embryos showed reduction of Dsx1 expression. In addition, transient overexpression of Vri in early female embryos up-regulated the expression of Dsx1 and induced male-specific trait. Targeted mutagenesis using CRISPR/Cas9 disrupted the enhancer on genome in males, which led to the reduction of Dsx1 expression. These results indicate that Vri was co-opted as a transcriptional activator of Dsx1 in environmental sex determination of D. magna. The data suggests the remarkably plastic nature of gene regulatory network in sex determination.
| Sex is widespread for reproduction of offspring in the animal kingdom. In the sex determination process, through interactions of several genes in a hierarchical manner, an initial cue leads to sex-specific expression of the major effector of sexual differentiation, Doublesex (Dsx). Although how genetic factors on sex chromosomes control Dsx expression has been extensively studied in model organisms such as mouse, fruit fly, and nematodes, little is known about dependence of Dsx regulation on environmental signals. We used the crustacean, Daphnia magna, owing to its advantages for analyzing environmental sex determination: 1) fully sequenced genome, 2) recent advancement of genome engineering and 3) artificial control of sex by juvenile hormone treatment. We found that early male embryos transiently express the bZIP transcription factor, Vrille (Vri), known to be a circadian regulator, before male-specific Dsx1 activation begins. Disruption of a potential Vri-binding site in the Dsx1 regulatory region, and gain- and loss-of-function analyses revealed that Vri regulates male-specific Dsx1 activation in Daphnia. We infer that a novel gene can be co-opted as a regulator of Dsx in environmental sex-determining pathway. Our results would expand our understanding about the diversity and evolution of the sex-determining pathways in animals.
| The diversity and evolution of sex-determining pathways among animals are fundamental issues in developmental and evolutionary biology. The primary cues to trigger sexual development have been varied across evolution [1,2], and can be broadly divided into two categories: a strict genetic cue or merely an environmental signal [3]. There are numerous studies about genetic sex determination (GSD) mechanisms from various model organisms, including the mouse, nematode, and fruit fly. These studies have shown that, through interactions of several genes in a hierarchical manner, initial cues finally lead to sex-specific expression of the major effector of sexual differentiation, a DM-domain gene that encodes a transcription factor containing a DNA binding domain called DM-domain [4]. In addition, pioneering studies using model organisms have demonstrated that sex-determining genes differ among species upstream of the hierarchies [4,5]. In contrast, little is known about the mechanisms of environmental sex determination (ESD) because organisms with ESD systems are poor genetic models.
The crustacean waterflea, Daphnia magna, has emerged as a model organism for understanding ESD because of its fully sequenced genome [6,7] and advances in genetic manipulations through RNAi [8], ectopic expression [9], CRISPR/Cas9 [10] and TALEN systems [11–13]. In healthy populations, Daphnia normally produces female clones through parthenogenesis, but switches to sexual reproduction when environmental qualities for growth and reproduction decline [14]. In unfavorable environments such as shortened photoperiod, lack of food and/or high population density, Daphnia produces clonal males that allow fertilization of haploid eggs, which results in the production of resting eggs as a survival strategy upon harsh conditions [15]. We and others have shown that juvenile hormone analogs (JHAs) induce male production in cladoceran crustaceans without environmental cues [16,17]. A developing oocyte is sensitive to JH or JHA and a period when eggs are destined to be males by these chemicals is four to ten hours before ovulation (Fig 1A) [16,17], suggesting that environmental cues for sex determination are converted to JH signals neuroendocrinically. We also found that, during embryogenesis, a DM domain gene name Doublesex1 (Dsx1) is exclusively expressed in male-specific tissues and regulates the male trait development in D. magna [18], which provides evidence that both GSD and ESD have the same origin and share similar genetic components in their sex-determining pathways.
To understand mechanisms of JH-dependent Dsx1 activation in D. magna, we had previously examined temporal change of its expression during embryogenesis [18]. Of the two Dsx1 mRNAs (Dsx1-α, Dsx1-β) which differ only at the 5′ UTR, zygotic transcription of Dsx1-α mRNA is largely divided into three phases (Fig 1A), non-sex specific transcription prior to early gastrula at 6-hour post ovulation (hpo) (initiation), male-specific activation during gastrulation from 6- to 9-hpo (up-regulation), and constant transcription during late embryogenesis (maintenance). Male-specific transcription of Dsx1-pemRNA starts three hours later than Dsx1-s mRNA (around 9-hpo) and thereafter become more abundant in male embryos. We also generated transgenic D. magna to visualize spatiotemporal expression patterns and discovered that male-specific Dsx1 expression starts in a presumptive primary organizer that migrates from the rostral to the caudal side on a ventral region at 11-hpo and thereafter gradually becomes specialized in male traits [19]. These previous findings suggest that JH activates Dsx1-α mRNA transcript in a specific population of gastrula cells. However, there is a significant time lag between the critical period of the JH action and onset of up-regulation of Dsx1-α mRNA levels (Fig 1), suggesting that Dsx1 is not a primary JH-responsive gene regulated by the JH receptor protein, Methoprene-tolerant (MET) [20], but unknown transcription factors control its male-specific up-regulation in gastrula.
In this study, we aimed to identify the transcription factor responsible for male-specific up-regulation of Dsx1-α mRNA transcription that starts at 6-hpo. We searched for potential transcription factor binding sites at a region upstream of the transcription start site of Dsx1-α transcript. We found a potential enhancer that contains a consensus sequence of the Dsx binding site and an overlapping element for binding of an ortholog of the bZIP transcription factors, Drosophila Vrille (Vri) and vertebrate E4BP4/NFIL3, which are known to be involved in various general development processes including growth [21,22], circadian clock regulation [23,24], metamorphosis [25], apoptosis [26], and human T cell function [27]. In D. magna, Vri showed male-specific transient expression at 6-hpo. Loss- and gain-of-function analyses showed Vri to be necessary and sufficient for Dsx1 activation. In addition, the disruption of the enhancer suggested Vri-dependent Dsx1 activation. We infer co-option of the transcription factor Vri to the environmental sex-determination cascade.
To find candidate transcription factors (TFs) that activate Dsx1 male-specific expression from 6-hpo, we analyzed a sequence within 7,899 base pairs upstream from the transcription start site of Dsx1-α mRNA. First, elements similar to known TF binding sites were searched with the TFBIND program [28] using the transcription factor database TRANSFAC R.3.4. Next, because Dsx1 up-regulation and maintenance phases suggest positive feedback regulation of this gene, we investigated consensus binding sites of Drosophila melanogaster Dsx. Of the thousands of potential TF binding sites found in this study, we focused on an element similar to the fat body enhancer of the Drosophila yolk protein gene 1 that contains a Dsx binding site and an overlapping bZIP protein binding site [29]. We confirmed conservation of its position and sequence in the related daphniid species Daphnia pulex (Fig 1B and 1C). In this Daphnia species, a binding site for bZIP protein in the potential Dsx1 enhancer matched to a consensus binding site for mammalian E4BP4/NFIL3, suggesting that an ortholog of E4BP4/NFIL3 may function as a transcriptional activator of Dsx1 in Daphnia. To investigate the existence of an E4BP4/NFIL3 ortholog in D. magna, we performed a BLAST search using an amino acid sequence of the human E4BP4/NFIL3 against the D. magna genome database and found one ortholog that shows high homology in the bZIP domain to E4BP4/NFIL3 proteins (S1 Fig). We determined the cDNA sequence by 5′ and 3′ RACE reactions and obtained a 2,394 bp nucleotide sequence that codes for 797 amino acids (S2 Fig). Phylogenetic analysis using bZIP domains from various animals revealed that the Daphnia E4BP4/NFIL3 ortholog is most closely related to the insect E4BP4/NFIL3 ortholog Vrille (S3 Fig). Therefore, we designated this gene as Vrille (Vri).
We then analyzed the temporal expression profile of Vri by qRT-PCR during embryogenesis (Fig 2). At early stages of embryogenesis (0, 3, and 6-hpo), Vri expression in males was higher than that in females. At 6-hpo, Vri transcripts transiently became more abundant and retained the sexually dimorphic expression pattern (Fig 2A). At later embryonic stages (18 and 36-hpo), Vri expression increased both in males and in females and lost its sexual dimorphism (Fig 2B).
The existence of a Vri binding site in the Dsx1 promoter sequence and the male-specific expression of Vri prior to Dsx1 up-regulation led us to hypothesize that Vri could regulate male-specific Dsx1 up-regulation in gastrula. To investigate this hypothesis, we performed RNAi-mediated knockdown analysis as described previously [8,30]. To confirm specificity of phenotypes induced by Vri RNAi, we designed two siRNAs, Vri_siRNA_1 and Vri_siRNA_2, which differ at their target sequences (S2 Fig). To observe the cells and tissues influenced by Vri RNAi during embryogenesis, we used transgenic H2B-GFP expressing Daphnia that allows us to visualize individual cells in an embryo [31]. We injected each siRNAs into the eggs induced to become males by exposure to the JH agonist Fenoxycarb.
Based on H2B-GFP expression patterns, development of both Vri_siRNA_1- and Vri_siRNA_2-injected embryos seemed to be normal at around 10 to 11-hpo. At 20-hpo, Vri_siRNA_1-injected embryos developed cephalic appendages such as second antennae but did not start thoracic segmentation in contrast to control embryos (Fig 3A). Vri_siRNA_2-injected embryos died because of more severe phenotypes in which the segmental structures were not formed. At 30-hpo, Vri_siRNA_1-injected embryos showed abnormal segmentation of thoracic appendages, and undeveloped posterior and anterior regions of the embryos (Fig 3A), which prevented us from investigating sex-reversal in sexually dimorphic structures such as the 1st antennae. These RNAi-dependent severe deformities were also observed in females (S5 Fig).
To exclude the possibility that the developmental defect affects Dsx1 expression, we analyzed Dsx1 expression levels in RNAi embryos at 11-hpo by qRT-PCR and validated that Vri expression level was negligible in both of the RNAi embryos (S4 Fig). qRT-PCR analysis also revealed that both Vri_siRNA_1 and Vri_siRNA_2 reduced Dsx1 expression (Fig 3B). To further analyze where Vri RNAi reduced Dsx1 expression, we used another transgenic Daphnia, a Dsx1 reporter strain that expresses mCherry, the red fluorescence protein under the endogenous Dsx1 promoter/enhancer [19]. At 20-hpo, in control male and female Daphnia, the mCherry fluorescence appeared exclusively in male embryos and is localized in the 1st antennae, which are the first organs to show a male-specific trait in Daphnia. In addition, mCherry-expressing cells could be seen in thoracic appendages, which may be supplied from the posterior growth zone [19] (Fig 3C). In Vri_siRNA_1-injected male embryos, mCherry signal could be seen only in the posterior growth zone but its signal was weaker. Vri_siRNA_2 injected embryos did not show any red fluorescence (Fig 3C, Table 1), although due to severe effect of Vri silencing on embryonic processes, we could not exclude the possibility that some of the structures which normally express the mCherry reporter were not properly formed when Vri was silenced.
To test whether transient expression of Vri in early embryos is sufficient to activate Dsx1 and trigger male development, we induced transient ectopic expression of Vri in females by delivering capped, polyadenylated mRNAs into ovulated eggs via microinjection. We first attempted to establish a system to mimic transient expression of Vri in early male embryos. We constructed GFP mRNAs harboring the 5′ UTR and 3′ UTR sequences obtained from Xenopus laevis β-globin gene and injected this chimeric GFP mRNAs into female eggs. This injection led to expression at early embryogenesis (3 to 10-hpo) but not in the later stages (S6 Fig). Therefore, we linked the X. laevis β-globin UTRs to the Vri CDS and injected this chimeric Vri mRNA into wild-type eggs that would develop into females.
Although this chimeric mRNA induced high embryonic lethality (Table 2), the juveniles that survived showed partial elongation of the 1st antennae in an mRNA concentration-dependent manner (Fig 4A, Table 2). Consistent with this masculinized phenotype, we could confirm up-regulation of Dsx1 expression levels in Vri RNA-injected daphniids by qRT-PCR at 48 to 50-hpo (Fig 4B). Low viability prevented us from evaluating further masculinization in injected female animals.
In addition, by using the Dsx1 reporter strain, we tested the effects of the same chimeric Vri mRNA on Dsx1 activation in females and detected high and widespread mCherry expression mainly in thoracic appendages at 50-hpo (Fig 4C, Table 3). To confirm whether Vri’s DNA binding activity was necessary for Dsx1 activation, we injected mRNA encoding a mutated form of Vri that lacked the bZIP domain (S2 Fig). This mutated Vri could increase Dsx1 expression levels but showed lower transactivation activity (Fig 4C, Table 3). Taken together, these loss-and-gain-of-function analyses show that Vri functions as a transcription activator for Dsx1 expression in D. magna.
To test whether the enhancer element is required for Dsx1 activation and male trait development, we tried to disrupt its sequence on the genome by using the CRISPR/Cas9 system. Because the low GC content (23%) of the enhancer prevented us from designing enhancer-targeting TALENs and gRNAs, we designed two separate gRNAs, gRNA-1 and gRNA-2, near to the enhancer (Fig 5A) and confirmed the functionality of the gRNAs by Cas9 in-vitro cleavage assay (S7 Fig). We then co-injected the two gRNAs with Cas9 protein into the Dsx1 reporter strain [19] that would develop into males and evaluated effects of enhancer disruption on Dsx1 and the morphological phenotypes. At 36-hpo, we found four different phenotypes from the 12 injected embryos. Four embryos (#1, #2, #3, and #4) exhibited phenotype-1 (Fig 5B), in which embryonic development was delayed and the embryos showed weaker mCherry fluorescence than control but at later stages, they could have normal male traits development. Two embryos (#5 and #6) showed phenotype-2 wherein egg development was disturbed and mCherry signal was weak with abnormal localization. Phenotype-3 was observed in three embryos (#7, #8, and #9) showing the most severe deformities and no mCherry expression. The remaining three embryos (#10, #11, and #12) showed no apparent change in phenotype compared to control (Phenotype-4). The abnormal development of these eggs prevents us from observing the sex-specific traits or feminized phenotypes.
To examine the correlation between the introduced mutations and the observed phenotypes, we extracted genomic DNA from each embryo and performed genomic PCR to amplify the enhancer region. Native PAGE electrophoresis of PCR products showed either bands of smaller sizes than what was expected from wild-type sequence, or wild-type bands of reduced intensity, suggesting that large and small deletions in the enhancer region had occurred (Fig 5C). We measured the intensity of each band and calculated the ratio of intensity of expected to smaller bands, and observed that the more severe the phenotype of injected embryo was, the higher was the ratio. These results indicate that the enhancer may be a cis-regulatory element for male-specific Dsx1 expression.
In addition, we attempted to generate the enhancer knockout mutants by injecting the Cas9 protein-gRNAs complexes and collecting offspring of the injected daphniids. In injection into eggs that develop into females, neither somatic nor heritable mutations were detected (S1 Table, S2 Table). In male daphniids, the injection led to high embryonic lethality (>90%) (S2 Table). We could collect offspring by feminizing the survived males using Dsx1 RNAi, but no mutant line was generated.
We had previously found that JH and Dsx1 are essential for environmental sex determination in D. magna [18]. JH drives commitment to male development in oocytes at 4 to 10 h before ovulation [32]. In response to JH signal, Dsx1 is up-regulated from early gastrula at 6 h post-ovulation and is maintained in late embryos for the control of male trait development [19]. However, the molecular mechanisms that mediate JH signaling and Dsx1 up-regulation have remained unknown. In this study, we identified the bZIP transcription factor, Vri as a candidate transcriptional activator by sequence analysis of the Dsx1 promoter/enhancer. Further studies involving expression pattern analysis, loss- and gain-of-function analyses and disruption of an enhancer harboring a Vri consensus binding site indicated that it is required for male-specific Dsx1 up-regulation. Our findings provided evidence that Vri has been co-opted as a component upstream of Dsx1 in the environmental sex-determining pathway.
Over the past several years, new sex-determining genes have been identified in genetic sex-determining pathways in several animals, which reveals the importance of gene co-option. Mechanisms for co-option of new sex-determining genes are largely divided into three categories: 1) allelic diversification, 2) duplication of genes related to sexual development and 3) recruitment of a novel gene with no homology to any known sexual regulators [33]. First, by allelic diversification, transcription factor SOX3 was recruited as a master regulator for sex determination in mice [34] and Indian ricefish [35]. By the same mechanism, the DM-domain gene Dmrt1 and the gonadal soma-derived growth factor (Gsdf) were also co-opted at the top of sex-determining pathways in birds [36] and Luzon ricefish [37] respectively. Second, in frog [38] and Medaka [39], the Dmrt1 gene was duplicated and one of the duplicates gained function as a master sex-determining gene. In insects, transformer orthologs that are conserved components of the sex-determining cascades, were duplicated in honeybee [5,40], resulting in upstream regulators named the Csd. These findings suggested that orthologous genes are repeatedly co-opted for genetic sex-determining pathways in independent animal lineages [33] even though, in the silkworm and the rainbow trout, novel factors, a piRNA [41] and the interferon regulatory factor irf9 [42] seems to have evolved as sex determiners. The Vri gene was previously identified as one of genes regulated by Dsx in male Drosophila [43]. As well as most of previously identified sex-determining genes, Vri might be repeatedly employed in the sex-determining regulatory networks. In environmental sex-determining D. magna, without allelic diversification and duplication, Vri would have been co-opted in upstream of Dsx1. Sex-related roles of Vri in various organisms should be examined in future.
Our findings indicate that Vri functions as an activator of the Dsx1 gene in Daphnia. In Drosophila, Vri regulates various developmental processes such as cell growth, proliferation and flight [21,44], as well as metamorphosis [25] and tracheal integrity [22]. In addition to these processes, Vri is required for circadian oscillation by repression of Clock transcription [24]. In mammals, the Vri ortholog E4BP4/NFIL3 is also reported as a clock-controlled gene. It competes for the binding site of the PAR-protein. Both Drosophila Vri and mammalian E4BP4/NFIL3 are well known as transcriptional repressors. However, in the human immune response system, E4BP4/NFIL3 was identified as an activator of the IL3 promoter [27] and was also shown to up-regulate IL-10 and IL-13 [45]. It is essential for lineage commitment of innate lymphoid cells (ILCs) [46]. In natural killer cell development, E4BP4/NFIL3 interacts with the histone ubiquitinase MYSM1 and maintains an active chromatin state at the Id2 locus [47]. In Daphnia sex determination, Vri works at the gastrulation stage when lineage commitment occurs. These similarities in regulation at the genetic and cellular levels may suggest that the molecular mechanism of Vri-dependent Dsx1 activation is similar to that of E4BP4/NFIL-3 function in human ILCs.
Based on the timing of action of JH, Vri, and Dsx1, we were able to propose a hierarchy of signal transduction in environmental sex determination (Fig 6A). In this hierarchy, JH first stimulates expression of Vri, which in turn activates Dsx1 expression. To examine the possibility that the JH-receptor MET directly regulates Vri activation, we searched for sequences similar to the MET-binding site for the Vri promoter/enhancer and found one candidate sequence that is conserved in two Daphnia species (Fig 6B), suggesting that this motif functions as an element to regulate the JH-dependent gene expression. However, because there is still time lag between JH signaling and Vri activation, there might be other molecules that respond to JH signal and then direct the male-specific Vri transcription. Thus, discovering these early response genes of JH signal may improve our understanding of hormonal signaling and the environmental sex determination pathway.
Interestingly, in the initiation phase of Dsx1 transcription, Dsx1 is transcribed both in males and in females at 3 to 6-hpo. We hypothesize that in males, Vri might form a heterodimer with Dsx1, bind to the enhancer and up-regulates Dsx1 expression at 6 to 9-hpo. Drosophila Dsx is known to form a heterodimer with the bZIP-domain transcription factor and binds to the fat body enhancer (FBE) of the yolk protein gene [29]. Transactivation of Dsx1 by a truncated Vri lacking the bZIP domain in this study also suggests that heterodimer formation allowed the mutated Vri to access the target binding site. These suggest that the heterodimeric combination of Dsx and bZIP transcription factors has functioned as a transcriptional regulator before divergence of insects and crustaceans. Even though we provide substantial genetic evidence of Dsx1 activation by Vri in early embryos, because the loss- and gain-of- Vri function led to embryonic lethality, we still cannot conclude that Vri is the sole upstream component acting as a Dsx1 activator that is necessary for male trait development. To understand more about Vri function in environmental sex determination, we will need to clarify localization of Vri in early embryos and perform knockdown/overexpression in cells that express Vri endogenously, which would avoid alternation of non-sex specific functions of Vri in later embryos.
In targeted mutagenesis using Cas9, we could not introduce any mutation into the Vri binding site at the Dsx1 promoter/enhancer on the genome in females. In contrast, this mutagenesis introduced deletion at the target site on the genome in males and reduced Dsx1 expression. These results suggest that this enhancer may be silenced via closed chromatin in females but is required for Dsx1 activation in males. We also found that deletion of the enhancer led to embryonic lethality in males although we could not shed light on the mechanism underlying this high mortality. However, these clear differences of phenotypes between males and females in our targeted mutagenesis experiments indicate a male-specific role of this enhancer. Further study is needed to understand the epigenetic regulation at Dsx1 locus.
In conclusion, we demonstrate co-option of the bZIP transcription factor Vrille upstream of the Dsx1 in the environmental sex-determining cascade of the crustacean D. magna. Vri is transiently expressed in early gastrula in response to juvenile hormone and controls male-specific up-regulation of Dsx1 in late gastrula. This is the first finding that Vri is recruited into sex determining pathways. Our finding reveals the remarkably plastic nature of Dsx regulation, which will contribute to understanding of the diversity and evolution of the sex-determining pathways in organisms.
All of the wild-type (WT) and transgenic lines share the same genetic background (NIES clone). They were cultured in ADaM medium [48] as described previously [49]. Male Daphnia were obtained by exposing female adults (2–3 weeks old) to 1 μg/L of the synthetic juvenile hormone analog, Fenoxycarb (Wako Pure Chemical; Osaka, Japan) [50]. We utilized previously established transgenic lines of D. magna. One of the transgenic lines (HG-1) expresses H2B-GFP protein under the control of D. magna Elongation Factor 1 α-1 (EF1α-1) promoter/enhancer [31]. Another was the Dsx1 reporter strain, which was generated by introducing mCherry gene upstream of Dsx1 coding sequence in the genome of the HG-1 [19].
Total RNA was extracted from female and male embryos in triplicates using Sepasol-RNAI solution (Nacalai Tesque; Kyoto, Japan). The RNA was subjected to cDNA synthesis using random primers (Invitrogen; Carlsbad, CA, USA) and the SuperScriptIII Reverse Transcriptase (Invitrogen). qPCR was conducted with the SYBR GreenER qPCR Supermix Universal (Invitrogen) using the Mx3005P real time (RT)-PCR system (Agilent Technologies; Santa Clara, CA, USA). Vri expression was quantitated and was normalized with the ribosomal protein L32 expression level using the primers listed in S3 Table. The primers used to amplify the Dsx1 and the ribosomal protein L32 gene were the same as described previously [18]. For normalization of Dsx1 expression level in the Vri knockdown using Vri_siRNA_1 and overexpression, expressions of three other reference genes, ribosomal L8 gene, β-actin gene and Cyclophilin gene [51] were analyzed using the primers listed in S3 Table. The geometric mean of the reference genes were calculated and used for normalization as described previously [52].
The Vri cDNA sequence was amplified from Daphnia by 5′ and 3′ rapid amplification of cDNA ends (RACE) methods as described previously [30]. The primer sequences used for cDNA fragment amplification were as follows: Vri 5′ RACE gene specific primer (5′-TGTTGCTGCCGATTGCGCTGACACTG-3′); Vri 5′ RACE nested primer (5′-CTCGGTCGAACGCCGTCCGCTACTG-3′); Vri 3′ RACE gene specific primer (5′-CCGGCCGTGTACTGCCGCTCAAACTA-3′); and Vri 3′ RACE gene nested primer (GGCTGCCGCTGTTCTGCTGACACTCA-3′). The resulting PCR products were excised from an agarose gel after electrophoresis, purified and were cloned into a TOPO vector (Invitrogen) for sequencing analysis. We then used the DNA sequence for homology search and phylogenetic analyses using BLAST and MEGA (version 7.0.21) as mentioned previously [30]. The Vri cDNA sequence is available from the DDBJ database (http://getentry.ddbj.nig.ac.jp/getentry/na/LC230164/?format=flatfile&filetype=html&trace=true&show_suppressed=false&limit=10) (Accession number LC230164).
To knockdown the Vri gene, 100 μM of Vri_siRNA_1 and Vri_siRNA_2 (sequences indicated in S2 Fig) were used. A previously used control siRNA (5′- GGUUAAGCCGCCUCACAUTT-3′) was utilized as a negative control [53]. The siRNA oligonucleotides were dissolved in DNase/RNase-free water (Life Technologies Inc.; Grand Island, NY, USA).
To overexpress the Vri gene, chimeric Vri cDNA harboring the 5′ UTR and 3′ UTR of X. laevis β-globin gene was designed and subcloned downstream to the T3 promoter on the pRN3 vector [54]. The Vri CDS of this plasmid was replaced with the CDS of GFP fused with minos transposase for preparation of control mRNA for investigating effects of β-globin UTRs on mRNA stability and/or translation efficiency. These plasmids were linearized by BsaAI restriction enzyme, purified with phenol/chloroform extraction and used as templates for mRNA synthesis. In vitro transcription by T3 RNA polymerase and poly-A tail addition were performed according to the manufacturers’ protocol of the commercial kits mMessage mMachine T3 kit (Life Technologies Inc.) and Poly(A) Tailing kit (Life Technologies Inc.), respectively. The synthesized mRNAs were column purified by RNeasy Mini kit (Qiagen; Tokyo, Japan), followed by phenol/chloroform extraction, ethanol precipitation, and dissolution in DNase/RNase-free water.
For the syntheses of gRNAs, the templates were prepared by the cloning free method [55]. The sense synthetic oligo contains three main parts: a T7 promoter (shown in bold), a variable targeting sequence (N18) and the first 20 nt of the Cas9 binding scaffold sequence. The full sequence is as follows: (5′- GAAATTAATACGACTCACTATAGGNNNNNNNNNNNNNNNNNNGTTTTAGAGCTAGAAATAGC-3′). The anti-sense oligo contains 80 nt full sequence of the Cas9 binding scaffold: (5′-AAAA GCACCGACTCGGTGCCACTTTTTCAAGTTGATAACGGACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAAC-3′) where the underlined nucleotides denote the complementary sequence between two oligo sequences. The PCR reaction was performed with PrimeSTAR polymerase (Takara Bio; Shiga, Japan). After purification by phenol/chloroform extraction, the DNA fragments were used as templates for in vitro transcription with the MEGAscript T7 kit (Life Technologies Inc.), followed by column purification with mini Quick Spin RNA columns (Roche diagnostics GmbH; Mannheim, Germany), phenol/chloroform extraction, ethanol precipitation, and dissolution in DNase/RNase-free water.
Microinjection was performed as described previously [8]. Eggs were obtained from adult Daphnia at 2–3 weeks of age, directly after ovulation and placed in ice-cold M4 media contained 80 mM sucrose. The specific RNAs for each experiment were mixed with either Alexa Fluor 568 dye (Life technologies Inc.) or Lucifer Yellow dye (Life technologies Inc.) with final concentrations of 0.01 μM and 1 μM respectively, as an injection marker. The microinjection was performed on ice and the injected eggs were incubated in a 96-well at 23°C for the appropriate time.
We mixed in vitro synthesized RNA with Cas9 protein to make gRNA-Cas9 complexes. Cas9 protein was prepared as described previously [56]. They were incubated for 5 min at 37°C and injected into wild type D. magna eggs, as described previously [8]. To characterize the somatic mutation on Vri binding site generated by Cas9 protein, target loci were amplified by PCR from genomic DNA isolated from each injected egg. To extract the genomic DNA, injected embryos were homogenized individually in 90 μL of 50 mM NaOH with zirconia beads. The sample was heated at 95°C for 10 min, followed by a neutralization step by adding 10 μL of 1 M Tris-HCl (pH 7.5). Before this DNA extract was used as a PCR template, the sample was centrifuged at 13,000 g for 5 min. The PCR was performed with HS Ex Taq polymerase (Takara Bio) using a primer pair designed as follows: Vri-bs forward (5′-GATGTCACGAAATCTGAGGTC-3′) and Vri-bs reverse (5′-GATCTAAACACCTTGGCGTAAC-3′), which amplified 214 bp including the enhancer region. The PCR products were analyzed with native PAGE gel electrophoresis. To characterize the heritable mutagenesis, injected Daphnia were cultured separately until they produced offspring. The offspring were pooled (up to 8–10 daphniids) and genomic DNA extraction and genomic PCR were performed as mentioned above.
|
10.1371/journal.ppat.1000049 | DC-SIGN and CD150 Have Distinct Roles in Transmission of Measles Virus from Dendritic Cells to T-Lymphocytes | Measles virus (MV) is among the most infectious viruses that affect humans and is transmitted via the respiratory route. In macaques, MV primarily infects lymphocytes and dendritic cells (DCs). Little is known about the initial target cell for MV infection. Since DCs bridge the peripheral mucosal tissues with lymphoid tissues, we hypothesize that DCs are the initial target cells that capture MV in the respiratory tract and transport the virus to the lymphoid tissues where MV is transmitted to lymphocytes. Recently, we have demonstrated that the C-type lectin DC-SIGN interacts with MV and enhances infection of DCs in cis. Using immunofluorescence microscopy, we demonstrate that DC-SIGN+ DCs are abundantly present just below the epithelia of the respiratory tract. DC-SIGN+ DCs efficiently present MV-derived antigens to CD4+ T-lymphocytes after antigen uptake via either CD150 or DC-SIGN in vitro. However, DC-SIGN+ DCs also mediate transmission of MV to CD4+ and CD8+ T-lymphocytes. We distinguished two different transmission routes that were either dependent or independent on direct DC infection. DC-SIGN and CD150 are both involved in direct DC infection and subsequent transmission of de novo synthesized virus. However, DC-SIGN, but not CD150, mediates trans-infection of MV to T-lymphocytes independent of DC infection. Together these data suggest a prominent role for DCs during the initiation, dissemination, and clearance of MV infection.
| Despite the availability of an effective vaccine, measles virus (MV) is still a major cause of childhood morbidity and mortality in developing countries. Almost all non-vaccinated children catch the highly contagious virus during an MV outbreak. This suggests an efficient route for primary infection. However, the main target cells for MV replication, CD150+ lymphocytes, are barely present in the respiratory tract where MV enters the body. Here we demonstrate an alternative route of MV transmission: dendritic cells that are abundantly present in the sub-epithelial tissues of the respiratory tract may capture MV through binding to either CD150 or DC-SIGN. Although some virus particles are processed for antigen presentation, others escape from degradation. After virus capture, DCs migrate to the lymphoid tissues where they encounter CD150+ lymphocytes and transmit the virus, after which viral replication is started. Our data provide new insights into the transmission of measles virus, and suggest a dual role for DCs in the pathogenesis of measles.
| Measles is a systemic disease, caused by measles virus (MV) infection of respiratory and lymphoid tissues. MV is a member of the Paramyxoviridae family, genus Morbillivirus. The virus is highly contagious and is spread via the respiratory route [1]. Although the course and symptoms of measles are well characterized, little is known about the cellular events underlying the disease. The target cells for MV at the site of transmission and during the systemic phase of the disease are still under debate [1]–[4]. Moreover, the interaction of MV with the immune system, paradoxically resulting in induction of strong MV-specific immunity, but also immunosuppression, has not been fully clarified.
It was previously thought that MV initially infects epithelial cells of the respiratory tract, and is disseminated during viraemia by infected monocytes [3],[5]. However, these cells only express CD46, the receptor for attenuated MV strains, but do not express CD150 (SLAM), the primary receptor for wild-type MV [6],[7]. CD150 is mainly expressed on subsets of lymphocytes, thymocytes, macrophages and mature dendritic cells (DCs) [7]. Moreover, we have recently shown that lymphocytes, but not monocytes, are the predominant cells infected in vivo during measles in macaques [2]. Moreover, lymphocytes are not in large numbers present at respiratory epithelial surfaces compared to lymphoid tissues and therefore we hypothesize that other cells are the target for MV at sites of entry.
DCs are professional antigen presenting cells (APCs) that have a sentinel function in the immune system; DCs capture antigens in the periphery and, upon activation, migrate to the lymphoid tissues to present the antigens to T-lymphocytes, resulting in a pathogen-specific immune response [8]. We hypothesize that DCs mediate transmission of MV: DCs capture MV in the respiratory tract, but instead of degradation the virus is protected and transported into the lymphocyte-rich area in the lymphoid tissues, where it is efficiently transmitted to CD150+ lymphocytes. A similar role for DCs has been described for HIV-1, where DCs capture HIV-1 via the C-type lectin dendritic cell-specific ICAM-3 grabbing non-integrin (DC-SIGN) and mediate transmission of HIV-1 to T-lymphocytes by de novo production of virus or transferring the virus particles directly to the T-lymphocytes (trans-infection) [9],[10]. We have previously shown that DC-SIGN also mediates binding of MV to DCs, which enhances DC infection through CD150 in cis [11]. Moreover, in infected macaques MV-infected DCs have been observed in conjunction with infected T-lymphocytes, suggesting transmission of virus between both cell types [2].
Here we set out to investigate the role of DC-SIGN and CD150 in both antigen presentation and MV transmission by DCs. MV capture by DCs leads not only to antigen presentation but also to efficient transmission to T-lymphocytes. Both the tissue distribution and functional studies demonstrate that CD150 and DC-SIGN have distinct functions in MV transmission by DCs. The identification of their function in antigen presentation and MV transmission will lead to a better understanding of MV pathogenesis.
MV enters the body in the respiratory tract; however the initial target cells at the site of entry remain unknown. DC-SIGN and CD150 are the major receptors for wild-type MV strains, of which only CD150 can function as entry receptor. DC-SIGN is abundantly expressed by DCs in peripheral tissues, such as the dermis, foreskin, gut and cervix and on DCs and specialized macrophages in the lymphoid tissues [9], [12]–[14]. However, little is known about the expression of DC-SIGN and CD150 in the respiratory tract. Therefore we investigated the presence of DC-SIGN+ and CD150+ cells in the different respiratory tissues by immunofluorescence microscopy.
DC-SIGN was abundantly present in buccal, pharyngeal, tonsillar, tracheal and bronchial sub-epithelial tissues (Figure 1A and S1, Table 1). Similar to previous reports [15], scattered DC-SIGN+ DCs were also observed in the lungs, mainly in the interstitium of the alveoli (Table 1).
Monocyte-derived DCs (moDCs) in culture express CD150, and expression levels are increased upon maturation [16],[17]. In situ, CD150 expression was detected on cells in the sub-epithelial tissues of the upper respiratory tract, but we observed very little expression in the lower respiratory tract (Figure 1A and S1, Table 1). Although some CD150+ cells were present in the epithelia of the tonsillar crypts and the pharynx, the expression of CD150 in these tissues was low compared to that in lymphoid tissues (Figure 1A and S1, Table 1). These cells might represent macrophages or lymphocytes that can be targets for MV infection and their infection might explain why infection is observed in epithelial tissues in vivo [2]. We also observed some autofluorescence in the epithelia of the tracheal, bronchial and tonsillar epithelium, probably caused by mucus (Figure 1A and S1). In the upper respiratory tract, we rarely detected co-localization of CD150 and DC-SIGN, suggesting that DC-SIGN+ DCs, abundantly present just below the epithelia in the respiratory tract, express no or low levels of CD150 (Figure 1A and S1, Table 1).
During viraemia, MV-infected cells enter the lymphoid tissues. Here, DC-SIGN+ DCs might facilitate infection of lymphocytes, similar to HIV-1 [9]. We therefore investigated the expression of DC-SIGN and CD150 in lymph node and tonsil. As previously described [13],[15], DC-SIGN+ cells were mainly located around the medullary sinuses and in the paracortex (T-cell areas) of the lymph node as well as in the inter-follicular tissue (T-cell area) of the tonsil (Figure 1B, Table 1). However, DC-SIGN+ cells were also found in the B cell follicles of the tonsil. Notably, DC-SIGN+ cells were located in close contact with the CD150+ cells (Figure 1B). Strikingly, strong co-localisation of DC-SIGN and CD150 was observed in the medullary sinuses (Figure 1B, Table 1). Together these data suggest that DC-SIGN+ cells are not only important in the initial phase of MV infection, but might also be involved in MV infection in the lymphoid tissues during the systemic phase of the infection.
DC-SIGN is an attachment receptor for MV and mediates infection of DCs through CD150 in cis [11]. We investigated whether MV capture by DC-SIGN+ DCs also leads to antigen processing and presentation to MV-specific CD4+ T-lymphocytes. We performed an antigen presentation assay using MV-specific CD4+ T cell clones. [18],[19]. As APCs we used moDCs, expressing high levels of DC-SIGN [20] or an autologous Epstein-Barr virus-transformed B-lymphoblastic cell line (BLCL). The APCs were incubated with different dilutions of MV that was UV-inactivated to exclude DC and T-lymphocyte infection. Subsequently the APCs were co-cultured with the MV-specific CD4+ T-cell clones. At the highest concentrations, UV-MV induced DC maturation, since CD86 was upregulated, whereas DC-SIGN was down-regulated (Figure 2A). Notably, in contrast to LPS stimulation, HLA-DR was not upregulated by MV. T-cell activation was measured by the detection of IFN-γ production by ELISPOT and ELISA. We used two different MV-specific T cell clones (GRIM99 and GRIM61) that matched and one that mismatched (GRIM 76) the HLA type of the donor DCs. DCs incubated with UV-inactivated MV specifically activated the HLA-matched MV-specific T cell clones, whereas the mismatched MV-specific T cell clone was not activated (Figure 2B and C). Moreover, an irrelevant T cell clone (LB5) was not activated by the DCs (Figure 2B). Thus, MV capture by DCs leads to specific antigen processing and presentation of MV peptides in the context of MHC-class II molecules. MV-derived antigen presentation by DCs was more efficient at low antigen concentrations than presentation by autologous BLCL (Figure 2B); while the peptide control response was not significantly different (239+/−16 versus 264+/−18).
To investigate whether DC-SIGN and CD150 are involved in antigen presentation, moDCs were pre-treated with specific blocking antibodies to CD150, DC-SIGN and with the C-type lectin inhibitor mannan. Both mannan and anti-DC-SIGN inhibited activation of the MV-specific T-cell clone, demonstrating that DC-SIGN supports MV antigen uptake and processing for antigen presentation (Figure 2B and C). Notably, antibodies against CD150 also inhibited T-cell clone activation, although to a lesser extent than antibodies against DC-SIGN. The role of DC-SIGN and CD150 was not dependent on post-entry effects, since addition of the antibodies just before the T-lymphocytes were added did not affect MV antigen presentation to the CD4+ T-cell clone (data not shown). Thus, viral uptake by DC-SIGN and to a lesser extent CD150 leads to virus degradation and antigen presentation of MV-derived antigens to CD4+ T-lymphocytes.
DCs capture HIV-1 via DC-SIGN [9], and facilitate the infection of T-lymphocytes by transferring the virus through the infectious synapse [21]. Since DC-SIGN+ DCs and CD150+ lymphocytes closely interact in the lymphoid tissues and the upper respiratory tract (Figure 1), we investigated the role of DCs in MV transmission in DC-T-lymphocyte co-cultures. To analyse viral transmission of DCs to T-lymphocytes we used the recombinant MV-IC323-EGFP strain. This MV strain has similar characteristics as its parental IC-B wild-type strain [22], but infected cells produce high amounts of EGFP. The concentration of EGFP in the cells is directly related to the level of virus replication. The entry receptor for this virus is CD150, and not CD46, similar as MV wild-type strains [6]. DCs were infected with MV-IC323-EGFP, and subsequently co-cultured with PHA-stimulated T-lymphocytes expressing high levels of CD150. After two days the infected cells were analyzed by fluorescence microscopy. MV infection induced the formation of large clusters, which contained multiple EGFP+ syncytia (Figure 3A). Most infected cells were observed in clusters, and notably long EGFP+ dendritic processes were observed that interconnected these clusters (Figure 3A). To investigate which cells were present in the clusters, either DCs or T-lymphocytes were stained with a red dye before infection. Staining of either cell type demonstrated that both infected DCs and T-lymphocytes were present in the EGFP+ clusters (Figure 3B), reminiscent of the in vivo infection of DCs and T-lymphocytes observed in lymphoid tissues of macaques [2].
To investigate whether DCs enhance the infection of lymphocytes, T-lymphocytes or DC-T-lymphocyte co-cultures were infected with different concentrations of MV-IC323-EGFP and subsequently analyzed by flow cytometry. MV could readily infect activated T-lymphocytes, but addition of DCs enhanced infection two-fold (Figure 3C and D). To investigate the role of DC-SIGN in MV transmission in a DC-T-lymphocyte co-culture, DCs were pre-incubated with mannan and infection was measured. Mannan partially prevented the increased infection of the lymphocytes in the DC-T lymphocyte co-cultures, demonstrating that DC-SIGN is involved in the enhanced infection in the DC-T lymphocyte co-culture, probably by increasing viral transmission to T-lymphocytes (Figure 3D).
To investigate whether DCs transmit MV to their target cells, DCs were incubated with MV-IC323-EGFP for two hours, washed extensively to remove unbound virus and subsequently co-cultured with activated T-lymphocytes. In DC cultures without T-lymphocytes, low percentages of MV-infected DCs were detected, whereas in the DC-T lymphocyte co-cultures large clusters of MV-infected cells and syncytia were observed (Figure 4A upper panels). These data strongly suggest that DCs capture MV and transmit the virus to T-lymphocytes independently of de novo synthesis of virus by infected DCs, since only a few infected DCs were observed (Figure 4A). This process is referred to as trans-infection. However, HIV-1 studies have shown that DCs can also mediate transmission of de novo synthesized HIV-1 [10]. DCs in the respiratory tract in situ express high levels of DC-SIGN and no or low amounts of CD150 (Figure 1), suggesting that DCs are not productively infected by MV. Therefore, we investigated whether DCs mediate trans-infection. As demonstrated, DCs transmit MV efficiently to T-lymphocytes in a co-culture (Figure 4A) but de novo synthesis in DCs cannot be excluded, since both DCs and T-lymphocytes are infected in the DC-T lymphocyte co-culture (Figure 4B). To exclude de novo synthesis of virus in DCs, we used the fusion inhibitor peptide (FIP, 200µM) [23], which was added to the co-cultures 2 hours after addition of the T-lymphocytes to MV-infected DCs. We observed large clusters of EGFP+ cells in the presence of FIP (Figure 4A). FIP prevents fusion of MV with cell membranes and of membranes of MV-infected cells with those of neighbouring cells. Therefore FIP blocks infection and syncytium formation [24]. Thus, T-lymphocytes expressing EGFP must have been infected during the 2 hours co-cultivation with MV-infected DCs before FIP was added (Figure 4B). This is a time frame that excludes de novo synthesis of MV by the DCs. In contrast to the condition without FIP, no syncytium formation was observed in DCs and T-lymphocytes cultured in the presence of FIP, confirming that FIP indeed prevented fusion. Moreover, incubation of DCs with different concentrations of MV demonstrated that T-lymphocytes were the major MV-infected cell population in the DC-T lymphocyte co-culture (Figure 4C), demonstrating that DCs mediate trans-infection. To determine the efficiency of the trans-infection, the absolute number of infected cells was calculated (Figure 4D). A 6-fold higher number of T-lymphocytes compared to DCs were infected in the co-cultures. This demonstrates that trans-infection of T-lymphocytes by DC-bound MV is more efficient than cis-infection of DCs. Thus, DCs efficiently mediate transmission of MV to T-lymphocytes, and this process primarily occurs independently of de novo synthesis of MV.
DC-SIGN and CD150 are both important for binding of MV to DCs and subsequent infection. Indeed, MV-infection of DCs is inhibited by antibodies against CD150 and DC-SIGN, as well as by mannan (Figure 5A) [11]. MV transmission by de novo synthesis of MV particles depends on infection of DCs and therefore these data suggest that transmission through de novo synthesis of virus is dependent on both CD150 and DC-SIGN.
To investigate whether DC-SIGN is involved in MV trans-infection of T-lymphocytes, DCs were pre-treated with mannan or antibodies against DC-SIGN and transmission was measured in the presence of FIP. In all donors, both mannan and antibodies against DC-SIGN inhibited trans-infection (Figure 5B), although donor variations were observed. Direct infection of T-lymphocytes after pre-treatment with the blocking agents demonstrated that these compounds do not interfere with the infection of the target cells (Figure 5B). Allogeneic DCs induce T-lymphocyte activation and as such might increase CD150 expression and subsequently MV infection of T-lymphocytes. Therefore, we used autologous T-lymphocytes for donors #5–7. In this setting, trans-infection was also observed and was dependent on DC-SIGN (Figure 5B), indicating that trans-infection is independent of T-lymphocyte activation. To analyze whether these differences were significantly different throughout the donors, data were normalized to the medium condition (Figure 5C), demonstrating that DC-SIGN is important for transmission of MV from DCs to T-lymphocytes (Figure 5C).
Since CD150 is important for MV binding to DCs and subsequent infection, we investigated whether this receptor is also important for trans-infection of T-lymphocytes. Strikingly, antibodies against CD150 did not block the trans-infection and even slightly increased the transfer to T-lymphocytes (Figure 5D). These results demonstrate that DC-SIGN mediates both DC infection in cis and trans-infection, whereas CD150 is only involved in infection, and thus CD150 and DC-SIGN have distinct roles in MV transmission by DCs to T-lymphocytes.
In peripheral blood of experimentally infected macaques, MV infection of both CD4+ and CD8+ T-lymphocytes was observed [2]. Therefore we investigated whether DCs mediate transmission of MV to both T-lymphocyte subsets in vitro. CD4+ and CD8+ T-lymphocytes were purified from PHA-stimulated human peripheral blood mononuclear cells (PBMCs), and both expressed high levels of CD150 (Figure 6A). DCs were infected with MV-IC323-EGFP, and after extensive washing, co-cultured with either CD4+ or CD8+ T-lymphocytes. In both co-cultures, EGFP expression was observed in large clusters containing EGFP+ syncytia, demonstrating that DCs mediate transmission to both T-lymphocyte subsets (Figure 6B). To measure whether DCs mediate trans-infection, FIP was added two hours after addition of the T-lymphocytes to the MV-infected DCs. Trans-infection to both subsets is efficient and mediated by DC-SIGN, since pre-treatment of DCs with mannan inhibited infection of T-lymphocytes in both cultures (Figure 6C). Thus, DCs transmit MV to both CD4+ and CD8+ T-lymphocytes.
Two hallmarks of measles are that the virus is highly contagious and infection results in strong MV-specific cellular immune responses. Both viral transmission and antigen presentation should therefore occur through efficient and robust processes. Since DCs are professional APCs and have been demonstrated to mediate transmission of several viruses, we have investigated the role of DCs in both processes for measles. Here we have shown that a subset of DCs, the DC-SIGN+ DCs, mediates both transmission and antigen presentation of MV to T-lymphocytes. The receptors DC-SIGN and CD150 are both involved in DC infection and antigen presentation, whereas only DC-SIGN is involved in MV trans-infection of T-lymphocytes.
DC-SIGN+ DCs have previously been identified in several sub-epithelial and lymphoid tissues [9],[13]. Here, we have identified DC-SIGN+ DCs in the sub-epithelial tissues of the human mouth, pharynx, trachea and bronchi, and the presence of scattered DC-SIGN+ cells in the lung. CD150 is expressed on in vitro cultured macrophages and DCs and is increased upon maturation [16],[17],[25]. However, CD150 was previously not detected on immature DCs in skin and lung in situ [26] and we could not detect CD150 on DC-SIGN+ DCs in respiratory epithelia, suggesting that these DCs are not susceptible to MV infection and DCs capture MV through DC-SIGN. However, expression of low levels of CD150 that might support MV infection cannot be excluded using immunofluorescence microscopy. DC-SIGN strongly enhances infection of DCs through CD150 [11], and therefore low levels of CD150 might be enough for efficient infection of DCs.
Different mechanisms are involved in virus transmission by DC-SIGN, since HIV-1 capture by DC-SIGN can result both in cis and trans-infection [9],[27]. However, DC-SIGN binding of HIV-1 also leads to virus degradation and presentation in the context of MHC [28],[29]. Interestingly, we observed that both DC-SIGN and CD150 are involved in MV processing and presentation of MV-derived peptides to the MV-specific CD4+ T-cell clone GRIM99. B cells are able to present MV antigens to T cells [30]. BLCL express high levels of CD150, but no DC-SIGN and MV capture leads to virus degradation and presentation of MV peptides to the autologous CD4+ T-cell clone and this was previously demonstrated to be dependent on endocytosis [30]. Notably, DCs were more efficient in antigen processing and presentation than the autologous BLCL and antibodies against DC-SIGN inhibited antigen presentation to a larger extent than antibodies against CD150. Thus, although both DC-SIGN and CD150 are involved, DC-SIGN is more important for antigen presentation of MV by DCs.
MV is a highly contagious virus [1] suggesting that it has a very efficient entry into the respiratory tissues. However, the number of CD150+ target cells in the respiratory tract is low, suggesting that these are not the first targets for MV at the site of entry. In contrast, the tissue is lined with DC-SIGN+ DCs and these DCs are better candidates since DC-SIGN+ DCs efficiently capture MV and mediate transmission of MV to T-lymphocytes in vitro. MV transmission to non-stimulated lymphocytes by DCs was inefficient (data not shown), which is in line with a previous report that demonstrated that transmission of MV from DCs to T cells isolated from blood is inefficient [31]. This is probably due to low expression of CD150 on blood-derived lymphocytes. However in lymphoid tissues, where DCs migrate to, CD150 is highly expressed on T and B-lymphocytes and on monocytes [7],[32].
Recently it was demonstrated that DCs specifically transmit HIV-1 to HIV-1-specific T-lymphocytes [29] suggesting that immunological synapse formation enhances viral transmission, due to prolonged interactions during antigen presentation and T cell activation. However, we did not observe differences between transmission to autologous- and allogeneic T-lymphocytes, suggesting that prolonged immunological synapse interactions occurring during allogeneic but not autologous T-DC interactions are not necessary for MV transmission. This might be due to differences in infectivity between both viruses.
MV transmission can occur independently of de novo synthesis of virus in DCs (referred to as trans-infection). Using specific blocking antibodies, we have demonstrated that trans-infection is dependent on DC-SIGN but not on CD150. This is physiologically relevant, since DC-SIGN+ DCs in the respiratory tract express no or low levels of CD150, and are therefore not susceptible to MV infection. Both DC-SIGN and CD150 are important for binding of MV to DCs [11]. However, our data show that binding of MV to DC-SIGN and CD150 results in different internalization pathways. Although both DC-SIGN and CD150 lead to virus degradation for antigen presentation, CD150 binding also results in viral entry, whereas only the interaction of DC-SIGN with MV leads to viral protection for trans-infection. Indeed, inhibition of CD150 resulted in enhanced trans-infection due to less degradation or fusion and therefore increasing the amount of virus for the DC-SIGN-mediated transmission. Several donors were tested for the involvement of DC-SIGN in trans-infection of MV. The contribution of DC-SIGN varied between the donors, suggesting that another attachment receptor might play a role, such as syndecan-3 for HIV-1 [33]. In macaques, both CD4+ and CD8+ T-lymphocytes are infected during the viraemic phase of measles disease [2]. It is unclear whether viral transmission can also occur to CD8+ T lymphocytes. Interestingly, we observed MV transmission by DCs to both CD4+ and CD8+ T-lymphocytes, indicating the formation of an infectious synapse between DCs and CD8+ T-lymphocytes, similar to DCs and CD4+ T-lymphocytes [21].
During MV infection in macaques, lymphoid tissues are major sites of MV replication [2]. In human lymphoid tissues, CD150+- and DC-SIGN+ cells are in close contact, which can contribute to massive replication of MV. This is supported by our in vitro observations that DC-SIGN+ DCs enhance infection of T-lymphocytes in co-cultures and MV infection is predominantly observed in the clusters of DCs and T-lymphocytes. Notably, long EGFP+ dendrites were frequently observed between clusters, suggesting that viruses spread between clusters through these dendrites. Although these dendrites might be a MV-specific effect, which facilitates virus spread, these dendritic processes might also have a physiological function in the immune system, such as the interchange of antigenic information. The fact that MV infection in DC-T lymphocyte co-cultures was observed in clusters was highly reminiscent to the infectious foci that we have previously observed in tissues of infected macaques [2]. This pattern and enhanced infection in DC-T lymphocyte co-cultures suggest that the virus is much more efficiently transmitted by direct cell-cell contact than as cell-free virus.
In contrast to the APCs in the peripheral tissues, the borders of the medullary sinuses of the lymph node contain a population of cells that express high amounts of both DC-SIGN and CD150. These DC-SIGN+ cells have been previously shown to express CD68 and lack the expression of DEC205, suggesting that these cells are medullary macrophages [13],[34]. These cells are in contact with the lymph and therefore encounter tissue-derived antigens. Based on the high expression of both CD150 and DC-SIGN, and the fact that DC-SIGN can enhance infection in cis [11], it is likely that these cells become infected during measles and might contribute to further virus transmission.
In conclusion, these data provide us with an alternative view on how MV might disseminate from the site of infection to their main target cells, the lymphocytes: DC-SIGN+ DCs, which are abundantly present in the sub-epithelial tissues of the respiratory tract, capture MV and process the virus for antigen presentation, but a part of the virus escapes from degradation. In previous studies dendrites of sub-epithelial DCs have been shown to pass the tight junctions of the epithelium of the gut and respiratory tract and sample the mucosal surfaces [35], which could result in efficient capture of the virus. Moreover, MV induces activation of the DC via TLRs [36], which will induce migration of DCs from the peripheral tissues to the lymphoid tissues. Although DCs might encounter CD150+ cells in the underlying mucosal tissues, the abundant expression of CD150 in lymphoid tissues strongly suggests lymphoid tissues as the major site of MV transmission and replication. A similar mechanism might play a role in spreading the virus throughout the body, even to privileged tissues such as the brain, and could therefore be involved in complications such as subacute sclerosing panencephalitis (SSPE). Moreover, during viraemia, DCs might increase infection and tissue destruction. In the future, in vivo studies will be required to prove the importance of MV transmission by DCs.
The following antibodies were used: CD150-specific mouse antibody 5C6 [37], DC-SIGN-specific mouse antibodies AZN-D1 and AZN-D2 [9], DCN46 and DCN46 conjugated with PE (BD Pharmingen, San Diego, CA, USA), CD3-specific mouse antibody SK-7 conjugated with PerCP (BD Pharmingen, San Diego, CA, USA), goat anti-mouse IgG antibody conjugated with PO (Jackson Immunoresearch, West Grove, PA, USA), HLA-DR- (Immu357) and CD86- (HA5.2B7) specific mouse antibodies conjugated with PE (Immunotech, Marseille, France), goat anti-mouse antibody conjugated with FITC (Zymed Laboratories Inc., South San Fransisco, CA. USA), Alexa488- or Alexa594-labeled anti-mouse antibodies (Molecular probes, Eugene, OR, USA).
Vero-CD150 cells [38] were grown in Dulbecco's Modified Eagle's Medium (DMEM; Gibco Invitrogen, Carlsbad, CA, USA) supplemented with 4500 mg/L glucose; 110 mg/l sodium pyruvate; 4 mM L-glutamine; 10% heat-inactivated fetal calf serum (FCS); 20 U/ml penicillin and 20 µg/ml streptomycin, at 37°C with 5% CO2. The CD4+ HLA-DQw1-restricted T-cell clone GRIM99 [18],[19] recognizes an epitope in the MV fusion protein (EVNGVTIQV). GRIM61 and GRIM99 are also MV-specific CD4+ T cell clones, for which the epitopes have not been mapped (Van Binnendijk JI 1989). LB5 is a CD4+ T cell clone of unknown specificity (unpublished). All clones were cultured in RPMI-1640 (RPMI 1640, Gibco Invitrogen, Carlsbad, CA, USA) supplemented with 4 mM L-glutamine; 10% heat-inactivated human AB serum (Sigma-Aldrich, St. Louis, MO, USA); 20 U/ml penicillin, 20 µg/ml streptomycin and 10−5 M 2-mercapto-ethanol in 96-well round bottom plates. Epstein-Barr virus-transformed B-lymphoblastic cell line (BLCL-GR) [18],[19] was used as autologous APC, and was cultured in RPMI 1640 supplemented with L-glutamine, penicillin, streptomycin and 10% FCS. Immature moDCs were cultured as described before [39]. In short, human blood monocytes were isolated from buffy coats by Ficoll density centrifugation, followed by selection of CD14+ cells using magnetic beads (MACS, Milteny Biotec GmbH, Bergisch Gladbach, Germany). Purified monocytes were cultured in RPMI-1640 medium supplemented with 4 mM L-glutamine; 10% FCS; 20 U/ml penicillin and 20 µg/ml streptomycin and differentiated into immature moDCs in the presence of IL-4 and GM-CSF (500 and 800 U/ml, respectively; Schering-Plough, Brussels, Belgium).
PBMCs were isolated from buffy coats by Ficoll density centrifugation, activated with phytohemagglutinin (3 µg/ml; Sigma-Aldrich, St. Louis, MO, USA), and cultured in complete RPMI-1640 medium. At day 3 the cells were washed and cultured with IL-2 (100 units/ml). As determined by flow cytometry >80% of the activated PBMC were CD3+ and therefore these cells are referred to as T-lymphocytes throughout the text. The CD4+ and CD8+ T-lymphocytes were enriched at day 3 after PHA stimulation by negative selection using MACS beads. To label DCs and T-lymphocytes, the cells were stained with the PKH26 red fluorescent cell linker kit (Sigma-Aldrich, St. Louis, MO, USA) for general cell membrane labelling, according to the manufacturers protocol.
The recombinant MV strain IC323-EGFP [40] was propagated on Vero-CD150 cells. For virus production, Vero-CD150 cells were infected with a multiplicity of infection (MOI) of 0.01 in DMEM supplemented with 2% of FCS. After 90 minutes cells were washed to remove unbound virus and were subsequently grown in DMEM supplemented with 10% FCS. Cells and supernatant were harvested when 80% cytopathic effect was observed. To release cell-bound virus, the cells were sonicated (3 times, 10 seconds, Sonicor Instrument Corporation, Copiaque, N.Y., USA). The cells were centrifuged (10 minutes, 1000 g) and the supernatant was snap-frozen in liquid nitrogen before titration on Vero-CD150 cells. The titer of the virus-stock used was 1×106 TCID50/ml. Purified MV Edmonston with a concentration of approximately 1 mg/ml, was inactivated by UV-irradiation (30 minutes, 15W 312 nm) and is referred to as UV-MV throughout the text.
Tissues of healthy human donors were obtained through the pathology department of the VU University Medical Center, according to the institutional ethical guidelines. Cryosections (7 µm) were fixed with 100% acetone and stained with primary antibody combinations against DC-SIGN (DCN46, IgG2b, 10 µg/ml) and anti-CD150 (5C6, IgG1, 10 µg/ml) or a buffer control for 18 hours at 4°C. Sections were counterstained with isotype-specific Alexa488- or Alexa594-labeled anti-mouse antibodies. Nuclei were stained with Hoechst (Molecular Probes, Eugene, OR, USA). After mounting, sections were examined with a Nikon Eclipse E800 fluorescence microscope and recordings were made with a digital NIKON DXM1200 camera. Two persons used the photographs to quantify the staining in the different tissues independently. To determine the density of DC-SIGN+- or CD150+ cells, the number of positive cells was divided by the total number of cells, based on the nuclei staining. To determine the co-localization, the double-stained cells were divided by the total number of stained cells. Based on the control sections, autofluorescence was often observed in the lower respiratory tract and is indicated in the pictures.
Monocytes were isolated from an HLA-DQw1-matched donor using CD14 MACS beads and differentiated into immature DCs as described above. Subsequently, these DCs or autologous BLCL-GR (5×103 cells) were used as APCs in an interferon-γ (IFN-γ) ELISPOT assay as previously described [41]. Briefly, APCs were plated into 96-well v-bottom plates in complete RPMI-1640 containing IL-4 and GM-CSF and pre-incubated with mannan (0,25 mg/ml) for 30 minutes at 37°C. Next, the cells were incubated with different dilutions of UV-MV at 37°C or a positive control peptide (EVNGVTIQV; 1 µM). After overnight incubation the CD4+ T-cell clones were added to the APCs (5×103 cells per well), the plates were briefly centrifuged (1 minute, 300 g) and subsequently incubated at 37°C for 1.5 hour. Subsequently the cells were transferred to nylon bottom plates (Nalge Nunc International, Rochester, NY) coated with a monoclonal antibody specific for human IFN-γ (1-D1K; Mabtech, Stockholm, Sweden), and incubated at 37°C for five hours. Finally, plates were washed with phosphate-buffered saline (PBS) containing 0.05% Tween 20 (Merck, Darmstadt, Germany). Spots were visualized by incubation with a secondary biotinylated mAb against IFN-γ (7-B6-1; Mabtech), followed by staining with streptavidin–alkaline phosphatase (Mabtech), and nitroblue tetrazolium–5-bromo-4-chloro-3-indolyl-phosphate (Kirkegaard & Perry Laboratories, Gaithersburg, MA, USA). Finally, the color reaction was stopped by washing the plates with water and spots were counted with an automated ELISPOT reader (automated ELISAspot assay video analysis systems; distributed by Sanquin Reagents, Amsterdam, The Netherlands).
In parallel, the same APCs were also used to stimulate the same T-cell clone for IFN-γ production in supernatant. Briefly, APCs (1×104 cells) were used to stimulate the T-cell clone (3×104 cells) in round-bottom plates, and were incubated at 37°C for 24 hours before supernatants were harvested. To determine the contribution of DC-SIGN and CD150, the APCs were pre-incubated with mannan (0,25 mg/ml), anti-DC-SIGN (AZN-D2; 20 µg /ml) or anti-CD150 (5C6; 20 µg/ml) for 30 minutes at 37°C. The IFN-γ concentrations in the supernatants were determined by ELISA (Biosource International, CA, USA).
For infection and transmission assays DCs (5×104 cells) were seeded in a V-bottom plate and pre-incubated with mannan (0,25 mg/ml), anti-DC-SIGN (AZN-D2; 20 µg/ml) or anti-CD150 (5C6; 20 µg/ml) for 30 min. at 37°C, before incubation with MV-IC323-EGFP at 37°C for 2 hours (5×104 TCID50, unless stated otherwise). After 2 hours the cells were washed and transferred to a flat-bottom plate. To measure transmission, activated T-lymphocytes (2×105 cells) were added. If indicated the fusion inhibitory peptide (FIP) Z-d-Phe-L-Phe-Gly-OH (Z-FFG; 0.2 mM; Bachem, Heidelberg, Germany) was added 2 hours later. For co-culture assays T-lymphocytes (2×105 cells), either or not together with DCs (5×104 cells), were pre-incubated with 0,25 mg/ml mannan and infected with different concentrations of MV-IC323-EGFP.
All cells were cultured for three days at 37°C in complete RPMI-1640 containing IL-4 and GM-CSF. The cultures were monitored using a Leica DMIL fluorescence microscope, and pictures were taken using a Leica DFC 320 camera (Leica Microsystems, Wetzlar, Germany). At day 3 the cells were harvested, washed and fixed with 4% PFA and EGFP expression was measured by flow cytometry. DCs had higher autofluorescence compared to T-lymphocytes the EGFP+ gate for both population was set at the uninfected control sample. To determine the infection of specific cell populations, the cells were stained with directly labeled antibodies against DC-SIGN and CD3 before analysis. The absolute number of infected DCs (DC-SIGN+/EGFP+) and T cells (CD3+/EGFP+) of the total counted sample by flow cytometry ( = 105 events) were used to determine the efficiency of transmission.
To determine the variation in MV transmission among the different DC donors, the infection of T-lymphocytes was normalized to the “medium” control condition, which was set at 100%. Subsequently, the percentages of infection from the different donors were used in a one-way analysis of variance (ANOVA). When the overall F test was significant, differences among the donors were further investigated with the post hoc Bonferroni test using Graphpad Prism software. A probability of p<0.05 was considered statistically significant. |
10.1371/journal.pgen.1000401 | Phylogenomics of Unusual Histone H2A Variants in Bdelloid Rotifers | Rotifers of Class Bdelloidea are remarkable in having evolved for millions of years, apparently without males and meiosis. In addition, they are unusually resistant to desiccation and ionizing radiation and are able to repair hundreds of radiation-induced DNA double-strand breaks per genome with little effect on viability or reproduction. Because specific histone H2A variants are involved in DSB repair and certain meiotic processes in other eukaryotes, we investigated the histone H2A genes and proteins of two bdelloid species. Genomic libraries were built and probed to identify histone H2A genes in Adineta vaga and Philodina roseola, species representing two different bdelloid families. The expressed H2A proteins were visualized on SDS-PAGE gels and identified by tandem mass spectrometry. We find that neither the core histone H2A, present in nearly all other eukaryotes, nor the H2AX variant, a ubiquitous component of the eukaryotic DSB repair machinery, are present in bdelloid rotifers. Instead, they are replaced by unusual histone H2A variants of higher mass. In contrast, a species of rotifer belonging to the facultatively sexual, desiccation- and radiation-intolerant sister class of bdelloid rotifers, the monogononts, contains a canonical core histone H2A and appears to lack the bdelloid H2A variant genes. Applying phylogenetic tools, we demonstrate that the bdelloid-specific H2A variants arose as distinct lineages from canonical H2A separate from those leading to the H2AX and H2AZ variants. The replacement of core H2A and H2AX in bdelloid rotifers by previously uncharacterized H2A variants with extended carboxy-terminal tails is further evidence for evolutionary diversity within this class of histone H2A genes and may represent adaptation to unusual features specific to bdelloid rotifers.
| Bdelloid rotifers are microscopic animals common in ephemeral freshwater environments throughout the world. They are unusual not only because they have been reproducing without males for millions of years, but also because they can survive long periods of complete desiccation at any life stage and exposure to levels of ionizing radiation that cause hundreds of DNA double strand breaks per genome. Canonical histones (H2A, H2B, H3, and H4) are highly conserved proteins that package DNA in the nucleus and are involved in the regulation of chromatin metabolism. Because the conserved histone variant of canonical H2A, H2AX, is involved in the repair of DNA double strand breaks, we tested the possibility that bdelloid H2A histones are unusual. Strikingly, we find that bdelloids lack both H2A and H2AX, the absence of which is in contrast to their ubiquitous presence in other eukaryotes. Instead, we find that bdelloid rotifers replaced their canonical H2A protein by H2A variants not found in any other eukaryote. These results gain particular interest in view of the extreme resistance of bdelloid rotifers to desiccation and ionizing radiation and their attendant ability, possibly unique among metazoans, to repair massive levels of DNA breakage.
| Rotifers of Class Bdelloidea are freshwater invertebrates of widespread occurrence that have attracted particular interest because their apparent lack of both males and meiosis suggests they are ancient asexuals [1]. Bdelloids are also of interest because of their extraordinary ability to survive and continue reproduction after desiccation at any life stage [2],[3] and doses of ionizing radiation that cause hundreds of DNA double-strand breaks (DSBs) per genome [4]. In contrast, rotifers of their sister class, the facultatively sexual monogonont rotifers, can survive desiccation only at a specific stage in their life cycle, as resting eggs [5],[6], and are not unusually resistant to ionizing radiation [4]. Since the H2A histones and some of their variants are well conserved in other eukaryotes and are involved in DSB repair and in certain meiotic processes [7], we speculated that bdelloid H2A histones may be unusual. In order to test this expectation, we investigated histone H2A genes and proteins of two species of bdelloid rotifers, Adineta vaga (A. vaga) and Philodina roseola (P. roseola), which represent two distantly related families within the Class.
Histones are architectural proteins that package eukaryotic DNA into nucleosomes, being essential in the maintenance, expression, and replication of the genome. The genes coding for the four canonical histones, H4, H3, H2B and H2A, which make up the nucleosome, are expressed during the S-phase of the cell cycle when the nuclear DNA is synthesized and are clustered in most metazoan genomes. These replication-dependent histone genes typically do not contain introns in animals and their mRNAs represent the only known cellular mRNAs that are not polyadenylated, ending instead in a highly conserved stem loop. The processing step in replication-dependent histone mRNA biosynthesis is a 3′-end endonucleolytic cleavage between the stem-loop and a so-called histone downstream element (HDE) [8]–[11]. In contrast, replication-independent histone genes encode variants of the canonical histones that are found outside of histone gene clusters and are expressed throughout the cell cycle. These variant histone genes may contain introns and their transcripts are polyadenylated [12].
Of the histone proteins, the H2A family includes the largest number of described variants and displays the greatest degree of diversity in carboxy-terminal tail length and sequence [12],[13]. One of these variants, H2AX, of wide occurrence in eukaryotes, is characterized by a unique and invariant C-terminal SQ(E/D)Φ-(end) motif, where Φ indicates a hydrophobic residue. It has been demonstrated, primarily in yeast and mammals [14],[15], that this highly conserved motif of H2AX is a consensus sequence for serine phosphorylation by PI3 kinases and that the serine residue is always located four amino acids from the C-terminal residue. In response to DSBs, H2AX becomes phosphorylated at this serine over large regions (∼2 Mb) surrounding the sites of breakage. Although the mechanistic implications of phosphorylation of the SQ(E/D) motif in H2AX are not fully understood, it is apparently required for normal DSB repair throughout the eukaryotic kingdom, being involved in the retention and accumulation of repair and checkpoint proteins to DNA breaks [for review], [ see 7], [14]–[19]. Moreover, it has been demonstrated in several eukaryotes that phosphorylated H2AX also plays a role in meiotic processes, including repair of meiotic DNA breaks made by SPO11 [20]–[23], prophase meiotic sex chromosome inactivation [23]–[25] and telomere movement [26]. With the exception of the nematode Caenorhabditis elegans, which lacks H2AX, it is ubiquitous throughout eukaryotes while the fruit fly Drosophila melanogaster has a H2AZ/H2AX chimeric H2A named H2AvD [16].
Considering the extreme radiation resistance of bdelloid rotifers [4] and the likelihood that, as in Deinococcus radiodurans, such resistance is an adaptation to repair and survive damage associated with desiccation including extensive DNA breakage [27],[28], one might expect in these organisms a high percentage of nucleosomal core H2A to be replaced by H2AX. For example in Saccharomyces cerevisiae, canonical H2A is replaced by H2AX and high levels of homologous recombination (and thus double strand breaks) occur [29]. Instead, we found that none of the H2A genes in the two bdelloid species have the H2AX-defining SQ(E/D) motif two amino acids from the C-terminal end and none of the bdelloid H2A genes is similar to canonical H2A. The absence of H2A and H2AX in bdelloid rotifers contrasts with their ubiquitous presence in other eukaryotes and with the presence of canonical H2A in the monogonont rotifer Brachionus plicatilis (B. plicatilis). The three different types of H2A genes we found in bdelloid rotifers are apparently unique to bdelloids and form distinct lineages that evolved from canonical H2As. We also found that the regions of the genome containing histone gene clusters are organized as two co-linear pairs, consistent with the degenerate tetraploidy of bdelloid rotifers [30],[31], with one pair lacking an H2A gene in the cluster while the other cluster contains an H2A gene, designated H2Abd, that has an unusual C-terminal tail. Although present in a cluster containing the canonical H4, H3 and H2B genes, it is not H2Abd that is highly expressed under normal conditions in the nucleosomes of both bdelloid species; instead, the principal H2A found in the nucleosomes is an unusual H2A variant coded by a gene designated H2Abd1 that is not located in the histone gene cluster.
It seems reasonable to speculate that the various unusual features of bdelloid H2A histones are associated with the adaptation of bdelloids to survive desiccation and perhaps also with their lack of meiosis.
Primary genomic fosmid libraries of A. vaga and P. roseola were separately probed with PCR amplification products obtained by using primers based on highly conserved regions of the canonical H3 and H2A histone genes (indicated in Figure 1B). Individual fosmids hybridizing to both probes should contain the clustered histone H3 and H2A genes while fosmids hybridizing only to the H2A probe would be expected to contain the H2AX variant or any other non-clustered H2A variant that may be present and which, like H2AX, has a primary sequence similar to that of canonical H2A. Fosmids that hybridized to either or both probes (∼120 in P. roseola and ∼225 in A. vaga) were tested by PCR and by direct sequencing, leading to the isolation of ∼80 fosmids from P. roseola and ∼180 fosmids from A. vaga containing histone H3 and H2A genes.
All fosmids from both bdelloid species containing canonical H3 fell into one of four categories, each coding for the same highly conserved H3 amino acid sequence (Figure S1), but clearly distinguishable at synonymous sites. One fosmid of each category from both A. vaga and P. roseola was fully sequenced (∼35 kb) and annotated giving contigs Avhis-1 (EU652315), Avhis-2 (EU850438), Avhis-3 (EU652316), Avhis-4 (EU850439) and Prhis-1 (EU850440), Prhis-2 (EU652317), Prhis-3 (EU652318) and Prhis-4 (EU850441) respectively (Figure 1A, Figure S1) [see also 30]. Two of these fosmids from each species, designated co-linear pair A, contain genes for the canonical histones H4, H3, H2B and a variant of H2A, and are highly similar. Fosmids of the other pair found in both species and designated co-linear pair B, are also highly similar to one another but lack the H2A gene (Figure 1A, Figure S1). Pair B is considerably diverged from pair A (Ks ca 45 percent) and some non-histone genes present in each pair are not present in the other. This pattern is consistent with the degenerate tetraploidy of bdelloid genomes [30],[31]. The H2A gene found in half of the histone clusters in both A. vaga and P. roseola and designated H2Abd, has a long and unique carboxy-terminal tail with a sequence unlike the C-terminal tail of any canonical or variant H2A known in other metazoans (Figure 1B, Genbank accession numbers for the two nucleotide copies in A. vaga EU853686, EU853685 and in P. roseola EU853693, EU853694).
Other fosmids in both bdelloid species were found to contain histone H2A genes not in clusters, but also with a uniquely long carboxy-terminal tail and are designated variants H2Abd1 and H2Abd2. The H2A gene H2Abd1 is present in four copies in both A. vaga and P. roseola, in two co-linear pairs of contigs, only one of which includes a gene for H2B (Figure 1A) and with about 50 percent synonymous divergence between gene copies in different pairs, again consistent with degenerate tetraploidy [30],[31] (Genbank accession numbers for the nucleotide copies a, b, c, d in A. vaga EU853687 to EU853690 and in P. roseola EU853695 to EU853698). H2Abd2, the third H2A variant, also found in both bdelloid species, is not clustered with any histone gene, and is present only as two closely similar copies within a co-linear pair of contigs (Figure 1A, Genbank accession numbers for the two nucleotide copies a and b in A. vaga EU853691, EU853692 and in P. roseola EU853699, EU853700).
The bdelloid histone H2A variants and the canonical H2A genes we found in the monogonont rotifer B. plicatilis are aligned along with the canonical H2A genes and H2A variants of other eukaryotes in Figure 1B. Only the variants H2AX and H2AZ, found in most eukaryotic lineages are represented in the alignment. The macroH2A and Barr-body deficient H2A (H2A Bbd) variants are not included because they are vertebrate-specific [13]. All three types of bdelloid H2A genes, H2Abd, H2Abd1 and H2Abd2, code for C-terminal amino acid sequences extending 28–43 amino acid residues beyond the canonical LLPKK motif and are typically longer than those found in other metazoans, with the exception of human macroH2A (Figure 1B and Table S1). Interestingly, none of the bdelloid-specific H2A C-terminal tails resemble those of other canonical H2As represented in Genbank and all lack the SQ(E/D)Φ-(end) motif characteristic of H2AX, indicating that both of these highly conserved proteins are absent from bdelloid rotifers.
Since all H3 and H2A-containing fosmids were examined and the same three H2A variant genes, in the same organization, were found in both A. vaga and P. roseola, it is likely that we have identified all copies of the H2A genes containing a canonical H2A core. The canonical H2A genes we found in the monogonont rotifer B. plicatilis closely resemble the canonical H2A genes present in most eukaryotes and differ substantially from the bdelloid-specific H2A variants (Figure 1B). The presence of all three H2A variants in species representing two different bdelloid families [32] but not in the monogonont suggests that they are characteristic of the entire class Bdelloidea and have arisen after the separation of bdelloids and monogononts but before the bdelloid radiation.
In order to confirm the absence of canonical histone H2A, and to determine which histone H2A replaces it in the nucleosomes of P. roseola and A. vaga, we compared the histones in the nucleosomal fraction of both bdelloid species with those of the monogonont rotifer B. plicatilis and human HeLa cells by denaturing gel electrophoresis (SDS-PAGE - Figure 2) and liquid chromatography-tandem mass spectrometry (LC-MS/MS) of peptides following enzymatic digestion.
Except for the bdelloid H2A, all rotifer histones displayed an electrophoretic mobility indistinguishable from to that of their human homologs. The identity of rotifer histones was further verified by LC-MS/MS, confirming the presence of the canonical histone proteins H4, H2B and H3 in bdelloid nucleosomes and all four canonical histones in monogononts. The region of the gels between rotifer H4 and H2B and the three prominent bands of mass greater than that of rotifer H3 (Figure 2) were also examined by LC-MS/MS. Only the first prominent band of greater mass than H3 from each bdelloid proved to be a histone (band H2Av in Figure 2). This band was identified as H2Abd1 by mass-spectrometric analysis of carboxyl terminal tails. Such detailed analysis also identified peptides coded by each of the four copies of the gene H2Abd1 in both bdelloids as well as a minor quantity of peptides corresponding to H2Abd from A. vaga (an example for H2Abd1 of P. roseola is given in Figure S2). No peptides from either bdelloid species coded by H2Abd2 were detected by LC-MS/MS.
The substantial mass difference observed in Figure 2 between band H2Av of A. vaga and that of P. roseola is consistent with the difference in C-terminus tail length and the calculated mass for proteins coded by H2Abd and H2Abd1 in A. vaga and P. roseola (Table S1).
The results of the SDS-PAGE/LC-MS/MS analysis corroborate the findings from the genomic sequence data: canonical H2A is absent from both bdelloid species and is replaced by the variant H2Abd1 and, at a much lower level, by H2Abd.
There is little similarity in amino-acid sequence among the three different H2A variant C-terminal tails in a given bdelloid species (Figure 1B). There is also considerable interspecies amino-acid difference between C-terminal regions of the same variant although a few short motifs are conserved. The C-tails of the histones coded by the four copies of H2Abd1 in both A. vaga and P. roseola were aligned and represented in Logos format with the ‘tallest’ residues representing the most conserved amino acids (Figure 3B). Within this C-terminal tail the framed block resembles a putative four-residue long S(T)PK(R)K(R) class of minor groove DNA binding motifs in which P has a strict position at i+1 [33]. A variety of these SPKK motifs have been found in termini of histone H1 and in the N-terminal tail of sea urchin histones H2B [34],[35] but also in the N-terminal tail of Drosophila centromeric H3 [33]. Until our study, its presence in the C-terminal of H2A was known only in plants [36]. In all these instances the SPKK motifs mediate histone interactions with linker DNA in the minor groove. The P. roseola H2Abd1 C-terminal tails contain such a motif while in A. vaga a variation of it appears to be present (Figure 3B).
The amino acid alignment of the two copies of both H2Abd and H2Abd2 in A. vaga and P. roseola is given instead of the Logos representation (Figure 3A, 3C). Although no similar SPKK class of DNA binding motifs was found in these variants, there are several conserved amino acids.
Excluding the distinctive C-terminal tails (starting at the arrow in Figure 1B), the bdelloid H2A genes all have an amino acid sequence closely similar to that of the canonical eukaryotic H2A. The H2A residues involved in histone-histone interactions, such as those of loop 1 that interacts with the other histone H2A and those of the docking domain that contacts the H3-H4 dimer, are well conserved between the three bdelloid H2A variants and canonical H2A (Figure 1B). Such conservation is also characteristic of H2AX, while all other H2A variants differ substantially from the canonical H2A [29],[37]. H2AZ diverges specifically in three different regions including the above histone-histone interaction zones (see Figure 1B). The alignment of Figure 1B suggests that the three unusual bdelloid H2A variants are closely related to canonical H2A over the entire histone fold domain (see also the phylogenetic analysis below).
We also examined the characteristics of the DNA sequence beyond the stop codon of the different bdelloid histone H2A variants and the canonical H3, H4 and H2B genes. Typically, only the replication-dependent histone genes expressed during DNA synthesis have a characteristic, unique 16-nucleotide stem-loop sequence in the 3′ untranslated region [8]–[11]. In contrast, all of the bdelloid histone genes depicted in Figure 1A except H2Abd2 have the same 16-nucleotide stem-loop sequences 40–80 bp beyond the stop codon (Figure 3D and Table S1). The stem loop is similar to those in other metazoans, consisting of a four-nucleotide loop (CTTT) and a six base-pair stem (Figure 3D). In addition to the 16 nucleotides of the stem-loop structure, there is a high degree of conservation of the ten nucleotides before the stem (not shown). There is also a conserved AT-rich region at the position of a putative histone downstream element (HDE), 4–12 bp downstream of the stem-loop in all of the histone genes of A. vaga and P. roseola except H2Abd2 (Figure 3D). A similar region is present in the replication-dependent histone genes of C. elegans and is believed to be involved in histone 3′ mRNA processing [11]. Both H2A variants H2Abd and H2Abd1 in both bdelloid species contain the stem-loop motif beyond the stop codon, as well as a putative polyadenylation signal similar to that of H2AX. This latter variant, ubiquitous in other eukaryotes but absent in bdelloids, is packaged in nucleosomes during DNA-replication and is also deposited preferentially in response to DNA double strand breaks [38]. H2Abd2 does not contain this stem-loop motif but has a putative polyadenylation signal.
Another unusual feature of bdelloid histones is the presence of introns in each of the various histone genes (canonical histones and variants) in one or both bdelloid species, except for the H2B gene adjacent to H2Abd1 (Table S1). The length of the introns ranges from 52 to 76 bp, which is typical of bdelloid introns [39]. This contrasts with the absence of introns in the canonical replication-dependent histone genes in other animals, although they are present in the canonical histone genes of plants and fungi [40],[41].
Inter-species comparisons of the ratio of amino acid changes to synonymous changes (Ka/Ks) in the bdelloid histone gene sequences indicate that the amino acid sequences of the clustered H2B, H3 and H4 genes (Figure 4A, right column) are highly conserved except for 2 amino acids in the N-terminal region of H2B. Such conservation is characteristic of eukaryotic canonical histones. The H2A variants H2Abd, H2Abd1 and H2Abd2 are also highly conserved, with the exception of their C-terminal tails and a short region in the N-terminal region of H2Abd (Figure 4A, left column).
In order to detect and measure selection on bdelloid rotifer H2A proteins we used the programs SELECTON [42],[43] and PAML [44],[45]. For both programs we used an amino acid based nucleotide alignment and corresponding phylogenetic tree (tree represented in Figure 4B). The H2A histone fold domains of the different bdelloid H2A variants, but not their C-terminal tails, are under strong purifying selection as detected by SELECTON (results not shown) and also seen in Figure 4A. The program PAML was used to search for signal of positive selection in the bdelloid H2A genes; it implements a likelihood ratio test for positive selection based on dN/dS rate ratios [46] on specific branches [47], on individual codon sites [48], or on both simultaneously (i.e., a branch-site method for testing positive selection on individual codons along specific lineages) [49]. No significant branch specific selection was found. The PAML models used for detecting site-specific selection were M1a and M7 for neutral evolution and M2a, M8 for positive selection. All models were significantly better than model M0 indicating that the dN/dS ratio varies along the sequence but no significant positive selection was detected, as the comparisons M1a-M2a and M7-M8 were not significantly different.
The phylogenetic tree represented in figure 4B clusters the two bdelloid species for each specific H2A variant. It therefore appears that the different H2A variants arose before the separation of the two bdelloid families and have been diverging since mostly through neutral evolution (as detected by SELECTON, results not shown).
A multiple amino acid alignment of canonical H2A genes and H2A variants was carried out in MAFFT [50],[51] (result not shown). The H2A variants included in this alignment are H2AX and H2AZ from distinct eukaryotic lineages, the vertebrate-specific H2A variants macroH2A and Barr-body deficient H2A (H2A Bbd), and the bdelloid H2A variants (H2Abd, H2Abd1 and H2Abd2). Based on this alignment we inferred the phylogeny of the H2A proteins using a maximum likelihood approach as implemented in Bootstrap Raxml [52] (Figure 5). Due to the highly different types of H2A genes included (belonging to different eukaryotic lineages), several parts at the root of the tree are unresolved but some distinct groups are apparent. Congruent with previous phylogenies [16],[29] the vertebrate-specific variant H2A Bbd (purple) forms a distinct cluster outside the canonical H2A group while vertebrate MacroH2A (light blue) is a distinct lineage within canonical H2As. The H2AZ variants (green), having a role in transcription [13] and thought to be universally conserved, clearly form a monophyletic group distinct from canonical H2A. The evolution of the H2AX gene (red) is different from the other H2A variants because it had multiple evolutionary origins within the eukaryotic kingdom as concluded previously [16],[29] and entirely replaced canonical H2A in fungi and Giardia (Figure 5). The H2A variants of the bdelloid rotifers (orange) form a distinct lineage, with H2Abd, H2Abd1 and H2Abd2 clustering with the canonical H2A genes of the monogonont rotifer B. plicatilis (Figure 5). This analysis demonstrates that the bdelloid-specific H2A genes are not closely related to any of the other H2A variants but evolved from a canonical H2A of a common monogonont-bdelloid ancestor.
We investigated in more detail the convergent evolution of H2AX found here and also by Li et al. [16] and Malik&Henikoff [29] by repeating the phylogenetic analysis in bootstrap Raxml using only H2A and H2AX sequences of specific plant, insect, vertebrate and fungi species in order to obtain a better alignment than in the previous analysis. It appears from this phylogeny (Figure 6) that the H2AX variants evolved multiple times but at a higher-order level than indicated in the previous phylogenies (Figure 6) [16],[29]. Indeed, the H2AX variants cluster together rather then with canonical H2As within the plants, vertebrates and insects and, hence, have a single origin within each of these groups.
While the bdelloid canonical histones H2B, H3 and H4 are highly similar to their counterparts in other eukaryotes, we find that the bdelloid complement of H2A histones is highly unusual, with carboxy-terminal tails that are much longer than those of canonical H2A and that are unlike any other eukaryotic H2A variants. The bdelloid H2A histones may be classified as heteromorphous variants because the extent of amino acid sequence change involves a large portion of the C-terminal tail and not merely a few changes as is typical of H2A isoforms [13]. Even the H2A gene H2Abd found in half of the histone clusters in both bdelloid species codes for an unusual C-terminal tail and it seems apparent that canonical H2A is absent from bdelloid rotifers. One of the bdelloid variants, the unclustered H2Abd1 gene, is highly expressed in the embryos of both species during normal growth, while expression of the clustered H2Abd gene was detected at a substantially lower level and only in A. vaga. The H2Abd and H2Abd1 sequences in both bdelloid species specify both the stem-loop motif characteristic of the replication-dependent histones in other eukaryotes and a putative polyadenylation signal in the 3′ UTR characteristic of replication-independent histone variants. Since both of these bdelloid-specific H2A variants H2Abd and H2Abd1 have a stem loop beyond the stop codon and were found in the nucleosomal fraction, they are probably the H2A proteins incorporated into nucleosomes during normal DNA replication in bdelloid embryos, the only life stage in which mitosis occurs in the somatic cells of these eutelic organisms. These bdelloid H2A variants may also be expressed at other times throughout the cell cycle, as observed for H2AX in other eukaryotes [38], while H2Abd2 variants in both species lack the stem-loop motif and the corresponding histones were not found in the protein analysis of nucleosomal fractions. Nevertheless, considering the strong purifying selection under which they have evolved, the histones coded by these variants are almost certainly expressed and incorporated into nucleosomes under some specific conditions. The monogonont rotifer B. plicatilis, belonging to the sister class of bdelloids, has canonical H2A and appears to lack variants similar to those present in bdelloids.
Since the three bdelloid H2A variants are found in each of the two studied species belonging to distantly related families, and as the bdelloid H2As group together with the canonical H2A of monogonont rotifers in the H2A phylogenetic tree (Figure 5), it seems likely that all these variants evolved from a canonical H2A ancestor. The evolutionary process that took place involved the alteration and extension of the H2A carboxy-tail by at least 25 amino acids and the appearance of conserved motifs. One of these motifs, found in the variant expressed in the nucleosomes (H2Abd1), seems related to the S(T)PK(R)K(R) class of DNA-binding motifs and may play a role in the interaction with linker DNA and the packaging of the nucleosome. Moreover, the protein extension of the H2A C-terminal tail is in a region of the nucleosome close to the binding site of histone H1 and hence may affect the structure or dynamics of the nucleosome [12],[16]. Indeed, H2A constitutes the largest heterogeneous family of histone variants that are active in distinct aspects of chromatin conformation and genomic function and the results presented here are consistent with the evolutionary diversity within the H2A family. The high degree of sequence conservation observed within the histone fold domains of the different bdelloid H2A variants is consistent with the general finding that H2A variability is largely confined to the carboxy-terminal domain, both in length and composition [12]. The inter-species variability found in the carboxy-tails of each bdelloid H2A variant (Figure 3A,B,C) could be the result of neutral evolution after the separation of the two families, as seen in the SELECTON analysis, and may suggest that the extension has a more significant role than its particular amino-acid sequence.
By analogy with the radiation- and desiccation-resistant bacterium Deinococcus radiodurans, in which prolonged desiccation causes extensive DNA breakage [27],[28], it is likely that bdelloid radiation resistance similarly reflects an adaptation to survive DNA breakage associated with the frequent desiccation events they experience in the ephemerally aquatic habitats they typically inhabit [4]. DNA breakage in other eukaryotes is accompanied by phosphorylation at the serine in the invariant S[−4]Q(E/D)(I/L/F/Y) motif found in H2AX of protists, fungi, plants and animals. In fungi and Giardia, the H2AX variant has completely replaced the canonical H2A. Considering the involvement of H2AX in the cellular response to DSBs and its ubiquitous occurrence in eukaryotes, we expected to find H2AX genes and a high percentage of H2AX proteins in bdelloid nucleosomes. Instead, none of the bdelloid H2A genes contain the S[−4]Q(E/D) motif characteristic of H2AX. Although SQ occurs at the final two residues of A. vaga H2Abd and 26 amino acids from the C-terminal end of P. roseola H2Abd2, it is notable that these SQ residues are present in a different H2A variant in each species. They are therefore not conserved across bdelloid families and may therefore not represent a functional motif. Further, in all other eukaryotes the SQ motif characteristic of H2AX in which the serine is phosphorylated requires an adjacent acidic residue (glutamic or aspartic acid) that follows the SQ, a carboxy terminal hydrophobic residue, and an invariant position with regard to the carboxyl terminus [37]. Since the bdelloid SQ sequences lack these defining characteristics, we may conclude that there is no H2AX variant in either bdelloid species. It therefore appears that H2AX is dispensable for bdelloid DNA DSB repair, representing an extraordinary exception to the ubiquity of H2AX across eukaryotes.
Although the functional significance of the unusual features of bdelloid H2A histone variants has not yet been investigated experimentally, the most plausible explanation of our findings is that they have evolved from canonical H2A as part of the ensemble of adaptations that have allowed bdelloid rotifers to survive desiccation and its attendant burden of DNA damage. One may speculate that differences in the conditions and possible nature of DNA breakage may have driven the evolution of different ensembles of H2A variants among eukaryotes. Such an explanation emphasizes the apparent evolutionary flexibility of H2A and its variants as compared to other histone genes and leads to the evolutionary question as to how H2AX and H2A variants, like those found in bdelloids, are reinvented in the mold of canonical H2A [29].
Fosmid genomic libraries were constructed from sheared genomic DNA of the bdelloid species Adineta vaga by J. Hur and Philodina roseola by K. Van Doninck and P. Wang. Genomic DNA was extracted from purified bdelloid eggs as described previously [4],[53] except that CsCl density-gradient purification was replaced by phenol∶chloroform extraction. The Epicentre CopyControl Fosmid Library Production Kit (EPICENTRE Biotechnologies) was used to construct fosmids of each bdelloid species as previously described [30].
PCR-derived histone probes, using primers based on highly conserved regions of H3 and H2A, were used to screen the genomic libraries of bdelloid rotifers. Each library of each of the above mentioned bdelloid rotifer species was separately hybridized with probes for H3 and H2A.The DNA of all the selected histone fosmids was extracted manually [54] and tested by PCR and direct sequencing of fosmid templates to confirm the presence or absence of each histone gene H3, H4, H2B and H2A, and to verify which type of H2A gene was present. Histone genes were characterized by BLASTX searches on the National Centre for Biotechnology Information non-redundant databases. Exons and introns were mapped by comparison to homologous amino acid sequences using the software genewise [55]. Gene prediction and the mapping of introns were also verified using the program genemark self-trained on the C. elegans genome [56] and the translation tool Expasy.
The canonical H2A primers used to make the probes for the bdelloid rotifers could also be used to amplify all H2A genes of the monogonont rotifer B. plicatilis containing a canonical H2A core.
For both bdelloid species, four fosmids containing the histone clusters and each copy of canonical H3 were selected for complete sequencing (∼35 kb). DNA from each of these fosmids was purified using Nucleobond Purification kits (BD Biosciences), sheared to a size range of 3–5 kb with a Genemachines Hydroshear (Genomic Solutions) and subcloned in TOPO vectors (Invitrogen) for shotgun sequencing. The resulting sequences were assembled into single contigs as described in [31] and the complete detailed annotation is published in a separate paper [30].
The multiple alignments of the H2A genes and their variants were done using the online version MAFFT v6 [50],[51] with the BLOSUM62 matrix. The “Mafft-homologs” option was enabled only for the alignments including less than 30 sequences. The algorithms used were G-INS-i and L-INS-i when macroH2A sequences were respectively excluded or included in the alignment. To determine which model of protein evolution would best fit our data we used ProtTest v1.4 [57]. The phylogenetic analyses were carried out with the maximum likelihood approach as implemented in the online version of Bootstrap Raxml [52] available on the CIPRES Portal (http://www.phylo.org/sub_sections/portal/). The following parameters were applied: Dayhoff substitution matrix (selected by ProtTest), empirical base frequencies, maximum likelihood search and 1,000 bootstrapping runs. The trees obtained were displayed using FigTree v1.1.2 (http://tree.bio.ed.ac.uk/software/figtree).
The bdelloid H2Abd1 amino-acid carboxy-terminus tails or the nucleotide sequences beyond the stop codon were also aligned using MAFFT and displayed as Logos with the interface Weblogo [58],[59] to emphasize the conserved motifs. The Logos format is a graphical representation of aligned sequences where the size of the letter is proportional to the frequency of that particular residue in that position.
For the sliding window analysis, coding sequences of histone genes were first aligned according to their translated peptide sequences with RevTrans 1.4 [60] using the Dialign 2 method. Sliding windows of the Ka/Ks ratio, among histone genes from the same co-linear pair but in different species, were then generated with DnaSP 4.0 [61] using 50 bp windows and 10 bp steps.
To measure the nature and magnitude of natural selection acting on bdelloid H2A genes, an amino acid based nucleotide alignment using MAFFT and a corresponding phylogenetic tree with Bootstrap Raxml were built and then used in the program SELECTON for a so-called “High-precision” analysis (http://selecton.tau.ac.il/) [42],[43]. This program evaluates the dN/dS ratio (ω) [46]. Neutral evolution predicts an ω = 1 whereas significantly higher and lower values than 1 are respectively interpreted as evidence for positive and purifying selection. Furthermore, we used PAML v4.1 [44],[45] to test for positive selection along sequences and branches. PAML tests different codon substitution models and performs a likelihood ratio test of positive selection based on the dN/dS ratio. We tested branch-specific selection for every internal branch in the tree. Site-specific selection was tested by comparing different models: “M0” which corresponds to a single dN/dS ratio along the sequence, “M1a” and “M7” for neutral evolution (dN/dS = 1) and “M2a” and “M8” for positive selection (dN/dS>1).
Histone proteins were extracted from A. vaga, P. roseola and B. plicatilis embryos following a modified protocol of Tops et al. [62]. The A. vaga and P. roseola rotifer cultures were bleached to obtain clean bdelloid eggs and embryos. B. plicatilis embryos were obtained from a filtered, snap-frozen B. plicatilis biomass (received from Terry Snell) washed with 0.1% SDS and then bleached. After the bleach treatment, the clean bdelloid and monogonont embryos were washed with extract buffer (10 mM HEPES, pH 7.1; 5 mM MgCl2, 2 mM DTT, 10% glycerol and complete protease inhibitor tablets (Roche)) and finally resuspended in 0.5 volume extract buffer. The suspension was dripped in N2 (l) and the resulting frozen egg balls were ground in a cold mortar. The obtained powder was thawed on ice and sheared using a chilled dounce homogenizer (30 strokes, tight pistol). The obtained crude extract was centrifuged for 10 min at 14000 rpm to separate the pellet (with nuclei and membranes) from the soluble cytosol. The pellet was washed twice with extract buffer and subsequently resuspended in extract buffer with 0.4N sulfuric acid and left at 4°C overnight. After centrifugation at 14000 rpm for 10 min at 4°C, the acid soluble histone proteins in the supernatant were precipitated with 20% Trichloroacetic acid (TCA). The obtained histone protein pellet was dried 10 min at 95°C and kept at −20°C or immediately resuspended in 1× alkaline sample buffer, heated at 100°C for 10 min and separated by electrophoresis on 15% Tris-glycine SDS-polyacrylamide gels as previously described [63]. Human cells were obtained from the Maniatis laboratory (MCB, Harvard University), washed with extract buffer (as above) and finally resuspended in 0.5 volume extract buffer. The following steps of the histone protein extraction of human cells were identical to the one described for bdelloid and monogonont embryos.
The Coomassie blue stained bands of bdelloids and monogononts on the SDS-PAGE gels were excised (from the lowest band of the gel up to ∼25 kD) and washed with 50% acetonitrile in water. Histones in gel slices were digested with trypsin and subjected to microcapillary reverse-phase HPLC nano-electrospray tandem mass spectrometry (LC-MS/MS) on an LTQ-Orbitrap mass spectrometer (ThermoFisher, San Jose CA). Acquired MS/MS spectra were then correlated with public sequences using the algorithm SEQUEST and custom programs developed in house. The MS/MS peptide sequences were then reviewed in detail for consensus with known proteins and the results manually confirmed for fidelity. The histone protein extractions, SDS-PAGE gels and mass spectrometry analyses of specific bands were done twice for each rotifer. In addition, de novo sequencing and additional LC-MS/MS analyses against the translated sequences known for bdelloid and monogonont histones were performed with alternative proteolytic enzymes (chymotrypsin, pronase and elastase) to extend the overall coverage of the carboxyl terminal tails of the histone H2A variants. Exhaustive coverage and redundant acquisition of peptides was maximized by the in-house program Enzyme Optimizer designed to choose a multi-enzyme strategy based on proteotypic peptide properties.
DNA sequences have been deposited at Genbank under accession numbers EU652315 to EU652318, EU850438 to EU850441 and EU853685 to EU853700.
|
10.1371/journal.pmed.1002403 | Associations between an IgG3 polymorphism in the binding domain for FcRn, transplacental transfer of malaria-specific IgG3, and protection against Plasmodium falciparum malaria during infancy: A birth cohort study in Benin | Transplacental transfer of maternal immunoglobulin G (IgG) to the fetus helps to protect against malaria and other infections in infancy. Recent studies have emphasized the important role of malaria-specific IgG3 in malaria immunity, and its transfer may reduce the risk of malaria in infancy. Human IgGs are actively transferred across the placenta by binding the neonatal Fc receptor (FcRn) expressed within the endosomes of the syncytiotrophoblastic membrane. Histidine at position 435 (H435) provides for optimal Fc–IgG binding. In contrast to other IgG subclasses, IgG3 is highly polymorphic and usually contains an arginine at position 435, which reduces its binding affinity to FcRn in vitro. The reduced binding to FcRn is associated with reduced transplacental transfer and reduced half-life of IgG3 in vivo. Some haplotypes of IgG3 have histidine at position 435. This study examines the hypotheses that the IgG3-H435 variant promotes increased transplacental transfer of malaria-specific antibodies and a prolonged IgG3 half-life in infants and that its presence correlates with protection against clinical malaria during infancy.
In Benin, 497 mother–infant pairs were included in a longitudinal birth cohort. Both maternal and cord serum samples were assayed for levels of IgG1 and IgG3 specific for MSP119, MSP2 (both allelic families, 3D7 and FC27), MSP3, GLURP (both regions, R0 and R2), and AMA1 antigens of Plasmodium falciparum. Cord:maternal ratios were calculated. The maternal IgG3 gene was sequenced to identify the IgG3-H435 polymorphism. A multivariate logistic regression was used to examine the association between maternal IgG3-H435 polymorphism and transplacental transfer of IgG3, adjusting for hypergammaglobulinemia, maternal malaria, and infant malaria exposure. Twenty-four percent of Beninese women living in an area highly endemic for malaria had the IgG3-H435 allele (377 women homozygous for the IgG3-R435 allele, 117 women heterozygous for the IgG3-R/H alleles, and 3 women homozygous for the IgG3-H435 allele). Women with the IgG3-H435 allele had a 78% (95% CI 17%, 170%, p = 0.007) increased transplacental transfer of GLURP-R2 IgG3 compared to those without the IgG3-H435 allele. Furthermore, in infants born to mothers with the IgG3-H435 variant, a 28% longer IgG3 half-life was noted (95% CI 4%, 59%, p = 0.02) compared to infants born to mothers homozygous for the IgG3-R435 allele. Similar findings were observed for AMA1, MSP2-3D7, MSP3, GLURP-R0, and GLURP-R2 but not for MSP119 and MSP2-FC27. Infants born to women with IgG3-H435 had a 32% lower risk of symptomatic malaria during infancy (incidence rate ratio [IRR] = 0.68 [95% CI 0.51, 0.91], p = 0.01) compared to infants born to mothers homozygous for IgG3-R435. We did not find a lower risk of asymptomatic malaria in infants born to women with or without IgG3-H435. Limitations of the study were the inability to determine (i) the actual amount of IgG3-H435 relative to IgG-R435 in serum samples and (ii) the proportion of malaria-specific IgG produced by infants versus acquired from their mothers.
An arginine-to-histidine replacement at residue 435 in the binding domain of IgG3 to FcRn increases the transplacental transfer and half-life of malaria-specific IgG3 in young infants and is associated with reduced risk of clinical malaria during infancy. The IgG3-H435 allele may be under positive selection, given its relatively high frequency in malaria endemic areas.
| Human immunoglobulin G (IgG) is critical for immunity to many infections including malaria. The active transport of IgG across the placenta during pregnancy contributes to protection against malaria and other infections during early infancy.
IgG is divided into 4 subclasses with different functions. IgG3 is unique among human subclasses in that it represents only 5% of total IgG, yet it is the most potent subclass for activating complement and cellular immune responses (opsonization). Its half-life in blood is 7 days, compared to 21 days for other subclasses.
The potent opsonizing characteristics of IgG3 may contribute to protection against malaria, particularly during infancy as a result of transplacental transport of IgG from the mother.
Recent studies have identified a single mutation in the binding domain of IgG3 that engages a receptor that regulates IgG persistence in blood and mediates IgG transport across the placenta.
In this study, we examined whether women with this mutation had increased transplacental transfer of malaria-specific IgG3, and whether their offspring had a prolonged half-life of serum IgG3 and increased protection against malaria.
We studied 497 pregnant women from Benin, west Africa, in a birth cohort study of their offspring, who were closely evaluated for the presence of malaria during the first year of life.
We found that 24% of women had a specific mutation for IgG3 that was associated with significantly enhanced transplacental transfer and extended persistence of malaria-specific IgG3 in infant blood, compared to women who lacked the mutation.
Offspring of women with this IgG3 mutation had a reduced risk of clinical malaria during infancy.
This study provides evidence that transplacental transfer of malaria-specific IgG3 antibodies contributes to protection against malaria.
The studied mutation in the IgG3 binding domain may have arisen to facilitate protection against malaria.
Malaria vaccines that stimulate IgG3 production may demonstrate enhanced efficacy.
| Plasmodium falciparum malaria remains a major cause of mortality and morbidity in children under 5 years of age in many endemic countries [1]. Death or severe disease typically arises from 1 to 3 overlapping syndromes: severe anemia, respiratory distress, or loss of consciousness [2]. Young infants under 6 months of age experience less clinical malaria compared to older children [3], but the mechanisms conferring immunity are not well understood [4]. One potential mechanism is active transplacental transfer of malaria-specific maternal immunoglobulin G (IgG) during gestation; however, the antigen targets and types of antibodies involved remain poorly defined [4]. Functional assays using cord blood plasma, such as antibody-dependent respiratory burst assays, show that immunoglobulins provide protection against severe malaria during the first 6 months of life [2]. This suggests that cytophilic immunoglobulin subclasses with potent opsonizing and complement fixing functions (e.g., IgG1 and IgG3) are important in protection against malaria. Recent studies have shown that malaria-specific IgG3 relative to other IgG subclasses is more strongly associated with malaria immunity (reviewed in [4]), and its passive transfer to the fetus may contribute to reduced risk of malaria in early infancy. IgG3 is notable compared to other IgG subclasses: it has higher affinity for complement component C1q and Fc gamma receptors (FcγRs), a 4-fold longer flexible hinge region, that are characteristics to enhance opsonization of malaria-infected erythrocytes and promotion of effector functions such as complement-dependent cytotoxicity, antibody-dependent cellular cytotoxicity, and respiratory burst phagocytosis [5,6]. In studies of malaria blood stage vaccine candidate antigens (e.g., GLURP), increased opsonic phagocytosis activity of IgG3 in vitro is strongly associated with reduced risk of febrile malaria, thus confirming the importance of IgG3 function in immunity to malaria [7].
Human IgGs are actively transferred across the placenta, with 30%–40% higher levels in newborn serum compared to maternal blood [8]. Preferential transport occurs for IgG1, followed by IgG4, IgG3, and IgG2 [9]. The neonatal Fc receptor (FcRn), expressed within endosomes in the syncytiotrophoblastic membrane, mediates IgG transfer by micropinocytosis. The interaction between the constant region of immunoglobulin heavy chain gamma (IGHG) and the FcRn is highly pH dependent. In particular, histidines at positions 310 and 435 (H310 and H435) on the IGHG Fc fragment become positively charged at pH 6.5, forming salt bridges with corresponding FcRn residues [10–12]. The FcRn is also a homeostatic receptor responsible for prolonging IgG half-life, by protecting it from lysosomal degradation and recycling it to systemic circulation [13]. Interestingly, IgG3 has a half-life of 7 days, compared to 21 days for IgG1 [14]. This difference in half-life has been attributed to the presence of arginine at position 435 (R435) in the IgG3 heavy chain instead of histidine (H435), as in IgG1, IgG2, and IgG4 [15]. In vitro studies demonstrate that IgG3-R435 has a lower binding affinity to FcRn [16]; consequently, IgG3-R435 is competitively inhibited by IgG1, resulting in reduced transplacental transfer and decreased half-life [5].
In contrast to other IgG subclasses, IgG3 is highly polymorphic, with distinct variants that give rise to allo-specific antibodies (allotypes). Three distinct IgG3 amino acid variants (IGHG3*17, IGHG3*18, and IGHG3*19), identified by allotypes G3m15 and G3m16, naturally contain a histidine at position 435. Both IgG3-H435 and IgG3-R435 may be present at different levels in the blood of heterozygous individuals [17]. The H435 variant is uncommon in Europeans (~1%), but is more prevalent in Asians (10%–25%) and in Africans (30%–60%) [18], suggesting it may be under positive selection in malaria endemic areas. The functional significance of this polymorphism for IgG3 transplacental transfer and IgG3 half-life, especially with respect to malaria, has not been much investigated [19]. Factors that impair or enhance transplacental transfer can alter the amount of pathogen-specific IgG in the newborn circulation and therefore susceptibility to infections, including malaria.
Here we examine the association between the IgG3-H435 polymorphism and transplacental transfer of P. falciparum–specific IgG1 and IgG3 in 497 Beninese mother–newborn pairs, and we investigate how this transfer is associated with the risk of malaria during infancy. Specifically, we explored the hypotheses that the maternal IgG3-H435 variant increases transplacental transfer of anti-malaria IgG3 to the newborn, prolongs IgG3 half-life in infant blood, and reduces the risk for malaria in infancy.
The University of Abomey Calavi institutional review board (Benin) and the Consultative Ethics Committee of the Institute of Research for Development (France) approved the study protocol for the Beninese cohort. Before enrollment, all women signed an informed consent that also included consent for their children. All methods were carried out in accordance with approved guidelines.
Between 1 June 2007 and 31 July 2008, 572 newborns were enrolled in a birth cohort study in Tori-Bossito, in south Benin [20], an area classified as mesoendemic for malaria [21]. Infants were actively and passively followed for malaria until 12 months of age, as previously described [22]. After delivery, thick blood films were prepared from blood obtained by scraping the wall of the incision from the maternal side of the placenta. The presence of P. falciparum (placental malaria) was determined by microscopy. Active surveillance was conducted with weekly home visits, at which time axillary temperature and symptoms related to malaria were assessed. Febrile children were further evaluated at the local health center using a malaria rapid diagnostic test and a thick blood smear. Any child who developed symptoms of malaria at other times was encouraged to attend the health center, where a similar evaluation was performed and recorded (passive surveillance). Symptomatic (clinical) malaria was defined as axillary temperature > 37.5°C and positive blood smear for malaria, as previously described [22]. Symptomatic malaria was treated with artemether and lumefantrine combination therapy, as recommended by the National Malaria Control Program. Every month, during one of the weekly visits at home, blood smears were systematically collected to determine asymptomatic carriage (active surveillance). To increase the probability that febrile illness was due to malaria, only those individuals with >2,500 parasites/μl were included in the analysis of symptomatic malaria. This cutoff value was selected because few of the asymptomatic participants had parasitemia levels of >2,500 parasites/μl. Maternal venous and cord blood was obtained at delivery, and venous blood samples were collected from infants every 3 months. Plasma was separated and stored. Of the 572 participants initially enrolled in the study, 27 infants were excluded because of follow-up problems (11 with doubtful identification, 12 with extensive missing data, and 4 with missing individual malaria exposure) [20], and an additional 48 infants were excluded because they had missing placental malaria information and/or insufficient quantity or quality of DNA for genotyping, thus yielding 497 individuals available for analysis.
Maternal hypergammaglobulinemia was defined as total maternal IgG of ≥1.6 g/dl. Gestational age was measured by the Ballard score [22], and premature birth was defined as birth at <37 weeks of gestation. Pregnant women were enrolled at delivery, when a questionnaire was administered to collect information on the women’s obstetric background and current pregnancy. Eighty-four percent of the pregnant women included in the study reported taking at least 1 dose of Fansidar (500 mg sulfadoxine and 25 mg pyrimethamine) for malaria chemoprophylaxsis [22]. Independent variables are listed in Table 1.
The homes of all infants were localized by GPS. Mosquitoes were collected at 4 houses in each study village (N = 9 villages) over 3 successive nights, every 6 weeks between 1 July 2007 and 31 July 2009. Rainfall was recorded twice a day with a pluviometer in each village during the entire follow-up [20,23]. In the analysis, the number of rainy days during the 10 days before a mosquito catch was considered. The following data were also collected during each catch: season, type of soil 100 meters around the house, presence of small pools of water from partial/abandoned construction 100 meters around the house, presence of a watercourse within 500 meters around the house, normalized difference vegetation index 100 meters around the house, ownership of bednets, use of insect repellent, and number of inhabitants per house. Based on these variables, a predictive regression model was developed, and an individual malaria exposure variable was constructed for use in this analysis. Details of this model are published elsewhere [20,23].
ELISA was performed on maternal, cord, and infant sera as described previously [24]. The concentrations of total human IgG and specific IgG1 and IgG3 directed at promising blood stage vaccine candidate antigens were measured. Antigens for study included apical membrane antigen 1 (AMA1), merozoite surface protein 1–19 (MSP119), MSP2 (2 allelic families, 3D7 and FC27), MSP3, and glutamate-rich protein (GLURP; 2 regions, R0 and R2), as detailed previously (supplementary information in [24]). A dedicated program in R statistical software [25] was derived from ADAMSEL FLP b039 (http://www.malariaresearch.eu/content/software) and was used to analyze antibody concentrations by ELISA optical density as described previously [24]. Briefly, censored values (below detection threshold or over saturation) were imputed using a (log) linear regression model, taking into account confounding variables. The model was fitted using a stochastic expectation maximization algorithm and was robust since the association between imputed and measured values was very good (R2≥0.91) [24].
Maternal DNA was extracted with a Qiagen kit according to manufacturer recommendations (Qiagen, Valencia, CA, US). At least 1 ng/μl DNA was amplified with a forward primer (FWD 5′-GTCGGGTGCTGACACATCTG-3′) and a reverse primer extended by a universal primer M13 (REV 5′-AGCGGATAACAATTTCACACAGGA | CTTGCCGGCYRTSGCACTC-3′) [26]. The additional M13 sequence extended the amplicon size in order to enhance sequence quality. The amplification was performed with AccuPrime Taq DNA Polymerase, High Fidelity (Fisher Scientific, Florence, KY, US) and the buffer II (delivered with the enzyme). PCR product purification was performed with the Wizard SV 96 PCR Clean-Up System (Promega, Madison, WI, US). For the sequencing, the amplification forward (FWD) and/or a second forward (FWD2 5′-AGGTCAGCCTGACCTGCCTG-3′) primer [26] was used for sequence accuracy and led to 805- and 277-nucleotide-long sequences, respectively (from bp 1,382 or 1,910 to bp 2,186; accession number NC_000014.9). In case of a nonconclusive sequencing, a third reverse primer was used (M13-REV 5′-AGCGGATAACAATTTCACACAGGA-3′). The sequencing was performed by Eurofins Genomics (Louisville, KY, US).
Prespecified confounding factors were maternal malaria at delivery, hypergammaglobulinemia, bednet use, malaria chemoprophylaxis during pregnancy, maternal weight and age, parity, newborn sex, gestational age, and birth weight. These factors were evaluated with respect to the IgG3-H435 polymorphism by Pearson chi-squared test (to test the difference between proportions) or Student unpaired t test (to test the difference between means) as appropriate.
The transfer of malaria-specific antibodies is defined by the cord-to-mother transfer ratio (CMTR, cord IgG level divided by maternal IgG level). The CMTR was used in the different analyses as dichotomized above and below the median. A univariate logistic regression was performed to compare the CMTR for IgG1 and IgG3 for all mothers, stratified by R435H polymorphism. Given the multiple comparisons in this analysis, significant difference was considered as p ≤ 0.007 after a Bonferroni correction was applied. A multivariate logistic regression model assessed the association of the IgG3-H435 polymorphism with the degree of transplacental transfer, adjusted for the potentially confounding variables described above. The competition between IgG1 and IgG3 for FcRn binding was tested by a multivariate logistic regression model, after stratification by maternal hypergammaglobulinemia. In this model, the dependent variable was the CMTR of malaria-specific IgG3, and the independent variables were malaria-specific maternal IgG1 level and the confounding variables listed below. The half-life of IgG3-H435 (versus IgG3-R435) in infant blood was evaluated by a mixed linear regression analysis using data from 302 infants for whom antibody levels at birth (cord blood) and 3 and 6 months of age were available, and for whom no evidence of a rise in malaria-specific antibody levels between 0 and 3 months or between 3 and 6 months of age was noted. A hierarchical model was used to account for multiple monthly measurements on the same individuals. The association between protection against malaria, defined as time to first malaria parasitemia, and the transferred IgG subclasses was illustrated by Kaplan–Meier curves and evaluated by log-rank test. A multivariate Cox analysis was performed to assess the association between the delay to first symptomatic malaria over the first year of life and the IgG3-H435 polymorphism. In this multivariate analysis, exposure to P. falciparum infection (as described above [23]) was used as an independent variable as were the other variables listed below. A Poisson regression was performed with the same independent variables as used for the multivariate Cox regression to evaluate the relationship of maternal IgG3-H435 (versus IgG3-R435) and the number of symptomatic malaria and asymptomatic malaria episodes between birth and 12 months of age. All multivariate analyses included the variables with p ≤ 0.20 in the univariate step. Statistical significance was set at p < 0.05. The variance inflation factor (VIF) values were tested in models where a collinearity could be suspected. All statistical analyses were performed using Stata, version 13.0 (StataCorp, College Station, TX, US). The original protocol and modifications are available in S1–S3 Protocols.
The overall frequency of the IgG3-H435 variant in the population group of Tori-Bossito, south Benin, was 0.12 compared to 0.88 for IgG3-R435. A total of 117 mothers were heterozygous and 3 were homozygous for IgG3-H435, resulting in 24% of study participants (pregnant women) carrying the allele (Table 1). The IgG3-H435 variant was not associated with prespecified confounding factors likely to influence transplacental antibody transfer and/or malaria risk in mothers and infants (Table 1).
We explored the hypothesis that the maternal IgG3-H435 variant increased transplacental transfer of anti-malaria IgG3 to the newborn: it was expected that (i) the maternal IgG3-H435 variant would be associated with a better transplacental transfer of IgG3 and (ii) this IgG3 transfer would be equivalent to IgG1 transfer, for which the variant is always IgG1-H435.
Overall, malaria-specific IgG1 and IgG3 levels were similar in mothers and neonates (S1 Fig). The transplacental transfer of malaria-specific IgG3 that recognized MSP2-3D7, GLURP-R0, and GLURP-R2 was significantly greater for mothers with IgG3-H435 compared to those with homozygous IgG3-R435 (Fig 1A). A similar pattern was observed for the other antigens, although these differences were not significant. Using multivariate analysis to adjust for other factors that could influence transplacental transfer (see Table 1), we found that transfer of anti-malarial IgG3 was 60% to 95% higher for IgG3-H435 than for homozygous IgG3-R435 (Fig 1B, all antigens p < 0.01 except MSP119). Of note, transplacental transfers of IgG3-H435 and IgG1 were similar for many of the antigens (Fig 1A), except for MSP2-FC27 (p < 0.01). For a subset of the antigens tested, other factors were independently associated with reduced transplacental transfer of malaria-specific IgG3, including the presence of placental malaria, increased maternal malaria-specific IgG3 levels, and increased levels of maternal total IgG (Table 2).
Prior studies have shown that, in vitro, IgG1 (always H435) competes with IgG3-R435 for binding to FcRn [5]. To examine whether this occurs in vivo, serum samples obtained from mothers at birth were assayed for level of malaria-specific GLURP-R2. Women with GLURP-R2-specific IgG1 above the median compared to those below the median demonstrated a 62% reduced transplacental transfer of GLURP-R2-specific IgG3 (odds ratio [OR] = 0.58 [95% CI 0.39, 0.87], p = 0.008, N = 405). This analysis was performed in a subset of women without hypergammaglobulinemia (<1.6 g/dl of total IgG, N = 405) because elevated IgG is an independent risk factor for impaired transplacental transport of IgG3 (Table 2). The competition between IgG1 and IgG3 for binding to FcRn and therefore for transplacental transfer is represented by the association between a high level of maternal IgG1 and a decreased transfer of IgG3. We noted a greater competition for transplacental transfer of IgG3 for GLURP-R2 in women with homozygous IgG3-R435 compared to IgG3-H435 (OR = 0.54 [95% CI 0.21, 1.43], p = 0.22), but this difference was not statistically significant. For the other antigens, the level of maternal anti-malaria IgG1 was not associated with decreased transplacental transfer of IgG3.
For mothers with high levels of malaria-specific IgG3 at birth, we detected higher levels of malaria-specific IgG3 in their infants, and for a longer period of time (Table 3). After adjustment for maternal IgG3 level, we found that offspring of mothers with IgG3-H435 had a 28% to 35% greater persistence of malaria-specific IgG3 between birth and 6 months of age, compared to offspring of mothers homozygous for IgG3-R435 (Fig 2). This difference was significant for AMA1, MSP2-3D7, MSP3, GLURP-R0, and GLURP-R2, but not for MSP119 and MSP2-FC27.
Prior studies have shown that antibody responses directed against GLURP, especially block 2 region, are associated with a reduced risk of clinical malaria in different populations [7,27]. In the present study, an increased transplacental transfer of GLURP-R2-specific IgG3 was associated with a delay in time to first symptomatic malaria in the first year of life (Fig 3; hazard ratio [HR] adjusted for maternal placental malaria and infant malaria exposure = 0.59 [95% CI 0.45, 0.77], p < 0.001). Similar results were observed for GLURP-R0 (HR = 0.73 [95% CI 0.56, 0.95], p = 0.02), but not for the other malaria antigens (S1 Table). This suggests that IgG3 directed to GLURP has greater functional activity in blood stage malaria infection.
Because women with IgG3-H435 have increased transplacental transfer of IgG3 and their offspring have IgG3 with prolonged half-life during infancy, we examined whether offspring of mothers with IgG3-H435 versus homozygous IgG3-R435 had a reduced risk of malaria in infancy. Using Cox proportional hazard analysis (Fig 4), there was a delay in time to first symptomatic (clinical) malaria infection as defined by a fever of >37.5°C and >2,500 parasites/μl among offspring of women with IgG3-H435 versus homozygous IgG3-R435 (HR = 0.69 [95% CI 0.37, 1.03], p = 0.08, adjusted for individual malaria exposure and placental malaria), but this difference was not statistically significant. The difference was most noticeable at 6–8 months of age, when survival curves diverged. To further explore the relationship of IgG3-H435, infant age, and susceptibility to malaria, we examined the data using a Poisson regression model stratified by 0–6 versus 6–12 months of age and whether the infant had clinical malaria or asymptomatic parasitemia (Table 4). Of note, this analysis was not prespecified in the initial protocol. Offspring of women with the IgG3-H435 allele had a 32% reduced risk (incidence rate ratio [IRR] = 0.68 [95% CI 0.51, 0.91], p = 0.01) of clinical malaria during infancy (0–12 months). This association was most pronounced among infants 6–12 months of age (IRR = 0.61 [95% CI 0.43, 0.87], 39% reduced risk, p = 0.007). There was no association between the IgG3-H435 allele and the risk of asymptomatic malaria (Table 4). Of note, altering the definition of clinical malaria using different parasitemia levels, >5,000/μl, >10,000/μl, etc., does not alter the association with protection; however, the confidence interval widens because the sample size diminishes.
Human IgG is the only antibody isotype that is actively transferred across the placenta, providing passive immunity for the newborn. With respect to all IgG subclasses, IgG3 is noteworthy for its shorter half-life and diminished capacity for transplacental transfer. Due to an arginine substitution for histidine at position 435, found exclusively in IgG3, binding to FcRn is reduced [5]. Why IgG3 differs from other IgG subclasses in several respects is unclear. One possibility is that by limiting the transplacental transfer of maternal IgG3 directed to fetal antigens, the risk of fetal alloimmune pathology is reduced [28]. Also, the potent pro-inflammatory properties of IgG3 could be tempered by its shorter half-life [29]. Yet the powerful effector function of IgG3 may be beneficial against some pathogens such as P. falciparum, where malaria-specific cytophilic IgGs have been most strongly associated with protection [7,30]. Since infants are most susceptible to malaria infection, enhanced transplacental transfer of IgG3 with prolonged half-life may be advantageous. In support of this hypothesis, we showed that 24% of Beninese women living in an area highly endemic for malaria possess the IgG3-H435 allele, which enhances IgG3 binding to FcRn [5]. We demonstrated that women with this allele have increased transplacental transfer of malaria-specific IgG3 to their fetus, and this IgG3 persists longer in infant blood than IgG3-R435. We showed that increased transplacental transfer of IgG3 directed to the malarial GLURP antigen is strongly associated with reduced risk of clinical malaria, and offspring of mothers with the IgG3-H435 allele have reduced risk of clinical malaria in infancy compared to offspring of women with homozygous IgG3-R435. Together, these data support the conclusion that malaria-specific IgG3 contributes to protection against clinical malaria during infancy.
To our knowledge, this is the first study to show that an antigen-specific IgG3-H435 amino acid polymorphism influences transplacental transfer of IgG3, prolongs its half-life, and is associated with protection against a major pathogen in vivo. Prior studies have shown extended persistence of IgG3-H435 compared to IgG3-R435 in agammaglobulinemic patients who received intravenous IgG replacement [5]. Other studies demonstrated, using an ex vivo placental perfusion model [6], that IgG3-H435 transferred as efficiently as IgG1 across the human placenta; this was confirmed in vivo in 6 healthy pregnant Chinese women with the G3m16+ allotype, which enabled expression of IgG3-H435 [19]. Thus, our work confirms and expands the observation that IgG3-H435 alters the biological characteristics of IgG3 by enhancing its transplacental transfer and prolonging its half-life in vivo.
Offspring of women with IgG3-H435 had a 31% to 39% reduced risk of clinical malaria during infancy based on 2 independent analytic models. There was no association with protection against asymptomatic malaria, suggesting that the association was not related to exposure, but to the ability to control the level of parasitemia. This association was most pronounced at 6 to 8 months of age, when most of the passively transferred antibody protection in an infant has typically waned. There are several possible explanations for a delayed protective effect. First, IgG3-H435 may persist longer than other IgG subclasses to extend the period of passive protection. This is supported in studies of X-linked agammaglobulinemic individuals treated with intravenous IgG from donors with IgG3-H435, where IgG3-H435 antibody persisted 30% longer than that of the other subclasses in vivo [5]. Second, malaria-specific IgG1 may mask the protective effect of IgG3-H435 early on, but once the more abundant IgG1 has bound and cleared malaria antigen, the persistent effect of IgG3-H435 is relatively more important. Third, offspring of women with IgG3-H435 are more likely to have inherited the variant themselves, which may contribute to stronger acquisition of natural immunity in the infant.
Levels of cytophilic antibody to merozoite surface antigens (i.e., levels of IgG1 and IgG3) often correlate more strongly with protection against malaria than levels of IgG from the other subclasses. Thus, an enhanced association with protection against malaria would be expected for higher levels of functional antibodies directed against blood stage antigens [7,31–33]. The most important characteristics of functional antibodies include the capacity to activate complement and the ability to opsonize merozoites or malaria-infected erythrocytes for uptake by phagocytic cells. Of the IgG subclasses, IgG3 exemplifies these characteristics in that it has the greatest ability to activate complement and the highest affinity for FcγRs [34]. Thus, a mutation enhancing transplacental transfer of IgG3 and prolonging its half-life may be under positive selection in malaria endemic areas, where infant mortality due to malaria can reach 20% [1]. Supporting this possibility is the observation that greater frequencies of the SNP rs4042056 that encodes IgG3-H435 have been recorded in populations where malaria historically is more frequent (e.g., South Asian and African, ~0.12), as compared to northern European populations (~0.01) [18,35]. More information regarding IGHG3 polymorphism frequencies in endemic populations will help to clarify potential selective pressures on this SNP and its role in malaria clinical outcomes.
Almost all of the women with the IgG3-H435 haplotype in the study were heterozygotes. Expression of immunoglobulin genes undergoes allelic exclusion in individual B cells, as only 1 IgG3 heavy chain is expressed in a specific B cell [36]. Thus, throughout clonal selection, B cell proliferation, and memory B cell formation, all progeny of an individual B cell with activity against a given malarial epitope express the same IgG3 heavy chain. However, other pre-B cells undergoing a different IgG recombination could retain a different IgG3 heavy chain allele. This would result in malaria-specific B cells expressing IgG3 with either H435 or R435 alleles, although since they would arise from different B cell precursors, they may not recognize the exact same epitopes. The presence of both IgG3-H435 and IgG3-R435 in the serum of a single individual has been confirmed by mass spectroscopy of IgG3 from individuals with the IgG3-H435 allele [17]. However, in any one person there may be quite different levels of IgG3-H435 and IgG-R435 to specific antigens. Thus, in the current study, individuals bearing the IgG3-H435 haplotype would have, on average, just half of their IgG3 with H435 heavy chain, thus diluting a potentially protective effect.
Studies examining immune correlates of protection against malaria are frequently hampered by the inability to adequately control for exposure. In this investigation, frequent active and passive surveillance for malaria and a detailed assessment of malaria exposure at the individual level mitigate this limitation. Moreover, the observation that a single point mutation of IgG3 can measurably be associated with transplacental transfer of IgG3 to the newborn and extend its half-life, independent of other risk factors for malaria, allows for a more precise assessment of the role of malaria-specific IgG3 in protecting against malaria. There are limitations to the study as well. Almost all of the women carrying the IgG3-H435 allele were heterozygous, and we could not assess how much malaria-specific IgG3-H435 relative to IgG3-R435 was present in a given individual. In addition, the proportion of malaria-specific IgG produced by infants relative to that acquired from their mothers was not assessed, and how the relative proportions contribute to protection during infancy remains undefined. Finally, we could not determine the genotypes of the infants with respect to IgG3-H435 because requisite DNA was unavailable.
In conclusion, our study shows that the IgG3-H435 variant can have a profound association with transplacental transfer of malaria-specific IgG3, and it prolongs IgG3 half-life in infant blood, thereby enhancing immunity to malaria in infancy. This suggests that cytophilic IgG3, with a longer half-life following exposure to P. falciparum, may be of critical importance in protection against malaria. Vaccine approaches whereby malaria-specific IgG3 antibody responses are enhanced may prove to be especially useful.
|
10.1371/journal.pntd.0001213 | Measuring Fitness of Kenyan Children with Polyparasitic Infections Using the 20-Meter Shuttle Run Test as a Morbidity Metric | To date, there has been no standardized approach to the assessment of aerobic fitness among children who harbor parasites. In quantifying the disability associated with individual or multiple chronic infections, accurate measures of physical fitness are important metrics. This is because exercise intolerance, as seen with anemia and many other chronic disorders, reflects the body's inability to maintain adequate oxygen supply (VO2 max) to the motor tissues, which is frequently linked to reduced quality-of-life in terms of physical and job performance. The objective of our study was to examine the associations between polyparasitism, anemia, and reduced fitness in a high risk Kenyan population using novel implementation of the 20-meter shuttle run test (20mSRT), a well-standardized, low-technology physical fitness test.
Four villages in coastal Kenya were surveyed during 2009–2010. Children 5–18 years were tested for infection with Schistosoma haematobium (Sh), malaria, filaria, and geohelminth infections by standard methods. After anthropometric and hemoglobin testing, fitness was assessed with the 20 mSRT. The 20 mSRT proved easy to perform, requiring only minimal staff training. Parasitology revealed high prevalence of single and multiple parasitic infections in all villages, with Sh being the most common (25–62%). Anemia prevalence was 45–58%. Using multiply-adjusted linear modeling that accounted for household clustering, decreased aerobic capacity was significantly associated with anemia, stunting, and wasting, with some gender differences.
The 20 mSRT, which has excellent correlation with VO2, is a highly feasible fitness test for low-resource settings. Our results indicate impaired fitness is common in areas endemic for parasites, where, at least in part, low fitness scores are likely to result from anemia and stunting associated with chronic infection. The 20 mSRT should be used as a common metric to quantify physical fitness and compare sub-clinical disability across many different disorders and community settings.
| Reduced physical fitness, which is a manifestation of the body's inability to maintain adequate oxygen supply to the tissues, can have many causes. In developing countries, a person's low physical fitness is often the result of anemia and undernutrition, which have multifactorial etiologies including poor diet and chronic infections such as malaria, hookworm and schistosomiasis. In past surveys, exercise tolerance has been measured using non-standardized tests that were poorly-suited to young children. In this study, we implemented a well-validated and reliable 20-meter shuttle run test in a low-resource area of Kenya. Results for 1950 children, aged 5–18 years, showed that impaired fitness was common and associated with anemia and poor growth with boys being more affected than girls. The 20 mSRT is a feasible and low-cost tool that can be easily delivered in low-resource settings to identify children who manifest the disabling but often sub-clinical manifestations of their disease. We propose its implementation as a standard fitness test in less developed areas to allow comparisons across morbidity studies assessing the impact of different interventions.
| In the context of chronic disease, exercise intolerance due to decreased physical fitness is a measurable outcome strongly related to decreased quality of life in many spheres of human performance. Among children, loss of physical fitness is associated with anemia, chronic inflammatory conditions, and inadequate nutrition leading to impaired growth [1]–[3]. Accurate, affordable measurement of physical fitness is expected to become a very valuable tool in gauging community burden of disease where such conditions are common. However, to date, no fitness-related standardized test has been widely adopted as a morbidity assessment tool.
Exercise intolerance due to anemia and chronic parasitic diseases such as schistosomiasis can lead to chronic disability in children and to decreased adult productivity. These important morbidities have been routinely underestimated in past disease burden assessments [4], [5].Quantifying physical fitness can be done by many different methods, but their implementation and frequency of use most often relate to the resources available at the site of testing. In developed countries, measuring school-age children's fitness is a common practice, and well-controlled studies have been carried out to measure the maximum aerobic capacity and to compare pre- and post-training fitness results [6]–[8]. By contrast, in low resource settings, adequate assessment of aerobic capacity has rarely been done outside the research laboratory environment, and even then, a variety of different methods have been applied, with sometimes conflicting results. Reduced aerobic capacity was documented in Ethiopian and Indian children using cycle ergometry [9], [10]. Other studies have utilized portable accelerometers [11] or the Harvard-Step test [12]–[15] as fitness measuring tools. But these were neither designed nor standardized for children under 17 years old. In Mozambique, a combination of observation and activity questionnaires [16] showed no correlation with morbidity outcomes such as undernutrition. Given the variability of results for the methods employed to date, there is an evident need for a rapid, affordable physical fitness test that can be uniformly implemented in many settings around the world.
In this paper we present the results from cross-sectional surveys in four parasite-endemic villages in coastal Kenya, during which participating children were uniformly tested using the well validated, low-technology multistage 20-meter shuttle run test (20 mSRT). Our working hypothesis, as shown in Figure 1, was that children affected by infection-related anemia or undernutrition would be exercise intolerant (with decreased aerobic capacity), and that this could be reliably quantified by the 20 mSRT. To the best of our knowledge this is the first time that the 20 mSRT has been used in a low-resource environment in parasitic disease research.
Ethical clearance was obtained by the Institutional Review Board at the University Hospital Case Medical Center of Cleveland and the Ethical Review Committee of the Kenya Medical Research Institute (KEMRI). Children were eligible if they were residents of the area for at least two years, were between 5–18 years old, and they had provided child assent and written parental consent.
This cross-sectional study was conducted in four Schistosoma haematobium (Sh) endemic rural villages (Nganja, Milalani, Vuga, and Jego, see Figure 2) in Msambweni and Kwale Districts in Coast Province, Kenya [17]. This sub-study is part of a larger project in the area studying the ecology of transmission of vector borne parasitic infections. 78% of all eligible children in Nganja (N = 240/309), 51% (N = 416/822) in Milalani, 74% in Vuga (N = 726/983) and 73% in Jego (N = 652/890) agreed to participate. Non-participating children either i) belonged to households that refused to take part in the study or ii) did not complete all the phases involved. A common reason for refusal during the recruitment period was fear of being tested for Human Immunodeficiency Virus (although this was not part of the study) or refusal to handle the biological samples necessary for parasitology testing.
Subjects were enrolled at the time of the village demography survey in February, August and November of 2009 for Nganja, Milalani and Vuga respectively and March of 2010 for Jego. After an initial interview with the head of the household, in which general information about living conditions was obtained, children were screened for the presence of Sh, Wuchereria bancrofti, malaria and geohelminths (Ascaris lumbricoides, Trichuris trichiura, and hookworms). Nutritional and fitness assessment were also done. When available, dates of birth were cross-checked with national identification cards for adults and vaccination cards for children.
Egg burden for Sh was assessed by the urine filtration method [18], and the presence of hematuria was also recorded. The subjects provided a single sample of a mid-morning urine specimen that was immediately processed.
The night prior to the parasitology survey, eligible subjects were given a stool container by local community health workers to provide a single stool sample. The following morning, the stool samples were taken to a central facility and examined in duplicate by Kato-Katz method for microscopic detection of parasite eggs.
Finger prick blood was used to obtain a hemoglobin measurement (Hemocue, Ångelholm, Sweden), and a rapid antigen test for P. falciparum for malaria (ICT Diagnostics, Australia) and W. bancrofti for filarial (Binax, Portland, ME). Testing was performed in all participating children except one. Anemia was categorized according to WHO criteria by age and sex [19].
Because growth is considered the best indicator of nutritional status in children, standardized measurements of height and weight were taken. Prior to working in the surveys, all technical staff who performed anthropometric measurements received standardization training followed by independent reliability assessment. Supervised by a trained anthropometrist, candidates performed duplicate measurements of height (agreement within 0.5 cm), weight (agreement within 1.0 kg), for ten healthy volunteer children on the same day. The results were then compared for inter and intra-observer reliability. The trainees' intra- and inter-examiner technical errors of measurement fell within the reference values [20]–[22] and were therefore considered accurate.
Eligible children were measured according to procedures described by Jeliffe [23] while wearing a kanga- a traditional light cloth- wrapped around their bodies. Weight was obtained by digital weight-scale (SECA model 803, Hanover, MD) and was rounded to the nearest 0.1 kg. Height was measured with the use of a stadiometer (SECA model 214, Hanover, MD) and measurements were read to the nearest 1.0 cm. Instruments were calibrated daily prior to use. Every measurement was performed twice and the mean values were used for analysis. Reference population Z-scores were calculated for each subject's height-for-age (HAZ) and body-mass index for age (BAZ) using two different reference populations for comparison: First, the standards included in the US Centers for Disease Control and Prevention's Epi Info™, Version 3.5.1 (CDC, Atlanta, GA) from the year 2000 [24] and second, the World Health Organization's Anthro-Plus for ages 5-19 years old (WHO, Geneva, Switzerland) with reference growth standards from the year 2006 [25], [26].
HAZ is considered an indicator of long-term linear growth whereas BAZ variations better reflect acute changes in nutritional status. According to WHO standards [27], stunting was categorized as an observed HAZ that was two or more standard deviations (SD) below average (HAZ score≤−2). Children were categorized as clinically wasted if their BAZ was more than two SD below average for their age (BAZ score≤−2). Children were further identified as severely wasted if their BAZ was≤−3.
The 20 meter shuttle run test (20 mSRT), initially validated in Canadian schoolchildren, was used to determine the maximal aerobic capacity of children enrolled in this study [28], [29]. The test was carefully explained to the participating children before the start. A demonstration was then performed by one of the recorders. Materials needed for the test included: a measuring tape to mark the 20 meter track, a battery-powered CD player, study-ID coded score sheets, and pencils. A table or desk was usually provided by the village school to facilitate data recording. The schematic diagram of the test is illustrated in Figure 3. Briefly, a group of 5–15 children were instructed to run back and forth on a 20 meter course that had been previously marked off in a flat, shady area of the village. A soccer field was used in 2 of the 4 villages. Although there was often a mixture of ages and body sizes, this did not interfere with the performance of the test, as every individual was able to run up to his or her personal limit. The number of children simultaneously performing the test was determined based on the number of recording staff available. After several pilot runs, it was concluded that every observer could effectively monitor 4–5 children per test. The subjects each lined up behind the starting line and waited for the starting beep on the pre-recorded CD. They then had to run towards the marked 20 m line, reaching it before (or at the same time as) the second beep sound signal from the CD. The interval of these signals progressively shortened, increasing the pace by 0.5 km h−1 every minute, from a starting speed of 8.5 km h−1. Each change of level, and thus speed, was announced every minute by the pre-recorded CD. Each cohort of children took between 5–11 minutes to complete the test. When the subject could no longer follow the pace, he/she was asked to stop, and their highest level obtained was recorded as the fitness score in a Study ID coded sheet. Higher scores correlated with better fitness, and these were later translated into standardized estimates of maximum oxygen consumption (VO2max) [28] for each child.
Demographic data collected in the field were double entered in hand-held devices (Dell Axim, Round Rock, Texas) using Visual CE 10 (Cambridge, MA) and a paper form. Data were then transferred in duplicate into ACCESS 2007 (Microsoft, Seattle, WA) and the databases compared for errors. Parasitology and anthropometric data were similarly entered to complete the database.
Exploratory analysis started with univariate distributions followed by bivariate analyses to explore the pairwise relationships of individual outcomes (Table S1). Egg counts were log-converted to adjust for their skewed distribution. Later multivariable analysis was used to assess associations controlling for age, gender, infection, co-infection status and intensity and a scale of socioeconomic status derived using Principal Component Analysis (PCA) of combined asset scores [30]. Linear regression modeling of fitness scores was performed to obtain estimates of their relationship to other subject variables measured in the study. Generalized Estimating Equation (GEE) modeling was used to account for household clustering effects. Multiple regression diagnostic analyses were performed (R2, C(p), MSE) to aid in the choice of the optimal model. Variance Inflation Factor (VIF) was also calculated to test for co-linearity among model variables.
The model outcome of interest was fitness level (as scored on the 20 m SRT) and the final GEE-linear regression models presented here were fit to establish its multiply-adjusted association with age, anemia, wasting, and stunting (using WHO age-and sex-based definitions). Additional alternative variables that were explored during model development included continuous variables for weight, height, other observed anthropometrics, hemoglobin, and/or hookworm and schistosomiasis burden (using log-transformed egg counts).We also explored the use of dichotomous/polyotomous variables for malaria or filaria infection, and infection intensity categories (light/moderate/heavy) for hookworm and schistosomiasis. All analyses were performed using SAS 9.2 (Cary, NC) and SPSS 17 (Chicago, IL).
Of the 2034 children 5–18 years old who participated in the surveys, 1950 children (95.9%) with complete parasitological data completed the 20 m SRT, and were included in the final analysis. Seven children refused to run, and 23 (1.1%) were unable to participate due to limiting physical conditions that included asthma, seizures, pregnancy, club-foot, leg wounds and feeling unwell. Thirty-six (1.8%) did not wait for fitness testing and left after providing their biological samples.
After a brief explanation and demonstration of the test, study participants were batched in groups of 5–15 individuals for testing. The running surface was dirt and most of the children ran barefoot, with which they typically felt most comfortable. There was a very good overall understanding of the test, with occasional need for repetition of instructions and a few false starts. When this happened, a rest period was instituted that lasted between 15–20 minutes. In performing the test, several situations were taken into consideration: i) If a child was lagging behind the recorded marked pace he/she was asked to stop and the level obtained was recorded accordingly, ii) If a false start happened, all children were asked to re-start, iii) If tripping occurred, the child was asked to stop and after a recovery period he/she restarted the test.
No significant differences were observed among the four villages in terms of sex or age distribution, however significant inter-village differences were observed in the prevalence of anemia, mean hemoglobin and malnutrition parameters as shown in Table 1. Village-specific parasite prevalence is illustrated in Figure 4. Prevalence of Sh was highest (25%–62%) followed by geohelminths (15–35%), P. falciparum (9–24%) and W. bancrofti (4–9%).
Both acute and chronic undernutrition were present in each of the four communities, as measured by wasting and stunting prevalence, respectively (see Table 1). Growth references from the CDC-2000 [24] as well as from the newer WHO-2006 [25] were used for comparison. When both results were compared, statistically significant differences in nutritional scoring were detected between these two standard references, with CDC growth standards indicating higher levels of undernutrition when compared to WHO growth standards (data not shown).
As was expected, based on growth physiology, there were marked gender differences in 20 m SRT performance scores, so the final analysis of fitness outcomes is stratified here by sex. (Mean scores per age group for boys and girls are shown in Table S2 in the supporting information files). Because inter-village variations in fitness outcomes were small, all village data were merged and included in a single morbidity model. A separate model was run to compare estimated VO2 max results for Kenyan children with the reference Canadian cohort [29]. Initial bivariate parameter estimates for each potential explanatory variable, along with their 95% confidence interval (accounting for household clustering) are shown in Table 2. Several multivariable modeling strategies were then explored, which revealed co linearity between infection and anemia consistent with the pathways to exercise intolerance shown in Figure 1. For this reason, infection status was ultimately excluded from the final model. The results from the best-fit Generalized Estimated Equation (GEE) multivariable modeling are summarized in Table 3.
For girls, in bivariate analysis, being anemic and harboring hookworm infection negatively correlated with fitness scores (estimates of −0.299 and −0.124 respectively, Table 2). Increasing age significantly associated with better aerobic capacity (estimate 0.080 per year of age). When adjusted for other covariates (Table 3), stunting reached significance (estimate −0.310) and both anemia and age remained as significant predictors of fitness score.
For boys, bivariate correlation analysis indicated a positive association between increasing age and better fitness scores (estimate 0.257 per year), as had been seen with girls. Malnutrition parameters and anemia negatively correlated with fitness scores, both in bivariate as well as multivariable analysis (stunting adjusted estimate-1.007; wasting adjusted estimate −0.829; anemia adjusted estimate −0.282). After this multivariable adjustment, no additional association was seen between fitness and parasite infection per se.
VO2 max outcomes. As detailed in Table S2, there were marked differences in mean fitness scores according to age and sex. Boys manifested a relatively constant VO 2 max (at about 51 ml kg−1 min−1) from ages 5–10 in all villages (Figure 5). This was, however, followed by an evident decline to 42 ml kg−1 min−1 by 18 years of age. The girls' results show a progressively descending VO2 max with age, from 53 ml kg−1 min−1 at 5 years to 32 ml kg−1 min−1 at 18 years old. Kenyan boys, when measured against a reference group of Canadian boys, performed comparably up to age 10, followed by a relatively sharp decline in VO2 max relative to the Canadian standard. Canadian and Kenyan girls had few differences across all age groups (Figure 5).
There is a need for a standardized aerobic fitness test in epidemiological surveys. In the past, fitness has been measured among children harboring parasitic infections in several different ways, but none have proven to be both sufficiently easy to perform and reliable so as to become a standard test for field studies [1], [11], [16], [32]. The Harvard Step Test (HST) has been one of the most widely implemented field tests in previous studies. For example, it has been used to monitor changes in fitness scores after treatment of schistosomiasis and geohelminths in Kenya [13], [33]. In Brazil, the HST has been used to document the association between diarrheal disease and decreased fitness [1]. In Zimbabwe it was used to assess physical work performance and improvement in adult cane cutters after anti-schistosomiasis treatment [14] and in China it was a useful tool in advanced adult Schistosoma japonicum disease [15]. However, the HST has never been formally validated in younger children. Because the HST requires accurate timing and pulse taking, the number of children who can be tested per day is limited, and there are questions about its operational limitations in terms of inter- and intra-observer variability. Our group pilot tested the HST before exploring other exercise options, and found that there was a wide inter and intra-recorder variation in the required pulse measurements at 1, 5, and 10 minutes post-exercise.
Accelerometers have also been used to measure spontaneous activity [11], but activity does not necessarily represent fitness [34]–[36]. Physical fitness relates to a normal physiological functional capacity allowing the individual to have adequate oxygen supply to the tissues. For the accelerometer approach, one must consider that cultural differences contribute to different activity patterns for boys and girls, and these differences need to be accounted for in the interpretation of activity data. Questionnaires have been used for activity assessments in Mozambique, but were found to have limited utility for estimating energy expenditure [16]. Other tests addressing strength and endurance have been used for testing adolescent girls in Senegal, and these have documented apparent fitness reductions in an undernourished group [37].
Our choice of the 20 mSRT was based on its simplicity as a field test and its validity vis a vis laboratory-based physiological testing [6]–[8], [38]. Since its development by Leger and Lambert in 1982 [38], the 20 mSRT has been widely used in the developed world as a means of estimating aerobic capacity in adults and children. Its prediction of the maximum aerobic capacity (VO2 max) is calculated by age and gender-adjusted linear regression from the maximum speed obtained in the test [29]. In this benchmark Canadian study, the accuracy of the 20 m SRT was validated against standard multistage treadmill testing (correlation of SRT score to the VO2 max attained was r = 0.90). Test-retest reliability coefficients were 0.89 for children and 0.95 for adults [29]. Since Leger and Lambert's initial description in 1982 [38], numerous studies (all in developed countries) have reported similar test-retest reliability with applicability for large scale fitness assessments [39], [40]. In Germany, Mechelen et al. performed a study in order to counter-validate the 20 mSRT in children against direct measurements of VO2 max using a multistage maximal treadmill test in boys and girls ages 12–14 years. They concluded that the 20 mSRT is a valid tool to be used to evaluate the maximal aerobic power in children [41].
With slight variations in protocol, the 20 mSRT may be the most widely used aerobic fitness field test among children and adolescents in industrialized countries [6], [7]. It has also been used in asthmatic children to monitor cardiorespiratory changes during an aerobic training program [40]. However, to our knowledge, the 20 mSRT has never been used as a disability measuring tool in a low-resource setting. Aandstad and colleagues from Norway used the 20 mSRT in Tanzania to compare aerobic fitness in Tanzanian and Norwegian children. However, this study did not take into consideration infection or other morbidity status of the participants [42].
Although we were not able to compare 20 m SRT results to formal laboratory or ergometer testing, our results conform to those obtained in Canada in correlating fitness with increasing age, increasing height, and gender differences. This strongly supports the validity and applicability of the 20 m SRT as a useful tool in gauging fitness in less-developed areas. Consistent with international testing results (summarized in a meta-analysis of 109 studies from 37 countries [8]) we found that boys, overall, had better scores than girls. Comparison between our cohorts reveal Kenyan scores to be, overall, below those obtained in Canada for children 5–18 years [28], based, primarily on boys being less fit after 10 years of age.
In our field-based studies of parasitic disease burden, we approached local validation of the 20 m SRT by examining whether sex-specific and age-adjusted fitness scores were significantly modified by physical conditions such as anemia that are known to limit physical fitness in other populations. As expected, anemic girls and boys scored significantly lower on the 20 mSRT than their non-anemic local counterparts, reflecting a measurable relative disability. Girls with hookworm infection had more exercise intolerance than boys, however this effect was not seen in the multivariable-adjusted model suggesting that the hookworm effect was mediated through anemia or nutritional variables in the model. Malaria or filarial infection were not significantly associated with aerobic capacity in either gender, but because these two infections were relatively rare, we might not have had sufficient power in the study to see an independent effect for each infection. The bivariate association seen in boys between schistosomiasis and increased fitness may reflect a selective increase in exposure to infected water in more active children, as previously described among South African boys [34]. We hope to further clarify this effect as we evaluate personal activity patterns in our expanded study.
Malnutrition parameters, in particular, stunting and wasting, emerged as strong predictors of decreased fitness, predominantly in boys. This is consistent with findings of previous studies about stunting [35], which indicate that children develop better biomechanical motor efficiency as their height increases. However, wasted boys were also significantly less fit, underscoring the importance of assessing both types of acute and chronic measures of nutritional status when measuring disability.
There were limitations to our study. First, this cross-sectional study can only indicate significant associations with reduced fitness, but cannot indicate causality. We did not approach the comprehensive measurement of fitness testing in terms of aerobic capacity, strength, flexibility, or body composition. However, we were able to establish that the 20 m SRT is feasible, and appears to be reliable in a rural, resource-limited setting, with minimal requirements for observer and participant training. As a limitation of our analysis, misclassification of infection prevalence by underdiagnosis of active infections may have limited our ability to establish a direct association between infection status and fitness. Given that our diagnoses were based on screening parasitology of a single day's blood, urine or stool specimens using methods known to have incomplete sensitivity [18], [43]–[46], parasite prevalence (as well as infection intensity) was, no doubt, underestimated in our study. In our future analysis, we plan to refine our estimates of past and present infection with specific anti-parasite antibody detection [47].
In summary, this study was able to link fitness, as measured by a low-technology field test, with prevalent anemia and growth stunting, which are known morbidity outcomes affecting children with chronic parasitic infections. We believe the results presented in this paper can serve as a point of reference for other projects aiming to measure fitness in low-resource settings. Prior international standardization of the 20 mSRT makes it an especially valuable tool for refining estimates of disease burden in less-developed countries. Beyond its use in epidemiological research, it could be easily implemented in rural schools as a screening fitness test to detect underlying health conditions, as is commonly done in industrialized countries. Through simplified detection of sub-clinical morbidity at the community level, children at risk for anemia and/or associated parasite infections could then be brought to earlier medical attention, thereby enhancing the impact of national control programs.
|
10.1371/journal.pntd.0004029 | Rapidly Escalating Hepcidin and Associated Serum Iron Starvation Are Features of the Acute Response to Typhoid Infection in Humans | Iron is a key pathogenic determinant of many infectious diseases. Hepcidin, the hormone responsible for governing systemic iron homeostasis, is widely hypothesized to represent a key component of nutritional immunity through regulating the accessibility of iron to invading microorganisms during infection. However, the deployment of hepcidin in human bacterial infections remains poorly characterized. Typhoid fever is a globally significant, human-restricted bacterial infection, but understanding of its pathogenesis, especially during the critical early phases, likewise is poorly understood. Here, we investigate alterations in hepcidin and iron/inflammatory indices following experimental human typhoid challenge.
Fifty study participants were challenged with Salmonella enterica serovar Typhi and monitored for evidence of typhoid fever. Serum hepcidin, ferritin, serum iron parameters, C-reactive protein (CRP), and plasma IL-6 and TNF-alpha concentrations were measured during the 14 days following challenge. We found that hepcidin concentrations were markedly higher during acute typhoid infection than at baseline. Hepcidin elevations mirrored the kinetics of fever, and were accompanied by profound hypoferremia, increased CRP and ferritin, despite only modest elevations in IL-6 and TNF-alpha in some individuals. During inflammation, the extent of hepcidin upregulation associated with the degree of hypoferremia.
We demonstrate that strong hepcidin upregulation and hypoferremia, coincident with fever and systemic inflammation, are hallmarks of the early innate response to acute typhoid infection. We hypothesize that hepcidin-mediated iron redistribution into macrophages may contribute to S. Typhi pathogenesis by increasing iron availability for macrophage-tropic bacteria, and that targeting macrophage iron retention may represent a strategy for limiting infections with macrophage-tropic pathogens such as S. Typhi.
| An adequate supply of iron is essential for both human hosts and their infecting pathogens. Hepcidin is the human hormone that controls the quantity and distribution of iron throughout the body. During infections, hepcidin activity may redistribute iron away from serum and into macrophages, potentially affecting pathogen replication, depending on the niche of the invading microbe. However, the involvement of hepcidin in human bacterial infections remains poorly investigated. Similarly, the pathogenesis of typhoid fever, caused by infection with Salmonella Typhi is also poorly understood. We therefore investigated the behaviour of hepcidin and other iron/inflammation-related parameters during the course of typhoid fever in human volunteers challenged experimentally with Salmonella Typhi. Hepcidin concentrations rose rapidly during acute typhoid infection, in parallel with fever. Hepcidin induction was accompanied by a rapid decline in serum iron concentrations, likely reflecting iron sequestration in macrophages (a preferred replication site of Salmonella Typhi). The extent of hepcidin upregulation associated with the extent of serum iron starvation. We hypothesize that hepcidin activity during acute typhoid infection in humans may elevate iron levels in the niche used by the pathogen for replication. Targeting macrophage iron retention should be evaluated as a potential strategy for limiting typhoid fever.
| Typhoid fever is a common infection that follows oral ingestion and invasion of the Gram-negative bacterium Salmonella enterica serovar Typhi (S. Typhi). An estimated 26.9 million cases occurred globally in 2010, disproportionately affecting children in resource-limited areas of sub-Saharan Africa and southeastern Asia [1,2].
S. Typhi is a human-restricted pathogen. Unlike non-typhoidal Salmonella infection, which is characterized by rapid-onset gastrointestinal inflammation and diarrheal illness in immunocompetent adults, S. Typhi causes a systemic infection. After ingestion, bacteria disseminate through the reticuloendothelial system, where they are thought to incubate for 7–14 days. Clinical illness then develops, characterized by fever and non-specific symptoms including headache, nausea and abdominal pain, and accompanied by bacteremia [3]. However, detailed understanding of typhoid pathogenesis remains limited, in part since convincing small-animal infection models are lacking [4]. An experimental human S. Typhi challenge model was recently reestablished, presenting a unique opportunity to investigate typhoid pathogenesis in a controlled setting in the natural host [5–7].
Conflict exists between hosts and invading pathogens over the control of the critical micronutrient, iron (reviewed in [8,9]). To limit free iron availability, mammalian hosts sequester iron using high-affinity iron-binding proteins including transferrin, lactoferrin, haptoglobin, hemopexin and the iron storage protein ferritin. To counteract this, many bacteria express higher affinity siderophores (e.g. enterobactin) that appropriate iron from host iron binding proteins; host-produced siderophore-binding proteins such as lipocalin-2 in turn counter these. A further host response to infection involves the rapid induction of hypoferremia, where iron becomes sequestered in reticuloendothelial macrophages and therefore excluded from serum [8,10]. This state may be disadvantageous to extracellular pathogens [11] but potentially could be exploited by intracellular, macrophage-tropic bacteria including S. Typhi [12]. Together, the host mechanisms aimed at sequestering iron from invading microorganisms are considered to contribute to innate protection against infection, often termed “nutritional immunity” [8].
In recent years, many genes involved in mammalian iron homeostasis have been discovered [13], meaning that the molecular basis of iron perturbations during infections can be investigated in a new light. Amongst these, hepcidin stands out as the central regulator of systemic iron balance [14]. Hepcidin dictates dietary iron uptake and recycling of red cell iron by binding and causing degradation of the sole known iron exporter ferroportin, which is expressed on duodenal enterocytes and iron-recycling macrophages [15]. Consequently, high hepcidin levels effect iron exclusion from serum, through blocking dietary iron uptake and preventing macrophage iron release. Hepcidin is induced homeostatically in response to increased plasma and liver iron [16,17], but is also an acute phase protein upregulated by inflammatory cytokines, notably IL-6 [18–20]. Thus, elevated hepcidin concentrations during inflammation and infections contribute to hypoferremia [11,21] and, if chronic, to iron-restricted erythropoiesis and anemia [22].
Hepcidin regulation is less well studied in the context of human infection. Analyses to date indicate that hepcidin behavior differs between infections. For example, it is upregulated during uncomplicated malaria [23–26], and during acute, chronic and advanced HIV-1 infection [27,28]; however, it is suppressed during Hepatitis C Virus infection [29], and in severe malarial anemia, where bone-marrow derived signals indicating erythropoietic iron demand likely dominate, suppressing hepcidin production [26,30]. Importantly, hepcidin remains remarkably poorly studied in human bacterial infections. This is despite iron representing a battleground of host-bacterial conflict important enough to have shaped both primate and bacterial genomes alike [31]. Here, we investigate the dynamics of hepcidin in relation to iron and inflammatory indices during acute experimental Salmonella Typhi infection in humans.
Human typhoid challenge was performed with healthy consenting adult volunteers (18–60 years) who had not previously received typhoid vaccination or resided in typhoid-endemic areas for >6 months [7]. Data from two sets of study participants are described: first, from the placebo arm of a vaccine/typhoid challenge study, where participants received an oral placebo vaccine (sodium bicarbonate solution and excipients) 28 days before oral challenge with S. Typhi (n = 30, Study A, Table 1 (baseline data from day of typhoid challenge shown)), and secondly, for more detailed longitudinal analysis, from a previously described cohort challenged orally with S. Typhi in a preliminary dose-escalation challenge model (n = 20, Study B, Table 1) [7].
Full details of the challenge model used in both studies are described in Waddington et al [7]. Briefly, participants ingested a single freshly prepared dose of S. Typhi (Quailes strain, 104 CFU) suspended in sodium bicarbonate solution. After challenge, study participants were reviewed daily for 14 days; blood samples were collected on alternate days, at typhoid diagnosis and intervals thereafter. In typhoid-infected and non-infected participants, the mean blood volumes collected during the 28 days following challenge were approximately 920mL and 600mL respectively. “Typhoid diagnosis” was defined a priori by clinical and/or microbiological endpoints: temperature ≥38°C sustained for ≥12 hours and/or blood culture evidence of S. Typhi bacteremia, respectively. Antibiotic treatment (ciprofloxacin, 500 mg twice daily, 14 days) was initiated upon attainment of either diagnostic criterion and in all remaining participants at Day 14. Actual challenge doses were determined, and quantitative blood culture was performed at typhoid diagnosis, as previously described [7].
Serum samples were filtered prior to analyses using Costar Spin-X low protein binding 0.22μm cellulose acetate membrane filters (Corning). Spin-filtering had no effect on hepcidin measurement (n = 4 samples, p = 0.36 (paired t-test)). Hepcidin was quantified by ELISA using the hepcidin-25 EIA kit (Bachem), with the manufacturer’s protocol modified to incorporate a 9-point, 2-fold serial dilution standard curve. Samples were diluted to 10% or 5% prior to analysis. The lower limit of detection (LLOD) was 0.08 ng/mL, calculated as described previously [27]. Samples returning a reading below LLOD were assigned the value (LLOD*dilution factor)/2.
Serum ferritin (Architect Ferritin Assay) was quantified using the Abbott Architect 2000R automated analyzer (Abbott Laboratories); C-reactive protein (CRP) (MULTIGENT CRP Vario Kit, with high sensitivity calibrators), serum iron and Unsaturated Iron Binding Capacity (UIBC, MULTIGENT Iron Kit) were quantified using the Abbott Architect c16000 automated analyzer (Abbott Laboratories). Transferrin saturation (Tsat) was calculated using the formula: Tsat = ((Serum Iron)/(Serum Iron + Unsaturated Iron Binding Capacity))*100. CRP concentrations above and below the assay limits of detection (160 and 0.1 mg/L) were assigned the values 160 mg/L and 0.05 mg/L respectively.
Plasma cytokine concentrations were measured in duplicate using a custom TNFα / IL-6 Luminex panel (MILLIPLEX MAP kit, Millipore) according to the manufacturer’s instructions. Readings with % Coefficient of Variance >30% were excluded; those falling below the LLOD were allocated the value 1.6 pg/mL (LLOD/2).
Hemoglobin, red blood cell counts, and mean corpuscular volume (MCV) were quantified by routine hematologic analysis.
Statistical analyses were performed using Prism (version 6, GraphPad Software Inc.), SPSS (version 16.0, IBM SPSS), STATA/SE13.1 (Statacorp) and R statistical language [32]. All raw data can be found in the file S1 Dataset.
For indices that were not normally distributed (hepcidin, ferritin, and CRP in all cases; additionally serum iron and transferrin saturation when considering data other than baseline data), geometric means were compared, or data were log-transformed prior to analysis. Differences between indices pre-challenge and at typhoid diagnosis (Study A) were evaluated using paired t-tests. In correlation analysis, Pearson correlation coefficients were computed; in cases where study participants contributed more than a single observation, correlation analyses were adjusted accordingly by using regression with clustered errors (STATA/SE13.1), which adjusts the confidence intervals of the regression coefficients to account for intra-cluster correlation, as is likely when multiple observations from the same individuals are included. Statistical tests returning p<0.05 were considered significant.
Time-course analyses were performed using the packages fields [33], nlme [34] and lme4 [35] within R statistical language [32] as follows. (i) Normalization of time series: Since the time between typhoid challenge and diagnosis (TD) varied between participants, the time variable “day relative to TD” was used, with TD = 0 being day of diagnosis. (ii) Smoothing spline regression: Mean analyte measurements across all subjects for each day relative to TD were calculated; samples from day of typhoid challenge (‘baseline’) and the final visit (Day 14 post-challenge) were grouped separately. A smoothing spline regression was applied with smoothness estimated from the data by generalised cross validation (GCV) [36]. 95% pointwise prediction intervals and conservative simultaneous Bonferroni bounds were calculated. (iii) Assessment of the effect of time relative to TD on analyte concentrations: linear mixed-effects models were fitted using a described model formulation [37] and computational framework [38]. The categorical variable “day relative to TD” was modeled by fixed effects; variability between individuals was captured using random effects. The null hypothesis that there is no significant difference in analyte levels over time after challenge was tested using the Wald test; specific pairwise comparisons between analyte concentrations at baseline (day of challenge) and later time-points were tested by t-tests, accounting for subject-specific variability. Tests returning p<0.05 were considered significant.
The National Research Ethics Service approved both studies (Oxfordshire Research Ethics Committee A, 10/H0604/53 and 11/SC/0302). They were performed in accordance with the principles of the ICH-Good Clinical Practice guidelines and amendments. All study participants provided written informed consent in accordance with the Declaration of Helsinki on at least one occasion, as previously described [7].
Characteristics of study participants from two experimental typhoid challenge studies are given in Table 1. The typhoid attack rates (percent typhoid-diagnosed participants by Day 14) in Study A and B were 67% (20/30) and 65% (13/20), while mean duration between challenge and typhoid diagnosis was 7.4 (95% Confidence Interval (CI): 6.6–8.3 days) and 7.7 days (6.7–8.7 days), respectively. There were no significant differences in the baseline characteristics of participants recruited to Study A or B, except that the challenge dose and transferrin saturations were marginally higher in Study B. Considering all participants from Studies A and B together, significant associations between log10-hepcidin and log10-ferritin levels (p<0.0001, r2 = 0.642; S1 Fig, panel A) and between hepcidin and both transferrin saturation (S1 Fig, panel B) and hemoglobin (S1 Fig, panel C) were found in baseline, pre-challenge samples. Male participants had significantly higher baseline hemoglobin, hepcidin and ferritin levels than females (S1 Fig, panels D-F). These observations are typical of healthy adult populations.
In univariate analyses, there were no significant differences in age, sex, weight or challenge dose, or in baseline hematological or iron-related parameters between those subsequently diagnosed or not diagnosed with infection, even when participants from the two studies were pooled together to increase power (S1 Table). Amongst individuals diagnosed with typhoid, we found no association between the time to typhoid diagnosis and baseline iron status as indicated by ferritin (r2 = 0.014, p = 0.505) or hepcidin (r2 = 0.015, p = 0.497). Amongst individuals from the two studies who were diagnosed with typhoid, increasing challenge dose was significantly negatively associated with time-to-diagnosis (r2 = 0.174, p = 0.016) and positively associated with the number of bacteria quantified at diagnosis (r2 = 0.241, p = 0.007).
To investigate the extent to which typhoid infection was associated with changes in hepcidin and other iron indices, we analyzed serum samples collected at baseline and on day of typhoid diagnosis in participants challenged in Study A (n = 19/20, 7 females and 12 males). Amongst these individuals, the mean time to typhoid diagnosis was 7.4 days (95% CI: 6.6–8.3 days); mean oral temperature was significantly higher at typhoid diagnosis than at baseline (diagnosis: 37.6°C [95% CI, 37.3–37.9°C]; baseline: 36.3°C [36.1–36.5°C]; Fig 1A).
Hepcidin concentrations at typhoid diagnosis were approximately 10-fold higher than at baseline (Fig 1B). This marked hepcidin response was accompanied by hypoferremia demonstrated by a significant decline in mean serum iron and transferrin saturation (Fig 1C and 1D). In contrast, there was a significant increase in the inflammatory markers, CRP and ferritin at diagnosis compared to baseline, although the relative change in ferritin concentration was less notable than that of hepcidin or CRP (Fig 1E and 1F). There were no significant differences in hemoglobin between measurements at baseline and at diagnosis (Fig 1G). Together, these data demonstrate that significant hepcidin upregulation and concurrent hypoferremia are features of the acute phase response to S. Typhi infection.
To assess the kinetics of alterations in hepcidin and other indices following S. Typhi challenge, we analyzed serial samples from Study B, firstly from the 7 participants who did not develop clinical disease following challenge, and secondly from the 13 individuals diagnosed with acute infection. For both groups, up to 7 time points from Day 0 (baseline, challenge day) onwards were analyzed (mean, 6.15 time points).
In those who did not develop typhoid infection, significant reductions in hepcidin, ferritin and hemoglobin concentrations, and in red blood cell counts, were observed during the 14-day study period (S2 Fig, panels A-D). There was also suggestion of decline in serum iron and transferrin saturation (S2 Fig, panels E/F). This likely relates to the repeated phlebotomy required by the study protocol, causing reduction of iron indices including hepcidin. CRP concentrations and oral temperatures remained low/normal throughout confirming the absence of a systemic inflammatory response in challenged but non-infected individuals (S2 Fig, panels G/H). Thus, the following time course data from typhoid-infected individuals must be interpreted in the light of these study protocol effects on hematological parameters.
In participants who developed typhoid fever, increases in temperature were measured from 48 hours prior to diagnosis (Fig 2A). A concomitant rise in hepcidin concentration was observed, maximal 2 days after diagnosis; temperature and hepcidin levels normalized towards baseline levels over approximately 4 days following treatment initiation (Fig 2B). Similarly, significant declines in serum iron (commencing prior to diagnosis and reaching a mean nadir of 4.9 μmol/L two days post-diagnosis, down from 14.5 μmol/L at baseline, Fig 2C) and transferrin saturation (mean 6% at nadir two days post-diagnosis, down from 28% at baseline, Fig 2D) were observed; these indices, like hemoglobin (Fig 2E), were lower at the final time point (Day 14) than at baseline (Fig 2C and 2D and 2E), likely reflecting the effect of venesection described above (S2 Fig). However, the possibility of a hepcidin-mediated block in iron absorption during infection contributing to this observation should not be excluded.
A significant induction of the acute phase protein CRP was also observed, escalating marginally later than the initial perturbations to hepcidin and transferrin saturation, but similarly peaking 2 days after typhoid diagnosis (Fig 2F). The iron storage protein ferritin, also an acute phase protein, was induced later than hepcidin or CRP and took longer to resolve towards baseline levels (Fig 2G).
Together, these data indicate that the kinetics of hepcidin perturbations and the associated hypoferremia during acute S. Typhi infection mirror typhoid-associated fever and CRP induction.
We next investigated relationships between hepcidin concentration and serum iron status in those exhibiting typhoid-related inflammation and those who were not. In this analysis, we included all data from the study from both diagnosed and non-diagnosed individuals, using regression with clustered errors, thereby accounting for the inclusion of multiple observations derived from the same individuals. When there was evidence of acute inflammation (defined as CRP >5 mg/L), significant negative associations between hepcidin and both serum iron and transferrin saturation were observed (Fig 3A and 3B). In contrast, when acute inflammation was absent (CRP <5 mg/L), significant positive associations between hepcidin and serum iron parameters were found (Fig 3A and 3B). Thus, larger hepcidin responses predicted more profound hypoferremia in the context of inflammation, but the opposite in non-inflamed samples, when they presumably reflected iron status. The latter effect was also noted in baseline challenge day samples (S1 Fig). Unlike hepcidin, ferritin did not correlate with the extent of hypoferremia during inflammation, although it did associate positively with serum iron parameters in non-inflamed samples (Fig 3C and 3D). These data indicate non-equivalence of these two indices of iron status, as noted in previous work [39], and suggest hepcidin may be more closely linked to hypoferremia in the context of the acute inflammation observed during typhoid infection.
Hepcidin upregulation during acute phase responses is typically associated with STAT3 activation following signaling by IL-6 and potentially other cytokines (e.g. IL-22) [18–20]. We therefore assessed IL-6 concentrations in plasma samples from the individuals from Study B who developed typhoid infection. We only observed a weak IL-6 response in a subset of individuals (Fig 4A, see S3 Fig for individual profiles); in the majority of individuals, IL-6 upregulation was not detected. Modest TNF-alpha responses were more consistent, with the highest levels recorded day 2 post-diagnosis in most individuals (Fig 4B, see S3 Fig for individual profiles). These data suggest the cytokine response during typhoid infection may have been blunted, as previously described [40], and that determinants other than serum IL-6 may be responsible for the hepcidin upregulation observed in this context.
Salmonella Typhi is a significant human pathogen, leading to a major global burden of disease particularly among children and younger adults in endemic settings [1,2]. Evolution from a common Salmonella ancestor is thought to have occurred ~50–100,000 years ago [41]. However, the basis for the evolution of its ability to evade host defenses and cause systemic infection remains poorly characterized. Understanding how S. Typhi interacts with the human host environment, including the macrophage niche, is crucial in deciphering its pathogenicity and for devising prevention or eradication strategies. The battle for iron is a key determinant of host-bacterial interactions [8,9,31]. Here, using an experimental human typhoid challenge model, we track for the first time in an invasive human bacterial infection the behavior of the iron regulatory hormone hepcidin and its relationship to perturbations in iron parameters, inflammatory markers, and fever: significant hepcidin upregulation, accompanied by a profound decline in serum iron was observed in participants diagnosed with typhoid infection.
Hepcidin has several characteristics reflecting a likely ancestry in immunity to infection. It is a liver-derived acute-phase peptide induced via the inflammatory JAK/STAT3 signaling pathway [18–20,42]. It structurally resembles antimicrobial beta-defensins and has modest antimicrobial activity itself [43,44]. Hepcidin’s involvement in human infection pathogenesis has been widely proposed, likely relating more to its ability to rapidly alter systemic partitioning of iron than its direct antimicrobial activity [12]. Despite this, its regulation and influence on the pathogenesis of human bacterial infection remains poorly investigated. In humans, significant hepcidin upregulation has been observed during sepsis [45,46], during tuberculosis (with and without HIV coinfection) [47,48], and to a less notable extent in children with concurrent Helicobacter pylori infection and iron deficiency anemia [49]. The longitudinal behavior of hepcidin has been assessed during experimental uncomplicated malaria (where modest increases in hepcidin and IL-6, associated with changes in systemic iron parameters, were observed) [23] and during the acute phases of HIV-1, Hepatitis B Virus and Hepatitis C Virus infections [27]. However, the nature of longitudinal perturbations in hepcidin during the acute phase of a bacterial infection in humans has never been investigated.
In study participants diagnosed with acute typhoid infection, we found a marked upregulation of hepcidin around the time of diagnosis, coincident with appearance of fever. Hepcidin concentrations remained high for at least 48-hours during acute infection irrespective of prompt antibiotic therapy, and resolved to normal levels from 4 days after diagnosis. We predict that hepcidin would remain high for considerably longer if the infection were left untreated. Significant elevations in the acute phase proteins, CRP and ferritin, and striking declines in serum iron and transferrin saturation (from 28% at baseline to 6% at nadir 2 days post-diagnosis) were also evident. The data suggested hepcidin activity from 1–2 days prior to typhoid diagnosis, and are consistent with the previous description of hypoferremia in the Maryland typhoid challenge in the 1970s [50]. Furthermore, our data indicated that, when acute inflammation was present, the extent of hepcidin upregulation significantly predicted the degree of hypoferremia; in contrast, in normal non-inflamed conditions, hepcidin positively associated with serum iron parameters. Given these data, hepcidin and the associated hypoferremia should be considered for investigation as potential biomarkers of acute infection.
Hepcidin upregulation in the context of inflammation/infection is typically linked to signaling via the IL-6/STAT3 pathway [19,42]. However, we only detected a modest elevation in plasma IL-6 around typhoid diagnosis, with several participants maintaining IL-6 levels below detectable levels at each sampling time point. When IL-6 was detected, it was at considerably lower levels than in other conditions where hepcidin is notably induced: IL-6 was typically one or more orders of magnitude higher during uncomplicated malaria [24], sepsis [45], or experimental endotoxemia [21]. Similarly, although TNF-alpha was induced, the levels detected were relatively low. Since IL-6 and TNF-alpha data were not available from the day of diagnosis or the following day, we cannot exclude the occurrence of a stronger, transient IL-6 induction during these two days. Similarly, we cannot exclude more local but significant cytokine effects in intestine, portal circulation and liver leading to hepcidin upregulation that may not be detected in systemic circulation. Nevertheless, one established feature of S. Typhi infection is a blunted cytokine response [40]. Several factors, most prominently the Vi-capsular polysaccharide, enable S. Typhi to evade innate immune responses (for example by enabling evasion of detection by complement [51]) and to establish systemic infections without clinical sepsis [40,52]. It is therefore possible signals besides IL-6 are involved in the significant acute phase response and hepcidin induction during acute typhoid infection. Thus, despite being an immunologically evasive infection, dramatic hepcidin up-regulation and hypoferremia remain features of typhoid in humans. Mechanistic links between hepcidin and hypoferremia should not, however, be concluded from observational data such as these. Nevertheless, based on data from other settings linking hypoferremia with hepcidin upregulation during infection and inflammation [11,53], we hypothesize that hepcidin plays a role in mediating hypoferremia and that the hypoferremia reflects rapid sequestration of iron in macrophages during acute typhoid infection.
The interplay between Salmonella enterica infection and iron has been well studied, typically through using in vitro or in vivo S. Typhimurium models. The iron exporter ferroportin is upregulated via NO-mediated Nrf2 activation in ex vivo S. Typhimurium-infected macrophages, reducing macrophage iron availability—a state that restricts bacterial replication [54–56]. Despite this mechanism, hepcidin induction and hypoferremia are still observed during invasive murine S. Typhimurium infection, associating with macrophage iron sequestration via reduced ferroportin activity; interference with hepcidin upregulation in this context, leading to reduced cellular iron levels, is protective for the host [55]. Correspondingly, hepcidin administration to infected ferroportin-expressing cells in vitro enhances bacterial replication [54].
As reflected by their different pathologies, there are key differences between non-typhoidal and typhoidal Salmonella enterica serovars, despite high degrees of sequence homology [57]. These include the expression of virulence determinants (most notably the Vi-capsular antigen) and inactivation of over 200 genes in S. Typhi compared with its cousin S. Typhimurium [57]. Interestingly, several of these inactivated genes relate to iron acquisition pathways [58]. There is evidence that S. Typhi relies heavily on the fepBDCG enterobactin ferric iron uptake system [57], which is upregulated in isolates from typhoid patients [59]; the upregulation of this system likely reflects the difficulty of obtaining iron from a host environment where iron availability is typically scarce.
In conclusion, during human S. Typhi infection, where hepcidin is strongly upregulated and a marked hypoferremia is observed, we hypothesize that hepcidin activity and macrophage iron retention are dominant over any macrophage cell-intrinsic protective mechanisms aimed at reducing cellular iron content. Stimulating a strong hepcidin response may represent another bacterial strategy for ensuring iron supply to facilitate replication. Therefore, in typhoid (and possibly other macrophage-tropic intracellular bacterial infections), hepcidin-induced hypoferremia may be actively disadvantageous to the host rather than being a stereotypical protective response to infection [11]. A recent study in humans demonstrated that spiegelmer-based hepcidin neutralization during experimental endotoxemia can prevent induction of hypoferremia [53]. Whether targeted manipulation of hepcidin and host iron distribution offers a potential strategy for treating intracellular infections should be investigated further, particularly in an era of increasing antibiotic resistance.
|
10.1371/journal.pntd.0001414 | Tracing of the Bile-Chemotactic Migration of Juvenile Clonorchis sinensis in Rabbits by PET-CT | Adult Clonorchis sinensis live in the bile duct and cause clonorchiasis. It is known that the C. sinensis metacercariae excyst in the duodenum and migrate up to the bile duct through the common bile duct. However, no direct evidence is available on the in vivo migration of newly excysted C. sinensis juveniles (CsNEJs). Advanced imaging technologies now allow the in vivo migration and localization to be visualized. In the present study, we sought to determine how sensitively CsNEJs respond to bile and how fast they migrate to the intrahepatic bile duct using PET-CT.
CsNEJs were radiolabeled with 18F-fluorodeoxyglucose (18F-FDG). Rabbits with a gallbladder contraction response to cholecystokinin-8 (CCK-8) injection were pre-screened using cholescintigraphy. In these rabbits, gallbladders contracted by 50% in volume at an average of 11.5 min post-injection. The four rabbits examined were kept anesthetized and a catheter inserted into the mid duodenum. Gallbladder contraction was stimulated by injecting CCK-8 (20 ng/kg every minute) over the experiment. Anatomical images were acquired by CT initially and dynamic PET was then carried out for 90 min with a 3-min acquisition per frame. Twelve minutes after CCK-8 injection, about 3,000 18F-FDG-labeled CsNEJs were inoculated into the mid duodenum through the catheter. Photon signals were detected in the liver 7–9 min after CsNEJs inoculation, and these then increased in the whole liver with stronger intensity in the central area, presenting that the CsNEJs were arriving at the intrahepatic bile ducts.
In the duodenum, CsNEJs immediately sense bile and migrate quickly with bile-chemotaxis to reach the intrahepatic bile ducts by way of the ampulla of Vater.
| Clonorchis sinensis adults habituating in the bile duct cause clonorchiasis endemic in East Asian countries, in which about 15–20 million people are supposedly infected. It has previously been reported that C. sinensis metacercariae excyst in the duodenum and that the juvenile flukes migrate to the bile duct through the ampulla of Vater in 4–7 hours. Recently advanced imaging technologies have enabled visualization of movements and localizations of parasites in mammalian hosts. From present study, we found the following: newly excysted C. sinensis juveniles (CsNEJs) were efficiently in vitro radiolabeled with 18F-FDG since CsNEJs have glucose transporters; CCK-8-induced gallbladder contraction was various rabbit to rabbit; CsNEJs promptly recognized bile and migrated up the duodenum to reach the intrahepatic bile ducts by way of the ampulla of Vater and the common bile duct as early as 7–9 minutes after inoculation. Some CsNEJs responding slowly to the bile delayed arriving at the distal bile capillaries. It was visualized for the first time that the CsNEJs migrate quickly within 10–20 minutes from the duodenum to the intrahepatic bile duct. These findings provide fundamental information on the migration of parasites living in the biliary passages of mammals.
| Human Clonorchis sinensis infections are endemic in East Asia countries, such as China, Vietnam, and Korea, where 15–20 million people are estimated to be infected [1]. In South Korea, clonorchiasis is currently the most prevalent parasitic infection and estimated to infect 1.3 million people [2]. C. sinensis infected patients suffer from abdominal pain, hepatomegaly, obstructive jaundice, indigestion, and complications of cholecystitis, cholelithiasis, and cholangiocarcinoma [3], [4]. Furthermore, recently, C. sinensis was categorized as a Group 1 biological carcinogen by the International Agency for Research on Cancer [5].
Humans are the final host and become infected by eating freshwater fish containing C. sinensis metacercariae. Ingested metacercariae excyst in the duodenum due to trypsin stimulation [6], and the newly excysted C. sinensis juveniles (CsNEJs) migrate to the intrahepatic bile duct. The migration route of CsNEJs has been previously examined in experimental animals. In rabbit experiments, the common bile duct was first ligated surgically then C. sinensis metacercariae were administered to the rabbits through a gastric tube. One month later, adult C. sinensis were searched for in the bile ducts, but were not found. Based on this finding it was suggested that CsNEJs migrate through the common bile duct to the intrahepatic bile ducts [7], and this has been taken to be the migration route of C. sinensis in mammalian hosts [3].
Parasites such as C. sinensis have specific in vivo migration routes in their hosts, which could be targeted for development of therapeutic and preventive interventions against parasitic diseases. Furthermore, in vivo imaging technologies have been recently developed for the clinical diagnoses of a wide range of diseases, and these techniques have a potential to monitor the movements of CsNEJs.
Molecular imaging has emerged as a discipline at the intersection of molecular biology and in vivo imaging. It enables cellular functions to be visualized and molecular processes to be followed in living organisms in a non-invasive manner. Recently, studies on the visualization of live parasite in hosts have been conducted. Using transgenic Plasmodium parasites, pre-erythrocytic development was visualized; Plasmodium sporozoites entered hepatic cells, developed in a large schizont, and released merozoites in liver [8], [9]. However, these techniques are not applicable to trematodes, because stable transgenic flukes are difficult to be generated.
In mammalian hosts, adult forms of trematodes consume large amounts of glucose to generate and supply energy by running the glycolytic pathway [10]. Adult schistosomes import exogenous glucose, equivalent to their dry body weight every 4 hours from host blood by using glucose transporters in their tegumental membranes [11], [12]. In C. sinensis, glucose transporter and Na+/glucose co-transporter are expressed abundantly in the adult stage but less so in the metacercarial stage as presented in the C. sinensis transcriptome [13]. Adult C. sinensis worms uptake glucose to produce energy in the anaerobic environment of the bile duct [14]. Therefore, we expected that C. sinensis could be labeled with 2-deoxy-2-[18F]fluoro-D-glucose (18F-FDG), a glucose analogue used for the radiolabeling and diagnostic imaging of cancer cells [15]. Thus, by ex vivo labeling CsNEJs with 18F-FDG, we hoped their migration in the final host could be traced in vivo by positron emission tomography-computed tomography (PET-CT).
In vivo imaging techniques have strong merits for the noninvasive tracing on pathogens moving within tissues of living animals, as they involve minimal manipulation and/or euthanasia of animals, and allow repetitive tracking in same animals. Furthermore, as was found in the present study, these techniques make it possible to monitor the distribution and migration of CsNEJs in vivo from the duodenum to the liver or distal bowel. This study was carried out to determine how CsNEJs find their way and how rapidly they migrate to the intrahepatic bile duct by using in vitro 18F-FDG radiolabeling and PET-CT in a rabbit model.
Topmouth gudgeons (Pseudorasbora parva), the second intermediate host of C. sinensis, were purchased at a fish market in Shenyang, Liaoning Province, People's Republic of China. Fishes were ground then digested in artificial gastric juice (8 g of pepsin 1∶10,000 (MP Biochemicals Co., Solon, OH, USA) and 8 ml of concentrated HCl in 1 liter of water) for 2 hr at 37°C [10]. To remove particulate matters, the digested soup was filtered through a sieve of 212 µm mesh. C. sinensis metacercariae (135–145 µm×90–100 µm) were then filtered out using seives of 106 and 53 µm meshes and washed thoroughly several times with 0.85% saline. C. sinensis metacercariae were collected under a dissecting microscope and stored in phosphate-buffered saline at 4°C until required [10].
The metacercarial cyst wall of C. sinensis is thick and can hinder glucose diffusion. Thus to maximize radiolabeling efficiency, metacercariae were excysted and juvenile worms were liberated from cysts. The C. sinensis metacercariae were excysted by treating them with 0.05% trypsin at 37°C for 5 minutes (Gibco, Grand Island, NY, USA) in 1× Locke's solution (150 mM NaCl, 5 mM KCl, 1.8 mM CaCl2, 1.9 mM NaHCO3), a maintaining medium of CsNEJs [16]. CsNEJs were washed 5 times with 1× Locke's solution, and used immediately. CsNEJs were divided into two groups of 10–270 juveniles each; one was of CsNEJs that excysted just before radiolabeling and the other was of the CsNEJs fasted for 24 hours. The two CsNEJ groups were radio-labeled with 18F-FDG by incubating them in 1× Locke's solution containing 74 MBq 18F-FDG at 37°C for 15, 30, or 60 min. After washing 3 times with 1× Locke's solution, radioactivity was measured for 10 min using a PET (GEMINI TF, Philips Healthcare, Cleveland, OH, USA). Numbers of CsNEJs were counted and labeling efficiency was calculated as counts per minute (cpm) divided by number of the CsNEJs. Radio-labeling efficiencies of the CsNEJs in both groups were measured 3 times and significant differences were determined using the student's t-test.
Rabbits (New Zealand White, male, 2.2–2.5 kg) were purchased from Samtako Bio Korea Inc. (Osan, Korea). Rabbits were cared for and handled according to guidelines issued by Chung-Ang University College of Medicine Animal Facility (an accredited facility) in accordance with AAALAC International Animal Care policy. Animal experiments were approved by the institutional review board of the Chung-Ang University animal facility (CAUMD 09-0024).
Gallbladder contraction and emptying time induced by cholecystokinin-8 (CCK–8) varied from rabbit to rabbit. To select rabbits that responded sensitively to CCK-8, cholescintigraphy and 99mTc-mebrofenin (3-bromo-2,4,6-trimethylphenyl carbamoylmethyl iminodiacetic acid) were used. Briefly, rabbits were fasted for 12 hrs and anesthetized with a 0.47 mg/kg Rompun (xylazine hydrochloride; Bayer Korea, Seoul, Korea) and 12.5 mg/kg Zoletil 50 (Zolazepam and Tiletamine; Virvac Korea, Seoul), intramuscular injection. 99mTc-mebrofenin (74 MBq) in 0.5 ml volume was then administered via an ear vein to each anesthetized rabbit. When full of 99mTc-mebrofenin, gallbladders were stimulated to contract by injecting CCK–8 intravenously at 20 ng/kg every 1 min. A dynamic image was taken every 1 min for 1 hour for each rabbit. All images were obtained with a rotating dual-headed gamma camera equipped with a low-energy, high-resolution collimator (Vertex TM, Philips Healthcare, Cleveland, OH, USA) using a 256×256-pixel matrix at an energy range of 20% at 140 keV.
Fresh CsNEJs (n = ∼3,000) were radio-labeled with 18F-FDG by incubating them in a maintaining medium containing 74 MBq 18F-FDG at 37°C for 15 min. CsNEJs were washed 3 times with 1× Locke's solution and then placed in 500 µl of 1× Locke's solution. The procedure was conducted as follows (Figure 1). A rabbit sensitive to CCK-8 was anesthetized with 0.47 mg/kg Rompun and 12.5 mg/kg Zoletil 50 by intramuscular injection and placed in restraints in a supine position on a plastic board. A catheter (5F Simmons II, Cook Co., Bloomington, IN, USA), equipped with a guidewire (0.035″ Radifocus®, Terumo, Tokyo), was inserted through the animal's mouth and its end positioned in the mid duodenum under guidance (Axiom Artis; Siemens, Erlangen, Germany). The rabbit was then moved with the catheter in situ and placed in PET-CT bed. To stimulate gallbladder contraction and bile juice release, 20 ng/kg of CCK–8 was injected intravenously every minute over this experiment [17]. After 12 minutes of CCK-8 injection, 18F-FDG-labeled CsNEJs in 500 µl of 1× Locke's solution were introduced into the mid duodenum through the catheter; residual CsNEJs in the catheter were flushed into the duodenum with 0.5 ml of 1× Locke's solution. One transmission CT image was obtained before the introduction of the 18F-FDG-labeled CsNEJs and a dynamic PET scan then was performed over 90 min with a 3-min acquisition per frame. Finally, one static PET image was scanned for 10 min. This procedure is depicted schematically as a flow-chart in Figure 1.
All photon data were collected using a dedicated PET-CT scanner. PET images were reconstructed after applying CT-based attenuation and scattering corrections using the ordered subset expectation maximization algorithm (2 interations, 16 subsets) with the point spread function. Image analysis was performed on a dedicated workstation using Extended Brilliance Workspace (ver. 3.5.2.2260, Philips Healthcare). A region of interest (ROI) was set on the whole liver in dynamic axial images while referencing corresponding coronal images and radiating photons were counted over each frame. PET images were subsequently visually evaluated for the presence of focal 18F-FDG uptake by radiolabeled CsNEJs. Migration of the CsNEJs to the intrahepatic bile ducts was estimated by semi-quantitatively analyzing photon counts from rabbit liver.
To confirm migration of the CsNEJs to the intrahepatic bile ducts, adult C. sinensis were recovered from the liver of the CsNEJ-inoculated rabbits. Four weeks after image scanning, rabbits were euthanized and C. sinensis adult worms were recovered from the bile ducts by carefully squeezing liver slices. For pathologic section slides, the liver was fixed in 10% neutral formalin, processed along a routine procedure and stained with hematoxylin and eosin. As a negative control, 18F-FDG-labeled CsNEJs were inoculated into two rabbits not injected with CCK-8. In these rabbits, bile is not released from the ampulla of Vater, neither attract the CsNEJs to the bile duct.
Fresh CsNEJs were labeled with 10,760, 7,726 and 13,842 cpm/worm after incubation in radiolabeling media for 15, 30, and 60 min, and fasted CsNEJs were labeled with 11,115, 8,043, and 12,318 cpm/worm when incubated for 15, 30, and 60 min, respectively (Figures 2A & B). Labeling efficiencies were similar in the two groups at all time points. For downstream experiments, fresh CsNEJs were radiolabeled with 18F-FDG at 37°C for 15 min.
To determine an appropriate time point to inoculate the 18F-FDG-labeled CsNEJs in the duodenum after CCK-8 injection, gallbladder contraction and 50% bile emptying times were determined using 99mTc-mebrofenin and cholescintigraphy. After 99mTc-mebrofenin injection, radioactivity increased immediately in the gallbladder to reach a peak at about 15 min, which was maintained for over 60 min (Figure 3A). When rabbits were injected intravenously with CCK-8, 99mTc-mebrofenin was rapidly released from the gallbladder and flowed down the small intestine (Figure 3B). Of the 16 rabbits tested for gallbladder contraction, 6 responded sensitively to CCK-8. On average, it took 11.5 min to evacuate 50% of the gallbladder volume after the first CCK-8 injection. The rabbits responding to CCK-8 were allowed one week to recover and were then included in the in vivo imaging experiments.
Under x-ray visualization and anesthesia, the end of a catheter was located in the mid duodenum (Figure S1). The rabbit was then positioned in the PET-CT bed; anesthesia was maintained with intravenous CCK-8 at a dose of 20 ng/kg every minute during PET-CT scanning. One abdominal CT image was obtained initially and then dynamic PET scanning was started. Three minutes after the initial PET scanning, the 18F-FDG-labeled CsNEJs were inoculated into the mid duodenum (Figure 1). Dynamic and static PET scans were carried out using PET-CT on migrating 18F-FDG-labeled CsNEJs in 6 rabbits, which included 2 controls.
Signals emitted from the 18F-FDG-labeled CsNEJs were detected in the intestine of the 4 experimental rabbits by PET, and thus, we were able to trace CsNEJ migration by in vivo imaging. When the 18F-FDG-labeled CsNEJs were injected through the catheter, signals were detected at end of the catheter in the duodenum and along the small intestine driven by peristalsis along the distal portion of the intestine (Figure S2). Signals of CsNEJs appeared in the liver as early as 7–9 min after inoculating the 18F-FDG-labeled CsNEJs into the duodenum (Figure 4A).
As time elapsed, some photon spots emerged in the liver region and enlarged whereas others faded. These spots appeared to be randomly and evenly distributed in the liver regardless of lobe structure (Figures 4A–D), and gradually increased in number to plateau at about 21 min after inoculation of the radiolabeled CsNEJs (Figures 4 & 5). Spots suggestive of CsNEJs moving through the common bile duct were not observed in PET-CT images. In static PET-CT images taken finally over 10 min, CsNEJs appeared to aggregate in central region of the liver (Figures 4E & F). Of the CsNEJs inoculated into the duodenum, some migrated up to the bile ducts and others down to the lower bowel driven by peristalsis (Figure S2).
In rabbits not injected with CCK-8 (the negative control group), signals of 18F-FDG-labeled CsNEJs were only observed in the small intestine in dynamic and static PET images.
At 4 weeks after the CsNEJs inoculation into the duodenum, adult C. sinensis worms were found to inhabit and to have provoked pathologic changes in the bile ducts. On average 1,077±806 adults were recovered from the biliary tracts of the rabbits (Figure S3).
In vivo the migration route of C. sinensis was indirectly determined by ligating the common bile ducts of hosts. Recently, live Schistosoma mansoni adults in mice were labeled with protease-activated fluorochrome or 18F-FDG and visualized, localized, and quantified using fluorescence molecular tomography or PET [18], [19]. In the present study, we applied the methodologies and investigated PET-CT as a new in vivo imaging method for monitoring the migration of CsNEJs and their localization in the rabbit liver. The rabbits are highly susceptible to and retain the C. sinensis infections long time to evaluate impact of the infection on the hepatobiliary system. The rabbits have the biliary system similar to that of human. Distribution of C. sinensis in the liver of the experimental rabbits was proportional to volume of the liver lobes [20]–[22]. We, therefore, expected the rabbit as a reliable experimental animal model to study bile-chemotactic migration of the CsNEJs, suggesting that findings obtained from the rabbits are applicable to human.
Trematodes import glucose through glucose transporter, and a large number of glucose transporters have registered in the C. sinensis transcriptome database [13]. 18F-FDG is a glucose analog tagged with isotope 18F, and is transported into cytoplasm by glucose transporters in cell membrane. In the cytoplasm, FDG is phosphorylated to FDG-6-phosphate by hexokinase, and FDG-6-phosphate is neither metabolized further nor able to diffuse out of cells. Thus, FDG-6-phosphate is trapped and accumulates in cells as the dephosphorylation of FDG-6-phosphate by glucose-6-phosphatase in cytoplasm is a slow process [15], [23]. We expected that fasted CsNEJs would uptake more FDG than fresh CsNEJs because CsNEJs should have consumed their reserve energy source, primarily glucose, during fasting in glucose-free 1× Locke's solution. However, FDG uptakes in both groups were similar, suggesting that FDG moved quickly into the tegument of CsNEJs through glucose transporter by facilitative diffusion, as was observed for schistosomes [24], [25].
During our studies, we have observed that CsNEJs move toward bile dose-dependently by chemotaxis in in vitro assays (unpublished data). Based on our data and the notion that C. sinensis juveniles migrate up through the common bile duct, it was essential that bile juice is released from the gall bladder to attract CsNEJs into the common bile duct.
Technetium labeled hepatobiliary radiopharmaceuticals has greatly facilitated studies on gallbladder function [26]. Since CCK-stimulated cholescintigraphy was first reported in 1979, gallbladder emptying function has been measured by using standard cholagogic stimulus agents by biliary excretion scintiography [27], [28]. Cholescintigraphy with 99mTc-iminodiacetic acid has been used to diagnose diseases in the biliary system, such as, bile duct obstruction, cholelithiasis, cholecystitis, and biliary fistula [29]–[31].
The gallbladder normally fills with hepatic bile during fasting and empties its contents into the duodenum in response to stimulation by CCK, either released endogenously following a meal or administered exogenously [32]. However, gallbladder emptying response to exogenous CCK varies among patients and experimental animals. In this study, gallbladder contraction and bile juice release was achieved by repeatedly injecting CCK-8. By cholescintigraphy, 99mTc-mebrofenin was found to be released rapidly from gallbladders after CCK-8 administration. Thus, this scheme enabled us to study in vivo bile-chemotactic behavior of CsNEJs in rabbits.
Using CsNEJ radiolabeling and bile excretion from gallbladder, images of CsNEJs migrating to the intrahepatic bile ducts in rabbits were obtained by PET-CT. The radiolabeled CsNEJs were inoculated into the mid duodenum, which is supposed to be an excystation site for C. sinensis metacercariae [3], [6]. We visualized 18F-FDG-labeled CsNEJs migrating to the liver in experimental rabbits using PET-CT. The first signals of CsNEJs arriving at the liver from the duodenum were detected by dynamic PET as early as 7–9 min after inoculating CsNEJs into duodena. At 21 minutes post-inoculation, photon signals emitted from CsNEJs in liver appeared to have stabilized though their intensities undulated, which suggested most CsNEJs responsive to bile immediately migrated up to the intrahepatic bile duct. Imaging was ended with a final static PET-CT image because signals were of greater intensity than on dynamic PET images, suggesting that some CsNEJs were late to arrive and accumulated in the intrahepatic bile ducts [3]. In in vitro assays, CsNEJs showing rapid bile-taxis were promptly re-activated and moved rapidly and continuously toward bile added to assay chambers, and slow responders responded slowly (unpublished data).
The artificial manipulation of CsNEJs employed in this study, that is, in vitro excystation and radiolabeling, and inoculation into the duodenum, may have reduced adaptation to body temperature, chemotactic response to bile, migration to the bile duct, and survival in bile juice. To compensate for this, in the present study, 3–5 times more 18F-FDG-CsNEJs than normally usual experiment was inoculated via catheter into the duodenum. We believe that slow responders arrived late at the intrahepatic bile ducts after PET scans, and increased numbers of C. sinensis adult worms recovered from the bile ducts [3]. When filet of the fresh water fish was minced by teeth and ingested by mammalian animals including human, the C. sinensis metacercariae could be released from the filet in the stomach after 1–2 hour, and then passed down to the duodenum. Considering immediate excystation of the C. sinensis metacercariae in contact with trypsin [6], human infection may take place within 2–3 hours after eating raw filet of the fresh water fish.
We searched for photonic signals from the common bile ducts in dynamic and static PET-CT images of experimental rabbits, but found no signal. The common bile duct is narrow and CsNEJs either passed rapidly or steadily in file, and thus, only small number of juveniles (not enough to create a PET-CT image) was captured in a given frame. Furthermore, anatomically the common bile duct is located in the deep abdomen under the liver, which hinders emitted photons.
Collectively, we report for the first time that CsNEJs were efficiently radiolabeled in vitro with 18F-FDG, and that CsNEJs migrate quickly with bile-chemotaxis to the intrahepatic bile duct as visualized in rabbits by PET-CT.
|
10.1371/journal.pgen.1007026 | Identification of genetic networks that act in the somatic cells of the testis to mediate the developmental program of spermatogenesis | Spermatogenesis is a dynamic developmental process requiring precisely timed transitions between discrete stages. Specifically, the germline undergoes three transitions: from mitotic spermatogonia to spermatocytes, from meiotic spermatocytes to spermatids, and from morphogenetic spermatids to spermatozoa. The somatic cells of the testis provide essential support to the germline throughout spermatogenesis, but their precise role during these developmental transitions has not been comprehensively explored. Here, we describe the identification and characterization of genes that are required in the somatic cells of the Drosophila melanogaster testis for progress through spermatogenesis. Phenotypic analysis of candidate genes pinpointed the stage of germline development disrupted. Bioinformatic analysis revealed that particular gene classes were associated with specific developmental transitions. Requirement for genes associated with endocytosis, cell polarity, and microtubule-based transport corresponded with the development of spermatogonia, spermatocytes, and spermatids, respectively. Overall, we identify mechanisms that act specifically in the somatic cells of the testis to regulate spermatogenesis.
| Sexual reproduction in animals requires the production of male and female gametes, spermatozoa and ova, respectively. Gametes are derived from specialized cells known as the germline through a process called gametogenesis. Gametogenesis typically takes place in a gonad and requires the germ cells to be surrounded by specialized somatic cells that support germline development. While many prior studies have identified germline specific genes required for gametogenesis few have systematically identified genes required in the somatic cells for gametogenesis. To this end we performed an RNAi screen where we disrupted the function of genes specifically in the somatic cyst cells of the Drosophila melanogaster testis. Using fertility assays we identified 281 genes that are required in somatic cyst cells for fertility. To better understand the role of these genes in regulating spermatogenesis we classified the resulting phenotypes by the stage of germline development disrupted. This revealed distinct sets of genes required to support specific stages of spermatogenesis. Our study characterizes the somatic specific defects resulting from disruption of candidate genes and provides insight into their function in the testes. Overall, our findings reveal the mechanisms controlling Drosophila melanogaster spermatogenesis and provide a resource for studying soma-germline interactions more broadly.
| Spermatogenesis is a highly choreographed developmental process that involves dramatic changes in cell morphology and the integration of multiple regulatory cues. Orchestrating such a process would be challenging even if spermatogenesis involved only one cell type, but it in fact includes two very different types of cells, somatic cells (soma) and germ cells (germline). Close association between the soma and the germline is a conserved feature of spermatogenesis in animal testes [1]. Importantly, interplay between these two tissues is critical for the normal progression of spermatogenesis and disruption of one cell type severely affects the other [2,3].
Spermatogenesis can be divided into three distinct developmental stages: the spermatogonial-stage, characterized by mitotic germ cells; the spermatocyte-stage, characterized by meiotic germ cells; and the spermatid-stage, characterized by the morphogenetic changes that form spermatozoa [1]. The spermatogonial population is sustained by germline stem cells (GSCs). When GSCs divide, the resulting daughter cells can produce either a GSC or a spermatogonial cell that proceeds through spermatogenesis. A specialized stem cell niche typically plays an instructive role in maintaining the balance between GSC self-renewal and the differentiation of spermatogonia [4,5]. During the spermatogonial-stage the germline initiates a program of differentiation and undergoes transit-amplifying mitotic divisions. These divisions are incomplete and the resulting germ cells retain cytoplasmic bridges connecting them as a ‘germ cell cyst’ [6,7]. Following the final transit-amplifying division the germline enters the spermatocyte-stage. Entry into this stage constitutes a “point-of-no-return” during spermatogenesis, as the genome is irreversibly altered and reduced from diploid to haploid by meiosis [8,9]. Progression into the spermatid-stage is marked by the completion of meiotic divisions. Spermatid development is characterized by dramatic morphological changes as the genome is compacted into a small high-density nucleus, and the cellular machinery required for fertilization and motility are formed. Finally, excess cytoplasm is removed and the intercellular bridges connecting the spermatids are severed, releasing them as individual spermatozoa [10,11].
The Drosophila melanogaster testis provides a powerful model to study the conserved process of spermatogenesis as it allows convenient imaging of the transitions between different stages of germline development [12–14]. The early stages of spermatogenesis occur at the closed, anterior end of the testis known as the ‘apical tip’. The apical tip of each testis holds the stem cell niche known as the hub which is made up of tightly clustered somatic cells attached to a dense accumulation of extra-cellular matrix [15]. The hub acts as a signalling centre that secretes multiple cell signalling cues to regulate the maintenance and behaviour of both germline stem cells (GSCs) and somatic cyst stem cells (CySCs). These signals include ligands for the Hedgehog (Hh) [16], Janus Kinase-Signal Transducers and Activators of Transcription (JAK-STAT) [17], and Bone Morphogenetic Protein (BMP) [18] pathways. Both GSCs and CySCs orient their centrosomes perpendicular to the hub during mitosis, resulting in predominantly asymmetrical divisions where one cell retains contact with the hub and the other is displaced and subsequently differentiates [19,20]. Displaced GSCs form gonialblasts that will go on to become spermatozoa, while displaced CySCs form cyst cells that support the germ cells throughout spermatogenesis.
A key event during the spermatogonial-stage is a process known as encapsulation whereby two cyst cells wrap each gonialblast and all three cells together form a ‘spermatocyst’. All further germ cell development takes place within the spermatocyst in the lumen formed between the two encapsulating cyst cells. Each encapsulated gonialblast proceeds through four rounds of transit-amplifying divisions resulting in two-cell, four-cell, eight-cell, and finally sixteen-cell germline cysts [21]. After reaching the sixteen-cell point, germ cells enter the spermatocyte-stage during which they undergo massive growth and expand approximately twenty-five times in volume as they move towards the basal end of the testis. The spermatocytes proceed through meiosis mid way down the testis and form sixty-four round spermatids [21]. During the spermatid-stage the germ cells polarize and elongate, their nuclei continuing to move basally while their flagellar axonemes grow to over 1800μm in length [21]. The two cyst cells encapsulating the germline also differentiate during spermatogenesis. The cyst cells undergo dramatic changes in gene expression, grow in size, and become two distinct types of cyst cells—head and tail cyst cells [3]. When the head cyst cell surrounding the spermatid nuclei reaches the end of the testis it attaches to a layer of somatic cells called the terminal epithelium. The individualized spermatozoa within the spermatocyst are then coiled at the base of the testis and threaded tail first through a narrow duct into the seminal vesicle [22]. Successful completion of spermatogenesis requires cooperation between cyst cells and germ cells. Eliminating cyst cells or disrupting their ability to encapsulate both result in the failure of germ cells to differentiate past the spermatogonial-stage [23–25].
Here we explore how the somatic cells of the testis contribute to germline development during spermatogenesis. We systematically identify genes that are required in the Drosophila melanogaster cyst cells for fertility. Subsequent analysis of the phenotypes resulting from somatic knockdown of candidate genes identified the stage of spermatogenesis for which their function was required. By employing bioinformatic approaches we find functional clusters of genes that mediate developmental events associated with progress through spermatogenesis. Detailed analysis of the phenotypes resulting from the disruption of several representative gene classes illustrates the diverse mechanisms through which somatic cells support the germline during spermatogenesis.
To date there has not been a large-scale systematic attempt to identify genes specifically required in the somatic cells of the testis for spermatogenesis. There are two key challenges in such an undertaking: First, the need to study the function of genes exclusively in the somatic cells, and Second, the compounding fact that many genes that mediate spermatogenesis are likely involved in other developmental processes. Consequently, a standard forward genetic screen for mutations causing male sterility would fail to identify many key players in spermatogenesis. With these challenges in mind we designed and optimized a genetic screen using somatic cyst cell specific RNAi-mediated knockdown to identify genes that are required for germline development (Fig 1A and 1B). Spermatogenesis was initially assessed using male fertility assays that were carried out in triplicate. RNAi knockdowns that produced no progeny in the majority of crosses were then retested to confirm the initial result; RNAi knockdowns that consistently showed infertility were analysed phenotypically via immunofluorescence and confocal imaging. The collection of RNAi lines that were screened includes a set of cytoskeletal genes and cytoskeletal regulators, which was initially used as a pilot [26]. Although we enriched the collection for genes associated with stem cell regulation [27–29] the full collection of RNAi lines screened were obtained from a broad array of sources and targeted a wide variety of different gene classes (see S2 Table).
Overall we screened 2597 RNAi lines representing 2004 genes, accounting for approximately 14% of the protein-coding genes in the Drosophila melanogaster genome [30] (Fig 1C). In total we identified 281 genes (14% of tested genes) as being required in the somatic cyst cells for fertility (see S1 Fig and S1 Table). Each of the 281 candidate genes was knocked down in the cyst cells and the phenotype characterized using markers for the somatic cells (Fig 2; tj>mGFP) and the germ cells (Fig 2; Vasa, bam-GFP, Boule, DonJuan::GFP). Using the information derived from this analysis each gene knockdown was classified into one of four distinct phenotypic categories (Fig 2B–2E). Although, there was some variability in penetrance of the phenotypes, which is typical for RNAi-mediated knockdown, the classification was based on the most penetrant manifestation of the phenotype. The first category, ‘cyst cells absent’, was characterized by the absence of both CySCs and cyst cells resulting in small rudimentary testes often containing undifferentiated germ cells (Fig 2B). In the second category, ‘spermatogonial defects’, cyst cells were present but the germ cells remained unable to progress past the mitotic stage of development and often over-proliferated as tumour-like growths (Fig 2C). The third phenotypic category, ‘spermatocyte defects’, contained cyst cells as well as germ cells that had begun to differentiate, but were arrested before the completion of the meiotic stage of development (Fig 2D). In the final phenotypic category, ‘spermatid defects’, cyst cells were maintained, but post-meiotic germ cell morphogenesis was disrupted resulting in a failure to produce mature spermatozoa (Fig 2E). Overall, 85% of the candidates were analysed and assigned a phenotypic category. Cyst cells absent represented the largest phenotypic category, accounting for nearly 53% of the classified candidate genes. The proportional breakdown for the other categories is as follows: 18% spermatogonial defects, 4% spermatocyte defects, and 26% spermatid defects (Fig 1E).
To gain more insights from the screen we performed a detailed bioinformatic analysis of the results. First, we compared our screen with similar studies that investigated spermatogenesis in Drosophila melanogaster (fly) and Mus musculus (mouse) (Fig 1D). For example, two small-scale follow-up screens, subsequent to larger screens, investigated the role of 221 [31] and 113 [32] genes in the somatic cyst cells of the fly testis. Of the shared genes between our screens (genes they identified and we tested), there was approximately a 68% overlap with the candidate genes identified in our screen (67/98 genes and 26/39 genes, respectively) (Fig 1D). Furthermore, we compared our results to a screen recently carried out in the somatic follicle cells of the fly ovary [33]; of the shared genes between our screens there was a 40% overlap in candidate genes (100/250 genes) (Fig 1D). Intriguingly, 271 of our candidate genes have a homologous gene in mice [34] and 110 of these homologs are also expressed in a stage-specific pattern in Sertoli cells during mammalian spermatogenesis [35] (Fig 1D). Together these comparisons confirm that our list of candidates is enriched for genes that regulate somatic cells during spermatogenesis.
To characterize possible molecular interactions and identify genetic networks in our list of candidate genes we built a protein-protein interaction map using the STRING-Database [36] (Fig 1F). Furthermore, we integrated Gene Ontology (GO) terms [37] into our map and found interacting clusters of genes mediating specific cellular functions. This approach identified clusters involved in, adhesion and cell polarity, endosomal transport, the tubulin and actin cytoskeleton systems, mRNA processing, protein ubiquitination and stability, and mitochondrial function (Fig 1F). To determine if any of these gene clusters were associated with specific phenotypes the DAVID algorithm [38] was utilized to identify enriched groups of related GO terms (Fig 1G). This analysis revealed that GO terms associated with septate junction assembly and regulation of the cell cycle were enriched in candidate genes yielding spermatogonial defects. Additionally, GO terms associated with cell polarity and oocyte axis specification were enriched in candidate genes yielding spermatocyte defects, while GO terms associated with the cytoskeleton and microtubule-based movement were enriched in candidate genes yielding spermatid defects (Fig 1G). Overall this demonstrates that the candidate genes identified in our screen contained several interacting gene networks and that these were enriched in specific phenotypic categories.
To elucidate why particular clusters of interacting genes were enriched in specific phenotypic categories representative genes were chosen for further study. The spermatogonial defect category was of special interest since this category likely contained genes that regulate cyst stem cells (CySCs) and germline stem cells (GSCs). Analysis of candidate genes whose knockdown phenotypes manifested during the spermatogonial-stage revealed an intriguing enrichment of genes associated with endocytosis including, Rab5, Chc, Vps16A, Sec15, and AP-1-2β. Since Rab5 has long been established as a master regulator of early endosomes [39] it was chosen for further study as a representative gene within this cluster (Fig 3). Somatic cyst cell specific knockdown of Rab5 resulted in a strong and penetrant phenotype of spermatogenesis arrest at the spermatogonial-stage (Fig 3C). This phenotype was distinct from those observed when other Rab GTPases, such as Rab11 and Rab7 were knocked down, which resulted in earlier and later defects, respectively (Fig 3B and 3C). Rab5 knockdown testes accumulated both somatic cell and germ cell based tumour-like growths over the course of 1 to 2 weeks (Fig 3D and 3E). Based on the observed phenotype we expected the germ cell growths to be arrested at the spermatogonial-stage. To confirm this we examined the morphology of fusomes that interconnect transit-amplifying spermatogonia. This analysis revealed that the germ cell growths associated with Rab5 deficient cyst cells had thin, branched fusomes similar to those found in spermatogonia (S2 Fig). To gain insight into the mechanism responsible for this terminal phenotype, knockdown was induced in adult testes using the conditional temperature-sensitive Gal80 system (Fig 3F and 3G). This approach allowed for the reconstruction of the cellular events preceding the production of the tumour-like growths. Specifically, we observed that when Rab5 was knocked down Zinc-finger homeodomain-1 (Zfh1), a transcription factor essential for CySC identity [40], was expressed in many cyst cells outside the stem cell niche (Fig 3G). By comparison, in control testes Zfh1 is down regulated outside of stem cell niche permitting CySC differentiation (Fig 3F). Sustained Zfh1 expression in cyst cells is linked to accumulation of ectopic CySCs and GSCs [40]. This raises the possibility that the signalling environment around the stem cell niche, which typically mediates the down regulation of Zfh1 to allow CySC differentiation, is disrupted upon Rab5 knockdown.
We envisaged two possible mechanisms by which Rab5 knockdown could disrupt the signalling environment around the niche such that cyst cell differentiation is blocked. The niche itself could be altered, thus blocking differentiation, or Rab5 knockdown could inhibit the ability of cyst cells to processes niche-derived signals. Two pieces of evidence argue the latter. First, the effect of Rab5 was cell autonomous as shown by clonal RNAi-mediated knockdown of Rab5 (Fig 4A and 4B). In spermatocysts containing somatic clones expressing rab5.RNAi the cyst cells encapsulated the germ cells but both tissues failed to fully differentiate (Fig 4A). This phenotype was less penetrant then that observed when Rab5 was knocked down in all cyst cells, which may reflect differences in the level of knockdown due to the use of an alternative Gal4 (c587-Gal4 in clones versus tj-Gal4 in the constitutive knockdown). Alternatively this could reflect differences due to the late induction of the knockdown specifically in adult clones. Nonetheless, the stem cell marker Zfh1 was often expressed in these somatic cell clones while the differentiation marker Eyes absent (Eya) [41] was not expressed (Fig 4B). Second, although the stem cell niche underwent abnormal growth in Rab5 knockdown flies, in testes from 1-day post eclosion (DPE) males the hub was only marginally larger than controls (Fig 4C and 4F). Taken together these results suggest that Rab5 modulates the ability of CySCs to process stem cell renewal signals and enables cyst cells to differentiate as they leave the niche.
Multiple niche-derived signals have been implicated in maintenance of cyst stem cell (CySC) fate and sustained Zfh1 expression (Fig 3A). In particular, it is known that JAK-STAT signalling maintains Zfh1 expression in CySCs [40]. Based on this, one prediction to explain the Rab5 knockdown phenotype is that JAK-STAT signalling levels are higher in the CySCs. Surprisingly this is not the case as measurement of STAT immunostaining revealed levels that were slightly lower than controls (Fig 4G–4I). However ectopic JAK-STAT signalling was detected in the tumour-like growths of cyst cells found in aged males, suggesting abnormal activation of this pathway could be driving Zfh1 expression outside the niche (S3E Fig). The Hedgehog (Hh) pathway also plays a role in maintaining Zfh1 expression and promoting CySC identity [42,43]. To measure Hh signalling we analysed the localization of the Hh receptor Patched as it has been shown to be a useful readout for Hh signalling in CySCs [44]. Loss of Rab5 induces a substantial increase in Hh signalling as judged by the size of Patched staining puncta (Fig 4J–4L). Furthermore Hh signalling was also detected in the tumour-like growths of cyst cells that form outside of the niche in aged males (S3D Fig). Overall this data suggests that Rab5 modulates signalling pathways, such as JAK-STAT and Hh, to enable cyst cell differentiation.
Both JAK-STAT signalling and Hh signalling have been implicated in controlling the expression of BMP ligands by CySCs [44,45]. Ectopic expression of the BMP ligand Decapentaplegic (Dpp) also leads to the over-proliferation of spermatogonia [46]. We therefore examined whether BMP signalling was elevated in the germ cells associated with Rab5 deficient cyst cells by immunostaining for the phosphorylated form of Mothers against decapentaplegic (pMad). While pMad is only weakly detected in controls it is robustly detected in spermatogonia when Rab5 is knocked down in the cyst cells (S3A–S3C Fig). Together with our previous results this demonstrates that Rab5 mediates the differentiation of cyst cells and regulates their expression of BMP ligands to facilitate spermatogonial development.
Of all the phenotypic categories defined in the screen, the spermatocyte defects category was the smallest, consisting exclusively of regulators of cell polarity and polarized protein trafficking. The phenotypes resulting from cyst cell specific knockdown of candidate genes in this functional cluster were analysed in detail to understand their role during the spermatocyte-stage. Specifically, our analysis focused on the role of a group of interacting proteins composed of Bazooka (Baz), Par-6, and atypical Protein Kinase C (aPKC). Together these proteins function as the ‘Par polarity module’ and has long been established as a central regulator of apical polarity in epithelial cells [47] (Fig 5B). Moreover, Par module genes are expressed in cyst cells during spermatogenesis [48,49]. When the Par module genes (Baz, Par-6, aPKC) were knocked down in cyst cells the spermatogonial-stage appeared normal and the germ cells were encapsulated similar to controls (Fig 5E and 5F, 5K–5L). However, although the germ cells often entered the meiotic spermatocyte-stage, as judged by the expression of Boule [50], spermatogenesis arrested shortly afterwards (Fig 5E and 5F). This phenotype does not represent a general consequence of disrupting cell polarity. For example, in agreement with previous results [51], we found that knockdown of the ‘Scribble polarity module’ genes (Dlg1, Scrib, L(2)gl) that regulate baso-lateral polarity in epithelial cells [52] gave rise to a distinct phenotype that manifested during the mitotic spermatogonial-stage (Fig 5H–5J). In contrast, the main phenotype observed upon cyst cell specific knockdown of Par polarity module genes (Baz, Par-6, aPKC) was a failure of the germ cells to develop or survive past the meiotic spermatocyte-stage (Fig 5E and 5F, 5K–5L).
To elucidate the specific cellular role of Par polarity module genes during the spermatocyte-stage, the subcellular distribution of Bazooka (Baz) was examined (Fig 6A–6C). Baz expression becomes prominent by the early spermatocyte-stage and is localized to the junctional belt between encapsulating cyst cells [53] (Fig 6A and 6B and S4 Fig). Furthermore, Baz was observed to occupy a distinct junctional complex compared to the Scribble module protein Discs large-1 (Dlg1) (Fig 6C). This shows that during the spermatocyte-stage, cyst cells exhibit a form of cell polarity that utilizes two spatially separated regulatory modules. A key role of the Par polarity module in epithelial cells is to coordinate the establishment and maintenance of apico-lateral adherens junctions [47]. In comparison, the Scribble polarity module is required to maintain baso-lateral septate junctions [52]. These roles were recapitulated in the cyst cells, even though these cells are not epithelia (Fig 6D–6G). Specifically, the Par module protein Bazooka (Baz::GFP) predominantly colocalizes with the adherens junction component DE-Cadherin (DEcad) (Fig 6D and 6E), while the Scribble module protein Discs large-1 (Dlg1::GFP) predominantly colocalizes with the septate junction component Coracle (Cora) (Fig 6F and 6G). Importantly, knockdown of Baz resulted in the disruption of adherens junctions (DEcad), but not septate junctions (Cora) in spermatocyte-stage cyst cells (Fig 6H and 6I). These results suggest that mechanisms regulating cell polarity are essential for spermatocyte development and survival within the adluminal compartment of the spermatocyst.
The final phenotypic category defined by the screen, the spermatid defects category, contained a striking enrichment of candidate genes associated with microtubule-based transport. Amongst the genes in this category were components of the dynein (Dhc64C) and dynactin (Glued) complex [54]. This category also contained the β3 (β-tubulin at 60D) and α2 (α-tubulin at 85E) microtubule subunits, both of which are expressed exclusively in the somatic cells of the testis in adult males [55,56] (Fig 7B and 7D). Knockdown of these genes in cyst cells resulted in fewer fully elongated spermatids, as judged by the expression of Don Juan [57], even after normal progression through the spermatogonial and spermatocyte-stages (Fig 7C, 7E and 7F). As a result of these defects, individualized spermatozoa were not extruded into the seminal vesicle (Fig 7H and 7I). In the germline, it is well established that microtubules and the dynein-dynactin complex are central to spermatid elongation [58,59]. To gain insight into the role of these proteins in cyst cells, the expression of β3tubulin was characterized during the spermatid-stage (Fig 8A). This revealed a striking rearrangement of the microtubules from the short arrays typical of ‘round spermatid-stage’ cyst cells to the extended, linear arrangement typical of ‘elongated spermatid-stage’ cyst cells (Fig 8A). While cyst cell specific knockdown of the dynactin component Glued did not disrupt the initial encapsulation of the germline, it did disrupt the dense arrangement of β3tubulin in spermatid-stage cyst cells (Fig 8B and 8C). These results were consistent with a role for microtubules and the dynein-dynactin complex in cyst cell morphogenesis during spermatid development. To understand the underlying phenotype resulting from knockdown of Glued in the cyst cells, spermatocyst integrity was analysed using a permeability assay. This assay measures the ability of fluorescently-conjugated dextran to access the surface of germ cells at different stages of development [53]. In control testes, the dye was unable to access the surface of spermatid-stage germ cells in the apical tip of the testis, consistent with an intact spermatocyst (Fig 8D). By comparison, when Glued was knocked down in cyst cells, the dye was able to access the surface of fully elongated spermatids that had reached the apical tip of the testis (Fig 8E). Overall, this data suggests that microtubules and the dynein-dynactin complex play a key role in maintaining the growth and integrity of cyst cells during spermatid development.
We performed an RNAi-mediated knockdown screen and identified candidate genes that are required in the somatic cyst cells of the Drosophila melanogaster testis for spermatogenesis. Analysis of the candidates revealed functional gene networks that were required for specific stages of spermatogenesis. Phenotypic analysis of these genes provides insight into the mechanisms that are essential for somatic cells to support each stage of germline development. In particular, we show that components of the endocytic machinery are critical for spermatogonial development. Endocytosis modulates the activity of niche-derived signals and is thus essential for regulating cyst stem cell differentiation. Furthermore, proteins that maintain cell polarity and regulate cell-cell junctions in epithelia serve a similar role in the cyst cells during spermatocyte development. Failure to maintain cell polarity leads to the disruption of adherens junctions and results in the death of spermatocytes within the adluminal domain of the spermatocyst. Finally, we demonstrate that microtubule-based transport is required in cyst cells for the completion of spermatid development. A specific set of microtubule subunits and the dynein-dynactin complex mediate the large morphological changes in cyst cells during spermatid elongation. Together these results provide mechanistic insight into the cell biological events that underlie spermatogenesis and identify functional clusters of genes that act in the somatic cells to mediate germline development.
Spermatogenesis involves a sequence of coordinated morphogenetic behaviours. Both germ cells and somatic cells undergo enormous changes in cell shape and size, but the molecular mechanisms underlying these events, particular those of the somatic cells, are not well understood. Our screen now provides entry points in attempting to understand these molecular mechanisms. It will be possible, starting with the candidate genes we identified, to mine protein interaction databases and identify other candidate genes that act during each stage of spermatogenesis. Importantly, although we have assigned each of our candidates a phenotypic category based on the stage at which spermatogenesis arrested, it is likely that many of these genes act during multiple stages of spermatogenesis. While our study assigns a function to these genes at a particular stage, this might only reflect the time at which RNAi mediated gene knockdown becomes effective. Also, it is important to note that the phenotypes we observe might reflect phenotypes that arise during earlier stages or in embryonic development, but only manifest later on. For these reasons, follow up experiments on these candidate genes will need to use, as we have done (see Figs 3G and 4B), clonal and inducible knockdown approaches [60] to confirm the stage at which they are required. Nonetheless, for the genes we identified in our screen that have known functions in spermatogenesis there was a good correspondence between their previously studied roles and the stage at which our analysis suggested they were required.
Our results suggesting that Rab5 functions in the somatic cyst cells to fine-tune niche-derived stem cell maintenance signals are supported by recent analysis of the role Rab5 plays in modulating expression of the BMP ligand Decapentaplegic (Dpp) [61]. While we show that Rab5 controls the levels of JAK-STAT and Hh signalling in cyst cells, Rab5 has similarly been shown to control the levels of JAK-STAT and Jun kinase (JNK) signalling as well as Dpp expression in cyst cells [61]. Rab5 and the endosomal machinery are likely to be key regulatory nodes for controlling how cyst cells receive, process, and transmit signals with the stem cell niche and the germline. For both Hh and JNK signalling it appears the predominant role of Rab5-mediated endocytosis is to attenuate signalling activity in the cyst cells. While the role Rab5 plays in regulating JAK-STAT signalling appears to be more complex. Perturbation of the Rab5 endocytic machinery modulates cyst cells sensitivity or responsiveness to niche-derived signals and maintains them in an undifferentiated state for a longer period of time. As a consequence of this, the cyst cells produce tumour-like growths exhibiting characteristics of stem cell identity, such as Zfh1 and Traffic Jam (Tj) expression [40,62]. Overall, these results demonstrate how cell-cell signalling regulates the transition from stem cell to differentiating cell, which represents a hallmark of the spermatogonial-stage. Indeed, another gene cluster enriched in this phenotypic category, septate junction components, has previously been shown to modulate soma-germline signalling in the testis [53,63]. We expect that analysis of other gene clusters enriched in this phenotypic category is likely to yield more insight into the mechanisms that mediate cyst cell differentiation.
Mechanisms that regulate cell polarity and adhesion are known to play a role in spermatogenesis; in particular they have important functions in the germline. During early spermatogenesis the Par module protein Baz associates with centrosomes in germline stem cells and this interaction is key for ensuring proper asymmetric cell division [64]. Another Par module protein, aPKC, has been shown to be essential for germline polarization during the spermatid-stage of development [49]. Although germ cells are not epithelial, they do exhibit a form of polarization with spermatid nuclei localized to one side of the spermatocyst surrounded by the head cyst cell, while the flagellar axonemes grow towards the other side surrounded by the tail cyst cell [11]. We find that in somatic cyst cells, polarization mechanisms are required even earlier during the spermatocyte-stage. Our work builds on these previous studies and adds three insights. First, we provide evidence that the polarity cyst cells exhibit is reminiscent of epithelial polarization, with multiple, spatially segregated polarity modules. Second, we find that in spermatocyte-stage cyst cells the Par polarity module is specifically required for the maintenance of adherens junctions. Third, we find that different polarity modules have different roles, with the Scribble module regulating spermatogonial development and the Par module regulating spermatocyte development. The Par module may mediate the substantial growth in volume of the spermatocyst lumen as this would likely require cell-cell junctional remodeling [65,66]. Together these findings suggest that the gene clusters enriched in this phenotypic category mediate the formation or maintenance of the cyst cell luminal environment.
Microtubules and the dynein-dynactin complex are known to play important roles in the germline during spermatogenesis. Mutations in the germ cell specific β2tubulin subunit [58] or components of the dynein–dynactin complex [59,67–69] affect the formation of the flagellar axonemes. It is also known that mutations in the somatic cell specific β3tubulin subunit results in male sterility [70]. We show that in cyst cells both β3tubulin-based microtubules and the dynein-dynactin complex are required to maintain the integrity of the spermatocyst during spermatid development. One possibility is that as spermatid elongation is independent of the soma [71] the cyst cells must grow and change shape simply to accommodate the dramatic increase in the length of the germ cells. Specific microtubule subtypes and the dynein-dynactin complex may act as the morphogenetic machinery that drives the elongation of cyst cells and in their absence the structure of the spermatocyst is compromised. Our findings are consistent with the interpretation that many of the gene clusters enriched in this phenotypic category are important regulators of cyst cell morphogenesis.
Our work adds to a growing list of studies that explore spermatogenesis using the powerful genetic and imaging tools available in Drosophila melanogaster. Two recent studies used similar RNAi-based approach to ours, but focused predominantly on genes required in the germline [31,32]. By contrast our study is the largest to focus specifically on how the somatic cells of the testis regulate spermatogenesis. Our work thus provides a comprehensive overview of the complex cooperation between the soma and the germline required for the remarkable process of spermatogenesis.
Female flies were collected from tj-Gal4,UAS-mGFP;UAS-Dcr2 or tj-Gal4;UAS-mRFP,UAS-Dcr2 stocks and bred to a library of males from UAS-RNAi stocks. Individual male progeny (tj>RNAi) were transferred to vials with three to five virgin w1118 females for fertility assays. Virgin females were collected from the Virginator stock [♀w1118/ w1118 and ♂w1118/Dp(2;Y)G,hs-hid], raised at 20°C and shifted to 29°C for 3 days during pupation to kill males. Individual male fertility assays were done in triplicate and listed as sterile if no larva/pupae were present 14 days post-mating. If the majority of fertility assays yielded sterile males, more males were tested and additional RNAi lines targeting the gene were screened.
Spermatogonia were identified by expression of Vasa, dense DAPI staining nuclei, and their small size; differentiating spermatogonia were identified by expression of both Vasa and bam-GFP. Spermatocytes were identified by expression of Vasa, dispersed DAPI staining, and their large size; meiotic spermatocytes identified by expression of both Vasa and Boule. Early spermatids identified by expression of Boule, polarization of their nuclei, and their elongated shape; individualizing spermatids identified by expression of DonJuan::GFP. Spermatozoa identified by expression of DonJuan::GFP and their accumulation in the seminal vesicle as individual cells. Phenotypes were assigned based on the stage of germline development that was disrupted in the majority of samples. Germline disruption was defined as over-proliferation, disintegration, death, or other disorders the germ cells including the failure of spermatozoa to reach the seminal vesicle.
Protein-protein interaction network (Fig 1F, 1G and S1 Fig) built using STRING-DB 10.0 [36]. Network map created using Cytoscape 3.0 [72] and the force-directed layout mediated by the ‘AllegroLayout’ plugin. Sterile genes were coloured by their functional class based on a summary of the associated GO terms and the relevant literature by the authors. GO term enrichment (Fig 1G) was assessed using DAVID 6.8 [38] by comparing sterile genes with either spermatogonial, spermatocyte, or spermatid defects to all the sterile genes identified. DAVID analysis included the ‘GO DIRECT’ terms for Biological Process (BP), Cell Compartment (CC), and Molecular Function (MF), as well as ‘UP KEYWORDS’ and ‘UP SEQ FEATURE’ (Fig 1G and S2 Table).
tj-Gal4 (P-element insertion {GawB}NP1624) obtained from the Bloomington Drosophila Stock Center [BDRC]). Additional lines from the BDSC included the Virginator (w1118/Dp(2;Y)G,hs-hid), UAS-Dcr2 (UAS-Dicer2), UAS-mGFP (UAS-mCD8::GFP), UAS-mRFP (UAS-mCD8::Tomato), UAS-RFP.nls (UAS-RedStinger), Baz::GFP (bazCC01941), Dlg1::GFP (dlg1YC0005), DonJuan::GFP (dj-GFP.S), and hh-LacZ (hhP30). Vasa::GFP (vas.EGFP.HA) provided by the Kyoto Drosophila Genetic Resource Center (DGRC). upd-LacZ (upd1-PD) courtesy of David Bilder, University of California Berkeley, USA. bam-GFP (bam-GFP-799/+133) courtesy of Christian Bökel, Center for Regenerative Therapies Dresden, Germany. UAS-RNAi lines obtained from the Transgenic RNAi Project (TRiP), the Vienna Drosophila RNAi Centre (VDRC), and the National Institute of Genetics (NIG). RNAi lines used in Figs 2–8 include UAS-CG10483.RNAi (TRiP.HMS01023), UAS-ance.RNAi (TRiP.HMS03009), UAS-AP-2μ.RNAi (TRiP.JF02875), UAS-syx7.RNAi (TRiP.JF02436), UAS-msn.RNAi (TRiP.HMJ02084), UAS-pyr.RNAi (VDRC.36523), UAS-rab5.RNAi (VDRC.103945 except Frt40A;VDRC.34096 in Fig 4A and 4B), UAS-rab11.RNAi (TRiP.JF02812), UAS-rab7.RNAi (VDRC.40338), UAS-baz.RNAi (VDRC.2914), UAS-par-6.RNAi (VDRC.108560), UAS-aPKC.RNAi (VDRC.2907), UAS-dlg1.RNAi (TRiP.JF01365), UAS-scrib.RNAi (TRiP.JF03229), UAS-l(2)gl.RNAi (TRiP.HMS01522), UAS-β3tub.RNAi (VDRC.34607), UAS-glued.RNAi (TRiP.JF02803), UAS-Dhc64C.RNAi (TRiP.JF03177). See S2 Table for all RNAi stocks.
Samples fixed using 4% PFA in PBS for 15 minutes. Antibodies incubated in PBS supplemented with 0.2% BSA and 0.3% Triton-X at 4° with the exception of rabbit-anti-STAT incubated at 20°. Primary antibodies included rat-anti-Filamin.C’terminus (1:1000, Lynn Cooley—Yale University, USA), rabbit-anti-Boule (1:1000, Steven Wasserman—University of California San Diego, USA), guinea pig-anti-Tj (1:2500, Dorothea Godt—University of Toronto, Canada), rabbit-anti-Zfh1 (1:1000, Ruth Lehmann—New York University, USA), guinea pig-anti-Zfh1 (1:1000, James Skeath—Washington University in St Louis, USA), rabbit-anti-STAT (1:1000, Erika Bach—New York University, USA), rabbit-anti-pMad (1:1000, Ed Laufer—Columbia University, USA), rabbit-anti-Baz (1:1000, Tony Harris—University of Toronto), rabbit-anti-β3tub (1:4000, Renate Renkawitz-Pohl—Philipps-Universität Marburg, Germany), chicken-anti-GFP (1:1000, ab13970—Abcam), goat-anti-Vasa (1:200, dC-13—Santa Cruz), rat-anti-dsRed (1:1000, 5f8—Chromotek), chicken-anti-LacZ (1:1000, ab13970—Abcam), mouse-anti-αSpectrin (1:100, 3A9—Developmental Studies Hybridoma Bank [DSHB]), mouse-anti-Eya (1:50, eya10H6—DSHB), mouse-anti-FasIII (1:1000, 7G10—DSHB), mouse-anti-Patched (1:100, Ptc Apa 1—DSHB), mouse-anti-Dlg1 (1:2, 4F3 anti-discs large—DSHB), mouse-anti-Cora (1:500, C566.9 and C615.16—DSHB), rat-anti-DEcad (1:20, DCAD2—DSHB). Secondary antibodies included CF405S, Alexafluor-488, Cy3, and Cy5 conjugates.
All flies were raised at 25°C unless otherwise stated. RNAi induction in adult males (Fig 3F and 3G) done using α1tub-Gal80ts;tj-Gal4,α1tub-Gal80ts/UAS-rab5.RNAi flies raised at 18°C and split into two cohorts 1–3 Days Post Eclosion (DPE) and raised at 18°C or 29°C for 14 days. Cyst cell clones made using c587-Gal4,UAS-RFP.nls,UAS-mGFP,hs-Flp;α1tub-Gal80,Frt40A/Frt40A;UAS-rab5.RNAi flies (Fig 4A and 4B) and c587-Gal4,UAS-mGFP,hs-Flp;α1tub-Gal80,Frt40A/Frt40A flies (S4 Fig) raised at 20°C. Clones induced 1–5 DPE using three 45-minute heat-shocks at 37°C, each separated by 90 minutes and then aged 14 days at 20°C. Hub size (Fig 4C–4F) measured as the maximum diameter of upd-LacZ positive cell nuclei. JAK-STAT signalling (Fig 4G–4I) measured as the fluorescent intensity ratio of each Zfh1-positive CySC nuclei within 15μm of the hub, to the average of 2–4 GSCs contacting the hub in a single image for each testis. Hedgehog signalling (Fig 4J–4L) measured as the maximum diameter of Patched staining puncta within 20μm of the hub in a single image for each testis. Cyst cell junctions (Fig 6B and 6C and S4B Fig) imaged by dissecting single spermatocysts from L3 larval testis [73] into Schneider's Drosophila Medium and allowing them to adhere to poly-lysine coated slides for 30-minutes prior to fixation and antibody staining. Polarity protein co-localization (Fig 6E and 6G) performed on full depth, z-projected images encompassing 250μm of the apical tip of the testis. ImageJ (NIH) and the ‘Coloc 2’ plugin were utilized for Pearson co-localization analysis. Terminal epithelium and seminal vesicle images (Fig 7H and 7I) taken from males aged 2 days without females. Permeability assay (Fig 8D and 8E) used 10kDa dextran conjugated to AlexaFluor-647 (Life Technologies) at a final concentration of 0.2μg/μl [53]. Statistical tests done with Prism (Graphpad), all student t-tests were two-tailed and applied Welch’s correction.
|
10.1371/journal.pbio.0050117 | Unmasking Activation of the Zygotic Genome Using Chromosomal Deletions in the Drosophila Embryo | During the maternal-to-zygotic transition, a developing embryo integrates post-transcriptional regulation of maternal mRNAs with transcriptional activation of its own genome. By combining chromosomal ablation in Drosophila with microarray analysis, we characterized the basis of this integration. We show that the expression profile for at least one third of zygotically active genes is coupled to the concomitant degradation of the corresponding maternal mRNAs. The embryo uses transcription and degradation to generate localized patterns of expression, and zygotic transcription to degrade distinct classes of maternal transcripts. Although degradation does not appear to involve a simple regulatory code, the activation of the zygotic genome starts from intronless genes sharing a common cis-element. This cis-element interacts with a single protein, the Bicoid stability factor, and acts as a potent enhancer capable of timing the activity of an exogenous transactivator. We propose that this regulatory mode links morphogen gradients with temporal regulation during the maternal-to-zygotic transition.
| Embryonic development is controlled by a complex interaction between maternal and zygotic activities. Maternal messenger RNAs and proteins are deposited in the unfertilized egg during oogenesis; after fertilization, the activation of the zygotic genome is accompanied by the degradation of a fraction of maternally supplied transcripts. This switch from maternal to zygotic control of development is characterized by a dramatic remodeling of gene expression, and represents a universal regulatory point during animal development. Because it is not usually possible to identify which genomes are contributing to these transcriptional changes, we have used chromosomal ablation to determine maternal versus zygotic contribution for each mRNA detectable on microarray in the Drosophila blastoderm. This has allowed us to distinguish transcriptional and post-transcriptional modes of regulation and to identify common cis-regulatory elements associated with different classes of transcripts. Our analysis revealed that although mRNA degradation does not involve a simple regulatory code, the activation of the zygotic genome is based on a simple mechanism, which links morphogen gradients with temporal regulation. It will be interesting to address whether similar mechanisms also operate in other animals.
| Embryonic development is controlled by a complex interaction between maternal and zygotic activities. Although maternal transcripts and proteins are deposited in the egg during oogenesis, the activation of the zygotic genome starts at different stages in different animals and is concomitant with the degradation of a fraction of maternally supplied transcripts [1–3]. Thus, during the maternal-to-zygotic transition (MZT), the embryo undergoes an extensive remodeling of gene expression and must integrate post-transcriptional regulatory mechanisms, which are the only ones operating during the previous maternal stages, with transcriptional regulation of its own genome. How this is achieved is poorly understood.
The concomitant degradation of maternal transcripts and activation of zygotic transcription has made it difficult in any animal to interpret changes in gene expression [4–6]. Whereas an increase in gene expression levels can be interpreted as a sign of zygotic transcription, a decrease or absence of change is also consistent with zygotic gene activation if it is accompanied by maternal mRNA degradation. One way to test whether a particular RNA is supplied maternally or zygotically is to compare its levels in embryos that have or do not have the corresponding DNA template. Under these conditions, differences in expression level indicate the relative maternal and zygotic contribution. Drosophila melanogaster offers the unique opportunity to perform such an experiment for the entire genome, as it is possible to use chromosomal rearrangements to produce embryos that lack specific arms or even entire chromosomes [7,8]. Such embryos develop normally until cycle 14 and then show defects characteristic of the chromosomal region deleted. The results of such experiments suggest that the Drosophila embryo develops under the control of maternally provided proteins until nuclear division 13. This stage, usually referred to as the mid-blastula transition (MBT), defines the point from which development comes to be controlled by the zygote's own genome [1]. The first morphological signs of the zygotic genome appear with the cellularization of the cortically migrating nuclei and the beginning of gastrulation. From a transcriptional point of view, the zygotic genome is silent until nuclear cycle 9–10 [9]. In the germline, this quiescence is maintained until later stages of development, arguing for specific regulation between the soma and the germline [10].
The molecular mechanisms linking the nuclear cycles to the activation of transcription are unknown and may involve the chromosomal squelching of negative regulators of transcription, as has been proposed for the Xenopus embryo [3]. Chromatin-based mechanisms may also play a role. In the mouse embryo, for example, at least one cycle of DNA replication is required to change the methylation state of the chromatin to a transcriptionally competent conformation [11]. However, in none of these organisms have the molecular players actually regulating activation of the zygotic genome been identified. Because such regulators must be maternally provided, they are not easily identifiable in genetic screens. On the other hand, the recent technological advances in genomics and bioinformatics may offer alternative strategies for elucidating this mechanism, especially if the identification of cis-regulatory elements can be coupled to a biochemical characterization of the factors that bind to them.
Here we took advantage of the phenotype generated by the removal of specific genes acting during cellularization to identify embryos lacking defined chromosomal arms, and analyzed their expression profiles using microarrays. Because this strategy allows discrimination between transcriptional and post-transcriptional regulation of gene expression, we describe here the first complete analysis of the MZT during animal development.
Earlier attempts to identify zygotically active genes in Drosophila relied on comparing mRNA levels at cycle 14 with those from unfertilized eggs or early 0–1-h-old embryos [12]. Although zygotic transcription begins already at earlier nuclear cycles (9–10), we also started our analysis by focusing on cycle 14 because this stage represents the earliest time point at which the mutant phenotypes associated with the deletion of each specific chromosome can be recognized. The time-course characterization of earlier time points will be presented in the section describing the activation of the zygotic genome. The temporal resolution of our measurements is at 1-h intervals covering the first 3 h of embryogenesis: (1) unfertilized eggs, (2) 0–1 h (cycles 1 to 10), (3) 1–2 h (cycles 10 to 13), and (4) 2–3 h (cycle 14).
Figure 1A plots the levels of mRNAs from visually staged 0–1-h eggs with those that have developed to cycle 14 (2–3 h). In principle, this type of measurement allows identification of the following categories of transcripts: (1) purely zygotic (transcripts that are not expressed at 0–1 h and are detected as present at 2–3 h), (2) maternal+zygotic (transcripts that are present at 0–1 h and whose level increases at 2–3 h), and (3) maternal or maternal+zygotic (transcripts that are present at 0–1 h and whose level either does not change or decreases in level at 2–3 h).
Transcripts expressed at the same level in both collections lie on the diagonal (Figure 1). A large fraction of transcripts deviates from the diagonal and are present at increased or decreased levels in cycle 14. Although mRNAs that increase can be most simply explained by new transcription, the existence of mRNAs whose levels go down suggests that post-transcriptional regulation may be too complex to make judgments about the maternal or zygotic source of a transcript based on measured mRNA levels alone. The decrease or stability in the level of mRNAs may reflect a complex balance between activation and degradation. Even the identification of purely zygotic transcripts can be problematic if the designation is based only on measurements at 2–3 h being above the background at 0–1 h. To address this problem, we undertook a genetic approach based on chromosomal deletions (in embryos that had developed exactly to the same stage) coupled to microarray analysis. We sought to evaluate the traditional interpretation of gene expression measurements, which considers up-regulated transcripts as zygotic, stable transcripts as maternal, and down-regulated transcripts as maternal-degraded (Figure 1B, model).
The left arm of the second chromosome represents approximately 20% of the entire genome and is predicted to contain approximately 2,500 open reading frames (BDGP4 annotation; Berkeley Drosophila Genome Project, http://www.fruitfly.org/). We compared mRNAs from embryos that lack the left arm of the second chromosome with similarly staged wild-type embryos. Such 2L− embryos can be recognized by their distinctive halo of lipid-rich cortical cytoplasm during cellularization [13], at the precise moment when major zygotic transcription begins. Figure 1C and 1D plot the result of this experiment.
Most mRNAs have similar levels in both collections, and lie on a diagonal (Figure 1). Deviations tend to be located towards the lower left of the diagonal, indicating that certain mRNAs are less abundant in the 2L− collection. There is a small number of mRNAs whose level increases when the 2L arm is removed. Altogether these changes can represent direct and/or indirect responses to the ablation of the 2L arm. Although primary responses must involve genes located on 2L, secondary responses are expected to be randomly distributed on the three major chromosomes. We plotted the chromosomal location of down-regulated and up-regulated genes at different cut-offs (Figure 1E and 1F). At a stringent fold-change cut-off value of ten, the number of deviant mRNAs is small, and all of them represent mRNAs that are less abundant than in wild type. Approximately 90% of these mRNAs are encoded by genes located on 2L, indicating that they are normally supplied by zygotic transcription: removal of 2L eliminates the DNA templates for such transcripts, and the transcripts are not made.
As we decrease the fold-change cut-off, the number of genes that deviate from the diagonal increases. A 2-fold cut-off identifies 378 genes on 2L whose levels depend on the presence of that chromosomal arm in the embryos. A 2-fold difference signifies that at least 50% of the total number of transcripts for each of these genes, present at cycle 14, are derived from zygotic rather than maternal transcription. The observation that even at this cut-off, approximately 60% of down-regulated genes are located on 2L strongly validates this procedure. Indeed, if the observed changes were due to random fluctuations of mRNA levels, such changes would be distributed over the entire genome, and 20% of them would be located on 2L. It should be noted that, in principle, the down-regulated transcripts might also include maternal mRNAs whose stability is regulated by zygotic transcription. However, the enrichment on 2L suggests that this applies to a very small fraction of genes. We therefore classify all down-regulated transcripts (on the deleted arm) as zygotic. The remaining 631 genes that are located on 2L and detected in cycle 14 embryos are not dependent on the presence of the left arm of the second chromosome in the embryo, and must therefore be supplied by maternal transcription. At the 2-fold cut-off, a second class of affected mRNAs appears. These mRNAs are expressed at a higher or lower level than the wild-type controls, and they mapped to other regions of the genome. We interpret these mRNAs as gene products whose levels depend indirectly on the left arm of the second chromosome. We therefore name these genes “secondary targets” of 2L removal. They may be targets of transcription factors encoded on 2L whose expression at cycle 14 depends on the presence of that arm. Alternatively, they might be post-transcriptionally regulated maternal transcripts whose stability or degradation depends on zygotic transcription. In order to discriminate between these mechanisms, we screened the entire genome and determined the maternal and zygotic contribution for each individual gene.
Using additional chromosomal rearrangements, we extended the analysis described in detail above for 2L to the rest of the genome, analyzing mRNA populations present in embryos deficient for the X chromosome, the entire second chromosome, or the entire third chromosome. In most cases, hybridizations were performed in quadruplicate using different batches of embryos. Mutant embryos were recognized under a compound microscope based on their specific abnormalities associated with defects in nuclear morphology, in actin-myosin dynamics, and organelle transport: nullo (chromosome X) [14], halo (Chromosome 2) [13], and bottleneck (Chromosome 3) [15]. These three phenotypes appear synchronously as the embryo enters cycle 14, thus allowing a precise staging protocol (Figure 2A–2J). In each case, we were able to identify mRNAs encoded on the deleted chromosome and whose levels depend directly on the presence of that chromosome. These mRNAs thus appear to be predominantly supplied by zygotic transcription (Figure 2K). For all subsequent analyses, we defined the class of down-regulated genes to be those genes with a fold-change of at least three and a p-value less than 0.001. In this range, between 60% to 80% of down-regulated genes map to the chromosomes removed. A more stringent cut-off would have increased specificity at the expense of secondary targets. In addition, we have built a simple online database, which provides access to the entire dataset at any (user-specified) fold-change and/or p-value cut-off (http://rd.plos.org/pbio.0050117).
A 3-fold cut-off identifies all mRNAs that are at least 67% supplied by zygotic transcription at cycle 14. Combining the data from all four manipulations, we estimate that such zygotically active genes represent about 18% of the genes detectable at cycle 14, i.e., 1,158 genes distributed on all four chromosomes (Table S1). The remaining mRNA species appear to be supplied predominantly by maternal transcription. When looking at the entire dataset, zygotically active genes appear to be uniformly distributed throughout the genome.
Each chromosomal manipulation also identified apparent secondary targets that mapped to other chromosomes. Similar to the results obtained from the 2L− experiments, levels of such mRNAs deviated at most 2- to 3-fold in either the positive or the negative direction from wild-type (WT) mRNAs. To test whether these genes were in fact transcriptional targets of genes on the removed chromosome, we asked whether third chromosomal or X chromosomal genes identified in the 2L chromosomal screen as secondary targets behaved as primary targets when the third chromosome or X was removed. This was true for 62% of down-regulated and 29% of up-regulated genes. Our four experiments identified a total number of 778 secondary targets of which only 28% are zygotic (Table S2). The remaining 72% (563) are mostly maternally supplied. We conclude from these observations that the expression level of most zygotically active genes was not influenced by other loci, and changed significantly only when the chromosome encoding them was removed.
The identification of 563 non-zygotic mRNAs (Table S3) whose level changed in response to the removal of a specific chromosome must represent post-transcriptional regulation of maternal transcripts. The stability or degradation of these transcripts may be regulated by transcription of certain factors (coding for RNA-binding proteins or regulatory RNAs) on the chromosomes that are removed. In agreement with this interpretation is the observation that ablation of each chromosome or chromosomal arm results in the misregulation of distinct targets. Thus, transcription at multiple loci regulates the stability of distinct maternal transcripts. For example, the degradation of String and Twine, two cell cycle regulators involved in timing the MZT [16], is regulated by zygotic transcription on the X and second chromosomes, respectively (Table S2).
Next, we characterized the relative contribution of maternal transcripts to the total cycle 14 expression level of zygotically active genes. We compared the mRNA levels of 1,158 zygotic genes at 0–1 h with that observed at 2–3 h (Figure 3A). In one third of the cases, transcripts could not be detected in 0–1-h embryos, and increased over the 2-h period that follows fertilization. Expression of these genes is therefore purely zygotic: all transcripts detected at cycle 14 are produced by transcription in the embryo itself (Table S4). Almost all of these transcripts (~90%, 300 out of 334) would have been detected as purely zygotic using the simple criterion “absent at 0–1 h–present at 2–3 h” (Table S5). On the other hand, if only this latter criterion had been used, an additional 268 genes would be scored as purely zygotic, even though the level of these transcripts does not change significantly when the chromosome harboring them is removed. When used on its own, the “absent-present” filter may be unreliable because it identifies zygotically active genes by comparing expression measurements at one stage with background levels at another. Our double-filter approach (change in response to the deletion of the DNA template + a “present-absent” value) yields a more stringent and accurate estimate of purely zygotic transcripts.
The remaining two thirds of the 1,158 zygotic genes were present in unfertilized eggs (Table S6). Because the overall levels of theses mRNAs either did not change significantly or decreased between 0 h and 3 h, the dependence of cycle 14 levels on zygotic transcription implies the specific degradation of maternal transcripts before that time. Thus, we conclude that an increase in gene expression over time is not a sufficient criterion to identify zygotic genes.
To follow the stability of maternal transcripts (which is obscured by the presence of newly supplied zygotic transcripts in WT embryos), we compared mRNA levels from early 0–1-h embryos (WT) with mRNA levels from embryos missing each chromosome, hand-selected from the same stock during cycle 14. The initial analysis was restricted to genes on the left arm of the second chromosome (Figure 3B); therefore, all 2L mRNAs detected at either stage must be supplied maternally. The relative change in expression levels between 0–1 h and 2–3 h provides a measure of their stability during that period. Consistent with their strictly maternal source, none of the 1,009 transcripts from 2L increased significantly between 0 h and 3 h. Approximately 65% remained constant, and 35% dropped more than 3-fold. We extended this analysis to the rest of the genome and estimated that of 6,485 total maternally supplied genes, 2,110 (33%) go down significantly by cycle 14. In 646 cases, the maternal degradation was at least in part compensated by zygotic transcription (Table S7). A representative list of maternal and zygotic transcripts known to be degraded or induced during the MZT and detected by our analysis is shown in Table 1.
One third of the zygotic transcripts we have identified are not expressed maternally and can be considered purely zygotic genes. These genes are enriched for transcription factors (“transcription factor activity” Gene Ontology (GO) category, p < 10−9). This may reflect the necessity of timing the activity of genes regulating the establishment of cell identity during differentiation. The remaining two thirds, those with maternal contribution, are not significantly enriched in any specific functional class. This raises the question as to why the embryo transcribes genes when the corresponding maternal transcripts are present. Two possible scenarios can be envisaged: (1) maternal transcripts must also be supplied by zygotic transcription, because they are degraded very quickly (i.e., they have short half-lives), and (2) zygotic transcription offers some advantages, such as precise spatial patterning, differential processing (e.g., splice variants), or intracellular localization. In the latter scenario, maternal mRNAs would be specifically degraded to ensure that zygotic transcripts are the only source of these genes at cycle 14.
Using data downloaded from the BDGP in situ database, we asked whether the zygotic genes we have identified are expressed in specific patterns at cycle 14 (Figure 3D). A total of 241 of the genes we found to be zygotic are annotated in the database. Of those, 59% were expressed in discrete patterns at cycle 14. Among the total number of genes in the database (1,227), only 27% were patterned at cycle 14. Thus, the zygotic genes we identified are enriched more than 2-fold (p < 10−32) in patterned expression compared to what would be expected by chance. Even among the 143 zygotic genes that initially had uniform maternal component, 29% evolved to patterned expression by cycle 14, a situation occurring for only 11% of the genes in the entire in situ database (p < 10−8). Because the expression level of these genes was either stable or decreased during the MZT, we conclude that coupling maternal degradation with zygotic transcription is part of the patterning mechanism. Indeed, one third of the genes expressed in patterns at cycle 14 required both zygotic transcription and degradation of uniform maternal mRNAs (Figure 3D).
We then asked whether the different categories defined above share common genomic regulatory elements, which could explain the behavior of an individual gene during the MZT.
We first investigated whether down-regulated maternal genes have over-represented motifs in their 3′ UTRs. A total of 1,095 maternal genes with annotated 3′ UTRs decreased in levels significantly between 0–1 h and 2–3 h. As shown in Figure 3C, we found several short sequences that are significantly enriched within these 3′ UTRs, compared to the entire set of annotated 3′ UTRs. None of them matched the 5′ extremity of any of the 78 known microRNAs (miRNAs) in D. melanogaster. A similar conclusion was drawn also by studying transcript stability in unfertilized eggs [17]. The sequences we found can be divided into two families, based on sequence similarity. The first family contains a UUGUU core, which resembles the target site for the PUF family of RNA-binding proteins (whose unique representative in the D. melanogaster genome is Pumilio). To further investigate the role of Pumilio in maternal mRNA degradation, we compared our down-regulated maternal genes to the list of 135 targets of Pumilio in fly embryos [18]. Although these targets do not all contain exactly the same sequence, 118 of the 135 target genes were identified as maternal in our experiments, and 63 of these (53%) were also down-regulated between 0–1 h and 2–3 h. On the other hand, only 23% of maternal genes decreased globally. Therefore, Pumilio targets are very significantly over-represented in maternal down-regulated genes (p < 10−12). Sequences from the second family match the AU-rich element (canonically defined as UAUUUAU), a known mediator of mRNA degradation [19]. Interestingly, an RNA interference (RNAi)-based screen performed in Drosophila S2 cells has suggested that several components of the miRNA processing pathway are required for degradation of AU-rich element–containing mRNAs [20].
We then investigated whether the zygotic transcripts share common DNA regulatory motifs in their upstream regions. We identified a highly over-represented 7-nucleotide–long sequence (CAGGTAG, which from now on we will refer to as the 7mer) and several of its variants within the 2 kilobase (kb) upstream regions of purely zygotic genes (Figure 3C). This motif has been previously identified in the upstream region of sisterless A and B, and Sex-lethal, three genes involved in sex determination that are expressed early during embryogenesis [21]. A more recent study identified this motif upstream of other genes expressed prior to cycle 14, thus suggesting a more general regulatory function [22].
The results described above are intriguing because the 7mer we found is present upstream of only a fraction of the zygotic genes at cycle 14. Although the major activation of the zygotic genome occurs at cycle 14, earlier reports indicated signs of zygotic transcription as early as cycle 10 when the embryonic DNA is still engaged in fast cycles of S-phases and mitoses without interphases [23]. We therefore asked whether the 7mer represents a general feature of genes expressed prior to cycle 14 and, in general, whether the zygotic genes we have identified are transcribed altogether during cycle 10 or whether different classes of transcripts respond differently to the embryonic cycles and DNA content.
We compared the expression profile of unfertilized eggs, 0–1-h freshly fertilized eggs (pre-pole cell formation, cycles 1–9) and 1–2-h embryos (post-pole cell formation and pre-cellularization, cycles 10–13). No significant change in expression levels was observed between unfertilized eggs and the 0–1-h eggs, indicating that neither transcription nor degradation has occurred (Figure S2). Importantly, in these experiments, we analyzed unfertilized eggs that had been aged for 1 h at most. Therefore, our results do not contradict previous reports describing the degradation of a subset of maternal transcripts in unfertilized eggs [17,24] since, in those studies, unfertilized eggs were aged for longer periods of time, and degradation was observed after 2 h, peaking between 2 and 4 h.
Between the 0–1-h to 1–2-h collections, a single group of 59 genes was significantly up-regulated (Figure 4A). These genes (including Snail, Zen, and Nullo, see Table S8 for a complete list) are expressed even prior to the gap and pair-rule genes, which in our measurements do not yet show significant increased levels at this time point. Expression of gap and pair-rule transcripts was detected at 2–3 h (Table S1), arguing that their transcripts accumulate with a slower kinetic. When searching for over-represented motifs in the 2-kb upstream regions of these genes, we found the same motif as for the pure zygotic genes, along with other overlapping or slightly distinct variants (Figure 4B). We found 91.5% of the 59 genes have at least one copy of any of these variants, whereas the expectation based on all genes in the genome is 40% (p < 10−15). Moreover, 28.8% of the 59 genes have four or more non-overlapping copies of these sequences, a situation occurring for only 1.6% of the Drosophila genes (p < 10−16). Thus we conclude that the activation of the zygotic genome starts from genes containing this motif. Interestingly, the occurrences of the 7mer within the 2-kb upstream regions tend to be much closer to the transcription start site than expected by chance (Figure 4C). Finally, we asked whether these genes share some additional features that increase the overall fitness of gene expression prior to cycle 14. We found that 70% of these genes do not contain introns (Figure 4A). Since intronless genes represent only 20% of the Drosophila genome, this result suggests an important selective advantage for the transcription of intronless genes in concomitance with fast-cycling nuclei.
The identification of a single highly over-represented cis-element in the 5′ region of the early zygotic genes suggests the existence of a single trans-acting factor involved in timing the activation of the zygotic genome. If such a factor exists, it is most likely maternally provided and loaded into the egg during oogenesis. To identify this factor, we undertook a biochemical approach. We performed sequential DNA affinity chromatography (see Materials and Methods for details) using the 7mer or, as negative control, the upstream activation sequence (UAS) (the consensus binding site of the yeast trans-activator GAL4). The result of this experiment is shown in Figure 5A. Only one band was detected in the 7mer elute, and no specific band was detected in the UAS control elute. Mass spectrometry sequencing identified this protein as the Bicoid stability factor (BSF), and Western blotting analysis confirmed this result (unpublished data).
BSF has been previously identified as a Bicoid mRNA binding protein involved in regulating the stability of Bicoid transcripts during oogenesis [25]. Our data suggest an additional transcriptional function for BSF in the embryo, and indeed, the human homolog of BSF has been shown to function as a transcriptional regulator [26].
In order to address the specificity of the 7mer/BSF interaction, BSF was expressed in rabbit reticulocyte in the presence of 35S methionine, and the binding to the 7mer or to a mutated oligo (in which the two GG at position 3 and 4 were mutated to TT) was tested. In vitro–synthesized BSF bound directly and specifically to the 7mer, and only background signal was retained on the beads coupled to the mutated oligo (Figure 5B).
Next, we analyzed the subcellular distribution of BSF in the embryo using immunostaining and confocal microscopy imaging. BSF was localized to both the cytoplasm as well as the nuclei of the blastoderm epithelium (Figure 5C and 5D). In the germ cells (pole cells), which at this stage are transcriptionally silent, BSF was retained in cytoplasmic puncta (Figure 5E and 5F). Thus, BSF is differentially compartmentalized between the soma and the germ line, and this compartmentalization may be important to maintain the transcriptional quiescence in the germline.
To test the function of BSF in the early embryo, it is necessary to remove the maternal contribution. (BSF transcripts are maternally provided and the protein is expressed during oogenesis [25].) To perform this experiment, we produced germline clones using a P element insertion that maps in the BSF open reading frame and is homozygous lethal. Flies containing such clones failed to lay eggs, and the ovaries were arrested at a very early stage of development, indicating that BSF is required also during oogenesis. This made it impossible to test the function of BSF in the early embryo. Therefore, we took an alternative approach with the aim to functionally characterize the activity of the 7mer. We considered two possible scenarios. One possibility is that the 7mer may have enhancer activity, sufficient to drive transcription on its own. Alternatively, it may play a permissive role by functioning in a combinatorial fashion with additional factors. To discriminate between these two possibilities, we set up conditions to measure gene expression using an assay based on the UAS/GAL4 system [27].
We generated embryos expressing green fluorescent protein (GFP) under the control of the UAS–heat shock minimal promoter either with or without five copies of the 7mer, and followed GFP expression using video microscopy. GFP was not detected in embryos unless GAL4 was also provided. Strikingly, the presence of the 7mer led to a more than 4-fold increase in the expression of GFP compared to controls (transgene without the 7mer), as shown in Figure 6A and 6B.
Next, we asked how early this stimulatory activity could be detected. We analyzed GFP transcripts using fluorescent in situ hybridization (FISH). This technology allows the visualization of nascent transcripts as they arise from the site of transcription [28]. Because we crossed males carrying the GFP transgene to females providing GAL4, only one chromosome in the embryo is expected to transcribe GFP. In agreement with this prediction, we detected only one major transcription focus, appearing as an individual dot, per nucleus (Figure 6C and 6D). We observed an increase in the number and size of dots at each nuclear division when the 7mer was present (Figure 6D). This difference could be detected as early as cycle 11 (Figure S1). By cycle 14, images are characterized by a high signal-to-noise ratio and showed an approximately 1.7-fold increase in the number of dots per embryo (Figure 6E). Thus, the presence of CAGGTAG increases the number of nuclei that are actually engaged in transcription. Because the size of each dot is also larger (Figure 6F), each dot most likely contains more transcripts. If this interpretation is correct, then it should be possible to quantify this difference by measuring the total amount of GFP transcripts.
Embryos were harvested either at the stage when the earliest GFP transcripts were expressed (cycle 10 to 13) or at cycle 14. Total RNA was extracted and subjected to reverse-transcription PCR (RT-PCR) (Figure 6G). As a staging control, we followed the expression of Snail, a known zygotic gene. We detected 7mer-driven transcription as early as cycle 12. In the absence of this motif, no GFP expression was detected. By cycle 14, we observed a 2-fold increase in GFP expression, which is in agreement with the FISH quantification. As a control, we also followed Snail mRNA, which was expressed at similar levels in both conditions, and its expression increased from cycle 12 to 14. Altogether, these results show that the CAGGTAG motif functions as an enhancer that cannot drive transcription on its own (Figure 6C), but can activate expression prior to cycle 14, in combination with a transcriptional activator. In agreement with this result is the finding that CAGGTAG and its variants are particularly abundant in enhancer sequences bound by Dorsal and Bicoid (Table S9), thus suggesting a combinatorial regulatory mode (see model in Figure 7 and Discussion).
The switch from maternal to zygotic control of early embryonic development is characterized by a dramatic remodeling of the transcriptional complexity present in the oocyte. We have genetically identified the relative maternal and zygotic contribution for the expression of each individual gene during the D. melanogaster mid-blastula transition. The criterion we used to identify zygotically expressed genes is strictly based on the direct relationship between the DNA template and the corresponding transcript. The specific phenotype generated upon removal of each chromosomal arm allowed us to collect a synchronous population of embryos just at the stage when the first morphological signs of the zygotic genome become visible. The location of the majority of down-regulated genes to the chromosomal arm that was ablated provided an excellent control for the entire experimental procedure we have undertaken.
In summary, our results indicate that zygotic transcription contributes to approximately 20% of the genes expressed at cycle 14, and as much as 30% of maternal transcripts become unstable during the mid-blastula transition. However, about a third of these transcripts are also supplied by zygotic transcription and, therefore, their expression levels at cycle 14 remain constant. Purely zygotic transcripts represent only a third of the total set of zygotically expressed genes. The remaining two thirds also had a maternal contribution and were present in unfertilized or 0–1-h eggs. The zygotic transcription of such maternally provided genes does not always result in an increase in the total amount of transcript, indicating specific degradation of the maternal counterpart. Thus, a change in gene expression over time is not a sufficient criterion to identify zygotically active genes, nor to measure the stability of maternal transcripts. These results have important implications for the definition of maternal and zygotic genes, and provide a genome-wide analysis, which will be instrumental for a molecular characterization of the MZT.
Our analysis shows that purely zygotic transcripts are enriched in transcription factors. Providing these genes through zygotic transcription, which in turn is related to the number of nuclei, ensures that correct number of cells is assigned to a specific fate and, ultimately, the establishment of the correct body proportion. The execution of a specific differentiation program represents a more complex problem, in that it involves the adjustment of the expression of genes involved in basic cell function. Therefore, these genes must be expressed during oogenesis, to support oocyte development, and their activity modulated through zygotic transcription.
Our data argue that zygotic transcription allows a large fraction of ubiquitously expressed maternal mRNAs to be expressed again in localized patterns at the blastoderm stage. The generation of these patterns involves the degradation of the maternal transcript and the corresponding activation of zygotic transcription. This result is in agreement with previous studies on individual genes (e.g., the maternally provided Cdc25 phosphatase string and the maternal-zygotic transcription factor hunchback), which were reported to undergo a similar MZT [29,30]. Altogether our analysis is consistent with the proposal that zygotic transcription provides the spatial precision at which important regulatory genes must be expressed during the differentiation of the developing embryo. Therefore, our results can be used in combination with chromosomal deficiency screening to quickly identify gene function at the mid-blastula transition by reducing the number of candidate genes contained in each deficiency to zygotic-dependent expressed genes. We have used this approach to identify the bearded genes as the zygotic genes regulating Notch signaling during mesoectoderm specification [31].
In addition, our results show that zygotic transcription is required for the degradation of a distinct subset of maternal transcripts. Because these transcripts do not share any statistically enriched common regulatory sequence and because each chromosomal manipulation targeted distinct transcripts, we propose that multiple zygotic activities must be involved in this regulation. One possibility is that the zygotic expression of miRNAs might be part of this mechanism. For example, in zebrafish, mir-430 was shown to control the degradation of a pool of maternal transcripts [32]. Although we have not detected any significant over-representation of known miRNA target sites in our data, the involvement of miRNAs in specific pathways can be tested once zygotic control regions have been more closely defined. We also identified maternal mRNAs that are degraded and not replenished by zygotic transcription. A fraction of these genes share sequences within their 3′ UTR, which resemble the known target site for the Pumilio RNA-binding protein. We showed that a very large fraction of the Pumilio targets in the embryo are indeed degraded during the transition from maternal to zygotic stages. Pumilio was first identified as an inhibitor of translation controlling posterior fate by promoting deadenylation of hunchback mRNA [33,34]. Our results suggest that Pumilio might also promote degradation of mRNA targets as shown for the yeast homolog Puf3 [35].
The transition from a silent to a transcriptionally active genome is one of the most dramatic events in a developing embryo and is subject to regulation at multiple steps. We have identified the CAGGTAG motif (and its variants) as an important player in this transition. We identified the BSF as the factor binding to this motif in the early embryo.
BSF has been previously identified as a protein binding to the 3′ UTR of Bicoid mRNA and involved in regulating Bicoid transcript stability during oogenesis [25]. However, the precise biochemical function of BSF is unknown. Mutation of this gene causes lethality, and induction of homozygous germline clones arrests oogenesis (see Results). Thus, BSF must have additional function other than the regulation of Bicoid transcripts because Bicoid itself is not required for oogenesis, and zygotic mutants are viable.
Interestingly, the human ortholog of BSF, the leucine-rich protein LRP130, has been shown to bind to a cis-regulatory sequence in the 5′ proximal region of the MDR1 gene and to act as a transcriptional regulator [26,36]. Although we could not genetically test the function of BSF, our analysis suggests that BSF is not able to drive transcription on its own, but must act in a combinatorial fashion. Indeed, we found that CAGGTAG and its variants are particularly abundant in enhancer sequences bound by Dorsal and Bicoid (Table S9). These transcription factors activate the expression of their target genes in a concentration-dependent manner and define distinct developmental units along the dorsal-ventral (D-V) and anterior-posterior (A-P) axes [37,38]. Interestingly, only a subset of the known targets for these transcription factors have this element, even within the same spatial unit. For example, in the D-V patterning system controlled by the Dorsal gradient, CAGGTAG is found in Snail and Tom, but not Neuralized. Both Snail and Tom are expressed prior to Neuralized (Snail and Tom are among the 59 genes induced during cycle 10–11) and must act before Neuralized to precisely position Notch signaling at the mesoderm–mesoectoderm boundary [31]. The inability of CAGGTAG to drive transcription on its own makes it an ideal timer, which links spatial gradient with temporal regulation (see model, Figure 7).
In conclusion, the experiments described in this work will be instrumental for studying the activation of the zygotic genome in other animals and for guiding embryonic stem cell differentiation. In the mouse, zygotic transcription begins by the two-cell stage, and a large number of maternal mRNAs persist beyond this stage [4]. The contribution of zygotic transcription to the expression of these mRNAs is still unknown. Further, the activation of the mouse genome is characterized by discrete wave-like patterns of gene expression. Similarly, the activation of the Drosophila genome starts with a battery of 59 transcripts induced from cycles 10–11 to cycle 14. Interestingly, 70% of these genes do not contain introns, and they all encode small proteins. This result is in agreement with two previous studies showing that genes transcribed early in development tend to be unusually short [39] and that the presence of a long intron (19 kb) limits the expression of the knirps-related gene (knrl) early during Drosophila development [40]. The use of intronless genes might reflect the necessity of expressing regulatory genes requiring a minimal response time. Intriguingly, many short genes in the human genome have been implicated in anti-sense–mediated gene regulation [41]. In the Drosophila embryo, the expression of intronless genes occurs when the nuclei are still engaged in rapid phases of DNA duplication and mitoses. Because the nuclear membrane is required for the assembly of the splicing machinery, the selection of intronless genes might ensure the production of functional transcripts in concomitance with nuclear divisions.
WT flies were Oregon-R; all stocks were maintained by standard methods at 18 °C, unless otherwise specified. Transgenic embryos over-expressing GFP were generated using P element–mediated germline transformation using w1118 as the recipient host. GFP was ectopically expressed using the 7merUAS-GFP line 4 (III) or UAS-GFP line 10 (II) and the matαTub-Gal4VP16 67C;15 driver. Embryos were collected at room temperature. The halo deficiency is Df(2L)dpp[s7-dp35] 21F1–3;22F1–2 and was balanced over CyO [13]. Compound chromosomes: embryos with no X chromosome were obtained by crossing attached-X/Y females to X/Y males. The stock used was C(1) DX, y f [7]. The compound II chromosomes RM(2L); RM(2R) = C(2)v, in which the two left arms or the two right arms segregate together, were used to generate 2L− and 2R− embryos [8]. The compound II C(2) EN and compound III C(3) EN st1, cu1, es, stocks (Bloomington 2974 and 1117) were used to generate embryos deficient for the entire second and third chromosome, respectively. The BSF P element insertion used for the germline clone experiments was obtained from the Szged stock center: FRT-l(2)SH1181. This P element insertion has been mapped to the BSF cDNA and is homozygous lethal. Moreover, it failed to complement a deficiency covering the BSF genomic locus Df(2L)M36F-S5.
Embryos were collected on apple juice-agar plates, visually staged under a compound microscope, dechorionated for 2 min in 5.25% sodium hypochlorite (Austin's bleach), washed in water, and then frozen in 1 ml of heptane (Sigma, http://www.sigmaaldrich.com) using a dry ice/ethanol chamber. Total RNA was extracted with TRIzol (Invitrogen, http://www.invitrogen.com), and 10 mg of RNA (approximately 100 embryos) was used to synthesize complementary RNA (cRNA) according to the Affymetrix protocol. Each array (standard format: Drosophila genome 1, Affymetrix) was probed with 15 mg of biotinilated cRNA for 16 h in a 45 °C oven. Arrays were washed and stained using the GeneChip Fluidics Station (Affymetrix, http://www.affymetrix.com) according to the EukGE-WS2 protocol. Subsequently the arrays were scanned with an Argon-ion laser scanner (Affymetrix). Graphs were generated using the GeneSpring software (Silicon Genetics/Agilent, http://www.chem.agilent.com). CEL files were loaded in R (http://www.r-project.org), and analyzed using the Bioconductor package [42]. The analysis followed the “Golden Spike” methodology described in [43]. Briefly, Present, Marginal, or Absent flags were computed using the MAS approach, with default parameters. Within a group of replicates of the same condition, a probe set was identified as present if it had a Present flag (not Marginal) in more than 50% of the replicates. Intensity values were corrected using the MAS approach, and arrays were normalized with respect to each other at the probe level using a loess function. PM probe intensities were corrected for unspecific hybridization using the MM probes and the MAS approach. Expression summaries were generated using the MedianPolish method. A second loess correction was subsequently applied to the expression summaries. Two-tailed statistical tests were performed on the replicates using the regularized t-test approach implemented in CyberT [44], with normalization constant set to five times the minimum number of replicates among the two populations analyzed, and the window size set to 101. Further set operations (unions and intersections) were performed at the probe set level, using custom R functions.
Probe sets from the DrosGenome1 Affymetrix platform were matched to the latest version of the D. melanogaster cDNAs downloaded from ENSEMBL [45], as of August 2005 (BDGP4). Briefly, each probe was matched to the cDNA set using BLAST, with seed length 11, and e-value threshold 1e-5. Only probes with 100% identity to their match were considered. A probe set was considered to match a transcript if at least ten (out of 14) of its probes matched the transcript. Transcript identifiers were collapsed into their corresponding gene identifier. Two-kilobase upstream regions (up to the TSS when available, or to the ATG codon otherwise) and 3′ UTRs were downloaded from ENSEMBL, also as of August 2005. Only the longest 3′ UTR for a same gene was retained. Motif finding within a set of genes of interest was performed using an exhaustive k-mer enumeration and over-representation approach. Briefly, each 7mer was considered in turn (6-, 8-, 9- and gapped k-mers were also examined, but yielded negative or similar results). The set of genes of interest (e.g., early zygotic genes) that have at least one copy of the 7mer was determined, with size denoted as s1. The set of all Drosophila genes that have at least one copy of the same 7mer, in their upstream (or 3′ UTR) region was then determined, with size denoted as s2. The size of the overlap between the two sets was then calculated, and denoted as i. A p-value of the size of overlap being greater than i (representing the over-representation of the k-mer within s1) was calculated using the cumulative hypergeometric distribution. All 7mers were sorted based on their p-value, corrected for multiple testing using the Bonferroni correction. Only 7mers with corrected p-values lower than 0.05 were considered significant and used for further analysis. Motifs derived from 3′ UTRs were systematically compared to the seed regions within the sequences of the 78 known Drosophila miRNAs, downloaded in August 2005 from the miRNA registry [46].
The 0–3-h. embryos were harvested, dechorionated for 2 min, washed in phosphate buffer saline (PBS) 0.1% Triton X-100, and frozen at −80 °C. Forty grams of packed embryos were diluted in 50 ml of Lysis buffer (25 mM Hepes [pH7.6]; 100 mM KCl; 12.5 mM MgCl2; 1 mM DTT; 10% glycerol; 500 μg of Poly DI-DC (Sigma), 1X Protease Inhibitor cocktail (P8340; Sigma) and homogenized in a Dounce tissue grinder at 4 °C. Total lysate was spun down for 5 min at 1,000 rpm in order to pellet the nuclei. Total membranes were spun down by ultracentrifugation in a 70 Ti rotor (Beckman, http://www.bioscience.com) at 40,000 rpm for 1 h at 4 °C. The supernatant from this centrifugation step was incubated for 5 h at 4 °C with 500 μl of immobilized streptavidin agarose beads (Pierce 20347; http://www.piercenet.com) coupled to 1 mg of double-stranded UAS oligo (51 bases long, three tandem copies of the GAL4 binding site) biotinylated at the 5′ end of the upper strand with a BioTEG (5′-BioTEGCGGAGTACTGTCCTCCGCGGAGTACTGTCCTCCGCGGAGTACTGTCCTCCG) and equilibrated in 100 mM KCl; 25 mM Hepes (pH 7.4). The flow true from this purification step was then loaded onto either 500 μl of beads containing the 7mer sequence (49 bases long, seven copies of the CAGGTAG repeat; (CAGGTAGCAGGTAGCAGGTAGCAGGTAG CAGGTAGCAGGTAGCAGGTAG) or onto 500 μl of UAS oligo beads prepared as described above and incubated for an additional 4 h at 4 °C. Columns were washed with 20 ml of wash buffer (100 mM KCl, 25 mM Hepes [pH 7.4]). Bound proteins were eluted in three fractions containing 5 mM EDTA, 50 mM Hepes (pH 7.4), 1 mM DTT and increasing concentration of KCl: first, 500 mM KCl; second, 1 M KCl, and third, 2 M KCl. Each elution step was performed by incubating the beads for 10 min in a total volume of 500 μl at room temperature. Each eluted fraction was concentrated and desalted on a centrifugal filter 3-kDa cut-off (Millipore CAT NO: 42403; http://www.millipore.com) and loaded on SDS-Page. The bound protein was cut and sequenced according to standard mass spectroscopy procedure.
BSF cDNA was in vitro transcribed and translated in the presence of 35S methionine using the T7-TnT Quick-coupled transcription/translation system (Promega CAT NO: TM045; http://www.promega.com). We incubalted 60 μl of radio-labeled BSF for 3 h at 4 °C with 50 μl of beads containing the 7mer sequence or a mutated sequence (CATTTAG) prepared as described above, in a total volume of 500-μl buffer containing 25 mM Hepes (pH 7.4), 100 mM KCl, 12.5 mM MgCl; 1mM DTT 1-mg/ml BSA, and 2.5-mg poly dI-dC. After four washes in 1 ml of buffer containing 25 mM Hepes (pH 7.4), 100 mM KCl, and 1 mM DTT, bound proteins were eluted in 80 μl of 1.5 M NaCl, 50 mM Hepes (pH7.4), 5 mM EDTA, and 1 mM DTT. Then 30 μl of the eluted proteins were loaded on SDS-page and processed for autoradiography.
All variants of the CAGGTAG motif with a significant p-value were retained, and aligned manually. All occurrences of these variants within the 59 early genes were determined, along with their position with respect to the TSS; overlapping occurrences were removed. The remaining occurrences were used to form a weight matrix, whose motif logo was drawn using WebLogo [47]. All Drosophila enhancers in the Redfly database [48] were downloaded, as of March 2006. All enhancers were searched for the above variants of CAGGTAG, but only those that had at least two copies were retained. The density of CAGGTAG variants per kilobase was calculated for all retained enhancers, and enhancers were sorted according to this density.
All available in situ data and corresponding annotation keywords were downloaded from the BDGP in situ database [49], as of August 2005. Only genes with detectable expression at stages 4–6 were retained for further analysis, resulting in a set of 1,227 genes. Based on the annotated keywords, gene expression at each stage was classified as uniformly expressed, patterned, or not detected.
Two complementary oligos each containing five copies of the CAGGTAG repeats flanked by a SphI restriction site at both the 5′ and 3′ extremities were in vitro synthesized, annealed, cut with SphI, and cloned into the SphI site of the pUasT-EGFP (1075) destination vector (Gateway; Invitrogen). Recombinant plasmids were identified using PCR and sequencing for correct orientation. For the quantification of GFP transcripts, 80 embryos were visually staged under a regular dissecting microscope, hand selected, and total RNA extracted using Trizol (Sigma). For each time point, 80 ng of total RNA was used in a one-step RT-PCR reaction (Invitrogen Superscript One-step RT-PCR). Twenty-five cycles of amplification at TM 55 °C ensured a linear range of amplification. The primers used were: GFP-F: ACGTAAACGGCCACAAGTTC; GFP-R: TGCTCAGGTAGTGGTTGTCG; Snail-F: CGGAACCGAAACGTGACTAT; Snail-R: GCGGTAGTTTTTGGCATGAT. Amplified reactions were loaded on a 1.2% agarose gel, stained with ethidium bromide, and quantified using a gel documentation system equipped with a CCD camera (FluorChem; Alpha Innotech, http://www.alphainnotech.com).
Embryos were dechorionated for 2 min in bleach and fixed in 4% paraformaldehyde (Electron Microscopy Science, http://www.emsdiasum.com/microscopy/)-heptane for 20 min. Embryos were blocked in 10% bovine serum albumin (BSA) in PBS, 0.1% Triton-X-100 (Sigma) for 1 h. Primary antibodies were incubated in PBS containing 5% BSA and 0.1% Tween-20 for 12 h at 4 °C. Embryos were washed five times in PBS, 0.1% Triton-X-100, and incubated with secondary antibodies for 2 h at room temperature in PBS, 5% BSA, 0.1% Tween-20. After five washes in PBS, stained embryos were mounted in Aquapolymount (Polyscience, http://www.polyscience.com). Antibodies: mouse anti-Armadillo (1:50); rabbit anti-myosin-2 (1:500). Secondary antibodies were Alexa-488 conjugated (1:500; Molecular Probes). For the detection of BSF, embryos were heat fixed and the goat anti-BSF antibody (P.M. Macdonald) was used at 1:500 dilution as described in [25].
FITC-UTP–labeled RNA antisense probes against GFP were generated by cutting the pBlue script II EGFP(N) plasmids with KpnI and transcribing with the T7 polymerase. Embryos were dechorionated for 2 min in bleach and fixed in 4% paraformaldehyde (Electron Microscopy Science)-heptane for 20 min. Hybridization was performed in 50% formamide (Roche, http://www.roche.com), 5× SSC (Sigma), 1× Denharts (Sigma), 1% (w/v) blocking agent (Roche), 10-mg/ml yeast tRNA (Sigma), 0.1% Triton X-100 (Sigma), 0.1% CHAPS (Sigma) for 16 h at 56 °C. Thereafter, embryos were blocked in 2× Western blocking reagent (Roche), PBS, 0.1% Triton-X-100 (Sigma) for 1 h at room temperature and incubated with mouse anti-FITC (Roche) 1:400 incubated in blocking buffer for 3 h. Embryos were washed five times in PBS, 0.1% Triton-X-100, and incubated with anti-mouse Alexa488-conjugated secondary antibodies (Molecular Probes) used at 1:500 dilution for 1.5 h at room temperature in blocking buffer. After five washes, nuclei were stained using Toto-3 dye (Molecular Probes), incubated in 70% glycerol for 30 min, and mounted in Aquapolymount (Polyscience). Images were acquired with a PerkinElmer spinning disk confocal microscope equipped with a 40× (numerical aperture [na] 1.3) oil immersion objective (Nikon, http://www.nikon.com) using the Ultra VIEW imaging system (PerkinElmer, http://www.perkinelmer.com). Serial sections were collected, and the number and size of nuclear dots was quantified as follows. Image analysis was performed using MATLAB v7 and the Image Processing Toolkit (http://www.mathworks.com). Pixel intensities were first linearly stretched using the imadjust and stretchlim functions. Whole-embryo boundaries were located within the images using a thresholding of 0.20, a morphological opening with disk of radius 1 (pixel) for removal of small artifacts, a filling of type “hole” with radius 4, a morphological opening with disk of radius 20 for removing larger artifacts, followed by a morphological closing with disk of radius 1 for obtaining sharper boundaries. The resulting single object (representing the embryo) was used as a mask for further analysis. Intensities within the masks were then thresholded using a 0.99 cut-off. A morphological opening operation was used to remove small artifacts, using a disk of radius 1. The remaining objects (spots) were extracted from the image using the bwlabel function, and their number and area calculated using the regionprops function. For the GFP quantification, masks for each of the ten WT and nine 7mer embryos within the time course were drawn using Photoshop (Adobe Systems, http://www.adobe.com), from the first frame in the movie. These masks were subsequently used to study only the regions of interests (single embryos) in later frames. For each frame in the time course, the median pixel intensity with each embryo boundary was calculated. Average median intensities and corresponding standard deviations across the ten WT and nine 7mer embryos were finally calculated.
|
10.1371/journal.pgen.1002850 | A Luminal Glycoprotein Drives Dose-Dependent Diameter Expansion of the Drosophila melanogaster Hindgut Tube | An important step in epithelial organ development is size maturation of the organ lumen to attain correct dimensions. Here we show that the regulated expression of Tenectin (Tnc) is critical to shape the Drosophila melanogaster hindgut tube. Tnc is a secreted protein that fills the embryonic hindgut lumen during tube diameter expansion. Inside the lumen, Tnc contributes to detectable O-Glycans and forms a dense striated matrix. Loss of tnc causes a narrow hindgut tube, while Tnc over-expression drives tube dilation in a dose-dependent manner. Cellular analyses show that luminal accumulation of Tnc causes an increase in inner and outer tube diameter, and cell flattening within the tube wall, similar to the effects of a hydrostatic pressure in other systems. When Tnc expression is induced only in cells at one side of the tube wall, Tnc fills the lumen and equally affects all cells at the lumen perimeter, arguing that Tnc acts non-cell-autonomously. Moreover, when Tnc expression is directed to a segment of a tube, its luminal accumulation is restricted to this segment and affects the surrounding cells to promote a corresponding local diameter expansion. These findings suggest that deposition of Tnc into the lumen might contribute to expansion of the lumen volume, and thereby to stretching of the tube wall. Consistent with such an idea, ectopic expression of Tnc in different developing epithelial tubes is sufficient to cause dilation, while epidermal Tnc expression has no effect on morphology. Together, the results show that epithelial tube diameter can be modelled by regulating the levels and pattern of expression of a single luminal glycoprotein.
| Epithelial tubes constitute the functional units of vital organs, and they undergo highly regulated changes in size and shape during development to accommodate the three-dimensional configurations optimal for organ physiology. Through studies of Drosophila melanogaster, we show that epithelial tube diameter can be modelled simply by regulating the levels and pattern of expression of a single glycoprotein. The protein is secreted into the tubular lumen, where it forms a dense matrix and acts in a dose-dependent manner to drive diameter growth. We suggest that deposition of the protein into the lumen promotes local expansion of the lumen volume, and thereby stretching of the surrounding tube wall. Such a mechanism could represent a general means to adjust tube diameter during epithelial organ development.
| Tube growth is a critical phase in the development of many organs and is tightly regulated to produce correct lumen dimensions. Tube-size maturation often occurs after the organ has acquired its basic layout and entails enlargement of the apical surface, and sometimes also expansion of the outer basal surface [1]. At the cellular level, tube growth is mediated by apical membrane growth, changes in cell shape and arrangement and cell proliferation. However, the signals that induce and steer these cellular changes to promote precise lumen size and shape are not fully understood.
It has become evident that lumen size can be influenced by the lumen environment itself. One example is through regulated osmotic pressure and fluid accumulation inside the lumen [2], [3], [4]. The resulting hydrostatic pressure will cause an increase in lumen volume and cellular remodelling within the surrounding epithelium to hold the larger lumen volume [5], [6], [7], [8]. A hydrostatic pressure can however not instruct a differential expansion of different regions of the same lumen, and requires that the epithelium has established a paracellular diffusion barrier [9], [10], [11]. Alternative means to shape organ lumens might involve luminal macromolecules. In the tracheal tubes of Drosophila melanogaster, lumen dilation depends on apical cell secretion [12], [13], but the formation of a uniform tube diameter and regular size of the apical cell domains requires a luminal chitin-based matrix that is transiently present during tube dilation [14], [15]. Moreover, lumen formation in the Drosophila retina requires Eyes Shut (Eys), a glycoprotein that is apically secreted by photoreceptor cells and causes separation of the apical membranes [16]. Similarly, the formation of a lumen during aortic tube formation in mouse requires de-adhesive functions of CD34-sialomucins that contribute to the apical glycocalyx and are thought to promote repulsion of the apical cell surfaces [17]. As the identity of luminal components in most developing organs has remained largely unknown, it is not clear to what extent they can contribute to the regulation of epithelial tube size.
In Drosophila embryos, it has been shown that mucin-type O-glycans are abundant in the lumen of many epithelial organs [18]. Mucin-type O-linked glycosylation is characterized by α-N-acetylgalactosamine (GalNAc) attached to the hydroxyl group of Serine and Threonine. In mucins, large domains containing repeats of Serine, Threonine and Proline (PTS-domains) become highly O-glycosylated and can form gel-like complexes upon binding to water due to the densely attached sugar residues [19], [20]. The Drosophila genome encodes several mucin-like proteins and, interestingly, a portion of these is dynamically expressed in embryonic epithelial organs [21], suggesting that they might be components of developing epithelial organ lumens with possible functions in tube growth. One such protein, Tenectin (Tnc), has indeed been shown to be secreted at the apical surface of the embryonic foregut, hindgut and tracheal tubes at mid-embryogenesis [22].
In this study, we explored a possible function for Tnc in epithelial tube growth, and found that Tnc is critical for diameter expansion of the hindgut. During hindgut growth, Tnc is observed as a dense striated matrix inside the lumen, and its luminal accumulation causes cell shape changes in the surrounding tube wall and tube expansion in a dose-dependent manner. Tnc exhibits limited spread along the tube axis, and can facilitate local dilation according to its spatial expression. The results suggest that Tnc drives volume expansion, and thereby tube dilation, and demonstrates a biological principle whereby the regulated expression of a single gene can steer the degree of lumen dilation along the tube length.
Tnc is a protein of 2788 amino acids and is predicted to include an N-terminal signal peptide and no transmembrane domains. It harbours two extensive PTS-domains that are flanked by cysteine-rich domains with similarity to the von Willebrand factor type C (vWC) domain (Figure 1A). The PTS-domains are present also in predicted orthologs of Tnc in other Drosophila species (Figure 1A), but show poor amino acid identity between the species, although they are of similar lengths and are rich in Serine, Threonine and Proline (Figure 1A). The domain organization of Tnc therefore resembles that of secreted gel-forming mucins, in which large PTS-domains are separated by cysteine-rich von Willebrand factor-like domains that mediate polymer formation. In mucins, the sequences of PTS domains are not conserved between species, supporting that their major function is as a scaffold for O-linked carbohydrates [19].
We used anti-Tnc to detail the distribution of Tnc in embryonic epithelial organs. Tnc-staining was most prominent in the developing embryonic hindgut, (Figure 1B), where it localized to the hindgut lumen (Figure 1D). The tnc transcript is also present in the foregut at stage 14 and in the tracheal dorsal trunks at stage 15 [22]. Consistently, Tnc was detected in the lumen of the foregut (Figure 1E), proventriculus, salivary gland ducts (Figure S1) and the tracheal dorsal trunks (Figure 1F). In addition, we observed Tnc in the lumen of the dorsal vessel at late stage 16 (Figure 1G), which correlated with tnc mRNA expression in clusters of cardioblasts that form the dorsal vessel proper (Figure S1). In all organs analysed, Tnc distributed to the entire lumen and thus behaved as a secreted intraluminal component.
In order to address a possible function for Tnc in epithelial organ development, we generated tnc mutant alleles by imprecise excision of P-element EY16369 inserted between the two transcriptional start sites of tnc [23]. One of the excision alleles, tnc13c, carries a deletion of both tnc transcription start sites. No tnc mRNA (Figure S1) or Tnc protein (Figure 1C) could be detected in embryos homozygous for tnc13c, arguing that tnc13c is a loss of function allele.
The shape of epithelial organ lumens of tnc13c mutant embryos was analysed by staining for Crumbs (Crb) that marks the apical epithelial surface. We found that tnc13c mutants had an unusually narrow hindgut lumen (compare Figure 2C′ and 2F). The same narrow hindgut was observed in embryos that carry tnc13c in trans to Df(3R)BSC655, a chromosomal deletion that lacks the tnc loci, and in embryos homozygous for tnc130a, an excision allele that lacks Tnc in the hindgut (Figure S2). Embryos homozygous for the viable tnc81b allele that corresponds to a precise excision of EY16369 have a hindgut similar to that of the wild type (Figure S2). It therefore appears that loss of tnc causes the narrow hindgut. We also noted that tnc13c mutant embryos exhibit slightly shorter tracheal dorsal trunks at stage 16, compared to the wild type (Figure S3), but could not detect any defects in other epithelial organs. Given the prominent expression and effect of Tnc in the hindgut, we focused on the function of Tnc in this organ.
The embryonic hindgut is an epithelial tube of ∼700 cells surrounded by a thin layer of visceral muscle cells [24], [25]. Hindgut formation commences with internalization of ectodermal cells at the posterior of the embryo [24], [25], [26]. The invagination elongates by mediolateral cell rearrangements to become narrow and J-shaped by stage 13 [24]. At this stage, the hindgut lumen of tnc mutants appeared similar to that of the wild type (compare Figure 2A′ and 2D).
By stage 14, the hindgut is divided into an anterior curved small intestine (Si), a posterior rectum and an in-between large intestine (Li). Li is further partitioned into a dorsal and ventral region, and all hindgut compartments are separated by Crb-expressing border-cells [27] (Figure 2G). Morphogenesis of the hindgut from stage 14 to 16 mainly involves tube elongation and expansion of tube diameter (Figure 2B′ and 2C′). Morphometric analyses showed that Li elongates by nearly 50% and Si almost triples in length from stage 14 to 16 (Figure S4). During the same period, Li and Si expand in diameter by about 40% and 95%, respectively (Figure 2H). The outer tube diameter, as visualized by anti-Dystroglycan (Dg), remained relatively constant during lumen growth (Figure 2A–2C), which is consistent with a concurrent flattening of the tubular epithelium [24]. In tnc mutant embryos, the hindgut lumen appeared slightly narrow at stage 14, when compared to the wild type (Figure 2B′ and 2E), and the difference in diameter increased until stage 16 (Figure 2C′ and 2F). The narrow lumen was not a result of the fixation, since it was also evident when analysing live embryos (Figure S2). At stage 16, Li diameter was similar to, and Si diameter was only 25% wider than that of the wild type at stage 14 (Figure 2H). The length of Li and Si lumens was however not reduced in tnc mutants, and Li was slightly longer in the mutants relative to the wild type (Figure S4). Assuming a circular lumen circumference, the hindgut lumen volume of tnc mutants would be less than 50% of that in the wild type at stage 16. Tnc is therefore critical for diameter expansion of the hindgut, after establishment of the basic organ layout at mid-embryogenesis.
Growth of the wild type hindgut occurs, from stage 13, through increase in cell size, change in cell shape and cell rearrangement [24]. To address the cellular changes that are associated with the narrow hindgut in tnc mutants, we stained for Drosophila epithelial cadherin (DECad) to reveal the apical cell circumferences. By counting cells within different segments of the hindgut epithelium, we could not detect differences in cell number (data not shown), but the cells of the mutant hindgut showed reduced apical circumferences (Figure 3A and 3B). The smaller apical cell domains were particularly evident in Si, where the number of cells covering identical sized areas was 1.8 times higher in tnc mutants than in the wild type (Figure 3C and 3D).
A close examination of the apical cell circumference in Li also revealed small aberrations in cell arrangement in tnc mutants, which were consistent with the narrow tube diameter: In the posterior Li, cells were less stretched along the lumen circumference (illustrated by identically sized brackets in Figure 3E and 3F), although the cells appear equally stretched along the tube length. In the anterior Li, there were more cells along the lumen length (11 versus 9 cells over 16 µm) and fewer cells surrounding the lumen circumference (Figure 3G and 3H), indicating a higher degree of cell intercalation in the this part of the Li.
The use of anti-Dg to visualize the basal hindgut surface revealed that also the outer tube diameter was consistently reduced in stage 16 tnc mutants. Serial z-stacked images spanning the entire hindgut were merged to yield a representative outer diameter (Figure 3I and 3J). Together, these observations argue that Tnc is required to expand the entire tube wall, by a mechanism associated with cell shape changes and slight cell rearrangements, similar to the effects caused by a luminal osmotic pressure [8].
In the above studies, we did not detect differences in the level or localization of Crb and DECad between the wild type and mutant hindgut epithelia. The mutant hindgut also showed normal level and localization of Fasciclin 3 (Fas3), a marker for septate junctions that is present along the apico-lateral cell surface (Figure 3K and 3L). Finally, staining for the transcription factor, Myocyte enhancer factor 2, revealed normal presence of visceral muscle cells surrounding the mutant hindgut epithelium, and expression of Delta and Engrailed in the ventral and dorsal halves of Li was unaffected in the mutants (data not shown). Tnc had therefore no detectable effects on the patterning or epithelial integrity of the hindgut tube.
Our analyses show that Tnc is required for expansion of the hindgut during stages 14 to 16, and that loss of Tnc resulted in a more severe reduction in Si diameter (6 µm) than in Li diameter (3 µm). To understand how these observations correlate with the presence of Tnc in the hindgut lumen, we followed Tnc expression during hindgut development (Figure 4A). From stage 13 to 16, Tnc appeared to gradually accumulate inside the lumen, consistent with a robust expression of tnc mRNA in the hindgut epithelium during stages 13 to 15. Transcript levels were highest in Si, and detection of Tnc as intracellular puncta in Si presumably represents Tnc protein under secretion. By stage 16, Tnc was abundant in the hindgut lumen. However, Tnc was not detected inside the cells and tnc mRNA expression had declined, indicating a cessation of Tnc synthesis and secretion into the lumen.
The stronger expression of tnc in Si versus Li, suggested that the degree of diameter expansion might correlate with the levels of tnc expression. To address this possibility we analysed the effect of Tnc over-expression in the hindgut. The P-element EY16369, inserted between the two transcriptional start sites of tnc, contains binding-sites for the GAL4 transcription factor and promotes transcription of tnc in the presence of GAL4 [23]. We first over-expressed Tnc uniformly in the hindgut epithelium using 69B-GAL4, which drives ubiquitous expression in the ectoderm from late stage 9 [28], [29], and bynGAL4, which drives strong expression in the hindgut starting at stage 7 [30] (Figure S5). Tnc over-expression resulted in an excessively dilated hindgut lumen at stage 16 (Figure 4B–4D). The effect was strongest with bynGAL4, yielding a Li diameter 1.5 times that of the wild type, while over-expression with 69B-GAL4 caused a 1.4 increase in Li diameter (Figure 4K). Si exhibited a ballooned appearance upon over-expression of Tnc, and the size of the Si lumen is therefore presented as an area. We found that both bynGAL4- and 69B-GAL4-driven tnc-expression caused a 1.2 times increase in Si area (Figure 4K). The excessively dilated lumen was accompanied by an increase also in outer tube diameter and by flattening of the tube wall (Figure 4G and 4H), and labelling with DECad showed enlarged apical cell circumferences in the hindgut and fewer cells along the tube, when compared to the wild type situation (Figure 4I and 4J). These effects are opposite to those observed in tnc mutant embryos. There was no significant alteration in hindgut length in embryos that over-express Tnc (Figure S6). It therefore appears that Tnc is sufficient to drive expansion of the hindgut lumen, and that it does so in a dose-dependent manner.
The GAL4-line, drmGAL4, drives expression of UAS-transgenes in the hindgut epithelium [31], but at a higher level in Si compared to Li (Figure S5). We used drmGAL4 to test if the pattern of tnc expression along the hindgut tube would be reflected by a differential degree in tube dilation. Indeed, expression of Tnc driven by drmGAL4 resulted in a 1.44 times Si area and a moderate 1.27 times Li diameter when compared to the wild type (Figure 4E and 4K). Thus, Tnc can cause local tube dilation according to its pattern and levels of expression.
The water-binding capacity of glycans can cause proteins with densely appended O-glycans to assume voluminous structures upon secretion. Given that Tnc has two large PTS-domains, we investigated if Tnc might be O-glycosylated. The first step in mucin-type O-glycosylation is the α-linked attachment of terminal N-acetylgalactosamine to Serine or Threonine to generate the so-called Tn-antigen, onto which Galactose can be added to generate the T-antigen. The Tnc protein is predicted to have a molecular mass of 290 kDa. When protein extracts from embryos and larvae were analysed, Tnc was retained in the stacking gel as molecular species substantially larger than the 250-kDa marker (Figure 5A), indicating that Tnc carries posttranslational modifications. To test if the larger size could be due to attached O-glycans, we treated embryonic extracts with deglycosylation enzymes. N-glycanase did not affect the migration of Tnc on the gel, but incubation with O-glycanase caused slightly faster migration of Tnc (Figure 5B), suggesting that Tnc is an O-glycosylated protein.
We next asked if accumulation of Tnc in the hindgut lumen contributes to detectable O-glycans. Using an antibody that detects the Tn antigen in embryos resulted in strong staining of many organ lumens, including that of the hindgut. Counter-labelling for Crb showed that anti-Tn stained both the luminal surface and the intraluminal compartment of the hindgut. The intraluminal staining was prominent at stages 15 and 16 (Figure 5C and 5D). When anti-Tn was applied to tnc13c mutant embryos, there was a marked reduction in intraluminal Tn-staining, although the apical epithelial surface stained at similar intensity in wild type and mutant embryos (Figure 5F and 5G). No other epithelial organs in tnc mutant embryos showed visible reduction in anti-Tn-staining (data not shown). The Vicia villosa lectin (VVA) also recognizes the Tn-antigen. Like anti-Tnc, labelling with VVA resulted in reduced staining of the hindgut lumen of tnc13c mutants, when compared to the wild type (Figure 5E and 5H). The fluorescence staining obtained with a conjugate of soybean agglutinin, which binds both terminal α- and β-linked N-acetylgalactosamine and galactopyranosyl residues, did not differ between wild type and mutant embryos (data not shown). These results argue that Tnc is an important carrier of mucin-type O-glycans in the hindgut lumen.
It has been recognized that formalin fixation fails to preserve the texture of glycan-rich matrices, and alcohol-based fixatives are required to demonstrate these structures in glycocalixes [32]. In our experiments, using formalin fixation, both anti-Tnc and anti-Tn resulted in a punctate staining of the hindgut lumen. The use of Clark's fixative with ethanol and acetic acid, however, resulted in dense staining for Tnc in the hindgut (Figure 5I). Upon close examination Tnc appeared as a striated structure that fills the entire hindgut lumen (Figure 5J). Also anti-Tn produced a dense and slightly striated staining of the hindgut lumen when using Clark's fixative (Figure 5I′). Thus, Tnc appears to form a glycan-rich matrix inside the expanding hindgut lumen.
The pan-luminal distribution of Tnc, its dose-dependent function and its ability to cause tube wall expansion, suggested that Tnc might drive tube dilation by causing a luminal pressure. To further investigate this possibility, we asked if misexpression of Tnc in other epithelial tubes, like the trachea, salivary glands and malpighian tubules, would be sufficient to cause tube dilation. The tracheal system arises from invagination of 20 ectodermal cell clusters. During stages 12 and 13, the cells rearrange to build six primary branches without further cell division [33]. At stage 14, branch fusion between neighbouring tracheal metameres form the two dorsal trunks (DTs) and, during stage 15, the trunks expand 3- to 5-fold in diameter [34]. We used btlGAL4 to drive expression of UAS-tnc in tracheal cells from stage 11, which is well before the endogenous onset of tracheal Tnc expression at stage 15. Such tracheal Tnc expression caused excessively dilated primary branches from stage 13 with enlarged apical cell circumferences (Figure 6A, 6B, 6D and 6E). At stage 15, the dorsal trunks in these embryos had more cells at the lumen perimeter than those of the corresponding wild type (Figure 6C and 6F), but the number of tracheal cells was comparable to the wild type situation.
Ectopic expression of tnc in salivary glands and malpighian tubules was achieved using 69B-GAL4 and drmGAL4, respectively. Each salivary gland arises from a cluster of ectodermal cells that invaginate and form elongated tubes without further cell division [35]. Ectopic expression of tnc in the salivary glands resulted in dilated glands, and the dilation became increasingly prominent as development proceeded (Figure 6G, 6H, 6J and 6K). The expansion was accompanied by enlarged apical cell circumferences and an increase in the number of cells encircling the lumen (Figure 6I and 6L). The malpighian tubules, arising from evagination of the hindgut anlage, initially have six to ten cells at the lumen circumference (Janning et al., 1986; Skaer and Arias, 1992) and elongate while the cells rearrange into thin tubes with two cells at the circumference (Skaer, 1993). Expression of tnc in the malpighian tubules also resulted in tube dilation (Figure 6M and 6N).
The effects of Tnc on epithelial cell organization in different tubular organs, prompted us to test if Tnc would have a similar influence on epithelial cells in a non-luminal context. We therefore used 69B-GAL4 and enGAL4 to express tnc ubiquitously in the epidermis or in epidermal stripes, respectively. The embryos were stained for Crb and the septate junction protein Coracle (Cora), in order to analyse epidermal morphology and apical cell circumference. The embryos were examined between stages 13 and 16, and we found no anomalies in the appearance of the developing epidermis or in epithelial cell shape of such embryos (Figure 6O–6R, and data not shown), although the embryos expressed Tnc at the apical surface (Figure S7). The ability of Tnc to promote apical surface growth and cell rearrangement when expressed in epithelial tubes, but not in the epidermis, would be consistent with a scenario where Tnc promotes tube dilation by generating an internal luminal pressure.
If Tnc drives tube expansion by adding volume to the lumen, it should promote lumen dilation after its secretion into the lumen and be able to affect cells other than those that produce the protein. In the hindgut, enGAL4 drives expression selectively in the dorsal Li (Figure 7A). When enGAL4 was used to over-express tnc, the Li lumen diameter became enlarged, while the size of the Si lumen was unaffected (Figure 4F and 4K). Although Tnc was over-expressed only in the dorsal Li, both the ventral and dorsal Li exhibited enlarged apical cell circumferences (Figure 7B and 7C). enGAL4 also drives expression of UAS-transgenes in a discrete cell cluster at one side of the salivary gland tube (Figure 7D). Expression of tnc in this cell cluster, driven by enGAL4, resulted in local dilation of the tube (Figure 7E). The cluster of en-expressing cells does not span the lumen perimeter but, nevertheless, all cells at the perimeter showed enlarged apical cell circumferences (Figure 7F). Thus, secretion of Tnc by one side of the tube wall promotes cellular changes also in the transverse side of the tube.
Interestingly, when enGAL4 was used to drive Tnc expression in salivary glands, Tnc was restricted to the dilated part of the tube, where it filled the lumen (Figure 7G). This observation indicates that secreted Tnc forms a local lumen-spanning complex with low mobility, around which dilation occurs. Such behaviour of luminal Tnc would also explain the regional effects of Tnc during normal hindgut tube dilation.
Here, we show that the luminal glycoprotein Tnc promotes diameter expansion of the Drosophila hindgut in a dose-dependent manner. The domain organization of Tnc, its contribution to detectable O-glycans in the hindgut lumen and its ability to form a dense luminal matrix suggest that Tnc has mucin-like characteristics. A possible involvement of mucin-like molecules in tubulogenesis has previously been recognized. The Caenorhabditis elegans let-653 is a secreted protein with a PTS domain of around 90–200 amino acids, depending on the splice variant. In mutants for let-653, the single-celled excretory canals develop massively enlarged lumen by an as yet unknown mechanism [36]. During cyst formation in Madin–Darby Canine Kidney (MDCK), it has been suggested that the initial separation of apical membranes involves de-adhesive properties conferred by large apically localized glycoproteins [37]. Candidate molecules are mucin 1 (MUC1) and the sialomucin Podocalyxin, which localize to the nascent lumens in MDCK cysts and in vivo [38], [39]. Recently, it was indeed shown that Podocalyxin is required to separate apical membranes during initial lumen formation in developing blood vessels. Podocalyxin is membrane-bound, and its negatively charged sialic acids are thought to cause electrostatic repulsion of the apical surfaces [17]. Tnc does however appear to function differently from these mucin-like molecules, since it is not required for lumen formation per se, but drives the subsequent step of tube diameter expansion.
The function of Tnc also differs from that of the chitinous matrix in the tracheal lumen, as the latter is not needed to increase the luminal volume during diameter expansion, but to shape a uniform diameter [14], [15]. A difference in action between the two luminal components is further supported by the slightly shorter tracheal tubes in tnc mutants, while loss of chitin causes too long tracheal tubes. We propose that Tnc-driven tube dilation represents a mechanism for shaping an epithelial tube, where the extent of tube wall extension and lumen volume expansion can be controlled by the intraluminal accumulation of a single protein.
During wing development, Tnc is found basal to the epithelium and is proposed to act as a ligand for PS2 integrin via RGD motifs in the vWC-like domains [23]. It is therefore possibly that luminal Tnc might cause tube wall remodelling by signalling through an apical cognate receptor(s). However, the results do not indicate a signalling function for Tnc: First, over-expression of Tnc in the hindgut causes an increase in tube diameter according to the levels of Tnc expression. Thus, a signaling function of Tnc would imply that Tnc is the limiting factor in the pathway. This is unlikely, since Tnc is abundant and fills the lumen of the wild type hindgut. Second, when Tnc was expressed at one side of the tube wall, all cells at the lumen perimeter were similarly affected. If Tnc signals via an apical receptor, the effects should be higher at the site of its secretion, given its strictly dose-dependent function. Third, the observed lumen-dependent function of Tnc implies that a putative receptor would have to be present in many epithelia in which Tnc is not normally expressed, but yet not ubiquitously, as Tnc had no effect on the epidermis.
Tnc-driven lumen expansion causes an increase in inner and outer tube diameter, associated with epithelial flattening. It is known that luminal volume expansion upon a hydrostatic pressure causes similar effects, for example during inflation of the zebrafish brain ventricle [4], [10], expansion of the mouse blastocyst [2] and in vitro growth of renal cysts [6], [7]. Our results would therefore comply with a mechanism whereby luminal accumulation of Tnc forces an increase in lumen volume and, thereby, expansion of the surrounding tube wall. Since luminal Tnc appears to be a major O-glycan with low mobility in the lumen, an attractive hypothesis is that Tnc forms supra-molecular complexes that cause volume expansion due to hydration of the attached O-glycans. Secretion of Tnc into a confined luminal space would then cause a pressure on the tube wall and lumen dilation. In an attempt to further evaluate if the effect of Tnc requires O-glycosylation of the PTS domains, we have analysed hindgut morphology and the size of Tnc in mutants that lack different glycosyl transferases. However, the results were inconclusive, showing effects on both Tnc levels and secretion (Z.S and A.U. unpublished).
The current study also show that Tnc can steer regional differences in tube diameter expansion along the tube axis, according to its pattern of expression. Such a regional effect of Tnc presumably occurs during normal hindgut development, where the amount of Tnc produced by Si is larger than the amount produced by Li. As a likely consequence, Si undergoes a higher degree of diameter expansion than Li, and it also shows a larger reduction in diameter upon loss of Tnc.
In summary, we have shown that Tnc forms a lumen-spanning complex that drives expansion of the surrounding tube wall. The local and dose-dependent effect of Tnc on tube dilation illustrates that a single protein can model differential lumen diameter along a tube. We suggest a model, were Tnc causes a luminal pressure upon secretion and promotes tube dilation according to its voluminous expansion (Figure 7H). Since the lumen of different epithelial organs have been shown to exhibit dynamic patterns of glycan distribution during development [18], [40], [41], it is possible that glycan-rich luminal components have a broad importance in shaping developing epithelial organs.
The P-element P{EPgy2}EY16369 contains UAS activating elements and is inserted between the two transcriptional start sites of tnc, allowing tnc transcription to be activated by GAL4 [23]. The tnc13c and tnc130a alleles were generated by imprecise excision of P{EPgy2}EY16369. Genomic sequencing revealed that the two transcriptional start sites of tnc are deleted in tnc13c, and that the second start site is deleted in tnc130a. tnc81b is a precise excision allele and is homozygous viable. A detailed analysis of the different tnc alleles generated by excision of P{EPgy2}EY16369 will be described elsewhere (LB and HB, unpublished). Ectopic-expression of tnc was driven by drumstick-GAL4 (drmGAL4), 69B-GAL4, engrailed-GAL4 (enGAL4) and enGAL4, UAS-GFP (all from the Bloomington Stock Centre, Indiana, USA) and BynGAL4 (Iwaki and Judith A. Lengyel, 2002) (from H. Skaer). The deficiency line Df(3R)BSC655 that uncovers tnc was also obtained from Bloomington Stock Centre.
Embryos were fixed with 4% formaldehyde for 20 minutes, dechorionated in methanol and stained according to standard procedures. The primary antibodies were: mouse monoclonal IgM 2A12 (1∶10, Developmental Studies Hybridoma Bank, DSHB), rabbit anti-GFP (1∶500; Molecular Probes, MP), rabbit anti-Tnc (1∶1000) [22], mouse monoclonal IgG1 anti-Crb (1∶10; DSHB), mouse monoclonal IgG2a anti-Fas3 (1∶10; DSHB), mouse monoclonal anti-α-Spectrin (1∶10; DSHB), Rabbit anti-Dg (1∶1000; Deng et al., 2003), rat anti-DECad (1∶20; DSHB), mouse monoclonal IgG anti-Delta (1∶10, DSHB), mouse monoclonal IgG1 anti-Engrailed/Invected (1∶20, DSHB), mouse monoclonal IgM anti-Tn 5F4 (1∶20; [42], mouse monoclonal IgG2b anti-Fas2 (1∶20; DSHB) and guinea pig anti-Cora [1∶2000; 43]. For fluorescent visualisation, secondary antibodies from Molecular Probes or Jackson ImmunoResearch were used at 1∶500. Staining with Texas Red-labelled Vicia villosa lectin (20 µg/ml; EY Laboratories, Inc. California), Alexa 488-conjugated soybean agglutinin (20 µg/ml; Molecular probes) and fluorescein-conjugated Chitin-binding Probe (1∶500, New England Biolabs) was performed according to manufacturers recommendations. For Clark's fixation, embryos were immersed in ethanol and acetic acid (1∶3) for 5 min, devitellinized and post-fixated in ethanol and acetic for 30 min. The embryos were washed several times in ethanol, rehydrated and subjected to normal staining. Visualization of the hindgut lumen in living embryos was achieved by injecting a 10-kDa dextran dye conjugated with Alexa 594 (Molecular Probes) into the hemolymph at the anterior end of the embryos. After 30 minutes, the dye had leaked into the hindgut lumen, allowing visualization of lumen size. Confocal imaging was done using a Bio-Rad Radiance 2000 system and a Leica DM5500B microscope was used to obtain wide field fluorescent images.
Proteins from embryos (50 embryos at stage 16) and larvae (5 third instar larvae) were extracted in RIPA lysis buffer. The samples were loaded on a NuPAGE Novex 4–12% Tris-Bis reducing gradient gel (Invitrogen) and the proteins were blotted onto a PVDF membrane. The blots were stained with antisera against the C-terminus of Tnc (1∶5000) or with mouse anti-α-Tubulin (1∶10000, Sigma) and developed using ECL plus (GE-Amersham). Deglycosylation was performed on protein extracts from 50 stage 16 embryos, using an enzymatic deglycosylation kit for N-linked and simple O-linked glycans (GK80110, GK80115, Prozyme, Hayward, CA) according to manufactures instructions.
Whole-mount in situ hybridization was performed with digoxigenin-labelled RNA anti-sense probes as previously described [14]. The RNA probe used for detection of tnc mRNA is the same as probe B of Mur96B/tnc [21] and corresponds to the second mucin-like domain (PCR-amplified from cDNA with 5′GACAATTCCCGAAATCTCCA and 5′CAGCATCCTGAGGAGACACA). A Nikon eclipse E1000 was used for imaging.
To measure embryonic hindgut lumen dimensions, embryos were labeled with anti-Crb and anti-DECad. Anti-GFP was used to recognize tnc13 homozygous embryos. The embryos were viewed from the dorsal side and z-stacks that spanned the entire hindgut lumen were obtained. Measurements were performed on two-dimensional projections of the z-stacks, using ImageJ. The area of the large intestine (Li) and the small intestine (Si) was determined by tracing the apical epithelial surface based on DECad-staining. The border-cells, highlighted by Crb-staining, were used to demarcate Li and Si. The length of Li and Si was estimated by tracing the left apical surface for Li, and the centre of the lumen for Si. Lumen diameter was derived from area divided by length. All embryos were measured at stage 16, exactly when the four lobes of the midgut had clearly formed and lied parallel to each other (spanning a time window of maximum 15 minutes, as assessed by live imaging at 25°C). The embryos were mounted in Methyl salicylate, which causes a small reduction in embryo size. Since normalization of the measured values to embryo size did not alter the results, the data are presented as true values. Cell numbers per unit length of Li was determined from confocal projections of sections that spanned the upper half of the hindgut tube of embryos labeled with anti-DECad.
|
10.1371/journal.pcbi.1003046 | A Mechanistic Understanding of Allosteric Immune Escape Pathways in the HIV-1 Envelope Glycoprotein | The HIV-1 envelope (Env) spike, which consists of a compact, heterodimeric trimer of the glycoproteins gp120 and gp41, is the target of neutralizing antibodies. However, the high mutation rate of HIV-1 and plasticity of Env facilitates viral evasion from neutralizing antibodies through various mechanisms. Mutations that are distant from the antibody binding site can lead to escape, probably by changing the conformation or dynamics of Env; however, these changes are difficult to identify and define mechanistically. Here we describe a network analysis-based approach to identify potential allosteric immune evasion mechanisms using three known HIV-1 Env gp120 protein structures from two different clades, B and C. First, correlation and principal component analyses of molecular dynamics (MD) simulations identified a high degree of long-distance coupled motions that exist between functionally distant regions within the intrinsic dynamics of the gp120 core, supporting the presence of long-distance communication in the protein. Then, by integrating MD simulations with network theory, we identified the optimal and suboptimal communication pathways and modules within the gp120 core. The results unveil both strain-dependent and -independent characteristics of the communication pathways in gp120. We show that within the context of three structurally homologous gp120 cores, the optimal pathway for communication is sequence sensitive, i.e. a suboptimal pathway in one strain becomes the optimal pathway in another strain. Yet the identification of conserved elements within these communication pathways, termed inter-modular hotspots, could present a new opportunity for immunogen design, as this could be an additional mechanism that HIV-1 uses to shield vulnerable antibody targets in Env that induce neutralizing antibody breadth.
| The Env glycoproteins, gp120 and gp41, are the viral targets of HIV neutralizing antibodies. Accordingly, vaccine studies have focused on eliciting broadly neutralizing antibodies against epitopes in these proteins. Sequence diversity and the conformational flexibility of Env have made vaccine design efforts difficult. It is well documented that mutations distant from defined epitopes can lead to escape from neutralizing antibodies. In such cases, allostery within the Env protein could play a dominant role. In this study, we characterized the dynamical network in gp120 in terms of how spatially distant regions communicate with each other. We introduced an approach based on coupling computer simulations to compare gp120 core structures of three different virus strains from two clades, clade B and C. Our study finds that the long-distance collective motions in the protein are functionally relevant and are conserved across diverse strains of gp120, the communication pathways associated with these motions are sensitive to its sequence. Importantly, we find that gp120 exhibits communication modules (communities) with key residues (hotspots) serving as conduits for communication between different communities, a possible strategy to exploit in future vaccine design efforts.
| The envelope (Env) glycoproteins, gp120 and gp41 are key vaccine components to induce antibody-mediated protection against HIV-1. Recently, monoclonal antibodies that can potently neutralize genetically diverse HIV-1 isolates have been recovered from a subset of HIV-1 infected individuals whose plasma exhibited exceptional neutralizing capacity [1]–[3]. All of these broadly neutralizing antibodies target conserved epitopes in either gp120 or gp41 to prevent viral entry into susceptible target cells. Furthermore, antibodies that bind to a conserved stretch of the gp120 variable loop (V1V2) domain conferred a modest level of protection against HIV-1 acquisition in the RV144 vaccine trial [4]. The humoral arm of the immune system is usually effective against viral infections and often contributes to complete clearance of a pathogen, resulting in the development of long-term immunity. However, in HIV-1, a delay in the induction of potent antibodies until well after the infection [5] has been seen along with viral evasion from neutralizing antibodies in natural infection through various mechanisms [6]–[9].
The extraordinary genetic diversity and the conformational plasticity of HIV-1 Env proteins, gp120 and gp41, present a formidable obstacle for effective immune control and vaccine design [3], [10], [11]. A rapid replication cycle, combined with the high error and recombination rates of the reverse transcriptase [12], [13] provide within-individual genetic diversity, which is then selected for immune evasion [6], [14]–[16]. Based on phylogenetic analysis, global HIV-1 sequences have been generally categorized into four groups (M, N, O and P), representing distinct introductions into humans, which can be further subdivided into clades and circulating recombinant forms [10], [17]–[19]. In addition, the clades tend to circulate in distinct geographical regions. The genetic diversity is driven by immune escape. When mutations occur within the antibody epitope, the mutations can directly reduce the binding affinity of the antibody to its target. In other cases, a mutation proximal to the epitope can change the glycosylation pattern of Env protein, creating a glycan shield that reduces accessibility of the epitope. Finally, escape mutations can occur in regions that are distal to the epitope [20]. These allosteric escape signatures take advantage of the conformational plasticity of Env proteins to evade antibody access to the epitope by changing the conformation or dynamics, and are thus much more difficult to identify and define mechanistically.
In a traditional sense, during allostery, a perturbation such as a mutation or ligand binding at an allosteric site induces a change in binding affinity of a second ligand at a distant active site. Allostery is often associated with a change in the conformation and/or dynamics of the protein [21], [22]. The energy landscape theory has been an effective tool to gain a mechanistic understanding of allostery. This theory states that a protein exists in more than one conformational state of comparable free energy in the absence of an allosteric effector [23], [24]. The binding of the allosteric effector changes the landscape and affects the relative populations of each of these states of the protein. The intrinsic dynamics of the protein cause conformational changes associated with differences in all of its regulatory states [23], [25]–[27]. In various neutralization studies of HIV-1, similar aspects of allostery have been observed. For example, antibodies may bind to known epitope (antibody-binding sites) in Env, but a spatially distant mutation alters the binding affinity of antibody and thereby leads to immune escape [20], [28]–[32].
In HIV-1, allosteric immune escape pathways are driven by the conformational plasticity of the Env subunits, gp120 and gp41, that associate in a trimeric fashion to form spikes in the viral membrane [33]. The Env subunits undergo large conformational changes following gp120 binding to receptors expressed by the target cell [34]–[39]. To enter a cell, gp120 initially binds to CD4, and then to a coreceptor (CCR5 or CXCR4), which are found together primarily on CD4 T cells [33], [38]–[40]. This viral entry process is highly allosteric in nature as gp120 binding to each receptor sequentially invokes a series of conformational changes. Thermodynamic studies indicate that significant changes in entropy are associated with the receptor binding process [35], [41]. The dynamics of the gp120 core by itself capture this inherent allostery [41], [42].
HIV-1 antibody neutralization profiles and associated allosteric immune escape often reflect sequence (or clade) specificity [8], [30], [32], [43]. It has been firmly established that clade B and C viruses exhibit different neutralization sensitivity and resistance patterns, even though only minor differences are seen in the three-dimensional structures of their gp120 proteins [44]. For example, bioinformatics analysis showed that certain residues in the α2 helix region of gp120 were under strong evolutionary pressure to evolve rapidly in clade C viruses, whereas this region was not under strong selective pressure in clade B viruses [30], [32]. Interestingly, only one neutralizing antibody epitope has been mapped exclusively to the α2 helix [45], [46], although residues in this region have contributed to other conformational epitopes [47], [48]. In addition, other studies suggest that there are immune escape mechanism(s) associated with α2 that have not been defined [30], [32]. Therefore, a mechanistic understanding of the conserved and variable features of allosteric communication in gp120, such as that exhibited by α2, could lead to the development of novel immunogen design strategies.
In the past, many different theoretical approaches have been employed to elucidate long-distance communication in proteins to obtain a mechanistic understanding of allostery. The lowest frequency normal modes of a simple elastic network have been successfully used to visualize the conformational changes associated with allostery [49]. However, normal modes are dependent on the structural fold of the protein and lack any sequence-specific effects. While the elastic network model assumes a harmonic approximation in the energy landscape of proteins, a quasi-harmonic analysis (also called Principal Component Analysis) of the fluctuations in the protein during equilibrium molecular dynamics (MD) simulations also identifies the coupled motion of a protein associated with allosteric conformational changes [50]–[54]. In addition, the dynamics of a protein in a MD simulation is sequence specific [55]–[58] and can be utilized to gain a molecular understanding of the conserved and variable features of allosteric communication in a protein family. Ranganathan and coworkers have used evolutionary analysis of proteins to show that a network of correlated residues plays a critical role in allosteric communication within several families of proteins [59], [60]. In spite of these studies, the exact role of sequence evolution in allosteric conformational pathways and the conservation of these pathways remain poorly understood.
In this study, we consider a topology-based analysis that utilizes network theory to capture the long-distance communication in gp120. Network analysis of macromolecules have been used in the past to identify the allosteric signaling pathways and conformational changes within proteins [27], [61]–[66]. These methods assume that energy transfer between local contacts leads to global communication in the macromolecule. The coupling in motion of residues in contact was used as a measure of information transfer between the two residues in the network as the motion in one residue can be used to predict the motion in the other residue [63], [67]. A number of communication pathways exist between an allosteric site and the active site in a protein, and these pathways can be identified using network theory. In the presence of multiple pathways for communication between distant sites, the role of each residue along the signaling pathway is queried. Recently, the modular nature of a network was studied in tRNA/protein complexes involved in translation [63]. This network was very dense within a module, while there were relatively fewer connections between different modules. It was shown that these inter-modular contacts were highly conserved by nature, and they existed in a majority of the suboptimal pathways within the macromolecule. This versatile method was then successfully applied to investigate the bottleneck for allosteric communication within metabolic protein-protein complexes [65], [68] and proteins involved in signaling [69]. However, the conservation of these modules in phylogenetically distant members of a similar structural fold has not been studied so far. As the dynamical network method utilizes information from MD simulations, it provides an ideal framework to investigate the conserved and variable aspects of the allosteric communication pathways in homologous proteins.
Here, we consider three different HIV-1 gp120 sequences. These gp120 proteins are chosen because their structures are highly conserved and are representative of the phylogenetic distances of HIV-1 Env sequences both within and between clades. The structures of each gp120 have been resolved in the CD4-bound conformation. Two structures are from clade B [70], [71] and the other is from clade C [44]. Clade B and C are the most prevalent phylogenetic forms of HIV-1 found in Europe/North America and Southern Africa respectively [17]. Historically, most studies of the structural aspects of Env gp120 to date have focused on clade B isolates, even though clade C alone accounts for more than 50% of global infections [72]–[74]. First, we employed a combination of correlation and principal component analyses to identify the dominant long-distance coupled motion between different functional regions in gp120 that are reminiscent of allosteric regulation. Then, network analysis was utilized to compare the communication pathways in gp120. We found that the shortest path for communication between distal regions is sensitive to the sequence of the individual protein; however, the modular structure of the allosteric network remains highly conserved. The inter-modular junctions (hotspots) form conduits for communication in the gp120 network and are associated with previously known antibody binding sites, and some of these residues are under high immune pressure to evolve rapidly. In addition, these hotspots have the potential to modify the dynamics of gp120 if a spatially proximal structural perturbation is introduced. Thus, we propose that reducing long-distance communication between distant regions of gp120 by appropriate choice of residues at the inter-modular junctions could be a viable strategy for structure-based immunogen design, as this approach would potentially expose vulnerable immune targets and control the conformational plasticity of the immunogen.
We performed long time scale unbiased all-atom molecular dynamics simulations of the gp120 structure from three different HIV-1 strains to characterize the communication network in these proteins and to identify the sensitivity of this network to its sequence. The HXB2 and YU2 strains both belong to clade B, while CAP210 belongs to clade C [75]. The high-resolution structure of the monomeric apo gp120 core (native monomer conformation without the CD4 ligand) remains unknown, although the structure of an HIV-1 gp120 homolog (SIV gp120) has been resolved in the absence of the CD4 receptor [76]. We find that the unliganded SIV structure does not lead to an optimal fit with the cryo-electron microscopy density maps of liganded and unliganded HIV-1 gp120 trimers on the viral membrane (Figure S1). This inconsistency was also noted in a previous study by Liu et al. [77]. Therefore, we carried out simulations of gp120 in the CD4-bound state and investigated the flexibility and dynamics of the protein in the absence of the CD4 receptor in these simulations. Additionally, we did not take into account the V1V2 hypervariable domain, as this region is expected to influence the gp120 core conformation [78] and the conformation of gp120 in the presence of the V1V2 domain remains unknown.
In all three simulations, gp120 protein did not undergo large conformational changes from the initial liganded metastable state during the timescale of our simulations (600 ns). This is consistent with the recent findings that the CD4-bound gp120 conformation is a reasonable approximation for the unliganded HIV-1 gp120 monomer core [78]. It was proposed that when the V1V2 domain and gp41 contacts are removed, the native gp120 core prefers a conformation that is similar to a CD4 bound conformation. A recent SAXS study demonstrated that when V1V2 is introduced onto a single monomer gp120 core, the core adopts a different conformation [41]. Therefore, by simulating a gp120 core without the V1/V2/V3 loops, we are considering a conformation that is representative of a preferred form of the gp120 core.
Initially, we explored the intrinsic dynamics of gp120 for the long-distance strain independent collective motions in the three simulations of gp120 cores. Coupled motions between distant sites would be consistent with the presence of communication pathways within this protein. Since the sequence independent trends observed are similar across all three simulations, we present below the results only for the YU2 strain.
In Figure 1A, the degree of coupled motion between different residues in the YU2 simulation was measured by normalizing the covariance matrix of fluctuations in all Cα-atoms. Residues that move in the same direction in a coordinated fashion are correlated, while those that move in the opposite direction are anticorrelated. The inner domain of gp120 is composed of three subdomains: (i) a five-stranded β-sandwich, (ii) an αβ bundle containing two α-helices and two β-strands, and (iii) the bridging sheet (Figure 1B). The outer domain is also composed of three subdomains: (i) a barrel with seven strands, (ii) a six-stranded barrel along with the α2 helix, and (iii) the variable loop V4 (Figure 1B). Besides local correlations within each subdomain of the protein, the motions in distant regions of the protein are coupled. The β-sandwich, the αβ bundle, and the seven-stranded β-barrel are correlated in motion during the simulation. These three subdomains are anti-correlated in motion with the bridging sheet (that includes the stem of the V1/V2 loop), the six-stranded barrel, and the α2 helix in the outer domain. In turn, the motion of bridging sheet is correlated to the six-stranded barrel and the α2 helix of the outer domain.
We show that these long-distance coupled motions are similar across the three gp120 sequences (Table S1). The patterns in the covariance matrix are highly similar across all three MD simulations (Table S1). Also, the covariance of motion between residues converged during the timescale of each simulation (Text S1 and Figure S2). Most significantly, the motions of regions that undergo a conformational change following CD4 binding are coupled: the bridging sheet and the CD4-binding loop undergo large conformational changes upon CD4 binding, and the motions in these regions are anticorrelated, indicating that they move in a coordinated fashion.
Principal component analysis (PCA) was used to identify the dominant long-distance coupled motions in gp120. Generally, these long-distance coupled motions are associated with functional regulation [50]–[54], [79]. In PCA, quasi-harmonic analysis is carried out on time series trajectories from MD simulations to identify the coupled motions that dominate protein dynamics. Here, we expect the collective motions to be somewhat different in the three gp120s since these motions depend on the individual sequences, but the collective motions associated with CD4 and co-receptor binding should be conserved regardless of the strain because the entry process is invariant. Any differences in dominant collective motions of this protein can be informative of strain-specific features in communication pathways within the protein. We performed PCA on a single trajectory formed by merging 600 ns trajectories from all three gp120 simulations to analyze the differences and similarities in the long-distance coupled motion. Approximately 65% of the fluctuations of gp120 observed in the three simulations are along the first three PCs.
The fluctuations along PC1 (40%) and PC3 (10%) are nearly equivalent in all three simulations (Figure 2C&D), indicating that the motion in these PCs might pertain to the common functional dynamics of the gp120 family. In the PC1, the motion of distant regions in the protein that exhibit large conformational changes upon CD4 binding is coupled. More precisely, the bridging sheet, the CD4 binding loop, and the α1-helix exhibit large fluctuations in PC1 (Figure 2A, red). The CD4 binding loop interacts with CD4 in the CD4-bound complex, while the bridging sheet is formed only in CD4-receptor bound conformations or in gp120 core monomer without the V1 and V2 loops. Similar motions are also observed along PC3 in which the motion of parts of the bridging sheet is coupled to that of the inner domain and a small portion of the CD4 binding loop in the simulation (Figure 2B, red). Consistent with experimental measurements, these regions are highly flexible in all three simulations (Text S1 and Figures S3 to S5). As the motions of distant functional sites in all three sequences of gp120 are coupled in PC1 and PC3, the dominant coupled motions in gp120 are reminiscent of allosteric coupling. Similar long-distance coupled motions have been associated with allosteric conformational changes in many proteins [50]–[54], [79]. However, the role of individual principal components in the conformational changes associated with gp120 binding to CD4 receptor binding are difficult to evaluate, since the high-resolution structure of the monomeric apo gp120 core is unknown, and allostery in the gp120 core may be entropic in nature.
In contrast to PC1 and PC3, the fluctuations of the CAP210 simulation along PC2 are much larger than those of the two B-clade simulations (Figure 3). In PC2, the motion of the bridging sheet is coupled to that of the outer domain spatially close to the α2-helix and to the motion of the loops in the outer domain leading to and returning from the bridging sheet. In addition, the outer domain exhibits moderate motion whereas the inner domain remains rigid. In other words, even though the dominant long-distance coupled motions of gp120 are conserved across different strains, there may be subtle strain-specific differences in these functional motions. The motion in PC2 might be more specific to certain gp120 sequences and could lead to sequence- or clade-specific antibody neutralization strategies. While we and others have established that the α2 helix of clade C gp120 is under higher immune pressure to evolve than the same region in clade B [8], [30], [32], [43], we illustrate here for the first time that sequence diversity in gp120 can also lead to subtle changes in the collective motions of this region in clade C gp120.
Both correlation and PCA of the gp120 simulations established that the long-distance coupled motions in gp120 are conserved to a large extent, and these motions dominate the intrinsic dynamics of gp120. Interestingly, the conserved motions in PC1 and PC3 couple the motions in the functional regions of gp120 – i.e., the CD4-receptor and the coreceptor binding sites. In addition, the PCA of these motions show that there are strain-dependent subtle changes to these long-distance coupled motions. It is often assumed that the global coupled motions in the protein occur due to transfer of energy between local contacts. Allosteric signatures involved in immune escape may also utilize these communication pathways. An important question is whether certain conserved aspects of these communication pathways can be targeted in immunogen design.
We utilized a dynamical network analysis method [63] to identify the routes associated with this signal transmission within gp120. This dynamical network analysis assumes that local coupled motion leads to long-distance coupled motion in the protein. In these networks, a node represents a residue in the protein, while edges connect nodes that are in contact for a majority of the simulation. The correlation of motion between residues in contact is used as a measure for the information transferred between these residues. The larger the edge distance, the lower the communication between the two nodes, since the nodes move more independently of one another during the simulation. In the dynamical network, the optimal pathway for communication is the one that is most coupled between distant sites in a protein. Suboptimal pathways refer to slightly less correlated pathways in the network that connect two regions in a protein. We initially analyzed whether the communication pathways in gp120 are conserved across all three strains to assess whether vaccine design strategies could be developed to reduce communication across these pathways.
Typically, the choices for the beginning and ending points for capturing a communication pathway are ligand binding sites or regions that are regulated in an allosteric fashion. In gp120, we chose one residue (HXB2# 353) in the CD4 binding loop as the beginning position (or source of information flow) due to its critical contribution to the CD4 receptor-binding site. In addition, the CD4-binding loop undergoes a conformational change upon binding to the CD4 receptor. We chose a second residue (HXB2 # 369) located at the C-terminal end of α2 helix as the endpoint, because this residue in the outer domain is the most distant from the CD4 binding loop (in terms of network distance). Also, conformations of outer domain varied less than the inner domain in all three simulations. The network in gp120 was examined to find the optimal (most coupled) and suboptimal paths for communication between the C-terminus of the α2 helix and the CD4-binding loop. In addition, sequence analysis of the α2 helix indicates that clade B and C sequences employ clade-specific mechanisms for immune escape and viral replication [8], [30], [32], [43]. Hence, we measured the most coupled pathway for communication between these regions in the dynamic networks generated for the three different sequences of gp120.
A large number of suboptimal or pre-existing paths exist for communication between these two sites [80]. It was hypothesized in a study by Sethi et al. [63] that a sequence change could convert a suboptimal path to an optimal path in the mutated protein network. In other words, the optimal pathway for communication between two distant sites in the protein can display sensitivity to the protein sequence, and may be subject to selective pressure. We show here that the optimal pathway for communication between three different gp120 sequences does indeed vary (Figure 4 and Table 1). In other words, different sequences of gp120 utilize distinct pathways for communication between the α2 helix and the CD4 binding site. For example, in the case of YU2 communication passes through β9, β10, and β11 whereas it does not pass through these structural elements in HXB2 and CAP210. A number of studies have focused on defining a single optimal pathway for communication between distant sites without also considering the suboptimal pathways [81], [82]. However, we show here that a suboptimal pathway for communication in one sequence can serve as an optimal pathway for communication in another sequence of a homologous protein (Table 2).
The community analysis was carried out on networks built from all three MD simulations of gp120. The modules (communities) are very similar in all three networks (Figures 5, S6 and Table 3). There are seven major communities in each network. The bridging sheet forms one community (Figure 5, green), while the α1 helix forms a second community along with the α5 helix (Figure 5, brown), which is close to the interface with the outer domain. In addition, the five-stranded β-sandwich forms the third major module in the inner domain (Figure 5, magenta). Due to the β1 strand unfolding during the timescale of the simulation, there is some splitting of the β-sandwich into two communities in the YU2 simulation, but to a large extent, the communities are conserved in this region between the different simulations. The outer domain is split into four major communities. In YU2 and HXB2, one community is formed by the C-terminal half of the α2 helix and the six-stranded barrel that interacts with it (Figure 5, blue), while a second community is formed by parts of the V4 loop and the N-terminus half of the α2 helix (Figure 5, red). Parts of the six-stranded barrel form the third community (Figure 5, white), and an additional community is formed by the rest of the seven-stranded barrel (Figure 5, lime). This community structure is highly conserved across all three sequences of gp120 that we have simulated.
Interestingly, in the CAP210 network, the six-stranded barrel forms a community (blue), but the α2 helix is not a part of this community. Instead, the α2 helix forms a separate community of its own (red). In other words, there are subtle changes in the CAP210 modules within this conserved structure. We also observed differences in the network in this region when we compared the modules (Table 3). This is consistent with the PCA from the three simulations discussed above and with the concept that B-clade (such as YU2 and HXB2) and C-clade (such as CAP210) envelopes use different mechanisms for immune escape near the C-terminus of the α2 helix [30], [32].
Conservation of subsections of the network is quantified based on the correlation in network properties (see Methods and [83]) within each subdomain in the protein (Table 3). A correlation value of +1 in the intramodular property across two different networks implies perfect conservation of the module in both networks, while a value of 0 denotes no conservation in the modules. As described above, the subdomains in the inner domain undergo changes during the simulation. Due to this relative instability of inner domain of unliganded gp120 in the CD4-bound conformation, the modules (especially, the bridging sheet) show more variability in network properties across the three different proteins. In addition, the YU2 simulation undergoes further changes during the timescales of our simulations; the β1 strand breaks away from the β-sandwich and the β2 and β3 strands break away from the bridging sheet. As a result, it is difficult to distinguish these inner domain modules with regard to the phylogenetic groups, as it appears that the CAP210 network is similar to the HXB2 network. In depth analysis of such distinctions between networks can be useful in deducing differences in macrophage and non-macrophage tropic Env proteins [66].
The communities are highly conserved in all three gp120 core sequences studied here, and these modules are more conserved in the outer domain of gp120 than the inner domain. Furthermore, the network in the outer domain is more highly conserved between the YU2 and the HXB2 networks as compared to the CAP210 network. This is consistent with the differential splitting of the α2 helix in this region into different communities in YU2 and HXB2 but not in CAP210 (Figure 5). The differences between the CAP210 and the HXB2 networks in these modules are more pronounced in correlations of clustering coefficient and maximum adjacency ratio (Table 3), as these are more sensitive to the overall structure of the network than connectivity and adjacency matrix. Further analysis suggests that the network in the α2 helix is more similar between the two B-clade sequences simulated here than either of the B-clade networks are to the C-clade network.
Nodes in the same community are highly interconnected and can communicate with one another very efficiently through multiple routes. Nodes belonging to different communities have fewer connections between them and could form a conduit for information transfer in the network. A previous study by Sethi et. al. found that some residues occurred in most of the suboptimal paths connecting distant regions and were highly conserved through evolution [63]. If communication through the inter-modular contacts is reduced or eliminated, the network becomes fragmented and the modules become independent of one another [63], [80]. Therefore, we term residues that form these inter-modular edges ‘hotspots’ (listed in Table S2).
Initially, we considered whether these hotspots occurred in regions where broadly neutralizing antibodies bind. There are two sites that are targeted by neutralizing antibodies in the core of gp120 – the CD4 binding site and the bridging sheet. Antibodies that bind to the CD4 binding site compete with binding to the CD4 receptor, thus blocking viral entry into host cells. Thus, antibodies that target this conserved and functionally critical region of gp120 are both broadly neutralizing and highly potent [1], [84]–[89] and the gp120 residues required for CD4 binding and recognition by CD4 binding site antibodies are well defined and often overlapping [1], [85], [90]. However, variation in sequences distant from the CD4 binding site can also influence sensitivity to neutralization at this site, but the exact mechanisms by which these mutations lead to immune escape are unknown [20], [28]–[32]. Antibodies that bind to the highly conserved bridging sheet can block viral entry into host cells by competing with the coreceptor for binding to gp120; however, neutralization by antibodies that target the bridging sheet is limited because this structure is only exposed after gp120 binding to the CD4 receptor [91]. It is likely that these antibodies are present in infected individuals and that they impose strong selective pressure on HIV-1 to remain dependent on CD4 for entry [92].
To investigate hotspots in terms of the CD4 and coreceptor binding sites, we based our analysis on four high-resolution crystal structures of monoclonal antibodies (b12, b13, F105, and VRC01) bound to the CD4 binding site [1], [85], [90] and one structure of monoclonal antibody 17b bound to the bridging sheet in gp120 [92]. We searched for the hotspot residues identified using our approach within the antibody-binding interfaces of gp120 in these structures. We noticed that multiple surface exposed hotspots occur close to the CD4 binding loop and the bridging sheet regions. In each of the five antibody-bound structures, we found between 4 and 8 hotspot residues at the interface of each antibody with gp120 (Table S3). In other words, several hotspot residues from each network are targeted during antibody binding to gp120. Thus, targeting residues that are critical to the integrity of the gp120 network may contribute to the high potency and breadth of these antibodies.
Next, we considered whether these hotspots occur in neutralization signature sites or regions under high selective pressure. A number of sequence-based studies have been performed recently by our group and others to identify genetic signatures associated with immune escape of HIV-1 in a population of infected individuals and to map out the effect of mutations on neutralization sensitivity to monoclonal antibodies [20], [28]–[32]. In particular, we performed a combination of experimental and bioinformatic analyses identified the genetic signatures associated with escape from the monoclonal antibody b12 that binds to the CD4 binding site [20]. In addition to signatures located at the interface of gp120 and b12, genetic signatures that were distant to the interface were also associated with escape from b12 neutralization. Of these distant signatures, only one residue (E268) was located within the gp120 core (i.e. not in a hyper-variable domain) was associated with immune escape from b12 antibody. Furthermore, residue E268 is located approximately 30 angstroms from the b12-gp120 interface. The fact that we also identified E268 as a critical residue on the CAP210 network here argues that a mutation at this position could impact the flow of information within gp120, in addition to decreasing b12 binding. Another study identified residues 456, 458, and 459 as neutralization signatures against the NIH45-46 antibody that also targets the CD4-binding site [93]. These residues occur distant from the antibody-binding site and residues 457 and 459 were also identified as hotspots in the CAP210 network in this study.
As immune escape mechanisms can be context (or clade) specific, studies have focused on identifying regions under high positive selection in a population infected with either the clade B or C virus [31], [48], [94], [95]. We recently performed a comparison of the sequence and structural characteristics of different regions of gp120 in clades B and C, and found that the V4 loop and α2-helix exhibit key clade-specific patterns in variation with antigenic implications [30]. A recent study with clade C viruses also reported that three residues (393, 397 and 413) in the V4 loop were associated with greater neutralization sensitivity [29], and two of these residues are hotspots in the CAP210 (clade C) network. We also reported evidence that five residues within the α2 helix were under high immune pressure to evolve rapidly in the clade C (335, 336, 337, 343, and 350) [30]–[32]. The residues 334, 335, and 349 in α2-helix were also identified as hotspots for communication in the clade C CAP210 network. In clade B, only residues 333 and 335 were identified for HXB2 in the N-terminus of α2-helix and residues 338 and 342 were identified for YU2 sequence.
The genetic signatures identified in the above studies do not correspond to commonly known neutralizing antibody binding sites in gp120, are distantly located to the broadly neutralizing antibody binding sites, and are modulated in an allosteric manner. Interestingly, some of these signatures affect antibody binding to gp120 in the CD4-binding site and/or the coreceptor-binding site presumably without affecting the entry function of gp120. While the exact mechanism(s) utilized by these residues to modulate antibody binding at distant sites remains unknown, we propose that these signature residues could mediate antibody escape by an allosteric mechanism via the gp120 communication network. In other words, if the virus can mediate antibody escape at a highly conserved, functional domain by making a change in a region that is more tolerant to diversity, then Env function is much less likely to be disrupted. Defining the biological contributions of these hotspots will require additional studies, some of which will need to include the full-length, glycosylated gp120-gp41 trimer. Nevertheless, the alteration of communication pathways in an Env immunogen could cause subtle changes in gp120 conformation that may in turn alter epitope exposure. The CD4 binding site in particular may be amenable to such interventions.
Finally, we independently verified that the inter-modular edge hotspot residues occur close to regions that affect the long-distance coupled motion of gp120 by using an alternate methodology based on binding leverage calculations. The binding leverage of a ligand-binding site measures the additional amount of stress introduced into the ligand-bound protein due to coupled motion along the ten lowest energy (or most dominant) normal modes of the apo protein [27], [96]. The binding of an antibody to a particular site on gp120 introduces new ligand-mediated contacts between residues in this region of the protein. The binding leverage measures the coupling between ligand binding and the functional dynamics of the protein. While mutations in the sequence of gp120 can lead to an increase or reduction in the number of contacts near the mutation site, binding leverage only measures the perturbations due to the addition of contacts when a ligand is introduced to the protein. The normal modes are calculated using an elastic network model and the collective motions in the lowest energy normal modes often dominate the motions involved in the regulatory conformational changes of the protein [27], [96]. Regions with high binding leverage could potentially have a greater effect on the motions in these normal modes after a perturbation (induced in the form of a ligand, or more generally, by a change in local structure or sequence) is introduced in this region. This approach has been utilized to accurately predict the allosteric sites in a known set of proteins [27], [96].
Even though the normal modes and binding regions are calculated based on the position of Cα atoms in gp120, a high correspondence between hotspots identified using the network analysis and regions of high binding leverage was observed. Between 70–80% of the regions identified during the binding leverage simulations contain at least one hotspot residue. In all three sequences, multiple hotspot residues tend to occur in the core of ligand binding sites with high-to-moderate binding leverage as compared to sites with low binding leverage (Tables S4, S5, S6 and Figure S7). Hotspot residues located in the CD4 binding site, the bridging sheet, and the β-barrel in the inner domain are conserved across all three structures and also have moderate to high binding leverage. In addition, a number of regions near the interface of the inner and outer domains contain conserved hotspots across all three networks and correspond to regions with moderate binding leverage. With the exception of one residue, the hotspots in all three networks have finite binding leverage and correspond to regions that could potentially affect the collective motions in the lowest normal modes of gp120.
Allosteric signal transmission involves the transfer of energy, leading to the communication of dynamic information between distant regions of the protein. This energy flows anisotropically through the residues in the protein leading to coupled motions in distant regions of the protein [97]. The HIV-1 Env gp120 protein employs an allosteric mechanism that is essential for entry of the virion into a CD4+ T-cell as well as for immune evasion. Here, we utilized a network analysis method based on the local correlation of motion between contacts in the protein [63] to identify the routes associated with this signal transmission within gp120. This dynamical network assumes that local coupled motion leads to global allosteric changes in gp120. We combined this network analysis approach with molecular dynamics simulations to deduce the conserved and variable features of the communication pathways in three known HIV-1 envelope gp120 protein from two different clades, B and C. The present study is also the first to investigate whether (i) the modules in a protein with the same structure change with sequence differences and (ii) changing the community structure in a protein can lead to different allosteric pathways.
First, we established the existence of long-distance coupled motions in gp120 with correlation and principal component analyses. These analyses demonstrated that the coupled motion between distant functional sites on gp120 is conserved and dominates its motions. Furthermore, these motions are reminiscent of allosteric regulation. We then utilized a network theory based approach to study the conserved and variable features of communication in three different gp120 networks, representing two HIV-1 subtypes. We show that many different pathways exist for communication between spatially distant sites in gp120 and a suboptimal pathway in one strain can serve as the optimal pathway in another strain (Tables 1 and 2). While the long distance coupled motions are highly conserved across the three gp120 cores considered, the shortest route for communication between spatially distant sites in gp120 varied with the sequence. Our analysis indicates that HIV-1 gp120 could retain its function and escape from antibody neutralization through mutations that allow it to utilize one of the suboptimal paths if the shortest pathway becomes blocked. This finding is consistent with the observation that genetically distinct, co-circulating HIV-1 variants within an individual commonly use different escape pathways to resist neutralization by the contemporaneous autologous antibody pool [6], [98]. More often than not, these escape pathways appear to protect conformational epitopes. Hence, blocking or altering dominant and suboptimal pathways for communication in gp120 should also be considered in vaccination strategies to increase exposure and/or immunogenicity of conserved epitopes to increase neutralization breadth.
Due to the redundancy of communication pathways in gp120, a natural question that arises is whether any conserved aspects of the network can be utilized for immunogen design. Here we investigated the modules of the network to answer this question. Residues within a module (also called communities) are highly connected, but residues in different modules contain relatively fewer edges between them. Thus, each community in the dynamical network is made up of residues that are in contact with each other and move in a correlated fashion during a MD simulation [63]. In contrast, the inter-modular junctions form conduits for information flow in the network, and by reducing information flow through these edges, one could potentially impose a larger impediment to communication through a protein network. An important question that we address here is whether the modules in dynamical networks are conserved across viral evolution and can therefore be targeted for therapeutic intervention. The community analysis of the network from our study revealed that modules were conserved across the three different gp120 strains. However, there were subtle changes in these communities, for example in the α2 helix region, that could lead to different allosteric immune evasion mechanisms or immunogenic properties in envelopes from phylogenetically distinct groups or clades. This is consistent with our previous studies demonstrating distinct mutational patterns in and around α2 between clade B and C gp120 [30]. These findings also support that vaccines designed for certain populations should include strategies that consider the dominant circulating clade or recombinant form.
Given the highly conserved modular nature of the gp120 network, the interface of these communities pointed to the presence of hotspots for long-distance communication in the protein. These hotspots exist at the junctions between modules in the network, and the communication between residues in two different modules has to flow through relatively fewer inter-modular edges. In this study, we found that a number of surface-exposed hotspots occur close to the functionally important CD4 binding loop and the bridging sheet region. This could be one of the reasons why some antibodies that target the CD4 binding site region exhibit broadly neutralizing character. Importantly, we show that these hotspots occur at residues that are part of well-defined epitope in gp120, as well as in sites distal to these epitopes that have been associated with neutralization resistance or immune escape. Furthermore, a number of hotspots occur along the α2 helix, and these residues were found previously to be under high selective pressure in a clade-specific manner in our earlier studies [8], [30], [32], [43]. Importantly, we verified the occurrence of these hotspots using an independent approach. This second analysis demonstrated that the perturbations near hotspots could potentially influence long-distance coupled motions (lowest energy normal modes) that dominate the intrinsic dynamics of gp120 core. Even though the dynamical network method as done in this study is more computationally intensive than the binding leverage-based method, the former is better suited to study sequence-specific allosteric communication mechanism compared to the Cα-atom based latter approach. Our studies suggest that even in the presence of multiple pathways for communication between distant regions in gp120, the conduits for information flow (hotspots) that we have defined could be exploited in new immunogen design strategies.
Finally, these communication hotspots could potentially be exploited to interfere with the flow of information across the allosteric network in gp120. The HIV-1 envelope has evolved multiple mechanisms to maintain an inherent level of neutralization resistance by protecting its most vulnerable and well conserved targets. The novel and rational immunogen design approach that we introduce here could be used in envelope-based vaccine strategies to focus the immune response on critical hotspot residues, or mutate those residues directly to expose conserved epitopes in gp120 (i.e. the CD4 binding site), in an effort to induce antibodies with neutralization breadth. Furthermore, this approach could inhibit long-distance immune escape pathways within gp120 should breakthrough infection occur by inducing antibodies against regions that contain these hotpots. Here, one can target vaccine-induced antibodies to hotspots to minimize the potential for immune escape via long-distance allosteric communication following a breakthrough infection. Many of the hotspots lie along the suboptimal paths that connect distant regions in gp120. If information flow through these hotspots could either be reduced or removed completely perhaps by mutating them, the allosteric network would become fragmented and the modules would function independently of one another. This could also lead to changes in conformation that expose otherwise hidden epitopes and increase neutralization breadth. Thus, this network-based approach could reduce the capacity of the HIV-1 envelope to shield its vulnerable neutralization targets, producing more effective immunogens.
The missing regions of the structures for the YU2, HXB2, and CAP210 sequences (PDB accession numbers 1G9M [71], 1RZK [70], and 3LQA [44]) lacking the V1–V3 loops were modeled using the MODELLER program [99]. The core of gp120 was similar in all these structures. It should be noted that 3LQA is the only high-resolution structure of a clade C gp120 sequence, while multiple clade B gp120 structures exist. Multiple templates were used because it has been shown that this creates a high-quality homology model. During modeling, disulfide constraints were added for the conserved cysteines present in all gp120 sequences. All sequence alignments used for modeling templates were based on sequences in the HIV-1 database (www.hiv.lanl.gov).
The starting conformations for the long timescale all-atom MD simulations were modeled using MODELLER [99] as described above. The protein was solvated in TIP3P water molecules [100] and neutralized in 150 mM NaCl salt. The MD simulations of the solvated proteins were performed using NAMD2 [101] with the CHARMM27 force field [102]. The protein was initially minimized and then heated to 298K with constraints added during these steps similar to the protocol in [103]. All simulations were performed with periodic boundary conditions using the NPT ensemble with pressure set to 1 atmosphere and temperature set to 298K. The pressure and temperature were maintained using the Langevin piston and the Langevin theromostat respectively. Electrostatics were calculated with the particle mesh Ewald method [104]. The van der Waals interactions were calculated using a switching distance of 10 Angstroms and a cutoff of 12 Angstroms. All the production runs were performed with 2 fs time step using the RATTLE [105] and SETTLE [106] algorithms. The proteins were equilibrated for 10 ns, and the initial burst in RMSD converged within this period. The YU2, HXB2, and CAP210 simulations were performed for a further 600 ns each. The number of atoms in each system was approximately 50,000 atoms. The coordinates were saved once every 1 ps in each simulation.
Correlations between all of the residues in gp120 were analyzed for the 600 ns production run using the normalized covariance:where denotes the covariance in motion of the Cα-atoms of residue i and j; while . The correlation matrix is also called the dynamic cross correlation matrix. The value of Cij is between the values of −1 and 1. If Cij = 1, then the residues are moving in a correlated fashion (same direction) during the simulation, while Cij = −1 implies that the residues are moving in an anticorrelation fashion (or in opposite directions). Residues that move independently of one another have a correlation value close to zero. However if residues move in a correlated fashion in perpendicular directions, their correlation value will also be close to zero. The frames are saved at an interval of every 1 ps, and a total of 600,000 frames were analyzed for the correlation matrices of each simulation.
To investigate the collective behavior within the complex, a standard principal component analysis (PCA) of the motions of the Cα atoms during the equilibration was performed as implemented in the program CARMA [107]. The unnormalized covariance matrix, Cov defined above, was diagonalized during PCA. The largest eigenvalues and their accompanying eigenvectors, capture the largest fraction of the observed variance in the motion. The contribution of each eigenvector to the observed motion is obtained using the projection matrix. On projecting the data from principal component i onto the Cartesian coordinates, the RMSD of each residue was calculated due to the ith principal component. The RMSD per residue plots give an estimate of regions that are highly coupled due to the ith principal component.
A network is defined as a set of nodes with connecting edges. Each amino acid residue in the protein is represented by a node in the network. Edges connect pairs of nodes if the corresponding residues are in contact, and 2 nonconsecutive monomers are said to be in contact if any heavy atoms (non-hydrogen) from the 2 monomers are within 6.5 Å of each other for at least 75% of the frames analyzed. The edges are weighted by the correlation in motion between the residues: [63]. The (anti-) correlation in motion is used as a measure for information transfer between the two residues in contact.
The length of a path Dij between distant nodes i and j is the sum of the edge weights between the consecutive nodes (k,l) along the path: . The shortest distance Dij between all pairs of nodes in the network is found by using the Floyd–Warshall algorithm. The betweenness of an edge is the number of shortest paths that cross that edge.
Although the shortest path is the most dominant mode of communication between the nodes, the number of paths within a certain limit of the shortest distance is a measure of the path degeneracy in the network. All suboptimal paths for communication between the active site and the identity elements are determined in addition to the shortest path. The tolerance value used for any alternate path to be included in the suboptimal path was , which is close to the average protein edge weight.
The network contains modules or communities of nodes that are more densely interconnected to each other than to other nodes in the network. The community structure is identified by using the Girvan–Newman algorithm [108]. In this algorithm, the shortest paths between all pairs of nodes in the network are calculated. The betweenness of an edge in the network is defined as the number of shortest paths that pass through it. The Girvan-Newman algorithm uses a top down approach to iteratively remove the edge with the highest betweenness and recalculate the betweenness of all remaining edges until none of the edges remain. The optimum community structure is found by maximizing the modularity value Q, which is a measure of difference in probability of intra- and intercommunity edges. As the algorithm divides the network into increasingly smaller communities, the modularity score is measured for each community division, and the maximum value corresponds to the optimal community distribution of the network. More recently, a number of algorithms have been developed that explore different strategies for dividing a network into community structures, but they are more complex and provide only subtle differences in the community architecture of these proteins. The variation of cutoffs used to define contacts was investigated by Sethi, et al., 2009 and showed that changes in the parameters (75% of frames and 4.5 Angstroms cutoff between any pair of heavy atoms in residues) defining the network contacts led to minor changes in the community distribution of the network.
To compare the sensitivity of different subdomains in the networks to the sequence of the protein, we calculated module preservation statistics defined in [83]. Briefly, the module preservation statistics measure the preservation of connectivity and weights of these connectivities between nodes within the modules in different networks. These calculations were performed only over the core of the protein (any position that was gapped in any of the structures were not considered to be part of the core), as the number of nodes for each subdomain has to be constant in the different networks. The HXB2 network was considered as the reference network for this analysis. However, the trends in Table 3 are independent of the reference network. There are four network measures that are considered in this analysis:
The calculation of binding leverage initially involves the identification of potential ligand binding sites in the protein as detailed in [27]. The ligand binding sites were identified using Monte Carlo docking simulations to the protein represented by its Cα-atoms. The probe contained 6 atoms in these simulations and the bond angles in the probe were allowed to vary between 90 and 180 degrees. The probe and protein interacted via a square well potential which was attractive for Cα-Cα distances between 5.5 and 8 Å. Distances shorter than 4.5 Å were forbidden. The probe binding sites were identified in 1000 docking simulations each containing 10000 Monte Carlo steps. Binding leverage is defined as the amount of distortion in the probe location due to the motions in the lowest ten normal modes. The normal modes were calculated using the anisotropic network model made from the Cα atoms in the protein [42]. Springs were introduced between any two residues in contact within the protein (default contact distance cutoff of 15 Å). The probe introduced additional contacts in the protein near the probe binding site. A spring was placed between all residue pairs in a probe location whose interconnecting lines pass through the ligand. The binding leverage was calculated as the change in potential energy due to the distortion in these springs induced by the first ten normal modes in the protein.
|
10.1371/journal.pcbi.1000536 | Antigenic Diversity, Transmission Mechanisms, and the Evolution of Pathogens | Pathogens have evolved diverse strategies to maximize their transmission fitness. Here we investigate these strategies for directly transmitted pathogens using mathematical models of disease pathogenesis and transmission, modeling fitness as a function of within- and between-host pathogen dynamics. The within-host model includes realistic constraints on pathogen replication via resource depletion and cross-immunity between pathogen strains. We find three distinct types of infection emerge as maxima in the fitness landscape, each characterized by particular within-host dynamics, host population contact network structure, and transmission mode. These three infection types are associated with distinct non-overlapping ranges of levels of antigenic diversity, and well-defined patterns of within-host dynamics and between-host transmissibility. Fitness, quantified by the basic reproduction number, also falls within distinct ranges for each infection type. Every type is optimal for certain contact structures over a range of contact rates. Sexually transmitted infections and childhood diseases are identified as exemplar types for low and high contact rates, respectively. This work generates a plausible mechanistic hypothesis for the observed tradeoff between pathogen transmissibility and antigenic diversity, and shows how different classes of pathogens arise evolutionarily as fitness optima for different contact network structures and host contact rates.
| Infectious diseases vary widely in how they affect those who get infected and how they are transmitted. As an example, the duration of a single infection can range from days to years, while transmission can occur via the respiratory route, water or sexual contact. Measles and HIV are contrasting examples—both are caused by RNA viruses, but one is a genetically diverse, lethal sexually transmitted infection (STI) while the other is a relatively mild respiratory childhood disease with low antigenic diversity. We investigate why the most transmissible respiratory diseases such as measles and rubella are antigenically static, meaning immunity is lifelong, while other diseases—such as influenza, or the sexually transmitted diseases—seem to trade transmissibility for the ability to generate multiple diverse strains so as to evade host immunity. We use mathematical models of disease progression and evolution within the infected host coupled with models of transmission between hosts to explore how transmission modes, host contact rates and network structure determine antigenic diversity, infectiousness and duration of infection. In doing so, we classify infections into three types—measles-like (high transmissibility, but antigenically static), flu-like (lower transmissibility, but more antigenically diverse), and STI-like (very antigenically diverse, long lived infection, but low overall transmissibility).
| There are two major principles by which pathogens avoid their elimination: escaping the host immune response via antigenic variation or immune evasion, or transmission to a new immunologically naive host. Directly transmitted pathogens which cause chronic diseases, such as many sexually transmitted infections (STIs), tend to rely more on the former, while many acute infections, for instance measles, rely more on high transmissibility. Indeed pathogens such as measles show very little antigenic diversity, with immune responses being strongly cross-reactive between strains. There are then those pathogens which have intermediate levels of both immune escape and transmissibility — such as influenza, rhinovirus and RSV (here referred to as FLIs — flu-like infections).
The evolutionary success of directly transmitted pathogens can also be seen to depend on the nature, frequency and structure of contacts between hosts. Infections transmitted to a small number of hosts (per time unit and infected individual) via intense contact (e.g., via fluids) are usually caused by pathogens of high antigenic diversity and long duration of infection, while those transmitted via casual contact (e.g., via aerosol) with a large number of hosts may typically have lower diversity and much shorter durations of infection. While many of the evolutionary constraints are different [1],[2], vector-borne infections typically fall in the former of these two classes [3],[4]. The relationship between so-called infection and transmission modes with respect to substitution rates of RNA viruses has been investigated in [5].
It is straightforward to explain the long duration of infection and consequent antigenic diversity of sexually transmitted or blood-borne infections: the frequency of relevant contacts between hosts is low, meaning infection needs to be extended to ensure the reproduction number (the number of secondary cases per primary case [6]) exceeds one. However, many childhood diseases (ChDs) — at least those caused by RNA viruses — would also seem to have the genetic potential to prolong their survival within one host via by generating antigenic variants. The fact this is not observed is much harder to explain. At its root are the tradeoffs between maximizing between-host transmissibility and within-host duration of infection, and these are what we focus on exploring in this paper.
The molecular genetic basis of transmissibility is still poorly understood for most pathogens. However, all other things being equal, the level of pathogen shedding by a host (whatever route is relevant) must be positively correlated with infectiousness. A first-pass analysis might therefore postulate that overall transmissibility (as quantified by the basic reproduction number, ) might be proportional to the total number of pathogen copies produced during an infection — the cumulative pathogen load. Past work using a simple model of the interaction between a replicating pathogens and adaptive host immune responses examine what rate of antigenic diversification within the host would maximize cumulative pathogen load [7]. This showed that the combination of resource-induced (whether nutrients or target cells) limits on peak pathogen replication rates and an ever more competent immune response mean that the optimal strategy is not to diversify as rapidly as possible, but instead to adopt an intermediate rate of diversification. In addition, there are further tradeoffs associated with high mutation rates — the ultimate being the error catastrophe associated with error rates in genome replication which exceed those seen in RNA viruses [8]–[11].
However, the assumption that transmission fitness (as quantified by ) is linearly proportion to total pathogen load is clearly naïve. The instantaneous hazard of infection for a susceptible host in contact with an infected host at a point in time may indeed be linearly related to pathogen load at that time, but going from this assumption to a calculation of the overall reproduction number is far more complex than simply calculating the area under the pathogen load curve. Integrating a hazard over the finite time of contact gives an exponential dependence between the probability of infection and pathogen load , i.e., . Such an expression fits experimental data [12] on the relationship between HIV viral load and transmission rates well (cf. Fig. 1). This means the parameter represents a pathogen load threshold below which the probability of infection declines rapidly, and above which it rapidly saturates to some maximal value. Hence can be thought of as the characteristic pathogen load required for transmission — though it is not a true minimum infectious dose — there is a finite probability of infection for , but that probability decays exponentially fast with reducing .
A key insight (and assumption) of the work presented here is that while we might expect pathogens to be able to evolve to reduce (or increase) , there are fundamental physical constraints imposed by transmission routes on the minimum value of attainable. An STI might have a minimum value of approaching a single pathogen particle (e.g. virion) but, for respiratory infections, the much lower proportion of all pathogen particles emitted from a host, which have any chance of contacting epithelial tissues of a susceptible host (even conditioning on a susceptible host being in the near vicinity of the infected individual), necessarily means that must be orders of magnitude larger for such pathogens.
We will show that there is a critical value of above and below which two different sets of pathogen types are evolutionarily favored (in terms of having maximal ). Within each set, the particular type which has maximal will be seen to depend on the local structure of the contact network between hosts.
Our approach is to construct a model of within-host pathogen dynamics which incorporates adaptive host immunity and antigenic diversification. The key output from this model is how pathogen load varies through time during an infection. We then calculate the basic reproduction number, , for that infection assuming a particular local contact network structure and frequency of contacts.
The within-host model developed here is an extension of a model studied earlier by one of us [7]. Our work builds on a range of past work examining the tradeoffs between within-host replication and persistence, antigenic variation and between-host transmission success, initiated by [13], and followed by [14],[15], which first include immune response and explore cross-immunity. More recent studies, to mention a few, investigate pathogen evolution under limited resources [16], include virulence [17], consider the immunological response in more detail [18], examine the impact of between-host contact structure on pathogen evolution [19],[20], and explore host-pathogen co-evolution [21],[22].
We use as our fitness measure for determining evolutionarily optimal phenotypic strategies. We do not explicitly model competition between pathogen strains with different phenotypes co-circulating in a host population, since for infinite populations, has been shown to be the fitness measure which determines the outcome of such competition [23]. This holds even when comparing strains with different rates of antigenic diversification — if the strain with lower induces no long-lived immunity in the host (giving SIS dynamics) and the higher strain induces life-long immunity, (giving SIR dynamics) the higher strain will still always (eventually) outcompete the lower strain. There are limitations to the use of as a fitness measure (further considered in the Discussion) — for instance, in situations where strains interact asymmetrically via cross-immunity, or when populations are small and stochastic extinction is significant. In addition, while we take account of local (egocentric) network structure in defining in our analysis, large-scale network structure might also affect the determinants of evolutionary fitness. However, we feel these limitations are outweighed for an initial analysis by the analytical and computational tractability afforded by use of a relatively simple transmission measure, and the consequent ability not to rely on unintuitive large-scale simulations.
We do not explicitly consider how a pathogen could evolve its biological characteristics to maximize transmission fitness (i.e. the evolutionary trajectory a pathogen would take through parameter space). There are undoubtedly many constraints on the possible paths which pathogens can take [24], however, and exploring how these affect, for instance, pathogen adaptation to a new host species, will be an important topic for future work.
The multi-strain model used extends past work [7] by adding cross-immunity between strains (see Methods for details). The infection within one host starts with a single strain, with further strains arising through random mutation. All strains compete for resources (e.g. target cells) to replicate. Immune responses to strains are assumed to be predominantly strain-specific, albeit with a degree of cross-immunity, the strength of which decays with the genetic distance between strains. Pathogen replication depletes resource, and independently from immunity, limits to pathogen growth are set by the replenishment rate of resource. This quantity only determines the short-term dynamics of the model whereas immunity is also responsible for the long-term behavior.
The dynamics of the model is characterized by an initial period of exponential growth of the pathogen load, which eventually slows due to immune responses and resource limitations. One observes a latency period and an initial peak. Pathogen load then declines exponentially. If the trough load of a pathogen strain drops below a threshold level we assume the pathogen is eliminated from the host (to avoid persistence at unrealistically low, fractional, loads). However if a novel strain emerges before the seed strain goes extinct, pathogen load can recover, so long as there is sufficient resource available and cross-immunity is not too strong — leading to a second, albeit lower peak in pathogen load. Further peaks in pathogen load can occur via the same mechanism. The rate at which new strains arise is the most important determinant of the number of pathogen load peaks seen and thus the overall duration of infection. Less intuitively, this rate also determines the size of the initial peak (discussed below).
Since mutation is modeled stochastically, we average over multiple realizations (e.g. Fig. 2A,B) of the model to calculate an average pathogen load distribution over time (Fig. 2C). The average distribution consists of a first latency period, a large initial peak, a second latency period and possibly an irregular oscillating part of low pathogen load. The point at which the viral load vanishes determines the duration of infection.
We systematically calculate average pathogen load curves from the within-host model for wide ranges of two biological parameters: the antigenic mutation rate (i.e., the rate of mutations which lead to antigenically novel strains) and the pathogen replication rate . These two parameters span what we call pathogen parameter space, in which evolutionarily favored pathogens are represented by points that are associated with maximal fitness values.
From the discussion in the introduction, we can immediately identify the cumulative pathogen load and duration of infection as epidemiologically relevant quantities. Fig. 3A,B show these as a function of the parameters and . In addition, Fig. 3C shows a quantity — interpolating between the two former — evaluated only for the initial period of the infection (utilizing the expression relevant for transmission, i.e., , quantified at the initial peak of the pathogen load ). We will see below that all the surfaces shown in Fig. 3A–C crudely represent fitness surfaces associated with three distinct pathogen types. The plots in Fig. 2 show the corresponding within-host dynamics for the different pathogen types.
The within-host dynamics generate a tradeoff between initial peak pathogen load and antigenic diversity: high initial peak load corresponds to low diversity and vice-versa (see Methods for more details). This tradeoff has implications for transmission, giving an enhanced spread of pathogens of low antigenic diversity during the initial peak of pathogen load. This effect explains the emergence of (ChD-like) infections with short durations of infection within our model framework (Fig. 3C vs. 3F). Long durations of infections (Fig. 3B) are also obtained, as expected, for pathogens with greater antigenic variation.
To calculate the reproduction number (i.e., the pathogen fitness), we model a dynamic contact network in the neighborhood of one initially infected host. The profiles of pathogen load over time obtained from the within-host model then determine the infectiousness of the infected host to its neighbors. (We utilize the mean-load profiles averaged over individual hosts.) Epidemiological dynamics are determined by 4 parameters. Two of these relate to properties of the transmission route: the infectiousness parameter and the contact rate between hosts . Together these define a two-dimensional parameter space we term transmission space. The other two define properties of the contact network between hosts: the replacement rate of neighbors and the cliquishness/clustering of the network (i.e., the proportion of pairs of contacts of a host who are also contacts of each other). These two parameters define what we term contact space.
We build a model (cf. Methods) incorporating these 4 parameters (plus implicitly the within-host pathogen space parameters) to calculate the number of first generation infections from an infected individual in an entirely susceptible population.
Varying the 4 parameters of transmission and contact space, we obtain three different classes of fitness landscapes over pathogen space — as represented by Fig. 3D–F. The maxima of each landscape differ with respect to their antigenic mutation rate (and hence the resulting level of antigenic diversity) and within-host pathogen replication rate. By changing the contact rate and keeping the other transmission as well as the contact space parameters fixed, one can shift between these classes. In general (as shown further below), low, intermediate, and high contact rates induce moderate, high, and low antigenic diversity, respectively, as evolutionarily favored outcomes (represented by the locations of the fitness maximum in Fig. 3D–F).
There are clear similarities between the three classes of fitness landscapes (Fig. 3D–F) and the different within-host infection characteristics plotted in Fig. 3A–C. Low contact rates induce landscapes that resemble the cumulative pathogen load, intermediate contact rates give landscapes resembling the the duration of infection surface, and high contact rates map onto the surface of Fig. 3C which characterizes the relative importance of the initial peak in the pathogen load profile. We classify the optima of these 3 classes of fitness landscape infection types, labeling them A, B, and C, respectively.
Varying the infectiousness parameter can also move the fitness landscape between these types — as (the STI limit; i.e., , ), the fitness landscape becomes more similar to the duration of infection surface (Fig 3B), while for (the FLIs limit; i.e., , ), it becomes more similar to the cumulative pathogen load surface (Fig. 3A); cf. (7) and (6). It is important to note that both of these limits involve substantial antigenic diversity — where transmission fitness is dominated by cumulative pathogen load (infection type A), while moderate antigenic diversity is seen, and when infection duration dominates fitness (infection type B), high antigenic diversity is selected for. Neither maps on to the special case of infection type C (Fig. 3F) in which optimal transmission fitness is achieved by a set of parameters giving very low antigenic diversity (in essence a single strain). For low antigenic diversity to be optimal, it is necessary for fitness to be dominated by the peak pathogen load achieved during primary infection (i.e., the first peak of pathogen load).
Varying the transmission and contact space parameters more systematically, one can map out the regions of parameter space for which particular infection types are optimal (Fig. 4). This shows how the emergence of pathogens of different types depends on the properties of the between-host contact network. Pathogens with low antigenic diversity (and thus short infectious periods) are favored by high network cliquishness (i.e., when an individual's contacts are contacts of each other — as is the case for household and school contacts), and the rate of turnover of network neighbors is low (again the case for household and school contacts).
So far we have assumed only the pathogen space parameters ( and ) can change during pathogen evolution. Now we examine making the infectiousness threshold a parameter which can evolve under selection — albeit with constraints on its lower bound set by the transmission route of the pathogen concerned. Fig. 5 shows the results as a function of contact rate for two different choices of contact space parameters and lower bounds on the infectiousness threshold parameter, suitable for a respiratory pathogen and an STI respectively. Reproduction numbers (Fig. 5B) lie in the expected range, and the three regimes of antigenic diversity corresponding to the types A/B/C) can be found in the evolutionarily optimal values of (Fig. 5A,C). Note that only type A and type C diversity is seen for the respiratory pathogen parameter choices, while only type B is seen for the STI parameter set. Indeed for the STI parameter set, the evolutionary stable state is independent of the contact rate, and is determined by evolving to its minimum value.
As expected, the evolutionary optimal value of the infectiousness parameter (Fig. 5B) is always close to the minimal attainable value, except in the type C pathogen regime (where cliquishness is necessary; cf. Fig. 4). The reason for the deviation from the minimum value lies in a reduced local network saturation, which is characteristic for type C: concentrating infectiousness over the shortest possible time period (and consequently lengthening the latent period) shortens the overlap between generations of infections, and this reduces the chance that the secondary cases of an index case infect remaining susceptible contacts of the index (before the index can infect them). The effect (which yields an enlarged susceptible number in (6)) is minor, however — the difference in between the optimal value of and the minimum bound set for a pathogen type is typically very small.
The evolutionarily optimal replication rate is always low for STI-like contact parameters (giving type B pathogens), reflecting the need for long-lived infections, but shows greater variability for respiratory pathogen parameter regimes (Fig. 5D) — being high in the type A regime, but low for type C. The latter result reflects a tradeoff between height of the initial peak in pathogen load and length of the latent period — longer latency, as explained above, can increase the number of direct infections caused by an index case by reducing the overlap between generations of infection. Only higher (minimal) infectiousness values — realistic for ChDs utilizing the respiratory transmission route — increase the optimal replication rate for type C infections (cf. Text S1, Sect. B.2). Note that these results are consistent with a recently formulated hypothesis on tradeoffs between reproductive rate and antigenic mutability [25], proposing a reciprocal relationship between these two (pathogen space) parameters in real-world infections.
Re-examining Fig. 4, it is clear that type A infections (green areas) only exist when the infectiousness parameter exceeds some minimum value (indicated on the graphs in Fig. 4 with an arrow). In the absence of constraints, selection for maximal transmissibility will clearly cause to evolve towards 0. Hence the effect of constraints on imposing a lower bound on has a critical effect on what range of pathogen types are expected. We define the value of the lower bound on infectiousness below which infection type A is no longer found the critical infectiousness threshold. Evolutionary dynamics show a phase transition at this point, as can be seen in Fig. 6 which maps the areas of contact parameter space for which different infection types are seen for choices of the lower bound on just above and below the critical point .
As discussed already, the transmission route is likely to be the most important determinant of the lower bound on , with STIs and other non-airborne pathogens, including those requiring a vector, being likely to achieve a much lower value of than respiratory pathogens (as assumed in Fig. 5). This is clear if one views as quantifying how much shed pathogen is typically wasted to achieve a single infectious contact. We therefore speculate that the critical infectiousness threshold may have a significant biological effect, with STIs — and also vector-borne infections — being within the sub-critical domain (Fig. 6B), and with ChDs and FLIs — not necessarily relying on a respiratory transmission route — being super-critical (Fig. 6A). Within the super-critical regime, the presence of low-diversity ChD-like type C infections depends less on the precise value of the critical infectiousness threshold and more on the contact rate and contact parameters. Infections of type C occur in contact networks with high cliquishness and low replacement rates — but not in the opposite case (cf. presence of blue areas in Figs. 4 and 5A). Vector-borne infections (representing contact networks of large neighborhood sizes or high replacement rates , and cliquishness not playing a role) are thus excluded to be type C. At first sight they seem to be type A, because of large reproduction numbers. Large , however, can also be the result of large neighborhood sizes or high replacement rates — immediate from (6) and (8). The quantity being important in this context is the lower bound on possible infectiousness values, which is small (i.e., sub-critical, ) — this identifies vector-borne infections as type B.
The work in this paper was motivated by a desire to understand why the most transmissible human pathogens — archetypal childhood diseases such as measles and rubella — show remarkably little antigenic variation, while less transmissible diseases — such as influenza (and many other respiratory viruses) and sexually transmitted diseases show substantial diversity. Addressing this question requires consideration of how evolvable parameters governing the natural history of infection within a host affect the transmission characteristics of a pathogen in the host population.
We developed a relatively simple multi-strain model of the within-host dynamics of infection. Pathogen particle consume resource to replicate, and their replication is inhibited by a dynamically modeled immune response with two components: strain-specific immunity, and cross-immunity. Cross-immunity was assumed to be the key fitness cost of antigenic diversity within the host; the benefit is a much enhanced duration of infection (and thus transmission). Pathogens which have a low rate of generating new antigenic variants are cleared from the host much faster than those with a high rate of antigenic diversification, but also maximize the initial peak level of parasite load reached prior to clearance (cf. Methods).
The second evolvable within-host parameter we considered was the within-host pathogen replication rate. Given the resource-dependent model of replication assumed, this has a more limited effect than in some models, but can set the timescale for pathogen load to initially peak and thus determine the effective latent period of the disease.
At the between-host level, we assume a simple relationship between pathogen load and infectiousness which has been shown to be appropriate to model HIV transmissibility [12], and incorporates the concept of a soft threshold level of pathogen load needed for a substantial level of transmissibility, . As argued above, this parameter is perhaps best viewed as the amount of excreted pathogen which is wasted to achieve an infectious contact. For a perfect pathogen, the value could correspond to a single pathogen particle, but in reality the physics of transmission will typically mean is much higher. We have considered to be an evolvable parameter, but introduced the concept of minimum possible value of which is transmission route specific — being intrinsically much higher for respiratory pathogens (where transmission occurs via virus filling a three-dimensional volume around the infected individual), and potentially much lower for sexually transmitted diseases where transmission occurs over a two-dimensional contact surface.
The final element we incorporate into the framework developed is contact between hosts, assumed to occur at some rate , within a contact network of hosts with a certain mean neighborhood size and cliquishness . We derive a simple model to calculate the reproduction number of a single infected host in this network allowing for local saturation effects in the network caused by clustering. It is the network-specific reproduction number we have used as our overall measure of pathogen fitness, and examine what within- and between-host pathogen characteristics maximize fitness for different types of transmission route and host contact network.
Putting these elements together, we found that optimizing reproductive fitness in this way leads to well-defined infection types A, B, C, as contact rates (and reproductive numbers) increase (cf. Fig. 5). Type A and B both represent infections with low , with A being influenza-like and B mapping more to sexually transmitted diseases. When contact rates are very low, only one of these two types is evolutionary stable, with the stable type being determined by the assumed minimum infectiousness threshold. The latter serves as an order parameter and determines the mode of transmission. Consistently, type A corresponds to a high minimum infectiousness threshold whereas type B results from a low minimum threshold. The change of the transmission mode as a function of transmission threshold is phase transition-like.
Infection type C represents childhood diseases with the highest values of . This regime is not possible for small network neighborhood sizes or low values of cliquishness (i.e. random networks). It relies on the existence of large, persistent and highly clustered contact neighborhoods. In this context, maximizing the number of secondary infections (and thus overall fitness) requires a pathogen strain able to (a) infect as many of the index host's contacts as possible in as short a possible time, and (b) minimize the extent to which generations of infections overlap. The latter constraint is a result of the network clustering — if secondary cases become infectious while the index case is still infectious, they may deplete susceptible from the contact neighborhood before the index case has the chance to infect them. A latent period of the same or longer duration as the infectious period results in more discrete generations and maximizes the reproduction number of the index case. The need for a long latent period results in the evolutionary optimal value of the within-host replication rate , being relatively low for type C pathogens.
The limited antigenic diversity and short infectious periods of type C pathogens are determined by the higher infectiousness threshold and the consequent need to maximize the peak pathogen load attained early in infection. When contact rates are high, the increase in duration of infection resulting from higher rates of antigenic diversity is insufficient to compensate for the reduction in peak pathogen load (and therefore infectiousness) caused by cross-immunity being generated against multiple pathogen strains simultaneously. A single strain pathogen generating a single immune response is able to generate a larger primary infection peak — though at the cost of being unable to sustain infection further.
It is encouraging to see that the classification of infection types our model predicts closely corresponds to many of the pathogen regimes identified in other work [24]. However, our focus has been slightly different from that work, which focused more on the effect of different intensities of cross-immunity on between host phylodynamics. In contrast, we have focused more on examining how differences in transmission routes and contact rates () determine pathogen characteristics — though the influence of different levels of cross-immunity could be explored in future work.
Furthermore, it is interesting to note that in the context of our model only the concept of a minimal infectiousness threshold — introduced to characterize transmission modes — is necessary to explain the findings of [25] on tradeoffs between reproductive rate and antigenic mutability. Reference to the host's age is not needed here.
The key limitation of our analysis is our highly simplified treatment of between-host transmission — namely using a network-corrected reproduction number as our measure of strain fitness. Doing so assumes evolutionary competition occurring in infinite (non-evolving) host populations in infinite timescales. It would clearly be substantially more realistic to explicitly simulate the transmission process in a large host population. The computational challenges are considerable — while large-scale simulations of influenza A evolution and transmission have been undertaken [7],[26],[27], these have not included within-host dynamics, and have simulated evolution for decades rather than millennia. Other work [20],[28] has simulated the evolution of pathogen strains on a contact network for longer time periods, but only in very small () populations, and without modeling within-host dynamics.
However, continuing advances in computing performance mean that it may now be feasible to explicit model multiple strains evolving within hosts and being transmitted independently in a large population. Such an approach would allow exploration of the relationship between antigenic diversity (and cross-immunity) within single hosts and strain dynamics at a population level. Perhaps even more importantly, it would allow extinction processes to be properly captured, while our current approach implicitly assumes fixation probabilities to be 1 even when fitness differences are marginal. Proper representation of finite population sizes and extinction will also allow the evolutionary emergence of childhood diseases (such as measles) as a function of early urbanization to be modeled.
A second limitation is that we only consider a single, highly simplified within-host model. Future work to test the sensitivity of our results to the choice of within-host model would be valuable (cf. Text S1, Sect. A, which investigates an extension of the model here). That said, we would argue that the key qualitative feature of our within-host model driving the evolutionary results is the tradeoff — mediated by cross-immunity — between the maximum value of parasite load attained in initial infection and the degree of antigenic diversity (and thus duration of infection).
Also a conceptual simplification must be pointed out here: our model assumes that mutations, controlled by , directly affect antigenicity. For real-world pathogens, however, the link between genetic and antigenic change is less clear. Measles, for example, has a mutation rate typical of RNA viruses [29], but its antigenic diversity is low. Instead of mutation rate controlling antigenic variability, a pathogen may evolve phenotypic robustness to genetic change.
Further, we have not attempted to capture specialized strategies pathogens have adopted for persistence within infected hosts, such as use of refuges from immune responses (HSV) or hijacking the immune system (HIV) — the model only reflects tradeoffs which may have contributed to pathogens adopting the range of persistence strategies seen in nature. An interesting addition to future work would also be the incorporation of pathogen virulence [30], which imposes an additional evolutionary constraint on within-host replication rates.
A last area which is a clear priority for future research is the relationship between within-host parasite load and infectiousness. We have assumed a relationship which has some support in data (Fig. 1), and indeed the HIV system is perhaps the best explored in terms of the possible evolutionary tradeoffs inherent in maximizing transmissibility [31]. Unfortunately, little comparable data is available for other (especially respiratory) pathogens.
The within-host dynamics are simulated by the following system of ordinary differential equations (see [7] for more details where this system is introduced without cross-reactive immunity):(1)(2)(3)representing (1) the load of pathogen strain , (2) the amount of the adaptive immune response specific to strain , and (3) the level of resource which all strains need to replicate; the number of equations, , corresponds to the number of strains present, where denotes the total pathogen load. For viral infections, for example, the load is assumed to represents the number of virions of strain , the immunity variable somehow the amount of specific antigen (produced by B cells), and the resource target cells (e.g., epithelial cells for flu or T cells for HIV) of maximal number .
Saturation effects, modifying linear dependency on and , are modeled with the Hill function . The resource limitations act via , where, for large loads (), growth is limited by the maximal pathogen capacity related with the resource, ; for small loads, the load is independent of the resource, . The adaptive immune response is given by the growth term , which increases in response to antigen quickly and reaches values at . For larger pathogen loads, growth stops slowly, limited by when . The parameter represents the critical load above which immunity saturates. Its value is chosen above the number of pathogen units released after one replication cycle per resource unit (see below).
Guided by values for RNA viruses, random mutations are assumed to occur with probability per pathogen replication, which happens at rate . Only a proportion of mutations generate new antigenic variants. We assume that all mutations not leading to new antigenic variants are deleterious. The emergence of new antigenic variants is modeled stochastically, where a Poisson distribution with expectation determines the number of mutant strains at time , with denoting the cumulative load. While back mutations are neglected in the equations above, they are taken account of in the numerical calculations.
New antigenic variants generated at time induce a specific immune response, . This grows so long as , and declines for downwards, but never goes below . These characteristics are determined by the structure of (2) and the parameter choice , where and define the base rates at which immunity is produced and declines, respectively.
We assume 5 loci with 3 alleles at each. (These numbers are small but sufficient for our analyses, cf. Text S1, Sect. C.) The distance between strains, , is defined as the number of loci at which strains and differ. The immune-related clearance rate of strain is given by , where and for and 0 otherwise. Here is the degree of cross-immunity, and is the parameter governing homologous clearance rates.
Independent of immunity, pathogen is cleared at a rate (chosen smaller than ; cf. (5) below). Pathogen growth is limited by resource, where defines the saturation point. As pathogen grows at rate , resource is consequently depleted at rate . Resource is replenished at rate , and its total is modeled to never exceed (chosen to represent a realistic number of target cells and thus give realistic pathogen loads; cf. Fig. 1 and the examples above).
The differential equations are solved using a Runge-Kutta algorithm with the initial values , and , starting with 1 strain. New antigenic variants are generated potentially after each time step, each with initial pathogen load (corresponding to 1 pathogen unit infecting 1 resource unit) and specific immunity , if generated stochastically at . The infection ends once pathogen load drops below the value (which is assumed to be the elimination threshold), or after 2 years (the latter cutoff being chosen for computational simplicity).
The parameter values (essentially , and ) and the regions of pathogen space (given by and ) have been chosen to produce load curves (with significant resource depletion at the load peak, i.e., ) that resemble measles characteristics (with latency periods of up to 10 days and significant pathogen loads for similar periods; cf. Fig. 2F) for small antigenic variation and small/intermediate reproduction rates. The duration of infection is adjusted by the strength of immunity (i.e., ), with the value used here selected to give infections of over 1 year duration for maximal antigenic variation.
This model is minimally complex, incorporating only the features essential to explain the tradeoff between transmissibility and antigenic diversity. A more realistic model is examined in Text S1, Sect. A. However, the key diversity-transmissibility tradeoff arises as a simple consequence of within-host cross-reactive immune responses raised to individual new strains and competition between strains for a common resource for replication, and is relatively independent of the model-specific form of implementation of these mechanisms.
The essential within-host dynamics of our combined within/between-host model is given by Eq. (1), which links pathogen replication to two inhibitors — host immunity and resource limitation. This equation quantifies the tradeoff for increasing antigenic diversity (the pathogen's survival strategy within the host) — namely the smaller initial pathogen load peak seen in Fig. 3C (and Fig. S1-2C in Text S1, Sect. B.1). The specific realizations for the acquisition of immunity and the replenishment of resource (modeled by Eqs. (2) and (3), respectively) are less important.
Let us consider the pathogen load dynamics soon after infection with one initial strain. Our numerical simulations have shown that the initial strain is much more prevalent (by orders of magnitude) than mutant strains produced up to the first peak, . This observation clarifies that resource limitation (as one inhibitor of pathogen growth) cannot explain the tradeoff discussed here — being of low prevalence, mutant strains are unlikely to deplete resource to an extent which results in significantly lower loads, and in any case all strains have the same intrinsic replication rate and use the same resource. But the specific immune response to mutant strains, provided it is partially cross-reactive, is able to reduce both the load of the initial strain and other strains, and can thus lower the total pathogen load. This result is largely independent of model implementation and only depends on the strain-specific immune response being generated at relatively low strain-specific pathogen loads, and being sufficiently cross-reactive to slow overall growth of pathogen load.
This verbal argument can be formalized. For simplicity we assume the load of the initial strain is a good approximation of the total pathogen load at the initial peak, . By applying as a condition for the initial peak, Eqn. (1) (with ) then yields a relation for the initial peak load,(4)where defines the immune response with respect to the strain number. Provided cross-reactive immunity is implemented (i.e., for some , so that ), the function is strictly increasing (independently of how cross-immunity is defined via the strain-distance weight function and the parameter ). This is based on the fact that, together with each newly generated strain , immunity is produced in a standardized way for the time period up to the initial peak when load is increasing and above a critical value, . This is the case in any setting where mutant strains have the same intrinsic replication kinetics as the initial strain. In our model, immune production happens at rates above (and below ) as long as , independently of the concrete acquisition rule in (2); see the modifications (Eqs. (S1-1,2)) and the corresponding result (Fig. S1-2C) in Text S1 for a more realistic but also more complicated mechanism.
As a consequence of resource limitation (i.e., the reduced growth ), (4) yields(5)Due to the monotony of , the function given by (5) is strictly decreasing. That means that the magnitude of the initial peak is inversely related to the number of (mutant) strains present. The result is independent of the specific functional form used for resource depletion (in (1)) and replenishment (in (3)), as is confirmed by considering the limit of large pathogen loads, where ; the resulting peak height, , shows the same monotonic dependence on as (5).
Finally, we examine what would happen if cross-immunity or resource limitation were not implemented in the model. Without cross-immunity, , and the initial peak is thus independent of the strain number (cf. Fig. S1-2I in Text S1, Sect. B.1). Without resource limitation, (4) degenerates, and the initial peak load cannot be compared for different values of antigenic variation.
As discussed in the text, we use the basic reproductive number of infected hosts as the measure of evolutionary fitness for infectious diseases [23]. For infections of finite duration ,(6)where denotes the number of susceptible hosts in the neighborhood (of assumingly constant size ) of one initially infected host, and is the transmission rate from the index case at time after infection. The pathogen-load dependence of the transmission rate is modeled by(7)where is the infectiousness threshold parameter and is the transmission coefficient, which critically depends on the contact rate . The parameter is the transmission probability per contact for a completely saturated pathogen load (), and lies between 0 and 1. This functional form is consistent with data for HIV (Fig. 1). The transmission dynamics in the entire susceptible contact neighborhood of an index case are given by(8)where . This equation models a local dynamic network (derived in the section below), where defines the transitivity or cliquishness of the network (proportion of neighbors of a node who are neighbors of each other) and the per-capita rate at which hosts in the neighborhood of the index case are replaced by new susceptible hosts. Here represents convolution, with . This expression corrects for the depletion of the local contact neighborhood of the primary case by individuals infected by the index case then infecting shared contacts of the index before the index case herself does. Such local saturation of the susceptible population is entirely a network effect and vanishes for .
It should be noted that the network dynamics are invariant for , bar a scaling of by . Enlarging the neighborhood size thus corresponds to effectively reducing cliquishness. This relation allows for incorporating vector-borne infections (characterized by large ) into our classification (as type B infections; cf. end of the section Infection types). Although our modeling framework has been designed for direct transmissions, it can formally be applied to vector-borne infections assuming that (due to relatively low ) the transmission delay through the vector is less important.
Here we derive Eqn. (8) of our between-host model, which also illustrates how the two parameters, and , characterize the host-contact network on local and on global scales, respectively.
The transmission dynamic in an initially entire susceptible contact neighborhood of one index case and fixed size, , can be reconstructed approximately in terms of average numbers of infectives and susceptibles ( and , resp.),(9)counting the (infinitesimal) number of new infections caused by the index case at time . We have included direct infections and secondary infections which, we assume, occur with likelihood in the contact neighborhood. The time delays, , as reflected by the transmission rates relevant for secondary infections, correspond to primary infections at . The integral covers the secondary infections caused by new infectives up to time , respecting the changing transmission rates resulting from time-dependent pathogen loads (cf. (7)).
Written exclusively in terms of susceptibles (while utilizing the notion of convolution), (9) reads(10)From here, Eqn. (8) is obtained by incorporating a constant (global) flow of individuals (referring to the entire host population) into the transmission model, quantified by the replacement rate of individuals in the considered neighborhood. This is readily confirmed by the formal replacement (of the ordinary derivative by a covariant version),(11)which models the recruitment of new susceptibles in exchange for old infectives.
|
10.1371/journal.pgen.0030078 | p53 Activation by Knockdown Technologies | Morpholino phosphorodiamidate antisense oligonucleotides (MOs) and short interfering RNAs (siRNAs) are commonly used platforms to study gene function by sequence-specific knockdown. Both technologies, however, can elicit undesirable off-target effects. We have used several model genes to study these effects in detail in the zebrafish, Danio rerio. Using the zebrafish embryo as a template, correct and mistargeting effects are readily discernible through direct comparison of MO-injected animals with well-studied mutants. We show here indistinguishable off-targeting effects for both maternal and zygotic mRNAs and for both translational and splice-site targeting MOs. The major off-targeting effect is mediated through p53 activation, as detected through the transferase-mediated dUTP nick end labeling assay, acridine orange, and p21 transcriptional activation assays. Concurrent knockdown of p53 specifically ameliorates the cell death induced by MO off-targeting. Importantly, reversal of p53-dependent cell death by p53 knockdown does not affect specific loss of gene function, such as the cell death caused by loss of function of chordin. Interestingly, quantitative reverse-transcriptase PCR, microarrays and whole-mount in situ hybridization assays show that MO off-targeting effects are accompanied by diagnostic transcription of an N-terminal truncated p53 isoform that uses a recently recognized internal p53 promoter. We show here that MO off-targeting results in induction of a p53-dependent cell death pathway. p53 activation has also recently been shown to be an unspecified off-target effect of siRNAs. Both commonly used knockdown technologies can thus induce secondary but sequence-specific p53 activation. p53 inhibition could potentially be applicable to other systems to suppress off-target effects caused by other knockdown technologies.
| Recent advances in sequence-based approaches to “knockdown” gene function have opened the door to an array of approaches to uncover functions for genes of interest. Vertebrate knockdown strategies—such as morpholinos (MOs) in zebrafish or RNA interference-based strategies in mammalian systems—have been demonstrated to be effective, rapid, and cost-efficient reverse-genetic approaches for studying gene function. However, their deployment has to date been limited by a number of technical (genomic, biological, and off-targeting) hurdles. One of the notable and unexpected findings from our work using MOs has been a series of observations surrounding unanticipated effects that are independent of the intended gene target. We have identified and characterized a recently described p53 induction pathway due to off-targeting that appears to be shared between knockdown technologies. This study reconciles a series of unexpected findings that show p53 upregulation at the transcriptional level in a subset of short inhibitory RNA- and MO-treated vertebrate systems. Moreover, concurrent p53 knockdown provides a new approach to facilitate the identification of previously hidden gene functions. This study provides both a new gene knockdown enhancement tool as well as additional insight into an important and conserved pathway implicated in cellular toxicity.
| Morpholino phosphorodiamidate oligonucleotides (MOs) [1] and short inhibitory RNAs (siRNAs) [2] have been instrumental to induce sequence-specific gene knockdown in multiple systems. However, the use of both technologies is sometimes limited by induction of off-target effects [3–7]. About 15–20% of MOs used in zebrafish show off-targeting effects [3], represented by a signature neural death peaking at the end of segmentation (1 day post-fertilization [dpf]). The affected embryos grow with smaller heads and eyes, exhibit somite and notochord abnormalities, and eventually display craniofacial defects. These MO-induced developmental defects are target-independent because they are not displayed by characterized mutants in the respective genes [3].
We show here that the off-target effects of MOs are mediated through p53-induced apoptosis. Concurrent knockdown of p53 with various MOs significantly alleviates off-target neural death. Importantly, however, p53 MO did not affect specific phenotypes induced by a variety of MOs. We propose the use of p53 knockdown as a tool to attenuate off-target effects and facilitate the study of specific loss of function phenotypes.
General morphological features of MO-induced off-target neural death have been previously described [3]. We further investigated the nature of this cell death and the mechanism of MO mistargeting. For this report, we focused primarily on MOs designed against two gene targets for which mutants have been previously described, wnt5/pipetail (ppt) [8] and smoothened/slow muscle omitted (smu) [9,10], to facilitate the discrimination between specific and nonspecific effects. A translational MO against smoothened (Smo MO) induces characteristic smu phenotype (spinal curvature, U-shaped somites) (Figure 1C). A splice-site wnt5 MO (Wnt5 MO1) induces tail and body-axis shortening and somite compression (Figure 1E), characteristic of the wnt5/ppt mutant (Figure 1K). What both Smo MO- and Wnt5 MO-injected embryos (morphants) have in common is an additional and very similar neural death (Figure 1, arrows). This neural death is target-independent, since it is not exhibited by the respective mutants (Figure 1K) [9,10]. Nonetheless, this neural death appears to be sequence-specific, since a completely different splice-site wnt5 MO (Wnt5 MO2) shows no neural death, but readily induces the characteristic wnt5/ppt phenotype (Figure 1G). We tested another type of knockdown molecule based on an alternating trans-4-hydroxy-L-proline/phosphate polyamide backbone called gripNA [11]. Interestingly, a gripNA targeting wnt5 (similar in sequence to Wnt5 MO1) also induces neural death along with the characteristic wnt5/ppt phenotype (Figure 1I). A gripNA against smoothened also causes additional neural death (unpublished data), supporting the idea that the off-targeting effects are not limited to the MO chemistry, but represent a common feature to these knockdown technologies.
The off-target neural death induced by MOs is highly reminiscent of the neural death induced by a published Mdm2 MO (Figure 1M). Mdm2 is a negative regulator of the tumor suppressor p53, the gene most frequently mutated in human cancers [12]. Mdm2 knockout in mice is an embryonic lethal [13] due to extensive p53 upregulation and p53-induced apoptosis. Mdm2-targeted MO in zebrafish was reported to induce apoptotic neural death [14]. We examined the mechanism of MO-induced off-target neural death by testing for apoptosis in multiple MO-injected zebrafish embryos using a transferase-mediated dUTP nick end labeling (TUNEL) assay (Figure 2) and by staining with acridine orange (unpublished data). Our results suggest that the neural death induced by off-targeting MOs is apoptotic in nature and was indistinguishable from the cell death observed in the Mmd2 knockdown [14]. We tested the specificity of Mdm2 MO-induced cell death by overexpressing a Mdm2 RNA construct. However, we did not observe any significant rescue of the Mdm2 MO-induced cell death with the Mdm2 RNA construct (unpublished data). Therefore, it is possible that the cell death phenotype induced by the Mdm2 MO is also primarily an off-targeting effect. We performed an in-depth analysis of MO off-targeting to examine the concordance between the phenotypes observed by light microscopy and apoptosis patterns observed by TUNEL staining. We analyzed zebrafish embryos injected with Wnt5 MO1 at 14 hpf (the onset of cell death, [3]), 22 hpf, 26 hpf, and 30 hpf (Figures 3 and 4; Figure S1). Brightfield images show the signature appearance of opaque-looking discolored tissue around the eyes and in the nervous system in embryos injected with Wnt5 MO1 (Figures 3 and 4; Figure S1). The extent of the opaque tissue increased at later time points and with increasing MO dose. The cell death could be more easily visualized using darkfield microscopy (Figures 3A, 3C, 3F, 3I, 3L, 3O, 3R, 3U, 3X, 4C, 4F, 4I, 4L, 4O, 4R, 4U, and 4X). This analysis shows the characteristic pattern of white tissue corresponding to the opaque structures seen in brightfield that is diagnostic of MO mistargeting in zebrafish embryos.
At 26 hpf, milder phenotypes displaying a characteristic anterior-ventral concavity and/or hindbrain depression could be observed, in the absence of the opaque/white tissue characteristic of the more severe cases of cell death (Figure 4D–4F). However, when analyzed by TUNEL staining, we observed that even the mild phenotypes (as seen by light microscopy) were associated with significant apoptosis (Figure 4P–4R). These mildly affected embryos usually recovered by 30 hpf, when they showed significantly less cell death, if any (unpublished data). The more severely affected embryos did not recover until day 2 or 3, at which time the characteristic apoptotic tissue was no longer apparent either through light microscopy or TUNEL analysis (unpublished data). However, these embryos lacked some neural tissue and developed with smaller heads and eyes (unpublished data).
The punctuated pattern of neural apoptosis seen at all time points and with increasing intensity in the more severe cases was strikingly different from the normal apoptotic pattern seen in uninjected control embryos (Figures 3 and 4). Developmentally regulated apoptosis has been described in detail [15] and was recapitulated by our analysis (Figures 3 and 4, uninjected embryos). However, at all studied time points the extent of developmentally regulated apoptosis was significantly less extensive than the apoptosis induced by MO off-targeting. In particular, at 30 hpf, little if any apoptosis was noted in control embryos. Therefore, we performed the TUNEL analysis at 30 hpf for all subsequent experiments to clearly differentiate between developmentally regulated apoptosis and apoptosis caused by MO off-targeting.
The neural apoptosis induced by a variety of MOs and the similarity to the phenotype induced by apparent p53 upregulation (Mdm2 MO) suggested the hypothesis that MO off-target effects can induce the p53 apoptosis pathway. Therefore, we tested whether p53 knockdown can rescue the off-target apoptosis phenotype induced by several MOs. Indeed, p53 MO attenuated the neural death induced by smoothened and Wnt5 MOs, as shown by morphology (Figure 1D and 1F), acridine orange (unpublished data), and the more specific TUNEL assay (Figure 2D and 2F). Similar results were observed for p53 knockdown rescue of Mdm2 MO-induced apoptosis (Figures 1N and 2M). Interestingly, p53 MO also alleviated the neural death induced by the Wnt5 gripNA, suggesting that this additional knockdown technology can upregulate the p53 pathway due to off-targeting (Figures 1J and 2I). However, as expected, p53 MO did not have any effect on wnt5/ppt mutant embryos (Figures 1K, 1L, 2J, and 2K). A second p53 MO of independent sequence also attenuated the off-target neural death, while a four-base mismatched MO did not show any effect (unpublished data).
Because neural death caused by MO-induced off-target effects is so frequent [3,16], we tested the p53 MO as a tool to alleviate off-target neural death. A good tool for this purpose should be effective, innocuous, and specific. The p53 knockdown by itself does not induce any significant defects, as p53 is not required for normal development in mammals or fish [17,18] (Figures 1B and 2B). Also, p53 MO does not affect the efficacy of gene-specific MOs, as it does not interfere with the penetrance of gene-specific phenotypes. To further confirm this, we tested whether p53 MO can affect the efficiency of splicing inhibition by Wnt5 MO1. Semi-quantitative reverse-transcriptase PCR (RT-PCR) analysis of Wnt5 RNA transcripts showed complete blockage of the splicing at exon 5–exon 6 boundary targeted by Wnt5 MO1, which was not affected by p53 MO (Figure 1H).
We also investigated whether p53 knockdown can affect specific cell death (other than neural death) and whether it affects phenotypes not associated with apoptosis (Figure 5). As shown by morphology and TUNEL staining (Figure 5A–5D), the p53 MO had no effect on the specific tail-cell death induced by the loss of function of chordin using a chordin-specific MO. In addition, p53 knockdown showed no effect on the MO-induced phenotypes of nacre (a pigment defect) (Figure 5G and 5H), no tail (a developmental patterning gene) (Figure 5I and 5J), or UROD (loss of function is visualized by fluorescence of red blood cells) (Figure 5K and 5L).
In conclusion, the p53 MO could be an efficient tool to attenuate off-target effects of MOs. We are currently coinjecting the p53 MO with all the MOs tested in a large-scale MO screen [16]. This strategy has greatly attenuated the neural death phenotypes and has notably eased the interpretation of the observed phenotypes, especially in craniofacial development (Figure 6; see below). p53 knockdown or the use of p53 null zebrafish [18] could potentially be of value for use in more traditional genetic approaches, such as chemical or insertional mutagenesis screens, to decrease the collateral tissue damage due to p53-induced cell death that potentially masks important phenotypes of particular interest to investigators.
Early neural death and loss of neural tissue caused by MO off-targeting could potentially affect later craniofacial development. This may generate numerous false positives in MO screening for genes important in craniofacial development. For example, we tested whether p53 co-knockdown could facilitate the analysis of craniofacial phenotypes, especially in the cases of unknown genes or where corresponding mutants are not available. For example, MOs that target three genes in our collection of novel proteins [16], SP2035, SP2054, and SP2063, caused neural death visible at 1 dpf and craniofacial defects visualized by Alcian Blue staining of the cartilage at 4 dpf.
The neural death caused by these MOs was attenuated by p53 co-knockdown (Figure 6A). The brightfield panels in Figure 6A show two types of milder neural defects that we have described in Figure 4; an anterior-ventral concavity for SP2054 and SP2063 (represented by a deficiency in the frontonasal tissue development, black arrows in Figure 6A) and a depressed hindbrain for SP2035 (represented by a lack/developmental delay of the hindbrain tissue, black arrowhead in Figure 6A). Interestingly, these milder defects were clearly associated with neural apoptosis, as shown by acridine orange staining (Figure 6A, fluorescence panels and quantified in corresponding graphs). At higher doses, the MOs against these targets showed a clear cell death pattern even in brightfield images (represented by opaque structures; unpublished data).
Later in development (4 dpf), the MO-injected embryos mentioned above also exhibited craniofacial defects (Figure 6B). We investigated whether these late craniofacial defects were due to the early loss of neural tissue (off-targeting) or to a specific role of the targeted genes in craniofacial development. To achieve this, we analyzed the effect of p53 MO on the cartilage structure at 4 dpf. The craniofacial defects in the SP2035 and SP2054 MO-injected embryos were not affected by p53 co-knockdown, while the craniofacial defects in the SP2063 MO-injected embryo were significantly diminished by p53 co-knockdown (Figure 6C). These results suggest that SP2035 and SP2054 are involved in craniofacial development, while the craniofacial defects seen in SP2063 MO-injected embryo are p53-dependent and thus may be due solely to off-targeting effects of the MO.
To further distinguish putative roles of SP2035, SP2054, and SP2063 in craniofacial development, we analyzed the expression patterns of these genes in zebrafish embryos (Figures 7 and S2). At 1 dpf, all three genes were expressed in the craniofacial region. But while the expression patterns of SP2035 and SP2054 were spatially restricted, SP2063 was more ubiquitously expressed. Interestingly, in subsequent days of development, SP2035 and SP2054 transcripts became specifically enriched in the pharyngeal arches primordia, while SP2063 became restricted to central nervous system structures (Figures 7 and S2). These expression patterns support a direct role of SP2035 and SP2054 in craniofacial development, while the role of SP2063 may be indirect, if any. The CNS expression of SP2063 may also explain the partial rescue of the SP2063 craniofacial phenotypes by p53 MO. If brain structures were affected by SP2063 MO injection, this may have influenced the mechanical structure of cartilage and contributed to the craniofacial phenotype, in conjunction with the loss of neural tissue caused by MO off-targeting.
In conclusion, the likely involvement of the studied novel genes in craniofacial development is supported by their expression pattern and corroborated with the dependence of craniofacial phenotypes on p53. Therefore, p53 co-knockdown can be used to help clarify craniofacial phenotypes induced by MOs against novel genes for which there are no mutant data available for comparison.
To understand better the mistargeting effects of MOs, we investigated other components of the p53 pathway. A direct target of the p53 transcription factor is p21/WAF/CIP [19]. We tested whether p21 transcription is induced in morphants with neural death. Using quantitative RT-PCR, we observed a significant increase in p21 RNA levels in morphants that show neural death (Smo MO and Wnt5 MO1) but no significant increase in morphants without off-target effects (Wnt5 MO2) (Figure 8). Very importantly, this increase in p21 expression was dependent on p53, since knockdown of p53 significantly decreased p21 RNA levels in respective morphants. Induction of p21 levels provides direct evidence for activation of the p53 protein. These results were similar to the induction of p21 in Mdm2 MO-injected embryos, which was dependent on p53, as expected (Figure 8) [14]. These results suggest that p53 protein is activated by injection of a selection of MOs, associated with off-target neural apoptosis. Consistent with this conclusion, p53 protein is not activated in a selection of morphants that do not exhibit any neural death.
Translational and post-translational mechanisms of p53 activation have been extensively documented [12]. A well-known mechanism for p53 induction is due to Mdm2 inactivation. Because Mdm2 is a ubiquitin-E3 ligase that targets p53 for proteasomal destruction, loss of function of Mdm2 leads to p53 protein accumulation and consequent apoptosis [14]. We investigated whether p53 transcriptional regulation is part of the p53 induction due to MO off-targeting. The p53 gene is known to express multiple isoforms as result of alternative splicing and internal promoters (Figure 9A) [20]. We designed primers to amplify a fragment specific to full-length p53 cDNA, which is the isoform we targeted by our p53 MO (Figure 9A) and was shown to be sufficient for neural death induction by co-knockdown experiments (Figures 1 and 2). We examined the levels of p53 transcription in embryos injected with various MOs by semi-quantitative RT-PCR. Interestingly, there was no significant increase in full-length p53 RNA levels in various MO-injected embryos (Figure 9B, top panel), suggesting that transcriptional induction of full-length p53 does not play a role in p53 activation by MO off-targeting. These results, together with our observations that knockdown of full-length p53 alleviates MO off-target effects, support a direct role of full-length p53 protein, but not of p53 transcriptional regulation, in neural death caused by MO off-targeting.
The p53 locus expresses multiple transcripts as result of alternative splicing and internal promoters [20]. For example, zebrafish have been recently reported to use an internal promoter in intron 4, conserved from flies to humans [21], to express an N-terminal truncated form of p53, Δ113 p53 (Figure 9A) [22]. The truncated p53 isoform is highly upregulated in zebrafish def mutants, specifically in the characteristic hypoplastic digestive organs [22].
We tested whether this truncated p53 isoform was induced under cell-death conditions through MO off-targeting. To discriminate between the two p53 transcripts, we designed primers to amplify specifically either the full-length p53 cDNA or that encoding the truncated p53 isoform (Figure 9A) and used semi-quantitative RT-PCR to examine p53 transcripts (Figure 9B). While full-length p53 RNA levels were not significantly increased in any of the MO-injected embryos (Figure 9B, top), the Δ113 p53 isoform was highly upregulated in MO-injected embryos with neural death and virtually absent in the MO-injected embryos with no neural death or in the uninjected controls (Figure 9B, middle).
We also performed microarray screens for the transcriptional consequences of various MOs. As shown in Figure 9C, when zebrafish embryos were treated with a MO or a gripNA [11] against the D. rerio homolog of the fhit tumor suppressor gene [23], we obtained evidence for increased transcription at the p53 locus. In five fhit knockdown microarrays, p53 transcripts were increased 7.9-fold with respect to control-injected embryos (t-test p-value = 0.00007). Remarkably, p53 and two other mRNAs among the top eight transcripts induced in the fhit datasets were common to the top eight induced genes in def zebrafish embryos [22]. Indeed, we have seen many of the same mRNAs coinduced by unrelated MOs (unpublished data). However, it is noteworthy that the probe used for microarrays binds to the 3′ UTR of p53, thus recognizing both full-length and the Δ113 p53 isoforms (Figure 9A).
We also conducted in situ hybridization experiments with a p53 riboprobe in embryos injected with the two Wnt5 MOs, one that showed neural death (MO1) and one that did not (MO2) (Figure 9D). Wnt5 MO1 showed increased p53 mRNA expression in the anterior part of the body (arrow, Figure 9D), while Wnt5 MO2 showed low ubiquitous p53 mRNA expression similar to the uninjected control (Figure 9D). In this case also, the riboprobe could bind both full-length and the Δ113 p53 isoforms (Figure 9A).
The RT-PCR experiments in Figure 9B showed that full-length p53 RNA levels were not increased in any MO-injected embryos, while the Δ113 p53 isoform was highly induced in embryos injected with off-targeting MOs. These results suggest that the increased p53 expression observed by microarray and in situ hybridization consists largely of Δ113 p53 RNA, and that transcriptional induction of full-length p53 does not contribute to p53 activation by MO off-targeting.
The p53 MO, which blocks neural cell death, was designed to knock down full-length p53 and would not be expected to affect the Δ113 p53 isoform (Figure 9A). To further test whether the highly induced Δ113 p53 mRNA is required for neural cell death, we designed a translational MO to specifically knock down this isoform (Figure 9A). We cannot design a splice-site blocker MO specific only for the N-terminal truncated isoform because all the splice junctions present in Δ113 p53 are also present in full-length p53 (Figure 9A). Coinjection of the Δ113 p53 MO with off-targeting MOs did not block cell death (unpublished data). This result is consistent with the fact that Δ113 p53 lacks the transactivation domain and part of the DNA binding domain, which are thought to be required for induction of apoptosis [21]. Thus, the Δ113 p53 isoform is most likely not the cause of cell death induced by MOs and may represent a diagnostic signature of off-target effects. In contrast, the full-length p53 protein is sufficient to cause neural death due to MO off-target effects, even if transcript levels are unchanged. More experiments are necessary to evaluate the significance of the Δ113 p53 isoform transcriptional induction beyond its use as a diagnostic for p53 activation.
We have shown that mistargeting MOs induce neural death via a pathway involving p53 activation. Curiously, ongoing synthesis of full-length p53 is required for cell death, while transcription of the Δ113 p53 isoform is a consistent and striking component of the mistargeting MO signature. We investigated various hypotheses for the mechanism underlying this off-target effect. The p53 pathway induction is independent of the intended gene target and appears to be sequence-specific, since two MOs of independent sequence, but targeted to the same gene, have strikingly different effects on p53 induction. This off-target effect is noted in both translational blockers and splice-site MOs, suggesting that the mechanism does not uniquely involve the transcription or the translation machinery. According to our analysis, MOs with off-target effects do not exhibit any overt primary sequence similarity to repeated elements such as rRNA genes or the zebrafish mitochondrial genome (unpublished data).
Although the mechanism of MO-induced p53 activation is still unclear, this pathway is activated by other knockdown technologies including gripNAs (Figures 1I and 1J, 2H and 2I, and 9C). Furthermore, related observations indicate that siRNAs can also induce off-target p53 activation. A recent study reports divergent changes in levels of p53 and p21 in cells subjected to ten different siRNAs targeted to menin [4]. The study shows that, while all the siRNAs knock down menin levels to various extents, some of the siRNAs cause a significant increase in p53 and p21 protein levels, independent of the levels of menin knockdown, while others have no effect on p53 or p21. One hypothesis is that the off-target effects caused by siRNAs are due to short sequence homology to other genes [5–7,24]. We have not observed any pattern of partial homology between off-targeting MOs and p53 or Mdm2 genes (unpublished data).
We have shown that certain MOs and gripNAs induce neural cell death in a manner that depends on synthesis of full-length p53 protein, but not on transcriptional activation of full-length p53. We also observed a diagnostic transcriptional induction of an N-terminal truncated isoform of p53. Interestingly, this truncated form is thought to act as a dominant negative molecule towards full-length p53, as it lacks the transactivation domain and part of the DNA binding domain [21]. The human homolog of Δ113 p53 was shown to be defective in promoting apoptosis and even to inhibit p53-mediated apoptosis [21].
Consistent with these results, a translational MO targeted specifically to the Δ113 p53 isoform did not alleviate the neural death induced by off-targeting MOs (unpublished data), although the full-length p53 knockdown did. Also, overexpression of Δ113 p53 RNA in zebrafish embryos did not cause neural death (unpublished data), suggesting that the Δ113 p53 isoform is insufficient to promote apoptosis. Potentially, the Δ113 isoform is transcriptionally induced secondary to p53-mediated apoptosis.
Transcriptional induction of the Δ113 isoform of p53 may represent a diagnostic signature for a specific type of cellular stress. High levels of the Δ113 isoform p53 transcription were observed in def [22] and fhit knockdown embryos and in morphants with off-targeting phenotypes, while lower levels of mRNA increase were observed in flathead embryos [25]. It remains to be determined whether off-targeting oligos target DNA, an RNA other than mRNA, or another cellular component, and whether the fhit knockdown profile is due to off-targeting or to a specific involvement in the stress response pathway.
A previous study also reported the presence of a shorter p53 transcript in zebrafish [14], with a size consistent with the predicted length of the Δ113 p53 isoform. Intriguingly, this shorter transcript was highly upregulated in zebrafish embryos under cell death–inducing conditions such as treatment with camptothecin or roscovitine or knockdown of the anti-apoptotic genes mdm2 and tsg1. Also noteworthy, the expression of the shorter p53 transcript seemed to be dependent on full-length p53 [14].
A very important issue for using p53 knockdown to mitigate neural death is specificity. In many cases, neural death can be a specific phenotype, and p53 MO rescue may suggest a specific interaction with the gene of interest. A key experiment to validate a MO phenotype is to observe rescue of the morphant phenotype with an RNA or DNA construct of the respective gene. If the neural death is rescued by the RNA/DNA construct, it is very likely that the gene of interest is specifically involved in cell death. If, however, the RNA/DNA rescue still yields a neural death phenotype, it is possible that the neural death is an off-target effect of the MO. For example, a recent study reported apoptosis and neuronal loss upon knockdown of presenilin enhancer Pen-2 in zebrafish embryos [26]. This neural death was significantly reduced by p53 co-knockdown, as in the case of off-targeting MOs. However, the authors clearly showed a rescue of the neural apoptosis by a Pen-2 RNA construct of a sequence not overlapping with the Pen-2 MO [26]. Together, these results support a true anti-apoptotic role of Pen-2 in promoting neuronal survival.
We have also attempted to rescue the Mdm2 MO-induced cell death phenotype with a Mdm2 RNA construct. However, we did not observe any significant rescue (unpublished data). One potential explanation is that this particular Mdm2 MO also has off-targeting effects. Five additional Mdm2 MOs have been reported to cause cell death that could be rescued by p53 MO [14], but we did not test any of these. We also tested whether the Wnt5 MO1-induced cell death is wnt5/ppt specific or a result of off-targeting. There are no indications from previous studies to suggest a role of wnt5/ppt in cell death. We observed no effect of a wnt5/ppt RNA construct [27] on the cell death specifically induced by Wnt5 MO1 (unpublished data), but not by Wnt5 MO2. However, we did not observe any rescue of the characteristic morphological defect associated with loss of wnt5/ppt either. This result is not so surprising, though, as there is no previous report on a successful RNA rescue of the body axis shortening phenotype caused by wnt5/ppt inactivation (either mutation- or MO-induced). It is possible that generalized overexpression of wnt5/ppt RNA may not be sufficient to compensate for decreased wnt5/ppt levels at the appropriate time and place.
In conclusion, if a cell-death phenotype caused by knockdown can be rescued by the respective RNA/DNA construct, it is likely that the gene of interest is involved in cell survival. If the RNA/DNA construct rescues the gene-specific phenotype but does not rescue the cell death phenotype observed in MO-injected embryos, it is likely that cell death represents an off-targeting effect of the MO. It is also possible that certain MO-induced phenotypes cannot be rescued by corresponding RNA/DNA overexpression, due to improper localization and/or timing of expression during development.
Ongoing work is geared to exploit p53 co-knockdown to alleviate off-target neural death of MOs and to discover the mechanism by which off-target MOs induce p53 activation as well as the signature Δ113 p53 transcript. Potentially, p53 knockdown by RNAi may also alleviate the off-target effects of siRNAs [7].
Wild-type zebrafish were purchased from Segrest farms (http://www.segrestfarms.com). Embryos were raised at 30 °C and spawning was carried out as described [28]. wnt5 mutant fish carrying the ppt hi1789b allele [29] were obtained from the Zebrafish International Resource Center (http://zebrafish.org/zirc).
MO and gripNA sequences are shown in Table 1. MOs were obtained from Gene Tools (http://www.gene-tools.com) and were prepared and injected in 1–4 cell stage embryos as described [30]. When two MOs were injected in the same embryo, we carried out both separate injections of the different MOs and single injections of MO mixtures, with very similar results. The only difference was a slightly increased mortality in the case of double-injected embryos as compared to single injections. In all cases, except where noted, p53 MO was injected 1.5-fold (w/w) to the other MO used. MO doses were: 3 ng of Smo MO, Wnt5 MO1, and Mdm2 MO; 4.5ng of p53 MO (except where noted otherwise); and 6ng of Wnt5 MO2. GripNAs were obtained from Active Motif (http://www.activemotif.com) and were prepared and injected similar to MOs. Wnt5 GripNA was injected at 2.25 ng, and coinjections with p53 MO were at 4 ng.
Embryos were visualized at 24–29 hpf, except where noted. Microscopy was performed on a Zeiss Axioplan 2 microscope (http://www.zeiss.com) fitted with differential interference contrast microscopy optics. Images were captured with a Nikon Coolpix 995 (http://www.nikonusa.com) or a Canon PowerShot G6 digital camera (http://www.canon.com), with multiple images combined using Adobe Photoshop software (http://www.adobe.com).
Embryos were dechorionated and fixed at 30 hpf or as indicated in 4% paraformaldehyde for 1 h at room temperature. They were then washed with PBS buffer twice and permeabilized with 0.1% sodium citrate and 0.1% TritonX for 2 min on ice. After washing twice in PBS buffer, embryos were incubated with the reaction mixture containing the terminal deoxynucleotidyl transferase and TMR-labeled nucleotides for 1 h in the dark at 37 °C. Reaction was stopped by washing with PBS three times. Terminal deoxynucleotidyl transferase catalyzes incorporation of labeled nucleotides to 3′-OH DNA ends in a template-independent reaction. The fluorescent signal was visualized and imaged using a Zeiss Axioplan 2 microscope coupled to an ApoTome, using AxioVision 4.2 software. z-stacks were superimposed using Extended Focus feature of the software.
Live embryos were immersed in 5 μg/ml acridine orange (Sigma, http://www.sigmaaldrich.com) for 10 min, then visualized and imaged for less than 60 s (the signal is quenched after 60-s exposure to fluorescence), as described for the TUNEL assay.
Total RNA was extracted from 32 hpf embryos using TRIZOL reagent (Invitrogen, http://www.invitrogen.com). Quantitative RT-PCR was carried out on 200ng of RNA using the LightCycler RNA Amplification kit SYBR Green (Roche, http://www.roche-diagnostics.us) in a LightCycler 2.0 Instrument, following manufacturer's protocols. The primers used are shown in Table 2. All expected PCR products span at least one intron (except the Δ113 p53 fragment), to ensure amplification solely from the cDNA and not from the genomic DNA. The primers for full-length p53 correspond to exon 4 (not present in the Δ113 p53 isoform) and exon 5. The primers for the Δ113 p53 isoform correspond to intron 4 (not present in the full length p53) and exon 5. The identity of the RT-PCR products was confirmed by sequencing. The samples were quantified by comparative cycle threshold (Ct) method for relative quantification of gene expression [31], normalized to β-actin. All experiments were performed with at least two different RNA preparations and at least three independent experiments for each RNA preparation.
cDNA for p53 probe was amplified using total RNA from 24 hpf zebrafish embryos injected with pax2 MO (Table 1) with primers shown in Table 2. The p53 riboprobe used in the in situ hybridization experiments spans exons 6–11, a region common to both full-length and Δ113 p53 isoforms. The cDNAs for SP2035, SP2054, and SP2063 were amplified from total RNA from 30 hpf zebrafish embryos, using primers indicated in Table 2. The PCR fragments for p53, SP2035, SP2054, and SP2063 were cloned into the pCRII TOPO vector (Invitrogen). The plasmids were then linearized with NotI (p53, SP2054 and SP2063) or Spe I (SP2035). DIG-labeled antisense RNA was synthesized using the SP6 polymerase (p53, SP2054, and SP2063) or T7 polymerase (SP2035) in conjunction with the in vitro DIG labeling kit (Roche). Zebrafish in situ hybridization was performed on 26–28 hpf embryos or indicated time points as previously described [32]. Microscopy was performed on a Zeiss Axioplan 2 microscope using DIC optics. Images were captured with a Canon PowerShot G6 digital camera.
Cartilage was stained with Alcian Blue using a modification of previously published protocols [33,34]. Anesthetized 4.5 dpf larvae were fixed in 4% phosphate-buffered paraformaldehyde overnight at 4 °C, then stained with 0.1% Alcian Blue (Sigma) in 70% ethanol and 0.37% hydrochloric acid for 4–6 hours at 4 °C. The embryos were cleared in 70% ethanol and 0.37% hydrochloric acid mixture, then rehydrated stepwise in PBS buffer. To enhance optical clarity, embryos were bleached with 3% H2O2 and 1% KOH for 20 min, then washed with PBS containing 0.2% Tween-20, then with PBS, and lastly with H2O. Embryos were stored in 50% glycerol with 0.25% KOH at 4 °C and were mounted in 2% methylcellulose for imaging.
Transcriptional profiling was performed by the Thomas Jefferson University Microarray Facility at the Kimmel Cancer Center. The spotted array contains 16,399 oligonucleotides (Compugen; annotated at http://giscompute.gis.a-star.edu.sg/~govind/zebrafish/version2). More than 100 β-actin oligonucleotides that serve as positive controls were present on each chip.
Zebrafish embryos were injected with phenol red control or 0.5 nl of 1 mM Fhit MO or 1 nl of 1 mM Fhit gripNA. Total RNA of 24 hpf phenol red control and MO-injected embryos were extracted by TRIZOL (Invitrogen). Gene expression was determined using biotin-labeled and in vitro–transcribed antisense RNA generated from the total RNA template. Each chip was scanned and quantified using a ScanArray Express laser scanner (PerkinElmer, http://www.perkinelmer.com). The signals on the oligo microarray were normalized by the median and regularized t-test was performed to determine significant differences between the controls and morphants. The p53 probe used in the microarrays corresponds to a short EST in the 5′ UTR of the gene (U60804) and consequently is common to both full-length and the Δ113 p53 isoform.
Accession numbers for the genes and gene products from the Ensembl D. rerio genome database (http://www.ensembl.org/Danio_rerio/index.html) are β-actin, NM 131031; chordin, NM 130973; Mdm2, NM 131364; nacre, NM 130923; no tail, NM 131162; p21, AL 912410; p53, NM 131327; pax2, NM 131184; smoothened, NM 131027; SP2035, NM 131401; SP2054, BX 901879; SP2063, NM 199847; UROD, NM 131347; and wnt5, NM 130937. |
10.1371/journal.pgen.1007289 | Degenerate Pax2 and Senseless binding motifs improve detection of low-affinity sites required for enhancer specificity | Cells use thousands of regulatory sequences to recruit transcription factors (TFs) and produce specific transcriptional outcomes. Since TFs bind degenerate DNA sequences, discriminating functional TF binding sites (TFBSs) from background sequences represents a significant challenge. Here, we show that a Drosophila regulatory element that activates Epidermal Growth Factor signaling requires overlapping, low-affinity TFBSs for competing TFs (Pax2 and Senseless) to ensure cell- and segment-specific activity. Testing available TF binding models for Pax2 and Senseless, however, revealed variable accuracy in predicting such low-affinity TFBSs. To better define parameters that increase accuracy, we developed a method that systematically selects subsets of TFBSs based on predicted affinity to generate hundreds of position-weight matrices (PWMs). Counterintuitively, we found that degenerate PWMs produced from datasets depleted of high-affinity sequences were more accurate in identifying both low- and high-affinity TFBSs for the Pax2 and Senseless TFs. Taken together, these findings reveal how TFBS arrangement can be constrained by competition rather than cooperativity and that degenerate models of TF binding preferences can improve identification of biologically relevant low affinity TFBSs.
| While all cells in an organism share a common genome, each cell type must express the appropriate combination of genes needed for its specific function. Cells activate and repress different parts of the genome using transcription factor proteins that bind regulatory regions known as enhancers. We currently have an incomplete view of how enhancers recruit transcription factors to yield accurate gene activation and repression. This problem is complicated by the fact that most animals contain over a thousand different transcription factors, and each can generally bind multiple DNA sequences. Thus, it is difficult to predict which transcription factors interact with which enhancers. To gain insights into this process, we focused on determining how an enhancer that activates a gene needed to make liver-like cells is regulated in a precise manner in the fruit-fly embryo. We demonstrate that the specific activity of this enhancer depends on weak and overlapping transcription factor binding sites. Furthermore, we demonstrate that computational models that include weak transcription factor interactions yield better predictive accuracy. These results shed light on how DNA sequences determine enhancer activity and the types of strategies that are most useful for predicting transcription factor binding sites in the genome.
| The control of gene expression is fundamental for defining a cell’s identity and ability to respond to environmental cues. At the transcriptional level, cis-regulatory modules (CRMs) act as platforms for transcription factors (TFs) that affect RNA polymerase activity [1, 2]. Hence, the number, organization, and affinity of TF binding sites (TFBSs) within a CRM convert information about cellular context conveyed by TFs into transcriptional activity [1, 3]. A typical strategy for predicting TFBSs is to use a model of TF binding specificity, such as a position-weight matrix, to score sequences and those with higher scores are predicted to have a greater likelihood of being functional TFBSs. However, this approach is called into question by the growing literature that reveals suboptimal TFBSs are often necessary for accurate biological function [4–11].
Evidence supporting biological relevance of suboptimal TFBSs can be summarized using four concepts [4, 12]. First, suboptimal TFBSs are more likely to differentiate between TFs with similar binding preferences. For instance, suboptimal Hox binding sites were empirically identified in the Drosophila shavenbaby (svb) enhancer and the non-consensus nature of these sites was critical to ensure svb is activated by abdominal, but not thoracic Hox factors [5]. Second, suboptimal TFBSs can be more sensitive to context (e.g. TF concentration). In a classic example, Caenorhabditis elegans genes associated with high-affinity PHA-4 TFBSs are expressed earlier in development when PHA-4 levels are low, whereas genes with low-affinity PHA-4 sites are induced by higher PHA-4 levels later in development [6]. Third, TFBS affinity can alter the ability of a TF to either activate or repress transcription. For example, Drosophila Hedgehog-responsive CRMs with a cluster of low-affinity Ci TFBSs activate transcription, whereas increasing the affinity of Ci TFBSs resulted in repression [7]. Fourth, CRM specificity may depend on suboptimal interactions between TFs. For instance, reporter assays interrogating the Otx-a enhancer in Ciona revealed suboptimal spacing between TFBSs promote enhancer specificity [8, 9]. These studies collectively demonstrate that low-affinity interactions between TFs and CRMs play an important role in accurate transcriptional regulation.
Since TFs have degenerate binding preferences and suboptimal sites are often biologically relevant, predicting functional TFBSs from background sequence is challenging. TFBS-prediction algorithms are typically binary classifiers: sequences are scored using a model of TF binding specificity (e.g. a PWM) and those that meet a threshold are classified as TFBSs. Moreover, the field has largely used arbitrary thresholds as default settings for TFBS-prediction algorithms, such as the 0.8 relative log-likelihood threshold—e.g. a recommended default on the JASPAR website [13]. How well these standard thresholds identify suboptimal TFBSs remains unclear, and the cost of lowering thresholds to identify suboptimal TFBSs (i.e. increased false-discovery rate) is largely unknown.
In this study, we used a well-characterized Drosophila CRM, Rhomboid-BAD (RhoBAD), to assess the role of suboptimal TFBSs for accurate gene regulation and tested the ability of algorithms to predict such sites. The rhomboid (rho) gene encodes a serine protease that triggers secretion of an Epidermal Growth Factor (EGF) ligand [14]. RhoBAD activates rho within specific abdominal sensory organ precursors (C1-SOPs), and thereby induces neighboring cells to form hepatocyte-like cells (oenocytes) essential for animal growth [15–20]. RhoBAD specificity is largely defined by a conserved 47 base-pair sequence (RhoA) that recruits activating and repressing TFs. Indeed, three copies of RhoA are sufficient to recapitulate the abdominal and C1-SOP specific activity of RhoBAD (Fig 1A and 1B) [21, 22]. In the abdomen, an Abdominal-A (Abd-A) Hox factor and the Extradenticle (Exd) and Homothorax (Hth) homeodomain proteins form a complex with the Pax2 TF to activate gene expression [22]. Thoracic segments, however, lack Abd-A expression and thereby allow the Senseless (Sens) TF to bind and repress RhoBAD [17]. Importantly, Pax2 and Sens expression are largely restricted to peripheral nervous system (PNS) cells in all segments. Thus, RhoA integrates both segment (Abd-A) and tissue-specific (Sens and Pax2) inputs to ensure accurate expression in abdominal C1-SOPs (Fig 1C and 1D).
Here, we show that RhoBAD requires overlapping low-affinity TFBSs for Pax2 and Sens to mediate accurate cell- and segment-specific output. Using transgenic reporters and DNA binding assays, we found that increasing Pax2 affinity results in gene activation in additional abdominal PNS cells, whereas increasing Sens affinity results in inappropriate repression. In addition, altering the TFBSs to allow simultaneous binding of activators and repressors impairs RhoBAD activity. Testing available TF binding preference models, however, revealed high degrees of variability in predicting these low affinity TFBSs. To define the source of this discrepancy, we developed a method that generates hundreds of PWMs by selectively sampling TFBSs based on predicted affinity. Surprisingly, we found that PWMs created from datasets depleted of high affinity sites were more accurate at predicting both low- and high-affinity Pax2 and Sens TFBSs from bacterial 1-hybrid (B1H), protein binding microarray (PBM), and mammalian ChIP-seq data than PWMs derived from high affinity sites. Altogether, these findings provide new insights into the functional roles of low affinity DNA binding sites and our ability to use computational approaches to identify TFBSs in complex datasets.
While empirical studies showed that five different transcription factors directly regulate RhoBAD, PWMs derived from published SELEX-seq assays [23] fail to predict the Sens and Pax2 TFBSs using the 0.8 relative-to-range log-likelihood (RLL) threshold (default setting on JASPAR [13]) (Fig 2A and 2D and S1 Fig). This finding suggests that the RhoA Sens and Pax2 TFBSs are low affinity and that the PWMs developed using these in vitro assays maybe too restrictive to accurately predict such functional low affinity TFBSs. To ascertain how affinity correlates with PWM RLL scores, we used electromobility shift assays (EMSAs) with purified Sens and Pax2 proteins to compare RhoA binding to nine randomly selected Pax2 and Sens sites from a published bacterial-1-hyrbid (B1H) study [24] (Fig 2B and 2E and S2 Fig). The selected B1H sites have a large range of RLL scores (Fig 2A and 2D) and were placed in the context of RhoA to maintain consistent flanking nucleotides. EMSAs revealed that the PWMs performed well in ranking TFBS affinity with Spearman’s rank correlations (ρ) of 0.65 and 0.85 for Pax2 and Sens, respectively, between predicted and observed binding (see Methods for details [25–27]) (Fig 2C and 2F). Moreover, these results revealed that the RhoA Pax2 and Sens sites (red in Fig 2C and 2F) were relatively low in affinity compared to the B1H sites (Fig 2B, 2C, 2E and 2F).
To determine if RhoBAD activity depends on a low-affinity Pax2 TFBS, we altered the Pax2 site to better match the consensus motif (RhoA-PS, Fig 3A). EMSAs using Pax2 confirmed a greater affinity for RhoA-PS than wildtype RhoA, without affecting Sens or Exd/Hth/AbdA binding (Fig 3B, and S3 Fig). Next, we integrated RhoBAD-lacZ and RhoBAD-PS-lacZ into identical loci and performed quantitative analysis on age-matched Drosophila embryos. Like RhoBAD-lacZ, RhoBAD-PS-lacZ drives high β-gal levels in abdominal C1-SOPs and weak levels in thoracic C1-SOPs, but with a small, statistically significant increase in all segments (Fig 3C–3E). In addition, RhoBAD-PS-lacZ embryos inappropriately increased β-gal expression in non-C1 PNS cells (Fig 3D, arrowheads). To determine if the ectopic activation of RhoBAD-PS reaches an "abdominal C1-SOP-like" level of activity, we defined a threshold equal to the 5th percentile of wild type RhoBAD-lacZ abdominal C1-SOP intensity and above the 100th percentile of thoracic C1-SOP intensity (red line in Fig 3E–3E'). Using this threshold, we found that β-gal is ectopically expressed in over 5 times more PNS cells in RhoBAD-PS-lacZ than RhoBAD-lacZ embryos (Fig 3F). As a control, no difference in intensity of Sens staining was observed in these embryos (S4 Fig). Thus, strengthening Pax2 binding results in increased RhoBAD activity within C1-SOPs as well as in additional abdominal PNS cells.
To assess if the increased activity of RhoBAD-PS can have functional consequences, we developed an oenocyte rescue assay. In the absence of rho, no EGF signal is sent from abdominal C1-SOPs, and thus adjacent ectodermal cells fail to differentiate into oenocytes [15, 17]. However, rho mutant embryos (rho7M) carrying a wild type RhoBAD-rhocDNA transgene can substantially rescue oenocyte (HNF4+) formation (Fig 3G) [18, 28]. Consistent with RhoBAD-PS having increased reporter activity, RhoBAD-PS-rhocDNA induced a significant increase in oenocyte numbers (Fig 3H and 3I). Moreover, analysis of rho heterozygous embryos carrying RhoBAD-PS-rhocDNA revealed that 30% (3/10) of the embryos had at least one segment with ectopic oenocytes whereas none were observed in embryos with wild type RhoBAD-rhocDNA (arrowheads in Fig 3J and 3K). Altogether, these oenocyte rescue data are consistent with RhoBAD-PS driving increased EGF signaling, potentially via non-C1-SOP cells.
Previous studies have found that increasing Sens TFBS affinity (RhoA-SS, Fig 4A) is sufficient to decrease abdominal RhoBAD activity [17] (Fig 4E). However, this experiment was performed prior to the discovery of an overlapping Pax2 site [22], and EMSAs reveal RhoA-SS not only increases Sens binding but also decreases Pax2 binding (Fig 4B and 4C). Hence, loss of RhoBAD-SS activity could be due to either increased repressor binding (Sens) or decreased activator binding (Pax2). To distinguish between these possibilities, we first compared the activity of RhoBAD-PM (which decreases Pax2 binding while leaving Sens binding) to the activity of RhoBAD-SS (which decreases Pax2 binding while simultaneously increasing Sens binding) (Fig 4B and 4C). Importantly, neither change substantially affects Exd/Hth/AbdA binding (S3 Fig). Comparative analysis of embryos with RhoBAD-PM-lacZ reveals a significant decrease, but not a complete loss, of abdominal C1-SOP activity; whereas RhoBAD-SS-lacZ embryos have a severe loss of activity in both abdominal and thoracic C1-SOPs (Fig 4E, 4F and 4H and S5 and S6 Figs). These results indicate that the lack of RhoBAD-SS activity is largely due to increased Sens binding, rather than loss of Pax2 binding. As a second test, we engineered a RhoA sequence with high affinity sites for both Sens and Pax2 (RhoA-PSSS). EMSAs reveal this sequence enhances Sens and Pax2 binding (Fig 4B and 4C) without affecting Exd/Hth/AbdA binding (S3 Fig). Reporter analysis demonstrates that RhoBAD-PSSS-lacZ embryos have nearly no activity in C1-SOP cells and behave much like RhoBAD-SS-lacZ (Fig 4G and 4H and S7 Fig). These results indicate that a strong Sens binding site can eliminate RhoBAD activity regardless of Pax2 affinity. Moreover, we found that RhoBAD-SS-rhocDNA and RhoBAD-PSSS-rhocDNA transgenes failed to rescue oenocyte development in rho mutant embryos (Fig 4J and 4K). By comparison, the wildtype RhoBAD-rhocDNA transgene significantly induced oenoctye formation in abdominal segments (Figs 3I and 4I). Altogether, these experiments demonstrate that segment-specific RhoBAD activity requires a low affinity Sens TFBS.
We previously found that abdomen-specific activity of RhoA is due to TF competition between a repressor (Sens) and activators (Pax2 plus Exd/Hth/Abd-A) [17, 22]. In this model, Exd/Hth/Abd-A and Pax2 bind RhoA in abdominal C1-SOP cells to both activate transcription and limit the binding of the Sens repressor (Fig 1C and 1D). Thoracic segments lack Abd-A expression, allowing Sens to bind and repress RhoBAD activity in the thorax. Moreover, the data in Fig 4 suggest that competition between activators and repressors is a key feature of regulating output, as raising Sens affinity results in dominant repressor binding and a loss of RhoBAD activity in abdominal SOPs.
To better understand the role of TFBS competition in segment-specific output, we created constructs that uncouple the repressor and activator TFBSs. To do so, we first tested a reporter with the sequences 3' to the Hox site randomly mutated (RhoA-RDM) and found it had similar activity as wild type RhoBAD-lacZ (Fig 5A and 5B and S8 Fig). Hence, this region can be altered without compromising RhoBAD activity. Next, we created a RhoA mutation that abolishes Sens binding (RhoA-SM) (Fig 5A and 5D) without altering activator binding (S3 Fig) [17]. Comparative analysis of RhoBAD-lacZ with RhoBAD-SM-lacZ revealed two expected results: First, loss of Sens binding resulted in a small but significant increase in thoracic expression compared to wild type RhoBAD-lacZ (Fig 5C and S9 Fig). Second, loss of Sens binding was not sufficient to equilibrate the C1-SOP levels between thoracic and abdominal segments. The latter result is consistent with thoracic segments lacking Abd-A, which plays an active role in stimulating transcription [22]. Unexpectedly, however, the Sens mutation also led to a small, but significant loss in abdominal SOP activity through mechanisms that are currently unclear (Fig 5C). Nevertheless, RhoA-SM eliminates Sens binding in vitro and alters RhoBAD activity in vivo, and thereby provides a platform for uncoupling the activator and repressor sites.
To generate sequences that lack TF binding competition, we created RhoA variants that lack an endogenous Sens site (RhoA-SM) and provide a new Sens site downstream of the Hox site. Three variants were tested: (1) a mutant Sens site (RhoA-SM/SM); (2) the wild type low affinity site (RhoA-SM/SWT); and (3) a high affinity Sens site (RhoA-SM/SS) (Fig 5A). EMSAs reveal that while Sens fails to bind RhoA-SM and RhoA-SM/SM, it binds the re-engineered RhoA-SM/SWT similarly to RhoA-WT and binds RhoA-SM/SS with greater affinity (Fig 5D). Importantly, moving the Sens site is sufficient to permit co-binding of activator and repressor TFs in vitro as Sens and Exd/Hth/Abd-A proteins simultaneously bind RhoA-SM/SS but not RhoA-SS, which has overlapping binding sites (Fig 5E).
Comparative analysis of the RhoBAD variants revealed that competition for overlapping TFBSs is essential for proper output (Fig 5F–5J and S10 Fig). First, we found that the re-engineered wild-type Sens site (RhoBAD-SM/SWT) is insufficient to repress reporter activity and behaves similarly as RhoBAD-SM and RhoBAD-SM/SM, which both lack Sens binding (Fig 5F–5H and 5J). Hence, a low affinity Sens site that is uncoupled from the activator sites (Pax2/Exd/Hth/Hox) is unable to repress either abdominal or thoracic SOP activity. In sharp contrast, the re-engineered high-affinity Sens site (RhoBAD-SM/SS) results in gene repression in both thoracic and abdominal SOPs (Fig 5I and 5J). This finding suggests that Sens can inhibit RhoBAD activation through mechanisms other than sterically blocking the binding of activator TFs. Moreover, these findings are consistent with the hypothesis that low affinity Sens sites are required to allow the abdominal Hox and Pax2 activators to stimulate RhoBAD expression. Hence, two features of RhoA are critical to yield segment-specific RhoBAD activity in abdominal C1-SOP cells: 1) low affinity Sens and Pax2 sites are required, and 2) the TFBSs overlap to ensure independent binding of activator versus repressor complexes.
Since CRM studies have increasingly found that low affinity sites are required for accurate output, we next assessed the utility of PWMs to predict such sites for Sens and Pax2. As described above, PWMs derived from SELEX-seq [23] assays for Sens and Pax2 failed to score RhoA above a 0.8 RLL threshold (Fig 2A). Additional published PWMs derived from bacterial-1-hybrid (B1H) assays (FlyFactorSurvey project) are available for Sens and Pax2 [24]. In B1H assays, “hits” are selected when a TF binds a sequence and activates survival gene expression in the presence of an inhibitor, such that increasing inhibitor concentrations select for higher affinity TFBSs [24]. For Sens, B1H assays were previously performed under “high" and “low” stringency conditions, whereas a single Pax2 B1H assay was conducted under low stringency (Fig 6A, top panel) [24]. Comparing the B1H and SELEX-seq PWMs revealed similar motifs, but with differences in their degree of degeneracy (Fig 6A and 6B, top panels). “Degeneracy” can be defined as the inverse of information content, which is measured in bits and represented by letter height in a sequence logo. We used each PWM to score RhoA and found that only the "low" stringency B1H derived PWMs successfully scored the Pax2 and Sens sites above the 0.8 RLL threshold (Fig 6A and 6B, top panel).
To more broadly assess the ability of each PWM to predict both high and low affinity sites, we analyzed published protein-binding microarray (PBM) data for vertebrate homologs of Pax2 and Sens (Danio rerio Pax2b and Homo sapiens Gfi1b) [29, 30]. These homologs share 88% and 85% sequence identity with the Pax2 and Sens binding domains, respectively; and thus, are likely good models for Pax2 and Sens binding specificities [23, 29]. A key advantage of PBM assays is binding strength (as measured by fluorescence intensity) positively correlates with binding affinity, thereby permitting scoring probes across a range of affinities. We scored bound probes of different fluorescent intensities using the B1H and SELEX PWMs, and used the Area-Under-the-Receiver-Operating-Characteristic (AUROC) to measure the ability of each PWM to discriminate bound probes (binned by fluorescence intensity) from non-specific sequences. For non-specific sequences, we randomly selected a matched number of probes from the bottom 50% fluorescence. Note, when AUROC values approach 0.5, the PWMs no longer reliably distinguish bound probes from non-specific sequences. This assessment surprisingly shows that the more degenerate PWMs (“B1H Low Stringency”) are not only more accurate in identifying low-affinity probes, but are also significantly more accurate in identifying high-affinity probes (Fig 6A and 6B, bottom panels).
The above findings suggest that DNA libraries that include low affinity TFBSs (i.e. B1H low stringency assays) produce lower information content PWMs with increased accuracy. To more thoroughly assess how the affinity of binding sites used to generate PWMs affects TFBS accuracy, we developed a method to sub-group B1H hits based on predicted affinity and compare the performance of PWMs generated from each sub-group (Fig 7). For this analysis, we hypothesized that B1H hits containing highly represented sequences are more likely to contain high affinity sites. Indeed, the number of times an 8-mer appears among all B1H hits correlates with the 8-mer E-scores derived from PBMs for related homologues (Spearman’s rank correlation >0.8) (S11 Fig). Therefore, we divided the 542 B1H hits for Sens and the 43,112 B1H hits for Pax2 into quartiles based on 8-mer frequency and derived 100 PWMs from each quartile by iteratively sampling 50 B1H hits from each quartile followed by the MEME algorithm [31] to generate PWMs (Fig 7A, see Methods for details). In such a manner, B1H hits within higher quartiles contain more highly represented sequences, and thus, generate PWMs with greater information content (Figs 7A and S12). As a control, we created a set of 100 PWMs by iterative sampling 50 B1H sequences from the unfiltered dataset (“control PWMs”). To determine how well PWMs from each quartile predict TFBSs compared to the control PWMs, we assessed their ability to score RhoA (Fig 7B), discriminate B1H hits from shuffled sequences (Fig 7C), and score PBM probes bound by the vertebrate Gfi1b and Pax2 factors (Fig 7D). We found that relatively low-information content PWMs from Quartile 2 performed significantly better than the control PWMs when scoring the RhoA sequence (Fig 7B). In contrast, PWMs generated using highly represented sequences (Quartiles 3 and 4) scored RhoA significantly worse than Quartile 2 and control PWMs. Moreover, a similar trend was found when assessing the accuracy of PWMs to discriminate B1H hits from shuffled sequences (Fig 7C) or when scoring PBM probes binned based on fluorescent intensity (Fig 7D). For example, Quartile 3 and 4 PWMs for both Pax2 and Sens have lower AUROC values for discriminating B1H hits from shuffled sequences and for discriminating even the most highly bound PBM probes compared to lower information content PWMs derived from either Quartile 2 or control PWMs (Fig 7C and 7D). It should be noted, however, that the very low information content PWMs from Quartile 1 do not perform as well, as evidenced by wider variance for RhoA score predictions (Fig 7B) and significantly lower AUROC values (Fig 7C and 7D).
Lastly, we investigated the ability of PWMs from each quartile to predict DNA binding events in cells using ChIP-seq data. While high quality ChIP-seq data are not available for Pax2 in either vertebrates or Drosophila, several published ChIP-seq experiments have been conducted for Gfi1 and Gfi1b in mammalian cells [32–40]. Using 10 published H. sapiens and M. musculus Gfi1 and Gfi1b ChIP-seq datasets, we assessed the ability of Sens PWMs to discriminate ChIP-seq peak sequences from an equal number of random, non-repetitive genomic DNA sequences. Initially we analyzed all called ChIP-seq peaks in each dataset and found that Quartile 2 PWMs out-performed all other PWMs in 7 of the 10 ChIP datasets, whereas all of the PWMs performed poorly (AUROC close to 0.5) on the remaining 3 ChIP datasets (S13 Fig). We next asked if the PWMs derived from higher affinity B1H hits (Quartile 3 and 4 PWMs) would perform better when analyzing only the strongest ChIP peaks. Therefore, we binned the top one thousand ChIP peaks from two representative ChIP-seq datasets [32, 33] based on fold enrichment and analyzed the bins separately. As expected, there is a general trend for all the PWMs to perform better as fold enrichment of ChIP peaks increases (Fig 7E). Interestingly, Quartile 2 PWMs out-performed almost all other PWMs in predicting ChIP peaks, regardless of fold enrichment (Fig 7E). Moreover, PWMs derived from the high-affinity B1H hits (Quartile 3 and 4 PWMs) had significantly less discriminatory power, even when predicting the most highly enriched ChIP-peaks. Thus, these findings suggest that, at least for Sens/Gfi1, using more degenerate PWMs derived from lower affinity sites better predicts TFBSs from both in vitro (B1H and PBM assays) and in vivo DNA binding data (RhoA and ChIP-seq).
In this study, we used quantitative DNA binding and transgenic assays to interrogate how Sens and Pax2 TFBSs contribute to cell- and segment-specific CRM activity and thereby EGF signaling in the Drosophila embryo. Our findings reveal that RhoA requires overlapping low-affinity TFBSs to accurately regulate transcription in abdominal SOP cells. In addition, we performed a computational analysis to interrogate the effectiveness of different PWMs for distinguishing TFBSs from background sequences. Taking the B1H, PBM, and ChIP-seq analysis together, our results demonstrate that low-information content PWMs better identify Sens and Pax2 TFBSs. Overall, these findings have important implications for two areas of biology: How CRM composition contributes to transcriptional outcome, and the properties of PWMs that best predict biologically meaningful TFBSs.
CRMs consist of TFBSs that integrate numerous inputs to determine transcriptional output. Three primary models for how CRMs regulate expression have been proposed [3] (A) the “Flexible Billboard” model posits that each TFBS independently recruits a TF that contributes to transcription in an additive manner, and thus the arrangement of TFBSs is of little importance [41]. (B) The “TF Collective” model posits that TFs work cooperatively, but that protein-protein interactions between TFs allow for flexible TFBS arrangement [42, 43]. (C) The “Enhanceosome” model posits that TFs form cooperative complexes that are constrained by the arrangement of TFBSs [44, 45]. While a few CRMs have been categorized according to these models, it is currently unclear what proportion of CRMs each of these models represent.
Our study reveals that the arrangement of TFBSs is important for the activity of the RhoBAD CRM. However, unlike the enhanceosome, which is constrained by cooperative TF complex formation, RhoBAD is instead constrained by competition for overlapping TFBSs. Hence, uncoupling repressor and activator TFBSs (RhoA-SM/SWT) in Drosophila melanogaster results in abnormal activity. This finding is consistent with a mechanism of steric exclusion, which thereby constrains the locations of the TFBSs. In fact, comparative analysis of the RhoA sequence across numerous Drosophilid species suggests that low affinity and overlapping Sens and Pax2 TFBSs are a conserved feature of rhomboid regulation (S14 Fig). Moreover, we found that the affinity of the overlapping sites is tuned to yield appropriate cell- and segment-specific outputs. Specifically, we show that enhancing Sens affinity to RhoBAD results in the loss of activation in abdominal segments, whereas increasing Pax2 affinity increases activation in a subset of ectopic PNS cells. In this way, RhoBAD combines features of two previously studied CRMs: the sparkling enhancer, which requires low affinity Suppressor of Hairless (Su(H)) sites for cell-specific expression in the Drosophila eye [46], and the shavenbaby enhancer that requires low affinity Hox sites to generate segment-specific outcomes in the Drosophila abdomen [5]. Hence, the overlap of low affinity sites for TFs expressed in the PNS (Sens and Pax2) and the abdomen (Abd-A) yields both cell- and segment-specific RhoBAD activity.
Our studies also have implications for how TFBS affinity can affect the mechanism used by a TF to ensure appropriate outputs. For example, Sens can repress RhoBAD activity via a high affinity TFBS, even if it does not overlap the activator sites. This finding indicates Sens uses a repressive activity that is not solely dependent on steric exclusion of activators. These results are consistent with studies demonstrating that the mammalian Sens homologues (Gfi1/Gfi1b) recruit repressive chromatin remodelers, such as HDAC-1 [47]. Moreover, Sens-mediated repression is dominant over a strong Pax2 TFBS, as demonstrated by the lack of activity of RhoBAD-PSSS (Fig 4). Hence, a low affinity Sens site and overlap with activator TFBSs are both required for proper CRM output, suggesting that these requirements constrain the ability of RhoBAD to tolerate sequence changes.
A common approach to predict functional TFBSs within CRMs has been to use large-scale in vitro DNA binding data from assays such as B1H, SELEX-seq, or PBM to create models of TF binding specificity [23, 24, 29, 48–50]. In addition, in vivo approaches, such as ChIP-seq or DamID assays, have been increasingly used to create PWMs from cells and tissues [40, 51]. Hence, for many TFs, several PWMs have been generated using data from different methods (see the CIS-BP database [29]); and for the biologist that wants to predict TFBSs, this raises the issue of which PWMs are best suited to identify functional TFBSs from genomic datasets?
Different PWMs for a given TF typically share common core sequences, but often vary in information content (i.e. degeneracy). A recent study compared a variety of algorithms to generate PWMs and found that in general those with lower information content performed better in predicting TFBSs [50]. Consistent with these results, we found an inverse relationship between PWM information content and accuracy for Sens and Pax2 –TFs from two distinct families (C2H2 Zinc Finger and Paired-Box TFs, respectively). Moreover, this relationship was observed with both published PWMs (Fig 6) as well as by selective sampling of B1H hits to create hundreds of PWMs (Fig 7). For example, by systematically comparing Sens PWMs generated using TFBSs of different predicted affinities, we determined that eliminating high affinity sites resulted in PWMs with increased predictive accuracy for both in vitro (B1H and PBM) and in vivo (RhoA and mammalian ChIP-seq data) DNA binding events. In contrast, using only B1H hits predicted to be of high-affinity resulted in over-representation of certain sequence motifs and, consequently, high-information content PWMs with poor accuracy. Thus, regardless of which DNA binding assay (SELEX, B1H, or PBM) is used to generate a library of sequences, care must be taken to ensure the library is sufficiently diverse to create PWMs that can accurately identify both low and high-affinity TFBSs. However, our approach also highlights that increasing PWM degeneracy has its limits, as highly degenerate PWMs created using the least-represented sequences (Quartile 1, Fig 7) resulted in highly erratic predictions. This finding may be due to the least represented sequences containing rare binding events and/or false positive sequences.
While TFs can interact with the genome over a range of affinities [52, 53] and CRMs with low-affinity sites have been identified [5–9, 45, 54–56], the prevalence of low-affinity interactions between TFs and DNA remains unclear. While our study does not definitively address this question, our analysis of Gfi1 and Gfi1b (mammalian Sens homologues) in vivo binding found that the same low information content PWMs that best discriminated RhoA, B1H hits, and PBM data from random sequences also performed significantly better at identifying potential TFBSs within ChIP-seq peaks (Fig 7). It should be noted that a consequence of degenerate DNA binding is that the number of high-affinity TFBSs within a genome are likely to be greatly outnumbered by the number of low affinity sequences in the genome. Moreover, protein-protein interactions between TFs can modify binding preferences [48]. Therefore, less restrictive models of TF binding may have greater accuracy for identifying TFBSs within ChIP-seq peaks because low affinity sites and modified binding preferences are less penalized than they are by more restrictive models of TF binding.
RhoBAD mutations (see sequences in Figures) were created using site-directed mutagenesis–primers available on request. Each mutation was cloned into pLacZattB, confirmed by DNA sequencing, and integrated into the same genomic location (51C) using ΦC31 (Rainbow Transgenics Inc.).
For quantitative expression analysis, all embryos were harvested, fixed, immunostained, and imaged under identical conditions. Each variant transgene was compared directly with an appropriate control: RhoBAD-lacZ or RhoBAD-SM-lacZ (in Fig 5J). The primary antibodies used were: Abd-A (Guinea Pig 1:500, [17]), Sens (Rat 1:125, [57]), β-gal (Chicken 1:1000, Abcam), and Pax2 (Rabbit 1:2000, [58]). Secondary antibodies conjugated to Alexa-Fluor molecules were purchased from Molecular Probes. Imaging was performed using a Nikon A1 LUNA inverted confocal microscope. Z-Stacks were mean-projected using Fiji (Bioformats plug-in to read ND2 files) [59–61]. NIS-Elements software was used to segment and quantify β-gal intensities in C1-SOPs. Raw measurements used to create graphs are provided in S6 Data.
His-tagged proteins were purified from BL21 cells using Ni-chromatography as previously described [62]: Abd-A [63]; Exd/Hth heterodimers [62]; Sens [17]; and Pax2 [22]. Proteins were confirmed using SDS-PAGE and Coomassie staining and concentrations measured by a Bradford assay. EMSAs were performed using native polyacrylamide gel electrophoresis [43, 64]. Probes were used at 0.36 μM, and protein concentrations were noted in figure legends. Acrylamide gels were imaged using the LICOR Odyssey CLx scanner and densitometry was performed using ImageJ. All quantitative EMSAs were done in triplicate. Predicted binding (Fig 2C and 2F) was calculated as follows:
PredictedBinding=M1+eE−μ
(1)
E is the PWM Energy Score for the site; μ is the chemical potential, which was derived by fitting to data using gradient descent (0.176 for Pax2 and 1.03 for Sens); and M is a scaling factor (equal to the maximum observed probe bound) [25–27].
Sens and Pax2 PWMs were derived from: (1) B1H PWMs were downloaded from the FlyFactorSurvey website (http://mccb.umassmed.edu/ffs/) [24]; (2) SELEX-seq PWMs were downloaded as position count matrices (PCMs) from Nitta et al. [23] (S1 Data). PCMs were converted to PWMs using a custom R script using a pseudo-count of n (where n is the number of observed nucleotides at a position). Sequence logos were created using the ggSeqLogo package for R [65].
To generate PWMs in Fig 7, B1H hits were (1) assigned an affinity score (defined below), (2) placed into quartiles, (3) 50 B1H hits were sampled from each quartile and (4) MEME was used to generate PWMs using the following parameters: -dna -nmotifs 1 –revcomp -mod oops [31]. These parameters indicate that (a) a DNA sequence has been inputted into MEME, (b) a single motif should be found, (c) the reverse complement is analyzed, and (d) the motif occurs only once per sequence. Steps 3 and 4 were repeated 100 times to generate 100 PWMs from each quartile. FlyFactorSurvey B1H hits are 25-mer (Pax2) or 27-mer (Sens) sequences. To calculate a predicted affinity score, each B1H hit was separated into un-gapped 8-mers and the number of occurrences of each 8-mer in the total pool of B1H hits was determined; the predicted affinity score is equal to the maximum occurrence of all 8-mers composing a B1H hit.
Using custom R scripts, we scored RhoA, B1H, PBM, and ChIP-seq sequences using the relative log-likelihood method [66], as follows:
RelativeLogLikelihood=x−SminSmax−Smin
x=log‑likelihoodscoreSmin=minimumpossiblescoreSmax=maximumpossiblescore
(2)
For all sequences, a sliding window was used to score the forward and reverse strands, and the score assigned is equal to the highest score produced. To allow partial matches to the PWM on the edges of sequence, two ambiguous nucleotides (i.e. “NN”) were added to both ends of each scored sequence. These ambiguous nucleotides receive a score of 0.
PWMs were assessed by their ability to discriminate known binding sites (score B1H hits, ChIP peaks, and high fluorescence PBM probes) from control sequences. This ability to discriminate was measured by the Area-Under-the-Receiver-Operating-Characteristic (AUROC)–a commonly used metric of the ability of a predictor (i.e. a PWM) to differentially score two sets of items (i.e. differentially score B1H hits, ChIP peaks, and high fluorescence PBM probes from control sequences). The more effectively a PWM scores can distinguish sequences containing binding sites from control sequences (regardless of the absolute score the PWM assigns to the sequences) the greater the AUROC value. In all instances, we assessed the AUROC using 10 sets of control sequences and assigned the median AUROC value to the PWM. Control sequences were the same number and length as the positive sequences.
B1H hits (“Unique Raw Sequence”) were downloaded from FlyFactorSurvey (http://mccb.umassmed.edu/ffs/) [24]. Control B1H sequences were created through mono-nucleotide shuffling.
H. sapiens and M. musculus Gfi1 and Gfi1b ChIP-seq peaks were downloaded from the NCBI GEO database as BED files [32–40]. Peaks were trimmed to the central 100 bp. For each ChIP dataset, control sequences were generated by randomly selecting an equal number of genomic loci that did not overlap with peaks in the ChIP dataset and did not overlap with UCSC Repeatmasker regions [67]. For two ChIP datasets (GSE50806 [32] and GSM552235 [33]) the same analysis was repeated using the top 1000 peaks, as defined by the MACS2 “fold enrichment” score [68] (Fig 7). Fold enrichment was calculated by downloading raw reads (SRA projects SRP029908 [32] and SRP002575 [33]); assessing read quality with FASTQC [69] (S2–S5 Data); mapping to the mm10 genome using Bowtie 2 [70]; and calling and scoring peaks using MACS2 [68].
PBM data were downloaded from CIS-BP (D. rerio Pax2b, M1499_1.02) and UniPROBE (H. sapiens Gfi1b, UP00592) [29, 30]. AUROC analysis was done on PBM probes binned by fluorescence (top 0.25%, top 0.5%-0.25%, top 1%-0.5%, top 2.5%-1%, top 5%-1%, top 10%-5%). For each bin, control sequences were generated by randomly selecting PBM probes in the bottom 50% of fluorescence.
Data processing was conducted in R using Bioconductor, Tidyverse, and AUC packages, and plotted using ggplot2, ggpubR, and gridExtra packages [71–77].
|
10.1371/journal.pgen.1002678 | SPE-44 Implements Sperm Cell Fate | The sperm/oocyte decision in the hermaphrodite germline of Caenorhabditis elegans provides a powerful model for the characterization of stem cell fate specification and differentiation. The germline sex determination program that governs gamete fate has been well studied, but direct mediators of cell-type-specific transcription are largely unknown. We report the identification of spe-44 as a critical regulator of sperm gene expression. Deletion of spe-44 causes sperm-specific defects in cytokinesis, cell cycle progression, and organelle assembly resulting in sterility. Expression of spe-44 correlates precisely with spermatogenesis and is regulated by the germline sex determination pathway. spe-44 is required for the appropriate expression of several hundred sperm-enriched genes. The SPE-44 protein is restricted to the sperm-producing germline, where it localizes to the autosomes (which contain sperm genes) but is excluded from the transcriptionally silent X chromosome (which does not). The orthologous gene in other Caenorhabditis species is similarly expressed in a sex-biased manner, and the protein likewise exhibits autosome-specific localization in developing sperm, strongly suggestive of an evolutionarily conserved role in sperm gene expression. Our analysis represents the first identification of a transcriptional regulator whose primary function is the control of gamete-type-specific transcription in this system.
| Stem cells give rise to the variety of specialized cell types within an organism. The decision to adopt a particular cell fate, a process known as specification or determination, requires the coordinated expression of all of the genes needed for that specialized cell to develop and function properly. Understanding the mechanisms that govern these patterns of gene expression is critical to our understanding of stem cell fate specification. We study this process in a nematode species that makes both sperm and eggs from the same stem cell population. We have identified a gene, named spe-44, that is required for the proper expression of sperm genes (but not egg genes). Mutation in spe-44 produces sterile sperm with developmental defects. spe-44 is controlled by factors that govern the sperm/egg decision, and its function in controlling sperm gene expression appears to be conserved in other nematode species.
| Stem cells have provoked tremendous interest because of their unique ability to differentiate into multiple cell types. The specification of a particular cell fate ultimately results in a program of cell-type-specific gene expression, and the identification and characterization of the regulators that mediate these transcriptional programs are a focus of intense research. Of particular note is the class of transcription factors that act as master switches; their activities are sufficient to dictate a particular cell fate by promoting, both directly and indirectly (via the regulation of additional transcription factors), the expression of the suite of cell-type-specific target genes. The canonical example is MyoD; heterologous expression is sufficient to convert a variety of cell types into myoblasts [1]. Master switch genes therefore specify as well as implement cell fate decisions.
The hermaphrodite germline of Caenorhabditis elegans offers an attractive model for investigating the regulation of stem cell fate specification and differentiation. Cell fate is restricted to a binary choice, sperm or oocyte, which greatly simplifies the analysis. The identical cellular milieu fosters the development of both types of gametes. The switch from spermatogenesis to oogenesis is genetically determined, but can be experimentally controlled using various temperature-sensitive mutations (reviewed in [2]) and chemical reagents [3]. Alternatively, germline stem cells can be manipulated to further expand their repertoire of potential fates, as recently demonstrated by their directed transdifferentiation into neurons [4].
The sexual fate of individual germ cells is specified by an elaboration of the same sex determination program that dictates male or hermaphrodite somatic development (reviewed in [5]). In the soma, that program culminates in the terminal regulator TRA-1, a homolog of cubitus interruptus and GLI transcription factors [6], [7]. TRA-1 promotes the hermaphrodite fate and inhibits male fate, and does so by direct repression of a number of transcription factors that, in turn, regulate sex-specific gene expression in a variety of somatic tissues including the intestine [8], the nervous system [9]–[11], the vulva [12], and the tail [13]. TRA-1 thereby acts as a classic master switch in specifying somatic sexual fate.
Within the germline of C. elegans, additional regulators modify this somatic sex determination program, in part to permit the production of male gametes in an otherwise female animal. In uncommitted germ cells, the FEM and FOG proteins function as the ultimate regulators of sexual fate. Sequence homology alone reveals little about the mode of FEM activity: fem-1 encodes a protein with ankyrin repeats, fem-2 a putative serine/threonine phosphatase, and fem-3 a novel protein [14]–. Proteomic analysis has been more enlightening and shown that the FEM proteins are components of a CUL-2-dependent E3 ubiquitin ligase complex that targets TRA-1 for degradation [17]. FOG-1 is homologous to cytoplasmic polyadenylation element binding proteins, and presumably regulates translation of transcripts that govern gamete cell fate [18]. FOG-3 shares homology with the Tob/BTG family of antiproliferation proteins, and functions in both the initiation and maintenance of spermatogenesis [19]–[21].
The output of this germline sex determination program is gamete-type-specific gene expression. Microarray screening has identified thousands of genes that are differentially expressed in the germline during sperm or oocyte development [22], [23]. The functional significance of the observed transcription regulation is validated by the detection of genes known to be required for gamete development. For example, a large number of Spe (spermatogenesis-defective) genes have been isolated in mutational screens for sperm-specific sterility (reviewed in [24]), and essentially all of those genes were classified as sperm-enriched by microarray data. Transgenic studies indicate that transcriptional control is the primary mode of regulation for sperm genes [25].
Although germline sex determination ultimately governs sperm and oocyte-specific transcription, the precise mechanism of that regulation remains enigmatic. The TRA-1 transcription factor is an attractive candidate for the job, but, despite its demonstrated role in the soma, it does not appear to directly mediate sex-specific gene expression in the germline. First, TRA-1 loss-of-function mutants produce sperm and oocytes that are competent for fertilization and embryonic development [26], [27]; the only logical interpretation is that all of the genes essential for sperm and oocyte function continue to be expressed in the absence of TRA-1. Second, epistasis analysis demonstrates that the FOG proteins regulate gamete cell fate subsequent to TRA-1 [28], [29]. That observation is supported by the identification of the fog-3 gene (and, most likely, fog-1) as a direct target of TRA-1 transcriptional regulation [30]. Finally, epistasis experiments likewise indicate that the FEM proteins govern germline sex determination in the absence of TRA-1 [31], [32]; as this function is obviously independent of their role in TRA-1 degradation, it is probable that additional targets exist. Therein lies the dilemma: the known transcription factor TRA-1 cannot directly regulate gamete-specific transcription, but the known downstream regulators do not encode transcription factors.
To date, the only identified regulator of C. elegans germline transcription is the GATA factor elt-1, which targets sperm genes [33]. However, that role does not appear to be its primary function. elt-1 expression is not limited to sperm, and the gene is required for embryonic and larval development in a variety of tissues [34], [35]. The ELT-1 binding element is present in less than 5% of sperm genes, the majority of which comprise the MSP multigene family, and elt-1 expression does not appear to be directly regulated by the sex determination program. Thus, the bulk of gamete-specific transcription is governed by as-yet-unidentified regulators.
This study provides a critical piece of the puzzle by identifying spe-44 as a key regulator of sperm-specific transcription. Deletion of spe-44 results in sperm-specific sterility and is associated with multiple defects in both the cell cycle and developmental programs of sperm differentiation. The spe-44 gene is expressed in the germline at the onset of spermatogenesis, and is regulated by the terminal germline sex determination factors FEM-1, FEM-3, and FOG-1. SPE-44 promotes expression of a large fraction of sperm-enriched genes, and that role is likely conserved among nematodes. Our work highlights the mechanistic differences between sex determination in the soma, where TRA-1 specifies and directly regulates sex-specific transcription, and the germline, in which SPE-44 implements the transcriptional program specified by the FEM and FOG proteins.
A candidate gene approach was used to identify potential regulators of sperm gene expression. Previous microarray data characterized sex-specific transcriptional profiles for both the germline and soma [22], [23]. Eleven homologs of transcriptional regulators were among the 1,343 genes enriched in the sperm-producing germline (Table S1). We further restricted our list of candidates to those for which pre-existing mutations were available for functional characterization, since sperm genes as a group are generally refractory to inactivation by RNA interference [for example, see 33]. We selected C25G4.4 for further study; the gene product is most similar to the mammalian glucocorticoid modulatory element-binding proteins GMEB-1 and -2 and encodes a SAND domain (for Sp100, Aire, NucP41/75, DEAF-1) [36]–[38]. SAND-containing proteins were first identified as transcriptional activators that bind to functional regulatory elements within the promoters of target genes [39], [40]. Solution and crystal structures of SAND domains reveal a novel DNA-binding fold centered on the highly conserved KDWK motif [41], [42]. The C25G4.4 gene product contains all of the conserved residues of the SAND domain, including KDWK, but otherwise possesses no identifiable domains.
The deletion allele ok1400 of C25G4.4 is almost certainly a null allele as it lacks 1577 of the 1797 base-pair coding region, including the conserved SAND domain (Figure 1A). The original strain isolated by the C. elegans Gene Knockout Consortium [43] produced both fertile and sterile progeny in roughly 3∶1 ratio, suggestive of heterozygosity for a locus conferring recessive sterility. When PCR was used to amplify the C25G4.4 interval from individual sterile and fertile hermaphrodites, the sterile phenotype proved to be tightly linked to the deletion allele; all sterile animals were homozygous for ok1400, whereas all fertile animals contained at least one copy of the wild-type gene (Figure 1B). No additional defects in development or morphology were observed in the ok1400 mutant, suggesting that C25G4.4 might function exclusively in the context of reproduction.
In self-fertilizing hermaphrodites, sterility can arise from defects in sperm and/or oocyte function. To discriminate among those possibilities, we set up reciprocal matings between ok1400 mutants and wild-type animals of the opposite sex (Figure 1C). Crosses between self-sterile ok1400 homozygous hermaphrodites and wild-type males yielded viable progeny, indicating that ok1400 hermaphrodites produce functional oocytes. By contrast, crosses between ok1400 homozygous males [marked with the tightly linked dpy-20(e1282) allele] and fem-1(hc17) hermaphrodites (which lack sperm) produced no progeny, whereas fem-1 hermaphrodites mated with homozygous dpy-20(e1282) males produced abundant outcross progeny. Microinjection of the C25G4.4 transgene rescued the sterility of unmated ok1400 homozygous hermaphrodites (Figure 1C, +Txgene), thereby confirming that the deletion allele is responsible for the observed sterility. Together these results indicate that the sterility of ok1400 results from a defect in sperm function. Based on the sperm-specificity of its sterile phenotype, the C25G4.4 gene was designated spe-44 (for spermatogenesis-defective).
Sperm cell development in C. elegans has been well characterized in both sexes [44]–[46]. Throughout the larval and adult stages, Notch signaling in the distal end of the gonad maintains a population of mitotically dividing germline stem cells with the potential to generate either oocytes or sperm [2]. During the third larval (L3) stage, the most proximal of these mitotic germ cells enter the meiotic cell cycle, an event that roughly coincides with the commitment of these cells to spermatogenesis [28]. By the fourth larval (L4) stage in both hermaphrodites and males, the various stages of spermatogenesis can be observed in a distal to proximal array within the gonad. Transcription of sperm genes typically initiates during the pachytene phase of the cell cycle [e.g., see 33]. Following the disassembly of the synaptonemal complex, the spermatocytes enter a karyosome stage during which global transcription ceases [46]. The primary spermatocytes then detach from the gonadal syncytium and undergo the two meiotic divisions. Cytokinesis during meiosis I can be complete or incomplete, producing secondary spermatocytes that contain one or two nuclei, respectively. Immediately following anaphase of meiosis II, cellular components that are not required for subsequent sperm function are partitioned away from the spermatids in a budding division and deposited in a residual body. In males, sperm production continues through adulthood, and immotile, spherical spermatids accumulate in the seminal vesicle. Upon insemination of hermaphrodites, the process of activation converts spermatids into mature, crawling spermatozoa that migrate from the uterus to the spermatheca. In hermaphrodites, gametogenesis switches abruptly from sperm to oocyte production at the L4/adult transition. Spermatids undergo activation into motile spermatozoa as they are propelled into the spermatheca ahead of the newly formed oocytes.
To characterize potential spermatogenesis defects, we examined adult male gonads, in which all stages of sperm development are readily visible, under differential interference contrast (DIC) optics. In wild-type gonads, spermatogenesis progressed in a distal-to-proximal array through the pachytene and karyosome stages of meiosis, followed by a meiotic division zone and then a region of packed spermatids (Figure 2A). The spe-44 gonads (Figure 2B) were similar in overall size, suggesting that germline proliferation occured normally. Developing spe-44 spermatocytes progressed appropriately through the pachytene and karyosome stages, although the mutant gametes were less refractile and thus smoother in appearance. However, the division zone was expanded, the cells misshapen, and normal spermatids failed to accumulate.
These differences were even more apparent in sperm spreads, which distinguish individual spermatocytes. In wild-type samples, the meiotic division zone could be further discriminated into spermatocytes undergoing anaphase I, anaphase II, and the post-meiotic budding division of spermatids from a residual body; numerous spermatids are also visible (Figure 2C). In the spe-44 samples, dividing spermatocytes often exhibited a variety of cleavage defects, including unequal divisions and the formation of multiple partially budded structures, and no budding divisions of residual bodies were observed (Figure 2D–2E). Aberrant spermatogenesis was also observed in hermaphrodites (Figure S1). These morphological defects prompted a more detailed characterization of the spe-44 phenotype.
To that end, we examined the chromatin and microtubule dynamics in male gonads (Figure 2F–2G). Similarly to DIC, no differences were observed between the germlines of wild-type and spe-44 males through the karyosome stage. The meiotic division zone in wild-type gonads contained a small number of spermatocytes with visible asters and a few cells undergoing the budding division (in which tubulin segregates to residual bodies), followed by a large number of spermatids (Figure 2F). By contrast, spe-44 gonads exhibit a greatly expanded meiotic division zone, suggesting a cell cycle arrest (Figure 2G).
Sperm spreads revealed a myriad of defects in the spe-44 spermatocytes within the division zone (Figure 2H). Although the nucleation of asters initiated normally during diakinesis (Dia), the asters were often larger and broader than their wild-type counterparts. The asters migrated correctly to set up the metaphase I spindle (MI), but chromosome segregation was aberrant with most or all chromosomes segregating to one spindle pole (MI-ab), particularly during the second meiotic division (MII-ab). Examples of unequal chromosome segregation and extensive aneuploidy were also evident in arrested spermatocytes (Arr), and some of the smallest cells had microtubules but no chromatin (data not shown). As in wild-type spermatogenesis, both complete and incomplete cytokinesis occurred during the meiotic divisions; however, we frequently observed multiple spindles (MS), as many as eight per spermatocyte, instead of the typical two or four. The presence of these supernumerary asters is suggestive of spindle overduplication and/or fusion of terminally arrested spermatocytes. The asters in these terminal spermatocytes were almost always found adjacent to the plasma membrane, and the microtubules in these spermatocytes were often unusually long. Ultimately, the mutant spermatocytes appeared to lyse; the number of arrested spermatocytes did not increase dramatically with age, and cellular debris accumulated within the gonad (data not shown).
Previous microarray data indicated that transcription of spe-44 is elevated during spermatogenesis [22], [23], so we utilized quantitative RT-PCR to assess the dynamics of spe-44 transcription during development. Expression of spe-44 was first detected in age-synchronized populations of wild-type hermaphrodites at the L3 larval stage, when gametes become committed to the sperm cell fate (Figure 5A). Transcript levels of spe-44 declined through the L4 and adult stages, when gametogenesis switches from sperm to oocyte production. In wild-type L4 and adult males, which produce sperm continuously, spe-44 expression levels were higher than in hermaphrodites at the same stages (hermaphrodites and males are morphologically similar prior to L4, so males were not assessed at earlier larval stages).
To determine the role of the germline sex determination program in spe-44 expression, we employed hermaphrodites that contained temperature-sensitive mutations in the terminal FEM and FOG regulators. When reared at the restrictive temperature, fem-3(gf) hermaphrodites produce only sperm. Expression of spe-44 was observed beginning at the L3 stage and, although the levels declined somewhat, remained elevated into adulthood (Figure 5A). Conversely, the fem-1(lf) and fog-1(lf) mutations, which cause hermaphrodites to make only oocytes, resulted in low spe-44 expression levels at all stages of development. Thus, spe-44 transcript accumulation correlates strongly with the production of sperm. Furthermore, peak expression is observed during L3 at the onset of sperm fate specification. This pattern stands in marked contrast to that observed for other sperm genes. For example, MSP, which is abundantly expressed, is not detectable until L4 [50]. Likewise, a developmental time course of global transcription demonstrates that essentially all sperm genes exhibit a sharp increase at the onset of L4 and corresponding decline at the transition to adulthood [22], [23]. Thus, the peak expression of spe-44 during L3 precedes that of other sperm genes, and is consistent with a role in the regulation of sperm gene expression. Finally, spe-44 expression is governed by terminal regulators of the germline sex determination pathway (fem-1, fem-3, and fog-1) that specify sperm or oocyte fate.
In situ hybridizations of dissected gonads were performed to ascertain the precise spatial pattern of spe-44 expression. The fem-3(gf) and fem-1(lf) mutations were used to restrict hermaphrodite gametogenesis to sperm or oocyte production, respectively. Abundant expression was first detected in fem-3(gf) animals during the L3 stage in the early meiotic germline (Figure 5B). DAPI staining of the chromatin indicated that expression coincides with early pachytene of meiosis I (Figure 5C). By contrast, spe-44 expression was undetectable in fem-1(lf) germlines at any stage of development (Figure 5D). Thus, spe-44 expression is restricted to the sperm-producing germline at a time and place consistent with the regulation of sperm gene expression.
Since SPE-44 shares homology with known transcriptional regulators, we performed DNA microarray screening to assess its role in governing gene expression. Comparison between wild-type and spe-44 L4 males at the onset of sperm production revealed statistically significant differences (p<0.05) in gene expression between the two samples. A total of 813 genes exhibited greater than two-fold changes in expression. The levels of 535 genes were reduced in the spe-44 males and 278 were elevated (Table S2).
We compared our spe-44 microarray data to previous transcriptional profiles that classified genes as sperm-enriched, oocyte-enriched, or germline-intrinsic (i.e., expressed in both sperm and oocyte) [22], [23]. Of the 535 genes that were down-regulated in the spe-44 strain, nearly two-thirds (343) were also classified as sperm-enriched (Figure 6). Note that the degree of overlap between the data sets is likely under-represented, in part due to technical differences between the microarray platforms that were employed. By contrast, very little concordance was observed with either the oocyte-enriched (four) or germline-intrinsic (25) classes. Therefore, loss of spe-44 results in a substantial defect in sperm-enriched gene expression. Those genes are hereafter referred to as spe-44 sperm targets.
We performed the same comparison with the 278 genes that exhibited increased expression in the spe-44 strain. No significant overlap was observed with any of the three germline categories (Figure 6). These results indicated that germ cell fate in spe-44 mutants is not shifted from sperm to oocyte; such a fate switch would be accompanied by an increase in oocyte-enriched transcription. Therefore, spe-44 is not a component of the sex determination pathway that specifies gamete cell fate; however, it is required to implement that fate by promoting the transcription of sperm genes.
We also examined the frequency of GO terms associated with the sets of genes that were down or up-regulated in the spe-44 mutant strain. Among the genes with reduced expression levels, three categories were over-represented: protein kinases (3.2-fold higher than expected) and associated terms (e.g., protein amino acid phosphorylation); protein phosphatases (4.3-fold) and associated terms; and structural molecules (8.6-fold) (Table S3). All of these categories are consistent with a defect in sperm gene expression. The overabundance of protein kinases and phosphatases among the sperm transcriptome has been reported previously [22]; the structural molecules largely consist of the MSP and SSP gene families, which are structural components of the sperm motility apparatus (see below). Similar characterization of the genes that were up-regulated in the spe-44 strain was not as informative. The only two over-represented terms were ATP-binding (2.3-fold higher than expected) and intracellular (2.2-fold) (Table S4).
The list of spe-44 down-regulated sperm targets contains a significant number of genes with demonstrable roles in sperm (shown in Table 1). The genes fall into four categories: 1) MSP (major sperm protein) is the structural polymer responsible for nematode sperm motility, and also serves as a signaling molecule to promote ovulation and oocyte maturation [51], [52]. A subset of the multigene MSP family (17 of 28 members) exhibits greater than two-fold reduction in transcript levels in the spe-44 mutant. 2) Members of the small sperm-specific protein family (formerly SSP; designated ssp/ssq/ssr/sss) are structurally similar to MSP and play a role in MSP polymerization [53], [54]; five are found among the spe-44 targets. 3) Spe (spermatogenesis-defective) and Fer (fertilization-defective) genes have been identified in genetic screens for sperm-specific sterility [55]–[57]; four Spe genes are targets of spe-44. 4) The elt-1 gene, which encodes a GATA transcription factor that regulates the transcription of a subset of sperm genes [33], is down-regulated in the spe-44 mutant strain. Because all of these genes have known functions within sperm, the reduced transcript levels observed in the spe-44 mutant may contribute to the variety of defects that occur during spermatogenesis (see Discussion).
The target genes identified by microarray might be directly or indirectly regulated by SPE-44. The multigene MSP family is likely to fall within the latter category. Prior work identified the GATA transcription factor ELT-1 as a direct regulator of MSP expression [33]. Our microarray data indicate that both elt-1 and the MSP genes are expressed at lower levels in the spe-44 mutant. The simplest model is a transcriptional cascade in which SPE-44 promotes elt-1 expression, whose product in turn promotes MSP transcription. Alternatively, SPE-44 might work in conjunction with ELT-1 to promote maximal levels of MSP transcription and/or to restrict expression to sperm.
We tested the ability of SPE-44 to directly promote transcription in a heterologous system, using a variant of the yeast one-hybrid assay [58]. We constructed a yeast lacZ reporter gene that contained the putative promoter region of spe-7 (Pspe-7::lacZ), a sperm target gene that is strongly down-regulated in the spe-44 mutant strain. Expression of spe-44 was controlled by the galactose-inducible GAL1 promoter. Induction of spe-44 resulted in Pspe-7::lacZ expression as measured by ß-galactosidase activity (Figure 6B). Expression of Pspe-7::lacZ was dependent upon spe-44, as ß-galactosidase activity was not detectable under non-inducing conditions. We also examined the specificity of SPE-44 transcriptional activation for its target promoter. No expression was observed from a lacZ reporter that contained the promoter region of spe-4, a sperm gene that is not a target of spe-44. Therefore, SPE-44 can function as a transcription factor to directly activate gene expression from its cognate promoter.
To determine if the SPE-44 protein localized to chromatin in vivo, we generated a SPE-44 antibody and stained dissected gonads. The timing and distribution agreed with our previous transcriptional analysis, and correlated perfectly with sperm production. Temporally, SPE-44 labeling was first detectable in L3 male germlines and persisted through adulthood (Figure S2 and Figure 7A). Spatially, SPE-44 labeling was absent in the distal stem cell niche and mitotic zone, initiated in the early meiotic zone, co-localized with chromosomes in the pachytene region, and became non-chromosomal before disappearing altogether in karyosome stage nuclei (Figure 7A; see inset for details). The observed staining pattern is dependent upon the SPE-44 protein, as it is absent in the spe-44 mutant strain (Figure S3).
Closer examination revealed that SPE-44 localized along the length of most but not all of the pachytene chromosomes (Figure 7A, inset, and 7B). The unpaired X chromosome was a likely candidate for SPE-44 exclusion: X is singularly devoid of sperm genes, and contains chromatin modifications consistent with transcriptional silencing [22], [59]. To test this hypothesis, we compared the distribution of SPE-44 and histone H3(lysine 9) dimethylation, a chromatin modification that specifically labels the X chromosome in the male gonad (Figure 7B). Co-incubation with both antibodies labeled all chromosomes, indicating that the chromosome that is unlabeled by SPE-44 is indeed the X (Figure 7B). The localization of SPE-44 to autosomes (which contain sperm genes) but not X (which does not) is consistent with a role as an early acting and positive regulator of sperm gene transcription.
For sperm targets of SPE-44, we predict that their distribution would be coincident with or subsequent to the appearance of SPE-44. Therefore, we examined the localization of SPE-44 in comparison to MSP, an abundant sperm-specific marker that is a target of SPE-44 (Table 1). Co-staining clearly demonstrated that SPE-44 is detectable prior to the accumulation of MSP (Figure 7C), which is first observed near the bend of the gonad at mid to late pachytene. Thus, SPE-44 exhibits the properties predicted for a regulator of sperm gene expression: the protein appears at the onset of sperm fate specification prior to the production of sperm proteins, is bound to the chromatin of committed and developing spermatocytes, and disappears in the karyosome stage coincident with the global cessation of transcription.
Labeling of SPE-44 in hermaphrodite gonads likewise correlates with sperm production. The protein is first detectable in L3 at the time of sperm fate specification (Figure S4) and in young L4 hermaphrodites undergoing spermatogenesis (Figure 7D), but is absent in adult hermaphrodites that are undergoing oogenesis (Figure 7E). We were particularly interested in the distribution of SPE-44 during the switch in gamete fate. An intriguing property of the C. elegans hermaphrodite gonad is its ability to cleanly transition from the production of spermatocytes to the production of oocytes without making sexually ambiguous gametes. To test whether SPE-44 might be functioning in this switch, we examined the pattern of SPE-44 localization in hermaphrodite gonads during this developmental transition. Remarkably, in the proximal gonads of hermaphrodites that contained spermatocytes directly adjacent to the first enlarging oocytes, we always observed a few SPE-44 staining nuclei interspersed among these initial oocytes (Figure 7F). Although we have not ruled out the possibility that these are residual SPE-44-positive nuclei displaced from the syncytial germline by enlarging oocytes, we favor the interpretation that these are developing spermatocytes, and that SPE-44 is a marker of gamete sexual fate that is regulated at the level of individual cells. As an aside, we note that anti-SPE-44 labeling of the hermaphrodite nuclei co-localizes with most but not all of the chromatin, and (as demonstrated in males) the unlabelled region presumably represents the two X chromosomes.
All nematode species share an unusual mode of sperm motility (amoeboid crawling) and a novel protein (MSP) that underlies that motility. Similarly, the regulation of sperm gene expression might be conserved among nematodes. Complete genomes are available for five species of the Caenorhabditis group, so those sequences were screened for spe-44-related genes. C. elegans itself contains a paralagous gene, gmeb-3, whereas each of the remaining species contains a single spe-44 homolog. These homologs encode proteins that are more similar to SPE-44 than GMEB-3 and likely orthologous, and the tree agrees well with the current phylogeny of these species (Figure 8A). Sequences related to spe-44 were also identified in more distant nematode species (data not shown); however, conservation was limited largely to the SAND domain, which is represented in several C. elegans genes, and the degree of similarity was insufficient to distinguish the relationship unambiguously. Although the extent of spe-44 conservation across the phylum is unknown, the gene is clearly conserved within the Caenorhabditis group.
An examination of sex-biased transcription in species of the Caenorhabditis group is currently underway (C. Thomas, manuscript in preparation), so those RNA-Seq data were examined for differences in spe-44 expression. In all four species under study (C. elegans, C. remanei, C. brenneri, and C. japonica), the spe-44 ortholog exhibited significant male-biased expression (Figure 8B). Note that the degree of male-biased expression may be greater than indicated by these data, which are derived from adult animals; in C. elegans, peak levels of spe-44 mRNA are observed in L3 larvae and decline through the L4 and adult stages (Figure 5A). Although the developmental profiles of spe-44 transcription in the other species are currently unknown, the data from adults clearly demonstrate that male-biased expression is a conserved property of spe-44.
The conservation of gene expression patterns among species can be reflected in the conservation of promoter elements that regulate transcription. Therefore, we performed multiple sequence alignment of the upstream regions from the five spe-44 orthologs. We discovered three highly conserved sequence elements (Figure 8C, in blue) within 150 nucleotides of the initiation codon, which could potentially represent binding sites for unknown regulatory factors. Thus, the sequence conservation of spe-44 is not limited to the coding sequences but extends to putative promoter elements.
Another distinctive property of SPE-44 in C. elegans is its restriction to meiotic autosomes in developing sperm. To determine if this distribution pattern is likewise conserved, isolated gonads from C. remanei males were immunostained with the anti-SPE-44 antibody (which recognizes a carboxy-terminal epitope whose sequence is conserved among the spe-44 orthologs). As in C. elegans, we observed SPE-44 labeling of the chromatin of pachytene nuclei, with the notable exception of the presumed X chromosome (Figure 8D, arrow in inset). The conservation of gene expression and protein distribution in both male/hermaphrodite (C. elegans) and male/female (C. remanei, C. brenneri, and C. japonica) species strongly suggests that the observed pattern represents the ancestral state among this group of nematodes. Although we have yet to directly demonstrate that spe-44 is required for spermatogenesis and sperm gene regulation in the other species, the degree of sequence conservation within the putative promoter and coding sequence, coupled with male-enriched expression and the pattern of protein distribution in the germline, makes a compelling case for an evolutionarily conserved role.
This work identifies spe-44 as a critical regulator of sperm-specific transcription in the C. elegans germline. Mutation of spe-44 results in sperm-specific sterility, a consequence of aberrant sperm development that includes cell cycle arrest and defects in sperm organelle assembly. The spe-44 gene is expressed in the sperm-producing germline at the onset of sperm fate specification, and its product is required for the appropriate expression of several hundred sperm-enriched transcripts. Targets of spe-44 include a number of genes known to be required for sperm development. The spe-44 gene and promoter sequence, male-biased expression pattern, and germline protein distribution are conserved among nematode species, consistent with an evolutionarily conserved role in the regulation of sperm gene expression. spe-44 lies at the terminus of the germline sex determination pathway, and its primary role is to implement cell fate specification by promoting sperm-specific transcription.
The spe-44 deletion phenotype is complex and multifaceted, as might be expected for an early acting factor that regulates a large number of sperm-specific genes. Sperm cell fate is specified appropriately in spe-44 mutants, but these gametes are unable to properly implement the programs of meiotic cell division and differentiation. Initiation of meiosis appears normal, but the ensuing defects in chromosome segregation suggest that spe-44 spermatocytes either lack, or are unable to properly regulate, key spindle and kinetochore components necessary for chromosome attachment. Ultimately, spe-44 spermatocytes arrest at M-phase and fail to undergo the budding division. The relationship between these two events is unclear. Cell cycle arrest might block progression of the budding division (e.g., via a checkpoint mechanism); alternatively, the segregation of cell cycle regulatory factors (perhaps in association with the meiotic spindles) to the residual body might be required for the completion of meiosis. In either case, the defects in fibrous body and membranous organelle assembly likely contribute to the failure of the budding division. FBs are composed of MSP assembled in a paracrystalline array, and MOs contain a variety of secretory and membrane proteins. FB-MO complexes are thought to facilitate the segregation of these critical components into the budding spermatid and away from the residual body, so the failure of this partitioning event in spe-44 mutants is not surprising.
The observed phenotypes presumably reflect reduced levels of multiple components necessary for specific aspects of spermatogenesis. The spe-44 mutation causes a defect in the expression of several hundred sperm genes, which are obvious candidates for those functions. For example, mutations in the spe-44 target spe-7 are associated with defects in fibrous body assembly similar to those observed in spe-44 mutants [M. Presler, K. Messina, and D. Shakes, pers. comm.]. The reduction in MSP expression might also contribute to this phenotype. However, any effect may be modest, as the spe-44 mutation targets only a subset of the MSP genes and their expression is reduced but not abrogated. Additional targets of spe-44 that might contribute to the membranous organelle assembly defects include spe-10 and spe-17, which are required for both the structural integrity of the membranous organelles and the proper segregation of components into the residual body during spermatid formation [60], [61]. Similarly, M-phase arrest might reflect reduced expression of the spe-44 target CDC14, a phosphatase that normally functions to inactivate cyclin-dependent kinase and promote M-phase exit [62].
SPE-44 was selected as a candidate transcriptional regulator on the basis of its SAND domain. SAND domain proteins exhibit intrinsic transcriptional activation ability in a variety of reporter assays, a property shared by SPE-44 (Figure 6C). However, these proteins appear to function primarily through interactions with other transcriptional regulators. In some cases, the binding partners are canonical transcription factors. For example, GMEB-1 and GMEB-2 function with the glucocorticoid receptor to modulate target gene expression [63], while DEAF-1 cooperates with the Hox gene Deformed [40]. SAND domain proteins have also been implicated in chromatin-mediated regulation of transcription. AIRE is part of a large protein complex that includes various chromatin components [64], SP100 protein interacts with HMG and HP-1 heterochromatin proteins [65], [66], and GMEB binds the histone acetylase CBP [63]. In plants, ULTRAPETALA1 has been shown to function in the Trithorax group chromatin remodeling complex to antagonize the repressive effects of histone methylation by the Polycomb complex [67]. The observed distribution pattern of SPE-44 suggests that it, too, interacts closely with chromatin markers; the protein localizes broadly along the length of the autosomes while being specifically excluded from the X chromosome, which is known to contain repressive chromatin modifications. Identification of SPE-44-interacting factors will be crucial in determining the mechanism(s) by which SPE-44 and other SAND-domain proteins regulate gene expression.
SPE-44 is not the sole transcriptional regulator that promotes sperm-specific gene expression. The list of spe-44 targets comprises only a subset of the 1,343 sperm-enriched genes identified previously [22], [23], and a majority of Spe genes with demonstrated roles in spermatogenesis are expressed appropriately in spe-44 animals. Ten additional transcriptional regulator homologs exhibit sperm-enriched expression, and those are potential mediators of sperm gene transcription. Although elt-1 expression is governed by SPE-44, the other transcription factors are not and could promote sperm gene transcription independently of SPE-44. In addition to those candidates, transcription factors that also function in somatic tissues might not have been identified as sperm-enriched but could nonetheless promote sperm gene expression in the germline. Alternatively, the observed sperm-enriched expression of some genes could reflect repression during oogenesis, and negative regulators might be predicted to exhibit oocyte-enriched, rather than sperm-enriched, expression.
The conservation of spe-44, and its presumed role in sperm gene transcription, may be somewhat surprising given the rapid evolution of reproduction within the Caenorhabditis group of nematodes. Hermaphroditism has evolved independently at least twice from the ancestral male/female species [68], [69]. Differences in the germline sex determination programs between C. elegans and C. briggsae include gene loss, gene gain, and reordering of the pathway components [70]–[72]. Nonetheless, the data strongly suggest that spe-44 function has been retained. The orthologous gene exhibits male-biased expression among the male/female species of the group. That expression likely reflects the conservation of sequence elements within the putative promoters of these orthologs. Most remarkable is the conserved pattern of SPE-44 protein localization in the male germline of C. remanei, including binding to the autosomes and exclusion from the presumptive X chromosome. Sperm genes are largely absent from the X chromosome in C. elegans, and chromatin markers indicate that X is transcriptionally silent at the time of sperm gene expression [22], [59]. Although the chromosomal distribution of sperm genes in nematodes other than C. elegans is currently unknown, germline silencing of the X chromosome is evolutionarily conserved among this group [59]. This observation strongly predicts a similar restriction of sperm genes to the autosomes, the site of SPE-44 binding, among these Caenorhabditis species. Taken together, the data are consistent with a conserved role for spe-44 in sperm gene expression.
One unanswered question is how the expression of spe-44 is itself regulated by the germline sex determination pathway. Our quantitative RT-PCR results clearly indicate that, in C. elegans, spe-44 is governed by the terminal components FEM-1, FEM-3, and FOG-1 at the level of transcription and/or mRNA stability. The FEM proteins are part of a CUL-2-dependent E3 ubiquitin ligase complex, so one potential mechanism would be via degradation of a repressor of spe-44 expression. That repressor is unlikely to be TRA-1, despite its demonstrated role as a target of FEM-dependent degradation and a repressor of male sexual fate in the soma. The FEM proteins specify gamete fate in the absence of TRA-1, and the spe-44 promoter region does not contain sequences that match the known TRA-1 consensus binding site [73]. However, there are highly conserved sequence elements upstream of spe-44 that might serve as binding sites for as-yet-unidentified regulatory proteins, which could be targets of FEM-dependent degradation. Similarly, FOG-1 is predicted to govern the translation of unknown factors that specify gamete fate, one or more of which might control transcription of spe-44. Alternatively, the putative polyadenylation-binding activity of FOG-1 might play a more direct role in regulating spe-44 transcript stability.
The timing and specificity of spe-44 expression indicate its critical role in implementing germline sexual fate. SPE-44 is one of the earliest known markers of sperm fate in C. elegans. Both temporally and spatially, SPE-44 precedes the production of other sperm-specific components. The precise, cell-specific pattern of SPE-44 localization during the spermatogenesis-to-oogenesis transition of hermaphrodites predicts that the transcription, translation, and protein stability of SPE-44 must be tightly regulated by a variety of feedback mechanisms. This exquisite restriction of SPE-44 to developing spermatocytes provides a powerful tool for the analysis of mutants that, due to defects in the germline sex determination program, produce gametes of sexually indeterminate character.
C. elegans strains were obtained from the Caenorhabditis Genetics Center, and are derived from the wild-type isolate N2 (Bristol). The spe-44(ok1400) deletion allele was isolated by the Gene Knockout Consortium [43] and backcrossed six times to N2 prior to analysis. Additional mutations include: dpy-20(e1282)IV, fem-1(hc17)IV, fem-3(q20gf)IV, fog-1(q253)I, let-92(s677)IV, unc-22(s7)IV. A linked spe-44(ok1400) dpy-20(e1282) mutant strain, balanced by let-92(s677) unc-22(s7), was generated to facilitate discrimination of homozygous lines. Homozygous spe-44 males were obtained by mating balanced heterozygous males to homozygous Dpy Spe hermaphrodites and picking Dpy male progeny. Sodium hypochlorite treatment of gravid adults was used to obtain embryos for age-synchronized analyses. Strains were propagated on OP50-seeded NGM plates and maintained at 15°C unless otherwise indicated. Genetic manipulations were carried out by standard methods [74]. Self-fertility assays were performed by total progeny counts of individual hermaphrodites; cross-fertility was assessed by mating individual hermaphrodites with four males for 24 hours, then counting total progeny.
Single-worm PCR detection of the spe-44(ok1400) deletion utilized primers HES-343, HES-364, and HES-502 (all primers listed in Table S5). Microinjection rescue was performed with plasmid pHS584, a 6.5 kbp fragment of the spe-44 genomic interval; included in the microinjection mix were plasmid pRF4, which contains the dominant rol-6(su1006) as a morphological marker, and genomic DNA, which enhances germline expression by formation of complex arrays [75], [76]. Rescue was determined by injecting balanced spe-44(ok1400) dpy-20(e1282)/let-92(s677) unc-22(s7) hermaphrodites, picking individual dumpy rollers from the F2 generation grown at 25°C, and counting total progeny. For qPCR, RNA was isolated via Trizol (Invitrogen) from triplicate samples (each ∼500 animals) grown at 25°C at the indicated developmental stage, reverse-transcribed with Superscript II (Invitrogen), and amplified with primers HES-502 and HES-539 (for spe-44) or AKA-70 and AKA-71 (for act-1). For yeast expression assays, the spe-44 cDNA was amplified with primers HES-531 and HES-532 and inserted into a derivative of pJG4-5 [77] that lacks the transcriptional activation domain. Putative promoter fragments were amplified from genomic DNA with primers HES-612 and HES-613 (spe-7) or HES-583 and HES-584 (spe-4) and inserted into a derivative of pLacZi (Clontech); the same spe-4 upstream sequence is sufficient to confer in vivo transgene rescue of spe-4 sterility [78]. Yeast media and manipulations followed standard protocols [79].
Intact gonads were obtained by dissection of individual worms. Sperm spreads were obtained by further applying slight pressure to the coverslip. Antibody staining of dissected gonads followed established protocols [e.g., see 46]. Specimens were incubated with primary antibodies for 2–3 hours at room temperature unless otherwise indicated. Primary antibodies included: FITC-labeled anti-alpha-tubulin (DMIA, Sigma) (used at 1∶100 dilution), anti-MSP (4D5, gift from D. Greenstein) (1∶200), anti-phospho-histone H3 (serine10) (Upstate Biotechnology) (1∶150), anti-MPM2 (DAKO Corp.) (1∶100 overnight at 4°C), anti-histone H3 dimethylated Lys 9 (Upstate Biotechnology) (1∶50), 1CB4 [49]. The SPE-44 antibody (1∶100) was generated in rabbits against a C-terminal peptide epitope and affinity-purified (Open Biosystems). Affinity-purified secondary antibodies (Jackson Immunoresearch Laboratories) (1∶100) included goat anti-rabbit TRITC-labeled IgG and FITC- or DyLight-labeled goat anti-mouse IgG. Images were acquired under differential inference contrast or epifluorescence using an Olympus BX60 microscope equipped with a Cooke Sensicam cooled CCD camera and IPLab software. In some cases, images were minimally processed to enhance contrast either with IPLab software or Photoshop.
In situ hybridization for spe-44 germ line expression was performed on dissected gonads following fixation [80]. Digoxigenin-labeled, single-stranded sense and antisense probes were generated from a 1.3 kb spe-44 cDNA fragment by linear amplification according to the manufacturer's protocol (Roche). Following hybridization, probe detection was by colorimetric assay with alkaline phosphatase-conjugated anti-digoxigenin antibodies and NBT/BCIP substrate.
Worms for microarray screening were obtained by hand-picking samples of 50 L4 males of the indicated genotype. RNA was isolated by Trizol treatment and ethanol precipitation using linear polyacrylamide (GeneElute LPA; Sigma-Aldrich) as carrier. RNA was amplified and labeled according to the manufacturer's protocol (Nugen). Microarray screening was performed in triplicate using GeneChip C. elegans genome arrays (Affymetrix). Microarray data were analyzed by Microarray Suite 5.0 (Affymetrix) and Genomics Suite (Partek) software, using a p-value threshold of 0.05 for differential expression. Sequences for comparative analysis of Caenorhabditis species were obtained from WormBase release WS220 [www.wormbase.org]. Multiple sequence alignments of spe-44 homologs and promoters were performed with CLUSTALW2 [81].
|
10.1371/journal.pbio.1000188 | Dual Regulation by Pairs of Cyclin-Dependent Protein Kinases and Histone Deacetylases Controls G1 Transcription in Budding Yeast | START-dependent transcription in Saccharomyces cerevisiae is regulated by two transcription factors SBF and MBF, whose activity is controlled by the binding of the repressor Whi5. Phosphorylation and removal of Whi5 by the cyclin-dependent kinase (CDK) Cln3-Cdc28 alleviates the Whi5-dependent repression on SBF and MBF, initiating entry into a new cell cycle. This Whi5-SBF/MBF transcriptional circuit is analogous to the regulatory pathway in mammalian cells that features the E2F family of G1 transcription factors and the retinoblastoma tumor suppressor protein (Rb). Here we describe genetic and biochemical evidence for the involvement of another CDK, Pcl-Pho85, in regulating G1 transcription, via phosphorylation and inhibition of Whi5. We show that a strain deleted for both PHO85 and CLN3 has a slow growth phenotype, a G1 delay, and is severely compromised for SBF-dependent reporter gene expression, yet all of these defects are alleviated by deletion of WHI5. Our biochemical and genetic tests suggest Whi5 mediates repression in part through interaction with two histone deacetylases (HDACs), Hos3 and Rpd3. In a manner analogous to cyclin D/CDK4/6, which phosphorylates Rb in mammalian cells disrupting its association with HDACs, phosphorylation by the early G1 CDKs Cln3-Cdc28 and Pcl9-Pho85 inhibits association of Whi5 with the HDACs. Contributions from multiple CDKs may provide the precision and accuracy necessary to activate G1 transcription when both internal and external cues are optimal.
| Eukaryotic cells grow and divide by progressing through carefully orchestrated stages of the cell cycle characterized by stage-specific patterns of gene expression, DNA replication, and scission. How stage-specific gene expression is coordinated with cell cycle progression is only partially understood. The phase known as G1 marks the initiation of the cell cycle (called START in yeast) and involves the coordinated expression of more than 200 genes regulated by two transcription factors, SBF and MBF. The activity of SBF and MBF is restrained by binding of the repressor protein Whi5 to the two transcription factors early in G1 phase. Phosphorylation of Whi5 by G1-specific forms of the cyclin-dependent kinase (CDK) Cdc28 promotes dissociation of Whi5 from SBF and its export from the nucleus; this, in turn, releases SBF to activate G1-specific transcription. This G1 transcriptional circuit is analogous to that defined in mammals by the E2F family of transcription factors and the retinoblastoma (Rb) tumor suppressor protein. Rb further contributes to the repression of G1-specific transcription in mammals by recruiting histone deacetylases (HDACs), which are chromatin remodeling complexes that regulate promoter accessibility. Here, we show that regulation of G1-specific transcription in yeast also involves repressor-mediated recruitment of HDACs. We demonstrate that repression by Whi5 is modulated by both Cln-Cdc28 and a second G1-specific CDK, Pcl-Pho85, and further show that both kinases regulate the interaction of Whi5 with HDACs. We propose that regulation of the repressor by more than one G1-specific CDK ensures definitive inactivation of Whi5, a critical event for appropriate cell cycle initiation.
| Cyclin-dependent protein kinases (CDKs) act as molecular machines that drive cell division, and cell cycle progression is dependent on oscillation between CDK active and inactive states. In S. cerevisiae, the CDK Cdc28 associates with nine different cyclin subunits to promote and coordinate a complex network of events necessary for smooth cell cycle transitions [1]. Irreversible commitment to a new round of cell division occurs toward the end of G1 phase in a process called Start in yeast. The analogous regulatory event is called the restriction point in mammalian cells [2],[3]. In yeast, three G1 cyclins, Cln1, Cln2, and Cln3, associate with Cdc28 to initiate events required for progression through Start. Passage through Start catalyzes a defined molecular program that initiates DNA replication, budding, spindle maturation, and chromosome segregation [3].
One key feature of Start in yeast, and G1 progression in other eukaryotic cells, is the induction of a transcriptional program involving over 200 genes, including those encoding the G1 (CLN1, CLN2, PCL1, and PCL2) and B-type cyclins (CLB5 and CLB6) [4],[5]. G1/S phase-specific transcription depends on two heterodimeric transcription factors called SBF (Swi4,6 cell cycle box binding factor) and MBF (MluI binding factor). These complexes share a common regulatory subunit, Swi6, which is tethered to DNA via its binding partners, encoded by SWI4 in SBF and MBP1 in MBF [5]. At the well-studied HO locus, binding of the zinc-finger transcription factor Swi5 is followed by recruitment of the Swi/Snf chromatin remodeling complex and the SAGA histone acetyltransferase complex [6]–[8]. These events set the stage for SBF binding and recruitment of the SRB/mediator complex [6]. Importantly, subsequent recruitment of PolII and transcription initiation is dependent on CDK activity [9]. Although any one of the three G1 cyclins is sufficient to drive Start, genetic studies indicate a key role for Cln3-Cdc28 in activating SBF and MBF. At the same time Cln1 and Cln2 are required for the proper execution of other Start-related events such as budding and DNA synthesis. Cells lacking CLN3 are large and severely delayed for onset of G1/S transcription, while ectopic induction of CLN3 in small G1 cells activates transcription and accelerates passage through Start [10].
Start does not occur until cells have passed a critical cell size threshold, a barrier modulated by nutrient conditions, among other regulatory inputs [11]. A systematic analysis of cell size profiles for the entire set of yeast deletion mutants uncovered many new regulators of Start including Whi5 and implicated it as an inhibitor of G1/S-specific transcription [12],[13]. Whi5 occupies specific promoters early in G1 phase when CDK activity is low. However, Cdc28-dependent phosphorylation of both Whi5 and SBF/MBF late in G1 phase results in disengagement from SBF and nuclear export of Whi5 consequently leading to activation of SBF- and MBF-dependent transcription [12],[13].
Whi5 is proposed to function in a manner analogous to the well-characterized Rb family proteins in metazoans. E2F, the functional analog of SBF/MBF, regulates G1-specific gene expression required for passage through the restriction point [14]. E2F activity is restricted to late G1 phase because of inhibition by the retinoblastoma protein (Rb). Rb associates with E2F to restrain its activity until late G1, at which point stepwise phosphorylation of Rb by two CDKs, cyclin D-Cdk4/6 and cyclin E-Cdk2, causes the dissociation of Rb from E2F [15]. This process appears to be regulated by a positive feedback loop in which Rb phosphorylation by cyclinE-Cdk2 leads to further dissociation of Rb from promoters and enhancement of G1-transcription. At the molecular level, Rb interacts with both E2F and chromatin remodeling complexes such as histone deacetylases (HDACs) [16]–[18]. Rb appears to repress transcription through at least three distinct mechanisms: (1) Rb can bind directly to the activation domain of E2F thereby blocking its activity [19]; (2) recruitment of Rb can block the assembly of the pre-initiation complex thus inhibiting the activity of adjacent transcription factors [20] and; (3) Rb can recruit remodelers such as HDAC1 and BRG1 to modify chromatin structure. BRG1 is one of the human Swi/Snf adenosine triphosphatases (ATPases) that remodel nucleosomes by utilizing ATP to weaken the interactions between DNA and histones [16],[17]. The specific roles of different CDKs in regulating E2F-Rb function, however, remain unclear.
Another yeast CDK Pho85 was originally discovered as a regulator of phosphate metabolism, but has since been shown to play numerous roles in the regulation of cell division and other processes [21]–[23]. Ten genes encoding Pho85 cyclins (Pcls) have been identified and they appear to dictate substrate and functional specificity of Pho85 [24]–[26]. Expression of three Pcls, PCL1, PCL2, and PCL9, is restricted to G1 phase of the cell cycle [25]. Specifically, PCL9 expression peaks early in G1, whereas maximal expression of PCL1 and PCL2 is observed at Start and is dependent largely on SBF [27]–[29]. Although Pho85 is not essential for viability, it is required for cell cycle progression in the absence of the Cdc28 cyclins CLN1 and CLN2 [29], and its absence leads to catastrophic morphogenic changes that culminate in a G2 arrest [30]. Consistent with this observation, inactivation of both Cdc28 and Pho85 CDKs specifically inhibits expression of G1-regulated genes involved in polarized growth [31].
As noted above, transcriptional repression by Rb has been linked to its interaction with histone modification complexes, in particular HDACs. Recent work highlights the importance of post-translational modifications of histone and nucleosome positioning in regulating gene expression [32],[33]. Histone acetylation neutralizes the positive charge generated by lysine-rich regions present in the N-terminal tails of histones, thereby disrupting nucleosome structure and increasing promoter accessibility [34]. As a result, many transcription activators have been shown to interact with histone acetyltransferases, whereas transcriptional repressors often associate with HDACs to promote nucleosome formation to occlude transcription factor binding [35],[36]. Histone deacetylation in S. cerevisae is mediated by a family of HDACs including Rpd3, Hda1, Hda2, Hos1, Hos2, and Hos3 [37]. Similar to their mammalian counterparts, some yeast HDACs are recruited to promoters by sequence-specific regulatory factors to repress gene expression. For example, the Rpd3 deacetylase complex is recruited to the INO1 promoter by the DNA binding protein Ume6 [35],[38]–[40]. This recruitment results in local histone deacetylation and repression of INO1 gene expression [41]. Hda1 is another example of this type of regulation, and is recruited to its target promoters by the repressor Tup1 [42].
In this study, we provide detailed mechanistic insights into Whi5-dependent regulation of G1-specific transcription and cell cycle progression. Specifically, we identify Whi5, to our knowledge, as the first demonstrated physiological substrate for the G1-specific Pcl9-Pho85 CDK and provide genetic and biochemical evidence supporting a direct role for Pho85 at Start. Furthermore, we show that in a manner similar to Rb in mammalian cells, Whi5-mediated repression involves the HDACs Rpd3 and Hos3. Dual phosphorylation of Whi5 by Cdc28 and Pho85 inhibits Whi5 activity in at least two ways. Both kinases appear to regulate interaction of Whi5 with different HDACs, whereas Cdc28 is also involved in disrupting Whi5 association with SBF and promoting its nuclear export [12],[13]. G1-specific CDKs thus are specialized to regulate different aspects of the same critical cell cycle event—inhibition of Whi5—resulting in definitive inactivation of the Whi5 repressor.
Synthetic dosage lethality (SDL) is a genetic assay that is based on the rationale that increasing levels of a protein may have no effect on the growth of an otherwise wild-type (wt) strain but may cause a measurable phenotype—such as lethality—in a mutant strain with reduced activity of an interacting protein [43],[44]. Previous studies suggest that SDL can be used effectively to identify novel enzyme targets and a genome-wide SDL screen in cells lacking Pho85 identified known targets of the CDK [23]. In addition to known substrates, several putative Pho85 targets were also identified, including the G1-specific transcription repressor Whi5 [24]. To further explore the role of Pho85 in G1 phase-specific transcription we examined the WHI5-PHO85 SDL interaction in greater detail. As noted previously, Pho85 activity and substrate specificity depends on its interaction with cyclin subunits known as Pcls [25]. To implicate specific Pcl-Pho85 complexes in modulating Whi5 function we examined the effects of WHI5 overexpression in cells lacking different Pcls (Figure 1). Similar to effects observed in cln3Δ and cln1Δ cln2Δ mutants [13], overexpression of WHI5 resulted in growth inhibition of pcl1Δ and pcl9Δ deletion strains and this growth defect was exacerbated in a pcl1Δ pcl9Δ double mutant (Figure 1). Unlike pcl1Δ or pcl9Δ mutants, strains lacking PCL2 or PHO80 cyclins were not adversely affected by increased WHI5 dosage suggesting that the WHI5-PHO85 genetic interaction is dependent on the PCL1,2 cyclin subfamily and more specifically on PCL1 and PCL9 (Figure 1). This observation is consistent with the fact that Pcl1 and Pcl9 (but not Pcl2) are the two G1-specific cyclins that localize to the nucleus [30],[45]. The growth phenotype seen in the plating assay was confirmed by measuring growth rates in liquid culture (unpublished data). On the basis of these results, Pcl1/9-Pho85 may contribute to Whi5 regulation in a manner similar to Cln3-Cdc28.
The genetic interactions described above suggest Whi5 may be a direct target of Pho85. Evidence supporting this hypothesis is provided by protein microarray assays where Whi5 is phosphorylated in vitro by Pcl1-Pho85 [46]. We characterized the Whi5-Pho85 interaction biochemically by performing in vitro kinase assays using recombinant Pcl-Pho85 CDK complexes and purified Whi5 as substrate (Figure 2A). Incorporation of [32P] into Whi5 was not detected in the absence of CDKs (Figure 2A, lane 4). However, Whi5 phosphorylation was observed in the presence of Pcl1- and Pcl9-Pho85 (Figure 2A, lanes 1,2) and when compared to Cln2-Cdc28 kinase activity, Pho85 and Cdc28 phosphorylated Whi5 at similar levels in vitro (Figure 2A, lanes 1–3).
Previous studies revealed multiple Whi5 slow-migrating isoforms that correlate with its phosphorylation state [12],[47]. We examined the effect of various cyclin or CDK mutants on Whi5 mobility (Figure 2B). Because of genetic redundancy of Pcl cyclins [27], we were unable to reproducibly detect changes in Whi5 phosphoforms in cyclin mutant strains. Therefore, a Pho85 mutant was used to asses the phosphorylation status of Whi5. Consistent with previous findings [12],[13], slow migrating Whi5 isoforms present in asynchronous wt extracts (Figure 2B, lane 1) were modestly reduced in cells lacking CLN3 (Figure 2B, lane 7) and completely absent in a cln1Δ cln2Δ double mutant (Figure 2B, lane 6), confirming that Whi5 phosphorylation depends on Cln-Cdc28 kinase complexes. Consistent with our SDL results and in vitro kinase assays, we observed a significant reduction in Whi5 mobility in extracts from a pho85 mutant strain (Figure 2B, lane 2). Thus, similar to Cdc28, phosphorylation of Whi5 also depends on Pho85 in vivo.
To determine if Whi5 physically associates with Pho85 in yeast, we first assayed Whi5FLAG immune complexes for kinase activity. A robust autophosphorylation activity was recovered from Whi5FLAG immunoprecipitates derived from wt cell extracts when radiolabeled ATP was added to the immunoprecipitated sample (Figure 2C, lane 2). This activity was partially dependent on both CDC28 and PHO85 (Figure 2C, lanes 3–5). We also confirmed a physical interaction between Whi5 and Pcls using a co-immunoprecipitation assay (Figure 2D). Immunoprecipitation of Whi5MYC from epitope-tagged cyclin extracts revealed a specific association between Pcl9 and Whi5 (Figure 2D, lane 4). We failed to reproducibly detect a physical interaction between Whi5 and Pcl1 (Figure 2D, lane 2) suggesting that Pcl9-Pho85 is the primary Whi5 CDK. Taken together, the phosphorylation and co-immunoprecipitation assays strongly suggest that, in addition to Cdc28, Pho85 also phosphorylates Whi5. Furthermore these results identify Whi5 as the first reported substrate for Pcl9-Pho85, one of two Pcls whose activity is restricted to early G1 phase.
Whi5 associates indirectly with G1 phase-regulated promoters through interaction with SBF and MBF. Interactions with these transcription factors and subsequent promoter binding are disrupted by CDK-dependent phosphorylation [12],[13]. Because Whi5 appears to be a Pho85 substrate, we assessed the occupancy of SBF promoters by Pcl9. To date, cyclins have not been detected at yeast promoters. Pcl9 is normally an unstable short-lived protein [27]; however, similar to other cyclins, Pcl9 turnover appears to be catalyzed in part by its cognate CDK, Pho85 (Figure 3A) [48]. Therefore, to test Pcl9 promoter localization in a more sensitive genetic background, we performed ChIP (Chromatin immunoprecipitation) experiments in a pho85Δ strain (Figure 3B). The highest levels of CLN2 promoter DNA were detected in Pcl9MYC immune complexes 30 min following release from a metaphase-anaphase arrest (Figure 3B). The Pcl9-chromatin association was no longer detectable 45 min after GAL-CDC20 induction indicating that the interaction is short-lived and transient as predicted for a regulator of Start. The association was Whi5-dependent since Pcl9 was not detected at the CLN2 promoter in a strain lacking Whi5 (Figure 3C). The localization of Pcl9 to CLN2, a G1 promoter, is consistent with a direct role for Pcl9-Pho85 in regulating G1 transcription.
As mentioned above, cln3Δ mutants arrest in G1 phase as large unbudded cells in response to increased WHI5 dosage, indicating that Whi5 is a dose-dependent regulator of Start. Therefore, if Pho85 and Cdc28 function analogously to inhibit Whi5 activity, we predict that elevated Pho85 kinase activity would antagonize the toxic effects of WHI5 overexpression and suppress the growth defects observed in a cln3Δ mutant. To test this prediction, high copy plasmids expressing PCL1, PCL2, PCL9, or PHO80 were introduced into a cln3Δ strain expressing WHI5 from a conditional MET25 promoter (Figure 4A). Plasmid-based expression of Pcls and Whi5 was confirmed by immunoblotting (Figure S1). Induction of WHI5 expression in a cln3Δ mutant resulted in cell death whereas overexpression of PCL1 or PCL9 partially suppressed this toxicity and restored growth (Figure 4A). Consistent with results from SDL analyses (Figure 1), this suppression was specific to PCL1 and PCL9 since neither PCL2 nor PHO80 were able to function effectively in the assay (Figure 4A). Furthermore, PCL1/9-mediated suppression was dependent on phosphorylation since growth of a cln3Δ mutant expressing a nonphosphorylatable form of WHI5 (Whi512A) [13] could not be restored (Figure 4A). These genetic results corroborate the biochemical evidence that Pcl-Pho85 regulates Whi5 activity through phosphorylation.
Given its effect on WHI5 overexpression, we next examined PCL effects on other CLN3-associated phenotypes. CLN3 is required to activate G1-specific transcription once cells have achieved a critical size [49]–[51]. A cln3Δ mutant exhibits a large cell size phenotype because of its inability to inhibit Whi5 and activate Start-specific transcription [12],[13]. Ectopic expression of PCL1 or PCL9 reduced cln3Δ cell size to an intermediate level between that of wt and cln3Δ cells (Figure 4B). Conversely, deletion of PCL9, PCL1, and the partially redundant cyclin PCL2 resulted in a cell size increase (Figure 4C). These results suggest that Pcl-Pho85 and Cln3-Cdc28 share a common role in cell cycle progression to regulate Whi5 activity and promote passage through Start.
To determine if Pcl-Pho85 and Cln3-Cdc28 might function in parallel to regulate Start, we first tried to test whether pcl9Δ cln3Δ or pcl1Δ pcl9Δ cln3Δ strains showed any synthetic growth defects. As expected, no growth defects were observed, probably because of the redundant effects of other Pcls [27]. Unlike the Cdc28 cyclins, which shows distinct cell cycle expression patterns, most Pcls are expressed at all cell cycle stages [25]. We then examined the phenotype of a pho85Δ cln3Δ double mutant. Cells lacking cln3Δ are larger than wt cells but do not display overt defects in growth rate while pho85Δ mutants are slow growing (Figure 5A). However, pho85Δcln3Δ double mutants exhibited a more pronounced growth defect compared to single mutants and analysis of DNA content revealed that the pho85Δ cln3Δ double mutant cells accumulated in G1 phase with predominantly unreplicated DNA (Figure 5A). Importantly, deleting WHI5 overcame both the cell cycle progression and growth defects observed in the absence of both CLN3 and PHO85. Notably, a pho85Δ cln3Δ whi5Δ triple mutant exhibited a growth rate similar to a cln3Δ single mutant indicating that Pcl-Pho85 and Cln3-Cdc28 function in separate yet converging pathways to regulate Whi5 function and, by extension, G1 cell cycle progression (Figure 5A). These observations also hold true under liquid growth conditions as shown. WHI5-dependent suppression appears to be specific to the pho85Δ cln3Δ phenotype because WHI5 deletion was unable to rescue 53 additional synthetic lethal interactions involving PHO85 (Table S1; D.Q. Huang and B.J. Andrews, unpublished data).
Given that Whi5 represses SBF- and MBF-specific transcription, we asked whether PHO85 affects SBF-driven reporter gene expression. A reporter gene consisting of tandem SCB consensus element repeats fused upstream of the HIS3 coding region was constructed and integrated into wt, cln3Δ, and pho85Δ strains. Previous work has shown that this reporter provides a highly specific read-out for SBF-dependent transcription [13],[52]. Growth on medium lacking histidine supplemented with 3-aminotriazole (3-AT) was used to assess SBF transcriptional activity (Figure 5B). Even though cells lacking PHO85 were moderately sensitive to higher concentration (5 mM) of 3-AT (unpublished data), both cln3Δ and pho85Δ mutants showed no growth in media containing 30 mM 3-AT indicating that SBF transcription is impaired in these mutants, whereas growth of wt cells was unaffected [13]. Furthermore, defects in SCB-driven gene expression were more pronounced in the pho85Δ cln3Δ double mutant (at 10 mM 3-AT, Figure 5B). Consistent with the genetic interactions described above (Figure 5A), SBF-dependent reporter activity was restored in pho85Δ cln3Δ mutants when WHI5 was deleted (Figure 5B). However, WHI5 deletion only partially rescued the growth defect in pho85Δ cells at 30 mM of 3-AT (Figure 5B). The Whi5-independent 3-AT sensitivity of pho85Δ cells may be due to unregulated Gcn4 in the absence of PHO85, since GCN4 is induced by 3-AT and Pho85 has been shown to regulate Gcn4 stability [53],[54]. Nonetheless, these data suggest that, like Cln3-Cdc28, Pcl-Pho85 modulates SBF activity through Whi5.
We next interrogated the effects of CDK activity on Whi5-mediated transcriptional repression (Figure 6). A construct expressing a LexA DNA binding domain fused to WHI5 was introduced into a strain harboring a LacZ reporter gene containing LexA binding sites in its promoter (Figure 6). Consistent with its role as a negative regulator of G1-specific transcription, a ∼10-fold reduction in β-galactosidase activity was observed in cells expressing the LexA-Whi5 fusion protein compared to a vector control (Figure 6). Overexpression of PCL9, CLN3, or CLN2 restored LacZ expression to intermediate levels indicating that activation of either CDC28 or PHO85 was capable of antagonizing Whi5 function in this assay (Figure 6). Consistent with suppression of WHI5-mediated growth defects (Figure 4), inhibition of Whi5 activity was dependent on phosphorylation since LacZ expression could not be restored in cells harboring an unphosphorylatable LexA-Whi512A fusion protein (Figure 6).
Cln2-Cdc28 activity was previously shown to disrupt recombinant Whi5-SBF complexes in vitro [13], but Cln3-Cdc28 and Pho85 kinases had not been assessed for this activity. A preassembled recombinant Whi5-Swi4FLAG-Swi6 complex bound to anti-FLAG resin was incubated with purified kinases in the presence of radiolabeled ATP and separated into soluble (Figure 7B, labeled “S”) and bound fractions (Figure 7B, labeled “B”). Equivalent amounts of kinase were approximated on the basis of in vitro kinase activity (Figure 7A, and Materials and Methods). As expected, Cln2-Cdc28 phosphorylation caused most of the SBF-bound Whi5 to be released into the soluble fraction (Figure 7B, lanes 3 and 4). In contrast, we failed to observe dissociation of Whi5 from SBF in the presence of Cln3- or Pcl9-CDK complexes (Figure 7B, lanes 5–10). In addition to negatively regulating the interaction of Whi5 with SBF, Cdc28 also controls its localization [13]. Unlike Cln-Cdc28 phosphorylation, which promotes Whi5 export from the nucleus, deletion of PHO85 did not dramatically affect the subcellular localization of Whi5 (Figure 7C). Together, these results suggest that Pho85 must regulate Whi5 function through alternate mechanisms.
We next explored what additional mechanism might explain Pcl- and Cln3-mediated regulation of Whi5 activity. Functional conservation clearly extends to Whi5 and its metazoan analogue Rb [14]. Since Rb represses transcription, in part, through recruitment of HDACs, we used a batch affinity chromatography assay to test for physical interactions between a Whi5GST ligand and tandem affinity tagged HDACs (Figure 8A). Specific interactions between Whi5 and Hos3, Rpd3, and, to a lesser extent, Hos1 were identified (Figure 8A, lanes 1, 5, 13) suggesting that, like Rb, Whi5-dependent transcriptional repression involves recruitment of HDACs. This observation is consistent with previous work that detected Rpd3 at the PCL1 promoter using a ChIP assay [55]. Furthermore, HOS3 and RPD3 were required for WHI5 dose-dependent effects on cell size. Like wt cells, strains lacking either HOS3 (Figure 8B, panel 1) or RPD3 (Figure 8B, panel 2) also exhibited a dose-dependent increase in cell size in response to WHI5 overexpression. However, additional cell size effects were not observed in strains lacking both HDACs, suggesting that Hos3 and Rpd3 regulate Whi5 function synergistically (Figure 8B, panel 3).
If HDACs are required for Whi5 function, then strains lacking HDAC function should be resistant to toxic effects associated with WHI5 overexpression. Consistent with this prediction, the growth defect caused by WHI5 overproduction in a cln3Δ was alleviated by the deletion of HOS3 and RPD3 (Figure 9A). Deletion of HOS3 alone rescued WHI5 toxicity in a pho85Δ strain while a cln3Δ mutant required deletion of both HOS3 and RPD3 in order to tolerate increased dosage of WHI5 (Figure 9A).
Given that Whi5 appears to be acting through HDACs, we predicted that deletion of HOS3 and RPD3 should phenocopy those genetic interactions seen in whi5Δ mutants. We first tested various HDAC deletion strains for suppression of the slow growth phenotype of a pho85Δcln3Δ mutant. As for WHI5, deletion of HOS3 and RPD3 partially suppressed the growth defect seen in the pho85Δcln3Δ double mutant strain (Figure 9B). Suppression was specific to HOS3 and RPD3 because deletion of other HDACs showed no suppression, and the growth rate of the pho85Δcln3Δhos3Δ strain was not improved by subsequent deletion of RPD3 and vice versa (Figure 9B).
We next asked if deletion of HDACs might overcome the Start arrest seen in cells lacking both CLN3 and BCK2, another regulator of G1 transcription that functions in parallel with CLN3 [56]. A cln3Δbck2Δwhi5Δ triple mutant grows as vigorously as wt, placing WHI5 downstream of both upstream activators of G1 transcription [13]. Interestingly, deletion of RPD3 partially restored growth in the cln3Δbck2Δ strain providing further evidence for an HDAC requirement in Whi5-mediated transcriptional repression (Figure 9C). Neither subsequent deletion of HOS3 nor deletion of other HDACs affected growth appreciably (Figure 9C). We also employed the SCB-HIS3 assays used above to explore SBF-driven reporter gene expression in the HDAC mutants (Figure 10). As expected, deletion of RPD3 rescued the growth defects of cln3Δ SCB-HIS3 cells in the presence of both 10 mM and 30 mM of 3-AT, whereas HOS3 gene knockout had a marginal but additive effect. In contrast, the growth of pho85Δ cells was slightly rescued by deletion of HOS3 but not RPD3 providing further evidence for Pho85 acting specifically through Hos3. Because of difficulties in detecting HDACs at promoters, we were unable to confirm these observations in vivo.
We also performed co-immunoprecipitation assays using affinity tagged RPD3 and HOS3 strains and observed an obvious decrease in Rpd3 and Hos3 in Whi5 precipitates from strains harboring increased levels of Pcl9, Cln2, or Cln3 cyclins (Figure 11A and 11B). Together, our genetic and biochemical results suggest that Pho85 may preferentially influence Whi5-Hos3 activity, whereas Cln3-Cdc28 is required for inhibition of both Rpd3 and Hos3.
Whi5 is a critical cell cycle regulator that links CDK activity in G1 phase to the broad transcriptional program that accompanies commitment to cell division. We provide substantial evidence that the multifunctional Pho85 CDK is an important regulator of Whi5 activity and G1 phase-specific transcription including: (1) Whi5 is phosphorylated and antagonized by Pho85 and is the first reported substrate for the G1-specific CDK complex, Pcl9-Pho85; (2) the activity of an SBF-dependent promoter is influenced by PHO85; (3) the Pcl9 cyclin binds to SBF-regulated promoters; (4) the repressor function of Whi5 is mediated through the HDACs Hos3 and Rpd3; and (5) HDAC-Whi5 association is regulated by G1-specific forms of both the Pho85 and Cdc28 CDKs. We therefore conclude that timely and efficient release from Whi5 inhibition and subsequent G1/S cell cycle progression requires the concerted activity of both Cdc28 and Pho85.
Several lines of evidence point to common roles for Pho85 and Cdc28. For example, a burst of both G1-specific Cdc28 and Pho85 activity is essential for cellular morphogenesis. A strain lacking the G1-specific cyclins, CLN1, CLN2, PCL1, and PCL2, undergoes a catastrophic morphogenic change and fails to establish polarized cell growth and cytokinesis [30]. Consistent with these observations, a chemical genomic analysis demonstrated that expression of genes involved in polarized cell growth was sensitive to simultaneous inhibition of both kinases, but not either single kinase [31]. A functional connection between Pho85 and Cdc28 is further supported by independent genetic and biochemical analyses that identify common targets phosphorylated by both kinases [45],[46],[48],[57]–[59].
Despite the clear functional overlap for G1-specific forms of Cdc28 and Pho85 in controlling morphogenesis, up to now, a direct role for Pho85 in cell cycle commitment and G1 phase-specific transcription has remained unclear. We discovered that, like Cdc28, Pho85 activates G1 transcription through inhibition of the Whi5 repressor. While the two kinases collaborate to control certain facets of Whi5 regulation, they are also specialized to modulate Whi5 function by distinct mechanisms. We have defined a novel HDAC-dependent mechanism that impinges on Whi5 function and implicates both Pho85 and Cdc28 as regulators of this process.
On the basis of these and other observations, we propose that Whi5 functional regulation involves perturbation of specific HDAC-Whi5 interactions and requires the concerted activity of both Cdc28 and Pho85 (summarized in Figure 12). Interestingly, our genetic observations support a model whereby Pcl-Pho85 preferentially targets the Hos3-Whi5 interaction illustrating a functional distinction between the two CDKs. While Pho85 associates with several cyclin subunits, only Pcl9 exhibits temporal expression and localization patterns compatible with such a function. PCL9 is expressed at the M/G1 phase transition and encodes a short-lived protein localized exclusively to the nucleus in early G1 phase [27],[60],[61]. Cln3 is also present in early G1 cells, but shows a complex localization pattern, with significant retention to the ER in early G1 cells, followed by chaperone-mediated release into the nucleus in late G1 phase [62]. How the specific features of Pcl9 and Cln3 localization might influence the timing of HDAC inhibition remains to be explored.
The second component of Whi5 regulation is predicated on previous studies indicating that G1/S gene expression is preceded by Whi5-SBF complex dissociation and subsequent nuclear export of Whi5 (Figure 12) [13]. Unlike early regulatory events, Cdc28 activity is both necessary and sufficient to drive these events since neither SBF binding to Whi5 nor nuclear localization of Whi5 was adversely affected in a pho85Δ mutant (Figure 7). Also, we are able to detect binding of SBF in vivo to CLN2 promoters when PHO85 is deleted (Figure 3C). However, both purified Cln3-Cdc28 and Pcl9-Pho85 failed to affect Whi5-SBF stability in vitro, while complex disruption was effectively achieved in the presence of Cln2-Cdc28 kinases (Figure 7). Cln3-Cdc28 and Pcl9-Pho85 may have a more pronounced effect on the Whi5-SBF complex in vivo. Alternatively, Cln3- and Pcl9-CDKs may act primarily as agonists of HDAC interactions while physical interactions with SBF and nuclear export are optimally mediated by the late G1 CDKs, Cln1- and Cln2-Cdc28. Indeed, recent work reveals activation of CLN2 expression while Whi5 remains bound to the promoter (H. Wang, L.B. Carey, Y. Cai, H. Wijnen, and B. Futcher, personal communication). Such a mechanism may serve to sharpen the onset, as opposed to the timing, of G1/S gene expression thus ensuring a sustained transcriptional burst and irreversible commitment to cell division [13]. Consistent with this idea, recent analysis of cyclin gene expression using a single cell assay affirms that positive feedback involving the Cln1 and Cln2 cyclins induces the G1/S regulon, and that this regulatory feedback is important for maintaining coherence of gene expression at Start [63].
SBF promoter recruitment depends on a series of well-organized chromatin remodeling events [7],[36]. SBF, in turn, regulates the recruitment of the general transcription machinery via a two-step process beginning with the mediator complex followed by CDK-dependent recruitment of RNA PolII, TFIIB, and TFIIH [9]. Previous studies suggested that this CDK requirement stems from Whi5, which in its unphosphorylated state, remains bound to SBF and occludes the basal transcription machinery from binding specific promoters [13]. We have extended this model to include a role for HDAC activity. We predict that Hos3 and Rpd3 contribute to Whi5 repression by preventing holoenzyme access to chromatin. During states of high CDK activity, Cdc28 and Pho85 abrogate Whi5-HDAC and Whi5-SBF interactions and initiate transcription. Consistent with our model, Pcl9 and Cln3 cyclins localize to G1 promoters and Whi5 remains associated with G1-specific promoters in the absence of HDAC-promoter interactions (Figure 3; H. Wang, L.B. Carey, Y. Cai, H. Wijnen and B. Futcher, personal communication). However, Whi5 may also repress transcription by additional mechanisms since its activity is partially retained in hos3Δ rpd3Δ mutants (Figure 9).
Rpd3 is a well-characterized HDAC that accomplishes most of its functions as part of a large protein complex [37]. The Rpd3-Sin3 deacetylase complex has long been implicated as a cell cycle regulator required for silencing HO gene expression to prevent mating type switching in newly budded cells [64],[65]. Our observations that Whi5 associates with Rpd3 and our genetic data linking G1 Cdks, Whi5, and Rpd3 reveal a more general role for Rpd3 in G1/S-phase specific transcription. These data are consistent with observations from Futcher and colleagues that the Rpd3 protein can be detected at the CLN2 promoter and that the amount of Rpd3 at the promoter is decreased when CLN3 is induced (H. Wang, L.B. Carey, Y. Cai, H. Wijnen and B. Futcher, personal communication). The Rpd3-Sin3 HDAC has also been connected to G1 transcription factors through the interaction of Sin3 with Stb1, a Swi6-binding protein [66]–[68]. Both Stb1 and Sin3 are required for repression of G1 transcription early in G1 phase [68]. Unlike Rpd3, Hos3 is largely uncharacterized, although a recent study suggests a role for Hos3 in yeast apoptosis upon exposure to oxidative radicals [69]. We have uncovered an additional role for Hos3 in Whi5-mediated transcriptional repression.
A question that arises from our observations is what advantage does combinatorial kinase regulation impart on specific biological processes such as G1/S cell cycle progression? Contributions from multiple CDKs may provide the precision and accuracy necessary for rapid definitive decisions that irreversibly affect cellular fate. Indeed, distributive multisite phosphorylation mechanisms exhibit ultrasensitivity with respect to kinase concentration, thereby creating a “switch-like” behavior in biological circuits [70]. Since cell cycle transitions typically display switch-like attributes, multisite phosphorylation by various kinase combinations may prove to be a rule rather than the exception amongst CDK targets, including key cell cycle regulators such as Whi5. In fact, a recent computational analysis showed enrichment of multiple closely spaced consensus sites for Cdc28 substrates in yeast, a pattern that proved predictive of likely CDK targets [71].
Although kinase combinations are likely necessary for cell cycle regulation, the contribution of each individual kinase may vary depending on specific signals and environmental stimuli. In certain environments, Pcl-Pho85 may have more dramatic, condition-specific effects on Whi5 function than Cdc28 analogous, perhaps, to the regulation of Rb that is required for quiescence and prevention of apoptosis [72],[73]. Previous studies indicate that Whi5 localizes to nuclei in stationary phase cells suggesting that Whi5 may also play a role in G0 [13]. Interestingly, Pho85 is required for survival in starvation conditions and plays an important role during stationary phase [74]–[76]. Furthermore, CDK5, the mammalian Pho85 homolog, induces apoptosis in neuronal cells via Rb phosphorylation [77]. Whether Whi5 activity is more prominently affected by Pcl-Pho85 in response to stationary or stress conditions requires additional investigation.
Similarities between metazoan and yeast cell cycle regulation are increasingly evident as we continue to characterize Whi5 function. For example, similar to proposed Pcl9/Cln3 “early” phase regulation (Figure 12), cyclinD-CDK4/6 phosphorylates Rb to promote HDAC dissociation and E2F transcriptional activation. E2F activation then leads to cyclin E expression, which, similar to Cln1/2 “late” phase regulation (Figure 12), may establish a positive feedback loop whereby cyclinE-CDK2 activity disrupts Rb-promoter interactions and stimulates G1-transcription further [15]. Despite these similarities, the importance of multiple regulatory components in both yeast and mammalian systems remains poorly understood and may be most fruitfully dissected using the yeast model.
The S. cerevisiae strains used are listed in Table 1. All gene disruptions and integrations were achieved by homologous recombination at their chromosomal loci by standard PCR-based methods and confirmed by PCR with flanking primers [78]. Standard methods and media were used for yeast growth and transformation. Two percent of galactose in the media was used to induce the expression of genes under the GAL1 promoter. Synthetic minimal medium with appropriate amino acid supplements was used for cells containing plasmids. Appropriate amounts of 3-AT were added to SD-HIS plates to assess the expression of HIS3 reporter genes. 10-fold serial dilutions (5–10 µl) of yeast cells were spotted onto plates with appropriate nutrition conditions to assess growth. Plasmids used in this study are listed in Table 2. In most cases, a DNA insert was amplified by PCR and inserted into a linearized vector by homologous recombination in yeast. Details of construction will be provided upon request.
The in vitro protein kinase assays monitored the incorporation of [32P] transferred from γ-32P-ATP to purified recombinant GST-Whi5. The reaction mixture for assays shown in Figure 2A contained 50 mM Tris-HCl (pH 7.5), 1 mM DTT, 10 mM MgCl2, and 1 µM ATP (including 20 µCi γ-32P-ATP) and 0.2 µg GST-Whi5 in 20 µl of total volume. 2 µl of a purified recombinant kinase (0.4 µg–0.8 µg) was added to the mixture and incubated at 30°C for 30 min. Purification of Cln and Pcl CDKs from insect cell expression systems have been previously described [13],[46]. Whi5 was then analyzed by SDS-PAGE and autoradiography. Kinase assays on immunoprecipitated proteins from yeast cell extracts were performed as described [13]. Kinase assays preceding the Whi5-SBF dissociation assay (Figure 7) were performed as described above except that 200 µM γ-32P-ATP was used instead of 1 µM. The final concentration of Cln3 and Pcl9 was 3 µM, and the final concentration of Cln2 was 60 nM (50-fold less).
Liquid β-galactosidase assays were performed as described [29]. Strains carrying appropriate plasmids were grown in synthetic minimal medium to mid-log phase, transferred to synthetic galactose medium, and incubated for 4 h. Cells were harvested and broken in lysis buffer (100 mM Tris-HCl [pH 8.0], 1 mM DTT, and 20% glycerol with protease inhibitors) with glass beads. The β-galactosidase activity was determined by adding 100 µl of total cell extract to 0.9 ml of Z buffer (100 mM Na2PO4, 40 mM NaHPO4, 10 mM KCl, 1 mM MgSO4, and 0.027% β-mercaptoethanol) and 200 µl ONPG (4 mg/ml) (Sigma). Units of β-galactosidase activity were determined as described [29].
The protein binding assay essentially followed the procedures described previously [13]. Briefly, 1 µl of insect cell lysate expressing SBF (Swi6-Swi4FLAG) was mixed with 1 µl of purified GST-Whi5 (∼0.1 µg) and 7 µl of M2 anti-FLAG resin (Sigma) in 8 µl of kinase buffer (50 mM Tris-HCl [pH 7.5], 1 mM DTT, and 10 mM MgCl2). The mixture was incubated at 4°C for 1 h with mixing. The beads bound to the SBF-Whi5 complex were then washed three times with kinase buffer, and mixed with various cyclin dependent kinases in kinase buffer with 0.2 mM ATP in a 20 µl volume. The kinase reaction was incubated at 30°C for 1 h. The soluble portion was taken out and mixed with 20 µl of 2×SDS-PAGE loading buffer. The beads in the tube were washed three times with kinase buffer before mixing with 15 µl of 2×SDS-PAGE loading buffer.
Strains containing galactose-inducible plasmids were grown to saturation in 2% raffinose media for 48 h. Expression of plasmids were induced by transferring into 2% raffinose 2% galactose media and liquid growth assays were performed as previously described over 36 h using a Tecan GENios microplate reader (Tecan) [79]. Average doubling (AveG) for each culture was calculated as previously described [79]. Growth rate for each mutant was calculated relative to the AvgG of the wt strain.
The localization of Whi5-GFP was monitored in wt, cdc28-4, and pho85Δ strains. Cells expressing pMET-GFP-WHI5 were grown to log phase in synthetic glucose medium without methionine. Cells were observed at a magnification of 1,000× using Nomarski optics and fluorescence microscopy and photographed by a Cascade 512B high-speed digital camera (Roeper Scientific) mounted on a Leica DM-LB microscope. Images were captured and analyzed by MetaMorph software (Universal Imaging Media).
The pho85Δ PCL9MYC GALpr-CDC20 and pho85Δwhi5ΔPCL9MYC GALpr-CDC20 cells were grown in YP-Galactose (YPG) medium to an optical density (OD600) of 0.4, blocked at M phase by growing in YPED medium for 3 h, and released into YPG medium. Samples were taken every 15 min after release and cross-linked with a final concentration of 1% formaldehyde. Wt and swi4Δ strains (for controls) were grown to OD600 of 0.6 in YPD. Formaldehyde cross-linking and preparation of whole-cell extracts were performed as previously described [80]. Immunoprecipitation were performed using 1∶200 dilution of α-myc monoclonal antibody (9E10), α-Swi6 or α-Swi4 polyclonal antibodies. The precipitates were washed twice with lysis buffer, once with LiCl detergent and once with Tris-buffered saline and processed for DNA purification. Enrichment at the CLN2 promoter sequence was quantified with real-time PCR, using a dual fluorogenic reporter TaqMan assay in an ABI PRISM 7500HT Sequence Detection System as previously described [13].
Recombinant GST-Pcl1 and GST-Whi5 were produced in a BL21 bacterial expression strain; other recombinant proteins were produced in insect cells infected with Baculovirus expression vectors [11],[46],[73]. Proteins were detected with 9E10 anti-Myc, 12C5 anti-HA, and M2 anti-FLAG monoclonal antibodies. FACS analysis of DNA content and cell size measurements were described previously [81].
|
10.1371/journal.pcbi.1005104 | Statistical Analysis of Tract-Tracing Experiments Demonstrates a Dense, Complex Cortical Network in the Mouse | Anatomical tract tracing methods are the gold standard for estimating the weight of axonal connectivity between a pair of pre-defined brain regions. Large studies, comprising hundreds of experiments, have become feasible by automated methods. However, this comes at the cost of positive-mean noise making it difficult to detect weak connections, which are of particular interest as recent high resolution tract-tracing studies of the macaque have identified many more weak connections, adding up to greater connection density of cortical networks, than previously recognized. We propose a statistical framework that estimates connectivity weights and credibility intervals from multiple tract-tracing experiments. We model the observed signal as a log-normal distribution generated by a combination of tracer fluorescence and positive-mean noise, also accounting for injections into multiple regions. Using anterograde viral tract-tracing data provided by the Allen Institute for Brain Sciences, we estimate the connection density of the mouse intra-hemispheric cortical network to be 73% (95% credibility interval (CI): 71%, 75%); higher than previous estimates (40%). Inter-hemispheric density was estimated to be 59% (95% CI: 54%, 62%). The weakest estimable connections (about 6 orders of magnitude weaker than the strongest connections) are likely to represent only one or a few axons. These extremely weak connections are topologically more random and longer distance than the strongest connections, which are topologically more clustered and shorter distance (spatially clustered). Weak links do not substantially contribute to the global topology of a weighted brain graph, but incrementally increased topological integration of a binary graph. The topology of weak anatomical connections in the mouse brain, rigorously estimable down to the biological limit of a single axon between cortical areas in these data, suggests that they might confer functional advantages for integrative information processing and/or they might represent a stochastic factor in the development of the mouse connectome.
| Tract-tracing depends on active axonal transport of tracers between nerve cells, indicating the anatomical connectivity between areas of the brain. Recent advances in tract-tracing technology have enabled reconstruction of the connectome or wiring diagram of mammalian cerebral cortex. Here, we propose a novel statistical model to account for the noise arising from automation of tract-tracing measurements and from injections of tracer into multiple cortical areas simultaneously. On this basis, we find that the strength of anatomical connectivity in the mouse brain varies over six orders of magnitude, with the weakest links between regions approximately representing a few axons. Including all weak links above the statistical noise thresholds, we find that the connection density of the mouse connectome (73%) is greater than previously reported. Many of the complex topological and spatial properties of the mouse brain network emerge on the basis of the strongest axonal projections, whereas the weakest links have a more random organization. We conclude that inter-areal connections mediated by a few axons can be rigorously distinguished from experimental sources of noise in contemporary tract tracing data. Such weak links could support integrated functions of the mouse brain network and/or could represent an element of randomness in its formation.
| Recently, there has been much interest in the connectome perspective of the brain, which aims to map the entire set of connections or interactions between brain regions, rather than the more traditional focus on individual regions and their connectivity [1, 2]. This view may offer new insights both into general rules underlying the pattern of connections of the healthy developing brain, and how these patterns are disturbed in disorders and disease [3–5]. Non-invasive techniques such as functional or diffusion magnetic resonance imaging, electroencephalography and magnetoencephalography allow for measuring these networks at the whole-brain scale. However, these techniques only indirectly measure the actual axonal connectivity between brain regions. Despite major advances in sophisticated statistical and computational methods to process these data, direct interpretation in terms of neurons and axons is infeasible. For this reason, there is great interest in carrying out such analyses in animal model systems where more direct measurements can be made.
In mammals, tract tracing is considered the gold standard for assessing axonal connectivity. This invasive technique allows for quantitative measurement of the strength or weight of axonal projections between cortical areas. However, the technique has been challenging to scale up, and historically most tract-tracing studies have necessarily focused on connectivity of a few regions experimentally studied. Meta-analytic collation of multiple primary studies in the literature was used for pioneering graph theoretical analysis of the cat and the macaque connectomes [6]; but there are many technical complications in combining published tract-tracing data, such as inconsistent usage of atlases and methodology across primary studies [7]. Recent years, however, have seen several institutes publish large-scale efforts to comprehensively map axonal connectivity in the mouse [8, 9] and the macaque [10] by collecting high resolution images of tracer propagation from a large number of injection experiments coordinated by a standard protocol. Initial analyses have already shown the potential value of these “next generation” tract-tracing datasets for informing our understanding of the organization of the mammalian connectome [11–15] and for identifying a large number of so-called “new found projections” or axonal projections below the limit of resolution of historical methods.
The gold standard for assessing connectivity from microscopy images is expert visual inspection and manual demarcation of signal from background noise. This process of expert curation of the images has been used to define axonal connections in a large dataset of tract-tracing experiments in the macaque [10, 16]. But expert curation is time-consuming and observer dependent so the Allen Institute for Brain Sciences (AIBS) has taken a more computational approach in analysing their large tract-tracing dataset in the mouse, using automated image analysis algorithms to quantify the amount of signal present in each target region [9]. This algorithmic approach may suffer from decreased sensitivity, i.e. greater risk of failing to detect a true signal when it is present, or decreased specificity, i.e. a greater risk of failing to rule out a true signal when it is absent.
Statistically principled methods are currently lacking to estimate connectivity from these datasets. However, the published large datasets contain repeated measurements for many connections. These allow for estimation of variability, an indicator of experimental noise and inter-animal differences, and can thus be used for principled estimation of uncertainty in connectivity estimates. In particular, the mouse tract-tracing data published by the AIBS might benefit from a more formal statistical approach, since the segmentation algorithm used for automated analysis was found to have low specificity or high false positive rate [9], when algorithmic assessment was benchmarked against expert evaluation of a subset of experiments. Moreover, such a method may be used to assess the impact of incomplete data, as even the comprehensive AIBS dataset does not include primary injections into all cortical regions.
An open question in systems neuroscience is what percentage of brain regions that could be connected are, in fact, connected. This is fundamentally a question about specificity and sensitivity of detection of connections or links. Recently, it has been claimed that 62% of all possible interregional intrahemispheric connections in macaque cortex exist [10, 17] (66% for a fully investigated subset of regions), which is much higher than previous estimates of 15%-45% cortical connection density [18–20], perhaps reflecting the superior sensitivity of next-generation tract-tracing technologies. Biologically, one would expect network density to increase with decreasing brain size, both because in smaller brains long-distance connections are relatively less costly [21], and because smaller brains generally have fewer regions. However, recent estimates of cortical network density are low for mouse [34%-41% (isocortex) [8]; 35.4%-53.5% (Table S5 from [9])] and rat [45.1% [13]].
Here, we provide a statistical framework for estimating axonal connectivity in large-scale tract-tracing datasets. First, we explore the reliability of different tract-tracing datasets by assessing the variability of repeated measurements of axonal connectivity in (i) the collated Cocomac macaque dataset (collated macaque) [22]; (ii) the multi-experiment dataset by Markov et al. on macaque (multi-experiment macaque) [10]: and (iii) the multi-experiment dataset by the AIBS on the mouse (multi-experiment mouse) [9]. Second, we propose a statistical model to estimate the mouse brain network from algorithmically segmented data with positive-mean noise and co-injected regions, estimating both connectivity weights and associated credibility intervals. Third, we use this approach to rigorously estimate cortical network density of the mouse brain and to enable graph theoretical analysis of connectome topology over the full range of axonal connection weights.
The connectivity weight can be defined in different ways. All definitions depend on the total volume of signal Oej found in the target region, but differ in how this value is normalized [9, 12]. Preferably, our definition of weight would not depend on the total volume Vi, Vj of regions involved. We will therefore estimate the “normalized connection density” nij: the amount of signal expected per unit volume of target region j, after injection of one unit volume of source region i. nij can be estimated from an experiment as O e j V j I e i. However, data on the amount of tracer injected, Iei, are only available for the multi-experiment mouse dataset. Therefore, in analyses comparing the different datasets, we will use the fractional weight fij, the fraction of the total signal measured in target regions that is contained in region j after injection of region i, introduced as the “fraction of labeled neurons” by Markov et al [16]. fij can be estimated from an experiment as Oej/∑k ≠ i Oek.
For each of these three datasets, we explored the variability of the measured weight of connectivity between regions. We focus on the continuous variable of weight rather than simply the (binary) presence or absence of connections as the mouse dataset shows an extremely low variability of binary connections (i.e. Oej > 0), due to its high false positive rate. For the mouse data, and the multi-experiment macaque data, we log-transformed the weights, to make them more comparable to the ordinal weights {1, 2, 3} assigned to each connection in the collated macaque dataset.
We constructed a measure of variability V as follows. First, we selected all pairs of regions for which at least one non-zero value was measured, and computed the mean connection weight mij for each pair. Second, we selected all pairs of regions for which at least two non-zero values were measured, and for each pair calculated the variance vij of the measured values. Last, we defined our measure of variability as the average of these variances, divided by the variance of the means
V = E ( v ¯ ) Var ( m ¯ ) (1)
This measure represents the variance in repeated measurements of the weight of the same connection scaled by the overall variance measured across all weights in the dataset. Note that this measure is invariant to affine transformations of the connection weights, which is important as the collated macaque labels {1, 2, 3} could be arbitrarily scaled to fit with more continuously quantitative measures of connection weight. To examine whether weaker connections are more variable, we compute V0.5 and V0.1, defined as above but with the variance taken over the 50% and 10% weakest weights. Finally, we perform sensitivity analyses on this metric to test robustness of results to selection of region pairs and exclusion of zero weights (S1 Appendix).
Two probability thresholds exist in the multi-experiment mouse tract-tracing data, which put a lower bound on the weight at which connections can still be identified. These are the noise threshold and the co-injection threshold. The noise threshold is the result of incorrectly identified signal in the absence of a connection. We quantify its region-specific probability density function gj as above. This distribution describes the density values to be expected for region j in the absence of a connection. A connection can only be identified if an injection into the source region generates a density in the target region that is higher than this noise threshold. The density generated depends both on the connection weight μij and the injection volume Iei, as an increase in either will increase the expected density due to this connection: log 10 ( I e i 10 μ i j ). We thus compute the noise threshold T n x ( i , j ) as the minimum connection strength μij needed such that the expected density of axonal connectivity of region j for at least one experiment is larger than the xth percentile of the noise distribution g:
T n x ( i , j ) = log 10 ( min e 10 g j x I e i )
with g j x the xth percentile of gj.
The co-injection threshold is due to injections not limited to one region. Such co-injections lead to a threshold very similar to the noise threshold described above. For example, if both region i1 and i2 are injected, a strong connection of i1 to region j would mask the connection from i2 to j. To make this effect visible, we calculate the co-injection threshold Tc(i, j) as the minimum μij needed to ensure that, for at least one experiment, the expected signal due to this connection is larger than the expected signal due to all other connections:
T c ( i , j ) = log 10 ( min e ∑ k ≠ i I e k 10 μ k j I e i ) (6)
We obtain a distribution of Tc(i, j) by computing this statistic for each sample from the MCMC, and obtain the 90% credibility interval (CI) by taking the appropriate percentiles.
We estimate the network density function d(x), defined as the percentage of all possible connections that are of at least weight x. We estimate this function separately for intra-hemispheric and for inter-hemispheric connections. To facilitate comparison of the datasets, we here take the weight to be the fractional weight fij introduced before, which can be obtained from our estimated parameters as V i 10 μ i j ∑ k V k 10 μ k j. d(x) gives the network density of the brain network if a cut-off of x is employed, i.e. setting all edges with weight smaller than x to 0. d is a monotonously decreasing function, and limx → ∞ d(−x) is the total network density. As there is no information about the weight of a connection if it is weaker than its associated thresholds, we compute d(x) using only those connections C(x) for which both the median of its noise threshold and of its co-injection threshold are smaller than x:
d ( x ) = 1 | C ( x ) | ∑ ( i , j ) ∈ C ( x ) 1 f i j > x (7)
with 1 the indicator function, and |C(x)| the size of the set C(x). d(x) can be calculated for each sample of the posterior from the MCMC run, thus providing the uncertainty in the network density estimates. To obtain a point estimate and CI for the overall network density, we evaluate d(x) at the x such that exactly 50% of connections have a threshold higher than x, i.e. at the point where still half of connections can be measured. This choice balances the competing requirements of obtaining the network density at the finest level, i.e. minimal x, and basing the network density estimate on as many connections as possible, i.e. maximal x.
The multi-experiment macaque dataset was expertly curated and, by definition, contains no demonstrable false positives. Furthermore, all injections are into a single region, so there is no co-injection threshold to consider. We thus estimate the fij as above, assuming no noise, obtaining the credibility interval for each connection weight. As only a subset of projecting neurons are counted, there is a threshold to consider; the smallest weight measurable for each experiment is 1/(number of neurons measured). Therefore, for each level x, we compute d(x) as before from all regions i such that the maximum number of neurons measured for any injection into this region is at least 1/x.
Finally, we should note that not all regions were (sufficiently) injected to estimate connectivity, both for mouse (31/43 regions injected) and macaque (29/91 regions injected). This incompleteness may cause a bias, e.g. regions are both more likely to be injected and to have many connections when they are larger. In S1 Appendix we assess the possible bias caused by the preferential injection of larger, more connected regions, by estimating the connectivity of the uninjected regions using regression on simple anatomical properties.
To better understand the possible role of the weak but above-threshold connections, we perform two graph theoretical analyses on the estimated mouse connectome. Graph theory is a mathematical discipline that abstractly represents and analyses the mouse brain network as a set of nodes (cortical regions) and edges (axonal connections) between them.
First, we separately investigated the set of 5% weakest and 5% strongest above-threshold connections. We mapped these connections in anatomical space and compared their spatial distance distribution, using the Euclidean distance between regional centroids.
Second, we estimated the topological metric of global efficiency for the whole network, using both a weighted graph that considers the estimated connection weights, and a binary graph model network where connectivity weights are thresholded so that edges are either absent or present. We investigated how global efficiency changes as edges are deleted below a continuously variable threshold weight. This procedure is very similar to the calculation of the network density d(x) above, except that the metric requires that all connection weights are known. We therefore restrict our analysis to those target regions that are also the source region for at least one experiment. We make the additional assumption that connections are identical for contralateral homologue regions, i.e. if i and k are homologues, and so are j and l, we have μij = μkl. As 31 regions were injected with at least half of total injection volume in the mouse cortex, we retain 62 regions and 62 × 61 connections.
For arbitrary threshold x, we can then construct the network whose edge weights eij are given by
e i j ( x ) = { 10 μ i j , if μ i j > x 0 , otherwise (8)
We define the distance between two nodes as the inverse, i.e. 1 e i j, which is infinite when the weight is zero. A shortest path between any two nodes A, B is then a sequence of nodes n1, …, nk such that n1 = A, nk = B, eni ni+1 > 0, i = 1, …, k − 1 and the length of the path L A B = ∑ i = 1 k - 1 1 e n i n i + 1 is minimal. Global efficiency is defined as the average inverse shortest path length [28]
G = 1 N ( N - 1 ) ∑ A ≠ B 1 L A B (9)
where N is the number of nodes in the network.
We also compute the fractional size of the largest component as a function of x. A component of the network is a subnetwork in which all nodes are directly or indirectly connected by edges. A fully connected network has only one component; in a network without edges each node is a component. The size of the largest connected component is the number of nodes it contains, which is a number between 1 and the total number of nodes in the network; divided by the number of nodes, this is the fractional size of the largest component, which ranges from 1 N to 1.
The number of region pairs for which there were at least 2 non-zero weights was 492 for the multi-experiment mouse dataset; 153 for the multi-experiment macaque dataset; and 300 for the meta-analytically collated macaque dataset. The corresponding weights are shown in Fig 2. For the mouse dataset, variance for a region pair was V = 0.35 times as large as the total variance in the dataset. The multi-experiment macaque dataset was less variable (V = 0.13), whereas the collated macaque dataset was more variable, with more within-pair variance than between-pair variance (V = 1.3). This result remained unchanged when we restricted the region pairs for which the mean was calculated, or when not log-transforming the measured weights (S1 Appendix). Fig 2 further shows that in the mouse dataset, variance is particularly large for weaker connections: this might well be because measured signal is dominated by false positive noise. Computing the same measure as before with variance calculated over the 50% or 10% weakest weights leads to a stark increase: V0.5 = 0.54, V0.1 = 1.3. This effect was not there for the macaque: V0.5 = 0.16, V0.1 = 0.12.
Fig 5 shows our estimates of overall intra- and inter-hemispheric network density d(x) in the AIBS dataset on the mouse and the Markov et al (2012) dataset on the macaque. To calculate d(x), we only consider those connections whose noise and co-injection threshold is lower than x. We see that overall the mouse has a higher network density than the macaque. We estimate the intrahemispheric network density as 73% (95% CI: 71%, 75%) for mouse and 59% (95% CI: 54%, 62%) for macaque. The interhemispheric network density for mouse was found to be 57% (95% CI: 54%, 59%). Note that our estimates are slightly lower than the sample mean weights [10] because we consider the log-transformed measurements, and E[log(X)] ≤ log(E[X]) (Jensen’s inequality). We find similar values (e.g. mouse intrahemispheric 71% (95% CI: 67%—75%)) when we adjust for the 12 uninjected cortical areas (S1 Appendix).
From Fig 5 it can be seen that the mouse network density increases slowly as the threshold x is lowered from 0 to -1, sharply and nearly linearly as x decreases from -1 to around -5, and then reaches a plateau. The threshold at which the plateau is reached is close to the estimated lower bound of the fraction of signal due to a single projecting neuron (Fig 5). The apparent decrease in network density around x = −7 is somewhat surprising, as the true network density is a decreasing function of x. However, our estimate of d(x) is based on only those connections that can be measured at this level, i.e. whose median noise threshold and median co-injection threshold are smaller than x. Thus, the estimate of d(x) is based on a decreasing number of connections as x decreases. The remaining connections at x = −7 are to the few target regions that have median noise threshold smaller than -7, which can bias the estimate of d(x). The macaque network density shows a similar first slow and then rapid linear increase, but does not reach a plateau. None of the macaque experiments has enough sensitivity to reach the estimated lower bound of the fraction of signal due to a single projecting neuron (Fig 5), possibly reflecting the small injection volume relative to the volume of the injected regions [10].
The 5% strongest (mean log weight (nij)∼10−0 mm-3) and the 5% weakest connections (mean log weight (nij)∼10−5 mm-3) were mapped separately in anatomical space (Fig 6). It is clear by inspection that the strongest connections were more locally and topologically clustered whereas the weakest connections were more random topologically and subtended longer connection distances spatially. The distance distribution is shifted to the right for the weakest connections and the degree distribution has a fatter tail, implying greater probability of high degree hubs, for the strongest connections.
Considering the topology of the whole connectome, we used weighted and binary graph models to investigate how global efficiency and the fractional size of the largest connected component (two metrics of network integration) behaved as a function of variable threshold for edge identification. For both analyses, the network becomes fully connected, i.e. the fractional size of the largest connected component becomes 1, at high thresholds (x ∼ 10−3; Fig 7). In the weighted graph analysis, the network reaches maximal efficiency similarly quickly, when only a relatively small proportion of strongly weighted edges have been included in the network. Addition of weak edges, by lowering the threshold, does not materially affect this metric of weighted graph topology. In the binary graph analysis, progressive lowering of the threshold is associated with a more gradual increase in efficiency, with incremental increases even at the lowest thresholds inclusive of the weakest connections.
We have assessed the variability of three large, quantitative tract-tracing datasets on the axonal connectivity of the mammalian cortex. We found lower variability for expertly curated data on the macaque than for algorithmically segmented data on the mouse; and highest variability for the meta-analytically collated dataset on the macaque. We then articulated a statistical framework to estimate the connection weights, and quantify the uncertainty in these estimates. We applied this approach principally to the the multi-experiment mouse dataset provided by the Allen Institute for Brain Science (AIBS) [9] and estimated the mouse connectome (S3 Table), accounting for false positive signals generated by algorithmic segmentation and for co-injection of tracer into more than one source region in a single experiment. Finally, we have explored some of the biological implications of these results. We found that the connection density of the mouse cortex was considerably higher than previously reported (73%) and that the weakest connections, probably representing no more than a few axonal projections, have a relatively random organisation with modest impact on the topology of the mouse cortical connectome.
Variability in repeated measurements of the same connection arises as a combination of inter-animal biological variability and measurement errors. We quantified the variability of each dataset as the mean variance of repeated measurements of the weight of the same connection normalized by the variance of all connection weights. Ideally, the former should be small compared to the latter. The multi-experiment macaque data from Markov et al [10] are clearly much less variable than the multi-experiment mouse data from Oh et al [9], with variability values of 0.13 and 0.35 respectively. This finding can be partly explained by the inhomogeneity of brain regions. Markov et al. performed repeat injections in the exact same position in the brain region specifically to assess variability, whereas Oh et al. performed injections in variable positions within each source region, specifically in order to increase coverage of cortex by tracer injections. Worryingly, the CoCoMac dataset has variability of 1.3, indicating that there is greater variance of repeated measurements than of all connection weights. This high value may be attributable to technical limitations in the algorithm employed to map all primary studies to the same anatomical atlas [23]. More fundamentally, it may reflect the inevitably lower reliability of measurements made in different labs using different experimental procedures (such as differing planes of section) to measure nominally identical connections [29]. In any case, it seems clear that the pioneering value of meta-analytic approaches to mammalian cortical connectomics has been surpassed by the much higher reliability of next-generation tract-tracing datasets compiled from multiple experiments conducted according to standard experimental protocols.
Our analysis has made clear how the mouse data provided by Oh et al. [9] suffers from at least two sources of noise. Firstly, the segmentation algorithm employed generates many false positive signals. We have quantified the distribution of this segmentation noise, constructing a per-region threshold for the minimum connectivity weight that is sufficient to refute the null hypothesis. Any connection to this region that would result in signal weaker than the threshold cannot be distinguished from positive-mean segmentation noise. The second source of noise in the AIBS datasets arises from co-injections. Among the large number of injection experiments provided by the AIBS—here, 489 experiments each involving a single localised cortical region of anterograde tracer—not all the injections have been constrained to a single anatomical region as pre-defined by an atlas or template. In some experiments, a third or more of total tracer volume will have been injected into one or more spatially adjacent cortical areas. Co-injection of two regions, for example, makes it difficult to disentangle the specific connectivity of each of these two source regions to the same target region. Statistically, this uncertainty about the weight of pair-wise axonal connectivity estimated for co-injected source regions can be represented by a co-injection threshold, which is unique for each pair of regions, and depends on the co-injection pattern of the experiments. Our statistical framework takes into account these two thresholds, and correctly reports high uncertainty when connections are weak relative to the two thresholds.
We have found the intra-hemispheric network density, i.e. the fraction of possible pair-wise connections that exists, of the mouse to be 73% (95% CI: 71%, 75%). This estimate appeared robust; adjusting for uninjected regions using a linear model that accounted for region size and neuronal density yielded a very similar density of 71% (95% CI: 67%—75%; S1 Appendix). These estimates of cortical connection density are higher than previous estimates in mouse [34%-41% (isocortex) [8]; 35.4%-53.5% [9]], but closer to recent estimates of cortical intra-hemispheric connection density in macaque [62% [10]]. There are two reasons we would a priori expect to find a somewhat higher network density for the mouse than macaque. Firstly, the cost of long-range connections increases at a higher rate for larger brains [21], thus we would expect higher network density for smaller brains. Secondly, the mouse has fewer identified cortical areas, and network density increases with decreasing number of brain areas. Thus, the finding of 73% network density seems to corroborate claims that the mammalian cortical network density has long been under-estimated [17]. As a technical caveat, we should note that inter-areal connections in mouse can be completely embedded in grey matter and that the anterograde tracer employed is sensitive to these axonal projections passing through a cortical area en route to termination in another area, which could potentially lead to an overestimation of the network density in mouse. Nevertheless, it is clear that advances in tract-tracing technology have enabled detection of axonal connectivity over 6 orders of magnitude with just-detectable connections constituting very weak links compared to the nearly million-fold greater connectivity weight of the most strongly weighted connections.
The biological meaning of such unprecedentedly weak axonal connectivity is not immediately clear. We computed a simple approximation of the connectivity weight attributable to a single axonal projection, by dividing the total number of neurons in mouse cortex (4 × 106) by the number of cortical regions (43) so that average total tracer signal from any region was attributable to at most 4 × 10 6 43 axonal projections. This single axon threshold corresponds to connection weights ∼10−5. The single axon threshold corresponds reasonably well with the point at which the estimated mouse network density, both intra- and inter-hemispheric, reaches a plateau. This observation is consistent with the log-normal distribution of connection weights [16, 30] extending to the scale of a single axonal projection, which would maximize the dynamic range of connection weights hypothesised to be important for brain function [31].
Weak axonal projections, down to the limit of a single axon connecting two cortical areas, should not be discounted as organisationally trivial. In social and other complex networks [32] weak links are well-recognised to serve important integrative functions in social functions, e.g., mediating information transfer, or gossip, between otherwise isolated cliques. In the macaque brain, it has been consistently argued that weak connections may have functionally important effects on brain dynamics by orchestrating oscillatory coherence of anatomically distributed cortical areas [10, 14, 16, 17, 33]. In the mouse connectome, the weakest links were topologically more random than the strongest links, and they traversed greater anatomical distances than the more locally clustered or lattice-like strongest links. These properties are conceptually consistent with the weakest links of the mouse connectome playing a similarly integrative role to weak links in social networks [34]. However, it is difficult to say how functionally important a single axonal projection to a cortical area comprising ∼105 neurons is likely to be in real-life. In a graph theoretical analysis weighted by axonal connectivity over 6 orders of magnitude, the topology of the network is dominated by strongly connected edges. The network is node-connected and has maximal global efficiency at a 10−2 weight threshold, equivalent to a network density of 7.5%. Further relaxation of the weight threshold does not substantially change the global topological properties established by the most strongly connected edges, consistent with results found in macaque [14]. However, if the network is modelled as a binary graph, effectively equalizing all true connection weights, then we can see theoretically expected incremental increases of network efficiency and integration as progressively weaker edges are added by relaxation of the edge threshold down to the minimum imposed by experimental noise. Future studies may explore this question in more depth by explicitly considering the topological properties of those weak connections that are estimated to exist with high confidence.
One way of thinking about these observations is that the weakest links add randomness to mouse connectome topology and integrative capacity to mouse connectome function. Prior results on retrograde fluorescent tract-tracing data on the macaque [10, 14, 16, 33] likewise found that the weaker connections (new found projections) were topologically integrative; but also demonstrated that the new found projections shared an anatomically specific profile of inter-areal connectivity in common with stronger (previously known) connections between cortical areas. The anatomical specificity of new found projections in the macaque cortex suggests that they are not as randomly organised as the weakest connections of the mouse cortex reported here. However, it is important to recognise that new found projections constituted approximately the weakest 36% of all inter-areal connections (on average over areas) in the macaque [33]; whereas we have focused on the weakest 5% of all connections in the mouse. It is plausible that the topological randomness of weak links may increase monotonically as a function of increasing weakness. In other words, the greater randomness of these results on the mouse may not be attributable to inter-species differences but rather to our focus on a smaller subset of more extremely weak links than the relatively larger subset of less weak new found projections. Future comparative studies would be useful to test this prediction more rigorously. Future studies of the mouse connectome might also explore the generative hypothesis that the random topology of the weakest links reflects their formation by stochastic processes of cortical network development. For example, synaptic connectivity of cortical neurons peaks in early post-natal years for mammals and there typically follows a prolonged phase of synaptic pruning or deletion of functionally unimportant, aberrant or over-exuberant synaptic connections and associated axonal projections. Therefore, one possible developmental mechanism for the randomly organised weak links of the adult mouse connectome could be that they reflect what remains after pruning of functionally prioritised connections. A testable prediction of this hypothesis is that the weight of the weakest adult connections should rise and then fall during post-natal connectome development.
In conclusion, we have provided a statistical framework to analyse tract tracing data, yielding estimates of connection weights and their associated uncertainty. We have provided these estimates and thresholds for the mouse connectome, such that they may be analysed by other researchers (S3 Table). We have drawn attention to the higher-than-expected maximum cortical connection density of the mouse, attributable to the precision of the measurements allowing resolution of tracer signals in the order of a single axon. These very weak inter-areal axonal projections are not yet completely understood biologically. Like weak links in social networks, they could be topologically integrative and important for global information transfer and/or the weakest (ultimately single axon) projections could be the lucky survivors of a developmental pruning process programmed to entirely eliminate them. It seems likely that functional importance will go with increasing axonal connectivity weight, so the very weakest connections are least likely to be functionally important for inter-areal communication; but the minimum weight needed for functional importance, or the functional threshold on axonal connectivity, is not yet known.
|
10.1371/journal.pbio.1000436 | Tracking Marsupial Evolution Using Archaic Genomic Retroposon Insertions | The Australasian and South American marsupial mammals, such as kangaroos and opossums, are the closest living relatives to placental mammals, having shared a common ancestor around 130 million years ago. The evolutionary relationships among the seven marsupial orders have, however, so far eluded resolution. In particular, the relationships between the four Australasian and three South American marsupial orders have been intensively debated since the South American order Microbiotheria was taxonomically moved into the group Australidelphia. Australidelphia is significantly supported by both molecular and morphological data and comprises the four Australasian marsupial orders and the South American order Microbiotheria, indicating a complex, ancient, biogeographic history of marsupials. However, the exact phylogenetic position of Microbiotheria within Australidelphia has yet to be resolved using either sequence or morphological data analysis. Here, we provide evidence from newly established and virtually homoplasy-free retroposon insertion markers for the basal relationships among marsupial orders. Fifty-three phylogenetically informative markers were retrieved after in silico and experimental screening of ∼217,000 retroposon-containing loci from opossum and kangaroo. The four Australasian orders share a single origin with Microbiotheria as their closest sister group, supporting a clear divergence between South American and Australasian marsupials. In addition, the new data place the South American opossums (Didelphimorphia) as the first branch of the marsupial tree. The exhaustive computational and experimental evidence provides important insight into the evolution of retroposable elements in the marsupial genome. Placing the retroposon insertion pattern in a paleobiogeographic context indicates a single marsupial migration from South America to Australia. The now firmly established phylogeny can be used to determine the direction of genomic changes and morphological transitions within marsupials.
| Ever since the first Europeans reached the Australian shores and were fascinated by the curious marsupials they found, the evolutionary relationships between the living Australian and South American marsupial orders have been intensively investigated. However, neither the morphological nor the more recent molecular methods produced an evolutionary consensus. Most problematic of the seven marsupial groups is the South American species Dromiciops gliroides, the only survivor of the order Microbiotheria. Several studies suggest that Dromiciops, although living in South America, is more closely related to Australian than to South American marsupials. This relationship would have required a complex migration scenario whereby several groups of ancestral South American marsupials migrated across Antarctica to Australia. We screened the genomes of the South American opossum and the Australian tammar wallaby for retroposons, unambiguous phylogenetic markers that occupy more than half of the marsupial genome. From analyses of nearly 217,000 retroposon-containing loci, we identified 53 retroposons that resolve most branches of the marsupial evolutionary tree. Dromiciops is clearly only distantly related to Australian marsupials, supporting a single Gondwanan migration of marsupials from South America to Australia. The new phylogeny offers a novel perspective in understanding the morphological and molecular transitions between the South American and Australian marsupials.
| The phylogenetic relationships among the four Australasian and three South American marsupial orders have been intensively debated ever since the small species Dromiciops was taxonomically moved from Didelphimorphia into the new order Microbiotheria and the cohort Australidelphia was erected based on ankle joint morphology [1]. Australidelphia comprises the four Australasian marsupial orders and the South American order Microbiotheria, a close relationship suggesting a complex ancient biogeographic history of marsupials. However, the exact phylogenetic position of Microbiotheria within Australidelphia has so far eluded resolution. Moreover, sequence-based attempts to resolve the positions of the South American opossums (Didelphimorphia) and the shrew opossums (Paucituberculata), which appear some few million years apart in the South American fossil layers close after the Cretaceous-Tertiary boundary [2], relative to Australidelphia have so far been futile (e.g., [3],[4]).
The two recently sequenced marsupial genomes of the South American opossum (Monodelphis domestica) [5] and a kangaroo, the Australian tammar wallaby (Macropus eugenii), provide a unique opportunity to apply a completely new approach to resolve marsupial relationships. The insertion patterns of retroposed elements, pieces of DNA that are copied via RNA intermediates and pasted randomly elsewhere in the genome, have successfully resolved the more than 130 million-year-old branch of therian mammals [6] and early placental mammalian divergences [7] as well as relationships within other mammalian orders [8]. Because the insertion sites are effectively random and parallel insertions or exact excisions are very rare [9], the shared presence of retroposed elements at identical orthologous genomic locations of different species, families, or orders is a virtually homoplasy-free indication of their relatedness. Thus, the interpretation of retroposon markers is simple and straightforward: the presence of one of these elements in the orthologous genomic loci of two species signals a common ancestry, while its absence in another species signals a prior divergence [10]. No other sequenced mammalian genome has shown as high a percentage of discernible retroposed elements as marsupials (52%) [5], an extremely large number of possible informative markers.
In addition, because young retroposed elements can insert into older elements, but older, inactive elements are not capable of inserting into younger ones, nested retroposon insertion patterns provide invaluable information about the relative times during which given retroposon families integrated into genomes. We used the transposition in transposition (TinT) application [11] to screen for such nested transpositions and to provide a complete picture of the succession of ancient retroposon activities so as to aid in the proper selection of element groups for resolving different parts of the marsupial tree.
After a complete screening of the opossum and kangaroo genomic sequences using the TinT algorithm, we recovered 8,245 and 4,499 nested retroposon insertions, respectively (Table S1). We then calculated the frequencies and time scales of short interspersed element (SINE) insertions using the likelihood approach implemented in TinT. The resulting pattern (Figure 1) revealed three different groups of retroposed SINEs: (1) elements specific for the lineage leading to opossum (RTESINE1, SINE1_Mdo, SINE1a_Mdo), (2) elements specific for the lineage leading to kangaroo (WALLSI1-4, WSINE1), and (3) a compiled group of elements active in both marsupial lineages. These three groups of elements were then used as a basis to screen for phylogenetically informative markers present in (1) the opossum lineage, (2) the branches leading to kangaroo, and (3) to find marsupial monophyly markers.
Three different search strategies (see Materials and Methods) revealed ∼217,000 retroposon-containing genomic loci. Highly conserved exonic primers were generated for 228 loci and experimentally tested on a small set of species. After carefully screening the sequences, we selected 32 loci based on criteria outlined in the Materials and Methods section for amplification in 20 marsupial species (Table S2). We carefully aligned and analyzed approximately 440 marsupial sequences to reveal 53 informative markers (Figure 2, Table 1).
Ten of the phylogenetically informative markers accumulated in the metatherian genome since their split from placental mammals, approximately 130 million years ago (MYA) [12],[13], and before the earliest divergence of the modern marsupial mammals, 70–80 MYA [3],[14]. All ten are absent in other mammals, significantly confirming the monophyly of marsupials (p = 2.0×10−5; [10 0 0] [15]).
The other 43 phylogenetically informative retroposon markers provide significant support for most of the basal splits within marsupials. The earliest marsupial divergence was previously impossible to resolve based on sequence data, which could not distinguish between Paucituberculata and Didelphimorphia as the sister group to Australidelphia [14],[16]–[19]. We identified two markers (MIR3_MarsA) in the South American shrew opossums (Paucituberculata) that were also present in all Australidelphia but absent in Didelphimorphia (Figure 2). Albeit not significant (p = 0.1111; [2 0 0]), this is the first molecular support for the earliest branching of Didelphimorphia, establishing it as the sister group to the remaining six marsupial orders. However, as significant support for this important marsupial branch requires three or more conflict-free markers [15], we attempted to find additional retroposons for the marsupial root. To find the third marker for the supported topology (Figure 2), a MIR3_Mars element present in kangaroo plus Paucituberculata but absent in opossum, we recovered ten additional loci from in silico screening; two contained the previously detected markers and eight contained new retroposons. Unfortunately, experimental verification showed that the absences of MIR3_Mars in opossum were due to non-specific deletions. On the other hand, we also did not find any loci with MIR3_Mars elements present in opossum plus Paucituberculata but absent in kangaroo, which would have supported the alternative of a close relationship between Didelphimorphia and Paucituberculata. We then screened for markers that would support the alternative hypothesis of Paucituberculata being the sister to all marsupials by performing an exhaustive in silico pre-screening for orthologous MIR3_Mars elements present in short introns of opossum and kangaroo. Starting from ∼6,000 potentially informative loci, we selected 39 highly conserved MIR3_Mars-containing introns. However, experimental verification showed that all of the elements were also present in the order Paucituberculata (Rhyncholestes), thus supporting the monophyly of marsupials (data not shown), but not the basal divergence.
Assuming that, in the entire genomes, there are more than just the two detected diagnostic insertions for the root, an expanded search including larger introns and conserved intergenic regions is required to find significant support for this branch. Such relaxed search conditions are expected to provide a huge number of additional markers spread over the entire marsupial tree, but will require extensive additional computational and experimental work.
Molecular estimates have placed the earliest divergences of Marsupialia in the Late Cretaceous, 65–85 MYA [3],[4],[14]. To resolve placental mammalian Cretaceous divergences [20], large amounts of sequence data were crucial to gain sufficient phylogenetic signal, which is a plausible explanation for the difficulties encountered in trying to resolve this branch in previous marsupial investigations [3],[4],[14]. However, morphological data have revealed several characters from the skull and postcranium, supporting Didelphimorphia as the sister to all marsupials [21], consistent with our two molecular markers.
Leaving the base of the tree for the time being, 13 of the original 53 markers were present in the South American Microbiotheria and the four Australasian orders but not in either Didelphimorphia or Paucituberculata, significantly supporting the monophyly of Australidelphia [1] (p = 6.3×10−7; [13 0 0]; Figure 2). The large number of phylogenetically informative markers indicates a long phylogenetic branch and/or a high degree of retroposon activity and fixation in the ancestral Australidelphia lineage. The branch separating Australidelphia from Didelphimorphia and Paucituberculata is one of the strongest supported and evolutionarily longest inter-ordinal branches in the marsupial tree [3],[4]. The fossil Australian marsupial Djarthia murgonensis is the oldest, well-accepted member of Australidelphia. Thus, combined with the lack of old Australidelphian fossils from South America, the most parsimonious explanation of the biogeography of Australidelphia is of an Australian origin [22]. However, the poor fossil record from South America, Antarctica, and Australia does not exclude that Djarthia, like Dromiciops, could be of South American origin and had a pan-Gondwanan distribution. Additional fossils from Australia or South America will shed more light on the early Australidelphian relationships and their biogeography.
Four markers significantly support the monophyletic grouping of the four Australasian orders to the exclusion of Microbiotheria (p = 0.0123; [4 0 0]; Figure 2). Several studies have presented evidence for the monophyly of the Australasian orders; these have typically been based solely on nuclear protein-coding genes such as ApoB, BRCA1, IRBP, RAG1, and vWF [4],[17],[19], albeit with relatively low support values. By contrast, other sequence-based studies, relying completely or partially on mitochondrial data, find the South American order Microbiotheria nested within the Australasian orders [3],[16],[23]. Thus, the two competing hypotheses, Microbiotheria nested within or outside Australasian orders, have confounded the search for a reliable marsupial phylogeny.
Two studies tried to combine the nuclear and mitochondrial data using different approaches to achieve a larger dataset with higher probability of resolving the marsupial phylogeny [14],[18]. Only R/Y-coding, removing of sites [18], or partitioning [14] reduced possible artefacts from the mitochondrial data enough to reach a topology consistent with the retroposon markers. However, both studies gave low support for the position of Microbiotheria, illustrating the difficulties in resolving a short branch using sequence data under difficult conditions, such as possible nucleotide composition bias problems and randomization of fast evolving sites. The support from two independent sources of phylogenetic information, our retroposon markers and nuclear genes [4],[17],[19], invalidates the mitochondrial results [3],[16],[23]. Complete mitochondrial genomes can give misleading signals, as was demonstrated for the incorrect position of Monotremata among mammals [24], and can even mislead phylogenetic reconstruction when mixed with nuclear data.
The position of Microbiotheria has been intensely debated since the cohort Australidelphia was first suggested based on tarsal evidence [1]. After decades of uncertainty derived from molecular and morphological data, we have uncovered four independent diagnostic retroposon insertions that finally place the South American order Microbiotheria at its correct place in the marsupial tree (Figure 2). Therefore, we propose the new name Euaustralidelphia (“true Australidelphia”) for the monophyletic grouping of the four Australasian orders Notoryctemorphia, Dasyuromorphia, Peramelemorphia, and Diprotodontia.
The relationship among the four Australasian orders is not resolved, and of special interest is the phylogenetic position of the marsupial mole, Notoryctes typhlops, which has been debated for a long time [3],[4],[14],[17]–[19],[21],[23]. The marsupial mole is the only burrowing marsupial and is found in the deserts of Australia. The eyes of the marsupial mole are vestigial and the fore- and hind limbs are morphologically derived due to the burrowing lifestyle. The derived morphology and the fact that the marsupial mole is the single species in the order Notoryctemorphia have complicated attempts to resolve its phylogenetic position relative to the other three Australian orders. Most analyses of molecular sequence data find the marsupial mole closely related to the orders Dasyuromorphia and Peramelemorphia, but the support values are generally weak [3],[4],[14],[17]–[19],[23], and the exact phylogenetic position relative to the other two orders is yet to be determined. During the retroposon screening one marker was found supporting a grouping of Notoryctes, Dasyuromorphia, and Peramelemorphia (p = 0.3333 [1 0 0]). The single retroposon marker is in agreement with the results from the sequence data. Extended screening of retroposons can provide additional evidence for the position of the marsupial mole among marsupials and which of the orders, Dasyuromorphia or Peramelemorphia, is the sister group.
Of the original 53 markers, 18 of them provide significant support for the monophyly of each of the five multi-species marsupial orders: five for Didelphimorphia (p = 0.0041; [5 0 0]), three each for Paucituberculata, Dasyuromorphia, and Diprotodontia (p = 0.037; [3 0 0]), and four for Peramelemorphia (p = 0.0123; [4 0 0]). Four of the remaining markers provide non-significant support for various intra-ordinal relationships of Diprotodontia (Figure 2). Two of them support the division between Vombatiformes (wombats and koala) and Phalangerida (kangaroos, possums) (p = 0.1111; [2 0 0]), challenging the results from mitochondrial sequence-based studies ([3], but see [25]), and one marker each supports the grouping of the possums Tarsipes and Pseudocheirus and that of the kangaroos Macropus and Potorous (p = 0.3333 [1 0 0]). One final marker supports the grouping of the Didelphis and Metachirus.
The outstanding advantage of using retroposon presence/absence data for phylogenetic reconstructions is the low probability of insertion homoplasy. Independent parallel insertions of identical elements or exact deletions are extremely rare [9], but nevertheless not completely negligible, especially after genome-wide in silico screening of rare informative loci. LINE1-mobilized elements, in particular, show a slight preference for a TTAAAA consensus insertion motif [26], but on the other hand, such elements are rare in the deep phylogenetic branches of marsupials (Figure 1; Figure S1). Excluding the more frequent near identical insertions or unspecific deletions requires careful aligning and interpretation of orthologous informative markers (see Materials and Methods and Dataset S1).
Another possible source of errors is incomplete lineage sorting (polymorphism during speciation) or ancestral hybridization that can affect any marker system. Particularly short internal branches of a tree (rapid speciation) and biased in silico pre-screening for potential phylogenetically informative loci are exposed to such effects [27].
The available genomes of the opossum and the kangaroo placed us in the advantageous situation of independently pre-screening two distant branches of the marsupial tree. All 53 experimentally verified markers confine a phylogenetic tree free of any marker conflicts. Fourteen of them were randomly inserted as a second marker in specific loci. For most internal branches we found significant support for the underlying prior hypothesis by three or more markers with a clear rejection of alternative hypotheses.
Given the limitations just mentioned, the retroposon marker system identified a clear separation between the South American and Australasian marsupials. Thus, the current findings support a simple paleobiogeographic hypothesis, indicating only a single effective migration from South America to Australia, which is remarkable given that South America, Antarctica, and Australia were connected in the South Gondwanan continent for a considerable time.
The search for diagnostic South American or Australidelphian marsupial morphological characters has been so far confounded by the lack of a resolved marsupial phylogeny [21],[22],[28]. The newly established marsupial tree can now be applied not only to morphological and paleontological studies but also to clearly distinguish genomic changes.
The marsupial classification of Aplin and Archer [29] has been followed throughout the text. Representatives of all seven marsupial orders were included for retroposon screening. Except for the two single-species orders, at least two species per order were investigated. For all orders except Didelphimorphia, representative species were chosen to cover the deepest splits within each order. Didelphimorphia: Monodelphis domestica (gray short-tailed opossum), Didelphis virginiana (Virginia opossum), Metachirus nudicaudatus (brown four-eyed opossum). Paucituberculata: Rhyncholestes raphanurus (Chilean shrew opossum), Caenolestes fuliginosus (silky shrew opossum). Microbiotheria: Dromiciops gliroides (monito del monte). Notoryctemorphia: Notoryctes typhlops (marsupial mole). Dasyuromorphia: Phascogale tapoatafa (brush-tailed phascogale), Dasyurus geoffroii (western quoll), Sminthopsis crassicaudata (fat-tailed dunnart), Myrmecobius fasciatus (numbat). Peramelemorphia: Macrotis lagotis (bilby), Perameles gunnii (eastern barred bandicoot), Isoodon obesulus (southern brown bandicoot). Diprotodontia: Tarsipes rostratus (honey possum), Pseudocheirus peregrinus (common ringtail possum), Trichosurus vulpecula (common brushtail possum), Macropus robustus (wallaroo), Potorous tridactylus (long-nosed potoroo), Vombatus ursinus (common wombat).
The marsupial genome harbors about 500 different families of interspersed repeats [30]. Several retroposon families were active around and after the split of Australasian [31] and South American marsupials and potentially encrypt information about their phylogeny. For successful and focused experimental retroposon screening it is invaluable to have, a priori, a map of the ancestral retroposon activities. The previously developed TinT method [11] relies on a numeral compilation (Table S1) of nested transpositions (TinT) extracted from RepeatMasker coordinates and visualized after calculating their maximal activity probabilities. For experimental application, 24 subtypes of small SINE elements, active over the range of marsupial evolution, were pre-selected for the TinT analysis (Figure 1). The complete statistics of SINE elements in M. domestica and M. eugenii are given in Figure S2.
The assembled genome of M. domestica (MonDom5) and the draft genome of M. eugenii were used to pre-select potential phylogenetically informative intronic retroposon loci. Three different in silico high-throughput strategies, implemented in specially developed C-scripts, were applied to extract the genomic information.
The 228 loci extracted by these three strategies were experimentally analyzed in a small subset of eight representative marsupial species (see strategy A). The sequences from the experimental screening were aligned and carefully inspected for (1) identical genomic insertion points of retroposed elements, (2) identical element orientation, (3) identical element subtypes, (4) as far as available, concurrent element flanking repeats, (5) shared diagnostic indels, and (6) the consistency of insertion in representative species. The 32 selected loci mentioned above (in A–C) were determined to be phylogenetically informative (elements present at orthologous genomic locations in two or more species) and were screened in a larger taxon sampling comprised of 20 marsupials covering all seven orders (see taxon sampling). After sequencing, 53 phylogenetically informative retroposon markers were identified from the 32 introns. More than one informative marker was recovered in each of 15 of the introns, due to independent retroposon insertions (Table 1), and an additional 18 autapomorphic insertions were found.
Total DNA was extracted from tissues using the standard phenol-chloroform protocol [32]. Approximately 10–50 ng DNA was used in each 25 µl PCR amplification using ThermoPrime Taq (ABgene, Hamburg) with 1.5 mM MgCl2. All PCR reactions were prepared for high throughput in 96-well plates and the DNA was amplified using the touchdown PCR strategy, decreasing the annealing temperature stepwise by 1°C for the initial ten cycles, followed by 25 cycles at 45°C annealing temperature (for primers see Table S3). The initial screening was performed using eight representative marsupial species (see above) and PCR products were visualized on 1% agarose gels to detect presence/absence patterns via the size shifts of fragments. The PCR products indicating such size shifts were purified and ligated into the TA cloning vector pDrive (Qiagen, Hilden). Ligations were left overnight at 7°C and transformed into XL1-Blue competent cells. Colonies were PCR screened using standard M13 primers. For each positive PCR product, at least two colonies were sequenced. All sequence alignments were conducted using Se-Al [33]. Sequences were screened for retroposons using the RepeatMasker program (http://www.repeatmasker.org/RMDownload.html) and a specific retroposon library (available upon request).
From the markers in Table 1 we built a presence/absence (1/0) data matrix of retroposons (Figure S3). The strict consensus, most parsimonious tree was reconstructed using the irrev.up option of character transformation implemented in PAUP*4.0b10 [34] in a heuristic search performed using 1,000 random sequence addition and tree-bisection and reconnection (TBR) branch swapping. Because strictly marsupial-specific retroposons were investigated, the hypothetical human outgroup was coded 0. The resulting tree had a length of 53 and a consistency index of 1. The tree topology shown in Figure 2 refers to the derived parsimony tree. Due to the complexity and randomness of retroposon insertions, there are an extremely large number of possible unique character states (insertion sites), and maximum parsimony analyses converge to maximum likelihood estimators [35]. Evidence from retroposon markers is considered to be statistically significant when three or more markers are found supporting one node (i.e., when p<0.05) [15].
|
10.1371/journal.pntd.0002661 | A Polarized Cell Model for Chikungunya Virus Infection: Entry and Egress of Virus Occurs at the Apical Domain of Polarized Cells | Chikungunya virus (CHIKV) has resulted in several outbreaks in the past six decades. The clinical symptoms of Chikungunya infection include fever, skin rash, arthralgia, and an increasing incidence of encephalitis. The re-emergence of CHIKV with more severe pathogenesis highlights its potential threat on our human health. In this study, polarized HBMEC, polarized Vero C1008 and non-polarized Vero cells grown on cell culture inserts were infected with CHIKV apically or basolaterally. Plaque assays, viral binding assays and immunofluorescence assays demonstrated apical entry and release of CHIKV in polarized HBMEC and Vero C1008. Drug treatment studies were performed to elucidate both host cell and viral factors involved in the sorting and release of CHIKV at the apical domain of polarized cells. Disruption of host cell myosin II, microtubule and microfilament networks did not disrupt the polarized release of CHIKV. However, treatment with tunicamycin resulted in a bi-directional release of CHIKV, suggesting that N-glycans of CHIKV envelope glycoproteins could serve as apical sorting signals.
| Polarized cells are found in many parts of the human body and are characterized by the presence of two distinct plasma membrane domains: the apical domain facing the lumen and the basolateral domain facing the underlying tissues. Polarized epithelial cells line the major cavities of our body, while polarized endothelial cells line the blood-tissue interface, both of which protect our body against the invasion of biological pathogens. Thus, many pathogens have to invade the monolayer of epithelial or endothelial cells in order to establish infection. During infection with Chikungunya virus, a mosquito vector bites a human host and inoculates the virus into the host's bloodstream. In recent epidemics of Chikungunya infection, more severe clinical manifestations such as neurological complications were observed. As such, we studied the infection of Chikungunya virus in polarized cells in an aim to provide explanations for the more severe pathogenesis observed.
| Chikungunya virus (CHIKV) belongs to the Alphavirus genus of the Togaviridae family. It is a spherical, enveloped virus of 60 to 70 nm diameter that consists of the major structural proteins Capsid, E2 and E1, and a single-stranded, positive-sense RNA genome of 11.8 kb [1]. CHIKV was first isolated in Tanzania in 1952 during the earliest recorded Chikungunya epidemic [2] and has since caused outbreaks in East Africa, South Africa and Southeast Asia [3]. CHIKV re-emerged in the recent epidemic outbreaks, including the largest documented outbreak of CHIKV in the Indian Ocean islands of Mayotte, Mauritius, La Réunion, and the Seychelles between 2005 and 2006 [3] and in India between 2006 and 2008 [4]–[6]. Since then, CHIKV has caused outbreaks in many parts of the world, including Singapore [7], Malaysia [8] and Europe [9], [10].
CHIKV infection causes a range of clinical manifestations including high fever, headache, erythematous skin rash and incapacitating arthralgia [2]. Chikungunya disease is generally a self-limiting illness. However, the symptoms of the illness, rheumatological manifestations in particular, may be chronic and persist for several months. Additionally, the recent outbreaks of Chikungunya are associated with unusual severity and neurological complications such as encephalitis [1], [11]–[13].
Upon being bitten by a CHIKV-infected mosquito, CHIKV enters the bloodstream of the human host. It is currently unknown how CHIKV infection leads to encephalitis in the recent re-emergences of Chikungunya disease. One postulation is that CHIKV may migrate across the blood-brain barrier from the blood capillaries into the brain cells in order to cause neurological complications. The key structural elements of the blood-brain barrier are the tight junctions between adjacent brain capillary endothelial cells, which act as a barrier to prevent the diffusion and invasion of blood-borne pathogens from the bloodstream into the brain tissues and protect the brain from blood-borne toxic compounds and pathogens [14], [15].
Polarized cells, including the endothelial cells lining the brain capillaries, are characterized by the presence of two distinct plasma membrane domains: the apical domain facing the lumen and the basolateral domain facing the underlying tissues. Sorting machineries within polarized cells recognise apical and basolateral sorting signals such as peptide motifs and post-translational modifications on proteins and transport them specifically to their respective domains. Following polarized sorting of proteins, the tight junctions at cell-cell contacts prevent the movement of proteins between the two domains and maintain the unique protein composition of each domain [16]. These discrete membrane domains function in the selective absorption and release of many proteins and pathogens.
Polarized epithelial cells line the major cavities of the body and polarized endothelial cells line the blood-tissue interface, both of which form a selective barrier against the invasion of many pathogens. In order to establish infection, many pathogens have to invade the monolayer of epithelial or endothelial cells [17]–[20]. Several viruses have been shown to display polarized entry and/or release in cellular models. For example, the entry and release of West Nile Virus [18], Hepatitis A Virus [21] and Simian Virus 40 [17] occur preferentially at the apical surface. In comparison, the entry and release of Semliki Forest Virus [22], [23] and Vesicular Stomatitis Virus [19] occur preferentially at the basolateral surface.
The polarized infection of CHIKV may provide insights to the pathogenesis of the virus and the mechanisms involved in how the virus crosses the polarized blood-brain barrier in the establishment of neurological complications. Thus, this study aims to establish a polarized cellular model for CHIKV infection in order to investigate whether the entry and release of CHIKV is polarized. We also examined host cell and viral factors that may be involved in the polarized sorting and release of CHIKV at specific domains of the host cell.
Non-polarized African Green Monkey kidney epithelial cells (Vero) and polarized Vero C1008, both from American Type Culture Collection, were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal calf serum (FCS). Human Brain Microvascular Endothelial Cells (HBMEC) from ScienCell were maintained in Endothelial Cell Medium (ECM) supplemented with 5% fetal bovine serum (FBS) and 1% endothelial cell growth supplement. The CHIKV strain used in this study, D1225Y08, was a kind gift from the Environmental Health Institute, National Environment Agency. D122508Y08 was isolated from the serum of a febrile patient during the 2007 to 2008 Chikungunya outbreaks in Singapore. The CHIKV122508 virus used in this study is a low passage virus that was cultured more than 5 passages in C6/36 cells derived from Aedes albopictus in Leibovitz's L15 medium supplemented with 2% FCS.
4×104 HBMEC cells (passage 2) were seeded on glass cover slips coated with 4 µg/cm2 fibronectin in 24-well plates. 5×104 Vero C1008 cells were seeded on uncoated glass cover slips in 24-well plates. The cells were infected with CHIKV at a multiplicity of infection (MOI) of 10. CHIKV-infected HBMEC cells were maintained in ECM medium supplemented with 5% FBS and the supernatant was harvested at 0, 12, 24, 36, 48, 72, 96 and 120 hours post-infection (h.p.i.). CHIKV-infected Vero C1008 cells were maintained in DMEM medium supplemented with 2% FCS and the supernatant was harvested at 6-hours intervals up to 48 h.p.i.. The supernatants were subjected to viral plaque assays to quantify the virus titer. CHIKV-infected HBMEC and Vero C1008 cells were viewed under the differential interference contrast microscope at 24-hours and 12-hours intervals, respectively, to observe for morphology changes post-infection. Immunofluorescence assay was performed to detect CHIKV protein expression in the CHIKV-infected HBMEC and Vero C1008 cells.
The integrity of HBMEC, Vero C1008 and Vero cell monolayers was assessed to ensure that the cell monolayers remained intact and to prevent exchange of materials between the apical and basolateral chambers such that a polarized infection of CHIKV on the cells can be set up and that the virus titers obtained from the apical and basolateral chambers would be representative of the viruses released from the apical and basolateral domain, respectively.The integrity of HBMEC, Vero C1008 and Vero cell monolayers was assessed by measuring the trans-epithelial electrical resistance (TEER) using the Millicell-ERS apparatus (Millipore). Vero and Vero C1008 cell monolayers with TEER values between 30 and 70 Ω/cm2 and HBMEC cell monolayers with TEER values of approximately 20 Ω/cm2 were used for polarized infection studies. The integrity of the cell monolayers was assessed again post-infection by measuring the TEER detecting for ZO-1 tight junction proteins by immunofluorescence assay, and assaying for the permeability of the cell monolayers to FITC-dextran. In brief, FITC-dextran was applied to the apical chamber and incubated for 20 minutes, after which the percentage of FITC-dextran flow-through was calculated by the fluorescence reading in the basolateral chamber over the fluorescence reading in the apical chamber. 100 ng/ml tumour necrosis factor (TNF) was applied to the cells to increase the cel monolayer permeability as a positive control.
2×105 HBMEC (passage 4), 5×104 Vero C1008 and 5×104 Vero cells were individually seeded on cell culture inserts with pores of 0.4 µm diameter (Greiner) and infected with CHIKV either apically or basolaterally at an MOI of 10. At 24 h.p.i., supernatants from the apical and basolateral chambers were collected separately for plaque assays to quantify the virus yields. To investigate the polarized entry of CHIKV, the total virus yield post-apical infection was compared to the total virus yield post-basolateral infection. The virus binding assay was also performed to further confirm the polarized entry of CHIKV. On the other hand, to investigate the polarized release of CHIKV, the amount of virus released into the apical and basolateral chambers were quantified separately by viral plaque assays and compared to determine whether the release of CHIKV in Vero C1008 and HBMEC cells is polarized. The cells were also fixed at 24 h.p.i., subjected to immunofluorescence assay and analyzed by confocal microscopy to visualize the localization of CHIKV protein expression.
The virus binding assay was performed to analyze of the polarized entry of CHIKV in Vero C1008 cells. Vero and Vero C1008 cells were seeded on porous cell culture inserts and apically- or basolaterally-infected with CHIKV at an MOI of 10 at 4°C. After 1.5 hours incubation to allow CHIKV to bind to cell surface receptors, the cells were washed with 1×PBS four times and fixed with 4% paraformaldehyde. Immunofluorescence assay was performed to label CHIKV virus particles with FITC fluorochrome to quantify the amount of CHIKV bound to the cells. The specimens were viewed under the confocal microscope and analysed with the Imaris software to calculate the average number of FITC flurochrome spots per cell (DAPI-stained).
Cells were fixed with 4% paraformaldehyde for 10 minutes and permeabilised with 0.01% Triton X-100 for 15 minutes. The cells were then incubated with the desired primary antibodies at 37°C for 1 hour, followed by species-specific secondary antibodies at 37°C for 1 hour. In the growth kinetics studies, the samples were probed with primary antibodies against Alphaviruses (Santa Cruz sc-58088, 1∶250 dilution), followed by FITC-conjugated secondary antibodies (Millipore, 1∶500 dilution). In the polarized infection studies, the samples were co-labeled with primary antibodies against CHIKV E2 glycoproteins (1∶100 dilution) and FITC-conjugated secondary antibodies (Millipore, 1∶250 dilution) together with primary antibodies against ZO-1 proteins (Invitrogen, 1∶500 dilution) and Dylight-594-conjugated secondary antibodies (Thermo Scientific, 1∶100 dilution). In the virus binding assay, the samples were probed with primary antibodies against Alphaviruses (Santa Cruz, 1∶250 dilution), followed by FITC-conjugated secondary antibodies (Millipore, 1∶500 dilution). Cell nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI) at room temperature for 20 minutes. 1×PBS washes were performed after each incubation step. The samples were subsequently mounted onto glass slides using DABCO and viewed under the Olympus IX81 inverted microscope or Olympus FV1000 confocal microscope.
Viral RNA was extracted (Qiamp viral RNA kit; Qiagen) from the virus supernatant collected from the upper and lower chambers of the cell culture insert after apical and basolateral infection with CHIKV. One-step SYBR green-based RT-PCR CHIKV viral RNA quantification was described previously [24] using the SYBR Green Quantitative RT-PCR Kit (Sigma-Aldrich, QR0100) in the ABI PRISM 7000 RT-PCR system. CHIKV samples were assayed with a concentration (250 nM) of the nsP2 primers (nsP2 forward primer: GGCAGTGGTCCCAGATAATTCAAG; nsP2 reverse primer: GTACATACCCCACCTAGATCTGTCG) in a 1× final concentration of SYBR green Taq Ready Mix for Quantitative RT-PCR (1× Taq DNA polymerase, 10 mM Tris-HCl, 50 mM KCl, 3.0 mM MgCl2, 0.2 mM dNTP, stabilizers) and 1× reference Dye. The RT-PCR conditions for the one-step SYBR green RT-PCR consist of a 20 minutes reverse transcription step at 44°C and then 2 minutes of Taq polymerase activation at 94°C, followed by 40 cycles of PCR at 94°C for 15 seconds (denaturation), 60°C for 1 minute (annealing and extension).
Cytochalasin B (Sigma) was used to reduce actin polymerisation rate by inhibiting the addition of actin monomers to the “barbed” end of microfilaments [25]. Nocodazole (Sigma) was used to inhibit the assembly of tubulin into microtubules [26]. Blebbistatin (Abcam) was used to inhibit myosin II by binding to myosin-ADP-Pi complex and intefering with phosphate release [27]. Tunicamycin (MP Biomedicals) was used to inhibit N-linked glycosylation of proteins [28].
The cell viability assay was first performed to assess the cytotoxicity of the various concentrations of cytochalasin B (0.1 to 4 µM), nocodazole (0.1 to 0.4 µM), blebbistatin (1 to 50 µM) and tunicamycin (0.2 to 4 µg/ml) by incubating Vero C1008 cells with the drugs in DMEM medium supplemented with 2% FCS for 24 hours. 0.1% DMSO was used as a non-treated control. After 24 hours incubation, one-tenth volume of alamarBlue reagent (Invitrogen) was added to the drug-treated cells and incubated for 2 hours at 37°C. Fluorescence measurements were performed at an excitation wavelength of 570 nm and an emission wavelength of 585 nm using a microplate reader (Infinite 200, Tecan). The percentage of cell viability was determined by comparing with the non-treated control.
Upon checking for cytotoxic effects, apically-infected Vero C1008 cells were fed with DMEM medium containing various concentrations of the drugs after the 1.5 hours virus adsorption period. Supernatants were collected after 24 hours for viral plaque assays to quantify the virus titer. The cells were analysed by immunofluorescence assay.
Two-tailed student t-test was used to analyse the difference in total infectious virus titer post-apical infection and that post-basolateral infection, as well as the difference in infectious virus titer in the apical chamber and that in the basolateral supernatant.
The susceptibility of Vero cells to CHIKV infection has been shown previously [29], [30]. However, it is not known if HBMEC and Vero C1008 cells are susceptible to CHIKV infection. As such, growth kinetics studies were performed to determine the susceptibility of HBMEC and Vero C1008 cells to CHIKV infection and to select appropriate time points to study the polarized infection of CHIKV.
For CHIKV-infected HBMEC cells, the cell density upon CHIKV infection was lower than that of mock-infected cells (Figure 1A). Moderate amounts of CHIKV protein expression were detected at 24 h.p.i.. However, the amount of CHIKV antigen detected remained low up to 120 h.p.i. (Figure 1B). The infectious virus titer gradually increased with time post-infection, with a maximum titer of 3.4×105 PFU/ml observed at 36 h.p.i. (Figure 1C). For CHIKV-infected Vero C1008 cells, extensive cytopathic effects (CPE) were observed from 36 h.p.i. onwards. The CHIKV-infected Vero C1008 cells were rounded and observed to be lifting off from the cell monolayer (Figure 2A). CHIKV antigen was detected as early as 12 h.p.i. and increased drastically at 24 and 36 h.p.i. (Figure 2B). The infectious virus titer also increased with time post-infection, with a maximum titer of 5.0×108 PFU/ml observed at 42 h.p.i. (Figure 2C).
The growth kinetics studies confirmed that both HBMEC and Vero C1008 are susceptible to CHIKV infection, as shown by the increasing infectious virus titer with time and detection of CHIKV antigen by immunofluorescence assay. Additionally, we selected 24 h.p.i. as an appropriate time point for subsequent studies on the polarized infection of CHIKV in HBMEC and Vero C1008 cells for several reasons. Firstly, maximum amount of viral antigens were detected at 24 h.p.i. for HBMEC and at 24 and 36 h.p.i. for Vero C1008 cells via immunofluorescence assays. However, CPE was observed at 36 h.p.i. of Vero C1008, which will affect the integrity of the cell monolayer during the subsequent polarized infection studies. In comparison, the HBMEC and Vero C1008 cells remained well spread out in a monolayer at 24 h.p.i.. Thirdly, the infectious virus titers were observed to be increasing at 24 h.p.i.. From the above descriptions, 24 h.p.i. was selected as an appropriate time point for subsequent studies on the polarized infection of HBMEC and Vero C1008.
Non-polarized Vero, polarized Vero C1008 and polarized HBMEC cells were seeded on cell culture inserts and infected with CHIKV either apically or basolaterally. The cell monolayers' integrity post-infection was determined by measuring the cell monolayers' trans-epithelial electrical resistance (TEER) and detecting ZO-1 protein expression via immunofluorescence assays. Upon apical and basolateral infection, the TEER of both Vero and Vero C1008 cell monolayers were within 40 and 70 Ω/cm2, which were comparable to that of mock-infected Vero and Vero C1008 cells (Figure 3A), demonstrating that the cell monolayers remained non-permeable post-infection to prevent exchange of materials between the apical and basolateral chambers. Thus, the infectious virus titer of the supernatant harvested from the apical chamber represents the amount of CHIKV released from the cells' apical domain, while the infectious virus titer of the supernatant harvested from the basolateral chamber represents the amount of CHIKV released from the cells' basolateral domain. Although the TEER of HBMEC cell monolayers were only between 20 and 30 Ω/cm2 post-apical and basolateral infection (Figure 3A), the TEER were comparable to that of mock-infected HBMEC cells. ZO-1 tight junction proteins were expressed in Vero, Vero C1008 and HBMEC cells at 24 h.p.i., and the expression of ZO-1 in infected cells was comparable to that in mock-infected cells (Figure 3B). The permeability of the Vero and Vero C1008 cell monolayers were also assayed by measuring the amount of FITC-dextran flow-through from the apical to the basolateral supernatant. The percentage of FITC-dextran flow-through was low for apically, basolaterally and mock-infected Vero and Vero C1008 cells as compared to the positive control cells treated with TNF (Figure 3C), further demonstrating that the integrity of the cell monolayers remained intact post-infection.
To investigate whether CHIKV enters Vero C1008 and HBMEC in a polarized manner, the total infectious virus titer at 24 hours post-apical infection was quantified by plaque assays and compared to the total infectious virus titer post-basolateral infection. Infection of non-polarized Vero cells was performed as a control. As expected, the total virus titer was similar between post-apical and post-basolateral infection of Vero cells. In contrast, the total virus titer post-apical infection of polarized HBMEC cells was 1.0 log units higher than that post-basolateral infection. Similarly, apical infection of Vero C1008 also resulted in a total virus yield of 1.2 log units higher than basolateral infection (Figure 4A). These data suggest that entry of CHIKV into Vero C1008 and HBMEC cells is polarized towards the apical domain. Furthermore, virus binding assays showed that the amount of CHIKV binding was higher upon apical inoculation of CHIKV as compared to basolateral inoculation of CHIKV in polarized Vero C1008 cells (Figure 4B). In contrast, the amount of CHIKV particles binding to non-polarized Vero cells were similar upon apical and basolateral inoculations. Therefore, these data further confirm the preferential entry of CHIKV at the apical domain of polarized Vero C1008 cells.
Next, to investigate whether the release of CHIKV from Vero C1008 and HBMEC is in a polarized manner, infectious virus titer released into the apical chamber (AiAc) was compared to the infectious virus titer released into the basolateral chamber (AiBc) post-apical infection, while the infectious virus titer in the apical chamber (BiAc) was compared to the infectious virus titer in the basolateral chamber (BiBc) post-basolateral infection. When non-polarized Vero cells were apically- or basolaterally-infected with CHIKV, the infectious virus titer in the apical and basolateral chambers were similar (Figure 5A). In contrast, at 24 hours post-apical and basolateral infection of polarized Vero C1008 cells, the infectious virus titer was 1.3 and 0.3 log units higher in the apical chamber than in the basolateral chamber, respectively. Similarly, at 24 hours post-apical and basolateral infection of polarized HBMEC cells, the infectious virus titer was 3.3 and 1.5 log units higher in the apical chamber than in the basolateral chamber, respectively (Figure 5A). These data suggest the polarized release of CHIKV towards the apical domain of Vero C1008 and HBMEC cells.
To further illustrate the polarized release of CHIKV, immunofluorescence assays were performed on cells post-polarized infection with CHIKV using antibodies against the CHIKV E2 viral protein to visually inspect the membrane domain at which CHIKV is released from. Immunofluorescence assays were only performed on non-polarized Vero and polarized Vero C1008 cells as the growth kinetics studies showed that high CHIKV viral antigen expression could be detected in CHIKV-infected Vero C1008 cells (Figure 2B) but only low amounts of CHIKV antigens could be detected in CHIKV-infected HBMEC cells (Figure 1B). CHIKV-infected Vero and Vero C1008 cells were co-labeled with antibodies against CHIKV E2 viral protein (arrows, Figure 5B–E) and ZO-1 proteins (arrowheads, Figure 5B–E). ZO-1 proteins are markers for tight junctions between adjacent cells, as well as an apical marker to discriminate between the apical and basolateral domains [31]. The Z-section images demonstrated that upon both apical infection (Figure 5B) and basolateral infection (Figure 5C) of polarized Vero C1008 cells, CHIKV was preferentially released from the apical domain, where the CHIKV particles localized on the same membrane domain as ZO-1. In comparison, upon apical infection (Figure 5D) and basolateral infection (Figure 5E) of non-polarized Vero cells, CHIKV was released bi-directionally from both the apical and basolateral domains of Vero cells. Thus, the results further demonstrated the polarized release of CHIKV from the apical domain of polarized Vero C1008 cells. Of note, the ZO-1 staining depicted in red in Figure 5D appears to be on the basolateral side of the Vero cells and not on the apical side as would be expected. This is because Vero cells are non-polarised, hence the ZO-1 proteins could be located on either sides of the cell. In addition, the shape of the cell might just be that they tapered off at the sides, making the ZO-1 staining seem like it is at the basolateral side.
The host cell cytoskeleton network and motor proteins have been shown to be play important roles in the polarized sorting of host cellular proteins and viral proteins [32], [33]. As such, we aimed to determine the involvement of host cell transport machineries in the polarized sorting of CHIKV towards the apical membrane for release. Cytochalasin B and nocodazole were used to inhibit actin polymerisation into microfilaments and tubulin polymerisation into microtubules, respectively. CHIKV-infected Vero C1008 cells were also treated with blebbistatin, a small molecule inhibitor of myosin II, to determine whether myosin II is involved in the apical sorting of CHIKV.
The alamarBlue assay was first performed to ensure that the drug concentrations used do not cause cytotoxic effects in Vero C1008. Results showed that the drug concentrations for cytochalasin B, nocodazole and blebbistatin did not result in cytotoxic effects (Figure S1). However, cell rounding was observed when 0.4 µM of nocodazole was added, which will disrupt the cell monolayer's integrity, allowing the exchange of materials between the apical and basolateral chambers. Hence, 0.1 to 0.3 µM nocodazole was used for the subsequent post-treatment studies.
Upon treating apically-infected Vero C1008 cells with cytochalasin B (Figure 6A), nocodazole (Figure 6B) and blebbistatin (Figure 6C), the infectious virus titers remained higher in the apical chamber than in the basolateral chamber. Additionally, confocal microscopy demonstrated that CHIKV was preferentially released from the apical domain of polarized Vero C1008 cells upon inhibition of actin polymerisation into microfilaments (Figure 6D), tubulin polymerisation into microtubules (Figure 6E), and myosin II (Figure 6F). These data suggest microfilaments, microtubules and myosin II are unlikely to be involved in the apical sorting of CHIKV in Vero C1008 cells, or may serve redundant functions in the apical sorting of CHIKV.
Besides the host cell factors, several apical sorting signals on apically-sorted proteins have been described to date, including glycosylphosphatidylinositol (GPI) membrane anchors, N-linked glycoproteins (N-glycans), NPXY motifs and YXXØ motifs [34]–[37]. Notably, the CHIKV envelope glycoproteins are N-glycosylated on asparagine residues N12 of E3 protein, N141 of E1 protein, and N263 of E2 protein [38]. Hence, to investigate viral factors involved in the sorting of CHIKV towards the apical domain of the host cells, CHIKV-infected Vero C1008 cells were treated with tunicamycin to inhibit N-glycosylation of the CHIKV envelope glycoproteins. The alamarBlue assay demonstrated that the concentrations for tunicamycin did not result in cytotoxic effects (Figure S1).
Interestingly, upon apical infection and tunicamycin treatment, similar infectious virus titers were obtained from the apical and basolateral chambers (Figure 7A). Additionally, confocal microscopy demonstrated that CHIKV was released from both apical and basolateral domains of polarized Vero C1008 cells in a bi-directional manner upon inhibition of N-glycosylation with tunicamycin (Figure 7B). qPCR assays were also performed to quantify the CHIKV RNA in the apical and basolateral chambers upon tunicamycin treatment as an alternative read-out for the amount of CHIKV particle released into the apical and basolateral chambers. Indeed, the qPCR results also demonstrated a non-polarized release of CHIKV upon tunicamycin treatment, where similar amounts of CHIKV RNA were detected in supernatants from both chambers (Figure 7C). These data suggest that the N-glycans of CHIKV glycoproteins may be involved in the polarized apical sorting of CHIKV.
Several polarized cell lines have been used in the study of the polarized infection of viruses, including polarized African Green Monkey kidney epithelial cells (Vero C1008), polarized human intestinal epithelial cells (Caco-2) and polarized Madin-Darby canine kidney epithelial cells (MDCK) [17]–[20], [39], [40]. However, there have yet been any studies on the polarized infection of CHIKV. In this study, the polarized infection of CHIKV was examined in HBMEC and Vero C1008 cells. In both HBMEC and Vero C1008 cells, the entry and release of CHIKV was polarized towards the apical domain of the cells. However, the growth kinetics studies of CHIKV on HBMEC and Vero C1008 (Figures 1–2) demonstrated that Vero C1008 is more permissive to CHIKV infection than HBMEC, as shown by the higher increase in infectious virus titer post-infection and higher amount of CHIKV protein expression. In agreement with our results, Sourisseau and coworkers also reported a low infectivity of human brain microvascular endothelial cell line hCMEC/D3, whereby only 1% of the cell population were infected with CHIKV [29]. As such, we established Vero C1008 as a suitable and more convenient in vitro cell model for subsequent studies on the polarized infection of CHIKV and elucidation of mechanisms and sorting signals involved in the polarized sorting of CHIKV towards the apical domain.
Epithelial and endothelial surfaces of the human body are a key component of the innate immune system because they act as a mechanical barrier against infection by pathogens. Polarized lung epithelial cells lining the respiratory tract are exposed to airborne pathogens. Polarized intestinal epithelial cells lining the small intestine are exposed to pathogens in the gastrointestinal tract. Polarized vascular endothelial cells lining the blood vessels are exposed to blood-borne pathogens. As such, these epithelial and endothelial surfaces are often the first point of contact between the human host and pathogens, and have to be breached by the pathogens in order to establish infection.
The polarized infection of viruses has been of research interest because it provides insights to the pathogenesis of the viruses. For example, infection with Simian Virus 40 (SV40) is characterized by the persistent infection of the rhesus monkey kidney, the presence of SV40 in the urine, as well as the absence of SV40 in the bloodstream [41]. This could be explained by the polarized entry and release of SV40 towards the apical domain of kidney tubular epithelial cells [17], thus restricting the infection of SV40 at the kidney. Additionally, the apical entry of Hepatitis A Virus (HAV) in polarized Caco-2 cells suggests that the intestinal epithelial cells can be infected by HAV present within the lumen of the gastrointestinal tract. Furthermore, the apical egress of HAV from Caco-2 cells may provide an explanation for the high shedding amount of HAV in the feces of HAV-infected patients. The apical egress of HAV from Caco-2 cells also suggests that epithelial cell infection is unlikely to result in penetration of HAV beyond the gastrointestinal epithelium. Thus, invasion of the liver by HAV may be dependent upon alternative mechanisms, such as transcytosis by specialized M cells in the distal ileum [21]. Furthermore, the entry of H1N1 and H5N1 Influenza viruses occurs bi-directionally in polarized alveolar epithelial cells, but releases predominantly at the apical domain facing the airways. The apical release of Influenza virus has been suggested to improve the transmissibility of Influenza virus by carrying the virus in respiratory droplets.
This study is the first to demonstrate the polarized entry and release of CHIKV towards the apical domain of polarized cells. The polarized entry of CHIKV at the apical domain of HBMEC and Vero C1008 cells suggests that the CHIKV receptors could be pre-dominantly sorted to the apical domain, allowing CHIKV to attach to the receptors and enter the polarized cells preferentially at the apical plasma membrane. Furthermore, the polarized release of CHIKV at the apical plasma membrane of HBMEC and Vero C1008 cells suggest that CHIKV structural proteins may contain apical sorting signals that direct their sorting towards the apical domain.
The polarized entry and release of CHIKV towards the apical plasma membrane of HBMEC may implicate that it is unlikely for CHIKV to gain access into the brain to cause neurological complications by apical entry from the blood into the brain microvascular endothelial cells and basolateral egress from the endothelial cells into the brain tissues. As such, other mechanisms could be explored to understand how CHIKV enters the central nervous system. For example, the entry and release of West Nile Virus (WNV) occurs at the apical domain of polarized cells, which is similar to CHIKV infection, and infections with WNV have been associated with neurological complications too. WNV causes neurological complications via the release of tumor necrosis factors, which transiently increases the permeability of the blood-brain barrier, thereby allowing WNV to diffuse across the blood-brain barrier from the bloodstream into the brain [42]. In another example, Venezuelan equine encephalitis virus (VEEV), a member of the Alphavirus genus, is also associated with encephalitis. VEEV infects the olfactory sensory neurons and spreads by retrograde neuronal dissemination into the brain to initiate viral replication in the brain initially. Subsequently, VEEV induces the biphasic opening of the blood-brain barrier and allows a second wave of VEEV from the periphery to enter the brain [43]. In addition, Couderc and coworkers [44] demonstrated that primary choroid plexus epithelial cells were highly susceptible to CHIKV infection in a polarized manner similar with what is demonstrated in our findings, with preferential entry via the apical domain. Thus, an alternative mechanism by which CHIKV infects the brain is through the cerebrospinal fluid produced through the choroid plexuses.
Nevertheless, the results of the polarized studies of CHIKV highlight that CHIKV infection could attain high viremia by apical entry into vascular endothelial cells lining the blood vessels, multiplying within the endothelial cells to high quantities, and releasing the newly synthesized CHIKV progenies back into the bloodstream at the apical domain. The high viremia in the bloodstream may aid the transmission of CHIKV as the mosquito vectors bite the CHIKV-infected patients and transfer the viruses to other human hosts during their next blood meal.
The cytoskeleton network and motor proteins in host cells have been shown to be involved in the polarized sorting of host cellular proteins and viral proteins [32], [33]. For example, the apical sorting of Measles Virus matrix protein is dependent on microfilaments [45] while apical sorting of West Nile Virus envelope proteins is dependent on microtubules [18]. In addition, myosin II motor proteins facilitate the apical sorting of Bile Salt Export Protein (BSEP), so that BSEP can perform its transporter function to secrete bile acids at the apical membrane domain [46]. However, this study demonstrated that individual inhibition of microfilament formation, microtubules formation and myosin II functions do not disrupt the apical sorting of CHIKV. This suggests that the microfilament and microtubule cytoskeleton networks and motor proteins may perform redundant roles in the apical sorting of CHIKV, such that the individual inhibition of one of the cytoskeleton or motor protein factors do not disrupt apical sorting of CHIKV. Furthermore, the apical sorting of CHIKV may involve other host cell mechanisms such as Rab proteins. For example, Rab4 proteins are involved in the redistribution of the transferrin receptor from the basolateral plasma membrane to the apical plasma membrane during the indirect sorting of transferrin receptor to the apical domain [47]. On the other hand, Rab 11 and Rab 14 proteins are involved in the apical sorting of the ribonucleoprotein and hemagglutinin of Influenza virus [48], [49]. Thus, the host cell factors involved in the apical sorting of CHIKV in polarized cells remain to be studied.
N-glycans have been widely described as sorting signals that direct the apical sorting of proteins [36], [50], such as the Bile Salt Export Protein [28]. Interestingly, inhibition of N-glycosylation by the addition of tunicamycin resulted in a bi-directional release of CHIKV from polarized Vero C1008 cells. These data suggest that N-glycans on the E1, E2 and E3 envelope glycoproteins of CHIKV could serve as apical sorting signals that direct the trafficking of CHIKV preferentially towards the apical domain. In comparison, previous studies have shown that release of Vesicular Stomatitis Virus (VSV) and Influenza virus remained polarized despite inhibition of glycosylation using tunicamycin [51]. Thus, the polarized sorting of each virus may depend on different apical sorting signals.
The mechanism(s) by which N-glycans mediate the apical sorting of glycoproteins is not well-understood currently. Nevertheless, vesicular integral-membrane protein 36 (VIP36), a type of mannose-binding lectin, has been proposed to be an apical sorting receptor that binds high-mannose-type N-glycans and facilitate apical sorting. VIP36 is predominantly located at the endoplasmic reticulum-Golgi intermediate compartment (ERGIC). The apical membrane content of VIP36 is twice as high as the basolateral content in MDCK cells, and the plasma membrane glycoproteins recognized by VIP36 are also twofold higher in the apical membrane compared to the basolateral membrane [52]. Thus, binding of VIP36 to high-mannose-type N-glycans may facilitate the sorting of glycoproteins carrying the high-mannose-type N-glycans at the endoplasmic reticulum into vesicles destined for the apical plasma membrane.
The drug treatment assays also showed that tunicamycin resulted in a reduction of total infectious CHIKV released. A similar reduction in VSV and Influenza infectious virus titer was also observed when the virus-infected cells were treated with tunicamycin [51], although the release of VSV and Influenza virus remained polarized towards the apical domain. This reduction in infectious virus titer could be due to the inhibition of N-glycosylation of the viral envelope glycoproteins, which might affect their ability to bind to the virus receptors on host cells during infection. Thus, tunicamycin might have inhibited the N-glycosylation of the envelope glycoproteins and interfered with the infectivity of the newly synthesized enveloped virus progenies, resulting in the production of lower infectious virus titer. As such, further studies is in progress to elucidate the role of N-glycans in the apical sorting of CHIKV in a polarized cell model.
|
10.1371/journal.pntd.0004464 | Vaccination with Replication Deficient Adenovectors Encoding YF-17D Antigens Induces Long-Lasting Protection from Severe Yellow Fever Virus Infection in Mice | The live attenuated yellow fever vaccine (YF-17D) has been successfully used for more than 70 years. It is generally considered a safe vaccine, however, recent reports of serious adverse events following vaccination have raised concerns and led to suggestions that even safer YF vaccines should be developed. Replication deficient adenoviruses (Ad) have been widely evaluated as recombinant vectors, particularly in the context of prophylactic vaccination against viral infections in which induction of CD8+ T-cell mediated immunity is crucial, but potent antibody responses may also be elicited using these vectors. In this study, we present two adenobased vectors targeting non-structural and structural YF antigens and characterize their immunological properties. We report that a single immunization with an Ad-vector encoding the non-structural protein 3 from YF-17D could elicit a strong CD8+ T-cell response, which afforded a high degree of protection from subsequent intracranial challenge of vaccinated mice. However, full protection was only observed using a vector encoding the structural proteins from YF-17D. This vector elicited virus-specific CD8+ T cells as well as neutralizing antibodies, and both components were shown to be important for protection thus mimicking the situation recently uncovered in YF-17D vaccinated mice. Considering that Ad-vectors are very safe, easy to produce and highly immunogenic in humans, our data indicate that a replication deficient adenovector-based YF vaccine may represent a safe and efficient alternative to the classical live attenuated YF vaccine and should be further tested.
| Live attenuated yellow fever vaccine (YF-17D) is an efficient and generally safe vaccine. Nevertheless, in recent years the reporting of serious adverse effects together with the given limitations in the use of this live vaccine in certain risk groups has spurred an interest in developing a more generally applicable and safer alternative. Using an adenovector platform and recombinant vaccines targeting both structural and non-structural YF antigens, we now demonstrate that non-replicating adenobased vaccines may be used to induce a state of host immunity, which like YF-17D vaccination encompasses both major arms of the adaptive immune system. Furthermore, in a murine challenge model, adenovector induced protection fully matched that induced by the current vaccine. Taken together our results strongly suggest that adenovectored vaccines targeting structural and non-structural viral antigens represent a viable and safe alternative to the existing live, attenuated YF vaccine.
| The design of vaccines against viral infections has evolved considerably with the advances in molecular biology, which have created many alternative approaches to the empirical development of live vaccines. Thus, the first generation of live attenuated vaccines and the second generation of subunit vaccines have now been followed by a third generation of vaccines based on recombinant DNA technology. The newly designed vaccines have several advantages compared to empiric attenuated live vaccines: their production is faster, cheaper and easier to control, and, importantly, their safety profile is considerably better than that of live viruses making them more appealing for use in humans. However, they have rarely shown the same immunogenicity as their live predecessors, and the biological mechanisms behind this difference have been the subject of extensive research.
The yellow fever (YF) vaccine, based on the live attenuated YF-17D virus, was developed in the 1930s by serial tissue culture passage of wild type YF virus (YFV) in mouse and chicken cell cultures [1–3]. A single vaccination with YF-17D can confer protection in more than 95% of the vaccinees, and immunity has been shown to last up to 40 years post vaccination and to correlate with presence of neutralizing Abs [4,5]. In spite of the clear success in preventing infection with YFV in many areas of the world, the YF-17D vaccine also has its dark side; rare, but often fatal vaccine-associated adverse events (SAEs) may be induced [5]. These SAEs mainly fall into two categories: vaccine-associated neurotropic disease (YEL-AND), which consists in a post-vaccinal encephalitis [5,6], and vaccine-associated viscerotropic disease (YEL-AVD), which is a pansystemic infection characterized by liver damage, similarly to infection with wild type YFV [7–9]. Interestingly, sequence analysis of the few isolates obtained from patients in whom adverse events following vaccination were fatal, demonstrated that the virus had not reverted to virulence, rather host genetic factors appeared to be responsible for the severe reaction to YF-17D virus [5,10]. Moreover, due to its live viral nature, the YF vaccine is contraindicated in pregnant women, infants, elderly, immunosuppressed and certain HIV infected individuals as well as in people with hypersensitivity to eggs in which the vaccine is still manufactured [5]. In this perspective, implementation of alternative vaccine strategies such as DNA-based vaccines has become desirable.
Recombinant DNA vaccines in which the antigen is encoded by an attenuated viral vector have demonstrated great potential, and very recently it has been found that a DNA vaccine encoding the envelope antigen of YFV may induce protection in murine studies [11]. However, the immunogenecity of naked DNA vaccines is substantially surpassed by that of replication deficient adenoviral vectors, which have been found to represent very efficient immunogens also in primates [12,13]. Thus, vaccination with recombinant adenovectors has been shown to elicit strong cellular immunity against antigens from several different viral pathogens such as Ebola virus, lymphocytic choriomeningitis virus, HIV, hepatitis C virus [14–19], as well as non-viral pathogens such as Listeria monocytogenes [20] and malaria [21]. In relation to YFV infection, recent results from our group strongly suggest that long-term clinical protection is mediated not only by neutralizing antibodies as previously believed [4], but also by virus-specific CD8+ T cells [22]. Therefore, with the dual purpose of using adenoviral vectors for further studies of immunity to YFV infection in our intracranial challenge mouse model and to explore the possibility of developing an adenobased vaccine alternative, we generated two vector constructs: an Ad serotype 5 (Ad5) vaccine vector encoding the YF-17D structural proteins core (C), membrane (M) and envelope (E), as well as an Ad5 vector encoding the viral non-structural protein 3 (NS3). The rationale behind this strategy was that the former vector would likely induce a humoral response towards the encoded YF-17D surface antigens, while the latter vector would induce cell-mediated immunity toward antigens present within YFV infected cells. Our results showed that vaccination with Ad-YF C,M,E or Ad-YF NS3 both resulted in long-standing protection of mice from severe YFV infection. While virus-specific CD8+ T cells appear to represent the major effectors when the viral NS3 protein was chosen as antigen, a combination of T-cell and B-cell immunity determined the outcome in mice vaccinated with the vector encoding the three viral structural proteins. Thus, besides strengthening the evidence for a role of CD8+ T cells in protection against severe YFV infection, our findings point to adenoviral vectors as potential vaccine candidates that may be used to develop a safer alternative for immunization against YFV infection.
Female C57BL/6 (wt B6) mice, B-cell (μMT/μMT) deficient mice, β2-microglobulin (β2m-/-) and MHC class II (Aβ-/-) deficient mice, as well as mice deficient in H-2Kb or in both H-2Kb/H-2Db molecules, were all obtained from Taconic Farms. Perforin (Prf) deficient and IFN-γ deficient mice were originally obtained from The Jackson Laboratory and IFN-γ/perforin double-deficient (IFN-γ/Prf) mice were generated locally through intercrossing of these strains, as previously described [23]. All mice used in this study were 7–10 weeks old and housed in a pathogen–free facility. All experiments were approved by the national animal ethics committee and performed in accordance to the national guidelines.
Experiments were conducted in accordance with national Danish guidelines (Amendment # 1306 of November 23, 2007) regarding animal experiments as approved by the Danish Animal Experiments Inspectorate, Ministry of Justice, permission number 2009/561-1679. The mice were housed in accordance with good animal practice as defined by FELASA.
pacCMV shuttle vectors encoding the structural proteins core, membrane and envelope (C, M, E), and the nonstructural protein 3 (NS3) from YF-17D virus (GenBank: X03700.1) were obtained from GenScript (Piscataway, NJ, USA). pacCMV vectors encoding the truncated form of NS3 (aa. 201–325) were also obtained from Gen Script (Piscataway, NJ, USA).
From all shuttle plasmids, human type 5 recombinant adenovirus (Ad5) vectors were produced through homologous recombination by standard methods [24]. After purification on a cesium chloride (CsCl) density gradient, adenoviral stocks were immediately frozen as single use aliquots in 10% glycerol at -80°C, and the infectivity of the stocks was determined using the Adeno-X Rapid Titer kit (BD Clontech). The presence of the inserted transgene was validated in all the recombinant Ad5 vectors by sequence analysis prior to their use for vaccination in mice. The genome of YF-17D virus with the position of the two known class I-restricted epitopes in B6 (H-2b) mice and the derived Ad-YF vectors, are depicted in Fig 1.
YF-17D virus (Stamaril, Sanofi Pasteur; reconstituted as recommended by the manufacturer) was propagated in Vero (ATCC CCL-81) cells grown in DMEM containing 10% FCS, glutamine and antibiotics (penicillin and streptomycin). Cell monolayers were seeded 24 hours earlier and infected with at a multiplicity of infection (MOI) of 0.001/cell in DMEM 2% FCS; infectious supernatants were harvested when cytopathic effect was more than 60% (day 6 p.i), freeze-thawed and clarified of cell debris at 2000 rpm for 15 min at 4C°. Viral stocks were stored as single use aliquots at -80C°. For determining virus infectious titers, we used an Immuno Focus assay (IFA) recently developed in our lab [22]. In brief, virus stocks were serially diluted (ten-fold) in DMEM and adsorbed for 1h at 37 C° onto Vero (ATCC CCL-81) cell monolayers in twenty four-well plates; cells were then overlaid with media containing 0.9% methyl cellulose and incubated at 37 C° for 3 days. Following fixation and permeabilization, a monoclonal antibody directed against NS3 from YF-17D virus was used for detection of virus antigens within infected cells; the foci of infection (analogous to plaques) were visualized using HRP-substrate reaction and counted. When virus titers in the brains were determined, the IFA was performed on serial 10-fold dilutions of homogenized and clarified 10% organ suspensions.
Neutralizing antibodies to YF-17D virus in sera from vaccinated mice were measured as described previously [22]. Briefly, appropriately diluted YF-17D virus (resulting in about 70–100 pfu per well) was incubated for 1 hour at 37°C with serial (2-fold) dilutions of mouse sera obtained from individual mice vaccinated with Ad-YF C,M,E or Ad-YF NS3. The virus-serum mixture was subsequently added to Vero cells monolayers in 24-well plates and incubated for 2 hours at 37°C, after which the overlay media containing 0.9% methyl cellulose was added. After 3 days of incubation at 37°C, the overlay media was removed and, following fixation and permeabilization, cells were stained as described above (IFA). Following counting, the neutralizing antibody titer, defined as the highest serum dilution neutralizing more than 50% of the viral plaques, was determined.
For adenovirus vaccination, 2x107 pfu of recombinant Ad-5 in a volume of 30 μl was given subcutaneously to isofluorane-anesthetized mice in the right hind foot pad; vaccination with YF-17D virus at the dose of 105 pfu in 300 μl was given subcutaneously (s.c.) at the base of the tail.
In YFV challenge studies, deeply anesthetized mice were injected intracranial with 104 pfu of YF-17D virus in a volume of 30 μl; when assessing clinical protection, mice were checked twice daily for two weeks following intracranial challenge and euthanized when development of illness (drop in body weight, ruffled fur, twitching) together with neurologic signs (hind leg paralysis and weakness) were observed. Mice developing sickness within 4 days post challenge were excluded from experiments.
A combination of two monoclonal antibodies (YTS 169 and YTS 156)[25,26] was used for in vivo depletion of CD8+ T cells from vaccinated mice prior to challenge. Mice to be depleted were injected i.p. with 100 μg of antibody one day prior to i.c challenge and at days 1 and 4 post challenge. The efficiency of the depletion was confirmed by flow cytometric analysis at day 7 post challenge.
Single cell suspensions of splenocytes were obtained by pressing the organs through a fine steel mesh (mesh size, 70 μm), followed by centrifugation and re-suspension in RPMI 1640 cell culture medium. The frequencies of antigen-specific CD8+ T cells in the spleen were determined by intracellular cytokine staining (ICS) performed after 5 h of incubation with relevant peptides (0.1 μg/ml of NS3 (268–275) or E (4–12)) in the presence of monensin (3 μM) at 37°C in 5% CO2. After incubation, the cells were stained with Abs for cell surface markers (peridinin chlorophyll protein-Cy5.5- or BrilliantViolet 421-CD8, APC-Cy7-CD44) and Ab for intracellular cytokine (APC- interferon gamma [IFN-γ]). Samples were run on an LSRII flow cytometer (BD biosciences) and analyzed using FlowJo software (TreeStar).
A nonparametric Mann-Whitney U test was used to compare quantitative data; survival after viral challenge was analyzed using the Log Rank test (*p < 0.05; **p < 0.01; (***p < 0.001). GraphPad Prism (version 6) software was used for statistical analysis.
As a first approach to evaluate our YF vaccine candidates, we analysed the immunogenicity of our structurally and non-structurally targeted vectors (see Fig 1) in terms of their capacity to induce effector CD8+ T cells and virus-specific antibodies in vaccinated mice. From previous studies it is known that both vectors contained T-cell epitopes of relevance in H-2b mice: a dominant H-2Kb-restricted epitope has been mapped to the NS3 protein, and an H-2Db-restricted subdominant epitope can be found in the viral E protein [27]. Consequently, to evaluate the CD8+ T-cell responses elicited in vaccinated mice, wt B6 mice were inoculated in the f.p. with 2x107 pfu of either of the two prototypic Ad5 vaccines, and the kinetics of the primary CD8+ T-cell responses was followed by flow cytometric analysis of ICS; a group of mice vaccinated with 105 pfu of YF-17D virus s.c. was included for comparison. As can be seen in Fig 2A, both constructs induced strong CD8+ T-cell responses against the YFV epitope encoded by the vector in question. Furthermore, it should be noted that the magnitude of T-cell responses generated in adenovector vaccinated mice markedly surpassed those elicited by the YFV vaccine; whether this is also the case in humans we do not know, but certainly adenobased vaccines are highly immunogenic also in primates [12,13].
Besides CD8+ T-cell responses, we also measured the antibody responses induced in vaccinated mice by an in vitro neutralization assay. As can be seen in Fig 2B, the Ad-YF C,M,E vector expressing the surface antigens of the YFV induced a clear virus-neutralizing response, albeit not quite at the level found subsequent to YF vaccination [22]. In contrast, mice vaccinated with the vector encoding NS3 did not generate any antibodies detectable in the neutralization assay. From these results we conclude that Ad-YF C,M,E induces both a humoral and a cell-mediated immune response of potential relevance to in vivo protection, while Ad-YF NS3 induces only a relevant cell-mediated response.
Having determined the type and magnitude of immune responses elicited by each of our adenoviral vectors, we proceeded to test the protective efficacy of a single immunization with either of the two Ad-YF vectors. Mice were inoculated with 2x107 pfu of Ad-YF C,M,E, Ad-YF NS3 or PBS in the f.p., and 14 days later, all animals were intracranial challenged with 104 pfu of YF-17D virus (Fig 3). The intracranial challenge model we adopted has previously been shown to induce a uniformly fatal neurotropic disease in naïve B6 mice, while YFV vaccinated mice are protected from disease [22]. We found that vaccination with Ad-YF C,M,E resulted in 100% protection from viral challenge, while all sham-vaccinated (PBS) mice succumbed to the infection (Fig 2). Interestingly, vaccination with Ad-YF NS3 resulted in 80% protection from subsequent YF challenge. Since this vector only encodes a non-structural viral protein, which is found exclusively in infected host cells, and since we have previously demonstrated that NS3 specific CD8+ T cells infiltrate the YF-infected brain [22], the latter observation strongly supported the idea that protection in this animal model of YFV infection may be accomplished by T cells alone.
To confirm the hypothesis that T-cell immunity suffices for protection, we first evaluated the outcome of YFV challenge in several mouse knock-out (KO) strains carrying targeted deficiencies in CD8+ T-cell responses or in their effector molecules. Thus, wt B6 mice, β2m KO mice, H-2kb KO mice and IFN-γ/Prf double KO mice were vaccinated with Ad-YF NS3; 14 days later, the mice were intracranial challenged with 104 pfu of YF-17D virus, and the clinical outcome was monitored. Additionally, MHC class II KO mice were included in the experiment to examine the role of CD4+ T cells in the Ad-YF NS3 induced response.
As can be seen in Fig 4A, we found that NS3 induced protection was completely abrogated in β2m KO mice (no survival), and profoundly impaired in H-2Kb KO mice, in which case only 10% of vaccinated animals survived the challenge, whereas the survival of wt B6 mice in this experiment was 100%. Given that the H-2Kb molecule is the restriction element for the dominant NS3(268–75) epitope [27], this result strongly support the idea that the CD8+ T cells specific for this epitope may be accountable for the vector-induced immunity; however, formally we cannot rule out the contribution of additional H-2Kb-restricted epitopes potentially present in the NS3 protein. Consistent with a dominant role for effector T cells, the combined deficiency of IFN-γ and perforin also resulted in complete abrogation of protection, with no survivors recorded among vaccinated IFN-γ/Prf double KO mice (Fig 4A). Thus, we concluded that the Ad-YF NS3 vector induces immunity to YFV via IFN-γ and/or perforin producing CD8+ T cells specific for viral H-2kb-restricted epitopes.
Notably, we also found that the absence of CD4+ T cells resulted in a total suppression of protective immunity (no survival in MHC class II KO mice) (Fig 4A). This cannot reflect lack of help to B cells, since the NS3 vaccine does not induce protective antibodies. However, a requirement for CD4+ T cell help for an effective CD8+ T cell response to Ad5-encoded viral antigens has previously been demonstrated [28]. Therefore, to test whether this was also the case for NS3 from the YF-17D virus, MHC class II KO mice, as well as wt B6 mice, were vaccinated with Ad-YF NS3, and the splenic CD8+ T cell response was analyzed by ICS at day 8 post vaccination. We found that MHC class II KO mice presented a significantly reduced NS3 (268–75) specific CD8+ T-cell response as evident by markedly lower numbers of IFN-γ producing CD8+ splenocytes in these mice compared to wt B6 animals (Fig 4B). Taken together, these findings imply that CD8+ T cells are the major effectors in the immune response induced by Ad-YF NS3 vaccination, and that CD4+ T cell help is essential in the priming phase. Whether CD4+ T cells are also important at later stages of the immune response cannot be established from our data, but “helpless” CD8+ T cells are known to be inferior with regard to survival and ability to undergo secondary expansion [29,30].
The fact that Ad-YF NS3 induced protection depends on cell-mediated immunity, suggested to us that, upon intracranial challenge, the virus inoculum would have to be cleared from the CNS of wt B6 mice by CD8+ effector T cells. As we had previously observed that lack of H-2Kb molecule resulted in a very drastic reduction of the Ad-YF NS3 induced protection in vivo (10% in H-2Kb KO mice vs. 100% in B6 mice), we wanted to confirm that NS3 vaccinated H-2Kb deficient mice would not be able to control YF-17D infection of the CNS following intracranial challenge. Finally, to further narrow down the identity of the involved T cells, we made an adenoconstruct encoding a truncated version of NS3 (Ad-YF NS3 (201–325)), and the protection associated with this vector was compared to that of the full length NS3 vector. Thus, we analyzed the viral loads in the brains of wt B6 mice that had been immunized with Ad-YF NS3 or Ad-YF NS3 (201–325) 14 days prior to intracranial YF challenge, as well as of H-2kb KO mice immunized with Ad-NS3 prior to intracranial challenge. Unvaccinated wt B6 mice served as controls. Viral titers were analyzed 7 days post challenge as this represents the latest time point for unbiased analysis of acutely challenged mice that would invariably succumb between day 8 and 9 p.c. As expected, the unvaccinated mice uniformly displayed a high viral burden in the CNS (106−107 pfu/g of brain), while the infection appeared to be partly controlled in the brains of Ad-YF NS3 immunized mice with viral titers about 3 logs lower than in the unvaccinated controls (Fig 5). Interestingly, vaccination of wt B6 mice with the truncated version of NS3 (Ad-YF NS3 (201–325)) also resulted in a significantly reduced viral burden in the brain as compared to unvaccinated mice (Fig 5), albeit not quite to the same degree as observed in full-length vaccinated mice. Notably, NS3 immunized H-2Kb deficient mice all displayed high viral loads in the CNS, almost at the same level as in unvaccinated mice. The same difference in the capacity to control YFV infection was observed when wt and H-2kb deficient mice had been vaccinated 60 days prior to challenge (see S1 Fig). These results confirmed that vaccination with the Ad-YF NS3 vector could induce protection through CD8+ T cells predominantly specific for H-2Kb restricted epitopes. Furthermore, the finding that mice given Ad5 encoding the truncated or the full length form of the NS3 protein displayed overlapping reductions in viral loads corroborated the hypothesis that the NS3 (268–275) specific CD8+ T cells represent a key subset of CD8+ T cells pivotal in Ad-YF NS3-induced antiviral immunity, albeit additional epitopes may also play a minor role.
Our best Ad-YF vaccine candidate in terms of in vivo protection appeared to be the vector encoding the viral structural proteins. Given that vaccination with this construct unlike the NS3 targeted construct induced a significant humoral response, we hypothesized antibodies were mainly responsible for the protection observed in Ad-YF C,M,E vaccinated mice. To test this hypothesis, we compared the outcome of intracranial YF challenge in vaccinated B-cell KO mice, H-2Kb/Db double KO mice and wt B6 mice. Mice were injected with 2x107 pfu of Ad-YF C,M,E in the f.p. two weeks prior to intracranial challenge, and the clinical outcome was monitored for 15 days after challenge. Interestingly, lack of B cells only resulted in diminished, but not abrogated protection (about 43% survival, Fig 6), relicating what we have recently observed regarding mice vaccinated with YF-17D [22]. In the case of MHC class I KO mice lacking both H-2kb and H-2Db molecules and with no known defects in B cell/antibody responses, we observed that protection was significantly reduced (40% survival) as compared to survival of all immunized wt B6 mice (Fig 6). Taken together these data suggested that a combination of CD8+ T-cell and B-cell responses are required for protection as induced by Ad-YF C,M,E, thus completely mimicking the protective response in YF-17D vaccinated animals [22].
Based on the above results it would appear that both humoral and T-cell mediated immunity were involved in protecting Ad-YF C,M,E vaccinated mice, while only CD8+ T cells contributed significantly in Ad-YF NS3 vaccinated animals. However, as the above result were obtained using KO mice and therefore could have been biased from events taking place during the priming phase, we decided to reassert our interpretation by performing acute depletion of CD8+ T cells prior to intracranial challenge of both Ad-YF C,M,E and Ad-YF NS3 vaccinated wt mice. To this end, animals were vaccinated as previously described and intracranial challenged with YFV 60 days later; unvaccinated wt mice were included as controls. Part of the vaccinated mice were depleted of CD8+ T cells in connection to the challenge and on day 7 p.c., brains from all mice were collected and viral titers were determined. As expected (Fig 7A), unvaccinated animals all exhibited high viral loads, while most Ad-YF C,M,E vaccinated mice had completely cleared infectious virus from the CNS by day 7 p.c. In these mice CD8+ T-cell depletion lead to a significantly increased viral load, but brain virus titers were still several logs lower in these animals than in the unvaccinated controls; this matches what we recently observed in YF-17D vaccinated mice [22]. In contrast, but still as expected, infectious virus was detected in the CNS of Ad-YF NS3 vaccinated, CD8+ T-cell replete animals, but with viral loads significantly lower than in unvaccinated controls, and antiviral protection was completely abrogated following CD8+ T-cell depletion (Fig 7A).
To define the precise effector mechanisms elaborated by NS3-specific CD8+ T cells, we next analysed the virus titers in brains of Ad-YF NS3 vaccinated IFN-γ KO, Prf KO, IFN-γ/Prf double KO as well as wild type mice that were intracranial challenged 14 days post vaccination; strain-matched unvaccinated animals served as controls (Fig 7B). We observed that vaccination with Ad-YF NS3 still induced a significant, but markedly reduced protection in the absence of IFN-γor Prf. On the other hand, vaccinated double KO mice could not control YFV infection of the CNS and displayed viral loads similar to those of strain matched unvaccinated control animals (Fig 7B) confirming our earlier finding regarding survival of vaccinated double KO mice (Fig 4A).
Finally we asked how long Ad vector-induced protection would last. To this end mice were vaccinated with each of the two adenoviral vectors as before, and 60 and 210 days later, the mice were intracranial challenged with YFV; 7 days later viral loads in the CNS of vaccinated animals were compared to those in unvaccinated controls. As can be seen in Fig 8, markedly reduced virus titers were found in both groups of vaccinated mice compared to unvaccinated controls 60 days post vaccination, and, notably, we did not observe any tendency towards a decline in vaccine-induced protection in the mice vaccinated 210 days earlier, indicating that even when acting on their own, CD8+ T cells have the capacity to provide long-term clinical protection in this murine model of YFV infection.
The first generation of viral vaccines comprised empirically attenuated live viruses, which have been documented to be very effective in preventing many infections. The yellow fever vaccine (YF-17D) belongs to this category, and with more than 500 million doses administered worldwide in the past 70 years, it stands as a paradigm for an effective vaccine [1,5]. For a number of years there was limited interest in further development of YF vaccines due to the successful application of the live attenuated vaccine. In the last decade, however, the use of the YF-17D virus has come into focus again due to the risk of serious adverse events following vaccination, and due to restrictions in immunizing certain categories, such as infants, pregnant women, elderly and immune-compromised people [31]. This has created a demand for alternative vaccine candidates [31], but only a limited number of studies have so far come up with potential new YF vaccines, and they have mostly been focused around making an inactivated vaccine inducing humoral immunity [32–34]. However, if as suggested by both of our studies [22], elicitation of an antiviral CD8 T-cell response is pivotal for full protection against YFV infection, an inactivated vaccine would not represent the best choice, whereas an Ad-vector based vaccine would seem an ideal solution. In a very recent study [11], a naked DNA vaccine encoding the E protein was tested in a murine model similar to ours and found to protect against a fatal outcome of intracranial infection; antibody levels were sufficient to explain the prevention of any mortality, whereas numbers of virus-specific T cells were quite low compared to what we obtain. Importantly, any functional importance of the induced T-cell response was not documented.
In this report, we have evaluated two prototypic Ad-based vectors encoding different YF-17D proteins. The choice of these proteins was based on our previous analysis of how the existing YFV vaccine works [22] and thus represents a translational approach to use the information previously gathered directly in the early design of a safer, but still highly efficient vaccine candidate. Our results clearly demonstrate that induction of YFV-specific CD8+ T-cell responses can lead to effective antiviral immunity in an infection model where protection until recently was ascribed entirely to neutralizing antibodies [4]. We demonstrate that a single vaccination with Ad-YF NS3, containing a conserved nonstructural viral protein, can induce almost complete protection from subsequent intracranial YFV challenge in wt B6 mice, and that CD8+ T cells specific for H-2Kb-restricted viral epitopes are the major effectors in the vaccine-induced immunity. These findings confirm and extend recent results obtained in similarly challenged YF-17D vaccinated mice [22]. Here we obtained strong evidence for a primary role of antibodies in the protection against YF disease in mice, but also found evidence pointing to a supporting role for virus-specific CD8+ T cells. Thus, an accelerated influx of virus-specific CD8+ T cells into the infected CNS was revealed in vaccinated mice, and CD8+ T-cell depletion significantly reduced antiviral protection [22]. In the present study we observed that CD8+ T-cell depletion completely abolished virus control in Ad-YF-NS3 vaccinated mice, and under these conditions we could show that absence of either IFN-γ or perforin significantly impaired virus control, whereas deficiency of both completely prevented a reduction in CNS viral load at 7 days post challenge. This is in slight contrast to our previous findings in YF-17D vaccinated mice in which case only mice lacking both IFN-γ and perforin were found to be significantly impaired in their ability to control infection of the CNS [22]. This apparent discrepancy probably reflects that YD-17D vaccinated mice unlike NS3 vaccinated mice have preexisting neutralizing antibodies, which also contribute to the observed virus control and hence lead to a blurring of the results [22]. Thus taken together our data convincingly demonstrate that T-cell released effector molecules together with antibodies represent an integrated component of the adaptive immune response to YFV in CNS.
To further address whether this focused T-cell response could be responsible for the in vivo protection conferred by Ad-YF NS3 vaccination, we designed a vector encoding a truncated version of the protein (Ad-YF-NS3 (201–325)). In preliminary experiments we observed that the truncated form of NS3 induced similar numbers of NS3 (268–275) specific T cells as did the full length protein (average of 5 mice/group: 107 tetramer+ CD8 T cells/spleen in Ad-YF-NS3 (201–325) vaccinated mice vs. 9x106 CD8 T cells/ spleen in Ad-YF NS3 vaccinated mice). However, the immune mediated virus control in the CNS was slightly reduced in mice vaccinated with the former construct, suggesting that while NS3 (268–275) specific T cells clearly represent the functionally most important subset of CD8+ T cells in NS3 immunized mice, there are likely to be additional minor specificities involved following immunization with the full length construct.
Notably, our attempt of generating a vector encoding the three YF viral structural proteins C, M and E resulted in an even more promising vaccine candidate. This was, to some extent, expected, as neutralizing Abs directed against the E protein have long been associated with protection from yellow fever [4]. However, upon immunization with Ad-YF C,M,E, we also noted a very strong CD8+ T cell response toward the subdominant viral epitope contained in the E protein, as compared to vaccination with YF-17D virus. CD8 T-cell depletion of these mice confirmed that antiviral protection reflected a combination of a humoral and a cell-mediated immune response similar to what we have recently found to be the case in YFV vaccinated mice [22]. Analysis of the neutralizing titers in Ad-YF C,M,E vaccinated mice revealed that such antibodies were induced, albeit not quite to the same level as found in YFV vaccinated mice.
In conclusion, our current results provide new independent support for our recent claim [22] that CD8+ T cells may play a significant role as effector cells against YFV infection. Still, the mechanisms of protection against intracranial challenge in mice may differ from those controlling the liver infection of humans. T-cell mediated immunity may be particularly important for controlling CNS infection owing to the blood/brain barrier, which initially limits antibody access. Most importantly, however, our findings demonstrate that by using the information previously gathered in studies of YFV vaccinated mice [22], it was possible to make promising vaccines candidates based on replication deficient adenoviral vectors, which might work—either alone (Ad-YF C, M, E) or combined (to broadened the T-cell response)—as potentially safe alternatives to the use of a live attenuated vaccine for induction of protection against YFV-induced disease; certainly, SAEs which represent visceral and neural dissemination of the attenuated YFV vaccine should be avoided. Notably, when it comes to human vaccinations there may be a need to substitute Ad5 with another adenovector due a high prevalence of antibodies against this serotype in certain human populations [35–37]. However, adenoviral vectors of non-human origin have already reached clinical testing (Ebola [38]) and several appear to match Ad5 in immunogenecity and safety [19,39]. Another critical issue may be the longevity of adenovector-induced protection in humans, e.g. adenoinduced protection towards Ebola infection is short-lived in non-human primates [18]. This should not represent a major problem, however, since numerous studies have demonstrated that adenovector priming works well in conjunction with heterologous boosting, e.g. using modified vaccinia Ankara [18]. In summary, our findings are very promising, and we believe that an adenovector based approach to develop a safer alternative for YFV vaccination deserves further evaluation in relevant animal models.
|
10.1371/journal.pbio.1002292 | Mechanical Coupling between Endoderm Invagination and Axis Extension in Drosophila | How genetic programs generate cell-intrinsic forces to shape embryos is actively studied, but less so how tissue-scale physical forces impact morphogenesis. Here we address the role of the latter during axis extension, using Drosophila germband extension (GBE) as a model. We found previously that cells elongate in the anteroposterior (AP) axis in the extending germband, suggesting that an extrinsic tensile force contributed to body axis extension. Here we further characterized the AP cell elongation patterns during GBE, by tracking cells and quantifying their apical cell deformation over time. AP cell elongation forms a gradient culminating at the posterior of the embryo, consistent with an AP-oriented tensile force propagating from there. To identify the morphogenetic movements that could be the source of this extrinsic force, we mapped gastrulation movements temporally using light sheet microscopy to image whole Drosophila embryos. We found that both mesoderm and endoderm invaginations are synchronous with the onset of GBE. The AP cell elongation gradient remains when mesoderm invagination is blocked but is abolished in the absence of endoderm invagination. This suggested that endoderm invagination is the source of the tensile force. We next looked for evidence of this force in a simplified system without polarized cell intercalation, in acellular embryos. Using Particle Image Velocimetry, we identify posteriorwards Myosin II flows towards the presumptive posterior endoderm, which still undergoes apical constriction in acellular embryos as in wildtype. We probed this posterior region using laser ablation and showed that tension is increased in the AP orientation, compared to dorsoventral orientation or to either orientations more anteriorly in the embryo. We propose that apical constriction leading to endoderm invagination is the source of the extrinsic force contributing to germband extension. This highlights the importance of physical interactions between tissues during morphogenesis.
| Embryos change shape dramatically during development. The genetic programs that drive the active behavior of cells underlying these changes are well understood, but little is known about how movements of neighboring tissues influence the shaping of a given tissue. We address this question for the anteroposterior elongation of the body axis (germband) of Drosophila embryos. We had previously shown that during elongation, the germband cells stretch along the anteroposterior axis, in addition to undergoing active rearrangements; this suggested that extrinsic tensile forces might be at play. In the current study we find that the start of main body elongation is synchronous with the invagination of both the mesoderm and the endoderm. We analyze mutants and find that cell stretching disappears in embryos lacking endoderm invagination but remains in those lacking mesoderm invagination. We then measure tension using laser ablation in acellular embryos that lack active cell rearrangements in the germband but undergo the initial stages of endoderm invagination. We find that tension is higher in the anteroposterior direction close to the invaginating endoderm. Our results indicate that endoderm invagination generates a tensile force that is transmitted to the germband, and contributes to its elongation. This study reveals how tissues interact during embryo morphogenesis.
| During development, many tissues extend in one orientation while narrowing in the orthogonal one. These so-called convergence and extension movements elongate the anteroposterior axis in bilateral animals during gastrulation, where they have been most studied [1–4]. Defects in convergence and extension movements at gastrulation have been linked to neural tube defects in mouse and human embryos [5]. Convergence and extension movements are also important later in embryo morphogenesis, for example for the elongation of the cochlear tube [6], the kidney tubules [7], and the limb and jaw cartilages [2].
Intracellular forces are key in convergence and extension and in most cases studied, drive polarized cell rearrangements [1,2]. These require planar polarization of proteins at cell membranes [3,8]. Planar polarization of actomyosin was first shown in Drosophila germband extension (GBE) to result in the selective shortening of dorsoventrally (DV) oriented cell contacts [9,10]. The cell biology of this process has since been extensively characterized, and planar polarization of several other components including Bazooka (the homologue of Par-3) and E-cadherin have been found to be required for active cell rearrangements [11–20]. These polarities are controlled by the anteroposterior (AP) segmentation cascade in Drosophila, the most downstream genes being the pair-rule genes, encoding transcription factors such as Even-skipped and Runt [9,10,21]. Recent work has found that a combinatorial code of Toll-like receptors is required for transducing the AP positional information from these transcription factors into the planar polarities required for polarized cell intercalation [22]. Recently, actomyosin-driven shortening of cell contacts has also been shown to be essential for convergence and extension movements in vertebrates [7, 23–25].
However, cell-autonomous behaviors might not be sufficient to fully explain axis elongation [26]. Stresses generated by neighboring morphogenetic movements or by the constrained geometry of the embryo could contribute to axis extension [27–29]. Evidence for extrinsic forces influencing tissue elongation has been reported: in Caenorhabditis elegans, body wall muscle contractions guide embryonic elongation [30]; in Drosophila oogenesis, the traction force produced by the follicle rotation is required for egg chamber elongation [31]; in the Drosophila developing wing, the contraction of the hinge produces a tensile stress that orients the cell behaviours required for wing blade elongation [32,33].
In the Drosophila embryo, we found previously that in addition to polarized cell intercalation, AP cell elongation contributes to GBE [34]. These cell shape changes are not a consequence of cell rearrangements: in the absence of polarized cell intercalation, the germband cells elongate even more in AP, a behavior most parsimoniously explained by an extrinsic tensile force acting on the tissue [34]. This gives us the opportunity to investigate how extrinsic factors can contribute to axis extension. Here, we search for the source of the extrinsic force acting on the germband by measuring the deformation of cells as a function of time, in the absence and presence of other morphogenetic movements. We find that blocking posterior endoderm invagination abolishes AP cell elongation. Furthermore, we present evidence that apical constriction leading to invagination of the posterior endoderm primordium produces a tensile force propagating from the posterior of the embryo. We conclude that this gastrulation movement at the posterior produces an AP tensile force contributing to the elongation of the main axis in Drosophila.
We analyzed apical cell shape changes using custom-made algorithms as previously [34,35]. We imaged embryos labeled with the junctional marker ubi-DE-cad-GFP on their ventral side by confocal time-lapse microscopy, acquiring images every 30 s at 20.5 ± 1°C, starting movies around morphological stage six and finishing around stage eight (Fig 1A, 1A’, 1C and 1C’). We segmented apical cell contours based on the fluorescent signal and linked cells in time, storing the coordinates of the centroid of each cell and of a polygon describing its outline, at each timepoint (Fig 1D and 1D’). To measure the cell shape changes, our algorithms consider small cell neighborhoods consisting of a central cell surrounded by one ring of its immediate neighbors (Fig 1B). Cell shapes for this neighborhood are measured by fitting an ellipse to each cell: strain rates are calculated over a 2 min window (±2 timepoints, see Fig 1B). To analyze specifically the AP component of cell shape change (the component that will contribute to axis extension), the strain rates were projected onto the AP embryonic axis. In our summary plots, we call this strain rate “AP cell length change,” expressed in proportion per minute (pp/min) (Fig 1E–1F’) and shorten it to “AP cell elongation” in the text thereafter. Note that from our measures of strain rates, we can also extract DV cell elongation and cell area change (see below). To consider only the deformation of cells from the germband (the tissue undergoing convergence and extension), we excluded any tracks from mesoderm and mesectoderm cells (Fig 1D and 1D’). These methods allow us to examine the patterns of AP cell elongation in living embryos, which we proposed to be a signature of an extrinsic force contributing to axis extension [34].
We had previously analyzed AP cell elongation in field of views that included the cephalic furrow as an anterior landmark (the cephalic furrow forms between the head and the germband) [34] (Fig 1A, 1A’ and 1C). These views show the anteriormost region of the ventral side of the germband and are thereafter called “anterior views” for simplicity. When visualizing AP cell elongation as a function of time and position along the AP axis in spatiotemporal heat maps, we noticed that the signal was higher towards the posterior edge of the field of view [34] (average for five movies, Fig 1E; individual movies, S1A Fig; tracking information, S1C and S1C’ Fig). This prompted us to image the ventral side of embryos more posteriorly, using the tail end of the embryo (as detected in apical optical sections) as a posterior landmark (Fig 1A’ and 1C’). Plotting spatiotemporal maps for these “posterior views” revealed that AP cell elongation becomes even stronger closer to the posterior tip of the embryo (average for four movies, Fig 1E’; individual movies, S1B Fig; tracking information, S1D and S1D’ Fig; example S1 Movie). Indeed, although AP cell elongation peaks around 10 min after GBE onset in both views, the magnitude is doubled in posterior views: 0.04 pp/min (average for four movies, Fig 1F’) compared to 0.02 pp/min in anterior ones (average for five movies, Fig 1F). Note that to be able to make fair comparisons between anterior and posterior views, we removed the tracks of ectodermal cells deformed by the cephalic furrow in anterior views (purple shaded region in Fig 1C, resulting tracks in Fig 1D), since these unrelated cell deformations would otherwise contribute to our measure of total AP cell elongation, as they did in our previous study [34]. All anterior views presented in this paper have been reanalyzed with this exclusion. We estimated that in wild-type embryos, the two fields of view overlapped by about 80 microns (Fig 1A’), and we concluded that the patterns of AP cell elongation detected in posterior views fully included the patterns seen in anterior views (Fig 1E and 1E’).
The AP cell elongation patterns appeared to form a gradient increasing from the anterior to the posterior. To ascertain this, we examined a short period around the peak of AP cell elongation, from 7.5 min to 12.5 min after GBE onset (Fig 2). This confirmed that AP cell elongation increased steeply towards the posterior of the embryo (Fig 2A–2D), forming a gradient over a distance of about 150 μm in posterior views (average for four movies, Fig 2D). Although the gradient is clearest in posterior views, some AP gradation was already detectable in anterior views (average for five movies, Fig 2C), consistent with the notion that we are visualizing the beginning of the gradient in anterior views. In posterior views, we also looked at snapshots of the gradient earlier in GBE, at 2.5, 5, and 7.5 min: the gradient was at first shallow and confined to the more posterior part of the field of view; it then expanded towards the anterior and became steeper with time (Fig 2E). These results suggested that a tensile stress deformed the tissue from a posterior source, starting at the onset of GBE and propagating towards the anterior of the embryo over time.
We also analyzed cell area change in addition to AP cell elongation (S2A and S2A’ Fig). When passively responding to planar extrinsic forces, cell apical areas are expected to change in opposite ways depending on whether cells are compressed or pulled: when pulled, cell areas should increase; in contrast, when compressed, cell areas should decrease. We had already noted in our previous study that AP cell elongation was accompanied by an increase in cell area in anterior views, supporting the idea that the germband was experiencing a planar tensile stress [34] (S2A Fig). This trend is even clearer for the posterior views: the patterns of AP cell elongation are matched by patterns of cell area increase, suggesting that the germband cells elongated in response to a tensile rather than compressive stress (compare S2A’ Fig with Fig 1E’). Note that in our analyses, we can observe changes in only the two planar axes defining the apical cell areas, but we expect the third axis, the cell length in Z, to increase or decrease in response to planar stress to keep the cell volume constant [37,38].
Around the onset of GBE, the germband cells are also subjected to a pull in the perpendicular direction, along DV, in response to the invagination of the mesoderm on the ventral side of the embryo [34] (mesoderm invaginates through a ventral furrow visible in Fig 1A’, 1C and 1C’). In both anterior and posterior views, we found that DV elongation of ectodermal cells in response to mesoderm invagination have patterns completely distinct from the AP cell elongation patterns that we are focusing on in this study: first, they are most prominent close to GBE onset and have disappeared by 10 min into GBE (whereas the AP cell elongation patterns peak just after 10 min), and second, they occur uniformly along the AP axis of the embryo (whereas the AP cell elongation patterns occur in a posterior gradient) (S2B–S2C” Fig). Note that AP and DV cell elongation patterns are both accompanied by an increase in cell area (S2A and S2A’ Fig), consistent with the idea that they are both the consequence of tensile forces. We concluded that germband cells are subjected to two independent tensile forces, one in the DV direction caused by mesoderm invagination (see also below), and one in the AP direction coming from the posterior of the embryo.
Together, our analysis of wild-type Drosophila embryos indicated that AP cell elongation formed an AP gradient consistent with a stress propagating from the posterior. We asked next what the origin of this tensile force was.
A stress propagating from the posterior seemed at odds with our previous model suggesting a role for mesoderm invagination in generating AP patterns of cell elongation [34]. This model was based on the analysis of anterior views, where we had previously found that AP cell elongation contributing to axis extension was reduced in twist (twi) mutants, which are defective for mesoderm invagination. Although we had proposed at the time that mesoderm invagination might contribute to the extrinsic tensile force deforming the germband, it was difficult to formulate a model for how it could do so [29,34]. We reanalyzed the data from anterior views after exclusion of the region deformed by the cephalic furrow (see above). We confirmed our previous results: in anterior views, AP cell elongation was significantly reduced in twi mutants compared to wild type (average for five movies, Fig 3A and 3A’; individual movies, S3A Fig; example S2 Movie). Next, we acquired new movies imaging the posterior ventral side of the embryo, using the posterior end of the imaged embryo as a landmark, as before for wild type. To our surprise, we found robust AP cell elongation in posterior views of twi embryos, with no statistical difference between the rate of AP cell length change between these mutant embryos and wild type (average for three movies, Fig 3B and 3B’; individual movies, S3B Fig). Elongating cells tended to increase in area in these posterior views, suggesting that they elongated in response to a tensile stress, as in wild type (S3C’ Fig). Note that in these cell area plots, the cell area increase in response to mesoderm invagination is absent (0 to 5–7 min), in posterior as in anterior views, demonstrating that the embryos we imaged are indeed defective for mesoderm invagination (compare S2A Fig with S3C Fig, and S2A’ Fig with S3C’ Fig). Further demonstrating this, DV cell elongation is gone in anterior and posterior views of twi mutant embryos (S3D, S3D”, S3E and S3E” Fig, compare with S2B, S2B”, S2C and S2C” Fig). This shows that whereas the early DV stretch of ectodermal cells is gone as expected in twi mutants (because there is no mesoderm invagination to pull the ectoderm in DV), the AP stretch of ectodermal cells is still present in posterior views (S3E and S3E’ Fig). This confirmed that DV and AP cell elongation were produced by two independent tensile forces, and that mesoderm invagination caused DV cell elongation in the germband. Refuting our previous model [34], this also indicated that mesoderm invagination did not cause the AP cell elongation contributing to GBE.
As before for wild type, we examined the gradient of AP cell elongation between 7.5 and 12.5 mins and confirmed that there is a significant difference with wild type for anterior views but no clear statistical difference when comparing posterior views (Fig 3C–3G”). This discrepancy suggested that the relative position of anterior and posterior fields of view are different in wild-type and twi mutants, leading to the detection of the AP cell elongation gradient in posterior views, but not in anterior views, in twi mutants. This is likely to be the result of several factors, one of which might be a difference in curvature on the ventral side of the embryo between the two genotypes. Indeed, we find that the outlines of twi embryos are less curved than wild-type ones in anterior views, and the embryos are wider, consistent with the notion that twi embryos are flatter (S4 Fig). A flatter ventral surface in twi mutants would make the posterior views more posteriorly located in twi mutants, because the position of the posterior landmark we use (the tip of the embryo in optical sections) will be influenced by curvature. A flatter surface could be a direct consequence of the failure of mesoderm invagination and the absence of a keel-like shape in twi mutants. Absence of invaginating mesoderm could not only affect the curvature of the embryo, but also change its mechanical properties and, for example, make it flatten more under a coverslip during imaging. Both factors would make the anterior and posterior fields of view further apart in twi mutants compared to wild type.
We concluded that a gradient of AP cell elongation was present in twi mutants and grossly similar to wild type in posterior views, showing that an event other than mesoderm invagination must be responsible for the AP extrinsic force deforming the germband.
We reasoned that candidates for generating a tensile stress at the posterior would be morphogenetic movements taking place at, or just before, the onset of GBE, because germband cells start to elongate in AP from the beginning of GBE [34] (Fig 1E and 1E’). To identify such events, we measured the timings of gastrulation movements relative to the start of GBE (Fig 4). Because some movements take place on the ventral surface (mesoderm invagination) and others on the dorsal surface (endoderm invagination, dorsal folding) (Fig 4A, see also Fig 1A and 1A’), we used light sheet microscopy (SPIM, selective plane illumination microscopy) to image the whole embryo volume through developmental time [39]. We labelled the cells with plasma membrane markers such as Spider-GFP and Resille-GFP and took timepoints every 30 sec (at 28–30°C). We examined three wild-type movies and three twi mutants defective for mesoderm invagination (Fig 4B).
We mapped the onset of GBE by identifying the first posteriorward displacement of ventral cells (Fig 4C and 4C’, and S3 Movie) and used the corresponding time-point as time zero for all the movies. To check that the development rates of all embryos imaged were comparable, we used patterned mitoses in the head as a developmental timer (Fig 4D– 4D”)[40]. We found that these mitoses start at 8.5 min, 10.5 min, and 11.5 min after GBE onset in the three wild-type movies and at 10.5 min, 11 min, and 12 min in the three twi movies (Fig 4B). This showed that there were no obvious differences in development rates between embryos and illustrates the temporal reproducibility of Drosophila early development.
Next, we mapped the timings of morphogenetic movements visible in the movies (Fig 4A) (for a review of the anatomy of these movements, see [29]). We concluded that the two morphogenetic movements most synchronous with GBE onset were mesoderm and posterior endoderm invaginations (Fig 4B). We mapped the onset of posterior endoderm invagination (also called posterior midgut invagination) by identifying in which movie frame the cells initiated apical constriction at the posterior of the embryo (Fig 1E and 1E’ and S3 Movie). Posterior midgut invagination preceded GBE by −3.5, −2, and −1.5 min in the three wild type, and by −2, −1.5, and −0.5 min in the three twi mutant embryos (Fig 4B). To map a clearly identifiable step of mesoderm invagination, we recorded the timepoint when the right and left sides of the mesoderm first met to begin forming the internal mesodermal tube, thereafter called “mesoderm sealing” (Fig 4F and 4F’ and S4 Movie). The times relative to the onset of axis extension were −0.5 min, −0.5 min, and +0.5 min for the 3 wild type movies (twi embryos fail to form a mesodermal tube) (Fig 4B). This confirms a remarkable synchrony between mesoderm sealing and GBE onset, which we had noted before [34] (see Discussion). We also looked at morphogenetic movements that occur on the dorsal side of the embryo. Dorsal folding occurs in two stereotyped locations under the control of the AP patterning system [41]. Although these folds start forming just before the onset of axis extension in wild type embryos, they initiate after GBE onset in two out of three twi embryos (Fig 4G and 4G’). Since AP cell elongation at the posterior end of the embryo are already high in twi mutants at GBE onset (Fig 3B), this suggests that the dorsal folds are not initiating these (although they could later contribute). Other dorsal movements include a dorsal contraction (Fig 4H, 4H’ and 4B) and the onset of amnioserosa cell flattening [42]. These occur respectively too early and too late, relative to the onset of GBE, to be key influences.
We conclude from this temporal mapping of morphogenetic movements that both mesoderm sealing and endoderm invagination are synchronous with the onset of GBE. Since we have refuted a role of mesoderm invagination in producing the gradient of AP cell elongation contributing to axis extension (see previous section), posterior endoderm invagination was the main candidate to generate a tensile stress during GBE.
To test a role of posterior endoderm invagination in AP cell elongation during axis extension, we examined folded gastrulation (fog) and torso-like (tsl) mutants that abolish endoderm invagination. Fog is a zygotic gene required for the apical constriction of the endoderm primordium cells arranged in a disc at the posterior, which leads to posterior midgut invagination [43]. The expression of fog in the posterior midgut primordium requires the zygotic gap genes huckebein and tailless, which themselves require the activity of the maternal gene tsl, an upstream component of the terminal patterning system [44]. In anterior views, no obvious AP cell elongation gradient was detected at the onset of GBE in fog mutants (compare S5B Fig with Fig 1E). However, fog mutants proved problematic to analyze because their extending germband form ectopic folds (arrows in S5A and S5B Fig and S7 Movie). These folds occur because the posterior end of the germband does not move away in these mutants, but polarized active cell intercalations still elongate the germband [43]. Folding stretched the germ-band cells locally and produced strong AP cell elongation, as seen on spatiotemporal maps from approximately 7 min after GBE onset (S5B Fig). As a consequence, the total AP cell elongation could not be meaningfully compared between wild-type and fog mutants.
To prevent folding, we analyzed one of the mutants that abolishes posterior midgut invagination, tsl, in combination with a mutant abolishing most of the active polarized cell intercalations in the trunk, Kruppel (Kr) [21,34]. To ask if tsl was required for the gradient of AP cell elongation, we compared Kr single mutants with these Kr; tsl double mutants. In posterior fields of views, AP cell elongations are slightly higher in Kr compared to wild type (Fig 5A). This was expected, since AP cell elongation increases in the absence of cell intercalation, presumably because in wild type, polarized cell intercalation acts to release some of the tensile stress in the germband [34]. The patterns of AP cell shape changes are, however, comparable in both genotypes (average for three movies, Fig 5B, compare with Fig 1E’; individual movies S5C Fig; example S5 Movie). As in wild type, the AP cell shape changes are accompanied by an increase in cell area, consistent with a tensile rather than compressive stress (Fig 5H; individual movies in S5D Fig). In double mutants Kr; tsl however, very little AP cell length change was detected (average for three movies, Fig 5C and 5D; individual movies S5E Fig; example S6 Movie). Note that the residual AP cell length change detected on the averaged spatiotemporal map (Fig 5C) was mainly present in one of the three individual movies (krtslCL040713, S5E Fig), and this signal was not accompanied by an increase in cell area, as would be expected for a tissue under tensile stress (S5F Fig). Consistent with this, there was no significant increase in cell area detected in double mutants Kr; tsl in the other two movies or in the averaged data (S5 Fig and Fig 5I). This indicated that the ectodermal cells in the posterior region of Kr; tsl mutants embryos were not under tensile stress.
We also examined in more detail the AP cell elongation gradient around its peak (from 7.5 to 12.5 min), in Kr versus Kr; tsl mutants. The steep gradient of AP cell elongation was abolished in Kr; tsl mutants (Fig 5E–5G). We concluded that posterior midgut invagination was required for the AP cell elongation contributing to axis extension in Drosophila.
To understand more precisely how posterior endoderm invagination could produce a stress that in turn leads to a gradient of AP cell elongation, we analyzed a simplified system, in the form of acellular mutant embryos. Several mutations are known that produce embryos, which fail to cellularize. In one such mutant, an endoderm-like invagination is still visible on the dorsal side of the embryo, suggesting that apical constriction of the endoderm primordium still occurs in acellular embryos [45]. Consistent with this notion, another acellular mutant was shown recently to undergo apical constriction of the mesoderm primordium, albeit at a slower rate (about 60% of the wild type) [46]. To confirm that apical constriction also happened for the endoderm primordium, we made movies of these acellular mutants expressing sqh-GFP [46], to visualize the actomyosin cytoskeleton (sqh encodes the non-muscle Myosin II Regulatory Light Chain) (S8 Movie). We observed a concentration of Myosin II in the region where apical constriction would normally occur in the presumptive posterior endoderm, close to where the pole cells (PC) are attached, in both live and fixed embryos (Fig 6A and 6B, and S6D’ Fig). We find that the acellular embryos go through the initial steps of wild-type endoderm invagination [43], with first the formation of a flattened plate at the posterior (S6D’ Fig), then constriction of the embryo’s surface leading to some degree of invagination (Fig 6A–6C) (see also Fig 3a in [46]).
In live embryos, we noticed that the concentration of Myosin II at the posterior is accompanied by flows of Myosin II towards it (S8 Movie, top panel). This suggested that apical constriction of the presumptive endoderm surface could exert a tensile stress on the surrounding apical surface of the embryo. We also saw flows towards the ventral region, presumably in response to apical constriction of the presumptive mesoderm. We confirmed the direction of these flows by tracking the Myosin II signal at the surface of the acellular embryos using Particle Imaging Velocimetry (PIV). In our example movie showing the whole lateral surface of the presumptive germband, we can clearly see by PIV both ventralward (towards mesoderm) and posteriorward (towards posterior endoderm) flows of Myosin II signal (S8 Movie, bottom panel). To confirm the existence of posterior flows, we acquired more movies of the posterior end of the embryo and visualized the flows by PIV. We found that all embryos analyzed showed posteriorward flows towards the presumptive posterior endoderm (n = 8, 2 examples in Fig 6A’–6B’).
To understand better how the Myosin II flows relate to the surface membranes of the acellular embryos, we compared the localization of Myosin II with those of the E-cadherin complexes. Just before gastrulation movements started, E-cadherin and Myosin II colocalized in a hexagonal-like pattern (estimated stage 5; S6A, S6A’, S6C and S6C’ Fig). These presumably correspond to the regions of the surface membrane that, in wild-type embryos, would normally invaginate and become furrow canals encircling each syncytial nucleus (for example, see [47]). Once gastrulation movements started in acellular embryos, this relatively regular organization became disrupted: E-cadherin and Myosin II still colocalized but now formed a meshwork at the surface of the embryo (estimated stage 7; S6B, S6B’, S6D and S6D’ Fig). Since E-cadherin complexes are presumably associated with membranes, we infer that Myosin II flows that we observe track the movement of surface membranes in these embryos.
The presence of posteriorward flows of Myosin II signal in acellular embryos suggested that apical constriction of the presumptive endoderm was able to pull the apical surface behind it and could generate an AP tensile stress, which in wild-type embryos could contribute to axis extension. We reasoned that acellular embryos provided an excellent system in which to physically probe this tension, since it is unlikely to exhibit more complex morphogenetic behaviours such as polarized cell intercalation, and so we could rule out a contribution of the latter to measured tensions. No planar polarization of Myosin II was recognizable in these embryos, confirming that the apical surface of the embryo was unlikely to undergo intercalation-like movements (S6A–S6D’ Fig).
To directly test our hypothesis that apical constriction of the endoderm primordium generated a tensile stress at the posterior, we carried out line ablations at the surface of the embryo. Using a near-infrared laser, we made 20 micron-long incisions oriented parallel to the AP or DV embryonic axes and at the posterior or the anterior of the presumptive germband, on the lateral side of sqh-GFP-labelled acellular embryos (Fig 6D and S6E Fig). If, as we proposed, a tensile stress propagated from the posterior endoderm, we predicted that the DV cuts at the posterior should show a faster relaxation than any of the other three types of cuts. We used fine-grained PIV to track the movement of the Myosin II network, as a proxy for surface motion, and measured the velocities of recoil in a small region around the cuts, subtracting the velocity of that region before the cut to correct for translation (see Materials and Methods) (S6F–S6G’ Fig). We found that, as predicted, the average relaxation velocity of the DV cuts at the posterior was significantly higher than for any of the other cuts (Fig 6E). At the posterior, there was a clear anisotropy in the relaxation velocities, the DV-oriented cuts relaxing much faster than the AP-oriented cuts, whereas at the anterior, there was no statistically significant anisotropy. This provided evidence for an increased AP-oriented tension at the posterior of the embryo, in response to apical constriction leading to invagination of the endoderm primordium in acellular embryos.
We have investigated the origin of the patterns of planar cell shape changes that we hypothesized previously were the signature of an extrinsic force acting during Drosophila axis extension [34]. We showed that the AP-oriented elongation of cell apices contributing to GBE are strongest at the posterior end of the embryo and decrease gradually towards the anterior. AP cell elongation is accompanied by an increase in cell area, suggesting that this gradient of cell shape change arises in response to a planar tensile stress coming from the tail end of the embryo. We found that the patterns of AP cell elongation and cell area increase are eliminated in the absence of posterior endoderm invagination (but not mesoderm invagination), suggesting that this morphogenetic movement is the source of the extrinsic force deforming the germband. We show that in acellular embryos, the cortical Myosin II meshwork flows towards the contracting posterior endoderm region, and that this is accompanied by an increased tension at the posterior. We conclude from these experiments that the apical constriction and invagination of the posterior endoderm primordium generates a tensile stress propagating to the germband and causing the AP cell elongation gradient that contributes to Drosophila axis extension (Fig 7).
We can think of two alternative explanations that could challenge this conclusion. First, AP cell elongations could be a secondary consequence of active cell intercalation. However, in AP patterning mutants such as Kr, where active polarized cell intercalation is diminished, AP cell elongation is increased rather than decreased [34]. This indicates that active cell intercalation (and AP patterning) is not required for AP cell elongation. Also, cell intercalation rates are high throughout the trunk [34], whereas AP cell elongation is found in a gradient culminating at the posterior (this paper). Therefore, these differing spatial patterns suggest that these two cell behaviours have independent origins. Also in acellular embryos, we observe posteriorward flows of the apical cortex associated with increased tension at the posterior, in absence of polarized cell intercalation. Together, this argues that polarized cell intercalation is not responsible for the gradient of AP cell elongation we observe.
Another possibility is that AP cell elongations are cell-autonomous, that is the result of an active spreading of the germband cells under the control of a genetic program. AP patterning is not required (see above), and the other patterning systems known to operate in the early embryo are the DV and terminal systems [44]. The observed gradient of AP cell elongation is orthogonal to the DV patterning axis and extends outside the terminal domain, so it cannot be explained simply by the activity of either of these systems. We conclude that the most parsimonious explanation is that the AP cell elongation patterns we observe are passive cell behaviours that occur in response to mechanical stresses.
We have found that the AP cell elongation gradient is still present in twi mutants in posterior views, refuting our previous model for a role of the mesoderm in producing these cell shape changes, which was based on analyzing anterior views [34]. We think that the source of the discrepancy is that the anterior and posterior views we imaged are further apart in twi mutants compared to wild-type, which means that the AP cell elongation gradient was mostly missed in twi anterior views. We identify at least one factor, curvature, to explain this difference. The difficulty in registering fields of view between these two genotypes precludes a more detailed comparison of the AP cell elongation gradient. Therefore, we cannot rule out a subtle contribution of mesoderm invagination to GBE. For example, mesoderm invagination, by changing the shape and perhaps the mechanical properties of the germband, might affect how the stress from endoderm invagination propagates throughout the ectoderm. This has some support from the analysis of the AP cell elongation gradient’s slope at specific timepoints, which appear shallower in twi mutant (see for example timepoint 7.5 min in Fig 3G”). To be able to compare the gradient of AP cell elongation between the two genotypes, we will need to perform apical cell deformation analysis in whole embryo movies such as the SPIM movies presented in this paper, in order to circumvent the problem of registering fields of view.
Our experiments identify the endoderm primordium as a source of tensile force. Using acellular embryos allowed us to explore how mechanical stresses could be produced by the posterior endoderm. Although they do not have cells, these mutant embryos are able to undergo the initial steps leading to both mesoderm [46] and endoderm invagination (this study). The apical surfaces of the embryo corresponding to the mesoderm and endoderm primordia are seen to enrich Myosin II, contract, and begin to invaginate ([46], this study), as in wild-type embryos [48]. A rigorous quantitative analysis on the mesoderm has demonstrated that the apical forces of constriction are transmitted to the underlying cytoplasm deep in the tissue and are sufficient to promote invagination, showing that cell individualization is dispensable, at least for the initial phases of invagination [46]. Our qualitative study suggests that the forces generated by apical constriction are also transmitted in the plane at the surface of acellular embryos. Using PIV, we visualized surface flows of Myosin II towards the mesoderm and endoderm primordia. Our laser ablation experiments indicate that the flows towards the endoderm primordium are accompanied by an increase in tension at the cortical surface of the acellular embryo. This suggests that apical cell constrictions of the endoderm primordium and the beginning of invagination are able to produce planar forces that pull the adjoining apical surfaces of the germband.
How do stresses transmitted from the apical cortex of constricting endodermal cells translate into a gradient of AP cell elongation in the elongating germband? Epithelial cells of the germband are thought to be connected mechanically to each other through the actomyosin cytoskeleton interacting with components of the apical adherens junctions such as the E-cadherin complexes [29,49]. Thus, tensile stresses caused by apical constriction should propagate through tissues and can conceivably result in mechanically stretching cells over some distance. We find here that germband cells elongate in AP over a distance of at least 150 μm from the site of endoderm constriction (See Fig 1E’). The gradation in AP cell elongation in response to endoderm invagination that we observe might be explained by friction or resistance from the cellular environment. These would prevent forces being instantaneously propagated throughout the germband. Since the germband tissue has to go around the posterior tip of the embryo to elongate, geometry might also have an impact on how forces are transmitted. Finally, we cannot exclude that spatial variation in stiffness of germband cells along the AP axis could cause them to respond differently to mechanical stress.
Endoderm and mesoderm invagination are both triggered by apical constrictions powered by apical networks of actomyosin [48]. We previously detected a stretch of the ectodermal cells in DV behind the invaginating mesoderm [34]. We confirm this in this paper, showing that DV elongation of the germband cells occurs for the first 5–7 min of GBE in wild-type. This is abolished in twi mutants in which mesoderm invagination is defective. Thus, germband cells are subjected to two independent tensile forces: one in the DV direction (around the onset of GBE) caused by mesoderm invagination, and another in the AP direction (during early GBE), caused by posterior endoderm invagination. Together, these observations show that the epithelial cells in the germband can respond passively to tensile stress generated in adjacent tissues apically constricting and invaginating, by stretching along the direction of stress.
The directionality of apical cell elongation is strongly constrained to AP for the patterns linked to endoderm invagination and to DV for those linked to mesoderm invagination. Indeed, the patterns of AP cell length change caused by endoderm invagination are not accompanied by much change in DV cell length and vice versa for mesoderm invagination (Compare S2C’ and S2C”Fig). Since both AP and DV cell elongation patterns are accompanied by an increase in cell area (S2A’Fig), this implies that the germband cells must shorten their z-axis if they are to maintain a constant cell volume. The maintenance of a constant cell volume throughout gastrulation appears likely, based on recent measurements [37,38]. We cannot access the Z dimension with our analyses of apical cell surface deformation and so verifying that cells do shorten along their z-axis will require tracking and analyzing cell shape changes in 3-D.
We had shown previously that the AP cell elongation patterns that we are observing in the germband contribute to axis extension [34]. This was shown by measuring strain rates (deformation) for the whole tissue and decomposing these into the strain rates caused by the cell length change and the strain rates caused by polarized cell intercalation [34,35]. We found that although the predominant behavior extending the germband is polarized cell intercalation, AP cell length changes are contributing significantly (about one-third of the total deformation) early in GBE. A question that remains is why AP cell elongation is temporally limited to early GBE, peaking around 10 min after the onset of GBE (Fig 1F and 1F’). In fact, AP cell elongation ceases rather abruptly at around 15 min after GBE onset (Fig 1E’). SPIM movies indicate that this developmental time (taking into account the difference in temperature for the acquisition of these movies, see Materials and Methods) corresponds to when the posterior midgut invagination becomes cup-shaped and appears to drop down from the surface of the embryo (S3 Movie; see schematics in Fig 4A and Fig 7). A possibility is that force generation from endoderm invagination ceases at this time, perhaps because apical constrictions in the primordium are completed. Alternatively, the presence or not of AP cell elongation in the germband could be a function of the balance between how much the actively elongating germband can push and how much endoderm invagination can pull. In other words, early, the pull from endoderm invagination might be stronger than the push from the extending germband, causing a stress in the germband tissue, which manifests as AP cell elongation. Late, the push from GBE versus the pull from endoderm invagination might be balanced: germband cells would not experience stress anymore and would cease to elongate.
In addition to producing cell shape changes contributing to axis extension, does the endoderm-generated tensile force have other roles in axis extension? The posterior pole of the embryo does not move dorsally in fog and tsl mutants and is associated with a buckling of the germband [43]. A possible interpretation of this phenotype is that the actively extending germband cannot intrinsically “push” round the corner (or displace presumptive endoderm). So endoderm invagination may have the role of guiding the germband around the posterior tip to overcome the obstacles posed by the surrounding tissues and the embryo geometry. The tensile stress from the endoderm might also facilitate polarized cell intercalation. Whereas DV-shortening of junctions is known to be caused by the intrinsic activity of the actomyosin cytoskeleton, it remains unclear how the AP-oriented nascent junctions elongate at the end of intercalation [9–19,22,26]. A possibility is that the extrinsic tensile force from the endoderm facilitates this AP junctional elongation either by directly exerting tension on the junctions or by indirectly “making space” for cells to intercalate, or in other words by displacing the boundary ahead of the self-deforming tissue [50]. It is also possible that an AP tensile stress could contribute to the nonreversibility of cell intercalation.
Finally, it is remarkable that three morphogenetic movements principally driven by cell-autonomous behaviours, GBE by polarized cell intercalation, and mesoderm and endoderm invagination by apical constriction, are happening so synchronously (Fig 4B). Furthermore, these movements are controlled by three distinct patterning systems: AP, DV, and terminal, respectively, that are understood to function independently of each other at these early stages [44]. It is unclear how the embryo can synchronize these three movements so precisely. One possibility is that there is a yet-undiscovered genetic cross talk between these pathways. However, our findings suggest an alternative explanation, that mechanical coupling between the invaginations of gastrulation and axis extension helps this synchronization. In vertebrate embryos, convergence and extension movements also happen at the same time as other morphogenetic deformations, for example epiboly [3] or neurulation [5], so understanding how morphogenetic movements interact is going to be important to fully understand how embryos are shaped.
Transgenic strains were spider-GFP, resille-GFP [51], ubi-DE-cad-GFP [52] and sqh-GFP[53]. Mutant alleles were Kr [1], twi [1], tsl [4], the tsl deficiency Df[3R]ED6076 (Flybase and Bloomington Stock Centre), and the acellular mutant characterized in [46].
Anterior movies are taken from [34]. Posterior movies were acquired as follows: late stage five embryos labeled with ubi-DE-cad-GFP were imaged ventrally every 30 sec at 20.5 ± 1°C, using a spinning disc confocal. Cell tracking, cell shape, and cell area analyses were performed as before using custom software (oTracks) written in IDL [34,35]. Best-fit ellipses are used to represent cell shapes and to calculate cell deformation. For statistics, we used a mixed-effects model as before [34].
Late stage five embryos labeled with spider-GFP and/or resille-GFP were mounted in 1.5% low melting point agarose and imaged using mSPIM [39]. Embryos were rotated to image four perpendicular views, which were reconstructed into a whole embryo image stack post-acquisition [54]. Image stacks were acquired every 30 sec at 28–30°C for 60 min. Reconstructed movies of three wild-type and three twi mutants were viewed in 4-D in custom software (Browser and Tracer) written in Interactive Data Language (IDL, Exelis) [55] to map timings of morphogenetic movements. Scatter graphs were plotted in Prism (GraphPad).
PIV was performed to visualize Myosin II flows at the embryo scale and also at a smaller scale to analyze relaxation of the tissue after laser ablation in acellular embryos.
Further details on the Materials and Methods are in S1 Text.
|
10.1371/journal.ppat.1004673 | Manipulating Adenovirus Hexon Hypervariable Loops Dictates Immune Neutralisation and Coagulation Factor X-dependent Cell Interaction In Vitro and In Vivo | Adenoviruses are common pathogens, mostly targeting ocular, gastrointestinal and respiratory cells, but in some cases infection disseminates, presenting in severe clinical outcomes. Upon dissemination and contact with blood, coagulation factor X (FX) interacts directly with the adenovirus type 5 (Ad5) hexon. FX can act as a bridge to bind heparan sulphate proteoglycans, leading to substantial Ad5 hepatocyte uptake. FX “coating” also protects the virus from host IgM and complement-mediated neutralisation. However, the contribution of FX in determining Ad liver transduction whilst simultaneously shielding the virus from immune attack remains unclear. In this study, we demonstrate that the FX protection mechanism is not conserved amongst Ad types, and identify the hexon hypervariable regions (HVR) of Ad5 as the capsid proteins targeted by this host defense pathway. Using genetic and pharmacological approaches, we manipulate Ad5 HVR interactions to interrogate the interplay between viral cell transduction and immune neutralisation. We show that FX and inhibitory serum components can co-compete and virus neutralisation is influenced by both the location and extent of modifications to the Ad5 HVRs. We engineered Ad5-derived HVRs into the rare, native non FX-binding Ad26 to create Ad26.HVR5C. This enabled the virus to interact with FX at high affinity, as quantified by surface plasmon resonance, FX-mediated cell binding and transduction assays. Concomitantly, Ad26.HVR5C was also sensitised to immune attack in the absence of FX, a direct consequence of the engineered HVRs from Ad5. In both immune competent and deficient animals, Ad26.HVR5C hepatic gene transfer was mediated by FX following intravenous delivery. This study gives mechanistic insight into the pivotal role of the Ad5 HVRs in conferring sensitivity to virus neutralisation by IgM and classical complement-mediated attack. Furthermore, through this gain-of-function approach we demonstrate the dual functionality of FX in protecting Ad26.HVR5C against innate immune factors whilst determining liver targeting.
| Adenoviruses are mostly considered self-limiting pathogens associated with respiratory, gastrointestinal and ocular infections; however, in immunocompromised subjects disseminated Ad infection can occur with life-threatening consequences. Many human Ads are capable of binding to coagulation factor X (FX). Following intravenous administration in animal models, FX binds directly to the major Ad capsid protein, the hexon, which subsequently results in virus accumulation in the liver. FX coating Ad5 also acts to shield against immune neutralisation via natural IgM antibodies and the classical complement system. Here we show that FX protection is not a conserved mechanism amongst Ads and identify the Ad5 hexon hypervariable regions (HVR) as the capsid proteins targeted by this host defense pathway. Furthermore, we show that genetic inclusion of Ad5 HVRs onto a native non-FX binder Ad26 to be sufficient to confer sensitivity to immune attack in vitro and in vivo. Using intravenous administration, we determine the significance of FX binding to the Ad5-derived HVRs with respect to defending the virus from neutralisation whilst mediating virus tropism. Our study gives new insight into the role of the viral HVRs and of FX at the interface between virus and host defense mechanisms.
| For the immunocompromised host, human adenoviruses (Ad) have emerged as a significant pathogen capable of exploiting the impaired immunological response and becoming invasive, manifesting in prolonged, severe and life threatening conditions [1–5]. There are seven human species (A-G) of this common non-enveloped, double-stranded DNA virus. Whilst in healthy individuals infections are self-limiting, targeting defined tissues such as the lung, eye and gastrointestinal system over a short time frame, disseminated Ad infections occur when immunity is low (e.g. sufferers of hereditary immune deficiencies, patients undergoing immunosuppressive treatment [1–5]). Systemic infections can culminate in serious and diverse clinical syndromes, ranging from fulminant hepatic failure, coagulopathy, hemorrhagic cystitis, myocarditis, encephalopathy, nephritis to multi-organ failure [2,6]. The immunocompromised patient population is expanding due to increased use of immunosuppressive therapies (e.g. cytotoxic drugs) and consequently Ads are gaining increased recognition as a clinical problem. The presence of a high viral load in the blood is often strongly indicative of a severe outcome [5,6]. Incidence of infection is approximately 2.5–47% in stem cell transplant recipients, whilst paediatric transplantation patients are more prone to the disseminated disease, with mortality rates reaching up to 70% [5,7,8]. Despite the high risk, there is no FDA approved drug specific to treating Ad infection and therapeutic options can be limited. Advancing our knowledge of the complex mechanisms underlying Ad5 infection in vivo is of great importance.
Using species C Ad2/5 as the prototype, the in vitro infection pathway has been very well documented. Studies have finely detailed the individual steps from virus binding via the fiber knob protein to the primary cell surface coxsackie and adenovirus receptor (CAR) [9], engagement of the Ad penton base with αvβ3/5 integrins leading to internalization [10] and subsequent trafficking from endosomes to nuclear import [11,12]. However, the lack of suitable animal models, which allow viral replication and closely mimic the human immune system, has challenged the study of Ad infection in vivo. Nevertheless an abundance of valuable information has been gained about viral spread, immune responses and methods to combat such, especially from its popularity as a viral vector for gene therapy, vaccination and virotherapy protocols. Of the 60+ human Ads identified [13], Ad5 is the virus-based gene transfer vector most frequently employed. Concomitantly, Ad5 is also one of the most seroprevalent of the family. The use of Ad5 as a viral vector has deepened our understanding of virus:host binding events, involvement of innate and adaptive immunity and of the factors leading to the substantial accumulation of Ad5 particles in the liver following bolus injection into the bloodstream. When administered intravenously (I.V.) the virus rapidly encounters a multitude of interactions with circulating blood components. These include virion neutralisation by pre-existing antibodies [14], sequestration by Kupffer cells [15], MARCO+-expressing splenic macrophages [16], polymorphonuclear leukocytes [17], natural IgM and complement opsonisation [18,19]. Binding to platelets [20,21], erythrocytes [22,23], and blood coagulation factors [24–26] all contribute to the substantial interplay between the virus and host. Dissecting the precise interactions which occur in vivo is key to our understanding of the virus infection pathways partnered with an individual’s defence mechanisms.
Previous work suggests that Ads belonging to species C (Ad1, Ad2 and Ad5) are more commonly associated with disseminated disease than other types and have been implicated in severe hepatic failure [5,8]. Hepatitis is a frequent and serious consequence of systemic Ad infections [27–29]. Coagulation factor X (FX) plays a fundamental role in determining the characteristic hepatic tropism of Ad5 [24–26]. Selectively blocking FX prevents Ad liver transduction in rodent and non-human primates following I.V delivery of virus [30–32]. FX binds with nanomolar affinity to the Ad5 hexon hypervariable regions (HVR), and acts as a bridge to attach the virus to N and O-linked heparan sulphate proteoglycans (HSPG) on the surface of hepatocytes [24,33,34]. Crystallographic and cryogenic electron microscopy identified contact points within and around Ad5 HVR5 and HVR7 which are responsible for interacting with the FX Gla (γ-carboxylated glutamic acid) domain [32]. Genetically swapping regions or specific amino acids within the Ad5 HVR5 and HVR7 for those of a non-FX-binding Ad (e.g. species D Ad48 or Ad26) has proven an effective strategy to abrogate the FX interaction and diminish liver transduction [32,35]. In addition to binding the FX Gla domain at the HVR7 amino acid motif T423-E424-T425, Ad5 is also capable of interacting with coagulation factor VII (FVII) at these points [36]. However, FVII does not support Ad5 transduction as it binds to the virus in an alternate orientation to FX, with dimerization of the FVII serine protease domains disguising potential HSPG receptor binding sites [36]. Further to its influence on liver transduction, Doronin et al. showed that FX coating the virus triggers recognition by the innate immune system via nuclear factor-κB (NFκB) activation and subsequent TLR4/TRAF6/NFκB-mediated inflammation, an effect absent when Ad5 was genetically manipulated to be devoid of FX binding [37].
Recent work by Xu et al. has produced additional insight into the function of FX in adenovirus biology [19]. They indicated a role for FX in protecting Ad5 from attack by natural IgM antibodies and the classical complement system upon exposure to murine blood [19]. In co-operation with IgM, complement activation acts as an innate host defense mechanism and has previously been shown to result in neutralisation of invading pathogens including Ads [18,19,38]. In vitro data demonstrated that FX can prevent Ad5 from this neutralisation when incubated with murine serum [19]. In contrast to studies in wild-type mice, the Ad5:FX interaction was not essential for liver transduction in mice deficient in natural antibodies or the complement components C1q and C4 [19]. Instead FX binding to Ad5 acted as a protective "shield", decorating the viral capsid and preventing natural IgM and classical complement mediated inhibition of Ad gene transfer [19]. This study has led to some speculation surrounding the impact of FX in determining Ad liver tropism.
It is evident that blood components influence Ad tropism, whilst other interactions (e.g. Kupffer cell uptake) remain dominant barriers to widespread Ad dissemination. Here, we studied the binding events and mechanisms deciding the fate of the virus in circulation. We attempted to dissect the importance of these interactions in determining viral cellular uptake and tropism. In this study we used genetically mutated Ad vectors to identify key hexon regions responsible for IgM and complement-mediated attack. The Ad5 HVRs were identified as the critical viral capsid components. We then incorporated these regions and FX binding capability onto a non-FX-binding Ad26 background. We utilised this novel vector to investigate the significance of FX and the role of the Ad5 HVRs in the interplay between viral immune recognition and tropism in vivo.
FX coating Ad5 shields the virus against IgM and classical complement-mediated immune attack [19]. Many other Ad types also bind to human FX (hFX) [24]. However it is unknown whether they are sensitive to neutralisation via the same pathway. Therefore, we compared the sensitivity of a selected panel of Ad vectors based on different species/types; the FX-binding Ad5, Ad35 and Ad50, and non-binding Ad48 and Ad26 [24] to neutralisation by murine serum in vitro. When Ad5 was incubated with C57BL/6 serum there was a significant increase in transduction compared to the media control, likely due to the presence of native FX in the murine serum (Fig. 1A). However when serum was pretreated with X-bp (binds the FX Gla domain blocking Ad:FX interaction [24]) Ad5 gene transfer was dramatically reduced, to levels significantly below control conditions, consistent with previous results [19] (Fig. 1A). This demonstrates that in the absence of FX Ad5 is sensitive to neutralisation by murine serum components [19]. In contrast, Ad35, Ad50, Ad48 and Ad26-mediated cell transduction was not affected by serum regardless of the presence of X-bp (Fig. 1A), indicating that these vectors are not sensitive to the same mechanism that mediates Ad5 neutralisation. It is noteworthy that the overall charge in the region of the Ad5 hexon hypervariable loops is more negative than that of Ad26/Ad35/Ad48/Ad50 (S1 Table). Protection from neutralisation by FX is also evident for Ad5 in rat serum (S1 Fig.).
Next we investigated the role of different capsid proteins in mediating Ad5 neutralisation. For this we employed a range of Ad5-based chimeric vectors (S2 Table). We found no role for the fiber or penton base, as swapping those of Ad5 for those of Ad35 (i.e. the Ad5.F35 and Ad5.F35P35 vectors), still resulted in vector neutralisation in the absence of FX (Fig. 1B). Next we assessed the contribution of the Ad5 hexon in enabling neutralisation. Here we utilised Ad5.HVR48(1–7), which we previously showed to lack FX binding [24] (S2 Table). Ad5.HVR48(1–7) was not inhibited by serum components regardless of FX, thus illustrating an essential requirement for the Ad5 hexon HVRs in mediating neutralisation via this mechanism (Fig. 1C). In addition, unlike the parental Ad5, the Ad5.HVR48(1–7) chimeric vector caused no induction of C3 activation in serum preincubated with X-bp (Fig. 1D). Hence, Ad5.HVR48(1–7) cannot bind to FX [24] and is not sensitive to neutralisation. Therefore these data indicate a pivotal role for the Ad5 HVRs in enabling immune attack via the IgM/classical complement pathway.
To investigate the contribution of different Ad5 HVRs in mediating neutralisation by mouse serum, we evaluated the sensitivity of Ad5 vectors containing a range of Ad26 HVR modifications. Ad26 was chosen for these studies as it does not bind FX [24] and is not susceptible to serum neutralisation in vitro (Fig. 1A). This series of Ad5/26 chimeric vectors (Ad5.HVR5(Ad26), Ad5.HVR7(Ad26) and Ad5.HVR5+7(Ad26)) (S2 Table) were previously shown to lack FX binding [32]. Unlike the Ad5 control, the Ad5 vectors with HVR(Ad26) swaps were all sensitive to inhibition by serum (Fig. 2A), suggesting the involvement of multiple HVRs in enabling neutralisation. Next, we employed a series of Ad5 vectors engineered with individual point mutations to alter specific amino acids to those found in Ad26 (S2 Table). These vectors were previously demonstrated to have reduced but not eliminated (e.g. Ad5.HVR5* (HVR5 mutations T270P and E271G) or abolished FX binding (e.g. Ad5.HVR7* (HVR7 mutations I421G, T423N, E424S, L426Y) and Ad5.HVR5*.HVR7*.E451Q (hereafter referred to as Ad5T*)) [32]. We evaluated vector sensitivity to neutralisation over a range of concentrations of C57BL/6 mouse serum (1–90% final volume) in vitro (Fig. 2B, S2 Fig.). In contrast to the media control, transduction of all FX-binding deficient Ad5 vectors was dramatically reduced in the presence of serum (Fig. 2B, S2 Fig.). This inhibitory effect was significantly lessened at lower serum concentrations (<25%) for the vectors engineered with point mutations, Ad5.E451Q, Ad5.HVR5*, Ad5.HVR7* and Ad5T* compared to vectors with entire HVR exchanges Ad5.HVR5+7(Ad26) (Fig. 2B). The latter vector remained highly sensitive to neutralisation even in the presence of 5% serum. This suggests that the sensitivity to serum is influenced by the location and extent of modifications to the Ad5 HVRs, and is, at least, partially independent of the complete loss of FX-binding capacity.
It has previously been shown that Ad5.HVR5* is capable of direct interaction with FX, albeit at lower levels than Ad5, however detailed affinity kinetics are not available [32]. We hypothesized that immune components may compete with FX for binding sites within related exposed Ad5 HVR loops. Notably we found the concentration of hFX in human serum to be ∼50% lower than that of normal plasma (S3 Table). To test whether increasing concentrations of FX could protect Ad5.HVR5* from immune attack, we spiked C57BL/6 mouse serum with 10 μg/mL hFX and examined the susceptibility of the non-FX-binding Ad5T* control vector and Ad5.HVR5* to neutralisation (Fig. 2C). As expected the presence of hFX did not affect the neutralisation of Ad5T* by serum and mediated no increase in gene transfer (Fig. 2C). However, hFX prevented neutralisation of Ad5.HVR5*, particularly in lower concentrations of serum (Fig. 2C). The presence of hFX also increased gene transfer of Ad5.HVR5* indicating its ability to both protect and act as a bridge to HSPGs despite sub-optimal FX:hexon binding conditions [24]. These data indicate that both hFX and the murine serum components can compete with one another, likely through binding to similar HVRs, and this is dependent on their relative concentrations and/or affinities.
To further dissect the role of the Ad5 HVRs and FX in protecting Ad from serum neutralisation, we engineered FX binding capacity into Ad26 by substituting the Ad5 HVRs into the Ad26 hexon. We attempted to generate a number of Ad26-based mutants however only three Ad26 chimeras were successfully packaged into mature viral vectors at high titer (Fig. 3A, S3 Fig., S4 Table). Specific hexon sequences were chosen for mutagenesis (S3 Fig.). These included the point mutant Ad26.Q461E (the corresponding Glu residue is conserved in all FX binding Ads) and the HVR exchanges Ad26.HVR5(Ad5).Q461E and Ad26.HVR5C [in which Ad26.HVR(1–3 and 5–7) were replaced by those of Ad5], respectively (Fig. 3A, S4 Table, S3 Fig.). To note, due to the genetic capsid modifications the vp:PFU ratio was ∼10 fold higher for both Ad26.HVR5C and Ad26.HVR5(Ad5).Q461E compared to the parental Ad26 vector (Fig. 3A). We next measured the ability of each virus to bind hFX by SPR. Ad26.HVR5C showed efficient binding when injected over a hFX biosensor chip, as did the positive control Ad5, while Ad26, Ad26.Q461E and Ad26.HVR5(Ad5).Q461E did not bind hFX (Fig. 3B). We then quantified affinity kinetics (Fig. 3C). The calculated association rate constant (ka) and dissociation rate constant (kd) values of the hFX for immobilized Ad26.HVR5C were 3.085x106 (1/Ms) and 1.068x10–2 (1/s), giving an overall equilibrium dissociation constant (KD) of 3.462x10–9 M. When Ad5 was immobilized and hFX was injected across the biosensor, ka and kd values were 1.308x106 (1/Ms) and 3.053x10–3 (1/s), giving a KD of 2.334x10–9 M (Fig. 3C), consistent with previously reported kinetics [24]. Therefore, incorporation of Ad5 HVR(1–3 and 5–7) into Ad26 generates de novo binding to FX by Ad26 at an affinity similar to Ad5.
Ad26.HVR5C provides a novel virus through which to ascertain whether inclusion of Ad5 HVR(1–3, 5–7) are sufficient for exposure to neutralisation and whether FX binding influenced the sensitivity of the parental virus to mouse serum. Therefore we assessed cellular transduction by the Ad26 mutants in the absence and presence of C57BL/6 or RAG2-/- murine sera +/- X-bp. RAG2-/- mice are immune-deficient, lacking mature T and B lymphocytes, closely related to the RAG1-/- strain in which Ad5 liver tropism was previously demonstrated to be FX independent [19]. Parental Ad26 was resistant to neutralization by both strains of serum regardless of FX, again demonstrating Ad26 is not sensitive to the same mechanism that mediates Ad5 neutralisation (Fig. 4A). Interestingly, Ad26.HVR5C was resistant to neutralization in the presence of FX but sensitive to neutralization when immune-competent C57BL/6 serum was pre-incubated with X-bp, a similar profile to that seen with Ad5 (Fig. 4A). RAG2-/- serum did not neutralize Ad26.HVR5C or Ad5 (Fig. 4A), indicating a requirement for efficient T and/or B cell antibody function [19]. Ad26.HVR5(Ad5).Q461E and Ad26.Q461E were unaffected by either sera regardless of the presence of FX. Furthermore, whilst there were differences amongst the basal levels of C3a between vectors, both Ad5 and Ad26.HVR5C, but not Ad26 enhanced C3a in a FX-dependent manner. Notably, the FX protection mechanism is not dependent on the heparin binding exosite in the FX serine protease domain, as blocking these HSPG interacting sites did not alter Ad induced C3a levels compared to the matched serum only control (Fig. 4C). Hence, inclusion of the Ad5 HVRs in Ad26.HVR5C not only leads to FX binding, but also sensitises the virus to attack in immune-competent mouse serum indicating the importance of these hexon regions in both FX binding and immune recognition.
We next investigated whether creating Ads engineered to bind FX influenced their ability to interact with cells in vitro and in vivo. We performed cell binding and transduction assays with Ad26.HVR5C to assess vector:cell interaction profiles (Fig. 5). SKOV3 cells were employed for these assays as they express low levels of CAR [39] and allow focus on the FX-mediated pathway. Both cell binding and transduction for Ad26.HVR5C were significantly increased in the presence of FX compared to the parental Ad26 which was unaffected by FX (Fig. 5A-B). This suggests that Ad26.HVR5C functionally binds FX leading to Ad:FX engagement with cellular HSPGs and subsequent gene transfer. Next, we examined whether the Ad26.HVR5C hexon:FX interaction generates hepatic tropism via FX bridging to hepatocytes [24,33,34]. Immune competent MF1 control mice were first treated with warfarin to deplete vitamin K-dependent coagulation factors, and then administered I.V. with 1011 vp of Ad5, Ad26 or Ad26.HVR5C.luc+. Luciferase expression was visualised by whole-body bioluminescence imaging and quantified at 72 h. Ad26 produced low level, widely biodistributed gene transfer and this was unaltered by warfarin (Fig. 6A). However, both Ad5 and Ad26.HVR5C produced selective transduction of the liver in non-warfarin treated mice (Fig. 6A-C) albeit the levels mediated by Ad26.HVR5C were significantly lower than those of Ad5. Despite the vp:PFU ratio of Ad26.HVR5C being ∼10 fold lower than Ad26 (Fig. 3A), there was a significant increase in liver luciferase levels for the Ad26.HVR5C vector compared to Ad26, in non-warfarin treated animals (Fig. 6C). For both Ad5 and Ad26.HVR5C, liver transduction was reduced to less than 0.5% of control following warfarin treatment. Quantification of viral genome accumulation in the livers of Ad5 and Ad26.HVR5C injected animals revealed a similar pattern of virus-mediated gene transfer (Fig. 6D). Thus, inclusion of FX binding into Ad26.HVR5C leads to profound retargeting effects following I.V. injection into MF1 mice.
We then performed the same experiment in immune-deficient RAG2-/- mice. Ad5 mediated FX-independent liver transduction in RAG2-/- mice, similar to that previously reported in RAG1-/- mice [19] (Fig. 7A-D). Ad26 demonstrated widespread gene expression in vivo which was equivalent in both non-warfarin and warfarin-treated mice (Fig. 7A). However, Ad26.HVR5C transduction was focused in the liver in non-warfarin treated RAG2-/- mice, whilst hepatic gene expression was dramatically decreased in the absence of FX (Fig. 7A-C). The transduction profile of Ad26.HVR5C in warfarin-treated mice was completely altered, no longer targeting the liver but instead exhibiting a more widespread biodistribution similar to the parental Ad26 (Fig. 7A). This demonstrates the ability of the hexon:FX interaction to determine Ad26.HVR5C hepatocyte uptake. Through this gain-of-function approach, incorporation of FX binding into Ad26, FX was shown to effectively alter vector tropism in vivo and dictate liver targeting of Ad26.HVR5C in control and immune-deficient mice.
The high levels of gene expression by Ad5 in warfarin treated RAG2-/- mice (Fig. 7A) is in contrast to the diminished luciferase expression (Fig. 6A) observed in warfarin treated MF1 mice. This is due to IgM and classical complement-mediated vector neutralisation, similar to that demonstrated previously [19]. Unlike Ad26 but in parallel to Ad5, gene expression by Ad26.HVR5C was also inhibited in warfarin treated MF1 mice but not warfarin treated RAG2-/- mice (Fig. 6A, 7A). This suggests Ad26.HVR5C was neutralised in immune competent mice as a direct consequence of the engineered Ad5 HVRs. This indicates that in the immune competent setting, in the absence of a FX protective mechanism, engineering the Ad5 HVRs into Ad26 confers sensitivity to immune attack and vector neutralisation in vivo.
Here, the role of the Ad5 HVRs and the significance of FX in defending the virus from host attack whilst determining liver targeting are described. We first identified the Ad5 HVRs as responsible for conferring sensitivity to serum neutralisation. We then genetically engineered the Ad5 hexon loops, onto a non-FX-binding Ad26 background to generate the novel vector Ad26.HVR5C. Employing this as an efficient tool, we deciphered the importance of Ad5 HVRs and FX in immune recognition, protection and biodistribution following systemic Ad administration.
Previous work reported that coagulation FX coats Ad5 to protect from IgM and classical complement-mediated immune neutralisation and that Ad5 liver uptake was not solely dependent on FX under immune-deficient conditions [19]. However the capsid proteins to which IgM and the classical complement components bind and initiate this cascade eventually leading to virus inactivation have not been identified. In addition to this work, Doronin et al., indicated that displacement of FX from the coagulation system through binding to Ad5 is sensed by the host and an inflammatory response is initiated [37]. Therefore, with both of these recent studies in mind, we attempted to identify the Ad capsid regions responsible for mediating immune neutralisation and further decipher the functional consequences of the Ad:FX interaction. We began by showing the FX protection mechanism was not conserved amongst human Ads. Ad5 was the only type tested susceptible to serum neutralisation in the absence of FX and using a series of genetic and pharmacological approaches this was shown to be reliant on the presence of the Ad5 HVRs. We then investigated the sensitivity of Ad5-based HVR mutants and FX-deficient vectors to neutralisation by serum from immune-competent mice and found that all viruses were inhibited, but interestingly there were differences amongst these inactivation levels. These variations indicate that it is more than simply the loss of FX binding or these contact points which likely dictates immune recognition. For our experiments we used fresh mouse serum. Unlike citrated or EDTA treated plasma there is no interference with calcium, which has previously been found to influence results [18]. During the coagulation process, a proportion of the coagulation factors will be activated and levels of intrinsic FX reduced. When we investigated the effects of spiking mouse serum with hFX and compared the null-binder Ad5T* to the FX-defective binder Ad5.HVR5*, we found hFX was capable of rescuing Ad5.HVR5* from neutralisation and further increasing gene transfer to levels above the media control. These data thereby suggests that hFX and the inhibitory murine serum components compete for binding within similar regions within the Ad5 hexon. Interestingly previous data has demonstrated using targeted PEGylation that similar Ad5 HVRs (1, 2, 5 and 7) are also responsible for interacting with scavenger receptors and subsequent Kupffer cell uptake, but in that case it has been suggested that FX and the scavenger receptors do not have overlapping binding sites [40].
Our work demonstrates a key role for the Ad5 HVRs in determining virus sensitivity to serum. We successfully engineered an Ad26 vector with Ad5-derived HVRs, Ad26.HVR5C, capable of FX-mediated cell binding and transduction in vitro. The species D Ad26 is a less seroprevalent virus than Ad5 [14]. It has been shown to utilise both CD46 [41] and CAR [42] as primary cellular receptors and is currently gaining a lot of attention as a vector-based vaccine against HIV [43]. Interestingly, the Ad5 HVR(1–3, 5–7) exchange in this Ad26-based vector suddenly made the virus susceptible to neutralisation by mouse serum, indicating the driving influence of the HVRs and the likelihood they possess a critical IgM recognising epitope. The smaller regions of exchange in Ad26.HVR5(Ad5).Q461E or the single point mutation in Ad26.Q461E were not sufficient to induce sensitivity to neutralisation. Conversely, incorporating the Ad26 HVR5+7 into Ad5 resulted in a vector (Ad5.HVR5+7(Ad26)) that remained susceptible to the inhibitory effects of serum. This indicates it is perhaps likely that multiple HVR loops working in conjunction which are responsible for mediating the crucial serum binding event. It will require further study to fine map the essential contact sites. IgM antibodies are broadly reactive, capable of binding to unrelated structures with low affinity [44] and mostly exist as a pentameric structure, containing ten potential antigen binding sites [45]. Therefore it is possible one IgM molecule may be capable of interacting across a wide hexon region. These natural antibodies are often directed against highly repetitive and charged motifs [46,47]. A recent study suggested the involvement of the electronegative potential of antigens in IgM recognition [47]. To note, the panel of recombinant Ad5-based FX-binding deficient viruses, with the point mutations were all modulated to have lower net negative charge, whilst the more sensitive entire HVR exchanges in Ad5.HVR5+7(Ad26) had more negative charge compared to the parental vector (S1 Table). In addition it is perhaps interesting that the net charge of the Ad5 hexon protein is more negative than the other vectors tested in this study, Ad26/Ad35/Ad48/Ad50. The HVRs of Ad types tested here differ significantly and charge-specific mutation analysis is required to further address this potential issue.
Engineering FX-binding into the Ad26 vector had profound effects on viral biodistribution in in vivo models of viral dissemination. In complete contrast to its parental virus, Ad26.HVR5C exhibited selective liver gene transfer in the presence of FX in immune-competent and deficient mice. In the absence of FX and immunity, Ad26.HVR5C reverted to its native tropism, with the same biodistribution pattern as Ad26. This demonstrates the tropism defining force instilled by the inclusion of FX binding. The interaction with FX instils new tropism in addition to its native Ad26 receptor usage (e.g. via CAR/CD46) [41,42]. It is evident from the work presented here, and from others [19], that Ad5 remains hepatotropic in warfarin treated immune-deficient mice. Therefore this suggests, that an alternative mechanism is utilised by Ad5 to transduce hepatocytes of immune deficient animals when FX binding is diminished. Further investigation is required to answer the important question of what is the precise mechanism determining Ad5 liver targeting in the absence of FX.
The relevance of the Ad:FX interactions and protection pathways in humans suffering from the disseminated disease is yet to be examined. Further study is therefore required to investigate whether human natural IgM can bind to Ads and if this initiates similar pathways as those in the mouse i.e. classical complement mediated neutralisation in the absence of FX [19]. It may also be beneficial to directly compare and interrogate the influences and interplay of complement, natural and acquired human immunoglobulins [48]. This may help to better predict the human immune defense to widespread Ad infection or when using Ads as gene therapy vectors. In the general population the majority of specific neutralising antibodies (e.g. IgG-based) are directed against the Ad5 hexon protein and particularly focused on the Ad5 HVRs [49,50]. Previous work has demonstrated that such neutralising human sera can prevent FX-mediated cellular binding and transduction in vitro [51]. In addition, in MF-1 mice preinjected with human sera, higher levels of pre-existing neutralising antibodies, correlated with decreased hepatocyte transduction in the presence of FX [51]. It would be interesting to determine whether different families of human immunoglobulins (IgG/IgM/IgA) can compete for binding sites within the Ad hexon HVRs, the relative affinities of these interactions and, further, whether FX coating of the hexon affects these processes. This would also be of clinical relevance in terms of gene therapy and for determining patient suitability or the type of Ad vector employed (FX-binding or non-FX-binding vectors) for intravascular Ad administration. Investigating the relevance of the FX protection mechanism from rodent to human models is an important next step for the field. Evidently, the complexity of Ad interactions with host factors is extensive and small animal models may only provide limited information. For instance, CAR-mediated Ad5 binding to human erythrocytes is an important clinical consideration however mouse erythrocytes are devoid of CAR expression, again highlighting the necessity for using human samples [22,52,53].
Despite the high morbidity and mortality rate associated with disseminated Ad infection there are currently no licensed specific anti-adenoviral drugs available for the management of the condition. Those that have been used in clinical settings, such as ribavirin, cidofovir, and vidarabiner, have yielded varied results and treatment options remain largely unsatisfactory [54,55]. A previous study identified small molecule pharmacological agents to block FX-mediated Ad5 infection following intravenous administration in a mouse model, and found it to be an effective strategy by which to prevent hepatic targeting [56]. However, in that case the pharmacological inhibitor was acting post the Ad5:FX binding event [56]. Doronin et al. used cryoEM and molecular dynamics flexible fitting simulations techniques to reveal an interaction between the FX Gla K10 residue and the Ad5 HVR7 residues E424 and T425 [37]. From our work it is evident that the host natural IgM and classical complement mediated defence mechanism is effectively governed via the Ad5 HVR loops, immune recognition is dependent on more than solely the loss of FX contact points, and FX functions in mediating liver uptake in both immune competent and deficient models. Identification of a small molecule inhibitor with favourable pharmacological properties capable of specifically blocking the Ad5:FX interaction at these contact points may enable a more robust host anti-viral immune response whilst limiting liver infection and thus be very valuable in treating systemic disease.
In summary, here we have identified the Ad5 HVRs as the key regions conferring sensitivity to the IgM and complement mediated host defence pathway. FX has both a pivotal role in protecting the virus from attack and has profound effects on Ad biodistribution. Through this work we gain a greater insight into the mechanism and significance of FX in protecting Ads against innate immunity and determining virus tropism.
All animal procedures were approved by the University of Glasgow Animal Procedures and Ethics Committee and performed under UK Home Office license PPL 60/4429 in strict accordance with UK Home Office guidelines. All efforts were made to minimize suffering.
HEK293 (human embryonic kidney: ATCC CRL-1573) and HeLa (human cervical adenocarcinoma: ATCC CCL-2) cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Invitrogen, Paisley, UK) and SKOV3 (human ovarian carcinoma: ATCC HTB-77) and A549 (human lung carcinoma: ATCC CCL-185) cells in RPMI-1640 medium (Invitrogen), with 2 mM L-glutamine (Invitrogen), 10% fetal calf serum (FCS; PAA Laboratories) and 1 mM sodium pyruvate (Sigma-Aldrich, UK) at 37°C 5% CO2. PER.C6/55K cells [57] were cultured in DMEM with 2 mM L-glutamine, 10% FCS, 1 mM sodium pyruvate and 10 mM MgCl2, at 37°C 10% CO2.
E1/E3 deleted Ad5, Ad35, Ad5/35 chimeras (Ad5.F35, Ad5.F35P35), Ad26, Ad50, Ad48 and Ad5/48 chimeric (Ad5.HVR48(1–7)) vectors encoding CMV-luciferase reporter genes were generated as described previously [14,24,58]. S2 and S4 Tables provide detail on each of the mutated vectors used in this study. E1/E3 deleted Ad5 and FX-binding deficient Ad5/Ad26 chimeric vectors (Ad5.HVR5+7(Ad26), Ad5.HVR5*, Ad5HVR7*, Ad5E451Q and Ad5T*) encoding CMV-lacz reporter genes were generated as described previously [32]. Hexon gene-modified Ad26 vector genomes were constructed in the context of pAd26.luc, a plasmid that contains a PacI site-flanked, full-length Ad26 vector genome with E1, E3, and E4 deletions/modifications [14]. The vector genome is further equipped, at the site of E1 deletion, with a CMV promoter-driven expression cassette for firefly luciferase. To generate the desired hexon gene-modified pAd26.luc plasmids, the concerning Ad26/Ad5 hexon modifications (S4 Table) were first made in the context of a smaller ‘hexon shuttle’ plasmid (gene synthesis and subcloning by GeneArt/LifeTechnologies), and then shuttled into a hexon gene-deleted derivative of pAd26.luc by homologous recombination in E.coli BJ5183 (Stratagene/Agilent Technologies), as described previously [32]. S3 and S4 Figs. further describe construction and Ad26/Ad5 sequence alignment, highlighting the regions targeted for mutagenesis.
Linearised Ad plasmids were transfected in PER.C6/55K using Lipofectamine 2000 (Invitrogen). Cells were harvested 10–14 days post-transfection. Viral particles (vp) were propagated and purified by CsCl gradients. Titers were determining by protein concentrations using micro-bicinchoninic acid (BCA) Protein Assay (Thermo Scientific). Titer calculations used the formula 1 μg protein = 4x109 vp and end-point dilution assays using PER.C6/55K for quantification of plaque forming units (pfu)/mL [59]. Purified Ads were analysed by SDS-PAGE followed by silver staining (Sigma-Aldrich) according to manufacturer’s instructions in order to verify the capsid composition and confirm the vector modifications did not interfere with the structural integrity of the particles.
Immobilized hFX: Performed using Biacore 2000 (GE Healthcare) as described [31]. Purified hFX was purchased from Cambridge Biosciences (Cambridge, UK). hFX was covalently immobilized onto the flowcell of a CM5 biosensor chip by amine coupling. Immobilized Ad: Performed using T200 (GE Healthcare). Virus was biotinylated using the EZ-link sulfo-NHS-LC biotinylation kit (ThermoFisher). The biotinylated products were coupled to streptavidin-coated sensorchips (SA; Biacore); Ad5 (482RU), Ad26.HVR5C (484RU). SPR was performed in 10 mM HEPES (pH7.4) 150 mM NaCl, 5 mM CaCl2, 0.005% Tween20 at a flow rate of 30 μL/min and sensorchips were regenerated by injection of 10 mM HEPES (pH 7.4) 150 mM NaCL, 3 mM EDTA, 0.005% Tween20. Kinetic analysis was performed using 2-fold serial dilutions (in duplicate, starting with 30 μg/mL) of hFX and fitted using a 1:1 binding model (Biacore Evaluation software, Biacore).
Cells were plated in 24-well formats (2x105 cells/well) and incubated overnight at 37˚C. Cells were cooled (4°C) for 30 min, washed with phosphate buffered saline (PBS) before adding 1000 or 5000 vp/cell Ad -/+ hFX. hFX was used at 10 μg/ml. Cells were incubated for 1 h at 4°C, washed with PBS, and harvested. DNA was extracted from cells using the QIAamp DNA mini kit (Qiagen) and quantified using Nanodrop (ThermoScientific). Viral genomes (200 ng DNA) were quantified by quantitative polymerase chain reaction (PCR) analysis (7900HT Sequence Detection System; Applied Biosystems) using Power SYBR Green PCR mastermix and CMV primers (Applied Biosystems).
Cells were plated in 96-well formats (1x104 cells/well) and incubated overnight at 37˚C. Cells were infected with 1000 or 5000 vp/cell Ad -/+ hFX. Cells were incubated for 3 h at 37°C, washed with PBS, maintained with medium and harvested 48 h post-transfection. Luciferase activity was measured using the luciferase assay (Promega, Southhampton, UK). Protein concentrations were calculated by BCA. Values expressed as relative light units (RLU)/mg of protein.
8–9 week old male MF1 outbred mice (Harlan, UK) and Rag2 knockout mice (on a C57BL/6 genetic background) (kind gift from Dr Alison Michie, University of Glasgow) were used. Mice were warfarin-treated (133 μg/mouse) prior to virus administration as previously described [24]. 1011 vp of Ad in 100 μL PBS were administered via tail vein injection. 72 h post-injection IVIS imaging was performed and animals were sacrificed. Livers were harvested and luciferase activity measured. DNA was extracted from livers using the QIAamp DNA mini kit. Viral genomes in 200 ng DNA were quantified by qPCR as above.
Cells were plated in 96-well formats (1x104 cells/well) and incubated overnight at 37˚C. Fresh serum from C57BL/6 mice, Rag2-/- mice or Wistar rats was separated from whole blood, diluted in RPMI-1640 and incubated with 2x1010 vp/mL Ad in a final volume of 50 μL. 40 μg/mL X-bp was added to samples to block FX. Where specified serum was spiked with 10 μg/mL hFX prior to Ad addition. Controls were vectors alone in serum-free medium. Vectors were incubated with serum or medium for 30 min at 37°C. Mixtures were diluted 200-fold in serum-free medium. 1000 vp/well was added for 2 h at 37°C, then replaced with medium containing 2% FBS. After ∼16 h, cells were harvested for determination of transgene expression and protein content.
Serum was separated from fresh murine blood. Virus (5x1010 vp/mL) was incubated with 50 μL of serum-/+ 40 μg/mL Xbp or 20 μg/mL NapC2 (a kind gift from Dr G. Vlasuk (Corvas International, San Diego, CA, USA)) for 90 min at 37°C, then 10 mM EDTA was added. Samples were frozen at -80°C until evaluation in a mouse C3a ELISA as previously described [18]. The capture antibody for ELISA was rat anti-mouse C3a (BD Pharmingen #558250) and the detection antibody used was biotin rat anti-mouse C3a (BD Pharmingen #558251).
The levels of hFX in human serum and plasma samples were measured using a Matched-pair antibody for ELISA of human Factor X FX-EIA (Quadratech Diagnostics, Surrey, UK) according to the manufacturer’s instructions. Purified hFX (Cambridge Bioscience) was used as the control.
Statistical significance was calculated using one-way ANOVA followed by Bonferroni post-hoc test with GraphPad Prism. In vitro results presented are representative data from three separate experiments with at least 3 experimental replicates per group. Each in vivo experiment was performed with a minimum of 3 animals per group (n = 4–6 for Ad treated groups, n = 3 for PBS groups). All error bars represent SEM.
|
10.1371/journal.pcbi.1007253 | Variation in plastic responses to light results from selection in different competitive environments—A game theoretical approach using virtual plants | Phenotypic plasticity is a vital strategy for plants to deal with changing conditions by inducing phenotypes favourable in different environments. Understanding how natural selection acts on variation in phenotypic plasticity in plants is therefore a central question in ecology, but is often ignored in modelling studies. Here we present a new modelling approach that allows for the analysis of selection for variation in phenotypic plasticity as a response strategy. We assess selection for shade avoidance strategies of Arabidopsis thaliana in response to future neighbour shading signalled through a decrease in red:far-red (R:FR) ratio. For this, we used a spatially explicit 3D virtual plant model that simulates individual Arabidopsis plants competing for light in different planting densities. Plant structure and growth were determined by the organ-specific interactions with the light environment created by the vegetation structure itself. Shade avoidance plastic responses were defined by a plastic response curve relating petiole elongation and lamina growth to R:FR perceived locally. Different plasticity strategies were represented by different shapes of the response curve that expressed different levels of R:FR sensitivity. Our analyses show that the shape of the selected shade avoidance strategy varies with planting density. At higher planting densities, more sensitive response curves are selected for than at lower densities. In addition, the balance between lamina and petiole responses influences the sensitivity of the response curves selected for. Combining computational virtual plant modelling with a game theoretical analysis represents a new step towards analysing how natural selection could have acted upon variation in shade avoidance as a response strategy, which can be linked to genetic variation and underlying physiological processes.
| Plants are able to respond to changes in the environment. Particularly, plants can show different structural traits (e.g. stem height and leaf size) in different planting densities. These trait changes are the result of so-called plastic responses that can be induced by changes in the light spectrum. Although a great part of the physiological processes underlying these plastic responses have been identified, it remains unclear how these plastic responses, and variations therein, can be the result of selection. In this paper we analyse selection on different plastic responses within different dynamic competitive environments. We use a 3D virtual plant model that simulates realistic plant growth based on light absorption, photosynthesis and specific light signals that induce changes in leaf growth. The 3D model simulated various competitive vegetation stands consisting of plants with different plastic responses at different planting densities. We conclude that selection in different densities results in different plastic responses. We advocate that our modelling approach allows for analyses related to selection on plastic responses itself, instead of on specific trait values in different environments, and is therefore an important new step forward in understanding the role of plastic responses for plant performance.
| In the course of evolution, plants have evolved traits often specific to a certain environment. When growth conditions change, the selection pressure on trait values change, which subsequently can change selection on genotypes. However, due to phenotypic plasticity one genotype can exhibit multiple phenotypes (i.e. multiple trait values) depending on environmental conditions [1–3], which helps a plant to survive across different environments. The extent to which plasticity is adaptive depends on the environmental conditions, the reliability of the cues that signal the (change in) environmental conditions, and the costs related to phenotypic changes and plasticity itself [4–7]. Although considerable genetic variation in plasticity has been documented in various species [8–11], it remains unclear how variation in plasticity is the result of direct evolutionary selection processes on a certain trait or a consequence of selection for other traits [12].
Evolutionary and ecological population models are widely used to explain genetic variation and species composition in different environments, and these models are often based on evolutionary game theoretical principles [13–17]. These models implicitly assume that variation in trait values is entirely due to genetic variation among genotypes. This implies that these models essentially predict selection for different genotypes in different environments when these environments select for different trait values. However, if plasticity would be considered in evolutionary game theoretical models as the ability of a genotype to change its trait value in response to environmental conditions, selection for different trait values in different environments would not necessarily lead to selection for different genotypes. In this paper we take the first step in exploring how variation in phenotypic plasticity could be the result of natural selection in different environments. Phenotypic plasticity is the result of plastic responses that are driven by physiological processes, and these responses are directly the result of environmental cues. To this end, the physiological processes underlying the plastic responses to environmental cues have to be quantitatively linked to trait values and to the performance of individual plants in various environments. Considering plasticity as a trait in itself and considering variation in plasticity across genotypes is required to analyse to which extent natural selection may have acted on variation in plastic responses.
In this study we focus on plastic responses to light competition in vegetation stands of varying planting density and associated neighbour-plant proximity. Plants growing at high density (close proximity to neighbour plants) typically exhibit greater elongation rates of leaf-supporting structures (i.e. internodes and/or petioles), reduced branching and greater leaf inclination angle than plants growing at low density [18–22]. One of the primary signals that induces these shade avoidance responses [23] is a reduction of the red to far-red ratio of light (R:FR), as plants selectively absorb red and reflect far-red light [20]. The reduction in R:FR light perceived by an individual plant is therefore considered a cue for neighbour proximity [24], reviewed in [23]; a low R:FR ratio signals that neighbours are close (high density) and a high R:FR ratio indicates that neighbours are farther away (low density). In addition, R:FR light conditions are also affected by the 3D structure of the canopy [22,25,26] and by self-shading [27]. These alternative causes of changes in the R:FR ratio may decrease the reliability of R:FR as cue for neighbour proximity and future light availability.
To be able to analyse to which extent natural selection may have acted on variation in shade avoidance responses to R:FR, it is required to consider the feedback between the R:FR cue and the plant phenotype. Changes in R:FR induce responses at the organ-level that cause changes in plant architectural phenotype, which in turn affects light capture for growth. The changed phenotype, in turn, changes the light environment and associated R:FR conditions, inducing new sets of responses and this continues throughout plant development. This feedback can be captured in so-called functional-structural plant models ([28], also called virtual plant models) that can mechanistically simulate the interaction between plant 3D structure, growth, and the light distribution within the canopy [29,30]. While taking into account the phenotype-environment feedbacks created by the vegetation itself, R:FR induced organ-level plastic responses and variation within these responses can be realistically scaled up to whole-plant performance at vegetation level [31,32]. We utilize a recently developed and validated virtual plant model [27,31] that simulates Arabidopsis (Arabidopsis thaliana) rosette growth based on light absorption for photosynthesis and growth and induces phenotypic changes via a plastic response curve allowing plants to dynamically change their phenotype during growth. The current model simulates the consequences of specifically petiole and lamina plasticity [33] for whole-plant performance in different environmental conditions based on an organ level plastic response curve that describes the sensitivity of the relative petiole and lamina response to R:FR. The response curve is treated as a trait itself, and different shapes of the curve represent different plasticity strategies with the value α (Fig 1). Petiole plasticity entails petiole elongation in response to decreasing R:FR and can put the leaves in a higher strata of the canopy to increase light capture. Lamina plasticity entails lamina growth reduction in response to decreasing R:FR and can negatively affect light capture because it reduces the total lamina area. Therefore these two organ-specific plasticities are considered antagonistic.
We combine this virtual plant modelling approach with evolutionary game theory (Box 1 and Fig 2) to analyse the extent to which variation in plastic responses could be the result of natural selection in different competitive environments. Different planting densities represent different competitive environments. Specifically, we search for convergence stable evolutionary stable strategies (cESS, [17,34]) at five different planting densities. A cESS (with strategy value α*) is a strategy that is evolutionary and convergence stable, which means that i) a resident population with trait α* cannot be invaded by a rare mutant with a trait value of α (both locally and globally) and ii) a mutant that has a trait value closer to α* than the trait value of the resident, can invade the resident population, if the resident population has any other value than α* [34,35]. In theory, the cESS definition also often requires the resident population to reach carrying capacity before invasion of a rare strategy [36,37]. However, in this study we assume that the resident population is at carrying capacity in the density tested as planting density is the environmental factor of interest.
To summarise, we ask the question to what extent natural selection may have acted on, or resulted in, variation in plastic responses by selecting different plastic shade avoidance response curves at different planting densities. We hypothesize that different plastic response curves will be selected at different planting densities if a given R:FR ratio signals a different level of future neighbour shading across these planting densities. In addition, we also explore the extent to which selection for plastic response curves depends on the cost trade-off associated with the petiole versus lamina responses to R:FR. The latter is expected to influence the selected plastic response curves because petiole and lamina responses have opposite effects on light capture and therefore influences plant performance.
To illustrate how plant growth in the virtual Arabidopsis model depends on planting density (Fig 3), we simulated the growth of plants within monomorphic vegetation stands consisting of plants that do not exhibit R:FR induced plasticity related to petiole elongation or lamina growth reduction (plasticity strategy α = 0 in Eq 1) at different densities. During canopy development, interaction between leaves of neighbour plants (see Methods and ref. [31]) resulted in increased leaf angles and lamina growth resulted in increased total leaf area index (Supporting Information S1 Fig). Together these processes decreased the average R:FR ratio perceived by the plants (Fig 3A). In addition to the decreasing R:FR ratio, light availability per individual plant also decreased during canopy development due to the presence of neighbour plants. Therefore, individual plant growth, represented by total accumulated biomass, was, at the end of canopy development, lower at high than at low densities (Fig 3B). Biomass allocation to different organs varied over time and varied with density because allocation of carbon to growing organs depended on the total available carbon due to light capture, and on the relative growth rates and the total number of growing organs at any time step (for model description see Methods). Under weak competition for light (low densities), the percentage of biomass allocated to the petioles decreased near the end of canopy development (Fig 3C) because sufficient carbon was available to reach potential growth of petioles and laminas and leftover carbon was additionally stored in laminas and the root. On the other hand, under strong competition for light (e.g. at high densities of 1600 and 6400 plants m-2) there was relative low carbon available for growth of all organs. From this low available carbon, a higher fraction was invested in petioles and relative little in laminas and the root, by which the percentage of carbon allocated to the petiole increased (although slightly at 6400 plants m-2).
To illustrate how the plasticity strategy (shape of the plastic response curve) affects the total plant biomass and the organ specific biomass allocation, monomorphic vegetation stands with different plasticity strategies were simulated (according to the Average scenario). Total plant biomass decreased, in general, with increasing plasticity strategy (Fig 4A), generally being highest for non-plastic individuals. This suggest that monomorphic populations (all individuals having the same plasticity) with low or no levels of plasticity generally perform better than populations with high levels of plasticity. There were exceptions to this trend, especially being that at the highest density (6400 plants m-2) biomass was higher at 0.2 plasticity level than for the 0 level (non-plastic plants). This result suggests that at very high densities some level of plasticity may have a benefit for population level light capture and performance.
During canopy development the plasticity strategy influenced the biomass allocation to petioles and laminas, which resulted in finally different percentages of biomass invested in petioles and laminas (Fig 4B & 4C). Plants with a high plasticity strategy (high α value) allocated relatively more biomass to the petioles and less to the laminas because these plants induced petiole and lamina plasticity at a relatively high R:FR earlier during canopy development. The differences between plasticity strategies on organ biomass allocation are represented in the organ-rank specific sizes, which embody the rosette phenotype of the plant (S2 Fig). The final percentage of biomass allocated to the petioles or laminas was different between planting densities because the dynamically changing R:FR influenced the plastic responses on top of the differences in light availability during growth and the organ specific growth rates, as illustrated previously (Fig 3).
To determine how natural selection may have acted on variation in the plastic response curve, and how this selection may depend on the competitive environment, we performed an evolutionary game theoretical analysis in which we searched for cESS (see introduction for the definition) at five planting densities. The virtual plant model simulated the performance of a mutant within a resident population for different combinations of plasticity strategies for the mutant and the resident population (see Fig 2C) at five planting densities (S3 Fig). These values were used to calculate the invasion exponents of the mutants, which were used to construct pairwise invasibility plots: positive and negative values of the invasion exponent relate to positive (blue) and negative (red) invasibility (Figs 5A–5E, 6A–6E and 6K–6O). To aid the interpretation of these pairwise invasibility plots and help identifying possible cESS, the values of the performance of the mutant within a resident population were interpolated with a non-linear smoother after which the invasion exponent of the mutant was calculated (Figs 5F–5J, 6F–6J and 6P–6T, see Methods for details).
By definition, the identity line in the invasibility plots represents the case where the performance and strategy value of the mutant are identical to the performance and strategy value of the residents (1:1 line). A second isocline (if present) represents the plasticity values where the performance of the mutant equals that of the resident but without the mutant and resident having the same plasticity value. The point where the identity line and the isocline intersect corresponds to a singular strategy that could represent a cESS. In graphical terms the singular strategy is a cESS (with value a*) when moving up or down from the identity line no mutant has higher performance compared to the resident performance (positive invasion exponent of the mutant); and when starting with a resident population that is left of a*, a mutant closer to a* should have a higher performance than the resident population (the same holds for residents right from a*). The extent to which the mutant can be closer to a* is determined by the second isocline. If the region between the isocline and the identity line does not include the horizontal line through a*, a singular strategy cannot invade all resident populations directly, but only through a series of stepwise mutations. Alternatively, the singular strategy could represent a branching point or evolutionary repeller (see [34] for an accessible treatment).
The pairwise invasibility plots based on the average petiole and lamina responses (Fig 5, see S4 Fig for the confidence intervals) show possible cESS at all densities, except for 6400 m-2. The cESS per density corresponds with different plasticity strategy values; higher plasticity values were selected at higher planting densities (Fig 5). This represents selection for plastic response curves with increased sensitivity for R:FR at higher densities. At the highest density (Fig 5J) a second isoclines is not present, by which a possible cESS could not be determined within our tested values. However, both the discrete and smoothed invasibility plots suggest that selection would result in plants with a high plasticity strategy (high R:FR sensitivity) probably beyond the tested values. At 1600 plants m-2 (Fig 5I) the pairwise invasibility plot has a complex shape around plasticity values of 0.5 and 0.6. The calculated values with their confidence intervals (Fig 5D and S4D Fig) show that the invasion exponents of plants with plasticity 0.5 and 0.6 in resident populations of 0.6 and 0.5 respectively are around zero, which means that more simulations are required to determine the precise value of the invasion exponents at that part of the pairwise invasibility plot.
Finally, we analysed the extent to which the above-mentioned cESS per density depended on the relation between petiole and lamina responses. In two additional scenarios, we either decreased (Weak scenario) or increased (Strong scenario) the lamina responses relative to petiole responses upon perception of R:FR (parameter n in Eq 2, see Methods). The balance between petiole and lamina responses, although both have the same plasticity strategy, relates to a trade-off regarding light capture during competition; increased light capture due to longer petioles versus decreased light capture due to smaller lamina area. Although not all pairwise invasibility plots identified a single cESS, the balance between petiole and lamina responses did clearly affect the mutant’s invasion exponents at different densities (Figs 5 and 6). When lamina responses were relatively small, there was selection towards plants with a slightly higher plasticity strategy value, evidently at the three lowest densities (compare Weak scenario Fig 6A–6J with Average scenario Fig 5). Higher plasticity strategy values relate to more sensitive plastic response curves. At 1600 plants m-2 (Fig 6I) there were three intersections between isoclines and the identity line, which suggests that the invasibility environment has a complex shape. However, similar as for the Average scenario, the confidence intervals of the invasibility exponents of plants with plasticity strategy value of 0.3 up to 0.6 included zero (S4I Fig), which suggest that strong conclusions about the invasibility of these mutants within their corresponding resident populations cannot be made with confidence based on the data. Invasibility exponent values around zero suggest that not one single plasticity strategy with a narrow range value would eventually dominate the population, but that plants with a broader range of plasticity values could persist together. At 6400 plant m-2 (Fig 6J) there was no second isoclines, which means that within the tested values no conclusive cESS could be determined. However, the invasibilty plot suggests that a potential cESS lies higher than the 0.7 tested here.
With relatively strong lamina responses there was selection for plants with a lower plasticity strategy compared to weak and average lamina responses (Figs 5 and 6). Although at 100 and 400 plants m-2 (Fig 6P and 6Q) there was no conclusive cESS, we expect that plants with a zero or negative plasticity strategy would be selected for. Negative plasticity strategy values would indicate that plants would invest less carbon in petioles and more carbon in laminas in response to lower R:FR. At 6400 plants m-2, the second isocline crossed the identity line multiple times (Fig 6T). This suggests again that the invasibility environment could be complex, and that not one single plasticity strategy would dominate a population at 6400 plants m-2. More simulations of various mutant-resident combinations within our tested range could verify the exact shape of the complex invasibility environment. Additional replication simulations could also verify that the observed complex invasion environments are not due to stochasticity in the model outcomes.
In this study, we analysed if variation in shade avoidance plasticity could be the result of natural selection in different environments. We did this by testing if different planting densities selected for different plasticity strategies that embodied petiole and lamina responses to a decrease in the ratio between red and far-red light (R:FR). Using a virtual plant model allowed us to scale from organ-level plastic responses and variation therein, to whole-plant phenotype and performance at population level. Organ-level plasticity was induced by changes in the locally perceived R:FR ratio by using a plastic response curve, and allowed plants to dynamically change their phenotype during the growing season, depending on the distance, size and plasticity strategy of neighbour plants. The study shows that different planting densities can select for different response curves, as reflected in the different convergence stable evolutionary stable strategies (cESS); with increasing density more R:FR sensitive plastic responses are selected for. This is in spite of the fact that plants use R:FR variation as a signal of plant density itself. Our findings suggest that variation in shade avoidance responses and variation in signal-sensitivity observed in plants could, in some cases, be the result of selection processes at different planting densities. Since density typically varies over growing seasons and in space and different densities select for different response curves, this could have acted to maintain genetic variation in plastic responses, and could help explain the large variation in plastic responses that is observed within species. In addition, the study shows that the balance between lamina and petiole responses affects the selected plastic responses curves because petiole and lamina responses had antagonistic influences on light capture. Plants with lower signal sensitivity were selected for when inducing plastic responses had relatively high negative consequences for light capture.
We hypothesized that different plasticity strategies would be selected at different planting densities if a given R:FR value signals different levels of future neighbour shading across planting densities. Our results show that selection would lead to different response curves at different planting density (Fig 5). We argue that this may be partly due to the fact that the severity of future neighbour shading (expressed in reduced photosynthetic active radiation) that is signalled by a given R:FR value, differs between planting densities. The amount of photosynthetic active radiation received by laminas whose petioles perceived a R:FR around 1.0 decreased with density (S5 Fig). In low density stands, a drop in R:FR during canopy development is mainly caused by self-shading, whereas in high density stands, R:FR reduction is more strongly determined by neighbour-shading [27]. This means that at different plant densities, a given R:FR ratio has different meanings regarding light availability and the level of impending light competition. R:FR is thus not a fully reliable cue for future shading over a range of densities. This prevents selection for one single plastic response curve over a range of densities. In low densities, when light competition is low and R:FR changes are mostly caused by the plant itself, a less sensitive response to a given R:FR is favourable to avoid relative long petioles and small laminas that have negative consequences for light capture and thus plant performance [19]. In high densities, a more sensitive response curve is favourable to create long petioles that can avoid neighbour shading and increase light interception for plant growth and performance.
Our conclusion, that selection at different densities can result in variation in phenotypic plasticity if the signal is unreliable for future environmental conditions, agrees with previous studies that concluded that signals can have different meanings in different environments, such as R:FR values in open versus closed-canopy forests [10,38,39], which could explain the observed variation in plasticity. Our model approach is different from these studies in that it focussed on a plastic response curve that represented the potential to induce relative trait changes instead of absolute trait values and value differences upon an environmental change. Our plastic response curve could be linked more readily to underlying genetics and physiological processes, as shown for Arabidopsis mutants deficient in transcriptional regulators or hormones [31]. Consequently, our modelling approach represents a step forward in linking selection processes to the genetic basis and physiological processes underlying phenotypic plastic responses.
We explored the extent to which selection for plastic response curves at different densities would depend on the trade-off between petiole and lamina responses that respectively would relate to increased and decreased light capture. Changing the strength of lamina responses relative to petiole responses influenced selection for the response curves over the full range of planting densities. Plants with higher plasticity strategy values were selected for when petiole responses were associated with lower lamina responses (Figs 5 and 6, Weak scenario compared to the other scenarios). This indicates that when inducing plastic responses has lower negative consequences due to lower lamina growth reduction, selection can result in higher R:FR sensitivity. In contrast, plants with lower plasticity strategies were selected for when petiole elongation was associated with a stronger reduction of lamina growth in response to R:FR (Figs 5 and 6, Strong scenario compared to other scenarios). Although at the lowest densities no cESS was found within the tested strategies, the pairwise invasibility plots suggest selection in the direction of plants with zero or negative responses to R:FR. A negative value would indicate that plants would invest less carbon in petiole growth and more in lamina growth in response to lower R:FR. However, based on experimental work [33] we conclude that these inverse petiole and lamina responses to decreasing R:FR are not plausible in Arabidopsis. In general, we conclude that plants without plastic responses to R:FR would be selected for (i.e. plants with low sensitivity for R:FR) when performance consequences of showing plasticity are too negative. This is in accordance with other theories. For example, studies related to the evolution of phenotypic plasticity [6,40] state that selection would favour non-plastic responses if costs of inducing plastic responses are high. In addition, error management theory [41] would predict that when phenotypic responses have high costs, selection would favour less sensitive response curves responsible for the plastic response. Altogether, our results are a quantitative example of the influence of cost trade-offs related to plastic responses on the selection for specific plastic responses.
The 3D virtual Arabidopsis model did not include other light signals than R:FR that can signal neighbour proximity and future shading [42]. For example the combination of low blue and low R:FR can indicate stronger shading and therefore can induce stronger shade avoidance responses than either light treatment alone [43]. Importantly, low blue and low R:FR signals can be created either by increased vegetation density of plants with roughly similar sizes or by tall trees in a closed-canopy forest. This suggests again that the reliability of an environmental cue to induce a response depends on the competitive environment. The location of signal detection on the plant can also affect reliability. For example, perception of low R:FR at the lamina tip is more reliable as cue for neighbour-proximity than R:FR perception at the petiole [27]. The regulation of multiple signals and their interactions are still poorly understood and need to be further studied to better understand selection for organ-specific plasticity under natural conditions.
In addition, besides petiole and lamina plasticity, our simulations did not consider responses to R:FR other than leaf angle increase (see Methods). Other responses such as specific leaf area increase [44], flowering time acceleration [8], root development reduction [45] and defence reduction [46] can also affect plant competition for light [32,47,48] and can thus influence selection for specific plastic response curves. We also did not consider any form of mechanical penalty on developing longer petioles, which would occur in natural systems and can affect plant performance in a density dependent manner. For example, the vulnerability to mechanical damage or hydraulic limitations for longer petioles can depend on density; in high density canopies, leaves can get mechanical support from surrounding leaves or protection against wind, by which plants have a lower risk of mechanical failure even if investment in supporting tissues is low. In low densities, this protection is low (or absent) and investment in longer petioles requires additional carbon allocation for petiole stability, which may affect final plant performance. If inducing phenotypic changes has higher fitness costs in low than in high planting density, we would expect an even larger effect of planting density on the selected response curves than what we found in the present study.
Although we considered a wide range of planting densities (regularly spaced), we did not consider that inter-plant distances within a natural vegetation is normally heterogeneous in space, which would make the light environment even more variable. Our result that different densities select for different plasticity strategies, suggests that when density is more variable in time and space transient evolutionary dynamics may prevail allowing different strategies to persist transiently [49]. Performing an analysis in which the distance between neighbour plants within the vegetation is heterogeneous or the density between successive generations is variable, could be the subsequent step towards identifying if and how many genotypes could persist over time.
In theory, the definition of an cESS requires a population to reach carrying capacity before invasiveness of a rare mutant is tested [36,37]. But in our analyses planting density was the environmental factor in question, and could thus not be changed as part of the analyses. We thus implicitly assumed that the different densities reflected different carrying capacities as determined by the overall environment (e.g. by resource availability). A consequence of the fact that different planting densities resulted in different cESS could be the following: if a given plastic response strategy is a cESS at a specific density, but the carrying capacity is at another density, then one may expect some kind of transience. Eventually it could be expected that different plastic response strategies will persist transiently if density changes strongly between years, or it could be expected that one single strategy would persist if the density stays constant around the carrying capacity of the population. It would be challenging and interesting to consider evolutionary and ecological dynamics while exploring the evolution of plastic responses in future studies.
Our analyses indicate that different planting densities can select for different plastic response curves that represent different R:FR sensitivity for petiole and lamina responses. This is consistent with the considerable genetic variation in shade avoidance responses observed in Arabidopsis and in several other species [8–11]. This result seems in part to be explained by the reliability of the R:FR signal as a cue for future neighbour-shading that varies per density. In addition, selection for specific shapes of the plastic response curve is influenced by the trade-off between responses that have generally positive (petiole elongation) versus negative (lamina growth reduction) consequences for light capture and therefore plant performance. Combining virtual plant modelling and evolutionary game theory is a new step toward analysing how phenotypic plasticity, and the underlying sensitivity to an environmental signal, can affect the composition of genotypes over a range of environments. Promising next steps could be including responses to multiple light signals, considering environmental dependence of inducing phenotypic plasticity and regarding environments to be more variable in time and space.
For this study, a functional-structural plant model of Arabidopsis rosette growth and development was utilised [27,31], built in the simulation platform GroIMP v1.5 (https://sourceforge.net/projects/groimp/). These types of virtual plant models have been proven strongly capable of predicting photosynthetic active radiation and R:FR distribution and plant architecture in stands of different planting densities for various species [32,50,51], including Arabidopsis [22,27,31]. This Arabidopsis model can simulate competitive interactions between Arabidopsis genotypes that differ in sensitivity to R:FR, a key element of this study, and has been validated [31]. The essence of this model is that all individual plants within the canopy grow as a function of the light they absorb, but at the same time create the light environment itself. Therefore, simulating various plant types with different phenotypic responses will determine the specific light environment within the canopy, which in turn will have repercussions for plant growth.
Here we summarise the model description and specify the most important model components, assumptions and choices. More details can be found in ref [31]. Simulated Arabidopsis plants emerged from seeds and grew for 46 days into an adult rosette plant with multiple leaves (petiole and lamina) produced in a spiral pattern, and a single root. The leaves captured light for photosynthesis and were also sinks for carbon. The root only functioned as carbon sink and had no effect on aboveground growth. Organ initiation (e.g. time between leaf emergence) and geometric representation such as orientation of the leaves and the shape of the leaves were simulated using empirical relationships. Plant growth was driven by total carbon assimilation and organ-specific carbon allocation and plastic responses induced by the R:FR environment.
The simulated light source emitted photosynthetic active radiation, red and far-red light and in each model time-step (representing 24 hours) these stochastic emitted light rays were reflected, transmitted and absorbed by the petioles and laminas individually according to their wavelength-specific spectral properties. The light source emitted an R:FR ratio of 2.3, photosynthetic active radiation with an intensity of 220 μmol m-2 s-1, and total daily light intensity was calculated based on 9 hours light per day, representing growth chamber conditions under which validation experiments had been carried out [31]. Plants were placed in regular places grids with different inter-plant distances to create different planting densities: 100, 400, 711, 1600 and 6400 plants m-2, which bracketed the densities in the validation experiment [31]. Border effects were minimized by using the plot replication functionality of GroIMP; canopies were replicated 20 times in the x and y direction and light conditions were calculated and averaged for these 400 canopies.
The local light environment perceived by the individual plants (at the organ level) was created by the specific 3D structures of all plants within the canopy itself. Every individual organ absorbed light which enabled the model to calculate the light partitioning over all individual plants within the canopy. This way, we did not have to make assumptions about light partitioning over individuals with different trait values, as has been done in most other game theoretical light competition models [52–54]. Thus, our model simulates competition for light as an emergent property and not an input parameter. Total accumulated biomass stored in root, laminas and petioles after 46 days of growth was used as a measure of plant performance. The rationale is that seasonal biomass scales with seasonal seed production, which for an annual plant like Arabidopsis equates to life time reproduction [55]. Generally for light-demanding species like Arabidopsis, under light competition this correlation is very strong [21,56,57]. Plants could not die during canopy development, they only stopped growing when light capture was insufficient.
The laminas individually absorbed photosynthetic active radiation that was converted into growth substrates via photosynthesis, assuming a negative exponential light response curve [58]. Total growth substrates per individual plant were summed into a central pool and subsequently partitioned over all growing organs through the relative sink strength principle [59]. The relative sink strength of an organ was expressed as a fraction of total plant sink strength, and determined the demand for substrates for each organ in relation to its age. Organ sink strength was defined as its potential growth rate and calculated using the beta growth function [60]. The beta growth function calculated the growth rate at organ age based on measured maximum organ size and duration of organ growth. Leaves senesced after 40 days of age. Thus primarily, the organs individually grew in time and 3D space based on the allocated substrates they received based on their relative sink strength. This means that when petioles receive more substrates due to an increase in relative sink strength, automatically other sinks such as leaves and roots will receive less.
Lamina and petiole growth was influenced by shade avoidance responses that were induced by changes in the R:FR ratio. In this study we focus on the petiole elongation and downregulation of lamina growth [33]. Leaves also showed leaf angle responses due to neighbour proximity and R:FR conditions, see ref [27,31], but this was not changed in between simulations. Leaves increased their angle with 16 degrees per time-step when the distance between neighbouring leaves was smaller than 2 mm (mimicking touching of leaves [22]) or when R:FR perception at the lamina was below a threshold of 0.5. These settings were chosen based on wild-type Arabidopsis responses and allowed plants to induce leaf angle increase depending on planting density. Petiole elongation and lamina growth downregulation responses occurred every time-step based on the response curve that illustrates the relationship between relative organ growth and R:FR perception at the organ itself (Fig 1). The response curve was found by fitting an empirical relationship through experimentally obtained petiole elongation data [31]:
F=min(2,(R:FRRFRcontrol)−α)
(1)
F is the relative organ growth factor. R:FR and RFRcontrol represent the actual experienced and the control R:FR ratio (RFRcontrol is 2.3, related to growth conditions in the validation experiment [31]). Parameter α (dimensionless) determines the curvature of the response curve and is referred to as the ‘plasticity strategy’ (See Fig 1 for variation in the curvature related to α values). Note that F will be 1 when the plasticity strategy α is 0, which will not induce petiole or lamina plastic responses. The higher the value of α the more sensitive the genotype is to R:FR decrease. We set a maximum of 2 to F, to prevent organs to change their own size more than twice within one day, since this has never been experimentally observed. Variation in the shape of the response curve represents variation in the physiological regulation of the response, as observed among response curves of Arabidopsis mutants [31]. Petiole elongation was simulated by multiplying petiole length with F, taking the R:FR perceived by the petiole itself as input [27]. This petiole elongation was calculated every model time-step after simulating petiole growth based on carbon allocation through the relative sink strength principle. Consequently, the petiole increased its demand (sink strength) for carbon substrates for the next time-step to account for the length increase. Lamina growth downregulation was simulated by decreasing the carbon demand (sink strength) (Eq 2):
D=(DpFn)
(2)
where D is the actual lamina carbon demand (mg day-1) as affected by R:FR perception, Dp the potential lamina carbon demand (mg day-1) as calculated by the beta growth function (see above), F the relative growth factor based on R:FR perceived by the lamina (calculated with Eq 1), and n a dimensionless coefficient that alters the strength of the lamina growth downregulation relative to the petiole elongation response. The default setting of the model had n = 2, also referred to as the Average scenario, see Model scenarios). An increase of n indicates that the demand for lamina growth decreases by which the lamina receives less carbon, while therefore more carbon is available for petiole growth. The n value only differed between scenarios. Note that the decreased carbon demand for a given lamina has direct consequences for the absolute carbon allocation to all growing organs; decreased carbon allocation to a lamina can be related to increased carbon allocation to a petiole. Plastic responses could only occur during the growth phase of the specific organ that is set by the beta-growth function.
In general, evolutionary game theoretical principles assume that over evolutionary time the strategy of a population can change when a rare individual with a different strategy than the population can invade the standing population. A rare mutant with a different strategy than the standing population (also called ‘resident’ population) can invade the standing population when it has a higher performance than the average performance of individuals of the standing resident population, and in theory this invasion will lead to replacement of the population, provided that the change caused to the resident phenotype by the mutant is sufficiently small. With the virtual plant model we captured this by simulating different canopies in which the middle plant of the canopy has the same strategy or a different strategy (also called mutant) than the surrounding resident population (Fig 2). Within this study the different strategies were represented by the shape of the plastic response curve with α ranging from 0 to 0.7 (See Eq 1 and Fig 1). A full matrix of combinations of mutant within a resident population has been simulated, resulting in 64 different canopies (Fig 2C). In total, we replicated these 64 canopies 20 times at five densities and for three scenarios (see below Model scenarios), resulting in a total of 19.200 simulations.
The simulated performances of mutants within resident populations were used to calculate the invasion exponents of the mutants, calculated by log(mean(mutant performance)/mean(resident performance)). These invasion exponents resulted in pairwise invasibility plots (Fig 2D). To smooth out stochastic variation simulated by the virtual model, and to be able to get a more detailed estimation of a possible cESS, the simulated performances of the mutants within the resident populations were interpolated with a non-linear smoother after which the invasion components were calculated (i.e. a general additive model [61] using the ‘mgcv’ package in R https://cran.r-project.org/web/packages/mgcv/mgcv.pdf, see S1 Script for details).
To explore if selection at different planting densities would result in different cESS, we simulated a full matrix of mutant-resident canopies with plasticity values ranging from 0 to 0.7 at five densities. In addition, we tested how the estimated cESS at different densities alter when changing the strength of lamina responses relative to petiole responses to R:FR. This was done through varying the parameter n (in Eq 2), which determined the sink strength of the lamina and therefore the potential lamina growth. In the Average scenario we used the model settings as describe above in which average lamina responses were relatively equal to petiole responses (n in Eq 2 is set to 2). In the Weak scenario, we simulated a weak downregulation of the sink strength of the lamina by setting the value of n in Eq 2 to 1, and in the Strong scenario we simulated a strong downregulation of the lamina sink strength by setting n to 4. Comparing the three scenarios gives insight in the relative importance of the responses that have antagonistic consequences for light capture and therefore plant performance; petiole elongation has generally beneficial consequences by placing the leaf in a higher strata of the canopy versus lamina growth downregulation has generally negative consequences because the size of the lamina that is responsible for capturing light will decrease.
|
10.1371/journal.pcbi.1004029 | Brain Network Adaptability across Task States | Activity in the human brain moves between diverse functional states to meet the demands of our dynamic environment, but fundamental principles guiding these transitions remain poorly understood. Here, we capitalize on recent advances in network science to analyze patterns of functional interactions between brain regions. We use dynamic network representations to probe the landscape of brain reconfigurations that accompany task performance both within and between four cognitive states: a task-free resting state, an attention-demanding state, and two memory-demanding states. Using the formalism of hypergraphs, we identify the presence of groups of functional interactions that fluctuate coherently in strength over time both within (task-specific) and across (task-general) brain states. In contrast to prior emphases on the complexity of many dyadic (region-to-region) relationships, these results demonstrate that brain adaptability can be described by common processes that drive the dynamic integration of cognitive systems. Moreover, our results establish the hypergraph as an effective measure for understanding functional brain dynamics, which may also prove useful in examining cross-task, cross-age, and cross-cohort functional change.
| The human brain is a complex system in which the interactions of billions of neurons give rise to a fascinating range of behaviors. In response to its changing environment—for example, across situations involving rest, memory, focused attention, or learning—the brain dynamically switches between distinct patterns of activation. Despite the wealth of neuroimaging data available, the immense complexity of the brain makes the identification of fundamental principles guiding this task-based organization of neural activity a distinct challenge. We apply new techniques from dynamic network theory to describe the functional interactions between brain regions as an evolving network, allowing us to understand these time-dependent interactions in terms of organizing characteristics of the whole network. We examine patterns of neural activity during rest, an attention-demanding task, and two memory-demanding tasks. Using network science techniques, we identify groups of brain region interactions that change cohesively together over time, both across tasks and within individual tasks. By developing tools to analyze the size and spatial distributions of these groups, we quantify significant differences between brain network dynamics in different tasks. These tools provide a promising method for investigating how the changing brain network properties of individuals correspond to task performance.
| An essential characteristic of the human brain is the ability to transition between functional states in synchrony with changing demand. A central focus in neuroscience involves quantifying this adaptability and understanding the underlying brain organization that supports it. Several studies have accomplished this with functional MRI techniques by delineating changes in regional blood-oxygen-level-dependent (BOLD) signal associated with different cognitive tasks, or between task states and task-free (resting [1], [2]) states [3], [4]. However, this approach, which examines the magnitude of brain activity alone, is unable to completely describe the complex correlation structure linking spatially segregated neural circuits. In particular, while providing crucial insight into the spatial structure and anatomical distribution of functional activity and how it differs between task and resting states, these methods are not well suited to probe the intrinsic organization of the dynamics of task-driven transitions between cognitive states, or co-evolving associations among brain regions throughout a particular task.
Recent advances in network science provide tools to represent and characterize the functional interactions between brain regions forming cognitive systems. In this formalism, brain regions are represented as network nodes and functional connections (estimated by statistical similarities between BOLD signals [5]) are represented as network edges [6], [7]. These approaches enable the statistically principled examination of large-scale neural circuits underlying cognitive processes, and have enabled quantitative comparisons between circuits [8], [9]. Indeed, a growing literature provides evidence that individual tasks may elicit specific functional connectome configurations [10], while maintaining a relatively stable functional backbone reminscent of the connectome configuration evident in the resting state [11].
Nevertheless, these studies have focused on examining task or cognitive states as separate and independent entities, and tools to quantify how brain networks reconfigure between these task states remain significantly underdeveloped. Initial efforts to examine reconfiguration properties of brain networks have focused on quantifying properties of dynamic functional connectivity at rest [12]. A relatively few studies have begun to examine reconfiguration properties during task states [13]–[17] or across a series of brain states accompanying behavioral change [18]–[21]. These studies have robustly demonstrated that functional connectome patterns change during task execution, and that individual differences in these reconfiguration properties have implications for task performance [13], [18]–[20].
In this paper, we ask a complementary set of questions that focus on sets of functional connections rather than on the entire functional connectome pattern. We ask whether sets of functional connections evolve independently within or across brain states, or whether they evolve cohesively, each set controlled by a common regulatory driver. To answer this question, we employ recently developed dynamic network science methods to estimate brain functional networks in one-minute time intervals as 86 participants engage in four task states: a task-free resting state, an attention-demanding state, and two memory-demanding states. We treat the evolving patterns of functional connectivity as temporal, or dynamic, networks [14], [15], [18], [19], [21], [22] and estimate the pairwise correlation between the strengths of functional interactions over time in order to identify groups of functional interactions which display similar changes in strength within and across task states. These groups of network edges with similar dynamic patterns, known as hyperedges, have been used to quantify the co-evolution in functional brain networks over the course of a learning task [23]. Our goal is to adapt this dynamic network science method to investigate the organization of evolving functional correlations both within and between task-specific cognitive states, using hyperedges as a measure of co-evolution. We hypothesize that overall, functional interactions between brain regions especially important for particular tasks are likely to be grouped in hyperedges with interactions between regions used strongly in other tasks, capturing co-evolution between task-specific functional networks as they turn off or on together when switching tasks. Furthermore, we expect that those functional correlations that link sets of brain regions whose coordination is crucial to a particular task will be more likely to co-evolve significantly during that task alone.
In this paper, we demonstrate the existence of hyperedges driven by significant co-evolution within groups of functional interactions, both within and across task states. We develop novel network diagnostics to characterize hyperedges according to their structure, anatomy, and task-specificity. These analyses provide a unique window into the adaptability of the brain as it transitions between states and offer quantitative statistics for the comparison of such adaptability across subject cohorts.
Informed written consent was obtained from each subject prior to experimental sessions. All procedures were approved by the University of California, Santa Barbara Human Subjects Committee.
Subjects engaged in a resting-state (task-free) period, as well as three separate tasks designed to engage different cognitive skills and task-specific brain networks: two separate functional runs of the same attention-demanding task, a memory task with lexical stimuli, and a memory task with face stimuli.
During the resting-state period, participants were asked to lie still with their eyes open and look at a blank screen. The attention task (Fig. 1) required subjects to view sequences of visual stimuli on a screen, with the goal of detecting the presence or absence of a target stimulus in each of several test displays. Before each test display, subjects were presented with a cue arrow whose color and direction provided probabilistic information on whether and where the target stimulus might appear. The test display was then flashed for approximately 50 ms, after which the subjects were required to choose whether or not the target stimulus had appeared. In both memory tasks (Fig. 1), 180 previously studied stimuli and 180 novel stimuli were presented to the subjects, who were asked to determine whether each stimulus was “old” or “new” – i.e., whether it had been previously studied. As in the attention task, the memory tasks included probabilistic cues: each stimulus was shown either in a particular color (lexical stimuli) or bordered by a color (face stimuli) which provided subjects with the probability that the stimulus was novel. Face stimuli were drawn from a variety of online faces databases [24]–[29]. For additional experimental details, see [30], [31], and supplemental information therein.
MRI data was acquired at the UCSB Brain Imaging Center from 116 healthy adult participants using a phased array 3T Siemens TIM Trio with a 12 channel head coil. Functional MRI data was taken while each participant engaged in the four tasks described above. This analysis combines two separate functional runs of the same attention task [30]. The sampling period (TR) was 2 s for the rest and attention tasks and 2.5 s for both memory tasks. In addition to functional data, a three dimensional high-resolution T1-weighted structural image of the whole brain was obtained for each participant.
Specific frequencies of oscillations in the BOLD signal have been associated with different cognitive functions. We focus our investigation on low frequency (0.06–0.125 Hz) oscillations in the BOLD signal that have proven useful for examining resting [43], [44] and task-based functional connectivity [18]. The task-related oscillations are posited to be specific to this frequency range, possibly due to a bandpass-filter-like effect from the hemodynamic response function [45]. We apply a Butterworth bandpass filter to isolate frequencies in the (0.06–0.125 Hz) range [46].
To construct a functional brain network, we use the 194 region hybrid atlas, where each region contains a roughly equal number of voxels. These 194 regions represent the network nodes. The , , and positions of each node are given by the centroid of the voxels which comprise the node. Edge weights in the functional brain network are computed by taking Pearson's correlations between the filtered time series within a defined time period for each pair of nodes [47].
Dynamic networks are constructed by taking the filtered time series in temporal windows of 60 seconds and computing a adjacency matrix of nodal correlations for each time window, where is the number of nodes. Each of these adjacency matrices represents the functional network over the 60 seconds in question. From this set of networks, we extract the edge weight time series by considering the correlation strength in each sequential network. We let be the total number of edges between the 194 nodes and construct an adjacency matrix X, where gives the Pearson correlation coefficient between the time series of edge weight for edges and . The entries of the adjacency matrix represent pairs of edges with correlated weight time series [23].
We consider a range of temporal window lengths from 40 to 120 seconds and find that our results for hyperedge size and spatial distributions are robust to changes in window length in this range. Because the TR varies between the memory tasks and the rest and attention tasks, windows of equal time length include different numbers of data points in different segments of the experiment. To ensure this does not affect our analysis, we conduct an analysis with the number of data points per window held constant, and obtain very similar results (see Figure 1 in S1 Text).
The cross-linked network structure, which contains information about groups of edges with similar time series (hyperedges), is extracted from the edge-edge correlation matrix X [23]. Fig. 2 provides a schematic illustration of the process of determining the cross-linked structure of a network. To exclude entries of X that are not statistically significant, we threshold X by evaluating the -values for the Pearson coefficient for each edge-edge correlation using a false discovery rate correction for false positives due to multiple comparisons [48]. If the -value for an entry satisfies the false discovery rate correction threshold, we set for our thresholded matrix . We set the thresholded entry of all other elements to zero. We binarize this thresholded matrix and obtain , where(1)
Each connected component in represents a hyperedge, a set of edges that have significantly correlated temporal profiles. The groups of nodes in Fig. 2(D) are examples of such connected components. A single hyperedge may include any number of edges between one (a singleton) and (the system size); these edges may be spatially clustered or at disparate locations throughout the brain. The set of all hyperedges defined in produces an individual hypergraph.
This hypergraph technique builds on recent trends in the wider field of network science. First, identifying groups of network edges that share similar properties, rather than the groups of nodes that have traditionally been the focus of community detection methods, has been recently shown to provide more intuitive representations of overlapping nodal communities and hierarchical structure [49]–[51]. Second, the idea of identifying functional groups based on the temporal patterns of their interactions has proven useful [51], [52]. Hypergraphs provide a straightforward method, both edge-based and intrinsically dynamic, of identifying and analyzing temporal patterns in network organization. In this work we focus on functional networks in the human brain, but the hypergraph-related diagnostics introduced below are easily generalizable to a broad variety of dynamic networked systems.
We use several methods to extract statistical features from individual hypergraphs and across the set of subjects.
Previous work identified regions with task-specific activity in rest, attention, and memory tasks [30]. Further understanding of the regions that have a correlation structure unique to one task provides insight into network structure differences between tasks. To investigate the task-specific hyperedge structure, we first group hyperedges that exhibit a significantly higher correlation within one task into task-specific sets. If a hyperedge is significantly correlated in two or more tasks, it is excluded from the task-specific hypergraphs. The task-specificity of hyperedges is calculated by comparing the correlation within a single task to the correlation over the same time length with time points chosen randomly from other tasks. This permutation test uses a Bonferroni correction for false positives due to multiple comparisons [53]. Task-specific hypergraphs are then used to construct task-specific hyperedge size distributions, hyperedge node degree distributions, and co-evolution networks.
To quantitatively probe the differences in spatial organization of dynamic functional co-evolution networks for the four tasks, we investigate two summary metrics that show significant variation across tasks. Choice of these measures is primarily motivated by observed coarse differences in co-evolution network structure.
The first “length-strength” metric is the Pearson correlation coefficient, R, between the strength of a connection in the co-evolution network and Cartesian distance between the two nodes linked by the connection (physical length). The Cartesian distance is computed by taking the , , and coordinates of each node and calculating the square root of the differences squared. The length-strength metric identifies a geometric property of the network, as well as a coarse estimate of the length of the strongest connections. Furthermore, connection length is related to network efficiency [54], [55], so differences in this measure could indicate varying levels of functional network efficiency corresponding to task states.
The second “position-strength” metric is the Pearson correlation coefficient, R, between the strength of the co-evolution network connection with the average anterior-posterior position of the two nodes. A measure of anterior-posterior position for each connection was found by taking the average position of the two nodes in the connection. Identifying the location of strong co-evolution network connections along the anterior-posterior axis provides a measure of where hyperedges are physically present in task states. Both the structural core [33] and a dynamic functional core area, comprised of sensorimotor and visual processing areas [19], are located in the posterior, so nodes in these regions have negative values. A larger negative position-strength value corresponds to a higher probability that hyperedges are active in these core areas.
The length-strength and position-strength metrics are evaluated for significance by comparing the correlation between length or position and connection strength to the same correlation performed on randomly chosen co-evolution connections. Again, the Bonferroni correction is performed to eliminate false positives due to multiple comparisons.
In Results, we discuss how these metrics reveal quantitative differences between task-specific networks. A more detailed analysis of the overlap between hyperedge co-evolution networks and relevant cognitive processing regions is also presented. In this analysis, we describe how delineated areas of higher hyperedge activity consistently correspond to recognized centers of task-specific activity.
In this analysis, we compare our results with two statistical null models based on measures for dynamic networks [22]. Hyperedges are formed from correlated edge time series; consequentially the null overall model randomly shuffles each edge time series over all experiments. This null model is designed to ensure that the hyperedges identified in our analysis can be attributed to the dynamics of the system, rather than some overall statistical property of the data set.
The other null test we perform, which we will refer to as the null within-task model, reorders each edge time series within each task, keeping tasks distinct. This is constructed in order to determine whether there are specific differences in the data between tasks.
We compile the results from the hypergraph analysis for each of the subjects and combine these results to obtain a size distribution, anatomical node degree distribution, and co-evolution network for the group. We then divide the data into task-specific hypergraphs and perform the previously mentioned analyses on the task-specific hypergraphs.
We construct a hypergraph for each individual and examine the cumulative distribution of hyperedge sizes ( from Equation 2), shown in Fig. 4. There is a distinct break in the slope between two branches of the distribution occurring at a size of approximately 100 edges, which we use to distinguish between “large” and “small” hyperedges. The total number of small hyperedges appears to roughly follow a power law with an exponent of approximately . The number of large hyperedges peaks around the maximum size, with relatively few in the middle range from 100 to 1000 edges. In Fig. 4, the sharp drop off in the distribution at large hyperedge sizes reflects the system size limitation on hyperedge cardinality.
There is a distinct partition in all individual frequency versus sizes distributions; one or two “large” hyperedges (), and many “small” hyperedges () that peak at the smallest size. A subject with relatively small maximum hyperedge size has hundreds of edges in this largest hyperedge, as well as multiple “small” hyperedges. The corresponding hypergraph of a subject with a maximum hyperedge near the system size is strongly dominated by the largest hyperedge, which contains almost all edges in the brain.
The null overall model shuffles the data over all tasks. There are no hyperedges greater than size one, so the results from this null model are not depicted in Fig. 4. These singletons signify no significant correlation with other edges. As a result, we performed no further analysis on this null model. The fact that no significant hyperedges were found in the null overall model validates the statistical significance of our results.
The null within-task model shuffles the data but ensures that task data stays within the same task. The size distribution of hyperedges from the null within-task model is shown in Fig. 4. The shape of the two distributions is similar, although the null within-task model has fewer hyperedges in the large regime and there are more singletons than in the original data. This indicates there is co-evolution structure across tasks because this structure corresponds to changes in edge states between two or more tasks. For example, if groups of edges have an overall high correlation in one task and a significantly lower correlation in another, it would induce a hyperedge across the tasks regardless of how the within-task time series are shuffled.
Examining the cumulative hyperedge size distribution provides information about the network topology but does not supply descriptive spatial information. Next, we quantify which anatomical locations in the brain participate in hyperedges, identifying differential roles in task-induced co-evolution. Fig. 5A depicts the hyperedge node degree on a natural log scale. The densest regions are located in posterior portions of the cortex, primarily in visual areas, while a second set of dense regions is located in the prefrontal cortex.
We construct a co-evolution network, as illustrated schematically in Fig. 3, where connection weight corresponds to the probability that two nodes participate in the same hyperedge. In Fig. 5B we present this co-evolution network over all individuals and all tasks. The graph includes sparse long-range connections between regions that are densely connected. Within the strongest 1% of connections, the high degree of bilateral symmetry indicates that corresponding nodes in the left and right hemispheres have a high likelihood of being placed together in a hyperedge. Dense areas of the graph include primary visual areas, portions of prefrontal cortex, and primary motor cortex.
The hypergraph algorithm groups together edges with significantly similar temporal behavior. However, this basic classification does not distinguish whether the correlation is present throughout the edge time series, or whether highly correlated sections of the time series drive the selection. We compute the average within-task edge correlation for each hyperedge and find that in some cases, strong edge correlation spans the tasks, while in other hyperedges, a strong correlation between edges within one task drives the hyperedge. An example of this task-specific correlation structure can be seen in Fig. 6. In the average within-task correlation on the left, there is a stronger average correlation in the word memory task than in any other task. Furthermore, the edge time series in the first hyperedge indicates it is driven mainly by a correlation within the word memory task.
To investigate this further, we construct task-specific co-evolution networks, composed of hyperedges with significantly stronger average correlation in one task than the others (see Methods). To identify these task-specific hyperedges for each task, we perform a permutation test on the edge weight time series, as described in Methods, and compare the total correlation within the task to the expected values. If a hyperedge displays significant edge correlation (determined by the Bonferroni correction on the -values from the permutation test) in only one task, we label it as a task-specific hyperedge. Hyperedges with two or more tasks exhibiting significant correlation are not included in the task-specific hypergraphs.
Fig. 7 illustrates the size distributions of all the task-specific results alongside the overall hyperedge size distribution. The sizes and spatial distributions of single task-driven hyperedges vary across tasks and incorporate significant information about functional network organization with respect to changing cognitive states. Attention has the greatest number of task-specific hyperedges, followed by face memory, word memory, and rest. In the small regime, the tasks follow a similar distribution. There are fewer large attention and rest hyperedges, while the face memory task closely mimics the overall distribution. The distinction in the distributions indicates that the tasks can be characterized by differing complexities of edge co-variations.
The spatial distributions of hyperedge node degree in each task, along with task-specific co-evolution networks, are shown in Fig. 8. The rest hypergraph has the least activity in posterior regions of the cortex, both in the hyperedge node degree plot and co-evolution network. In the attention network, long connections connecting the front and back of the brain distinguish it from the rest network. Furthermore, the concentration in the occipital lobe is larger in the memory co-evolution networks than in the rest or attention networks. We characterize these observed differences with two statistics, which are described in more detail in Methods. The length-strength metric is a correlation between connection length and strength in the co-evolution network. The position-strength metric is a correlation between connection position (anterior-posterior) and strength. The results of this analysis over the full unthresholded co-evolution network are in Fig. 9. All correlation values are negative, indicating that, in all tasks, stronger connections in the co-evolution network are located in posterior portions of cortex and are physically shorter.
We compare these values across tasks by performing pairwise permutation tests to determine which networks have statistically different properties. Fig. 9 depicts the -values from these tests, where the horizontal axis represents the statistic being tested and the vertical axis corresponds to the task being tested against. The black squares in this figure represent significant values, which are summarized in the following list:
These results delineate significant differences in co-evolution network structure between the tasks, confirming that the hypergraph analysis is a useful method for distinguishing between task states. Additional features of the task-specific co-evolution networks are described in more detail below.
Progress in understanding functional brain network topology provides significant insight into broad neuroscience questions regarding the brain's organization and ability to effectively transition between cognitive states. Quantifying complex network dynamics in the brain will further understanding in these areas and has promising applications to behavioral adaptation and learning [18], [19], [21]. We apply hypergraph analysis, a tool from dynamic network science, to functional brain imaging data in order to determine co-evolution properties of the brain as subjects perform a series of tasks. A previous application of this method to neuroscience uses hypergraphs to analyze how functional network structure changes over a long term learning task [19]. The learning experiment considers hypergraphs constructed over 6 weeks of training while subjects acquire a new motor skill, while our analysis compares hypergraphs over three different tasks performed within an interval of hours. Our analysis shows that hypergraphs are a useful tool for investigating shorter time scales and differentiating between task-specific networks.
Instead of analyzing the time-dependent behavior of groups of nodes, the hypergraph investigation considers the edge weight time series, where edges with statistically significant similarities in their temporal profiles are grouped into hyperedges. This approach is advantageous because it considers all edges, regardless of correlation strength, unlike previous methods which focus exclusively on strong correlations [30], [56]. The use of a data-driven analysis also allows us to investigate the dynamic changes in brain function over a series of tasks without prior assumptions of the structure of the connectivity network. This is a significant advantage over methods that characterize task states based on their differences with respect to the rest network [3], [4]. A comparison between the hypergraph analysis and these methods in a future analysis could reveal how the concentration of hyperedges varies in known task-positive or task-negative areas and determine whether this variation has an effect on task performance.
We demonstrate the existence of hypergraph structure in functional brain dynamics and statistically characterize the hyperedge distributions in comparison to appropriate null models. Shuffling the time series over all time produces no significant hyperedges, while shuffling within each task results in a size distribution that resembles the overall size statistics in shape, but with far fewer hyperedges. The distinct differences between the two null models and our results based on the original time series establish the significance of our findings. Furthermore, the existence of hyperedges after the within-task shuffling indicates the presence of activity in some edges that is differentiated between tasks. Since there are fewer large hyperedges after the within-task shuffling, we can also confirm that there are hyperedges caused by edge dynamics within tasks. This work primarily concentrates on hyperedges correlated within a particular task, but future analyses to understand the properties of hyperedges that are grouped due to other general properties would supplement our results.
The hyperedge size distribution is comprised of “small” and “large” hyperedges, where the size distribution of the small hyperedges follows a power law and the large hyperedges peak at the system size. We explore the overall spatial hyperedge distribution by constructing a hyperedge node degree plot, and find that the majority of the most densely connected nodes lie in the posterior portions of the brain. To better observe spatial hyperedge properties, we develop a co-evolution network, where connection weights correspond to the probability that a hyperedge will include the connection. The top 1% of connections in the network with the highest probability of inclusion in a hyperedge are most concentrated in the occipital lobe and prefrontal cortex. These are expected areas of hyperedge concentration, consistent with the visual nature of the tasks, as well as the coordination of quick decision making and the selection of specific motor responses.
We find there are hyperedges that are more correlated in one task and hyperedges that have a distinct profile across the tasks. Our results suggest that edges with a high probability of inclusion in task-specific hyperedges are often found in previously identified brain areas associated with the corresponding tasks, as discussed in detail below, confirming that the approach captures relevant information about task networks. In some cases, brain regions expected to show strong co-variation in a certain task are not included among the strongest connections of that task-specific co-evolution network; we also discuss examples of this in detail below. Repeating the analysis and grouping hyperedges that are significantly correlated in two tasks might lend insight into whether brain systems relevant to a certain task contain hyperedges that are correlated in another task and thus are rejected from our task-specific analysis.
In all tasks, stronger connections in the co-evolution network tend to be located in posterior portions of cortex and to be physically shorter. The higher probability of posterior edges to be included in hyperedges is consistent with the identification of a core set of highly structurally connected regions centered in the posterior of the brain, thought to play an important role in integrating large-scale functional connectivity [19], [33]. The tendency of strong connections to be physically shorter suggests high efficiency in task-specific co-evolution networks. This may reflect efficient wiring properties associated with minimal wiring for rapid processing and low energy expenditures found in structural brain networks and shared by some other biological and technological networked systems [57].
Because they consider both strong and weak edges with no thresholding, hypergraphs are well-suited for identifying groups of brain regions that, for example, initially have uncorrelated activity but become more correlated in synchrony (or vice-versa), as we expect task-associated cognitive networks to do as a result of switching between tasks. In order to extract these dynamic patterns, the hypergraph technique considers strong and weak edges equally, ignoring any offset between the average correlation strengths of different edge time series. This is intended to provide a complementary method to the common thresholding approach of separating or ignoring network edges with correlation strengths weaker than some critical value [30], [56]. Since weak edge connectivity has been shown to contain functionally relevant and predictive information in various contexts, retaining these edge weights is desirable [44], [81], [82]. There is also evidence that mean edge correlation values can be driven by non-biological artifacts such as head motion, even after applying standard motion-correction techniques [20]; by remaining indifferent to edge weight offsets, a hypergraph analysis avoids this concern.
In applications where the overall correlation strength of network edges is nevertheless important, it may be useful to supplement the dynamic information given by a hypergraph analysis with a measure that retains this edge weight information. Efforts to make quantitative comparisons between the hypergraph analysis and other dynamic graph theoretical methods in the context of the human brain are ongoing. We are currently investigating whether dynamic community detection on weighted brain networks, a node-based analysis which relies on edge correlation strength, provides complementary information to the hypergraph analysis.
Because we choose a linear measure to compute correlations between edge weight time series, our analysis as presented here does not account for time lag in these correlations. However, our framework could be extended to nonlinear measures that include time-lag information.
It is important to note that our method of computing a dense matrix of edge-edge correlations and thresholding according to significance does not necessarily identify direct conditionally-dependent correlations between time series, or correlations that represent the underlying structural connectivity of the brain. As with any method that infers a network structure from correlation data simply by thresholding, we expect many of these correlations to be indirect. For example, a significant correlation between two edge weight time series may occur because both edges are being controlled by a third, more central edge – and not because the two edges are directly connected either causally or structurally. In this sense, the edge-edge correlation structure does not capture relations that necessarily reflect the underlying control structure or the physical architecture of the brain. Our hyperedge analysis moves the focus away from such indeterminate dyadic relationships, considering only groups of all edges that share similar dynamic patterns without any intra-group organization or structure.
It is also possible, as in any fMRI analysis, that edge-edge correlations arise from task-induced indirect drivers, such as visual stimuli. Two regions that are both activated by a visual stimulus may show strong functional connectivity with one another in a single time window. Moreover, such regions may show similar changes in functional connectivity over time if their activation profiles to the stimulus evolve similarly during the experiment. As with any measurement of functional connectivity based on the Pearson correlation coefficient [83], a common and robust measurement of functional connectivity, such indirect drivers of functional connectivity are not distinguished from other more direct drivers of communication or interaction.
Throughout this work, we observe a significant amount of individual variability in the hypergraph properties of interest. In this manuscript, we have completed a group-level analysis and focused on investigating task-related differences in hypergraph structure. However, individual variability may be related to differences in cognitive ability and provide additional insight into the role of hyperedges in task performance, which is a topic of future research.
In this paper, we use hypergraph analysis to identify significant co-evolution between brain regions in task-based functional activity and develop new tools to summarize the spatial patterns of these co-evolution dynamics over the group of subjects. By isolating task-specific hyperedges, we quantify significant differences between the spatial organization of co-evolution dynamics within different tasks. This hypergraph analysis adds a crucial perspective to previous treatments of task-based brain function, describing temporal similarities between spatially segregated neural circuits by specifically examining the organization of connections that co-evolve in time. It provides a promising approach for understanding fundamental properties of task-based functional brain dynamics, and how individual variation in these properties may correspond to differences in behavior and task performance.
|
10.1371/journal.pbio.1000556 | A Mitochondrial Superoxide Signal Triggers Increased Longevity in Caenorhabditis elegans | The nuo-6 and isp-1 genes of C. elegans encode, respectively, subunits of complex I and III of the mitochondrial respiratory chain. Partial loss-of-function mutations in these genes decrease electron transport and greatly increase the longevity of C. elegans by a mechanism that is distinct from that induced by reducing their level of expression by RNAi. Electron transport is a major source of the superoxide anion (O⋅–), which in turn generates several types of toxic reactive oxygen species (ROS), and aging is accompanied by increased oxidative stress, which is an imbalance between the generation and detoxification of ROS. These observations have suggested that the longevity of such mitochondrial mutants might result from a reduction in ROS generation, which would be consistent with the mitochondrial oxidative stress theory of aging. It is difficult to measure ROS directly in living animals, and this has held back progress in determining their function in aging. Here we have adapted a technique of flow cytometry to directly measure ROS levels in isolated mitochondria to show that the generation of superoxide is elevated in the nuo-6 and isp-1 mitochondrial mutants, although overall ROS levels are not, and oxidative stress is low. Furthermore, we show that this elevation is necessary and sufficient to increase longevity, as it is abolished by the antioxidants NAC and vitamin C, and phenocopied by mild treatment with the prooxidant paraquat. Furthermore, the absence of effect of NAC and the additivity of the effect of paraquat on a variety of long- and short-lived mutants suggest that the pathway triggered by mitochondrial superoxide is distinct from previously studied mechanisms, including insulin signaling, dietary restriction, ubiquinone deficiency, the hypoxic response, and hormesis. These findings are not consistent with the mitochondrial oxidative stress theory of aging. Instead they show that increased superoxide generation acts as a signal in young mutant animals to trigger changes of gene expression that prevent or attenuate the effects of subsequent aging. We propose that superoxide is generated as a protective signal in response to molecular damage sustained during wild-type aging as well. This model provides a new explanation for the well-documented correlation between ROS and the aged phenotype as a gradual increase of molecular damage during aging would trigger a gradually stronger ROS response.
| An unequivocal demonstration that mitochondria are important for lifespan comes from studies with the nematode Caenorhabditis elegans. Mutations in mitochondrial proteins such as ISP-1 and NUO-6, which function directly in mitochondrial electron transport, lead to a dramatic increase in the lifespan of this organism. One theory proposes that toxicity of mitochondrial reactive oxygen species (ROS) is the cause of aging and predicts that the generation of the ROS superoxide should be low in these mutants. Here we have measured superoxide generation in these mutants and found that it is in fact elevated, rather than reduced. Furthermore, we found that this elevation is necessary and sufficient for longevity, as it is abolished by antioxidants and induced by mild treatment with oxidants. This suggests that superoxide can act as a signal triggering cellular changes that attenuate the effects of aging. This idea suggests a new model for the well-documented correlation between ROS and the aged phenotype. We propose that a gradual increase of molecular damage during aging triggers a concurrent, gradually intensifying, protective superoxide response.
| Mitochondrial function has been linked to the aging process in a number of ways [1]. In particular, mitochondria are crucial in energy metabolism and as such have been implicated in the aging process by one of the very first theories of aging [2], the rate-of-living theory of aging [3], which suggested that the rate of aging is proportional to the rate of energy metabolism (reviewed in [4]). Mitochondrial function in animals is also known to decline with age [5],[6], which, together with the finding that mitochondria are an important source of toxic reactive oxygen species (ROS), has led to the oxidative stress (or free radical) theory of aging [7],[8].
Two types of mutations that affect mitochondrial function have been found to affect the rate of aging in C. elegans, mutations that shorten lifespan, such as mev-1 [9] and gas-1 [10], and mutations that lengthen lifespan, such as clk-1 [11], isp-1 [12], lrs-2 [13], and nuo-6 [14]. lrs-2 encodes a mitochondrial leucyl-tRNA-synthetase, and its effect on the function of mitochondrial electron transport is likely relatively indirect, via partial impairment of mitochondrial translation. However, clk-1 encodes an enzyme necessary for the biosynthesis of ubiquinone, a lipid antioxidant and an electron transporter of the respiratory chain [15], and mev-1, gas-1, isp-1, and nuo-6 all encode subunits of mitochondrial respiratory complexes. On the strength of the oxidative stress theory of aging it has been suggested, and supported by a number of observations (reviewed in [16],[17]), that the mev-1 and gas-1 mutations reduce lifespan by increasing mitochondrial oxidative stress, and clk-1, isp-1, and nuo-6 increase lifespan by reducing it.
In addition to genomic mutations that affect mitochondrial proteins, it has been found that knockdown by RNA interference of C. elegans genes that encode subunits of mitochondrial complexes, including isp-1 and nuo-6, also prolongs lifespan [13],[18],[19]. Although the effect of RNAi on ETC subunits, which is conserved in Drosophila [20], was initially believed to be similar to that of the mutations [21],[22],[23], it was recently found that it is in fact distinct and separable [14].
A recent study analyzed patterns of gene expression in isp-1 mutants together with those in clk-1 and cyc-1(RNAi) [23] and suggested that the overlap between these patterns could define the biochemical processes that underlie the effect of all interventions that impact mitochondria. However, our recent findings that isp-1(qm150) and isp-1(RNAi) trigger fully separable mechanisms suggests that the overlapping gene expression changes identified by Cristina et al. [23] might not be sufficient to prolong lifespan. Rather some of the gene expression changes that are specific to each type of intervention are necessary for their effect on lifespan and can act additively.
isp-1 mutants show a trend toward low levels of oxidative damage to proteins, increased expression of the cytoplasmic Cu/Zn superoxide dismutase (SOD-1) and of the mitochondrial Mn superoxide dismutase (SOD-2) [24], and increased resistance to acute treatment with the prooxidant paraquat [14]. However, although knocking down the genes encoding the major superoxide dismutase by RNAi results in normal or elevated levels of oxidative damage, it had no effect on the lifespan of the mutants [24], suggesting that the reduced oxidative damage found in isp-1 mutants is not responsible for their longevity. Furthermore, the notion that mitochondrial oxidative stress could be the cause of aging has recently been challenged by a number of studies in C. elegans [24],[25],[26],[27],[28], in Drosophila [29], and in mice (reviewed in [30]).
ROS are not just toxic metabolites that lead to oxidative stress but are also signaling molecules that are believed to be involved in a mitochondria-to-nucleus signaling pathway that could impact aging [1],[31],[32],[33]. Interfering with mitochondrial function has the potential to alter the rate and/or the pattern of production of ROS by mitochondria, including in counter-intuitive ways. For example, reducing oxygen concentration increases ROS production by mitochondrial complex III in vertebrate cells [34],[35], and the knockout of sod-2 in C. elegans can lead to normal [25] or increased lifespan in spite of increased oxidative damage [26]. Here we examined ROS production by mitochondrial mutants and found that isp-1 and nuo-6 mutants have increased generation of the superoxide anion but not increased levels of other ROS and that this increase is necessary and sufficient for longevity, suggesting that superoxide triggers mechanisms that slow down aging, presumably at the level of gene expression.
To measure changes in mitochondrial ROS generation that could affect signaling, it is not adequate to measure the level of ROS damage, as a change in ROS damage levels can be brought about by changes in detoxification of ROS, in protein turnover, or in damage repair. However, it is notoriously difficult to directly visualize or measure ROS generation and ROS levels in intact organisms including in living worms. To overcome this difficulty we have adapted a technique originally developed for vertebrates that uses flow cytometry to sort isolated intact mitochondria and measure ROS levels with indicator dyes (Figure S1) [37]. Mitochondria were extracted from worms by standard techniques and loaded with either one of two fluorescent indicator dyes, H2DCFDA, a dye that is sensitive to a variety of ROS but rather insensitive to superoxide [38],[39], and MitoSox, a dye that is exclusively sensitive to superoxide [40]. The prooxidant paraquat (PQ) induces mitochondrial superoxide generation [41], and the antioxidant N-acetyl-cysteine (NAC) has an antioxidant effect on all types of ROS [42],[43]. As expected, when purified mitochondria were treated with PQ, the fluorescence of both H2DCFDA and MitoSox increased, and the fluorescence of both decreased when treated with NAC (Figures 1A, 1B, and S1B). One limitation of this technique is the need for a rather large amount of mitochondria. For example, a sufficient amount of worms is not readily obtained from worms treated by RNAi, and we have therefore focused on long-lived mutants only.
We used the cytometry technique to determine the generation of mitochondrial superoxide and of overall mitochondrial ROS in a number of long-lived mutants. Both isp-1 and nuo-6 mutations did not affect H2DCFDA fluorescence (overall ROS) significantly, but both showed elevated MitoSox fluorescence (superoxide) (Figure 1C and 1D). Mutants of four other genes (clk-1(qm30), eat-2(ad1116), daf-2(e1370), and sod-2(ok1030)) were also tested (Figure 1E and 1F). clk-1 mutants showed an elevation of overall ROS-associated fluorescence but not of superoxide-associated fluorescence. daf-2 mutants were most similar to the mitochondrial respiratory chain mutants with an elevation of superoxide-associated fluorescence but no significant elevation in overall ROS-associated fluorescence. Finally, eat-2 and sod-2 mutants showed no significant elevation in either signal but only a trend for low overall ROS in the case of eat-2 mutants and a trend for increased superoxide in the case of sod-2 mutants.
The elevated MitoSox signal in isp-1, nuo-6, and daf-2 corresponds mostly to increased superoxide generation, as all three mutants are known for elevated levels of the mitochondrial SOD-2 and SOD-3 [12],[14],[24],[44], whose activity would prevent the accumulation of superoxide. Elevated superoxide detoxification, however, should not prevent measuring increased superoxide generation as superoxide is generated at prosthetic electron carriers such as ubiquinone in complex III [45],[46] and FMN in complex I [47],[48], which are at least partially buried in the complexes. Thus a small molecular weight dye that has access to these sites can trap the superoxide before it has the opportunity to diffuse toward the SOD-2 and SOD-3 proteins. There is no increase in the H2DCFDA signal in these mutants likely because this dye is not particularly sensitive to superoxide [49]. It appears therefore that in the presence of efficient detoxification the level of overall ROS is not significantly increased by the increased superoxide generation that we observe. This is consistent with the finding that these mutants do not have increased oxidative damage [14],[24].
sod-2 deletion mutants do not show a significant increase in the MitoSox signal (Figure 1E and 1F), indicating that decreased detoxification does not lead to an easily measurable increase in this signal in purified mitochondria. The signal from H2DCFDA, a dye which has very broad sensitivity but is not very sensitive to superoxide [49], is also unchanged, suggesting that, at least in isolated worm mitochondria, electron transport is not the main source of the type of ROS to which H2DCFDA dye is significantly sensitive. The level of superoxide generation in these mutants might also be kept moderately low because of their reduced electron transport [26], although low electron transport could in principle also result in elevated superoxide as we have observed in isp-1 and nuo-6 mutants.
clk-1 mutants have only a small deficit in electron transport [24],[50],[51], in spite of a strongly altered content in quinones [51],[52],[53],[54]. Indeed, while wild-type animals contain endogenously synthesized UQ9 as well as a small amount of dietary bacterial UQ8, clk-1 mutants contain only the dietary ubiquinone and no UQ9. Here we found that clk-1 mutants have normal superoxide generation but enhanced overall ROS levels, which suggests that the antioxidant function of UQ9 is a crucial sink for mitochondrial ROS, whose absence appears to lead to an increase of overall ROS even in the absence of increase superoxide generation. eat-2 mutants are long-lived because of reduced food intake (dietary restriction) [55]. Although dietary restriction has been found to impinge on mitochondrial function in other systems, no changes in mitochondrial superoxide and overall ROS signals were observed.
To determine how the elevated superoxide affects the lifespan of mutants, we treated worms with 10 mM of NAC and scored their survival (Figure 2 and Table 1). The treatment had no effect on the survival of the wild type (Figure 2A), which shows that it is not toxic for lifespan at the concentration used. However, NAC treatment fully abolished the increased longevity of nuo-6 and severely limited that of isp-1 (Figure 2B and 2C). The lesser effect on isp-1 is consistent with the larger increase of superoxide in these mutants (Figure 1D), given that the effect of NAC is gradual (1 mM has less effect than 8 mM, which has less than 10 mM; Table S1). At high concentration (>10–15 mM) NAC can be deleterious even on the wild type, but at the concentration used (10 mM) NAC had no effect on the apparent health of the mutants, whose overall aspect after treatment was indistinguishable from that of the untreated worms (Figure S2A). We have also quantified several phenotypes, including defecation, swimming, brood size, and post-embryonic development, after NAC treatment of the wild type and of nuo-6, which is the mutant that is most sensitive to NAC (10 mM NAC completely abolishes its increased longevity). Treatment with 1 mM vitamin C also significantly shortened the lifespan of both isp-1 and nuo-6 mutants without affecting the wild type (Table S1). Most effects of NAC were quite small (Figure S2B–E), except on the post-embryonic development of the wild type (Figure S2C). Furthermore, for defecation, brood size, and post-embryonic development, the effect of NAC on the mutant produced a change in the same direction as on the wild type but of a lesser extent. Only for swimming is the effect greater on the mutant. But the effect consists of swimming faster after NAC treatment and thus bringing the mutant phenotype closer to the wild-type. We conclude that there is little evidence of an indirect deleterious effect of NAC.
NAC had only a moderate effect on the lifespan of the insulin-signaling daf-2 mutants (Figure 2E), suggesting that only a small part of the increased longevity of these mutants requires elevated mitochondrial superoxide. However, NAC fully abolished the increased lifespan of sod-2 mutants (Figure 2F), suggesting that, although increased generation of superoxide and other ROS as detected by our techniques were not significantly altered in these mutants, their increased lifespan depends on an elevation of superoxide or some other ROS. NAC did not shorten the lifespan of clk-1 mutants at 10 mM (Figure 2D), or even at 15 mM (Table S1), indicating that ROS metabolism is relatively irrelevant to the aging phenotype of these mutants. The effect of NAC on the lifespan of eat-2 could not be scored because NAC treatment rendered the animals unable to lay their eggs and they died from internal hatching at a young age. The origin of this effect is unknown. We also could not score the effect of NAC on RNAi-treated worms because 10 mM NAC was excessively damaging to the dsRNA-producing bacterial strain (HT115).
To determine whether an elevation in mitochondrial superoxide generation is sufficient to increase lifespan, we used the superoxide generator PQ. Treatment of C. elegans with high concentration of PQ (>0.2 mM) is severely deleterious. We thus first tested the ability of PQ to increase ROS damage in the animals at a very low concentration (0.1 mM). We found that this treatment indeed measurably increased the level of oxidative damage to proteins at the young adult stage as assessed by determination of protein carbonylation (Figure 3A) and increased the expression of both the main cytoplasmic (SOD-1) and the main mitochondrial (SOD-2) superoxide dismutases (Figure 3B and 3C). We then tested whether PQ could increase the lifespan of the wild type at three different concentrations (0.05, 0.1, and 0.2 mM) and found that at all three concentrations both the mean and maximum lifespan were increased, with a maximal effect at 0.1 mM (Figures 3D and 4A, and Tables 1 and S1). The effect of 0.2 mM was less pronounced than that of 0.1 mM and similar to that of 0.05 mM, likely because at 0.2 mM a toxic effect starts to balance the pro-longevity effect. The effect does not depend on the exact chemical structure of paraquat, as benzyl-viologen, a compound with similar activity as PQ but structurally different, also increases lifespan (Table S1). A small effect of the prooxidant juglone under different conditions has also been documented previously [56]. The effect did not depend on an effect of PQ on the E. coli (OP50) food source, as the effect was also observed with heat-killed cells (Table S1). Finally, the effect was not confined to development or adulthood as PQ prolongs lifespan whether provided only during adult lifespan or only during development (Table S1).
PQ treatment failed to significantly prolong the lifespan on nuo-6 and isp-1 mutants (Figure 4B and 4C, and Tables 1 and S1). This experiment is equivalent to genetic epistasis analysis and suggests that nuo-6, isp-1, and PQ increase lifespan by the same mechanism. It also suggests that the maximum level of lifespan extension that can be obtained by increasing mitochondrial superoxide generation is already reached in these two mutants and further increase of superoxide generation through PQ treatment cannot increase lifespan further. This was not due to a resistance of these mutants to PQ as 0.2 mM PQ shortened the lifespan of the two mutants (Table S1). sod-2 mutants, whose longevity is suppressed by NAC, are not as long-lived when untreated as wild type animals that are treated with PQ. However, treatment with PQ makes the sod-2 mutants live as long as wild type animals treated with PQ (Figure 4G). This absence of additivity suggests that the longevity of sod-2 mutants is indeed due to a small increase in superoxide, as expected from the function of SOD-2, and the suppressing effect of NAC on the mutant lifespan. In contrast to what we observed with nuo-6, isp-1, and sod-2, PQ treatment dramatically enhanced the lifespan of clk-1 and eat-2 mutants, significantly beyond the longevity increases induced by the mutations alone or by PQ applied to the wild type (Figure 4D and 4E, and Tables 1 and S1). This indicates that the effects of these mutations and the effect of superoxide are mechanistically distinct and additive, as expected from the finding that clk-1 and eat-2 mutants did not show increased mitochondrial superoxide levels (Figure 1F) and that the lifespan of clk-1 mutants could not be shortened by NAC treatment (Figure 2D). PQ treatment had only a small lifespan-lengthening effect on daf-2 (Figure 4F, and Tables 1 and S1), which is consistent with the finding that daf-2 mutants already show a substantial increase in superoxide generation.
We sought to determine whether the mutations and the PQ treatment had other common effects on mitochondrial function that could be responsible for the increased lifespans, besides elevation of superoxide levels. Work in other systems has suggested that increased mitochondrial biogenesis could impact lifespan positively [57],[58],[59], and mitochondrial defects in C. elegans have been found to stimulate mitochondrial biogenesis, resulting in a denser mitochondrial network [13]. We have examined mitochondrial density in the two mitochondrial mutants and in PQ-treated worms with Mitotracker Red, which is specific to mitochondria in mammalian cells [60],[61], which stains worms uniformly, and whose staining fully overlaps with that of mitochondrially targeted green fluorescent protein (GFP) (Figure S3). We found that isp-1 and nuo-6 display a denser mitochondrial network, as expected (Figure 5). However, this was not observed in wild type worms treated with PQ (Figure 5), indicating that the mechanism by which the superoxide triggers longevity does not require increased mitochondrial biogenesis. We also tested the effects of PQ and NAC treatment on oxygen consumption and ATP levels in the wild type and in the two mitochondrial mutants (Figure S4). NAC treatment increased oxygen consumption in the wild type and in the mutants. This result uncouples oxygen consumption from lifespan as NAC has no effect on the lifespan of the wild type, and its effect on the oxygen consumption of isp-1 mutants is larger than on that of nuo-6 mutants, although its effect on aging is smaller. PQ had an effect only on nuo-6, and it was small. Thus the effect of PQ on oxygen consumption also did not mirror its effect on lifespan. For ATP levels the only effect observed was a reduction by PQ of the elevated ATP levels that are observed in nuo-6 mutants.
daf-2 mutants have elevated superoxide levels, and they are sensitive to NAC (lifespan shortening by 15%). However, the level of superoxide in daf-2 appears not to be sufficient for a maximal effect as these mutants remain somewhat sensitive to PQ (lifespan lengthening by 9%). To further study how superoxide plays a role in the lifespan of daf-2 we studied genes that function downstream of daf-2. At least three genes are known to be required for the full lifespan extension of daf-2, that is, daf-16, aak-2, and hsf-1 [62],[63],[64]. If one of these genes were necessary for an activity that mediates the small effect of PQ on daf-2 mutants, PQ should not be able to prolong the lifespan of mutants of such a gene. In fact, however, we found that PQ prolonged the lifespan of all three mutants tested (Table 1). The lifespan increase upon PQ treatment of daf-16 (35% increase) and aak-2 (29% increase) is not as large as upon treatment of the wild type (58% increase). This suggests that part but not all of the lifespan increase determined by superoxide requires daf-16 and aak-2. These findings are consistent with the observations that the lifespan extension provided by nuo-6 and daf-2(e1370) are only partially additive (Table S1), similarly to what was found previously for isp-1 and daf-2 [12], and that elimination of daf-16 partially shortens the lifespan of isp-1 [12].
We also tested the sensitivity to PQ of mutants that are diagnostic of a variety of pathways of aging. In particular mutants of genes that, based on their known functions in C. elegans or that of their homologues in other systems, might encode the targets of superoxide signaling or be otherwise necessary for implementing superoxide signaling. The c-Jun N-terminal kinase 1 (JNK-1) is involved in stress responses in vertebrate cells and is a positive regulator of DAF-16 that acts in parallel to the effect of daf-2 on daf-16 [65]. We treated jnk-1(gk7) mutants with PQ and obtained a particularly large lifespan increase (Table 1). Although it is not clear what activities lie upstream of jnk-1 in C. elegans nor whether it has other targets than daf-16, its activity does not appear necessary for the effect of superoxide. The transcription factor SKN-1 defends against oxidative stress by mobilizing the conserved phase II detoxification response and can delay aging independently of DAF-16 [66]. Although PQ induces oxidative stress and induces enzymes that protect from oxidative stress (Figure 3), it was still able to prolong the lifespan of skn-1(zn67) mutants (Table 1), indicating that skn-1 does not act downstream of superoxide. wwp-1 encodes a conserved E3 ubiquitin ligase that is necessary for lifespan extension by dietary restriction [67]. Treatment of wwp-1(ok1102) with PQ prolonged lifespan of these mutants, which is consistent with our finding that PQ can considerably extend the lifespan of eat-2 mutants (Figure 4E). This confirmed that the lifespan increase produced by the superoxide increase in mitochondrial mutants is distinct from the mechanisms that support the lifespan effects of dietary restriction [14]. hif-1 encodes a worm homologue of the vertebrate hypoxia inducible factor 1α (HIF-1α), a transcription factor involved in a number of protective mechanisms. In C. elegans hif-1 is necessary for a lifespan pathway that involves proteolysis and that is distinct from insulin signaling [68] and has also been involved in the dietary restriction pathway [69]. In vertebrates HIF-1α is positively regulated by mitochondrial ROS [34],[35], which would make it an interesting candidate to mediate the effects of superoxide. However, PQ was fully capable of increasing the lifespan of the hif-1 mutants (Table 1).
Several of the genes whose mutants remain sensitive to PQ, including daf-16, have been involved in stress responses, including oxidative stress, yet they do not seem necessary for the effect of PQ. Similarly we have shown previously that although the expression of SOD-1 and SOD-2 are elevated in isp-1(qm150) mutants, the elevation is not necessary for the extended lifespan of these mutants [24]. nuo-6(qm200) mutants also show elevated SOD-1 and SOD-2 expression [14], but this too is unnecessary for the longevity of the mutants, as RNAi against sod-1 an sod-2, which we have shown to be efficient in reducing enzyme levels [24], does not shorten the lifespan of nuo-6 mutants (Figure S5). We conclude that the mitochondrial mutants protect from an aspect of the aging process that has not yet been studied through mutants that affect stress. In addition, our observations suggest that the lifespan effect we observed is not hormetic, as neither superoxide-detoxifying enzymes, nor the regulatory factors that are involved in protection from oxidative stress, are crucially implicated.
We have shown previously that mutations in isp-1 and nuo-6 prolong lifespan by a common mechanism [14]. Using direct measurement of ROS and superoxide we find here that this mechanism involves an increase in mitochondrial superoxide generation that is necessary and sufficient for the longevity of these mutants. As ROS, including superoxide [70],[71],[72], are known to be intracellular messengers, the increased superoxide might trigger a signal transduction pathway that ultimately results in changes in nuclear gene expression [23]. Superoxide is highly reactive and could trigger such a signal by modifying proteins in the mitochondria or in the nearby cytosol after having escaped from the mitochondria through an appropriate channel [73],[74]. Although no superoxide sensor has yet been identified, a similar type of mechanism, in which a highly reactive, quickly diffusing, molecule modifies a signal transduction protein, has been evidenced for nitric oxide (NO), which covalently and permanently modifies guanylyl cyclases. Similarly, hydrogen peroxide (H2O2), the product of superoxide dismutation, can inactivate phosphatases involved in signal transduction. Future work will aim at using forward and reverse genetic screens in C. elegans to uncover the molecular machinery that reacts to the superoxide signal, as well as the transcription factors that are needed to regulate nuclear gene expression in response to the pathway's activation. In addition, the pattern of gene expression that results in increased lifespan in these mutants could be defined very specifically by identifying changes in the gene expression patterns that are common to isp-1, nuo-6, and PQ treatment and that are suppressed by treatment with NAC.
A number of studies in C. elegans have explored hormesis by treating animals with sub-lethal but clearly deleterious treatments for a short period of time and observing subsequent prolongation of lifespan [75]. These hormetic effects are different from what we have observed and describe here, as both the genetic mutations and the very low level PQ are present throughout life and as only a part of the effect we observe might require the insulin signaling pathway. Furthermore, although in nuo-6 and isp-1 mutants the expression levels of the superoxide dismutases SOD-1 and SOD-2 are elevated, likely in response to the elevated superoxide generation, and as one expects in the hormetic response, these elevations are not necessary for the lifespan prolongation of nuo-6 (Figure S5) or isp-1 [24].
CLK-1 is a mitochondrial protein that is required for ubiquinone biosynthesis and its absence affects mitochondrial function [50], although it could potentially affect many other processes as ubiquinone is found in all membranes. Furthermore, ubiquinone is both a prooxidant as co-factor in the respiratory chain and an anti-oxidant. Interestingly, the mechanism of lifespan prolongation induced by clk-1 appears to be entirely distinct from, but particularly synergistic with, that induced by elevated superoxide. Indeed, clk-1 mutants do not show elevated superoxide generation and are not affected by NAC. Furthermore, although double mutant combinations of clk-1 with nuo-6 and isp-1 are not viable (unpublished data) the lifespans of clk-1 mutants treated with PQ (Figure 4D), or of sod-2;clk-1 mutants [26], or of clk-1;daf-2 mutants [76] are much greater than expected from simple additivity of the effects of individual mutations or treatments.
Studies in yeast [77] and in worms [78] have suggested that an increase in ROS from mitochondria might also be important in triggering the lifespan extension produced by glucose restriction. However, our results here with an eat-2 mutation, one of the ways in which global dietary restriction can be produced in worms, as well as with a wwp-1 and hif-1, which may function downstream of dietary restriction, did not reveal an involvement of superoxide signaling, providing further evidence for a distinction between the mechanisms of glucose restriction and dietary restriction. It remains possible, however, that DR could lead to superoxide or ROS production when it is induced by other methods than the use of an eat-2 mutant, as it is well documented that different types of DR induce different molecular mechanisms [79].
One question that our current experiments do not address is whether the mitochondrial dysfunction in the mutants, or the effect of PQ, is necessary in every tissue in order to increase longevity. There are indications for both the insulin signaling pathway mutants [80],[81] and dietary restriction [67],[82] that the entire effect might be mediated by action in particular cells that influence the physiology of the whole organism. Similarly, the presence or absence of the germline is sufficient for a dramatic effect on lifespan [83]. For mitochondrial dysfunction the question could be addressed in the future by mosaic analysis and by purifying and analyzing mitochondria from specific tissues using our flow cytometry technique to purify mitochondria expressing GFP in a tissue-specific manner.
The oxidative stress theory of aging has been one of the most acknowledged theories of aging for the simple reason of the strikingly good correlation between the levels of oxidative stress and the aged phenotype [8]. A number of recent results in worms and in mice, however, have suggested that oxidative stress cannot be the cause of aging [24],[25],[26],[30]. Our findings suggest a conceptual framework for why oxidative stress and the aged phenotype are so tightly correlated [31]. In this model mitochondria, like the rest of the cell, sustain a variety of age-dependent insults (not only and not even principally from oxidative stress) that trigger an increase in superoxide, which acts as a signal that induces general protective and repair mechanisms. However, aging in most animals is clearly irreversible, indicating that the protective mechanisms, which must have evolved to control damage in young organisms, are unable to fully prevent the accumulation of age-related damage. Thus, as superoxide is a reactive molecule as well as a signal, and as age-dependent damage cannot be fully reversed, it is possible that at high ages the chronically elevated superoxide will participate in creating some of the damage itself. This could explain the strong tendency for aged animals to have high oxidative stress and high oxidative damage, although it does not imply that ROS cause aging or even that they are a major source of age-dependent damage. In this model, the nuo-6 and isp-1 mutations lead to increased longevity because they turn on the stress signal prematurely and thus slow down the entire process.
Eggs were placed on plates at 20°C and left for 1 h to hatch. Larvae that had hatched during that period were placed onto fresh plates and monitored once daily until death. The animals were transferred once daily while producing eggs to keep them separate from their progeny. Animals were scored as dead when they no longer responded with movement to light prodding on the head and tail. Missing worms and worms that have died because of internal hatching (bagging) were replaced from a backup group. Survival was scored every day.
Drugs were added into NGM media from a high concentration stock solution (500 mM for NAC, 1 M for PQ, and 500 mM for vitamin C) before pouring of the plates. Plates were made fresh each week. Gravid adult worms were transferred from normal NGM plates to drug plates and left to lay eggs for 3 h. With each transfer of worms a substantial amount of bacteria was also transferred onto the new plates. The progeny was then scored for different phenotypes.
All dyes except MitoSox were diluted in DMSO at high concentration (all at 5 mM except H2DCFDA, which is at 10 mM) and frozen at −20°C as a stock. MitoSox was prepared fresh at 5 mM for each use. Before staining stocks were diluted in M9 buffer at a 1∶1000 dilution. Young adult worms were transferred into staining solution and stained for 20 min. Worms were mounted on a thick layer of half-dried agar pad on microscopic glass slides and then subjected to confocal microscopy (Zeiss LSM 510 Meta). Pictures were taken by Zeiss LSM Imaging software and analyzed by Volocity V4.0 software.
Five young adult worms (1st day of adulthood) were placed into 0.25 µl M9 buffer in a 0.5 µl sealed chamber at 22°C. A fiber optical oxygen sensor (AL300 FOXY probe from Ocean Optics) was inserted into this chamber and oxygen partial pressure was monitored for 15 to 30 min. Oxygen consumption measured in this way was normalized to body volume. For this worms were photographed at each measurement day under a binocular microscope and their cross-section was calculated with ImageJ software. Worm volume was determined by the formula: volume (nl) = 1.849 • 10–7 (nl/µm3) • area 1.5 (µm3) [84].
After RNAi treatment, 100 young adult worms of each genotype were picked, lysed in 2× loading buffer, and subjected to electrophoresis in 12% SDS–polyacrylamide gels (SDS–PAGE), and then blotted onto nitrocellulose membrane (Bio-Rad). After applying primary antibody (1∶1000, rabbit polyclonal antibody against worm SOD-1 or SOD-2) and secondary antibody (1∶10,000 mouse anti-rabbit IgG, Invitrogen), the membranes were incubated with the ECL plus detection reagent (Amersham Biosciences) and scanned using a Typhoon trio plus scanner. Band densities were analyzed by ImageQuant TL V2003.03.
For fluorescence activated cell sorting [37], adult worms grown on large NGM plates were collected and washed 3 times with M9 buffer. Worms were then suspended in 5× isolation buffer (200 mM mannitol; 120 mM sucrose; 10 mM Tris; 1 mM EGTA; pH 7.4) and set on ice. Worms were broken up with a 5 ml glass-glass homogenizer and centrifuged at 600 g for 10 min, the supernatant was collected and re-centrifuged at 7,800 g for 10 min, and the pellet was washed once with isolation buffer and then suspended in isolation buffer and kept on ice. Different dyes were added from stocks into the analysis buffer (250 mM sucrose; 20 mM MOPS; 100 uM KPi; 0.5 mM MgCl2; 1 uM CsA pH 7.0) at a 1∶1000 dilution before staining. 100 µl of mitochondria was added to 900 µl of analysis buffer with dye and substrate and incubated for 1 h at room temperature. Mitochondria were recollected by 7,800 g centrifugation and then suspended in 500 µl analysis buffer. A FACSCalibur instrument equipped with a 488 nm Argon laser and a 635 nm red diode laser (Becton Dickinson) was used. Data from the experiments were analyzed using the CellQuest software (Becton Dickinson). To exclude debris, samples were gated based on light-scattering properties in the SSC (side scatter) and FSC (forward scatter) modes, and 20,000 events per sample were collected, using the “low” setting for sample flow rate (Figure S1).
200 age-synchronized young adult worms were collected in M9 buffer and washed three times. Worm pellets were treated with three freeze/thaw cycles and boiled for 15 min to release ATP and destroy ATPase activity, and then spun at 4°C and 11,000 g for 10 min. ATP contents were measured with a kit (Invitrogen, Carlsbad, California, USA; Cat: A22066). The ATP content value was then normalized to the soluble protein level of the same preparation, measured with the protein assay from Bio-Rad.
Mitotracker green (Invitrogen M7514) stock concentration 5 mM; H2DCFDA (Invitrogen D399) stock concentration 10 mM; Mitotracker red (Invitrogen M7512) stock concentration 5 mM.
|
10.1371/journal.ppat.1005233 | Targeting HIV Reservoir in Infected CD4 T Cells by Dual-Affinity Re-targeting Molecules (DARTs) that Bind HIV Envelope and Recruit Cytotoxic T Cells | HIV reservoirs and production of viral antigens are not eliminated in chronically infected participants treated with combination antiretroviral therapy (cART). Novel therapeutic strategies aiming at viral reservoir elimination are needed to address chronic immune dysfunction and non-AIDS morbidities that exist despite effective cART. The HIV envelope protein (Env) is emerging as a highly specific viral target for therapeutic elimination of the persistent HIV-infected reservoirs via antibody-mediated cell killing. Dual-Affinity Re-Targeting (DART) molecules exhibit a distinct mechanism of action via binding the cell surface target antigen and simultaneously engaging CD3 on cytotoxic T lymphocytes (CTLs). We designed and evaluated Env-specific DARTs (HIVxCD3 DARTs) derived from known antibodies recognizing diverse Env epitopes with or without broadly neutralizing activity. HIVxCD3 DARTs derived from PGT121, PGT145, A32, and 7B2, but not VRC01 or 10E8 antibodies, mediated potent CTL-dependent killing of quiescent primary CD4 T cells infected with diverse HIV isolates. Similar killing activity was also observed with DARTs structurally modified for in vivo half-life extension. In an ex vivo model using cells isolated from HIV-infected participants on cART, combinations of the most potent HIVxCD3 DARTs reduced HIV expression both in quiescent and activated peripheral blood mononuclear cell cultures isolated from HIV-infected participants on suppressive cART. Importantly, HIVxCD3 DARTs did not induce cell-to-cell virus spread in resting or activated CD4 T cell cultures. Collectively, these results provide support for further development of HIVxCD3 DARTs as a promising therapeutic strategy for targeting HIV reservoirs.
| Current HIV therapies prevent AIDS by dramatically reducing, but not eliminating, HIV infection. A reservoir of HIV-infected cells persists during long-term antiviral therapy, and individuals are at increased risk to develop non-AIDS illnesses, e.g., accelerated heart, bone, or kidney disease. Novel strategies are thus needed to safely kill HIV-infected cells and reduce or eliminate the HIV reservoir. An emerging strategy to kill HIV-infected cells involves antibodies (Abs) that bind the HIV envelope protein (Env). Env can distinguish HIV-infected cells from uninfected cells, and some Env-specific Abs can kill HIV-infected cells by recruiting immune cells, e.g., NK cells and macrophages. Here, we developed a strategy to kill HIV-infected cells that is complementary to Env-specific Abs. We designed and evaluated Dual-Affinity Re-Targeting (DART) molecules that incorporate Env-binding specificities with a CD3-binding specificity to recruit and activate cytotoxic T cells. We report that HIVxCD3 DARTs potently and selectively kill HIV-infected cells. Furthermore, HIV DARTs perturb resting and activated viral reservoirs in cells isolated from individuals on antiviral therapy. This novel strategy may be an important element of future antiviral therapies that target the HIV reservoir.
| Advanced regimens of combination antiretroviral therapy (cART) prevent AIDS and suppress HIV replication to nearly undetectable levels in over 90% of treatment-naïve participants [1–3]. However, in nearly all cases, cART interruption results in resumption of viral replication [4,5], which indicates that current cART is not sufficient to eliminate the HIV reservoir and cure persistent infection. The ability of HIV to establish latency in a subset of infected CD4 T cells limits the ability of cART to reduce the reservoir [6]. Latency is characterized by the presence of integrated but transcriptionally silent proviral HIV DNA, which makes the infected cells invisible to the immune system and resistant to innate antiviral defenses [6,7].
Proviral DNA has been detected in multiple immune cell subsets that are permissive to HIV infection, but the best characterized reservoir exists in long-lived resting memory CD4 T cells [7,8]. The rare pool of latently infected memory CD4 T cells capable of producing infectious virus upon activation is believed to be maintained by homeostatic proliferation of memory T cells and/or intermittent antigen-driven clonal expansion [9]. Low levels of HIV replication confined to lymphatic tissues and undetectable in the periphery may also contribute the HIV reservoir [10,11]. Additionally, there is evidence that persistently infected cells capable of expressing low but detectable levels of HIV protein exist [12,13]. Herein, the HIV reservoir is defined to encompass: latently infected cells that are transcriptionally silent, persistently infected cells that express HIV protein basally, and cells that can be activated to increase expression of HIV protein. The extended decay rate of HIV reservoirs in peripheral blood lymphocytes indicates that life-long treatment with current cART regimens is unlikely to cure HIV infection [7].
Despite the success of cART in reducing viremia, HIV can be detected in participants on suppressive cART using sensitive single-copy assays [14]. Antiviral drugs do not prevent viral antigen expression in HIV-infected cells, which may contribute to chronic immune activation and inflammation in participants on cART [15–17]. Together, persistent HIV infection and associated immune dysfunction increase the long-term risk for non-AIDS morbidities including accelerated cardiovascular disease, liver and renal disease, non-AIDS-associated cancers, and neurocognitive impairment [18–20]. Thus, therapeutic interventions are needed that could substantially reduce or eliminate the HIV reservoirs or, alternatively, lead to host-mediated control of HIV without cART [10]. One proposed strategy is to combine pharmacologic activation of latent HIV expression with immune-mediated elimination of infected cells. Various classes of latency reversal agents such as HDAC inhibitors or TLR7 agonists have demonstrated the ability to activate the quiescent reservoir and increase viral gene expression ex vivo and/or in vivo [21,22]. HIV envelope protein (Env) is an attractive target for immune-mediated killing of infected cells because it is unique to the virus. Potent broadly neutralizing anti-Env IgG antibodies (bNAbs) with preserved Fc-dependent effector function have provided preliminary in vivo evidence for reservoir reduction [23,24]. Bi-specific antibodies that combine an anti-Env arm with an anti-CD3 arm to recruit CTLs are an alternative strategy to IgG-mediated killing. This strategy could recruit CTLs with any T cell receptor specificity to selectively kill HIV-infected cells.
Bi-specific antibodies have demonstrated the ability to potently kill low-frequency target cells in humans, e.g., residual disease in B-cell lymphoma [25,26]. Several bi-specific platforms are in various stages of clinical testing, primarily for oncology-based therapeutic applications. Bi-specific T-cell engagers (BiTEs) represent one of the most advanced bi-specific antibody platforms, as blinatumomab (CD19xCD3 BiTE) was recently approved for the treatment of acute lymphoblastic leukemia [27]. An alternative platform, termed Dual-Affinity Re-Targeting (DART) currently has several constructs in clinical development for oncology indications: MGD006, a CD123xCD3 DART, is being evaluated in a Phase 1 trial in patients with refractory acute myeloid leukemia [28] (NCT02152956); MGD007, a gpA33xCD3 DART in MP3 format for enhanced pharmacokinetic (PK) properties, is being evaluated in a Phase 1 trial in patients with colorectal cancer (NCT02248805); MGD011, a CD19xCD3 DART in MP3 format, is being evaluated in a Phase 1 trial in patients with B-cell hematological malignancies (NCT02454270). DART molecules also are being pursued to attenuate autoimmune disorders: MGD010, a CD32BxCD79B DART in MP3 format, is designed to simultaneously bind both targets on individual B cells and inhibit their activation; it is being tested in a Phase 1 trial in normal volunteers (NCT02376036).
To examine the potential of employing the DART platform to treat participants with HIV, we describe here the design and characterization of a series of HIVxCD3 DARTs derived from diverse broadly reactive anti-Env monoclonal antibodies that are non-neutralizing or neutralizing. The DART constructs induced a potent and specific CD8 T cell-dependent elimination of primary resting CD4 T cells infected with multiple HIV isolates in vitro. In addition, HIVxCD3 DARTs were capable of reducing the level of virion production ex vivo in cells isolated from infected participants on suppressive antiretroviral therapy. Together, the generated data support further development of these bi-specific T-cell redirecting molecules for targeting the HIV reservoirs in cART-treated participants.
A series of bi-specific antibody constructs that bind simultaneously to HIV Env and human CD3 receptor were generated using the basic DART platform (Fig 1A) [29,30]. To maximize the breadth of Env recognition across multiple HIV isolates, complementarity determining regions (CDRs) from four bNAbs (PGT121, PGT145, 10E8, VRC01) [31–33] were incorporated into the Env-recognizing arm of HIVxCD3 DARTs. These bNAb-derived DARTs were compared with ones with HIV Env arms derived from two non-neutralizing antibodies (A32, 7B2) that bind broadly conserved residues in Env and efficiently induce antibody dependent cell-mediated cytotoxicity (ADCC) [34–41]. Each of the six designed bi-specific Abs recognizes a distinct epitope on the surface of HIV Env protein (Fig 1B). CDRs derived from palivizumab [42], an antibody recognizing the fusion protein of respiratory syncytial virus (RSV), were used to construct a negative control DART (RSVxCD3) that does not bind to HIV Env. The CD3-recognizing arm, which was identical in all DARTs, was derived from hXR32 [28], a humanized mouse anti-human CD3ε antibody, which cross-reacts with nonhuman primate CD3ε. The HIVxCD3 DARTs in basic format were produced by expression in stably transfected CHO cells and purified. The formation of properly assembled molecules was confirmed by reducing and non-reducing SDS-PAGE and analytical SEC; the average purity of the assembled HIVxCD3 DART molecules was 95%.
DART binding via the CD3 arm to soluble recombinant human CD3 receptor was equivalently efficient for all HIVxCD3 DARTs as well as for the RSVxCD3 control (Fig 2A). DART binding via the HIV Env arm to soluble recombinant HIV JRFL Env (gp140) monomer ranged from efficient (7B2, VRC01, PGT121) to less efficient (A32) to undetectable (PGT145, 10E8) (Fig 2B). These patterns were recapitulated in the bi-specific binding assay, which measures simultaneous binding to JRFL gp140 and human CD3 (Fig 2C). Weaker gp140 binding by the A32 arm is consistent with the CD4-inducible nature of its epitope [34,36], as binding was assessed in the absence of CD4. Lack of gp140 monomer binding by the PGT145 arm is consistent with the quaternary nature of its epitope (V1-V2 loop), which exists only in mature Env trimers [32]. Lack of gp140 monomer binding by the 10E8 arm may be due to the dependence of its epitope, located in the membrane-proximal external region (MPER), on the fusion-intermediate conformational state of gp41 [44]. In general, the binding of the HIVxCD3 DARTs to JRFL gp140 mimicked that of the corresponding parental IgGs (Fig 2D). All 6 HIVxCD3 DARTs exhibited binding to CM244 and/or 92Th023 Env (gp140) presented on the surface of Env-transfected HEK293 cells (S1 Fig). DARTs with A32, 7B2, VRC01, or 10E8 arms bound efficiently to both cell lines, while DARTs with PGT145 or PGT121 arms bound only to the CM244 Env-expressing cell line (moderately for the PGT145 arm and weakly for the PGT121 arm). These latter binding patterns appear to reflect attenuated recognition of one or both of these particular Env isolates by the PGT121 in the DART format, because the parental antibodies bind efficiently to cells expressing Env from other HIV isolates (S1B Fig) [45–47]. In summary, our data demonstrate that all 6 HIVxCD3 DARTs exhibit binding to CD3 and Env (in the form of monomeric gp140 and/or cell surface Env), and the differential binding pattern across different systems is likely related to the ability of specific DARTs to recognize Env antigens from different HIV strains.
The DARTs were evaluated in a FACS-based cytotoxicity model to assess their ability to mediate CD8 T cell-dependent killing of HIV-infected cells [48]. A resting CD4 T cell in vitro model of HIV infection was developed to create targets that are not actively dividing, as latently infected cells are likely to be in a resting state in vivo [49,50]. Briefly, unstimulated primary CD4 T cells purified from healthy human participants’ peripheral blood mononuclear cells (PBMCs) were spinfected with replication-competent HIV isolates. After 6 to 7 days in culture, typically 1% to 3% of the total CD4 population expressed intracellular p24 and surface Env proteins (Fig 3A). HIV-infected cells exhibited a resting phenotype, indicated by a low level of CD69 surface expression (Fig 3B). Furthermore, HIV-infected cells exhibited decreased expression of CD4, consistent with Nef-mediated CD4 down-regulation [51]. Detection of surface Env expression by FACS required biotinylation of a high-affinity anti-Env mAb (e.g., PGT121), suggesting that the HIV-infected target cells in this model have relatively low levels of Env expression, which we consider a key attribute for relevant testing of the Env-targeting mechanisms. Addition of cART at the time of infection inhibited the expression of HIV proteins at day 6 post-infection, indicating that the FACS detected de novo viral protein expression rather than the presence of virions from the inoculum on the surface of CD4 T cells (Fig 3C).
To generate effector cells, autologous unstimulated CD8 T cells were purified from the same participant’s PBMCs as the CD4 T cells infected with HIV. CD8 T cells were then co-cultured with HIV-infected CD4 T cells at varying ratios and duration in the presence or absence of HIVxCD3 DARTs ranging from 0.6 pM to 2,000 pM. The percentage of DART-induced cytotoxicity was determined by measuring the reduction in CD8- p24+ cells with an active (HIVxCD3) or control (RSVxCD3) DART compared with a no DART control (S2 Fig). HIVxCD3 DART-mediated killing of HIV-infected CD4 T cells plateaued at a CD8:CD4 T cell ratio of 2:1 after a co-culture period of 72 hours (S3 Fig). The kinetics of DART-dependent CD8 T-cell killing of HIV-infected CD4 T cells are similar to those reported for bispecific molecules that redirect resting T-cells at low E:T ratios against other targets [52,53]. HIVxCD3 DARTs-mediated killing of HIV-infected CD4 T cells required the presence of CD8 T cells, and the control RSVxCD3 DART did not induce any appreciable killing of HIV-infected CD4 T cells (S2 Fig). HIVxCD3 DARTs did not mediate killing of uninfected p24-negative CD4 T cells (S2 Fig). To confirm that the reduction in percent p24-positive CD4 T cells mediated by HIVxCD3 DARTs was indicative of death of HIV-infected cells, both cell-associated vRNA and vDNA were measured in cell pellets that were collected at the same time as p24 measurements. Reduction in the percent p24-positive CD4 T cells correlated well with the reduction in the number of vRNA-positive and vDNA-positive cells co-cultured with HIVxCD3 DARTs but not with RSVxCD3 control DART (S4 Fig). The reduction in percent p24-positive CD4 T cells was slightly larger than the reduction in cell-associated vRNA or vDNA, which may indicate that a fraction of HIV RNA and DNA-positive cells express sufficient Env protein to be recognized and killed by HIVxCD3 DARTs. Analysis of cell culture media confirmed that DARTs remained stable under the cell killing assay conditions (S5 Fig).
HIVxCD3 DART constructs derived from PGT121, 7B2, and A32 demonstrated potent killing of CD4 T cells infected with each of the three tested diverse CCR5-tropic HIV-1 isolates with divergent envelope sequences: BaL, IN/93/905 (IN), and 92/RW/008 (RW) (Fig 4). A32xCD3 showed the most consistent potency for inducing redirected killing activity with EC50 values of 4.2 to 5.4 pM across all three HIV isolates (Table 1). On the other hand, PGT121xCD3 induced more consistent maximal killing activity with Emax values of 86–93% (Table 1). PGT145xCD3 potently killed IN and RW, but was less potent against BaL. These findings are consistent with the ability of the parental PGT145 IgG to potently neutralize IN and RW but not BaL (S6 Fig). In contrast to these HIVxCD3 DARTs, 10E8xCD3 and VRC01xCD3 did not potently kill CD4 T cells infected with the HIV isolates that were tested. The parental 10E8 IgG and VRC01 IgG did bind HIV BaL-infected cells (S1B Fig). In addition, VRC01 IgG neutralized each HIV isolate tested, while 10E8 IgG weakly neutralized two of the three isolates (S6 Fig). In summary, HIVxCD3 DARTs containing CDRs from PGT121, PGT145, A32, or 7B2 mediated picomolar killing at levels > 90% in a model with primary unstimulated CD4 T cells infected with multiple HIV isolates.
Combinations of HIVxCD3 DARTs were profiled with the goal of maximizing recognition of diverse Env antigens expected to be present in participants infected with various HIV strains and subtypes. Pairwise combinations of PGT121xCD3, A32xCD3, and 7B2xCD3 were evaluated against BaL- and IN-infected cells. These three DART constructs performed well individually and would not be expected to compete for binding to Env, as they recognize spatially distinct Env epitopes (Fig 1B). For PGT121xCD3 paired with either A32xCD3 or 7B2xCD3, the combined potency and maximal level of killing observed for both HIV isolates were not substantially different from the effect of PGT121xCD3 alone, while the A32xCD3 plus 7B2xCD3 combination was slightly less effective against the CD4 T cells infected by the HIV IN isolate (Fig 5). For any of the tested pairs, there was no apparent synergistic benefit of the combination, a result that may be due to the very high potency and level of killing already achieved with each HIVxCD3 DART alone.
The in vitro model of HIV infection used above reproduces an important aspect of HIV reservoirs in that it uses resting CD4 T cells, but it cannot fully recapitulate the latency that is found in participants treated with suppressive cART. Importantly, HIV-infected cells derived from participants on cART are likely to be less frequent and express lower levels of HIV proteins, if any, compared with the resting cells infected in vitro. These biological differences make it critical to evaluate the efficacy of HIVxCD3 DARTs ex vivo using cells isolated from HIV-infected participants with prolonged virus suppression by cART. We used two different models to determine whether DARTs can impact HIV-infected cells isolated from participants. In the first model, we determined whether HIVxCD3 DARTs could reduce HIV virion production from unstimulated cells from HIV-infected cART-suppressed participants. This resting ex vivo model was designed to represent how HIVxCD3 DARTs may be initially evaluated alone in vivo, similar to the Phase 1 trial with bNAb 3BNC117 [54]. In the second model, we tested whether treatment with HIVxCD3 DARTs in combination with a latency reversal agent can affect the response of the viral reservoir to re-stimulation with the same agent. This stimulated model was designed to represent how HIVxCD3 DARTs could be evaluated in combination with a compound that enhances HIV protein expression.
PBMCs were isolated from cART-treated HIV-infected participants whose viral load was undetectable for a minimum of 12 months. PBMCs were used to evaluate DARTs ex vivo rather than purified CD4 and CD8 T cells. PBMCs may more closely mimic the biological diversity of the reservoir and ratios of effector T cells, and PBMCs do not require additional purification steps that reduce the yield of limited materials. In the unstimulated ex vivo model, a combination of PGT121xCD3 and 7B2xCD3 DARTs or the RSVxCD3 control DART was added to PBMCs in the absence of any activating agent. Cultures were maintained in the presence of antiretroviral agents (ARVs) to prevent HIV transmission to uninfected cells, and levels of supernatant vRNA were quantified by qRT-PCR after 8 days and 14 days of culture (S7 Fig). Supernatant vRNA was measured as a functional indication of virion production to assess the impact of DARTs on the HIV reservoir. Using this model, we have previously demonstrated that supernatant HIV RNA can be pelleted by ultracentrifugation, indicating that this HIV RNA is primarily contained within virions [55]. On Day 8 of culture, HIVxCD3 DARTs reduced vRNA supernatant levels by 29% to 49% compared with RSVxCD3 control DART in 4 of 4 participants. These reductions did not reach statistical significance (Fig 6A). By Day 14, HIVxCD3 DARTs reduced supernatant vRNA levels by 26% to 92% in 4 of 4 participants. In 3 participants, these reductions did reach statistical significance (Fig 6B). These ex vivo data suggest that a portion of the HIV reservoir may express sufficient basal levels of Env to enable targeting by HIVxCD3 DARTs.
In the stimulated ex vivo model, a combination of two HIVxCD3 DARTs or the control RSVxCD3 DART were added to PBMCs that were untreated or activated with the PKC agonist indolactam, which robustly activated HIV in all of the participants tested (S8A Fig). After 6 days, CD4 T cells were isolated from PBMCs and activated with indolactam for an additional 3 days (S7 Fig). This re-stimulation model was designed to measure the effect that HIVxCD3 DARTs may have had on the inducible HIV reservoir. In 2 of 4 participants that were tested, the HIVxCD3 DARTs significantly reduced the level of supernatant HIV RNA that was induced following the re-stimulation with indolactam, while the control DART had no appreciable effect on the re-stimulation of inducible reservoir (Fig 6C). In 4 additional participants that were tested in this model, the initial stimulation with indolactam was sufficient, in the absence of DARTs, to reduce the level of supernatant HIV RNA induced by subsequent re-stimulation with indolactam. The addition of HIVxCD3 DARTs or control DART did not demonstrate additional benefit to reduce the inducible reservoir response in these participants (S8B Fig). Taken together, the results from these ex vivo studies indicate that HIVxCD3 DARTs are capable of targeting either unstimulated or stimulated primary HIV-infected cells from a subset of virologically suppressed participants.
The MP3 DART format was developed to prolong the short circulating half-life of basic format DARTs [28]. MP3 DARTs contain a human IgG1 Fc domain that has been mutated (L234A/L235A) to inactivate effector function via binding to FcγRs and/or complement, while retaining binding to the neonatal FcR (FcRn) to engage the IgG salvage pathway (Fig 7A) [56,57]. The MP3 DART format would likely be preferable for clinical applications, as it would reduce dosing frequency while maintaining optimal exposure levels.
A total of four HIVxCD3 DARTs containing Env binding CDRs derived from A32, 7B2, PGT121, and PGT145 were constructed in MP3 format, produced in CHO cells and purified. The formation of properly assembled molecules was confirmed by reducing and non-reducing SDS-PAGE and analytical SEC; the average purity of the assembled HIVxCD3 MP3 DART molecules was 96%. The binding properties of these four HIVxCD3 MP3 DARTs and the matching control RSVxCD3 MP3 DART were evaluated. Binding to recombinant CD3 and JRFL gp140 were slightly reduced for the MP3 DARTs compared with the corresponding basic DARTs (S9A, S9B and S9C Fig). Similarly, binding to the surface of HEK293 cells expressing HIV CM244 env was slightly reduced with the MP3 DARTs compared with the same corresponding basic DARTs (S9D and S9E Fig). However, when administered to human FcRn transgenic mice, a model that offers reliable predictions of antibody pharmacokinetics in monkeys and humans [58,59], the A32xCD3 MP3 DART exhibited a major improvement in serum half-life and exposure compared with a DART in basic format, which was cleared in only a few hours (Fig 7B). The pharmacokinetic parameters for the A32xCD3 MP3 DART in human FcRn transgenic mice approximate those observed with IgG1 molecules.
HIVxCD3 DARTs in MP3 and basic format were compared for their ability to mediate redirected CD8 T cell-dependent killing of HIV-infected cells. PGT121xCD3 DARTs in basic and MP3 format were compared side-by-side in the in vitro model of resting CD4 T cell infection using cells prepared from three independent participants. Importantly, both DART formats exhibited similar potency and maximum elimination of HIV-infected cells (Fig 8). These results support further investigation of the extended half-life MP3 DART format.
A bi-specific Ab that simultaneously targets HIV Env and CD3 expressed on CD4 T cells has the potential to enhance cell-to-cell spread of HIV by binding Env on an HIV-infected cell and CD3 on an adjacent uninfected CD4 T cell. This interaction could activate uninfected cells and, in turn, make them more susceptible to cell-to-cell viral transmission. To evaluate the potential for HIVxCD3 DARTs to activate CD4 T cells, we measured the expression of the cell surface activation markers CD25, CD69, and HLA-DR on the uninfected p24-negative and HIV-infected (p24-positive) CD4 T cells following HIV infection in the presence of DARTs. In uninfected and HIV-infected CD4 T cells, PGT121xCD3 MP3 DART increased the frequency of CD69-positive and CD25-positive CD4 T cells, but not HLA-DR-positive CD4 T cells compared with the RSVxCD3 MP3 control DART (S10A Fig). To evaluate the potential for this partial activation to contribute to cell-to-cell HIV spread, we developed two in vitro infection models using primary CD4 T cells that were either unstimulated or activated. In an activated CD4 T cell spreading infection model, the PGT121xCD3 MP3 or A32xCD3 MP3 DART increased neither the frequency of HIV-infected cells nor virus production following incubation for 5 days. On the contrary, PGT121xCD3 MP3, but not A32xCD3 MP3, profoundly reduced the fraction of p24-positive cells in the activated model of cell-to-cell transmission, a finding that is consistent with the neutralizing ability of PGT121 (S10B Fig). These data indicate that HIVxCD3 DARTs derived from bNAbs can exert a direct antiviral effect in the absence of effector CD8 cells likely by neutralizing virus or blocking cell-to-cell spread. In the resting model of HIV infection, there was no evidence for enhanced viral replication in the presence of PGT121xCD3 MP3 when compared with the RSVxCD3 MP3 control DART (S10C Fig). Taken together, these results indicate that HIVxCD3 DARTs are unlikely to pose a risk for enhancing the spreading of HIV infection, and the constructs derived from neutralizing antibodies may, in fact, protect uninfected cells.
Despite the established clinical efficacy of suppressive cART, HIV reservoirs persist throughout the life of treated participants. In longitudinal studies, HIV sequence stability suggests that cART effectively prevents active virus replication and spread [60]. cART does not eliminate HIV-infected cells or prevent HIV antigen expression, and persistent HIV antigen exposure may contribute to chronic immune dysfunction and accelerated non-AIDS diseases [61]. Therefore, there is a clear unmet medical need to eliminate HIV reservoirs. Reduction of reservoirs to levels that restore normal immune function may also provide a health benefit. Several HIV cure strategies target HIV Env because it is selective and may be exposed on the surface of infected cells. Env-targeting bNAbs have been able to reduce viral reservoirs in preclinical models of persistent HIV infection, as measured by declines in proviral DNA and delays in viral rebound [23,24,62,63]. bNAbs have the potential to eliminate HIV-infected cells via FcγR-mediated effector functions that engage immune effector cells, such as monocytes, macrophages, natural killer cells, and neutrophils [64–67]. The success of bNAbs in animal studies has stimulated interest in conducting investigational clinical trials aimed at targeting Env in HIV-infected participants using either passively administered IgG, e.g., VRC01, sustained in situ production of bNAbs by AAV-vectored approaches, e.g. PG9, or vaccination approaches designed to elicit bNAbs [68].
CD3-targeted bi-specific DARTs are antibody-based molecules that elicit a cell-mediated killing mechanism that is distinct from Fc effector-competent bNAbs. HIVxCD3 DART molecules specifically redirect cytotoxic T cells to Env-expressing target cells and induce their lysis. This process involves simultaneous binding to the surfaces of HIV-infected cells and CD3-expressing polyclonal T cells by the anti-Env and anti-CD3 arms, respectively. Bi-specific antibody-mediated redirected T cell killing of target cells is concomitant with effector T cell activation, proliferation, and upregulation of granzyme B and perforin in a target-dependent manner [28,69], which may prime CTL for serial cytotoxicity [53]. These bi-specific T-cell redirecting molecules are effective in vivo at doses many-fold lower than those typically employed for mAbs [70]. For example, blinatumomab, a CD19xCD3 BiTE, has been shown to be clinically potent and efficacious with an acceptable safety profile when administered at doses of 28 μg/day and was approved in 2014 for the treatment of relapsed or refractory B-precursor acute lymphoblastic leukemia [26]. In nonhuman primate studies with DARTs, treatment with a CD123xCD3 basic DART at doses ranging from 0.1 to 1.0 μg/kg/day resulted in profound depletion of circulating CD14- CD123+ cells, the intended target cells [28], and treatment with a CD19xCD3 MP3 DART at doses of 5–10 μg/kg/week resulted in profound depletion of CD19+ cells in the periphery and in lymphoid tissues [57]. While anti-drug antibody (ADA) responses against DARTs have been observed in these cynomolgus monkey studies, the frequencies appear to be comparable to those observed with other human or humanized monoclonal antibodies in monkeys, and, importantly, it is well established that immunogenicity in cynomolgus monkeys is not predictive of immunogenicity in human subjects [71]. The immunogenicity of DARTs for oncology and autoimmune indications in human subjects is being monitored in multiple clinical studies (NCT02152956, NCT02248805, NCT02376036, NCT02454270).
DARTs have inter-chain disulfide bonds at their C-termini and are structurally compact and well suited for forming stable cell-to-cell contacts between CTL and target cells. Additionally, DARTs exhibit greater potency than BiTEs in side-by-side comparisons [30,72]. The enhanced potency of bi-specific DARTs may be particularly relevant in regards to HIV reservoirs, which represent a low-frequency target that likely express Env at low densities [73]. Furthermore, the importance of CD8 CTL in controlling the HIV reservoir is evident in elite controllers, who have demonstrated the ability to suppress viremia in the absence of cART [48,74,75]. Unlike therapeutic HIV vaccine strategies that can only enhance HIV-specific CTL, bi-specific HIVxCD3 DARTs can redirect polyclonal CTL to kill HIV-infected cells. This is an important distinction because HIV-specific CTL in some cART-treated participants may have an anergic or senescent phenotype characterized by defects in cytotoxic function [76,77]. This would also be useful in cases where viral epitopes have evolved to escape HIV-specific CTL killing, as is typically the case for cART-suppressed HIV participants initially treated during chronic infection [78].
Here, we report that HIVxCD3 DARTs with different Env specificities elicit potent and specific elimination of HIV-infected cells. For these studies, we utilized an in vitro model based on the infection of unstimulated primary CD4 T cells with wild-type HIV isolates [12,50]. As opposed to dividing cell lines or mitogen-stimulated latency models, a resting primary cell model may better approximate the resting state and corresponding low levels of surface Env expressed on reservoir cells from HIV-infected participants on cART. Primary unstimulated CD4 T cells may also respond to killing signals from effector cells in a more relevant manner than activated CD4 T cells. For these same reasons, the effector cells used in this model were unstimulated, autologous CD8 T cells. Multiple Env-specificities were evaluated in our in vitro model to determine whether the spatial location of the Env epitope might influence the efficiency of redirected lysis. Bi-specific constructs were derived from both bNAbs and broadly reactive, non-neutralizing Abs. These classes likely recognize different Env forms, and it is unclear whether neutralization activity is preferred for recognizing cellular membrane forms of Env and inducing efficient redirected lysis [43,79]. bNAbs selected for these studies recognize spatially distinct Env domains [80] and are some of the broadest, most potent mAbs available (PGT121, V3-glycan; PGT145, V1/V2 loop [32]; VRC01, CD4bs [81]; 10E8, gp41 [82]). DARTs with bNAbs-derived arms may preferentially, or in some cases exclusively, e.g., PGT145, bind the mature Env trimers [31,83].
In our in vitro HIV infection model, PGT121- and PGT145-derived HIVxCD3 DARTs exhibited higher maximal killing and more potent killing of infected cells than those derived from VRC01 and 10E8 (Fig 4 and Table 1). VRC01xCD3 and 10E8xCD3 DARTs were able to bind Env-transfected cells (S1A Fig), and in single cycle neutralization assays, the parental VRC01 IgG, but not 10E8 IgG, demonstrated potent neutralization of the HIV isolates tested in the cytotoxicity assays (S5 Fig). Neutralization of an HIV isolate by the parental IgG may not correlate with corresponding DART cytolytic activity, given that virions and infected cells may express different forms and levels of Env. Recognition and binding of Env is required, but it may not be sufficient for potent DART-mediated killing. For example, VRC01 and 10E8 IgGs each bound Env on HIV-infected cells (S1B Fig), suggesting that spatial conformation of certain Env epitopes, and the corresponding geometry of the DART-mediated synapses that form, may determine cytolytic activity in some cases. In our model, DARTs targeting the V1V2 domain or the V3 glycan domain of Env induced more efficient redirected lysis than DARTs targeting the CD4 binding site or MPER domains. However, it is premature to generalize these results before testing additional antibodies from each Env domain class against a larger number of HIV isolates.
In contrast to bNAbs, HIVxCD3 DARTs with arms derived from broadly reactive, non-neutralizing mAbs, such as A32 and 7B2, may be expected to preferentially recognize their epitopes in the context of nonfunctional forms of Env [43,46,79,84]. For example, 7B2 preferentially binds gp41 when gp120 dissociates [82], and A32 binds to a CD4-induced epitope that is exposed on the functional Env trimer only after CD4 binding during entry [40]. However, the A32 epitope (C1 domain) is expressed on the surface of infected cells early and thus may be an efficient ADCC epitope [37]. In the RV144 vaccine trial, for example, A32 blocked ADCC activity in 96% of cases where ADCC was induced, suggesting that potentially protective ADCC responses were directed to epitopes preferentially exposed on non-functional Env forms [85,86]. Nonfunctional Env forms may in fact be the predominant Env expressed on infected cells, as fully cleaved, trimeric forms accounted for only 10% of total cellular Env for 3 HIV isolates examined [82]. This may partially explain why DARTs derived from the non-neutralizing Abs 7B2 and A32, which that target nonfunctional Env forms, bound to HEK293-D371 and HEK293-D375 with higher MFIs than DARTs derived from the bNAbs PGT121 and PGT145 (S1 Fig). Notably, A32xCD3 and 7B2xCD3 DARTs both exhibited potent and robust killing of HIV-infected primary resting CD4 T cells that was comparable to those of PGT121xCD3 and PGT145xCD3 DARTs (Fig 4 and Table 1).
Given the ability of HIVxCD3 DARTs to kill HIV-infected cells in vitro, we evaluated DARTs ex vivo using PBMCs isolated from HIV-infected cART-treated participants with suppressed viral load. PBMCs may better represent the physiological diversity of HIV reservoir and the ratio of target and effector cells. Additionally, material from HIV participants is always limited, and use of PBMCs reduces loss of cells during isolation of individual cell subsets, enabling more robust execution of key experiments by including a larger number of replicates. Ex vivo models have potential advantages compared to in vitro models. For example, HIV-infected cells ex vivo may express more biologically relevant levels and forms of Env. These cells may also respond differently to cytotoxic signaling than cells infected in vitro. In addition, effector CD8 T cells from HIV-infected participants on cART may have altered functions relative to CD8 T cells from healthy individuals. Reduction of cell-associated proviral vDNA is likely a more definitive indication of HIV-infected cell death than reduction in supernatant vRNA. However, available data suggest that it may not be biologically feasible to reduce vDNA in most subjects. For example, sequence data suggest that 88.3% of HIV proviruses in the reservoir of cART-suppressed participants are defective, and maximal activation of resting CD4 T cells in vitro induced infectious virus from <1% of proviruses [87]. In a subsequent study, only 1.5% of proviruses were induced by mitogen to produce virion, as measured by supernatant vRNA [88]. There are also technical difficulties that limit the ability to demonstrate small reductions of vDNA in rare cell populations by quantitative nucleic assays. Prior to ex vivo studies, pair-wise HIV DART combinations were evaluated in vitro to ensure that combinations predicted to increase coverage of diverse strains would not be antagonistic (Fig 5). Given the distinct and complimentary binding properties of bNAbs and broadly reactive, non-neutralizing Abs discussed above, the PGT121xCD3 + 7B2xCD3 combination was selected for ex vivo testing in two models using unstimulated CD4 T cells and CD4 T cells treated with a potent latency reversal agent. Prior to ex vivo evaluation, we demonstrated in vitro that this HIVxCD3 DART combination was capable of reducing cell-associated p24, vRNA, and vDNA (S4 Fig). In the unstimulated ex vivo model, this HIVxCD3 DART combination effectively reduced the level of supernatant vRNA in 3 of 4 participants tested compared with the control DART (Fig 6B). These findings suggest that basal levels of Env expression on unstimulated CD4 T cells may be sufficient for HIVxCD3 DART-mediated reservoir reduction. This result is consistent with the reduction in cell-associated vDNA that was observed in SHIV-infected non-human primates (NHP) treated with ARVs in combination with parental PGT121 [23]. For the stimulated ex vivo model, we selected the PKC agonist indolactam, as it effectively activated HIV in all of the participants tested (S8 Fig). While mitogens, e.g. anti-CD3/anti-CD28 or PMA/ionomycin, may be more effective LRAs than PKC agonists, mitogens can also induce T cell proliferation leading to potential reservoir expansion and/or resistance to DART-mediated cell killing. Our ex vivo stimulated model was designed to mimic conditions similar to those potentially employed in vivo to evaluate HIVxCD3 DARTs. LRAs with mitogenic activity cannot be used in vivo for safety reasons and it is unlikely that an LRA would be available for clinical testing with a level of HIV activation matching that of mitogenic activators. Importantly, PKC agonists as a class have demonstrated the ability to strongly and consistently activate HIV without inducing T cell proliferation [89–92]. In our stimulated ex vivo model, we observed that indolactam treatment combined with HIVxCD3 DARTs reduced the subsequent reactivation of reservoir compared with the control DART in 2 of 4 participants who did not demonstrate a reduced re-stimulation with indolactam alone (Fig 6C). This stimulated model may be less sensitive to reductions if latent viruses that were not activated by the first stimulus are stochastically activated by the second stimulus, as previous work suggests [87]. Additionally, a PKC agonist may not be the optimal latency reversal agent to combine with HIVxCD3 DARTs for reservoir reduction. For example, PKC activation alone may enhance T cell survival by inhibiting apoptosis [93,94]. Additional studies are therefore needed to select optimal latency reversal agents to combine with HIVxCD3 DARTs to enhance elimination of HIV-infected cells. Taken together, results from our ex vivo models demonstrate reduction of virus production that is suggestive of infected cell killing, but ultimate proof of the reservoir reduction would have to be obtained by in vivo testing of DARTs.
In summary, we have demonstrated potent and Env-specific HIVxCD3 DART-mediated killing of HIV-infected cells in vitro and reduction of viral protein expression ex vivo. HIVxCD3 DARTs that target either broadly neutralizing Env epitopes or broadly reactive, non-neutralizing Env epitopes were effective, as was an extended half-life MP3 DART format. Taken together, these results provide support to evaluate this platform in an animal model of HIV latency to determine whether the HIV reservoir can be safely reduced in vivo, as was recently demonstrated by PGT121 IgG in SHIV-infected NHPs.
HIV-infected participants were enrolled into the study at the Quest Clinical Research (QCR) in San Francisco, CA. The study was approved by the Western Institutional Review Board. Informed, written consent was obtained from participants prior to any study procedures.
HIV-infected participants participating in the study were selected based on sustained plasma viral load suppression (<50 copies/mL for >12 months), CD4 counts (>350 cells/mL), and absence of co-infection with hepatitis B or C virus.
HEK293-D371 and HEK293-D375 cell lines with doxycycline-inducible expression of HIV CM244 (subtype AE) gp140 and 92Th023 (subtype AE) gp140, respectively, were obtained from Dr. John Kappes (University of Alabama at Birmingham); cells were maintained in complete RPMI 1640, 20% fetal bovine serum (FBS), 1% Pen/Strep, and 1 μg/mL doxycycline was added for at least 1 day to induce Env expression. Primary cell isolation and culture and HIV-1 isolates used for in vitro cytotoxicity assays are subsequently described.
Basic HIVxCD3 DARTs consist of two covalently linked polypeptide chains: Chain 1: CD3VL-HIVVH-ASTKG-E-coil, and Chain 2: HIVVL-CD3VH-ASTKG-K-coil. The oppositely charged E/K-coil domain [95], located at the carboxyl terminus of each chain and containing an interchain disulfide bond, drives heterodimer formation. HIVxCD3 MP3 DARTs consist of three covalently linked polypeptide chains: Chain 1: CD3VL-HIVVH-ASTKG-E-coil-Fc, Chain 2: HIVVL-CD3VH-ASTKG-K-coil, and Chain 3: Fc. The Fc (human IgG1) sequence was modified by point mutations (L234A/L235A) to greatly reduce or abolish binding to activating FcγRs and complement [56]. Chains 1 and 2 form a heterodimer by virtue of the E/K-coil dimerization domain and interchain disulfide bond. Chains 1 and 3 are covalently linked by two disulfide bonds in the Fc hinge region. HIV arms were based on the VH and VL sequences of the following anti-Env mAbs: A32, 7B2, PGT121, PGT145, VRC01, and 10E8 (GenBank accession numbers listed at the end of the Materials and Methods). The CD3 arm was derived from hXR32, a humanized mouse anti-human CD3ε mAb, which cross-reacts with CD3ε from cynomolgus and rhesus macaques [28,57]. A control DART was similarly constructed by replacing the HIV arm with an irrelevant specificity from palivizumab (an anti-RSV mAb) [42].
For the basic DARTs, Chain 1 and Chain 2 coding sequences were cloned into the bicistronic CET1019AD UCOE vectors (EMD Millipore), transfected into CHO cells to generate stable cell lines, and the basic DART proteins were purified as described previously [29]. For the MP3 DARTs, Chain 1 and Chain 2 coding sequences were cloned into a modified CET1019AD vector that contains a neomycin resistance gene, and Chain 3 sequence into the monocistronic CET1019AS UCOE vector. The two plasmids were co-transfected into CHO cells for the generation of stable cell lines. The MP3 DART proteins were purified by affinity chromatography using Protein A Sepharose and followed by SEC when necessary. Approximate size and homogeneity of purified DART proteins in basic or MP3 format were analyzed by SDS-PAGE (NuPAGE Bis-Tris gel system; Invitrogen) and analytical SEC (TSK GS3000SWxL SE-HPLC; Tosoh Bioscience).
DART binding to antigen proteins was measured by ELISA. For monospecific binding assays, a MaxiSorp microtiter plate (Nunc) coated with recombinant protein (human CD3ε/δ heterodimer or JR-FL gp140ΔCF) in bicarbonate buffer was blocked with 3% BSA and 0.1% Tween-20. DART proteins were applied, followed by sequential addition of biotinylated anti-EK coil antibody (recognizes the E/K heterodimerization region of DART proteins) and streptavidin-HRP (BD Biosciences). For bi-specific DART binding assays, the plate was coated with JRFL gp140ΔCF, and DART application was followed by sequential addition of biotinylated CD3ε/δ and streptavidin-HRP. For binding assays with anti-Env IgGs, the plate was coated with JR-FL gp140ΔCF, and IgG application was followed by sequential addition of biotinylated anti-human IgG1 Fc antibody and streptavidin-HRP. HRP activity was detected with SuperSignal ELISA Pico chemiluminescent substrate (Thermo Scientific).
DART binding to cell lines expressing HIV-1 Env was measured by flow cytometry. DARTs at 4 μg/mL were incubated with 105 cells in 200 μL FACS buffer containing 10% human AB serum for 30 minutes at room temperature. After washing, cells were resuspended in 100 μL of 1 μg/mL biotin-conjugated mouse anti-EK antibody, mixed with 1:500 diluted streptavidin-PE and incubated in the dark for 45 minutes at 2–8°C. Cells were washed, resuspended with FACS buffer, and analyzed by flow cytometry using a FACSCalibur (BD Biosciences) and FlowJo software (TreeStar).
PBMCs from healthy participants (AllCells) were isolated by Ficoll-plaque gradient. Total CD4 T cells were isolated from PBMCs using an EasySep Human CD4+ T cell Enrichment Kit (Stemcell Technologies). CD4 T cells (5 × 107) were infected with HIV-1 strain BaL or with the HIV-1 clinical isolates HIV 92/RW/008 or HIV IN/93/905 (NIH AIDS Reagent Program). Infection was done with 50–100 ng p24/million CD4 T cells by spinfecting at 1200 ×g for 2 hours [49,50]. Cells were incubated for 5 days at 37°C in RPMI plus 10% FBS with 30 U/mL IL-2 (Invitrogen). PBMCs drawn and isolated on the same day were frozen in 90% heat-inactivated FBS and 10% DMSO. PBMCs were thawed 1 day prior to co-culture, and cells were rested overnight in media at 37°C. On the day of co-culture, CD8 T cells were isolated from thawed PBMCs using an EasySep Human CD8+ T cell Enrichment Kit (Stemcell Technologies). CD8 T cells were co-cultured with infected CD4 T cells at a CD8 T cell:CD4 T cell ratio of 2:1 with varying concentrations of DARTs for 3 days at 37°C.
PBMCs were isolated by Ficoll-plaque gradient from leukapheresis samples of HIV-infected participants treated with cART and resuspended at 3 million cells/ml in RPMI with 10% FBS, penicillin/streptomycin, and 100 nM elvitegravir/100 nM efavirenz to prevent new rounds of infection. For experiments with unstimulated cells, combination of 200 pM PGT121xCD3 and 200 pM 7B2xCD3 DART or 400 pM RSVxCD3 alone (negative control) DART were added to cells and incubated in a 37°C, 5% CO2 incubator for 14 days. Every 3 to 4 days, media was removed and added back with the appropriate DARTs and antivirals. For experiments with stimulated cells, a combination of 200 pM PGT121xCD3 and 200 pM 7B2xCD3 DART or 400 pM RSVxCD3 alone (negative control) DART were added together with 1 μM indolactam, and cultures were incubated in a 37°C, 5% CO2 incubator for 7 days. Total CD4 T cells were then purified by negative selection from each culture using an EasySep Human CD4+ T cell Enrichment Kit (Stemcell Technologies). CD4 T cells were plated in 24-well plates at 1 million cells/mL in 2 mL of media containing 100 nM elvitegravir and 100 nM efavirenz and incubated for 3 days. To measure HIV RNA levels, plates were spun at 500 ×g for 5 min, and 1 mL of culture supernatant was analyzed by a robotic COBAS Ampliprep/Taqman system (Roche Diagnostics), which extracts total nucleic acid and quantifies HIV RNA in copies per milliliter using the HIV Test, v2.0 kit (Roche Diagnostics).
Total CD4 T cells from healthy participants’ PBMCs were purified by negative selection using EasySep magnetic beads (Stemcell Technologies). 50×106 CD4 T cells were infected with 50 ng-100 ng p24/ml of lab strain BaL by spinfecting at 1200 ×g for 2 hours [48,49]. Cells were washed twice post spinfection and incubated at 37°C with 30 U/mL IL-2 (Invitrogen). MP3 DARTs were added 24 hours post spinfection at varying concentrations. Cells were stained 72 hours post addition of DARTs with live/dead Fixable Aqua Dead Cell Stain (Invitrogen), then with APC-labeled antibody to CD25 (BD Biosciences), PE-labeled antibody to CD69 (BD Biosciences), and v450-labeled antibody to HLA-DR (BD Biosciences). Cells were fixed and permeabilized with Intracellular Fixation & Permeabilization Buffer Set (eBioscience), stained with PE labeled antibody to Gag p24 (Beckman Coulter), and analyzed by flow cytometry using a LSRFortessa (BD Biosciences) and FlowJo software (TreeStar).
Total CD4 T cells were isolated from healthy participants’ PBMCs using EasySep Human CD4+ T cell Enrichment Kit. Cells were divided into 2 aliquots post isolation, where one half of the cells were blasted with 5 μg/mL PHA (Sigma Aldrich) and 100 U/mL IL-2 (Invitrogen) at 2x106 cells/mL for 3 days at 37°C. The other half remained in culture with 30 U/mL IL-2 at 37°C. Three days post activation, cells were infected with 50–100 ng p24/million cells of lab strain BaL by spinfecting at 1200 × g for 2 hours [48,49]. Cells were washed twice post spinfection. The unstimulated portion of cells was labeled with Cell Trace CFSE (Invitrogen) on day 5 and co-cultured with infected blasted cells at 1:1 ratio with or without MP3 DARTs. Cells were stained 48 hours post addition of DARTs with live/dead Fixable Aqua Dead Cell Stain, then with APC-labeled antibody to CD25, PE-labeled antibody to CD69, and v450-labeled antibody to HLADR. Cells were fixed and permeabilized with Intracellular Fixation & Permeabilization Buffer Set (eBioscience), stained with PE labeled antibody to Gag p24 (Beckman Coulter), and analyzed by flow cytometry using a LSRFortessa (BD Biosciences) and FlowJo software (TreeStar).
Female mice (strain B6.Cg-Fcgrttm1Dcr Tg(FCGRT)276Dcr; Jackson Laboratories) were injected intravenously with A32xCD3 MP3 DART diluted in phosphate-buffered saline at a dose level of 5 mg/kg (total of 6 animals). Blood samples for serum were collected from subgroups of 3 animals per time point over a period of 21 days. Concentrations of DART in serum were quantitatively measured by ELISA with immobilized goat anti-hXR32 antibody, which recognizes the anti-CD3 domain (hXR32) of the DART, for capture and goat anti-human IgG Fc-biotin together with streptavidin-horseradish peroxidase (SA-HRP) for detection. The pharmacokinetic parameters were determined using WinNonlin software (Pharsight).
A32 [3TNM_H, 3TNM_L]; 7B2 [AFQ31502, AFQ31503], PGT121 [JN201894.1, JN201911.1], PGT145 [JN201910.1, JN201927.1], VRC01 [GU980702.1, GU980703.1] and 10E8 [JX645769.1, JX645770.1].
|
10.1371/journal.pgen.1000781 | Increased Expression and Protein Divergence in Duplicate Genes Is Associated with Morphological Diversification | The differentiation of both gene expression and protein function is thought to be important as a mechanism of the functionalization of duplicate genes. However, it has not been addressed whether expression or protein divergence of duplicate genes is greater in those genes that have undergone functionalization compared with those that have not. We examined a total of 492 paralogous gene pairs associated with morphological diversification in a plant model organism (Arabidopsis thaliana). Classifying these paralogous gene pairs into high, low, and no morphological diversification groups, based on knock-out data, we found that the divergence rate of both gene expression and protein sequences were significantly higher in either high or low morphological diversification groups compared with those in the no morphological diversification group. These results strongly suggest that the divergence of both expression and protein sequence are important sources for morphological diversification of duplicate genes. Although both mechanisms are not mutually exclusive, our analysis suggested that changes of expression pattern play the minor role (33%–41%) and that changes of protein sequence play the major role (59%–67%) in morphological diversification. Finally, we examined to what extent duplicate genes are associated with expression or protein divergence exerting morphological diversification at the whole-genome level. Interestingly, duplicate genes randomly chosen from A. thaliana had not experienced expression or protein divergence that resulted in morphological diversification. These results indicate that most duplicate genes have experienced minor functionalization.
| The relationship between morphological and molecular evolution is a central issue to the understanding of eukaryote evolution. In particular, there is much interest in how duplicate genes have contributed to morphological diversification during evolution. As a mechanism of functionalization of duplicate genes, differentiation of both gene expression and protein function are believed to be important. Although it has been reported that both expression and protein divergence tend to increase as a duplication ages, it is unclear whether expression or protein divergence in duplicate genes is greater in those genes that have undergone functionalization compared with those that have not. Here, we studied 492 duplicate gene pairs associated with various degrees of morphological diversification in Arabidopsis thaliana. Using these data, we found that the divergence of both expression and protein sequence were important sources for morphological diversification of duplicate genes. Although both mechanisms are not mutually exclusive, our analysis suggested that expression divergence is the minor contributor and protein divergence is the major contributor to morphological diversification. However, the expression or protein sequence of randomly chosen duplicate genes did not show significant divergence that resulted in morphological diversification. These results indicate that most duplicate genes experienced minor functionalization in the genome.
| Duplicate genes rarely exhibit de novo functions (neofunctionalization); more usually, the functions of the original gene are split into multiple functions among the duplicate genes (subfunctionalization) [1]–[5]. Such functionalization through gene duplication is considered to be an important source of diversification in complex organisms [6]. As a mechanism of functionalization in duplicate genes, differentiation of both gene expression and protein function are thought to be important. In particular, differential patterns of gene expression among paralogs are widely believed to play a prominent role in morphological diversification, because such differences are essential for development [7]–[10]. However, substantial amounts of data support morphological diversification through divergence of protein function [11].
Many researchers have studied divergence of either expression or protein function in duplicate genes at the genome scale [12]–[24]. Although divergence of either expression or protein sequence tends to increase as a duplication ages, it is unclear whether either expression or protein divergence in duplicate genes has been elevated by functionalization. Therefore, it is of interest to compare the divergence rate of either expression pattern or protein sequence of duplicate genes of the same age that have and have not undergone functionalization. If divergence of both expression and protein function are important sources for functionalization, the divergence rate of both should be higher in duplicate genes that have undergone functionalization compared with those that have not.
A. thaliana is an excellent model organism for addressing the above issue because it has a highly duplicated genome and many knock-out mutants have been generated. Here, to address how duplicate genes have contributed to morphological evolution, we classified Arabidopsis duplicate genes into high, low and no morphological diversification groups based on knock-out data, and examined the divergence rates of both expression pattern and protein sequence among the three morphological diversification groups.
From the literature and from our earlier work (see Materials and Methods) [25],[26] we identified 398 pairs of duplicate genes in which the knock-out mutant of either gene in a pair induced abnormal morphological changes relative to wild type. Abnormal morphological changes were classified into seed, vegetative and reproductive phenotypes on the basis of the definition of Meinke et al [27]. When the knock-out phenotype is totally different between genes in a paralogous gene pair, it is reasonable to assume that functionalization occurred after gene duplication (Figure 1A). For example, the knock-out mutant of AT4G09820 and AT5G41315 genes induced a yellow seed coat in the reproductive stage and a reduction of trichomes in the vegetative stage, respectively. Therefore, the knock-out phenotype is completely different between AT4G09820 and AT5G41315 because two abnormal phenotypes appeared in different developmental stages. Thus, paralogous genes with different phenotypes (morphological differences between phenotypes) are defined to have high morphological diversification. It is more common, however, to observe knock-out phenotypes that are similar or identical between paralogous genes (Figure 1B). For example, the knock-out mutants of AT1G62830 and AT3G10390 genes both induced late flowering. Although the knock-out phenotype of the two genes is similar, there would appear to be functionalization in such paralogous genes because a morphological change resulting from the deletion of one gene occurs when there is no or little functional redundancy between the paralogous genes. We, therefore, thought that such paralogous genes had some degree of functionalization after gene duplication. However, it is likely that similar or identical phenotypes indicate paralogous genes that have lower functionalization compared with paralogous genes with different phenotypes. Therefore, paralogous genes with either similar or identical phenotypes (morphological changes within phenotypes) were defined to have low morphological diversification. In this study, we identified 163 and 235 paralogous gene pairs associated with high and low morphological diversification, respectively. As a control set, we focused on paralogous gene pairs in which abnormal morphological changes are observed only upon the deletion of multiple paralogous genes but deletion of each gene separately did not induce abnormal morphological changes (Figure 1C). For example, the double knock-out mutant of AT3G58780 and AT2G42830 exhibits fruit dehiscence but knock-out of each gene alone did not induce abnormal morphological changes. Such paralogous gene pairs are likely to have some degree of functional redundancy. We, therefore, defined these paralogous gene pairs as having no morphological diversification. The number of paralogous gene pairs identified without morphological diversification was 94. Thus, we identified a total of 492 paralogous gene pairs associated with the three kinds of morphological diversification (Table S1).
To examine the expression pattern divergence for a paralogous gene pair, we obtained intensities of gene expression by microarray analysis under 634 conditions. Expression divergence in a pair of genes is usually inferred by 1 minus R (Pearson's coefficient of correlation) of the expression intensities among experimental conditions. Here, we transformed the value as log((1−R)/(1+R)), because the transformation is more sensitive for examining expression differences [19]. When we applied the log((1−R)/(1+R)) values to paralogous gene pairs among the three morphological diversification groups, the log((1−R)/(1+R)) values increased as morphological diversification increased (Figure S1). However, the relationship may be strongly influenced by duplication age (sequence divergence) in the case that morphological diversification increases as sequence divergence increases. We, therefore, investigated sequence divergence in paralogous gene pairs by examining synonymous (Ks) and nonsynonymous (Ka) distance among morphological diversification groups [28]. Consequently, both synonymous and nonsynonymous distances increased as morphological diversification increased (P<0.01 by Wilcoxon's test; Figure S1 and Table S2). To minimize the effect of duplication age, log((1−R)/(1+R)) was divided by Ks. This is because expression divergence is expected to increase as duplication timing becomes earlier and Ks increases in a nearly linear fashion with duplication age [17],[19],[24]. Ed (log ((1−R)/(1+R))/Ks) is an indicator of the expression divergence rate between a paralogous gene pair: high and low Ed indicates high and low expression divergence at the same duplication age, respectively. When we calculated Ed between a paralogous gene pair in the three morphological diversification groups, Ed increased as morphological diversification increased (Figure 2A). Ed differed significantly between each pair of morphological diversification groups (P<0.01 by Wilcoxon's test; Table S2), suggesting that expression divergence is an important source for morphological diversification of duplicate genes.
There are genetic and epigenetic factors that are the source of expression divergence. Since the differentiation of cis-regulatory elements can be a major genetic effect, we examined the proportion of known cis-regulatory elements that overlap in the promoter regions of paralogous gene pairs [29]. The proportion of cis-regulatory elements that overlap decreased as morphological diversification increased (Figure S2). The proportion of overlapping cis-regulatory elements differed significantly between each pair of morphological diversification groups (P<0.05 by Wilcoxon's test; Table S2 and Figure S2), indicating that the divergence of cis-regulatory elements contributes to morphological diversification. With respect to epigenetic factors, we investigated the proportion of methylated cytosines to non-methylated cytosines in the promoter regions of paralogous genes [30]. The proportional difference in paralogous gene pairs did not significantly differ between each pair of morphological diversification groups (Table S2 and Figure S2), indicating that an epigenetic effect through methylation is unlikely to contribute to morphological diversification. Taken together, expression divergence led by the differentiation of cis-regulatory elements is an important source for morphological diversification in duplicate genes.
Because duplication age (sequence divergence) between paralogous gene pairs increased as morphological diversification increased (Figure S1), we examined divergence rates of protein sequences of the same duplication age. Divergence rates of protein sequences are commonly inferred from selection pressure in coding sequences, i.e. the ratio of the non-synonymous substitution rate (Ka) to Ks. High and low Ka/Ks ratios indicate high and low protein divergence rates at the same duplication age, respectively [28]. When we applied the Ka/Ks ratio to paralogous gene pairs within the three morphological diversification groups, the Ka/Ks ratio increased as the morphological diversification increased (Figure 2B). The Ka/Ks ratio differed significantly between each pair of morphological diversification groups (P<0.01 by Wilcoxon's test; Table S2), suggesting that protein divergence is an important source for morphological diversification of duplicate genes.
To analyze the kinds of amino acid replacements that have occurred during morphological diversification, we classified all amino acid replacements as either ‘chemical radical’ or ‘conservative’ on the basis of an amino acid classification generated in an earlier report [31]. We examined the ratio of the radical nonsynonymous substitution rate (Kr) to the conservative nonsynonymous substitution rate (Kc). Interestingly, the Kr/Kc ratios of all types of paralogous gene pairs were similar (Figure 2C and Table S2), indicating that paralogous gene pairs with either high, low or no morphological diversification tend to have the same level of radical protein divergence. The Kr/Kc ratio based on this amino acid classification is significantly correlated with the Ka/Ks ratio at the whole genome level [31]. Therefore, radical changes become restricted in paralogous gene pairs with higher morphological diversification. One explanation for this restriction is that radical changes do not affect morphological diversification. However, some reports have shown that radical changes significantly influence functional divergence [23],[32]. Therefore, it does not seem to be a reasonable explanation. Another explanation is that radical changes may induce serious functional errors. To maintain duplicate genes that encode functional proteins, radical changes may be too deleterious. Therefore, paralogous gene pairs involved in higher morphological diversification may be subject to purifying selection against radical amino acid changes.
To compare the divergence rate of expression pattern with that of protein sequence in paralogous gene pairs associated with morphological diversification, we focused on paralogous gene pairs without morphological diversification because the divergence rate of expression pattern and/or protein sequence in these duplicate genes has little effect on morphological diversification. Therefore, the top 5% of Ed and Ka/Ks ratios for paralogous gene pairs without morphological diversification were defined to be the threshold of higher divergence rate of expression pattern and protein sequences, respectively. We then counted the numbers of paralogous gene pairs with a higher divergence rate in each of the high and low morphological diversification groups (Table 1). To make the relative roles clear, we simply compared the observed ratio between paralogous gene pairs with only higher expression divergence and those with only higher protein divergence, assuming no bias between expression and protein divergence in either high or low morphological diversification groups. Interestingly, the number of paralogous gene pairs (37 in either high or low morphological diversification groups) with a protein divergence but no expression divergence was significantly higher than the number of paralogous gene pairs (62 in either high or low morphological diversification groups) with a higher expression divergence but no protein divergence, as determined by the chi-square test (P<0.05). These results indicate that paralogous gene pairs with a higher divergence rate of protein sequence contribute to morphological diversification more effectively than those with a higher divergence rate of expression. The inference from these results is that protein sequence plays the major role (59–67%) and expression plays the minor role (33–41%) in morphological diversification.
We performed the same analysis using the top 10% of Ed and Ka/Ks ratios of paralogous gene pairs without morphological diversification as the threshold of higher divergence rate of expression pattern and protein sequences, and obtained essentially the same results (Table S3). Therefore, we believed that the relative rates of expression and protein divergence are stringent in morphological diversification.
Finally, we addressed to what extent duplicate genes were associated with expression or protein divergence exerting morphological diversification at the whole genome level. To examine this question, we randomly chose 1000 pairs of paralogous gene pairs. We then compared Ed and Ka/Ks ratios among the 1000 random paralogous gene pairs and among paralogous gene pairs with high, low or no morphological diversification (Figure 2). Both Ed and Ka/Ks ratios for the random paralogous gene pairs were significantly lower compared with that for the paralogous gene pairs with high or low morphological diversification but were significantly higher compared with that for the paralogous gene pairs without morphological diversification (P<0.01 by Wilcoxon's test, (Figure 2A and 2B and Table S2). However, the Kr/Kc ratio was not different between any pair in the four categories (P>0.05 by Wilcoxon's test, Figure 2C and Table S2). As discussed earlier, the Kr/Kc ratio is not an indicator for functionalization, therefore, no difference is reasonable. These results suggest that duplicate genes have not experienced divergence of expression or protein sequence exerting morphological diversification on a genome-wide scale. It is, therefore, likely that most duplicate genes have experienced only minor functionalization, at least in A. thaliana.
To understand to what extent molecular changes in duplicate genes have contributed to morphological diversification in A. thaliana, we examined the divergence rate of either expression pattern or protein sequence in duplicate genes associated with morphological diversification and found that both divergences are important sources in morphological diversification. Although both mechanisms are not mutually exclusive, our analysis suggested that changes of protein sequence play the major role and changes of expression pattern play the minor role in morphological diversification. However, randomly chosen duplicate genes have not experienced divergence of expression or protein sequence exerting morphological diversification. These results indicate that most duplicate genes have experienced minor functionalization and only a few duplicate genes are likely to be crucial to morphological evolution.
We used data from the available literature and from our bank of previously generated T-DNA insertional mutants [25],[26], to identify 1203 duplicate genes whose knock-out induced abnormal morphological changes relative to wild type. The nucleotide sequences of A. thaliana (TAIR7) were obtained from TAIR (www.arabidopsis.org). Duplicate genes were defined as proteins that matched other proteins in a BLAST search with E<1×10−4 [33]. We then classified the 1203 duplicate genes into 786 gene families by the Markov clustering algorithm (http://micans.org/mcl/). In every pair of each family, we examined the amino acid identity and the coverage (percentage of alignable regions). We found 405 paralogous gene pairs with amino acid identity >0.3 and coverage >0.5. Since tandem duplicates have a higher chance of exhibiting similar expression due to leaky expression or conserved sequences by gene conversion than non-tandem duplicates [34]–[36], we removed tandem duplicates from the 405 paralogous gene pairs. As reported earlier [37], tandem duplicates were defined as genes in any gene pair, T1 and T2, that (1) belong to the same gene family, (2) are located within 100 kb of each other, and (3) are separated by at most 10 nonhomologous (not in the same gene family as T1 and T2) genes. In this definition, we identified 7 tandem paralogous gene pairs. After removing these tandem paralogous gene pairs, we used 398 non-tandem paralogous gene pairs in this study. Note that each knock-out mutant of paralogous genes induced abnormal phenotypic changes.
To examine the degree of morphological diversification between the genes of the paralogous gene pairs, we classified morphological changes into seed, vegetative and reproductive phenotypes, according to the definition of Meinke et al [27]; the changes were defined as high (morphological changes between phenotypes) and low (morphological changes within phenotypes) morphological diversification. Briefly, seed, reproductive and vegetative phenotypes show visible changes in development. We identified 163 paralogous gene pairs associated with high morphological diversification and 235 associated with low divergence (Table S1).
As a control set, we identified from the literature165 duplicate genes that did not show morphological diversification. Absence of morphological diversification was defined as the observation of morphological change only upon the deletion of multiple paralogs; deletion of each gene separately did not induce morphological change. After removing tandem paralogous gene pairs, we found 95 paralogous gene pairs with amino acid identity >0.3 and coverage >0.5 (Table S1).
We obtained Affymetrix ATH1 data from the AtGenExpress expression atlas at TAIR (http://www.arabidopsis.org/). We compiled 1280 microarray datasets under 634 conditions, consisting of 82 different developmental stages, 72 biotic treatments, 285 abiotic treatments, 11 nutrient treatments, 81 hormone treatments, 40 chemical treatments, 21 cell cycle stages and 42 different genotypes. The array intensities were processed with the Bioconductor (http://www.bioconductor.org) affy package in the R software environment (http://www.r-project.org). Specifically, the array intensities were adjusted to reduce background with the mas5 function, and the normalize quantiles function was used for between-array normalization. The background-corrected and background-normalized intensities were used for further analysis.
We obtained the mapping data of known cis-regulatory elements in 1 kb promoter regions of all A. thaliana genes at ATCOECIS (http://bioinformatics.psb.ugent.be/ATCOECIS/) [29]. To examine the divergence of cis-regulatory elements in each paralogous gene pair, we used the proportion of overlapping cis-regulatory elements (the number of overlapping cis-regulatory elements over the number of observed cis-regulatory elements). To examine divergence of methylation in paralogous gene pairs, we obtained the mapping data of bisulfite-treated DNA sequences in the TAIR7 genome at NCBI Gene Expression Omnibus (GSM276809) [30]. The bisulfate-treatment converts cytosine to uracil in unmethylated cytosine sites but does not affect cytosine in methylated cytosine sites. Since the methylation of each cytosine site was determined multiple times, a methylated cytosine site was defined when that site is more often methylated than not. We calculated the proportion of methylated cytosine sites (the number of methylated cytosine sites over the number of observed cytosine sites) in promoter regions (500 bp upstream from either start codon or transcriptional start site) of all A. thaliana genes because the methylation of 500 bp upstream regions is considered to be sensitive for gene expression [30]. The proportional difference of methylated cytosine sites in a paralogous gene pair was used to represent the methylation divergence in a paralogous gene pair.
Nucleotide sequences of A. thaliana (TAIR7) were obtained from TAIR (www.arabidopsis.org). Pairwise alignment was performed with the program CLUSTALW to align coding regions [38]. Ks and Ka between paralogous genes were estimated by the modified Nei–Gojobori method [28]. The transition/transversion ratio was estimated for each paralogous gene pair, and the ratio was then used to estimate Ka and Ks. To infer the ratio of the radical non-synonymous substitution rate (Kr) to the conservative non-synonymous substitution rate (Kc), we classified amino acids according to Hanada et al. 2007 [31]. Radical and conservative changes were defined as amino acid replacements between and within groups, respectively. The ratio of Kr to Kc for each paralogous gene pair was estimated by the Zhang method [39].
We randomly chose genes from the total set of annotated A. thaliana genes (TAIR7). For a chosen gene, similarity searches were conducted against all annotated A. thaliana genes using BLASTP [33]. We aligned the chosen gene and all homologous genes identified in the BLASTP search using CLUSTALW and estimated the amino acid similarity among them [38]. We calculated the amino acid identity and the coverage (percentage of alignable regions) between the chosen gene and the matched gene with the highest identity. If the paralogous gene pair had amino acid identity >0.3 and coverage >0.5, we added the pair to a random set. We repeated this procedure until we obtained 1000 paralogous gene pairs.
|
10.1371/journal.pgen.1000569 | A Strand-Specific RNA–Seq Analysis of the Transcriptome of the Typhoid Bacillus Salmonella Typhi | High-density, strand-specific cDNA sequencing (ssRNA–seq) was used to analyze the transcriptome of Salmonella enterica serovar Typhi (S. Typhi). By mapping sequence data to the entire S. Typhi genome, we analyzed the transcriptome in a strand-specific manner and further defined transcribed regions encoded within prophages, pseudogenes, previously un-annotated, and 3′- or 5′-untranslated regions (UTR). An additional 40 novel candidate non-coding RNAs were identified beyond those previously annotated. Proteomic analysis was combined with transcriptome data to confirm and refine the annotation of a number of hpothetical genes. ssRNA–seq was also combined with microarray and proteome analysis to further define the S. Typhi OmpR regulon and identify novel OmpR regulated transcripts. Thus, ssRNA–seq provides a novel and powerful approach to the characterization of the bacterial transcriptome.
| We have applied a novel, strand-specific variation of RNA–seq (ssRNA–seq) to an analysis of the prokaryotic enteric pathogen Salmonella enterica serovar Typhi, the causative agent of Typhoid fever. Strand-specific data facilitated a high-resolution analysis of RNA transcription at a whole genome level with base-pair resolution. Using this technique, we were able to resolve overlapping transcripts of many genes, identify novel small RNAs, improve the accuracy of annotation, verify operon structure, and identify both transcriptionally active and inactive regions. We have compared the ssRNA–seq approach to standard RT–PCR and microarrays, validating the data. ssRNA–seq was used to redefine the OmpR operon that contributes to the pathogenicity of Typhi, identifying several novel OmpR regulated genes and operons. Finally, we have linked the ssRNA–seq data to the proteome and have provided simple open-access informatics tools to simplify interrogation of the data.
| DNA sequencing has been exploited to determine the whole genome sequence of hundreds of prokaryotic and eukaryotic species, facilitating gene identification, transcriptomics and the linkage of genotype to phenotype. To date, genome-wide analysis of the transcriptome has relied to a significant degree on the use of DNA microarrays. However, recent advances in DNA sequencing technologies have facilitated the determination of nucleotide sequence with a genomic read depth several orders of magnitude greater than was previously possible. These novel sequencing approaches have been successfully applied to studies on the transcriptome of eukaryotic genomes [1],[2] and to chromatin immunoprecipitation analysis [3],[4].
Bacterial genomes are relatively small and have a high density of coding sequences (CDS) in comparison to most eukaryotes. For example, the chromosome of Salmonella enterica serovar Typhi (S. Typhi), the causative agent of typhoid fever, is ∼4.8 Mbp in length, with ∼4,700 genes currently defined in available annotation [5],[6]. S. Typhi, unlike most Salmonella serotypes that have a broad host range and are associated with localised gastroenteritis, are host adapted (human restricted) and cause systemic infections. The genome of S. Typhi harbours many interesting features including horizontally acquired genetic islands specific to this serotype and ∼220 pseudogenes. These pseudogenes are potentially inactivated but many are intact in related host-promiscuous serovars such as S. Typhimurium [7]. S. Typhi express a polysaccharide, known as the Vi capsule, which is encoded on a large composite element known as Salmonella pathogenicity island (SPI)-7 that resembles a conjugative transposon [8]. S. Typhi also harbours several putative prophage elements, some of which are absent from S. Typhimurium and E. coli [9]. Prophages can encode virulence-associated cargo genes, which are not essential for phage viability [9].
In this study we exploit a novel ssRNA-seq method to identify the transcriptional template strand for both coding and non-coding sequences of S. Typhi Ty2 at a whole genome level using Illumina-platform high-throughput sequencing. We have identified many putative novel small non-coding RNAs (ncRNAs) and characterised mRNA expressed by pseudogenes. Strand-specific analysis has facilitated the re-annotation of a number of genes and by combining transcriptomic with proteome analyses, we have validated the expression of previously hypothetical genes. Further, we quantify differences in gene transcription in an ompR mutant and identify novel regions under control of this virulence-associated locus.
In order to characterise the S. Typhi transcriptome using ssRNA-seq, RNA was prepared from S. Typhi Ty2 grown to mid-log phase in LB broth. Since 16S and 23S rRNA was anticipated to be the most abundant RNA species these were depleted prior to sequencing by oligonucleotide hybridisation-mediated selective capture and separation using magnetic beads. The depleted RNA was reverse transcribed to cDNA and sequenced on an Illumina GAI. The resulting 36-base reads were mapped to the S. Typhi Ty2 genome. The sequence coverage per base was subsequently plotted and visualised using the genome browser Artemis and DNAplotter, [10],[11] (Figure 1A). To generate the transcript map each base on each strand of the genome was assigned a value derived from the alignment of sequence reads generated from each S. Typhi cDNA sample (Table 1). The method employed yields RNA transcript reads in a strand-specific manner and is particularly powerful because it can be used to identify small RNAs and to resolve transcripts originating from overlapping DNA sequences in a manner not possible using low-density microarrays.
To facilitate analysis of the S. Typhi Ty2 transcriptome we calculated the arithmetic mean per base-pair (AM) of mapped sequence reads for the predicted coding strand of the genes currently annotated on the genome and subtracted that of the putative non-coding strand to identify outliers (Figure 1B). Ninety one percent of the reads mapped to the previously annotated S. Typhi Ty2 coding strand, providing evidence for successful deconvolution of the nucleotide sequence in a strand-specific manner. This is almost certainly an underestimate of the strand specificity of this method as the remaining 9% of sequences mapped to the unannotated strand were either upstream of a CDS encoded nearby on the opposite strand, putatively identifying riboswitches and promoter regions, or mapped to unannotated or potentially mis-annotated regions. Examples of analyses where strand specific reads were readily identified are shown in Figure 2. As previously reported for ssRNA-seq analysis of eukaryotic RNA [2], the sequence coverage varied across each CDS, indicated by peaks and troughs (Figure 2). However, this profile was remarkably consistent between biological replicates. Importantly, many intergenic regions and 34% of the annotated CDS had few (AM<1) or no mapped reads. For sequence data mapped to a region where CDS orientation is highly “mosaic” the plots align predominantly to the predicted annotation (Figure 2B), further illustrating the strand-specific nature of the ssRNA-seq data. Sequence reads that mapped to non-coding strands may represent transcriptionally active but previously unannotated features of the genome. Indeed, these data enabled us to identify putative errors in the annotation of some genes, including many genes annotated as hypothetical, such as the locus t2145 (Figure 2C). A large amount of sequence data mapped to the opposite strand from this annotated CDS (Figure 2C), which suggests this may not be a hypothetical gene at all but instead the 5′ region, and a putative novel cis-regulatory element, of the gltA (t2146) gene [5],[6]. In support of this hypothesis, the intact reading frame for this predicted gene is not conserved outside the Salmonella but the DNA sequence itself is conserved in this location in many enteric bacteria, including Escherichia, Klebsiella and Enterobacter.
In order to provide an overview of the gene classes active in the genome-wide transcriptome, we determined the AM for each CDS and by using the previously assigned functional group classification [5] we assigned the number of sequence reads to each of the 12 functional classifications, which were then normalised, relative to total genome content of each class (Figure 3). A ratio of >1 represents a more highly transcriptionally active class. The ratios for outer membrane/surface structures, regulators, conserved hypotheticals and central intermediary metabolism were ∼1. Interestingly, transcriptional reads for CDSs associated with energy metabolism, information transfer and pathogenicity/adapation/chaperones were over-represented in the transcriptome with the ratio ranging from 1.57 to 2.23. As may be expected, transcriptionally silent prophage elements (ratio ∼0.75) are under represented with most of this transcript mapping to putative phage cargo genes (discussed later). Genes predicted to encode proteins required for degradation of both macromolecules and small molecules are under-represented in the reads, which is consistent with these genes being under tight transcriptional control in rich media. Interestingly, although pseudogenes represent 4.6% of the predicted CDS of S. Typhi Ty2, only 0.69% (ratio ∼0.15) of the entire ssRNA-seq generated transcriptome mapped to these genes, indicating a 10-fold reduction in expected transcriptional activity.
Many annotated predicted genes of the published S. Typhi genomes were assigned in the absence of clear protein homologies or direct evidence for transcription or translation into a protein product. We therefore carried out a comprehensive survey of transcript sequence and proteomic analysis of predicted genes in S. Typhi Ty2. We determined the AM sequence coverage and mapped peptides identified from fractions of S. Typhi Ty2 using LC-MS, for each predicted gene in the available annotation. The proportion of predicted genes in each functional class with AM sequence coverage >1 and those with at least one mapped peptide from proteomic analysis were determined (Figure 4, Table S1). The groups with the greatest proportion of transcriptionally active genes were those of information transfer (transcription and translation) and degradation of macromolecules, as would be expected for actively dividing bacteria in mid log phase. Pseudogenes, phage/IS elements, and ‘unknown’ genes were the least transcriptionally active and few peptides were mapped to products of genes from these classes. Relatively few pseudogenes had significant transcription and very few peptides mapped to these interrupted genes, suggesting that following pseudogene formation, transcription and translation are rapidly lost. It is also evident that functionally unassigned predicted hypothetical genes are frequently transcriptionally inactive with few mapped peptides. This may be because these are genes are only activated by specific environmental signals such as in vivo within host tissues or they are not true CDS and represent mis-annotation.
We also identified a variable [12] and previously unannotated region of the S. Typhi Ty2 genome to which mapped a number of transcriptome sequence reads (Figure 5). In this region, transcript data identified six transcribed CDS allowing us to refine the annotation of this region. These highly transcribed genes encode (from left to right) a protein with similarity to hypothetical proteins from a number of sequenced enteric bacteria (previously annotated as t0869); a protein with similarity only to a single protein from a marine Bacillus sp NRRL B-14911; two proteins with similarity to a restriction endonuclease/methylase pair (t0872 and t0871); and a conserved hypothetical protein with a broad phylogenetic range of matches. The last transcribed gene in this region is a putative phage integrase, which is adjacent to a conserved tRNA-Asn; the whole region is bounded by a 24 bp direct repeat of the 3′ end of the tRNA gene, indicating that this region is likely to be capable of independent integration and excision. Also encoded in this region are two duplicated type-IV pilV-like proteins, MobA-family and TraD-family conjugal transfer proteins, and a number of other genes of unknown function that show little evidence for transcription under these conditions.
Forty-two of the 82 ncRNAs annotated by Rfam in S. Typhi Ty2 generated transcripts that were detectable by ssRNA-seq analysis, with a range of AM between 1.2 and 438.18 reads/bp (Figure 6A). In addition, many sequence-reads mapped to novel intergenic regions of the S. Typhi Ty2 genome that were not previously annotated. Further analysis of such reads has allowed us to annotate an additional 55 genomic features as small ncRNAs based on these data. Many of these novel ncRNAs were not unique to S. Typhi Ty2, since putative homologues were identified in other S. enterica serovars and other bacterial species (Table S2). We also identified 25 CDS that were preceded by putative 5′ UTR transcripts, 13 of which were more than 150 bp in length (Table S2), as well as 5 novel putative 3′ UTR, 2 of which are adjacent to sprB and ramA. Subsequently, we determined the AM for each predicted ncRNA (Figure 6B, Figure S1A, S1B), showing that 91 of the 245 elements had an average AM>1. Taken together, this sequence data suggests that there may be many previously unidentified functional ncRNAs present in S. Typhi, that are conserved in other bacteria.
Recently a survey of Hfq bound ncRNAs was reported [4]. We mapped 52 ncRNAs identified in this study to S. Typhi Ty2 using annotation from the Rfam database [13]. Five of these had an average AM>1, the minimum of which was 1.75 (istR-2) and the maximum 13.38 (STnc250). The lack of overlap between these two datasets may be due to different experimental strategies for isolating and enriching RNA species. The RNA isolation method used in this study was optimised to remove contaminating proteins and may therefore remove RNA species bound to Hfq. Furthermore, the RNA preparation includes a rRNA depletion step, Hfq is known to bind sRNA to mRNA to regulate ribosomal initiation. This has been reported to occur both pre- [14] and post-ribosome initiation [15]. It is therefore possible that HFQ-bound sRNAs are being removed, prior to sequencing.
Transcription and translation in prokaryotes are commonly regulated by changes to the conformational structure of cis-acting ncRNAs called riboswitches. These RNAs generally bind metabolites related to the function of their associated downstream genes [16],[17],[18],[19],[20] and have been identified using bioinformatics methods based on sequence conservation of the 5′ UTR. Several known riboswitches, such as btuB [17],[21] and TPP [19] were highly represented in the ssRNA-seq data (Table S2). To analyse the putative ncRNAs for the potential to act as putative cis-regulatory elements we used a combination of ssRNA-seq, sequence conservation, secondary structure conservation and further homology search using covariance models. Subsequently, RNAz was used to rank candidates. One of the most interesting regions encoding putative novel ncRNAs was Salmonella pathogenicity island 1 (SPI-1) [22] (Figure 4C). Two of these SPI-1 associated transcripts were identified [23] as candidate riboswitches, here designated RUF_220c (1) and RUF_219c (2) (Figure S2A, S2B). RUF_220c and RUF_219c are located directly upstream of the araC-like regulators sprA (t2988) and sprB (t2987), respectively. A third candidate element, which is predicted to be a 3′UTR designated RUF218c (3), is encoded on the antisense strand of the sitD gene, an iron transport protein [24], and a hypothetical protein O30622 (t2767), which may have been acquired independently of the rest of SPI-1 [24]. The sequence of RUF_218c is conserved across cyanobacteria, firmicutes and proteobacteria. The sitA gene maps sequenced transcripts (average AM = 1.27) and 7 uniquely mapping peptide hits, whereas sitB, sitC and sitD have slightly lower levels of expression (AM = 0.37, 0.38 and 0.61, respectively) and map no sequenced peptides in our proteome preparations. It is possible that RUF218c is an antisense repressor of these proteins, as it is predicted to form a moderately stable minimum free energy secondary structure compared to a shuffled ensemble of sequences that have the same di-nucleotide composition (p = 0.0090). The fourth candidate element, named RUF_221 (4), maps to the 5′ UTR of iagA (t2999) (Figure 6C), an invasion regulator [25]. The predicted structure of this RNA (34% G+C) is not supported by other analyses (RNAz probability of 0.0037 and shuffling p value = 0.2627) (Figure S1C). There was also high sequence coverage in the 5′ UTR of t3658 (STY3917 in CT18), an orthologue of glmS of E. coli, a glutamine fructose-6-phosphate amino transferase. In certain Gram-positive micro-organisms a riboswitch has been characterised in the 5′ UTR of the glmS gene, that also encodes a glutamine fructose-6-phosphate amino transferase [26], suggesting that a candidate cis-regulating element is also present in S. Typhi.
A further previously unidentified putative non-coding feature is RUF_107c (complement strand, base range 101116..101223), which is highly expressed in these S. Typhi Ty2 samples. This element, predicted to be highly structured by RNAz (p = 0.9396), has approximately 115 paralogues in Salmonella (Figure S1A). Further, it is conserved across ∼82 bacterial species but is chiefly restricted to Enterobacteriaceae. The genomic context of RUF_107c and its paralogues is not consistent with a cis-regulatory or a transposable element, as the sequence does not consistently co-occur with either CDSs or near transposases, respectively.
S. Typhi harbours a number of distinct prophage, whose complement can vary between the different evolutionary lineages [9],[12]. Such prophages are regarded as being predominantly transcriptionally silent in the genome and can harbour horizontally acquired ‘cargo’ genes potentially encoding factors that modify the virulence of the host bacteria [27],[28]. Our analysis confirms that most of the resident prophage are indeed predominantly transcriptionally inactive (Figure 7) but it is worth noting that the ssRNA-seq mapping was sufficiently sensitive to highlight low level transcription across phage regions involved in maintaining lysogeny (Figure 7). However, we noted that four of the prophages did harbour transcriptionally active regions and that some of these mapped over well-known cargo genes such as sopE encoded by the SopE phage (Figure 7A). Cargo genes are non-essential for phage proliferation but may confer fitness to the lysogenised host bacterium [29],[30],[31]. Similar analysis of this prophage and others within the S. Typhi Ty2 genome highlights several transcriptionally active regions, which may encode novel cargo genes. Bioinformatics analysis of these regions, in some cases, supports this hypothesis, in that the genes do not encode known phage proteins and have differing GC content than other S. Typhi genes [9]. The SopE prophage expresses another region distinct from sopE that could encode three putative cargo genes, which are similar to hypothetical proteins found in Vibrio cholerae (Figure 7A, t4323, t4324 and t4325). Database searches using the transcriptionally active regions in the ST35 prophage (base range 3500845..3536809) reveal sequence similarity to hypothetical genes found in E. coli O157 (Figure 7B genes t3414, t3415). The ST46 prophage (4666742..4677430) encodes three transcriptionally active genes; two have sequence similarity to protein kinases and the third is a candidate threonine/serine kinase (Figure 7D, t4519, t4520, t4521). Thus, these methodologies may provide a novel approach to identifying phage cargo genes expressed during the lysogenic phase. A total of 73 peptides mapped to the four prophages (Figure 7, Table S4) and 59 (81%) mapped to highly transcribed regions containing known or putative cargo genes. Of these remaining, 5 peptides mapped to the highly transcribed cI repressor gene required for maintenance of lysogeny.
S. Typhi, in common with other host-adapted pathogens, harbours a large number (∼220) of putatively inactivated pseudogenes [5],[6],[12]. Genome degradation may contribute to host restriction by inactivating pathways essential for infections in the non-permissive host. Theoretically, putative pseudogenes can still express a functional truncated protein domain, as for example has been demonstrated for truncated cytotoxin in Chlamydia trachomatis [32],[33]. Based on comparative sequence analysis of 21 S. Typhi and two S. Paratyphi A genome sequences, the 220 pseudogenes of S. Typhi strain Ty2 have been assigned to four groups based on their predicted relative age [34]. We were able to identify nine pseudogenes in S. Typhi Ty2 that exhibited high levels of transcription, a property that was independent of their predicted age (Figure S3), suggesting that transcription may be being maintained to express functional domains as RNA or peptides. We did identify one pseudogene, hdsM (t4575) with sequenced peptide data, mapping to the open reading frame upstream of the inactivating stop codon. This represents the only significant evidence in this study of peptides mapping to putative pseudogenes. This, combined with the relative lack of transcription from other S. Typhi pseudogenes, supports the current interpretation that the majority of these genes are no longer active.
The global regulator OmpR is known to regulate the levels of transcription from a number of distinct loci within the S. Typhi genome, including the viaB locus [35] associated with Vi capsule production and the outer membrane porins ompC and ompS [36],[37],[38]. OmpR is also known to interact with the endogenous two-component regulator of SPI-2, ssrAB in S. Typhimurium [39],[40]. The complete OmpR regulon in S. Typhi has not been fully defined. We therefore prepared RNA from S. Typhi Ty2 and an otherwise isogenic S. Typhi Ty2 ompR mutant growing simultaneously in LB broth to mid-log phase (OD = 0.6). This RNA was then subjected to ssRNA-seq analysis and supporting conventional microarray analysis as control (see Methods).
To perform a quantitative ssRNA-seq comparison between sequenced products from the S. Typhi and ompR mutant RNA pools the AM was determined for all CDSs and, for the purposes of this analysis, these values were treated as intensity values similar to those derived by microarray scanning. We did not compare expression of ncRNA in this analysis. Using the AM and the LIMMA package for microarray analysis the data were quantile normalized [41]. Prior to Benjamini-Hochberg false discovery rate estimation and correction (BH-FDR), 305 genes had significantly different levels of sequenced transcript (2-fold change and p-value<0.05) (Figure S4) in S. Typhi Ty2 compared with ompR mutant derivative. Following application of BH-FDR, differences in sequence transcript was significant in fifteen of these genes (2-fold, adj p<0.05), all exhibiting a significant decrease in transcription in the ompR mutant. Consistent with previous reports, the entire viaB locus including tviABCDE, vexABCDE is represented in these 15 genes, [35] as well as envZ, the sensing component of the OmpR regulon. The four remaining genes were slsA (t3757), hyaA (t1458) and hypothetical genes t1459 and t1641 that are discussed below. Importantly, we confirmed that these genes were differentially expressed in the ompR mutant by a further method, quantitative PCR assays (data not shown).
Many of the 305 genes with significantly different transcript levels in the ompR mutant before BH-FDR correction, such as ssrAB, ompC and ompS, were reported previously to be OmpR regulated in Salmonella [36],[38],[39]. Furthermore, 71 of these genes appear to be encoded as 28 separate operons with similar differential expression patterns (Table S4). Indeed, some of these clusters of genes including ttrRS, ssrAB, cheBY, narZY, flgMN, flgHIJ, modAB, phnUV, hycGH, rplEXN, rplFR, aceBAK, have been previously confirmed as operons. This suggests ssRNA-seq identified blocks of differentially transcribed genes increasing our confidence in these findings despite exclusion following application of BH-FDR. Other examples of genes identified by ssRNA-seq include ssrA and ssrB (expression ratios of 0.09 and 0.31 respectively) known regulators of SPI-2 previously reported to be influenced by OmpR [39],[40]. Furthermore, the flagella genes fliH, fliI, fliM, with expression ratios of 2.08, 2.49 and 2.26 respectively, were in the original list of 305 genes.
Since ssRNA-seq is a new approach to mRNA expression analysis we performed independently a classic microarray analysis on the mRMA prepared from wild type S. Typhi Ty2 and ompR mutant derivatives as described in Methods and compared the data. We confirmed similar differential expression of 38 of the original 305 genes (two-fold, p<0.05) identified by ssRNA-seq independently by this DNA microarray expression analysis (Figure S4, Table S3). Of the 17 genes of this type that were decreased in expression in both experiments, ssRNA-seq reported a greater difference (ompR/WT) in the expression levels of these apparently down-regulated transcripts compared with microarrays (Figure S4). Genes previously characterised as OmpR-regulated with decreased levels of expression in S. Typhi Ty2 ompR mutants were tviABCDE, vexABCDE, ompC and ompS. Genes not previously described in the OmpR regulon identified in these data included the slsA (t3757) gene that is encoded within SPI-3 (also confirmed by direct RT-PCR), that is conserved throughout Salmonella and a putative inner membrane associated isochorismatase hydrolase. Isochorismatase hydrolase has been characterised in the phenazine biosynthesis pathway in Pseudomonas aeruginosa, potentially involved in antimicrobial activity and induced neutrophil cell death [42]. The hydrogenase uptake gene, hyaA2 (t1458) is also under represented in the S. Typhi Ty2 ompR ssRNA-seq data, the microarray data and direct RT-PCR assays. Salmonella encodes three predicted hydrogenase operons, two hydrogenase 1 operons (hyaACDEFt1048 and hyaA2B2C2D2E2F2t1454) and a hydrogenase 2 operon (hybOABCDEFG) that are important factors in respiration. Interestingly, two subunits of each operon, hyaA and hyaB2, are pseudogenes in S. Typhi Ty2 and CT18 [5],[6]. All three of these operons contribute to virulence in the S. Typhimurium murine model [43]. Furthermore, expression of the gene divergently transcribed from hyaA2, a putative secreted choloylglycine hydrolase (t1459) is also significantly decreased. The family of choloylglycine hydrolases cleave carbon-nitrogen bonds, exclusive of peptide bonds, and include conjugated bile acid hydrolase and penicillin acylase [44].
Intriguingly, 21 genes were increased in expression by the loss of ompR as determined by both ssRNA-seq and microarray analysis. Two contiguous flagellin regulatory genes, flgN (t1749) and flgM (t1748) were increased in expression in the ompR mutant. FlgM is a negative regulator of flagella biosynthesis and a mutation in this gene attenuates virulence in S. Typhimurium [45]. FlgN is required for the efficient initiation of filament assembly [46]. The glyoxylate shunt genes (aceBAK) are also increased in expression in S. Typhi Ty2 ompR (confirmed for aceA by RT-PCR) and fatty acid catabolism by isocitrate lyase is crucial for macrophage persistence in Mycobacterium tuberculosis [47]. Three genes t3544, t3543 and t3538 that are predicted components of a ribose/arabinose transport operon were also increased in expression. Furthermore, predicted genes t1788-90 were greatly increased in expression in the ompR mutant. These genes are contiguous and encode proteins with sequence similarity to a sialic acid transporter, a secreted protein and a sialic acid lyase respectively and are not present in E. coli. Molybdate transport is a crucial co-factor for anaerobic metabolism and transcription from two genes, modAB, required for its transport were increased in the ompR mutant. The levels of transcription of dppA, cstA, ybeJ, ybfM, glnH and t1709 were also increased in S. Typhi Ty2 ompR and these encode proteins annotated as periplasmic dipeptide transporter, carbon starvation response, glutamate transport, putative outer membrane, glutamine transport and a hypothetical protein, respectively.
We show here that Illumina-based ssRNA-seq sequencing technology allows the analysis of the transcriptome of the bacterial pathogen S. Typhi at the whole genome level and in a strand-specific manner. This technology therefore provides a powerful new approach to studies on bacterial gene expression, pathogenicity and mechanisms involving gene regulation at the level of transcription. By converting RNA to DNA it is possible to profile expression at a genome-wide level in such detail that even subtle features such as regulatory RNA features and small RNA sequences can be readily identified. Indeed, we were readily able to identify known attenuators and similar features in front of the Threonine (thr), Tryptophan (trp) and other operons (Figure S5). The depth of sequence analysis is sufficient to differentiate levels of expression, facilitating studies on bacteria or their mutant derivatives growing in different environments or conditions. Visualisation and interpretation of the transcript map was simplified by the exploitation of powerful bioinformatics mapping software and a modified version of the genome browsing tool Artemis [48]. Further, the transcriptome analysis was linked to the proteome, providing validation for a number of previously hypothetical genes. Indeed, the analysis was a useful tool for improving the annotation of the genome, redefining the limits of genes and transcripts and identifying novel small CDSs. Our analysis confirmed the expression of many known riboswitches that have recently been characterised and identified many more candidates. Indeed, we have mapped significant sequence data to the 5′ UTR of over 127 genes using ssRNA-seq. Many of the currently annotated riboswitches were predicted bioinformatically and their functionality was previously assessed through in vitro phenotyping assays [16],[17],[19]. Our genome-wide survey predicts such elements on a whole genome level providing candidates for further biological analysis. Three of these candidate regions were located in SPI-1, where they may impact on the expression of virulence genes.
Pseudogenes have contributed to apparent genome degradation in a number of host-adapted pathogens. Pseudogenes harbour potentially inactivating mutations that are normally identified through genome annotation programmes. However, the exact mechanisms by which pseudogenes impact on Salmonella pathogenesis is not fully understood but is believed to involve a loss of pathways that diversify the mechanisms the pathogen uses to survive in different hosts and tissues [49]. In this report, we demonstrate that the transcription of many pseudogenes is low or absent in a manner that is independent of the predicted age of pseudogene acquisition. However, we did identify several pseudogenes that are transcribed in rich media and peptides mapped to one of these. However, overall the evidence supports the concept that most S. Typhi pseudogenes are indeed null mutations.
Analysis of the prophage like elements encoded within the S. Typhi genome demonstrates that these are largely transcriptionally silent regions. Even so, the analysis was sensitive enough to identify genes that contribute to maintenance of the prophage state such as repressors of lysogeny. However, using ssRNA-seq analysis we are able to highlight transcriptionally active regions within largely inactive prophage elements. Some of these correlated with previously characterised cargo genes such as sopE that can contribute to pathogenicity. We postulate that other transcriptionally active regions within these prophage elements may be novel “cargo genes”. Peptides were mapped back to some of these regions.
Finally, we believe that the approaches described here are potentially applicable to any bacterium and provide a simple route towards the analysis of gene expression. The method, as outlined, has the advantage of providing strand-specific analysis allowing high resolution transcription maps to be generated. The method described is generic in that it can be performed with relatively minimal manipulation of nucleic acid and all the bioinformatics tools described are freely available. Further work will be required to optimise the use of ssRNA-seq for routinely analysing transcription in bacteria. For example, we do not yet know how accurate the quantitative analysis will be at a genome level in different bacteria. Our comparisons between ssRNA-seq and DNA microarray analysis for comparing differential gene expression using S. Typhi wild-type and ompR mutant derivatives indicates that the two approaches may be complementary but that they may not yield completely overlapping data. Indeed, previous work has shown that different microarray platforms are subject to considerable variability in reported transcription [50],[51]. Nevertheless, by combining both approaches we have identified sets of both known and novel OmpR regulated genes.
The bacteria used were all derivatives of S. Typhi Ty2 [6]. The ompR null mutant was made in the S. Typhi Ty2 by the red recombinase method [52] using the kanamycin resistance plasmid, pkd13, and primers ggatcgtctgctgacccgtgaatctttccatctcatgggtgtaggctggagctgcttc and gtctgaatataacgcggatgcgccggatcttcttccacattccggggatccgtcgacc. Cultures were grown in LB to OD600 = 0.6, fixed with 2∶1 volumes of RNAprotect Bacteria (Qiagen) and harvested. RNA was isolated from the pellet using SV RNA isolation kit (Promega) according to manufacturers instructions. RNA quality was determined using Bioanalyser (Agilent) and quantified using the ND-1000 (NanoDrop Technologies) after every manipulation step. 23S and 16S rRNA were depleted using MicrobExpress kit (Ambion). Genomic DNA was removed with two digestions using Amplification grade DNAse 1 (Invitrogen) to below PCR-detectable levels. The effect of incomplete DNAse treatment was a general increase in background (Figure S6). RNA was reverse transcribed using random primers (Invitrogen) and Superscript III (Invitrogen) at 45 C for three hours and heat denatured at 70 C for 15 minutes. Second strand synthesis was omitted in order to retain strand specific sequence determination; validation of this method is presented in full elsewhere (Croucher N, Fookes M, Perkins T. et al, submitted). Highly transcribed genes fliC, tviB and vexA, with a maximum amplicon of 250 bp, were used as targets for a PCR as a positive control for reverse transcription.
Sequencing libraries for the Illumina GA platform were constructed by shearing the enriched cDNA by nebulisation (35psi, 6 min) followed by end-repair with Klenow polymerase, T4 DNA polymerase and T4 polynucleotide kinase (to blunt-end the DNA fragments). A single 3′ adenosine moiety was added to the cDNA using Klenow exo- and dATP. The Illumina adapters (containing primer sites for sequencing and flowcell surface annealing) were ligated onto the repaired ends on the cDNA and gel-electrophoresis was used to separate library DNA fragments from unligated adapters by selecting cDNA fragments between 200–250 bps in size. Fragmentation followed by gel-electrophoresis were used to separate library DNA fragments and size fragments were recovered following gel extraction at room temperature to ensure representation of AT rich sequences. Ligated cDNA fragments were recovered following gel extraction at room temperature to ensure representation of AT rich sequences. Libraries were amplified by 18 cycles of PCR with Phusion polymerase. Sequencing libraries were denatured with sodium hydroxide and diluted to 3.5 pM in hybridisation buffer for loading onto a single lane of an Illumina GA flowcell. Cluster formation, primer hybridisation and single-end, 36 cycle sequencing were performed using proprietary reagents according to manufacturers' recommended protocol (https://icom.illumina.com/). The efficacy of each stage of library construction was ascertained in a quality control step that involved measuring the adapter-cDNA on a Agilent DNA 1000 chip. A final dilution of 2 nM of the library was loaded onto the sequencing machine.
We used the computational pipeline developed at the Wellcome Trust Sanger Institute, (http://www.sanger.ac.uk/Projects/Pathogens/Transcriptome/). We mapped all reads to the S. Typhi Ty2 genome using MAQ and discarded all reads that did not align uniquely to the genome. The quality parameter (−q) used in MAQ pileup was 30. MAQ pileup prints an array of delimited information formatted as one line per genomic base. Each base is assigned a value for the number of piled sequences and the mapped strand for each read, represented by a “.” (forward) and “,” (reverse). For example, Forward strand: all_bases, 7887, G, 45, @.............................................; Reverse strand: all_bases, 914, G, 6, @,,,,,,,; Overlapping Strands: all_bases, 7690, G, 38, @,,,,,.,.,,..,,,,,,.................... These data were then mapped strand specifically using the perl script maqpileup2depth.pl returning a plot file with two columns which can be read into Artemis as a graph by using commands “Graph, Add User Plot”.
Candidate ncRNA sequences from Salmonella enterica subsp. enterica serovar Typhi Ty2 complete genome (EMBL ACC: AE014613.1) were searched against RFAMSEQ (a subset of the EMBL nucleotide database) using the Rfam search pipeline based upon WU-BLAST filters followed by covariance model (CM) scoring [13]. CMs have been proven to be vastly more accurate than BLAST for scoring ncRNAs [53]. Reliable matches were subsequently aligned and a consensus RNA secondary structure predicted folded using WAR [54]. Covariance models (CMs) were built for each resulting alignment; these researched searched against RFAMSEQ using the Rfam pipeline until there were no new reliable hits [13]. The subsequent alignments and secondary structures were inspected and modified by hand where improvements could be made. The secondary structure diagrams [55] and phylogenetic trees were built from these results. The alignments were then screened with the RNAz suite of tools for de-novo ncRNA prediction tool [23]. The original candidate sequences from S. typhi Ty2 were also analysed for individual secondary structure content using a permutation test. One thousand shuffled sequences with the same di-nucleotide content were generated for each native sequence. The distribution of predicted minimum free-energy (MFE) values of folding for the shuffled ensembl of sequences was used to determine the significance of the MFE value for the native sequence. There is an extensive literature on this approach with mixed success, the method is best suited to highly stable structures such as microRNAs [56],[57],[58].
AM per base pair was determined using the script tram.pl and this value used as an expression value like fluorescence intensity on a microarray. The data from both microarray and ssRNA-seq were quantile normalised and differential analysis performed using the LIMMA package [41].
We isolated RNA from three biological replicates and for each, four slides were hybridised using 16 µg of RNA and compared to the same amount of BRD948 RNA. The dyes were swapped for two arrays in each replicate. Low density spotted microarrays were used. Design, hybridisation and scanning were performed as previously described in Doyle et al [59] and array data submitted to Array Express. Overall 216 genes were identified as being differentially transcribed (2-fold, adj p-value<0.05) and 73 of these were reduced in transcription compared with BRD948.
Whole cells were fractionated as previously described by Hantke [60]. Protein samples were reduced and alkylated with iodoacetamide prior a separation in a 4–12% NuPAGE Bis-Tris gel (Invitrogen). Gels were stained with colloidal Coomassie blue (Sigma) and bands were excised and followed by in-gel digestion by trypsin (sequencing grade; Roche). The extracted peptides were analyzed with on-line nano LC-MS/MS on an Ultimate 3000 Nano/Capillary LC System (Dionex) coupled to a LTQ FT Ultra mass spectrometer (ThermoElectron) equipped with a nanoelectrospray ion source (NSI). Samples were first loaded and desalted on a PepMap C18 trap (0.3 mm id×5 mm, Dionex) then separated on a BEH C18 analytical column (75 µm id×10 cm) [7] over a 30 or 45 or 60 min linear gradient of 4–32% CH3CN/0.1% FA based on the gel band's size and intensity. The mass spectrometer was operated in the standard data dependent acquisition mode controlled by Xcalibur 2.0. The survey scans (m/z 400–1500) were acquired on the FT-ICR at a resolution of 100,000 at m/z and the three most abundant multiply-charged ions (2+ and 3+) with a minimal intensity at 1000 counts were subject to MS/MS in the linear ion trap. The dynamic exclusion width was set at ±10 ppm. The automatic gain control (AGC) target value and maximum injection time were set at 1×106 and 1000 msec for FT and 1×104 and 250 msec for ion trap respectively. The instrument was externally calibrated. The Raw files were processed by BioWorks 3.3 and then submitted to a database search in Mascot server 2.2 (www.MatrixScience.com) against an in-house built Typhi Ty2 genomic 6-frame translated database [61]. All peptides with a posterior error probability (probability that an individual peptide was identified by chance alone) of 1% or less were accepted for subsequent analysis, resulting in an overall false discovery rate of about 0.1%. The analysed proteomic data has been submitted to EBI PRIDE database (www.ebi.ac.uk/pride/) with and can be viwed under PRIDE accession number 9770–9774.
The peptide sequences were mapped to all matching positions in a 6-frame translation of the entire genome and only peptides that mapped to one region of the genome were included in these data.
|
10.1371/journal.ppat.1002987 | Myeloid-Related Protein-14 Contributes to Protective Immunity in Gram-Negative Pneumonia Derived Sepsis | Klebsiella (K.) pneumoniae is a common cause of pneumonia-derived sepsis. Myeloid related protein 8 (MRP8, S100A8) and MRP14 (S100A9) are the most abundant cytoplasmic proteins in neutrophils. They can form MRP8/14 heterodimers that are released upon cell stress stimuli. MRP8/14 reportedly exerts antimicrobial activity, but in acute fulminant sepsis models MRP8/14 has been found to contribute to organ damage and death. We here determined the role of MRP8/14 in K. pneumoniae sepsis originating from the lungs, using an established model characterized by gradual growth of bacteria with subsequent dissemination. Infection resulted in gradually increasing MRP8/14 levels in lungs and plasma. Mrp14 deficient (mrp14−/−) mice, unable to form MRP8/14 heterodimers, showed enhanced bacterial dissemination accompanied by increased organ damage and a reduced survival. Mrp14−/− macrophages were reduced in their capacity to phagocytose Klebsiella. In addition, recombinant MRP8/14 heterodimers, but not MRP8 or MRP14 alone, prevented growth of Klebsiella in vitro through chelation of divalent cations. Neutrophil extracellular traps (NETs) prepared from wildtype but not from mrp14−/− neutrophils inhibited Klebsiella growth; in accordance, the capacity of human NETs to kill Klebsiella was strongly impaired by an anti-MRP14 antibody or the addition of zinc. These results identify MRP8/14 as key player in protective innate immunity during Klebsiella pneumonia.
| Neutrophils are phagocytes that are well known for their capacity to engulf and kill microbial pathogens. It has become increasingly clear that neutrophils also kill or inhibit growth extracellularly by releasing neutrophil extracellular traps (NETs), chromatin fibers decorated with neutrophil derived proteins. MRP8/14 has been identified as one of the major antimicrobial proteins herein. Previous investigations have shown that endogenously released MRP8/14 is also sensed by the host as a danger signal and able to potentiate the harmful systemic inflammatory response syndrome. Indeed, in the setting of fulminant systemic inflammation, such as induced by endotoxin or Escherichia coli administration, MRP8/14 contributed to organ injury and mortality. The clinical scenario of sepsis however, involves an initial infection at the primary site followed by bacterial spreading to other organs. In the present setting of pneumonia-derived sepsis using the common human respiratory and sepsis pathogen Klebsiella pneumoniae MRP8/14 clearly served a beneficial role in antimicrobial defense. We here provide a likely mechanism by showing that MRP8/14 plays a role in phagocytosis and that its presence is critical in both murine and human NETs to inhibit bacterial growth.
| Klebsiella (K.) pneumoniae is a frequent causative pathogen in pneumonia [1], [2] and the second most common cause of gram-negative sepsis [3], [4]. Klebsiella infection presents a significant burden on healthcare and is associated with high morbidity and mortality rates. Effective treatment of this microorganism is even more challenging due to the emergence of microbial resistance to (last-resort) antibiotics [5], [6]. It is therefore of great importance to expand our understanding on host defense mechanisms that influence the outcome of Klebsiella pneumonia. Such knowledge may eventually help in the development of new therapies.
Invasive infection and accompanying inflammatory mechanisms can cause tissue damage that is associated with release of endogenous “alarm” proteins. These proteins, also known as Damage Associated Molecular Patterns (DAMPs), are recognized by pattern recognition receptors and perpetuate inflammatory responses [7], [8]. Among these DAMPs, the S100 proteins MRP8 (myeloid-related protein, S100A8) and MRP14 (S100A9) have gained increasing interest [9], [10]. They are mainly and constitutively expressed in neutrophils where they comprise 45 percent of total cytoplasmic protein [11]. MRP8 and MRP14 are able to dimerize with a clear preference for the most stable and biologically relevant MRP8/14 heterodimer (or calprotectin), which can be actively released into the extracellular space [12]–[15]. MRP8/14 induces a variety of host responses and the extent of expression correlates with clinical [9], [10] and experimental [16] disease activity. Previous investigations have pointed to a complex role of MRP8/14 in severe infection, which may either be protective or harmful to the host. MRP8/14 can enhance inflammation via activation of Toll-like receptor (TLR)4, by amplifying tumor necrosis factor (TNF)-α release in response to lipopolysaccharide (LPS), the immunostimulatory component of the gram-negative bacterial cell wall. In the setting of fulminant systemic inflammation such as induced by high dose LPS or Escherichia (E.) coli administration, endogenous MRP8/14 contributes to lethality [15]. On the opposite site, MRP8/14 may be important for innate defense against microorganisms by virtue of its involvement in leukocyte migration [17]–[20] and its direct antimicrobial effects [21]–[23]. In addition, recent studies revealed that MRP8/14 is a major component of neutrophil extracellular traps (NETs) [24], DNA-networks released by neutrophils that trap microorganisms and facilitate interaction with antimicrobial proteins and thereby bacterial killing [25], [26]. Although at present the importance for NET associated MRP8/14 for bacterial killing is unknown, the presence of MRP8/14 was found to be crucial for the clearance of fungi by NETs in vitro [24], [27].
In the present study, we aimed to characterize the role of MRP8/14 during pneumonia-originating sepsis caused by K. pneumoniae. For this we used mrp14 deficient (mrp14−/−) mice, which due to instability of MRP8 in the absence of its binding partner MRP14 are considered deficient for MRP8/14 at protein level [28]–[30]. We here show that MRP8/14 deficiency results in enhanced bacterial dissemination, increased distant organ damage and a reduced survival during Klebsiella pneumonia. Using in vitro models, we show that mrp14−/− macrophages have a reduced capacity to phagocytose this bacterium. We further demonstrate that MRP8/14 directly reduces the growth of Klebsiella and in addition is essential in NET-mediated growth inhibition of this pathogen. These results identify MRP8/14 as an important protective mediator in the innate immune response to bacterial pneumonia caused by a clinically relevant pathogen.
To gain a first insight into the potential role of MRP8/14 in gram-negative pneumonia, we intranasally infected Wt mice with K. pneumoniae (104 cfu) and measured local and systemic MRP8/14 concentrations 6, 24 and 48 hours thereafter. MRP8/14 levels became detectable in BAL fluid at 24 hours after infection; high levels were found at 48 hours (median 745 ng/ml; Fig. 1A). In whole lung homogenates, MRP8/14 was detectable at low levels in uninfected mice and concentrations did not increase during the first 6 hours after infection; high MRP8/14 levels were detected at later time points (median 200 µg/ml at 24 hours; Fig. 1B). Plasma MRP8/14 levels also increased during the course of the infection, reaching peak concentrations at 48 hours (median 440 ng/ml; Fig. 1C).
To obtain insight into the cellular source of MRP8/14, we stained lung tissue slides obtained from naïve and infected Wt mice for MRP8 and MRP14. Bronchial epithelium of naïve lungs showed a faint staining for MRP8 (but not MRP14) that did not intensify during Klebsiella pneumonia. A small number of MRP8 and MRP14 positive cells, mainly residing macrophages, were already present. During the course of the disease, expression of both MRP8 and MRP14 increased strongly, primarily as a consequence of infiltrating neutrophils (Fig. 1D–I).
To investigate the functional role of MRP8/14 in host defense during gram-negative pneumonia, we infected mrp14−/− and Wt mice with 104 viable K. pneumoniae and harvested lungs, blood, spleen and livers at predefined time points for quantitative cultures, seeking to collect data representative for local defense, at the primary site of infection, and subsequent dissemination. Initially, mrp14−/− mice showed slightly lower bacterial loads in lungs compared to Wt mice (Fig. 2A; p<0.001). At later time points, mrp14−/− mice tended to have higher bacterial burdens in their lungs. Remarkably, at late stage infection (after 48 hours, shortly before the first deaths occurred) mrp14−/− mice displayed increased bacterial loads in blood, liver (both p<0.01) and spleen (p<0.05), suggesting that MRP14 deficiency is associated with enhanced dissemination of the infection (Fig. 2B–D). To investigate the impact of the reduced antibacterial defense in mrp14−/− mice on survival, we performed an observational study instilling the same infectious dose (104 cfu) as used in the experiments determining bacterial growth and spreading (Fig. 2E). This inoculum rapidly led to death shortly after 48 hours in all mice; notably, mrp14−/− mice tended to die earlier in this lethal model (p = 0.10 versus Wt mice). Arguing that the infectious challenge might have been too high to reveal a detrimental effect of MRP14 deficiency on survival, we repeated this observational study with a 10-fold lower inoculum (103 cfu; Fig. 2F). This infectious dose resulted in a 56% lethality in Wt mice 10 days after inoculation; mrp14−/− mice displayed a higher mortality rate and eventually 94% of mrp14−/− mice died (p = 0.0008 versus Wt mice). These data establish the important protective role of MRP8/14 in K. pneumoniae pneumonia as reflected by an increased dissemination of bacteria and a reduced survival.
The fact that MRP14 deficiency in particular influenced bacterial loads in distant organs led us to hypothesize that mrp14−/− mice would also show enhanced bacterial growth after direct intravenous injection of K. pneumoniae. Indeed, 48 hours after intravenous infection with 2×103 Klebsiella cfu, mrp14−/− mice demonstrated higher bacterial burdens in spleen, liver and lungs (all p<0.01; Fig. S1). These data suggest that MRP8/14 importantly contributes to systemic protection against K. pneumoniae infection.
Bacterial pneumonia is associated with neutrophil migration to the lung parenchyma, which is considered to be an essential component of a protective innate immune response [31], [32]. Previous studies indeed have documented that neutrophils play an important role in innate defense early after Klebsiella airway infection [33], [34]. MRP8/14 has been implicated as an important mediator of neutrophil recruitment in various inflammatory conditions [17]–[19], [35], including pneumonia [20]. Therefore, we determined the extent of neutrophil influx in mrp14−/− and Wt mice at 6, 24 and 48 hours after intranasal challenge with K. pneumoniae by assessing the number of Ly-6G positive cells in lung tissue sections and by measuring MPO concentrations in whole lung homogenates (Fig. 3). Neither the number of Ly-6G positive cells, nor whole lung MPO concentrations differed between mrp14−/− and Wt mice at any time point; if anything, mrp14−/− mice tended to have more Ly-6G positive cells and higher MPO levels in their lungs than Wt mice. Hence, these data strongly argue against a role for MRP8/14 in neutrophil influx into the lungs during Klebsiella pneumonia.
This model of K. pneumoniae pneumonia is associated with profound lung inflammation [36]–[38]. Although the presence of MRP8/14 did not affect lung bacterial loads or neutrophil recruitment at later time points, we wondered whether lung pathology would be influenced by the (proinflammatory) effects of MRP8/14 itself during Klebsiella pneumosepsis. We therefore analyzed HE-stained lung tissue slides obtained from infected Wt and mrp14−/− mice using the semi-quantitative scoring system described in the Materials en Methods section (Fig. 4). Already at 6 hours after infection mild interstitial inflammation and pleuritis were found in all mice; at later stages endothelialitis, bronchitis and edema became apparent. Interestingly, after 24 and 48 hours, mrp14−/− mice showed exaggerated lung pathology with enhanced interstitial inflammation, bronchitis and larger surfaces of confluent inflammation infiltrate. The enhanced lung pathology may be a reflection of the increased disease severity in mrp14−/− mice and suggest that the presence of MRP8/14, though proinflammatory, is not essential for the induction of lung inflammation during Klebsiella pneumosepsis.
Extracellular MRP8/14 has been shown to amplify the TNF-α response upon LPS stimulation and mrp14−/− bone marrow cells demonstrated a reduced responsiveness to LPS in vitro [15]. In vivo this correlated with lower plasma TNF-α levels in mrp14−/− mice challenged with LPS [15]. To study the impact of MRP14 deficiency on cytokine release in gram-negative pneumonia, we measured the levels of cytokines (TNF-α, IL-1β, IL-6, IL-10) and chemokines (MIP-2, KC) in lung homogenates and plasma (TNF-α, IL-1β and IL-6 only) harvested from mrp14−/− and Wt mice after intranasal infection with Klebsiella. Surprisingly, overall differences between mouse strains were limited. In whole lung homogenates, mrp14−/− mice displayed reduced MIP-2 levels after 6 hours only; at other time points and for other mediators, levels were similar between groups (Fig. 5A–F). Similarly, plasma cytokine levels did not differ between Wt and mrp14−/− mice at 6 or 24 hours post infection; after 48 hours however the plasma levels of IL-6 and IL-1β were even higher in mrp14−/− mice compared to the Wt mice (Fig. 5G–I).
We next determined the role of endogenous MRP8/14 in the inflammatory response to Klebsiella infection in vitro. To this end, we measured TNF-α levels after incubating whole blood from Wt, mrp14−/− and tlr4−/− mice with log-increasing loads of viable, growth-arrested Klebsiella for 6 hours. In consistence with our in vivo data, whole blood obtained from Wt and mrp14−/− mice displayed a similar cytokine response. Only upon exposure to the highest bacterial concentration, mrp14−/− whole blood showed a modestly reduced cytokine response. Whole blood from tlr4−/− mice showed almost no TNF-α release in response to Klebsiella (Fig. S2). Together, these data indicate that endogenous MRP8/14 has no, or little contribution to the TLR4 mediated cytokine response to K. pneumoniae infection.
This model of gram-negative pneumonia-derived sepsis is associated with hepatocellular injury during late stage infection [38]. In our time point experiments, MRP14 deficiency was associated with enhanced bacterial dissemination to distant organs, including the liver. We were thus interested to what extent this enhanced bacterial dissemination influenced hepatocellular injury in these animals. Microscopic examination revealed dramatically enhanced pathology in livers from mrp14−/− mice after 48 hours, reflected by more advanced liver necrosis accompanied with thrombi and more (micro)abscesses compared to Wt mice (Fig. 6A–C). Clinical chemistry findings confirmed the existence of more extensive hepatocellular injury, i.e. mrp14−/− mice had higher plasma levels of ALT and AST, in particular 24 hours after infection (Fig. 6D,E). At this time point, mrp14−/− mice also showed higher plasma LDH concentrations (indicative for cellular injury in general) relative to Wt mice (Fig. 6F).
A recent report has shown that addition of MRP14 is able to enhance bactericidal effects of human neutrophils by means of improving bacterial phagocytosis capacity [39]. We wondered whether this would correspond to impaired K. pneumoniae phagocytosis in mrp14−/− murine phagocytes. To investigate this possibility, we harvested whole blood and macrophages from naïve Wt and mrp14−/− mice and compared neutrophil and macrophage ability to internalize CFSE-labelled, viable, growth-arrested K. pneumoniae by FACS. Although MRP8/14 is abundantly present in neutrophils, mrp14−/− neutrophils only displayed a modest reduction in their capacity to phagocytose Klebsiella compared to Wt neutrophils (Fig. 7A). Mrp14−/− macrophages however, were significantly reduced in their capacity to internalize Klebsiella bacteria (Fig. 7B). In addition to increased dissemination, the reduced capacity of macrophages to phagocytose the bacteria most probably contributed to the enhanced bacterial outgrowth in mrp14−/− mice during Klebsiella infection in vivo.
MRP8/14 has been shown to inhibit the growth of several microorganisms by binding of divalent cations [22], [24], [27]. To study whether K. pneumoniae growth is affected by MRPs, we grew bacteria in medium for up to 24 hours in the presence or absence of MRP8/14 heterodimer, MRP8 homodimer or MRP14 homodimer (all 50 µg/ml). Addition of MRP8/14 heterodimer almost abolished growth of K. pneumoniae, while MRP8 and MRP14 homodimers had no effect (Fig. 8A). The growth inhibitory effect of MRP8/14 was dose dependent (Fig. 8B). To check whether MRP8/14 induced growth inhibition was due to a metal chelating effect, zinc was added to medium treated with 10 µg/ml MRP8/14 prior to incubation with bacteria; in this experiment (Fig. 8C) growth was restored when zinc was added in increasing amounts. Thus, MRP8/14 inhibits growth of K. pneumoniae through chelation of metals.
Earlier studies have demonstrated that Klebsiella pneumoniae induces NET formation in vivo [40]. Using immunofluorescence technique, we found similar decondensation of nuclei of neutrophils, strongly indicating the formation of NETs in lungs of both Wt and mrp14−/− mice (Fig. S3). It has recently been shown that MRP8/14 is abundantly present in NETs and that its presence is critical for fungal clearance [24], [27]. To test whether MRP8/14 is important in NET-mediated growth inhibition of K. pneumoniae, we stimulated Wt and mrp14−/− neutrophils with PMA to form NETs and then incubated these with viable bacteria. NETs from Wt neutrophils inhibited the growth of Klebsiella, while mrp14−/− NETs did not (Fig. 9A). To investigate the antibacterial properties of MRP8/14 in human NETs, we induced NET formation in neutrophils from healthy donors and then coincubated these with viable Klebsiella in the presence or absence of a neutralizing polyclonal anti-MRP14 antibody. In line with results obtained with other pathogens [24]–[27], human NETs effectively inhibited the growth of Klebsiella. This effect was strongly dependent on MRP8/14: addition of an anti-MRP14 antibody, blocking the chelating effect of MRP8/14 [27], almost completely restored growth of K. pneumoniae. These data confirm that NET-mediated growth inhibition of K. pneumoniae in a human system is MRP8/14 dependent. This growth inhibition may rely on the chelation of divalent cations, since addition of zinc in excess led to the same, if not an even stronger effect on growth of Klebsiella (Fig. 9B). The growth-inhibiting role of MRP8/14 within NETs also applied to other bacteria that are sensitive to NETs. Both gram-positve S. aureus and gram-negative Pseudomonas aeruginosa, showed an increased outgrowth in the presence of anti-MRP14 compared to control IgG (Fig. S4).
Gram-negative sepsis is a major challenge in the care of critically ill patients. Despite the availability of effective antimicrobial therapy and supportive care, mortality rates remain up to 30–50% [3], [41]. Severe sepsis is associated with the release of MRP8/14 [16], which in models of endotoxic shock and fulminant sepsis contributes to organ injury and mortality [15], [16]. The clinical scenario of sepsis, however, involves an initially localized infectious source with subsequent spreading of bacteria to distant body sites. We argued that in this setting, quite different from its previously described detrimental role in acute systemic sepsis models, MRP8/14 may be an important component of protective innate immunity at least in part because of its antimicrobial properties [21]–[23]. Thus, in the present study we aimed to determine the role of MRP8/14 during gram-negative sepsis originating from the lungs, using an established clinically relevant pneumonia model characterized by gradual growth of bacteria at the primary site of infection followed by dissemination, tissue injury and death [36]–[38], allowing to study a potential role of MRP8/14 in both the initial immune response as well as the subsequent harmful systemic inflammation phase. In summary, we found that intranasal K. pneumoniae infection resulted in local and systemic MRP8/14 release and that MRP14 deficiency led to an increased mortality, most likely as a consequence of enhanced bacterial outgrowth and organ damage. The beneficial role of this heterodimer may result from its involvement in phagocytosis and its strong growth inhibitory effect, while it hardly influenced the TLR4 mediated cytokine response to Klebsiella.
Severe human sepsis results in systemic release of MRP8/14 irrespective of the primary source of infection [16]. Patients with sepsis caused by pneumonia display the highest protein levels [16]. In the present mouse study, systemic MRP8/14 levels gradually increased during the course of the infection. Our current finding of high MRP8/14 concentrations at the primary site of infection from 24 hours onward is in accordance with a previous investigation from our group reporting high local levels of MRP8/14 in patients and mice with bacterial peritonitis [16]. Of note, however, systemic MRP8/14 levels in mice with E. coli peritonitis were at least ten times higher after one day than in mice with Klebsiella pneumonia studied here, which most likely reflects the fulminant nature of the septic syndrome induced by intraperitoneal E. coli administration. In these acute challenge models, such as is also produced by high dose LPS injection, MRP8/14 was suggested to act as a danger signal enhancing the cytokine response of LPS via TLR4, thereby potentiating the harmful systemic inflammatory response syndrome [15], [16]. Indeed, after either E. coli or LPS administration, MRP14 deficiency attenuated systemic inflammation and consequently improved survival in E. coli induced peritonitis and LPS-induced shock, starting to occur from 20 and 6 hours respectively [15]. In contrast, in the more clinically relevant sepsis model used here, a “cytokine storm” such as detected after LPS administration was not induced and mortality only started to occur after 2 days, allowing the gradually increasing levels of MRP8/14 to serve its beneficial role in antimicrobial defense.
Of note, whereas MRP14 deficiency clearly reduced systemic cytokine release after LPS or E. coli injection [15], [16], such an effect was not seen during Klebsiella pneumonia. In contrast, plasma concentrations of proinflammatory cytokines were higher in mrp14−/− mice during late stage infection, whereas lung cytokine and chemokine levels were largely similar in mrp14−/− and Wt mice. The increased plasma levels of IL-1β and IL-6 in mrp14−/− mice likely reflect the increased bacterial loads in blood, providing a more potent proinflammatory stimulus. In accordance with our in vivo data, MRP14 deficiency had little or no effect on the TLR4 mediated TNF-α response in a 6 hour incubation of whole blood with growth-arrested K. pneumoniae. Together, these data suggest that MRP8/14 minimally contributes to the cytokine response in the context of a gradually growing bacterial load, i.e. can be compensated for by other mechanisms. Furthermore, lungs from mrp14−/− mice showed increased inflammation, most likely as a consequence of the more severe infection in these animals and indicating that MRP8/14 is not a critical inducer of lung inflammation during gram-negative pneumosepsis.
In spite of high lung concentrations of MRP8/14, this heterodimer did not play a significant role in the control of local infection considering that pulmonary bacterial loads only slightly differed between mrp14−/− and Wt mice after 24 and 48 hours. The lack of MRP8/14 did however result in strongly increased bacterial burdens in blood, spleen and especially the liver. This led us to postulate that the apparent antimicrobial effect of MRP8/14 primarily resided in organs distant from the lung. Indeed, mrp14−/− mice administered with K. pneumoniae directly intravenously, thereby bypassing a potential effect of MRP8/14 in the airways, displayed strongly increased bacterial growth in multiple body sites, including the spleen and liver. We found a reduced capacity of mrp14−/− macrophages, but not mrp14−/− neutrophils, to phagocytose Klebsiella in vitro, which may partially explain enhanced bacterial outgrowth in these organs. Indeed, resident spleen macrophages form an important barrier to blood-borne pathogens and facilitate clearance in systemic infection [42]. The vast majority of bacteria entering the bloodstream are cleared by the liver [43]. Resident liver macrophages (Kupffer cells), constitute 80–90 percent of total tissue macrophages in the body [44] and have been attributed to clear the bulk of bacteria that are taken up by this organ [43]. A number of more recent studies, however, show that certain pathogens, including K. pneumoniae, can be cleared even in animals that lack Kupffer cells and suggest that immigrating neutrophils crucially contribute in hepatic clearance of circulating bacteria [45]–[47]. Hence, the exact contribution of MRP14 mediated phagocytosis by macrophages in our in vivo model remains to be elucidated in further detail.
Impaired antibacterial defense in the liver may have led to enhanced formation of (micro)abscesses. Abscesses, contain high levels of MRP8/14 and could be essential in the control of K. pneumoniae infection [22], [48]. We therefore hypothesize that liver abscesses deficient for MRP8/14 promote bacterial outgrowth of Klebsiella and may be the source for further bacterial dissemination. As a consequence of uncontrolled liver infection, enhanced hepatocellular damage occurred as reflected by the increased plasma levels of liver transaminases and enhanced liver pathology.
Several animal studies have implicated MRP8/14 as a mediator of neutrophil recruitment. In murine air pouch models, pretreatment with blocking antibodies directed against MRP8 and MRP14 significantly reduced leukocyte migration in response to LPS [17] or monosodium urate crystals [18]. Anti-MRP8 and anti-MRP14 antibodies also attenuated leukocyte influx into the pulmonary compartment during S. pneumoniae pneumonia [20]. Chemotaxis by MRP8/14 may partly act via upregulation of adhesion molecule expression and induction of CXC chemokines [35]. In the present study we did not find evidence for a role for MRP8/14 in chemotaxis: MPO levels in whole lung homogenates and the number of Ly-6G positive cells in lung tissue slides were not affected by the loss of MRP8/14 in spite of reduced levels of the neutrophil attracting chemokine MIP-2 early after infection. Similarly, we showed earlier that neutrophil numbers in the peritoneal cavity during E. coli induced peritonitis did not differ between mrp14−/− and WT mice [16]. MRPs may also mediate other neutrophil functions like, degranulation phagocytosis and respiratory burst [39], [49]. We however, found a similar capacity of mrp14−/− and Wt neutrophils to mount a respiratory burst (data not shown) and to phagocytose when incubated with Klebsiella bacteria.
Recently, MRP8/14 was found to inhibit Staphylococcus (S.) aureus growth through chelation of zinc and manganese [22]. Divalent ion-chelation also reduced the enzymatic activity of superoxide dismutase thereby inhibiting bacterial virulence [23]. In accordance, mrp14−/− mice showed higher bacterial loads after intravenous S. aureus injection [22]. In contrast, mouse studies investigating abdominal sepsis or urinary tract infection caused by E. coli [16], [50] or pneumonia caused by S. pneumoniae [20] did not point to an antimicrobial role for MRP8/14. In the present study we showed that the growth of K. pneumoniae was dose dependently inhibited by MRP8/14 while neither MRP8 nor MRP14 homodimers affected growth in any way. The antimicrobial effect of MRP8/14 toward Klebsiella could be overcome by addition of zinc implicating chelation of divalent cations by MRP8/14 as a key process herein. During infection, neutrophils can kill pathogens through different mechanisms, including by the release of NETs composed of chromatin decorated with neutrophil derived proteins [24]–[26]. We here observed the presence of decondensated nuclei of neutrophils in the lungs strongly indicating the formation of NETs. Histones have been implicated as the predominant antibacterial component of NETs, responsible for a reduction in bacterial counts in vitro already after 30 minutes [25]. Such a role in a short time span was not found for MRP8/14: incubation of Klebsiella with NETs for one hour resulted in reduced outgrowth as well, but this was not influenced by coincubation with anti-MRP14 (data not shown). A previous investigation established that MRP14 is not required for NET formation by neutrophils [24]. We here show that MRP8/14 is an important player in growth inhibition in late stage infection of mouse NETs and that NETs void of MRP8/14 are unable to inhibit the growth of K. pneumoniae. NETs in the Klebsiella-infected lungs, which have been documented in a previous study [40], might contribute to decreased dissemination into spleen and liver in an MRP8/14-dependent manner via this growth inhibitory effect. In accordance, human NETs strongly inhibited Klebsiella growth, which was almost completely reversed by anti-MRP14 antibodies blocking the chelating effect of MRP8/14 or by the addition of zinc. Our current data are the first to indicate that the capacity of NETs to kill a bacterium is highly dependent on MRP8/14 and that MRP8/14 exerts its antimicrobial effects in NETs on Klebsiella through metal chelation. A similar MRP8/14 dependent mechanism was recently shown for killing by NETs of the fungi Candida albicans [24] and Aspergillus nidulans [27].
In conclusion, we here document that MRP14 deficiency leads to increased bacterial growth and dissemination accompanied by enhanced organ damage and mortality in K. pneumoniae sepsis originating from the lungs. MRP8/14 exerts its essential protective role by its involvement in macrophage phagocytosis and by directly inhibiting the growth of K. pneumoniae through divalent cation chelation. This study shows for the first time that MRP8/14 within NETs is critical in both a murine and human system controlling bacterial infection.
Experiments were carried out in accordance with the Dutch Experiment on Animals Act and approved by the Animal Care and Use Committee of the University of Amsterdam (Permit number: DIX100121, DIX101223) or carried out according to the recommendations in the guide for the care and use of laboratory animals conformed to Swedish animal protection laws and applicable guidelines (djurskyddsmyndigheten DFS 2004:4) and approved by the local Ethical Committee (Dnr A 29-09).
C57Bl/6 Wild type (Wt) mice were purchased from Charles River Laboratories Inc. (Maastricht, the Netherlands). Mrp14−/− mice, backcrossed >10 times to a C57BL/6 background were generated as described [28] and bred in the animal facility of the Academic Medical Center (Amsterdam, the Netherlands). The Animal Care and Use Committee of the University of Amsterdam approved all experiments.
Mice were intranasally inoculated with 104 K. pneumoniae serotype 2 (ATCC 43816 Rockville, MD) in a 50 µl saline solution (n = 7–8 per strain) and sacrificed 6, 24 or 48 hours thereafter [36], [37]. In an additional study, mice were intravenously injected with K. pneumoniae (2×103 colony forming units (cfu)) in the tail vein and sacrificed 48 hours after infection. Collection and handling of samples were done as previously described [36], [37]. In brief, blood was drawn into heparinized tubes and organs were removed aseptically and homogenised in 4 volumes of sterile isotonic saline using a tissue homogenizer (Biospec Products, Bartlesville, UK). To determine bacterial loads, ten-fold dilutions were plated on blood agar (BA) plates and incubated at 37°C for 16 h. In survival studies mice (n = 12 to 16 per strain) were intranasally inoculated with 103 or 104 K. pneumoniae and monitored for up to 10 days after infection. Bronchoalveolar lavage (BAL) fluid was obtained from a separate group of infected Wt mice (n = 6) at indicated time points. The trachea was exposed through a midline incision and cannulated with a sterile 22-gauge Abbocath-T catheter (Abbott Laboratories, Sligo, Ireland). Bilateral BAL was performed by instilling two 0.5 ml aliquots of sterile phosphate buffered saline (PBS) as described earlier [36]. 0.9–1 ml of BAL fluid was retrieved per mouse.
Lung homogenates were prepared for immune-assays as described before [36], [37]. MRP8/14 levels were measured by ELISA [15]. Lung cytokines and chemokines TNF-α, interleukin (IL)-1-β, IL-6, IL-10, Keratinocyte-derived chemokine (KC) and macrophage inflammatory protein 2 (MIP-2)(all R&D systems, Minneapolis, MN) and Myeloperoxidase (MPO; Hycult Biotechnology BV, Uden, the Netherlands) were measured using specific ELISAs according to manufacturer's recommendations. Plasma TNF-α, IL-6 and IL-1β were measured by cytometric bead array flex set assay (BD Biosciences, San Jose, CA) in accordance to the manufacturer's instructions. Lactate dehydrogenase (LDH), aspartate aminotransferase (AST) and alanine transaminase (ALT) were measured in plasma with kits from Sigma (St. Louis, MO), using a Hittachi analyzer (Boehringer Mannheim, Mannheim, Germany).
Lung and liver pathology scores were determined as described before [36]–[38]. In brief, lungs and livers were harvested at the indicated time points, fixed in 10% buffered formalin, and embedded in paraffin. 4 µm sections were stained with haematoxylin and eosin (HE) and analyzed by a pathologist blinded for groups as described earlier. To score lung inflammation and damage, the entire lung surface was analyzed with respect to the following parameters: bronchitis, edema, interstitial inflammation, intra-alveolar inflammation, pleuritis, endothelialitis and percentage of the lung surface demonstrating confluent inflammatory infiltrate. Each parameter was graded 0–4, with 0 being ‘absent’ and 4 being ‘severe’. Livers were scored according to the following parameters: number of thrombi, number of (micro)abscesses, presence and degree of inflammation, and presence and degree of necrosis. Each parameter was graded 0–3, with 0 being absent and 3 being severe. The total pathology score for lungs and livers was expressed as the sum of the score for all parameters. Granulocyte staining was done using FITC-labeled rat anti-mouse Ly-6 mAb (Pharmingen, San Diego, CA) as described earlier [51]. Ly-6G expression in the lung tissue sections was quantified by digital image analysis [52]. In short, lung sections were scanned using the Olympus Slide system (Olympus, Tokyo, Japan) and TIF images, spanning the full tissue section were generated. In these images Ly-6G positivity and total surface area were measured using Image J (U.S. National Institutes of Health, Bethesda, MD, http://rsb.info.nih.gov/ij); the amount of Ly-6G positivity was expressed as a percentage of the total surface area. MRP8 and MRP14 staining of lung tissue were performed as described previously [28]. For immunostainings, specimens were processed similarly as described previously [24]. Briefly, samples were deparaffinized, rehydrated in decreasing concentrations of EtOH, and subjected to antigen retrieval by cooking in 10 mM citrate buffer, pH 6.0, for 10 min. Specimens were blocked with 2% BSA and mouse Ig blocking reagent according to manufacturer's protocol (Vector Laboratories, Burlingame, USA) in PBS/0.1% Triton for 1 h at room temperature. Subsequently, specimens were incubated with primary antibodies directed against myeloperoxidase (MPO) (A0398, Dako) and histone H1 (clone AE-4, Acris) diluted in blocking solution over night at 4°C. Primary antibodies were detected with Alexa Fluor 488- and 568-conjugated secondary antibodies (Life Technologies) diluted in 2% BSA in PBS/0.1% Triton. DNA was visualized with DAPI (Life Technologies) and slides were mounted with fluorescence mounting medium (Dako). Pictures were taken with a Nikon C1 confocal microscope and presented as maximum intensity projections from parts of Z-stacks.
Growth-arrested bacteria were prepared as described [53] In brief, K. pneumoniae were cultured and washed with pyrogen-free sterile saline and resuspended in sterile PBS to a concentration of 2×109 bacteria/ml. The concentrated K. pneumoniae preparation was treated for 1 h at 37°C with 50 µg/ml Mitomycin C (Sigma-Aldrich; Zwijndrecht, the Netherlands) to prepare alive but growth-arrested bacteria. Subsequently, the growth-arrested K. pneumoniae preparation was washed twice in ice-cold sterile PBS by centrifugation at 4°C, and the final pellet was dispersed in ice-cold PBS in the initial volume and transferred to sterile tubes. Undiluted samples of these preparations failed to generate any bacterial colonies when plated on BA plates, indicating successful growth arrest. Bacteria were washed and resuspended in RPMI and diluted to ten-fold lower bacterial concentrations (2×104–7 cfu/ml). 100 µl of heparinized whole blood obtained from 4 individual Wt, mrp14−/− and tlr4−/− [54] mice were then incubated with 100 µl of the different bacterial concentrations in a 96 wells plate and incubated for 6 hours at 37°C, 5% CO2. After incubation, plates were centrifuged at 4°C and supernatant was harvested for determination of TNF-α.
Phagocytosis of K. pneumoniae was determined as described before [55]. In brief, growth-arrested bacteria were prepared as described above and labeled with carboxyfluorescein succinimidyl ester (CFSE, Invitrogen, Breda, the Netherlands). 50 µl heparinized whole blood was incubated with 50 µl bacteria in IMDM (Gibco) (end concentration of 1×107 CFU/ml) at 37°C (n = 6 per group) or 4°C (n = 3 per group). After 20 minutes, samples were put on ice to stop phagocytosis. Afterwards, red blood cells were lysed using isotonic NH4Cl solution (155 mM NH4Cl, 10 mM KHCO3, 100 mM EDTA, pH 7.4). Neutrophils were labeled using anti-Gr-1-PE (BD Pharmingen, San Diego, CA) and washed twice in FACS-buffer (0.5% BSA, 0.01% NaN3, 0.35 mM EDTA in PBS) for analysis. Peritoneal macrophages (derived from 3 mice) were pooled, washed twice and resuspended in IMDM containing 2 mM L-glutamine and 10% fetal calf serum (FCS) (Gibco). 1×105 cells per well were seeded in a 96-well flat-bottom plate in 250 µL to adhere overnight at 37°C, 5% CO2. The following day, macrophages were washed twice with pre-warmed medium to wash away non-adherent cells. Growth-arrested bacteria were opsonised in 10% normal mouse serum before added to cells at a multiplicity of infection of 100 in a volume of 100 µL. Bacteria and macrophages were spun at 1000 RPM for 5 minutes and incubated at 37°C (n = 8 wells per strain) or 4°C (n = 4 wells per strain). After 1 hour, samples were washed twice with ice-cold PBS, then thoroughly scraped from the bottom and washed again in FACS-buffer. The degree of phagocytosis was determined using FACSCalibur (Becton Dickinson Immunocytometry, San Jose, CA.) The phagocytosis index of each sample was calculated as follows: geo mean fluorescense×% positive cells.
Recombinant mouse MRP8 and MRP14 homodimers as well as MRP8/14 heterodimers were generated as previously described [56]. To test growth inhibitory effects of MRPs on K. pneumoniae, bacteria were grown to log phase and diluted to approximately 10.000 cfu/ml in HEPES buffered RPMI. 100 µl of this bacterial suspension was added to 100 µl of recombinant murine MRP8 or 14 homodimer or MRP8/14 heterodimer in HBSS (end concentration 50 µg/ml unless indicated otherwise; n = 4–6) without Ca2+ and Mg2+ (Gibco). Bacteria and MRPs were incubated for 0, 2, 4, 8 or 24 hours at 37°C. To test reversibility of growth inhibitory effects on K. pneumoniae, increasing concentrations of ZnSO4 were added to a 10 µg/ml MRP8/14 solution. Growth was assessed by plating out ten-fold dilutions of bacterial concentrations on BA plates and incubation at 37°C for 16 h.
Murine neutrophils were isolated from Wt and mrp14−/− mice; animals used for these experiments were bred in the animal facility of the Umeå University (Umeå, Sweden). Experiments were carried out according to the recommendations in the guide for the care and use of laboratory animals conformed to Swedish animal protection laws and applicable guidelines (djurskyddsmyndigheten DFS 2004:4) and were approved by the local ethics committee (Dnr A 29-09). Mature murine neutrophils were isolated from bone marrow as previously described [57]. Briefly, bone marrow cells from tibia and femur were singularized by using a 70-µm cell strainer and separated by centrifugation for 30 minutes at 1500 g on a discontinuous Percoll gradient with 52% (vol/vol), 69% (vol/vol), and 78% (vol/vol). Neutrophils harvested from the distinct layer between 69% and 78% were resuspended in HBSS without Ca2+ and Mg2+ until use. Murine NETs were induced as described earlier [24], [27], [57]. In a 24-well plate, 5×105 Wt or mrp14−/− mouse neutrophils in 500 µL RPMI with 1% (vol/vol) mouse serum were stimulated with 100 nM phorbol 12-myristate 13-acetate (PMA, Sigma Aldrich, St. Louis, MO) for 20 hours at 37°C, 5% CO2 to induce NET formation. The supernatant was discarded; NETs were washed once with RPMI and incubated with 500 µL RPMI containing approximately 5000 cfu K. pneumoniae. Subsequently plates were centrifuged for 5 minutes at 300 g and incubated for 7 hours at 37°C.
Human neutrophils were isolated from peripheral blood of healthy donors using Polymorphprep (Axis-Shield, Oslo, Norway) according to the manufacturer's instructions. Neutrophils were harvested and washed with HBSS without Ca2+ and Mg2+. Remaining red blood cells were lysed using sterile isotonic NH4Cl solution without EDTA for 10 minutes. After lysis of red blood cells, neutrophils were washed and resuspended in HBSS without Ca2+ and Mg2+ until use. In a 96-well plate, 3×105 human neutrophils resuspended in 50 µL HEPES buffered RPMI (Gibco) were stimulated with 100 nM PMA for 4 hours at 37°C, 5% CO2 to induce NET formation [25]. Sytox Green (Molecular Probes, Carlsbad, CA) confirmed the presence of NETs (data not shown). The supernatant was discarded and approximately 100 cfu log phase grown K. pneumoniae, P. aeruginosa or S. aureus, resuspended in 200 µL HEPES buffered RPMI were added to the wells and spun down for 5 minutes at 300 g. To test growth inhibiting properties of endogenous MRP8/14 in human NETs, samples were pre-incubated for 30 minutes with 15 µg/ml rabbit polyclonal anti-MRP14 antibody (H00006280-D01P; Abnova, Taipei, Taiwan), unspecific rabbit polyclonal control antibody or an excess of ZnSO4 (1 mM), before the addition of bacteria. Overnight bacterial growth was assessed by plating out ten-fold dilutions of bacterial concentrations on BA plates and incubation at 37°C for 16 hours. Bacterial growth of Klebsiella is expressed as percentage of control values (K. pneumoniae growth in media without neutrophils with or without anti-MRP14, control antibody or ZnSO4).
Data are expressed as box-and-whisker diagrams depicting the smallest observation, lower quartile, median, upper quartile and largest observation unless indicated otherwise. Differences between mrp14−/− and Wt mice were analyzed by Mann-Whitney U test. Survival was compared by Kaplan-Meier analysis followed by a log rank test. Analyses were done using GraphPad Prism version 5.0, Graphpad Software (San Diego, CA). Values of p<0.05 were considered statistically significant different.
|
10.1371/journal.pgen.1006771 | Ribosomal DNA copy number loss and sequence variation in cancer | Ribosomal DNA is one of the most variable regions in the human genome with respect to copy number. Despite the importance of rDNA for cellular function, we know virtually nothing about what governs its copy number, stability, and sequence in the mammalian genome due to challenges associated with mapping and analysis. We applied computational and droplet digital PCR approaches to measure rDNA copy number in normal and cancer states in human and mouse genomes. We find that copy number and sequence can change in cancer genomes. Counterintuitively, human cancer genomes show a loss of copies, accompanied by global copy number co-variation. The sequence can also be more variable in the cancer genome. Cancer genomes with lower copies have mutational evidence of mTOR hyperactivity. The PTEN phosphatase is a tumor suppressor that is critical for genome stability and a negative regulator of the mTOR kinase pathway. Surprisingly, but consistent with the human cancer genomes, hematopoietic cancer stem cells from a Pten-/- mouse model for leukemia have lower rDNA copy number than normal tissue, despite increased proliferation, rRNA production, and protein synthesis. Loss of copies occurs early and is associated with hypersensitivity to DNA damage. Therefore, copy loss is a recurrent feature in cancers associated with mTOR activation. Ribosomal DNA copy number may be a simple and useful indicator of whether a cancer will be sensitive to DNA damaging treatments.
| The ribosomal DNA encodes the RNAs needed to make ribosomes for protein synthesis and cellular proliferation. However, ribosomal DNA has been excluded from most mammalian genome-wide studies due to challenges associated with its analysis. We find that both the sequence and copy number of the ribosomal DNA can change in human cancer genomes. mTOR is a kinase that senses the nutritional environment and is often over-active in cancer. Given mutational evidence for mTOR activation in the human cancer genomes with loss of ribosomal DNA copies, we analyzed ribosomal DNA in hematopoietic stem cells derived from mice under conditions of mTOR activation. Like the human cancer genomes, the ribosomal DNA copy number contracts in mTOR activated hematopoietic stem cells relative to normal stem cells. Loss is associated with high rates of cellular proliferation, rRNA production, and protein synthesis, but compromised survival in the presence of DNA damage. Contractions are a recurrent feature in cancer genomes with overactive mTOR and may predict sensitivity to DNA damaging chemotherapeutics. Ribosomal DNA may be altered in other disease contexts.
| Repetitive regions in the human genome include ribosomal DNA (rDNA), telomeres, and centromeres, and these regions have profoundly important functions for genome stability [1,2,3]. While challenging to study, genetic variability of these regions is ripe for selection. The rDNA encodes the functional RNAs of the ribosome, the most conserved and utilized genes in the genome. Three of the four RNAs are transcribed initially as a single transcript by RNA polymerase I. The repeats encoding the RNAs are present on five different regions on human acrocentric chromosomes, with each region composed of several copies of the gene, referred to as 45S. 45S is processed into three separate RNAs (28S, 5.8S, and 18S). The fourth RNA is encoded by the 5S genes, located on human chromosome 1 and transcribed by RNA polymerase III. The number of 5S copies correlates with 45S [4]. The loci that encode the rRNAs are transcribed at incredibly high rates, with rRNA often constituting as much as 80% of the RNA in actively proliferating cells, to meet the demand for ribosome production. Even so, only about 50% of the repeats are transcribed.
The number of repeats is variable in human genomes [5], with one recent computational study estimating between 14–410 copies of 45S [4]. The variability in number combined with five different chromosomal locations makes the 45S pattern unique in each human genome. The human 45S genes are highly recombinogenic, with an estimated 10% recombination in a single generation [6]. The sequence of the 45S repeat within a species is highly conserved due presumably to high rates of recombination which allow for concerted evolution [7]. While repeats exist mainly in a head to tail tandem arrangement, palindromic arrangements have also been observed, especially in syndromes with increased genomic instability [8].
One can imagine that one way for a cell to achieve high rates of protein synthesis would be to expand the number of rDNA copies, especially given that ribosomal RNA can be limiting for ribosome biogenesis [9]. In fact, increases in copy number occur in different developmental and disease contexts. Expansions occur in frog oocytes [10] and the macronucleus of tetrahymena [11] as part of the normal course of development, presumably to help achieve high rates of ribosome biogenesis and protein synthesis. Southern blotting experiments demonstrate that recombination at the rDNA is common in adult lung and colorectal cancer [12], suggesting the possibility that the rDNA may be affected in the cancer genome. However, copy number of the rDNA in mammalian development and disease has not been examined.
The TOR (target of rapamycin) kinase detects the nutritional environment and promotes ribosome biogenesis when amino acids are plentiful. The PTEN (phosphate and tensin homolog) phosphatase is a negative regulator of mTOR activity and acts as a tumor suppressor. Somewhat paradoxically, overfeeding flies, which overstimulates mTOR, has been associated with contraction of the rDNA copy number in the germline in Drosophila, transmitted transgenerationally [13]. TOR signaling was required for the amplification from low copy number back to normal copy number in budding yeast [14]. Together these results suggest there may be a connection between TOR signaling and the rDNA copy number, but this has not been explored in the mammalian genome. Furthermore, if the copy number or sequence changes in response to the environment, then rDNA may act as both a sensor and adaptor under stress conditions.
In addition to its role in ribosome biogenesis, the rDNA has many extra-ribosome functions, such as regulating gene expression, chromosome organization, and titrating chromatin factors [15,16,17,18]. The nucleolus, which contains the ribosomal DNA, is a hub for many signaling pathways and can act as a stress sensor. Chromosomal domains from most human chromosomes are associated with the nucleolus [19,20], highlighting its role as an organizer. Dramatically reducing the copy number in yeast does not cause reduction in rRNA levels, but causes all repeats to become active (sometimes termed “compensation”) and the yeast are hypersensitive to DNA damage [21]. In contrast, low copy compromises protein synthesis and development in the bobbed mutants in Drosophila [22]. Furthermore, copy number of the rDNA titrates position effect variegation in flies [15,18,23]. Therefore, modulation of copy number has the potential to regulate signaling, genome stability, chromosome organization, and gene expression without necessarily affecting ribosome biogenesis.
Recent technological developments enable accurate and high throughput measurements of the copy number of the rDNA. Copy number can be estimated from whole-genome sequence (WGS) and droplet digital PCR (ddPCR). The plethora of cancer genome sequencing projects allows comparison between the copy number in the tumor genome versus the normal genome from a single individual. We used these new technologies to analyze how rDNA copy number changes in different tissues and in cancer. Surprisingly, rDNA copy number is reduced in cancer genomes with mTOR activation, providing evidence that copy number can be altered in a natural system. We also discovered co-occurring increased copy number of additional genes, and evidence for single nucleotide variation in the repeats. Using cancer stem cells derived from a mouse model for leukemia driven by loss of Pten, we find that copy loss does not compromise proliferation, rRNA production, or protein synthesis, but cells are hypersensitive to DNA damage. These data suggest that the ribosomal DNA can act as a sensor and adaptor to cancer-associated stress. Copy number may be a biomarker with predictive potential.
In the mouse genome, the 45S repeats are located on 5 different chromosomes (12, 15, 16, 18, 19) whereas all the 5S repeats are located on chromosome 8. The copy number for the two repeats co-varies and has been previously estimated using a computational approach from sequence data derived from a collection of laboratory and wild mice, showing a 10-fold range with values for 45S of 31–289 and for 5S of 32–224 [4]. We wanted to extend these findings by developing a ddPCR method to directly measure the copy number of the 45S repeat with high accuracy in commonly used laboratory strains including the inbred strains C57BL/6 and DBA/2J, and the outbred strain CD-1. In addition, we wanted to know whether copy number varied by tissue given different metabolic requirements. Mitochondrial DNA (mtDNA) copy number has been reported to have an inverse relationship with rDNA copy number in human blood samples [17] and is thought to fluctuate with metabolic demand.
To answer whether rDNA copy number is variable across different mouse strains, tissues, and organs, we collected genomic DNA (gDNA) from fifteen tissues of three mouse strains, and performed ddPCR. The assay is designed such that the copy number of 45S and the single copy gene Gapdh1 are measured in the same reaction using fluorescent probes. We found the average copy number for C57BL/6 was 156 based on 57 tissue measurements from eight mice (standard deviation is 13.1), and DBA/2J was 123 based on two mice (standard deviation is 12.3) (Fig 1, S1 Fig). The mouse to mouse variation in the inbred strains was ~10%, with no obvious gender difference (S1 Fig). While the mean copy number for CD-1 was similar (145 based on 5 mice), the variation between individuals was much higher (standard deviation is 43.2), as much as 2-fold in the individuals tested. This demonstrates that mice from inbred strains tend to have a much more similar copy number to each other than individuals from the outbred CD-1 strain. The copy number variation in CD-1 is more like what has been reported in the human population.
Next we addressed how copy number varies across tissues and organs. To first address the level of variation in the ddPCR protocol, we isolated gDNA from the same tissue after dividing it into three samples. In this case the variability is usually less than 5% (S2 Fig). The results suggest that the accuracy of the measurements is appropriate to assess levels of variation higher than 5%. rDNA copy number differs by up to 15% between tissues from the same individual, and this is true in all three strains (Fig 1, S1 Fig). This variation is similar to individual to individual differences from the two inbred strains. From these studies, we conclude that the changes reported in mtDNA copy number in various tissues are not mirrored by equivalent changes in rDNA copy number, since the tissue to tissue variation in a single individual for rDNA is low but for mtDNA is high. Given the result that tissues from the same individual have very similar rDNA copy numbers, this opens the possibility of using matched cancer-normal genome data to explore rDNA copy number using previously developed computational methods that use genome coverage for normalization [17].
We analyzed high-coverage tumor-normal matched WGS data from 162 individuals across eight tumor types, including 19 childhood acute lymphoblastic leukemia samples (ALL, phs000341) [24], 36 medulloblastoma samples (phs000409) [25], 16 core-binding factor acute myeloid leukemia samples (CBF-AML, phs000414), 40 prostatic neoplasm samples (phs000447) [26], 12 AIDS-related lymphoma samples (phs000530), 12 intestinal neoplasm samples (Liver/Small bowel, phs000579) [27], 13 osteosarcoma samples (phs000699) [28], and 14 esophageal adenocarcinoma samples (phs000598) [29]. The normal WGS data was derived from either blood, solid normal tissue, or germline tissue (S1 Table). We also analyzed the rDNA copy number in WGS data derived from blood of 143 normal individuals (phs000424, GTEx). These datasets were chosen for three reasons: 1) we wanted representation of both solid and blood tumors, 2) WGS data was available and 3) permission to use the data for this analysis was granted. The use of ~16,000 exons for normalization and the comparison of many tumor-normal pairs helps to offset concerns regarding how aneuploidy in the cancer genomes might skew results (see Materials and methods).
To first verify the bioinformatics method, we compared our copy number measurements between the three different regions of the 45S gene for all the normal samples. We found good pairwise correlations between 18S, 5.8S, and 28S genes (Fig 2A), similar to those previously reported with this method [17]. We next calculated the copy number for the normal and tumor matched samples. The copy number of 18S, 5.8S, and 28S for the normal samples was subtracted from the tumor for each individual and plotted as the normalized copy number. We found three cancer genome projects, osteosarcoma, AIDS-related lymphoma, and esophageal adenocarcinoma, for which there was a statistically significant reduction in copy number (~70–90 copies) for the tumor genomes (Fig 2B–2D). The other five genome projects did not show significant changes (S3 Fig), suggesting the method can discern loss vs. no loss. We found this result remarkable because it suggests that over the course of tumor development a lower copy number at the rDNA is selected in some cancers. This is opposite to the idea that more copies might be needed in cancer to reach high rates of ribosome biogenesis required for high rates of protein synthesis and proliferation. Instead, some other aspect of high proliferation may select for a lower copy number, for instance, efficient DNA replication. These findings represent the first demonstration that rDNA copies can be lost as part of cancer development.
We used the ~16,000 exons selected for normalization, which are scattered across all chromosomes, to examine whether the chromosomes with rDNA (13, 14, 15, 21, 22) are present at lower levels relative to other chromosomes in the tumor genomes, since this would indicate chromosome loss as one mechanism to lose rDNA copies. These exons were preselected based on the criteria that they were derived from the largest exon (300 bps or larger) of a single gene, mapped uniquely in the genome, and did not have sequence similarity to each other. Using sequencing coverage, we found evidence that chromosomes in the tumor genomes show more variable coverage than in the normal genomes (S4A Fig), presumably reflecting increased aneuploidy. However, the chromosomes with rDNA are not lost at higher rates in genome projects that are positive for rDNA loss compared to the genome projects negative for rDNA loss (S4B Fig). These results suggest preferential loss of rDNA chromosomes is not the predominant mechanism underlying loss of rDNA copies in the cancer genome projects showing a net loss of rDNA copies.
To further analyze copy number variation in the cancer genome projects, we used the copy number of the preselected exons used for normalization to discover additional genes whose copy number co-varied with the ribosomal DNA. Exons were selected that displayed a significant copy number difference based on results of a paired t test for all the tumor versus all the normal values for each exon for each genome project (FDR< = 0.05). For the three genome projects with low rDNA copy number in tumor versus normal (phs000530, phs000598, phs000699) there were 353 such exons. We then used hierarchical clustering to identify exons with similar trends in copy number (gain or loss) in the three “positive” projects compared to the five “negative” projects. The “gain” cluster was the most striking, with ~100 exons (Fig 3A). These exons are spread across all chromosomes and are not confined to the chromosomes with 45S genes, indicating global rather than local structural variations accompany the loss of copies. This result suggests there may be a global concerted genomic signature associated with decreased rDNA copies in cancer. The GO terms associated with the genes from which these exons were derived include DNA damage response and metabolic regulation, among others (Fig 3B, S2–S4 Tables), suggesting the increase in copy number of these genes could affect cellular physiology. While none of these terms reach statistical significance when adjusted for multiple hypothesis testing, there are conflicting views on whether this adjustment is necessary for GO term analysis. Unfortunately, RNA-seq data was not available for these cancer genome projects, which would have allowed us to evaluate if there was a corresponding gene expression signature.
The rDNA represents a prime example of concerted evolution, which refers to the fact that within a species there is very little sequence variation, but there is significant variation between species [7]. The homogeneity is thought to be maintained via frequent and continuous sweeps of recombination. However, variation in the ribosomal DNA sequence has not been assessed in the human genome. We identified single nucleotide variants (SNVs) and indels in tumor and normal genomes relative to a common reference. Indels were rare so we focused on SNVs. Within a single individual, the majority of SNVs (~90%) detected relative to the reference were shared between the tumor and normal genome. However, subtraction of the shared SNVs still yielded several unique SNVs. The number of unique SNVs were similar in tumor and normal genomes across the 8 different projects (Fig 4A, S5 Table). When taken in aggregate, the number of SNVs in the cancer genomes is not higher than in the normal genomes. These results suggest that as cells divide, SNVs can arise at the rDNA and propagate at a similar level in tumor and normal cells.
One might predict the coding portion of the locus to experience selective pressure to generate functional ribosomes for high proliferation. SNVs were binned for each region of the repeat (Fig 4B). Previous work in yeast has suggested the most variable region is the non-transcribed portion, because mutations here might have the least functional consequences [30]. The human repeat is ~30% transcribed (ETS, ITS, 18S, 5.8S, 28S) and ~70% non-transcribed (IGS). We found that the regions encoding 18S and 5.8S had virtually no SNVs in either the tumor or normal genomes, suggesting these sequences cannot tolerate mutations, consistent with a previous report in yeasts [30]. Surprisingly, SNVs were detected in the 28S region, with a similar number in tumor and normal genomes. We detected hotspots for variation; the 189 SNVs were located at only 22 unique positions (Fig 4C). SNVs in the gene encoding 28S rRNA have the potential to affect ribosome function if the corresponding repeats are transcribed.
All non-coding regions of the human repeat had SNVs, including the 5’ and 3’ external transcribed spacer regions, the internal transcribed spacers, as well as the large intergenic region, suggesting these regions can tolerate SNVs. Interestingly, the IGS did not appear to be mutated at a significantly higher rate than the overall repeat (Fig 4B, S6 Table). SNVs in the transcribed non-coding regions have the potential to affect transcription and processing of rRNA but are difficult to functionally evaluate given our limited understanding of the sequence features of the human repeat. In summary, variation in the non-coding region is higher than in the coding region, regardless of transcription. Moreover, the amount of variation is generally similar in the normal and tumor genomes.
Another way to quantify variation that pertains specifically to a multi-copy sequence in the genome is allele number. By this we mean that individual nucleotide positions could have sequence evidence for more than one base, which we refer to as alleles. For the shared SNVs, we evaluated the number of alleles present in the cancer and normal genomes, to ask whether there was differential presence of alleles. We found three genome projects with evidence for more minor alleles in the tumor as compared to normal genomes (Fig 4D). The modest increase observed in prostate neoplasms (phs000447) could be due to the handful of cancer genomes in this project with very high copy number. Most notably, the esophageal adenocarcinoma (EAC) genomes have an increased allele frequency relative to the matched normal genomes (phs000598), despite having a lower copy number. Therefore, in the EAC genomes, there are more alleles of the 45S gene sequence despite fewer copies. The initial publication describing the analysis of the EAC genomes reported mutational complexity with accumulated mutations including activating mutations in PIK3CA, and loss of function mutations in PTEN, PIK3R1, AKT2, and AKT3, together implicating the mTOR-PTEN pathway in some instances of this cancer [29]. Our results reveal that the mutational complexity of this cancer also encompasses variation in 45S. Overall, our sequence analysis of the human 45S repeats indicates a surprising level of SNVs with newly identified hotspots in the human 28S gene, and some cancer genomes with evidence for increased alleles.
In the osteosarcoma genome project the tumors are driven by increased mTOR pathway activity, which can occur via multiple different mutations, including loss of PTEN, a negative regulator of mTOR [28]. Loss of PTEN in prostate cancer is associated with tumor aggression and poor outcome. In fact, mutations that activate the PI3K (phosphatidylinositol-3-kinase)-mTOR pathway are frequently found in many cancers, including AIDS-related lymphomas [31], and esophageal cancer, as mentioned above [29,32,33,34]. PTEN is widely known as a tumor suppressor and a guardian of the genome [35]. Because the samples used for the cancer genome projects have a mixed population of mutations, may have genetic instability, and may be derived from mixtures of cell types, we wanted to specifically address whether high mTOR function could influence rDNA copy number in the context of a uniform genetic alteration (loss of Pten) in a pure cell population.
We used a mouse model for Pten-/- leukemia in the C57Bl6 background to assess rDNA copies in hematopoietic stem cells (HSCs)[36]. Pten is excised in bone marrow HSCs postnatally, but remains intact in all tissues other than the progeny of null HSCs. HSCs were isolated from mice using flow sorting of Lineage-, Sca-1+, c-Kit+, and CD34- cells from bone marrow prior to progression to leukemia; these cells are cancer stem cells which will eventually develop to myeloid proliferative disorder and leukemia in vivo or upon transplant [37]. These cells are not aneuploid, based on analysis of a model with similar progression to leukemia [38]. Pten null cells isolated from bone marrow show more colony forming units and reduced quiescence, demonstrating increased proliferation [36,37]. We verified that loss of Pten in HSCs results in a pathway signature indicative of mTOR activation, including higher levels of phosphorylated mTOR, Rps6, and S6K1 (S5 Fig).
Remarkably, the Pten-/- HSCs have significantly lower copy number compared to two different controls, tail samples from the same mice which retain Pten and matched Pten+/+ HSCs (Fig 5A and 5B). To examine rDNA copy number, we collected WT and Pten-/- HSCs from 20 single-cell derived clones from two different mice per genotype (see Materials and methods), and extracted gDNA from these samples for the ddPCR assay. Each Pten-/- clone had lost about 40 copies, approximately 20% of their total. HSC clones derived from three additional WT-null pairs showed similar results (S5 Fig), for a total of 5 pairs of female mice. Therefore, loss of Pten in HSCs was associated with significant loss of rDNA copies while the copy number in the tails was comparable.
We constructed and sequenced libraries from genomic DNA derived from 20 WT and 20 Pten-/- clones derived from two age and sex matched females. Using our computational method, counting of the 5.8S and 18S sequences showed a loss of ~30 copies in the Pten-/- clones, and 28S showed a loss of ~20 copies (S6 Fig). Given that the libraries were made using a transposase method that has some bias, the agreement between the computational and ddPCR methods is quite good, and lends confidence in the overall trend toward loss in Pten-/- HSCs. This trend is consistent with the lower rDNA copies observed in the genomes of osteosarcoma, AIDS-related lymphoma, and esophageal adenocarcinoma. WT and Pten-/- clones have similar ploidy (S6 Fig), indicating that chromosome loss is not the mechanism by which rDNA is lost in this context. Further analysis of the DNA sequences from the 40 clones revealed 110 SNVs relative to the consensus sequence that were shared between WT and Pten-/- clones. There were a handful of SNVs that occurred uniquely in Pten-/- (9) or WT clones (28), with most occurring in the large IGS region (5 for WT and 17 for Pten-/-). No SNVs were detected in 18S or 5.8S, but 3 were detected in 28S. Overall the SNV pattern is similar to that observed in the human genome. Together these results suggest that in some cancers, and specifically in loss of PTEN, loss of rDNA copies may provide some selective advantage for the cancer genome. Furthermore, in the Pten-/- HSC model, loss of copies is a relatively early event, prior to progression to leukemia or aneuploidy.
PTEN plays a critical role in the maintenance of chromosome stability, preventing double-strand breaks (DSBs) [39]. Deletion of PTEN in prostate cancer cells is associated with sensitivity to DNA damaging agents including ionizing radiation, mitomycin-C, UV, H2O2, and methyl methanesulfonate (MMS)[40]. Loss of PTEN in HCT116 colon cancer cell lines confers sensitivity to ionizing radiation [41]. We found that the Pten-/- HSCs were more sensitive than WT HSCs to DNA damage, including bleomycin and ionizing radiation (Fig 5C and 5D, S5 Fig), and MMS and hydroxyurea at the higher concentrations tested (Fig 5E and 5F, S5 Fig), consistent with Pten protecting against DNA damage in HSCs. HSCs were stained with antibodies to the nucleolar proteins fibrillarin and nucleolin, and nucleoli were imaged. Despite loss of copies, nucleolar size is not significantly different between Pten-/- and Pten+/+ HSCs (S7 and S8 Figs).
To further examine how loss of copies affects cell function, we compared Pten-/- and control HSCs in culture. Pten-/- single cell derived colonies have more cells than controls (Fig 6A), consistent with previous reports that Pten-/- HSCs proliferate faster. Furthermore, we examined the production of rRNA using 3H-uridine incorporation. Because rRNA is the most abundant RNA product, this method has been widely used as a proxy to measure rRNA synthesis. We found that rRNA production is more robust in the Pten-/- HSCs compared to Pten+/+ HSCs despite reduced rDNA copy number (Fig 6B). This result is consistent with previous reports that PTEN normally represses both RNA polymerase I- and RNA polymerase III-dependent transcription [42,43]. Moreover, global protein synthesis, as measured by 35S-methionine incorporation, was significantly higher in the Pten-/- HSCs (Fig 6C), consistent with a previous study [44]. Therefore, loss of copies and DNA damage sensitivity occurs in the context of more robust proliferation, rRNA production, and protein synthesis.
To determine whether the DNA damage sensitivity was due to fewer copies or the activation of mTOR, we treated HSCs with INK128 to block TOR activity and asked whether the Pten-/- HSCs were still sensitive to DNA damage. For these experiments HSCs were derived from 3 age matched pairs of male mice. First we identified a concentration of INK128 that would effectively block TOR activity in the HSCs (Fig 7A). Second, we measured copy number by ddPCR, finding loss of ~30 copies in Pten-/- HSCs that occurred independent of treatment with INK128 (Fig 7B). The copy number in the tails of mice was similar across animals and similar to the copy number in the WT HSCs (Fig 7C). Next we measured rRNA production (Fig 7D) and protein synthesis (Fig 7E). These outputs depended on mTOR activity, as expected, and were effectively normalized between the genotypes in the presence of INK128. Together with the results in Fig 6, these results demonstrate similar behavior between HSCs derived from male and female mice, with 8 pairs examined in total. Finally, we examined the sensitivity of the mTOR-blocked HSCs to bleomycin. Importantly, the Pten-/- HSCs are more sensitive to bleomycin than the WT HSCs (Fig 7F), suggesting that the sensitivity is not due to differential activation of mTOR, but could derive from the lower copy number.
Finally, we asked whether copy number can change rapidly in human retinal pigment epithelial (RPE) cells grown in culture. RPE cells have a normal karyotype such that the droplet digital PCR method, which uses a single copy reference gene to calculate copy number, can be applied. RPE cells were transfected with 3 different siRNAs to knockdown PTEN, or a control siRNA. The knockdown of PTEN was confirmed by Western blot (S9 Fig). The copy number of the RPE cells was determined at 80 hours post-transfection, or after about 3 doublings. Under these conditions the copy number is not altered (S9 Fig). This result suggests that loss may require more doublings, or may require other environmental factors provided in an animal, such as cell-cell competition. This finding suggests that the loss of copies is not an immediate event upon loss of PTEN and is consistent with the result in the HSCs that short term treatment with an mTOR inhibitor is not sufficient to alter the copy number.
Our findings suggest that in hematopoietic cancer stem cells, cell growth and ribosome biogenesis can occur robustly with ~30–40 fewer rDNA repeats. Interestingly, in budding yeast, loss of rDNA copies is associated with sensitivity to DNA damage [21]. Our results show that mouse HSCs without Pten function have fewer copies and are similarly sensitive to DNA damage. With only half of the rDNA repeats normally transcribed, loss of repeats can be compensated by increasing the fraction transcribed in yeast. However, the binding of factors that normally associate with the inactive repeats to maintain the stability of the locus may be compromised, increasing the sensitivity of yeast with fewer repeats to DNA damage [45]. It remains to be determined whether compensation occurs in the mammalian genome, and whether the absence of silenced repeats could cause DNA damage sensitivity. Nevertheless, copy number may be a useful predictor of DNA damage sensitivity.
We have used both computational approaches and ddPCR to analyze the copy number and sequence of the rDNA in mammalian cells. We find tissue to tissue variation in copy number in a single mouse is relatively low, as is individual to individual variation within an inbred strain. However, in an outbred mouse strain, the level of variation is higher, more resembling the situation in human genomes. We report for the first time that some cancer genomes, and in particular genomes associated with high mTOR activity, tend to have fewer copies than the matched normal genomes, a finding replicated in mouse cancer stem cells, and consistent with a previous report demonstrating transgenerational loss of rDNA copies in D. melanogaster with overactive TOR [13]. The low copy cancer genomes show concerted copy number changes in additional genes, suggesting a structural signature for these genomes. rDNA sequence variation can also occur in cancer genomes. Low copy in the Pten-/- HSCs is associated with sensitivity to DNA damage, extending previous reports that PTEN guards against genome instability. Our results suggest that copy number and sequence can change in the mammalian genome in cancer, and that loss of copies may have both costs and benefits. Future studies further analyzing the mechanisms and functional consequences of genomic alterations at the rDNA in mammals are warranted. Given our findings, it seems possible that ribosomal DNA could exhibit changes in other contexts.
Loss of rDNA repeats may affect genome function and chromosomal processes via the release of protein factors. The rDNA locus houses many pluripotency factors in mammalian cells, including Oct4 [46]. The rDNA also contains binding sites for Myc [47], a key transcription factor for proliferation that can also affect DNA replication [48], and CTCF [49], a key chromosome organization protein. The rDNA chromatin contains many different histone modifications [49], which could sequester chromatin readers and writers. Finally, nucleolar associated domains are enriched in regions displaying heterochromatin signatures in Arabidopsis [50] and rDNA copy number titrates position effect variegation, a heterochromatin based silencing phenomenon in Drosophila [23]. Losing repeats has the potential to liberate factors for re-distribution to the rest of the genome which could affect chromosome organization, gene expression, and replication.
Interestingly, PTEN has been shown to directly control the function of the DNA replication factor MCM2 during DNA replication stress [51]. Replication stress occurs at the rDNA and has been reported as a potent driver of functional decline in HSCs [52]. In a mouse model for cancer driven by deficiency in Mcm2, genomic deletions can occur [53]. Spontaneous DSBs also occur at higher levels upon loss of PTEN [39]. Together these findings suggest that one possible reason that rDNA repeats are lost in the absence of PTEN is that DSBs and replication stress are handled without MCM2 function, and repair events are required. The repair events that result in deletions may be selected for, since the rDNA is difficult to replicate and loss of repeats might facilitate a successful cell cycle. We suggest that rDNA may both sense and adapt to genomic stress, with PTEN-mTOR normally guarding against copy loss.
Extrachromosomal expansions of rDNA have been reported in frog oocytes [10]. This expansion is thought to facilitate the production of rRNA for ribosome biogenesis and the translational requirements in these cells. Based on this finding we predicted that cancer cells might expand the repeats due to similar requirements. Instead, we found recurrent evidence for contractions. We note that there is a key difference between the oocyte and the cancer cell—rounds of DNA replication. We speculate that the requirement to replicate the rDNA in cancer may select for the loss of repeats while the lack of replication in the oocyte may allow the expansion to be tolerated. The timing of the loss of the repeats relative to the development of cancer cannot be determined from the cancer genome projects. However, in the mouse HSCs loss occurs prior to the development of leukemia and aneuploidy, at the stage of a cancer stem cell. Further characterization of the selective forces and timing of loss will be interesting questions for the future.
Linking the rDNA to both genome stability and ribosome biogenesis may enable it to act as a critical molecular sensor. Changes in translation are associated with cancer [54]. Myc and PI3K-AKT-mTOR are major oncogenic signaling pathways that promote reprogramming of translation. Studies to date have focused on mRNA regulatory elements, tRNA function and codon usage bias, and adaptations to stress that affect translation, but not ribosomal DNA copy number as a factor under the control of TOR signaling. Protein synthesis is quite tightly regulated in HSCs and defects that arise via loss of Pten or silencing of rDNA repeats cause functional decline and aging [44,52]. Our study suggests that rDNA copy number and sequence can be altered in human cancers. HSCs with fewer copies are more sensitive to DNA damage, but are not compromised for rRNA production, proliferation, or protein synthesis. We speculate that the extra inactive copies may normally serve in part to counteract rDNA instability. Together, our data and others show that multiple mechanisms regulate protein synthesis and genome stability to control aging and prevent leukemogenesis in HSCs, and that these processes may be linked using the rDNA as a sensor.
In summary, the rDNA copy number and sequence can change in cancer, with high mTOR activity associated with contractions and DNA damage sensitivity. With this recognition comes the possibility to target these loci. Cancer stem cells with low copy number may be more sensitive to DNA damaging agents. We speculate that this single copy number measurement could be used as a proxy detector for the variety of mutations that can occur in the PI3K-AKT-mTOR pathway in cancer and a predictor of whether DNA damaging drugs would selectively target the cancer stem cells. These findings may be applicable to cancer diagnosis and therapeutic choice.
All animals were handled in strict accordance with good animal practice as defined by the relevant national and/or local animal welfare bodies, and all animal work was approved by the Stowers Institute for Medical Research, Institutional Animal Care and Use Committee.
Animals were sacrificed by carbon dioxide administration. All mice were 8 week old males or females, as noted. Fifteen tissues were isolated from three mouse strains including C57BL/6, DBA/2J, CD1. All the tissues were stored at -80°C. Most tissues or organs were cut into 2–3 small pieces to purify gDNA. For mouse HSCs, about 105 cells were used to extract gDNA. gDNA samples were extracted using the Maxwell® 16 Tissue DNA Purification Kit in Promega Corporation. The concentration of gDNA samples was measured by Qubit dsDNA HS Assay. For digital droplet PCR, 1 ng template was used per reaction.
Digital droplet PCR was used to measure rDNA copy number, performed per the manufacturer’s protocol (Bio-Rad). Briefly, isolated gDNA was digested with the restriction endonuclease Hae III. Reaction mixtures were made with target copy number variable sequence (45S rDNA) and internal control (Gapdh1 for mouse, TBP for human). Droplet generation was performed, followed by endpoint PCR. Each PCR product is detected by a fluorescent probe. Droplets were read by QX200 droplet reader, and quantitation was performed using Quantasoft software.
The Pten tamoxifen-inducible SCL-Cre mouse model has been previously described [36]. In brief, mice at 6–8 weeks of age were given 2 mg tamoxifen for 5 days. The 8 pairs of mice used for these experiments were age and sex matched. Both WT and PtenloxP/loxP (Lesche); HSC-Scl-Cre-ErT+ (Göthert) were treated with tamoxifen. HSCs were collected 12 days after the final injection. At this point these cells are considered leukemic stem cells. The timing of development of a clinically defined leukemia in vivo varies in this model, generally occurring within 3–4 months post-induction, but sometimes sooner. The Pten-/- HSC cause leukemia development after transplant. HSCs were sorted (in this case defined as lineage negative, Sca-1+, c-Kit+, and CD34- cells) by flow cytometry into methylcellulose semi-solid medium (M3434 media from Stem Cell Tech.). Single HSCs are sorted into individual wells in a 96-well plate. These single cells will (in about 35–50% of the wells) form a large colony of mainly myeloid and erythroid hematopoietic cells all derived from the single HSC. The individual colonies are harvested by incubating a flooded well in PBS for 20–30 minutes, and pipetting up and down to disassociate the colony.
Cells were rinsed once with ice-cold PBS and lysed in ice-cold lysis buffer (buffer A: 50 mM HEPES-KOH (pH 7.4), 2 mM EDTA, 10 mM pyrophosphate, 10 mM β-glycerophosphate, 40 mM NaCl, 1% Trition X-100 and one tablet of EDTA-free protease inhibitors (Roche) per 25 mL). The soluble fraction of the cell lysate was isolated by centrifugation at 12,000 g for 10 min. Sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) was performed using NuPAGE Novex 4%–12% Bis-Tris precast gels (Invitrogen). Western blotting was performed per standard protocol using a nitrocellulose (Whatman, Protran) membrane. The following antibodies were used: mTOR (Cell Signaling Technology, #2972), phospho-mTOR (Ser2448) (Cell Signaling Technology, #2971), phospho-S6K1 (Thr389) (Abcam company, ab126818), S6K1 (Cell Signaling Technology, #9202), phospho-RPS6 (Ser235/236) (Cell Signaling Technology, #2211), RPS6 (Santa Cruz Biotechnology, sc-74459), α-tubulin (Sigma Inc., T6199), phospho-Akt (Ser473) (Cell Signaling Technology, #4060), Akt (Cell Signaling Technology, #9272), PTEN (Cell Signaling Technology, #9188). Secondary antibodies were HRP linked, anti-rabbit IgG (from donkey) and anti-mouse IgG (from sheep), (GE Healthcare, NA934V, NA931V, and NA935V, respectively).
The 105−106 Pten+/+ and Pten-/- HSCs were seeded on 8-well chamber slides, with poly-L-lysine pre-coating overnight at 4°C. The HSCs were cultured for 24 hours for cell attachment. Next, the samples were washed in PBS, fixed for 10 min at room temperature 20–22°C with 4% paraformaldehyde, permeabilized for 5 min in PBS containing 0.5% Triton X-100 and washed in PBS. After blocking for 30 min in PBS containing 1% BSA at room temperature, the preparations were incubated overnight at 4°C with the following antibodies diluted in PBS containing 1% BSA: mouse anti-fibrillarin with 1:200 dilution, rabbit anti-nucleolin (H-250; Santa Cruz) with 1:200 dilution. The coverslips were then washed in PBS and incubated for 30 min at room temperature with the secondary antibodies Alexa Fluor 488 goat anti-mouse and Alexa Fluor 555 donkey anti-rabbit (Molecular Probes) diluted in PBS containing 1% BSA and then washed in PBS. The coverslips were mounted on slides and analyzed by fluorescence imaging with a Zeiss Axioplan II confocal microscope. The quantification of nucleolar size was performed in the fibrillarin and nucleolin stained regions using Image J software. Nucleolar size distributions and average sizes were determined using custom ImageJ plugins essentially as in [55] but with minor modifications to account for higher image resolution. Firstly, confocal images were maximum intensity projected. Briefly, a rolling ball background with a radius of 5 pixels was subtracted from each image. Next, images were smoothed with a Gaussian blur of 2 pixels (standard deviation). Finally, images were thresholded at 0.25 times the maximum processed intensity in the image. Each resulting distinct spot was considered a nucleolar region for area measurements.
Methods for rRNA labeling were derived from a previous report [55,56]. The cultured Pten+/+ and Pten-/- HSCs (105−106) were washed in PBS twice, and switched to fresh Dulbecco’s Modified Eagle’s Medium (Sigma) supplemented with 200 μM L-cysteine (BSA, Sigma), 50 μM 2-mercaptoethanol (Sigma), 1 mM L-glutamine (Gibco) and 0.1% bovine serum albumin (Sigma). HSCs were pulse labeled with 3H-uridine (5 μCi) for the indicated amount of time (0, 30, 60, 120 min) per sample. Total RNA was isolated with TriZol reagent (Invitrogen, U.S.A) and the concentration of each RNA sample was measured by Qubit RNA assay. 1 μg of each sample was counted in a Beckman LS 6500 multipurpose scintillation counter to determine new rRNA production based on 3H-uridine incorporation. Three replicates were used to derive the standard deviation from two pairs of mice (WT and Pten-/-) from the same litter. Significance was calculated using an unpaired t test.
Pten+/+ and Pten-/- HSCs (105−106) were plated in 100 μl of methionine/ cysteine -free Dulbecco’s Modified Eagle’s Medium (Sigma) supplemented with 200 μM L-cysteine (Sigma), 50 μM 2-mercaptoethanol (Sigma), 1 mM L-glutamine (Gibco) and 0.1% BSA. The HSCs were pre-cultured with the fresh medium for one hour to deplete endogenous methionine. HSCs were pre-treated with 10 μM MG-132, a proteasome inhibitor, for 1 hour, and then were labeled with 30 μCi of 35S-methionine for 1 hour. Cells were lysed in RIPA buffer and proteins were precipitated by the addition of hot 10% TCA. After centrifugation, the precipitate was washed twice in acetone. The precipitate was dissolved in 100 μL of 1% SDS and heated at 95°C for 10 min. An aliquot of the SDS extract was counted in Esoscint for 35S radioactivity in a liquid scintillation counter to determine the amount of 35S-methionine incorporated into proteins.
To measure growth in the presence of DNA damage, mouse HSCs (3X104) were plated in 100 μl fresh medium containing various DNA damage stresses at the indicated dosages. After culturing for five days, cell viability was assessed based on total cell number and trypan blue staining.
Cancer genome data was obtained with permission from dbGaP. The analysis for rDNA copy number is similar to a previously published method [17]. Briefly, the human consensus 45S rDNA sequences was obtained from NCBI (accession: U13369). Raw fastq whole-genome DNA sequence reads were downloaded from dbGaP. Reads were mapped to the 45S locus, and a set of 16022 pre-selected putative single-copy exons (the largest from each gene) using Bowtie2 v2.1.0 with default parameters. Only concordant read pairs are kept in the down-stream analysis. The rDNA copy number of 18S, 5.8S, and 28S was calculated as the mean coverage in the respective regions. To make samples comparable to each other, the rDNA copy number was further normalized to the background genome coverage, which is calculated as the median coverage of the single copy exons.
For sequence analysis, the predominant sequence of each individual genome was compared to the human reference sequence for 45S. Each nucleotide position was either a match to the consensus 45S sequence, or if not a match, then the position was called a SNV. SNVs were called using samtools mpileup (v1.2) after the duplicated reads were removed from the aligned read files. SNVs were further filtered with the quality score (>20, “PASS”). If the SNVs of the cancer and normal matched genome pair were identical, this was termed a shared SNV. If the SNV was not identical, this position was termed a unique SNV. For the shared SNVs in a pair, each position was evaluated for sequence evidence of 1–4 different nucleotides. The allele numbers for the shared SNVs were used to calculate an average allele number for each genome. The average of the number of alleles at the shared SNVs for normal was subtracted from tumor to generate the allele difference for each matched pair, plotted as an individual point in the box plot. GO terms were calculated using the R package GOstats.
DNAseq libraries were generated from 1 ng of genomic DNA as assessed by the Qubit 2.0 Fluorometer (Life Technologies). Libraries were made according to the manufacturer’s directions for the Nextera XT Library Prep Kit (Illumina, Inc.) and purified using the Agencourt AMPure XP system (Beckman Coulter). Resulting libraries were checked for quality and quantity using the LabChip GX (Perkin Elmer) and Qubit. Equal molar libraries were pooled, re-quantified and sequenced as 125 bp paired reads on the Illumina HiSeq 2500 instrument using HiSeq Control Software 2.2.58. Following sequencing, Illumina Primary Analysis version RTA 1.18.64 and Secondary Analysis version bcl2fastq2 v2.17 were run to demultiplex reads for all libraries and generate FASTQ files.
Human RPE1 cells (hTERT-immortalized retinal pigment epithelial cell line from ATCC), at low passage number and ~70% cell confluence, were transfected by three different PTEN-siRNAs or the control siRNAs (Ambion) using Lipofectamine RNAiMAX Transfection Reagent from Thermo Fisher Scientific company. 24 hours post-transfection, cells were placed in fresh Dulbecco's Modified Eagle Medium plus 10% Fetal Bovine Serum. At 80 hours post-transfection, cells were harvested. Genomic DNA was extracted using the Maxwell® 16 Blood DNA Purification Kit from Promega Corporation. The DNA concentration was measured using the Qubit dsDNA HS assay, and normalized to 1ng/μL.
|
10.1371/journal.pntd.0006735 | The effect of temperature increase on the development of Rhodnius prolixus and the course of Trypanosoma cruzi metacyclogenesis | The increase in the global land temperature, expected under predictions of climate change, can directly affect the transmission of some infectious diseases, including Chagas disease, an anthropozoonosis caused by Trypanosoma cruzi and transmitted by arthropod vectors of the subfamily Triatominae. This work seeks to study the effects of temperature on the development of the life cycle, fertility and fecundity of the insect vector Rhodnius prolixus and on the metacyclogenesis of T. cruzi. All of the variables were subjected to 3 temperatures: 26°C, 28°C and 30°C. Hatching time was evaluated, along with time to fifth instar, time to adult, fecundity studied using the e-value, and egg viability during the first 3 reproductive cycles. In addition, the amounts of metacyclic trypomastigotes of the TcI and TcII DTUs in R. prolixus were evaluated from days 2 to 20 at two-day intervals and from weeks 6 to 8 post-infection. Decreases were observed in time to hatching (15–10 days on average) and in time to fifth instar (70–60 days on average) and transition to adult (100–85 days on average). No significant differences in egg viability were observed in any of the reproductive cycles evaluated, but an increase in fecundity was observed at 30°C during the third reproductive cycle. At 30°C, there was also an increase in the number of infective forms and a decrease in the time at which metacyclic trypomastigotes were detected in the rectal ampulla of the insects for both TcI and TcII. According to these results, the expected temperature increase under climate change would cause an increase in the number of insects and a greater probability of infection of the parasite, which affects the transmission of Chagas disease.
| Chagas disease is an anthropozoonosis caused by the flagellated protozoan Trypanosoma cruzi and mainly transmitted through the infected faeces of insects of the subfamily Triatominae. Because these insects are sensitive to climatic conditions, it is expected that disease transmission may be affected by the increase in global land temperature, predicted under climate change. Therefore, we wanted to evaluate the effect of temperature increase on the development, viability of eggs and fertility of R. prolixus, the most important vector insect in Colombia, and on the development of the parasite within this insect. We observed a decrease in the development time of R. prolixus and an increase in the number of infectious forms of T. cruzi in the insect as the temperature increased. These results suggest that if the temperature increases as expected, there may be an increase in the number of insects that can transmit the disease, as well as an increase in the likelihood of infection due to the increase in the number of infectious forms. Our data contributes to the understanding of the possible effects of the expected temperature increase under climate change on Chagas disease transmission and can be used to make predictive models that can more accurately predict the future of Chagas disease.
| According to climate prediction models, it is expected that by the end of the 21st century, the global terrestrial temperature will increase between 0.3°C and 4.8°C with respect to the average temperature observed between 1986 and 2005 [1]. This increase can affect the transmission of infectious disease agents transmitted by vectors because both insects and vertebrate reservoirs are sensitive to climatic conditions [2]. The alteration in transmission occurs because the temperature and rainfall expected under the effect of climate change can affect the range, proliferation, viability and maturation rates of vectors, pathogens and reservoirs. [3]. Several studies have reported variations in the transmission of diseases such as malaria [4], dengue [5,6] and leishmaniasis [7,8]. In the case of malaria, it has been observed that the time taken by Plasmodium, within the vector mosquito, to pass from gametocytes to infective sporozoites decreases when the temperature of the environment increases [9]. Likewise, a decrease in the development time of the insect vector has been reported, which leads to an increase in the number of insects per season and consequently to a higher transmission rate [10]. In some countries, such as Colombia, Venezuela, Guyana and Peru, there has been a resurgence or intensification of endemic and epidemic malaria that correlates with the phenomenon of the El Niño Southern Oscillation (ENSO) [11, 12]. Likewise, changes in the distribution of leishmaniasis and Dengue virus vectors that may lead to an increase in the risk of transmission of these diseases have been predicted [5, 7].
Another parasitic disease transmitted by vectors that could be affected by climate change is Chagas disease, an anthropozoonosis caused by the flagellated protozoan Trypanosoma cruzi, which is transmitted to humans and other mammals mainly through the infected faeces of insects of the subfamily Triatominae [13]. The disease is a serious public health problem; it is estimated that more than eight million people are infected in Latin America [14]. Despite efforts to control vectors in various regions of Latin America, the World Health Organization estimates that 5,274 new cases occur annually due to vector transmission in Colombia, 933 in Honduras and 873 in Venezuela, countries in which Rhodnius prolixus is the main transmitting vector of the disease [15].
More than 150 species of triatomines have been reported in Latin America, of which 10 are considered primary vectors of the parasite because they colonize houses, while another 20 are considered secondary vectors because they invade human habitations from their peri-domestic or wild habitat. [16] R. prolixus is present in both domestic and wild transmission cycles, which is why it is imperative to design very specific vector control strategies [17]. The life cycle of this vector includes 5 nymphal stages and an adult stage, all obligate haematophagous. None of the nymphs can develop into another stage without having fed at least once. In addition, adults cannot produce eggs without ingesting blood, as it is essential for this process [18]. R. prolixus is considered the most important vector of T. cruzi transmission in Colombia, Venezuela and most Central American countries [13,19]. Also, different outbreaks of oral transmission have been reported in Colombia and Venezuela incriminating R. prolixus [20, 21, 22].
The effect of temperature over R. prolixus life cycle has been barely studied. It is well known that under laboratory conditions, the temperature range for eclosion and molting of R. prolixus was reported to be between 16–34°C [23].No development was observed at 15°C and 35°C [24, 25]. Development is generally studied at constant temperatures between 25 and 28°C and about 70% humidity or unspecified conditions of ambient temperature and humidity. However, temperature and humidity in insect habitats may differ considerably and vary according to circadian and seasonal patterns. R. prolixus is found mainly in Colombia and Venezuela from 0 to 2,600 m above sea-level, in regions with annual median temperatures from 11 to 29°C and 250 to 2,000 mm annual precipitation [26, 27, 28].
The biological cycle of T. cruzi in the vertebrate host begins with contact of the faeces of a triatomine infected with metacyclic trypomastigotes. The parasite penetrates the cells by forming a vacuole bound to lysosomes, then escapes from the vacuole and differentiates into amastigotes, which reproduce in the cytoplasm by binary fission. Subsequently, amastigotes differentiate into metacyclic trypomastigotes that are mobile. These latter forms of the parasite lyse the host cell, disperse and can infect other cells. When insects ingest blood from an infected mammal, trypomastigotes in the blood differentiate into epimastigotes, and some develop into spheromastigotes. The epimastigotes divide in the midgut by binary fission and migrate to rectal ampulla, where they are transformed into metacyclic trypomastigotes (a process called metacyclogenesis), which are eliminated in the faeces of the triatomine [29]. T. cruzi, in turn, has high genetic variability, which has been subdivided by international consensus into six Discrete Typing Units (DTUs), TcI—TcVI, including a DTU associated with anthropogenic bats and called TcBat [30, 31, 32]. In the northern countries of the southern cone, TcI and TcII are the most frequent DTUs in humans, vectors and reservoirs [30]. In Colombia, it is observed that TcI and TcII are the most frequent DTUs in Rhodnius prolixus [17].
Efforts to predict the future of transmission of Chagas disease under the effects of climate change have been scarce and have focused, above all, on trying to establish possible changes in the distribution of vectors through mathematical models [33, 34, 35, 36, 37, 38]. In general, the results obtained have been varied and seem to depend on the vector species, which is why more studies are needed on the physiology of the parasite and the vector to enable more reliable predictions. The objective of this work is to study the effects of temperature, expected under predictions of temperature increase on the development of the life cycle of the insect vector R. prolixus and on the metacyclogenesis of T. cruzi under controlled laboratory conditions.
The T. cruzi scenarios of transmission have been previously studied by our group in Maní, Casanare (Colombia) and is characterized by dense forests of Palms. Infestation rates of 100% in a transect of 120 studied palms were reported [39]. Insect were collected manually in the the axils of A. butyracea palms and transported to our laboratory in the same day of capture.
Specimens of R. prolixus collected in Attalea butyracea palms were kept in incubators that simulated the same average annual humidity and temperature conditions of the axils of the palms where they were collected (26°C, 80% RH and photoperiod 12:12). Temperature was measured using an "EXTECH RHT 20: Humidity / Temperature Datalogger" in different palms. Previous field reports stated that the temperature in palms axils is 26°C-27°C according to Urbano et al. [40]. On the other hand, these devices were also used to measure and control the three temperatures (26, 28, 30°C) in the incubators, during the whole time of the study.
The triatomines maintained in the laboratory were fed chicken (Gallus gallus) blood once every 15 days. Veterinary medical services of Universidad de los Andes–Biological Sciences bioterium, (mice (Mus musculus) blood-chicken (Gallus gallus) blood).
Taking into account the minimum temperature increase expected under climate change and the average temperature of the collection site, three temperatures were chosen for monitoring parasitemia in mice (Mus musculus) and metacyclogenesis in insect vectors: 26°C, 28°C and 30°C. These temperatures were adjusted in closed incubators provided with constant air flow, a constant relative humidity of 80% and a 12:12 light: dark photoperiod.
Taking into account that in Colombia, the most prevalent DTUs of T. cruzi are TcI and, less frequently, TcII, blood trypomastigotes of the strains MHOM/CO/04/MG (TcI) and MHOM/BR/53/Y (TcII) were used to conduct the experiments. These reference strains were previously characterized by 24 microsatellite markers and 10 mitochondrial markers (mMLST) and were maintained in successive passages from mouse to mouse every 15 days, following the recommendations of the Institutional Committee for the Care and Use of Laboratory Animals of the University of the Andes.
For each temperature, 4 groups of 30 newly oviposited eggs were formed. Each group was kept in a plastic container (Diameter 10.5 cm, Height 17 cm) with a filter paper base containing faeces of uninfected insects to ensure the presence of the microbiota in the insects to be studied. For each group, the mortality by stage, the average time to hatching, change to fifth instar and change to adult were recorded.
For each temperature, 20 females and 20 virgin males obtained from the study of the life cycle were taken. The sample size was estimated for each experiment, taking into account the variability and the average of experimental data previously reported [41, 42]. The weights of the females were recorded before and after feeding as reported elsewhere [41], and reproductive pairs were formed, verifying copulation by observing the "spermatophore casing". This procedure was performed during 3 reproductive cycles, taking each cycle as the 21 days after feeding. The eggs oviposited per cycle were transferred to a Petri box with a filter paper base and were kept there to observe hatching. The e-value was calculated, as indicative of the capacity of a female to use the blood ingested towards egg production [43], as was the hatching rate for each pair in each reproductive cycle. The e-value was calculated as follows:
e−value=Totalnumberofeggslaid21dayspostmealInitialfemaleweightxbloodmealweightx1000
(1)
Sixty mice were inoculated intraperitoneally with 0.2 mL of infected blood with each of the selected T. cruzi strains. Fifteen days post-inoculation, the mice were anaesthetized intraperitoneally with pentobarbital and were then exsanguinated by cardiac puncture. This procedure was previously approved by the research ethics committee and CICUAL of the University of the Andes in ruling 318 of 2014. The concentration of parasites in the blood and their viability were verified by light microscopy using a Neubauer chamber. If necessary, the infected blood was diluted in pathogen-free mouse blood to a concentration of 1×10 6 parasites/ml.
For each of the selected strains, 186 fifth-instar nymphs of R. prolixus were fed heparinized mouse blood at a concentration of 1×10 6 parasites/mL using an artificial feeder. After feeding, these nymphs were separated into groups of 62 individuals and were subjected to each of the temperatures. (Fig 1)
To establish the possible differences in the time at which metacyclic trypomastigotes were observed in the rectal ampulla, 2 nymphs were dissected per temperature and strain, from the second post-infection day and at intervals of 2 days until day 20. Similarly, to determine if there were differences between temperatures in the number of infectious forms in the rectal ampulla, 2 nymphs were dissected daily per temperature and strain from week 6 to week 8 post-infection. To perform the dissection, a cut was made in the last abdominal segment of the insect, and the rectal ampulla was carefully separated from the rest of the intestinal tract. The contents of the ampulla were macerated and resuspended in 100 mL of physiological saline (0.9%) at room temperature. Ten microlitres of this solution was used to quantify the number of metacyclic trypomastigotes in the Neubauer chamber.
All procedures with animals were conducted according to the Guide for the care and use of laboratory animals (8 ed)–National Research Council EEUU and the Institutional Animal Care and Use Committee Guidebook of OLAW. The Universidad de los Andes APLAC, in ruling 318 of 2014, approve all animals protocols used in the present work.
All of the statistical analyses were performed in GraphPad Prism 7 software. First, the normality of the data was evaluated using the Shapiro-Wilk test; when the data were not normal, non-parametric tests were used. The Kruskal-Wallis test was used to evaluate the effects of temperature on the life cycle, fecundity and viability of R. prolixus eggs. Dunn’s multiple comparisons post hoc test was used to determine which temperatures were responsible for the significant difference found. In addition, a Chi-square test was perform to establish if there were differences in the mortality of the insects subjected to different temperatures. A Krustal-Wallis test was performed to evaluate if there were statistically significant differences between the temperatures evaluated and the DTUs used. Two tests were necessary, before 20 days post-infection (DPI) and after 35 DPI. Additionally, followed by an analysis of multiple comparisons to determine the groups that showed these differences.
The time to hatching, change to fifth instar nymph and change to adult were significantly different between temperatures (Kruskal-Wallis test, P < 0.0001). In evaluating hatching time and time to adult, it was found that each temperature differed significantly from the other (Dunn's test, P < 0.0001). However, for the time required to change to fifth instar, significant differences were only found between 26°C and other temperatures (Dunn's test, P < 0.0001), but not between 28°C and 30°C. Compared with 26°C, the control temperature, the development time from egg to adult was reduced by 13% when the insects were kept at 30°C and by 9% when the eggs were kept at 28°C (Fig 2).
No significant differences were found in insect mortality between temperatures (Chi-square P = 0.67). In general, mortality was low for all temperatures (6.6% for 30°C, 5.83% for 28°C and 4.16% for 26°C) in comparison with previous studies [44], and was introduced especially during the change from fifth instar to adult, representing 75% of the mortality at 30°C, 83.3% at 28°C and 60% at 26°C.
The e-value, as an indicator of fecundity of the females, was not significantly different between reproductive cycles for the insects subjected to 26°C (Kruskal-Wallis test, P = 0.1080) and 28°C (Kruskal-Wallis test, P = 0.7409). At 30°C, the reproductive cycle did affect the e-value (Kruskal-Wallis test, P = 0.0001), with the first cycle being shorter and significantly different from the second (Dunn's multiple comparisons test, P = 0.0083) and third cycles (Dunn's multiple comparisons test, P = 0.0001). No significant differences were observed in the e-values between temperatures during the first (Kruskal-Wallis test, P = 0.2367) and second reproductive cycles (Kruskal-Wallis test, P = 0.3118), but there were differences during the third (Kruskal-Wallis test, P = 0.0043), with 30°C significantly different from 26°C (Dunn's multiple comparisons test, P = 0.0092) and 28°C (Dunn's multiple comparisons test, P = 0.0092) (Fig 3A).
However, the hatching rate was not affected by either the reproductive cycles (Kruskal-Wallis test, 26° P = 0.0533, 28° P = 0.1687 and 30°C P = 0.2032) or the temperatures to which the eggs were subjected (Kruskal-Wallis test, 26° P = 0.3689, 28° P = 0.2511 and 30°C P = 0.1333) (Fig 3B).
For TcI, the appearance of metacyclic trypomastigotes was observed in the rectal ampulla on the sixth day post-infection at 30°C, while at 26°C and 28°C, it was observed on days 10 and 14, respectively. (Fig 4A). Similarly, for TcII, the appearance of infective forms was observed on day 16 post-infection at 30°C, while at 28°C and at 26°C, metacyclic trypomastigotes were not observed during the first 20 days post-infection. (Fig 4B). All the data evaluated here exhibited a non-parametric distribution. Then, a Krustal-Wallis test was used to evaluate the difference between temperatures and DTUs. A first analysis carried out up to 20 DPI showed that there was no statistically significant difference in the concentration of parasites between the temperatures analyzed for each DTUs, and the same when making a comparison between both DTUs. Otherwise it happened with the analysis carried out after the 35 DPI, where a statistically significant difference of the concentration of parasites was observed between the temperatures analyzed for both DTUs (p = <0.0001). When performing the multiple comparisons test, the temperature of 30° C presented a clear difference with respect to 26° and 28°, given by an increase in the concentration of parasites. These results were maintained for both DTUs. An analysis between DTUs showed a higher concentration of parasites in the insects infected with the TcI DTU with respect to TcII. These results were verified statistically (p = <0.0001), and were maintained throughout the three temperatures.
From weeks 6 to 8 after inoculation, there were significant increases in the numbers of infective forms of TcI and TcII found in the rectal ampulla of insects subjected to 30°C compared with insects subjected to 26°C and 28°C (Fig 4C and 4D). In general, a much greater amount of metacyclic trypomastigotes of the TcI strain compared with the TcII strain was observed. (Fig 4)
In this study, the effects of increasing temperature, as expected under predictions of climate change, on the life cycle, fecundity and viability of R. prolixus eggs and on the development of T. cruzi in R. prolixus were evaluated. Our results indicate that increasing temperature from 26°C to 30°C has effects on the time of development (Fig 2), the fecundity of the insect (Fig 3A) and the development of the parasite (Fig 4). However, no significant effect on egg viability was observed (Fig 3B).
The negative relationship between the temperature and the development time of R. prolixus obtained was in agreement with findings previously reported by Clark [45] and Luz et al [46]. This faster development of insects can be a result of increased metabolic rate caused by the increase in temperature [47], a relationship already demonstrated for R. prolixus [48]. In Fig 2B and 2C, showing the results for changes to fifth instar and to adult, one can see that the times to such changes in some insects are equal to those for other temperatures. These observations could be explained by the frequency of feeding and the amount of blood ingested by each specific individual. Although a food source was offered weekly, it was observed that some insects did not feed or did not ingest enough blood to change, which generates an increase in the life cycle duration [18].
The temperature at which the greatest mortality was observed was 30°C; however, the percentage was similar to that reported by Arevalo et al [41] and was less than observed by Gomes et al [49] for R. prolixus under laboratory conditions. In general, the greatest mortality occurred at the change to fifth instar, which has been reported by other investigators for R. prolixus [41, 49,46] and other triatomine species, such as R. robustus [50], T. infestans [51] and Meccus picturatus [52]. Despite of our results, acclimation capability to temperature must be considered. This is a long-step process that might take several years and in our study we were not able to consider that variable. This is relevant in the light of previous studies that show the three main sensitive parameters for Chagas disease transmission (mortality rate, density of vectors and bite rate) [53]. We explored two of them (mortality and density). However, in future studies the bite rate must be considered because with the increase of temperature, there will occur a higher metabolic rate and in consequence a higher bite rate which could drive the increase in the transmission of T. cruzi. Therefore, in the future bite rate should studied.
Although the presence of Spermatophore casings was checked to verify copulation, the number of copulations per couple was not taken into account. Therefore, it is possible that both the differences observed in the e-value between reproductive cycles for 30°C and the low hatching rate observed during the first reproductive cycle at this temperature can be explained by a difference in the number of copulations. For T. brasiliensis, it is known that females that have multiple copulations produce more eggs with a greater percentage of fertility than females that copulate only once [54]. However, it would be interesting to study fluctuations in fecundity during more reproductive cycles to establish if the differences observed in the third reproductive cycle are maintained or if they are only the result of the number of copulations. If these differences were maintained between reproductive cycles, it would be possible to think that at 30°C, the number of eggs laid per female would increase, and as a consequence, there would be a greater number of insects per reproductive cycle that would be available to transmit T. cruzi. This has been studied by Schilman and Lazzari in 2004 [55], where they found that females oviposit across a range of temperatures from 22 to 33°C with a peak at 25–26°C in accordance with our findings. Nevertheless, they cannot discern whether R. prolixus females actively choose certain oviposition substrates according to temperature or whether they oviposit where they find themselves. These results, both in the life cycle and in the fecundity and fertility of the insects, suggest that the expected temperature increase under climate change could increase the insect density in the palms of A. butyacea. This possibility is not very different from field reports that state that the density of insects in these palms is higher in summer seasons than in rainy seasons, when the temperature decreases [39].
The time at which infectious forms of both TcI and TcII are observed in the rectal ampulla of R. prolixus is shorter at 30°C than at the other temperatures. These results are consistent with previous data reported for Triatoma infestans, where metacyclic trypomastigotes were observed to be faster at 28°C than at 20°C [56]. Likewise, the number of infective forms observed for TcI from week 6 to week 8 was greater at 30°C than at the other temperatures. This relationship between the temperature and reproduction of the parasite has been previously reported under in vitro conditions for the epimastigote stage [43,57]; however, it must be taken into account that under in vitro conditions, the parasite is not subject to the immune factors and the microbiota of the insect [58]; therefore, it is difficult to extrapolate and compare the in vitro results with the results obtained in this study.
The amount of metacyclic trypomastigotes observed in the rectal ampulla of R. prolixus was markedly higher for TcI than for TcII. This difference is likely due to the presence of trypanolytic factors in the haemolymph of R. prolixus that differentially affect TcII but not TcI, which is the main reason why this species of vector is considered to lack the capacity to transmit said DTU [59, 60]. However, it is interesting to note that the temperature may be affecting this interaction, since at 30°C, an increase in the number of infective forms was observed for TcII (Fig 4B–4D). Therefore, in the future, it would be important to evaluate whether this increase in the number of infective forms, due to temperature, could enhance the capacity of R. prolixus to transmit T. cruzi II.
In conclusion, this study showed that as temperature increases (26 to 30°C), there is a more rapid appearance and an increase in the number of infective forms of T. cruzi in R. prolixus, along with a significant decrease in the development time of said vector. These results could suggest that under the effects of climate change, the probability of infection with T. cruzi could increase. However, it is necessary to study the effects of more climatic and ecological factors and the effects of such factors on parasite-vector interactions to predict the future of Chagas disease with better accuracy.
|
10.1371/journal.pntd.0001699 | Rickettsiae Induce Microvascular Hyperpermeability via Phosphorylation of VE-Cadherins: Evidence from Atomic Force Microscopy and Biochemical Studies | The most prominent pathophysiological effect of spotted fever group (SFG) rickettsial infection of microvascular endothelial cells (ECs) is an enhanced vascular permeability, promoting vasogenic cerebral edema and non-cardiogenic pulmonary edema, which are responsible for most of the morbidity and mortality in severe cases. To date, the cellular and molecular mechanisms by which SFG Rickettsia increase EC permeability are largely unknown. In the present study we used atomic force microscopy (AFM) to study the interactive forces between vascular endothelial (VE)-cadherin and human cerebral microvascular EC infected with R. montanensis, which is genetically similar to R. rickettsii and R. conorii, and displays a similar ability to invade cells, but is non-pathogenic and can be experimentally manipulated under Biosafety Level 2 (BSL2) conditions. We found that infected ECs show a significant decrease in VE-cadherin-EC interactions. In addition, we applied immunofluorescent staining, immunoprecipitation phosphorylation assay, and an in vitro endothelial permeability assay to study the biochemical mechanisms that may participate in the enhanced vascular permeability as an underlying pathologic alteration of SFG rickettsial infection. A major finding is that infection of R. montanensis significantly activated tyrosine phosphorylation of VE-cadherin beginning at 48 hr and reaching a peak at 72 hr p.i. In vitro permeability assay showed an enhanced microvascular permeability at 72 hr p.i. On the other hand, AFM experiments showed a dramatic reduction in VE-cadherin-EC interactive forces at 48 hr p.i. We conclude that upon infection by SFG rickettsiae, phosphorylation of VE-cadherin directly attenuates homophilic protein–protein interactions at the endothelial adherens junctions, and may lead to endothelial paracellular barrier dysfunction causing microvascular hyperpermeability. These new approaches should prove useful in characterizing the antigenically related SFG rickettsiae R. conorii and R. rickettsii in a BSL3 environment. Future studies may lead to the development of new therapeutic strategies to inhibit the VE-cadherin-associated microvascular hyperpermeability in SFG rickettsioses.
| Rickettsial diseases are serious human infections. Some spotted fever group (SFG) rickettsial pathogens are bioterror agents. A major clinical hallmark of SFG rickettsial disease is the infection of endothelial cells leading to enhanced vascular permeability. Previous studies show that SFG rickettsiae cause dose-dependent hyperpermeability, which was associated with disruption of intercellular adherens junctions (AJs). The underlying molecular mechanism by which the junctional complexes are disrupted, ultimately causing changes in the endothelial paracellular milieu during rickettsial infection, remains largely unclear. The available evidence suggests that inflammatory stimuli can trigger tyrosine phosphorylation of various components of AJs, mainly the vascular endothelial–cadherin (VE-cadherin). This causes gaps at AJs, partially due to phosphorylation-induced destabilization of VE-cadherins at the plasma membrane and increased endocytosis, greatly increasing paracellular leaks. Here, we hypothesize that infection by SFG rickettsiae induces endothelial cells to develop altered VE-cadherin in association with phosphorylation of tyrosine residues. Utilizing nano-mechanical studies with atomic force microscopy and biochemical analysis of the major AJ protein VE-cadherin, we have implicated that phosphorylation of VE-cadherin directly attenuates homophilic interactions between VE-cadherins. The experimental approach advances a new way of studying rickettsial infection. This strategy should prove useful in uncovering novel therapeutic strategies for virulent arthropod-borne rickettsioses.
| Spotted fever group (SFG) rickettsioses are composed of over 25 species of rickettsiae that are causative agents of a wide spectrum of diseases, ranging from the virulent Rocky Mountain spotted fever (Rickettsia rickettsii) and severe systemic Mediterranean spotted fever (R. conorii) to the recently identified R. parkeri rickettsiosis (R. parkeri) and non-pathogenic R. montanensis [1], [2]. The main target cells of SFG rickettsiae are the endothelial cells that line the entire vasculature [3]–[5]. The most prominent pathophysiological effects of rickettsial infection are increased microvascular permeability, promoting vasogenic cerebral edema and non-cardiogenic pulmonary edema, which are responsible for most of the severity and mortality in Rocky Mountain spotted fever and Mediterranean spotted fever [6]. The cellular and molecular mechanisms by which Rickettsia increase endothelial cell permeability are largely unknown. Previous studies show that R. rickettsii and R. conorii cause dose-dependent hyperpermeability, which was associated with disruption of intercellular adherens junctions (AJs) after infection [5], [7], [8]. The underlying molecular mechanism by which the junctional complexes are disrupted, ultimately causing changes in the endothelial paracellular milieu during rickettsial infection, remains unclear [6], [9].
The available evidence suggests that inflammatory stimuli such as histamine, tumor necrosis factor (TNF), and vascular endothelial growth factor (VEGF) can trigger tyrosine phosphorylation of various components of AJs, mainly the vascular endothelial–cadherin (VE-cadherin), β-catenin, and p120-catenin complex, consequently dissociating catenins from the complex [10]–[12]. This causes gaps at AJs, partially due to phosphorylation-induced destabilization of VE-cadherins at the plasma membrane and increased endocytosis [10], [13], [14], greatly increasing paracellular leaks in cultured endothelial cells [15].
Here, we hypothesize that infection by SFG rickettsiae induces endothelial cells to develop altered junctional protein VE-cadherin in association with phosphorylation of tyrosine residues, so that the Ca2+-dependent, homophilic cis and trans interactions with their extracellular regions [16], [17] are affected or even eliminated, resulting in aberrant properties of junctional complexes. In order to test this hypothesis, detailed information about the biomechanical properties of protein–protein interactions as well as protein–cell interactions at the molecular level is required. Atomic force microscopy (AFM) is ideally suited for these studies because it has a unique capability to measure the interactive forces between receptors and ligands with piconewton resolution [18]–[22]. Within the last decade, this technique has been developed to exert and measure inter- or intra-molecular forces, revealing detailed insights into the functional mechanics of biomolecules [23]–[27]. AFM has been employed to study different cadherin interactions in vitro, in order to mimic different aqueous physiological conditions in vivo [27]–[30].
In the present study, we used single-molecule AFM techniques to study the nanomechanical properties of the interactive forces between VE-cadherin and living human cerebral microvascular endothelial cells upon infection with R. montanensis, which is genetically similar to R. rickettsii and R. conorii and displays a similar ability to invade cells in vitro and can be experimentally manipulated in the Biosafety Level 2 (BSL2) environment [1], [2], [31]. In addition to AFM techniques, we applied routine immunofluorescent (IF) staining, immunoprecipitation (IP) phosphorylation assay, and in vitro endothelial permeability assay to study the biochemical mechanisms that may participate in the enhanced vascular permeability as an underlying pathologic alteration of SFG rickettsial infection. Our experiments help elucidate the molecular mechanism by which SFG rickettsial infection may trigger tyrosine phosphorylation of VE-cadherins, thus destabilizing homophilic molecular interactions at AJs and altering endothelial biophysical features to enhance paracellular leaks.
Recombinant human VE-cadherin Fc chimera was purchased from R&D Systems (Minneapolis, MN). Cell culture medium Prigrow I and fetal bovine serum were obtained from Applied Biological Materials (Richmond, BC, Canada). Unless otherwise indicated, all reagents were purchased from Thermal Fisher Scientific (Waltham, MA).
To allow us to employ BSL2 procedures, we utilized a BSL2 rickettsial species, R. montanensis (strain M/5–6), was used for the present study, obtained from the laboratory of David H. Walker. A 10% yolk sac suspension of R. montanensis from infected eggs diluted in sucrose-phosphate-glutamate (SPG) buffer (0.218 M sucrose, 3.8 mM KH2PO4, 7.2 mM K2HPO4, 4.9 mM monosodium l-glutamic acid, pH 7.0) was propagated through two passages in Vero cells [32], [33]. R. montanensis cells were harvested from 180-cm2 tissue culture flasks containing confluent monolayers of infected Vero cells. The infected Vero cells were harvested from each flask surface with scraper, diluted in 10 ml of supplemented medium, and centrifuged at approximately 13,000×g for 5 min at room temperature. The pellet from each flask was suspended in 15 ml of supplemented media, and was transferred to a precooled 50-ml tube containing 5 g of 3-mm glass beads, and vortexed vigorously for 30 s in order to disrupt the Vero cells. Vortexing was repeated two times with 60-s intervals of incubation on ice between each 30-s vortexing. The lysates were centrifuged at approximately 800× g for 10 min to remove unbroken Vero cells and cellular debris. The supernatant, containing released R. montanensis cells, was transferred to a tube, and the rickettsiae were pelleted by centrifugation at 15,000×g for 25 min at 4°C. Purified rickettsiae were frozen in SPG buffer at −80°C. Rickettial content of the frozen stocks was determined by plaque assay and TCID50 assays on Vero cells, and yielded approximately 1×109 bacterial cells per ml. Uninfected Vero cells were processed by the same procedure as normal control material.
Immortalized human cerebral microvascular endothelial cells (h-CMEC; Applied Biological Materials, Richmond, BC, Canada) were grown in Prigrow I medium supplemented with 10% heat-inactivated fetal bovine serum in 5% CO2 at 37°C. All experiments were performed between passages 15 and 18, and cells were fed with Prigrow I medium with 1% fetal bovine serum.
h-CMEC were cultured on round glass coverslips (12 mm diameter, Ted Pella, Redding, CA) for AFM studies and IF assay until confluent at 90%. The cells were then infected with R. montanensis at a multiplicity of infection (MOI) of 10. After 24, 48, and 72 h, the cells on the coverslips were washed three times in phosphate-buffered saline (PBS) before the downstream studies were performed.
The mechanical properties between VE-cadherin functionalized AFM tips and cell monolayers were studied using AFM that consisted of a detector head (Digital Instruments, Tonawanda, NY) mounted on top of a single axis piezoelectric positioner with a strain gauge sensor (P841.10, Physik Instrumente, Auburn, MA). This system has a z-axis resolution of a few nm and can measure forces in the range of 5–10,000 pN [34]. The monitoring of the force reported by the cantilever and the control of the movement of the piezoelectric positioners are achieved by means of two data acquisition boards (PCI 6052E, PCI 6703, National Instruments) and controlled by custom-written software (Wavemetrics, Portland, OR). In order to measure the interactive forces, we used cantilevers with a 10 µm latex bead glued to the tip (Novascan Technologies, Ames, IA). We incubated the cantilevers with 50 µl of recombinant human VE-cadherin/Fc (R&D Systems, Minneapolis, MN) at 100 µg/mL in 0.1 M NaHCO3 (pH 8.6) overnight at 4°C. Unbound proteins were removed by rinsing with PBS. Bovine serum albumin (BSA, Sigma, St. Louis, MO) at 500 µg/ml in PBS was used to block the exposed surface of the latex bead. The spring constant of each individual cantilever was calculated using the equipartition theorem [21]. The cantilever spring constant varied between 20–50 pN/nm. Interactive forces were measured by pressing the cantilever onto the cell monolayer for ∼500 ms and then stretching for several hundred nm. We used a serum-free Hank's Balanced Salt Solution (HBSS) supplemented with 10 mM HEPES, 2 mM CaCl2 and 1 mM glucose. In the experiments using blocking antibodies, cells were pretreated with different antibodies (25 µg/ml) for 15 min before the AFM measurements. Unless noted, the pulling speed of the different force-extension curves was about 1.0 µm/s.
The permeability of h-CMECs upon infection with R. montanensis at a MOI of 10 was determined using an in vitro vascular permeability assay (Millipore, Billerica, MA) as previous described [12], [14], [35], [36]. Briefly, h-CMECs were seeded onto type I rat-tail collagen-coated polycarbonate Transwell filters (6.5-mm diameter and 3-µm pore size; Millipore, Billerica, MA) and confluent monolayers were inoculated with R. montanensis or mock-infected control material cells. At different time points post-infection (p.i.), hCMEC permeability was assessed by adding 0.5 mg/ml of fluorescein isothiocyanate (FITC)-dextran (40 kDa; Sigma, St. Louis, MO) to the top chamber above the filter. After 3 hours, FITC-dextran present in the bottom compartment was assayed by using a BioTek Synergy 2 multi-mode microplate reader (485 nm excitation, 530 nm emission). The fold-change in fluorescence intensity over the basal permeability of monolayers was used as an indicator of paracellular permeability of assessed monolayers. Experiments were performed in sets of four.
Cells were fixed with cold methanol at 24, 48, or 72 h after infection. Each experiment was repeated three times. The primary antibodies, a mouse monoclonal IgG against VE-cadherin (1/500) (Clone TEA1/31, Meridian Life Science, Saco, ME) and a rabbit polyclonal IgG antibody against SFG rickettsiae (1∶5000), were added and incubated for 2 h. VE-cadherins and rickettsiae were detected with secondary goat anti-mouse Alexa 488 and goat anti-rabbit Alexa 594 conjugated antibodies (Invitrogen, Carlsbad, CA), respectively. IF images were taken and analyzed with an Olympus BX51 imaging system.
In experiments following IF studies, in vitro cellular expression of VE-cadherin was analyzed by Western immunoblotting according to established methods [37], [38]. After infection with R. montanensis for 24, 48, or 72 hr in T75 flasks, whole-cell extracts of infected and mock control cells were prepared by lysis in RIPA buffer (Santa Cruz Biotechnology, Santa Cruz, CA) containing 1× PBS, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate (SDS), aprotinin, and phenylmethylsulfonyl fluoride (PMSF). The concentration of total protein was determined before equal amounts of soluble protein (50 ug/lane) were subjected to SDS–polyacrylamide gel electrophoresis (SDS-PAGE) (10% acrylamide) (Invitrogen, Carlsbad, CA). Proteins were transferred onto a polyvinylidene difluoride (PVDF) membrane and then incubated with mouse monoclonal anti-VE-cadherin antibody (dilution 1∶1000; Meridian Life Science, Saco, ME), followed by incubation with a secondary antibody at 1∶2000 for 30 min. Blots were visualized by using a chemiluminescence kit (Pierce, Rockford, IL). Data were analyzed densitometrically using 1D scan EX software (BD Biosciences, Rockville, MD). A Western blot for α-tubulin served as loading control to verify equal loading and transfer.
To study VE-cadherin phosphorylation, cell lysates were prepared with an ice-cold RIPA lysis buffer (the same as for Western immunoblot assays). After centrifugation at 12,000×g for 20 min, the protein supernatant was collected. Equal amounts of protein with optimal Dynabead Protein G (Invitrogen, Carlsbad, CA) conjugated with anti-VE-cadherin antibody were incubated for 2 h at room temperature. The Dynabead-antibody-antigen complex pellets were precipitated and separated using DynaMag-2 (Invitrogen, Carlsbad, CA). The pellet was washed three times with PBS, and resuspended in 20 µL of SDS sample buffer (Invitrogen, Carlsbad, CA) and heated for 10 min at 70°C. Samples were then separated by gel electrophoresis followed by immunoblotting. A mouse monoclonal anti-phosphotyrosine antibody, 4G10 (Millipore, Billerica, MA) was used at dilution of 1∶500 for detection of proteins containing phosphotyrosine. All experiments were performed in sets of three.
Values are reported as mean ± SD. The data were analyzed using Student's paired t-test (Sigmaplot, Sigma Stat, Jandel Scientific Software, San Rafael, CA). Statistical significance was determined at P<0.05.
The abnormal VE-cadherin expression induced upon infection of rickettisae was visualized under fluorescence microscopy. At earlier time points, no detectable difference was noted in IF studies of VE-cadherin compared to normal controls, although rickettsiae were detected at 24 hr and 48 hr post-infection (Figure 1). As the infection progressed, VE-cadherin's distribution appeared disrupted after 72 hr at endothelial cell contacts in confluent cell layers when compared to controls (Figure 1).
To determine if disorganized or reduced VE-cadherin at endothelial AJs are relevant to endothelial paracellular barrier dysfunction, we assessed cell permeability by an in vitro vascular endothelial cell permeability assay. As seen in Figure 2, in h-CMEC monolayers, infection of R. montanensis induced a 1.58-fold increase in para-endothelial cell permeability at 72 hr post-infection compared to control. There was no significant change in hCMEC monolayers at 24 hr and 48 hr post-infection compared to normal controls.
To determine the possible biochemical basis of increased endothelial permeability, we examined VE-cadherin using Western immunoblotting. Initial studies failed to reveal any alterations in expression of general VE-cadherin between control and infected experimental monolayers at 24, 48, and 72 h post-infection (Figure 3A). Therefore, we used an IP-phosphorylation assay to focus on tyrosine phosphorylation of VE-cadherin because phosphorylation is thought to be an important event leading to destabilization of the AJ complex [10], [11], [14]. Infection with R. montanensis at a MOI of 10 stimulated increased tyrosine phosphorylation of VE-cadherin at 48 hr, with even greater phospholyration at 72 hr post-infection (Figure 3B and 3C). This time corresponds to the increased endothelial permeability, suggesting that modulating VE-cadherin activity through phophorylation is one of the mechanisms regulating VE-cadherin-related endothelial monolayer paracellular permeability.
For AFM experiments, human VE-cadherins were immobilized on beads attached to cantilever tips. Subsequently, this functionalized cantilever was gently brought into contact with the surface of a confluent monolayer of human cerebral microvascular ECs. The maximum compression force was set to approximately 200 pN. The contact time was critical in this study, and was constantly kept about 0.5 s before the cantilever was retracted at a constant pulling speed of 0.5 um/s. AFM mimicked the binding of intramembrane VE-cadherins between the ECs under physiological conditions. By monitoring the cantilever deflection and retraction cycle, the binding, stretching, and rupture of VE-cadherin-VE-cadherin complexes can be monitored in terms of forces and distance as a function of time.
Typical force-extension patterns of the interactions between VE-cadherin and ECs are shown in Figure 4A. The force curves represent the force changes on the cantilever as a function of its travel distance. By integrating the areas underneath the force curve and above the baseline (dashed lines to present zero force, Figure 4A), we calculated the work that is required to break any interactive bonds between the cantilever and the ECs. Normal ECs always show a strong binding to the VE-cadherin functionalized cantilever. To our surprise, rickettsial-infected ECs showed a dramatic decrease in binding affinity to VE-cadherin as early as 48 hrs post-infection. The average work from infected cells decreased to ∼20% of that of uninfected cells (Figure 4B). The level of work in infected cells remained low after 72 hrs.
To confirm the specificity of this VE-cadherin-EC interaction, several control experiments were performed. For example, an excess of antibodies to VE-cadherin was added, or BSA was added to block the interactions between the VE-cadherin and ECs. As shown in Figure 4, the presence of a blocking monoclonal antibody against VE-cadherin (25 µg/ml) resulted in a significant decrease in the adhesion interaction between normal cells and the probing cantilever. The detected force was small, only about 30 pN, which is close to the detection limit of our instrument (∼20 pN). Another control experiment was to probe the normal cells with a cantilever covered by BSA at 100 µg/ml in PBS. No force was detected in this experiment since the exposed surface of the cantilever was completely blocked by BSA (data not shown).
SFG rickettsial diseases are serious human infections. Some SFG rickettsial pathogens are bioterror agents [40]. A major clinical hallmark of SFG rickettsial disease is the infection of EC leading to enhanced vascular permeability [6]. The cellular and molecular mechanisms by which SFG rickettsiae increase endothelial permeability are largely unknown [4]. The endothelial cells that line all blood vessels function to regulate the influx and efflux of solutes and fluids between the vessel lumen and the surrounding interstitium. The movement of vessel contents is mediated by two broad mechanisms, the paracellular and transcellular routes. Relatively little is known about the role of the second route in microvascular hyperpermeability during inflammation [14]. The paracellular pathway, which is generally accepted to be dominant in inflammatory pathological states, is controlled by the dynamic opening and closing of endothelial junctions, mainly mediated by transmembrane proteins VE-cadherin at AJs and claudin at tight junctions (TJs) [10], [41], [42]. VE-cadherin initiates cell-cell adhesion and promotes its maintenance through its transmembrane domains [17]. VE-cadherin may also form a signaling complex through its cytoplasmic tail, interacting with β-catenin and p120-catenin [43]. However, it is hard to clearly separate these two aspects. VE-cadherins are linked to a large variety of intracellular partners that mediate intracellular signaling and modulate the organization of the actin cytoskeleton to provide the dynamic forces necessary for appropriate tissue morphogenesis [10]. VE-cadherin-deficient mice die at mid-gestation due to defective vascular remodeling [44]. The primitive vascular plexus initially forms, but beyond embryonic day 9 these vessels regress and disintegrate. VE-cadherin-blocking antibodies disrupt cell-cell adhesion, increase permeability, and enhance transmigration of leukocytes [45], [46]. However, VE-cadherin's role in the mechanism responsible for enhanced microvascular permeability during SFG rickettsioses needs to be elucidated.
In an earlier study, a remarkable observation was made regarding discontinuities in the endothelial localization of AJ proteins after a prolonged period of R. conorii infection [9]. Similar findings were made by our group using IF studies in mouse models of intravenous infection by R. conorii. Endothelial cells lining cerebral and pulmonary microcirculation display significantly diminished AJ and TJ proteins at day 5 after infection with a lethal dose of rickettsiae (unpublished observations). Furthermore, in an in vitro functional study, enhanced microvascular endothelial permeability has been described, which is correlated with dissociation of AJs (β-catenin and p120) during 24, 48, and 72 hr post-infection by R. rickettsii [7]. In the present study using a human cerebral mirovascular endothelial model, we observed aberrant structures of inter-ECs VE-cadeherin at 72 hr post-infection by R. montanensis, in which enhanced microvascular permeability was documented using an in vitro endothelial cell permeability assay.
VE-cadherins engage in Ca2+-dependent homophilic interactions in which a VE-cadherin molecule on one cell binds to an identical VE-cadherin molecule on an adjacent cell [43]. After binding, cadherins aggregate laterally in trans and cis at cell–cell junctions and form a zipper-like structure along the cell border that promotes tight adhesion between endothelial cells [16], [17]. In the present study, we used AFM to directly examine alterations in protein-protein adhesion forces that underlie this paracellular dysfunction following SFG rickettsial infection. AFM experiements revealed a dramatic reduction in the interactive forces between VE-cadherin and EC after 48 hr of infection. This decreased protein-EC interaction took place prior to the enhanced microvascular permeability detected by in vitro endothelial permeability assay at 72 hr p.i. This fact indirectly supports the idea that Ca2+-dependent homophilic interactions between VE-cadherin molecules on adjacent cells are the target during SFG rickettsial infection-induced endothelial hyperpermeability.
There are many mechanisms that regulate VE-cadherin, including modulating VE-cadherin activity through phosphorylation and controlling VE-cadherin availability at the endothelial surface [10], [47]. Stimuli such as histamine, thrombin, tumor necrosis factor (TNF), and vascular growth factor (VEGF) induce tyrosine phosphorylation of VE-cadherin, in which Src and Rac play a role as key pathway mediators to promote kinase-regulated phosphorylation of VE-cadherin on different residues attenuating stability at endothelial AJs [10], [48]. Evidence has been established that mediators of inflammation signal through Src and Rac to trigger the tyrosine phosphorylation of VE-cadherin, leading to the endocytosis of VE-cadherin in a β-arrestin-dependent fashion [49]. Thus, kinase-mediated phosphorylation coordinates with the destablizing barrier function of VE-cadherin at endothelial AJs. By competing with phosphorylate kinase, binding of p120-catenin may prevent VE-cadherin endocytosis from the plasma membrane, stabilizing it at the endothelial AJ [50]. Previous studies have demonstrated that the catenin class of proteins, β-catenin and p120-catenin, dissociate from the interendothelial cell junctions in response to SFG rickettsial infections [7], [9]. Furthermore, a study using in vitro human endothelial-targeted R. rickettsii and human cerebral microvascular endothelial cells showed that the addition of pro-inflammatory stimuli essential to rickettsial immunity enhances rickettsia-induced microvascular permeability in a dose-dependent manner [7]. Taken together, this evidence suggests that SFG rickettsial infection may cause endothelial paracellular barrier dysfunction in association with phosphorylation of VE-cadherin, thus destabilizing endothelial AJs. A major finding of the present study is that upon infection by SFG rickettsiae, tyrosine phosphorylation of VE-cadherin was activated in human cerebral microvascular endothelial cells, which started at 48 hr and increased at 72 hr post-infection, although no difference was detected for general VE-cadherin expression at the same time. Given that the in vitro endothelial permeability assay showed enhanced microvascular permeability at 72 hr post-infection and the AFM studies showed a dramatic reduction in the adhesive force between VE-cadherin and endothelial cells at 48 hr, we suggest that upon infection by SFG rickettsiae, phosphorylation of VE-cadherin directly attenuates homophilic protein-protein interactions at the endothelial AJs, leading to endothelial paracellular barrier dysfunction and microvascular hyperpermeability.
In the present study, we present data to support the association between phosphorylation of endothelial AJ proteins and enhanced microvascular permeability during SFG rickettsial infection. However, it is not established whether activated phosphorylation is a direct consequence of rickettsial infection of the endothelial microvasculature, or whether it is a consequence of less specific physiological responses such as inflammation. We will utilize atomic force microscopy in future studies involving pathogenic SFG R. conorii and R. rickettsii to help to address the potential role of rickettsiae as a trigger mechanism to alter major AJ components that affect vascular permeability.
In summary, our results indicate that phosphorylation of VE-cadherin directly attenuates homophilic interactions between VE-cadherins. Our nano-mechanical and biochemical studies of the major endothelial AJ protein VE-cadherin have implicated attenuated VE-cadherin-endothelial cell interaction as an underlying cause of enhanced microvascular permeability that occurs at one prolonged stage upon infection by R. montanensis. Our experimental approach advances a new way of studying rickettsial infection and will allow similar studies of the closely related SFG rickettsiae R. conorii and R. rickettsii. This strategy should prove useful in uncovering novel therapeutic strategies for virulent arthropod-borne rickettsioses.
|
10.1371/journal.pntd.0006054 | Rapid clearance of Schistosoma mansoni circulating cathodic antigen after treatment shown by urine strip tests in a Ugandan fishing community – Relevance for monitoring treatment efficacy and re-infection | Schistosomiasis control and elimination has priority in public health agendas in several sub-Saharan countries. However, achieving these goals remains a substantial challenge. In order to assess progress of interventions and treatment efficacy it is pertinent to have accurate, feasible and affordable diagnostic tools. Detection of Schistosoma mansoni infection by circulating cathodic antigen (CCA) in urine is an attractive option as this measure describes live worm infection noninvasively. In order to interpret treatment efficacy and re-infection levels, knowledge about clearance of this antigen is necessary. The current study aims to investigate, whether antigen clearance as a proxy for decreasing worm numbers is reflected in decreasing CCA levels in urine shortly after praziquantel treatment. Here CCA levels are measured 24 hours post treatment in response to both a single and two treatments. The study was designed as a series of cross-sectional urine and stool sample collections from 446 individuals nested in a two-arm randomised single blinded longitudinal clinical trial cohort matched by gender and age (ClinicalTrials.gov Identifier: NCT00215267) receiving one or two praziquantel treatments. CCA levels in urine were determined by carbon-conjugated monoclonal antibody lateral flow strip assay and eggs per gram faeces for S. mansoni and soil-transmitted helminths by Kato-Katz. Significant correlations between CCA levels and S. mansoni egg count at every measured time point were found and confirmed the added beneficial effect of a second treatment at two weeks after baseline. Furthermore, presence of hookworm was found not to be a confounder for CCA test specificity. Twenty-four hours post treatment measures of mean CCA scores showed significant reductions. In conclusion, removal of CCA in response to treatment is detectable as a decline in CCA in urine already after 24 hours. This has relevance for use and interpretation of laboratory based and point-of-care CCA tests in terms of treatment efficacy and re-infection proportions as this measure provides information on the presence of all actively feeding stages of S. mansoni, which conventional faecal microscopy methods do not accurately reflect.
ClinicalTrials.gov NCT00215267
| Large scale efforts to control schistosomiasis in several sub-Saharan countries are in progress. In order to accurately monitor the effect of interventions, we need diagnostic tools which are highly specific, sensitive, affordable and easy to use and implement. For Schistosoma mansoni detection the circulating cathodic antigen (CCA) is an attractive option as this antigen is a measure of actively feeding worms and can be measured in urine samples. However, knowledge about how fast this antigen is cleared in response to treatment with praziquantel is necessary for interpretation of consecutive measures based on CCA in order to use this tool optimally. Here we show that CCA is already significantly reduced 24 hours after treatment both in single and double treatment regimens in a community sample from Musoli Village, Uganda. Furthermore, the data supports interpretation of trace measures of CCA as positive since the majority of individuals with trace measures respond to treatment. These observations provide a basis for extended use of CCA-based tools in monitoring treatment efficacy and possibilities for logistically advantageous prevalence screening strategies.
| The ambitious London declaration prompted by the WHO 2012 roadmap to combat neglected tropical diseases commits to global schistosomiasis control by 2020 [1, 2]. Mass drug administration with praziquantel is the most widely implemented intervention for control. However, the necessity of an inter-disciplinary approach incorporating other strategies, such as community health education, improved safe water supply, sanitation and control of intermediate hosts, has become evident for targeting elimination of this poverty associated disease [3]. For elimination strategies, where low infection intensities and focal epidemiology must be addressed in the later stages of a control programme, a test-then-treat approach based on point-of-care tests, preferably incorporated in integrated disease control programmes, becomes relevant in contrast to continuous reliance on mass drug administration only protocols.
In order to accurately monitor the progress of interventions aiming at reducing, controlling or even regionally eliminating transmission, we need highly specific, sensitive, feasible and affordable diagnostic tools, which optimally live up to all the ASSURED criteria (Affordable, Sensitive, Specific, User-friendly, Rapid and robust, Equipment-free and Deliverable to end-users) [4, 5]. Schistosoma spp. worm antigens, which are released by living worms, are attractive targets compared to specific antibodies as they enable identification of currently infected individuals compared to individuals with previous infection. This is very relevant for assessing treatment efficacy as praziquantel kills adult schistosomes and mature eggs, but does not affect the immature worm and egg stages [6–8].
Circulating cathodic antigen (CCA), which is a mucin-type glycoprotein antigen [9], regurgitated by Schistosoma spp. [10–14], has been shown to be a good measure for active Schistosoma mansoni infection [15]. CCA is excreted into the bloodstream as soon as schistosomules start actively feeding and at least in high dose mouse models the antigen level is positively correlated to worm burden [16]. In humans the earliest verifiable detection point so far demonstrated is four weeks post infection [17]. In addition, CCA can be detected in sera and breast milk from infected individuals [18]. CCA detectable in human urine was first described by Carlier et al. (1975) and Deelder et al. (1976) [11, 19]. However, the potential of this antigen was not recognised until recently [20].
Two robust immunochromatographic methods for direct CCA detection in human urine are now published; a laboratory based strip test (using colloidal carbon labelling) and a commercially available point-of-care test (POC-CCA, Rapid Medical Diagnostics). Rapid antigen clearance after praziquantel treatment is essential for CCA-based diagnostic tools to be used for interpreting treatment efficacy, re-infection dynamics and potential development of drug resistance. Furthermore, knowledge of antigen clearance is very relevant for identifying target groups harbouring persistent active infection such as egg negative individuals. In this study the laboratory based lateral flow urine strip test was used to investigate whether changes in CCA score were detectable as early as 24 hours after treatment in urine samples from a community living in a S. mansoni high endemic area by Lake Victoria, Uganda. Furthermore, the relationship between S. mansoni egg output and CCA score in relation to age groups and responses to a single versus double treatment schedules were assessed.
The current study is based on a series of cross-sectional sampling points nested in a two-arm cohort randomised single blinded longitudinal clinical trial on “The effect of praziquantel treatment on Schistosoma mansoni morbidity and re-infection along Lake Victoria, Uganda” (ClinicalTrials.gov Identifier: NCT00215267). A questionnaire based census formed the basis for random recruitment of participants above seven years of age stratified by gender and age [21, 22]. Sample size for the clinical trial set-up was estimated to 552 individuals using cure rate data from Lake Albert, Uganda [23] with a power of 90%, significance level of α = 0,05 and an anticipated drop-out from baseline (2005) to finalisation (2007) of 40%. Randomisation to either the arm receiving one dose of praziquantel treatment (1 Tx) or the arm receiving a second dose two weeks (2 Tx) after baseline was done by a scientist not involved in treatment and laboratory work using computer generated random numbers [24]. The study was blinded for examiners, but participants were fully informed of treatment regimen. Urine samples were investigated at five time points and stool samples at two time points after baseline (Fig 1). Sample sizes for each analysis are stated; no-show or difficulty delivering stool/urine for individuals at some time points account for missing data (S1 Dataset).
Study participants were recruited from Musoli Village, Mayuge District in south-east Uganda, which is located along Lake Victoria. Mayuge District is located 1161m above sea level and receives annual precipitation ranging from 600-1100mm with average temperatures from 19–27°C [25]. This area of Lake Victoria displays high perennial S. mansoni transmission and no S. haematobium transmission (personal communication N. Kabatereine). The only water source available to Musoli Village for both domestic and recreational use was Lake Victoria, hence exposing the population to schistosomiasis. Albeit, that Mayuge District authority had implemented praziquantel mass drug administration for control purposes, Musoli Village had not yet received treatment at the time of the study. Besides fishing, the local economy relies on subsistence farming.
Ethical approval was obtained from the Higher Degrees Research and Ethics Committee of the School of Public Health, Makerere University. Ethical clearance was granted by the Uganda National Council of Science and Technology (Reference number: UNCST:HS59).
Informed consent written in Lusoga, which is the most commonly used local language, was obtained from each adult participant. Assent was obtained from participants under 15 years of age, which also were required to present a signature from a parent or guardian. Participants were free to withdraw from the study at any point in time without facing any form of repercussion.
All participants were treated with standard 40mg/kg single dose praziquantel (Distocide 600mg, Shin Poong Pharmaceuticals) at baseline irrespective of infection status. Furthermore, 400mg of albendazole (Alzental 400mg, Shin Poong Pharmaceuticals) was given to each participant for treatment of soil-transmitted helminths. One participant group received a second standard single dose of praziquantel (40mg/kg) two weeks after baseline. Before treatments all participants were offered a snack and a drink in order to minimise adverse events. After treatment participants were observed for two hours for management of possible adverse events. An experienced nurse directly observed all treatments [24]. After finalisation of the clinical trial in 2007, the whole community including the study participants were treated with a single standard dose of both praziquantel and albendazole following national guidelines.
All data was double-entered in Microsoft Excel 2003 spreadsheet software, where after the data file was exported to SPSS (IBM) for further analyses. Graphical illustrations were made using GraphPad Prism 6 software. For all statistical analyses p values of <0.05 were considered significant and two tailed tests applied. Hookworm species, A. lumbricoides and T. trichiura results were only recorded as positive/negative during data collection. Spearman’s correlation co-efficient (rho) was used to describe relationships between CCA score and KK egg intensity. Kruskall-Wallis/Mann-Whitney was used to compare mean EPG for 6KK/individual slide EPG and CCA score stratified by treatment regimen and time points. Mann-Whitney was also used to investigate associations between CCA and hookworm status. Paired t-tests were applied for comparing the means of CCA scores with 24 hour intervals. Binary logistic regression models were used to assess the odds for being CCA positive when positive for S. mansoni (controlled for hookworm, gender and child/adult) or hookworm eggs.
The total number of recruited participants with CCA and S. mansoni egg count data was 446 individuals, of which 228 (51,1%) were male and 218 (48,9%) female, originating from 243 different households. The median age was 23 years (range 7–76 years). Participants were divided in two treatment arms stratified by gender and age; 240 individuals received praziquantel only at baseline while the other arm of 206 individuals received a second dose 2 weeks after baseline. Demographic, occupational, S. mansoni morbidity and infection intensity (KK) data including dose related treatment effect were reported in Tukahebwa et al. 2013 and Koukounari et al. 2013 [21, 22]. Prevalence of S. mansoni infection for each sampling time point based on detection of CCA in urine (trace -/+) or eggs in faeces is shown in Table 1.
In the total sample, the cured proportions determined based on CCA in urine at baseline (trace considered positive) are 32,3% (104/322; trace = positive) and 41,0% (132/322; trace = negative) at baseline +24 hours and 36,3% (121/333; trace = positive) and 50,5% (168/333; trace = negative) at two weeks. For two weeks +24 hours the cured proportions are 50,0% (67/134; trace = positive) and 66,4% (89/134; trace = negative) in the two treatments arm compared to baseline (trace considered positive). The one treatment group at the two week time point show cured proportions in response to a single treatment at baseline of 36,4% (43/118; trace = positive) and 43,2% (51/118; trace = negative). Correspondingly, when considering trace as negative at baseline, the following cured proportions were observed; baseline +24 hours 30,0% (92/307; trace = positive) and 38,4% (118/307; trace = negative), two weeks 34,9% (111/318; trace = positive) and 48,7% (155/318; trace = negative), and for the two treatments arm at two weeks +24 hours 47,2% (60/127; trace = positive) and 64,6% (82/127; trace = negative). For the one treatment arm the cured proportions are 34,2% (39/114; trace = positive) and 41,2% (47/114; trace = negative) when assuming trace is negative for baseline measures.
CCA score (0, ½, 1, 2, 3) and S. mansoni eggs (EPG) were both obtained at baseline, nine weeks and two years. There was no significant differences between treatment arms for CCA scores or egg intensities at baseline (CCA p = 0,792, n = 444; eggs p = 0,553, n = 445) or two years after treatment (CCA p = 0,951, n = 326; eggs p = 0,716, n = 346). However, significantly lower CCA scores (p = 0,001, n = 399) and fewer eggs (p<0,001, n = 428) were observed at nine weeks in the two treatments arm compared to the one treatment arm. For S. mansoni egg positive participants the geometric mean egg count was 253 EPG; 95% CI[212–301] at baseline (n = 395); 15 EPG, 95% CI[12–19] at nine weeks (n = 170) and 41 EPG; 95% CI[33–52] at two years (n = 232). S. mansoni prevalence and geometric mean egg counts for each KK slide and 6KK at baseline, 9 weeks and 2 years stratified by treatment regimen can be found in Table A in S1 Table. The arithmetic mean CCA score for positive participants were 2,2; 95% CI[2,1–2,3] at baseline (n = 365), 1,3; 95% CI[1,2–1,4] at nine weeks (n = 210) and 1,7; 95% CI[1,6–1,8] at two years (n = 246) (trace considered positive). CCA data is restricted to a narrow range of defined assay band intensity related observations (0–3), which limits demonstration of potential variation and average measures might not be truly representative, but they are used here in lack of a better method of illustration. Arithmetic mean CCA score and geometric mean egg counts for the three time points stratified by age groupings and treatment regimens are shown in Fig 2.
Positive correlations between CCA score and S. mansoni egg intensity were found at all measured time points (see Table B in S1 Table for CCA score/egg intensity category distribution data). The correlation was stronger at baseline (p<0,001, n = 443, rho = 0,717) and after two years (p<0,001, n = 320, rho = 0,691) than at nine weeks (p<0,001, n = 395, rho = 0,525). Similar findings are obtained when stratifying according to treatment regimen with better correlation at two years post treatment for both the two treatments arm (p<0,001, n = 153, rho = 0,762) and one treatment arm (p<0,001, n = 167, rho = 0,611) than at nine weeks for two treatments (p<0,001, n = 185, rho = 0,430) and one treatment (p<0,001, n = 210, rho = 0,554).
The prevalence of A. lumbricoides was very low. Only one individual was found infected at baseline (n = 430), none at 9 weeks (n = 426) and at two years a different single egg positive individual was found (n = 341). No individuals were at any time point found to be T. trichiura egg positive. The hookworm species prevalence was 44,3% (193 of 436) at baseline, 5,6% (24 of 425) at nine weeks and 28,0% (97 of 346) at two years. Of the individuals with hookworm data both at baseline and nine weeks, 180 were positive at baseline of which 20 remained positive at nine weeks. Of these thirteen individuals remained hookworm infected at two years post treatment (76,5%, n = 17).
Among S. mansoni egg negative individuals there was no association between hookworm and being CCA positive at baseline (p = 0,569, n = 48), nine weeks (p = 0,990, n = 235) or two years (p = 0,348, n = 104) post treatment. In binary univariate logistic regression models the following odds ratios were obtained for being CCA positive when hookworm egg positive at baseline (n = 48, OR = 1,5; 95% CI[0,4–5,5] (trace = positive)/OR = 1,9; 95% CI[0,4–8,7] (trace = negative)), nine weeks(n = 235, OR = 1,1; 95% CI[0,3–4,6] (trace = positive)/OR = 0,5; 95% CI[0,06–3,8] (trace = negative)), and two years (n = 104, OR = 0,7; 95% CI[0,3–1,5] (trace = positive)/OR = 0,8; 95% CI[0,3–1,8] (trace = negative)) for the S. mansoni egg negative (6KK) subgroups.
Twenty four hours after baseline 74,9% of CCA positives (n = 339) have a decline of ≥1 score unit after treatment (Table 2, Table C in S1 Table), whereas only 3,5% have an increase of ≥1 score unit. 12,4% had unchanged positive CCA scores. With respect to trace scores 5,3% and 3,2% showed a decline or increase of ≥1 score unit, respectively, whereas only 0,6% remained unchanged as trace positives. Fifteen individuals had trace positive score at baseline whereof 80% (12) were negative 24 hours later, two remained as trace positive and only one observation showed an increase in score to “1” (Table C in S1 Table). The mean CCA scores at baseline and 24 hours after treatment were significantly different (p<0.001). Mean CCA score [±95%CI] stratified by age groups for baseline and plus 24 hours is shown in Fig 3A.
The extend of daily fluctuations in measured CCA levels can be observed at two weeks where 18,8% (n = 101) of CCA positives only treated at baseline (1 Tx) had a decline in CCA score of ≥1 at plus 24 hours and 26,7% had an increase of ≥1 (Table 2, Table C in S1 Table); 35,6% had an unchanged positive score in this group. Comparably, 34,3% had a decline and only 7% had an increase in CCA score of ≥1 at 24 hours post treatment in the arm being treated at two weeks (2 Tx). In the two treatments arm 28,3% had an unchanged positive score. However, the arm only having received treatment at baseline (n = 147) show no significant difference in means at two weeks and two weeks plus 24 hours (p = 0,568). In contrast, the arm receiving a second treatment at two weeks (n = 157) has a significantly lower mean CCA score 24 hours post-treatment (p<0,001). At the two week time point 25 individuals scored trace positive in the two treatments arm and 13 responded to treatment and scored negative 24 hours later (Table C in S1 Table). Eight remained trace positive and only four showed an increase in score. In comparison seven individuals from the one treatment arm showed trace positive scores at two weeks and a negative score at two weeks +24 hours. Only one individual remained trace positive and 5 had an increased score at two weeks + 24hours. Mean CCA score [±95%CI] stratified by age groups for two weeks and two weeks plus 24 hours is shown in Fig 3B.
CCA scores at the baseline +24 hours and two week time points both reflect measurements of antigen in urine after a single treatment at baseline only. Table 3 shows the number of individuals (n) with a given score at baseline+24hrs after treatment and the corresponding score at two weeks post-treatment. Forty-three percent (n = 359) of individuals had an unchanged score from baseline +24hours to the two week time point. When considering change in score at sample level, only 36 discrepancies are observed (Table 3). Out of these score differences 19 are positive at both time points with a trend of lower score at two weeks, leaving only 17 (5%) observations positive at baseline +24hrs after treatment, which are negative at two weeks, giving an overestimation of prevalence compared to two weeks post-treatment (60% vs 55%, trace considered positive).
In order to design the best possible interventions aiming at breaking S. mansoni transmission, availability of accurate diagnostic tools detecting live worms and thus active infection is crucial. Such tools can improve the understanding of the actual efficacy of treatment and proportion of re-infection both of which are currently based on faecal egg counts (KK). When evaluation of treatment efficacy is based on classical faecal egg counts alone, a portrayal of poor treatment efficacy may be the result as eggs produced by newly matured worms and eggs lodged in tissue and being released over a period of time, despite clearance of worms, will present as treatment failure. For antigen-based diagnostic tools, it is necessary to know whether the detected antigen is removed from circulation in response to treatment, as it is otherwise impossible to interpret whether this tool can be used to evaluate re-infection levels and treatment efficacy. To our knowledge this is the first report describing a consistently detectable decline of CCA after only 24 hours in response to praziquantel treatment in a community based study using the urine strip test. The decreased level of CCA in response to this first round of mass drug administration in this community could be observed both after baseline treatment and again at two weeks in the two treatments arm, albeit to a lesser extent quite possibly due to the already lowered worm burden. This supports the hypothesis that positive CCA measures 24 hours after treatment reflect the presence of CCA derived from immature worms, which are not susceptible to praziquantel [8], as well as from adult worms surviving PZQ treatment. A proportion of these immature worms will then have matured by the two week time point and treatment will take effect at the second administration. Murine experimental work support this view as studies have shown that CCA is detectable in liver tissue as early as two weeks post infection depending on infection dose [16, 32]. Our own observations (personal communication A. Kildemoes) show detectable CCA in mouse urine based on the POC-CCA test from 2,5 weeks post-infection depending on infection intensity. These observations point towards CCA being excreted as soon as the immature worms start feeding in the bloodstream. Our observations of consistently declining CCA short term levels in response to praziquantel treatment in the vast majority of this community sample indicate that CCA measures can bear more weight in terms of interpreting treatment efficacy, potential resistance and re-infection patterns. Repeated CCA measures in individuals could potentially be used to assess how susceptible adult worms are to praziquantel and provide information on potential up-concentration of less susceptible worm populations. The consistent decline in CCA measures was observed despite the fact that differences in clearance rate in response to treatment are to be expected due to varying initial infection intensity, daily CCA level fluctuations and host metabolism. For individual diagnosis this tool should ideally be used in combination with other diagnostic measures such as egg counts or schistosomal DNA in faecal matter in order to gain a clearer picture of the true burden of infection and related disease manifestations in the host.
Both the lateral flow test used here and the POC-CCA test have showed good specificity and sensitivity for S. mansoni in large studies including sub-Saharan multi-country settings [33, 34], however the interpretation of trace measures need elucidation. More stable performance by CCA than KK on this study’s data is shown by estimated sensitivities and specificities for both diagnostic methods fitted in a latent markov model with no gold standard assumed and published in Koukounari et al. (2013) [21]. Traditionally calculated comparisons of CCA to KK performance can be found in Table A in S1 Supporting information. Lamberton et al. (2014) argues that trace measures should be considered positive based on a study carried out in children from an endemic area in Uganda comparing a 6KK as standard with a single urine sample tested with POC-CCA [35]. The data presented here support this view as we observe a response to treatment in the majority of trace positives within 24hrs at both treatment time points (Table C in S1 Table).
The strip test used for this study has a comparable sensitivity to the POC-CCA, hence the current results point towards trace measures in POC-CCA to be considered as positives in populations comparable to the sample studied here. This is particularly relevant in high prevalence areas in a mass drug administration context as the additional drug cost and minimal praziquantel adverse effects can be perceived as acceptable even if a small proportion of trace positives are indeed S. mansoni negative. However, more knowledge on sensitivity and specificity in terms of potential false positives due to cross-reactivity mediated by other helminth infections is needed, when operating in low-endemic settings and/or when elimination is targeted and a test-then-treat approach is implemented. When testing cross-reactivity on sera and urine from people infected with parasites it is important to take both biological compartmentalisation and geographical epidemiology into account. Worms with stages feeding within the human host would logically heighten the possibility of presence of regurgitated antigens. Some of these could potentially share epitopes with CCA. Furthermore, killing of tissue dwelling or migratory stages could release antigens to the same host tissue compartments as schistosomes. This could be the case with migratory or tissue dwelling stages of parasites such as filarial nematodes, Fasciola spp. or in cysticercosis. The monoclonal IgG antibody, which recognise CCA from adult and immature actively feeding schistosomes and not egg-derived CCA [29, 36, 37], used in this study and in the POC-CCA, has been tested for cross-reactivity on sera from people infected with a range of parasites [15, 38]. However, the sample sizes for each parasite is very small and further studies are needed to identify to which extent false-positives occur. Here we showed that the often co-endemic hookworm infection is not likely to be a confounder for the assay as there is no association observed between hookworm and CCA positivity (irrespective of traces being interpreted as positive or negative). Recently, caution in terms of interpreting test read-outs of the POC-CCA when used in pregnant women has been raised [39]. Pregnant women may present with a changed Lewis-X moiety level in urine and this could possibly increase the background. However, while more research is needed in order to elucidate potential cross reactions, current data support the application of the available CCA tests into the existing schistosomiasis epidemiology and diagnostics toolbox [40, 41]. Application of the tests should be combined with guidelines for use and interpretation of test results, which are suited for mass drug administration based control programmes and individual diagnosis in test-then treat approaches, respectively. Both the current study based on urine strip CCA test in a community sample and POC-CCA studies in children even in a low endemic sub-Saharan settings have presented less variability and better or comparable sensitivity of CCA than a single KK alone [21, 42–44], despite the fact that daily CCA fluctuations exist [42].Furthermore, application in combination with other tools in a migrant and traveller context with histories of exposure to fresh water in S. mansoni endemic regions is also relevant as it opens up for earlier detection of infection and better clinical management [45]. Potential discrepancies in interpretation of colour reaction development is particularly relevant in the context of trace/no trace scoring in terms of inter-reader variability [26].
For monitoring treatment efficacy purposes, it is relevant to compare the prevalence outcomes obtained at baseline+24hrs and the two weeks after treatment time points. An overall prevalence deemed similar at these two time points equals a logistical advantage for screening strategies. 43% of individuals, who gave urine samples at both time points (n = 359), had an unchanged score (Table 3). Only 10% of observations differ at overall sample level at the two time points in terms of intensity of infection. The majority of these discrepant scores are positive at both time points, which doesn’t affect prevalence estimations at population level. The remaining observations with a positive score at baseline+24 hours and score 0 at the two week time point constitute only 5% resulting in a marginal overestimation of prevalence. The finding that a prevalence estimate made 24hrs after baseline treatment based on a single urine strip CCA test presented only a small over-estimation of prevalence compared to the two week time-point as result of a single treatment, provide prospects of using this tool as a cost-effective prevalence screening method [46]. A team could gain the same information on a single overnight field site visit compared to deploying a team a second time. There might also be positive compliance outcomes of such a strategy as people are already on site combined with the more socially acceptable and ease of delivery urine sample compared with faecal sampling. Furthermore, this would provide basis for administration of a second dose of praziquantel to individuals with residual high scores to maximize the efficacy of treatment either immediately and/or administered two to three weeks later by a community worker in order to allow immature worms to mature and become drug susceptible.
A consistently measurable decline in CCA levels is seen in urine already 24 hours after both a single and a second praziquantel treatment in a community sample. The decline in CCA in response to treatment occurs in the majority of individuals even for trace measurements, which supports that trace should be considered positive in populations and transmission zones comparable to this study. These observations inform use of and provide weight to further interpretation of CCA based diagnostics and support extended applicability of the CCA based tools for S. mansoni in control programmes and on individual diagnostics level.
|
10.1371/journal.pgen.1007543 | Single-strand annealing between inverted DNA repeats: Pathway choice, participating proteins, and genome destabilizing consequences | Double strand DNA breaks (DSBs) are dangerous events that can result from various causes including environmental assaults or the collapse of DNA replication. While the efficient and precise repair of DSBs is essential for cell survival, faulty repair can lead to genetic instability, making the choice of DSB repair an important step. Here we report that inverted DNA repeats (IRs) placed near a DSB can channel its repair from an accurate pathway that leads to gene conversion to instead a break-induced replication (BIR) pathway that leads to genetic instabilities. The effect of IRs is explained by their ability to form unusual DNA structures when present in ssDNA that is formed by DSB resection. We demonstrate that IRs can form two types of unusual DNA structures, and the choice between these structures depends on the length of the spacer separating IRs. In particular, IRs separated by a long (1-kb) spacer are predominantly involved in inter-molecular single-strand annealing (SSA) leading to the formation of inverted dimers; IRs separated by a short (12-bp) spacer participate in intra-molecular SSA, leading to the formation of fold-back (FB) structures. Both of these structures interfere with an accurate DSB repair by gene conversion and channel DSB repair into BIR, which promotes genomic destabilization. We also report that different protein complexes participate in the processing of FBs containing short (12-bp) versus long (1-kb) ssDNA loops. Specifically, FBs with short loops are processed by the MRX-Sae2 complex, whereas the Rad1-Rad10 complex is responsible for the processing of long loops. Overall, our studies uncover the mechanisms of genomic destabilization resulting from re-routing DSB repair into unusual pathways by IRs. Given the high abundance of IRs in the human genome, our findings may contribute to the understanding of IR-mediated genomic destabilization associated with human disease.
| Efficient and accurate repair of double-strand DNA breaks (DSBs), resulting from the exposure of cells to ionizing radiation or various chemicals, is crucial for cell survival. Conversely, faulty DSB repair can generate genomic instability that can lead to birth defects or cancer in humans. Here we demonstrate that inverted DNA repeats (IRs) placed in the vicinity of a DSB, interfere with the accurate repair of DSBs and promote genomic rearrangements and chromosome loss. This results from annealing between inverted repeats, located either in different DNA molecules or in the same molecule. In addition, we describe a new role for the Rad1-Rad10 protein complex in processing fold-back (FB) structures formed by intra-molecular annealing involving IRs separated by long spacers. In contrast, FBs with short spacers are processed by the Mre11-Rad50-Xrs2/-Sae2 complex. Overall, we describe several pathways of DSB promoted interaction between IRs that can lead to genomic instability. Given the large number of IRs in the human genome, our findings are relevant to the mechanisms driving genomic destabilization in humans contributing to the development of cancer and other diseases.
| Double strand breaks (DSBs) in DNA result from the interactions of DNA with environmental agents or cellular metabolites, and are a major source of genetic instability (reviewed in [1, 2]). The efficient repair of DSBs acquired over time is critical for cell survival. Since some DSB repair pathways result in mutations and chromosomal rearrangements and others do not (reviewed in [3–5]), the choice of safer repair pathways is important to maintain genomic integrity.
The two main types of DSB repair pathways, non-homologous end-joining (NHEJ) and homologous recombination (HR), are conserved from yeast to humans [2, 6]. NHEJ occurs by re-ligation of broken DNA ends (Fig 1A). HR proceeds via DNA 5’-3’ end resection followed by repair using a homologous DNA sequence (e.g., a sister chromatid or a homologous chromosome) for repair (Fig 1B). HR pathways include synthesis-dependent strand annealing (SDSA), double-Holliday junction (dHJ) repair, break-induced replication (BIR) and single-strand annealing (SSA) [1, 7]. Both SDSA and dHJ repair can lead to gene conversion (GC) (Fig 1B-i,ii). dHJ can also lead to GC associated with crossing-over (CO) and it is frequently used for the repair of meiotic DSBs [2]. While, SDSA predominates in mitotic cells and rarely leads to crossing-over (reviewed in [1, 2]), BIR is a pathway used in situations where only one broken DNA end can find homology for repair (reviewed in [8–10]). BIR is initiated by invasion of a 3’ single-stranded-DNA (ssDNA) end into the homologous sequence, followed by an unusual type of DNA synthesis copying large, often chromosomal sized DNA regions (Fig 1B-iii). Instead of a replication fork, BIR proceeds via a migrating DNA bubble, which causes frequent mutations and large chromosomal changes including deletions, duplications and translocations [10–14]; (reviewed in [8]). Thus, BIR is more deleterious than SDSA, which very rarely leads to genomic instability (reviewed in [1, 2]). SSA proceeds via annealing between DNA repeats that become single-stranded following resection of broken DNA ends. SSA between direct DNA repeats can lead to deletions (Fig 1B-iv), while SSA between inverted repeats promotes the formation of unusual DNA structures namely inverted, sometimes dicentric, chromosome dimers (ID) and fold-backs (FB) (Fig 1B-v) [15–18]. Since the potential for genomic destabilization varies between various DSB repair pathways, it is important to understand how the choice between the pathways is made by living cells. When DSBs are initiated in yeast cells in such a way that both ends possess the homology to the homologous template, repair predominantly proceeds (in more than 90% of the cells) by SDSA or dHJ leading to gene conversion (GC), while repair by BIR is rare [19, 20]. The suppression of BIR leading to the predominance of GC outcomes was proposed to result from a recombination execution checkpoint (REC) that stimulates the choice of healthier DSB repair mechanisms [19]. In addition, several factors, including the type of DNA damage, chromatin structure, the extent of DSB resection, and the age of the cells can influence the choice of DSB repair pathway [21, 22–24], and also reviewed in [1, 7]. For example, older yeast cells use BIR more frequently as compared to younger cells [21]. Also, multiple DNA breaks caused by gamma irradiation of yeast, frequently result in genomic rearrangements via BIR [25]. Such rearrangements could result from breaks introduced at the position of repeated elements, as well as it is also possible that multiple breaks suppress REC, and thus promote the repair of individual breaks via BIR.
The presence of inverted repeats (IRs) in the vicinity of a DSB can also affect DSB repair [17, 26, 27]. IRs are characterized by a symmetry that allows them to switch between inter- and intra-strand base pairing (reviewed in [28]) to form secondary DNA structures, such as cruciforms, hairpins, or fold-backs which can promote genetic instabilities. For example, IRs are implicated in the formation of rearrangements associated with several diseases including cancer, Emanuel syndrome, and X-linked congenital hypertrichosis [29–31]. IRs induce genomic instability by various mechanisms, including impediment of DNA replication at the position of non-B DNA structure formed by IRs or by affecting the repair of DNA. For example, we demonstrated that IRs located in cis to HO-induced DSB affected its repair [17]. In particular, we observed that DSBs that were introduced by HO-endonuclease promoted SSA between IRs located 30-kb centromere proximal to these breaks on two different sister chromatids, and this led to the formation of inverted dicentric dimers (Inverted dimers, IDs) [17]. Mitotic breakage of these dicentric chromosomes leads to chromosome rearrangements including translocations, deletions, and duplications. We established that IR-mediated inter-molecular SSA requires Rad52, but is Rad51-independent, similar to SSA involving direct DNA repeats [24]. Our experiments also suggested that inter-molecular SSA can potentially re-route DSB repair from one pathway to another. Specifically, we observed that inter-molecular SSA can successfully compete with DSB repair by allelic BIR and re-route it into ectopic BIR. The problem, however, was that allelic BIR in our previous system was a very slow and inefficient process. A much more interesting question is whether an efficient DSB repair pathway, allelic gene conversion, can also be outcompeted by IR-mediated SSA. However, this question could not be answered using our original experimental system where IRs were located 30-kb away from the DSB-site [17], as it took more than 7 hours to complete SSA due to the need to resect over 30 kb to initiate SSA, whereas GC takes only 2 to 3-hours [20]. In addition, it remained unclear which properties of the IRs played a role in the ability of the IRs to undergo SSA and to promote rearrangements. For example, it has been previously shown that during DNA replication, IRs can form covalently closed hairpin-capped DNA molecules called fold-backs (FBs) [15, 16, 28]. In yeast, the Mre11-Rad50-Xrs2(MRX)/Sae2 complex can cleave FBs with short (12-bp) hairpin loops generating DSBs that can lead to chromosomal rearrangements [16]. The replication of unprocessed FBs leads to the formation of inverted dicentric dimers promoting breakage-fusion-bridge cycles [15, 16] leading to chromosomal rearrangements including DNA amplifications, deletions, and translocations [15, 18]. While the formation of FBs by IRs during replication is well-studied [18], the formation and processing of FBs following DSB induction is not well investigated. It remains unknown whether formation of FBs depends on the structure of IRs and how FBs containing long hairpin loops are formed and processed. Also, the molecular mechanisms of IR-mediated genome destabilization remain unclear.
We here present a new experimental system, with IRs located close (~3-kb) to the DSB site. Using this system, we tested whether DSBs can promote IR-mediated SSA, and whether it can compete with pathways leading to the formation of GC. Our data suggests that IRs can participate in both inter- and intra- molecular SSA leading to the formation of inverted dicentric chromosome (ID) and fold-back (FB) structures, respectively. We show that the choice between inter- and intra- molecular SSA is determined by the length of the spacer separating the IRs. In strains where IRs are separated by a long 1-kb spacer, inter-molecular SSA is more efficient, while intra-molecular SSA dominates in strains containing IRs separated by a short 12-bp spacer. We also demonstrate that different protein complexes are involved in the processing of FBs containing long and short hairpin loops. Specifically, FBs with short (12-bp) hairpin loops are processed by the MRX-Sae2 complex, whereas FBs with long (1-kb) loops are processed by the Rad1-Rad10 complex. In addition, we demonstrate that the presence of IRs in the vicinity of DSBs promotes the repair of these DSBs via BIR, while decreasing the frequency of GC. Taken together, our results indicate how inverted DNA repeats influence the choice and outcome of DSB repair and can result in chromosomal rearrangements.
Our goal was to examine the effect of inverted DNA repeats on the repair of nearby DSBs. To accomplish this, we generated a haploid strain (IR-1000) with an IR consisting of two 2-kb-long sequences separated by a 1-kb-long spacer and located 3 kb centromere proximal to MATa, on chromosome III (Fig 2A-i). This IR was shorter than the Ty elements that we previously used to describe IR-mediated SSA [17], but was within a length range that was previously used by various groups to study SSA genetics and IR-mediated GCRs [18, 25, 32–34]. The IR in this study was created by inserting a 2-kb long copy of PHO87 gene in an inverted orientation next to the endogenous PHO87 gene originally located at this chromosomal position.
A DSB was introduced at MATa by the HO-endonuclease expressed under the control of a galactose-inducible promoter [35]. To follow the kinetics of DSB repair, samples from yeast liquid culture were collected at various time-points after DSB induction, and repair was analyzed using contour-clamped homogeneous electric field (CHEF) gel electrophoresis followed by Southern blot hybridization using ADE1-specific sequence as a probe (Fig 2A-i, ii, Fig 2B). A 450-kb DSB repair product was accumulated between 2 and 5 hours following DSB induction (Fig 2B, wt). Based on the size of this repair product, we hypothesized that it represents an inverted dimer (ID) formed by single strand annealing (SSA) between IRs located on different sister chromatids, similarly to previously described [17] (Fig 2A-ii, see also Fig 1B-v-a). This was confirmed by digesting genomic DNA with AvrII restriction enzyme, followed by gel electrophoresis and Southern blot hybridization using a probe specific to RBK1 gene, a sequence located near and centromere proximal to the IRs (Fig 2C, probe P-1, 2D). We observed that the 23-kb original fragment (OF), present in chromosome III before the induced DSB in the 0-hr time-point, disappeared after the induction of a DSB (Fig 2C–OF, 2D-wt), whereas at the 0.5-hr time-point a 9-kb fragment corresponding to the DSB cut fragments (CF) appeared, and finally at the 6-hr time point a 7-kb fragment appeared consistent with the molecular-size of ID (SSA repair product) fragment. The amount of ID product detected at 6 hours after DSB induction was 76% of the amount of broken chromosome-III (cut-fragment, CF) in wt cells (Fig 2D-wt, S1A Fig), indicative of a high efficiency of this ID pathway resulting from inter-molecular SSA. The cut-fragment band (0.5-hr time-point post DSB induction) was used to calculate the amounts of repair products because it represented the total amount of the chromosome that was broken, following DSB induction. In addition, the high molecular weight of the original-chromosome fragment (OF) at 0-hr time-point made the Southern-transfer of this fragment inefficient and therefore unreliable for calculations. Overall, our results indicated that inter-molecular SSA leading to the formation of ID is the predominant pathway of DSB repair in this haploid strain. Similar results were obtained in the same strain in the presence of nocodazole, which prevents mitotic divisions and excludes the possibility of dicentric formation after replication of a fold-back structure resulting from intra-molecular annealing between inverted repeats. As predicted, formation of IDs occurred much faster in the strain containing IRs that are located 3 kb away from MATa compared to what we observed in our previous experimental system where IRs were located > 30-kb away from the DSB site [17]. The formation of IDs requires Rad52, but not Rad51 (Fig 2B), consistent with the current models for SSA [24, 36]. We also asked whether ID formation requires the flap endonuclease Rad1 known to be involved in SSA between direct DNA repeats [37]. In the absence of Rad1, the level of IDs was reduced more than 10-fold as compared to RAD1 (wt) strain (Fig 2B-5-hr wt vs rad1Δ). The analysis of genomic DNA digested with AvrII endonuclease (Fig 2C) also showed that the amount of ID also was significantly decreased in rad1Δ (3%) compared to RAD1 (wt, 76%) (Fig 2D; see also S1A Fig). Overall, we conclude that inter-molecular SSA between IRs is a predominant pathway of DSB repair when the DSB occurs near IRs. Surprisingly, the analysis of DSB repair in rad1Δ revealed the formation of an additional 3.5-kb band at 6-hr time-point (Fig 2D-rad1Δ). We hypothesized that this fragment represented a fold-back hairpin molecule formed by annealing of IRs within the same resected 3’-single stranded DNA with the spacer between IRs forming a single stranded loop at the tip of the hairpin (Fig 1B-v-b), similar to those reported previously when inverted repeats were included in a single-stranded DNA region [15, 16] although the role of Rad1 in the accumulation of FBs was not reported before.
To confirm that the 3.5-kb repair product detected in IR-1000-rad1Δ strain (Fig 2D) had a structure expected of a FB molecule (Fig 1B-v-b), we used a combination of native and denaturing gel electrophoresis of genomic DNA digested with AvrII and SphI enzymes (Fig 3A and 3B). The two AvrII sites flanked the IR region, while the SphI site was located in the spacer DNA between the IRs and was expected to be refractory to cutting by Sph1 following FB formation due to single-stranded nature of the loop formed by the spacer DNA between IRs in the FB molecule (Fig 1B-v-b, Fig 3A-FB). Due to these reasons, such an FB-fragment is expected to open into an open-FB molecule with a size that is twice the size under denaturing conditions compared to the size of the FB molecule under native electrophoresis conditions. Indeed, denaturing gel electrophoresis shows a new 7-kb band, expected from the denaturing of a 3.5 kb hairpin molecule (Fig 3A and 3B-denaturing gel). Moreover, the 3.5-kb band was consistently detected in our experiments in IR-1000-rad1Δ strain at 6-hr time-point following hybridization to either the RBK1-specific probe P-1 (Fig 2C and 2D-rad1Δ) or the PHO87- specific probe P-2 (Fig 3A and 3B-native gel). Further, the 3.5-kb band did not show hybridization to a probe specific to the 3’-flap sequence, including the region located between IRs and MATa (S1B Fig, probe P-3), which was expected to be removed prior to FB formation (S1C Fig). These results suggest that the 3.5-kb repair product observed in IR-1000-rad1Δ strain has a structure that is consistent with FB hairpin molecule. Based on our data we propose that Rad1 is responsible for the processing of FBs with 1-kb loops, and when Rad1 is missing, FBs accumulate in the cells.
Another possibility was that the accumulation of FBs in the absence of Rad1 in our IR-1000-rad1Δ strain resulted from the channeling of inter-molecular SSA intermediates that are incapable of forming IDs in the absence of Rad1 into intra-molecular SSA generating FBs. This possibility was addressed by repeating the experiment in cells arrested at the G1 stage of the cell cycle when inter-sister SSA cannot occur. We asked whether FBs can be detected in RAD1+ (wt) cells that are arrested in G1 stage where there is no sister chromatid to form IDs. The experiment was performed in RAD1+ and rad1Δ strains with IRs separated by a 1-kb spacer that were also bar1Δ (to facilitate cell cycle arrest at G1 by α-factor pheromone) [38]. The complication for our experiment was that resection of DSB ends required for the formation of FBs is normally defective in cells arrested at G1. To overcome this problem, we made the cells ku70Δ similar to described in ([39]), and this restored resection to an extent. Following the experiment in G1-arrested cells, we detected accumulation of FBs in rad1Δ, but not in RAD1 cells (Fig 3A and 3C). It took about 12-hours following DSB induction to detect FBs in this rad1Δ strain. This slower rate of FB formation likely resulted from slower DSB resection in G1-arrested cells. The key result is that even in G1-stage of the cell cycle where no inter-molecular SSA could occur, we still detected an accumulation of FBs in rad1Δ, but not in RAD1 cells. These results suggested that FB represents an independent pathway of its own rather than the result of channeling from an unsuccessful ID formation and supported our hypothesis of direct involvement of Rad1 in processing of FB structures containing long spacers.
To confirm that processing of FBs by Rad1 is not sequence-specific, we modified the construct by replacing its inter-IR spacer with DNA of bacteriophage lambda that was 1-kb or 1.5-kb long (S2 Fig). In both these strains, FBs were detected in the rad1Δ background, but not in the isogenic RAD1+ strains. This is consistent with Rad1 being involved in the processing of FBs with long (≥1-kb) ssDNA loops regardless of their DNA sequence.
The novel role of Rad1 in processing of FBs that we observe in our IR-1000-rad1Δ strain is similar to the known function of the MRX-Sae2 complex in FB processing [16]. However, we did not observe FBs in the sae2Δ derivative of our IR-1000 strain (Fig 3D and 3E- sae2Δ). This could be due to the difference in the properties of IRs in our strain as compared to IRs used in previously published strains [16]. In particular, the role of MRX-Sae2 in FB processing was previously described in strains where IRs were separated by short (~12-bp) spacers, while the distance between IRs in our strain is 1000 bp. We therefore reduced the length of the spacer between IRs in our strain from 1000 bp to 12 bp generating the IR-12 strain. Consequently, we observed a robust accumulation of FBs in sae2Δ derivative, but not in rad1Δ derivative of IR-12 strain at 6-hours following DSB induction (Fig 4A and 4B- sae2Δ, 4B-rad1Δ). In particular, at 6-hr after DSB induction, the intensity of a 3-kb FB band observed in IR-12-sae2Δ was 86% of the intensity of the cut fragment measured at 0.5-hr (Fig 4B-sae2Δ, S1D Fig), demonstrating that the FB is the predominant outcome for IR-12. The FB structure of this product was also confirmed by a combination of native and denaturing gel electrophoresis of genomic DNA digested with BseYI restriction enzyme (Fig 4C and 4D). In particular, the FB molecule formed a 2.4-kb band following the native gel electrophoresis, but on denaturing gel electrophoresis a 4.8 kb band was observed. This is expected from the denaturing and opening of a 2.4-kb hairpin shaped FB molecule to form 4.8-kb open-FB molecule (Fig 4C). (Please note that some 2.4-kb band remained present in 6-hr sample following denaturing gel electrophoresis as the remaining OF, CF & ID also generate 2.4-kb band).
We also observed the accumulation of FB structures in IR-12-rad1Δsae2Δ strain (Fig 4B). However, the amount of FBs accumulated in IR-12-rad1Δsae2Δ was 3-fold lower as compared to the amount of FBs accumulated in IR-12-RAD1sae2Δ (Fig 4B, S1D Fig). The fold-back structure of the repair products accumulated in rad1Δsae2Δ was also confirmed by a combination of native and denaturing gel electrophoresis (Fig 4C and 4E) as described before. This demonstrates that Rad1 is also involved in the removal of 3’-flaps prior to the formation of a covalently-closed hairpin FB structure.
Since rad1Δ and sae2Δ showed opposite effects on accumulation of FBs for IRs with large and small spacers, we examined their effect on IRs with intermediate spacers. In particular, we analyzed DSB repair in strains containing IRs separated by 100-bp and 500-bp spacers. In these experiments, neither sae2Δ nor rad1Δ single mutants showed accumulation of FBs (Fig 5A–5D). Accumulation of FBs was detected only in double rad1Δsae2Δ mutants. The yield of FB product in rad1Δsae2Δ strains containing 100-bp and 500-bp spacers was respectively 18% and 17% of the broken chromosome III (S1D Fig). Thus, our data suggests that although Rad1 and MRX-Sae2 have unique roles in processing FBs with long (~1-kb) and small (~12-bp) loops respectively, both proteins could process FBs with intermediate size loops.
In our experiments, the length of the spacer separating IRs determined not only the proteins involved in FB processing, but also the choice between inter-molecular SSA (leading to ID formation) and intra-molecular SSA (leading to FB formation). In particular, the change of the spacer length from 1-kb to 12-bp led to the decrease in the amount of IDs from 76% to 28% (Figs 3E-wt, 4B-wt; S1A Fig-IR1000 & IR-12), and to a corresponding increase in FB formation from 4% in 1kb-rad1Δ to 86% in 12bp-sae2Δ (Figs 3E-rad1Δ, 4B-sae2Δ; S1D Fig-IR-1000-rad1Δ & IR12-sae2Δ). The intermediate distance between IRs (100-bp or 500-bp), was associated with intermediate level of IDs (~50% of the broken chromosome III (Fig 5B-wt, 5D-wt, S1A Fig)), and intermediate amounts of FBs (~18% of broken chromosome III (S1D Fig, Fig 5B-rad1Δsae2Δ, Fig 5D-rad1Δsae2Δ)). Together, these results suggest that ID is the prominent repair product in strains with a long (1-kb) spacer, while FB is predominant in strains with small (12-bp) spacer. This difference in the efficiency of FB formation at least in part explains why the effect of sae2Δ on FB formation, observed in strains with small (12-bp) spacer, was so much more pronounced as compared to the effect of rad1Δ that we observed in strains with long (1- kb) spacer.
Several proteins, including Rad10, Saw1 and Slx4 are known to collaborate with Rad1 in the removal of 3’-flaps during SSA involving direct repeats [40, 41]. We used the IR-1000 strain with deletion of RAD10, SAW1, or SLX4 genes to ask whether the corresponding genes also accumulate FBs similar to that observed in rad1Δ strain, suggesting these proteins assist Rad1 for processing FB structures. We observed that each of these deletions led to the accumulation of unprocessed FBs (Fig 6A and 6B). On the other hand, deleting RAD14, which is involved (along with RAD1) in nucleotide excision repair (NER) (reviewed in [42]) did not produce detectable FBs (Fig 6B). Thus, processing of FBs in our experiments requires a 3’-flap removal by Rad1-Rad10 complex rather than its role in NER. Additionally, since the amount of FBs accumulated in rad1Δ in our experiments was low (4% (S1D Fig)), we asked whether other known yeast nucleases may work in parallel with Rad1-Rad10 in processing FBs. We expected that deleting the gene encoding for such a nuclease may also lead to accumulation of FBs, while deletion of this gene in a rad1Δ background will increase the amount of FBs as compared to the rad1Δ alone. We observed that deleting genes encoding several known yeast nucleases (SLX1, MUS81, YEN1 and RAD2 (reviewed in [43, 44]) did not lead to FB accumulation, and did not visibly change the amount of FB accumulated when these genes were deleted in rad1Δ background (Fig 6B). Therefore, it is unlikely that Slx1, Mus81, Yen1 or Rad2 participate in the processing of FB structures. This result also suggested that these four endonucleases were unlikely to be the unknown proteins involved in the removal of 3’- flaps in the absence of Rad1.
Finally, we asked whether Rad52, known to promote annealing between single-stranded homologous DNA in yeast [45] is required for the formation of fold-back structures. We observed that deleting RAD52 prevented the accumulation of FBs in both IR-12-sae2Δ and IR-1000-rad1Δ strains, where accumulation of FBs occurred otherwise in IR-12-sae2Δ and IR-1000-rad1Δ strains containing RAD52 (Fig 6A, 6C, 6D and 6E). Importantly, formation of FBs was not affected by deleting RAD51, encoding a strand invasion protein, in IR-12-sae2Δ and IR-1000-rad1Δ strains (Fig 6C and 6E). We conclude that Rad52 (but not Rad51) is required for DSB-induced annealing between inverted repeats located in the same DNA molecule regardless of the distance separating the repeats.
We have previously demonstrated that DSBs introduced in such a way that both broken ends bear homology to the allelic position of the homologous chromosome are predominantly repaired by pathways leading to allelic GC, and rarely by BIR, a deleterious DSB repair mechanism that operates when only one of the broken DNA ends is available for repair (reviewed in [8, 9, 46, 47]). Here we asked whether the presence of IRs close to a DSB position leads to IR-mediated SSA and promotes BIR.
To test this idea, several diploid strains were generated by crossing MATa strains carrying IRs (IR-12 or IR-1000) or not carrying IRs, with a MATα-inc strain (AM476, S1 Table), which contains a mutation (α-inc) that prevents from cutting of MAT by HO endonuclease (similar to [48], Fig 7A-i). In addition, the MATα-inc strain did not contain IRs, but carried a URA3 gene inserted at the position of the THR4 gene. DSBs were initiated at MATa by plating cells on a galactose-containing media, and the repair outcomes were analyzed (see Materials and Methods for details) and classified (Fig 7A- ii, iii, iv). We observed that in the absence of IRs, DSB repair predominantly led to GC (>90%) with MATa replaced by MATα-inc. Of all repair events, approximately 5% of repair events had phenotypes indicative of BIR. In particular, they were represented by either Ade+ Thr- Ura+ full colonies or by sectored colonies, where one sector was Ade+ Thr- Ura+ while another was Ade+ Thr+ Ura+ (Fig 7A-iii and 7B-BIR). In addition, <0.5% of all events were Ade-Thr-Ura+ and resulted from the loss of the broken chromosome (Fig 7A-iv and 7B-Loss).
We observed a 3-fold increase of BIR events in diploids containing inverted repeats (IR-1000 or IR-12) in the vicinity of a DSB as compared to the control without inverted repeats (Fig 7B-BIR). The presence of IRs also led to a significant increase of chromosome loss (approximately 4% of chromosome loss in IR-1000 and IR-12 versus no detected cases of chromosome loss in no-IR strain (Fig 7B-Loss)), which suggested that the repair of DSBs introduced in the vicinity of IRs was more likely to fail as compared to strains without IRs. Overall, our results suggested that IRs located in the vicinity of a break channel some DSB repair from GC into BIR and increased chromosome loss. We hypothesized that the decrease of GC resulted from IR-mediated SSA disrupting the coordinated behavior of two broken DSB ends, which is known to be important for the successful GC repair [26]. This hypothesis was supported by our observation of a concomitant accumulation of IDs along with GC repair products that we observed in a time-course experiment performed in IR-1000 diploid strain (Fig 8A). This observation confirmed that inter-molecular SSA can indeed successfully compete with allelic GC repair. Interestingly, ID formation was observed in the IR-12 diploid as well, where they could be either formed by inter-molecular SSA or by replication of unprocessed FB structures. The latter possibility is less likely since no accumulation of FBs was observed in this strain, which suggested that FBs were efficiently processed by the MRX-Sae2 complex.
We previously demonstrated that inverted dicentric dimers resulting from inter-molecular SSA between IRs can break in mitosis and the broken fragments are often repaired by ectopic BIR using non-homologous chromosomes generating translocations ([17, 49]). Here we asked whether the BIR occurring in diploids containing IR-1000 or IR-12 was allelic or ectopic. Allelic BIR proceeds via invasion of a broken fragment into the homologous chromosome at an allelic position. Ectopic BIR proceeds by strand invasion into a non-homologous chromosome at repeated sequences such as Ty or delta elements. CHEF-gel electrophoresis analysis of DNA isolated from Ade+ Thr- Ura+ repair outcomes (Fig 7B-BIR) from IR-containing diploids revealed that 66% (71/108) and 70% (52/74) resulted from allelic BIR in IR-1000 and IR-12 strains respectively (Fig 8B and 8C). Further, 26% (28/108) and 19% (14/74) of analyzed repair outcomes in IR-1000 and IR-12, respectively, contained an unusually sized ADE1- hybridizing band indicative of ectopic BIR repair (similar to [13, 17]) (Fig 8B and 8C). The frequency of ectopic BIR was slightly less (6%) in No-IR strain as compared to strains with IR, even though the increase of ectopic BIR proved to be statistically significant only in a case of IR-1000 as compared to No-IR strain (P = 0.02 by Fisher’s exact test). Surprisingly, our CHEF analysis of Ade+ Thr- Ura+ events also demonstrated that approximately 10% of these outcomes contained repair band that was similar in size to the original, recipient chromosome (Fig 8B and 8C-Long-GC), and therefore could represent cases with a very long (>16-kb) gene conversion, known to be mechanistically similar to BIR [19].
Inverted DNA repeats are a known source of chromosomal rearrangements including those leading to cancer (reviewed in [50, 51]), Emanuel syndrome, and X-linked congenital hypertrichosis. Two molecular mechanisms promoting IR-mediated genomic destabilization include DNA breaks introduced into secondary DNA structures formed by IRs as well as the interruption of DNA replication at the positions of secondary structures formed by IRs, both of which have been the subject of multiple studies [15, 16, 18, 28, 52]. In the present study, we provide several insights into, yet another mechanism of genetic destabilization mediated by IRs, where IRs mis-route the repair of DSBs introduced in their vicinity. First, we demonstrate that such DSBs can be channeled into two types of IR-mediated single-strand annealing: intra-molecular SSA and inter-molecular SSA and that the choice between these two pathways is dictated by the size of the DNA spacer separating two halves of the IR. Second, we demonstrate that different protein complexes participate in the processing of FB structures and the choice of the protein complex is dictated by the distance between the two arms of the IRs. Finally, we demonstrate that both IR-mediated mechanisms channel DSB repair into pathways that are prone to genomic rearrangements.
According to our model (Fig 9A and 9B), 5’- to 3’ resection of DSB ends allows inter-molecular or intra-molecular SSA between IRs leading to the formation of inverted dicentric dimers or fold-back structures. The 3’ clipping activity of Rad1-Rad10 complex is required during both of these processes, but these two pathways differ from each other in their level of Rad1-Rad10 dependency. In particular, the inter-molecular SSA requires clipping of two non-homologous 3’-flaps (Fig 9A and 9B; Inter-molecular SSA pathway), which cannot proceed without Rad1-Rad10, thus making inter-molecular SSA fully Rad1-Rad10 dependent. However, formation of fold-backs that requires cleavage of only one 3’-flap, shows only partial dependence on Rad1-Rad10 complex. This observation indicated that besides Rad1-Rad10 complex, some other protein (s) can perform 3’ flap cleavage, which is consistent with several other studies [53, 54]. At present, the identity of this hypothetical protein remains unknown. Based on our results, it seems unlikely that this protein is Rad2, Slx1, Yen1, or Mus81 though we cannot exclude that several nucleases from this list may perform this function in a redundant fashion.
The predominance of intra-molecular SSA for IRs separated by short (12-bp) rather than long (1-kb) spacer might be explained by a higher probability of interaction for IRs in close proximity to each other as compared to those located further away. Another possible contributing factor is thermodynamic stability of FB structures, which was shown to be inversely proportional to the length of the spacer separating IRs [55, 56]. Importantly, this data is consistent with the results obtained by [57–59] who previously demonstrated that the efficiency of intra-molecular annealing between IRs is inversely proportional to the distance between them. The facilitation of inter-molecular SSA between IRs separated by longer spacers may be promoted by sister chromatid cohesion, which is known to facilitate inter sister recombination [60, 61], while shorter spacers could make it difficult for IRs located on different sisters to contact each other. Importantly, our data demonstrates that Rad52 is involved in the formation of FBs containing both long and short spacers, which refuted a previously formulated hypothesis that the formation of FBs separated by short spacers could be Rad52-independent [16]. The Rad52 dependence (and Rad51-independence) of FB formation, is consistent with an SSA pathway responsible for FB formation requiring a direct involvement of Rad52 in the annealing between ssDNA regions.
We propose that when FBs are formed, they can be processed by two different protein complexes. Specifically, FBs containing short single-strand loops are processed by MRX-Sae2 complex (consistent with previous observations [16, 62, 63]), while FBs with long spacers (≥ 1-kb) are processed by Rad1-Rad10 complex. Interestingly, two nucleases appear partially redundant when it comes to processing of FBs with intermediate-length spacers. We speculate that longer ssDNA spacers appear to be similar to a long 3’-flap that therefore can be recognized and cut by the Rad1-Rad10 complex. Previously, it has been shown that binding of RPA to ssDNA stimulates the cutting activity of XPF protein, a human homolog of Rad1 [64, 65]. Perhaps the binding of larger ssDNA loops by RPA can stimulate the activity of Rad1-Rad10 complex. Conversely, RPA binding might inhibit processing of ssDNA by MRX-Sae2 protein complex. Together, RPA binding to the larger ssDNA loops might play a role in the choice of proteins for ssDNA loop processing. However, it remains unclear how different loops of larger size are distinguished from each other. For example, both 500-bp and 1000-bp loop are expected to be bound by RPA; yet the former can be cut by MRX-Sae2 complex, while the latter not, thus suggesting the existence of some other, presently unknown, factors contributing to the distinction between these loops in respect to their recognition by MRX-Sae2 complex, even though both can be processed by Rad1-Rad10 complex. Finally, our data obtained in diploid cells suggests that both types of IR-mediated SSA (inter- and intra-molecular) cause shifting of DSB repair from GC into BIR. To explain this, we propose that a DSB end proximal to IRs is frequently engaged in IR-mediated SSA which disrupts coordinated action of the two broken DNA ends required for successful GC. In particular, we propose that inter-molecular SSA leads to the formation of inverted dicentric dimers, which are subsequently broken during mitotic division (Fig 9). The new broken ends are essentially “one-ended” and therefore prone to repair via BIR (reviewed in [8]). In a case of FB formation, the increased BIR can be explained by one of two possible scenarios. The FB could be processed by either MRX-Sae2 or by Rad1-Rad10 nuclease complexes, and the resulting broken end can be repaired by either GC (if the second end of the original, HO-induced DSB end is still available), or by BIR (if the second DSB end is degraded or engaged in another recombination event). Alternatively, if the FB remains unprocessed, then following DNA replication, it could be converted into an ID (similar to [18, 28]). The latter possibility is consistent with the results of our time-course analysis using diploid strains containing IR-12, where significant accumulation of IDs was detected.
Here we characterized two pathways of DSB repair involving annealing between IRs that can result in genomic destabilization. IRs are especially frequent in the 17% of the human genome comprised of LINEs and Alu repeat elements that commonly form clusters, where individual repeats often have opposite orientations [66, 67]. Alu elements (~300-bp in size) are the most abundant class of large dispersed DNA repeats in humans, while LINEs (~6-kb in size) are autonomous transposable elements that are present in thousands of copies in human genome [68, 69]. LINEs correlate in size of repeat to yeast Ty elements (~6-kb) that are responsible for the majority of IR-mediated GCRs reported in yeast [17, 18, 32, 34, 57]. Additionally, Alu elements promote genome destabilization similar to that induced by yeast Ty elements [58]. The length of IRs used in our study was within the length of naturally occurring IRs as well as within the lengths used previously by various groups to study IR-mediated GCRs, as well as to investigate the genetics of SSA [17, 18, 24, 25, 32, 36, 37, 57–59]. In addition, the majority of previous studies believed that only IR pairs separated by very short (<20-bp) spacers promote high levels of genomic instabilities [58, 68, 70]. However, clusters of repeat elements do not always have spacers of the short <20-bp size [66, 67]. The results obtained in our study also point towards genome destabilization that can result from IRs separated by much longer (~ 1kb) spacers.
With IRs being abundant in the human genome (see for example in [28, 50, 71]), we propose that the mechanisms described here are likely to occur in human cells. Specifically, it has been demonstrated that DNA amplifications described in various human cancers often are associated with IRs [50, 72–74]. Further, genetic destabilizations observed in cancer cells often result from mis-routed DSB repair (reviewed in [75, 76]), so it is possible that gene amplifications and/or genomic destabilizations observed in cancer cells can result from inter- or intra-molecular SSA between IRs. Additionally, it is possible that inter- or intra-molecular SSA between IRs induced by DSBs can contribute to genetic rearrangements leading to several other human diseases. For example, Emanuel syndrome is known to result from recurrent and non-recurrent translocations induced by palindromic AT-rich repeats (PATRRs) [77]. These translocations result from DNA breaks introduced in the middle of the palindrome, likely at the tip of a secondary structure (cruciform or hairpin formed by the palindrome) (reviewed in [78]). We propose that the mechanism described in our study can facilitate formation of secondary structures by PATRR, which in turn can facilitate breakage leading to formation of translocations. It is also possible that the molecular mechanisms described here could contribute to the formation of β-thalassemia, known to originate from large palindrome-induced deletions [30].
In the future, it will be important to compare the structural properties of inverted DNA repeats promoting different disease-associated rearrangements, since it may reveal that different IR-mediated SSA pathways are involved in their formation. It is also important to characterize the respective roles of human homologs of Rad1-Rad10 and MRX-Sae2 complexes in the formation of DSB and IR-promoted chromosomal rearrangements since it may shed light on the specific mechanisms of chromosomal rearrangements leading to various pathologies in humans.
The genotypes of all strains used in this study are shown in S1 Table. The haploid yeast strain AM1102 contains two 2-kb long PHO87 sequences in inverted orientation to each other, separated by a 1-kb-long spacer, and located 3-kb centromere proximal to MATa. AM1102 was derived from JKM111 (S1 Table) in three steps using delitto perfetto approach [79]. First, AM1036 was constructed by transformation of JKM111 with a DNA fragment generated by PCR amplification of the pGSKU plasmid [79] using two primers:
OL556: 5’AGATTGGGAGTTGGTAGACCTTTTGGTCGTTAATGAAATTGAGGGTCTTCtagggataacagggtaatccgcgcgttggccgattcat-3’
and OL557: 5’-GTACTTCAGGGCTTTCGTGCGAACAGAAAAGCACCCCTCTCGAACCCAAAttcgtacgctgcaggtcgac-3’
Upper-case letters correspond to the sequences upstream of the position 193891-bp and downstream of 194040-bp of chromosome III (according to Saccharomyces cerevisiae genome database). Lower-case letters correspond to the sequences specific to the pGSKU cassette [79]. Subsequently, AM1050 was constructed by transforming AM1036 with a DNA fragment generated by PCR amplification of the PHO87 region using genomic DNA of JKM111 as a template. The primers used for the amplification were as follows:
OL554: 5’-AGATTGGGAGTTGGTAGACCTTTTGGTCGTTAATGAAATTGAGGGTCTTCctcacactttctcaaatacaacgc-3’;
and OL555: 5’-GTACTTCAGGGCTTTCGTGCGAACAGAAAAGCACCCCTCTCGAACCCAAAccagccgattccataaggttttaa-3’
Upper-case letters correspond to the sequences upstream of 193891-bp and downstream of 194040-bp positions of chromosome III, respectively. The lower-case letters correspond to the sequences specific to the PHO87 region. This transformation resulted in formation of a 2-kb-long inverted repeat with a 1-kb-long spacer sequence separating IRs located 3-kb centromere proximal from MATa.
Subsequently, a series of strains containing the same 2-kb long repeat of PHO87, but separated by shorter spacers of various lengths were constructed by delitto perfetto protocol in two steps [79]. First, AM1354 was constructed by transformation of AM1050 with a DNA fragment generated by PCR amplification of the plasmid pGSKU using primers:
OL1107: 5’-TTCATTGACCATTCAAAGAAAAGGTGCTGCTGAAAGCATGCCACTGTATAAAGATGTTCAGAtagggataacagggtaatccgcgcgttggccgattcat-3’
and OL1108: 5’-GGTGTGTAGAGTCACAAATAGAAAGTGCTTTTGGATCGTCCGGTGAAATTGCAGTAATACttcgtacgctgcaggtcgac-3’.
The upper-case letters correspond to the sequences located upstream of 194287-bp and downstream of 194950-bp position of chromosome III and located inside the spacer separating inverted repeats in AM1050. Lower case letters correspond to the sequences specific to the pGSKU cassette.
Subsequently, a series of strains with 500-bp (AM1398), 100-bp (AM1399) and 12-bp (AM1648) spacer separating IRs were constructed by deleting the pGSKU cassette and varying amounts of spacer DNA sequence. This was carried out by transformation of AM1354 with following denatured oligonucleotides as described in [79]:
For the construction of AM1398 with 500-bp-long spacer between 2-kb inverted repeats:
OL1258: 5’-GTATTGCACATGAACACTAGGCAAGAATTAATAGAAAGTGAATGGAATGGCTTCACTAGCCCATAAGAAAGCACAGCATTCTAC-3’
For construction of AM1399 with 100-bp-long spacer between 2-kb inverted repeats:
OL1135: 5’-CGAGAGGGGTGCTTTTCTGTTCGCACGAAAGCCCTGAAGTACCACAGCATCGATAGAATACCTGTAGGCAGAGCGACAGCAAAA-3’
For construction of AM1648 with 12-bp-long spacer between 2-kb inverted repeats:
OL1214: 5’-TCCATGCAGTACTTAAAACCTTATGGAATCGGCTGGTTTGGGTTGCCCCCAGCCGATTCCATAAGGTTTTAAGTACTGCATGGA-3’
Finally, AM1757, AM1760 & AM1762 were constructed from AM1398, AM1399 and AM1648 respectively by transformation with ade3::GAL::HO cassette using pop-in-pop-out approach [80].
AM2290 and AM2295 were constructed from AM1102 using delitto perfetto approach in two steps. First, AM1352 was constructed from AM1102 by inserting pCORE cassette into the spacer region between 2-kb IRs. This was carried out by transforming AM1102 with DNA fragment generated by PCR amplification of the spacer DNA sequence and pCORE plasmid [79] using the following primers:
OL1109: 5’–TTCATTGACCATTCAAAGAAAAGGTGCTGCTGAAAGCATGCCACTGTATAAAGATGTTCAGAgagctcgttttcgacactgg– 3’
OL1110: 5’–GTGTGTAGAGTCACAAATAGAAAGTGCTTTTGGATCGTCCGGTGAAATTGCAGTAATACtccttaccattaagttgatc– 3’
Upper-case letters correspond to the sequences upstream of 194287-bp position and downstream of 194950-bp position on chromosome III. The lower-case letters correspond to the sequences specific to the pCORE cassette. Finally, AM2290 and AM2295, with inverted repeats separated by bacteriophage lambda spacers that were 1-kb (in AM2290) or 1.5-kb (in AM2295) in length were constructed by transformation of AM1352 with DNA fragments generated by digestion of plasmids–pAM-36 & pAM-37 with HindIII and EcoRI restriction enzymes, generating fragments containing 1-kb or 1.5-kb DNA sequence from bacteriophage lambda DNA in between PHO87 inverted repeat sequence.
The majority of single-gene deletion mutants were constructed by transformation with a PCR-derived KAN-MX module (see the strain list in S1 Table for genes deleted by KAN-MX) flanked by terminal sequences homologous to the sequences flanking the open reading frame of each gene [81]. The resulting constructs were confirmed by PCR and by phenotype analysis. To disrupt RAD1, a pJH551plasmid was digested with SalI and transformed into recipient strains, and transformants were selected on a leucine-dropout media. To construct AM1362, AM1401, AM2207, RAD52 was disrupted using plasmid pJH181 that was digested with BamHI and transformed into recipient strains, and transformants were selected on a leucine-dropout media. The full list of oligonucleotides used in this study is available upon request.
Yeast were grown at 30°C. Yeast extract-peptone-dextrose (YEPD) media, and synthetic complete medium, with nitrogenous bases or amino acids omitted were prepared as described in [82]. YEP-lactate (YEP-lac) and YEP-galactose (YEP-gal) contained 1% yeast extract and 2% Bacto peptone that was supplemented with either 3.7% lactic acid (pH 5.5) or 2% (w/v) galactose, respectively. 5-fluoroorotic acid (5-FOA) was added to synthetic complete medium with trace amounts of uracil (30mg/mL). For selection of cells containing insertion of KAN::MX cassette G418 (Calbiochem) was added to YEPD media to the final concentration of 0.3 g/L. The arrest of cells at G1 cell cycle stage was achieved by addition of α-factor (Zymo research) to the final concentration of 5μM.
To determine the distribution of DSB repair events in yeast diploids, yeast cultures were grown in leucine drop-out medium for ~20 hours, followed by ~18 hours in YEP-lactate, and then diluted and plated on YEP-Gal plates similar to [13, 48]. Colonies that grew on YEP-Gal plates were then replica plated onto appropriate omission media to determine the fraction of DSB repair events with the following phenotypes: (i) Ade+ Thr+ Ura+ (gene conversion; GC) (ii) Ade+ Thr+/- Ura-/+ (GC associated with crossing-over); (iii) Ade+ Thr- Ura+ (BIR); (iv) Ade- Thr- Ura+ (chromosome loss). The frequency of individual DSB repair outcomes were determined as described in [48] based on at least three experiments. Cell viability following HO induction was derived by dividing the number of colony-forming units (CFUs) on YEP-Gal by the number of CFUs on YEPD.
The kinetics of DSB repair was analyzed in experiments where cell culture aliquots were removed every 30 or 60 minutes following galactose addition, similarly to described previously [20, 48, 83]. In brief, yeast cells were grown for approximately 20 hours in YEP-lac media to a final concentration of approximately 1–2 x 107 cells per ml. To induce HO breaks, galactose was added to the final concentration of 2%. We used 50-ml culture aliquots for the extraction of DNA by glass-bead-phenol-sodium dodecyl sulfate protocol [84] that was used for Southern blot analysis. For CHEF gel electrophoresis, chromosomal plugs were prepared as described in [48]. For Southern blot analysis, the DNA was digested with appropriate restriction enzymes, and the resulting fragments were separated using 1% agarose gel. CHEF was performed using genomic DNA embedded in plugs of 1% agarose. The DNA was subsequently analyzed using Southern blot analysis. The membranes containing separated DNA fragments were hybridized to the chromosome-specific DNA probes, produced by labeling of chromosome-specific DNA fragments with P32. The ADE1-specific probe was prepared using pJH879 plasmid digested with SalI [85]. All other probes were generated by PCR amplification using genomic DNA of AM1102 as a template and were specific to the following yeast sequences: 1) Probe P-1 specific to RBK1 region of chromosome III located between 193591–193891 bps positions. 2) Probe P-2, specific to PHO87-IR specific located between 196893–197172 bps positions of chromosome III. 3) Probe P-3, specific to BUD5 region of chromosome III located between 197181–197488 bps positions; 4) Probe P-4, specific to the IRC7 region of chromosome VI located between 265004–265341 bps positions. The chromosomal coordinates for these probes were derived from SGD. The full list of oligonucleotides that were used to generate each probe is available upon request. Blots were analyzed by using ImageQuant TL 7.0 software from GE healthcare. The amounts of DSB repair products obtained by combination of CHEF and Southern blot analysis were calculated as the percentage of original chromosome intensity before DSB induction that was converted to the repair product. Quantification of the fold-back (FB) and Inverted Dimer (ID) levels in the restriction digestion analysis were calculated as the percentage of cut fragment-CF (detected at 0.5-hr time point) that was converted into FBs and IDs respectively. All the gels were run with appropriate molecular-size markers that allowed to estimate the size of every band visualized in every image. All quantifications were based on the results of at least three independent experiments.
|
10.1371/journal.pntd.0002546 | High Prevalence of Skin Disorders among HTLV-1 Infected Individuals Independent of Clinical Status | Human T-cell lymphotropic virus type 1 (HTLV-1) infection can increase the risk of developing skin disorders. This study evaluated the correlation between HTLV-1 proviral load and CD4+ and CD8+ T cells count among HTLV-1 infected individuals, with or without skin disorders (SD) associated with HTLV-1 infection [SD-HTLV-1: xerosis/ichthyosis, seborrheic dermatitis or infective dermatitis associated to HTLV-1 (IDH)].
A total of 193 HTLV-1-infected subjects underwent an interview, dermatological examination, initial HTLV-1 proviral load assay, CD4+ and CD8+ T cells count, and lymphproliferation assay (LPA).
A total of 147 patients had an abnormal skin condition; 116 (79%) of them also had SD-HTLV-1 and 21% had other dermatological diagnoses. The most prevalent SD-HTLV-1 was xerosis/acquired ichthyosis (48%), followed by seborrheic dermatitis (28%). Patients with SD-HTLV-1 were older (51 vs. 47 years), had a higher prevalence of myelopathy/tropical spastic paraparesis (HAM/TSP) (75%), and had an increased first HTLV-1 proviral load and basal LPA compared with patients without SD-HTLV-1. When excluding HAM/TSP patients, the first HTLV-1 proviral load of SD-HTLV-1 individuals remains higher than no SD-HTLV-1 patients.
There was a high prevalence of skin disorders (76%) among HTLV-1-infected individuals, regardless of clinical status, and 60% of these diseases are considered skin disease associated with HTLV-1 infection.
| HTLV-1 infection may increase the risk of developing skin disorders. A total of 193 HTLV-1 infected subjects were studied, including asymptomatic carriers and HAM/TSP patients. Of the subjects, 76% had an abnormal skin condition, with a high prevalence both among HTLV-1 asymptomatic carriers and HAM/TSP patients. The most prevalent SD-HTLV-1 was xerosis/acquired ichthyosis (48%), followed by seborrheic dermatitis (28%). Patients with SD-HTLV-1 were older (51 vs. 47 years), had a higher prevalence of myelopathy/tropical spastic paraparesis (HAM/TSP) (75%) and an increased first HTLV-1 proviral load compared with patients without SD-HTLV-1. When excluding HAM/TSP patients, the first HTLV-1 proviral load of SD-HTLV-1 individuals remains higher than no SD-HTLV-1 patients. Thus, skin diseases are highly prevalent among HTLV-1-infected individuals.
| Adult T-cell leukemia/lymphoma (ATLL), HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) and infective dermatitis associated with HTLV-1 (IDH) are the main diseases caused by human T-cell lymphotropic virus type 1 (HTLV-1) infection [1]–[3]. However, several other clinical conditions have been associated with this viral infection, such as uveitis, thyroiditis, arthritis and polymyositis [4]–[6].
There are an estimated 5 to 10 million HTLV-1 infected individuals worldwide and Brazil is considered a highly endemic area for HTLV-1 infection, with the largest absolute number of HTLV-1 infected individuals, with more than one million people living with this virus [7]–[9]. Despite this high prevalence, only a few studies on the dermatological aspects of HTLV-1 infection have been described in this country [10].
There is a lack of surrogate markers to assess the infected patients who have a higher risk for HTLV-1 associated skin disorders. Moreover, there are few immunological studies among HTLV-1-infected persons who are simultaneously suffering from HAM/TSP and skin diseases. The aim of this study is to evaluate the prevalence of skin disorders in HTLV-1-infected individuals and to correlate this prevalence with the initial HTLV-1 proviral load, and initial CD4+ and CD8+T cell count.
In the last 18 years, a cohort of HTLV-infected subjects has been followed in the HTLV-outpatient clinic at the Institute of Infectious Diseases “Emilio Ribas” (IIER), with the support of nurses, nutritionists and physical therapists. From a total 450 HTLV-1-infected individuals, including asymptomatic carriers and HAM/TSP patients, 193 of them were consecutively evaluated for skin disorders from January 2008 to July 2010 by the same dermatologist, blinded for the clinical status to minimize information bias. Demographical and clinical dates were collected, and dermatological examinations were carried out. The HIV co-infected individuals were excluded, but HCV co-infected subjects were included.
Ethics Statement: Written informed consent was obtained from all participants, and the IIER ethical board approved the protocol.
The patients underwent laboratory studies, including HTLV-1 serological diagnosis, initial CD4+ and CD8+T cell counts, and an initial HTLV-1 proviral load. In accordance with previous studies, the following skin disorders associated with HTLV-1 infection (SD-HTLV-1) were considered: xerosis/acquired ichthyosis, seborrheic dermatitis and infective dermatitis associated with HTLV-1 (IDH) [1], [10]–[12]. HAM/TSP was diagnosed according to previously established criteria [13]. Skin culture and punch skin biopsies were performed when clinical examination was not sufficient for a dermatological diagnosis. Antibodies to HTLV-1/2 were detected by a diagnostic enzyme-linked immunosorbent assay (ELISA) and confirmed by Western blot analysis and polymerase chain reaction (PCR), which are capable of discriminating between HTLV-1 and HTLV-2 [14]. To determine the counts of CD4+ and CD8+ T-cell subsets, fresh whole blood specimens were collected in EDTA tubes and subjected to flow cytometry (Coulter EPICS® XL-MCLÔ Flow Cytometer - Beckman Coulter, Fullerton, CA), using human monoclonal antibodies anti-CD3, anti-CD4, and anti-CD8, labeled with fluorochrome.
The results of the first HTLV-1 proviral load were available in the database. Quantitative proviral DNA levels were detected by a real-time automated PCR method, using TaqMan probes for the pol gene. The albumin gene served as the internal genomic control, and MT2 cells were used as a positive control. The results are reported as copies/10000 PBMCs, and the detection limit was 10 copies [15].
Data were analyzed using SPSS 17.0 software. The association between independent variables and the outcome was analyzed either by Student's t-test or ANOVA (normal distribution variables), or by Mann-Whitney test (non-normal distribution variables), while the association between categorical variables and the outcome was assessed by the X2 test. HTLV-1 proviral load was log-transformed to obtain a normal distribution. Correlations between HTLV-1 proviral load and SD-HTLV were performed using Spearman's rank correlation. Data are expressed as mean ± standard deviation (normal distribution variables) or median and interquartile range (non-normal distribution variables). Statistical significance was set at a p value<0.05.
One hundred ninety-three HTLV-1-infected subjects (43% of all HTLV-1-infected patients at the HTLV outpatient clinic from the Emilio Ribas Institute cohort) underwent a dermatological exam. Mean age of patients was 49.4±12.3 years. Female gender had a higher prevalence of HTLV-1 infection (72%). Regarding the presence of neurological involvement, 38% of the patients had a diagnosis of HAM/TSP. The dermatological examination revealed a high prevalence of skin disorders among the HTLV-1-infected patients (76%). Sixty-five individuals (34%) had one dermatological condition, and 42% (n = 81) of the patients had two or more dermatological conditions. Among the 147 patients that had an abnormal skin condition, 79% (n = 116) had a skin disorder associated with HTLV-1 infection (SD-HTLV-1) (xerosis/ichthyosis or seborrheic dermatitis) and 21% (n = 31) had other dermatological diagnoses. The most prevalent skin disorder associated with an HTLV-1 diagnosis was xerosis/ichthyosis (48%), followed by seborrheic dermatitis (28%).
Table 1 shows the prevalence of skin disorders in the HTLV-1 patients based on a diagnosis of HAM/TSP or asymptomatic carriers. SD-HTLV-1 were more prevalent on HAM/TSP patients (xerosis/acquired ichthyosis (p = 0.007; seborrheic dermatitis (p = <0.0001); IDH (p = 0.022). Patients with SD-HTLV-1, including asymptomatic carriers and HAM/TSP, are older, have a higher prevalence of HAM/TSP, and have a higher first HTLV-1 proviral load (performed for 109 patients, p = 0.009), compared with patients without SD-HTLV-1 (Table 2). Note that 75% of the SD-HTLV-1 group was made up of HAM/TSP individuals.
Table 3 depicts the presence of SD-HTLV-1 in subjects that are asymptomatic for neurological symptoms (absence of HAM/TSP). Mean age of SD-HTLV-1 patients was 51 years, as compared with 44 years for HTLV-1 patients without SD (p = 0.002), regardless of gender, CD4+ and CD8+ T cell counts (p = 0.489; p = 0.824). The initial HTLV-1 proviral load was significantly higher for the group with SD-HTLV-1 as compared with that for the group without SD-HTLV-1 and asymptomatic for neurological symptoms (p = 0.021).
Notably, 76% of the HTLV-1-infected asymptomatic carriers and 88% of the HAM/TSP patients showed some skin disorder in our study. These findings are similar to those described in two previous studies involving asymptomatic carriers and HAM/TSP subjects [10], [12]. Thus, it is important to stress that HTLV-1 infection may have an etiological link to skin disease. In fact, skin disorders are highly associated with HTLV-1 infection, regardless of neurological symptoms, and they may represent a clinical warning sign for the diagnosis or progression of this infection [10]–[12].
Excluding HAM/TSP cases, subjects with a diagnosis of SD-HTLV-1 are older than groups with other types of skin disorders or individuals with a normal dermatological exam. However, no significant association was observed with gender. These findings suggest that older HTLV-1-infected individuals, who probably have a longer duration of their viral infection have had more time to develop SD-HTLV-1.
Although IDH is the only skin disease in which HTLV-1 infection is a criterion for diagnosis, other skin disorders could also be associated with HTLV-1 infection, including xerosis/acquired ichthyosis and seborrheic dermatitis, as previously demonstrated [10]–[12], [16]. In fact, in our study, these illnesses were the most prevalent skin disorders associated with adult HTLV-1-infected individuals, regardless of the clinical status. However, studies with a longer follow-up should be performed to assess the hypothesis that these skin manifestations are related to HTLV-1 infection.
The exact pathogenic mechanism of IDH still needs to be made clear, but the current view is that a diagnosis of HTLV-1 infection is necessary, leading in susceptible individuals to immune deregulation, with subsequent immunosuppression and superinfection with Staphylococcus aureus and beta-haemolytic streptococci, what additionally leads to chronic antigenic stimulation and persistent inflammation of the skin. Genetic, host and environmental factors have been shown to be associated [17].
Xerosis and acquired ichthyosis have been described as the main dermatological manifestations associated with HAM/TSP patients [11], [12], [18]. Xerosis is characterized by dryness of the skin, and acquired ichthyosis is clinically characterized by cutaneous xerosis and the formation of polygonal thin flat scales of varying sizes, mainly on the extremities [19].
Acquired ichthyosis is a consequence of hypohydrosis that may be secondary to the involvement of the autonomic nervous system, affecting directly the HTLV-1-infected skin cells [12], [20]. On the basis of histopathological and immunohistochemical analyses of skin fragments of acquired ichthyosis from HAM/TSP individuals, it was concluded that keratinocytes are activated, probably as a result of cytokines that are liberated from HTLV-1-infected lymphocytes. This activation leads to an interference in keratinocyte differentiation and migration, resulting in a defect in the processes of normal desquamation, accumulation and retention of corneocytes [18]. We noticed that more than 50% of the SD-HTLV-1 cases involved xerosis/acquired ichthyosis and decided to include them in the same group because the acquired ichthyosis showed a similar clinical dermatological pattern, comparable with higher degree xerosis, and so they were clinically difficult to differentiate [12].
HTLV-1 proviral load is a laboratorial risk marker for the development of HAM/TSP and other diseases related to HTLV-1 infection [21], [22]. The initial HTLV-1 proviral load was higher in the group with SD-HTLV-1 (p = 0.009) as well as in the SD-HTLV-1 HAM/TSP-free group (p = 0.021), both of which were statistically significant. Although the proviral load may differ greatly among individuals, it is relatively stable during the course of the HTLV-1-related disease [23]. In HAM/TSP patients this finding suggests that proviral load reaches a stable level determined by the relationship between viral expression and the immune response against the virus [23].
As previously shown, HTLV-1 was identified by PCR on skin cells in addition to lymphocytes in HTLV-1-infected persons, regardless of their clinical status 10. Because of these findings, several authors believe that HTLV-1 can modify the function of infected cells, resulting in skin disorders that are caused directly by the presence of the virus in the infected cells [10], [16], [24]. Another possible mechanism of skin disorders among HTLV-1-infected individuals is the production of cytokines in HTLV-1-infected lymphocytes, promoting a functional disturbance on skin cells [18]. This lack of association may be explained by the presence of HTLV-1 in specific sites that occur during HAM/TSP and the proviral load in the cerebral spinal fluid (CSF) [25]. Lezin et al. reported that the proviral load quantified in CSF was able to distinguish clearly between healthy groups of HTLV-1 carriers and patients presenting HAM/TSP [25].
Finally, for the first time, a high prevalence of skin disorders (76%) independent of clinical status was disclosed among HTLV-1-infected individuals. These findings may have important implications in the clinical setting in places where this infection is endemic and dermatologists, infectious diseases specialists and general clinicians should be aware of skin presentations of the HTLV-1 infection. Moreover, the influences of demography and co-morbid conditions may be relevant, but have not been fully studied. Thus, data derived from a cohort of referred patients followed in a specialized HTLV clinic may not be a representative sample of the whole population of HTLV-1 patients and therefore unlikely to reflect the prevalence of skin conditions among all HTLV-1 patients.
|
10.1371/journal.pntd.0001716 | Prophylactic Platelets in Dengue: Survey Responses Highlight Lack of an Evidence Base | Dengue is the most important arboviral infection of humans. Thrombocytopenia is frequently observed in the course of infection and haemorrhage may occur in severe disease. The degree of thrombocytopenia correlates with the severity of infection, and may contribute to the risk of haemorrhage. As a result of this prophylactic platelet transfusions are sometimes advocated for the prevention of haemorrhage. There is currently no evidence to support this practice, and platelet transfusions are costly and sometimes harmful. We conducted a global survey to assess the different approaches to the use of platelets in dengue. Respondents were all physicians involved with the treatment of patients with dengue. Respondents were asked that their answers reflected what they would do if they were the treating physician. We received responses from 306 physicians from 20 different countries. The heterogeneity of the responses highlights the variation in clinical practice and lack of an evidence base in this area and underscores the importance of prospective clinical trials to address this key question in the clinical management of patients with dengue.
| A low platelet count is a common feature of dengue infection. It is thought that the platelet count correlates with the severity of the infection and may contribute to the risk of developing haemorrhage, a well-recognised complication of dengue. As a result of this platelet transfusions are used in some settings to reduce the risk of haemorrhage. There is currently no evidence to support this practice, and platelet transfusions are costly and sometimes harmful. We conducted a survey assessing the use of platelets in dengue. Respondents were all physicians involved with the treatment of patients with dengue. Respondents were asked that their answers reflected what they would do if they were the treating physician. We received 306 responses from 20 different countries. The striking feature of the survey responses was the heterogeneity of approaches to the use of platelets in dengue. These findings highlight the variation in clinical practice and lack of an evidence base in this area and underscore the importance of conducting prospective clinical trials to address this key question in dengue clinical management.
| Dengue is globally the most important arboviral infection and threatens an estimated 2.5 billion people worldwide [1]. Thrombocytopenia is almost universally observed in dengue infection [2]. This results from both reduced production and increased destruction of platelets [3]–[5]. It is thought that severe thrombocytopenia correlates with disease severity and may contribute to the risk of developing haemorrhage [6], [7]. The 2009 WHO dengue guidelines do not advocate the use of prophylactic platelet transfusions, whereas the 2011 regional WHO guidelines for South East Asia suggest prophylactic platelets may be considered in those with a platelet count less than 10×109/L [8], [9]. Some dengue-endemic countries support the use of prophylactic platelet transfusions to prevent haemorrhage in patients with thrombocytopenia, for example India (<10×109/L), whereas others, such as Brazil, do not [10], [11]. However platelet transfusions are costly, potentially dangerous and their use in dengue lacks an evidence base [12]–[15].
We conducted a survey among physicians directly involved in the care of dengue patients in order to determine how platelets are used in the clinical management of dengue. The majority of respondents were practicing physicians in dengue-endemic areas. The exceptions to this were respondents from Africa, where dengue is emerging, and the UK where the respondents were infectious disease physicians who regularly see patients who have recently travelled to dengue-endemic areas. A questionnaire containing nine case histories and an additional question about prophylactic platelet transfusion thresholds was emailed to physicians with experience in managing dengue patients and known to us. Respondents were specifically asked that their responses reflect what they would do if they were the treating physician. Email recipients were invited to further disseminate the questionnaire within their own clinical networks. The complete list of questions is available as a supplementary file (Questionnaire S1).
The case histories were based on real clinical cases seen at the Hospital for Tropical Diseases in Ho Chi Minh City, Vietnam. Four case histories describe patients with clinically non-severe dengue but varying levels of thrombocytopenia. Case 1 describes an 18-year-old female with platelets of 23×109/L and no bleeding. Case 2 describes a 28-year-old male with platelets of 29×109/L. He had no bleeding but a past history of a perforated peptic ulcer. Case 3 describes a 29-year-old female with a rapid fall in platelets to 22×109/L. She had no bleeding. Case 4 describes a 30-year-old male with platelets of 3×109/L and no bleeding. Five case histories describe patients with different manifestations of severe dengue associated with varying levels of thrombocytopenia. Case 5 describes a 19-year-old male with platelets of 18×109/L. He had dengue hepatitis but no bleeding. Case 6 describes a 20-year-old female with platelets of 17×109/L. She had suspected dengue encephalitis but no bleeding. Case 7 describes a 24-year-old male with platelets of 31×109/L. He had hepatic failure thought to be secondary to dengue but no bleeding. Case 8 describes a 23-year-old female with platelets of 8×109/L. She had shock, epistaxis and vaginal bleeding. Case 9 describes a 23-year-old male with platelets of 33×109/L. He had shock and mucosal bleeding. The final question aimed to determine thresholds at which a physician would consider transfusing platelets as prophylaxis against haemorrhage. Respondents were asked to select a single option.
In total, 306 physicians from 20 different countries responded within a specified time period. The responses from Asia were 52 from Indonesia, 7 from Bangladesh, 5 from the Philippines, 9 from Singapore, 8 from Cambodia, 18 from Malaysia, 3 from Thailand, 20 from Vietnam and 12 from India. The responses from Latin America and the Caribbean were 13 from Cuba, 10 from Brazil, 17 from Paraguay, 6 from Peru, 1 from Mexico, 1 from Bolivia, 1 from Martinique and 81 from Colombia. The responses from Africa were 37 from Nigeria and 2 from South Africa. In addition there were 3 responses from the UK.
Among the 4 case histories describing patients with clinically non-severe dengue associated with varying levels of thrombocytopenia, 16–24% of respondents recommended platelet transfusion at platelet concentrations of 22–29×109/L., but approximately one-third of the respondents would transfuse platelets if the count fell to 3×109/L. (Table 1)
Among the 5 case histories describing patients with different manifestations of severe dengue associated with varying levels of thrombocytopenia, more respondents would transfuse platelets if the patient was in shock and bleeding (case histories 8 and 9). There were substantial differences in the responses from physicians in Africa than those from Asia and America for all 9 cases (Table 1).
The final question aimed to determine thresholds at which a physician would consider transfusing platelets as prophylaxis against haemorrhage. Respondents were asked to select a single option. 31 (10%) respondents would consider a prophylactic platelet transfusion if the platelet count was below 50×109/L. 8 (2.6%) respondents would consider a prophylactic platelet transfusion if the platelet count was below 40×109. 10 (3.3%) respondents would consider a prophylactic platelet transfusion if the platelet count was below 30×109. 17 (5.6%) respondents would consider a prophylactic platelet transfusion if the platelet count was below 20×109. 46 (15%) respondents would consider a prophylactic platelet transfusion if the platelet count was below 10×109. 190 (62%) respondents would only consider transfusing platelets in patients with signs of haemorrhage.
The responses categorised by global region are summarised in Table 1.
Our study has limitations. There is an element of selection bias in the way the survey was conducted, as the physicians who distributed the survey within their countries were known to have an interest in dengue. The survey is subject to response bias meaning that the answers may not accurately reflect clinical practice in the respective countries. In addition, the country representation is not balanced.
Despite these limitations the striking result of this survey is the heterogeneity of approaches to the use of prophylactic platelet transfusions in dengue. 112/306 respondents would consider transfusing platelets prophylactically at various levels of thrombocytopenia. When the responses are categorised by region (Table 1) African respondents would advocate platelet transfusions more frequently, perhaps reflecting more limited experience with dengue and experience with other haemorrhagic fevers. The choice to use prophylactic platelet transfusions may be influenced by cost and availability of platelets, as well as individual experience in managing dengue and other medical conditions that affect the platelet count. There is considerable variability within countries suggesting an individual's practice may differ from recommendations in guidelines. For example 6/12 Indian respondents and 7/10 Brazilian respondents would consider the use of prophylactic platelets. The responses reflect wide variation in clinical practice and are indicative of the paucity of clinical evidence to guide practice in this area.
At present there is limited evidence to support the use of prophylactic platelet transfusions in dengue despite their inclusion in some national guidelines. As the global reach of dengue continues to expand the need to conduct clinical trials to construct an evidence base to guide the appropriate use of platelets in dengue becomes ever more pressing.
|
10.1371/journal.ppat.1001068 | Burkholderia Type VI Secretion Systems Have Distinct Roles in Eukaryotic and Bacterial Cell Interactions | Bacteria that live in the environment have evolved pathways specialized to defend against eukaryotic organisms or other bacteria. In this manuscript, we systematically examined the role of the five type VI secretion systems (T6SSs) of Burkholderia thailandensis (B. thai) in eukaryotic and bacterial cell interactions. Consistent with phylogenetic analyses comparing the distribution of the B. thai T6SSs with well-characterized bacterial and eukaryotic cell-targeting T6SSs, we found that T6SS-5 plays a critical role in the virulence of the organism in a murine melioidosis model, while a strain lacking the other four T6SSs remained as virulent as the wild-type. The function of T6SS-5 appeared to be specialized to the host and not related to an in vivo growth defect, as ΔT6SS-5 was fully virulent in mice lacking MyD88. Next we probed the role of the five systems in interbacterial interactions. From a group of 31 diverse bacteria, we identified several organisms that competed less effectively against wild-type B. thai than a strain lacking T6SS-1 function. Inactivation of T6SS-1 renders B. thai greatly more susceptible to cell contact-induced stasis by Pseudomonas putida, Pseudomonas fluorescens and Serratia proteamaculans—leaving it 100- to 1000-fold less fit than the wild-type in competition experiments with these organisms. Flow cell biofilm assays showed that T6S-dependent interbacterial interactions are likely relevant in the environment. B. thai cells lacking T6SS-1 were rapidly displaced in mixed biofilms with P. putida, whereas wild-type cells persisted and overran the competitor. Our data show that T6SSs within a single organism can have distinct functions in eukaryotic versus bacterial cell interactions. These systems are likely to be a decisive factor in the survival of bacterial cells of one species in intimate association with those of another, such as in polymicrobial communities present both in the environment and in many infections.
| Many bacteria encounter both eukaryotic cells and other bacterial species as a part of their lifestyles. In order to compete and survive, these bacteria have evolved specialized pathways that target these distinct cell types. Type VI secretion systems (T6SSs) are bacterial protein export machines postulated to puncture targeted cells using an apparatus that shares structural similarity to bacteriophage. We investigated the role of the five T6SSs of Burkholderia thailandensis in the defense of the organism against other bacteria and higher organisms. B. thailandensis is a relatively avirulent soil saprophyte that is closely related to the human pathogen B. pseudomallei. Our work uncovered roles for two B. thailandensis T6SSs with specialized functions either in the survival of the organism in a murine host, or against another bacterial cell. We also found that B. thailandensis lacking the bacterial-targeting T6SS could not persist in a mixed biofilm with a competing bacterium. Based on the evolutionary relationship of T6SSs, and our findings that B. thailandensis engages other bacterial species in a T6S-dependent manner, we speculate that this pathway is of general significance to interbacterial interactions in polymicrobial human diseases and the environment.
| Bacteria have evolved many mechanisms of defense against competitors and predators in their environment. Some of these, such as type III secretion systems (T3SSs) and bacteriocins, provide specialized protection against eukaryotic or bacterial cells, respectively [1], [2]. Gene clusters encoding apparent type VI secretion systems (T6SSs) are widely dispersed in the proteobacteria; however, the general roles of these systems in eukaryotic versus bacterial cell interactions are not known [3], [4].
To date, most studies of T6S have focused on its role in pathogenesis and host interactions [5], [6], [7]. In certain instances, compelling evidence for the specialization of T6S in guiding eukaryotic cell interactions has been generated. Most notably, the systems of Vibrio cholerae and Aeromonas hydrophila were shown to translocate proteins with host effector domains into eukaryotic cells [8], [9]. Evidence is also emerging that T6SSs could contribute to interactions between bacteria. The Pseudomonas aeruginosa HSI-I-encoded T6SS (H1-T6SS) was shown to target a toxin to other P. aeruginosa cells, but not to eukaryotic cells [10]. Unfortunately, analyses of the ecological niche occupied by bacteria that possess T6S have not been widely informative for classifying their function [3], [4]. These efforts are complicated by the fact that pathogenic proteobacteria have environmental reservoirs, where they undoubtedly encounter other bacteria. The observation that many bacteria possess multiple evolutionarily distinct T6S gene clusters–up to six in one organism–raises the intriguing possibility that each system may function in an organismal or context-specific manner [3].
The T6SS is encoded by approximately 15 core genes and a variable number of non-conserved accessory elements [4]. Data from functional assays and protein localization studies suggest that these proteins assemble into a multi-component secretory apparatus [11], [12], [13]. The AAA+ family ATPase, ClpV, is one of only a few core proteins of the T6S apparatus that have been characterized. Its ATPase activity is essential for T6S function [14], and it associates with several other conserved T6S proteins [15], [16]. ClpV-interacting proteins A and B (VipA and VipB) form tubules that are remodeled by the ATPase, which could indicate a role for the protein in secretion system biogenesis. Two proteins exported by the T6SS are haemolysin co-regulated protein (Hcp) and valine-glycine repeat protein G (VgrG). Secretion of these proteins is co-dependent, and they may be extracellular components of the apparatus [10], [13], [17], [18], [19], [20].
Burkholderia pseudomallei is an environmental saprophyte and the causative agent of melioidosis [21]. Infection with B. pseudomallei typically occurs percutaneously via direct contact with contaminated water or soil, however it can also occur through inhalation. The ecological niche and geographical distribution of B. pseudomallei overlap with a relatively non-pathogenic, but closely related species, Burkholderia thailandensis (B. thai) [22]. The genomes of these bacteria are highly similar in both overall sequence and gene synteny [23], [24]. One study estimates that the two microorganisms separated from a common ancestor approximately 47 million years ago [24]. It is postulated that the B. pseudomallei branch then diverged from Burkholderia mallei, which underwent rapid gene loss and decay during its evolution into an obligate zoonotic pathogen [25]. As closely related organisms that represent three extremes of bacterial adaptation, this Burkholderia group offers unique insight into the outcomes of different selective pressures on the expression and maintenance of certain traits.
B. pseudomallei possesses a large and complex repertoire of specialized protein secretion systems, including three T3SSs and six evolutionarily distinct T6SSs [3], [26], [27]. The genomes of B. thailandensis and B. mallei contain unique sets of five of the six B. pseudomallei T6S gene clusters; thus, of the six evolutionarily distinct “Burkholderia T6SSs,” four are conserved among the three species. Remarkably, T6SSs account for over 2% of the coding capacity of the large genomes of these organisms. For the current study, we have adopted the Burkholderia T6SS nomenclature proposed by Shalom and colleagues [28].
To date, only Burkholderia T6SS-5, one of the four conserved systems, has been investigated experimentally. The system was investigated in B. mallei based on its co-regulation with virulence determinants such as actin-based motility and capsule [27]. B. mallei strains lacking a functional T6SS-5 are strongly attenuated in a hamster model of glanders. Preliminary studies suggest that T6SS-5 is also required for B. pseudomallei pathogenesis [28], [29]. In one study, a strain bearing a transposon insertion within T6SS-5 was identified in a screen for B. pseudomallei mutants with impaired intercellular spreading in cultured epithelial cells [29]. The authors also showed that this insertion caused significant attenuation in a murine infection model.
Herein, we set out to systematically define the function of the Burkholderia T6SSs. Our study began with the observation that well-characterized examples of eukaryotic and bacterial cell-targeting T6SSs segregate into distant subtrees of the T6S phylogeny. We found that Burkholderia T6SS-5 clustered closely with eukaryotic cell-targeting systems, and was the only system in B. thai that was required for virulence in a murine model of pneumonic melioidosis. The remaining systems clustered proximally to a bacterial cell-targeting T6SS in the phylogeny. One of these, T6SS-1, displayed a profound effect on the fitness of B. thai in competition with several bacterial species. The function of T6SS-1 required cell contact and its absence caused sensitivity of the strain to stasis induced by competing bacteria. In flow cell biofilm assays initiated with 1∶1 mixtures of B. thai and Pseudomonas putida, wild-type B. thai predominated, whereas the ΔT6SS-1 strain was rapidly displaced by P. putida. Our findings point toward an important role for T6S in interspecies bacterial interactions.
We conducted phylogenetic analyses of all available T6SSs to examine the evolutionary relationship between eukaryotic and bacterial cell-targeting systems. The phylogenetic tree we constructed was based on VipA, as this protein is a highly conserved element of T6SSs that has been demonstrated to physically interact with two other core T6S proteins, including the ClpV ATPase [15]. In the resulting phylogeny, the systems of V. cholerae and A. hydrophila, two well-characterized eukaryotic cell-targeting systems, clustered closely within one of the subtrees, whereas the bacteria-specific P. aeruginosa H1-T6SS was a member of a distant subtree (Figure 1 and see Figure S1) [8], [9], [10]. In an independent analysis, Bingle and colleagues observed a similar T6S phylogeny, and termed these subtrees “D” and “A,” respectively [3].
Next we examined the locations of the six Burkholderia T6SSs. Interestingly, T6SS-5, the only Burkholderia system previously implicated in virulence, clustered within the substree containing the V. cholerae and A. hydrophila systems (Figure 1). Four of the remaining Burkholderia systems clustered within the subtree that included the H1-T6SS, and the final system was found in a neighboring subtree. These data led us to hypothesize that T6SSs of differing organismal specificities are evolutionarily distinct. Apparent contradictions between organismal specificity based on our phylogenetic distribution and studies demonstrating T6S-dependent phenotypes were identified, however these instances are difficult to interpret because specificity was not measured and cannot be ascertained from available data.
We chose B. thai as a tractable model organism in which to experimentally investigate the role of the Burkholderia T6SSs. Due to our limited knowledge regarding the function and essentiality of each gene within a given T6SS cluster, we reasoned it prudent to inactivate multiple conserved genes for initial phenotypic studies. Strains lacking the function of each of the five B. thai T6SSs (Burkholderia T6SS-3 is absent in B. thai) were prepared by removing three to five genes, including at least two that are highly conserved (Figure 1A). When possible, polar effects were minimized by deleting from a central location in each cluster.
To probe the role of the Burkholderia T6SSs in virulence, we utilized a recently developed acute pneumonia model of melioidosis [30]. The survival of mice infected with approximately 105 aerosolized wild-type or mutant bacteria was monitored over the course of ten days. Consistent with previous studies implicating T6SS-5 in B. mallei and B. pseudomallei pathogenesis, mice infected with ΔT6SS-5 survived the course and displayed no outward symptoms of the infection (Figure 2A) [27], [29]. On the other hand, those infected with the wild-type strain or strains bearing deletions in the other T6SSs succumbed by three days post infection (p.i.).
The B. thai T6SS-5 locus is adjacent to bsa genes, which encode an animal pathogen-like T3SS. Inactivation of the bsa T3SS secretion system also leads to dramatic attenuation of B. thai in the model we utilized [26]. The regulation of these secretion systems appears to be intertwined; a recent study in B. pseudomallei showed that a protein encoded within the bsa cluster strongly activates T6SS-5 of that organism [31]. To rule out the possibility that attenuation of ΔT6SS-5 was attributable to polar effects or changes in regulation of the bsa T3SS, we generated a strain bearing an in-frame deletion of a single gene in the cluster, tssK-5 (Figure 1A). A tssK-5 ortholog is readily identified in nearly all T6S gene clusters and it shares no homology with known regulators. Like the T6SS-5 deletion, ΔtssK-5 completely attenuated the organism (Figure 2B). Genetic complementation of this phenotype further confirmed that T6SS-5 is an essential virulence factor of the organism.
To investigate whether the retention of virulence in the ΔT6SS-1,2,4 and 6 strains could be attributed to either compensatory activity or redundancy, we next constructed a strain bearing inactivating mutations in all four clusters and measured its virulence in mice. Mice infected with this strain succumbed to the infection with similar kinetics to those infected with the wild-type, indicating that T6SS-5 is the only system of B. thai that is required for virulence in this model (Figure 2C). In summary, these data indicate that T6SS-5 is a major virulence factor for B. thai in a murine acute melioidosis model, whereas the remaining putative T6SSs of the organism are dispensible for virulence.
To more closely examine the requirement for T6SS-5 during infection, we monitored B. thai wild-type and ΔtssK-5 c.f.u. in the lung, liver, and spleen at 4, 24, and 48 hours following inoculation with approximately 105 bacteria by aerosol. At 4 hours p.i., no differences were observed in c.f.u. recovered from the lung (Figure 3A). After this initial phase, lung c.f.u. of ΔtssK-5 gradually declined, whereas wild-type populations expanded approximately 100-fold. Both organisms spread systemically, however significantly fewer ΔtssK-5 cells were recovered from the liver and spleen at 24 and 48 hours p.i. (Figure 3B).
Thus far, our findings did not distinguish between a specific role for T6SS-5 in host interactions, such as escaping or manipulating the innate immune system, versus the alternative explanation that T6SS-5 is generally required for growth in host tissue. To discriminate between these possibilities, we compared the virulence of ΔtssK-5 in wild-type mice to a strain with compromised innate immunity, MyD88−/− [32], [33]. Mice lacking MyD88 were unable to control the ΔtssK-5 infection and succumbed within 3 days (Figure 3C). The differences in virulence of the Δtssk-5 strain in wild-type and MyD88−/− infections suggest that T6SS-5 is required for effective defense of the bacterium against one or more innate immune responses of the host. Altogether, these data strongly support the conclusion that T6SS-5 has evolved to play a specific role in the fitness of B. thai in a eukaryotic host environment.
Earlier work by our laboratory has shown that T6S can influence intraspecies bacterial interactions. We showed that the H1-T6SS of P. aeruginosa targets a toxin to other P. aeruginosa cells [10], and that in growth competition assays, toxin-secreting strains are provided a fitness advantage relative to strains lacking a specific toxin immunity protein. Based on this information and the locations of the B. thai T6SSs within our phylogeny, we postulated that one or more of these systems could also play a role in interbacterial interactions. Preliminary studies indicated that T6S did not influence interactions between B. thai strains, thus we decided to test the hypothesis that the B. thai T6SSs play a role in interspecies bacterial interactions.
Without information to guide predictions of specificity, we developed a simple and relatively high-throughput semi-quantitative assay to allow screening of a wide range of organisms for sensitivity to the B. thai T6SSs. The design of the assay was based on two key assumptions for T6S-dependent effects – that they are cell contact-dependent and that they impact fitness (as measured by proliferation). To facilitate measurement of T6S-dependent changes in B. thai proliferation in the presence of competing organisms, we engineered constitutive green fluorescent protein expression cassettes into wild-type B. thai and a strain bearing mutations in all five T6SSs (ΔT6S) [34]. Control experiments showed that the lack of T6S function did not impact growth or swimming motility (Figure 4A and 4B). To test the assay, we conducted competition experiments between the GFP-labeled wild-type and ΔT6S strains against the unlabeled wild-type strain. The GFP-expressing cells were clearly visualized in the mixtures, and, importantly, wild-type and ΔT6S competed equally with the parental strain (Figure 4C; BT).
We next screened the B. thai strains against 31 species of bacteria. Most of these were Gram-negative proteobacteria (5α; 3β; 18γ), however two Gram-positive phyla were also represented (4 Firmicutes; 1 Actinobacteria). Although we endeavored to screen a large diversity of bacteria, many taxa could not be included due to specific nutrient requirements or an unacceptably slow growth rate under the conditions of the assay (30°C, Luria-Bertani (LB) medium). The outcomes of most competition experiments were independent of the T6SSs of B. thai. T6S-independent outcomes varied; in most instances, B. thai flourished in the presence of the competing organism (Figure 4C). However, a small subset of species markedly inhibited B. thai growth (Figure 4C; PAt, PAe, SM, VP). Interestingly, B. thai proliferation was reproducibly affected in a T6S-dependent manner in competition experiments against 7 of the 31 species tested. All of these were Gram-negative organisms, and in each case, B. thai ΔT6S was less fit than the wild-type. T6S-dependent competition outcomes fell into two readily discernable groups; the first included three γ- and one β-proteobacteria (Figure 4C; BA, EC, KP, ST). In competition with these organisms, B. thai ΔT6S displayed only a modest decrease in proliferation relative to the wild-type. Differences in the size and morphology of assay “spots” containing wild-type or ΔT6S were noted in several instances for this group of organisms. Quantification of c.f.u. verified that these differences were reflective of a minor, but highly reproducible fitness defect of ΔT6S (data not shown).
The second group consisted of three γ-proteobacteria: P. putida, P. fluorescens, and S. proteamaculans. The proliferation of B. thai grown in competition with these organisms appeared to be highly dependent on T6S (Figure 4C; PP, PF, SP). For further analyses, we focused on this latter group; henceforth referred to as the “T6S-dependent competitors” (TDCs).
The next question we addressed was whether one or more of the individual T6SSs were responsible for the TDC-specific proliferation phenotype of B. thai ΔT6S. To determine this, we inserted a GFP over-expression cassette into our panel of individual B. thai T6SS deletion strains, and performed plate competition assays against the TDCs. In competition with each TDC, ΔT6SS-1 appeared as deficient in proliferation as ΔT6S, whereas the other strains grew similarly to the wild-type (Figure 5A). The dramatic differences in the competition outcomes between the strains were also discernable by the naked eye. Competition experiments that included B. thai lacking T6SS-1 had a morphology similar to a mono-culture of the TDC, whereas co-cultures possessing an intact T6SS-1 were more similar in appearance to B. thai mono-culture.
It remained possible that the effects of T6SS-1 on the fitness of B. thai in competition with other bacteria were either non-specific or unrelated to its putative role as a T6SS. As mentioned earlier, one common observation from detailed studies of T6SSs conducted to date is that its effects require cell contact [8], [9], [10]. This has been postulated to reflect a conserved mechanism of the apparatus akin to bacteriophage cell puncturing [18]. To address whether the apparent fitness defect of ΔT6SS-1 involves a mechanism consistent with T6S, we probed whether its effects were dependent upon cell contact. A filter (0.2 µm pore diameter) placed between B. thai and TDC cells abrogated the T6SS-1-dependent growth defect (Figure 5B). In control experiments, the three TDCs were directly applied to an underlying layer of the B. thai strains. In each case, a zone of clearing was observed in the ΔT6SS-1 layer, while no effect on wild-type proliferation was noted. From these data we conclude that cell contact is essential for the activity of T6SS-1.
We next sought to quantify the magnitude of T6SS-1 effects on B. thai fitness in competition with TDCs. To ensure the specificity of T6SS-1 inactivation in the strains used in these assays, we generated a B. thai strain bearing an in-frame clpV-1 deletion, and a strain in which this deletion was complemented by clpV-1 expression from a neutral site on the chromosome. In plate competition assays, the ΔclpV-1 strain displayed a fitness defect similar to ΔT6SS-1, and clpV-1 expression complemented the phenotype (Figure 5C). Measurements comparing B. thai and TDC c.f.u. in the competition assay inoculum to material recovered from the assays following several days of incubation confirmed that inactivation of T6SS-1 leads to a dramatic fitness defect of B. thai (Figure 5D). Depending on the TDC, the competitive index (c.i.; final c.f.u. ratio/initial c.f.u ratio) of wild-type B. thai was approximately 120-5,000-fold greater than that of the ΔclpV-1 strain. All TDCs out-competed ΔclpV-1 (0.0021<c.i. <0.015); on the contrary, wild-type B. thai was highly competitive against P. putida (c.i.: 5.8) and P. fluorescens (c.i.: 61), and its relative numbers decreased only modestly in assays with S. proteamaculans (c.i.: 0.24). In summary, our findings indicate that T6SS-1 plays an important role in the interactions of B. thai cells in direct contact with other bacteria. T6SS-1-dependent effects are species-specific, and in some cases, can be a major determinant of B. thai proliferation.
Three models could explain the T6SS-1-dependent effects we observed on B. thai fitness in competition with the TDCs: (i) T6SS-1 inhibits TDC proliferation, thereby freeing nutrients for B. thai; (ii) T6SS-1 prevents TDC inhibition of B. thai growth; or (iii) T6SS-1 performs both of these functions. To distinguish between these possibilities, we compared B. thai and TDC growth rates following inoculation into either mono-culture or competitive cultures on 3% agar plates. Our prior experiments indicated that T6SS-1-dependent effects on B. thai were similar in competition assays with each TDC (Figure 4F and Figure 5), therefore we utilized P. putida to represent the TDCs in this and subsequent experiments. Surprisingly, we found that the proliferation of P. putida and wild-type B. thai was largely unaffected in competition assays (Figure 6A–C). However, ΔclpV-1 proliferation was severely hampered in the presence of P. putida. Indeed, B. thai ΔclpV-1 c.f.u. expanded by only 2.1-fold during the first 23 hours of the experiment, whereas wild-type c.f.u. increased 220-fold. Consistent with earlier results in P. aeruginosa [10], the effects of T6SS-1 on the fitness of B. thai in co-culture with P. putida were not observed in liquid medium (Figure 6D and 6E).
The proliferation defect of B. thai ΔclpV-1 could be attributable to P. putida-induced growth inhibition, cell killing, or a combination of these factors. We reasoned that if killing was involved in the ΔclpV-1 phenotype, the difference in cell death between wild-type and ΔclpV-1 would be most pronounced at approximately 7.5 hours following inoculation of the competition assays, when wild-type B. thai are rapidly proliferating and ΔclpV-1 cell numbers are not expanding. At this time point, we identified similar numbers of dead cells in wild-type and ΔclpV-1 competitions, suggesting that T6SS-1 inhibits stasis of B. thai induced by P. putida (Figure 6F).
In our plate competition assays, low moisture availability impairs bacterial motility, and artificially enforces close association of B. thai with the TDCs. To determine whether T6SS-1 could provide a fitness advantage for B. thai under conditions more relevant to its natural habitat, i.e., where nutrients are exchanged and dehydration does not drive interbacterial adhesion, we conducted mixed species flow chamber biofilm assays.
Previous studies in E. coli and V. parahaemolyticus have implicated T6S in the inherent capacity of these organisms to form biofilms [35], [36]. Furthermore, additional T6SSs are activated during biofilm growth or co-regulated with characterized biofilm factors such as exopolysaccharides [14], [37], [38], [39], [40]. Thus, prior to performing mixed species assays, we first tested whether inactivation of T6SS-1 influenced the formation of monotypic B. thai biofilms. Wild-type and ΔT6SS-1 strains adhered equally to the substratum and formed indistinguishable monotypic biofilms that reached confluency after four days (Figure 7A), indicating T6SS-1 does not play a role in the inherent ability of B. thai to form biofilms.
Next we seeded biofilm chambers with 1∶1 mixtures of B. thai and P. putida. In mixed biofilms, the B. thai strains again adhered with similar efficiency, however a dramatic difference between the capacity of the strains to persist and proliferate in the presence of P. putida became apparent within 24 hours (Figure 7B). At this time point, wild-type B. thai microcolonies had expanded and dispersed throughout the P. putida-dominated biofilm, whereas B. thai ΔclpV-1 microcolonies had diminished in number. Consistent with the results of our plate assays, P. putida growth was not noticeably impacted by the activity of T6SS-1 at early time points in the experiment. As the biofilm matured, wild-type B. thai gradually displaced P. putida, and by four days after seeding, B. thai microcolonies accounted for most of the biofilm volume. These data suggest that T6SS-1 can provide a major fitness advantage for B. thai in interspecies biofilms.
Our findings suggest that the highly conserved T6S architecture can serve diverse functions. We found T6SSs within B. thai critically involved in two very distinct processes – virulence in a murine infection model and growth in the presence of specific bacteria. The systems involved in these diverse phenotypes, T6SS-5 and T6SS-1, respectively, are distantly related, and cluster phylogenetically with other T6SSs of matching cellular specificity. We were unable to define the function for three of the B. thai T6SSs, however their clustering in the H1-T6SS subtree suggests that they could have a role in interbacterial interactions. These systems may not have been active under the assay conditions we utilized, they might be specific for organisms we did not include in our screen, or their activity may not affect proliferation. Phylogenies have proven to be powerful tools for guiding researchers studying complex protein secretion systems [41], [42]. However, determining whether T6S phylogeny holds promise as a general predictor of organismal specificity will require more studies that evaluate the significance of individual systems in both eukaryotic and bacterial cell interactions.
Although B. thai is not generally regarded as a pathogen, our data suggest that Burkholderia T6SS-5 plays a role in host interactions that is conserved between this species and its pathogenic relatives, B. pseudomallei and B. mallei [27], [28], [29], [43]. We postulate that T6SS-5, like many other virulence factors, evolved to target simple eukaryotes in the environment. The benefit T6SS-5 provides the Burkholderia in a mammalian host could have been one factor that allowed B. mallei to transition into an obligate pathogen. Based on our results implicating T6SS-1 exclusively in interbacterial interactions, the role of this system in the lifestyle of B. mallei is more difficult to envisage. Indeed, the cluster encoding T6SS-1 is the most deteriorated of the T6S clusters of B. mallei and is unlikely to function [27]. Of the 13 conserved T6S-associated orthologous genes, 8 of these appear to be deleted in B. mallei T6SS-1, however the remaining T6S clusters of the organism are largely intact (0–3 pseudogenes or absent genes).
Of the 33 organisms screened, the effects of B. thai T6SS-1 were most pronounced in competitions with P. putida, P. fluorescens, and S. proteamaculans. Whether these organisms are physiologically relevant B. thai T6SS-1 targets is not known, however P. putida and P. fluorescens have been isolated from soil in Thailand [44], [45], and the capacity of these organisms to form biofilms is well documented [46], [47], [48]. P. putida and P. fluorescens are recognized biological control agents, suggesting that the rhizosphere could be one habitat where antagonism with B. thai might occur [49]. Notably, we did not observe T6SS-dependent effects on B. thai proliferation in the presence of the five Gram-positive organisms included in our screen. The number and diversity of organisms we tested were too low to ascribe statistical significance to this observation, however it is tempting to speculate that the effects of T6S might be limited to Gram-negative cells. This would not be unexpected given the structural relatedness of T6S apparatus components to the puncturing device of T4 bacteriophage [18], [19], [20].
We found that T6SS-1 allows B. thai to proliferate in the presence of the TDCs. This surprising and counterintuitive finding raises the question of what inhibits B. thai ΔclpV-1 growth, and is it an intrinsic (derived from B. thai) or extrinsic (derived from the TDC) factor? Our data indicate that the activity or production of this factor manifests in the absence of T6SS-1 function only when a TDC is present and intimate cell contact occurs. If the factor is intrinsic, we postulate that its activity is inappropriately triggered by ΔT6SS-1 in the presence of the TDCs, but that its function serves an adaptive role for wild-type B. thai. For example, under circumstances where it is not advantageous for B. thai to proliferate, such as when it is exposed to particular organisms, antibiotics, or stresses, this factor could initiate dormancy. There is evidence that T6S components can participate in cell-cell recognition in bacteria. Gibbs et al. recently reported the discovery of an “identification of self” (ids) gene cluster within Proteus mirabilis that contains genes homologous to hcp (idsA) and vgrG (idsB) [50]. Inactivation of idsB caused a defect in recognition of its parent, resulting in boundary formation between the strains.
If the factor is extrinsic, T6SS-1 might be more appropriately defined as a defensive, rather than an offensive pathway. T6SS-1 could provide defense by either influencing the production of the extrinsic factor within the TDC, such as by repressing expression, or it could provide physical protection against the factor by obstructing or masking its target. If the fitness effect that T6SS-1 provides B. thai depends on a specific offensive pathway present in competing organisms, the presence of this pathway in an organism could be the basis for the apparent specificity we observed in our screen. Future studies must address whether the determinants of T6SS-1 effects are intrinsic, extrinsic, or a combination of the two. The design of our competition screen was limited in this regard; we measured T6SS-1 activity indirectly, and we were able to test only a modest number of species. Understanding the mechanism of action of T6SS-1, for example by identifying its substrates, will provide insight into the specificity of the secretion apparatus.
While it is widely accepted that diffusible factors such as antibiotics, bacteriocins, and quorum sensing molecules are common mediators of dynamics between species of bacteria, an analogous cell contact-dependent pathway has yet to be defined [51]. We found that T6S can provide protection for a bacterium against cell contact-induced growth inhibition caused by other species of bacteria. Given that most organisms that possess T6S gene clusters are either opportunistic pathogens with large environmental reservoirs or strictly environmental organisms, we hypothesize that T6SSs are, in fact, widely utilized in interbacterial interactions. Bacteria-targeting T6SSs may be of great general significance to understanding interactions and competition within bacterial communities in the environment and in polymicrobial infections.
All research involving live animals was conducted in compliance with the Animal Welfare Act and other federal statutes and regulations relating to animals and experiments involving animals, and adhered to the principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council, 1996. All work involving animals was approved by the Institutional Animal Care and Use Committee at the University of Washington.
B. thai E264 and E. coli cloning strains were routinely cultured in Luria-Bertani (LB) broth or on LB agar at 37°C. All bacterial species used in this study are listed in the legend of Figure 4. The medium was supplemented with trimethoprim (200 µg/ml), ampicillin (100 µg/ml), zeocin (2000 µg/ml), irgasan (25 µg/ml) or gentamicin (15 µg/ml) where necessary. For introducing in-frame deletions, B. thai was grown on M9 minimal medium agar plates with 0.4% glucose as a carbon source and 0.1% (w/v) p-chlorophenylalanine for counter-selection [52].
B. thai T6SSs were inactivated utilizing a previously described mutagenesis technique based on the suicide plasmid pJRC115 containing a mutated phenylalanine synthetase (pheS) gene for counter-selection [52]. Unmarked in-frame deletions of three to five T6SS genes per T6SS gene cluster (at least two of which are core T6SS genes; see Figure 1) were constructed by splicing by overlap PCR of flanking DNA [53]. The open reading frames were deleted except for 4–8 codons at the 5′ end of the upstream gene and 3′ end of the downstream gene, and the insertional sequence TTCAGCATGCTTGCGGCTCGAGTT was added as previously described [14]. E. coli SM10 λpir was used to deliver the deletion constructs into B. thai by conjugational mating and transconjugants were selected on LB agar plates supplemented with trimethoprim and irgasan.
The conserved T6SS genes tssK-5 (BTH_II0857) and clpV-1 (BTH_I2958) were deleted using the in-frame deletion mutagenesis technique described above. For single copy complementation, the mini-Tn7 system was utilized [34]. For this, the B. thai ribosomal promoter PS12 sequence was cloned into the suicide vector pUC18T-mini-Tn7T-Tp using complementary oligonucleotides to yield pUC18T-mini-Tn7T-Tp-PS12 [54]. The tssK-5 and clpV-1 open reading frames along with 16–20 bp upstream were amplified and inserted into pUC18T-mini-Tn7T-Tp-PS12. The resulting plasmids and the Tn7 helper plasmid, pTNS3, were introduced into appropriate deletion strains by electroporation using a previously described protocol [52], [54]. Transposition of the Tn7-constructs into the chromosome of B. thai was determined by PCR as described previously [55].
The mini-Tn7 system was utilized to integrate green fluorescent protein (GFP) and cyan fluorescent protein (CFP) expression cassettes into the chromosome of B. thai and P. putida, respectively [55], [56]. To construct a mini-Tn7 derivative for constitutive expression of GFP, the GFP cassette was amplified from pQBI-T7-GFP (Quantum Biotechnologies) without the T7 promoter region as previously described and inserted into KpnI and StuI sites of pUC18T-mini-Tn7T-Tp-PS12 [27]. This plasmid was then introduced into relevant B. thai strains and insertion of Tn7-GFP into the chromosome was verified as described above. To construct a GFP-labeled ΔclpV-1 complemented strain, we made use of the fact that two Tn7 insertion sites (attTn7) are present in the genome of B. thai. The chromosomally integrated Tn7 Tpr resistance cassette of ΔclpV-1 complemented was excised using pFLPe2, which expresses a Flp recombinase, before introducing pUC18T-mini-Tn7T-Tp-PS12-GFP. Insertion of Tn7-GFP into the other attTn7 site was confirmed by PCR as described previously [55], [56]. To engineer CFP labeled P. putida, the mini-Tn7(Gm)-CFP plasmid and the helper plasmid pUX-BF13 were introduced into the strain by electroporation as previously described [56].
Growth kinetics of B. thai strains were measured in LB broth using the automated BioScreen C Microbiology plate reader (Growth Curves) with agitation at 37°C. Three independent measurements were performed in triplicate for each strain.
Swimming motility of B. thai strains was analyzed in 0.25% LB agar. Swimming plates were stab-inoculated with overnight cultures and incubated at 37°C for 48 h. Two independent experiments were performed.
Specific-pathogen-free C57BL/6 mice were obtained from Jackson Laboratories (Bar Harbor, ME). MyD88−/− mice were derived by Dr. Shizuo Akira (University of Osaka) and backcrossed for at least 8 generations to C57BL/6 [57]. Mice were housed in laminar flow cages with ad lib access to sterile food and water. The Institutional Animal Care and Use Committee of the University of Washington approved all experimental procedures. For aerosol infection of mice, bacteria were grown in LB broth at 37°C for 18 hours, isolated by centrifugation, washed twice, and suspended in Dulbecco's PBS to the desired concentration. An optical density at 600 nm (OD600) of 0.20 yielded approximately 1×108 CFU/ml. Mice were exposed to aerosolized bacteria using a nose-only inhalation system (In-Tox Products, Moriarty, NM) [30]. Aerosols were generated from a MiniHEART hi-flo nebulizer (Westmed, Tucson, AZ) driven at 40 psi. Airflow through the system was maintained for 10 minutes at 24 l/min followed by five minutes purge with air. Immediately following aerosolization, the pulmonary bacterial deposition was determined by quantitative culture of left lung tissue from three to four sentinel mice. Following infection, animals were monitored one to three times daily for illness or death. Ill animals meeting defined clinical endpoints were euthanized. At specific time points after infection, mice were euthanized in order to quantify bacterial burdens and inflammatory responses. To determine bacterial loads, the left pulmonary hilum was tied off and the left lung, median hepatic lobe, and spleen each were removed and homogenized in 1 ml sterile Dulbecco's PBS. Serial dilutions were plated on LB agar and colonies were counted after 2–4 days of incubation at 37°C in humid air under 5% CO2.
Overnight cultures of B. thai and competitor bacteria were adjusted to an OD600nm of 0.1 and mixed 5∶1 (v/v). For competitions using fluorescent strains, 2.5 µl of the mixture was spotted on 3% w/v LB agar and fluorescence was measured after approximately one week following incubation at 30°C. For quantitative competitions using non-fluorescent strains, 10 µl of the mixture was spotted on a filter (0.22 µm; GE Water & Process Technologies) and cells were harvested and enumerated at the indicated time points. Colonies of the competing organisms were distinguished from B. thai strains using a combination of colony morphology, growth rate and inherent antibiotic susceptibility.
Growth competitions of B. thai against P. putida were performed on filters as described above. At 7.5 h after initiating the experiment, the filters were resuspended in 200 µl LB broth and cell viability was measured using the LIVE/DEAD BacLight Bacterial Viability Kit for microscopy according to the manufacturer's protocol (Invitrogen). The number of dead cells was determined for five random fields per competition using fluorescence microscopy. Two independent experiments were performed in duplicate.
Biofilms were grown at 25°C in three-channel flow-chambers (channel dimensions of 1×4×40 mm) irrigated with FAB medium supplemented with 0.3 mM glucose. Flow-chamber biofilm systems were assembled and prepared as previously described [58]. The substratum consisted of a 24×50 mm microscope glass cover slip. Overnight cultures of the relevant strains were diluted to a final OD600nm of 0.01 in 0.9% NaCl, and 300 µl of the diluted bacterial cultures, or 1∶1 mixtures, were inoculated by injection into the flow chambers. After inoculation, the flow chambers were allowed to stand inverted without flow for 1 h, after which medium flow was started with flow chambers standing upright. A peristaltic pump (Watson-Marlow 250S) was used to keep the medium flow at a constant velocity of 0.2 mm/s in the flow-chamber channels. Microscopic observation and image acquisition of the biofilms were performed with a Leica TCS-SP5 confocal laser scanning microscope (CLSM) (Leica Microsystems, Germany) equipped with lasers, detectors and filter sets for monitoring GFP and CFP fluorescence. Images were obtained using a 63×/1.4 objective. Image top-down views were generated using the IMARIS software package (Bitplane AG). The flow-chamber experiment reported here was repeated twice, and in each experiment each mono-strain or mixed-strain biofilm was grown in at least two channels, and at least 6 CLSM images were recorded per channel at random positions. Each individual image presented here is therefore representative of at least 24 images.
Annotated genomes were downloaded from the Genome Reviews ftp site (ftp://ftp.ebi.ac.uk/pub/databases/genome_reviews/, January 2010, 926 bacterial genomes (1814 chromosomes and plasmids) [59]. Protein sequences from all genomes were aligned with rpsblast [60] against the COG section of the CDD database (January 2010) [61]. Only proteins showing an alignment covering at least 30% of the COG PSSM with an E-value ≤10−6 were retained. To avoid any errors in COG assignments, we discarded all hits that overlap with another hit with a better E-value on more than 50% of its length. We considered the following 13 COGs as ‘T6SS core components’: COG0542, COG3157, COG3455, COG3501, COG3515, COG3516, COG3517, COG3518, COG3519, COG3520, COG3521, COG3522, COG3523 [3], [4]. Two genes were considered neighbours if they are separated by less than 5000 bp. Only clusters containing the VipA protein (COG3516) and genes coding for at least five other T6SS core components were included in the analyses. The Edwardsiella tarda (EMBL access AY424360) system was added manually because the complete genome sequence and annotation of this organism was unavailable in Genome Reviews.
In three of the 334 T6SS clusters, two VipA coding genes were identified. Manual inspection of two of these clusters in Acinetobacter baumannii (ATCC 17978) and Vibrio cholerae (ATCC 39541) revealed that they resulted from apparent gene fissions; in both cases we kept the longest fragment corresponding to the C-terminal part of the full length protein. In the third case, Psychromonas ingrahamii (strain 37), the two VipA coding genes resulted from an apparent duplication event: one of the two copies showed a high mutation frequency and was discarded. In total, we included 334 VipA orthologs in T6SS clusters. The 334 VipA protein sequences were aligned using muscle [62]. Based on this alignment, a neighbour-joining tree with 100 bootstrap replicates was computed using BioNJ [63].
|
10.1371/journal.pcbi.1000951 | Synchronization of Firing in Cortical Fast-Spiking Interneurons at Gamma Frequencies: A Phase-Resetting Analysis | Fast-spiking (FS) cells in the neocortex are interconnected both by inhibitory chemical synapses and by electrical synapses, or gap-junctions. Synchronized firing of FS neurons is important in the generation of gamma oscillations, at frequencies between 30 and 80 Hz. To understand how these synaptic interactions control synchronization, artificial synaptic conductances were injected in FS cells, and the synaptic phase-resetting function (SPRF), describing how the compound synaptic input perturbs the phase of gamma-frequency spiking as a function of the phase at which it is applied, was measured. GABAergic and gap junctional conductances made distinct contributions to the SPRF, which had a surprisingly simple piecewise linear form, with a sharp midcycle break between phase delay and advance. Analysis of the SPRF showed how the intrinsic biophysical properties of FS neurons and their interconnections allow entrainment of firing over a wide gamma frequency band, whose upper and lower frequency limits are controlled by electrical synapses and GABAergic inhibition respectively.
| Oscillations of the electrical field in the brain at 30–80 Hz (gamma oscillations) reflect coordinated firing of neurons during cognitive, sensory, and motor activity, and are thought to be a key phenomenon in the organization of neural processing in the cortex. Synchronous firing of a particular type of neuron, the inhibitory fast-spiking (FS) cell, imposes the gamma rhythm on other cells in the network. FS cells are highly interconnected by both gap junctions and chemical inhibition. In this study, we probed FS cells with a synthetic conductance stimulus which mimics the electrical effect of these complex connections in a controlled way, and directly measured how the timing of their firing should be affected by nearby FS neighbours. We were able to fit a mathematically simple but accurate model to these measurements, the “synaptic phase-resetting function”, which predicts how FS neurons synchronize at different frequencies, noise levels, and synaptic connection strengths. This model gives us deeper insight into how the FS cells synchronize so effectively at gamma oscillations, and will be a building-block in large-scale simulations of the FS cell network aimed at understanding the onset and stability of patterns of gamma oscillation in the cortex.
| Rhythmic oscillations of concerted electrical activity can occur in the neocortex and hippocampus at gamma frequencies (30–80 Hz), and are thought to be associated with a variety of cognitive tasks including sensory processing, motor control, and feature binding [1], [2]. A striking feature of gamma oscillations is their ability to be generated locally in the neocortex. Local gamma oscillations can be produced by pharmacological [3], [4], electrical [5] or optogenetic [6] stimulation. In vivo, synchronous gamma oscillations may be highly localized or widely distributed, even between hemispheres, with or without phase lags between different areas and layers [1]. It appears, therefore, that local neocortical circuits have an intrinsic capability for generating gamma oscillations, while sensory inputs and connections from other brain regions may shape the complex spatial patterns of oscillatory interaction.
Synchronized firing of cortical inhibitory interneurons has been implicated in the production of these rhythms in many experimental and modeling studies. During spontaneous network activity of the neocortex in vivo, the power of intracellular voltage fluctuations at frequencies higher than 10 Hz is dominated by inhibitory postsynaptic potentials, which are correlated with the extracellular gamma rhythm, and which synchronously inhibit nearby pyramidal cells [7]. A recent study using conductance injection in neocortical pyramidal cells indicated that gamma-frequency-modulation of firing is almost completely determined by their inhibitory input [8]. In the hippocampus and cortex, models of interneuron activity suggest that network oscillations depend on mutually inhibitory synaptic conductances [9], [10], [11].
Fast-spiking (FS) inhibitory interneurons are coupled by electrical synapses in addition to mutual and autaptic inhibitory synapses [12], [13], [14], [15]. Electrical synapses alone [12], [13] or in combination with GABAergic synapses [14] can produce synchronous firing in pairs of these interneurons in vitro. In addition, the biophysical properties of FS neurons appear to be ideally suited to generating gamma rhythms: they have a hard (“type 2”) onset of regular firing at about 30 Hz [16], which means that they can be easily entrained at this frequency. They also show a strong intrinsic drive for spike generation at gamma frequencies when stimulated with broadbrand conductance noise [17]. Recently, selective optical stimulation of FS interneurons, but not of pyramidal neurons, was shown to cause gamma oscillations [6]. Electrical synapses amongst mutually inhibitory interneurons have been found to increase the precision of synchrony in simulation studies [18], [19], [20]. However, the relative roles of chemical inhibition and gap-junctional coupling in shaping synchronous oscillations in the cortex are still unclear.
The theory of synchronization of coupled oscillators uses the concept of phase dynamics to evaluate the stability of the relative phase of coupled oscillators in time [21], [22]. The key to this approach is to determine the effect of a very small perturbing input on the phase of oscillation (“phase resetting”), as a function of the point in the oscillation cycle at which it occurs. This is most often used, under the assumptions of weak coupling and linear summation of phase shifts, to account for how the relative phase of presynaptic and postsynaptic cells evolves from cycle to cycle.
However, as described above, FS cells in the cortex are actually coupled quite strongly to other FS neighbours, with large postsynaptic conductance changes caused by each presynaptic action potential. Here, we have used synthetic conductance injection, or dynamic clamp, to directly measure the phase-resetting response to conductance inputs mimicking the effects of presynaptic action potentials, while systematically varying the relative strengths of electrical and GABAergic inhibitory conductances. The compound synaptic connections between FS neurons, together with the intrinsic spike-generating properties of FS neurons, give rise to a distinctively-shaped phase-resetting relationship, or “synaptic phase-resetting function”, which ensures rapid and precise synchronization over a large gamma-frequency range.
FS cells in rat somatosensory cortical slices were identified by their morphology, action potential shape and characteristic firing pattern in response to depolarizing current injection [12], [13], [23], [24]. FS cells fired high frequency, nonadapting trains of action potentials during depolarizing current steps, occasionally interrupted by pauses with subthreshold oscillations, particularly around threshold [16] (see Methods). We used conductance injection/dynamic clamp [25], [26] to reproduce the effects of electrical and chemical synapses (Fig. 1, see Methods). In FS cells, both gap junctions and GABAergic synapses from neighboring cells are located perisomatically [14], so that point conductance injection at the soma should reasonably reproduce the electrical effects of synaptic inputs. Gap junctions were implemented as a static conductance between the recorded cell and a “voltage-clamped” trajectory of “presynaptic” membrane potential. This “voltage-source” approximation, importantly, allowed us to characterize a functional mapping between the presynaptic spike time and the influence on postsynaptic membrane potential, without considering any reverse effect of gap-junctional current on the presynaptic cell. This is valid as long as the presynaptic cell is considered to be much more strongly controlled by its other inputs, as when it is already part of a synchronous assembly (see Discussion). It is estimated that each FS cell is gap-junction coupled, directly or indirectly, with a measurable coupling, to between 20 and 50 other FS neurons [27], so that if the presynaptic cell is quite strongly-driven by a major proportion of these inputs, then the effect of any one can be neglected. At rest, this gap-junctional input produced a small postsynaptic spikelet (Fig. 1a, left), very similar in size and shape to those observed with natural electrotonic coupling [12], [13]. We also measured coupling coefficients (the ratio of postsynaptic to presynaptic potential change) for gap-junctional type conductance. These were similar to physiological values, and larger for step inputs (0.05–0.22) than for spike inputs (0.01–0.05), owing to low-pass filtering by the combined effects of gap junctional conductance and membrane resistance and capacitance [28].
Many pairs of FS cells are connected by both GABAergic (GABAA, chloride conductance) and electrical synapses [12], [13], [14]. We simulated GABAergic synaptic input using conductance injection (Fig. 1a, middle). The GABA reversal potential (EGABA) was set to −55 mV, based on gramicidin-perforated patch measurements in this cell type [10], [29], considerably more depolarized than in pyramidal neurons [30]. Thus, inhibition is shunting in the range of membrane potentials between spikes during repetitive firing (Fig. 1b). Starting from the resting potential, the “IPSP” is a small depolarisation lasting about 40 ms, again very similar to natural IPSPs in these cells. At the resting potential, a stimulus with both electrical and GABAergic components produces a biphasic depolarizing response (Fig. 1a, right) with the gap-junctional potential visible just before the larger GABAergic potential. Unlike the gap-junctional spikelet, though, the amplitude of the GABAergic potential can change sign in the subthreshold, interspike range of membrane potentials, reversing around EGABA [12].
To determine how this compound synaptic input shifts the timing of periodic firing in an FS cell, we applied conductance inputs during periodic firing elicited by a maintained excitatory stimulus, a step of excitatory conductance reversing at 0 mV. An example response to a compound “synaptic” perturbation is shown in Fig. 1b. In phase-resetting analysis of synchronization, the state of the neuron is characterized by a single quantity, the phase angle, , which – in the absence of any perturbations - increases linearly with time, and which is reset to zero whenever it reaches , corresponding to the occurrence of a spike [21]. The variability of interspike intervals can be represented by adding additional noise, due to stochastic gating of ion channels and other intracellular sources of variability, to the rate of change of . To measure the phase resetting, or shift in the phase, produced by synaptic-like conductance inputs, we applied isolated single inputs during long trains of periodic firing. Fig. 1c shows the relationship between the time tp at which an input (in this case a compound gap/GABA input) is applied, relative to the time of the preceding spike, and the time until the next spike occurs (tn). This clearly deviates from the line of slope −1 (dotted line) expected in the absence of any input, and has two approximately linear regions separated by a sharp transition. Note the characteristic progressive decrease in the variability of this relationship, as tp increases – this is because the earlier the input arrives, the more time is left for integrating the effects of noise before the next spike.
From this relationship, we can estimate the phase at the moment that each input is applied, and the amount of phase resetting produced by the input (see Methods), as shown in Fig. 1d, in which is plotted as a function of . This relationship - the total phase-resetting effect of a synaptic input as a function of the phase at which it arrives – we will refer to as a synaptic phase-resetting function (SPRF), to distinguish it from a classical phase response or phase-resetting curve, which normally describes responses to very small, brief inputs, whose effects can be considered to sum linearly. We examined how the parameters of the synaptic input determine the shape of the SPRF, by varying the magnitude of gap-junctional and GABAergic conductance, applied individually or together (Fig. 2a–f). These components vary physiologically, since FS cells' interconnections can be purely GABAergic (one-way or reciprocal), purely gap-junctional or both [12], [13], [14]. In addition, there is a wide range of electrical synaptic strengths [28].
Purely GABA input produced a phase delay early in the cycle, which increased during the cycle until an abrupt critical point, beyond which it had no effect (Fig. 2a). Introducing a small (250 pS) gap junction, caused a linear region of phase advance (Fig. 2b), as in Fig. 1d, which had an abrupt onset at a phase of about . A sharp transition marks the boundary between this region and the first, phase delay part of the phase cycle. The slope of the phase advance region became more negative, and the boundary between the regions, designated the critical phase , shifted earlier in the cycle, as gap junctional conductance increased (Fig. 2c, d, e). With no GABAergic input, a phase advance region produced by gap junctional input is seen in isolation (Fig. 2f).
Thus, GABAergic input retards, and gap-junctional input advances the phase of firing. For the compound gap/GABA input, the early region of phase delay has a slope determined by the amplitude of inhibition, gi (see Methods), and switches abruptly, midcycle, to a region of decreasing phase advance, whose slope is determined by ge, with no detectable sign of cancellation of the two regions in midcycle. The only clear interaction between the electrical and GABAergic components was that a larger gap junctional conductance shifted to earlier in the cycle.
To quantify the goodness of fit of the piecewise linear SPRF, we performed a chi-square test of 130 phase response curves (in total 6111 data points, 10 cells). For each SPRF, variance of phase was estimated from an unperturbed spike train within the same experiment (median = 0.021 (rad/2π)2. 111 of 130 SPRFs contained no significant difference between the model fit and experimental result (p<0.05). The average reduced chi-square value was 0.80, meaning that the overall fit of the model is extremely good, given the measured degree of variance in the phase. On the whole, the relatively simple piecewise linear model performs remarkably well.
The dependencies of the slopes and breakpoint on the strengths of gi and ge were also fitted by linear relationships (Figure 3). The negative slope of the region of phase delay was proportional to inhibition (, Fig. 3b), the negative slope of the phase advance region was proportional to excitation (, Fig. 3a), while was weakly sensitive to ge (, Fig. 3c). Average values of a and b of this piecewise linear model for the SPRF were a = 0.16/nS (n = 7 cells, 3 cells providing insufficient data for analyzing this dependency), b = 0.69/nS (n = 10 cells). c and d were more variable from cell to cell, and the pooled data in fact showed little overall dependence on ge (not shown). Nevertheless (e.g. Fig. 3c), the weak relationship is clear within individual cells.
Having established that conductances resembling the synaptic input of neighboring FS cells can consistently modify spike timing, we next tested the ability of FS cells to synchronize to, or to be entrained by this input. To visualize the time course of entrainment, we examined responses stroboscopically [22], sampling the phase of the FS cell at the times of periodic stimuli. Figure 4 shows such an experiment. Before the conductance pulses are switched on (open circles), the phase changes in a “sawtooth” pattern, reflecting detuning - the continuously growing phase difference between two oscillators of different frequencies. After the conductance transients begin (Fig. 4, filled circles), the phase quickly converges on a fixed value relative to the stimulus, at about (dashed line), which matched the expected equilibrium phase difference from solving Equation 2 with parameters for this cell. Thus the FS cell becomes phase-locked and frequency-locked to the stimulus train, with spikes occurring around 0.6π before, or equivalently 1.4π after each stimulus. After the end of the stimulation train, the phase reverts to the drifting detuned state.
The piecewise linear SPRF could also account for the frequency band over which synchronization was possible. Fig. 5 shows an experiment in which an FS neuron firing at a steady frequency F was stimulated repeatedly with a periodic synaptic conductance input at frequency f, and an index of the synchrony of the cell with the input (S, varying between 0 and 1, see Methods) was measured over a range of frequencies. As seen in Fig. 5a, this changes from a low level when f is very different from F, to a high value approaching 1, when . Because of the effects of noise in the neuron, there is no absolute phase locking (S<1), and the change in synchrony with input frequency does not have abrupt boundaries, but falls away continuously as the difference between f and F grows. It is clear that the central region of high synchrony lies below the unperturbed or natural firing frequency F when only inhibition is applied (Fig. 5b), above F when only gap-junctional conductance is applied (Fig. 5c), or both above and below F when a compound input is applied (Fig 5a). This observation was duplicated by the piecewise linear model of the SPRF, analysis of which (see Methods) predicted the 1∶1 synchronized frequency bands shown in gray, for the deterministic (noise-free) case – in this neuron, these boundaries corresponds to a synchrony of about 0.7. The synchronized frequency band is much narrower for either gap-junctional stimulation alone (Fig. 5b) or GABAergic inhibition alone (Fig. 5b). Iterations of the noisy stroboscopic map derived from the fitted SPRF (Eq. 2) showed that it could also reproduce the distribution of S adequately (black curves in Fig. 5a–c). Thus the piecewise linear model of the SPRF appears to account very well, both for the frequency range and degree of synchronization in noise.
We next used the SPRF to predict the frequency ranges of entrainment for different strengths of inhibition and electrical coupling (Fig. 6), by analyzing the bifurcations at the onset of synchrony in the stroboscopic map of the phase, i.e. the map of the phase of the postsynaptic cell at successive presynaptic spike times in a regular train (see Methods, equation 2). For the deterministic (zero noise) case, 1∶1 entrainment corresponded to a stable fixed point of the map, labelled in the example shown in Fig. 6a. As the amount of detuning (difference between f and F) varies, the map shifts vertically, so that at certain stimulus frequencies, the fixed point disappears (at a “corner-collision” bifurcation [31]). Thus, it is possible to plot the regions in which there is synchronization in the plane (Fig 6b) or the plane (Fig. 6c,d). These form Arnol'd tongues [22] in which the frequency range of entrainment shrinks as the synaptic strength is reduced.
This analysis shows a number of effects which are relevant to the physiological function of FS neurons. Increasing strongly increased the upper frequency limit of entrainment and weakly increased the lower limit (Figs. 6b). When it is impossible to entrain firing with f<F. Conversely, with , it is impossible to entrain for f>F, and increasing strongly reduces the lower frequency limit of entrainment (Fig. 6c,d).
Since physiologically, entrainment must occur in the face of considerable noise, we also investigated the effect of adding noise to the phase map. It is possible to define stochastic bifurcation points of the map F, at which there is a qualitative change in the nature of the stochastic dynamics. These points coincide with the deterministic bifurcation frequencies [32] for (see Methods for details). We examined the frequency extents of this kind of stochastic entrainment at different noise levels (Fig. 6b–d). In all cases, increasing the noise in the phase shrinks the region of entrainment. For rad/2π, which was a typical noise level in these cells in vitro, the area of stochastic entrainment shrank to a third or less of the noise-free case. This noise-induced distortion is not symmetrical in the frequency axis. For example, Fig. 6d shows that in the absence of electrical coupling, the lower frequency limit of entrainment was highly susceptible to noise while the upper limit was not. The greater the level of electrical coupling (), the more the upper limit was reduced by noise.
The SPRF makes several predictions. First, FS cells receiving purely electrical synaptic input will synchronize effectively when driven at frequencies higher than F. Higher frequencies can be followed with stronger electrical input. Second, cells will synchronize to purely inhibitory input at frequencies lower than F, and stronger inhibition allows lower frequencies to be followed. Third, combined electrical and inhibitory input allows cells to synchronize to frequencies both above and below their unperturbed frequency. Although noise diminishes the frequency band of synchronization, sometimes asymmetrically, these conclusions remain valid in the presence of noise. For typical strengths of combined electrical-inhibitory synaptic connections, 20 Hz or greater bandwidths of stochastic synchronization persist even in quite high levels of noise (σ = 0.1).
A number of previous theoretical and experimental studies have examined the phase-resetting properties of cortical neurons. Ermentrout and Kopell developed a theoretical approach to calculate what they termed the “synaptic interaction function” based on phase response curves and the assumption of weak coupling [33]. Reyes and Fetz (1993) stimulated synaptic inputs to regularly-firing pyramidal neurons to measure the phase resetting produced by EPSPs [34], while Stoop et al. (2000) used similar measurements to predict input frequency regions for entrainment and chaos [35]. Netoff et al. used dynamic-clamp to measure phase-resetting (or spike-time response curves) by artificial excitatory or inhibitory conductances in excitatory stellate cells of medial entorhinal cortex, and oriens-lacunosum-molecular interneurons in the CA1 region of hippocampus [36], and were able to demonstrate synchronization in pairs of neurons connected by artificial conductances mimicking synaptic connections, or between biological neurons and simulated neurons. In fast-spiking inhibitory cells, Mancilla et al. (2007) measured phase-resetting relationships for small current pulses (weak coupling) and showed that they could account quite well for synchronization of pairs of gap-junction coupled FS cells, both experimentally and in a biophysical model of FS neurons [37]. In this paper, we go further, by using conductance injection (dynamic clamp) to reproduce the combined effect of gap-junctional and strong synaptic connections, and using this to predict the resulting synchronized frequency bands, and their dependence on synaptic strength, including the effect of noise in the synaptic phase-resetting function on synchronization.
The conductance pulses which we have used are based on the physiological properties of the synaptic connections between FS neurons. In FS neurons of a basket morphology, APs initiate in the axon [38] arising usually from a proximal dendrite, [39] and receive many of their inhibitory connections and gap junctions from other fast-spiking interneurons perisomatically [14]. Thus, dynamic clamp recordings at the soma should provide a reasonably realistic simulation of the natural gap-junctional and fast inhibitory input.
In order to carry out this analysis, we have made the approximation that, between spikes, the presynaptic voltage of the gap-junctional input was held at a resting potential of −70 mV, . In other words, we have focused on the effect of gap-junctional current flow associated with the discrete event of the presynaptic spike. This approach does not take account of the way in which presynaptic membrane potential would gradually depolarize between spikes, if firing periodically. We have also ignored the two-way nature of coupling between cell pairs. In other words we model entrainment of one cell by another, rather than synchronization of a symmetrical coupled pair. Although both electrical and inhibitory coupling can often be asymmetrical [13], [40], they may also be quite symmetrical. However, the entrainment studied here models the situation where the presynaptic cell is already imperturbably-driven as part of a strong synchronously-firing assembly of FS neurons, so that the phase and frequency of its firing will be clamped to that of its predominant input. Thus, the SPRF that we measure should be an effective model for describing recruitment of new cells to such a synchronous assembly.
It is expected that the preferred firing frequency F of the postsynaptic cell may also affect the form of the SPRF, since the timing of intrinsic ion channel kinetics will shift relative to phase as the cycle length changes. In a few experiments where we were able to address this issue, we indeed found evidence of a change in the parameters of the SPRF model. a, the dependence of phase delay on gi, increased quite strongly as firing frequency increased, and shifted earlier in the cycle as firing frequency increased. The dependence of b and d on firing frequency was not marked. The relatively strong effect on a may partly reflect the long duration of the IPSP conductance relative to the period of the cycle.
The synaptic phase-resetting function, or SPRF, for compound input was distinguished by the following features: an extremely abrupt midcycle switch from phase delay to phase advance, which shifted weakly towards the early part of the cycle as the strength of electrical coupling was increased; amplification of the phase delay region by increasing inhibition; and amplification of the phase advance region by increasing gap-junctional coupling. We found that these qualitative features were also present in a biophysical model of firing in fast-spiking cells [41] (see Methods), incorporating voltage-gated sodium, Kv1.3 and Kv3.1/3.2 potassium channels, and stimulated with exactly the same inputs as used experimentally (Fig. 7). In this fully-deterministic model, we also observed a very fine local structure of fluctuations around the main relationship, particularly in the phase delay. Despite these qualitative similarities between the model and experimental results, there were also major differences. In experiments, phase advance was produced exclusively by gap-junctional conductance and phase delay exclusively by inhibition, while in the model, gap-junctional input did affect phase delay strongly early in the cycle – this was never observed experimentally. This deficiency of the biophysical model suggests that additional conductances expressed in FS neurons somehow help to confer a complete immunity to gap-junctional stimulation in the early, phase-delay part of the cycle. We surmise that the voltage-gated potassium conductance in this part of the cycle may actually be much higher than in the model, and that this may allow phase delay and advance to be regulated completely independently. Also, because of their relative timing, the effect of inhibition will outlast that of the gap-junctional current transient – thus phase delays caused by inhibition starting early in the cycle may in fact be caused more by their persistence until later in the cycle. In addition, the model shows a pronounced curvature in the phase delay region of the SPRF which was not noticeable in any experimental recordings. This might reflect the presence of other voltage-dependent conductances in real FS cells which effectively linearize this part of the relationship.
The sharp discontinuity between phase delay and advance which emerges at high synaptic strengths is a result of the particular intrinsic biophysical properties and the nature of the synaptic perturbation. It appears to be related to the “class 2” nature of the FS neuron threshold [16], and may be sensitively determined by the potassium conductance densities and kinetics [42], [43]. It was not observed for example in a class 1 excitable Morris-Lecar model. The discontinuity is a critical decision point, or threshold, in the progression of the membrane potential towards spike initiation, at which hyperpolarization and depolarization both exert their maximal influence. The effect of this shape of SPRF is to ensure very rapid synchronization of the cell. Maximal phase shift occurs in the middle of the cycle when the phase difference is high - the postsynaptic cell either advances or delays its phase to achieve nearly immediate in-phase firing when detuning between pre- and postsynaptic cell is small. This extremely sharp midcycle transition is not observed in conventional phase-resetting relationships to weak brief inputs in these cells [37], [44], and is a consequence of the integration of the strong compound input.
The piecewise nature of the SPRF, with the phase advance contributed exclusively by gap-junctional input, and the delay component contributed exclusively by chemical inhibition, mean that these two types of connection have complementary roles in synchronization: gap junctions are necessary to entrain the firing of the postsynaptic cell to a frequency higher than its preferred frequency, while inhibitory synapses are necessary to entrain firing to a frequency lower than the preferred frequency (as seen in Figures 5 and 6). This can be seen as follows. Let H be the phase difference between postsynaptic and presynaptic cells (). The change in H over one period of the input, i.e. from input i to input i+1, is: . Therefore, when entrainment is achieved, , and so if F>f, then , and if F<f, then .
Using the SPRF to model entrainment assumes that the effect of each stimulus in the train is the same as if it was applied in isolation. The success of the SPRF in predicting entrainment shown here demonstrates that it is at least a good approximation for this purpose, and that the arithmetic of adding effects of multiple sequential synaptic inputs behaves reasonably linearly. The SPRF assumes that the entire dynamical state of the neuron may be represented by just a single number at any time, the phase, which would imply that its dynamical state always lies on a limit cycle, along which it is kicked instantaneously forwards and backwards by the synaptic inputs. The complex dynamics of a real neuron containing a large number of different voltage-dependent conductances distributed in a complex morphology, and the strong and non-instantaneous nature of the perturbation mean that this is a considerable simplification of the reality. An indication of whether the phase approximation is reasonably valid, is to test whether there is any higher-order phase resetting, i.e. changes in the interspike interval following that during which the input is applied, or in subsequent intervals. When we analysed second order shifts, we found that they were sometimes detectable, but very small in relation to the first-order SPRF (See Figure S1), in line with the short memory of FS cells for input conductance fluctuations [17].
FS cell firing is suspected to be directly and primarily responsible for producing gamma oscillations in the neocortex [6], [7], [8]. Different fine-scale subnetworks of mutually-exciting pyramidal cells in layers 2 or 3, which are driven by specific subsets of local layer 4 inputs, appear to interact with other such subnetworks via the inhibitory interneuron network [45]. Synchronization of FS cells, therefore, may be essential for linking responses of pyramidal cells very rapidly to specific features of the synaptic input, as hypothesized to occur in sensory “binding” [2]. We have shown that the effect of conductance inputs which realistically mimic single synaptic connections on the phase of FS firing is very powerful, and is capable of entraining the postsynaptic cell even against strong noise. The strikingly sharp discontinuity between phase delay and advance in the SPRF causes a very rapid jump to nearly in-phase firing.
The relative strengths of electrical and inhibitory components can vary greatly from connection to connection [12], [13], and some pairs of FS cells connected by gap junctions can synchronize their firing, while others cannot [14]. The strengths of these components will also vary dynamically. Electrical synapses can exhibit plasticity through G protein-coupled receptor activation, intracellular calcium and phosphorylation [46], and the GABAergic connections show strong short-term depression [12], [13], [14]. These effects presumably help to shape the spatiotemporal dynamics of synchronous firing. The model that we introduce here could easily accommodate independent plasticity rules for inhibition and gap junctions, by additional rules for modifying the slopes of the corresponding regions of the SPRF. In addition to such modulation, the GABAA receptor is also the target of many important neuroactive drugs, such as benzodiazepines, barbiturates and ethanol. These will be expected to influence the shape of the SPRF, and the synchronization behavior of FS cells in the gamma frequency range. The SPRF, therefore, may be a useful tool for characterizing the action of such compounds on pathological network states treated by such drugs.
Firing is considerably more variable in vivo than in vitro [47], and it is important to consider the consequences of the SPRF in strong noise. The stochastic bifurcation analysis that we carried out (Fig. 6) delineated a well-defined boundary between entraining and non-entraining frequencies, based on a qualitative change in the nature of the motion of the phase [32] (see Methods). The stronger the noise, the smaller the frequency region of stochastic entrainment – in line with intuition, noise acts to break down synchronization. The strength of the noise effect in controlling the boundary of the synchronized region is not symmetrical around F – thus noise can effectively shift, as well as shrink the synchronized frequency band.
In conclusion, the synaptic phase-resetting function of FS cells firing at gamma frequencies, as characterized here, is very well-suited to achieving rapid synchronization, and demonstrates complementary roles of the two types of synaptic connection in determining the frequency range of synchronization. It provides a simple yet surprisingly accurate model for predicting synchronization of these cells, and should be a useful component in network models aimed at understanding the complex spatiotemporal properties of locally-synchronized gamma-frequency firing in the cortex.
300 µm sagittal slices of somatosensory cortex were prepared from postnatal day 13–19 Wistar rats, using a vibratome (DSK Microslicer Zero 1, Dosaka EM, Kyoto), in chilled solution composed of (in mM): 125 NaCl, 25 NaHCO3, 2.5 KCl, 1.25 NaH2PO4, 2 CaCl2, 1 MgCl2, and 25 glucose, oxygenated with 95% O2, 5% CO2 gas. Slices were then held at room temperature for at least 30 minutes before recording. The tissue was visualized with an Olympus BX50WI upright microscope (Olympus UK, London) using infrared differential interference contrast videomicroscopy. During recording, slices were perfused with oxygenated solution identical to the slicing solution, at 31–35°C (8 cells analysed in detail) or 23°C (4 cells). 10 µM 2-(3-carboxypropyl)-3-amino-6-(4-methoxyphenyl)-pyridazinium bromide (SR95531; gabazine), 10 µM D-2-amino-5-phosphonopentanoic acid (AP5), and 10 µM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) were usually added, to block chemical synaptic transmission mediated by GABAA, N-methyl-D-aspartic acid (NMDA),and α-amino-3-hydroxy-5-methyl-4-isoxazole proprionic acid (AMPA) receptors, respectively. Whole-cell recordings were made from the somas of nonpyramidal neurons in cortical layers 2/3, 4, and 5. Cells identified as FS neurons had a mean input resistance of 202±87 MΩ (n = 12). Data from 10 fast-spiking neurons (taken from 8 animals) were used for analysis, with a further 12 cells showing consistent results, but which were not complete enough for analysis. The number of synaptic phase-resetting functions with different parameters of the conductance perturbations (see below) which could be constructed for each cell was limited by the lifetime of the recording, typically 20 to 40 minutes.
Patch pipettes of 3–5 MΩ resistance were pulled from borosilicate capillary glass and filled with an intracellular solution containing (in mM): 105 K-gluconate, 30 KCl, 10 HEPES, 10 phosphocreatine, 4 ATP, 4 MgCl2, and 0.3 GTP, adjusted to pH 7.3 with KOH. Current-clamp recordings were performed using an Axon Multiclamp 700A or in a few cases, an Axopatch 200A amplifier (Axon Instruments, Foster City, CA). Membrane potentials were corrected for nulling of the liquid junction potential before seal formation. Signals were filtered with a four-pole low-pass Bessel filter at −3dB cutoff frequency of 5 kHz, sampled at 20 kHz, and recorded with custom software written in MATLAB (The Mathworks, Natick, MA).
Recorded neurons were stimulated using artificial conductance injection [25], [26], [48]. An effective conductance is inserted in the recorded cell by injecting a current I according to Ohm's law, I = g(V−Erev), where g is the conductance, V is the membrane potential of the cell, and Erev is the reversal potential of the conductance. A conductance injection amplifier [49] or digital signal processing system (SM-1 or SM-2, Cambridge Conductance, Cambridge, UK) [50] with response times of less than 200 ns or 10 µs respectively, were used to calculate and produce the current command signal in real time for the current-clamp amplifier.
Steady trains of action potentials at gamma frequencies were elicited by steps of AMPA-receptor like ohmic conductance, reversing at 0 mV, to which perturbing conductances were added as follows. Stimuli that mimicked action potentials filtered through electrical synapses were generated. An action potential (AP) waveform was produced using a conductance-based model of an FS cell, identical to that of [41], except that the leak conductance was reduced to better fit the stimulus-response curves of actual FS cells (see Fast-spiking cell conductance-based model (section below)
This AP waveform was then used as the time-varying Erev signal for a constant conductance ge, representing the electrical synapse. The conductance of a unitary synaptic GABA event was modelled as a difference of exponentials , where is the scaling amplitude of the inhibitory conductance, and was 7 ms, and was 0.5 ms. In compound stimuli, the start of the GABA event was delayed by 3 ms from the start of the simulated action potential to represent synaptic latency. The reversal potential EGABA was usually set to −55 mV [29].
Spike times were determined as the times of positive-going threshold crossings of the membrane potential at a threshold set at 10 mV below the peak of action potentials. The phase at which a stimulus was applied was calculated from the time elapsed from the preceding spike, relative to the unperturbed firing period. Variability of phase was characterized by the phase order parameter, or synchrony , which varied between 0 (phases distributed uniformly between 0 and ) and 1 (phases all identical). The change in phase () caused by a stimulus was calculated as follows. Let be the phase reached at the moment of perturbation, the phase immediately after, the time after the previous spike at which the perturbation is applied, the time elapsed after the perturbation before the next spike, and the average interspike interval. Then , and .
The synaptic phase-resetting function (SPRF, see Fig. 2) was approximated by the piecewise linear relationship:(1)where conductance values are in nS, -α is the slope in the phase advance section, -β is the slope of the phase delay section, and is the breakpoint. SPRFs were fitted to experiments by least-squares, and using Grubbs' test for outliers, to delete occasional outlying points (in most cases none, but no more than three per SPRF).
Entrainment of periodic spiking to periodic stimulation was simulated by the noisy map describing the evolution of the phase from stimulus n to stimulus n+1:(2)where f is the stimulus angular frequency, F is the unperturbed (natural) angular frequency of the cell, and is a Gaussian-distributed noise term, with variance . The biophysical simulations of Fig. 7 were carried out using the model specified by [41], modified slightly as described above (see Conductance injection).
Bifurcation points, where 1∶1 entrained fixed points of the map given by Eq. 2 appear, were solved for directly. To determine the points of stochastic bifurcation, we used the definition of [32]. The stochastic map of the phase between successive stimuli on a unit circle S is represented by a Markov operator p on the phase distribution, where is the conditional probability density function of the phase at stimulus i+1, given a phase of at stimulus i.and the distribution of phase advances from stimulus n to stimulus n+1 according to:p is approximated by a stochastic transition matrix, and the onset of stochastic entrainment is defined by the point where the second eigenvalue of this stochastic transition matrix changes from real to complex. This definition of a stochastic bifurcation coincides with the deterministic case as the noise level approaches zero, is clearly defined even when the steady-state phase distribution hardly changes, and incorporates the dynamics of the phase: the first eigenfunction gives the stationary or invariant distribution of the phase, while the second eigenfunction can be thought of as forming the principal component of the average time course of relaxations from an initial random phase distribution.
A model of fast-spiking cell membrane potential (V) dynamics was used (as above for generating action potentials for gap-junctional stimulation) which was slightly modified, with a different leak conductance, from that specified in Erisir et al., 1999 [41] (also correcting typographical errors in the published description of the model). Sodium (Na), Kv1 (K1) and Kv3 type potassium and static leak (L) conductances were used in a single electrical compartment of capacitance C, as follows (units of mV for voltage, ms−1 for rates):Exactly the same conductance stimuli were applied to the model as to cells experimentally (see Conductance injection section above).
|
10.1371/journal.pmed.1002277 | Risk prediction models for selection of lung cancer screening candidates: A retrospective validation study | Selection of candidates for lung cancer screening based on individual risk has been proposed as an alternative to criteria based on age and cumulative smoking exposure (pack-years). Nine previously established risk models were assessed for their ability to identify those most likely to develop or die from lung cancer. All models considered age and various aspects of smoking exposure (smoking status, smoking duration, cigarettes per day, pack-years smoked, time since smoking cessation) as risk predictors. In addition, some models considered factors such as gender, race, ethnicity, education, body mass index, chronic obstructive pulmonary disease, emphysema, personal history of cancer, personal history of pneumonia, and family history of lung cancer.
Retrospective analyses were performed on 53,452 National Lung Screening Trial (NLST) participants (1,925 lung cancer cases and 884 lung cancer deaths) and 80,672 Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO) ever-smoking participants (1,463 lung cancer cases and 915 lung cancer deaths). Six-year lung cancer incidence and mortality risk predictions were assessed for (1) calibration (graphically) by comparing the agreement between the predicted and the observed risks, (2) discrimination (area under the receiver operating characteristic curve [AUC]) between individuals with and without lung cancer (death), and (3) clinical usefulness (net benefit in decision curve analysis) by identifying risk thresholds at which applying risk-based eligibility would improve lung cancer screening efficacy. To further assess performance, risk model sensitivities and specificities in the PLCO were compared to those based on the NLST eligibility criteria. Calibration was satisfactory, but discrimination ranged widely (AUCs from 0.61 to 0.81). The models outperformed the NLST eligibility criteria over a substantial range of risk thresholds in decision curve analysis, with a higher sensitivity for all models and a slightly higher specificity for some models. The PLCOm2012, Bach, and Two-Stage Clonal Expansion incidence models had the best overall performance, with AUCs >0.68 in the NLST and >0.77 in the PLCO. These three models had the highest sensitivity and specificity for predicting 6-y lung cancer incidence in the PLCO chest radiography arm, with sensitivities >79.8% and specificities >62.3%. In contrast, the NLST eligibility criteria yielded a sensitivity of 71.4% and a specificity of 62.2%. Limitations of this study include the lack of identification of optimal risk thresholds, as this requires additional information on the long-term benefits (e.g., life-years gained and mortality reduction) and harms (e.g., overdiagnosis) of risk-based screening strategies using these models. In addition, information on some predictor variables included in the risk prediction models was not available.
Selection of individuals for lung cancer screening using individual risk is superior to selection criteria based on age and pack-years alone. The benefits, harms, and feasibility of implementing lung cancer screening policies based on risk prediction models should be assessed and compared with those of current recommendations.
| In the United States, lung cancer screening is currently recommended based on age, pack-years smoked, and years since smoking cessation, the criteria used to select participants for the National Lung Screening Trial (NLST).
A number of recent investigations suggest that using lung cancer risk prediction models could lead to more effective screening programs compared to the current recommendations.
External validation and direct comparisons between risk models are often limited due to insufficient numbers of events or methodological limitations.
Various performance characteristics of nine risk prediction models for lung cancer incidence or mortality were assessed using data from two randomized controlled trials on lung cancer screening: the NLST and the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO).
The calibration performance of the models was satisfactory, but discrimination ranged widely between models. However, all models had a higher sensitivity—and some models had a slightly higher specificity—than the NLST eligibility criteria.
Using risk prediction models to select individuals for lung cancer screening is superior to currently recommended selection criteria.
The benefits, harms, and feasibility of using risk prediction models to select individuals for lung cancer screening should be assessed and compared with current recommendations.
| The National Lung Screening Trial (NLST) found that screening with low-dose computed tomography (CT) can reduce lung cancer mortality by 20% [1]. Based on an evidence review, including the results of the NLST and a comparative microsimulation modeling study, the United States Preventive Services Task Force (USPSTF) recommended lung cancer screening for current and former smokers aged 55 through 80 y who smoked at least 30 pack-years and, if quit, quit less than 15 y ago [2–4]. To our knowledge, only the United States has implemented lung cancer screening policies. Although the province of Ontario, Canada, recommends screening individuals at high risk for lung cancer through an organized program, no program has yet been established [5]. Cancer Care Ontario (the provincial cancer agency of Ontario) is currently evaluating the feasibility of implementing such a program [6]. European countries have not yet made any recommendations on lung cancer screening, as the final results of the Dutch-Belgian Lung Cancer Screening Trial (Nederlands-Leuvens Longkanker Screenings Onderzoek [NELSON] trial), potentially pooled with high-quality data from other trials, are still awaited [7–9].
The screening eligibility criteria used in the current USPSTF recommendations are based on age and pack-years, a measure of cumulative smoking exposure. Thus, these recommendations do not take other important risk factors into account, such as family history, nor other relevant aspects of smoking, such as smoking duration or intensity. Recently, a number of investigations have suggested that determining screening eligibility using an individual’s risk based on age, more detailed smoking history, and other risk factors such as ethnicity and family history of lung cancer could lead to more effective screening programs compared with the USPSTF recommendations [10–13]. Indeed, some lung cancer screening guidelines already encourage assessment of an individual’s risk to determine screening eligibility [14].
While various lung cancer risk prediction models have been developed, external validation and direct comparisons between models have been limited due to insufficient numbers of events or methodological limitations [15–21]. Such validations are essential, as risk prediction models generally have optimistic performance within their development dataset [15–17]. This study aims to externally validate and directly compare the performance of nine currently available lung cancer risk prediction models for stratifying lung cancer risk groups and determining screening eligibility.
No identifiable information was used; therefore, no institutional review board (IRB) approval was needed. Nonetheless, a determination of exempt was given by the University of Michigan IRB (HUM00054750), and a determination of this not being human subjects research was given by the Fred Hutchinson Cancer Research Center (former affiliation of J. J.) IRB (6007–680).
We used data from two large randomized controlled screening trials: the NLST and the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO) [1,22–24]. All participants in the CT arm (n = 26,722) and chest radiography (CXR) arm (n = 26,730) of the NLST and ever-smoking participants in the CXR arm (n = 40,600) and control arm (n = 40,072) of the PLCO were included in the analysis. Never-smokers in the PLCO were not considered, as (1) not all lung cancer risk prediction models can be applied to never-smokers and (2) never-smokers are unlikely to reach levels of risk that allow them to benefit from screening [13,25].
Data on the predictor variables in each trial were collected through epidemiologic questionnaires administered at study entry and harmonized across both trials. Reported average numbers of cigarettes smoked per day above 100 were considered implausible and recoded as 100 cigarettes per day (n = 11). Furthermore, body mass index values less than 14 and over 60 kg/m2 were considered implausible for enrollment in both trials and recoded as 14 (n = 5) and 60 kg/m2 (n = 18), respectively. Lung cancer diagnoses (1,925 in the NLST and 1,463 in the PLCO) and lung cancer deaths (884 in the NLST and 915 in the PLCO) that occurred between study entry and 6 y of follow-up were included in the final dataset and were considered as binary outcomes.
Our study includes nine risk prediction models for lung cancer incidence or death that have been used frequently in the literature. Risk prediction models were not considered for this investigation, if they (1) were developed for specific ethnicities and are therefore not broadly applicable [26–28], (2) used information on biomarkers or lung nodules and are therefore not readily applicable for the prescreening selection of individuals [29–33], (3) were developed for identifying symptomatic patients [34,35], (4) did not incorporate smoking behavior [36], (5) did not provide information on parameter estimates (e.g., baseline risk parameters) necessary to allow replication of the model [11,12], or (6) had poor discriminative ability in their development dataset [37].
Nine models remained and were investigated: the Bach model, the Liverpool Lung Project (LLP) model, the PLCOm2012 model, the Two-Stage Clonal Expansion (TSCE) model for lung cancer incidence, the Knoke model, two versions of the TSCE model for lung cancer death [10,38–44], and simplified versions of the PLCOm2012 and LLP models. The characteristics of these models are shown in Table 1. The TSCE and Knoke models consider only age, gender, and smoking-related characteristics as risk factors [40–43]. The Bach model considers asbestos exposure as an additional risk factor, while the LLP and PLCOm2012 models consider multiple additional risk factors [10,38,39]. The simplified versions of the PLCOm2012 and LLP models considered only age, gender, and smoking variables. A detailed description of each model can be found in S1 Appendix.
Data on frequency and intensity of asbestos exposure, used in the LLP and Bach models, was not available for the PLCO participants and could not be accurately derived for the NLST participants [38,39]. Therefore, we assumed that none of the participants were exposed to asbestos, even though this assumption may lead to biased estimates [45]. However, as the potential number of individuals with asbestos exposure was low (less than 5% of the NLST participants reported ever working with asbestos), this bias is expected to be minor [46].
The LLP model incorporates age at lung cancer diagnosis of a first-degree relative: early age (60 y or younger) versus late age (older than 60 y) [38]. However, while both the PLCO and the NLST had information about the occurrence of family history of lung cancer (yes/no), neither had information on the age of diagnosis for the affected relative(s). Since the median age of lung cancer diagnosis in the United States is 70 y and the majority of lung cancers occur after the age of 65 y (68.6%), we assumed that lung cancer in first-degree relatives in the PLCO and the NLST always occurred after the age of 60 y [47,48].
In addition, the LLP model incorporates a history of pneumonia as a risk factor [38]. While information on this risk factor was available in the NLST, it was not available in the PLCO. Therefore, we assumed that none of the PLCO participants had a history of pneumonia for the complete case analyses. While 22.1% of NLST participants had a history of pneumonia (Table 2), the association of a history of pneumonia with a lung cancer diagnosis within 6 y was not clear (p = 0.3378 in the CT arm and p = 0.0035 in the CXR arm). Missing history of pneumonia for PLCO participants was imputed by using information from the NLST participants [49].
To assess the performance of the risk prediction models, several metrics were employed: calibration, discrimination, and clinical usefulness (net benefit over a range of risk thresholds) [50]. The performance of the investigated risk prediction models was assessed in each trial arm separately, for both lung cancer incidence and lung cancer mortality. We assessed both lung cancer incidence and mortality in both arms of both trials for all investigated risk models, as these outcomes may be influenced differently by screening. Screening may affect the predictive performance for lung cancer incidence, due to the advance in time of detection due to screening (lead time) and the detection of cancers that would never have been detected if screening had not occurred (overdiagnosis) [51–53]. Furthermore, CT screening reduces lung cancer mortality compared to CXR screening, which may influence the predictive performance of models for lung cancer mortality in the CT arm of the NLST [1]. Furthermore, the sensitivity and specificity of each model in the PLCO cohorts were compared to the sensitivity and specificity of the NLST/USPSTF smoking eligibility criteria (being a current or former smoker who smoked at least 30 pack-years and, if quit, quit less than 15 y ago). Model performance was assessed by varying follow-up duration and outcome (5- and 6-y lung cancer incidence or mortality) to investigate the effect of follow-up duration on the discrimination performance of each model [54]. The 5- and 6-y time frames were chosen because the LLP and PLCOm2012 models were calibrated to these respective time frames, and complete follow-up of NLST participants was limited to 6 y [10,38]. Since performance was similar for 5- and 6-y outcomes, only the results of the 6-y outcomes are presented. Performance was evaluated for the risk prediction models as presented in their original publication, without any recalibration or reparameterization to the NLST and the PLCO. The only exception is the PLCOm2012 model, which was originally developed based on data from the control arm of the PLCO [10]. All analyses were performed in R (version 3.3.0) [55].
Calibration plots were constructed for the observed proportions of outcome events against the predicted risks for individuals grouped by similar ranges of predicted risk [56]. Perfect predictions should show an ideal 45-degree line that can be described by an intercept of 0 and a slope of 1 in the calibration plot [57]. The calibration intercept quantifies the extent to which a model systematically under- or overestimates a person’s risk; an intercept value of 0 represents perfect calibration in the large. The calibration slope was estimated by logistic regression analysis, using the log odds of the predictions for the single predictor of the binary outcome [50]. For a (near-)perfect calibration in the large, a calibration slope less than 1 reflects that predictions for individuals with low risk are too low and predictions for individuals with high risk are too high [50]. The calibration plots, calibration in the large, and calibration slopes for each model were obtained using the R package rms [58].
Discrimination reflects the capability of a model to distinguish individuals with the event from those without the event; the risk predicted by the model should be higher for individuals with the event compared with those without the event [59]. The area under the receiver operating characteristic curve (AUC) was used to assess discrimination, which ranges between 0.5 and 1.0 for sensible models. The AUCs for each model were obtained using the R package rms [58].
While discrimination and calibration are important statistical properties of a risk prediction model, they do not assess its clinical usefulness [50,54,59]. For example, if a false-negative result causes greater harm than a false-positive result, one would prefer a model with a higher sensitivity over a model that has a greater specificity but a slightly lower sensitivity, even though the latter might have a higher AUC [60].
In the context of selecting individuals for lung cancer screening, a model is clinically useful if applying that model to determine screening eligibility yields a better ratio of benefits to harms than not applying it. Decision curve analysis has been proposed to assess the net benefit of using a risk prediction model [60,61]. Decision curve analysis evaluates the net benefit of a model over a range of risk thresholds, i.e., the level of risk used to classify predictions as positive or negative for the predicted outcome. For example, for the PLCOm2012 model, a risk threshold of 1.51% has been suggested, meaning that individuals with an estimated risk of 1.51% or higher are classified as positive (and thus eligible for screening) and individuals with an estimated risk lower than 1.51% as negative (and thus ineligible for screening) [13].
The net benefit is defined as:
net benefit= true positive count−(false positive count * weighting factor)number of individuals assessed for screening eligibility
where the weighting factor is defined as:
weighting factor= risk threshold(1−risk threshold)
This weighting factor represents how the relative harms of false-positive (classifying a person as eligible for screening who does not develop, or die from, lung cancer) and false-negative (classifying a person as ineligible for screening who develops, or dies from, lung cancer) results are valued at a given risk threshold, i.e., the ratio of harm to benefit, and is estimated by the threshold odds. For example, a risk threshold of 2.5% yields the following weighting factor:
weighting factor= 0.025(1−0.025)=139
This weighting factor implies that missing one case of lung cancer that could be detected through screening is valued as 39 times worse than unnecessarily screening one person, or that one case should be detected per 40 screened persons. Consequently, the less relative weight one gives to detecting a lung cancer case, the higher the risk threshold one will favor.
The net benefit can then be interpreted as follows: if the net benefit at a risk threshold of 2.5% is 0.002 greater compared with screening all persons eligible according to the NLST criteria, taking the weighing factor into account, this is equivalent to a net improvement in true-positive results of 0.002 × 1,000 = 2 per 1,000 persons assessed for screening eligibility, or a net reduction in false-positive results of 0.002 × 1,000/(0.025/0.975) = 78 per 1,000 persons assessed for screening eligibility [60]. Thus, if the risk model has a positive net benefit at the preferred risk threshold, this indicates that applying the model at this risk threshold provides a better ratio of benefits to harms than current screening guidelines based on pack-years. Decision curves visualize the net benefit over a range of risk thresholds, allowing one to discern whether and at which risk thresholds applying the risk model can be clinically useful [61]. Decision curves were used to determine at which range of risk thresholds applying the models provides a net benefit over using the NLST eligibility criteria for selecting individuals for lung cancer screening.
Finally, we identified the risk threshold for each model in the PLCO cohorts that selected a similar number of individuals for screening as the NLST eligibility criteria, on which most lung cancer screening recommendations are currently based. We then assessed the sensitivity (the number of individuals with lung cancer incidence or death classified as eligible for screening divided by the total number of individuals with lung cancer incidence or death) and specificity (the number of individuals without lung cancer incidence or death classified as ineligible for screening divided by the total number of individuals without lung cancer incidence or death) for each model compared to the NLST criteria at the chosen risk threshold, as reported before by Tammemägi et al. [13].
Multiple imputation of missing data for all considered risk factors was performed through the method of chained equations using the R package MICE [62]. History of pneumonia was not measured in the PLCO but was measured in the NLST; therefore, data from the NLST were used to impute history of pneumonia for PLCO participants [49]. Analyses were performed using 20 imputations, and the results were pooled through applying Rubin’s rules [63]. The results of the analyses with imputation of missing variables were similar to those obtained from complete case analyses. The Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines suggest applying multiple imputation when missing data are present, as complete case analyses can lead to inefficient estimates [64,65]. Therefore, all analyses reported here were performed with multiple imputation of missing values.
An overview of the characteristics of the four study cohorts (two trial arms in each trial) is given in Table 2, stratified by 6-y lung cancer incidence. A similar table stratifying participants by 6-y lung cancer mortality is provided in S2 Appendix. An overview of the proportion of individuals with complete information on all risk factors, stratified by trial arm and 6-y outcome, is given in S3 Appendix. Overall, approximately 93% of the study population had complete information for all considered risk factors.
The risk prediction models included in this study were developed in different populations (Table 1) and incorporate risk factors, specifically smoking behavior, in different ways (S1 Appendix). In addition, some models predict lung cancer incidence, while others predict lung cancer mortality. Therefore, the estimated absolute risk for the same individual varies between models [66]. Fig 1 shows the estimated 6-y risk of lung cancer incidence or mortality (depending on the target outcome of the model) across the models for five individuals with different risk factor profiles. This difference in estimated absolute risk between models suggests that specific risk thresholds might be needed for each model.
Overall, all models showed satisfactory calibration performance (S4 Appendix). The models showed the best calibration performance when they were applied to their target outcome, i.e., lung cancer incidence rather than lung cancer mortality for lung cancer incidence models. The calibration was better for all models in the PLCO datasets than in the NLST datasets.
The discriminative performance of the models (Figs 2–5) was better in the PLCO datasets (AUCs ranging from 0.74 to 0.81) than in the NLST datasets (AUCs ranging from 0.61 to 0.73). The discriminative performance of most models was better for lung cancer mortality than for lung cancer incidence (i.e., the AUCs of most models were higher for lung cancer mortality than for lung cancer incidence) in all datasets, except for the PLCO control arm. The PLCOm2012 model (and its simplified version), the Bach model, and the TSCE incidence model showed the best discriminative performance across all datasets regardless of the type of predicted outcome. The discriminative performance of the models was similar for 5- and 6-y time frames, as shown in S5 Appendix.
Decision curve analysis for each risk prediction model provided a range of risk thresholds that yield a positive net benefit compared with the NLST eligibility criteria. Table 3 shows the lower and upper bounds for these ranges of risk thresholds for 6-y lung cancer incidence across all datasets. Overall, the lower and upper thresholds varied by model, but the ranges were roughly consistent across models, going from approximately 0.1% to 16.7%. This suggests that applying the models is useful for determining screening eligibility if missing one case of lung cancer that could be detected through screening is perceived as being between 999 and 5 times worse than unnecessarily screening one person. More detailed results for the decision curve analyses for both lung cancer incidence and mortality are shown in S6 Appendix.
Applying the NLST eligibility criteria yielded a sensitivity of 71.4% (95% confidence interval: 68.0%–74.6%) and a specificity of 62.2% (95% confidence interval: 61.7%–62.7%) for 6-y lung cancer incidence in the PLCO CXR arm (Fig 6; Table 4). The sensitivity and specificity of each of the risk prediction models were higher than those of the NLST eligibility criteria. The PLCOm2012 model, in particular, followed by the Bach model and the TSCE incidence model had the highest sensitivities (all three models >79.8%) and specificities (all three models >62.3%) among all evaluated models. Fig 6 also shows the risk thresholds for each model that select a similar number of individuals for screening as the NLST eligibility criteria. Similar results were found for the PLCO control arm and for using 6-y lung cancer death as the outcome measure (S7 Appendix).
This study assessed the performance of nine lung cancer risk prediction models in two large randomized controlled trials: the NLST and the PLCO. The models had satisfactory calibration, had modest to good discrimination, and provided a substantial range of risk thresholds with a positive net benefit compared with the NLST eligibility criteria. Given appropriate model-specific risk thresholds, all risk prediction models had a better sensitivity and specificity than the NLST eligibility criteria. This implies that lung cancer risk prediction models, when coupled with model-specific risk thresholds, outperform currently recommended lung cancer screening eligibility criteria (Tables 3 and 4; Fig 6).
The risk prediction models considered in this study were developed in various cohorts for different outcome measures (lung cancer incidence versus mortality), with fundamental differences in model structures. Consequently, the absolute risk estimates differed between models, which led to differences in calibration performance between the models, specifically in the NLST cohorts. In addition, there were clear differences in discriminative ability between the models. The discriminative ability of all models was better in the PLCO cohorts than in the NLST cohorts, which may be caused by the higher heterogeneity in risk factor profiles among individuals in the PLCO compared with the NLST [67,68]. The NLST required individuals to have smoked at least 30 pack-years and included only current and former smokers (who quit less than 15 y ago), whereas the PLCO did not have any criteria for enrollment with regards to smoking history. In line with these criteria, the average NLST participant had a higher lung cancer risk than the average PLCO participant. The results of our investigation suggest that the discriminative ability of the evaluated models may be lower in groups at elevated risk, which may be due to the lower heterogeneity in risk among participants in these groups [67,68]. However, randomized clinical trials suggest that the results of CT screening may provide an opportunity to improve risk stratification in these groups. In the NLST, participants with a negative prevalence screen had a substantially lower risk of developing lung cancer than participants with a positive prevalence screen [69]. Similarly, in the NELSON trial, the 2-y probability of developing lung cancer after a CT screen varied substantially by pulmonary nodule size and the volume doubling time of these pulmonary nodules [8]. Therefore, incorporating the results of CT screening could improve the risk stratification in groups of individuals at elevated risk. Finally, while there was little difference in specificity between the models at risk thresholds similar to the NLST eligibility criteria, there was a clear difference in sensitivity. In particular, the PLCOm2012 model, followed by the Bach model and the TSCE incidence model, had the best performance across all aspects investigated in this study.
Previous studies have also compared the performance of different lung cancer risk prediction models [20,21]. D’Amelio et al. examined the discriminatory performance of three risk prediction models for lung cancer incidence in a case–control study and found modest differences between the models [20]. However, this study considered a limited number of participants (1,066 cases and 677 controls) and did not consider other aspects of model performance such as calibration or clinical usefulness. Li et al. examined four risk prediction models for lung cancer incidence in German participants of the European Prospective Investigation into Cancer and Nutrition cohort [21]. They found that while the differences between most of the evaluated models were modest, generally only the Bach and the PLCOm2012 models had similar or better sensitivity and specificity compared to the eligibility criteria used in the NLST and other eligibility criteria that were used in various European lung cancer screening trials (which applied less restrictive smoking eligibility criteria than the NLST). This cohort consisted of 20,700 individuals, but fewer than 100 lung cancer cases occurred, which limits statistical power for external validation [18,19].
In contrast to these previous studies, we performed a comprehensive validation, including aspects of calibration, discriminative ability, and clinical usefulness, for many models, in a large sample (n = 134,124) with 3,388 lung cancer cases and 1,799 lung cancer deaths. In addition, while our study supports earlier findings that risk prediction models outperform the NLST eligibility criteria, it also suggests that the PLCOm2012 model followed by the Bach and TSCE incidence models perform better than other models in all investigated aspects.
Our study has some limitations. While our results provide indications regarding at which risk thresholds the investigated risk models can be clinically useful, the optimal thresholds to apply remain uncertain. Determining optimal thresholds requires information on the long-term benefits (such as life-years gained and mortality reduction) and harms (such as overdiagnosis) of applying these thresholds [60]. Natural history modeling may provide further information on the trade-off between the long-term benefits and harms for screening programs with different risk thresholds, similarly to how our previous study informed the USPSTF on its recommendations for lung cancer screening [2].
Another limitation is that information on some of the predictor variables included in the evaluated risk prediction models was not available in the NLST and the PLCO, e.g., asbestos exposure was missing in both cohorts. However, only a few variables were unavailable. Furthermore, some of the evaluated models that used only age, gender, and smoking behavior, such as the TSCE models and the Knoke model, performed similarly to the other models that used additional information on risk factors, suggesting that age, gender, and smoking behavior are the most important risk factors for lung cancer. Thus, the improved performance of these models over the NLST eligibility criteria may primarily be due to the inclusion of detailed smoking behavior in these models. The NLST eligibility criteria use a dichotomized criterion for accumulated pack-years, e.g., an exposure of at least 30 pack-years, which leads to a loss of information for continuous variables [70]. Furthermore, pack-years are estimated by smoking duration and intensity (cigarettes per day), and previous studies indicate that both components contribute independently to an individual’s risk for developing lung cancer; an aggregation of both may not fully capture the effects of smoking on lung cancer risk [10,43,71].
We chose to evaluate the models for varying follow-up lengths (5- and 6-y time frames) to investigate the effect of follow-up duration on the discrimination performance of each model [54]. Although the discriminative performance of the models was similar for 5- and 6-y time frames (S5 Appendix), this may not be the case for more disparate time frames.
A number of pertinent questions remain with regards to the implementation of lung cancer screening [9]. Current guidelines like the USPSTF recommendations suggest that individuals should be asked, at a minimum, about their age and smoking history [3]. A number of the models evaluated in our study use information on additional risk factors, such as personal history of cancer, which could be a potential barrier for implementing lung cancer screening based on risk prediction models. However, the LLP and PLCOm2012 models were successfully used to recruit individuals for the UK Lung Cancer Screening Trial (UKLS) and the Pan-Canadian Early Detection of Lung Cancer Study (PanCan), respectively, through short questionnaires [33,72]. This suggests that acquiring information on the risk factors required for these models does not pose a major barrier for implementation. Furthermore, for some risk models, such as the Bach and PLCOm2012 models, online calculators are available, which provide opportunities for fast risk estimation in clinical practice [73–76]. For example, the PLCOm2012 model has been embedded in a lung cancer screening decision aid that has been widely adopted and that can be used to satisfy the Centers for Medicare & Medicaid Services reimbursement requirement for shared decision making [75–77].
In conclusion, our study suggests that lung cancer screening selection criteria can be improved through the explicit application of risk prediction models rather than using criteria based on age and pack-years as a summary measure of smoking exposure. These models might also be helpful for improving the shared decision-making process for lung cancer screening recommended by the USPSTF and required in the US by the Centers for Medicare & Medicaid Services [3,75,78]. However, recommendations for the implementation of risk-based lung cancer screening require a thorough evaluation of the benefits and harms of risk-based screening, as well as an assessment of the feasibility of implementing strategies based on risk models. Therefore, future studies need to evaluate the long-term benefits and harms of applying risk prediction models at different risk thresholds, while considering the potential challenges for implementation, and compare these with the expected benefits and harms of current guidelines.
|
10.1371/journal.pgen.1006904 | Tex19.1 promotes Spo11-dependent meiotic recombination in mouse spermatocytes | Meiosis relies on the SPO11 endonuclease to generate the recombinogenic DNA double strand breaks (DSBs) required for homologous chromosome synapsis and segregation. The number of meiotic DSBs needs to be sufficient to allow chromosomes to search for and find their homologs, but not excessive to the point of causing genome instability. Here we report that the mammal-specific gene Tex19.1 promotes Spo11-dependent recombination in mouse spermatocytes. We show that the chromosome asynapsis previously reported in Tex19.1-/- spermatocytes is preceded by reduced numbers of recombination foci in leptotene and zygotene. Tex19.1 is required for normal levels of early Spo11-dependent recombination foci during leptotene, but not for upstream events such as MEI4 foci formation or accumulation of H3K4me3 at recombination hotspots. Furthermore, we show that mice carrying mutations in Ubr2, which encodes an E3 ubiquitin ligase that interacts with TEX19.1, phenocopy the Tex19.1-/- recombination defects. These data suggest that Tex19.1 and Ubr2 are required for mouse spermatocytes to accumulate sufficient Spo11-dependent recombination to ensure that the homology search is consistently successful, and reveal a hitherto unknown genetic pathway promoting meiotic recombination in mammals.
| Meiosis is a specialised type of cell division that occurs during sperm and egg development to reduce chromosome number prior to fertilisation. Recombination is a key step in meiosis as it facilitates the pairing of homologous chromosomes prior to their reductional division, and generates new combinations of genetic alleles for transmission in the next generation. Regulating the amount of recombination is key for successful meiosis: as this process involves making many simultaneous breaks in the DNA, too much will likely cause mutations, chromosomal re-arrangements and genetic instability; whereas too little causes defects in homologous chromosome pairing prior to the meiotic divisions. This study identifies a genetic pathway required to generate robust meiotic recombination in mouse spermatocytes. We show that male mice with mutations in Tex19.1 or Ubr2, which encodes an E3 ubiquitin ligase that interacts with TEX19.1, do not generate sufficient meiotic recombination. We show that the defects in these mutants impact on recombination early in meiosis when programmed DNA double strand breaks are being made and processed. This defect likely contributes to the chromosome synapsis and meiotic progression phenotypes previously described in these mutant mice. This study has implications for our understanding of how this fundamental aspect of genetics and inheritance is controlled.
| Recombination plays key roles in meiosis and gametogenesis through facilitating the pairing and reductional segregation of homologous chromosomes, and by increasing genetic variation in the next generation. Meiotic recombination is initiated when programmed DNA double strand breaks (DSBs) are generated during the leptotene stage of the first meiotic prophase. Meiotic DSBs recruit a series of recombination proteins visualised cytologically as recombination foci, and initiate a search for homologous chromosomes thereby promoting homologous chromosome synapsis during zygotene. Recombination foci continue to mature while the chromosomes are fully synapsed in pachytene, and eventually resolve into crossover or non-crossover events. Crossovers exchange large tracts of genetic information between parental chromosomes, increasing genetic diversity in the population. Furthermore, these crossovers, which physically manifest as chiasmata, hold homologs together after they desynapse in diplotene and help to ensure that homologous chromosomes undergo an ordered reductional segregation at anaphase I [1,2].
Meiotic DSBs have a non-random distribution across the genome, and their frequency and location play an important role in shaping the recombination landscape [2,3]. In male mice, a few hundred meiotic DSBs are generated during leptotene, around 20–25 of which mature into crossovers. The positions of meiotic DSBs across the genome are determined by PRDM9, a histone methyltransferase that mediates trimethylation of histone H3 lysine 4 (H3K4me3) at recombination hotspots [4,5]. Meiotic DSBs are generated by an endonuclease that comprises SPO11 and TOPOVIBL subunits [2,3,6]. In mice, mutations in Spo11 result in fewer DSBs during leptotene and zygotene, and defects in the pairing and synapsis of homologous chromosomes [7–9]. The overall amount of SPO11 activity appears to be dynamically controlled at multiple levels during meiotic prophase. At the RNA level, Spo11 is alternatively spliced into two major isoforms whose relative abundance changes as meiotic prophase proceeds [10–12]. There also appears to be regulation of SPO11 activity at the protein level: negative feedback mechanisms acting through the DNA damage-associated protein kinase ATM prevent excessive Spo11-dependent DSBs from being generated during meiosis, potentially limiting any genetic instability caused by errors arising during repair of the DSBs and meiotic arrest caused by unrepaired DSBs [13]; and chromosome synapsis feeds back to locally inhibit SPO11 activity in chromosomal regions that have already synapsed during zygotene [14].
Mutations in genes involved in regulating early stages in meiotic recombination in mammals might be expected to phenocopy Spo11-/- mutants to some extent in having reduced numbers of DSBs in leptotene, and arrest at pachytene with chromosome asynapsis. One group of genes that is required for chromosome synapsis in mouse spermatocytes, but whose mechanistic role in meiosis is poorly defined, is the germline genome defence genes [15]. These genes are involved in suppressing the activity of retrotransposons in developing germ cells, and mutations in many of them cause defects in progression through the pachytene stage of meiosis [15]. Mutations in one of these germline genome defence genes, Mael, which encodes a conserved component of the piRNA pathway, causes de-repression of retrotransposons and a considerable increase in Spo11-independent DNA damage [16]. The Spo11-independent DNA damage generated in these mutants could potentially reflect the activity of the retrotransposon-encoded endonucleases that generate nicks or breaks in the host DNA to mediate mobilisation of these genetic elements [16]. In contrast, spermatocytes carrying mutations in the DNA methyltransferase accessory factor Dnmt3l also de-repress retrotransposons, but have relatively normal levels of DSBs that are aberrantly distributed across the genome [17–19].
The germline specificity in expression of at least a subset of the germline genome-defence genes is achieved through tissue-specific promoter DNA methylation [20]. One of the most methylation sensitive of these genes is Tex19.1 [20,21]. Tex19.1 was originally identified in a screen for testis-specific genes [22], and is one of two rodent paralogs of this mammal-specific gene family [23]. Although TEX19.1 was described as being a nuclear factor with potential roles in maintenance of stem cells or pluripotency [23], subsequent functional studies demonstrated that TEX19.1 is predominantly cytoplasmic in the mouse germline, where it has roles in meiosis and repression of retrotransposons [24,25]. TEX19.1 physically interacts with UBR2 [25], inhibiting the activity of this E3 ubiquitin ligase towards its normal cellular substrates in the N-end rule pathway [26] and promoting its activity towards retrotransposon-encoded proteins [27]. Tex19.1 mutant spermatocytes progress into the pachytene stage of meiotic prophase but frequently contain asynapsed chromosomes and accumulate retrotransposon RNA [24,25]. Thus, Tex19.1 mutants arrest at a similar stage of meiosis as Dnmt3L and Mael mutants, despite expressing different retrotransposon RNAs [16,17,24]. However, meiotic chromosome synapsis requires multiple upstream events to be executed correctly, and it is not clear if the meiotic defects in Tex19.1 mutant spermatocytes are similar to the defects present in Dnmt3L and Mael mutant spermatocytes, or if they arise through different mechanisms.
In this study we elucidate why loss of the germline genome defence gene Tex19.1 results in chromosome asynapsis in male meiosis. We show that loss of Tex19.1 generates a meiotic phenotype distinct from either Mael-/- or Dnmt3l-/- mutants. Rather loss of Tex19.1 phenocopies hypomorphic Spo11 mutants and impairs Spo11-dependent recombination during the leptotene stage of meiotic prophase. Furthermore, we show that mice lacking the TEX19.1-interacting protein UBR2 phenocopy the recombination defects seen in leptotene Tex19.1-/- spermatocytes. These data show that Tex19.1 and Ubr2 are required for sufficient SPO11-dependent recombination to ensure robust identification and synapsis of homologous chromosomes in meiotic spermatocytes.
Tex19.1 is a DNA methylation-sensitive germline genome defence gene whose expression is primarily restricted to germ cells and pluripotent cells in the embryo [20,22–24]. We and others have previously reported that Tex19.1-/- males have defects in spermatogenesis on a mixed genetic background, and that around 50% of pachytene spermatocytes in Tex19.1-/- testes have asynapsed chromosomes, but the molecular explanation for this defect remains unknown [15,24,25]. Synapsis requires the accurate and timely execution of a number of events in the preceding stages of the first meiotic prophase, including the generation of meiotic DNA double-strand breaks (DSBs) in leptotene, followed by homolog pairing and assembly of the synaptonemal complex (SC) in zygotene [1]. To investigate the molecular basis for the chromosome asynapsis in pachytene Tex19.1-/- spermatocytes we sought to test whether each of these events occurs normally in the absence of Tex19.1.
First, we confirmed that the meiotic chromosome asynapsis phenotype persists in Tex19.1-/- spermatocytes after backcrossing onto an inbred C57BL/6 genetic background: 65.4% ± 1.3 of Tex19.1-/- pachytene spermatocytes from three animals were asynapsed in this genetic background, significantly higher than the 3.7% ± 1.3 of Tex19.1+/±- control pachytene spermatocytes that were asynapsed (Student’s t-test, p<0.001) (Fig 1A). This is similar to the asynapsis in 50% of Tex19.1-/- pachytene nuclei previously reported for a mixed genetic background [24,25]. To assess whether the chromosome asynapsis in Tex19.1-/- spermatocytes represents defects in SC assembly rather than pairing of homologous chromosomes, we scored the configuration of the asynapsed chromosomes in these asynapsed pachytene Tex19.1-/- nuclei. Defects in assembly of the SC transverse filaments results in asynapsed chromosomes that are aligned in their homolog pairs whereas defective recombination or pairing between homologous chromosomes manifests as isolated asynapsed single chromosomes, partial synapsis between non-homologous chromosomes, and incomplete synapsis between homologous chromosomes [7,8,28,29]. Asynapsed chromosomes in Tex19.1-/- spermatocytes are present in multiple configurations consistent with defects in recombination or homolog pairing, but do not present as asynapsed aligned homolog pairs (Fig 1A, Fig 1B).
To confirm that the asynapsis phenotype in Tex19.1-/- spermatocytes does not represent a primary defect in SC assembly, we quantified the effect of Tex19.1 on the number of SC fragments assembled independently of recombination in a Spo11-/- genetic background [30]. Spo11-/- spermatocytes arrest with a zygotene-like SC configuration with complete axial element formation but limited synapsis [7,8]. Spo11-/- Tex19.1+/± and Spo11-/- Tex19.1-/- spermatocytes are able to assemble similar amounts of SC in this assay (Fig 1C, Fig 1D), suggesting that loss of Tex19.1 does not severely impair recombination-independent SC assembly. Taken together, these data suggest that the chromosome asynapsis in Tex19.1-/- spermatocytes is likely primarily caused by defects in meiotic recombination and/or homolog pairing rather than a direct defect in SC assembly.
We next investigated whether loss of Tex19.1 impaired the abundance of recombination intermediates required for homologous chromosome pairing and synapsis. Chromosome spreads were immunostained for SYCP3, a component of the axial and lateral elements of the SC [31], and SYCE2, a component of the SC central element [32], to identify zygotene nuclei. Recombination foci associated with the chromosome axes were visualised by immunostaining for the single-stranded DNA binding proteins RPA, DMC1 and RAD51 [33]. The number of RPA, DMC1 and RAD51 foci in control Tex19.1+/± zygotene nuclei (Fig 2) are all within the ranges previously reported for wild-type zygotene spermatocytes (150–250 for RPA foci, 100–250 for DMC1 and RAD51 foci) [33]. However, zygotene Tex19.1-/- spermatocytes have fewer DMC1 and RAD51 foci than their littermate controls, with DMC1 and RAD51 foci frequency reduced to 87% and 67% of control levels respectively (Fig 2). Interestingly, the number of RPA foci is not statistically different from zygotene control nuclei (Fig 2), which could potentially reflect RPA foci being a later marker of recombination than RAD51 and DMC1 [34]. The differential behaviour of RAD51 and DMC1 foci in Tex19.1-/- spermatocytes suggests that the generation, repair, or maturation kinetics of recombination foci is perturbed in the absence of Tex19.1.
Meiotic recombination is initiated during leptotene [9], therefore we next investigated whether loss of Tex19.1 might perturb recombination foci frequency at this earlier stage of meiotic prophase. Counts of RPA, DMC1 and RAD51 foci in leptotene nuclei revealed a severe reduction in the frequency of each of these in the absence of Tex19.1 (Fig 3). The numbers of RPA foci, DMC1 foci and RAD51 foci in leptotene Tex19.1-/- spermatocytes were reduced to 63%, 30%, and 60% of those present in control spermatocytes (Fig 3). Thus, loss of Tex19.1 results in reduced numbers of meiotic recombination foci in leptotene spermatocytes, a defect that precedes the chromosome asynapsis at pachytene.
The reduced numbers of recombination foci in Tex19.1-/- spermatocytes could potentially decrease the efficiency of the DSB-dependent homology search and contribute to chromosome asynapsis in this mutant. Analysis of Spo11 hypomorphs suggests that reduced numbers of meiotic DSBs impairs the initiation of synapsis and manifests as reduced numbers of SC fragments during late leptotene/early zygotene stages [14]. We therefore analysed the extent of synapsis in zygotene Tex19.1-/- nuclei to assess whether the initiation of synapsis might similarly be impaired in these mutants. Chromosome spreads were immunostained with axial and central element SC markers and the percentage synapsis assessed in each zygotene nucleus (S1 Fig). In the absence of Tex19.1, most zygotene nuclei contained very low amounts of synapsis (<10%), whereas the majority of control zygotene nuclei contained intermediate levels of synapsis (10–70%, S1 Fig). The extent of synapsis in Tex19.1-/- nuclei is more consistent with these mutants exhibiting a widespread block or delay in the initiation of synapsis throughout the nucleus, rather than defects in synapsis of specific chromosomes or progression of synapsis along the chromosome axes once it has initiated. Thus, as described for Spo11 hypomorphs [14], the reduced numbers of recombination foci in Tex19.1-/- sperrmatocytes during leptotene could potentially cause defects in homologous chromosome synapsis during zygotene resulting in asynapsis persisting in pachytene.
The reduced number of RPA, DMC1 and RAD51 foci in leptotene Tex19.1-/- spermatocytes might reflect fewer Spo11-dependent DSBs in these cells, or defects in the processing and resection of those DSBs to form the single-stranded DNA ends that recruit RPA, DMC1 and RAD51, or accelerated repair of SPO11-induced DNA damage. Phosphorylation of the histone variant H2AX to generate γH2AX occurs in response to Spo11-dependent DSB formation [9], and is not impaired in spermatocytes proposed to be defective in subsequent processing of those DSBs [35]. We therefore tested whether loss of Tex19.1 affects γH2AX abundance in leptotene spermatocytes. In both control and Tex19.1-/- leptotene nuclei, γH2AX is present as a diffuse cloud of staining over regions of the nucleus (Fig 4A). Interestingly, quantification of the γH2AX signal showed that the amount of γH2AX in leptotene Tex19.1-/- nuclei was around half that in Tex19.1+/± controls (Fig 4B). Taken together, the reduced numbers of recombination foci and the reduced intensity of γH2AX immunostaining in Tex19.1-/- spermatocytes suggests that loss of Tex19.1 likely causes defects in early stages of Spo11-dependent recombination, or accelerated repair of SPO11-induced DNA damage.
The bulk of the γH2AX generated in spermatocytes reflects the generation of Spo11-dependent meiotic DSBs, however small amounts of γH2AX are generated independently of Spo11 in these cells [9,36–39]. The extent of the decrease in γH2AX abundance in Tex19.1-/- spermatocytes is arguably more consistent with reduced abundance of Spo11-dependent DSBs, but it is possible that loss of Tex19.1 also affects Spo11-independent γH2AX generated during leptotene. To test directly whether loss of Tex19.1 affects Spo11-independent γH2AX we quantified γH2AX abundance as well as DMC1 foci in Spo11-/- Tex19.1-/- double mutant spermatocytes. The relatively low levels of γH2AX present in Spo11-/- spermatocytes typically manifests as a pseudo sex body, a cloud of γH2AX associated with a subset of asynapsed axes undergoing meiotic silencing of unsynapsed chromatin [38,39]. In addition to the pseudo sex body, smaller additional flares of chromosome axis-associated γH2AX staining termed L-foci are also present [36,37]. Spo11-/- Tex19.1-/- spermatocytes displayed similar γH2AX staining patterns and similar numbers of γH2AX L-foci as Spo11-/- Tex19.1+/± controls (Fig 4C, Fig 4D). Thus, pseudo sex body formation and Spo11-independent γH2AX L-foci frequency are independent of Tex19.1. In addition, although loss of Tex19.1 impairs DMC1 foci frequency in a wild-type Spo11 background (Fig 2, Fig 3), loss of Tex19.1 has no detectable effect on DMC1 foci frequency in a Spo11-/- mutant background (Fig 4E, Fig 4F). Thus, loss of Tex19.1 appears to reduce the amount of Spo11-dependent recombination present in spermatocytes. In this respect the Tex19.1-/- phenotype bears some resemblance to hypomorphic Spo11 mutants [14,40]
Mutations in the genome-defence gene Mael have been reported to result in the accumulation of large amounts of Spo11-independent DNA damage as assessed by γH2AX staining and the presence of axis-associated RAD51 foci in late zygotene Mael-/- Spo11-/- double mutant spermatocytes [16]. However, in contrast to Mael-/- Spo11-/- double mutants [16], zygotene-like Tex19.1-/- Spo11-/- double mutant spermatocytes do not accumulate γH2AX (Fig 4C, Fig 4D) or axis-associated RAD51 foci (S2 Fig). Therefore, Tex19.1 and the piRNA pathway component Mael appear to have different effects on Spo11-independent DNA damage in meiotic spermatocytes.
SPO11 is locally regulated in the nucleus, and feedback controls are thought to allow SPO11 to continue to generate DSBs on asynapsed regions of the chromosomes in late zygotene [14]. Spo11 hypomorphs are still able to generate DSBs on asynapsed chromatin [14]. To assess whether asynapsed chromatin is similarly able to accumulate high levels of DSBs in Tex19.1-/- mutants, we counted the number of RPA foci associated with the sex chromosomes, which remain largely asynapsed during pachytene. In the population of pachytene Tex19.1 spermatocytes that successfully synapse all their autosomes, sex chomosomes were still able to accumulate similar numbers of RPA foci as control pachytene nuclei (S1 Fig). Thus, loss of Tex19.1 does not prevent the accumulation of RPA foci on asynapsed chromatin.
Loss of Tex19.1 results in female subfertility as well as male infertility [24]. However, loss of Tex19.1 has sexually dimorphic effects on progression through meiotic prophase, and in contrast to its effects on spermatocytes, loss of Tex19.1 does not cause defects in chromosome synapsis in female meiosis [26]. Nevertheless, it is possible that loss of Tex19.1 could still cause a reduction in the number of early recombination foci in female meiosis that might not be sufficient to result in chromosome asynapsis. We therefore analysed RAD51 foci in E14.5 Tex19.1-/- foetal oocytes to test whether loss of Tex19.1 affects recombination in female meiosis. However, the number of RAD51 foci in late leptotene Tex19.1-/- oocytes is not significantly different from late leptotene Tex19.1+/± littermate controls (S3 Fig). Therefore Tex19.1 is not required for accumulation of RAD51 foci in female meiosis and has a sexually dimorphic role in early meiotic recombination.
We next investigated whether the reduced frequency of Spo11-dependent recombination foci in leptotene Tex19.1-/- spermatocytes might reflect defects upstream of Spo11 in meiotic recombination. The requirements upstream of Spo11 for meiotic DSB formation are relatively poorly understood in mammals, however SPO11 activity likely depends on the recruitment of the conserved axis-associated protein MEI4 to the chromosomal axes in leptotene [41]. We therefore quantified MEI4 foci in leptotene Tex19.1-/- nuclei to test whether this event is perturbed by loss of Tex19.1. Control leptotene Tex19.1+/± spermatocytes possess an average of 218 axis-associated MEI4 foci (Fig 5A, Fig 5B), similar but slightly lower than the average 309 foci per leptotene nucleus reported previously [41]. Leptotene Tex19.1-/- nuclei possess similar numbers of MEI4 foci to leptotene Tex19.1+/± controls (Fig 5A, Fig 5B). Thus, the reduced frequency of recombination foci seen in Tex19.1-/- leptotene spermatocytes appears to be a consequence of defects acting downstream or independently of MEI4 localisation to chromosome axes.
Spo11 function is also influenced by the activity of the histone methyltransferase PRDM9, which targets SPO11 to recombination hotspots [2,4,5]. Mutations in Prdm9 result in reduced anti-H3K4me3 immunostaining in P14 spermatocytes, a failure to enrich H3K4me3 at Prdm9-dependent recombination hotspots, a reduction in recombination foci during early prophase, and meiotic chromosome asynapsis [5,42,43]. We therefore tested whether loss of Tex19.1 might impair Prdm9 function by assessing anti-H3K4me3 immunostaining intensity in leptotene nuclei. However, we could not detect a difference in the amount of anti-H3K4me3 immunostaining between Tex19.1+/± and Tex19.1-/- leptotene nuclei (Fig 5C, Fig 5D). To test whether the distribution of H3K4me3 rather than its total abundance might be altered in the absence of Tex19.1 we performed H3K4me3 chromatin immunoprecipitation (ChIP)-qPCR on P16 testes. H3K4me3 is enriched at transcriptional start sites (TSSs) of active genes in addition to meiotic recombination hotspots [44], and as expected both Tex19.1+/± and Tex19.1-/- testes show enrichment of H3K4me3 at Gapdh and Polr2a active TSSs, but not at a Polr2a intragenic region (Fig 5E). However, loss of Tex19.1 does not perturb the accumulation of H3K4me3 at Prdm9-dependent recombination hotspots (Fig 5E). Thus, the defects in Spo11-dependent recombination seen in Tex19.1-/- spermatocytes does not appear to be a downstream consequence of impaired Prdm9 activity.
Tex19.1 plays a role in repressing retrotransposons in testes and placenta [21,24,45], and Tex19.1-/- testes have increased abundance of MMERVK10C retrotransposon RNA, but not RNAs encoding IAP or LINE-1 retrotransposons [24,45]. To test if the increase in MMERVK10C RNA is a consequence of transcriptional de-repression we also analysed retrotransposon sequences in the P16 testis H3K4me3 ChIP. Interestingly, the LTR driving MMERVK10C expression, but not IAP LTRs or LINE-1 5' UTR sequences are enriched in anti-H3K4me3 ChIP from Tex19.1-/- testes relative to Tex19.1+/± controls (Fig 5F). Thus the increase in MMERVK10C retrotransposon RNA abundance previously reported in Tex19.1-/- testes [24,45] reflects, at least in part, transcriptional de-repression of this element. However, the 2-fold increase in H3K4me3 abundance at MMERVK10C LTR sequences does not detectably interfere or compete with enrichment of H3K4me3 at Prdm9-dependent recombination hotspots.
TEX19.1 physically interacts with the E3 ubiquitin ligase UBR2 [25] and regulates its activity [26,27]. TEX19.1 protein is undetectable in Ubr2-/- testes, suggesting that much of the TEX19.1 protein in the testis requires UBR2 for its stability [25]. Ubr2 is implicated in the ubiquitylation and degradation of N-end rule substrates and previous reports suggest that loss of Ubr2 causes variable defects in spermatogenesis possibly depending on the strain background [46]. Some Ubr2-/- spermatocytes are reported to progress into pachytene and arrest due to defects in the accumulation of ubiquitylated histone H2A at the sex body and meiotic sex chromatin inactivation during pachytene [47,48]. Ubr2-/- spermatocytes are also reported to arrest and apoptose in prophase I due to defects in the repair of DSBs, homologous chromosome pairing, and SC formation [46,48]. Given the lack of detectable TEX19.1 protein in Ubr2-/- testes, we tested whether the reported defects in homologous chromosome pairing and SC formation in Ubr2-/- spermatocytes [46] might reflect earlier defects in the initiation of meiotic recombination similar to Tex19.1-/- spermatocytes. We generated Ubr2-/- mice carrying a premature stop codon in the N-terminal region of UBR2 within the UBR domain that binds N-end rule substrates. The Ubr2-/- mice analysed here have no detectable UBR2 protein in their testes (S4 Fig), a 68% reduction in testis weight (S4 Fig), and no detectable sperm in their epididymis (S4 Fig), consistent with Ubr2-/- spermatogenesis defects reported previously [46]. The seminiferous tubules in Ubr2-/- mice contain reduced numbers of post-meiotic round and elongated spermatids, and accumulations of pyknotic and zygotene-like nuclei consistent with meiotic defects (S4 Fig) as reported previously [46,48]. Similar to Tex19.1-/- testes [24], some round and elongated post-meiotic spermatids are detectable in Ubr2-/- testes suggesting that any meiotic defects present do not completely block spermatogenesis. Furthermore, loss of Ubr2 phenocopies the specific retrotransposon derepression seen in Tex19.1-/- testes [24]: MMERVK10C, but not LINE-1 or IAP, retrotransposon RNAs are derepressed in Ubr2-/- spermatocytes (Fig 6A, Fig 6B).
We tested whether the meiotic defects in Ubr2-/- spermatocytes might resemble the asynapsis seen in Tex19.1-/- spermatocytes (Fig 1A and 1B). Chromosome spreads from Ubr2-/- testes confirm that this Ubr2 mutant allele causes defects in progression through meiotic prophase, and very few spermatocytes progress through pachytene into diplotene (Fig 6C). Furthermore, around 40% of pachytene Ubr2-/- spermatocytes had at least one asynapsed autosome pair when staging SYCP3-positive nuclei for meiotic progression under low magnification (Fig 6C, Fig 6D). At higher magnification, 65.9% ± 2.5 Ubr2-/- pachytene nuclei from three animals have some autosomal asynapsis, compared to 11.3% ± 2.5 pachytene nuclei from three Ubr2+/+ animals (p<0.001, Student’s t-test). Like in Tex19.1-/- spermatocytes (Fig 1A, Fig 1B), these asynapsed chromosomes are present in multiple configurations consistent with defects in recombination or homolog pairing (Fig 6E). Similar to Tex19.1-/- spermatocytes, the asynapsis in Ubr2-/- spermatocytes is also associated with earlier defects in meiotic recombination. γH2AX abundance, DMC1 foci frequency and RAD51 foci frequency are reduced to around 50%, 52% and 58% respectively of those seen during leptotene in Ubr2-/- mutants (Fig 7), which contrasts with a previous report that γH2AX staining, and RAD51 and RPA foci frequency are unaffected in leptotene Ubr2-/- spermatocytes [48]. Consistent with the decrease in leptotene recombination foci frequency reported here, DMC1 and RAD51 foci frequency remain around 66% and 86% respectively of that seen in control spermatocytes during zygotene (Fig 7). As there appeared to be some qualitative similarity between the defects in recombination foci frequency in Ubr2-/- and Tex19.1-/- spermatocytes, we tested whether this meiotic recombination defect would be sufficient to delay or impair the initiation of chromosome synapsis in Ubr2-/- spermatocytes. Measurement of the extent of chromosome synapsis in zygotene Ubr2-/- spermatocytes suggests that, like in Tex19.1-/- mutants (S1 Fig) and Spo11 hypomorphs [14], synapsis is delayed in the absence of Ubr2 (S4 Fig). These data suggest that the defect in progression to pachytene previously reported in Ubr2-/- mutants [46] may reflect loss of TEX19.1 protein and earlier defects in the meiotic recombination in these mutants. Furthermore, these data show that Ubr2 and Tex19.1 are both required to allow sufficient early recombination foci to accumulate to drive robust homologous chromosome synapsis in mouse spermatocytes.
This study aimed to elucidate the mechanistic basis of the chromosome synapsis defect in male mice carrying mutations in the germline genome defence gene Tex19.1 [24]. We have shown that the pachytene chromosome asynapsis in these mice, and in mice carrying mutations in the TEX19.1-interacting protein UBR2, is likely a downstream consequence of reduced meiotic recombination earlier in meiotic prophase. Wild-type mice generate around 10-fold more meiotic DSBs than there are chiasmata, and the large numbers of DSBs generated in leptotene and zygotene appear to be important to drive pairing and synapsis of homologous chromosomes [2,7,8]. Allelic series of Spo11 activity suggest that reducing the number of meiotic DSBs to around 50% of normal levels is sufficient to cause chromosome asynapsis [14,40]. The reduction in early recombination foci seen in leptotene Tex19.1-/- and leptotene Ubr2-/- spermatocytes, is similar to this threshold and could be sufficient to account for the chromosome asynapsis seen in these mutants. Notably, Tex19.1-/- spermatocytes and Ubr2-/- spermatocytes do not exhibit a severe asynapsis phenotype: only a proportion of pachytene Tex19.1-/- or Ubr2-/- spermatocytes have asynapsed chromosomes, and there is some progression to post-meiotic spermatid stages in both these mutants. Thus the ~50% reduction in leptotene DSB frequency caused by loss of Tex19.1 or Ubr2 could be sufficient to cause the level of asynapsis present in these spermatocytes.
Interestingly, the frequency of recombination foci in zygotene Tex19.1-/- spermatocytes is closer to wild-type levels than that seen during leptotene, suggesting additional recombination foci are accumulating during zygotene that allow the Tex19.1-/- spermatocytes to catch up with wild-type cells. It is possible that DSB generation is delayed in Tex19.1-/- spermatocytes, or that repair of DSBs is accelerated in leptotene but not zygotene Tex19.1-/- spermatocytes, or that this compensation of the Tex19.1 recombination deficiency during zygotene reflects control mechanisms that regulate DSB frequency in meiotic cells [14]. An overall delay in germ cell development is probably not causing a delay in meiotic recombination relative to axial element assembly as previous analysis of gene expression profiles in P16 Tex19.1-/- testes does not exhibit enrichment of genes expressed in more immature germ cells such as spermatogonia or leptotene spermatocytes [24,45]. In hypomorphic Spo11 mice, DSBs are generated on asynapsed regions of the chromosomes during zygotene, potentially stimulating homology search and synapsis in these regions [14]. However, although any additional early recombination foci that accumulate in zygotene in Tex19.1-/- spermatocytes might be rescuing asynapsis to some degree, they are not sufficient to allow the majority of Tex19.1-/- spermatocytes to complete synapsis.
Tex19.1 is one of a group of germline genome defence genes which cause retrotransposon de-repression and defects in meiotic chromosome synapsis [15]. Although a common mechanism could link de-repression of retrotransposons and chromosome asynapsis in these mutants, mutations in different germline genome defence genes seem to have distinct effects on DNA damage and recombination during meiosis. Mutations in Mael cause a striking increase in Spo11-independent DNA damage in meiotic spermatocytes which has been proposed to represent DSBs generated by retrotransposon-encoded endonucleases [16], but the absence of any detectable increase in Spo11-independent DNA damage in Spo11-/- Tex19.1-/- spermatocytes reported here contrasts markedly with the phenotype of Spo11-/- Mael-/- spermatocytes. Moreover, zygotene recombination foci frequency is reduced in Tex19.1-/- spermatocytes, but are not perturbed by mutations in Dnmt3l [18,19]. Although Tex19.1 has been proposed to be part of the piRNA pathway [49], the phenotypic differences between the meiotic defects in Tex19.1 mutants and different germline genome defence mutants, indicate that distinct mechanisms may be causing asynapsis in each of these mutants.
The spectrum of retrotransposons de-repressed in Tex19.1-/- spermatocytes differs from those de-repressed in Mael-/- testes and Dnmt3l-/- testes [16,17,24,45]. It is therefore possible that some of the differences between the meiotic phenotypes of these mutants reflects differences in the type of retrotransposon de-repressed or the mechanism of de-repression. Data from Dnmt3l-/- mice suggests that transcriptional activation of LINE-1 retrotransposons alters the distribution of meiotic recombination foci and induces recombination at LINE-1 elements leading to interactions between non-homologous chromosomes [19]. It is not clear if transcriptional activation of MMERVK10C elements in Tex19.1-/- spermatocytes causes a similar re-distribution of meiotic recombination foci. Neither is it clear if loss of Tex19.1 perturbs recombination at all meiotic recombination hotspots equally. Thus, we cannot rule out the possibility that altered distribution of meiotic recombination is contributing to chromosome asynapsis in Tex19.1-/- spermatocytes. However, recombination foci abundance in Tex19.1-/- spermatocytes is reduced to a level similar to that seen in hypomorphic Spo11 mutants, which also have defects in chromosome synapsis [14,40]. Thus, reduced meiotic recombination is likely the primary cause of chromosome asynapsis in Tex19.1-/- spermatocytes.
The data presented here suggest that both Tex19.1 and Ubr2 are required for sufficient meiotic recombination to drive robust chromosome synapsis in spermatocytes. We have shown defects in the number of early recombination foci, and in the amount of γH2AX present during leptotene in both these mutants. Further experiments are required to delineate which stage in early recombination is disrupted in these mutants. It is possible that SPO11 activity and DSB formation itself is reduced or delayed. Alternatively early stages in processing SPO11-dependent DSBs or signalling the SPO11-induced DNA damage could be perturbed in these mutants. Or SPO11-dependent DSBs or recombination intermediates could be repaired more rapidly in the absence of Tex19.1 or Ubr2. In contrast to males, Tex19.1 was not required for synapsis in oocytes, or for the generation of normal numbers of RAD51 foci in female meiosis. This could reflect a difference in the genetic requirement for Spo11-dependent recombination between male and female meiosis, or alternatively could reflect some redundancy between Tex19.1 and its paralog, Tex19.2 at this stage of development [20,23].
UBR2 was previously suggested not to have a role in the initiation of meiotic recombination as it did not localise to recombination foci and was not required for normal recruitment of RAD51 or RPA to recombination foci during leptotene [48]. Immunocytologically-detectable enrichment at recombination foci is probably not a requirement for UBR2 to directly or indirectly influence the initiation of meiotic recombination. However, the effect of Ubr2 on recombination foci and γH2AX during leptotene reported here does contradict the previous description of the Ubr2-/- leptotene spermatocyte phenotype, although representative images and quantitative analysis of recombination foci in leptotene Ubr2-/- spermatocytes were not shown in that study [48]. Differences between mouse strain background or Ubr2 allele being studied may contribute to this, and the delay in synapsis initiation during zygotene (S4 Fig) could also complicate meiotic prophase substaging during analysis of Ubr2-/- spermatocytes leading to differences between studies. However, reduced numbers of recombination foci in zygotene Ubr2-/- spermatocytes have been reported previously [48] and are consistent with the data presented here. Our data indicate that the reduction in zygotene recombination foci in Ubr2-/- spermatocytes is a consequence of earlier defects in the meiotic recombination during leptotene, and that reduced meiotic recombination is contributing to the Ubr2-/- meiotic phenotype.
The phenotypic similarity between Tex19.1-/- mutants and Ubr2-/- mutants, in combination with the physical interaction between TEX19.1 and UBR2 proteins [25], and the requirement for Ubr2 for TEX19.1 protein stability [25], suggests that TEX19.1 and UBR2 are functioning in the same pathway to promote meiotic recombination. However, the molecular mechanism underlying the genetic requirement for Tex19.1 and Ubr2 in meiotic recombination is not clear. It is possible that the defects in the initiation of meiotic recombination in Ubr2-/- spermatocytes described here reflects the absence of TEX19.1 protein in these cells and uncharacterised downstream functions of TEX19.1 in promoting meiotic recombination. Alternatively, it is possible TEX19.1 is regulating the activity of UBR2 [26], a RING-domain E3 ubiquitin ligase, and that TEX19.1’s effects on meiotic recombination reflect a role for at least one of UBR2’s substrates in this process. Indeed, loss of Tex19.1 might have effects on multiple UBR2 substrates and that could be responsible for the different aspects of the Tex19.1-/- phenotype in different developmental stages and tissues. UBR2 has been shown to ubiquitylate histone H2A and histone H2B [47], and is implicated in degrading the C-terminal fragment of the REC8 cohesin subunit generated by separase-dependent cleavage [50]. Therefore it is possible that loss of Tex19.1 or Ubr2 affects meiotic chromosome organisation or the meiotic chromatin substrate on which the SPO11 endonuclease is acting. It is also possible that UBR2 ubiquitylates SPO11, or one of its regulators, and that loss of Tex19.1 or Ubr2 affects the amount of SPO11 activity rather than its chromatin substrate in leptotene spermatocytes. These possibilities are not mutually exclusive and further work is required to elucidate how the Tex19.1-Ubr2 pathway influences meiotic recombination. However, the data presented here demonstrates that Tex19.1 and Ubr2 are genetically required to ensure that sufficient recombination is present in male meiosis.
Animal experiments were carried out under authority of UK Home Office Project Licence PPL 60/4424 after ethical approval by University of Edinburgh Animal Welfare and Ethical Review Body.
Tex19.1-/- animals backcrossed three times to a C57BL/6 genetic background were bred and genotyped as described [24]. Spo11+/- heterozygous mice [7] on a C57BL/6 genetic background [18] were inter-crossed with Tex19.1+/- mice. Animal experiments were carried out under UK Home Office Project Licence PPL 60/4424. Noon on the day that a plug was found was designated E0.5, day of birth was designated P1, and adult mice were typically analysed at between 6–14 weeks old. Tex19.1+/+ and Tex19.1+/- animals have no difference in testis weights or sperm counts [24] and did not show any difference in leptotene recombination foci frequency (171 ± 7 and 185 ± 7 DMC1 foci; 217 ± 7 and 217 ± 8 RPA foci for wild-type and heterozygous mice respectively, no significant difference by Mann-Whitney U test, n = 40, 21, 10, 40 respectively). Therefore data from these control genotypes were pooled as Tex19.1+/± to reduce animal use. Epididymal sperm counts were determined as described [24]. Ubr2-/- mice were generated by CRISPR/Cas9 double nickase-mediated genome editing in zygotes [51]. Complementary oligonucleotides (S2 Table) targeting exon 3 of Ubr2 were annealed and cloned into plasmid pX335 [52], amplified by PCR, then in vitro transcribed using a T7 Quick High Yield RNA Synthesis kit (NEB) to generate paired guide RNAs. RNA encoding the Cas9 nickase mutant (50 ng/µl, Tebu-Bio), paired guide RNAs targeting exon 3 of UBR2 (each at 25 ng/µl), and 150 ng/µl single-stranded DNA oligonucleotide repair template (S2 Table) were microinjected into the cytoplasm of C57BL/6 × CBA F2 zygotes. The repair template introduces an XbaI restriction site and mutates cysteine-121 within the UBR domain of Ubr2 (Uniprot Q6WKZ8-1) to a premature stop codon. The zygotes were then cultured overnight in KSOM (Millipore) and transferred into the oviduct of pseudopregnant recipient females. Pups were genotyped and the mutant Ubr2 allele back-crossed to C57BL/6.
Chromosome spreads from testes were prepared as described by Peters et al. [53] for the Spo11-/- and Tex19.1-/- Spo11-/- analyses, or by Costa et al. [32] for all other analyses. Chromosome spreads were prepared from foetal ovaries as described [54]. For immunostaining, slides were blocked and antibodies diluted in PBS containing 0.15% BSA, 0.1% Tween-20 and 5% goat serum as indicated in S1 Table. The anti-MEI4, anti-SYCE2, and anti-RPA primary antibodies used were as reported [41,54,55]. Alexa Fluor-conjugated secondary antibodies (Invitrogen) were used at a 1:500 dilution, and 2 ng/μl 4,6-diamidino-2-phenylidole (DAPI) was used to fluorescently stain DNA. Slides were mounted in 90% glycerol, 10% PBS, 0.1% p-phenylenediamine. Three or four channel images were captured with iVision or IPLab software (BioVision Technologies) using an Axioplan II fluorescence microscope (Carl Zeiss) equipped with motorised colour filters. Immunostaining was performed on spreads from at least three experimental and three control animals unless otherwise stated. Statistical analysis was performed in R [56], means are reported ± standard error, and n is reported as total number of spreads analysed in each experiment.
Nuclei were staged by immunostaining for the axial/lateral element marker SYCP3 [31]. Nuclei with short fragments of axial element but no synapsis were classified as leptotene, nuclei containing some regions of axial element undergoing synapsis along with some regions of axial element not undergoing synapsis were classified as zygotene, and those with complete autosomal synapsis as pachytene. Immunostaining for the central element component SYCE2 [32], or the transverse filament component SYCP1 [57], were included in some experiments to monitor synapsis. Asynapsed pachytene nuclei [24] were identified due to the presence of at least one completely synapsed pair of autosomes and at least one incompletely synapsed pair of autosomes exhibiting asynapsis along at least half its length. Nuclei from Spo11-/- and Spo11-/- Tex19.1-/- mice that had complete axial element formation were classified as zygotene-like regardless of the extent of synapsis. For analysis of RAD51 foci in late leptotene oocytes in chromosome spreads from E14.5 foetal ovaries, oocytes with extensive linear SYCP3 staining, indicating axial element formation, and an absence of clear interactions between these axes, were judged to be in late leptotene.
Recombination foci in leptotene and zygotene nuclei were imaged by capturing z-stacks using a piezoelectrically-driven objective mount (Physik Instrumente) controlled with Volocity software (PerkinElmer). These images were deconvolved using Volocity, a 2D image generated in Fiji [58], and analysed in Adobe Photoshop CS6. DMC1, RAD51 and RPA foci were counted as recombination foci when they overlapped a chromosome axis. To measure leptotene γH2AX or H3K4me3 signal intensity, nuclear area was delimited using the DAPI signal, and signal intensity in that area quantified and corrected for background non-nuclear signal in 16 bit grayscale images using Fiji software. To assess the extent of synapsis in zygotene nuclei (S1 Fig, S4 Fig), the total length of completely assembled SC was estimated by SYCP1 or SYCE2 staining and expressed relative to the total length of SYCP3-containing axial/lateral element in that nucleus. For this and all immunocytological scoring, images were scored blind with respect to genotype by pooling control and knockout images, randomly assigning new filenames to each image, then decoding the filenames after scoring.
Decapsulated P16 testes were macerated with razor blades in ice-cold PBS, tissue fragments were removed by allowing to settle, and testicular cells pelleted at 860g for 5 minutes at 4°C. The cells were resuspended in PBS and cross-linking ChIP performed essentially as described [59]. 5µl rabbit anti-histone H3K4me3 antibody (Millipore) was coupled to 20µl Dynabeads-Protein A (Life Technologies) for each ChIP. DNA was purified using MinElute PCR Purification Kits (Qiagen), eluted in 20µl buffer EB, and diluted 1:10 for quantitative PCR (qPCR) using SYBR Select Master Mix (Applied Biosystems). ChIP and input samples from three biological replicates of Tex19.1+/± and Tex19.1-/- P16 testes were assayed in triplicate by qPCR using SYBR Green Master Mix (Roche) and a LightCycler 480 (Roche). ChIP enrichment was calculated relative to 10% input samples, and normalised to enrichment for the β-actin (Actb) transcriptional start site. Primers used for qPCR are listed in S2 Table. Primers for Prdm9-dependent recombination hotspots were derived from DMC1 ChIP-seq data [5].
Histology of Bouin’s-fixed testes, and in situ hybridisation of MMERVK10C probes to Bouin’s-fixed testis sections were performed as described [24].
RNA was isolated from macerated mouse testes using TRIzol (Invitrogen) and treated with Turbo DNAse (Ambion) to digest any genomic DNA contamination. 1µg DNAse-treated RNA was used to synthesise cDNA using Superscript III (Invitrogen). The cDNA was used as a template for qPCR using SYBR Select Master Mix (Applied Biosystems), and the relative quantity of RNA transcript calculated using the standard curve method as described by the supplier. The qPCR was performed on the LightCycler 480 (Roche), retrotransposon RNA levels were measured relative to β-actin, and normalised to control samples. Each biological replicate was assayed in triplicate, and alongside no reverse transcriptase and no template control reactions to confirm the absence of genomic DNA contamination.
P16 testes were homogenised in 2× Laemmli SDS sample buffer (Sigma) with a motorised pestle, boiled for 2–5 minutes and insoluble material pelleted in a microcentrifuge. Lysates were resolved by electrophoresis through pre-cast Bis-Tris polyacrylamide gels (Life Technologies) and Western blotted to PVDF membrane using the iBlot Dry Blotting System (Life Technologies). PBS containing 5% skimmed milk and 0.1% Tween was used to block the membrane and dilute antibodies. Primary antibodies for Western blotting were mouse anti-UBR2 (Abcam, 1:1000 dilution) and mouse anti-β-actin (Sigma, 1:5000 dilution). HRP-conjugated secondary antibodies (Cell Signaling Technology; Bio-Rad) were detected with SuperSignal West Pico Chemiluminescent Substrate (Thermo Scientific).
|
10.1371/journal.pbio.1000254 | A Cost of Sexual Attractiveness to High-Fitness Females | Adaptive mate choice by females is an important component of sexual selection in many species. The evolutionary consequences of male mate preferences, however, have received relatively little study, especially in the context of sexual conflict, where males often harm their mates. Here, we describe a new and counterintuitive cost of sexual selection in species with both male mate preference and sexual conflict via antagonistic male persistence: male mate choice for high-fecundity females leads to a diminished rate of adaptive evolution by reducing the advantage to females of expressing beneficial genetic variation. We then use a Drosophila melanogaster model system to experimentally test the key prediction of this theoretical cost: that antagonistic male persistence is directed toward, and harms, intrinsically higher-fitness females more than it does intrinsically lower-fitness females. This asymmetry in male persistence causes the tails of the population's fitness distribution to regress towards the mean, thereby reducing the efficacy of natural selection. We conclude that adaptive male mate choice can lead to an important, yet unappreciated, cost of sex and sexual selection.
| In many species, females are frequently subject to harassing courtship from males attempting to mate with them. These persistent male behaviors can result in females incurring substantial direct fitness costs. We set out to examine how these costs may influence adaptive potential in a species that also exhibits male mate choice, i.e., a preference by males for females exhibiting certain traits. We found that harmful courtship behaviors were directed predominantly towards females of greater reproductive potential (and away from females of lesser potential), resulting in a reduction in the variation of lifetime reproductive successes among females in the population. This change in distribution of realized fitnesses represents a previously unappreciated consequence of sexual conflict–adaptive male mate preference can slow the rate of accumulation of beneficial mutations and speed the rate of accumulation of harmful mutations, thereby creating a “sexual conflict adaptive load” within a species.
| Historically, most studies of mate choice have focused on mate preference by females, because this sex typically has higher levels of parental investment and lower variance in realized fitness [1]–[4]. Mate choice by males, however, is a common feature of many species [5]–[8], yet its adaptive consequences are far less commonly considered [7],[9], and are typically only studied in species with reversed sex roles [10]–[12]. Here, we focus on species with “typical” sex roles that also experience sexual conflict due to antagonistic male persistence (e.g., unrelenting courtship and repeated mating attempts) that arises because the optimal outcomes of mating interactions often differ for males and females [13],[14]. It is well established that females can suffer substantial fitness costs from receiving too much male attention [15]. For instance, male sexual persistence is an important source of female mortality and/or reduced fecundity in a number of species, e.g., frogs, Crinea georgiana [16]; toads, Bufo bufo [17]; feral sheep, Ovis aries [18]; lizards, Lacerta vivipara [19]; ducks, Anas platyrhynchos [20]; orangutans, Pongo pygmaeus [21]; water striders, Gerris odontogaster [22]; and fruit flies, Drosophila melanogaster [23],[24]. In this study, we focus on a different harmful consequence of females being subject to male persistence that only occurs when males evolve a mate preference for high-fecundity females: a reduced rate of adaptation.
We first develop a graphical model in which a single quantitative trait is a reliable, direct indicator (rather than an indirect indicator, like a costly ornament) of a female's “intrinsic” fecundity (i.e., fecundity in the absence of costly male persistence). For example, in a wide diversity of taxa, variation in female fecundity is strongly correlated with body size [2],[4] because larger females have more resources to invest in fecundity. Henceforth, we arbitrarily assume that a female's body size is the phenotypic trait correlated with fecundity, but our logic applies to other indicator traits that directly influence her fecundity, such as parasite load [25] or abdomen size in many insects [7]. Males are expected to evolve a mating preference for larger females whenever this preference increases their own lifetime reproductive success [14]. Such an adaptive male mate preference will cause larger, intrinsically high-fecundity females, to receive more antagonistic male persistence, compared to smaller, intrinsically low-fecundity females. The fitness consequences of this relationship will depend upon how female resistance to male-induced harm scales with the indicator trait (in this case, body size). Assuming that a female's resistance to the harmful male persistence does not rise sufficiently fast with increasing body size, the male preference should reduce the fecundity of large females and increase that of small females, thereby reducing the standing variance in fitness (Figure 1). As a result, the selective advantage of any beneficial genetic variation that makes females more competitive for limiting resources, and hence more fecund, will experience a smaller selective advantage than if harmful male persistence was randomly applied to females throughout the population. Such nonrandom male persistence will cause adaptive evolution in females to be slowed whenever female fecundity is: (i) heritable, (ii) genetically correlated with the indicator trait, and (iii) a major determinant of her lifetime fitness that does not strongly trade off with her other fitness components. A reduced rate of adaptive evolution by females can also be deduced from Fisher's fundamental theorem [26], so long as the male-induced reduction in the phenotypic variation in female fecundity also leads to a reduction in the additive genetic variation among females. Furthermore, when there is a positive genetic correlation for fitness between females and males, adaptive male mate choice is expected to reduce the rate of adaptation in both sexes. Although a counteracting effect could occur if male preference for high-fecundity females increases the variance in male fitness, or if male preferences lead to positive assortative mating for fitness, here we focus on female fitness and the potential for male mate preferences to reduce its heritable variation.
The conclusion that adaptive male choice leads to a reduced rate of adaptation by females can also be deduced by focusing on mutations at a single arbitrary locus. Let the mutation rate to new beneficial mutations be UBen and the selective advantage of the mutation, expressed as a selection coefficient and averaged across the sexes, be s. Assuming approximate additivity (i.e., little dominance), the probability of the mutation becoming fixed can be calculated using the diffusion approximation [27] as 2s(Ne/N), where N is the population size and Ne is the effective population size. With recurrent mutation to new beneficial mutations, the rate of advance of adaptive evolution is approximated by:(1)(2)
If we partition selection between the sexes and let the selective advantage of a mutation be s♂ in males and s♀ in females then,(3)(4)
Next, we assume that the expression of the beneficial mutation also increases the attractiveness of females to males (e.g., as a result of increasing her body size), so that those females expressing the beneficial mutation receive an excess of antagonistic male persistence. We express this cost with an additional selection coefficient s♀biased-persist, which is applied only to females,(5)
Comparison of Equations 4 and 5 demonstrates that the rate of adaptive evolution will always be slower whenever males bias their antagonistic persistence towards fitter females and cause s♀biased-persist to be negative; i.e., when there is adaptive male mate choice and increased male persistence is harmful to females. Increased male persistence directed towards more fecund females that express the beneficial allele reduces the selective advantage of those females and thereby reduces the variance in fitness among females in the population.
The primary prediction from our models is that, in species with antagonistic male persistence, adaptive male mate preference leads to a “cost of being an attractive female.” This cost reduces the selective advantage of females expressing more beneficial genetic variation (and hence are larger, on average) and increases the fitness of females expressing less of this variation (and hence are smaller, on average). Put more simply, adaptive male mate preference causes the tails of the population's distribution of female lifetime fecundity to regress towards the mean (Figure 1). This prediction is contingent on four assumptions that must be met in order for our model to operate: (i) lifetime fecundity and net fitness are strongly genetically correlated, (ii) body size and fecundity are positively correlated, both phenotypically and genetically, (iii) more antagonistic male persistence is directed towards females with higher intrinsic fecundity (i.e., potential fecundity in the absence of costly male persistence), and (iv) female resistance to male-induced harm does not rise sufficiently fast with increasing body size. We tested the major prediction of the model, and its underlying assumptions, using a laboratory population of the model species D. melanogaster. In this population, assumption (i) is well established [28],[29], so here we focus on testing whether our population meets assumptions ii–iv, before experimentally assessing whether male mate preference for high-fitness females causes the tails of the distribution of lifetime fecundity to regress towards the mean.
Joint measures of female body size and lifetime fecundity in our base population (LHM) of D. melanogaster indicated that these two phenotypic traits are strongly correlated. As predicted from past studies of many taxa [2],[4],[30], fecundity was higher in large females compared to small females (Figure 2). This result was found both when females experienced minimal exposure to males (mean ± standard error [SE]: large females, 27.3±1.59; small females, 18.42±1.07; t-test t = 4.64, df = 98, p<0.0001; p-values reported throughout the manuscript are two-tailed) and when male exposure was continuous (large females, 16.5±1.06; small females, 12.62±0.72; t-test t = 3.01, df = 98, p<0.003). We tested for a genetic correlation between body size and fecundity in a separate study in which two populations each were artificially selected for either large or small body size. After 83 generations of divergent selection, lifetime fecundity was significantly higher in the lines selected for large body size compared to the lines selected for small body size (mean ± SE: large females, 31.38±1.81; small females, 15.25±1.58; t-test t = 6.69, df = 2, p = 0.02). Since body size was the only target of artificial selection, the large divergence in fecundity between treatments demonstrates a strong positive genetic correlation between body size and fecundity, a result consistent with other research [31].
To test this assumption, we first measured how the persistence (courtship behaviour) of individual males was allocated between two nonvirgin females (differing in eye colour phenotype, brown or red, for ease of individual identification). We performed a two-way ANOVA on the amount of persistence behaviour directed towards each female, with the body size of that “target” female (large or small), the body size of the competitor female present in the test tube (large or small), the eye colour of the target female (red or brown), and all possible interactions as predictor variables. This analysis was significant overall (F7,232 = 7.50, p<0.0001), with significant effects of the target female body size (F1,232 = 38.37, p<0.0001) and the body size of the competitor female (F1,232 = 13.53, p = 0.0003), but no effects of eye colour (F1,232 = 0.41, p = 0.52), or any of the interactions (all p>0.60). When individual males were housed with two nonvirgin females differing in body size, males directed more persistence towards the larger female than towards the smaller female (paired t-tests, p≤0.0002, Figure 3). When males were housed with two nonvirgin females of similar body size (both small or both large), the levels of male persistence directed towards the red- and brown-eyed females were not significantly different (paired t-tests, p≥0.50, Figure 3). Further evidence of a male mate preference for nonvirgin females of larger body size was obtained from mating assays conducted under conditions that more closely mimicked the normal culture environment of the LHM population (16 males combined with 16 females during the “adult competition” phase of the life cycle [28],[32]). When presented with a choice of nonvirgin females differing in body size, males mated with large-bodied females at a greater rate than with small-bodied females (generalized linear model [GLM] with binomial error terms; pconsensus = 1.17×10−5, Replicate 1: χ21,18 = 5.90, p = 0.015; Replicate 2: χ21,74 = 22.87, p<0.0001; Figure 3). These remating results are unlikely to have arisen from large females possessing a greater receptiveness to male courtship effort because males kept under “no-choice” mating conditions (where either only large or only small nonvirgin females were present) mated with small females more frequently than with large females (GLM with binomial error terms; pconsensus<1×10−6, Replicate 1: χ21,28 = 14.79, p = 0.0001; Replicate 2: χ21,28 = 7.16, p = 0.0075; Figure 4).
To test this assumption, we compared the reduction in lifetime fecundity of large and small females when they were either minimally or continuously exposed to males (Figure 2). Continuous male exposure harmed large females more than small females (two-way ANOVA, interaction between body size and male exposure, F1,196 = 4.75, p = 0.031), indicating that larger females were not more resistant to the harmful male persistence that they received.
Finally, we tested the model's key prediction by comparing the mean fecundities of large- and small-bodied females used in the choice and no-choice mating assays described earlier. In the no-choice assays, where all females were either of large or small body size, male preference for large female body size could not cause them to direct their antagonistic persistence away from smaller females and towards larger females. In contrast, in the choice assays, where females of different body sizes were simultaneously present, a redirection of antagonistic male persistence towards larger females was possible. We found that the difference in the mean fecundities of large- and small-bodied females was smaller when males could direct their antagonistic persistence towards large females (Figure 5, pconsensus = 0.012, interaction tests for each replicate: F1,46 = 2.79, p = 0.1 for the smaller, first replicate; F1,101 = 4.92, p = 0.03 for the larger, second replicate).
The results of our male mate preference tests clearly demonstrate that males have mate preferences for larger nonvirgin females—a result consistent with earlier work on virgin females [33]. Rather than displaying an “undiscriminating eagerness” [34] to mate, when given a choice between females differing in body size, male D. melanogaster preferred to court and mate with large, high-fecundity females over small, low-fecundity females. Given the significant fecundity differences associated with female body size described above, this mate preference is likely to be adaptive from the male's perspective, as mating with larger, more fecund females is likely to yield greater direct, as well as indirect [35], benefits. It is unlikely that this male mate preference is adaptive from the female's perspective, as several studies have established that chronic male persistence in the LHM population is very harmful to females, and it is not sufficiently compensated by indirect genetic benefits [36]–[38]. Our experiments demonstrate that larger females receive more harmful male persistence but do not reveal the specific mechanism by which this harm accrues. Further work will be needed to resolve the degree to which this increased harm is due to harassment during courtship [24], damage associated with copulation [39], and/or the activity of products transferred in the male's seminal fluid [23],[40].
Having experimentally ascertained that the LHM population of D. melanogaster satisfied all the assumptions necessary in which to test the key prediction our model, we were able to meaningfully assess the fitness consequences of adaptive male mate preferences. When males had the ability to bias their antagonistic persistence towards large-bodied females, we saw a decrease in the mean fecundity of these preferred females, compared to those large females that were in an experimental environment where all females were of similar size, and biases of antagonistic male persistence were not possible. In contrast, small-bodied females were, on average, able to realize relatively higher fecundities when they were housed with larger females (which, our study indicates, were attracting more harmful male persistence) than they were when they were housed in an environment in which males had no other choice of mates. Although our study found that males directed more courtship towards large females and also mated them more frequently, both of which can be harmful in and of themselves [24], the observed cost to large females might also have occurred because large females were mated, on average, to more harmful males [30],[41] than were smaller females. Irrespective of the mechanism of this cost, together these assays revealed how male mate preferences will ultimately cause the tails of the distribution of fecundity to regress towards the mean. Since adult lifetime fecundity is strongly correlated with lifetime fitness in females of the LHM population [42], this male-driven sexual selection is expected to reduce the rate of adaptive evolution of any trait that is positively correlated with female body size. It is common for deleterious mutations to reduce body size in D. melanogaster [43], and it is reasonable to assume that many beneficial mutations will cause their carriers to be more competitive as larvae, allowing them to garner more resources during the larval competition phase of their life cycle and become larger, more fecund, adults. As a consequence, male mate preference for larger females is expected to commonly interfere with both progressive evolution and to increase the population's mutational load by interfering with purifying selection. For example, suppose that environmental change led to selection for alleles conferring higher desiccation tolerance. If more desiccation-tolerant females had a competitive advantage such that they grew to a larger size prior to reproduction (e.g., [44]), then a male preference for these females would reduce their relative fecundity and increase that of smaller, less desiccation-tolerant females. As a result, the population may be less responsive to environmental change, become an inferior competitor species, and be at a greater risk of extinction.
Collectively, our results support our model's key prediction that male mate preference for high-fitness females reduces the selective advantage of larger, more fecund females and increases that of smaller, less fecund females. This finding, obtained in a laboratory population, is likely to apply to natural populations for two reasons. First, the study was done on a large, outbred population that has been maintained in a competitive laboratory environment, at continuous large size, for over 400 generations [28],[32]. Over this period of time, the opportunity for adaptation to the laboratory environment should have been substantial, permitting the flies to be experimentally assayed under conditions to which they are highly adapted. Second, we measured natural variation in body size, rather than inducing extreme body size variation via nutritional deprivation and/or excessive larval crowding. This was accomplished using a sieve shaker device (developed by ADS and WRR), which enabled us to quickly sort thousands of adult flies based on natural variation in their body size, and obtain the largest and smallest individuals to use in our experiments. Flies from these two body size groups differed markedly in fecundity, with the larger females producing over 30% more eggs than small females under both minimal and continuous male exposure conditions. Although our assays of male mate preference support a directional preference for large-bodied females, in one assay (Figure 2), males could only choose between females of large and small body size. Thus, there is the possibility that the true male preference function favours females of intermediate size. However, in our second assay (Figure 3), males were able to choose between large or small females versus random females (average), and these data support the conclusion that male preference is monotonic for larger females.
Our model of adaptive male mate choice in the context of harmful male persistence has important limitations. First, we have implicitly assumed that the increased male persistence (directed toward larger, more intrinsically fecund females) does not cause larger females to have lower than average fecundity. Second, male condition may be more variable in nature compared to the laboratory, and condition-specific patterns of male persistence could either enhance or reduce the bias of male persistence toward larger females. Third, we have ignored complicating factors such as size-assortative mating interactions, e.g., smaller females receiving persistence predominantly from smaller or poor-condition males. Fourth, we have assumed that male mate choice is based on a female trait that directly influences her fecundity, such as body size. Theory predicts that this type of male mate preference will lead to a monotonic preference for larger females [45],[46]. However, when the preferred female trait is a costly indicator of fecundity, such as an energetically expensive ornament, then males can evolve to prefer intermediate trait values in females [45],[46], and our model would not apply. Fifth, our model may not apply to species where females obtain direct net benefits from increased mating rates, such as those with nuptial feeding [47]. Lastly, we have assumed a static male preference and female indicator trait. In many contexts, these two traits can be expected to coevolve, and this dynamic is not included in our model. Nonetheless, our empirical work suggests that the requisite conditions for the model to operate, at least transiently, can feasibly be achieved.
Our finding of harmful effects of adaptive male mate choice represents a previously unappreciated cost of sexual reproduction in species with antagonistic male persistence. Rather than simply showing that male-induced harm reduces overall female fecundity, we have shown that biases in the distribution of this harm among mates reduces the selection differential between females with intrinsically high and low fecundity. This reduced efficacy of natural selection will retard a population's rate of adaptive evolution and increase both its equilibrium mutational load and its stochastic accumulation of harmful mutations. The cost of adaptive male mate choice, however, only applies when males can reliably ascertain a female's fecundity using a trait that is heritable and correlated with heritable fitness variation. In Drosophila, female body size represents such a trait since it is influenced by both genotype [48],[49] and a number of environmental factors (including temperature, nutrition and larval crowding conditions [50]), and responds rapidly to directional selection. In species with little or no heritability for body size, however, an adaptive cost of male preference for high-fecundity females would not apply. Nonetheless, given the prevalence of male mate preferences [7], this new cost that we describe may be a widespread evolutionary phenomenon. For this reason, it should be considered in the broader context of the ongoing debates over the interfering or reinforcing role that sexual selection plays in the process of adaptation, and whether sexual selection increases or decreases the risk of extinction of populations and species [51].
For all male–female interaction assays, we used D. melanogaster adults obtained from the wild-type LHM population [28],[29] or from a replicate population (LHM-bwD) in which a dominant brown-eyed marker (bwD) had been introgressed through repeated backcrossing into the LHM genetic background. The LHM population is maintained on a 14-d culture cycle with a 12-h L∶12-h D diurnal cycle at 25°C in humidity-controlled incubators. Briefly, each generation begins with eggs placed in 56 “juvenile competition” vials (150–200 eggs per vial; each vial containing 10 ml of cornmeal/molasses medium). After 11.25 d, emerging adults are lightly anesthetized with CO2, mixed among vials, and transferred to “adult competition” vials (16 pairs of males and females per vial), which are seeded with 6.4 mg (dry weight) of live yeast. After 2 d of adult competition, the flies are transferred to “oviposition” vials, and then discarded after laying eggs for 18 h. The eggs laid in these oviposition vials are culled to a density of 150–200 eggs per vial and become the “juvenile competition” vials of the next generation. Because only eggs from the oviposition phase of the life cycle are used to propagate the next generation, and populations have been consistently maintained under these culture conditions for over 400 generations, the number of eggs laid during the 18-h oviposition phase represents a meaningful measure of lifetime fecundity in these populations. As such, experiments were designed to mimic these culture conditions as closely as possible. Detailed culturing protocols for these large populations (adults n>1,800 per generation for LHM and n>1,300 per generation for LHM-bwD) can be found elsewhere [28],[29]).
We altered the quality of potential female mates by collecting females of differing adult body size, a phenotypic trait that is frequently positively correlated with fecundity [2],[30],[52]. We collected flies from the ends of the normal distribution of body sizes that are produced under typical lab culture conditions. Flies were sorted by size with the use of a sieve shaker device (Gilson Performer III, Gilson Company) which mechanically separates anesthetized flies on the basis of their ability to pass through a series of 20 electroformed sieves, in which the diameter of the holes in each sieve was 5% larger than the diameter of the holes of the sieve below (diameter of top sieve holes = 1,685 µm; diameter of bottom sieve holes = 800 µm). Flies were placed into the column (under light CO2 anaesthesia), and were agitated at a rate of 3,600 vibrations min−1 for 2 min. By using this technique, it was possible to quickly sort hundreds of flies simultaneously on the basis of their body size. For all experiments, “small” flies were defined as those that were small enough to pass through the 1,095-µm diameter sieve, whereas “large” flies were those that were too large to pass through the 1,281-µm diameter sieve.
To assess the phenotypic correlation between body size and fecundity, we collected adult flies from the LHM population as they eclosed as virgins on day 9 of their life cycle. Flies were separated by sex, and on the following day, females were sorted by size using the sieve sorter protocol described above. One hundred female flies each of large and small body size were then placed individually (under light anaesthesia) into small test tubes that had been seeded with 0.4 mg of yeast (the amount of yeast per female experienced under normal culture conditions). Into each of these vials, three adult males were placed for a period of 2 h, during which time all virgin females were observed to have mated once. Males were then removed randomly from half of the vials to create 50 adult competition vials with minimal male exposure and 50 with continuous male exposure for each female body size category. Maintaining flies under these two conditions allows us to confirm that there is an intrinsic difference in fecundity between females of different sizes that is independent of the negative net fitness effects of continuous male presence. Matching the normal culturing protocol of the flies, vials were returned to the incubator for an additional 2 d, at which time flies were transferred to oviposition vials containing fresh medium (with a scored surface to encourage oviposition) for a period of 18 h before being discarded. The number of eggs laid in each vial was counted, and mean fecundities were compared using t-tests for females differing in size in each male-exposure treatment. The complete dataset was also used to test the assumption that female resistance to male-induced harm does not rise sufficiently fast with increasing body size, by examining whether or not female flies of one size were harmed more by continuous male exposure.
In order to verify that there was a genetic correlation between body size and fecundity, we assessed the fecundity of females obtained from populations of D. melanogaster that are part of an ongoing experimental evolution project (of ADS and WRR) in which females had been artificially selected for either large or small body size using a size-sorting procedure similar to that described above. These populations are otherwise cultured in a manner similar to the LHM population from which they were all originally derived. At the time of the assay, the artificial selection had been operating for 83 generations in each of two replicate populations per treatment, and there had been considerable divergence in body size (mean female diameter [µm] ± SE: large treatment, 1,218.5±37.67; small treatment, 786.7±43.5; t-test t = 7.51, df = 2, p<0.01). For this assay, 72 virgin females were obtained at random from each of the four experimental populations. On day 11 of their life cycle, these females were placed in adult competition vials in groups of 16, along with 16 males taken randomly from the LHM population, for a period of 2 h, during which time all females were observed to have mated once. Males were removed from the vials, and after 2 d in the incubator, females were transferred to individual oviposition vials containing fresh medium (with a scored surface) for a period of 18 h before being discarded. The number of eggs laid in each vial was counted and the mean fecundity of the two replicates of each treatment was compared using a t-test (with population as the unit of replication). Since the selected trait in these experimental populations was body size, any consistent change in fecundity between the two treatments must be due to a genetic correlation between the two traits.
In order to test whether males have mate preferences, a series of behavioural assays were conducted. Nonvirgin flies from both the LHM and LHM-bwD populations were collected on day 11 of their life cycle, and females were sorted by size to isolate large- and small-bodied individuals. Pairs of female flies differing in eye colour (to aid individual identification) were placed into small, adult competition vials (test tubes) in all possible combinations of body size. After a 1-h anaesthesia-recovery period, a single unanaesthetized adult male fly was added to each test tube, which were then placed on their sides in a well-lit room. Over the course of 11 sessions, spaced 40 min apart, the male in each test tube was observed. Male persistence behaviour was defined as being located within 5 mm of a female and oriented towards her [53]–[55]. Data on the frequency of the male persistence behaviour was collected for each type of female in each treatment. A total of 30 replicate test tubes per treatment were scored.
In these assays, nonvirgin adult female LHM flies were collected on day 11 of their life cycle and sorted by size (see above) to isolate large and small individuals. Females were then placed into one of two types of adult competition vials (a vial containing fresh medium seeded with 6.4 mg of live yeast). In the first, choice experiment, either eight large or eight small red-eyed LHM females were placed into an adult competition vial along with eight randomly collected LHM-bwD females and 16 LHM-bwD males. In the second, no-choice treatment, either 16 large or 16 small red-eyed LHM adult females were placed into an adult competition vial along with 16 LHM-bwD males. These vials were kept in the incubator (on their sides) for 24 h, at which time males were removed. The vials, containing females only, were then returned to the incubator for an additional 24 h. Remating rates were assayed by placing all females into individual oviposition vials (test tubes) containing fresh, scored medium for the purpose of measuring the paternity of her offspring. Eighteen hours later, the adult flies were discarded, and the test tubes containing eggs were incubated for 11 d. At this time, the presence and number of red-eyed and brown-eyed progeny in each brood were scored to ascertain whether the female had remated. The proportion of females in each adult competition vial that produced brown-eyed offspring (indicating a remating event) was recorded. To examine remating rates in relation to female body size and treatment, we constructed GLMs that used a logit link function and binomial error distribution, where the number of females that remated is the dependent variable and the total number of females assayed is the binomial denominator. We tested whether male mate preferences caused the tails of the distribution of female lifetime fecundity to regress towards the mean by performing a two-way ANOVA, with body size, remating treatment, and their interaction as predictor variables. A significant interaction term (that was associated with a smaller difference between the mean fecundity of large and small females when male preference was possible) would indicate that the tails of the fecundity distribution had regressed toward the mean. Each type of remating assay was repeated twice. The first, choice assay was comprised of ten adult competition vials (the unit of replication) for each body size treatment, whereas the second replicate was comprised of 38 adult competition vials in the large body size treatment and 37 in the small body size treatment. Both replicates of the no-choice assay were comprised of 15 adult competition vials for each body size treatment.
|
10.1371/journal.pmed.1002179 | Genetic Predisposition to an Impaired Metabolism of the Branched-Chain Amino Acids and Risk of Type 2 Diabetes: A Mendelian Randomisation Analysis | Higher circulating levels of the branched-chain amino acids (BCAAs; i.e., isoleucine, leucine, and valine) are strongly associated with higher type 2 diabetes risk, but it is not known whether this association is causal. We undertook large-scale human genetic analyses to address this question.
Genome-wide studies of BCAA levels in 16,596 individuals revealed five genomic regions associated at genome-wide levels of significance (p < 5 × 10−8). The strongest signal was 21 kb upstream of the PPM1K gene (beta in standard deviations [SDs] of leucine per allele = 0.08, p = 3.9 × 10−25), encoding an activator of the mitochondrial branched-chain alpha-ketoacid dehydrogenase (BCKD) responsible for the rate-limiting step in BCAA catabolism. In another analysis, in up to 47,877 cases of type 2 diabetes and 267,694 controls, a genetically predicted difference of 1 SD in amino acid level was associated with an odds ratio for type 2 diabetes of 1.44 (95% CI 1.26–1.65, p = 9.5 × 10−8) for isoleucine, 1.85 (95% CI 1.41–2.42, p = 7.3 × 10−6) for leucine, and 1.54 (95% CI 1.28–1.84, p = 4.2 × 10−6) for valine. Estimates were highly consistent with those from prospective observational studies of the association between BCAA levels and incident type 2 diabetes in a meta-analysis of 1,992 cases and 4,319 non-cases. Metabolome-wide association analyses of BCAA-raising alleles revealed high specificity to the BCAA pathway and an accumulation of metabolites upstream of branched-chain alpha-ketoacid oxidation, consistent with reduced BCKD activity. Limitations of this study are that, while the association of genetic variants appeared highly specific, the possibility of pleiotropic associations cannot be entirely excluded. Similar to other complex phenotypes, genetic scores used in the study captured a limited proportion of the heritability in BCAA levels. Therefore, it is possible that only some of the mechanisms that increase BCAA levels or affect BCAA metabolism are implicated in type 2 diabetes.
Evidence from this large-scale human genetic and metabolomic study is consistent with a causal role of BCAA metabolism in the aetiology of type 2 diabetes.
| Higher circulating levels of isoleucine, leucine, and valine, i.e., the branched-chain amino acids (BCAAs), are strongly associated with the risk of future type 2 diabetes.
It is not known if this association reflects a causal relationship.
It is important to assess the aetiologic nature of this relationship. If it is causal, then intervening on BCAA levels or metabolism may reduce the risk of diabetes.
We used a human genetics framework known as “Mendelian randomisation” to study this question. Mendelian randomisation postulates that if a biomarker is causally implicated in a disease, then genetic variants specifically associated with that biomarker should also be associated with the disease.
In a meta-analysis of 1,992 incident cases of type 2 diabetes and 4,319 non-cases, we found strong associations between higher levels of each of the BCAAs and a higher risk of type 2 diabetes.
In a genome-wide meta-analysis of 16,596 individuals, we identified five genomic regions where common genetic variants were associated with BCAA levels.
In a meta-analysis of genetic association studies including 47,877 cases of type 2 diabetes and 267,694 controls, a genetically predicted difference of one standard deviation in amino acid level was associated with an odds ratio of type 2 diabetes of 1.44 (95% confidence interval 1.26–1.65) for isoleucine, 1.85 (1.41–2.42) for leucine, and 1.54 (1.28–1.84) for valine.
Evidence from this large-scale human genetic and metabolomic study is consistent with a causal role of BCAA metabolism in the aetiology of type 2 diabetes.
Possible limitations of the study included the possibility of non-specific associations of the genetic variants included in the study (i.e., “pleiotropy”) and the relatively low proportion of heritability in BCAA levels explained by the identified genetic variants.
| Early evidence of impaired branched-chain amino acid (BCAA) metabolism in insulin resistance and obesity dates back to more than 40 years ago [1]. More recently, in investigations of the human metabolome, multiple research groups have described associations of higher levels of isoleucine, leucine, and valine with insulin resistance, obesity [2], and a higher risk of future type 2 diabetes [3]. These associations have since been replicated in several studies [4–6] and are corroborated by a growing body of experimental evidence [2,7–10]. However, it remains unknown whether BCAA metabolism is causally implicated in type 2 diabetes [7] or whether observational associations simply reflect reverse causality, whereby differences in BCAA levels are a consequence of pathophysiological processes that precede the development of the disease [11].
Genetic approaches to causality assessment provide an opportunity for rapid and cost-effective prioritisation of targets for interventional studies [12–20]. Therefore, genetic variants associated with BCAA levels can be used to study the aetiologic links of the BCAA metabolic pathway with type 2 diabetes. A previous smaller-scale investigation that was limited to one genetic variant associated with 3-methyl-2-oxovalerate, a by-product of BCAA metabolism, was inconclusive [21].
Therefore, in this study, we used genome-wide association studies (GWASs) coupled with large-scale metabolomic measurements to investigate the aetiologic relationship between BCAA metabolism and type 2 diabetes, an issue of high clinical and public health relevance due to the strength of the associations between BCAA levels and type 2 diabetes [2,3] and the global burden of this condition [22,23].
We adopted a Mendelian randomisation approach to evaluate possible causal relationships between BCAA metabolism and type 2 diabetes (Fig 1). Inherited DNA variants are randomly assigned during meiosis and remain fixed throughout the lifetime. Therefore, they are unlikely to be affected by disease processes (i.e., reverse causality) and non-genetic confounding. Mendelian randomisation exploits these properties in order to estimate the relationship of modifiable risk factors with disease outcomes without the limitations of traditional observational epidemiology [12–15]. This approach postulates that if a biomarker is causally implicated in a disease, then genetic variants associated with that biomarker (but not other disease risk factors) should be associated with the disease in the direction predicted by observational analyses [12–15].
Therefore, we conducted meta-analyses of (a) GWASs of plasma levels of isoleucine, leucine, and valine; (b) studies of the associations of BCAA-raising genetic variants with type 2 diabetes; and (c) prospective studies of the associations of baseline isoleucine, leucine, and valine levels with incident type 2 diabetes.
The investigations described in this study were approved by local ethical committees, and participants provided their informed consent.
We performed GWASs of the plasma levels of isoleucine, leucine, and valine in the Fenland study (n = 9,237; S1 Text; S1 Table). Fenland GWAS results were then meta-analysed with publicly available results from a meta-analysis of the KORA and TwinsUK studies [24]. The total sample size of the genome-wide meta-analysis was 16,596 individuals.
We estimated the association with type 2 diabetes of the lead single nucleotide polymorphism (SNP) at each BCAA-associated genomic locus. We meta-analysed SNP association results from the DIAbetes Genetics Replication And Meta-analysis [25] and the EPIC-InterAct [26], GoDARTs [21], and UK Biobank studies [27] (S1 Text; S2 Table). The total sample size of this analysis was 47,877 type 2 diabetes cases and 267,694 controls.
We conducted a systematic review of the literature of prospective studies of the association of BCAA levels and incident type 2 diabetes (S1 Text). The results of five prospective population-based studies [3–6] were meta-analysed with the unpublished results of the EPIC-Norfolk case-cohort study using fixed effect models (S3 Table) [28]. The I-squared statistic was used to quantify heterogeneity. In the EPIC-Norfolk study, the association of metabolite levels with incident type 2 diabetes was estimated using multivariable Cox proportional hazards regression with Prentice weighting and robust standard errors. The total sample size of the observational analysis was 1,992 incident cases of type 2 diabetes and 4,319 non-cases.
In the Fenland study, BCAA levels were measured using liquid chromatography coupled with tandem mass spectrometry (S1 Fig; AbsoluteIDQ p180 Kit, Biocrates Life Sciences [29]). In the KORA and TwinsUK studies, BCAA levels were measured as part of untargeted metabolomic measurements using gas chromatography mass spectrometry or ultra-performance liquid chromatography coupled with tandem mass spectrometry (UPLC-MS/MS) [24]. In the EPIC-Norfolk study, BCAA levels were measured by UPLC-MS/MS (S1 Fig; DiscoveryHD4 platform, Metabolon [30]). This platform also measured an additional 15 metabolites related to BCAA metabolism. In the Southall And Brent REvisited Study (SABRE), the levels of BCAAs were measured by nuclear magnetic resonance at 0 and 120 min during the course of an oral glucose tolerance test (OGTT) [31,32].
Genome-wide association analyses were conducted using SNPTEST v2 [33], and study results were meta-analysed using METAL [34], which was chosen as the analysis software because studies included in the meta-analysis were focused on individuals of European ancestry. All genetic association results are reported per allele and assuming an additive effect. In the meta-analysis of GWASs of BCAA levels, we meta-analysed Z-scores rather than betas and standard errors in order to minimise the influence of the different platforms used to measure amino acid levels in participating studies. The conventional threshold for genome-wide statistical significance (p < 5 × 10−8) was used to identify associated loci. The SNP with the lowest p-value within a 1 million base-pair window was chosen as the lead SNP at a given genomic locus.
For each amino acid, we estimated the variance in amino acid level explained by the identified lead SNPs using linear regression models in the Fenland study. We used the BOLT-REML algorithm of the BOLT-LMM v2.2 software [35] to estimate chip-based heritability. This was estimated directly from variants captured across the genome on the genome-wide genotyping chip. We also used estimates of family-based genetic heritability from twin studies from Shin and colleagues [24]. We calculated the proportion of heritability explained by the lead SNPs by dividing the explained variance estimates by the familial and chip-based heritability estimates. Using genome-wide association results and LD score regression [36], we estimated genetic correlations of BCAA levels with type 2 diabetes, continuous glycaemic traits (fasting glucose, fasting insulin, glucose at 2 h in a 75-g OGTT, HbA1c, HOMA-B, and HOMA-IR), and anthropometric traits (body mass index [BMI] and waist-to-hip ratio) on the LD Hub platform (http://ldsc.broadinstitute.org/#; accessed 6 October 2016) [37]. Isoleucine estimates were not generated by the software due to low levels of genetic heritability estimated by the LD Hub platform.
We estimated the association between a genetically predicted difference of 1 standard deviation (SD) in isoleucine, leucine, or valine plasma level and type 2 diabetes risk. For each of the BCAAs, we constructed weighted genetic risk scores including the lead SNP (independent genetic variants analysis) or genome-wide significant SNPs in imperfect linkage disequilibrium (r2 < 0.8 for all pairwise SNP comparisons; correlated genetic variants analysis) at each locus identified by the genome-wide meta-analysis for that metabolite. Fenland study weights were used to scale the effect of each genetic score to 1 SD. We used Fenland study weights as this was the largest study and we had access to individual-level data, allowing the standardisation of genetic effect estimates. Lead SNPs (nearest gene) included in the genetic scores used in the independent variants analysis were as follows: rs7678928 (PPM1K), rs75950518 (DDX19A), rs58101275 (TRMT61A), and rs1420601 (CBLN1) for isoleucine; rs1440581 (PPM1K) for leucine; and rs1440581 (PPM1K) for valine. Analyses of correlated genetic variants modelled the estimates of 12 SNPs at four loci (PPM1K, DDX19A, TRMT61A, and CBLN1) for isoleucine, seven SNPs at the PPM1K locus for leucine, and eight SNPs at the PPM1K locus for valine (details in S1 Text).
We combined estimates of “SNP to metabolite” and “SNP to type 2 diabetes” associations to calculate estimates of each “genetically predicted metabolite level to type 2 diabetes” association [13,14]. Estimates of multiple SNPs contributing to a given genetic score were pooled using an inverse-variance-weighted method [13,14]. For the analysis of correlated genetic variants, estimates were pooled with a weighted generalised linear regression method that accounts for the correlation between genetic variants [13]. The correlation values were obtained using the SNAP software [38]. Results were scaled to represent the odds ratio [OR] per 1-SD genetically predicted difference in amino acid level. Given that Mendelian randomisation assumes no pleiotropic effect beyond that on the risk factor of interest (i.e., BCAA levels), we excluded the lead SNP (rs1260326) at the known pleiotropic [39–43] GCKR locus from the isoleucine genetic score (Section 1 of S2 Text).
In order to assess the specificity of the genetic scores, we performed large-scale association testing of BCAA-raising alleles with the levels of blood metabolites: 175 metabolites in the Fenland study, 453 metabolites in KORA and TwinsUK, and 18 metabolites of the BCAA pathway in the EPIC-Norfolk study (S1 Text). We also investigated the association of BCAA-raising alleles with cardiometabolic traits in large-scale GWAS meta-analyses of up to 328,036 individuals [39–43].
We investigated other potential links between BCAA levels and type 2 diabetes. We tested the association with BCAA levels of a genetic predisposition to higher BMI, greater insulin resistance, and impaired insulin secretion using previously validated unweighted genetic scores for these traits [44–46] (S4 Table). Using linear regression models, we estimated the association of fasting insulin with BCAA levels during the course of an OGTT in the SABRE study.
We investigated changes in levels of PPM1K gene expression in skeletal muscle during the course of an OGTT. Fifty age-matched men with either normal glucose tolerance (n = 25) or type 2 diabetes (n = 25) were recruited at the Karolinska Institutet, Sweden. Following an overnight fast, a skeletal muscle biopsy was taken from the vastus lateralis muscle under local anaesthesia with a Weil-Blakesley conchotome tong instrument (Agntho’s). Participants ingested a standardised solution containing 75 g of glucose, and 2 h later a second biopsy was taken from the vastus lateralis of the contralateral leg. Biopsies were frozen immediately and stored in liquid nitrogen until processed. Total RNA was extracted from biopsies using the mirVana miRNA Isolation Kit (Thermo Fisher). Equal amounts of total RNA were used to synthesize cDNA using random primers and the High Capacity cDNA Reverse Transcription Kit (Thermo Fisher). Quantitative PCR was performed with a ViiA 7 Real-Time PCR System with Fast SYBR Green Master Mix (Thermo Fisher). The ViiA 7 Software (version 1.1) was used to determine threshold cycle (Ct) values, and relative gene expression was calculated with the comparative Ct method relative to a reference gene, with dCt being the difference in expression between the gene of interest and the reference gene. The most suitable reference genes were determined to be GUSB, RPLP0, and TBP (out of four measured genes) using the NormFinder algorithm (Multid Analyses) [47]. Statistical analysis was performed on dCt values since these were normally distributed.
Both Mendelian randomisation and observational association estimates are reported per 1-SD increase in metabolite level. Standardisation was necessary in light of the different platforms used to measure BCAA levels. Analyses were conducted using STATA v13.1 (StataCorp) and R (https://cran.r-project.org/). Further details of the methods used in this study are reported in S1 Text.
Genome-wide meta-analyses of 10.5 million genetic variants in 16,596 individuals revealed five independent genomic loci associated with BCAA levels at the genome-wide level of statistical significance (Table 1; Figs 2, S2 and S3). The strongest signal was 21 kb upstream of the PPM1K gene on Chromosome 4q22.1, a locus previously reported for BCAA levels [24,48] (Fig 2).
PPM1K encodes the mitochondrial phosphatase that activates the branched-chain alpha-ketoacid dehydrogenase (BCKD) complex [49–51]. This catalytic complex is responsible for the rate-limiting step of BCAA catabolism, i.e., the irreversible oxidative decarboxylation of branched-chain alpha-ketoacids [49–51]. In addition to PPM1K, we identified four novel loci associated with isoleucine (Table 1; Fig 2). Lead SNPs at these loci were also associated with the levels of the other two BCAAs in a consistent direction, albeit not at the genome-wide level of significance (S5 Table).
We estimated that identified lead genetic variants explain 7.5%, 6.3%, and 5.3% of the chip-based heritability of isoleucine, leucine, and valine levels estimated in the Fenland study using BOLT-LMM [35] and 2.2%, 0.8%, and 1.2% of the family-based heritability of isoleucine, leucine, and valine levels estimated in twin studies [24], respectively.
A genetic predisposition to a higher level of isoleucine, leucine, or valine was strongly associated with higher odds for type 2 diabetes (for isoleucine, OR 1.44, 95% CI 1.26–1.65, p = 9.5 × 10−8; for leucine, OR 1.85, 95% CI 1.41–2.42, p = 7.3 × 10−6; for valine, OR 1.54, 95% CI 1.28–1.84, p = 4.2 × 10−6; Fig 3A). Relative risk estimates from genetic analyses based on the identified genetic variants were similar to those from the prospective studies of the association between baseline amino acid levels and incident type 2 diabetes (Figs 3A, S4 and S5; Section 2 of S2 Text). For isoleucine, the results of the genetic analyses were almost identical when considering genetic variants at the PPM1K locus only or genetic variants at all four loci (S6 Table). At the PPM1K locus, both the lead SNPs in the BCAA association analyses, rs1440581 (r2 = 0.88) and rs7678928 (r2 = 0.70), were in linkage disequilibrium with the lead type 2 diabetes SNP at the locus, rs1975393 (Fig 3B). There was a dose-response relationship between the association with amino acid level and the relative increase in diabetes risk for the four lead SNPs that were used for the construction of the isoleucine genetic score (Fig 3C). Similarly, the association between isoleucine level and incident type 2 diabetes in prospective studies showed a graded dose-response relationship (Fig 3D; S7 Table).
The isoleucine, leucine, and valine genetic scores were specifically associated with the three BCAAs and not with any of the remaining 172 metabolites measured in the Fenland study (Fig 4). Consistent with these results, the rs1440581 SNP at PPM1K was associated with only eight out of 453 metabolites studied by Shin et al. [24], all of which belonged to the BCAA pathway (S8 Table). These results illustrate the high specificity to BCAA metabolism of these genetic scores.
The BCAA-raising genetic scores and their constituent SNPs were not associated with continuous metabolic traits in large-scale meta-analyses. There was an association of the isoleucine-raising allele of rs1420601 near CBLN1 with higher BMI (S9 Table), but this did not affect the main analysis results (OR for type 2 diabetes after exclusion of rs1420601 = 1.50, 95% CI 1.25–1.80, p = 0.000013).
We assessed genome-wide genetic correlations of BCAA levels with type 2 diabetes and continuous metabolic traits and found statistically significant (p < 0.05) positive correlations of (a) leucine level with type 2 diabetes (rgenetic = 0.34, p = 0.0004), HbA1c (rgenetic = 0.34, p = 0.0038), BMI (rgenetic = 0.25, p = 1.5 × 10−5), and waist-to-hip ratio (rgenetic = 0.29, p = 2.6 × 10−6) and of (b) valine level with type 2 diabetes (rgenetic = 0.54, p = 8.8 × 10−5), fasting insulin (rgenetic = 0.42, p = 0.013), BMI (rgenetic = 0.43, p = 1.8 × 10−5), waist-to-hip ratio (rgenetic = 0.47, p = 6.4 × 10−6), HOMA-B (rgenetic = 0.40, p = 0.016), and HOMA-IR (rgenetic = 0.45, p = 0.012).
In the Fenland study, both higher BMI and higher fasting insulin level, a measure of fasting-state insulin resistance, were associated with higher levels of BCAAs (S10 Table). In addition, genetic predispositions to higher BMI and insulin resistance, but not to impaired insulin secretion, were associated with higher levels of BCAAs (S11 and S12 Tables; S6 Fig). The association of genetic predisposition to greater adiposity with high BCAA levels appeared to be mediated by insulin resistance (S11 Table). In the SABRE study, a glucose challenge resulted in a reduction of circulating BCAA levels. Individuals with higher fasting insulin showed a diminished reduction of all three BCAA levels in response to a glucose challenge (S7 Fig; S13 and S14 Tables). In muscle biopsies collected during an oral glucose challenge, the expression of PPM1K increased at 2 h in normoglycaemic individuals, but not in age-matched patients with type 2 diabetes (S8 Fig). These results are in line with what has been reported in previous investigations [52,53] and suggest that greater adiposity and impaired insulin sensitivity result in exposure to higher levels of BCAAs both in the fasting state and after glucose intake, which could be mediated at least in part by impaired BCKD activation.
Because BCAA-raising alleles were strongly associated with metabolites specific to the BCAA pathway, we hypothesized that studying their pattern of association might provide additional insights into the mechanisms affected by these genetic variants. Using untargeted metabolomics in up to 8,693 individuals, we investigated the association of BCAA-raising alleles with the levels of 18 metabolites involved in BCAA metabolism (Tables 2 and S15; Fig 5). BCAA-raising genetic variants were strongly associated with higher levels of the branched-chain alpha-ketoacids and other metabolites upstream of the irreversible branched-chain alpha-ketoacid dehydrogenation (Fig 5). Metabolites downstream of this important biochemical reaction were largely unaffected by BCAA-raising alleles (Tables 2 and S15; Fig 5).
Higher levels of branched-chain alpha-ketoacids and other metabolites upstream of BCKD action were also strongly associated with a higher risk of incident type 2 diabetes in the EPIC-Norfolk case-cohort study, whereas the associations of downstream metabolites were inconsistent (Tables 2 and S15; Fig 5).
In this large-scale human genetic study, BCAA-raising polymorphisms identified with a genome-wide approach were associated with a higher risk of type 2 diabetes. Our findings suggest that mechanisms leading to impaired BCAA metabolism are implicated in the pathophysiology of type 2 diabetes. Given the strong and specific association with the BCAA pathway observed after testing more than 500 metabolic phenotypes, it is unlikely that the identified genetic variants affect type 2 diabetes risk via mechanisms outside of this metabolic pathway. The relative increase in diabetes risk associated with these genetic variants appeared to be proportional to the size of the association with amino acid levels, which would further support a possible causal link. The diabetes risk increase estimated in genetic analyses was also consistent with the direction and magnitude of association between baseline amino acid levels and incident diabetes in observational prospective studies.
A genetic predisposition to insulin resistance was associated with higher plasma BCAA levels, suggesting that previously reported associations between genetic susceptibility to higher BMI and BCAA levels [11] may be mediated by insulin resistance mechanisms. Consistent with recent studies in mice [54], the expression of PPM1K in muscle biopsies during an oral glucose challenge failed to increase in people with type 2 diabetes, suggesting that the link between insulin resistance and higher BCAA levels may be partly mediated by impaired BCKD activation. Therefore, part of the contribution of insulin resistance mechanisms to type 2 diabetes may be exerted via impaired BCAA metabolism.
In metabolome-wide investigations of BCAA-raising alleles, we found evidence of an accumulation of BCAAs and BCAA-derived metabolites upstream of the oxidative dehydrogenation of branched-chain alpha-ketoacids. This reaction is the irreversible and rate-limiting step in BCAA metabolism, catalysed by the mitochondrial BCKD complex [55]. The pattern of association observed in this study mirrors that observed in maple syrup urine disease (MSUD), an inborn error of metabolism caused by rare loss-of-function mutations in genes encoding components of the BCKD complex [55] or its regulatory phosphatase [56]. In our GWAS of BCAA levels, the strongest association signal was located 21 kb upstream of the PPM1K gene, which encodes the mitochondrial phosphatase that activates BCKD [49–51]. Loss-of-function mutations of PPM1K in humans [56] or the knock-out of its ortholog Ppm1k in mice models [35] results in impaired BCKD activity and high levels of BCAAs and branched-chain alpha-ketoacids, a pattern that resembles that observed for common PPM1K genetic variants in our study. Therefore, it is plausible that the genetic variants identified in this study act by impairing the catabolism of BCAAs, hence leading to higher circulating levels of these amino acids. We found that levels of all metabolites accumulated upstream of BCKD action were associated with incident type 2 diabetes. This further supports the hypothesis that reduced BCKD activity could be one of the mechanistic links between BCAA metabolism and type 2 diabetes.
Our findings have public health and clinical implications as they indicate that modulation of BCAA metabolism may impact diabetes risk. The activity of BCKD is a major determinant of the rate of BCAA catabolism and is amenable to modulation by pharmacological intervention [57–62]. Improving insulin sensitivity may also ameliorate BCAA metabolism, resulting in reduced diabetes risk. This approach is supported by the well-described reduction of amino acid levels following the administration or secretion of insulin [63,64], by our finding of higher BCAA levels in individuals with a genetic predisposition to insulin resistance, and by the observation of reduced BCAA levels following insulin-sensitising interventions [65–67].
Future studies will clarify the molecular mechanisms linking impaired BCAA metabolism and increased risk of type 2 diabetes. The lack of a strong association between BCAA-raising alleles and continuous metabolic traits in publicly available GWAS meta-analyses possibly reflects a lack of statistical power or the unavailability of data about specific glycaemic traits altered by these alleles, similar to several other type 2 diabetes risk variants [25]. We found positive genetic correlations of BCAA levels with type 2 diabetes and several glycaemic and anthropometric traits, which points to shared genetic determinants between BCAA levels and greater adiposity, hyperinsulinaemia, and hyperglycaemia.
A growing body of evidence from human, cellular, and animal models is beginning to shed light on the possible mechanisms linking BCAA metabolism and diabetes risk [2,7,8,68]. Emerging mechanistic explanations include a synergistic interference of BCAAs and lipids with the response of peripheral tissues to insulin [8]. In animal feeding studies, BCAA supplementation requires the background of a high-fat diet to promote insulin resistance [8]. The rs1440581-C variant, i.e., the lead BCAA-raising allele in our genome-wide association analysis, was found to be associated with less weight loss and insulin sensitisation following a calorie-restricted, high-fat diet in a dietary intervention study in humans [69]. Recent research has shown that higher BCAA levels are associated with a gut microbiome pattern characterised by enriched BCAA biosynthetic potential, including Prevotella copri and Bacteroides vulgatus species, pointing to a possible role of the gut flora in the relationship between BCAA levels and insulin resistance [70]. BCAAs and their by-products have also been linked to beta-cell dysfunction [7–9]. The knock-down of PPM1K in clonal INS-1 cell lines has been shown to impair glucose-stimulated insulin secretion [9]. It has also been suggested that a chronic exposure to high levels of BCAAs may result in a constant hyperinsulinaemic response, eventually leading to beta-cell exhaustion [8]. The increased oxidative stress associated with the accumulation of BCAA-derived alpha-ketoacids is consistent with the link between superoxide generation and beta-cell dysfunction [71,72].
Another aspect to be elucidated is whether patients with MSUD have altered glycaemic metabolism. Alterations of glucose metabolism were not reported as a clinical feature in a review of the phenotype of the disease [55]. However, Mogos et al. reported hypoglycaemia in a case of MSUD, which could perhaps be consistent with the hypothesis of BCAA-mediated hyperinsulinaemia [73]. MSUD can be fatal early in life, is often associated with severe cognitive impairment, and is managed with a tightly controlled therapeutic diet [55]. Therefore, MSUD may not be an optimal model to study the risk of a late-life disease with important dietary determinants such as type 2 diabetes.
Limitations of the genetic approach used in this study affect its interpretation. Mendelian randomisation assumes that the genetic variants used as instruments are associated with the disease exclusively via the risk factor of interest. For this reason, we excluded the pleiotropic GCKR locus from analysis. We also assessed the association of genetic variants with more than 500 phenotypes, without finding evidence of pleiotropy. While this reduces the possibility that pleiotropy has influenced our findings, we cannot entirely exclude the possibility of pleiotropic associations. Similar to other complex traits, the genetic variants identified in this study explain only a fraction of the heritability of BCAA levels. Therefore, it is possible that only some of the mechanisms that increase BCAA levels or affect their metabolism are implicated in type 2 diabetes.
While our genetic approach investigated the relationships between circulating BCAA levels and diabetes risk, BCAA metabolism is a complex biological phenomenon closely linked to other metabolic pathways. It is plausible that changes in this metabolic pathway may influence diabetes risk only in a certain metabolic context. Some suggestions come from the emerging biology of the glucokinase regulatory protein (GCKR), the genetic variants of which had to be excluded from our study for methodological reasons. GCKR is known to switch liver metabolism from production of glucose towards that of triglycerides, in particular triglycerides with lower carbon content and double bonds [74,75]. These lipid species are associated with higher diabetes incidence in observational studies [76]. Therefore, the activation of specific pathways (such as those that increase circulating BCAAs or short-chain, highly saturated triglycerides) may typically increase type 2 diabetes risk, except when such activation is caused by GCKR. These findings suggest that (a) the mechanisms by which the concentrations of these biomarkers are raised, not their higher concentrations per se, are implicated in higher metabolic risk and (b) higher BCAA levels might be tolerated as long as other protective pathways are activated. Larger-scale genetic studies and randomised controlled trials will help to further qualify the causal nature of the relationship between BCAA metabolism and type 2 diabetes risk.
In this study, BCAA-raising polymorphisms identified with a genome-wide approach were associated with a higher risk of type 2 diabetes, consistent with a causal role of BCAA metabolism in the aetiology of this common complex disease.
|
10.1371/journal.pgen.1007487 | Role of Bicaudal C1 in renal gluconeogenesis and its novel interaction with the CTLH complex | Altered glucose and lipid metabolism fuel cystic growth in polycystic kidneys, but the cause of these perturbations is unclear. Renal cysts also associate with mutations in Bicaudal C1 (Bicc1) or in its self-polymerizing sterile alpha motif (SAM). Here, we found that Bicc1 maintains normoglycemia and the expression of the gluconeogenic enzymes FBP1 and PEPCK in kidneys. A proteomic screen revealed that Bicc1 interacts with the C-Terminal to Lis-Homology domain (CTLH) complex. Since the orthologous Gid complex in S. cerevisae targets FBP1 and PEPCK for degradation, we mapped the topology among CTLH subunits and found that SAM-mediated binding controls Bicc1 protein levels, whereas Bicc1 inhibited the accumulation of several CTLH subunits. Under the conditions analyzed, Bicc1 increased FBP1 protein levels independently of the CTLH complex. Besides linking Bicc1 to cell metabolism, our findings reveal new layers of complexity in the regulation of renal gluconeogenesis compared to lower eukaryotes.
| Polycystic kidney diseases (PKD) are incurable inherited chronic disorders marked by fluid-filled cysts that frequently cause renal failure. A glycolytic metabolism reminiscent of cancerous cells accelerates cystic growth, but the mechanism underlying such metabolic re-wiring is poorly understood. PKD-like cystic kidneys also develop in mice that lack the RNA-binding protein Bicaudal-C (Bicc1), and mutations in a single copy of human BICC1 associate with renal cystic dysplasia. Here, we report that Bicc1 regulates renal gluconeogenesis. A screen for interacting factors revealed that Bicc1 binds the C-Terminal to Lis-Homology domain (CTLH) complex, which in lower eukaryotes mediates degradation of gluconeogenic enzymes. By contrast, Bicc1 and the mammalian CTLH complex regulated each other, and Bicc1 stimulated the accumulation of the rate-limiting gluconeogenic enzyme even in cells depleted of CTLH subunits. Our finding that Bicc1 is required for normoglycemia implies that renal gluconeogenesis may be important to inhibit cyst formation.
| Autosomal dominant polycystic kidney disease (ADPKD) is an incurable inherited chronic disorder characterized by progressive kidney enlargement and frequent end-stage renal disease due to numerous fluid-filled cysts that are induced by mutations and loss of heterozygosity in PKD1 or PKD2 genes [1]. Complexes of the corresponding transmembrane proteins polycystin-1 (PC1) and polycystin-2 (PC2) are activated at the cell surface by specific WNT ligands to mediate influx of extracellular Ca2+, or by mechanical stimulation of primary cilia that induces Ca2+ release from intracellular stores in the endoplasmic reticulum [reviewed in 2]. Intracellular Ca2+ spikes dampen the levels of cAMP by inhibiting adenylate cyclases V (AC5) and VI (AC6) and by stimulating phosphodiesterase (PDE) activity [2–4]. PC2 and PC1 also negatively regulate AC6 at the level of its mRNA and protein expression [4, 5]. AC6 is important for cystic growth [6] since accumulation of excess cAMP combined with Ca2+ restriction stimulates proliferation and cyst enlargement through several pathways, including PKA, EGFR, Src, B-Raf/Erk and mTORC1 signaling [2]. In addition, cystic growth in PC1-deficient kidneys is fueled by a Warburg-like metabolic switch characterized by increased anaerobic glycolysis at the expense of oxidative phosphorylation [7]. How PKD1 attenuates glycolysis is unknown [8].
Large cystic kidneys showing hyperactivation of EGFR, Src and mTORC1 also develop in homozygous mutant bpk mutant mice owing to a frame-shifting mutation in one of two alternative transcripts of Bicc1 [9–12]. Consistent with a role in polycystin signalling, Bicc1 is required to increase PC2 mRNA and protein levels, and its own accumulation is inhibited in Pkd1 mutant kidneys [13, 14]. Maintaining normal Bicc1 expression is likely important since mutation or loss of a single copy of BICC1 in human is sufficient to provoke renal cystic dysplasia [15]. Similar to ADPKD kidneys, Bicc1 null mutant kidneys also upregulate AC6 protein levels, accumulate cAMP, and secrete excess Fetuin-A [16, 17]. However, a role of Bicc1 in human ADPKD remains to be defined.
Bicc1 consists of three RNA-binding K homology (KH) and two KH-like domains at the N-terminus, a Gly- and Ser-rich intervening sequence (IVS) and a sterile alpha motif (SAM) at the C-terminus. SAM domains are found in over 4000 proteins (SMART, http://smart.embl-heidelberg.de), often forming dimers or oligomers by head-to-tail self-association [18]. In bpk mutant mice, the frame-shifted C-terminus of Bicc1 is abnormally elongated by 149 aberrant amino acids that disrupt SAM-SAM interactions, thus underlining the importance of polymerization for Bicc1 function in vivo [19]. Bicc1 and its Xenopus homolog xBic-C are thought to inhibit canonical Wnt signal transduction by sequestering cytoplasmic Dishevelled [19, 20] or by directly binding Wnt11 mRNA, respectively [21]. Inhibition of Dishevelled is potentiated by SAM-mediated stabilization of Bicc1 polymers in cytoplasmic foci, whereas KH domains recruit specific mRNAs [16, 19, 20]. Drosophila Bicaudal-C in “Malpighian” tubules (the structures corresponding to renal tubules) has been shown to bind myc mRNA and negatively regulates d-Myc protein levels [22]. Direct Bicc1 targets in kidneys are elusive, but likely include AC6 and protein kinase inhibitor (PKI) α mRNAs to mediate their loading unto miRNA-induced silencing complexes in a process that strictly depends on SAM domain polymerization [16, 19]. However, this is not the only mechanism how Bicc1 regulates mRNA translation. In particular, when bound to, a 3'UTR fragment of Cripto-1 mRNA in early Xenopus embryos, xBic-C directly repressed 5' cap-dependent translation, independently of the SAM domain and of miRNA [23]. Furthermore, binding of Drosophila Bicaudal-C to its own mRNA has been shown to promote poly(A)-tail deadenylation by the CCR4-NOT complex, or to inhibit it, depending on the developmental context [24]. Consistent with potential roles in translation activation, Bicc1 has also been shown to increase the translation of Pkd2 mRNA, in this case by protecting the 3'UTR against miR-17 [14]. Furthermore, a novel translation-activating function of Bicc1 mediated by binding to eIF3 on mRNAs specifically at centrosomes seems to require fine-tuning by the orofacial-digital syndrome protein-1 (OFD1) to suppress renal cyst formation [25]. Thus, besides being epistatic to PC1 and PC2, Bicc1 likely mediates multiple functions that could be relevant in PKD and other cystic kidney diseases.
An unbiased approach to identify Bicc1-dependent processes is to analyse its protein interactome. In Drosophila ovaries, Bic-C interacts with CCR4-NOT deadenylase complex and with cytoplasmic polyadenylation element binding protein (CPEB) [24, 26]. In addition, Drosophila Bic-C co-purified a complex of Me31B, Tral, PABP and Cup that is important to correctly secrete and localize the TGF-α homolog Grk during oogenesis [27]. By comparison, little is known about Bicc1 partners in mammals. In HeLa cells, Bicc1 co-immunoprecipitated with several other KH domain-containing proteins such as Sam68, GLD-1, GRP33, and Qk1 [28]. More recently, Bicc1 was identified as a binding partner of the ankyrin repeat- and SAM domain-containing protein ANKS6, which is mutated in cystic kidneys of a subset of nephronophthisis patients [29, 30]. ANKS6 also binds the related Bicc1-interacting protein ANKS3, suggesting that all three proteins may act in a common cilia signalling pathway [30, 31]. In keeping with a role in cilia signalling, Bicc1 has been detected in primary cilia from proximal tubule epithelial LLC-PK1 cells [32].
Here, we conducted a proteomic screen to identify novel Bicc1-interacting factors. Among 243 proteins that were enriched at least 2-fold by tandem affinity purification (TAP) of Bicc1, we focused on an interaction of Bicc1 with the CTLH complex because the orthologous complex in S. cerevisae triggers glucose-induced degradation of gluconeogenic enzymes. We show that Bicc1 maintains the levels of the gluconeogenic enzymes fructose-1,6-biphosphatase (FBP1) and phosphoenolpyruvate carboxykinase (PEPCK/PCK1), while inhibiting CTLH complex accumulation in mouse kidneys. Epistasis analysis shows that Bicc1 also stimulates FBP1 protein expression in mIMCD3 and LLC-PK1. Even though the CTLH complex did not inhibit FBP1 in these cell-based models, it negatively regulated Bicc1 protein levels, apparently by competing for the same SAM domain surfaces that are also required for self-polymerization and the silencing of Bicc1 target mRNAs. Our results suggest that Bicc1 is a target of CTLH complex and regulates metabolic processes.
To identify Bicc1-binding proteins, we purified tandem affinity-tagged Bicc1- StrepII-HA (Bicc1-SH) from a Flp-In T-REx 293 knock-in cell line and analyzed co-purifying proteins by Liquid Chromatography-Mass Spectrometry (LC/MS) (Fig 1A). Doxycycline-induced Bicc1-SH expression in HEK293T cells normalized to γ-tubulin reached levels comparable to those of endogenous Bicc1 in mouse inner medullary collecting duct mIMCD3 cells (top panel), but below those of transiently transfected HA-Bicc1 (bottom panel) (Fig 1B). To validate that Bicc1-SH was functional, we analyzed its mRNA silencing activity using luciferase mRNA reporters that contain target 3'UTR fragments of AC6 or PKIα [16]. We found that Bicc1-SH and HA-Bicc1 silenced both reporters to a similar extent (Fig 1C). LC/MS analysis of Bicc1-SH purified by sequential Strep-tag and anti-HA pull-down identified 243 co-purifying proteins that were enriched at least 2-fold (S1 Table). Highly enriched proteins comprised the NOT1 subunit of the CCR4-NOT complex and ANKS3, indicating that our screen detected known Bicc1-interacting factors [24, 31, 33]. Among the novel interactions, we observed 6.5- to 25-fold enrichment of C-terminal to Lissencephaly Homology (CTLH) complex subunits [34], including WD repeat-containing protein WDR26, Ran-binding protein-9 (RanBP9), Macrophage Erythroblast Attacher (MAEA), Two-hybrid-Associated protein With RanBPM 1 (TWA1), Required for Meiotic Nuclear Division 5 homolog A (RMND5a), and 2.3-fold enrichment of Armadillo repeat containing 8 (ARMC8). Only the CTLH subunits Muskelin-1 (MKLN1) and GID4 (formerly C17orf39) were not enriched by Bicc1-SH under these conditions (Table 1).
Several subunits of the mammalian CTLH complex and the orthologous Gid complex in S. cerevisiae contain Lissencephaly Homology (LisH), CTLH and C-terminal CT11-RanBP9 (CRA) domains, or transducin/WD40 and armadillo repeats [34, 35]. In yeast growing on glucose, the Gid complex consists of 9 subunits required to degrade gluconeogenic enzymes via the proteasomal Pro/N-end rule pathway or via lysosomes [36–38]. The orthologous mammalian CTLH complex has no known function and contained no GID4 under the conditions examined in HEK293 cells [34]. To validate binding to Bicc1, we performed co-immunoprecipitation assays in HEK293T cells that do not express endogenous Bicc1. Anti-Flag immunoprecipitation of DDK-tagged WDR26 co-immunoprecipitated RanBP9, TWA1 and HA-Bicc1 (Fig 1D). WDR26 also co-precipitated MKLN1, whereas anti-MKLN1 pulled down HA-Bicc1 together with RanBP9 (Fig 1E). Moreover, immunoprecipitation of endogenous Bicc1 in extracts of mIMCD3 cells enriched endogenous RanBP9, WDR26, and Twa1 (Fig 1F). These results confirm that mammalian CTLH complexes bind Bicc1.
To evaluate potential interactions with Bicc1 in vivo, we compared the expression levels of CTLH subunits in wild-type (WT) and Bicc1-/- newborn mice. Western blot analysis revealed 2-fold higher WDR26 protein levels in the knockout compared to WT (n = 5, P≤0.01) and a similar trend for GID4 and RanBP9, but not for TWA1 (Fig 2A–2D). By contrast, no changes were observed in WDR26 or GID4 mRNAs (Fig 2E). These results indicate that Bicc1 attenuates the levels of some but not all CTLH subunits in vivo.
In S. cerevisiae, the Gid complex polyubiquitinates FBP1 and PEPCK and induces their degradation to switch from gluconeogenesis to glycolysis in high glucose [37, 39]. To assess glucose metabolism, we first measured blood glucose in WT and Bicc1-/- neonates. Average glucose levels decreased below 60 mg/dl compared to 80 mg/ml in wild-type (Fig 3A). Furthermore, RT-qPCR and Western blot analysis revealed 3.4-fold less PEPCK mRNA and 10-fold less protein, whereas FBP1 decreased 2-fold specifically at the protein level and only in Bicc1-/- kidneys (Fig 3B and 3C). By contrast in livers, where Bicc1 localizes to cholangiocytes and not to hepatocytes [19], FBP1 and PEPCK expression were unchanged (Fig 3D). These results establish that Bicc1 increases PEPCK and FBP1 expression in newborn kidneys and maintains normoglycemia.
Since Bicc1 did not affect FBP1 transcription, we chose FBP1 to further assess how Bicc1 influences its protein level. Ectopic expression of HA-Bicc1 in HEK293T cells increased the accumulation of FBP1 (Fig 4A), whereas RNAi depletion of Bicc1 in mIMCD3 cells [16, 40] decreased it (Fig 4B). Alternatively, we inactivated Bicc1 by CRISPR/Cas9 editing using single guide RNA. While Western blot confirmed the loss of Bicc1 protein, FBP1 levels varied among independent sgBicc clones (Fig 4C, S1 Fig). To evaluate whether Bicc1 regulates FBP1 through the CTLH complex, we depleted WDR26 or Twa1 or both by RNAi. FBP1 levels did not increase but rather decreased, and only in sgBicc cells, pointing to potentially complex layers of inhibitory mechanisms (Fig 4D). Depletion of WDR26 similarly failed to enhance FBP1 accumulation in LLC-PK1 proximal tubule cells, even though RNAi of Bicc1 inhibited it (Fig 4E). FBP1 also was not significantly stabilized in cells treated with the proteasome inhibitor MG132 or with H+-ATPase inhibitor Bafilomycin A1, or both (S1 Fig). Thus, Bicc1 does not upregulate FBP1 simply by inhibiting the CTLH complex.
To assess the contribution of Bicc1 polymerization, we also tested the regulation of FBP1 by the non-polymerizing D913;K915;E916/AAA mutant Bicc1 MutD [19]. Bicc1 MutD failed to alter FBP1 expression (Fig 4F), indicating that the effect of wild-type Bicc1 described above is specific and likely depends on SAM domain-mediated polymerization.
Rather than degrading FBP1, CTLH complex may target Bicc1. In keeping with this idea, depletion of CTLH subunits in mIMCD3 cells significantly increased endogenous Bicc1 protein, but not its mRNA (Fig 5A). To assess Bicc1 degradation, we treated mIMCD3 cells for 4 hrs with MG132 or Bafilomycin A1 or empty vehicle. Treatment with MG132 increased Bicc1 protein levels without inducing slow migrating intermediates characteristic of polyubiquitination (Fig 5B). By contrast, an apparent increase in Bicc1 levels after treatment with the H+-ATPase inhibitor Bafilomycin A1 was not statistically significant, and both drugs combined gave variable results, reflecting increased toxicity.
Previously, we have shown that Bicc1 is stabilized in large cytoplasmic foci by self-polymerization of its SAM domain [18, 19]. To distinguish whether the CTLH complex interacts with Bicc1 in cytoplasmic foci or with a non-polymerized pool, or both, we performed co-immunostainings of HA-Bicc1 and endogenous RanBP9 in HEK293T cells. We only observed co-localization with RanBP9 peripheral to and outside bright Bicc1 foci and only at increased laser intensities (Fig 5C), indicating that access of CTLH complexes to macroscopically visible Bicc1 self-polymers may be limited.
To estimate the size of Bicc1-CTLH complexes, we fractionated HA-tagged Bicc1 on sucrose density gradients, using CNOT1 as a control (Fig 6A). HA-Bicc1 shifted a peak of CNOT1 from fractions 7–9 to fractions 9–11, all of which also contain ribosomal protein S6 (RPS6) [19] (S2 Fig). By contrast, RanBP9, TWA1, ARMC8, and MKLN1 peaked with low molecular weight (LMW) HA-Bicc1 in fraction 5, and GID4 and WDR26 peaked in fractions 1–3 or 7, respectively. These results indicate that the average size of Bicc1-CTLH complexes is smaller than the ribosome-sized complexes of Bicc1 with CNOT1. Since CTLH complexes did not accumulate in cytoplasmic Bicc1 foci (Fig 5C) or in high molecular weight (HMW) fractions (Fig 6A), we tested whether they bind Bicc1 independently of its SAM domain. Coimmunoprecipitation experiments in transfected HEK293T cells showed that deletion of the SAM domain inhibited Bicc1 binding to CTLH subunits, whereas deletion of the KH domains did not (Fig 6B). To validate that binding involves the SAM domain, we used GST-SAM fusion protein in pull-down assays in HEK293T cell extracts. Irrespective of the presence or absence of exogenous HA-Bicc1, GST-SAM pulled down endogenous TWA1, whereas GST alone did not (Fig 6C). As an additional control, we analyzed another SAM domain protein using Flag-tagged ANKS3. Coimmunoprecipitation analysis revealed no Flag-ANKS3 binding to RanBP9, WDR26 or TWA1 above background levels (Fig 6D). However, co-transfection of full length ANKS3 or of the truncated mutant ANKS3ΔC lacking the C-terminal region distal of its SAM domain inhibited CTLH complex binding to HA-Bicc1 (Fig 6E), indicating that ANKS3 competes with CTLH complex for Bicc1. Previous structure analysis suggests that the SAM domain of ANKS3 interfaces with the ML and EH surfaces of the Bicc1 SAM domain, thereby allowing the bulky C-terminal region of ANKS3 to block Bicc1 polymer elongation [41]. To distinguish whether ML or EH surfaces of the Bicc1 SAM domain or their oligomerization also mediate CTLH complex binding, we mutated them individually [19] (S3 Fig). Control mutations MutA and MutF outside the ML and EH surfaces did not inhibit Bicc1 binding to any of the CTLH subunits analyzed, including WDR26, TWA1 and ARMC8. Binding of ARMC8 was also unaffected by mutations MutC or MutE that disrupt the ML or EH surface, respectively. In sharp contrast, disruption of either of these surfaces impaired WDR26 and TWA1 binding, and this defect was not rescued even if Bicc1 MutC and MutE were coexpressed to permit the formation of MutC/MutE dimers (S3 Fig). These data suggest that SAM domain oligomerization increases CTLH complex binding.
To further map the topology of CTLH and Bicc1 complexes, we conducted yeast two-hybrid assays. The strongest interactions among CTLH subunits were those of TWA1 with ARMC8, RMND5A, and RANBP9, whereas RanBP9 and RMND5A in turn strongly bound WDR26 and MAEA, respectively, (Fig 6F, S4 Fig). GID4 interacted with ARMC8 (GID5), albeit more weakly than its yeast homolog [42]. In contrast, no CTLH subunit alone directly associated with Bicc1, or vice versa. To test whether docking sites are buried in Bicc1 polymers, we coexpressed individual CTLH subunits with polymerization mutant Bicc1 MutD. Compared to wild-type, Bicc1 MutD showed increased binding of ARMC8 in two-hybrid assays, but not of other individual CTLH subunits (Fig 6F, S5 Fig). To independently validate the influence of SAM domain polymerization, we repeated the TAP-tag purification using Bicc1 MutD, and at lower salt concentrations that allowed Bicc1/CTLH complexes to retain MKLN1 (Table 2). We found that compared to WT, Bicc1 MutD co-purified significantly less CTLH complex, even though binding to other proteins such as CAD, FASN and ACACA was not inhibited (Table 2).
Taken together, these data suggest that the polymerization interfaces of the Bicc1 SAM domain act together with SAM-independent ARMC8 binding to recruit CTLH complexes (Fig 6G).
Cystic growth in ADPKD reportedly involves anaerobic glycolysis accompanied by a decrease in the expression of gluconeogenic enzymes [7]. Here, a proteomic screen and validation by coimmunoprecipitation, yeast two-hybrid and density gradient fractionation assays revealed that Bicc1 interacts with the CTLH complex, the mammalian ortholog of the Gid complex that mediates degradation of gluconeogenic enzymes in S. cerevisiae. CTLH complexes recruited Bicc1 both independently of SAM domain polymerization via ARMC8 and through SAM domain surfaces that can be blocked by ANKS3 or buried in large Bicc1 self-polymers, suggesting CTLH complexes target Bicc1 oligomers devoid of an ANKS3 cap. Our analysis in cultured mIMCD3 cells indicates that CTLH complex curbs the levels of Bicc1 but not of the gluconeogenic enzyme FBP1. Nevertheless, deletion of Bicc1 in mice led to hypoglycemia and diminished the expression of FBP1 and PEPCK specifically in kidneys but not in liver, correlating with increased accumulation of CTLH complex. These results reveal multi-layered regulation of FBP1 expression by Bicc1 and CTLH complexes: While Bicc1 overall mediates a net increase in FBP1, inhibition of self-polymerizing Bicc1 by the CTLH complex and vice versa provides a mechanism to adjust the levels of Bicc1.
Bicc1 and the ankyrin and SAM domain proteins ANKS3 and ANKS6 all suppress renal cysts and can coprecipitate each other [29–31]. In our interactome screen, stringent co-purification with TAP-tagged Bicc1 significantly enriched ANKS3 but not ANKS6, confirming that Bicc1 more tightly binds ANKS3 and independently of ANKS6 [43]. Since recent metabolomic analysis links ANKS3 to amino acid metabolism [44], future studies should investigate whether this process also depends on Bicc1. TAP-tag purification of Bicc1 also enriched several CCR4-NOT deadenylase subunits including CNOT1, NOT-5, -6/6L and -9. In addition, we found that Bicc1 can bind to the multifunctional enzyme carbamoyl-phosphate synthetase 2, aspartate transcarbamylase, dihydroorotase (CAD), which is rate-limiting for pyrimidine synthesis [45], and acetyl-CoA carboxylase alpha (ACACA) and fatty acid synthase (FASN) which function sequentially in fatty acid synthesis [46]. While regulation of these enzymes by Bicc1 remains to be tested, these interactions point to potential roles in several metabolic pathways.
Here, we focused our attention on the CTLH complex. The core of the homologous Gid complex in S. cerevisiae is formed by Gid1, Gid5 and Gid9, held together by Gid8. In turn, Gid5 and Gid9 recruit Gid2 [47, 48]. Gid2 and Gid9 are RING domain E3 ubiquitin ligases. Each contains a CTLH motif that is also found in Gid1 and Gid8 and in mammalian RanBP9 (GID1), RMND5a (GID2), TWA1 (GID8), MAEA (GID9), WDR26 (GID7) and in MKLN1 [34, 47]. The CTLH subunit most enriched by TAP tag-purified Bicc1 was WDR26, a cytoplasmic WD40 domain protein that can recruit substrates to the E3 Ub ligase CUL4 and inhibit MAPK-induced activation of serum response transcription factors [49, 50]. Bicc1 also enriched the CUL4-binding protein DDB1, but no CUL4. Possibly, WDR26 and/or DDB1 are substrate-specific adaptors for more than one E3 Ub ligase. RanBP9 was discovered as a partner of Ran, which regulates mitotic spindle assembly and nuclear and ciliary protein import [51–53]. RanBP9 also binds the KH domain protein Fragile Mental Retardation (FMRP) [54]. However, two-hybrid assay detected no RanBP9 interaction with Bicc1 KH domains.
Bicc1 significantly diminished the accumulation of the CTLH subunit WDR26, and we noted a similar trend for RanBP9 and GID4. In S. cerevisiae, the Gid complex is activated by the accumulation of Gid4 within minutes after addition of glucose to degrade gluconeogenic enzymes [37, 38]. A function for human GID4 is elusive. We detected GID4 in HEK293T cells and found that it co-fractionated with other CTLH subunits in sucrose density gradients. GID4 also interacted with RanBP9 and RMND5a in two-hybrid assays. Although this topology differs from S. cerevisiae where Gid4 is recruited by Gid5 [42], these data show that human GID4 can bind the mammalian CTLH complex. TAP-tagged Bicc1-SH copurified CTLH complexes that contained no detectable GID4. Possibly, Bicc1 reduces GID4 expression or inhibits GID4 recruitment to mammalian CTLH complex. Although, since GID4 expression was low and since binding to CTLH complexes is below detection even in the absence of Bicc1 [34], we could not conclusively test this hypothesis.
We found that depletion of CTLH complex subunits by RNAi increased endogenous Bicc1 levels in mIMCD3 cells. In the simplest model, the CTLH complex accelerates Bicc1 turnover by direct binding. In keeping with this idea, Bicc1 interacted with the CTLH subunit ARMC8 in two-hybrid assays. ARMC8 has been shown to reduce the accumulation of E-cadherin and of α- and β-catenin to promote cell invasiveness in various cancers [55–57]. ARMC8 targets α-catenin for proteasomal degradation independently of polyubiquitination, despite associating with E3 ligases of the CTLH complex [47, 58]. In two-hybrid assays, ARMC8 bound polymerization mutant but not wild-type Bicc1 or its isolated KH or SAM domains. Furthermore, in mammalian cells, ARMC8 and other CTLH subunits clearly segregated from HMW Bicc1 polymers during density gradient fractionation. SAM domain sequestration in HMW polymers may reduce direct access of ARMC8 to the intervening sequence between KH and SAM domains. In keeping with this model, only a diffusely distributed Bicc1 pool peripheral to microscopically visible cytoplasmic polymers overlapped with RanBP9 foci. Nevertheless, the Bicc1 SAM domain and its polymerization interfaces reinforced binding to assembled CTLH complexes, pointing to cooperativity among its subunits. In keeping with this idea, truncated Bicc1ΔSAM or polymerization mutants only inefficiently coimmunoprecipitated the subunits WDR26 and TWA1, and ARMC8 binding to polymerization mutant Bicc1 MutD during 2-step tandem affinity purifications was weakened compared to wild-type Bicc1. CTLH complex recruitment was also inhibited by overexpressed ANKS3, which competes for free Bicc1 SAM domain interfaces to cap their self-polymerization [43]. Taken together, these observations suggest that ARMC8 primes Bicc1 for recruiting additional CTLH subunits to exposed SAM domains. While the CTLH complex is not the only regulator of Bicc1 degradation, our finding that SAM domain polymerization interfaces mediate this interaction suggests that CTLH complexes preferentially target LMW Bicc1 oligomers at equilibrium with larger polymers when they are uncapped of ANKS3 (Fig 6G). It will be interesting to determine whether LMW forms of Bicc1 outside cytoplasmic foci may associate with cilia or centrosomes, and whether their availability is regulated by cell metabolism or vice versa.
A switch of glucose metabolism from glycolysis to gluconeogenesis is important during fasting and at birth when newborns feed on milk, which is low in glucose. During prolonged fasting, up to 25% of gluconeogenesis derives from extrahepatic sources such as kidney and intestine [59]. Prompted by its interaction with the CTLH complex, we tested whether Bicc1 regulates glucose metabolism. Despite hypoinsulinemia, blood glucose levels in Bicc1-/- neonates decreased more than 1.3-fold below wild-type levels. Among possible causes, we considered a role for Bicc1 in gluconeogenesis. Whereas PEPCK converts oxaloacetate into phosphoenolpyruvate, FBP1 catalyzes the rate-limiting step of fructose 1,6-bisphosphate hydrolysis to fructose 6-phosphate. RT-qPCR analysis revealed a 3.4-fold decrease in PEPCK mRNA levels specifically in Bicc1-/- kidneys but not in liver, and both PEPCK and FBP1 were further down-regulated at the protein level. In the absence of glucose, ATP production largely depends on fatty acid oxidation. A resulting increase in the levels of ketone bodies can lead to systemic acidosis if kidneys cannot keep up with acid secretion. Acid secretion into urine requires ammonium production by glutaminolysis, combined with increased gluconeogenesis that drives bicarbonate synthesis and transport into the blood to buffer plasma pH [60]. Normally, hypoglycemia stimulates glucagon-induced upregulation of PEPCK expression and gluconeogenesis through cAMP/PKA signalling. Failure to induce renal PEPCK expression and gluconeogenesis despite elevated cAMP levels as observed in Bicc1 mutants predicts a defect in acid-base balance, which in turn likely accelerates disease progression. Indeed, accumulation of ammonium in kidney cortex combined with impaired urinary secretion stimulates interstitial nephritis, and it promotes cyst formation in the Han:SPRD rat model of ADPKD [61–63].
In S. cerevisiae, the Gid complex inhibits gluconeogenesis by targeting PEPCK and FBP1 for proteasomal degradation by the proline N-end rule [37, 64]. FBP1 levels decreased almost 2-fold in Bicc1 mutant kidneys without a corresponding decrease at the mRNA level. Depletion of Bicc1 by RNAi also decreased FBP1 protein in cultured mIMCD3 and LLC-PK1 cells, whereas ectopic expression of Bicc1 in HEK293T cells increased it. To test whether endogenous Bicc1 stabilizes FBP1 by inhibiting the CTLH complex, we depleted the subunits TWA1 or WDR26. Knockdown of the CTLH complex by RNAi did not stabilize FBP1. We also observed no accumulation of FBP1 intermediates characteristic of polyubiquitination in cells treated with MG132. In basal-like breast cancer, FBP1 expression can be transcriptionally repressed by Snail1 to stimulate glycolysis and confer a metabolic advantage to cancer stem cells [65], and nearly all renal cell carcinomas deplete FBP1 to promote their aggressiveness [66]. Future studies should investigate whether this process involves Bicc1 or its interaction with CTLH complexes.
For Western blot analysis, the following antibodies were used according to manufacturer's instructions: Anti-PEPCK 1:1000 (Abgent), anti-MAEA 1:1000 (R&D systems), anti-C17orf39 1:500 (Aviva), anti-WDR26 1:1000 (Bethyl Lab), anti-Twa1 1:1000 (Proteintech), anti-FBP1 1:500 (Sigma), anti-RanBP9 1:1000 (Abcam), anti-Armc8 1:500 (Santacruz), anti-Bicc1 1:1000 (raised by Proteogenix), anti-HA 1:1000 (Sigma), anti-γ-Tubulin 1:1000 (Sigma), anti-Actin clone C4 1:1000 (Merck). We used densitometric analysis for protein band quantification.
Bicc1+/- heterozygous mice on a C57Bl/6 mouse genetic background were maintained in ventilated cages at the animal facility of the Ecole Polytechnique Fédérale de Lausanne (EPFL) [20]. All animal experiments were approved by the Veterinary Service of the Swiss canton of Vaud or by the Institutional Animal Care and Use Committee and adhered to the guidelines in the Guide for the Care and Use of Laboratory Animals (National Research Council. 2011. Guide for the care and use of laboratory animals, 8th ed. National Academies Press, Washington, DC). Newborn mice were decapitated at postnatal day P2, and total blood glucose was measured using the Accu-Check Aviva Nano glucometer (Roche). Kidney and livers were immediately snap frozen in liquid nitrogen after dissection. For use, the tissues were grinded with a pestle in lysis buffer and sonicated using a Bioruptor device (Diagenode). After centrifugation, protein concentration was quantified using Bradford assay (Bio-Rad).
To generate TAP-tagged MmBicc1 and MmBicc1-D913K915E916/AAA (Bicc1 MutD), a previously described HA-tagged mouse Bicc1 cDNA [20] and HA-tagged mouse Bicc1 MutD [19] were amplified by PCR and fused in-frame as a Kpn I restriction fragment to the SH tag in the inducible expression vector pcDNA5/FRT/TO (courtesy of Aebersold Lab). Alanine substitutions of residues D913, K915 and E916 in HA-Bicc1 have been described previously [19]. Briefly, the mutated DNA fragment was amplified by overlap extension PCR and then subcloned between Bgl II and Xba I sites of pCMV-SPORT6::HA-Bicc1. DNA fragments for Y2H assays were cloned by PCR in plasmid pACT2 distal to the GAL4 transactivation domain coding sequence, or in plasmid pGBKT7 distal to the GAL4 DNA-binding domain coding sequence using primers with unique restriction sites. The luciferase reporter plasmids Luc-AC6 3’UTRprox and Luc-PKIα 3’UTRprox in the vector pCS2+ have been described [16].
HEK293T and mIMCD3 cells were cultured in DMEM medium (Sigma) supplemented with 10% fetal bovine serum (FBS; Sigma), glutamine (1%; Invitrogen), and gentamicin (1%; Invitrogen). LLC-PK1 cells were cultured in DMEM with 5.5 mM D-glucose, supplemented with 10% fetal bovine serum, 100U/ml penicillin, 100ug/ml streptomycin. These and all other cell lines in this study were negative for mycoplasma as determined by an ELISA-based assay (Roche). Expression vectors were transfected using jetPEI transfection reagent (Polyplus) according to the manufacturer's instructions. Small interfering RNAs against MmWDR26 (TAAAGGCTTTAGCTCATTCAGGTCA), MmTwa1 (CCGACTCATCATGAACTAC) and MmBicc1 (CCAACCACGUAUCCUAUAATT) or SsWDR26 A (CCTCATGCAAGAGTCAGGATGTCGT), SsWDR26 B (AATAGGACAGCACTTGAATGG) or scrambled control (Microsynth) were transfected during 48 hrs using INTERFERin transfection reagent (Polyplus). CRISPR/Cas9 editing of Bicc1 mIMCD3 cells was carried out using the guide sequences 5’-GCGAGCGCAGCACCGACTCGCCGG-3’ were cloned into the expression vector PX458 containing GFP-tagged Cas9 [67]. The resulting sgRNA/Cas9 expression vector were transfected and after 24h, the cells were trypsinized, washed with PBS and resuspended in PBS/1% FBS for single cell sorting for GFP by FACS into 96-well plate containing complete medium.
HEK293T cells were transfected with 1 μg of HA-Bicc1 in 6 well plates. After 24 hrs, the cells were plated on coverslips. At 48 hrs post-transfection, cells were washed with phosphate-buffered saline (PBS) and methanol-fixed during 10 mins at -20°C. Cells were washed again with PBS and incubated with blocking solution containing 1% BSA for 1 hr at room temperature. Mouse anti-HA and rabbit anti-RanBP9 antibodies were diluted 1:500 and 1:100, respectively, in blocking solution and added to the cells during 2 hrs at room temperature. After washing with PBS, cells were incubated with Alexa 647-conjugated secondary donkey anti-mouse and Alexa 488-conjugated donkey anti-rabbit IgG during 1 hr at room temperature in the presence of 4’,6-diamidino-2-phenylindole (DAPI). Pictures were acquired by confocal microscopy using a Zeiss LSM700 microscope.
To generate the inducible MmBicc1-SH stable cell line, we co-transfected Flp-InT-REx HEK293 cells (Invitrogen) with pOG44 Flp recombinase expression plasmid (Invitrogen) and with pcDNA5/FRT/TO vector containing MmBicc1-SH. Correct integration by Flp-mediated homologous recombination gave rise to hygromycin-resistant clones that were isolated and validated for MmBicc1-SH expression after 24 hrs of doxycycline induction. For tandem affinity purification [68], 3 dishes of Bicc1-SH T-Rex cells were treated without (control) or with doxycycline (1 μg/ml) during 24 hours. Cells were rinsed and scraped from dishes in Tris-buffered saline (TBS) containing 150 mM NaCl in 50 mM Tris.HCl pH 7.4, followed by extraction with TBS containing 1 mM DTT, 0.05% NP-40, phosphatase inhibitor cocktail 3 (Sigma), RNAse inhibitors (Promega) and protease inhibitor cocktail (Roche). After sonication and centrifugation at 10'000 × g for 10 min, supernatants were incubated on a rotating wheel for 2 hrs at 4°C with Strep-Tactin Sepharose resin (IBA). After incubation, resins were loaded in Mobicol columns (MoBiTec) and washed with 10 ml 0.05% NP-40 in TBS, and eluted on ice using desthiobiotin 3X (containing 400 mM NaCl) during 15 min as described [69]. For the second purification, eluates were incubated with anti-HA agarose beads (Sigma, A2095) during 2 hrs at 4°C. After incubation, beads were loaded on Mobicol columns, rinsed with 10 ml Tris.HCl pH 7.4 (10 mM) containing 100 mM NaCl, 2.5 mM MgCl2, 1 mM DTT, and 0.02% NP-40, and eluted using 125 mM HCl into a vial containing 50 μl of Triethylammonium bicarbonate (TAEB) neutralization buffer (Sigma). Eluates were concentrated in 40 μl using Amicon Ultra 0.5 ml 3K (Millipore). TAP-tag purification of Bicc1-MutD-SH was done with buffer containing Tris.HCl pH 7.6 (30 mM), 150 mM NaCl, 1 mM MgCl2, 0.05% NP-40, phosphatase inhibitor cocktail 3 (Sigma), RNAse inhibitors (Promega) and protease inhibitor cocktail (Roche). Strep elution was done using desthiobiotin 2X (containing 300mM). Elution from HA bead was done as described above. Laemmli buffer was added to the eluates and boiled samples were loaded in a 4–15% Mini-PROTEAN TGX precast Gel (BioRad).
LC-MS/MS analysis was done at the Proteomic core facility (PCF) at Ecole Polytechnique Fédérale de Lausanne (EPFL). Peptides were extracted from Coomassie Brilliant blue R-250 stained gel slices and subjected to tryptic digestion. Reverse phase separations were performed on a Dionex Ultimate 3000 RSLC nano-UPLC system (Thermo Fisher Scientific) connected online an Orbitrap Elite Mass Spectrometer (Thermo Fisher Scientific) piloted with Xcalibur (version 2.1) and Tune (version 2.5.5), as described in [70]. Samples were analyzed using Mascot (version 2.6, Matrix Science, Boston, MA, USA) set up to search the human subset of the UniProt database (Release 2017_02). For visualization and validation, MS/MS data was entered in Scaffold 4 (Proteome Software Inc., Portland, OR). The peptide identification threshold, and the protein identification threshold based on matches with at least 2 identified peptides were set at 1% FDR. The MS analysis was conducted in duplicates from two biological replicates. For the semi-quantitative analysis of the MS data, a fold enrichment has been calculated by comparison with the background found in the control condition. The values have then been normalized over the fold enrichment obtained for the bait protein. The presented values correspond to the average of the normalized fold enrichments obtained in the two replicates.
To measure luciferase expression, HEK293T cells were plated in 24-well plates and transfected on the following day with the indicated plasmids (100 ng/well) and with a lacZ expression vector (50 ng/well) in triplicate samples using jetPEI (Polyplus Transfection). 24 hrs after transfection, cells were lysed in buffer containing 25 mM Tris-phosphate, pH 7.8, 2 mM DTT, 2 mM 1,2-diaminocyclohexanetetraacetic acid (CDTA), 10% glycerol, and 0.5% Triton X-100. Cell extracts were diluted 20-fold, and luminescent counts detected in a Centro XS3 LB960 microplate luminometer (Berthold Technologies). Values were normalized to β-galactosidase activity measured by spectrophotometry in a Safire2 microplate reader. Results represent mean values from at least three independent experiments. Error bars show standard errors of the mean (SEM). Student's t-test was used to calculate P values.
HEK293T cells were transfected with the indicated expression plasmids in 10 cm dishes (we used 2 μg/dish). 24 hrs after transfection, cells were washed with PBS and proteins were extracted with lysis buffer containing 1X TBS, 1mM DTT, 0.05% NP-40, 1x protease inhibitor cocktail (Roche) and phosphatase inhibitor cocktail 3 (Sigma). Endogenous Bicc1 was immunoprecipitated from mIMCD3 cell extracts. In brief, confluent cells in 10 cm dishes were washed with PBS and resuspended in extraction buffer as described above. After sonication and centrifugation at 10'000 × g for 10 min, supernatants were incubated on a rotating wheel for 2 hrs at 4°C with anti-HA agarose antibody or anti-FLAG M2 affinity gel (Sigma), or with protein G Sepharose beads (GE Healthcare) precoated with Bicc1 custom antibody [19] or pre-immune IgG (R&D Systems). Beads loaded on Mobicol columns were washed with 10 ml of Tris.HCl pH 7.4, 100 mM NaCl, 2.5 mM MgCl2, 1 mM DTT, and 0.02% NP-40 washing buffer and resuspended in Laemmli buffer. Elutions were fractionated on SDS-PAGE gels and analyzed by Western blotting.
Recombinant GST-Bicc1 SAM domain was produced in the E. coli BL21 strain (Novagen) as previously described using pGEX-1λT expression vector [16]. GST fusion protein were purified from cell extract under native conditions, using Glutathione Sepharose 4B (GE Healthcare) in 50 mM Tris.HCl pH 8, 200 mM NaCl, 1 mM DTT according to manufacturer's instructions, rinsed twice with 20 mM Tris.HCl pH 7.4; 750 mM NaCl; 1 mM DTT washing buffer followed by Tris.HCl pH 7.4, 20 mM; NaCl 200 mM; DTT 1 mM washing buffer and eluted by addition of 20 mM glutathion. Confluent HEK293T cells in 10 cm dishes were lysed in TBS buffer containing 1 mM DTT, 0.05% NP-40, Phosphatase inhibitor cocktail 3 (Sigma), RNAse inhibitors (Promega) and Protease inhibitors cocktail (Roche). After sonication and centrifugation at 10'000 × g for 10 min, supernatants were incubated on a rotating wheel for 2 hrs at 4°C with glutathione-Sepharose 4B beads saturated with GST-Bicc1 SAM or GST alone (negative control). After washing on Mobicol columns with 5 ml of 20 mM Tris.HCl pH 7.4 containing 200 mM NaCl, 2 mM MgCl2, 1 mM DTT, and 0.05% NP-40, the beads were resuspended in Laemmli buffer. Eluates were fractionated on SDS-PAGE gels and analyzed by Western blotting. Loading of GST fusion proteins was validated indirectly by Ponceau staining of proteins retained in the gel.
Subconfluent HEK293T cells were transfected with HA-Bicc1 or empty vector in 10 cm dishes (2 μg DNA/dish) using 3 plates per condition. Cell extracts were prepared after 24 hrs as described above for the GST pull-down assay. Continuous 15 to 60% sucrose gradients were prepared as described previously [19]. Fractions were recovered manually starting from the top, fractionated on SDS-PAGE gels, and analyzed by Western blotting. γ-tubulin was used as a control.
Total RNA from mIMCD3 cells and kidneys was isolated using TRIzol (Life Technologies) and RNeasy plus mini kit (Qiagen) following manufacturer's instructions. Total RNA was digested with RQ1 DNase (Promega). Reverse transcription of cDNA was carried out using SuperScript III reverse transcriptase and Oligo dT (Life Technologies) according to the manufacturer's recommendations. The qPCR was performed in a QuantStudio 6 Flex real-time PCR systems (Applied Biosystems) using goTaq qPCR 2x Master Mix (Promega). Samples were analyzed as triplicates, and expression levels were calculated with the manufacturer’s software using the ΔΔCt method. The PCR primers are described in S2 Table.
To assess binding of each CTLH complex subunit to WT Bicc1 or Bicc1- D913K915E916/AAA (MutD), they were each fused to the DNA-binding domain of the GAL4 transcription factor (GAL4‐BD) in the pGBKT7 plasmid (Clontech) as bait proteins. In parallel, we fused them each to the activation domain of the GAL4 transcription factor (GAL4‐AD) in the pACTII plasmid (Clontech) as prey proteins. The reporter gene used in this study is the HIS3 gene required for histidine biosynthesis. To monitor bait and prey interactions, appropriate pACTII (LEU2) and pGBKT7 (TRP1) plasmids were transformed into haploid cells from strain CG1945 (mat a; ura3‐52, his3‐200, ade2‐101, lys2‐801, trp1‐901, leu2‐3, 112, gal4‐542, gal80‐538, cyhr2, LYS2::GAL1UAS‐GAL1TATA‐HIS3, URA3::GAL417‐mers(x3)‐CYC1TATALacZ) and strain Y187 (mat α; gal4, gal80, ade2‐101, his3‐200, leu2‐3,112, lys2‐801, trp1‐901, ura3‐52, URA3::Gal1UAS GAL1TATA‐LacZ), respectively, using the lithium acetate method [71]. After crossing on YPD medium, diploid cells were selected on media suitable for double selection (Leu‐, Trp‐) and then plated on media suitable for triple selection (Leu‐, Trp‐, His‐). Where indicated, 3‐Amino‐1, 2, 4‐triazol (3‐AT) was added as a competitive inhibitor of histidine synthesis to evaluate the strength of the interactions. Growth was assessed after three days of incubation at 30°C. The interactions were confirmed in two independent experiments.
Error bars represent the standard error of the mean (SEM). Two-tail student-t test was used to compare the differences between 2 conditions to calculate p-values. 1-way ANOVA and Turkey’s multiple comparison test was used to compare groups of unpaired values and determine the significance (p-value) of every mean compared to every other mean.
|
10.1371/journal.pgen.1005475 | SmD1 Modulates the miRNA Pathway Independently of Its Pre-mRNA Splicing Function | microRNAs (miRNAs) are a class of endogenous regulatory RNAs that play a key role in myriad biological processes. Upon transcription, primary miRNA transcripts are sequentially processed by Drosha and Dicer ribonucleases into ~22–24 nt miRNAs. Subsequently, miRNAs are incorporated into the RNA-induced silencing complexes (RISCs) that contain Argonaute (AGO) family proteins and guide RISC to target RNAs via complementary base pairing, leading to post-transcriptional gene silencing by a combination of translation inhibition and mRNA destabilization. Select pre-mRNA splicing factors have been implicated in small RNA-mediated gene silencing pathways in fission yeast, worms, flies and mammals, but the underlying molecular mechanisms are not well understood. Here, we show that SmD1, a core component of the Drosophila small nuclear ribonucleoprotein particle (snRNP) implicated in splicing, is required for miRNA biogenesis and function. SmD1 interacts with both the microprocessor component Pasha and pri-miRNAs, and is indispensable for optimal miRNA biogenesis. Depletion of SmD1 impairs the assembly and function of the miRISC without significantly affecting the expression of major canonical miRNA pathway components. Moreover, SmD1 physically and functionally associates with components of the miRISC, including AGO1 and GW182. Notably, miRNA defects resulting from SmD1 silencing can be uncoupled from defects in pre-mRNA splicing, and the miRNA and splicing machineries are physically and functionally distinct entities. Finally, photoactivatable-ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP) analysis identifies numerous SmD1-binding events across the transcriptome and reveals direct SmD1-miRNA interactions. Our study suggests that SmD1 plays a direct role in miRNA-mediated gene silencing independently of its pre-mRNA splicing activity and indicates that the dual roles of splicing factors in post-transcriptional gene regulation may be evolutionarily widespread.
| microRNAs (miRNAs) are a class of small regulatory RNAs that fine-tune gene expression by reducing protein output from their target messenger RNAs and are implicated in myriad physiological and pathological processes. miRNAs are generated from long primary transcripts via sequential actions of the Drosha/Pasha and Dicer ribonucleases. Mature miRNAs are incorporated into the miRISC effector complexes that contain AGO family member proteins and serve as specificity determinants to guide miRISCs to their target RNAs. Previous studies suggested that select proteins implicated in the processing of messenger RNAs are required for the miRNA production/function, but the underlying molecular mechanism is not well understood. Here we show that SmD1, an essential protein implicated in the processing of messenger RNAs, directly interacts with both Pasha and primary miRNA transcripts and is required for optimal miRNA production. Furthermore, SmD1 associates with multiple components of the miRNA effector machinery and is required for miRNA function. Finally, our analysis reveals that defects in the miRNA pathway can be uncoupled from those in messenger RNA processing, and that the miRNA biogenesis and messenger RNA processing machineries are physically and functionally distinct entities. Our data thus suggests that SmD1 modulates the miRNA pathway independent of its role in messenger RNA processing.
| miRNAs are a class of ~22–24 nt endogenous regulatory RNAs present in all cell types of multicellular organisms [1,2]. By regulating the expression of diverse target RNAs, miRNAs play a key role in myriad biological processes including development, homeostasis, and innate immunity. In Drosophila, canonical miRNA biogenesis starts with RNA polymerase II-mediated transcription of long stem-loop primary miRNA transcripts (pri-miRNAs). They are processed in the nucleus by the Drosha/Pasha ribonuclease III (RNase III) microprocessor complex into ~60–70 nt precursor miRNAs (pre-miRNAs), via a reaction referred to as cropping [3,4,5]. Precursor miRNAs are subsequently exported to the cytoplasm by Exportin5/Ran-GTP and further processed in a dicing reaction by a second RNase III complex, the Dicer 1 (Dcr-1)/Loqs-PB complex, thereby liberating ~22–24 nt miRNA duplexes [6,7,8,9,10,11,12]. The mature miRNA strand of the duplex is predominantly incorporated into Argonaute 1 (AGO1)-containing miRNA-induced silencing complexes (miRISC). miRISC in turn engages target mRNAs via complementary base pairing between the seed region of miRNAs (positions 2–8) and miRNA-binding sites (primarily in the 3’ UTR of target mRNAs), and represses gene expression post-transcriptionally by promoting target mRNA destabilization and/or translation inhibition [13,14,15,16,17,18].
Over a decade of investigation has identified a collection of core components of the miRNA biogenesis and functional machineries, and delineated the framework of the molecular mechanism underlying the miRNA pathway. However, our knowledge of the miRNA pathways is far from complete, and many accessory factors that modulate miRNA biology await identification and functional characterization. A number of recent studies highlight extensive crosstalk between pre-mRNA splicing and small RNA-mediated gene silencing pathways. For example, the core RISC component AGO2 has been implicated in modulating pre-mRNA splicing both in mammals and in Drosophila [19,20]. Conversely, select splicing factors have been shown to impact small RNA-mediated gene silencing pathways. It has been reported that mutations in genes encoding a subset of splicing factors compromise RNAi in the fission yeast Schizosaccharomyces pombe [21], and that in plants, mutations in splicing factor genes compromise small RNA biogenesis [22]. In addition, the multifunctional human RNA-binding protein hnRNP A1, which regulates alternative splicing, impacts the processing of miR-18a and let-7a [23,24]. Furthermore, the KH-type splicing regulatory protein (KSRP) has been shown to positively regulate the biogenesis of miR-155 and let-7 [25,26]. Moreover, genome-wide RNAi screens conducted in C. elegans and cultured Drosophila cells show that depletion of certain splicing factors compromises RNAi [27,28,29]. Finally, a recent analysis of phylogenetic conservation of candidate RNAi factors suggests that select splicing factors are required for small RNA-mediated gene silencing [30].
SmD1, together with six other small ribonucleoprotein particle (snRNP) proteins (SmB, SmD2, SmD3, SmE, SmF and SmG), form a heptameric ring structure surrounding the U-rich small nuclear RNAs (snRNAs) [31]. These snRNP proteins constitute core components of the snRNP and play a key role in pre-mRNA splicing [32]. We recently showed that SmD1 depletion in cultured Drosophila cells compromises small interfering RNA (siRNA) biogenesis and function independently of its role in pre-mRNA splicing [33]. In the current study, we investigate the role of SmD1 in the miRNA pathway. We find that SmD1 depletion leads to a reduction in levels of mature miRNAs, which is accompanied by a derepression of the corresponding target messenger RNAs and a concomitant accumulation of primary miRNA transcripts. In addition, SmD1 associates with the microprocessor component Pasha and is required for optimal microprocessor activity. In contrast, SmD1 is dispensable for Dicer-mediated processing of pre-miRNAs into mature miRNAs. Furthermore, our analysis reveals that SmD1 is required for miRNA function besides its role in miRNA biogenesis. Specifically, SmD1 associates with the miRISC components AGO1 and GW182, and is required for optimal miRISC function. Moreover, we show that select splicing factors such as SmD1, but not pre-mRNA splicing per se, modulate the miRNA pathway, as defects in miRNA biogenesis and in pre-mRNA splicing can be uncoupled, and that SmD1 inactivation does not affect the expression of canonical miRNA pathway components. Finally, PAR-CLIP analysis identifies numerous SmD1-binding events across the transcriptome and reveals direct SmD1-miRNA interactions. Taken together, our study identifies SmD1 as a new modulator of the miRNA pathway at multiple levels, and provides direct evidence to support and extend the notion that select splicing factors are critical modulators of small RNA pathways in complex multicellular organisms, beyond the context of the spliceosome.
We first examined small RNA expression profiles in control S2 cells or cells depleted of Drosha, Dcr-2 or SmD1 (Fig 1A and S1 Table). As expected, compared with control samples, depletion of Drosha led to a reduction in the proportion of miRNAs in the total small RNA population. As a consequence, the proportions of all classes of endogenous siRNAs expanded, including those derived from the Flock House Virus (FHV), transposable element (TE), convergently transcribed RNAs (cis-NAT) and hpRNAs. In contrast, depletion of the siRNA biogenesis enzyme Dcr-2 caused the opposite phenotype. Notably, SmD1 knockdown led to a shrinkage of the miRNA population and an artificial expansion of endo-siRNAs. Considering our previous findings showing that SmD1 is required for siRNA biogenesis and SmD1 depletion led to a reduction in levels of endo-siRNAs [33], our data strongly suggest that SmD1 depletion can cause a comparable or even stronger degree of reduction in levels of miRNAs than in levels of siRNAs. To test this directly, we measured levels of endogenous siRNAs and miRNAs by Northern blot in SmD1 knockdown cells. Consistent with our recent finding [33], SmD1 depletion led to a marked decrease in levels of the endogenous siRNA esi-2.1 (Fig 1B and 1E). Importantly, levels of several miRNAs, including miR-33, miR-34, miR-276a, miR-317, miR-2b, miR-184 and miR-bantam, were significantly reduced upon SmD1 knockdown (Fig 1B–1G), reminiscent of the phenotype elicited by the loss of the canonical miRNA biogenesis enzyme Drosha. As a negative control, depleting canonical siRNA pathway components such as Dcr-2 or AGO2 predominantly affected levels of esi-2.1, but not miRNAs (Fig 1B–1E). To rule out potential off-target effects associated with dsRNAs, we tested an independent dsRNA against SmD1 and observed a similar impact on small RNA levels (S1 Fig). We also detected a significant increase in steady-state levels of a subset of target mRNAs for several miRNAs implicated in cell proliferation and apoptosis (Reaper, E2f1 and Socs36E as targets for miR-2b, miR-184 and miR-bantam, respectively) (Fig 1H) [34,35], consistent with the notion that miRNAs repress target gene expression by inhibiting mRNA translation and promoting target mRNA decay. We conclude that SmD1 is required for optimal miRNA biogenesis and/or stability.
Besides the afore-mentioned miRNAs, which are constitutively expressed in S2 cells, we also examined levels of the miRNA let-7, which is not expressed in naïve S2 cells but becomes highly induced upon treatment with 20-hydroxyecdysone (20-E) [36]. We detected lower levels of let-7 in SmD1-depleted cells (S2 Fig). These data reinforce the notion that SmD1 is required for optimal miRNA biogenesis. let-7 is highly expressed in the Drosophila heart, and is required for Drosophila heart function. Interestingly, depletion of SmD1 specifically in cardiac cells of adult Drosophila causes several defects in cardiac function (S3 Fig). These observations suggest that the cardiac phenotype elicited by SmD1 depletion is linked, at least in part, to defects in let-7 biogenesis.
miRNA biogenesis consists of multiple steps, including Drosha-mediated processing of pri-miRNAs into pre-miRNAs and Dcr-1-mediated conversion of pre-miRNAs into mature miRNAs, referred to as cropping and dicing, respectively (Fig 2A). To further define the biochemical step(s) of miRNA biogenesis that require SmD1, we examined the impact of SmD1 inactivation on levels of pri-miRNAs and pre-miRNAs. As expected, we detected a marked accumulation of several pri-miRNAs in S2 cells depleted of the microprocessor component Drosha (S4 Fig). Importantly, SmD1 phenocopies Drosha, albeit to a moderate extent (Fig 2B), suggesting that SmD1 is required for microprocessor-mediated processing of pri-miRNAs. To assess whether SmD1 is required for Dcr-1-mediated processing of pre-miRNAs into mature miRNAs, we treated S2 cells with dsRNAs targeting the firefly luciferase (as control) or SmD1, together with dsRNAs targeting Dcr-1 (to facilitate the detection of pre-miRNAs), and performed Northern blot to measure levels of pre-miRNAs. We found that compared to control samples, loss of SmD1 caused a reduction in levels of pre-miR-184 (Fig 2C). These observations strongly suggest that SmD1 is dispensable for Dcr-1-mediated processing of pre-miRNAs into mature miRNAs, as otherwise we would have detected an accumulation of pre-miRNAs upon SmD1 inactivation. We note that it remains formally possible that SmD1 is required for the processing of pre-miRNAs into mature miRNAs, and the observed decrease in the overall pre-miRNA levels upon SmD1 knockdown is an accumulative effect of impaired microprocessor activity (leading to less pre-miRNA production) and less efficient Dcr-1-mediated pre-miRNA processing (leading to accumulation of pre-miRNAs). Next, to definitively elucidate the requirement for SmD1 in different steps of miRNA biogenesis, we prepared total or cytoplasmic lysates from S2 cells treated with various dsRNAs, incubated lysates with radiolabeled pri-miRNAs or pre-miRNAs, and measured the microprocessor or Dcr-1 activities by monitoring the production of pre-miRNAs or mature miRNAs, respectively. Our analysis revealed that SmD1-deficient cell lysate is as competent as the control lysate in carrying out Dcr-1-mediated processing of pre-miRNAs into mature miRNAs (Fig 2D). In contrast, SmD1-deficient cell lysate displays significant defects in Drosha-mediated conversion of pri-miRNAs into pre-miRNAs (Figs 2E–2G and S5). These data demonstrate that SmD1 is specifically required for cropping, but is dispensable not the dicing step during miRNA biogenesis.
Having shown that SmD1 is required for optimal microprocessor activity, we next examined whether SmD1 associates with components of the microprocessor. Consistent with the requirement for SmD1 in the cropping step, we found that at Flag-tagged SmD1, but not the control protein Ran, co-immunoprecipitated with the endogenous microprocessor component Pasha (Fig 3A). In addition, the recovery of Pasha in the SmD1 complex is resistant to RNase A treatment, indicating RNA-independent protein-protein interactions. It remains to be determined whether the observed SmD1-Pasha interaction is direct or through protein intermediates. While we were unable to detect endogenous Drosha in the SmD1 complex by immunoblotting, most likely due to the low sensitivity of the Drosha antibody, moderate but clearly above background levels of microprocessor activity can be recovered from immunopurified endogenous SmD1 complex (Fig 3B). These observations underscore the functional relevance of the observed SmD1-Pasha interaction. Next, we performed RNA immunoprecipitation (RIP) assays to assess whether SmD1 could associate with pri-miRNAs, the substrates for the microprocessor. Our analysis revealed significant enrichment in the SmD1 complex all six pri-miRNAs that we have examined, the degree of which matched or even exceeded that of cognate SmD1-binding small nuclear RNAs (Fig 3C). In contrast, no significant enrichment of the control mRNA rp49 was observed. The observed interactions of SmD1 with Pasha and pri-miRNAs suggest that SmD1 might modulate miRNA biogenesis by serving as a molecular bridge to facilitate the recognition of pri-miRNAs by the microprocessor. To test this possibility, we performed RIP assay using a stable cell line expressing TAP-tagged Pasha and detected a significant enrichment of pri-miRNAs in immunopurified TAP-Pasha complex (S6 Fig). Importantly, recovery of pri-miRNAs in the Pasha complex was severely blunted upon depletion of SmD1 (Fig 3D). Interestingly, we also observed a moderate degree of enrichment of the control mRNA rp49 in the Pasha complex. This is consistent with the notion that the microprocessor complex potentially associates with and regulates the expression of a number of cellular mRNAs (S6 Fig) [37,38]. However, the association of the rp49 mRNA with Pasha seems to be largely unaffected by SmD1 depletion (Figs S6 and 3D).
To investigate the molecular detail of the SmD1-Pasha interaction, we next sought to map the protein domains responsible for this interaction. We generated a series of T7-tagged full length and truncated Pasha proteins and examined their capability of interacting with Drosha and SmD1 by performing co-immunoprecipitation assays. As expected, we found that full length Pasha was able to co-immunoprecipitate with endogenous Drosha in cultured Drosophila S2 cells (Fig 3E, lane 7). In addition, the C-terminal fragment of Pasha (334–642) is also capable of pulling down Drosha (Fig 3E, lane 10). Our data are consistent with previous reports showing that the C-terminal fragment of DGCR8 interacts with Drosha in mammals [39,40]. Note that T7-tagged Pasha127–642 was unable to co-immunoprecipitate with endogenous Drosha, possibly due to inefficient folding and/or presentation of the T7 epitope in native T7-Pasha127–642 protein in total cell lysate, as T7-Pasha127–642 was poorly recovered in the anti-T7 immunoprecipitate (Fig 3E, lane 11, lower panel). In contrast, T7-Pasha127–642 was readily detectable in input samples, most likely because the T7 epitope can be efficiently recognized by the anti-T7 antibody in denatured T7-Pasha127–642 (Fig 3E, lane 5, lower panel). Next, we examined whether various truncated Pasha proteins can co-immunoprecipitate with SmD1. We found that Flag-tagged SmD1 was able to pull down both full length and a number of truncated Pasha mutants containing the region spanning amino acids 127–333 (Fig 3F, lanes 7–9, 11). Thus, it appears that distinct regions of Pasha are required for the Pasha-Drosha and Pasha-SmD1 interactions (Fig 3G). The stoichiometry of the SmD1-microprocessor complex is currently not clear, and it remains to be determined whether the observed Drosha-Pasha and SmD1-Pasha interactions are mutually exclusive, or all three proteins are present in the same complex. Collectively, these data demonstrate that SmD1 associates with both components of the miRNA biogenesis machinery and primary miRNA transcripts, and that SmD1 is required for optimal recognition of the pri-miRNAs by the microprocessor.
Several lines of evidence point to a possible role of SmD1 in the effector phase of the miRNA pathway (i.e., miRISC assembly and function) besides its involvement in miRNA biogenesis: 1) SmD1 modulates siRISC assembly and function, and associates with several siRISC components, including AGO2 [33]; 2) SmD1, but not the control protein Ran, co-immunoprecipitates with miR-2b [33]; 3) in mammals SNRPD1 (ortholog of Drosophila SmD1) and AGO2 (a canonical miRISC component) interact with each other [33]; and 4) a considerable fraction of SmD1 is present in the cytoplasm (Fig 4A), where miRISC assembly and function primarily take place. To examine this possibility, we transfected SmD1-depleted S2 cells with a synthetic let-7 miRNA duplex together with a Renilla luciferase reporter construct carrying 8 copies of imperfect let-7-binding sites in the 3’ UTR. A firefly luciferase reporter lacking miRNA-binding sites serves as control. Defects in miRISC assembly/function are expected to be reflected as an increase (de-repression) in reporter activity compared to the negative control. It is worth noting that employing a synthetic mature miRNA duplex in the assay effectively circumvents the confounding factor that SmD1 is required for miRNA biogenesis. In addition, we chose let-7 because of its extremely low basal expression in S2 cells, thereby reducing background. Our analysis revealed that compared to control knockdown cells, SmD1-depleted cells display a marked de-repression of the let-7 reporter, resembling the phenotype elicited by depletion of the core miRISC component AGO1 (Fig 4B). These data indicate that SmD1 is required for miRNA function. Consistent with this notion, we found that Flag-tagged AGO1, but not the control protein Ran, is capable of co-immunoprecipitating with endogenous SmD1 (Fig 4C). Furthermore, endogenous AGO1 as well as GW182, another component of the miRISC, were detected in SmD1 complex, but not in the control Ran complex or the control immunoprecipitates using a non-immune serum (Fig 4D and 4E). It appears that the SmD1-AGO1 and SmD1-GW182 interactions are not strongly affected by RNase treatment (Fig 4D and 4E). These data demonstrate the interaction between SmD1 and components of the miRISC.
Next, to determine the functional relevance of the SmD1-AGO1 interaction and to directly assess the role of SmD1 in miRISC assembly/function, we examined whether SmD1 depletion impairs the function of miRISC by measuring the slicer activity of AGO1-miRISC programmed by the let-7 miRNA duplex. To circumvent the confounding factor that AGO2 is a much more robust slicer than AGO1 and thus could mask the weak slicer activity of AGO1, we first established stably transfected S2 cells expressing TAP-tagged AGO1. Then we immunopurified and immobilized the AGO1 complex onto agarose beads and incubated the AGO1 complex with cytoplasmic lysates (from either control cells or SmD1 knockdown cells) together with the let-7 miRNA duplex. The beads were subsequently washed thoroughly and the slicer activity of the bead-bound let-7-AGO1 miRISC against an mRNA substrate carrying a perfect let-7 binding site was measured. This assay revealed a significantly weaker slicer activity of the AGO1-miRISC assembled in SmD1-depleted cell lysate than that assembled in control lysates (Fig 4F and 4G), suggesting that SmD1 is required for efficient miRISC assembly and/or function. To examine whether SmD1 is required for the loading of miRNAs into AGO1, which is the first step of miRISC assembly, we depleted SmD1 in TAP-AGO1-expressing S2 cells, and measured the levels of endogenous miRNAs in immuno-purified TAP-AGO1 complex (Fig 4H). Note that as a control for the amount of TAP-AGO1 expressed from the transgene across different samples, we used identical amount of cell lysates in each immunoprecipitation. Compared with controls, the absolute levels of both miR-2b and miR-184 were markedly reduced in AGO1 miRISC recovered from SmD1-depleted cells (Fig 4H and 4I). This is in part due to lower levels of miRNAs in input samples from SmD1 knockdown cells (Fig 4H, compare the input samples). In addition, we calculated AGO1 loading index by normalizing levels of AGO1-bound miRNAs against those in the input. This analysis revealed a moderate decrease in AGO1 loading index upon SmD1 knockdown, suggesting that SmD1 is required for efficient loading of miRNAs into miRISC (Fig 4J). Taken together, these data demonstrate that SmD1 is required for efficient miRISC assembly/function besides its role in miRNA biogenesis.
To comprehensively identify direct SmD1-RNA interaction events across the transcriptome, we optimized the PAR-CLIP protocol in Drosophila S2 cells, recovered and deeply sequenced SmD1-bound RNAs (S7A and S7B Fig) [41,42]. Mapped reads are derived from various classes of RNAs (S7C Fig). Our initial analysis identified 2180 SmD1 binding sites (clusters) across the transcriptome. Of these, 1729 unique peaks have passed non-adaptive filtration (see Materials and Methods) and were used for subsequent analyses (S2 Table). Among them, 96 map to unannotated genomic regions. For the remaining 1633 clusters, 1553 (95%) and 80 (5%), respectively, map to the annotated coding and non-coding RNAs (Figs 5A and S8; S3 and S4 Tables). As expected, snRNAs, the cognate binding partners for SmD1, were abundantly present in the dataset, so were sequences derived from the endogenous siRNA esi-2.1 precursor CG4068, consistent with our previous report (Fig 5B and S2 Table) [33]. In addition, SmD1 binding events in the vicinity of splice junctions were found (S5 Table), thereby providing supporting evidence for well-documented role of SmD1 in regulation of pre-messenger RNA processing. Interestingly, we also identified a substantial number (24) of SmD1-interacting snoRNAs. Importantly, sequences derived from several pri-miRNAs, including Bantam, miR-2 family, miR-11, miR-33 and miR-34 were also recovered, thus demonstrating direct interactions between SmD1 and pri-miRNAs (Figs 3C, 5C, 5D and S9; S6 Table). Interestingly, among the primary miRNA transcripts that we have examined so far, miR-33 and miR-34 are the top two highly enriched in immunopurified SmD1 complex (Figs 3C, 5C and 5D). Notably, a major SmD1 binding peak directly overlaps with mature miR-34 (Fig 5D), suggesting that SmD1 may directly associate with mature miRNAs (in the context of miRISC). This is consistent with our previous findings that immunopurified SmD1 complex contains mature miRNAs [33]. An alternative and non-mutually exclusive possibility is that SmD1 could bind the sequence segment corresponding to mature miR-34 within the primary miRNA transcript.
It is conceivable that SmD1 depletion indirectly impacts the miRNA pathway by modulating the expression of genes encoding canonical miRNA factors. We previously performed pair-end RNA-sequencing in both control cells and cells depleted of SmD1 [33]. Consistent with the role of SmD1 in pre-mRNA splicing, our analysis revealed that the splicing pattern of ~25% cellular mRNAs is altered in SmD1-depleted cells. Importantly, we found no significant changes in the mRNA levels of canonical miRNA factors except an increase in levels of Drosha and a decrease in loqs-RD, whose encoded product is dispensable for the miRNA pathway (S10 Fig). Most importantly, immunoblotting assays revealed no significant changes in protein levels of canonical miRNA factors in SmD1-depleted cells except for an increase in levels of Loqs-PB and a concomitant reduction in Loqs-PD (Fig 6A) [33]. Given our findings that SmD1 impacts the cropping and miRISC assembly/function (Figs 2E–2G, 4 and S5), which do not require Loqs proteins, it is unlikely that the observed changes in levels of Loqs protein isoforms could account for the miRNA biogenesis and function defects elicited by SmD1 depletion. We conclude that SmD1 depletion does not functionally impact the expression of genes encoding canonical components of the miRNA pathway.
To determine if a broader link exists between the miRNA pathway and splicing, we examined additional snRNP (Sm) proteins for their potential involvement in the miRNA pathway. We found that depletion of several snRNP proteins led a reduction in levels of miR-2b, reminiscent of the SmD1 knockdown phenotype (Fig 6B). In contrast, levels of miR-2b remained unchanged in SmF knockdown cells. Furthermore, SmF knockdown did not impact in microprocessor activity either, even though SmF-depleted cells displayed obvious alteration in the splicing pattern of the CG13887 pre-mRNA, to the same extent as that observed in SmD1-depleted cells (Figs S11, 6C and 6D). These data clearly show that the miRNA defects elicited by SmD1 depletion can be uncoupled from pre-mRNA splicing defects, as the miRNA pathway appears to be intact in SmF-depleted cells, even though these cells display profound defects in pre-mRNA splicing.
To further dissect the relationship between the miRNA and pre-mRNA splicing machineries, we examined whether additional snRNP proteins are capable of interacting with Pasha. We found that besides SmD1, SmD2 is also capable of co-immunoprecipitating with Pasha (Fig 6E). In contrast, no interaction can be detected between Pasha and SmB or SmF, even though both SmB and SmF are capable of pulling down endogenous SmD1, most likely the SmD1 fraction in the spliceosome (Fig 6E). Furthermore, upon examining levels of pri-miRNA and snRNA species in various immuno-purified Sm protein complexes, we found that SmD1 consistently outperforms SmF in binding to various pri-miRNAs. In contrast, we detected significantly higher levels of the U4 snRNA in the SmF complex compared to those present in the SmD1 complex (Fig 6F). We note that Flag-tagged Sm proteins were employed in these assays and that these exogenous proteins have to complete with their endogenous counterparts for binding to their RNA partners. However, since we are measuring levels of various RNA cargos in the same sample, it allows us to make a direct comparison regarding the relative affinity between various Sm proteins and their RNA partners. These data suggest that compared with SmF, SmD1 appears to display a more prominent role in miRNA biogenesis, whereas SmF seems to be more dedicated to pre-mRNA splicing. In addition, these data also indicate that the spliceosome and miRISC are functionally distinct entities. To address this further, we immunopurified the microprocessor or the snRNP complexes by TAP-Pasha IP or SmD1 IP, respectively, and examined the presence of various snRNAs and pri-miRNAs in these complexes. As expected, snRNAs were highly enriched in the SmD1 complex (Fig 3C). In contrast, snRNAs were largely absent in the microprocessor (Fig 6G). On the other hand, several primary miRNA transcripts were highly enriched in the microprocessor (Figs 3D and S6). Taken together, these data demonstrate that select splicing factors, but not splicing per se, influence the miRNA pathway, and that the miRNA biogenesis machinery and the spliceosome are physically and functionally distinct.
Accumulating evidence suggests extensive crosstalk between pre-mRNA splicing and small RNA-mediated gene silencing pathways. For example, the core component of the RNAi effector machinery AGO2 plays a key role in the regulation of pre-mRNA splicing [19,20]. Conversely, a number of splicing factors have been implicated in modulating small RNA-mediated gene silencing [21,22,23,24,25,26,27,28,29,30]. We previously showed that the core snRNP splicing factor SmD1 is required for optimal biogenesis and function of siRNAs [33]. In the current study we report that SmD1 also plays a key role in modulating the miRNA pathway. Specifically, SmD1 interacts with the microprocessor component Pasha and with primary miRNA transcripts, and is selectively required for efficient recognition of primary miRNA transcripts by the microprocessor and the conversion of pri-miRNAs into pre-miRNAs. In addition, SmD1 associates with the miRISC components AGO1 and GW182, and is required for miRISC assembly/function. We further show that only select splicing factors, but not pre-mRNA splicing per se, modulate the miRNA pathway, since defects in pre-mRNA splicing can be uncoupled from defects in the miRNA pathway, and that the molecular machineries executing pre-mRNA splicing and miRNA biogenesis and function are physically and functionally distinct entities.
Our analysis reveals that SmD1, but not a closely related snRNP protein SmF, is required for optimal miRNA biogenesis. These observations indicate that SmD1 belongs to a select group of splicing factors that modulate small RNA pathways beyond the context of the splicing machinery. SmD1 associates with the N-terminus of Pasha, whereas the C-terminal domain of Pasha is sufficient to interact with Drosha. It is currently unclear whether the SmD1-Pasha and Drosha-Pasha interactions take place in a mutually exclusive manner, or alternatively there exists a Drosha-Pasha-SmD1 complex. Our observation that the immunopurified SmD1 complex is capable of carrying out the conversion of pri-miRNAs into pre-miRNAs lends support to the latter possibility (Fig 3B). Of note, several splicing factors have been shown to co-purify with the microprocessor [3]. Our study, together with a recent report that implicates FUS in miRNA biogenesis [43], demonstrates that select splicing factors are functionally associated with the microprocessor in higher multicellular organisms. While it is possible that SmD1 couples splicing and processing of pri-miRNA transcripts by recruiting the microprocessor to nascent pri-miRNA transcripts that undergo co-transcriptional splicing, our observations that depletion of SmF, a related snRNP splicing factor, does not impact miRNA biogenesis (Figs 6B and S11), that neither SmB nor SmF is able to co-immunoprecipitate with Pasha (Fig 6E), and that immunopurified SmD1 and Pasha complexes harbor overlapping yet distinct sets of cargo RNAs (Figs 3C, 6F and 6G), argue against this possibility.
While SmD1 is clearly required for miRNA biogenesis, it may operate in the context of multimeric protein complexes to execute this function. For example, our data show that SmD2 is as competent as SmD1 in pulling down primary miRNA transcripts and Pasha, and depletion of SmD2 led to a reduction in both miRNA levels and microprocessor activity (Figs 6B, 6E, 6F and S11). Together with previous studies that report the presence stable SmD1-SmD2 subcomplex, [31,44], our data raise the possibility that the SmD1/D2 sub-assembly may execute a moonlighting function during miRNA biogenesis beyond the context of the spliceosome. We found that adding back purified recombinant SmD1 or lysates from cells over-expressing SmD1 was not sufficient to restore the microprocessor activity or miRISC assembly/function in SmD1-depleted cell lysate (S12 and S13 Figs). These observations suggest that additional SmD1 co-factor(s) (such as SmD2) in the context of multimeric complexes of appropriate stoichiometry may be necessary to functionally impact the miRNA pathway. An alternative possibility is that SmD1 depletion leads to altered expression of unknown protein(s) involved in miRNA biology, which underlies the defects in the miRNA pathway elicited by SmD1 knockdown. However, our findings lend strong support to the former scenario.
Interestingly, our data uncover a differential dedication of various Sm proteins to the miRNA pathway and pre-mRNA splicing. For example, SmD1 and SmD2 consistently outperform SmF in binding to various pri-miRNAs, whereas the U4 snRNA is significantly more enriched in the SmF complex than in the SmD1 complex. (Fig 6F) This is consistent with our findings showing that SmD1 and SmD2, but not SmF, is required for the miRNA pathway. These observations also indicate the presence of sub-spliceosomal assemblies or novel SmD1-containing complexes that impact the miRNA pathway. Further supporting this notion, our co-immunoprecipitation assay reveals that Flag-tagged SmD1 is capable of pulling down endogenous SmD1 (Fig 6E, lane 8). Most likely this interaction takes place beyond the context of the spliceosome, as the snRNP ring structure contains only a single copy of each Sm protein. Interestingly, depleting either SmB or SmD2, immediate neighbors to SmD1 in the snRNP ring structure, led to a reduction in miRNA levels and microprocessor activity. However, only SmD2, but not SmB, is capable of co-immunoprecipitating with Pasha (Fig 6E). Identification of the complete collection of microprocessor-associated splicing factors and unraveling the stoichiometry of the microprocessor/SmD1-containing complexes should provide insights into the function of microprocessor-associated splicing factors in miRNA biogenesis.
Besides its involvement in the initiation phase of the miRNA pathway (miRNA biogenesis), SmD1 is also required for the effector phase of the miRNA pathway, as SmD1 depleted cells display defective miRISC assembly/function. This is manifested in part by defects in the loading of miRNAs into AGO1-miRISC (Fig 4H–4J). It remains to be determined whether SmD1 is similarly required for additional steps of miRISC assembly, including miRNA duplex unwinding and miRNA star strand removal.
We report here that the Drosophila SmD1 associates with the microprocessor and impacts the cropping step of miRNA biogenesis (Fig 6H). Interestingly, in C. elegans the SmD1 ortholog SNR-3 co-purifies with Dcr-1 [45]. It would be informative to address whether SNR-3 similarly impacts miRNA biogenesis in worms. In addition, we show here that SmD1 associates with the core miRISC component AGO1 in flies and impacts miRNA function (Fig 6H). An analogous observation has been made regarding the role of SmD1 in the siRNA pathway, where it associates with the siRISC component AGO2 and is required for siRISC assembly/function. Furthermore, we report previously that the human orthologs of SmD1 and AGO2 associate with each other [33]. These findings raise the possibility that the role of SmD1 in modulating small regulatory RNA biogenesis and Argonaute-RISC assembly and function may be evolutionarily conserved.
Drosophila S2 and S2-NP cells were maintained in Schneider’s medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin (Invitrogen). S2-NP cells stably expressing Flag-SmD1 were generated by transfection with pRmHa-3-Flag-SmD1 and the selection marker plasmid pHS-neo using the calcium phosphate method, followed by selection in medium containing 400 μg/mL G418 (Calbiochem). dsRNA treatment was performed as described previously [27,46]. Briefly, ~2 × 106 S2 or S2-NP cells were seeded in 6-well plates for 24 h and then transfected with 3 μg of the appropriate dsRNA. Two days later, the cells were harvested, replated in 6-cm plates for 24 h, and then treated again with 9 μg dsRNA. Three days later, the cells were harvested and used in assays. For Renilla luciferase reporter assays, transfections were performed in a 384-well format using HiPerFect (Qiagen).
DNA fragments encompassing the coding regions of SmD1, SmB, SmD2, SmF, Drosha, AGO1, and full length and truncated Pasha, together with Flag, T7 or TAP epitope tags were amplified by PCR and cloned into pRmHa-3 or pMK33-NTAP vectors. Anti-SmD1 antibody was generated by immunizing rabbits with a synthetic peptide (ProSci Inc.).
Cells were lysed in lysis buffer (20 mM Tris-HCl (pH 7.6), 150 mM NaCl, 2 mM EDTA, 10% glycerol, 1% Triton X-100, 1 mM DTT, 1 mM orthovanadate) supplemented with protease inhibitor cocktail (Roche). Cleared total lysates were immunoprecipitated with antibodies against Flag (Sigma). Both input and immunoprecipitated samples were analyzed by SDS-PAGE followed by immunoblotting with antibodies against Flag (Sigma), T7 (Novagen), AGO1 (Abcam), Pasha [5] or GW182 [47]. Other antibodies employed in this study include Drosha and Dcr-1 (gift from Dr. Greg Hannon), Loqs [48] or Tubulin (Sigma), as indicated. RNase treatment of the immunoprecipitates was performed as previously described [27,46]. To detect protein-RNA interactions, RNA was extracted from the immunoprecipitates by treatment with TRIzol and subject to RT-qPCR analysis using gene-specific primers.
Northern blotting was performed as previously described [46,48]. In brief, total cellular or co-immunoprecipitated RNA was isolated with TRIzol (Invitrogen). Samples of 15 μg RNA were separated on 15% denaturing polyacrylamide gels and transferred to Hybond-N+ membranes (Amersham Biosciences) in 1X TBE buffer. Small RNAs were UV crosslinked to the membranes, and the membranes were prehybridized in hybridization buffer for 2 h. DNA probes complementary to the appropriate strands were 5′ radiolabeled and incubated with membranes overnight at 37°C. Membranes were washed twice in 1X SSC with 0.1% SDS at 42°C, and then exposed to Phosphorimager screens for 12–48 h. Membranes were stripped by the addition of boiling 0.1% SDS solution and incubated for 30 min.
Cropping, dicing and slicing assays were performed as previously described, with minor modifications [49,50]. Briefly, for the cropping assay, total S2-NP cell lysates were prepared by sonicating cell suspension in lysis buffer (30 mM HEPES-KOH, pH 7.0, 100 mM potassium acetate, 2 mM magnesium acetate, 5 mM DTT, 20% glycerol, and 1X EDTA-free protease inhibitors (Roche)) for 5 times (5 sec each with 2 min interval at 30% duty cycle). Primary miRNAs were synthesized using a T7 MEGAscript in vitro transcription kit (Ambion) with α-32P-GTP, and gel purified. Aliquots of 10 μl of cell lysates containing the same amount of total protein were incubated in a final volume of 20 μl reaction mixture (30 mM HEPES-KOH, pH 7.0, 100 mM potassium acetate, 2 mM magnesium acetate, 5 mM DTT, 10% glycerol, 1 mM ATP, 10 mM creatine phosphate, 0.06 U/μl creatine kinase (Roche), 0.1 U/μl ribonuclease inhibitor (Promega), 10 ng/μl yeast tRNA, and 2000–10,000 cpm pri-miRNA substrate) at 25°C for 2.5 h.
For the dicing assay, cytoplasmic extracts from frozen S2-NP cells were prepared by thawing cells in a hypotonic buffer composed of 10 mM HEPES-KOH (pH 7.0), 2 mM magnesium acetate, 0.1% β-mercaptoethanol, and 1X EDTA-free protease inhibitors (Roche). Radiolabeled pre-miRNA substrate was prepared by incubating the pre-let-7 synthetic RNA oligo with T4 polynucleotide kinase (New England Biolabs) and γ-32P-ATP and subsequently gel-purified. Aliquots of 6 μl of cell lysates containing the same amount of total protein were incubated in a final volume of 10 μl reaction mixture (20 mM HEPES-KOH, pH 7.0, 2 mM DTT, 2 mM magnesium chloride, 1 mM ATP, 25 mM creatine phosphate, 0.06 U/μl creatine kinase (Roche), 0.8 U/μl ribonuclease inhibitor (Promega), and 2000–10,000 cpm pre-let-7 substrate) at 25°C for 1 h.
For the slicing assay, the capped Renilla luciferase mRNA substrate containing 1 copy of perfect let-7 binding site in the 3’ UTR was synthesized using a MEGAscript T7 in vitro transcription kit, incubated with Vaccinia virus capping enzyme (New England Biolabs) and α-32P-GTP, and gel purified. Minimal AGO1-miRISC was prepared by incubating TAP-AGO1 cell lysates with IgG beads with gentle rocking at 4°C overnight. The beads were thoroughly washed in hypotonic buffer and incubated with cytoplasmic lysates from either SmD1 knockdown or control cells in a final volume of 50 μl reaction mixture (8 mM HEPES-KOH (pH 7.0), 60 mM potassium acetate, 5 mM DTT, 1 mM ATP, 25 mM creatine phosphate, 0.03 U/μl creatine kinase, 0.2 U/μl ribonuclease inhibitor, and 1 mM let-7 miRNA) at 25°C for 30 min. The beads were subsequently thoroughly washed in hypotonic buffer and incubated in a final volume of 50 μl reaction mixture (8 mM HEPES-KOH (pH 7.0), 60 mM potassium acetate, 5 mM DTT, 1 mM ATP, 25 mM creatine phosphate, 0.03 U/μl creatine kinase, 0.2 U/μl ribonuclease inhibitor, 10 ng/μl yeast tRNA, and 2000–10,000 cpm cap-labeled mRNA substrate) at 25°C for 2 h. After the final incubation step, the cropping, dicing or slicing reaction mixtures were then added to 200 μl proteinase K buffer (200 mM Tris-HCl (pH 7.5), 25 mM EDTA, 300 mM sodium chloride, 2% w/v SDS, and 50 μg/mL proteinase K), incubated at 65°C for 30 min, and extracted with phenol/chloroform (1:1). RNA was precipitated from the supernatant and resolved by 6% (for cropping and slicing) or 15% urea-PAGE (for dicing).
PAR-CLIP procedure and library construction were conducted as previously reported [41,42]. Sequenced reads were trimmed of 3’ and 5’ adaptors (3’ = "TGGAATTCTCGGGTGCCAAGG"; 5’ = "ATCTCGTATGCCGTCTTCTGCTTG") using flexbar package [51]. Reads that were less than 15 nucleotides in length were discarded. In order to accurately analyze each binding event without the confounding bias introduced by repetitive regions, we identified and removed reads that align to rRNA, tRNA and RepeatMasker sequences (UCSC BDGP R5/dm3 and FlyBase FB2014_03). Elimination of reads mapped to repetitive sequences has significantly affected the overall read yield available for further analyses, thus complicating discovery of SmD1 binding sites. The remaining reads were aligned to Drosophila genome with bowtie algorithm [52]. Mapped locations were only reported for the optimal mismatch-stratum for each read up to a maximum of ten different locations. The resulting alignment was processed with PARALYZER tool (v1.1) [53]. All clusters that have two or more T to C conversion locations were reported. The discovered clusters were further filtered to exclude those where a) read coverage was lower than 10, b) ModeScore< = 0.6, and c) number of unique locations having at least one conversion event exceeded 2. The above non-adaptive filtration helped to remove potential false positive binding events in low coverage regions, where PARALYZER’s signal-to-noise estimation becomes less reliable. The location that a cluster mapped to, relative to a known transcript, was determined based on the FlyBase genome annotation (release 5.57). To assess the extent of SmD1 binding in miRNA precursor regions, ±10Kb around the known miRNA locus were examined and overlapping read clusters detected with BEDTools suite. To evaluate the role of SmD1 on regulation of mRNA splicing, coordinates of exon junctions annotated in FlyBase were extracted. Next, regions of interest around exon–intron junctions at the 5′ and 3′ ends of introns were determined. These loci included: a) 15 bp region around 5’ splice site, of which 5bp were inside the intron, and b) 28bp region around 3’ splice site, where only 3bp overlapped with the 3’ exon. Finally, the intersection between SmD1 binding and mRNA splice sites was computed.
Small RNA libraries were constructed from gel purified 19–24 nt RNA samples using the TruSeq small RNA sample kit according to manufacturer’s manual (Illumina), and sequenced on a GA-II machine. Reads were trimmed of 3’ and 5’ adaptors (3’ = "TGGAATTCTCGGGTGCCAAGG"; 5’ = "ATCTCGTATGCCGTCTTCTGCTTG") with cutadapt package (http://dx.doi.org/10.14806/ej.17.1.200) and mapped to the fly genome (rel. 5.57) with Bowtie software suite (v.1.1.1) [52] in a sequential manner {e.g. progressively relaxing constraints for permissive number of mismatches until the limit of 2 was reached (–v[0–2]-best)}. The alignment was processed with HTSeq [54] and read abundances assigned to various genomic elements calculated. To assess read numbers mapping to transposable elements loci we used only LINE-like and LTR elements annotated in RepeatMasker (UCSC BDGP R5/dm3 and FlyBase FB2014_03). Annotation and genomic coordinates for 3p-CIS-NAT, esiRNA loci used to estimate siRNA abundances were obtained from Eric Lai (personal communication). Reads that failed to map to the fly genome were aligned to the flock house virus (FHV) genome and counted.
See S7 Table.
SmD1 was knocked down in the Drosophila heart by cardiac-specific RNAi using the UAS/Gal4 system [55]. Female tinCΔ4-Gal4 flies [56] were crossed to UAS-shSmD1 (Bloomington stock 34834) and UAS-shGFP (gift from Dr. Norbert Perrimon) respectively. The female F1 progeny was aged for 3 weeks at 25°C and heart function was analyzed using high-speed recordings of semi-dissected hearts [57].
|
10.1371/journal.pntd.0000397 | Assays to Detect β-Tubulin Codon 200 Polymorphism in Trichuris trichiura and Ascaris lumbricoides | The soil-transmitted helminths (STH) Ascaris lumbricoides and Trichuris trichiura are gastrointestinal parasites causing many disabilities to humans, particularly children. The benzimidazole (BZ) drugs, albendazole (ALB) and mebendazole (MBZ), are commonly used for mass treatment for STH. Unfortunately, there is concern that increased use of anthelmintics could select for resistant populations of these human parasites. In veterinary parasites, and lately in filarial nematodes, a single amino acid substitution from phenylalanine to tyrosine, known to be associated with benzimidazole resistance, has been found in parasite β-tubulin at position 200. We have developed pyrosequencer assays for codon 200 (TTC or TAC) in A. lumbricoides and T. trichiura to screen for this single nucleotide polymorphism (SNP).
Pyrosequencing assays were developed and evaluated for detecting the TTC or TAC SNP at codon 200 in β-tubulin in A. lumbricoides and T. trichiura. Genomic DNA from individual worms, eggs isolated from individual adult worms or from fecal samples with known treatment history and origin, were sequenced at β-tubulin by pyrosequencing, and genotypes were confirmed by conventional sequencing. The assays were applied to adult worms from a benzimidazole-naïve population in Kenya. Following this, these assays were applied to individual worms and pooled eggs from people in East Africa (Uganda and Zanzibar) and Central America (Panama) where mass anthelmintic drug programs had been implemented. All A. lumbricoides samples were TTC. However, we found 0.4% homozygous TAC/TAC in T. trichiura worms from non-treated people in Kenya, and 63% of T. trichiura egg pools from treated people in Panama contained only TAC.
Although the codon 200 TAC SNP was not found in any of the A. lumbricoides samples analyzed, a rapid genotyping assay has been developed that can be used to examine larger populations of this parasite and to monitor for possible benzimidazole resistance development. The TAC SNP at codon 200, associated with benzimidazole resistance in other nematodes, does occur in T. trichiura, and a rapid assay has been developed to allow populations of this parasite to be monitored for the frequency of this SNP. Sample sizes were small, anthelmintic efficacy was not assessed, and treated and non-treated samples were from different locations, so these frequencies cannot be extrapolated to other populations of T. trichiura or to a conclusion about resistance to treatment. The occurrence of the TAC SNP at codon 200 of β-tubulin in T. trichiura may explain why benzimidazole anthelmintics are not always highly effective against this species of STH. These assays will be useful in assessing appropriate treatment in areas of high T. trichiura prevalence and in monitoring for possible resistance development in these STH.
| The soil-transmitted helminths Ascaris lumbricoides and Trichuris trichiura are gastrointestinal nematodes causing many disabilities in tropical parts of the developing world. Control programs, such as “The Focussing Resources on Effective School Health” (FRESH) Partnership, have been implemented to remove human soil-transmitted nematodes through large-scale use of benzimidazole anthelmintic drugs for school-aged children in developing countries. The benzimidazole drugs albendazole and mebendazole are commonly used as a single annual treatment in areas where the burden is high. In veterinary nematodes, repeated use of these anthelmintics has selected for resistant populations. Resistance to benzimidazoles is commonly associated with a single amino acid substitution from phenylalanine to tyrosine in the β-tubulin gene at position 200. In this study, we have developed pyrosequencing assays for codon 200 in A. lumbricoides and T. trichiura to screen for this single nucleotide polymorphism (SNP) in β-tubulin. The 200Tyr SNP was detected at low frequency in T. trichiura from non-treated people from Kenya and at high frequency in T. trichiura from treated people from Panama. The presence of the resistance-associated SNP may play a role in the sometimes low and variable efficacy of benzimidazole anthelmintics against T. trichiura.
| The soil-transmitted helminths (STH) are gastrointestinal nematodes widely distributed throughout the tropical and subtropical parts of the developing world. More than a billion people are infected with at least one species and 300 million are estimated to have severe infections with more than one of these parasites [1]. Infection often causes chronic disability but in certain instances may precipitate death. School age children are the most at risk of infection with STH and can show symptoms of malnourishment, experience growth stunting and intellectual retardation, with cognitive and educational deficits [1]. Control programs such as the “Focussing Resources on Effective School Health” partnership (FRESH) have been implemented in endemic countries to reduce the morbidity of school-aged children by using a single annual treatment with benzimidazole drugs : either albendazole or mebendazole [2].
BZ drugs are broad spectrum anthelmintics that bind to β-tubulin, causing interference with tubulin polymerization and destabilization of microtubules [3]. Mass drug administration (MDA) programs reduce the incidence and intensity of infections; however they can also cause a selection pressure on the parasites to develop resistance. Many studies have demonstrated that the widespread and frequent use of anthelmintic drugs in veterinary nematodes has led to the development of resistance [4],[5]. This resistance is usually due to a single nucleotide polymorphism (SNP) which causes an amino acid substitution from phenylalanine (Phe, TTC) to tyrosine (Tyr, TAC) in parasite β-tubulin at codon 200 [6]. A similar SNP at codon 167 (Phe167Tyr) or a glutamate to alanine change at codon 198 (Glu198Ala) can also occasionally be associated with benzimidazole resistance [7]. In human STH, and especially in hookworms, reports have suggested the development of benzimidazole resistance but failed to provide conclusive evidence [8]. If resistance against albendazole and mebendazole occurs, it will be a major threat to the mass de-worming programs in developing countries.
The objectives of this research were: (i) to determine the genomic sequences of the β-tubulin around codon 200 in A. lumbricoides and T. trichiura, (ii) to develop pyrosequencing assays for the detection of phenylalanine 200 or tyrosine 200 in the β-tubulin gene of each of these nematodes, and (iii) to assay DNA of individual adult worms and pooled eggs from the field to determine whether the Phe200Tyr SNP can be found and to obtain initial assessments of the frequency in areas that are either naïve to benzimidazole treatment or have been subject to benzimidazole treatment.
In this study, all parasite samples were from school-age children who have been infected naturally by A. lumbricoides and/or T. trichiura. Ethical approval was obtained from the Research Ethics Board, McGill University, the London School of Hygiene and Tropical Medicine and the Kenya Medical Research Institute. Samples of adult worms and/or eggs were collected from different locations with different treatment histories. Individual adult worms, 39 T. trichiura (20 males and 19 females) and 38 A. lumbricoides (19 males and 19 females), were collected from children from Kisumu, Kenya (latitude: 00°03′ S, longitude: 35°5′ E). These children were naïve for anthelmintic treatment; however, they received a single dose of “Combantrin” (pyrantel) in order to expel the adult worms in feces. These worms were considered as control parasites and stored at −80°C until needed.
Faecal samples with A. lumbricoides and T. trichiura eggs were collected in Panama in the Comarca Ngobe Bugle region (latitude: 8°3′ N, longitude: 86°12′ W) from 29 children in 3 different schools. These children had received a single dose of ALB (400 mg). Prior treatment, before commencement of the study, may have occurred, but was not documented. Stool samples were preserved in 70% alcohol and stored at 4°C.
From another study carried out in Zanzibar and Uganda [9], we received 91 DNA samples from individual A. lumbricoides adult worms recovered after being expelled with “Combantrin” from patients who lived within areas where large scale interventions with benzimidazoles were ongoing. The DNA was preserved in alcohol and stored at 4°C.
The DNA assay was applied to eggs from non-treated (BZ-naïve subjects) and eggs from subjects in treated areas. A. lumbricoides eggs from Kenya were recovered from the uterus of adult female worms. Before the dissection of each female worm, the anterior end of the parasite was separated from the body to avoid contamination of adult DNA with DNA from eggs. Female worms were then opened longitudinally. At approximately one quarter length along the body, the uterus is attached to the genital pore and subsequently divides into two branches. Each uterine branch was cut and opened to release the eggs. From the study carried out in Panama, 29 pooled A. lumbricoides egg samples and 8 pooled T. trichiura egg samples were recovered from stool samples by using a flotation technique [10].
Following the recovery of eggs, DNA was extracted from pooled A. lumbricoides eggs collected from the uterus of adult female worms. It was also extracted from individual A. lumbricoides female and male worms and from individual T. trichiura adult worms using the DNeasy Blood and Tissue Extraction Kit (Qiagen) according to the manufacturer's protocol. For the DNA extraction of A. lumbricoides and T. trichiura eggs from stools samples from Panama, the QIAamp DNA stool mini kit (Qiagen) was used according to the manufacturer's protocol.
Genomic sequence (GenBank AF034219) was available for T. trichiura β-tubulin, [11], but was not available for A. lumbricoides, so it was necessary to generate this sequence. Total RNA was extracted from adult A. lumbricoides from Kenya by TRIzol Reagent (Invitrogen Life Technologies, Burlington, ON) according to the manufacturer's protocol. Then, the total RNA was reverse transcribed with the oligo-dt (12–18) primer according to the manufacturer's instruction.
For the initial isolation of the A. lumbricoides β-tubulin gene, cDNA was amplified with degenerate primers. These primers were designed based on the conserved region of β-tubulin of six related nematodes: Haemonchus contortus (GenBank, M76493 - isotype 1), Brugia malayi (GenBank, AY705382), Necator americanus (GenBank, EF392851), Trichuris trichiura (GenBank, AF118385), Teladorsagia circumcincta (GenBank, Z69258 - isotype 1) and Onchocerca volvulus (GenBank, AF01886) (the alignments of β-tubulin cDNAs for 11 nematodes are given in Figure S1). Two sets of primers were designed for a nested PCR approach. In the first round PCR, the cDNA was amplified with the outer sense primer (5′-3′) CAAAGTGGAGCKGGHCACAACTGGC and the outer antisense primer (5′-3′) CGBAGATCHGCATTCAGCTGHCCAGG. The PCR product from the first round was then used as a template for the subsequent amplification using the nested primers, sense (5′-3′) CTYGGTGGAGGYACMGGWTC and antisense (5′-3′) CGBAGATCHGCATTCAGCTGHCCAGG. For both rounds, the PCR conditions were an initial denaturation at 95°C for 3 min, followed by 35 cycles of 95°C for 45 s, 60°C for 45 s, 72°C for 1 min and a final extension at 72°C for 5 min. Prior to cloning, a 5 µl aliquot of PCR product from the nested reaction was examined on a 1% agarose (TAE) electrophoresis gel to confirm the size of the product. The amplified PCR products were cloned into the pCR2.1 TOPO vector using a TOPO-TA-Cloning kit (Invitrogen, Life Technologies, Burlington, ON), as per the manufacturer's instructions and then sequenced.
Based on the sequenced fragment, gene-specific primers were designed for the 5′ and 3′ rapid amplification cDNA ends (RACE) reactions. For the 5′ RACE reaction two primers were designed, (5′-3′) CGTTGAGCGCCCTGTATGC and (5′-3′) CAACACCACATCAGAAACCT and used in a semi-nested PCR reaction with the nematode splice leader sequence SL1 (5′-3′) GGTTTAATTACCCAAGTTTGAG [12]. The amplification conditions for the primary reaction were 3 min at 94°C, followed by 40 cycles at 94°C for 45 s, 59.4°C for 45 s, 68°C for 1 min and a final extension at 68°C for 5 min. For the semi-nested reaction the amplification conditions were as outlined above with an annealing temperature of 55.8°C.
To isolate the 3′ end of β-tubulin cDNA, two gene specifics primers were designed (5′-3′) CCACGTCTTCACTTCTTCATG and (5′-3′) GTACGACATCTGCTTCAGGACCCTG, and used in a nested PCR reaction with oligo adaptor primers B1 (3231)(5′-3′) CCTCTGAAGGTTCACGGAT and B2 (3232)(5′-3′) CACGGATCCACATCTAGAT, respectively. The PCR conditions for both reactions were as described above with an annealing temperature of 60°C for the first reaction and 55.6°C for the nested reaction. The resulting fragments after the second round of each reaction were purified using the QIAGEN PCR purification kit according to the manufacturer's protocol. They were subsequently ligated into pGEM-T cloning vector (Promega, Madison, WI) and sequenced from both directions with SP6 and T7 vector primers. Phylogenetic analysis was undertaken with the UPGMA method and performed using the Mac vector program in order to identify the relationship between β-tubulin sequences of 13 nematodes.
To optimize the pyrosequencing DNA assay, two control plasmids were constructed for each species and contained either the “sensitive” codon TTC for the wild type or a mutation, generated by site-directed mutagenesis (TAC - mutant type), inserted at position 200. These plasmids were based on the amplification of a small portion of β-tubulin genomic sequence from A. lumbricoides and T. trichiura worms that were naïve for benzimidazole treatment. Gene-specific primers were designed to amplify a small fragment of genomic DNA (GenBank, FJ501301) surrounding the codon 200. In A. lumbricoides, primers were based on the sequence of the fragment of A. lumbricoides β-tubulin cDNA. All specific primers were designed with the software gene runner in the exonic region, sense (5′-3′) GGTGGAGGCACAGGATCTGGC, antisense (5′-3′) GCAGCCGCTCCTCG. For T. trichiura, primers were directly designed from the genomic sequence (GenBank AF034219), sense (5′-3′) GGTTTCAGATACAGTTGTAG (position 1212-1231, located 76 amino acids upstream of the first T of the codon 200 TTC) and antisense (5′-3′) CAAATGATTTAAGTCTCCG (position 1356–1374, located 146 amino acids upstream of the first T of the codon 200 TTC). PCR reaction conditions were an initial denaturation at 94°C for 3 min, followed by 35 cycles of 94°C for 45 s, an annealing temperature of 52°C in A. lumbricoides and 56°C in T. trichiura, for 45 s, and 68°C for 1 min and a final extension at 68°C for 5 min. Resulting fragments were cloned and sequenced as described above.
The site directed mutagenesis strategy consisted of the amplification of two overlapping fragments using outer and inner mutagenesis primer pairs. Then, the opposing PCR strands were annealed at overlapping regions, extended, and amplified by PCR to produce the desired full-length strand. Two pairs of primers were designed, in A. lumbricoides outer sense primers (5′-3′) GTTTCTGATGTGGTGTTGGAG, antisense (5′-3′) CAAATGGTTGAGGTCTCCG and inner mutagenesis primer, sense (5′-3′) CGATGAAACCTACTGCATTGACAATG, antisense (5′-3′) CAAATGGTTGAGGTCTCCG. The PCR conditions were as outlined above with an annealing temperature of 54°C. PCR products were electrophoresed though agarose gels and then purified. Ten µl (10–100 ng DNA) of each purified PCR product were mixed and denatured at 94°C for 3 min. The second PCR reaction had an annealing temperature of 53°C and allowed the production of the desired full-length strand. In T. trichiura the protocol used to generate the mutant type plasmid was as described above with outer sense (5′-3′) GGTTTCAGATACAGTTGTAG (position 1013–1031) and antisense (5′-3′) CAAATGATTTAAGTCTCCG (position 1356–1374) and inside mutagenesis primers, sense (5′-3′) CGGACGAAACATACTGCATAGATAATG, antisense (5′-3′) CATTATCTATGCAGTATGTTTCGTCCG (position 1277–1301). The annealing temperature for the first PCR was 54°C and 50°C for the second PCR. The fragments obtained with the desired mutation for both parasites were purified, cloned and subsequently sequenced.
Pyrosequencing was used to detect a possible SNP in the genomic DNA from the field samples. First, a smaller fragment of DNA from the control plasmids that surrounded the position 200 was amplified. Subsequently, we amplified the same portion of β-tubulin DNA of eggs isolated from individual A. lumbricoides adult worms from Kenya, Zanzibar, Uganda and DNA from pools of eggs from Panama and Kenya. For T. trichiura, DNA of individual worms from Kenya and pooled eggs from Panama was also amplified. A fragment of 158 bp of A. lumbricoides β-tubulin DNA was amplified with primers: sense (5′-3′) AGGTTTCTGATGTGGTGTTGGA and antisense (5′-3′) TATGTGGGATTTGTAAGCTTCAG. For T. trichiura, a fragment of 163 bp was amplified with gene-specific primers: sense (5′-3′) AGGTTTCAGATACAGTTGTAG (position 1211–1231), antisense (5′-3′) CAAATGATTTAAGTCTCCG (position 1356–1374). The antisense primer was biotinylated (Invitrogen, Life technologies, Burlington, ON) at its 5′ end. For both reactions, the thermal cycling conditions included an initial incubation at 94°C for 3 min, followed by 50 cycles of 94°C for 45 s, an annealing temperature of 58.7°C for A. lumbricoides, 55°C for T. trichiura, 68°C for 1 min and a final extension at 68° C for 6 min. Biotinylated PCR products were immobilized on streptavidin-coated Sepharose beads (Amersham Biosciences, Piscataway, NJ) and sequencing primers used for SNP analysis in the PSQ96MA instrument (Biotage AB, Charlottesville, VA) were: (5′-3′) GAGAACACGGACGAAACAT (position 1270–1288) for T. trichiura and (5′-3′) GAGAACACCGATGAAACCT for A. lumbricoides.
Genotype sequences obtained by pyrosequencing were confirmed by conventional sequencing at the Quebec/McGill University Genome Centre.
Fresh RNA from A. lumbricoides, naïve for BZ-treatment, allowed us to generate a high quality partial length β-tubulin cDNA (GenBank, EU814697). The length of the portion sequenced was 1137 pb. The translation product revealed a putative sequence of 378 amino acids. The phylogenetic tree (Figure 1) showed that the A. lumbricoides sequence did not have a close relationship with the Strongylida order which includes the hookworms, A. duodenale and N. americanus as well as the veterinary nematodes H. contortus and T. circumcincta. It was also the case for the Trichocephalida which includes T. trichiura. In fact, the A. lumbricoides cDNA sequence seemed to be more closely related to the Spirudida order which includes the filarial nematodes.
Based on A. lumbricoides and T. trichiura β-tubulin, diagnostic SNP assays, optimized with control plasmids, were applied. An alignment of the portion around the codon of interest in the translated β-tubulin protein sequences of wild type (WT) and the mutant type (MT) plasmids with sequences of other nematodes highlighted a high conservation of the protein sequence. It also showed the (T→A) substitution (Phe200Tyr at the amino acid level, Figure 2) that is associated with BZ resistance in veterinary nematodes [13].
Pyrosequencing assays were designed to genotype the single nucleotide polymorphism (SNP) at codon 200 of β-tubulin. This technique allows sequencing of short fragments of DNA in a very short period of time. The pyrosequencer used can process 96 samples in one hour, and the technique is highly accurate, making the results reliable and easy to interpret. This test was first evaluated with the help of the control plasmids. For A. lumbricoides and T. trichiura, we obtained distinct pyrogram profiles with the WT and MT plasmids. Pyrograms of each Ascaris and Trichuris control plasmid showed a single peak for the WT and MT plasmid that identified the “susceptible” TTC codon and the ‘resistant’ TAC, respectively, in the codon 200 position. Thus, these results confirmed that the diagnostic test was efficient and could easily and clearly identify the genotype of A. lumbricoides and T. trichiura worms for this position in the β-tubulin gene.
Once we obtained the expected genotype profiles, we applied the pyrosequencing diagnostic assay to A. lumbricoides adult worms and pooled eggs, and to T. trichiura adult worms from Kenya from people who had not received treatment with BZ drugs. For A. lumbricoides all individual females, males and pools of eggs were “sensitive” T/T. In contrast, we found more diversity in the T. trichiura population with regard to the (TAC) SNP. The frequency of heterozygote and homozygote worms was almost the same for male and female worms. Out of 20 males, 11 were homozygous T/T (55%), 8 were heterozygous T/A (40%) and 1 was homozygous A/A (5%). Among the 19 female worms, 11 were homozygous T/T (58%), and 8 were heterozygous T/A (42%).
In order to investigate if repeated exposure to BZ treatment had an impact on the polymorphism in β-tubulin at codon 200, associated with resistance in veterinary nematodes, we screened the β-tubulin genes of A. lumbricoides and T. trichiura from different areas where mass drug administration (MDA) programs had been in operation . Pyrosequencing assays, designed previously for each species were applied to detect the presence or absence of TTC or TAC at the SNP of interest. DNA from pooled eggs or DNA from individual worms was analysed. TTC and TAC at codon 200, in the parasite samples, were confirmed by conventional sequencing conducted at the Quebec/McGill University Genome Centre.
The genotype frequencies obtained for all species are shown in and are as follows:
The possible development of resistance to ALB and MBZ in human nematodes is a threat to MDA programs [8]. A number of studies have reported a reduced efficacy of ALB and MBZ anthelmintics in human STH after repeated treatments [9],[14]. Resistance in other nematodes is known to be associated with mutation in β-tubulin preventing the binding of BZ drugs [15]. Consequently, there is a need to investigate and monitor the frequency of SNPs in the β-tubulin gene [16],[17] that have been associated with BZ resistance. Bennett and co-workers used conventional sequencing on 72 individual T. trichiura, mostly from untreated individuals, and did not observe polymorphism at codon 200 in β-tubulin [18]. Using real time PCR and pyrosequencing, Schwab and colleagues have developed rapid genetic assays for individual Wuchereria bancrofti microfilaria [19]. Schwenkenbecher and co-workers have used real time PCR to assay for resistance SNPs in hookworms [20], but did not report finding the resistance associated SNP. However, the β-tubulin gene of A. lumbricoides had not been previously analysed. In this study, we developed pyrosequencing assays to detect SNPs in the β-tubulin genes of A. lumbricoides and T. trichiura, and analyzed samples obtained from naïve or benzimidazole treated individuals. Control plasmids, WT (TTC) and MT (TAC) were analyzed using the pyrosequencing method. The expected sequences, TTC (“sensitive”) and TAC (“resistant”) were obtained indicating that the assay was efficient. The development of these genetic assays allows a rapid screening method for the detection of possible resistance alleles in human parasites. This development is consistent with the aim of the Consortium for Anthelminthic Resistance SNPs (CARS) to develop panels of molecular markers for anthelmintic resistance in human and veterinary nematodes [15]. To date, there are not many reliable and accurate tools for diagnosing resistance in STH. Biological tests are used to assess resistance [21],[22] but have limited application as they generally can detect resistance only if the proportion of resistant worms is more than 25% [23]. The development of molecular assays for SNPs in the β-tubulin genes of STH gives new hope for monitoring for anthelmintic resistance. This is particularly important considering the difficulty in measuring resistance in helminth parasites of humans compared to helminth parasites of animals [24].
We genotyped STH species from Kenya before the implementation of MDA with ALB or MBZ in order to get baseline information on the SNP frequencies in the β-tubulin genes. In A. lumbricoides samples, heterozygous TTC/TAC or homozygous TAC/TAC were not found in samples from non-treated subjects. However, it would be interesting to do more sampling to confirm that the homozygous “sensitive” genotype predominates in different A. lumbricoides populations. It would also be interesting to analyse parasites from the same population of hosts after several rounds of treatment with ALB to see if repeated treatment over some years would result in the Tyr200 SNP being detected in A. lumbricoides worms.
In contrast, in T. trichiura, the SNP with TAC at codon 200 of β-tubulin was present in samples from the same Kenyan population. In T. trichiura, only one β-tubulin gene has been identified even with low stringency Southern blots [11]. This suggests that T. trichiura may carry only one isotype of the β-tubulin and that molecular change in this β-tubulin alone might result in resistance. In some veterinary nematodes, the sensitivity or resistance to BZ can be modulated by a second β-tubulin isotype [25]. Samples with both TTC and TAC, as well as others with TAC alone were identified in the T. trichiura population. Nevertheless, our sample size was low: out of 39 T. trichiura individual worms, 18 were heterozygous TTC/TAC and one individual was homozygous TAC/TAC. However, these frequencies may not be representative of a wider population of T. trichiura. Taking into account that the study was carried out in an area where MDA with BZs had not previously been implemented, the finding of a moderately high frequency of heterozygotes, as well as a low frequency of the “resistance” homozygote genotype, raises a concern should BZ deworming programs be implemented in the same region. However, even though the SNP was found as heterozygous in many parasites, a resistance phenotypic may not be apparent as the resistance may be recessive as has been reported for the Phe200Tyr SNP associated with resistance in Teladorsagia circumcincta [26]. This may delay the appearance of a resistance phenotype. However, further work will be necessary to correlate susceptibility/resistance phenotypes and genotypes to confirm the role of the codon 200 SNP with resistance in T. trichiura and to determine whether resistance is recessive, semi-dominant or dominant. However, repeated exposure with multiple rounds of treatment could lead to the loss of the susceptible allele as has already been demonstrated in veterinary nematodes [8],[27].
Even though the “resistance” (TAC) SNP was found in T. trichiura from MDA-naïve subjects, it is important to take into consideration that deworming programs are already common and have existed for many years in other regions of Kenya [28]–[30]. Communities or regions targeted are quite close to each other and people from treated areas could easily travel to non-treated areas [31].
It is interesting to note that many studies have demonstrated the lower efficacy of BZ drugs against T. trichiura compared with A. lumbricoides [9],[32],[33]. However, the factors involved in this difference of efficacy are not known, but could include pharmacokinetics, given the different locations of A. lumbricoides and T. trichiura, or differences in genetic predisposition to BZ susceptibility. Our current results could support the idea that the difference in sensitivity, between these species, could be due to the occurrence of different alleles with alternatively TTC or TAC at codon 200 in the β-tubulin gene within the gene pool of T. trichiura. In veterinary nematodes a number of factors including the frequency of anthelmintic use, proportion of the parasite population exposed to treatment, parasite turnover and other factors may contribute to the rate of development of resistance [34]. Implementation of very large scale control programs for STH and lymphatic filaria could increase drug selection pressure and possibly the frequency of the Tyr200 SNP in T. trichiura. Further analyses on the Phe200Tyr SNP in β-tubulin of T. trichiura worms and correlation of its frequency with benzimidazole efficacy will be important to determine so that the possible development of drug resistance as part of MDA programs for STH could be monitored by these assays. If the Tyr200 SNP is confirmed to be associated with BZ resistance in T. trichiura, it will also be important to investigate aspects of population dynamics which impact on the rate of change in SNP frequency in populations under drug pressure.
The final objective of our study was to determine if the frequency of the β-tubulin SNP varies after repeated treatment with BZ and if the resistance-associated SNP was high in A. lumbricoides and T. trichiura samples from areas where there had been MDA with ALB or MBZ. None of the A. lumbricoides samples from Panama, Uganda or Zanzibar, examined by pyrosequencing, carried the TAC mutation at codon 200 in the β-tubulin gene. However, in pooled egg samples of T. trichiura from Panama, we found the Phe200Tyr SNP in egg pools from hosts who were reported to have been treated with ALB. For the T. trichiura pooled egg samples from Panama, both mixed TTC/TAC and TAC alone, in different pools, were found. It is important to point out that as we used pooled eggs, we could not determine the frequency of different genotypes in the worm population. This means that SNP frequencies refer to between T. trichiura egg pools and not within a particular egg pool. Because of this, and the small number of samples, the SNP frequencies must be interpreted with caution. Based on the experience in veterinary nematodes where benzimidazole resistance appears to be recessive [26], a high frequency of the homozygous “resistance” genotype could affect the cure rate and the drug efficacy, and repeated treatment may increase the frequency of homozygous “resistance” genotypes and lead to a rapid development of drug resistance. A study carried out in South Africa on the drug efficacy of 400 mg ALB demonstrated a low cure rate for T. trichiura and the authors concluded that this drug was not appropriate for a deworming program in this region [35]. There is an urgent need for studies correlating drug efficacy with genotype.
Knowledge of the β-tubulin sequences will enable us to develop similar pyrosequencing assays for alternative SNPs in β-tubulin of STH, at codons 167 and 198, known to be involved in drug resistance in cyathostome nematodes [26],[36] and occasionally in H. contortus [7],[37].
Studies such as ours are required to help control program managers to make appropriate decisions for the design of treatment programs against these harmful parasites. In this study, we described a reliable, fast and easy DNA assay based on a pyrosequencing technique for two soil-transmitted nematodes of humans. This technique allows the SNP analysis of large numbers of egg samples or other parasite stages, in a short period of time. For the first time, we characterized the partial β-tubulin cDNA and genomic DNA sequence of A. lumbricoides. Knowledge of the β-tubulin sequences is important as little is known about resistance to benzimidazole drugs in human helminths and also because of the rapid development of drug resistance in veterinary nematodes. The SNP with TAC was found in individual worms of T. trichiura from non-treated people in Kenya, and in T. trichiura egg pools from treated people in Panama. These findings provide a possible explanation for the sometimes low efficacy of benzimidazole anthelmintics against T. trichiura and an important warning of the possibility that resistance may develop, particularly in T. trichiura. It is crucial to continue monitoring for the frequency of the codon 200 TTC/TAC SNP in areas under MDA and to confirm whether the TAC allele confers BZ resistance in these STHs.
|
10.1371/journal.ppat.1001291 | The Rubella Virus Capsid Is an Anti-Apoptotic Protein that Attenuates the Pore-Forming Ability of Bax | Apoptosis is an important mechanism by which virus-infected cells are eliminated from the host. Accordingly, many viruses have evolved strategies to prevent or delay apoptosis in order to provide a window of opportunity in which virus replication, assembly and egress can take place. Interfering with apoptosis may also be important for establishment and/or maintenance of persistent infections. Whereas large DNA viruses have the luxury of encoding accessory proteins whose primary function is to undermine programmed cell death pathways, it is generally thought that most RNA viruses do not encode these types of proteins. Here we report that the multifunctional capsid protein of Rubella virus is a potent inhibitor of apoptosis. The main mechanism of action was specific for Bax as capsid bound Bax and prevented Bax-induced apoptosis but did not bind Bak nor inhibit Bak-induced apoptosis. Intriguingly, interaction with capsid protein resulted in activation of Bax in the absence of apoptotic stimuli, however, release of cytochrome c from mitochondria and concomitant activation of caspase 3 did not occur. Accordingly, we propose that binding of capsid to Bax induces the formation of hetero-oligomers that are incompetent for pore formation. Importantly, data from reverse genetic studies are consistent with a scenario in which the anti-apoptotic activity of capsid protein is important for virus replication. If so, this would be among the first demonstrations showing that blocking apoptosis is important for replication of an RNA virus. Finally, it is tempting to speculate that other slowly replicating RNA viruses employ similar mechanisms to avoid killing infected cells.
| Among the variety of defense systems employed by mammalian cells to combat virus infection, apoptosis or programmed cell death is the most drastic response. Some large DNA viruses encode proteins whose sole function is to block apoptosis. Conversely, very little is known about whether RNA viruses encode analogous proteins. In many cases, RNA viruses are able to replicate before cell death occurs, which may be one reason why so little thought has been given to this topic. However, a number of RNA viruses, some of which are important human pathogens, have slow replication cycles and it stands to reason that they must block apoptosis during this time period. Here we show that the multifunctional capsid protein of Rubella virus is a potent inhibitor of apoptosis. Data from reverse genetic experiments suggest that the anti-apoptotic function of a virus-encoded protein is important for replication of an RNA virus. We anticipate that other slowly replicating RNA viruses may employ similar mechanisms and, as such, these studies have implications for development of novel anti-virals and vaccines.
| Rubella virus (RV) is an enveloped positive strand RNA virus in the family Togaviridae and is the sole member of the genus Rubivirus (reviewed in [1]). Humans are the only natural host for RV and in most cases the virus causes a systemic infection the symptoms of which include maculopapular rash, lymphadenopathy, low-grade fever, conjunctivitis and sore throat. However, RV infections can be complicated by the appearance of acute or chronic arthralgia, arthritis, thrombocytopenia and encephalopathy. In utero infection during the first trimester of pregnancy often results in a characteristic series of birth defects known as congenital Rubella syndrome. Worldwide, RV is thought to cause more birth defects that any other infectious agent yet, very little is known about molecular aspects of viral pathogenesis. A number of studies suggest that viral persistence may underlie some of the most serious aspects of infection including congenital Rubella syndrome and arthritis [2], [3], [4], [5], [6].
Among the togavirus family, RV is unique in that its replication is associated with mitochondria. The link between RV infection and this organelle first became apparent when analysis of purified virions revealed that cardiolipin, a phospholipid that is only found in mitochondria, is a significant component of the RV envelope [7]. Subsequently, it was discovered that RV infected cells exhibit striking mitochondrial defects. Virus infection induces clustering of mitochondria in the perinuclear region as well as formation of electron-dense plaques between apposing mitochondrial cisternae: structures that have been termed confronting membranes [8], [9]. The function of these structures is not known but expression of capsid protein in the absence of other RV proteins is sufficient to induce their formation [10]. A large pool of the capsid protein localizes to the surface of mitochondria [11] and the inter-mitochondrial plaques [12] but given that assembly of RV virions occurs primarily on Golgi membranes, the targeting of the capsid to this organelle likely reflects a nonstructural function of this protein.
The studies described above underscore the close link between the capsid protein and mitochondria in RV biology and form the basis for our central hypothesis; that association of the RV capsid protein with mitochondria is important for virus replication. All viruses must contend with host cell anti-viral mechanisms and large DNA viruses have the luxury of harboring in many cases, multiple genes devoted to thwarting host cell defenses (reviewed in [13]). In contrast, simple RNA viruses express a very limited number of proteins, most of which are directly involved in replication and virus assembly. Accordingly, it is beneficial if not essential that these viral proteins have multiple functions.
It is well documented that togavirus infection often results in apoptotic death of mammalian cells (reviewed in [14], [15]) and to our knowledge, there are no published studies showing that members of this virus family inhibit programmed cell death pathways. With the exception of RV, togavirus replication and virus egress from vertebrate cells occurs within 4–6 hours followed by extensive death by apoptosis within 24 hours. Accordingly, for most togaviruses, preventing apoptosis is likely not required in order for efficient replication to occur. In contrast, the replication cycle for RV is unusually slow; the eclipse period is at least 12 hours and viral titers peak virion secretion occurs between 48–72 (reviewed in [16]). RV-induced apoptosis in mammalian cells has been reported but generally, extensive cytopathic effect is not a hallmark of RV infection. When apoptosis does occur, it is not until 5–7 days post-infection that maximum levels are reached [17] and this is well past the peak virus production phase. In the present study, we report that the capsid protein blocks apoptosis in RV infected cells most likely to allow sufficient time for virus replication. This process occurs at the level of mitochondria through a Bax-dependent pathway.
We reasoned that in order for RV replication and virion secretion to increase through 48 hours and beyond, programmed cell death must be inhibited during this period. Accordingly, we compared the levels of apoptosis in RV and mock-infected A549 cells by indirect immunofluorescence using an antibody specific for activated caspase 3. Interestingly, less than 5% of infected cells exhibited signs of apoptosis 48 hours post-infection (Figure 1A and B). Moreover, when challenged with the kinase inhibitor staurosporine, a potent inducer of apoptosis [18], RV infected cells were significantly more resistant to apoptosis than mock infected cells. Specifically, the percentage of caspase 3-positive cells was almost three fold lower in the infected samples. This was not due to detachment of infected cells as data in Figure 1C show that treatment with staurosporine did not cause significant loss of infected cells. Finally, Figure 1D shows that even after 72 hours, RV infection does not significantly affect the percentage of viable A549 cells. Together, these data indicate that RV-infected A549 cells are resistant to programmed cell death.
Next, we sought to determine which viral protein(s) was primarily responsible for protecting infected cells against apoptosis. Previous studies have indicated that expression of the nonstructural proteins p150 and p90 are cytotoxic [19], [20] and therefore, we focused our attention on the virus structural proteins. Plasmids encoding glycoproteins E2 and E1 or capsid, were transiently tranfected into A549 cells and at 40 hours post-transfection, cells were induced to undergo apoptosis by treatment with anti-Fas. Samples were processed for indirect immunofluorescence (Figure 2A) and the numbers of active caspase 3-positive transfectants were determined. Data in Figure 2B show that the levels of apoptosis were similar in cells expressing the viral glycoproteins E2 and E1 and the negative control protein eGFP. In contrast, expression of the RV capsid protein was just as protective against anti-Fas as the well-characterized anti-apoptotic protein Bcl-XL [21]. Compared to eGFP or E2E1 transfectants that were treated with anti-Fas, the percentage of apoptotic cells among capsid transfectants was three fold lower (Figure 2B).
We also examined whether RV infection and/or capsid expression protects Vero cells from apoptosis. This cell line is used extensively to study RV replication and similar to what was observed with A549, infection of Vero cells with RV, or transient expression of capsid protein conferred protection from staurosporine-induced apoptosis (Figure S1A, B, arrowheads). Because Vero cells do not respond to anti-Fas treatment, it was not possible to determine the effect of capsid expression on death receptor pathways. These data appear to be at odds with a previous study which reported that the RV capsid was pro-apoptotic in RK-13 cells [22]; a cell line that is exquisitely sensitive to RV-induced apoptosis [23]. Accordingly, we assayed intrinsic and extrinsic apoptotic pathways by staurosporine and anti-Fas treatment of RK-13 cells at 48 and 72 hours post-transfection. In both cases, expression of the capsid protein conferred resistance to apoptosis similar to Bcl-XL (Figure S2). Together, these data indicate that the RV capsid is an anti-apoptotic protein that protects cells from multiple apoptotic stimuli.
We next endeavored to identify what step in apoptotic signaling was blocked by capsid protein. For these experiments, lentiviral transduction was used to create A549 cells that stably express capsid protein under the control of a doxycycline-regulated promoter. Results from indirect immunofluorescence showed that less than 50% of the polyclonal population of transduced cells expressed RV capsid following doxycyline treatment (Figure 3A). Similar to results shown in Figure 2, induction of capsid expression protected the stably transduced A549 cells against staurosporine- and Fas-mediated activation of caspase 3 (Figure S3). To further confirm that apoptotic stimuli do not activate caspases in these cells, we measured the appearance of the downstream caspase 3 substrate, cleaved Poly(ADP-ribose) polymerase (PARP). Figure 3B shows that expression of capsid protein results in decreased anti-Fas-induced cleavage of PARP compared to luciferase-expressing cells. These data indicate that capsid protects A549 cells from staurosporine and anti-Fas treatment by blocking caspase activation.
We next determined where upstream of caspase 3 activation, that capsid protein acted. Both staurosporine- and anti-Fas- can trigger apoptosis through the mitochondrial pathway, so we tested the ability of capsid protein to block depolarization of mitochondrial membranes in response to apoptotic stimuli. Doxycycline-treated A549 cells expressing capsid protein or luciferase were challenged with staurosporine or anti-Fas and then stained with the membrane potential sensitive dye TMRM. Samples were analyzed by FACS, after which the relative specific cell death levels for each sample were calculated (Figure 3C). Data in Figure 3D show that compared to luciferase, expression of capsid protein reduced the relative specific death induced through intrinsic (staurosporine) or death receptor-dependent pathways (Fas) by 20–35%. However, because less than 50% of the lentivirus-transduced cells express detectable levels of capsid protein, these numbers likely underestimate the true level of protection afforded by stable expression of the RV capsid protein.
Bax and Bak are two key apoptotic molecules that form oligomers on mitochondria [24], [25], [26] and apoptosis occurs when the mitochondrial outer membrane is permeabilized by these pore-forming molecules [27]. Accordingly, we next focused our efforts on these Bcl-2 family members starting with Bax. Normally, Bax is an inactive monomer found in the cytosol or loosely bound to the mitochondrial outer membrane of healthy cells [28], [29]. In response to apoptotic stimuli, Bax activation is characterized by a multi-step process whereby it undergoes a conformational change [30], [31], integrates into the mitochondrial membrane [28], [32] where it forms higher order oligomers [33]. It is the large Bax oligomers that are linked to the formation of membrane pores that facilitate release of mitochondrial cytochrome c and downstream caspase activation [33], [34]. Of these multiple steps, Bax conformational change can be detected by immunoreactivity with a conformation-specific antibody, 6A7 [35], [36]. We observed that RV infection induces Bax conformational change, however cytochrome c remained associated with mitochondria (Figure 4A, arrows). Moreover, Bax conformational change as detected by 6A7 staining was evident in the majority (76%) of cells expressing capsid protein (Figure 4B arrows). In contrast, among cells expressing the viral glycoproteins E2 and E1, only 6% contained activated Bax. Despite initial stimulation of Bax, similar to infected cells, no loss of cytochrome c from mitochondria was observed in capsid-expressing cells. Because capsid protein stimulates Bax in a manner that does not produce functional pores that mediate efflux of cytochrome c, we initially thought that capsid protein blocks oligomerization of Bax. However, data in Figure 5A indicate that this is not the case. Rather, our results suggest that capsid protein and Bax form mixed large hetero-oligomers even in the absence of apoptotic stimuli. Indeed, reciprocal co-immunoprecipitation experiments confirmed that capsid forms a stable complex with Bax (Figure 5B). Staurosporine treatment enhanced the formation of the capsid:Bax hetero-oligomers but evidently did not facilitate the assembly of functional Bax pores as the cells were not apoptotic. Interestingly, we found no evidence that capsid protein binds to Bak (Figure 5C) suggesting the interaction of this viral protein with Bcl-2 family proteins is highly specific. Together, these data suggest that capsid protein and Bax form mixed oligomers that do not function as pores.
Since capsid protein forms a complex with Bax, we next tested whether its expression could inhibit Bax-mediated apoptosis. Over-expression of either Bax or Bak induces cell death in the absence of other apoptotic stimuli [37]. A549 cells were co-transfected with plasmids encoding GFP-Bax and capsid, Bcl-XL (positive control) or vector alone (negative control) and at 24 hour post-transfection, samples were stained with the membrane-potential specific dye TMRM and then subjected to flow cytometric analyses (Figure 6A). As a second control, we transfected cells with a plasmid encoding a capsid deletion construct (CapNT) that is not targeted to mitochondria (see below). Loss of TMRM staining as a result of depolarization of mitochondrial membranes was used as the measure of apoptotic cell death. Quantitation of the data (Figure 6C) revealed that expression of capsid protein reduced the level of Bax-induced cell death by more than 60% compared to CapNT or vector alone. Similar results were observed for cells expressing Bcl-XL, a protein which has previously been shown to block the effects of Bax over-expression [38]. Data in Figure S4 show that capsid expression also protects primary human embryonic fibroblast (HEL-18) cells [17] from Bax-mediated apoptosis. The anti-apoptotic activity of capsid protein was specific to Bax as evidenced the fact that it did not attenuate Bak-mediated apoptosis (Figure 6B, C).
To further understand how capsid functions to block apoptosis, we determined whether expression of this viral protein inhibits Bax-induced release of cytochrome c. A549 cells were co-transfected with plasmids encoding GFP-Bax and capsid or empty vector. Localization of cytochrome c was monitored by fluorescence microscopy at 24 hours post-transfection. As expected, in cells expressing GFP-Bax and vector alone, there was marked loss of cytochrome c from mitochondria (Figure 7A, asterisks). In contrast, in cells that expressed both capsid protein and GFP-Bax, cytochrome c remained associated with this organelle (Figure 7A, arrows). However, consistent with data shown in Figures 5 and 6, capsid did not block GFP-Bak-induced loss of cytochrome c from mitochondria (Figure 7B arrows).
Based on the assumption that association of capsid protein with mitochondria is critical for its anti-apoptotic function, we next mapped the region of capsid protein that is required for targeting to this organelle. Analyses of the RV capsid protein sequence with web-based algorithms such as PSORT II Prediction (http://psort.nibb.ac.jp/form2.html) indicated that conventional mitochondrial targeting signals are absent. We therefore constructed a series of capsid deletion mutants whose localizations were determined by expression in A549 cells (Figure 8A). From the indirect immunofluorescence data shown in Figure 8B, it can be seen that the 23 amino acid residue E2 signal peptide which forms the hydrophobic carboxyl-terminus of capsid protein, is required for association with mitochondria. Moreover, the observation that a pool of CapCT overlaps with cytochrome c indicates that the carboxyl-terminal region of capsid protein contains information that is sufficient for targeting to mitochondria. Intriguingly, expression of the CapCT construct caused extreme compaction of the mitochondrial network to the perinuclear region, much more so than in cells expressing full-length capsid protein.
We next determined whether association of capsid with mitochondria correlated with its ability to block apoptosis. Transfected cells expressing the various capsid constructs were challenged with staurosporine or anti-Fas, and then apoptosis induction was assessed using the activated caspase 3 assay. The amino-terminal capsid construct (CapNT) neither associates with mitochondria nor protects against apoptosis (Figures 8, 9). Conversely, CapCT, a pool of which is targeted to mitochondria, protects as well as full-length capsid protein against staurosporine and anti-Fas challenge. CapΔRSP, which lacks the hydrophobic E2 signal peptide and a membrane proximal arginine-rich (R) motif, is not targeted to mitochondria and does not block staurosporine or anti-Fas-mediated induced activation of caspase 3. Interestingly, although CapΔSP does not localize to mitochondria, it did confer resistance to both Fas- and staurosporine-induced apoptosis (Figure 9). This observation suggests that the membrane-proximal R motif is important for the anti-apoptotic function of capsid. Table 1 summarizes the localization and anti-apoptotic properties of the capsid deletion mutants.
To investigate if the arginine residues in the membrane-proximal R motif were important for the anti-apoptitic function of capsid protein, we created a point mutant (CapCR5A) in which five arginines in this motif were changed to alanine residues (Figure 10A). This capsid mutant was targeted to mitochondria where it activated Bax and stimulated cytochrome c release in the absence of apoptotic stimuli (Figure 10B and C asterisks); indicating that the arginine residues within the R domain are critical for the anti-apoptotic activity of capsid protein. Moreover, it would appear that mutation of these arginine residues unmasks an intrinsic pro-apoptotic activity of capsid protein, which may explain why it alone can stimulate Bax conformational change and membrane insertion. Next, we compared the Bax-binding ability of the CR5A mutant relative to wild type capsid and capsid deletion constructs. The observation that more CapΔSP is recovered in anti-Bax coimmunoprecipitations than CapΔRSP (Figure 11A) suggests that the R domain is important for interaction with Bax. However, ablation of the arginine residues in the R domain did not affect binding to Bax indicating that the arginine residues per se in this motif are not essential for interaction with Bax (Figure 11A). Binding between Bax and CapCT or CapNT was not detected in our assays (Figure 11B). Indirect immunofluorescence analyses revealed that unlike wild type capsid and CapCR5A, neither CapNT, CapCT, CapΔSP nor CapΔRSP induced the 6A7-specific conformation change in Bax (data not shown). Together, these results suggest that capsid protein employs a multi-step mechanism to block apoptosis. Specifically, binding to Bax through the R domain and/or the carboxyl terminus stimulates a conformational change in Bax; but pore formation and/or functionality is blocked by the arginines in the R motif of capsid protein.
We introduced the CR5A mutations into the capsid gene of a RV infectious clone in order to determine if the membrane-proximal arginine-rich (R) domain in capsid protein is required for blocking apoptosis during infection. Our hypothesis was that early onset apoptosis would result in decreased replication and virus particle production. A549 cells were infected with wild type or CR5A strains of RV and virus replication and apoptosis induction were analyzed. Data in Figure 12A show that cells infected with CR5A virus were significantly more susceptible to Fas-dependent apoptosis. Moreover, in non-treated (control) samples, the level of virus-induced apoptosis was four fold higher in cells infected with the CR5A mutant. Similar results were obtained with infected Vero cells (data not shown). Next, we compared the levels of RV proteins in CR5A and wild type (WT) RV infected cells as a function of time. Figure 12B shows that in cells that were infected with WT RV, the level of virus nonstructural (p150) and structural proteins (capsid) peaked at 72 hours. In contrast to p150 levels which were only moderately lower, steady state levels of capsid protein were dramatically lower in CR5A infected cells at all time points. To control for the possibility that CR5A capsid was unstable in the infected cells, we also determined the relative levels of another structural protein, E1. Similar to capsid protein levels in CR5A infected cells, levels of E1 were much lower than in WT virus infected cells; suggesting a defect in synthesis of structural proteins in CR5A infected cells. Consistent with this theory, data in Figure 12C show that secretion of CR5A virions is severly impaired. This was not because the CR5A capsid is misfolded as data in Figure S5 show that this mutant capsid protein functions as well as wild type capsid in driving assembly and secretion of Rubella virus-like particles.
Nonstructural proteins are translated directly from the 40S genomic RNA whereas capsid and other structural proteins are made from a subgenomic RNA. Accordingly, it is possible that virus transcription and replication are impaired in the CR5A mutant. Quantitative RT-PCR with p90 specific primers was used to determine the relative levels of genomic RNA in the WT and CR5A infected samples (Figure 12D). From these data, it can be seen that replication of viral RNA was severely affected in CR5A infected cells. This was not due to decreased infection efficiency because at six hours post-infection, there was on average >50% more genomic RNA in CR5A infected cells (Table 2). Moreover, as demonstrated by plaque assays, cells infected with CR5A virus did release infectious virus (Figure 12E). Interestingly, the CR5A plaques were larger and had a spotty appearance compared to wild type virus-produced plaques which were smaller and clearer.
Although data in Figure S5 indicate that Cap5RA is not misfolded, without additional investigation, we could not completely rule out the possibility that the replication defects associated with the CR5A strain virus were due to other inherent defects of the mutant capsid protein. Therefore, we attempted to artificially block apoptosis by over-expression of Bcl-XL or adding the caspase inhibitor Z-VAD-FMK to CR5A infected cells. Over-expression of Bcl-XL did not rescue the CR5A replication but this result was non-informative as further investigation revealed that this anti-apoptotic protein was unable to protect mitochondria from the effects of CapC5RA in transfected cells (data not shown). In contrast, addition of Z-VAD-FMK did have a modest effect on production of viral proteins in CR5A infected cells (Figure 13A). The effect was most pronounced at 72 hrs post-infection where levels of p150 and capsid were considerably higher in Z-VAD-FMK treated cells. In contrast, blocking caspase activity in cells that were infected with wild type RV did not appreciably alter the steady state levels of viral proteins. Finally, it can be seen from the data in Figure 13B that Z-VAD-FMK treatment had a modest effect on production of CR5A virus. Compared to CR5A-infected Vero cells treated with DMSO alone, addition of Z-VAD-FMK resulted in a modest increase in viral titers as evidenced by increased clearing of RK-13 monolayers. Together, these data are consistent with our hypothesis that the anti-apoptotic function of capsid is important for virus replication.
Apoptosis is a common defense mechanism used by host cells to limit the spread of viral infections and consequently, a number of viruses have developed mechanisms to disrupt programmed cell death pathways. With few exceptions, all known viral apoptosis inhibitors are accessory proteins that are encoded by DNA viruses and therefore, a great deal of effort has focused on these proteins (reviewed in [39]). Ironically, even though RNA viruses cause the vast majority of viral diseases in humans, comparatively little is known about if or how these types of viruses interfere with apoptotic signaling. Among the exceptions are picornaviruses, a number of which encode “security” proteins (leader protein and 2BC) that can block apoptosis [40], [41]. The mechanisms by which these proteins block apoptosis are not known and interestingly, caspase activation can still occur normally. These proteins may not actually prevent cell death per se, but rather, shift the balance toward necrotic cell death as opposed to apoptosis.
Hepatitis C virus (HCV) is known to modulate apoptotic signaling but unlike picornaviruses, this virus is not cytolytic and readily establishes persistent infections in vivo. HCV encodes a number of proteins that reportedly exhibit anti-apoptotic activity. For example, the nonstructural proteins NS2 and NS5A interfere with programmed cell death by different mechanisms [42], [43]. The functions of HCV structural proteins in apoptotic signaling events are less clear; in particular, the core/capsid protein. The majority of data suggest that this protein acts to induce apoptosis although a number of published studies suggest otherwise (reviewed in [44]). Similarly, with one exception [45], expression of HCV E2 glycoprotein reportedly acts in a pro-apoptotic manner [46], [47], [48]. By and large, these studies involved plasmid-based expression of individual HCV proteins and indeed the data provide much to ponder with respect how this virus interfaces with apoptotic pathways. However, it is still not clear how individual HCV proteins or those of any other RNA virus affect cell death during infection.
Multiple laboratories have reported that RV infection induces programmed cell death in a variety of cultured cell lines [17], [19], [23], [49], [50] but it is worth noting that in virtually all cases, maximum synthesis of viral macromolecules and release of virions occur well before extensive apoptosis is observed. For example, in Vero cells, robust expression of structural proteins is first detected at 16 hours post-infection and secretion of infectious virions peaks 32 hours later [51]. Conversely, late apoptotic events such as DNA fragmentation and expression of pro-apoptotic proteins p53 and p21 does not peak until 5–7 days post-infection [17]. This indicates that that the majority of programmed cell death occurs long after the peak of virus production. Consistent with these observations, we show that RV infected cells are in fact, resistant to apoptosis for at least 48 hours post-infection.
Here, we provide evidence that in addition to functioning in virus assembly, the RV capsid protein is a potent inhibitor of apoptosis. With the possible exception of HCV capsid and E2, structural proteins of RNA viruses have been found to cause apoptosis rather than prevent cell death (reviewed in [39], [44], [52]). As far as we are aware, this is the first example of a structural protein from an RNA virus that functions to block cell death pathways through interactions with Bax. Mapping studies suggest that expression of the virus nonstructural proteins is the cause of RV-induced cell death [19], [20]. Accordingly, counteracting apoptotic pathways that become activated by expression of these early proteins may be essential for efficient replication; a theory that is supported by data from reverse genetic experiments with the CR5A mutant.
Our data appear to be in disagreement with previously published data showing that capsid protein expression induces apoptosis [22]. An important distinction between the previous study and the present work is that we assayed the ability of capsid protein to protect against various apoptotic stimuli rather than assaying whether or not capsid expression induces apoptosis in the absence of stimuli. In addition, we found that capsid protein blocks apoptosis in multiple cell lines (including a primary human cell line) whereas in the afore-mentioned study, capsid protein was reported to be pro-apoptotic in RK-13 but not other cell lines. Data in the present study are also consistent with the fact that stable cell lines that express high levels of RV structural proteins (including capsid protein) are readily established in a variety of cells types [19], [53], [54]. Importantly, the results of the reverse genetic experiments suggest that the anti-apoptotic function of the capsid protein is critical for RV replication. Although we cannot absolutely rule out the possibility that mutations in the R domain of capsid directly affect its functions in replication and assembly this seems unlikely. First, the CapCR5A mutant was able to drive particle assembly and CR5A virions efficiently delivered viral RNA to host cells. Second, the region of capsid that complements p150 function in virus replication is in the amino-terminal one third of the protein [55]. Accordingly, the most logical conclusion is that the anti-apoptotic role of capsid protein is necessary to promote survival of the host cell during the long replication cycle. To our knowledge, this has never been demonstrated before for an RNA virus but it is tempting to speculate that other slowly replicating RNA viruses employ similar mechanisms to avoid killing infected cells.
Although capsid protein may interfere with apoptosis by more than one mechanism, because the Bax-dependent pathway is a critical feature of mitochondrial apoptosis in most human cell types, interfering with the pore-forming ability of this protein is likely the key anti-apoptotic function of capsid protein. Binding of capsid protein to Bax induces a major conformational change, which interestingly, seems to promote activation and oligomerization of Bax. It is not clear if this phenomenon is related to the anti-apoptotic activity of capsid or if it is an inconsequential effect of complex formation with Bax. Figure 14 depicts a model in which capsid protein interferes with formation of functional Bax pores. In some critical aspects, the RV capsid protein may function analogously to the cytomegalovirus accessory protein vMIA, a putative Bcl-2 homolog that forms mixed oligomers with Bax [56]. However, confirmation of this theory is dependent upon determining the structure of the RV capsid protein.
Mapping studies localized the anti-apoptotic activity to the carboxyl-terminal region of capsid protein. The E2 signal peptide, which is required for membrane association of capsid protein [57], [58], is also essential for targeting to mitochondria but not for blocking apoptosis. Conversely, while the membrane proximal arginine-rich (R) motif in capsid is dispensable for targeting to mitochondria, it is required for protection from intrinsic and extrinsic apoptotic stimuli (Table 1). The R motif (RSARHPWRIR) of RV capsid bears remarkable similarity to the Bax-binding motif (RRHRFLWQRR) in vMIA [59] which is critical for blocking Bax- but not Bak-dependent apoptosis [60]. Despite the apparent similarities between vMIA and RV capsid protein, one difference worth noting is that the arginine-rich motif in capsid is not required for binding to Bax.
As mentioned above, it is possible that capsid protein blocks apoptosis through other mechanisms, at least one of which does not involve Bax. For example, CapCT, which neither binds to nor activates Bax, protects cells from staurosporine and Fas-dependent apoptosis. However, unlike full-length capsid, CapCT does not protect cells from Bax over-expression. Capsid binds two other pro-apoptotic proteins p32 and Par-4 [61] and through sequestration into non-functional complexes, it is possible that the functions of these proteins in apoptotic signaling are mitigated. Although we have no direct evidence to support this theory, binding to membrane-associated capsid protein may prevent Par-4 from engaging in pro-apoptotic complexes in the nucleus or cell surface [62], [63], [64]. Finally, we showed that translocation of the capsid-binding pro-apoptotic protein p32 into mitochondria is inhibited by RV infection [12]. Because targeting of p32 to mitochondria is critical for its function in programmed cell death pathways [65], [66], reducing the levels of mitochondrial p32 would be expected to reduce apoptosis.
To summarize, we describe a novel mechanism by which a viral capsid protein potently blocks apoptosis. Our data suggest that this function of capsid is important for virus replication and it is also tempting to speculate that establishing and/or maintaining persistent infections in vivo also requires this activity. RV is known to persistently infect lymphocytes and capsid-dependent inhibition of Fas-dependent apoptosis may allow the virus to disseminate through the body using apoptosis-resistant lymphocytes as conduits. It will be of interest to examine proteins from other slowly replicating RNA viruses to determine if capsids or other proteins double as inhibitors of apoptosis.
The following reagents were purchased from the respective suppliers: Protein A and G Sepharose from GE Healthcare Bio-Sciences Corp (Princeton, NJ); General lab chemicals from Sigma Chemical Co. (St. Louis, MO); Media and fetal bovine serum (FBS) for cell culture from Life Technologies-Invitrogen, Inc. (Carlsbad, CA); A549, HEK 293T, Vero, and RK-13 cells from the American Type Culture Collection (Manassas, VA.). Hel-18 cells [17] were obtained from Dr. Eva Gonczol, (National Center for Epidemiology, Budapest, Hungary).
A549 and HEK 293T cells were cultured in Dulbecco's minimal essential medium (high glucose) containing 10% FBS, 2 mM glutamine, 1 mM HEPES, and antibiotics. Vero cells were cultured in Dulbecco's minimal essential medium (high glucose) containing 5% FBS, 2 mM glutamine, 1 mM HEPES, and antibiotics. RK-13 cells were cultured in minimum essential medium containing 10% FBS, 2 mM glutamine, 1 mM HEPES, 0.1 mM non-essential amino acids, and antibiotics. Hel-18 cells were cultured in RPMI-1640 medium containing 10% FBS, 2 mM glutamine, 1 mM HEPES, 0.1 mM non-essential amino acids, and antibiotics. Cells were incubated at 37°C in a humidified atmosphere with 5% CO2. RV stocks were diluted with cell culture media and then added to cells that had been washed with PBS. Cells were incubated with the virus (1 ml/35 mm dish) for 4 hours at 35°C after which time the inoculum was replaced with normal growth media. Infected cultures were kept at 35°C until experimental analyses.
To investigate the effect of blocking apoptosis on virus replication, Vero cells were infected with M33 (wild type) or CR5A strains of RV (MOI: 1) in the presence of 50 µM pan-caspase inhibitor Z-VAD-FMK (Promega, Madison, WI) which was initially made as a 20 mM stock solution in dimethyl sulfoxide (DMSO). Z-VAD-FMK or control vehicle (DMSO) was added to infected cells every 24 hrs. Samples were processed at the indicated time points and virus titers were determined by plaque assay [67].
Plasmids for expression of pCMV5-Capsid, pCMV5-CapCT, pCMV5-CapΔSP and pCMV5-E2E1 have been described previously [58], [67], [68]. An expression plasmid encoding amino acid residues 1–152 of capsid (CapsidNT) was constructed by polymerase chain reaction (PCR) using a forward primer with a EcoRI site and a Kozak consensus ribosome-binding site (5′-CGGAATTCGCCACCATGGCTTCCACTACCCCCATCACC-3′′) and a reverse primer with a BamHI site and stop codon (5′-CGCGCGGATCCCTAGGCCTCAGTGGGTGC-3′). The restriction sites are underlined in the primer sequences. The CapΔRSP cDNA which encodes amino acid residues 1–267 of capsid, was constructed by PCR using the forward EcoRI and Kozak site-containing forward primer (5′-CGGAATTCGCCACCATGGCTTCCACTACCCCCATCACC-3′′) and a reverse primer with a BamHI site and stop codon (5′-CGCGCGGATCCCTACTCGGTGGTGTGAGGG-3′). The template cDNA for the CapNT and CapΔRSP PCR reactions was pCMV5-capsid. The CapC5RA expression plasmid was prepared by PCR using a forward primer containing an EcoRI site and a Kozak site (5′-TCACGGAATTC-3′) and a reverse primer containing a BamHI site (5′-TCAGGATCCCTAGGCGCGCGCGGTGC-3′). The template DNA was pBRM33-CR5A. The CapsidNT, CapsidΔRSP, and CapsidC5RA cDNAs were resulting products were sublconed into the EcoRI and BamHI sites of the mammalian expression vector pCMV5 [69] to produce pCMV5-CapNT, pCMV5-CapΔRSP and pCMV5-CapCR5A respectively.
For establishing capsid and luciferase expressing stable cell lines, the Lenti-X-tet-On advanced inducible expression system (Clontech Laboratories, La Jolla, CA) was utilized. A cDNA encoding full-length capsid was amplified by PCR using a forward primer with a BamHI site and Kozak consensus ribosome binding site (5′-TAGGATCCGCCACCATGGCTTCCACTACCCCCATCACC-3′) and a reverse primer with a EcoRI site (5′-GGCCGAATTCCTAGGCGCGCGCGGTGC-3′) respectively, where the restriction sites are underlined. The DNA used as a template was pCMV5-capsid. The PCR product was digested with BamHI and EcoRI, and subcloned into the pLVX-Tight-Puro vector. Lentivirus-production in HEK 293T and transduction of A549 cells were performed as per the manufacturer's instructions. At 48 hours post-transduction, cells were split 1∶2 into medium containing G418 (500 µg/ml) and puromycin (0.5 µg/ml). Surviving cells were tested for inducible expression of capsid by indirect immunofluorescence and immunoblot analyses. The resulting polyclonal stable cell lines A549-Luciferase and A549-Capsid were maintained in media containing G418 (250 µg/ml) and puromycin (0.25 µg/ml). To induce capsid or luciferase gene expression doxycycline (1 µg/ml) was added to the culture medium.
Codons for arginine-to-alanine mutations in the C-terminus of capsid were introduced into the RV M33 infectious clone (pBRM33) [70] by a two step cloning procedure. A 421 base pair synthetic fragment (Epoch BioLabs Inc, Sugarland, TX) containing five arginine-alanine substitutions (R264, 268, 271, 275, 277) was used to replace the analogous region in pCMV5-24S [68]. The resulting plasmid was named pCMV5-24S-CR5A. Next, the BsrGI/BamHI fragment from pCMV5-24S CR5A was used to replace the analogous region (BsrGI/BamHI) in pBRM33 resulting in the infectious clone pBRM33-CR5A.
Total RNA samples were isolated with TRI Reagent (Ambion) from Vero cells infected with M33 (wild type) or CR5A strains of RV (MOI: 1) according to the manufacturer's instructions. Prior to the RT-PCR reaction, 1 µg of total RNA was treated with 2 U of amplification grade DNase I (Invitrogen) as per manufacturer's recommendations. The DNase I-treated RNA samples were reverse transcribed to single-stranded cDNA using qScript Flex cDNA synthesis kit and Oligo (dT)20 primer (Quanta Biosciences, Gaithersburg, MD) as per manufacture's instructions.
Quantitative PCR reactions were conducted on a MX3005P thermocycler (Stratagene, La Jolla, CA) using a PerfecCTa SYBR green supermix low Rox real-time PCR kit (Quanta Biosciences). Reactions were carried out by triplicate in a total volume of 25 µl containing 5 µl of cDNA and 0.2 µM of each oligonucleotide primer. Primers used to amplify RV nucleotides 5520–5706 from the RV p90 non-structural protein coding region of the RV genome were as follow: RV-F (5′-AGGTCATGTCTCCGCATTTC-3′) and RV-R (5′-GTCCCGAGTAGCAAGGGTCT-3′). The amplification cycles for p90 samples consisted of an initial denaturating cycle at 95°C for 3 min, followed by 40 cycles of 15 s at 95°C, 30 s at 58°C, and 20 s at 72°C. Fluorescence was quantified during the 58°C annealing step, and the product formation was confirmed by melting curve analysis (57°C to 95°C). As an internal control, levels of the house keeping gene product cyclophilin A determined. Amplification was performed using the following primers, CYP-F (5′- TCCAAAGACAGCAGAAAACTTTCG-3′) and CYP-R (5′-TCTTCTTGCTGGTCTTGCCATTCC-3′). The amplification cycles for Cyclophilin A consisted of an initial denaturating cycle at 95°C for 3 min, followed by 40 cycles of 15 s at 95°C, 20 s at 60°C, and 40 s at 72°C. Fluorescence was quantified during the 60°C annealing step, and the product formation was confirmed by melting curve analysis (57°C to 95°C).
Quantification of the samples was performed using the two standard curves method [71], and the relative amount of RV genomic RNA was normalized to the relative amount of Cyclophilin A mRNA. Three independent PCR analyses were performed for each sample.
A549 cells (1×105) in 35 mm culture dishes were infected with the M33 strain of RV (MOI = 2) and then incubated for 48 hours at 35°C prior to lysis. Alternatively, A549 cells were transiently transfected with pCMV5-capsid, pCMV5-CapNT, pCMV5-CapCT, pCMV5-CapΔSP, pCMV5-CapΔRSP or pCMV5-CapCR5A using Lipofectamine 2000 (Invitrogen). Cells were lysed in 1% (wt/vol) CHAPS, 150 mM NaCl, 50 mM Tris, pH 8.0 containing Complete EDTA-free protease inhibitors (Roche) or 1% NP-40, 150 mM NaCl, 2 mM EDTA, 50 mM Tris, pH 7.4 containing protease inhibitors. Cell lysates were clarified by centrifugation at 10,000× g for 10 minutes at 4°C. Immunoprecipitation was performed with clarified lysates and 1 µg/ml of mouse monoclonal anti-Bax6A7 (Sigma), or 1∶1000 of rabbit polyclonal anti-capsid serum (7W7), or 2 µg/ml of rabbit anti-Bak (Millipore) antibodies overnight at 4°C with rotation. Fifteen microliters of protein A or protein G sepharose (50% suspension) were added and then samples were rotated for 1 hour at 4°C before washing; twice with lysis buffer and once with PBS. Proteins were eluted from the beads by boiling in protein gel sample buffer, separated by SDS-PAGE, and then transferred to polyvinylidene fluoride (PVDF) membranes (Immobilon-P Millipore, Bedford, MA). Membranes were incubated for 1 hour at room temperature with the following antibodies and dilutions: 1∶1000 rabbit anti-RV capsid (7W7) [61],1∶1000 mouse anti-capsid (H15C22), 1∶1000 goat anti-RV (Meridian Life Science Inc, Saco, Maine), 1∶1000 rabbit anti-Bak (Cell Signaling), 1∶1000 rabbit anti-Bax (Abcam) or 1∶5000 mouse anti-Bax (YTH-2D2, Trevigen Inc, Gaithersburg, MD). To detect E1 glycoprotein by immunoblotting, it was necessary to perform SDS-PAGE under non-reducing conditions. E1 was detected using a 1∶1000 dilution of a mouse monoclonal antibody (H2C213) obtained from Abbott Labs (Abbott Park, IL). After three washes with Tris-Buffered-Saline-Tween (TBS-T), the membranes were incubated with either goat anti-rabbit, goat anti-mouse or rabbit anti-goat horseradish peroxidase-conjugated IgG (Bio-Rad Hercules, CA) for 1 hour. Membranes were washed four times with TBS-T and immunoreactive proteins were detected using Supersignal West Pico chemiluminescent substrate (Pierce Biotechnology, Rockford, IL) followed by exposure to X-ray film (Fuji Photo Film Co, LTD, Tokyo, Japan).
A549-Capsid or A549-Luciferase cells were cultured in the presence of doxycycline and after 36 hours, anti-human Fas activating clone CH11 antibody (0.12 µg/ml) (Millipore, Temecula, CA) and cycloheximide (10 µg/ml) were added to the cultures for 6 hours. Cells were then lysed in 1% NP-40 buffer containing a cocktail of protease inhibitors. The lysates were centrifugated at 10,000× g for 10 min at 4°C, and protein concentrations were determined by BCA protein assay (Pierce Biotechnology, Rockford, IL) using bovine serum albumin as a standard. Equivalent amounts of protein (60 µg) from each lysate were resolved in 8% SDS-PAGE and transferred to PVDF membranes followed by immunoblotting with mouse monoclonal anti-cleaved PARP (Asp214) clone 19F4 antibody (Cell Signaling).
A549 and Vero cells cultured on glass coverslips were infected with RV (MOI = 1) or transiently transfected with pCMV5-capsid or pCMV5-CapNT or pCMV5-CapCT or pCMV5-CapΔSP or pCMV5-CapΔRSP or pCMV5-E2E1 or pcDNA3.1-Bcl-XL, and peGFP-Bax or peGFP-Bax (Gift of Dr. Michele Barry, University of Alberta) using Lipofectamine 2000 (Invitrogen). After 24 or 48 hours post-infection or post-transfection as indicated, cells were fixed in 4% paraformaldehyde for 20 min, followed by quenching with PBS containing 50 mM ammonium chloride. Cell membranes were permeabilized with PBS containing 0.2% Triton-X-100 for 5 min before incubation with primary and secondary antibodies. All the washes were done in PBS containing 0.1 mM CaCl2 and 1 mM MgCl2.
RV proteins were detected with rabbit anti-capsid (7W7), mouse anti-capsid (H15C22), mouse-anti E1 (H2C213), goat anti-RV, or human anti-RV (GB) which has been described previously. Mitochondria were detected using rabbit anti-cytochrome c (from Dr. L. Berthiaume, University of Alberta) or with a mouse anti-complex II monoclonal antibody (Mitosciences, Eugene, OR). Activated isoforms of Bax and capsase 3 were detected with a Bax-specific mouse monoclonal antibody 6A7 (Abcam) or caspase 3-specific rabbit monoclonal antibody (BD Pharmingen) respectively. Primary antibodies were detected with Alexa Fluor 594 chicken anti-mouse, Alexa Fluor 488 donkey anti-rabbit, Alexa Fluor 488 donkey anti-mouse, Alexa fluor 637 Donkey anti-rabbit and/or Alexa 594 goat anti-rabbit (Molecular Probes, Invitrogen, Carlsbad, CA). Coverslips were mounted onto microscope slides using ProLong Gold antifade reagent with 4'-6-Diamidino-2-phenylindole (Molecular probes, Invitrogen). Samples were then examined using Zeiss 510 confocal microscope or a Zeiss Axioskop2 microscope equipped with a CoolSNAP HQ digital camera (Photometrics).
A549-Cap or A549-Luciferase cells were cultured for 36 hours in the presence of doxycycline, followed by incubation with 1 µM staurosporine (Sigma-Aldrich) or anti-Fas antibody (0.12 µg/ml) and cycloheximide (10 µg/ml) for 6 hours. Cells were then homogenized in ice-cold mitochondria isolation buffer containing 200 mM mannitol, 70 mM sucrose, 10 mM Hepes, and 1 mM EGTA (pH: 7.5) using a dounce homogenizer with a loose fitting pestle. Unbroken cells and nuclei were pelleted by centrifugation at 500× g for 10 min at 4°C. The supernatants were then centrifuged at 10,000× g for 20 min at 4°C to obtain crude mitochondrial pellets that were cross-linked with 10 mM bis(maleimido)hexane (BMH; Thermo Scientific) for 30 min at room temperature. Samples then were separated on 4–12% polyacrylamide gels and then processed for immunobloting with rabbit polyclonal antibodies to Bax antibody (Abcam, Cambridge, MA) and capsid (7W7).
Expression of capsid or luciferase in A549-cap or A549-luc cells respectively was induced with doxycycline for 36 hours, followed by incubation with staurosporine (1 µM) or anti-Fas antibody (0.12 µg/ml) to induce apoptosis. Cells then were stained with 0.2 µM Tetramethylrhodamine methyl ester (TMRM) (Invitrogen, Molecular probes) for 30 min at 37°C before analyses by flow cytometry (FACScan, Becton Dickinson). For each sample, 10,000 events were acquired. Data were analyzed using CellQuest software. The percentage of killing was calculated as the number of TMRM-negative cells divided by the total number of cells, and standard deviations were determined from three independent experiments.
For Bax or Bak killing assays, A549 or Hel-18 cells were transfected with peGFP-Bax or peGFP-Bak together with pCMV5, pCMV5-CapNT, or pCMV5-capsid using Lipofectamine 2000 or Lipofectamine LTX respectively (Invitrogen). After 24 hours, cells were stained with 0.2 µM TMRM for 30 min at 37°C prior to analyses by two-color flow cytometry. TMRM fluorescence was detected through the FL-2 channel equipped with a 585-nm filter and eGFP fluorescence was measured using the FL-1 channel equipped with a 489-nm filter. Data were acquired on 10,000 eGFP-positive cells per sample, and analysis was performed using CellQuest software. The relative specific cell death was calculated as the number of eGFP-positive TMRM-negative cells divided by the total number of eGFP positive cells. Standard deviations were generated from three independent experiments.
Vero cells (1×105/35 mm dish) were transiently transfected with pCMV5-capsid or pCMV5-CapCR5A and pCMV5-E2E1 using Lipofectamine 2000. Assembly and secretion of RV-like particles was assayed after 48 hours of transfection as described elsewhere [58].
Data from FACS and indirect immunofluorescence-based apoptosis assays were subjected to statistical analyses (student's t test or one-way analysis of variance (One-way ANOVA)) using Predictive Analytics Software (version 17.0.3) (SPSS Inc, Chicago, Il).
|
10.1371/journal.pgen.1001071 | Tuberous Sclerosis Complex 1 Regulates dE2F1 Expression during Development and Cooperates with RBF1 to Control Proliferation and Survival | Previous studies in Drosophila melanogaster have demonstrated that many tumor suppressor pathways impinge on Rb/E2F to regulate proliferation and survival. Here, we report that Tuberous Sclerosis Complex 1 (TSC1), a well-established tumor suppressor that regulates cell size, is an important regulator of dE2F1 during development. In eye imaginal discs, the loss of tsc1 cooperates with rbf1 mutations to promote ectopic S-phase and cell death. This cooperative effect between tsc1 and rbf1 mutations can be explained, at least in part, by the observation that TSC1 post-transcriptionally regulates dE2F1 expression. Clonal analysis revealed that the protein level of dE2F1 is increased in tsc1 or tsc2 mutant cells and conversely decreased in rheb or dTor mutant cells. Interestingly, while s6k mutations have no effect on dE2F1 expression in the wild-type background, S6k is absolutely required for the increase of dE2F1 expression in tsc2 mutant cells. The canonical TSC/Rheb/Tor/S6k pathway is also an important determinant of dE2F1-dependent cell death, since rheb or s6k mutations suppress the developmentally regulated cell death observed in rbf1 mutant eye discs. Our results provide evidence to suggest that dE2F1 is an important cell cycle regulator that translates the growth-promoting signal downstream of the TSC/Rheb/Tor/S6k pathway.
| Tuberous Sclerosis Complex genes 1 (TSC1) is a downstream component of the Insulin Receptor signaling pathway that is often deregulated in many tumors. In this study, we discovered that the fruit fly homolog of TSC1 regulates E2F transcription factor by controlling protein expression. E2F family proteins are key regulators of cellular division, and other tumor promoting events are previously shown to regulate E2F activity. Our findings demonstrate the importance of altering the E2F activity during tumorigenesis and provide new insights into the crosstalk between tumor promoting events.
| Retinoblastoma (Rb) family proteins are important regulators of cell cycle progression and survival (reviewed in [1], [2]). Orthologs of Rb exist in all metazoans where their functions are evolutionarily conserved (reviewed in [3]). Their best-known molecular function is to physically interact with E2F family proteins and transcriptionally repress E2F-regulated target genes. Genome-wide expression studies revealed that genes involved in various biological processes, such as cell cycle progression, survival, and development, are regulated by E2F family proteins [4]–[6]. As a consequence, the loss of Rb family genes in mice results in developmental defects that are often associated with uncontrolled S-phase entry and ectopic cell death [7]–[9]. Importantly, reducing E2F activity largely suppresses the Rb mutant phenotypes, indicating that deregulated E2F activity is responsible for the defects [10], [11]. Overall, E2F family proteins are the key molecular targets of Rb family proteins and responsible for the developmental consequence of Rb inactivation.
The long-term consequence of Rb inactivation in mammals is tumorigenesis. In humans, the loss of Rb is believed to be a critical step for retinoblastoma development. Moreover, Rb is believed to be functionally inactivated in most, if not all, cancers (reviewed in [12]). In mice, Rb heterozygosity (Rb+/−) results in the formation of pituitary and thyroid tumors [7], [13]–[16]. The wild type copy of the Rb gene is lost in these tumors, illustrating the importance of Rb as a tumor suppressor gene. Moreover, conditional knockout of Rb and an additional member of the Rb family gene, p107 or p130, in mouse retina is sufficient to promote retinoblastoma development [17]–[20]. Similar to the developmental phenotype, deregulated E2F plays a major role during tumorigenesis in Rb mutant mice. In a pituitary tumor model, the loss of E2f-1 or E2f-3 reduces the frequency of tumor development [21], [22]. More recently, the importance of E2F family proteins in human cancer is further illustrated by the findings that E2F family proteins themselves are often deregulated in many types of cancers (reviewed in [23]). Clearly, E2F family proteins play a critical role during tumorigenesis and also contribute to the developmental defects observed in Rb mutant animals.
Although it is clear that studying the function of E2F is crucial to understand the biology of Rb mutant animals and cancers, it has been difficult to dissect the in vivo roles of E2F family genes in mammals. One of the difficulties is the fact that E2F family proteins can functionally compensate for each other, which is particularly true for the subset of E2F proteins called “activator E2Fs” (reviewed in [24]). This is best demonstrated by a recent study showing that a single “activator E2F”, E2F-3a, is sufficient to support embryonic and post-natal development in mice, and the expression of E2F-3b or E2F-1 under the control of E2F-3a promoter can perform the role of E2F-3a [25]. This study suggests that the unique developmental functions of “activator E2Fs” are largely determined by their expression patterns and not by the differences of their protein sequences. Interestingly, Drosophila melanogaster has only a single “activator E2F”, dE2F1. The function of dE2F1 is evolutionarily conserved and represents the three “activator E2Fs” in mammals. dE2F1 is required for cellular proliferation and controls DNA damage-induced cell death, activities that are shared by the three “activator E2Fs” in mammals (reviewed in [3]). Since dE2F1 is the sole member carrying out the function of three E2Fs in mammals, it is possible that the regulation of dE2F1 expression is more complex and tightly controlled in flies. However, the regulatory mechanism that controls dE2F1 expression in Drosophila is poorly understood.
Like Rb, RBF1 is the major regulator of dE2F1 in flies. Most of the rbf1 mutant phenotypes are believed to be due to deregulated dE2F1 and can be rescued by a hypomorphic mutant allele of de2f1 [26]. Because of its simplicity and conserved developmental function, the Drosophila Rb/E2F is considered as a simplified version of mammalian Rb/E2F. Although rbf1 mutations are not sufficient to promote tumor phenotype in Drosophila, recent genetic studies revealed that RBF1/dE2F1 plays a crucial role when proliferation and/or survival are compromised by various tumor-promoting mutations. For example, dE2F1 is required by hippo mutant cells to overcome the developmentally regulated cell cycle arrest in eye imaginal discs [27]. Moreover, dE2F1-dependent cell death limits the growth promoting effect of the archipelago mutations in the eye, and cooperates with low EGFR activity to promote cell death [28], [29]. Interestingly, although the Drosophila p53 (dp53) does not genetically interact with rbf1 during development, dE2F1 and p53 cooperate to promote DNA damage- induced cell death as they do in mammalian systems [30]. Overall, RBF1/dE2F1 can either promote and/or limit the proliferation of cells that carry tumor-promoting mutations in flies.
Tuberous Sclerosis Complex 1 (TSC1) is a tumor suppressor gene that is mutated in benign tumors (reviewed in [31]). The in vivo function of TSC1 was first identified in Drosophila melanogaster as a regulator of cell size and proliferation (reviewed in [32]). TSC1 is a negative regulator of the Ras Homolog Enriched in Brain (Rheb), which is an activator of Target of Rapamycin (Tor). The canonical TSC/Rheb/Tor pathway has been established as a central network governing cell size and growth regulation. Although initial reports clearly demonstrated that TSC1 inactivation perturbs the cell cycle profile, less is understood about the mechanism by which TSC1 controls the cell cycle as well as cell size. Here, we demonstrate that tsc1 mutations cooperate with rbf1 mutations to promote both unscheduled S-phase entry and cell death during Drosophila eye development. This cooperative effect between tsc1 and rbf1 mutations can be explained, at least in part, by the observation that dE2F1 expression is post-transcriptionally increased in tsc1 mutant cells. A dE2F-reporter construct, PCNA-GFP, is activated in tsc1 mutant cells, and de2f1 mutations completely suppress the ectopic cell death observed in the rbf1 and tsc1 double mutant cells, indicating that dE2F1 is activated by tsc1 mutations and required for cooperative effect between rbf1 and tsc1 mutations. We further demonstrate that Rheb and Tor control dE2F1 expression, and s6k mutations completely abolish the increase of dE2F1 expression observed in tsc2 mutant cells. These results demonstrate that the TSC/Rheb/Tor/S6k pathway is an important regulator of dE2F1 expression during development and cooperates with RBF1 to regulate cell cycle progression and survival.
Ectopic S-phase entry and cell death are well-established Rb loss-of-function phenotypes. To address the question whether growth-promoting mutations could alter the Rb mutant phenotypes, we sought to determine the effects of inactivating the Drosophila ortholog of Tuberous Sclerosis Complex 1 (TSC1) in an rbf1 mutant background. To test this, tsc1 mutant clones were generated in wild type or rbf1 mutant eye discs (Figure 1). Since homozygous rbf1 null flies die at the first instar larval stage, we used an rbf1 hypomorphic allele, rbf1120a. Mitotic tsc1 mutant clones were generated by expressing Flippase (FLP) with an eye-specific driver and marked by the absence of GFP. Thus, GFP negative clones in wild type background have only tsc1 mutations while GFP negative clones in the rbf1120a background have both rbf1 and tsc1 mutations. Third instar larval eye discs were dissected and immunostained with anti-BrdU antibodies. During normal eye development in Drosophila, S- phase cells, which can be labeled with BrdU, are found at the anterior portion of the eye imaginal disc where cells are asynchronously dividing, and immediately posterior to the Morphogenetic Furrow (MF) where some cells undergo an extra S-phase called the Second Mitotic Wave (Figure 1A). At the MF, asynchronously dividing precursor cells arrest in G1 and begin differentiation process. Therefore, normally, there is no BrdU incorporating cells at the MF. Surprisingly, in clones that are double mutant for rbf1 and tsc1, ectopic S-phase cells were readily observed at the MF (Figure 1C). Since we can occasionally detect rbf1 mutant cells entering S-phase at the MF, we compared the number of ectopic BrdU positive cells at the MF between rbf1 single and rbf1 tsc1 double mutant clones. We normalized the number of ectopic BrdU positive cells by the clone size, which is measured by the number of the pixels in images taken at the same magnification. Clones that do not contain ectopic BrdU positive cells are excluded from the analysis. We determined that, on average, 3.7±2.2 ectopic S-phase cells/1000 pixels are present in the rbf1 clones while 12.4±5.6 ectopic S-phase cells/1000 pixels cells are present in the rbf1 tsc1 double mutant clones, showing more than 3 fold increase. This result indicates that RBF1 and TSC1 cooperatively regulate G1 to S- phase transition. Next, we stained for dying cells with anti-cleaved Caspase 3 antibodies (C3). rbf1 mutant cells undergo apoptosis at the anterior region of the MF, and this is not observed in the wild type eye disc (Figure 1B). We had previously reported that this developmentally regulated cell death in rbf1 mutant eye discs is dE2F1-dependent [29]. tsc1 mutant cells also undergo apoptosis just anterior to the MF though the level of cell death is much lower than what is observed in rbf1120a eye discs. However, in clones that are double mutant for rbf1 and tsc1, we observed a great increase in C3 staining at the MF and the anterior region of the eye disc (Figure 1B and 1C). Therefore, we concluded that RBF1 and TSC1 synergistically promote survival as well as G1 arrest during Drosophila eye development.
RBF1 is best characterized as a regulator of dE2F1 transcription factors whose activity promotes both S-phase entry and apoptosis. Since we observed that tsc1 mutations are able to enhance both ectopic S-phase entry and cell death phenotypes in rbf1 mutant cells, we sought to determine if dE2F1 itself is deregulated by tsc1 mutations. Eye discs containing tsc1 mutant clones were generated as described previously and immunostained with an anti-dE2F1 antibody. We observed that the intensity of dE2F1 staining is clearly stronger in tsc1 homozygous mutant clones throughout the eye disc, both in dividing and differentiating cells (Figure 2A and Figure S1A). Moreover we detected similar increase in antenna and wing discs, indicating that the effect on dE2F1 protein expression is not tissue-specific (Figure S1B and S1C). Importantly, the intensity of dE2F2 staining, the only other member of the E2F family in Drosophila, is unchanged in tsc1 mutant cells (Figure 2A), indicating that the effect of tsc1 mutations on dE2F1 expression is specific. To confirm the immunostaining result, we performed immunoblot assays using protein extracts from eye imaginal discs comprised mostly of tsc1 mutant cells (see Materials and Methods). Consistent with the immunostaining experiments, dE2F1 protein level is higher in tsc1 mutant eye discs than in control discs while no difference is detected in dE2F2 protein level (Figure 2B). To determine whether TSC1 regulates the level of de2f1 RNA, we performed real-time quantitative PCR (RTq-PCR). RNA was isolated from eye discs of the same genotypes used for immunoblot. We designed de2f1 specific primers that span an intron and amplified portions of two exons (second and third exons or fourth and fifth exons) to distinguish the PCR products from cDNA and genomic DNA. charybdis (chrb), a previously reported TSC1 regulated gene is used as a positive control [33]. Similar to the published result, we observed that the level of chrb RNA is increased by 11-fold in tsc1 mutant eye discs (Figure 2C). However, we could not detect any significant changes in de2f1 RNA level in tsc1 mutant eye discs (Figure 2C). Therefore, we concluded that TSC1 regulates dE2F1 expression post- transcriptionally.
Next, we examined whether the transcription of a dE2F1 target gene is activated in tsc1 mutant cells. To address this question, we used a reporter construct, PCNA-GFP, whose GFP expression is under the control of the PCNA promoter, a well-established dE2F1 target gene. As shown in Figure 3, GFP expression is increased in tsc1 mutant cells in the posterior portion of the eye disc, suggesting that, at least in this region, the increase of dE2F1 protein is sufficient to activate the transcription of a target gene. Importantly, the abnormal BrdU positive cells observed in the same region of tsc1 mutant clones are scarcely present (Figure 1A), indicating that the increase in dE2F1-reporter activity is not an indirect consequence of ectopic S-phase cells. We also sought to determine if tsc1 mutations could further activate dE2F1 target gene expression in rbf1 mutant cells. Our attempt to compare dE2F1 target gene expression between rbf1 single and rbf1 tsc1 double mutant eye discs by RTq-PCR did not provide any conclusive results (data not shown). This was somewhat expected since a substantial number of rbf1 tsc1 double mutant cells, presumably cells with hyperactive dE2F1, undergo cell death (Figure 1B and Figure S2A). Therefore, we decided to perform an in situ hybridization experiment, hoping to detect specific changes in a subset of surviving rbf1 tsc1 double mutant cells. Expression patterns of dE2F1 target genes (rnrS, Cyclin E, and PCNA) were determined using antisense RNA probes. In wild type eye discs, the expression pattern of these target genes resembles that of BrdU staining since their transcription is activated during the G1/S phase transition (Figure 3B left panel). In rbf1 mutant eye discs, dE2F1 target genes are strongly expressed at the MF where dE2F1 protein expression is normally high (Figure 3B middle panel). It is probable that, in rbf1 mutant eye discs, dE2F1 target gene expression is mainly controlled by dE2F1 protein level since cell cycle-dependent regulation by RBF1 is absent. Interestingly, in rbf1 tsc1 double mutant eye discs, dE2F1 target genes are strongly expressed both at the MF and in the anterior region of the eye disc (Figure 3B right panel). We reasoned that, since rbf1 mutant cells at the MF already express a high level of dE2F1 protein (previously shown in [29]), there is only a small margin for dE2F1 target gene expression to be further activated by tsc1 mutations. However, in the anterior region of the eye disc where the dE2F1 protein expression is normally kept low [29], tsc1 mutations can have a greater effect on dE2F1 activity and target gene expression. As a consequence, dE2F1 target genes are strongly expressed both at the MF and in the anterior region of rbf1 tsc1 double mutant eye discs, reaching the threshold of expression before undergoing cell death. Supporting this idea, ectopic cell death in rbf1 tsc1 double mutant eye discs is mainly observed at the MF and in the anterior region of the eye disc (Figure S2A). Interestingly, we could not detect much increase in dE2F1 target gene expression in the posterior region of rbf1 tsc1 double mutant eye discs, somewhat contradicting the result obtained by the PCNA-GFP reporter construct (Figure 3A). One explanation is that the in situ hybridization experiment is not as sensitive and quantitative as the PCNA-GFP reporter construct. We also found that the residual RBF1 proteins in the hypomorpic rbf1120a allele are mostly expressed in the posterior region of the MF, explaining why cells in this region do not show much an increase in dE2F1 target gene expression (Figure S2B). Nevertheless, these results indicate that tsc1 mutations can activate dE2F1 target gene expression in the wild type and rbf1 mutant backgrounds.
To determine if the cooperative effect on cell death by rbf1 and tsc1 mutations is dE2F1- dependent, we generated an allele with an FRT chromosome carrying both tsc1 and de2f1 mutations. For this allele, we used the tsc1f01910 allele that contains a piggyBac transposable element inserted in the intron 6 of the tsc1 locus. Generating tsc1f01910 clones in rbf1120a eye discs produces a similar increase in the level of ectopic cell death observed in Figure 1 (Figure 4A). When tsc1f01910 and de2f1729 double mutant clones are generated in rbf1120a eye discs, we noticed that the sizes of tsc1 de2f1 double mutant clones are much smaller than that of tsc1 single mutant clones (compare Figure 4A and 4B). The sizes of tsc1 de2f1 double mutant clones in the wild type background are also small (data not shown), indicating that the loss of de2f1 severely compromises proliferation of tsc1 mutant cells. Occasionally, we were able to obtain rbf1120a mutant eye discs with substantial sizes of the tsc1 de2f1 double mutant clones. We performed C3 staining to measure the level of cell death in rbf1, tsc1, and de2f1 triple mutant cells in these eye discs. Interestingly, the prevailing cell death phenotype observed in rbf1 tsc1 double mutant cells at the MF is no longer present in rbf1 de2f tsc1 triple mutant cells (Figure 4B). This result demonstrates that the increased level of ectopic cell death observed in rbf1 tsc1 double mutant cells is dE2F1-dependent.
Next, we asked if the known downstream regulators of TSC1 could regulate dE2F1 expression. We first determined the effect of rheb loss-of-function mutations on dE2F1 expression by generating mitotic mutant clones of rheb in the eye disc. Rheb is a Ras superfamily GTPase whose activity is negatively regulated by TSC1. As shown in Figure 5A, dE2F1 protein level is reduced, though not absent, in rheb mutant cells. This is best observed at the MF where dE2F1 expression is normally high [34]. We then asked if Rheb is required for the increased dE2F1 expression in tsc1 mutant cells. dE2F1 protein level is also reduced in tsc1 rheb double mutant cells (Figure 5A), indicating that Rheb is an important downstream regulator of TSC1 controlling dE2F1 expression. We concluded that, although not essential, Rheb regulates dE2F1 expression during eye development, and is clearly required for dE2F1 upregulation in tsc1 mutant cells. Since Rheb controls dE2F1 expression, we next tested if Rheb is also required for dE2F1-dependent cell death. To test this, we generated rheb mutant clones in the rbf1120a mutant eye disc where deregulated dE2F1 produces a stripe of apoptotic cells at the anterior region of the MF (Figure 1A and [29], [35]). As shown in Figure 5B, this stripe of cell death is interrupted by rheb mutant clones. Moreover, the ectopic cell death observed in rbf1 tsc1 double mutant cells is completely suppressed by rheb mutations. These results indicate that Rheb is an important regulator of dE2F1-dependent cell death as well as dE2F1 expression.
Rheb activates the Tor serine/threonine kinase, which through phosphorylation, can either inhibit 4EBP or activate S6k. We examined whether these proteins downstream of Rheb also participate in dE2F1 regulation. To address this question, Tor, s6k, and 4ebp mutant clones were generated in the eye disc. Similar to what is observed in rheb mutant clones, dE2F1 expression is reduced, but not absent, in Tor mutant clones, indicating that Tor participates in regulating dE2F1 expression during eye development (Figure 6A). Importantly, dE2F2 expression is unchanged in Tor mutant clones (data not shown). Based on this observation, we had hypothesized that dE2F1 expression levels would decrease in s6k mutant clones and/or increase in 4ebp mutant clones. Surprisingly, dE2F1 expression is unchanged in either 4ebp or s6k mutant clones (Figure 6B). These results suggest that Tor is required for proper dE2F1 expression during eye development while 4EBP and S6k are dispensable.
The fact that the loss of neither 4ebp nor s6k has an effect on dE2F1 expression might indicate a functional redundancy between the two genes. Alternatively, an unknown factor downstream of Tor might regulate dE2F1 expression during development. Nevertheless, we assessed whether S6k is required for the increase of dE2F1 expression observed when TSC1 is inactivated. We aimed to generate mitotic clones that are double mutants for tsc1 and s6k. However, because tsc1 and s6k are on the opposite arms of the third chromosome, we used a mutant allele of tsc2 (or gig in Drosophila), which is on the same chromosomal arm as s6k. TSC1 and TSC2 function together as a heterodimer, and mutations of tsc1 or tsc2 yield very similar phenotypes [36]–[38]. As expected, dE2F1 expression is elevated in gig mutant clones (Figure 7A). Furthermore, similar to what was observed in tsc1 mutant clones in the rbf1120a mutant background, the level of ectopic cell death was increased in gig mutant clones generated in rbf1120a mutant eye discs (Figure 7B). Surprisingly, the effects of gig mutations on dE2F1 expression and ectopic cell death are completely suppressed by s6k loss-of-function mutations. We observed that the level of dE2F1 expression in s6k gig double mutant clones is unchanged compared to the control (Figure 7A), and the ectopic cell death observed in rbf1 gig double mutant cells is completely absent in rbf1 gig s6k triple mutant cells (Figure 7B). Moreover, we observed that the basal level of dE2F1-dependent cell death normally present in the rbf1120a mutant eye disc (the stripe of cell death, Figure 1B) is also suppressed (Figure 7B). These results indicate that s6k is required for both the elevation of dE2F1 expression upon TSC inactivation and the increased level of cell death in rbf1 gig double mutant cells. In summary, our genetic studies led us to conclude that TSC1 and TSC2 regulate dE2F1 expression and dE2F1-dependent cell death via the canonical Rheb/Tor/S6k pathway during Drosophila eye development.
The loss of Rb leads to hyperactivation of E2F family proteins, which is a crucial event during tumorigenesis. Here, we demonstrate that the Drosophila ortholog of TSC1 tumor suppressor cooperates with RBF1 to regulate dE2F1 activity during development. TSC1 post- transcriptionally regulates dE2F1 expression, and the loss of tsc1 cooperates with rbf1 mutations to promote unscheduled S-phase entry and cell death. This effect of tsc1 mutations on dE2F1 expression requires the components of canonical TSC/Rheb/Tor pathway that are major regulators of cellular growth. Our study provides evidence to suggest that dE2F1 is an important protein that couples growth signals to cell cycle progression.
Recent studies have identified that pro-proliferative and pro-apoptotic activities of dE2F1 are engaged by various Drosophila tumor suppressor genes, such as hippo and archipelago [27], [28]. Our findings add tsc1/2 tumor suppressor genes to this list. Previously, dE2F1 or Cyclin E overexpression is shown to bypass starvation induced G1 arrest at least in endoreduplicating tissues [39]. Moreover, similar to dE2F1, expression of Cyclin E is elevated in tsc1 mutant cells in eye imaginal discs. [36]–[38]. Perhaps, restricting the expression of cell cycle regulators, such as dE2F1 and Cycline E, is a part of the molecular mechanisms by which nutrient deprivation induces G1 arrest. Interestingly, overexpression of dE2F1 or Cycline E does not overcome starvation-induced G1 arrest in larval neuroblasts, indicating that, in mitotic cells, neither dE2F1 nor Cycline E is the limiting factor [39]. Consistent with this observation, we could not observe any appreciable increase in the size of rheb or Tor mutant clones in rbf1 mutant background, suggesting that multiple factors contribute to the proliferative defect observed in rheb or Tor mutant cells in imaginal discs.
Interestingly, despite the elevated level of dE2F1 and Cyclin E, tsc1 mutant clones have relatively normal patterns of BrdU staining at the MF and a limited amount of ectopic cell death. We believe that the activity of dE2F1 in tsc1 mutant cells is normally restricted by the presence of RBF1. The fact that the increase in ectopic S-phase entry and apoptosis by tsc1 mutations can be only observed in the rbf1 mutant background supports this idea. We propose that the TSC/Rheb/Tor pathway during development modulates the amount of dE2F1 needed for cellular division in proportion to the cell size. Supporting this idea, previous studies have demonstrated that tsc1 or tsc2 mutant cells spend less time in G1, a phenotype commonly observed in cells with elevated dE2F1 activity [36]–[38], [40]. It is conceivable that the elevated level of dE2F1 proteins in tsc1 or tsc2 mutant cells allows them to go through G1 to S-phase transition faster where RBF1 is normally inactivated by Cyclin Dependent Kinases (CDKs).
Despite being the only “activator E2F” in Drosophila, it is still unclear how dE2F1 expression is regulated during development. A recent study reported that Cul4(Cdt2) E3 ubiquitin ligase mediates destruction of dE2F1 in S-phase, a mechanism that regulates dE2F1 expression in a cell cycle dependent manner [41]. Our findings here suggest that the expression of dE2F1 is also regulated by a growth-controlling network. However, at this point, we do not know the exact molecular mechanism by which dE2F1 protein level is post- transcriptionally controlled by the TSC/Rheb/Tor pathway. The finding that S6k is involved in this process supports the idea of translational control since S6k directly phosphorylates and regulates proteins involved in translation, such as RpS6, eIF4B, and eEF2K to list a few (reviewed in [42]). However, it is also equally possible that the TSC/Rheb/Tor pathway controls dE2F1 protein stability. In S2 cells, neither tsc1 RNAi nor Rapamycin (Tor inhibitor) treatment in S2 cells had the same effect on dE2F1 expression observed in imaginal discs (Figure S3). It is probable that S2 cells lack factors necessary for dE2F1 regulation that are present in vivo. Nevertheless, it is important to note that this effect on dE2F1 expression is specific since dE2F2 expression is unchanged in tsc1, rheb or Tor mutant cells (Figure 2A and data not shown). Curiously, the requirement of S6k to regulate dE2F1 is limited to the context in which TSC is inactivated. The loss of s6k in the wild type background has no effect on dE2F1 expression while rheb or Tor mutations reduce the level of dE2F1 proteins in the eye disc (Figure 5A and Figure 6). In mammals, it has been demonstrated that the translation of specific mRNA can be mTor-dependent but not S6k- dependent [43]. The molecular mechanism in which S6k promotes dE2F1 expression only when TSC is inactivated is presently unclear and warrants further investigation.
Another interesting finding from our study is that s6k mutations suppress the dE2F1-dependent cell death normally present in rbf1 mutant eye discs (Figure 7). s6k mutations alone did not alter the dE2F1 expression level at least in the wild type background. Although it is not formally tested, this raises a possibility that the TSC/Tor/S6k pathway controls dE2F1-dependent cell death without altering dE2F1 expression. Interestingly, the crosstalk between the InR/Tor and the EGFR signaling pathways during Drosophila eye development has been recently established [44]. InR/Tor signaling regulates the timing of neuronal differentiation in the eye disc by modulating EGFR activity. Since the EGFR pathway is an important determinant of dE2F1-dependent cell death [29], S6k might promote dE2F1-dependent cell death by modulating the EGFR pathway. We speculate that the cooperative effect between tsc1 and rbf1 mutations is the consequence of multiple changes that include the increase in dE2F1 expression.
In cancer cells, it is generally thought that the loss of Rb function is the most common mechanism of deregulating E2F activity. However, in some types of cancers, amplification of E2F genes or overexpression of E2F family proteins have been observed (reviewed in [23]). Moreover, in a subtype of human retinoblastoma where Rb is already deficient, E2f-3 proteins are also overexpressed [45]. These observations suggest that E2F family genes themselves can be directly targeted and deregulated during tumorigenesis. It will be interesting to investigate if TSC1/2 or other tumor suppressors and oncogenes regulate the expression of E2F family proteins to promote tumorigenesis.
All crosses have been performed at 25°C. The rbf1 mutant allele, rbf1120a, and de2f1 allele, de2f1729, are described previously [15], [16]. The tsc1 alleles used in this study are tsc1R453X, a gift from Dr. Hariharan [38], and tsc1f01910 (Exelixis collection, Harvard Medical School). The mutant alleles of the TSC/Rheb/Tor pathway used in this study are as follows: Tor2L19 FRT40A and 4ebpnull are gifts from P. Lasko [46], [47]. s6k l-1 FRT80B is a gift from D.J. Pan [48]. The gig56 FRT80B, FRT82B rheb2D1, and s6kl-1 gig192FRT80B alleles were kindly provided by J.M. Bateman [44]. The 4ebpnull FRT40A, FRT82B de2f1729 tsc1f01910, and FRT82B rheb2D1 tsc1R453X alleles were generated by meiotic recombination. For the double mutant alleles, presence of both mutations is verified by genetic complementation tests using multiple mutant alleles. For example, presence of both s6k and gig mutations in s6k l-1gig192 FRT80B alleles were verified by crossing the alleles to gig52, gig192, s6k l-1 and s6kp{PZ}07084 alleles individually.
Flippase (FLP) was expressed from the eyeless promoter to generate mitotic clones in the eye. To examine clones in rbf1 mutant animals, the X chromosome carrying rbf1120a and an ey-FLP transgene was used. Followings are the full genotypes of larvae analysed.
y w eyFlp/+ or Y; FRT82B GFPubi/FRT82B tsc1R453X
y w eyFlp/+ or Y; FRT82B GFPubi/FRT82B rheb2D1
y w eyFlp/+ or Y; FRT82B GFPubi/FRT82B rheb2D1 tsc1R453X
y w eyFlp/+ or Y; GFPubi FRT40A/Tor2L19 FRT40A
y w eyFlp/+ or Y; GFPubi FRT80B/s6k l-1 FRT80B
y w eyFlp/+ or Y; GFPubi FRT40A/4ebpnull FRT40A
y w eyFlp/+ or Y; GFPubi FRT80B/gig56 FRT80B
y w eyFlp/+ or Y; GFPubi FRT80B/s6kl-1 gig192 FRT80B
w rbf1120a eyFlp/Y; FRT82B GFPubi/FRT82B tsc1R453X
w rbf1120a eyFlp/Y; FRT82B GFPubi/FRT82B tsc1f01910
w rbf1120a eyFlp/Y; FRT82B GFPubi/FRT82B de2f1729 tsc1f01910
w rbf1120a eyFlp/Y; FRT82B GFPubi/FRT82B rheb2D1
w rbf1120a eyFlp/Y; FRT82B GFPubi/FRT82B rheb2D1 tsc1R453X
w rbf1120a eyFlp/Y; GFPubi FRT80B/gig56 FRT80B
w rbf1120a eyFlp/Y; GFPubi FRT80B/s6kl-1 gig192 FRT80B
y w eyFlp/Y; FRT82B [W+] l(3)cl-R3/FRT82B (controls)
y w eyFlp/Y; FRT82B [W+] l(3)cl-R3/FRT82B tsc1R453X
w rbf1120a eyFlp/Y; [W+] l(3)cl-R3/FRT82B
w rbf1120a eyFlp/Y; [W+] l(3)cl-R3/FRT82B tsc1R453X
y w eyFlp/PCNA-GFP; FRT82B LacZarm/FRT82B tsc1R453X
The antibodies used in this study are: anti-dE2F1 (1/1000) [29], anti-dE2F2 (1/1000) [34], anti-RBF1 (1/100) from Dyson Lab, anti-C3 (1/200, Cell Signaling), anti-GFP-FITC (1/200, abcam), anti-β-galactosidase (Developmental Studies Hybridoma Banks [DSHB]), and anti- ELAV (DSHB). For immunostaining, third-instar eye discs were fixed in 4% formaldehyde for 20 minutes at room temperature (eye discs immunostained for anti-dE2F1 were fixed at 4°C for 30 minutes) and washed twice with 0.3% PBST (0.3% Triton X-100 in PBS) and once with 0.1% PBST (0.1% Triton X-100 in PBS). Fixed eye discs were incubated in primary antibody with 0.1% PBST and 5% normal goat serum (NGS) at room temperature for 3 hours. After four washes with 0.1% PBST, eye discs were incubated in secondary antibody with 0.3% PBST and 5% NGS at room temperature for 2 hours. Immunostained eye discs were then washed five times with 0.1% PBST at room temperature and mounted for confocal microscopy imaging (Zeiss LSM).
For in situ hybridization experiments, eye-antennal discs were prepared as described previously [26]. Anti-sense RNA probes were generated using cDNA clones LD41588, LD17578, and LD45889 for rnrS, CycE, and PCNA respectively. After hybridization, Alkaline Phosphatase conjugated anti-DIG antibodies were used to detect DIG labeled anti- sense RNA probes. For each target genes, more than 20 eye antennal discs were analyzed and the representative images were chosen to be presented.
40 eye discs of tsc1 mutant and control animals were dissected and used for Western blot as previously described [29].
The average of three independent experiments of triplicate-PCR reaction is presented. Total RNA was isolated from 40 eye-antenna eye discs with RNeasy Mini kit (QIAGEN) according to manufacturer's protocol, and reverse transcribed using DyNAmo cDNA Synthesis Kit (Finnzymes) according to manufacturer's instructions. Quantitative PCR reactions were performed with DyNAmo Flash SYBR Green qPCR Kit (Finnzymes). Quantification was determined by comparative threshold cycle method (CT) on Bio-Rad CFX Manager software. Both rp49 and β-tubulin were used as normalization controls in a single experiment. All primers were designed with Primer3 (Whitehead Institute fozr Biomedical Research primer3 shareware [http://frodo.wi.mit.edu/primer3/]). Primer pairs used are:
chrb1F (AACTGCAGGCTCAGCTACG)
chrb1R (CGCTCTCGAACTCAATGAAG)
de2f12-3F (CAGCACCACCACCAAAATC)
de2f12-3R (ACTGCTAGCCGTATGCTTCTG)
de2f15-6F (TACAGCCATGACCGCAAC)
de2f15-6R (GTTCAGCGCATACGGATAGTC)
tubulin-F (ACATCCCGCCCCGTGGTC)
tubulin-R (AGAAAGCCTTGCGCCTGAACATAG)
Rp49-F (TACAGGCCCAAGATCGTGAAG)
Rp49-R (GACGCACTCTGTTGTCGATACC)
|
10.1371/journal.pmed.1002044 | Exclusive Breastfeeding and Cognition, Executive Function, and Behavioural Disorders in Primary School-Aged Children in Rural South Africa: A Cohort Analysis | Exclusive breastfeeding (EBF) is associated with early child health; its longer-term benefits for child development remain inconclusive. We examine the associations between EBF, HIV exposure, and other maternal/child factors and the cognitive and emotional-behavioural development of children aged 7–11 y.
The Vertical Transmission Study (VTS) supported EBF in HIV-positive and HIV-negative women; between 2012 and 2014, HIV-negative VTS children (332 HIV exposed, 574 HIV unexposed) were assessed in terms of cognition (Kaufman Assessment Battery for Children Second Edition [KABC-II]), executive function (Developmental Neuropsychological Assessment Second Edition [NEPSY-II]), and emotional-behavioural functioning (parent-reported Child Behaviour Checklist, [CBCL]). We developed population means by combining the VTS sample with 629 same-aged HIV-negative children from the local demographic platform. For each outcome, we split the VTS sample into scores above or at/below each population mean and modelled each outcome using logistic regression analyses, overall and stratified by child sex. There was no demonstrated effect of EBF on overall cognitive functioning. EBF was associated with fewer conduct disorders overall (adjusted odds ratio [aOR] 0.44 [95% CI 0.3–0.7], p ≤ 0.01), and there was weak evidence of better cognition in boys who had been exclusively breastfed for 2–5 mo versus ≤1 mo (Learning subscale aOR 2.07 [95% CI 1.0–4.3], p = 0.05). Other factors associated with better child cognition were higher maternal cognitive ability (aOR 1.43 [95% CI 1.1–1.9], p = 0.02, Sequential; aOR 1.74 [95% CI 1.3–2.4], p < 0.001, Planning subscales) and crèche attendance (aOR 1.96 [95% CI 1.1–3.5], p = 0.02, Sequential subscale). Factors positively associated with executive function were home stimulation (aOR 1.36 [95% CI 1.0–1.8], p = 0.04, Auditory Attention; aOR 1.35 [95% CI 1.0–1.8], p = 0.05, Response Set) and crèche (aOR 1.74 [95% CI 1.0–3.0], p = 0.05, Animal Sorting). Maternal mental health problems and parenting stress were associated with increased emotional-behavioural problems on the total CBCL (aOR 2.44 [95% CI 1.3–4.6], p = 0.01; aOR 7.04 [95% CI 4.2–11.9], p < 0.001, respectively). Maternal HIV status was not associated with any outcomes in the overall cohort. Limitations include the nonrandomised study design and lack of maternal mental health assessment at the child’s birth.
EBF was associated with fewer than average conduct disorders and weakly associated with improved cognitive development in boys. Efforts to improve stimulation at home, reduce maternal stress, and enable crèche attendance are likely to improve executive function and emotional-behavioural development of children.
| The benefits of exclusive breastfeeding (EBF) in early life are well established and include optimal nutrition and protection from infectious diseases.
The longer-term benefits of EBF on child development and behaviour are less clear, and studies in low-income settings have shown conflicting results depending on the design of the study and whether other factors known to influence development, such as HIV exposure, have been taken into account.
There is a dearth of evidence examining the development of HIV-uninfected children born to HIV-infected mothers and whether these children are disadvantaged compared to those born to HIV-uninfected mothers.
This study was established in 2012 to investigate the development (cognitive and emotion-behaviour) of 1,536 HIV-negative children, born to HIV-infected and HIV-uninfected mothers, in rural South Africa, taking into account a range of current and early life factors known to be associated with child development.
Duration of EBF was associated with a reduction in conduct disorders in girls and boys, but there was no association with cognitive development in the overall sample, when allowing for other factors. Maternal intelligence quotient (IQ) was strongly associated with children’s later cognitive development.
Children born to HIV-infected mothers performed as well as the children born to HIV-uninfected mothers.
Promoting, protecting, and supporting EBF may result in fewer conduct disorders, in addition to the established benefits of improved nutrition and reduced morbidity and mortality.
Reducing conduct disorders is important because they may lead to aggressive or disruptive behaviours and are associated with an increase in later criminal behaviour and poor long-term mental health and academic achievement.
| There is strong evidence that exclusive breastfeeding (EBF) for 6 mo, as recommended by the World Health Organization (WHO) [1], optimises infant nutrition and substantially decreases mortality and morbidity from infectious diseases [2,3]. The relationship between EBF and cognitive development is less clear [4,5], although studies in high-income settings [6–8], a randomised trial from Belarus [9], and a recent study from Brazil [10] have shown positive associations. A large systematic review showed conflicting results depending on the study design and rigour, as well as the number of factors adjusted for [4]. The few studies from resource-limited settings were almost twice as likely to find no association. This suggests that confounding variables, including socioeconomic status and maternal cognitive ability, affect the choice to breastfeed and the positive effects found. Measuring the duration of EBF accurately is challenging because of factors related to definition, timing, and duration of recall [11,12], and many of the studies were limited by poor documentation of breastfeeding patterns [13,14]. Further limitations included small sample sizes [15,16] and predominantly Caucasian populations, with only one small study from Africa [16], which found no association with cognitive development but some advantages for child behaviour in breastfed infants. There was no evidence from HIV-prevalent areas where the long-term effect of EBF on child development remains unquantified.
Studies exploring the link between EBF and later development have focused on core cognitive development, sometimes termed the intelligence quotient (IQ). However, higher-order cognitive function, termed executive function, is critical for later development, particularly the ability to function in society [17]. Executive function coordinates and controls information processing, which is important for a child’s ability to manage emotions and behaviour, to follow rules, to concentrate, and to form friendships. Thus, executive function influences educational and social success [18]. Executive function is susceptible to environmental influences and therefore an important intervention target [19]. In addition, few breastfeeding studies have examined emotional-behavioural development, an important outcome affected by early life factors, which predicts later educational achievement.
The Vertical Transmission Study (VTS) (2001–2006) supported HIV-positive and HIV-negative women to practice EBF in a rural area of South Africa before antiretroviral treatment became available [20], providing the first evidence that EBF reduced the risk of postnatal HIV transmission [21] and was associated with significant benefits for children’s health and growth until up to 2 y of age [22,23] (Registration: NCT01948557, National Institute of Health, ClinicalTrials.gov). Here we investigate the association between EBF, HIV exposure, and other early and current life factors and later cognitive development, executive function, and emotional-behavioural development in VTS children now aged 7 to 11 y. We accounted for maternal cognitive function, home stimulation, crèche attendance, and maternal/caregiver stress and mental health and hypothesized that EBF would result in improved longer-term development in children, despite exposure to HIV and poverty.
Ethics permission for this study was granted by the Biomedical Research Ethics Committee (BREC), University of KwaZulu-Natal, South Africa (BF184/12). Women were contacted by telephone or a home visit to ask if they would be interested in this study. Those who agreed were then visited by a field worker who explained, and provided written details of, the study and obtained written informed consent from the mother and primary caregiver (if this was not the child’s mother).
The VTS, a nonrandomised, prospective, intervention cohort study, was implemented between 2001 and 2006 from the Africa Centre for Population Health, which also hosts a Demographic Surveillance System (DSS) platform [24]. This lay-counsellor, home-based intervention was designed to support mothers to practice EBF for the first 180 d of life [25]. Between 2012 and 2014, we re-enrolled HIV-negative children (aged 7–11 y) born to HIV-positive (“exposed”) and HIV-negative (“unexposed”) mothers from the VTS cohort; HIV-infected children were not re-enrolled because they have different developmental trajectories [26].
To establish a comparative population mean for the developmental outcomes (in the absence of appropriate normative data for validated cognitive assessments), we assessed 630 (485 unexposed; 145 exposed) same-aged HIV-negative children from the DSS, not included in the VTS, and combined these with the VTS sample. Mothers in the DSS group were exposed to the same antenatal care at local clinics, including messages regarding HIV and infant feeding, but not to the home-based intervention to support EBF. We aimed to rerecruit all available 1,289 VTS children meeting the inclusion criteria, of whom 935 (75%) were enrolled and 906 (70%) fully assessed. To establish a robust population mean, we used the population platform of the Africa Centre for Population Health surveillance to identify all 1,226 resident children who were matched for age and HIV exposure to the VTS children but had not been exposed to the VTS intervention. Of these, 844 children met eligibility criteria, 657 (77%) enrolled, and 630 (75%) completed assessments.
This analysis includes the developmental outcomes of the VTS children, for all of whom we have accurate data on infant feeding and HIV exposure; their outcomes are related to the population means. Children were enrolled if both the mother and child were alive, the child was a resident in the research area, the mother and child’s current HIV status was known, and, for the DSS children, if the HIV status during pregnancy was known, the mother received antenatal services in the study community, and the maternal-held child Road-to-Health Card was available.
All psychometric measures had been previously used in the population; clinical algorithms for depression and anxiety were used.
The home environment was assessed using a locally adapted version of the Home Observation for Measurement of the Environment (HOME) inventory [36]. Maternal cognitive ability was assessed using the Standard Raven’s Progressive Matrix [37].
Data were collected over three visits between September 2012 and September 2014. Study consent was obtained at Visit 1, current socioeconomic and health data and mothers’ mental health and cognitive ability at Visit 2, and children’s cognition and executive function at Visit 3. When the mother was not the primary caregiver, mental health assessments were completed by the child’s primary caregiver. Assessments were conducted by graduate-level research assistants with 3–5 y of child developmental assessment experience. The median number of days between Visit 2 and Visit 3 was 18 d.
Analyses were based on data extracted on 30 October 2014 and conducted using STATA version 13. For each outcome, we calculated a population mean from all VTS and DSS children and then created a binary indicator by splitting the VTS group into those scoring above or at/below the mean. For the HOME assessment and Raven’s score, we created a low/high indicator consisting of equal-size groups by splitting the VTS sample based on their median. In the VTS, daily feeding data were collected at weekly intervals. We defined EBF as the total number of days in the first 6 mo that the child received only breastmilk and then divided this number by 30, into months, irrespective of whether the days were sequential. We previously reported [25] that approximately 40% of VTS women interrupted EBF at some point in the first 6 mo, mostly by giving water or formula milk, of whom approximately 60% returned to EBF within 2 d. We considered that the total number of days of EBF in the first 6 mo was more likely to have an impact on child development than whether the days were sequential or not, and we did not wish to exclude children who had received breastmilk for nearly all 180 d except for 1 or 2 d when they received breastmilk and other fluids. Based on the existing literature and theoretical and conceptual reasoning, we identified relevant factors, including child, maternal, early life, socioeconomic, and household factors related to child development; we did not apply any stepwise regression techniques. For selection of the most relevant socioeconomic variable, we used principal components analysis to identify the top variables that explained the overall variance. We modelled each of the outcomes using complete case logistic regression analyses, accounting for intramother correlation (for twins). We included child sex, child age, birth order, and maternal age; early life factors including birthweight, maternal HIV status during pregnancy, months of EBF, urban/rural residence, ownership of a fridge (wealth indicator), maternal education, and whether the mother was the main income provider at the time of the child’s birth; and other factors including maternal current HIV status, cognitive ability, mental health and parenting stress, crèche exposure, current indicators of perception of household wealth, and HOME assessment score. Sex differences in cognitive development exist at the primary school age [38], and we also estimated sex-stratified logistic regression models, using the same outcomes.
We explored several approaches to modelling the developmental data, including continuous outcomes (see S1 Table) and upper versus lower quintiles, as well as other methods of categorising the EBF data, including cumulative, sequential, days of EBF, and ever/never EBF, but the results were not substantially different.
By end of the VTS follow-up in September 2006, when children were aged 2 y, 1,289 HIV-negative children were alive, of whom 941 were eligible for re-enrolment and 906 were assessed (Fig 1). Compared to HIV-unexposed children, exposed children were more likely to have a mother who was older and the main income provider, less likely to have been exclusively breastfed until 6 mo of age and to have attended crèche, and more likely to have a primary caregiver with a current mental health disorder (Table 1). Compared to children included in the current analysis, the 383 VTS children excluded were more likely to have a younger HIV-uninfected mother with more years of education at the time of pregnancy and were less likely to have a low birthweight (Table 1).
None of the cognitive development measurements were significantly associated with EBF or maternal HIV status in adjusted analyses (Table 2). In multivariable analyses, the only variable significantly positively associated with performance on all cognitive subscales was maternal cognitive ability (measured using the Standard Raven’s Progressive Matrix) (Table 2). Boys were approximately 30% less likely than girls to score above the mean in the Sequential Processing subscale, which tests audio and visual memory and memory span (adjusted odds ratio [aOR] 0.71 [95% CI 0.5–0.9], p = 0.03), whilst children who had attended crèche were almost twice as likely to score above the mean (aOR 1.96 [95% CI 1.1–3.5], p = 0.02). Children who were older at assessment performed worse on Riddles (aOR 0.40 [95% CI 0.2–1.0], p = 0.05).
None of the executive function measures were significantly associated with EBF duration, maternal HIV, or child sex (Table 3). Compared to children whose mothers were aged less than 20 y, those with mothers aged 20–29 y at their birth were almost twice as likely to score above the mean on the Animal Sorting subtest (aOR 1.82 [95% CI 1.2–2.8], p = 0.01), as were children whose mother was the main income provider during their infancy (aOR 1.81 [95% CI 1.0–3.1], p = 0.03) and those who had attended crèche (aOR 1.74 [95% CI 1.0–3.0], p = 0.05). For the Auditory Attention subtest, compared to children aged 8 y, those aged 9 and 10 y were over three (aOR 3.38 [95% CI 1.6–7.3], p = 0.01) and four times (aOR 4.56 [95% CI 2.1–9.9], p < 0.001) more likely to perform above average, respectively. Children with better stimulation at home (i.e., a HOME score above the median) were more likely to perform above the mean in the Auditory Attention (aOR 1.36 [95% CI 1.0–1.8], p = 0.04) and Response Set subtests (aOR 1.35 [95% CI 1.0–1.8], p = 0.05).
Being born in an urban environment and having a primary caregiver with high parenting stress were associated with more emotional and behavioural problems (higher scores on the Internalising and Externalising subscales and the Total score) (urban Total score: aOR 1.62 [95% CI 1.2–2.2], p = 0.01; parenting stress Total score: aOR 7.04 [95% CI 4.2–11.9], p < 0.001). Children whose caregiver had a current mental health disorder were more likely to score above the mean for Internalising (aOR 1.92 [95% CI 1.1–3.4], p = 0.03) and Total scores (aOR 2.44 [95% CI 1.3–4.6], p = 0.01) (Table 4). Boys were more likely to score above the mean for Internalising (aOR 1.53 [95% CI 1.1–2.0], p = 0.01), whilst children who attended a crèche were approximately twice as likely to score above the mean in Externalising (aOR 2.15 [95% CI 1.2–3.9], p = 0.01) and Total scores (aOR 1.96 [95% CI 1.0–3.8], p = 0.05). EBF and the mother’s antenatal or current HIV status were not significantly associated with Externalising, Internalising, or Total CBCL score.
Exploring the six DSM disorders (Table 5), EBF was significantly associated with lower scores (fewer problems) for conduct disorders. Those who were exclusively breastfed for 6 mo compared to 1 mo or less were approximately half as likely to score above the mean for conduct disorders (aOR 0.44 [95% CI 0.3–0.7], p < 0.01). Caregiver mental health and stress were associated with increases in all six disorders. Urban residence was associated with increases in somatic, attention deficit hyperactivity disorder (ADHD), and oppositional problems. Boys were less likely to be anxious (aOR 0.64 [95% CI 0.5–0.9], p < 0.01) but more likely to have somatic (aOR 1.34 [95% CI 1.0–1.8], p = 0.05) or oppositional (aOR 1.52 [95% CI 1.1–2.2], p = 0.02) disorders.
Contrary to the finding in the overall cohort, boys, but not girls, who were exclusively breastfed for more than 1 mo were twice as likely as those who were exclusively breastfed for a very short period to score above the mean for Learning Ability (aOR 2.07 [95% CI 1.0–4.3], p = 0.05) and half as likely to score above the mean for Externalising problems (aOR 0.48 [95% CI 0.2–1.0], p = 0.05) (Tables 7 and 9). However, girls who were exclusively breastfed for less than 1 mo were more likely to score above average on Auditory Attention compared to those who were exclusively breastfed longer. The finding of an association between maternal cognitive ability and improved performance on all four cognitive subscales in the overall cohort held for boys but not for girls. Boys whose mother’s cognitive ability was above the median score (Standard Raven’s Progressive Matrix) were twice as likely to score above average for the cognitive subscales (e.g., Planning subscale aOR 2.79 [95% CI 1.8–4.4], p ≤ 0.001). Maternal HIV status was not significantly associated with cognitive development, executive function, or emotional-behavioural problems overall. However, boys whose mothers became infected with HIV after pregnancy were more likely to score below the mean on the Planning Ability subscale than those whose mothers remained HIV negative (aOR 0.55 [95% CI 0.3–1.0], p = 0.05). Boys born to HIV-positive mothers, compared to those born to HIV-negative mothers, were more likely to score above average for reasoning and language ability (Riddles subtest) (aOR 1.92 [95% CI 1.1–3.3], p = 0.02). For girls, being born to an HIV-positive mother was associated with scoring below the mean for Planning Ability (aOR 0.53 [95% CI 0.3–0.9], p = 0.01) and below the mean (fewer problems) for Externalising (aOR 0.56 [95% CI 0.2–0.9], p = 0.03) and Total (aOR 0.56 [95% CI 0.3–1.0], p = 0.03) CBCL scores.
Girls, but not boys, with a normal birthweight were significantly more likely to score above the mean for Learning Ability (aOR 2.40 [95% CI 1.2–5.0], p = 0.02) and Planning Ability (aOR 2.04 [95% CI 1.1–3.9], p = 0.03) and below the mean (fewer problems) for Total CBCL score (aOR 0.42 [95% CI 0.2–0.9], p = 0.02). Boys of a birth order of five or more were less likely than firstborns to score above average on Animal Sorting (inhibition, planning, and cognitive flexibility) (aOR 0.40 [95% CI 0.2–1.0], p = 0.04); girls of a birth order of five or more were significantly more likely to have a lower Externalising CBCL score (aOR 0.38 [95% CI 0.2–0.9], p = 0.03).
In this stratified analysis, the associations between maternal education and cognitive outcomes and HOME scores and executive function held for girls but not for boys. Likewise, girls were approximately three times more likely to score above the median for higher emotional-behavioural problems if their mothers had a mental health problem (Total CBCL aOR 3.28 [95% CI 1.4–8.0], p = 0.01) or high parenting stress (Total CBCL 4.63 [2.1–10.1], p < 0.001).
In the full model, interactions between sex and EBF were significant for Conduct (p = 0.02) and CBCL Internalising (p = 0.02) outcomes and marginally significant for Learning (p = 0.07) and Auditory Attention (p = 0.08).
In summary, while duration of EBF was not associated with child cognitive development or executive function in the overall sample (including girls and boys), it was associated with fewer conduct disorders, and, when stratified by sex, there was weak evidence of improved cognitive development in boys.
To our knowledge, this is the first study examining EBF, HIV-exposure, and child cognition, executive function, and emotional-behavioural outcomes at primary school age in Africa.
Our finding that duration of EBF was associated with fewer conduct disorders has significant implications. Conduct disorders lead to aggressive or disruptive behaviours that interfere with learning and peer relationships, in turn leading to low self-esteem and further behavioural problems [39]. Conduct disorders in childhood are associated with an increase in violent criminal behaviour and poor long-term mental health and academic achievement in later life [40]. While to our knowledge there have been no formal analyses of the economic costs of conduct disorders in low-middle-income countries, the evidence from carefully conducted studies in high-income countries is that the costs are enormous [41]. For example, a report from the United Kingdom stated that the total cost of crime attributable to people who had a conduct disorder in childhood was estimated to be £60 billion per annum [42]. Given these costs to individuals, families, and society, it is highly relevant that this study has shown EBF to be associated with reduced likelihood of conduct disorders at this critical stage of development. Further, for boys, a longer EBF duration was weakly associated with a doubling of the odds of better cognitive development on the Learning subscale, which assesses ability for maintaining focused attention while coding and storing complex auditory and visual stimuli simultaneously and generating strategies to learn efficiently. Identifying potential strategies to improve the life chances of boys is important, and, in our context, EBF appears to be associated with a longer-term advantage for boys. There is increasing evidence that some groups of children are more susceptible to the effects of negative rearing and, importantly, that they may benefit more from the effects of positive rearing and interventions. This is known as the “differential susceptibility hypothesis” [42]. Boys may be more susceptible in a number of domains of development to the effects of negative rearing—for example, in language development [43,44]—and therefore, in line with our findings, may benefit more from early EBF.
The finding that EBF did not have an effect on children’s cognitive development overall is in accord with studies from other resource-limited settings that also found no association [4], including studies in India (n = 514, aged 9–10 y), China (n = 442, aged 3 y), Malaysia (n = 1,394, aged 9 y), Chile (n = 784, aged 5.5 y), and the Philippines (n = 1,984, aged 8.5 and 11.5 y), although these studies did not stratify by sex.
An important strength of this study is the investigation of a wide range of key determinants of child development. In our cohort, maternal cognitive ability was strongly positively associated with children’s performance across all cognitive subscales, but not with executive function, although higher maternal cognitive ability was associated with improved performance on the Planning subscale, which assesses ability to focus attention, make decisions, and apply working memory, abilities considered to reflect executive function [27]. Interestingly, while having attended a crèche was associated with improvement in children’s cognitive development (sequential scores that depend on practice, on which boys scored particularly low) at the primary school age, it was independently associated with poorer behaviour, with similar associations found for children born in urban and rural settings. This apparent paradox in relation to crèche exposure is consistent with data from high-income countries [45].
Executive function is a high-order cognitive function that coordinates and controls information processing [18]. It is critical in enabling a child to successfully integrate into environments such as school [46]. We found a positive association between executive function and better stimulation at home, older maternal age, and a mother who was the main income provider at the time of the child’s birth. These findings are in accord with evidence suggesting that executive function is susceptible to environmental factors [47]. Executive function is increasingly thought of as the core ingredients (creativity, flexibility, self-control, and discipline) that will determine a child’s success in life. There is evidence that executive function is negatively affected by stress, emotions, poor physical fitness, and childhood obesity [39]. Encouragingly, a growing body of evidence suggests that simple, low-cost interventions such as noncomputerized games, aerobics, martial arts, and mindfulness may support improvements in executive function [19].
Our finding that poor maternal mental health and high parenting stress, measured at the time of the child assessments, were associated with increased emotional-behavioural problems in children was unsurprising but useful to quantify in an African setting. The link between parental mental health and child behaviour outcomes is well-established, with children of depressed, anxious, or highly stressed mothers known to be at increased risk of psychological and behavioural problems [48,49].
There is a dearth of evidence examining the developmental trajectories of HIV-exposed children, in particular longitudinal studies including HIV-unexposed controls. A recent systematic review [26] examining HIV exposure and child development found only 11 studies (1,591 children aged 0–18 y) and showed that HIV-exposed children are disadvantaged compared to their HIV-unexposed peers, in particular in emotional-behavioural and, to a lesser extent, cognitive development. However, results are not consistent across research settings or age groups, with most of the current evidence being based on small samples with wide heterogeneity in outcome measures [26]. Our results, based on a large sample of children including HIV-unexposed controls suggest that, overall, HIV-exposed children performed as well as HIV-unexposed children in the domains examined and that the findings of other small studies may be overstated.
Limitations include the nonrandomised study design of the original intervention, the lack of maternal mental health measurement at the child’s birth, and no assessment of cognition or emotional/behavioural problems at earlier time points in the children’s lives. Infections of the developing brain and childhood malnutrition also affect later cognitive ability but were not included in the analyses. In addition, caution must be taken in interpreting the sex-stratified models, as examining subgroup effects increases uncertainty and is more likely to produce larger effect estimates. Further, because of the complexities in measuring cognitive and executive function and the need to model each of the outcomes separately, multiple tests were used, and there may be some false positive significant results. However, our results are in line with findings from other studies.
Strengths include a large cohort of HIV-exposed and HIV-unexposed children, population normal values, adjustment for a wide range of confounders including current maternal IQ, and a battery of culturally appropriate developmental assessments including executive function and behavioural outcomes. A unique strength in this study is the accurate documentation of daily EBF data. Both HIV-positive and HIV-negative women had high rates of EBF, with few breast health problems, likely due to the quality of lay breastfeeding counselling [25,50]. Given that all women received intensive support, it is possible that this may have limited our ability to detect differences between EBF and non-EBF children.
In conclusion, while EBF was not significantly associated with cognitive development at the primary-school age, there was an association between EBF and a reduction in conduct disorders and, for boys, weak evidence of positive associations both in aspects of cognitive development and behavioural problems more generally. Given the number of adverse factors in these families’ environments, including poverty and high HIV prevalence, the fact that these benefits were evident into the crucial early school years is important. Child outcomes were associated with a range of other key factors. While core cognitive development was principally associated with maternal cognitive abilities, executive function was associated with a number of modifiable environmental factors including home stimulation and crèche attendance. Child’s emotional development was largely associated with caregiver mental health. These findings highlight a number of avenues for potential interventions.
|
10.1371/journal.pgen.1000963 | Caenorhabditis elegans SMA-10/LRIG Is a Conserved Transmembrane Protein that Enhances Bone Morphogenetic Protein Signaling | Bone morphogenetic protein (BMP) pathways control an array of developmental and homeostatic events, and must themselves be exquisitely controlled. Here, we identify Caenorhabditis elegans SMA-10 as a positive extracellular regulator of BMP–like receptor signaling. SMA-10 acts genetically in a BMP–like (Sma/Mab) pathway between the ligand DBL-1 and its receptors SMA-6 and DAF-4. We cloned sma-10 and show that it has fifteen leucine-rich repeats and three immunoglobulin-like domains, hallmarks of an LRIG subfamily of transmembrane proteins. SMA-10 is required in the hypodermis, where the core Sma/Mab signaling components function. We demonstrate functional conservation of LRIGs by rescuing sma-10(lf) animals with the Drosophila ortholog lambik, showing that SMA-10 physically binds the DBL-1 receptors SMA-6 and DAF-4 and enhances signaling in vitro. This interaction is evolutionarily conserved, evidenced by LRIG1 binding to vertebrate receptors. We propose a new role for LRIG family members: the positive regulation of BMP signaling by binding both Type I and Type II receptors.
| Bone morphogenetic protein (BMP) family members, small secreted signaling molecules, play diverse roles in development and homeostasis. Uncontrolled BMP signaling results in a variety of disorders and diseases. BMPs signal to receiving cells through two receptor types, which act together to propagate the BMP signal within cells. To understand how BMP signaling is controlled, we used the nematode Caenorhabditis elegans to identify conserved regulators of BMP signaling. Here, we characterize SMA-10, the first extracellular positive regulator of DBL-1/BMP receptor-mediated signaling. SMA-10 is a new member of a family with leucine rich repeats and immunoglobulin-like domains (LRIG). SMA-10 physically binds the two types of DBL-1/BMP receptor. We demonstrate conservation of LRIG function by showing that a Drosophila melanogaster LRIG can functionally substitute for loss of C. elegans SMA-10/LRIG, that C. elegans SMA-10 can directly promote mammalian BMP signaling in cells, and that mammalian LRIG1 interacts with BMP receptors. Our work establishes a role for LRIGs in BMP regulation through binding both types of BMP receptor.
| Bone morphogenetic protein (BMP) receptor serine/threonine kinases (BMPRs) are pivotal signal transducers for the small, secreted BMP morphogens, members of the transforming growth factor β (TGF- β) superfamily (comprising subfamilies of TGF- βs, BMPs, activins, and others) [1], [2]. BMP dimers released from neighboring cells are received by these receptors, which leads to an intracellular cascade of transcriptional events. Depending on the specific pathway, cell type and milieu, these events result in a diverse array of cellular processes, from dorsal-ventral specification to cell cycle control and programmed cell death [3]. Understanding how growth factor pathways are regulated may lead not only to a better understanding of their normal physiological roles, but may also lead to potential treatments for a wide range of disorders and diseases [4], [5].
Secreted BMP dimers travel through the extracellular matrix to activate their receptors. Originally thought to be a process of simple diffusion, the movement of TGF-β superfamily members is now recognized to be highly regulated [6]. Many factors play a role in facilitating or preventing BMP ligand access to receptor. Post-translational processing and proteolysis of ligand, as well as seclusion of ligand by extracellular matrix (ECM) components like integrins and proteoglycans, for example, determine whether a ligand dimer can interact with its receptors [7]. Not only is the BMP's progress exquisitely controlled, but the receptors themselves are also subject to regulation [6], [8]. Inside the cell, receptor phosphorylation is inhibited by phosphatases, and SARA and Smurf proteins target receptors for polyubiquitination and degradation [8]. Outside the cell, the receptor complex can be inactivated by the decoy type I receptor BAMBI [9]. Coreceptors betaglycan/TGFβ receptor II (TGFβR3) and endoglin can bind certain BMPS and deliver them to receptors [10]–[14]. Endoglin also associates with select type I and type II receptors [13].
Pioneer studies in Caenorhabditis elegans and Drosophila melanogaster have identified components of the pathway and furthered understanding of BMP signaling [15], [16]. These studies have identified the conserved core of the signaling pathway, including the ligand, the type I and type II receptors, and the Smads. In C. elegans, a BMP-like pathway controls body size and male tail development (the Sma/Mab pathway). The receptors for the ligand DBL-1 are SMA-6 (type I) and DAF-4 (type II) [17], [18]. Receptor signals are transduced through the Smads SMA-2, SMA-3, and SMA-4 [19]. As in mammals, co-transcription factors that act with the Smads have been discovered, including SMA-9/Schnurri and RNT-1/RUNX [20], [21]. The body size phenotype of the Sma/Mab pathway is very sensitive to dose, indicating that the core pathway is tightly controlled [17], [18], [22], [23]. The only identified extracellular regulator of this pathway has been LON-2, a conserved heparan sulfate proteoglycan that binds DBL-1/BMP and attenuates pathway signaling [24].
To identify novel components of BMP signaling, we performed traditional forward genetic screens for mutations affecting the Sma/Mab signaling pathway ([25], unpublished results). From these genetic screens, we identified, cloned, and characterized sma-10. SMA-10 is a positive regulator of Sma/Mab signaling, and is absolutely and specifically required for body size regulation. It is a member of a cell surface localized family of proteins with leucine rich repeats and immunoglobulin-like domains (LRIG). The function of SMA-10 in BMP pathway signaling is conserved, as the Drosophila ortholog lambik rescues the body size defect of sma-10(lf) animals. Furthermore, SMA-10 promotes BMP signaling in mammalian cells. SMA-10 binds the pathway receptors SMA-6 and DAF-4 but not the BMP DBL-1, and a mammalian ortholog, LRIG1 (leucine-rich and immunoglobulin-like domains-1), also binds both type I and type II receptors. These studies identify a uniquely acting positive regulator of BMP signaling, SMA-10/LRIG, that directly interacts with type I and type II receptors from C. elegans to mammals.
The first small C. elegans mutants were identified in a large-scale screen for morphology and mobility mutants [26]. Their role in BMP signaling was elucidated when sma-2, sma-3 and sma-4 were characterized [19]. In an effort to identify additional genes that act in BMP signaling, we performed two genetic screens. From the first screen, in which body size mutant F2 animals were isolated from mutagenized N2/wild type P0 animals, two sma-10 alleles, wk26 and wk66, were identified [25]. From a lon-2(e678) suppressor screen, three additional alleles, wk88, wk89, and wk90, were identified and confirmed by complementation and sequencing.
The Sma/Mab pathway regulates both body size and male tail development. A reduction of dbl-1 pathway activity results in animals that are 55%–85% wild-type length [22],[23]. sma-10(lf) animals share the small body size defect, ranging from 79% to 88% the length of wild-type animals (Figure 1A and 1B, Table 1). The dbl-1 pathway also regulates the development and patterning of male tail structures [17]–[19], [22], [23], with mating spicules and sensory rays 5–7 being primarily affected. We therefore asked if sma-10 is also involved in patterning the male tail. Our studies revealed that all five alleles of sma-10, including a presumed null (wk88), have wild-type male tail rays and spicules (data not shown). These data suggest that SMA-10 is specifically involved in BMP signaling to control body size but not male tail development or patterning.
Because small body size is a hallmark phenotype of mutants in the Sma/Mab pathway, we asked if sma-10 genetically interacts with Sma/Mab pathway members. To determine the functional order of SMA-10 relative to Sma/Mab pathway members, we performed epistasis tests. We generated animals with mutations in sma-10 and DBL-1/BMP pathway genes that alone give opposite phenotypes (small and long, respectively), and asked what the terminal phenotype was for these strains (small or long). In a regulatory pathway, the terminal phenotype is a result of the loss of gene product with the most downstream effect bypassing the requirement for the second, upstream gene product. We created double mutant animals of sma-10 with lon-2, overexpressed dbl-1, or over expressed sma-6. We found that the small body size phenotype of sma-10(lf) was the terminal phenotype in double mutant animals with either lon-2 or over expressed dbl-1, but not for overexpressed sma-6 (Table 2). These results suggest that SMA-10 functions in the unique position between the receptor SMA-6 and the ligand DBL-1.
To understand the molecular nature of sma-10, we mapped and cloned the gene. sma-10 is located at LG IV: −26.82. Cosmid T21D12 conferred rescue of the sma-10(wk66) small phenotype, as does a sma-10 cDNA (Table S1). sma-10 encodes an 881 amino acid protein of the LRIG family (leucine rich repeats and immunoglobulin-like domains). SMA-10 is composed of an N-terminal signal sequence (amino acids 1–20), fifteen leucine-rich repeats (LRRs) flanked by an LRR N-terminal domain at amino acids 24–56 and an LRR C-terminal domain ending at amino acid 499, three immunoglobulin domains (spanning amino acids 503 to 802), a transmembrane domain (amino acids 839 to 861), and a short (19 amino acid) intracellular domain from amino acids 862 to 881 (Figure 2). The C-terminal tail is not conserved between C. elegans, Drosophila and mammals. This protein structure, which is largely extracellular and is transmembrane-bound, is consistent with the order of gene function, which places SMA-10 between the secreted DBL-1/BMP and its membrane-bound receptors.
Sequencing DNA of mutant sma-10 strains confirmed lesions within T21D12.9 (Figure 2). sma-10(wk26) contains a T to C bp change at position 572, a substitution that changes a conserved leucine to a phenylalanine in the second LRR at amino acid 102. A deletion of 251 bps generates sma-10(wk66) (on cosmid T21D12 from base pairs 21092 to 21342, beginning in the gene at the end of exon 9). This transcript is predicted to result in a truncated protein with 609 a.a. of SMA-10 and 10 a.a. of novel sequence before terminating, deleting sequences after the first immunoglobulin-like domain. wk88 introduces a stop codon into amino acid position 112 in the LRR2, resulting in a severely truncated protein, and is a presumed null allele. wk89 is a 975 bp deletion from position 21301 to 22275 in T21D12, starting in intron 9 and ending in intron 11, deleting the protein after amino acid 611 and creating a frame shift and premature termination sequence, removing sequences after the first immunoglobulin-like domain. wk90 changes sequence encoding Trp 286 to a stop codon, deleting sequences after LRR9.
Multiple splice variants are predicted by cDNA sequencing (Wormbase.org). yk352c5, a full-length 2.6 kb cDNA of the longest splice variant T21D12.9a, driven by 1.2 kb of upstream sma-10 promoter sequence, is sufficient to confer rescue of sma-10(wk66) animals (Table S1). A function for the shorter splice variants (T21D12.9b, T21D12.9c.1, and T21D12.9c.2) is not known.
Database searches reveal orthologs in other metazoans. The hallmark of this family is the presence of fifteen leucine-rich repeats (LRRs) and three immunoglobulin-like repeats (Ig-like) followed by a transmembrane domain. In Drosophila, there is one SMA-10 ortholog, Lambik (www.flybase.org), and in vertebrates, there are three known sma-10 orthologs, LRIG1, LRIG2, and LRIG3 [27]. lambik corresponds to CG8434, but no phenotypic characterization has been published. A distantly related, distinct family is composed of the insect-specific Kekkon members, which have six LRRs and one Ig–like domain. One of these, Kekkon 1 (kek1), plays a role in inhibiting the epidermal growth factor receptor (EGFR) [28]. Another Kekkon, Kekkon5 (kek5), interacts genetically with a BMP signaling pathway [29].
Although core Sma/Mab pathway components are expressed in many tissues, they are all required in the hypodermis for body size regulation. To determine if SMA-10 is also needed in these cells, we first asked where sma-10 is expressed. We created a functional translational fusion of SMA-10 with GFP at the C-terminus and expressed it using the sma-10 promoter. Expression of SMA-10:GFP was visible, though faint, in the hypodermis (Figure 1C and 1D), consistent with SMA-10's genetically identified role as a regulator of the DBL-1 signaling pathway, which functions in the hypodermis to regulate body size. Expression in the hypodermis of other DBL-1 pathway genes is also low [24], [30]–[33].
Bright fluorescent expression was also observed in the pharynx of animals from embryos to adults (Figure 1E). Pharyngeal expression was localized to the muscle segments pm1 (anterior cell of the procorpus), pm4 (metacarpus/anterior bulb), the anterior part of pm5 (isthmus), and pm7 (terminal bulb) (Figure 1E). Expression of SMA-10:GFP localized to both the cell surface and to puncta within the cells (Figure 1E). Expression of mRNA corresponding to the sma-10 locus has also been reported in intestine and renal gland cells in both larvae and adult animals [34]. Digestive tract expression (in the pharynx and intestine) of other Sma/Mab pathway members has been reported [18], [32], [33], [35], [36], and may reflect a proposed role for this pathway in an innate immune response [37], [38].
We then asked where SMA-10 is required for body size regulation. Because of its expression pattern and previous experiments showing that DBL-1/BMP pathway components are required in the hypodermis, we drove the expression of full-length genomic sma-10 from pharyngeal- or hypodermal-specific promoters in sma-10(wk66) animals and assayed for rescue of body size. We found that expression of sma-10(+) in hypodermis (using the rol-6 promoter), but not pharynx (using the myo-2 promoter), was sufficient to rescue animals to wild-type length (Table 3).
Lambik is a Drosophila LRIG that is orthologous to vertebrate LRIGs and to SMA-10. Although the function of Lambik is currently unknown, we asked whether lambik could functionally substitute for sma-10 in C. elegans. We drove Drosophila lambik cDNA from the sma-10 promoter in sma-10(wk66) animals. Transgenic animals were rescued to the wild-type body size (Table 4), thereby showing functional conservation between divergent Drosophila and C. elegans LRIGS.
Our genetic data demonstrate that SMA-10 acts within a nematode BMP-like pathway to promote signaling. To test whether this function is conserved, we asked if SMA-10 could regulate BMP signaling in mammalian cells. We used a standard reporter assay in a BMP-responsive human cell line (HepG2), the BMP-response element from the Smad7 gene driving luciferase [39]. BMP2 induced reporter activity significantly over the controls, whereas in the presence of SMA-10, the response of the promoter was increased 4.2-fold over BMP2 alone (Figure 3A). Thus, SMA-10 can directly promote mammalian BMP signaling.
The genetic and molecular data strongly support a model in which SMA-10 acts on the extracellular surface of the plasma membrane, and could act via physical interactions with ligand, receptors, or ligand and receptors. To examine these possibilities and to provide mechanistic insights into SMA-10's function, we asked whether SMA-10 directly binds to ligand or to the DBL-1 receptors, SMA-6 and DAF-4. For the first experiment, we employed affinity labeling and substituted BMP2 for DBL-1, since DBL-1 protein is not readily available and BMP2 has been shown to physically interact with other DBL-1 pathway members [17], [24]. SMA-10 was FLAG-tagged and BMP2 ligand was radio-iodinated with 125I. After anti-FLAG immunoprecipitation and blotting, the blot was exposed to film. However, we failed to detect any binding of BMP2 to SMA-10 (Figure S1). We next asked whether SMA-10 interacted with the receptors. FLAG-tagged SMA-10 and HA-tagged receptors were transfected into 293T cells. To isolate SMA-10 and associated proteins from cells, lysates were first immunoprecipitated with anti-FLAG antibody and blotted. The blot was then probed with anti-HA antibody to determine whether receptors were bound to SMA-10. Under these conditions, both SMA-6 and DAF-4 receptors co-immunoprecipitated with SMA-10 (Figure 3B). Thus, SMA-10 physically interacts with receptors and not with the ligand.
Given that SMA-10 binds the Sma/Mab pathway receptors, we tested whether this interaction was conserved in a mammalian system. Using the mammalian ortholog LRIG1 and mammalian BMP receptors, we assayed the ability of LRIG1 to bind the receptors. FLAG-tagged LRIG1 and HA-tagged versions of BMP receptors were transfected into mammalian cells. Lysates were immunoprecipitated with anti-FLAG antibody to pull down LRIG1, blotted, and probed with anti-HA antibody to determine whether LRIG1 binds any BMP receptors. We found that LRIG1 interacted strongly with type I receptor ALK6 and more weakly with the type I receptors ALK1, ALK2, ALK3, and ActRIB (Figure 3C). In addition, we detected weak interactions with type II receptors ActRII and ActRIIB (Figure 3C). Thus, LRIGs interact with BMP receptors and represent a new class of BMP receptor-associated proteins.
Here we describe a conserved, novel regulator of BMP signaling, SMA-10, which is absolutely required for body size signaling in C. elegans. Loss of SMA-10 function results in a phenotype similar to a loss of any of the Sma/Mab pathway components. Molecular and genetic analyses place its function at the level of the cell membrane between the ligand and the receptors. Several lines of evidence support a model in which SMA-10 and its orthologs are conserved regulators of BMP signaling. The Drosophila homolog lambik rescues sma-10(lf) animals. Consistent with genetic evidence that SMA-10 is a positive effector of BMP signaling, SMA-10 stimulates a positive BMP-specific response in human cells. Furthermore, SMA-10 and LRIG1 physically interact with BMP type I and type II receptors. LRIGs have previously been shown to regulate EGF and Met receptor tyrosine kinase signaling [40], [41], and recently the insect-specific kekkon5 was shown to genetically interact with a BMP signaling pathway [29]. Our work reveals a new, conserved role for the LRIG subfamily in BMP receptor signaling.
We propose that SMA-10 acts at the hypodermal membrane surface, where the Sma/Mab receptors are located, to facilitate receptor signaling. Our genetic evidence shows that the SMA-10 protein acts between the secreted ligand and the transmembrane receptors (Table 2). The structure of SMA-10 indicates that it contains a secretion signal and a transmembrane region (Figure 2). Subcellular localization of a rescuing translational SMA-10:GFP fusion shows SMA-10 in hypodermal tissues and at cell membrane surfaces (Figure 1C and 1E). The requirement for sma-10 expression in the hypodermis to rescue the sma-10(lf) small body size further supports the model that SMA-10 acts at the hypodermal membrane to facilitate DBL-1 binding to its receptors (Table 3). We predict that the extracellular domain is responsible for this rescuing effect, as SMA-10's C-terminal intracellular tail is short (19 amino acids) and not conserved with Lambik, which has SMA-10 function, and SMA-10 is able to rescue when tagged with GFP at the C-terminus. We suggest that this intracellular sequence is not critical for transducing BMP signals.
Mutant animals for all previously characterized Sma/Mab core-signaling components have alterations in both body size and male tail defects. However, sma-10 only affects body size. This can be explained if sma-10 acts in a tissue-specific manner, namely the hypodermis, where DBL-1activated pathway signaling is required for body size control. Previous studies of targets downstream of this pathway (lon-1, mab-21, and mab-23) and one upstream regulator (LON-2) have shown that functions in the body size and male tail pathways are separable [22], [24], [30], [31], [42]. Given that sequencing reveals that one of the sma-10 alleles (wk88) is a presumed null, we favor the model that SMA-10 acts in a tissue-specific manner to enhance signal strength in the cells contributing to body size but not to the male tail. Consistent with this reasoning, differences in the requirement of signal strength between the tail and the hypodermis are seen with a hypomorphic allele of sma-6, where body size is reduced while tail morphology is normal [18]. It remains to be seen whether differential responses to Sma/Mab pathway dosage differences are sufficient to explain this tissue specificity or if there exists another male-specific factor that functions, like SMA-10 in the hypodermis, to positively regulate Sma/Mab signaling in the male tail. We note that all the sma-10 alleles result in small animals that are not as small as animals with loss-of-function mutations in previously identified core-signaling components. If tissue specificity in other organisms is a common feature, then this could explain why sma-10 orthologs were not previously identified as BMP signaling modulators. We note that mammals have three LRIGs, and mutating a single LRIG (LRIG1, a.k.a. LIG-1) results in a mild hyperplasia of epidermal cells [43]. This is a seemingly minor defect, but loss of LRIG1 function may be partially compensated by the other two functional LRIGs. Therefore, redundancy may also explain why LRIGs have eluded identification as regulators of BMP signaling in higher organisms.
The LRR and Ig-like domains exist singly in many other proteins, but only in the LRIG family do they exist in the same protein. The number of protein domains in the SMA-10/LRIG family, which excludes the insect Kekkon subfamily, is invariant. These domains have been shown to be involved with protein-protein interactions [44], [45]. In these studies, we show that SMA-10 and LRIG1 bind BMP type I and type II receptors, and our genetic evidence and studies based on overexpression in mammalian cells suggests that these proteins are positively required for BMP family signal transduction.
Kekkon 1 has been shown to be a negative regulator of the EGF pathway in Drosophila and acts by binding to the receptors [28]. Given the involvement of Kekkon 1 in EGF signaling, LRIG1 was tested for EGFR regulation in mammalian tissue culture [40]. LRIG1 was shown to also bind EGFRs and enhance their degradation [40]. Signaling of EGFs through EGFRs promotes cell proliferation. LRIG1 negatively regulates cell proliferation by down-regulating EGFR responsiveness by increasing activated receptor ubiquitination [28], [40]. LRIG1 binds the E3 ubiquitin ligase c-Cbl, bringing it into the EGFR complex. EGFR then phosphorylates c-Cbl, thereby activating it and promoting ubiquitination and degradation of both LRIG1 and EGFRs. LRIG1's amino acids 900 to 930 contain the binding site for c-Cbl [40], within its intracellular domain that is not shared by other LRIGs, SMA-10, or Lambik. An interaction of c-Cbl with LRIG2 or LRIG3 has not been demonstrated [46]. The ectodomain of LRIG1 alone inhibits EGFR signaling, but does so in a ubiquitin-independent fashion, showing that EGFR inhibition by LRIGs is not exclusively through ubiquitination [47]. EGFR signaling in C. elegans, mediated by a single EGFR (LET-23), directs several embryonic and larval cell fates and also ovulatory contractions in adult hermaphrodites [48]. We did not see any obvious defects associated with EGFR in sma-10(lf) animals that might suggest an interaction with EGFR. LRIG1 also binds to and inhibits signaling by hepatocyte growth factor receptor (Met), a tyrosine kinase that in many known cancers is mutated or misregulated to promote invasive growth, though its mechanism of inhibition is ubiquitin-independent [41]. C. elegans has no recognized Met receptor tyrosine kinase [49]. Another Kekkon, Kekkon5, affects BMP signaling, but the model, based on genetic and structure-function analyses, proposes that Kekkon5 regulates ligand distribution or activity rather than acts directly on receptors, as we show here for both C. elegans LRIG SMA-10 and mammalian LRIG1 [29].
Although their expression appears to be universal, human LRIGs are differentially expressed in various cancer cell types, being down regulated in many types studied, but being upregulated in others [46]. Various explanations have been proposed based on the current understanding of LRIG function [46]. Our research showing LRIG interaction with BMP receptors (BMPRs) leads us to propose a new model, where the cell-specific levels of BMP and EGF activity, which negatively and positively regulate cell growth, respectively, determine the cell's response to LRIG exposure.
The observation that SMA-10 localizes to intracellular puncta (Figure 1C) is reminiscent of endocytic vesicles, and suggests a model of action for SMA-10 and the Sma/Mab signaling pathway [50]. In other systems, there is evidence that TGF-β superfamily pathway signaling can be activated via receptor monoubiquitination and receptor complex endocytosis into early vesicles (reviewed in [51], [52]). SMA-10 may thus promote signaling by facilitating receptor internalization into early endosomes.
This work identifies a new, conserved component of BMP signaling, SMA-10/LRIG. Mammalian members play known roles in some receptor tyrosine kinase pathways, and this work identifies a new role for this family in BMP receptor serine/threonine kinase signaling. We have shown that two members of this family physically interact with both the type I and type II BMP receptors, and C. elegans SMA-10 enhances signaling in both the nematode and in mammalian cells.
Animals were maintained according to standard protocols [26]. All mutant strains used in this study were derived from the wild-type Bristol strain N2, and some mapping was accomplished using the wild-type Hawaii isolate CB4856. Alleles used include sma-10(wk26, wk66, wk88, wk89, wk90) [25], lon-2(e678), sma-6(wk7), ctIs40 [ZC421 (dbl-1(+)) + pTG96 (sur-5::gfp)] [23], bxIs16 [tph1::gfp + cat-2::yfp] [42]; and nIs128 [pkd-2::gfp] [53]. Arrays made for this study are wkEx47 (sma-10p::sma-10:gfp + pUC18 filler DNA, 50 ng/µl each), wkEx91 (myo-2p::sma-10(+)), wkEx92 (rol-6p::sma-10(+)), wkEx93 (sma-10p::lambik), texEx190 (sma-6p::sma-6(+):gfp), and texEx195 (sma-10p::sma-10(yk352c5).
Creation of transgenic arrays was performed by standard microinjection techniques [54]. Genomic sma-10 or Drosophila lambik cDNA was cloned into nematode expression vector pPD95.75 with the appropriate promoter sequence. wkEx47 was made by removing the stop codon of the sma-10 genomic sequence and fusing the gfp sequence in-frame at the 3′ end. Transgenic animals were generated by germline microinjection, using constructs at 50 ng/µl (HW480 sma-10p::lambik, CZ10.2 sma-10p::sma-10:gfp, and CZ9.1 sma-10(yk352c5)) or 0.5 ng/µl (HW469 rol-6p::sma-10 and HW477 myo-2p::sma-10) with the co-injection marker ttx-3p::rfp or ttx-3p::gfp at 50 ng/µl (with HW480 and CZ9.1) or 100 ng/µl (with HW469 and HW477). ttx-3p drives expression in AIY interneurons. One representative stable line for each transgene was measured.
Isolation of sma-10(wk26) and sma-10(wk66) was previously described [25]. In an effort to identify additional alleles of genes that act BMP signaling, lon-2(e678) hermaphrodites were mutagenized with 50 mM ethyl methanesulfonate (EMS) using standard procedures [26]. Mutagenized P0 animals were transferred to plates and allowed to segregate self-progeny. F1 animals were transferred to new plates to segregate progeny, which were then scored for a small phenotype in a quarter of the population. From about 9,000 mutagenized genomes screened, three additional alleles of sma-10 were isolated, wk88, wk89, and wk90. These alleles, as well as the two alleles isolated in the Sma screen [25], were outcrossed five times before further analyses were done.
To measure body size, animals were picked at the L4 stage and photographed as young adults about 24 hours later. Images from individual animals were captured from a dissecting microscope using an Optronics MagnaFire CCD camera system and software (Optronics, Goleta, CA). Perimeters (Table 1) or lengths (Table 2, Table 3, Table 4) of animals were determined by using Image-Pro Plus measurement software (Media Cybernetics, Inc., Silver Spring, MD). The images for Figure 1C and 1D were captured using an Axiovert 200 M microscope (Carl Zeiss MicroImaging, Oberkochen, Germany) equipped with a digital CCD camera (C4742-95-12ER, Hamamatsu Photonics, Hamamatsu, Japan) and were deconvolved with AutoDeblur software (AutoQuant Imaging, Watervliet, NY). The Figure 1E image was obtained on an Olympus IX81 with a Carv Nipkow disk confocal unit (Atto Biosciences, Rockville, MD) and SensiCam QE camera (Cooke Corp., Auburn Hills, MI).
We performed statistical analyses on these measurements. Individual measurements from each strain were averaged. We determined the ratio and 95% confidence interval of the average measurement (mean) of each strain to the wild-type strain mean. To verify the significance of our findings, we tested the null hypothesis that the ratios of the compared means are the same. The ratios compared were the double mutant strain mean/wild type mean to the single sma/transgenic mutant strain mean/wild type strain mean. For populations measured on the same day, the denominator (the average of wild-type measurements) was the same and a Student's t-test was used to test the null hypothesis. For populations measured on different days, the Welch-Satterthwaite equation was used to calculate the effective degrees of freedom for this 2-tailed t-test for two ratios. We determined the value of t with the calculated degrees of freedom and compared the t-value to the Student's t table to construct the probability (p) of the null hypothesis.
sma-10(wk66); lon double mutant animals were constructed by crossing heterozygous sma-10(wk66) males with lon-2(e678) or lon-1(wk50) hermaphrodites, with animals overexpressing an integrated transgene with wild-type dbl-1 (ctIs40), or with animals overexpressing an extrachromosomal array encoding functional SMA-6 (texEx190) [15]. Wild-type F1 were isolated and small and long F2 animals were picked to individual plates. The F3 generation was then examined for the presence of long animals from a small F2 parent or small animals from a long F2 parent.
sma-10(wk66) was previously mapped to chromosome IV by two-factor crosses [25]. We further refined its position by standard three-factor mapping and single nucleotide polymorphism mapping [26], [55]. We used microinjection and germline transformation rescue [54] to discover that YAC Y80C9 and cosmid T21D12 rescue the small phenotype of sma-10(wk66) animals. Each predicted gene on T21D12, with at least 1.5 kb of promoter sequence, was amplified by the polymerase chain reaction (PCR). The DNA product was purified and injected into sma-10(wk66) animals. Only T21D12.9 rescued the small phenotype. PCR confirmed altered or deleted sequences in all five sma-10 alleles.
293T cells were cultured in Dulbecco's modified Eagle's medium containing high glucose and supplemented with 10% fetal bovine serum (FBS). HepG2 cells were cultured in minimal essential media supplemented with 1% non-essential amino acids and 10% FBS. Cells were transfected by calcium phosphate and lysed 48 hours after transfection in TNTE buffer containing 0.5% Triton-X-100 (150 mM NaCl, 50 mM Tris ph 7.4, and 1 mM EDTA) [56]. Immunoprecipitations were carried out using M2 anti-FLAG (Sigma) or 12CA5 anti-HA (made in-house) followed by incubation with Protein-G Sepharose beads (Amersham Biosciences, Uppsala, Sweden). Immunoprecipitates were then washed four times with lysis buffer containing 0.1% Triton-X-100. Proteins were separated by SDS-PAGE and immunoblotted with anti-HA (12CA5) or anti-FLAG (M2).
Transcriptional assays were carried out using a previously described BMP-responsive element from the mouse Smad7 gene (I-BRE) driving firefly luciferase [39]. Luciferase assays were carried out as described [39]. Renilla luciferase expressed from the CMV promoter was used as an internal control for transfection efficiency. Firefly luciferase values were normalized using Renilla luciferase values.
1.2 kb promoter sequence 5′ of the sma-10 open reading frame driving expression of genomic sma-10 was sufficient to rescue body size defects in sma-10(lf) animals, and this same promoter region was fused to Drosophila lambik/CG8434 cDNA to address conservation of LRIG family function. Tissue specific expression was determined by driving expression of sma-10 genomic sequence with myo-2 (pharyngeal expression) or rol-6 (hypodermal expression) promoters [57]. We generated GFP-tagged SMA-10 by fusing the eGFP sequence to the 3′ terminus of wild-type sma-10 genomic sequence that lacked its termination codon.
The rescuing sma-6 construct was made by fusing the mCherry sequence in frame to the 3′ end of genomic sma-6 and expressing it using 1260 bp sma-6 promoter sequence.
3x FLAG SMA-10 was made by fusing the full-length sma-10 cDNA from the first Sal I site in frame to the 3x FLAG signal peptide containing vector p3xFLAG-CMV8 (Sigma). Similarly, we constructed the 3x FLAG-tagged LRIG-1 by subcloning the LRIG cDNA from an endogenous Sal I restriction site to the C-terminus into p3xFLAG-CMV8. This construct encodes a protein with the 3x FLAG tag between the N-terminal signal sequence and the first LRR of LRIG1. SMA-6 was C-terminally tagged with the HA1 epitope of influenza virus hemagglutinin (HA) and was cloned into the mammalian expression vector pCMV5 (T. Reguly and J. Wrana, unpublished work). DAF-4DC-HA is HA-tagged DAF-4 with a deletion of its C-terminal tail (T. Reguly and J. Wrana, unpublished work). ALKIHA, ALK3HA, ALK6HA, ALK2HA, ActRIBHA, ActRIIHA, ActRIIBHA have been previously described [58]–.
|
10.1371/journal.pbio.0050317 | Novel Roles of Formin mDia2 in Lamellipodia and Filopodia Formation in Motile Cells | Actin polymerization-driven protrusion of the leading edge is a key element of cell motility. The important actin nucleators formins and the Arp2/3 complex are believed to have nonoverlapping functions in inducing actin filament bundles in filopodia and dendritic networks in lamellipodia, respectively. We tested this idea by investigating the role of mDia2 formin in leading-edge protrusion by loss-of-function and gain-of-function approaches. Unexpectedly, mDia2 depletion by short interfering RNA (siRNA) severely inhibited lamellipodia. Structural analysis of the actin network in the few remaining lamellipodia suggested an mDia2 role in generation of long filaments. Consistently, constitutively active mDia2 (ΔGBD-mDia2) induced accumulation of long actin filaments in lamellipodia and increased persistence of lamellipodial protrusion. Depletion of mDia2 also inhibited filopodia, whereas expression of ΔGBD-mDia2 promoted their formation. Correlative light and electron microscopy showed that ΔGBD-mDia2–induced filopodia were formed from lamellipodial network through gradual convergence of long lamellipodial filaments into bundles. Efficient filopodia induction required mDia2 targeting to the membrane, likely through a scaffolding protein Abi1. Furthermore, mDia2 and Abi1 interacted through the N-terminal regulatory sequences of mDia2 and the SH3-containing Abi1 sequences. We propose that mDia2 plays an important role in formation of lamellipodia by nucleating and/or protecting from capping lamellipodial actin filaments, which subsequently exhibit high tendency to converge into filopodia.
| Cell motility is a cyclic process, with the protrusion of the leading edge followed by retraction of the rear. Protrusion is driven by polymerization of actin filaments, with the spatial organization of these filaments determining the shape of the protrusions. For example, the spike-like filopodia contain bundles of long actin filaments, whereas the sheet-like lamellipodia contain branched actin networks. In biochemical assays, two stimulators of actin polymerization, Arp2/3 complex and formins, induce branched or individual filaments, respectively. In cells, Arp2/3 complex and formins also appear to be implicated in the formation of lamellipodia and filopodia, respectively. However, when we investigated the role of mDia2 formin by functional approaches, we unexpectedly found that it is essential, not only for filopodia, but also for lamellipodia. Moreover, functions of mDia2 in lamellipodia and filopodia appeared intimately linked. We recorded behavior of cells by light microscopy and then used electron microscopy to study actin architecture in the same cells. We found that an activated form of mDia2 was first recruited to lamellipodia, where it induced many long, unbranched filaments, and from there, drove formation of filopodia through gradual convergence of these lamellipodial filaments into bundles. These data demonstrate a strong relationship between structurally different actin filament arrays and molecular machineries involved in their formation.
| Cell motility is a cyclic process consisting of protrusion of the leading edge followed by retraction of the rear. Actin filament polymerization provides a driving force for protrusion, whereas the shape and dynamics of protrusive organelles depend on the spatial organization of underlying filaments and activity of accessory molecules. Spatially restricted and temporally controlled actin filament nucleation is critical for generating membrane protrusions. Arp2/3 complex and formins are two major actin filament nucleators acting as convergent nodes of signaling pathways leading to initiation of actin-based motility (reviewed in[1,2]. Arp2/3 complex nucleates branched actin filaments [3,4]. Conversely, formins nucleate single linear filaments, binding to and protecting from capping their growing barbed ends [5–7]. Therefore, Arp2/3 complex and formins are thought to have nonoverlapping functions in cells in the formation of dendritic networks and linear bundles, respectively. These distinct roles have been clearly demonstrated in yeast [8], but appear to be conserved also in mammals. Thus, Arp2/3 complex is a key nucleator during lamellipodia extension [9–12] and endocytosis, where it functions downstream of WAVE [13] and N-WASP [14] Arp2/3 activators, respectively. Formins mDia1 and mDia2, instead, play a role in stress-fiber formation [15] and filopodia [16,17], respectively. Dictyostelium dDia2 is also both necessary and sufficient for filopodia extension [18].
The concept of functional separation of Arp2/3 complex and formins during leading-edge protrusion to generate dendritic networks in lamellipodia and parallel bundles in filopodia, respectively, is challenged by the observation that filopodia may arise by reorganization of the lamellipodial network [10,19,20]. In this process, termed the convergent elongation, Arp2/3-dependent nucleation was proposed to supply filaments for filopodial bundles. However, other studies point to a nonessential role of Arp2/3 complex in filopodia [12,21] favoring an alternative model whereby formin mDia2 is sufficient to generate a filopodial bundle not necessarily associated with lamellipodia [17,22]. Thus, the mechanisms of filopodia formation and contribution of different actin nucleators to the generation of protrusions in mammalian cells are far from being elucidated.
The domain organization of two related formins, mDia1 and mDia2, and functions of individual domains are well characterized [23,24]. The functional module consisting of FH1 and FH2 domains is responsible for nucleation, barbed-end binding, and anticapping protection of formins in vitro [5–7]. In cells, the FH1FH2 domain of mDia1 can travel on polymerizing barbed ends over significant distances until it stops at the cell membrane [25]. The N-terminal regulatory region upstream of the FH1FH2 module contains the GTPase-binding domain (GBD), the Diaphanous inhibitory domain (DID), the dimerization domain (DD), and the coiled coil region (CC), whereas the Diaphanous autoinhibitory domain (DAD) is present at the C-terminus of the molecule. Full-length Diaphanous-related formins are autoinhibited through intramolecular interaction between DID and DAD, which is thought to be relieved by specific small GTPases binding to GBD and displacing DAD from DID [26]. Disruption of this intramolecular interaction by deleting DAD or DID sequences generates constitutively active mutants of these formins [27].
In this study, aiming to characterize the mDia2-dependent mechanism of filopodia formation, we investigated a role of mDia2 in the leading-edge protrusion by loss-of-function and gain-of-function approaches followed by detailed analyses of the phenotypes using light and electron microscopy (EM) and a combination of both. Unexpectedly, we found that mDia2 was implicated in lamellipodia formation, suggesting that the Arp2/3 complex, despite being essential, was not sufficient for this function. Furthermore, the constitutively active mDia2 mutant efficiently induced filopodia, but these filopodia were formed, not from a focal spot away from lamellipodia, but by the gradual reorganization of lamellipodial filaments converging into filopodial bundles. We also found that specific targeting to the membrane was essential for mDia2 functions in protrusion. The targeting required a multifunctional scaffolding protein Abi1, which also participates in other actin-remodeling events, including regulation of Arp2/3-dependent nucleation [13]. Thus, two types of actin-based protrusions, lamellipodia and filopodia, despite containing structurally distinct actin filament arrays, display considerable dynamic relationship with each other and share mDia2 as a key molecular player.
Mouse melanoma B16F1 cells form broad lamellipodia interspersed by a varying number of filopodia (Figure 1A). During their lifetime, filopodial bundles might remain within lamellipodia and are often termed microspikes, or protrude beyond the leading edge, becoming bona fide filopodia [19]. We determined the intracellular localization of endogenous mDia2 by immunostaining, and found that it concentrated at filopodial tips, but also was distinctly enriched in lamellipodia (Figure 1A). Filopodial localization is consistent with previously shown targeting of activated mDia2 [16,17] and Dictyostelium formin dDia2 [18] to filopodial tips, but lamellipodial enrichment was unexpected, and suggested that mDia2 might function in both types of protrusions.
We investigated mDia2 functions in cells using the RNA interference (RNAi) approach. Transfection of B16F1 cells with mDia2 short interfering RNA (siRNA) significantly depleted endogenous mDia2, but not other proteins involved in actin-based protrusion (Figure 1B), such as the p34-Arc subunit of Arp2/3 complex, a filopodial actin-bundling protein fascin [28], or VASP, an actin-binding protein localizing to filopodia tips, lamellipodial edges, and focal adhesions [29]. Depletion of mDia2 caused a significant delay in cell spreading (Figure 1C), cell migration (Figure 1D), and a striking inhibition of both lamellipodia and filopodia, as compared to control, scrambled siRNA-treated cells (Figure 1E–1H; Videos S1 and S2). Quantitative analysis of phalloidin-stained cells (Figure 2B) revealed that less than 10% of the cell periphery in mDia2-depleted cells was occupied by actin-rich fringes morphologically resembling lamellipodia (see Materials and Methods for quantification criteria), as compared to approximately 50% in control cells. Furthermore, the presence of lamellipodia sharply declined with decreasing levels of mDia2 (Figure S1). Accordingly, the lamellipodial markers, p16-Arc (Arp2/3 subunit) (Figure 1H), capping protein, Abi1, or VASP (unpublished data) were no longer enriched at the cell edge in mDia2 siRNA-treated cells. Thus, control cells had 53.1 ± 11.1% (n = 11) of their perimeter stained with p16-Arc antibody, whereas mDia2 knockdown cells had only 4.4 ± 1.6% (n = 10) of Arp2/3-positive edges. Notably, the remaining lamellipodia in mDia2-depleted cells were abnormal morphologically, as they were very narrow in the direction perpendicular to the edge and often resembled small ruffles, whereas classical broad and flat lamellipodia were not observed. These lamellipodia were also abnormal kinetically, as they quickly switched to retraction after transient protrusion (Figure 1G), and their protrusion rate and persistence were significantly reduced (Figure 1E).
Some filopodia also remained after transfection with mDia2 siRNA, but they appeared less rigid than normal, frequently bending and protruding in a curvy path. Although fascin was present in these filopodia, the number of fascin-containing structures was significantly decreased (Figure 1H). No obvious changes in stress-fiber levels or organization were detected (Figure 2A). Depletion of mDia2 also caused accumulation of pigment granules, consistent with a role of mDia2 either in membrane trafficking [30,31] or transcription [32].
The effects of mDia2 siRNA on lamellipodia and filopodia were specific because they could be rescued by RNAi-resistant full-length green fluorescent protein (GFP)-mDia2 (FL-mDia2*) (Figures 1E, 2, and S2). Although GFP-FL-mDia2* localized mainly cytoplasmically, as reported earlier for wild-type FL-mDia2 [27], it was sometimes enriched in lamellipodia and at filopodial tips, especially in rescued cells (Figure 2A), apparently because of less competition from the endogenous untagged protein. As additional specificity tests, we attempted to rescue the mDia2 knockdown phenotype by expressing proteins with potentially redundant activities, a closely related formin GFP-mDia1 and GFP-VASP [33,34]. However, neither of them produced a significant rescue, suggesting nonredundant functions with mDia2. To test a possibility that inhibition of lamellipodia involved down-regulation of Rac1, we expressed constitutively active Rac1V12 in mDia2 knockdown cells. Although Rac1V12 increased lamellipodia and filopodia in control cells, it did not restore their amount in mDia2 siRNA-treated cells. Interestingly, FH1FH2-mDia2 also did not rescue the phenotype (see below).
Platinum replica EM was used to reveal specific defects caused by mDia2 depletion in the architecture of the protrusive organelles (Figure 3). Normal lamellipodia in B16F1 cells are filled with branched actin network consisting of combination of long and short filaments, whereas filopodia contain tight bundles of long actin filaments (Figure 3A and 3B) [4,19]. In mDia2-knockdown cells, actin filament organization at the cell periphery was severely disrupted (Figure 3C–3G). Consistent with light microscopic data, only a small fraction (8.7 ± 8.9%, n = 5) of cell perimeter in these cells contained dendritic actin network characteristic for lamellipodia. The rare remaining lamellipodia contained patches of dense network poorly connected to the rest of the cytoskeleton and formed by short branching actin filaments (Figure 3F), while long filaments were not obvious there. This morphology suggests that Arp2/3-dependent actin nucleation likely remains functional [4], but anticapping protection and/or linear filament elongation was compromised [33]. Remaining filopodia, instead of a regular bundle of long, parallel filaments, contained less-uniform actin arrays (Figure 3G).
Collectively, these results reveal a novel role of formin mDia2 in the formation of lamellipodia, likely involving facilitated formation of long filaments, and corroborate its function in filopodial protrusion [16–18] by an alternative approach.
In addition to loss-of-function analysis by RNAi, we also tested mDia2 functions by a gain-of-function approach using a constitutively active mDia2 mutant (ΔGBD-mDia2) lacking GBD and a part of DID [15,16]. We reasoned that any unusual features induced by this mutant would point to specific molecular mechanisms of mDia2 involvement in cell protrusion.
In B16F1 cells, GFP-ΔGBD-mDia2 localized to a subset of lamellipodia (Figures 4A and 5) and to filopodial tips (Figure 6) with relatively low cytoplasmic fluorescence. This localization is consistent with the functional importance of the endogenous protein for these protrusions. Compared to endogenous mDia2, ΔGBD-mDia2 had narrower distribution at the very edge of lamellipodia. ΔGBD-mDia2–positive lamellipodia contained normal lamellipodial components, Arp2/3 complex, capping protein, VASP, and Abi1 (Figure 4A). However, at edges with high levels of ΔGBD-mDia2, capping protein was not detected, possibly reflecting anticapping activity of formins [7]. Conversely, the pattern of ΔGBD-mDia2 at the leading edge frequently correlated with that of Abi1. However, Abi1 only partially colocalized with the endogenous mDia2, which formed a broader band at the leading edge (Figure S3). An unusual feature of ΔGBD-mDia2–positive lamellipodia was the presence of fascin (Figure 4A), which is normally not detected in B16F1 lamellipodia [19]. Analysis of dynamic behavior of ΔGBD-positive lamellipodia revealed that their protrusion is remarkably persistent, whereas the rate of protrusion is slower than that of control cells (Figure 4B).
Correlative EM analysis ΔGBD-mDia2–positive lamellipodia (Figure 5) revealed that they had unusual abundance of long, parallel, unbranched actin filaments (Figure 5D). Quantitative analysis of filament orientation revealed a narrow distribution of angles in ΔGBD-mDia2–positive lamellipodia compared to very broad distribution in control lamellipodia (Figure 5E–5G). Enrichment in long filaments after expression of constitutively active mDia2 complements the siRNA data showing apparent filament shortening after mDia2 depletion. These long filaments frequently displayed apparently free (not engaged in branch formation) proximal ends (Figure S4), which are very rare in normal lamellipodia [35], suggesting that active mDia2 might nucleate these linear filaments within lamellipodia. However, ΔGBD-mDia2–positive lamellipodia also contained some branched filaments (Figure S4) and Arp2/3 complex (Figure 4A), suggesting that the lamellipodial network represents a mixture of linear and dendritic actin filament arrays. The atypical abundance of long, parallel filaments induced by constitutively active mDia2 might cause the peculiar recruitment of fascin.
Thus, the gain-of-function approach showed that the constitutively active mDia2 mutant was recruited to lamellipodia and induced structural reorganization of the lamellipodial actin network by promoting formation of long filaments, as well as increased the persistence of lamellipodia protrusion. These results corroborate the findings of mDia2 knockdown experiments, which suggested mDia2 function in lamellipodia.
Our findings of mDia2 involvement in lamellipodia formation apparently contrast with previous reports showing a role of mDia2 [16,17] and Dictyostelium dDia2 [18] in filopodia. However, we found that the lamellipodial localization of GFP-ΔGBD-mDia2 was transient and especially evident at early stages of transfection. At later times, ΔGBD-mDia2 induced abundant filopodia with ΔGBD-mDia2 localizing at their tips (Figure 6; Video S3), consistent with other reports [16–18]. We noticed, however, that these filopodia frequently had an unusual club-like shape with a thick actin-rich distal domain on a thin stalk (Figure 6A, 6C, and 6D). Nonetheless, they contained conventional filopodial markers: fascin, VASP, and myosin X [36], albeit with a somewhat altered localization (Figure 6A). Fascin was concentrated predominantly in thick distal parts of filopodia. VASP colocalized with ΔGBD-mDia2 at filopodial tips in some cells, but high levels of ΔGBD-mDia2 seemed to displace VASP from tips to the filopodial shafts, possibly reflecting VASP's bundling ability [37,38]. Abi1, which usually localizes to lamellipodial edges and filopodial tips [39], was faint at the filopodial tips enriched in ΔGBD-mDia2.
Filopodia induced by ΔGBD-mDia2 frequently were very long, flexible, not attached to the substratum, and occasionally appeared on the dorsal cell surface (Figure S5), resembling filopodia induced by a small GTPase Rif through mDia2 [17]. During protrusion, ΔGBD-mDia2 remained at filopodial tips (Figure 6D), consistent with the formins' ability to ride on elongating barbed ends [25]. The thick termini of filopodia moved forward, remaining approximately of the same length, suggesting actin filament treadmilling within these domains (Figure 6D). Consistent with this idea, GFP-actin speckles moved retrogradely in ΔGBD-mDia2–induced filopodia and disappeared upon or after exiting the thick terminal domains (Figure 6F–6H; Video S4). The protrusion rate of ΔGBD-mDia2–positive filopodia was slightly lower than that of control filopodia (Figure 6E).
By EM analysis, ΔGBD-mDia2–mediated filopodia contained actin bundles, which looked normal in their distal domains, but displayed informative differences in the proximal regions (Figure 7). In normal filopodia (Figure 3A), filaments in their roots commonly splay apart, reflecting the convergent elongation process of filopodia formation, and frequently terminate at branch points in the surrounding network [19]. Conversely, many ΔGBD-mDia2–induced filopodia displayed tapered bundles at the rear, with numerous unbound proximal ends (Figure 7A and 7B), which are predicted to be pointed ends based on their orientation. Since actin network is assembled predominantly at the leading edge and then treadmills backwards, undergoing little rearrangement except depolymerization, the structural organization of the network to some extent portrays its immediate history [40]. Therefore, tapered roots of ΔGBD-mDia2–induced bundles suggest that these filaments were nucleated by ΔGBD-mDia2 and/or they were nucleated by Arp2/3 complex, but then debranched and depolymerized from the pointed ends. Notably, some ΔGBD-mDia2–induced filopodia contained bundles with splayed roots (Figure 7A and 7C), and some filaments in these filopodia originated from a branch point (Figure 7C), suggesting that Arp2/3-nucleated filaments might contribute to bundle formation.
Together, these results confirm the previous observations that mDia2 plays a role in filopodia generation, and localizes to filopodial tips. We extend prior observations by showing that although filopodia induced by active mDia2 share many features with normal filopodia, they also display an abnormal abundance of prematurely terminated filaments in the proximal parts of filopodial bundles.
We next asked whether mDia2 roles in lamellipodia and filopodia are related. Kinetic analysis of filopodia initiation showed that virtually all ΔGBD-mDia2–induced filopodia formed from ΔGBD-mDia2–positive lamellipodia (Figure 8; Video S5). In this pathway, a line of ΔGBD-mDia2 fluorescence at the lamellipodial leading edge gradually condensed into dots at the tips of newly formed filopodia (Figure 8A), similar to VASP behavior in naturally occurring filopodia [19]. Full-length GFP-mDia2 displayed similar behavior (Video S6), although high cytoplasmic fluorescence of this autoinhibited protein significantly decreased contrast of the leading edge signal. Newly formed ΔGBD-mDia2–induced filopodia might subsequently fuse with each other, generating larger filopodia and gradually acquiring a club-like shape (Figure S5). Some filopodia subsequently translocated to the dorsal surface of lamella; the majority of dorsal protrusions were formed by this pathway (Figure S5).
To understand the mechanism of lamellipodia-to-filopodia transition, we performed correlative EM for cells with known live history (Figure 8B–8J; Video S7–S9). Figure 8 shows two examples of filopodia formed during live imaging. One filopodium began to form approximately 30 s before fixation, as judged from condensation of GFP fluorescence at the leading edge (Figure 8D and 8E). In the corresponding EM image, this filopodium consisted of long filaments originating from a broad area at the base and coming together at the tip (Figure 8H). Projection of the life history onto EM image revealed that filaments converged in parallel with condensation of GFP fluorescence (see Figure 8 legend for detail), similar to normal filopodia [19].
The second example (Figure 8F and 8G) shows a slightly older filopodium formed from a line of GFP fluorescence visible in the first frame of the movie. By approximately 50–60 s before fixation, the line gradually condensed into a dot, suggesting that the filopodial bundle was formed by that time. Subsequently, the dot moved approximately 1 μm as the filopodium protruded. In the corresponding EM image (Figure 8I and 8J), the filopodium contained a spindle-shaped actin bundle that tapered both toward the front (perhaps because of tighter filament bundling) and toward the rear due to gradual termination of filaments. The projection of the fluorescence history of this filopodium onto the EM image showed a sparse actin network proximally from the position of the leading edge at time 0:00 (yellow line in Figure 8I). Such morphology might be explained by nucleation of a filament bundle by ΔGBD-mDia2 at a local spot in actin-poor region, followed by elongation of the bundle, as proposed [22]. However, in this case, we would see an emergence and gradual advance of a dot of ΔGBD-mDia2 fluorescence in the time-lapse movie instead of a fluorescent line converging into a dot (Figure 8F). As we showed above (see Figure 5), the linear distribution of ΔGBD-mDia2, as it appears at the 0:00 time point for this filopodium (Figure 8F and 8I), always associates with dense lamellipodial network of long, aligned filaments. Thus, the tapered shape of the filopodium root among sparse actin network is more consistent with the alternative possibility that actin depolymerization eliminated much actin from the filopodium rear in parallel with the filopodium protrusion at the front.
Correlative analysis of six cells imaged live for 90–120 s showed that all nascent filopodia (31 total) emerged by condensation of linear lamellipodial fluorescence. At EM level, 14 out of these 31 filopodia were similar in structure to the first example in Figure 8; five were similar to the second example; nine displayed an intermediate organization, having a fully formed distal bundle with well-preserved splayed filaments in the root; they looked similar to the filopodium marked by an asterisk in Figure 7; and one filopodium contained converging filaments like in the first example, but showed significant actin depletion at the root, like in the second example.
These data, collectively, indicate that induction of filopodia by active mDia2 temporally and mechanistically follows the mDia2 appearance in lamellipodia. Long, unbranched actin filaments induced by ΔGBD-mDia2 in the lamellipodial network seem to play an important role in this process by expressing a high tendency to converge into bundles, and leading to filopodia induction in association with pre-existing lamellipodia.
FH1FH2 constructs of many formins are sufficient to nucleate actin filaments and bind barbed ends [24], whereas sequences upstream of this module are required for targeting of some formins [41,42]. Enrichment of ΔGBD-mDia2 at the leading edge and filopodial tips (Figures 4 and 6) may be mediated simply by binding to barbed ends through the FH2 domain, or involve an additional targeting mechanism. We tested this idea using FH1FH2-mDia2 mutant (Figure 9).
GFP-FH1FH2-mDia2 was not able to rescue the mDia2 siRNA phenotype (Figure 2) and, when expressed in wild-type B16F1 cells, displayed different behavior as compared to ΔGBD-mDia2 (Figure 9; Video S10): (1) it induced fewer and shorter, evenly thin, but not club-shaped, filopodia; (2) it localized mostly in the cytoplasm and poorly at filopodial tips; and (3), in contrast to ΔGBD-mDia2, GFP-FH1FH2-mDia2 was easily removed by detergent extraction (Figure 9B). Truncation of the FH2 domain from the C-terminus or its complete removal from FH1FH2-mDia2 abrogated filopodia-inducing ability and localization to filopodia tips (Figure S6). Thus, although mDia2 nucleating and barbed end-binding module FH1FH2 is required for and capable of tip localization and weak filopodia induction, additional motifs of ΔGBD-mDia2 are needed for robust mDia2 targeting to the leading edge and efficient induction of filopodia. This is strengthened by the observation that an analogous mutant of mDia1 localized differently (Figure S6). Similar to previous reports [43,44], active mDia1 displayed strong cytoplasmic fluorescence with weak targeting to the edges.
Searching for a molecular mechanism of mDia2 targeting, we concentrated on proteins displaying robust localization to the leading edge of lamellipodia. Ena/VASP proteins were likely not involved, as ΔGBD-mDia2 localized properly in Ena/VASP-deficient MVD7 cells (Figure S6). Next, we considered the WAVE/Abi1/Nap1/PIR121/HSP300 (WANP) complex [11,45], a major activator of Arp2/3 complex in lamellipodia, because Abi1 [39] and WAVEs [46] localize robustly to the lamellipodial leading edge. Depletion of any member of WANP complex causes degradation of its other components [11,47]. Thus, we tested whether mDia2 would associates to any of the members of the WANP complex.
Ectopically coexpressed GFP-FL-mDia2 and Abi1 (Figure 10A), but not WAVE (Figure 10B), Nap1, or PIR121 (unpublished data), readily coimmunoprecipitated. Furthermore, recombinantly produced full-length Abi1 associated in vitro with FL-mDia2 (Figure 10C). Finally, a C-terminal Abi1 fragment containing the SH3 domain was able to associate in in vitro binding assays with FL-mDia2, ΔGBD-mDia2, and an N-terminal fragment of mDia2 containing residues up to the beginning of the FH1 domain, but not with FH1FH2-mDia2 or FH1-mDia2 (Figure 10D and 10E). These data map the Abi1-interacting region on mDia2 to amino acids 258–518 within the N-terminal regulatory region (part of DID + DD + CC). Lack of interaction with FH1FH2-mDia2 and FH1-mDia2 argues against a possibility of nonspecific interaction between the proline-rich FH1 domain of mDia2 and SH3 of Abi1.
To assess the physiological relevance of this interaction, we used a previously characterized stable line of Abi1-RNAi–interfered (Abi1KD) HeLa cells with approximately 90% depletion of Abi1 (Figure S7), which fail to form lamellipodia in response to a variety of stimuli, but can spread and form filopodia [11]. We first examined filopodia in Abi1KD cells quantitatively and by EM (Figures 11 and S7). Control HeLa cells formed a mixture of filopodia and small lamellipodia or ruffles, whereas Abi1KD cells lacked ruffles and displayed long filopodia (Figures 11A and S7). Since filopodial bundles are normally partially embedded into a lamellipodial actin network, inhibition of lamellipodia likely exposed these internal parts, leading to apparent elongation of filopodia. Despite having longer filopodia, Abi1KD cells had fewer filopodia per cell than control cells (Figure 11A and 11D), indicating that Abi1 and/or its interacting partners, such as other members of WANP complex, may play a role in filopodia formation. Interestingly, despite almost complete absence of a lamellipodial dendritic network, EM analysis revealed multiple branched filaments in the roots of filopodia in Abi1KD cells (Figure S7), suggesting a preferential formation of filopodia at the sites with residual activity of Arp2/3 complex.
We tested next whether the human ortholog of mDia2 (DIAPH3) plays a role in filopodia formation in control and Abi1KD HeLa cells using both loss-of-function and gain-of-function approaches. Similar to B16F1 cells, DIAPH3 siRNA significantly depleted DIAPH3 expression and impaired cell spreading and protrusive activity in control and Abi1KD HeLa cells (Figure 11). DIAPH3 depletion decreased filopodia length and number in Abi1KD HeLa cells, and inhibited ruffles and filopodia in control HeLa cells, confirming in the human system that DIAPH3 is an important part of the mechanism of lamellipodia and filopodia formation.
Similar to B16F1 cells, expression of ΔGBD-mDia2 in HeLa cells induced long, robust filopodia (Figure 12A and 12C). Interestingly, the number of filopodia decreased after ΔGBD-mDia2 expression (Figure 11B), which might occur because of extensive fusion of neighboring filopodia, a well-known feature of filopodia dynamics [40]. In contrast, Abi1KD cells formed slender and significantly shorter filopodia in the same conditions (Figure 11A and 11C), whereas their number slightly, but significantly, decreased compared to ΔGBD-mDia2–expressing control cells (Figure 12B). Furthermore, ΔGBD-mDia2 in Abi1KD cells localized cytoplasmically, with only faint signal at the filopodial tips, which was more sensitive to detergent extraction than in control HeLa cells (Figure 12D). Thus, the ΔGBD-mDia2 in Abi1KD cells, similar to FH1FH2-mDia2 in B16F1 cells, was deficient in robust targeting to the edge and efficient induction of filopodia, suggesting a role of Abi1 in ΔGBD-mDia2 targeting.
In this study, by combination of loss-of-function (RNAi) and gain-of-function (constitutively active mutant) approaches, we investigated a role of mDia2 formin in the actin-based protrusion in motile cells. Although our original goal was to characterize the mechanism of mDia2-dependent filopodia formation, we unexpectedly found that mDia2 is important for lamellipodial protrusion and induces filopodia in association with lamellipodia. Based on our data, we propose a model suggesting that mDia2 is recruited to the lamellipodial leading edge in an Abi1-dependent manner, where it nucleates new filaments and/or maintains filament elongation by protecting barbed ends from capping. In the course of elongation, lamellipodial filaments converge and become bundled, generating filopodia.
We found that mDia2 can be recruited to the leading edge by at least two mechanisms, one of which relies on binding of the mDia2 FH1FH2 module to barbed ends and another on N-terminal mDia2 sequences interacting with Abi1. The FH1FH2 module contributes to localization to filopodial tips by binding to and riding on elongating barbed ends, which accumulate at the membrane after encountering the boundary conditions [25]. However, FH1FH2-mDia2 was not sufficient to fully restore formin's functions in mDia2-interferred cells and to efficiently induce filopodia, likely reflecting inferior membrane targeting of FH1FH2-mDia2 as compared to ΔGBD-mDia2 or full-length mDia2. In the second mechanism, binding of N-terminal regulatory sequences of mDia2 to Abi1 provides a potential additional surface for mDia2 proper localization to leading edges. This is also consistent with the observation that similar regions were previously shown to control targeting of other formins [41,42]. It is also possible that Abi1 not only recruits mDia2 to the membrane, but also participates in its regulation, but we currently do not have data to evaluate this idea.
The majority of Abi1 has been shown to be engaged in complex with WAVE, Nap1, and PIR121 [11,45,48]. However, we did not detect interaction of mDia2 with WAVE, Nap1, or PIR121, indicating that mDia2 and Abi1 may form a distinct macromolecular signaling unit with respect to the WANP complex. Accordingly, only a minor fraction of Abi1 interacted with mDia2 in coimmunoprecipitation experiments. The ability of Abi1 to act as a scaffolding molecule driving the formation of additional signaling units, such as those containing N-WASP [14] or Eps8 [49,50], has been previously observed. The Abi1-mDia2 assembly adds to the versatility of Abi1, which by entering distinct complexes is capable of controlling diverse processes and/or coordinating related activities, such as the formation of lamellipodia and filopodia protrusions reported here. The precise regulatory events governing the relationship between these units remain to be defined and represent one of the challenges for future investigation.
Specific targeting of proteins is a common way to confine the protein activity to a certain area. Therefore, mDia2 targeting to lamellipodia and filopodia points to an idea that mDia2 may function there. Although mDia2′s role in filopodia is expected [16,17], the lamellipodial function contrasts with the current belief that formins and Arp2/3 complex are responsible for different actin structures. Two complementary sets of data in our study suggest that mDia2 is an important player in lamellipodial protrusion. First, mDia2 inhibition by siRNA severely impaired lamellipodia and reduced long filaments in the remaining lamellipodial network. Second, constitutively active ΔGBD-mDia2, conversely, induced numerous long filaments in lamellipodia. Similar complementarity was observed regarding the persistence of lamellipodia protrusion, which decreased after mDia2 depletion and increased after expression of the constitutively active mutant. The lamellipodial function seems specific for mDia2; other formins, which were supposedly present in knockdown cells, were not sufficient. Moreover, overexpression of a related formin, mDia1, did not rescue lamellipodia formation. Interestingly, 3T3 cells lacking mDia2 [15] still make lamellipodia. We found, however, that 3T3 cells express another related formin, mDia3, which is absent in B16F1 cells (unpublished data), raising a possibility that mDia3 may be partially redundant with mDia2.
Biochemical activities of mDia2 suggest that it may contribute to lamellipodia formation through nucleation and/or anticapping protection of actin filaments (Figure 13A). Our ΔGBD-mDia2 data are consistent with both possibilities. Thus, high frequency of linear filaments with unlinked “pointed” ends, and fast depolymerization of actin filaments from the rear (see Figure 8) are consistent with formin, but not with Arp2/3-mediated nucleation. The anticapping activity of mDia2 is also employed during lamellipodia formation. Indeed, the dissociation of mDia2 from barbed ends is very slow, at least in vitro [51], suggesting that it would protect the nucleated filaments from capping for long time. In addition to this consideration, we found that (1) capping protein was diminished in ΔGBD-mDia2–rich lamellipodial regions; (2) ΔGBD-mDia2 enrichment at the leading edge correlated with the presence of long actin filaments in EM images; and (3) mDia2-depleted lamellipodia contained very short branched filaments. The latter observation also suggests that mDia2 may protect from capping branched filaments, apparently nucleated by Arp2/3 complex. The ability of ΔGBD-mDia2 to displace VASP from barbed ends (Figures 6A and S6) is also consistent with this idea. Notably, long unbranched filaments may also form in the course of debranching of Arp2/3-nucleated and mDia2-protected filaments without nucleation by ΔGBD-mDia2. Importantly, lamellipodial filaments nucleated and/or protected from capping by mDia2 may be responsible for the presence of long filaments [52], and two kinetically distinct actin subpopulations in lamellipodia [53] noted earlier.
It is unclear why mDia2 has such a significant role in lamellipodial protrusion, considering the well-established involvement of Arp2/3 complex in this process [1]. One possibility is that mDia2-nucleated filaments may help to initiate lamellipodia by nucleating “mother” filaments, which are required to Arp2/3-dependent nucleation. Additionally, mDia2 may be a better “device,” compared to VASP, for example, to maintain fast and processive barbed-end growth and thus to ensure persistent protrusion. Long filaments induced by mDia2 in lamellipodia may also provide better connection of lamellipodia to the rest of the cytoskeleton. In any case, our results suggest that two different actin nucleators, Arp2/3 complex and mDia2, jointly contribute to generation of lamellipodia.
A role of mDia2 in filopodia formation in mammalian cells [16,17] and of dDia2 in Dictyostelium was reported [18], but the mechanism of formin-mediated filopodia induction has not been fully investigated. The previous model suggested that mDia2 initiates filopodia from a focal spot not necessarily associated with lamellipodia [17,22], thus putting this mechanism at an apparent conflict with the convergent elongation model [19]. In this study, we provide evidence to solve this contradiction. By investigating the spatiokinetic mechanism underlying filopodia initiation by active mDia2, we found that, in our experimental system, lamellipodia are a highly preferred site for filopodia initiation, although subsequently, filopodia may lose association with lamellipodia, for example, by moving to the dorsal surface. However, we cannot exclude a possibility that in other experimental systems or conditions [12,54], formins may nucleate actin bundles from focal spots without association with lamellipodia or other dendritic arrays.
The overall process of filopodia initiation by mDia2 was very similar to that described for naturally emerging filopodia [19] and involved gradual convergence of lamellipodial filaments (Figure 13B). These results show that the convergent elongation mechanism is applicable to formin-induced filaments. Importantly, FL-mDia2 displayed similar convergence of lamellipodial fluorescence into dots, as ΔGBD-mDia2, suggesting that convergence behavior is not entirely due to excessive activity of ΔGBD-mDia2. Initiation of filopodia by ΔGBD-mDia2 was very efficient, whereas the lamellipodia from which the filopodia arose were quite transient. This can be explained by the favorable combination of biochemical activities in mDia2, which may drive the efficient transformation of lamellipodia into filopodia, not only by nucleating and elongating filaments, but also by cross-linking them [55] and thus promoting the initiation of filament convergence. A downside of the excessive activity of ΔGBD-mDia2 is that filopodia acquired an abnormal club-like shape, in which the majority of filaments did not extend all the way to the cell body. Based on the known mechanism of formin-mediated nucleation and our data obtained by correlative light microscopy and EM (Figure 8), and by observing GFP-actin dynamics (Figure 6), we interpret this phenotype as fast depolymerization of mDia2-nucleated filaments from unprotected pointed ends.
Because of formation of atypical club-like filopodia, we suppose that ΔGBD-mDia2 does not fully recapitulate the physiological pathway of filopodia formation. Based on the structural data, we previously proposed that Arp2/3 complex nucleates filaments for filopodial bundles [19]. We have confirmed this idea by a functional siRNA approach showing that Arp2/3 depletion inhibits filopodia (F. Korobova and T. Svitkina, unpublished data. These two sets of data together suggest that Arp2/3 complex and mDia2 may jointly contribute to filopodia formation. In such case, one would expect that the balanced activities of these two nucleators are important for the normal filopodial shape and functions, whereas any imbalance will likely lead to aberrant filopodia. Club-like filopodia induced by constitutive activation of mDia2 may serve as an illustration for this contention. Abnormal filopodia were also formed after depletion of mDia2. However, the relative roles of formins and Arp2/3 complex in filopodia await further investigation.
In summary, we have shown that mDia2 plays a role in formation of both lamellipodia and filopodia. However, these roles are not separate, but related to each other; mDia2 first participates in lamellipodia formation and then induces filopodia from lamellipodia. A most likely scenario is that mDia2-nucleated actin filaments and/or filaments nucleated by Arp2/3 complex and protected by mDia2 are initially dispersed in the lamellipodial network, and subsequently, they elongate persistently and gradually converge, eventually segregating themselves into filopodial bundles. Although regulation of actin dynamics is the most straightforward way to explain how mDia2 may function in protrusion, we cannot exclude a possibility that other mDia2 activities, such as regulation of microtubule dynamics [56], membrane trafficking [30,31], surface blebbing [57], or transcription [32], also contribute to generation of observed phenotypes.
Plasmids. Full length mDia2 in a Bluescript cloning vector (Stratagene) and pEFm-EGFP-ΔGBD-mDia2 (amino acids [aa] 258–1,171) [15] were gifts of A. Alberts (Van Andel Research Institute). The mRFP1-ΔGBD-mDia2 construct was prepared by substituting EGFP with mRFP1 [58], which is a gift from R. Tsien (University of California San Diego). Full-length mDia2 was cloned into pEGFP-C1 vector (Clontech). DNA fragments for FH1 (aa 519–600), trFH1FH2 (aa 519–909), FH1FH2 (aa 519-1007), and NT-mDia2 (aa 1–550) were amplified by PCR, cloned into pEGFP-C1 vector, and confirmed by sequencing. EGFP-VASP was obtained from F. Gertler (Massachusetts Institute of Technology [MIT]) and cloned into pECFP-C1 vector (Clontech); myosin X [36] from R. Cheney (University of North Carolina-Chapel Hill); and full length mDia1 and ΔN2-mDia1 [44] from S. Narumiya (Kyoto University).
Antibodies. The following polyclonal antibodies were obtained as gifts: mDia2 from A. Alberts [15] and H. Higgs (Dartmouth Medical School); CPβ2 C terminus (R26) from D. Schafer (University of Virginia); Arp3 from T. Uruno (Holland Laboratory); cofilin from J. Condeelis (Albert Einstein College of Medicine); and VASP from F. Gertler (MIT). Antibody against Abi1 was previously described [59]. The following antibodies were from commercial sources: fascin (DAKO), α-tubulin (Sigma), cortactin 411F (Upstate), and secondary antibodies (Molecular Probes or Jackson Laboratories). All other reagents were from Sigma unless indicated otherwise.
siRNA. siRNA for mDia2 coding region (5′-ataagagagcagtatttcaaa-3′) and control siRNAs (5′-aagaaatagggaaggtggaac-3′ and 5′-aaatttacaggacttcagtca-3′) were obtained from Dharmacon, and siRNA for DIAPH3 coding region (5′-aaccttcggatttaaccttag-3′) was from Ambion. siRNAs were Cy3-labeled using a Silencer siRNA labeling kit (Ambion) and used at 20 nM concentration. The efficiency of siRNA transfection was approximately 90%. Effects of siRNA were analyzed 2 to 3 d post-transfection. For rescue experiments, silent mutations were introduced into siRNA-targeted region of GFP-mDia2 (5′-ataagagaAcagtaCttcaaa-3′) using the Quikchange site-directed mutagenesis kit II (Stratagene).
B16F1 mouse melanoma cells [19], HeLa cells, and Abi1KD HeLa cells [11] were cultured as described. Transient transfection of DNA and siRNA was performed using Fugene6 or Lipofectamine 2000 (Invitrogen), respectively. Lipofectamine 2000 was used for cotransfection of siRNA and DNA. Light microscopy was performed using Eclipse TE2000-U inverted microscope (Nikon) equipped with Planapo 100 × 1.3 NA or 20 × 0.75 objectives and Cascade 512B CCD camera (Photometrics) driven by Metamorph imaging software (Molecular Devices). For live-cell imaging, cells were transferred into phenol red–free L-15 or DMEM medium (Gibco) supplemented with 10% FBS and kept on the microscope stage at 35 °C during observation.
Samples for platinum replica EM were processed as described [60] and analyzed using JEOL 1200EX transmission electron microscope operated at 120 kV. Cells expressing ΔGBD-mDia2 for EM analysis were identified by correlative EM or isolated by FACS. Immunostaining was performed after cell extraction for 5 min at room temperature with 1% Triton X-100 in PEM buffer (100 mM PIPES-KOH [pH 6.9] 1 mM MgCl2, 1 mM EGTA) containing 2% polyetheleneglycol (MW 35,000) and 2 μM phalloidin, followed by fixation with 0.2% glutaraldehyde and quenching with NaBH4. Texas red phalloidin (0.033 μM; Molecular Probes) was used for actin staining. Fascin staining was performed after methanol fixation of extracted cells. For staining with VASP or mDia2 antibody, cells were extracted/fixed by a mixture of 0.25% glutaraldehyde and 0.5% Triton X-100 in PEM buffer for 20 min. To evaluate cytoskeletal association of mDia2 constructs, cells were first imaged live, then culture medium was replaced by the extraction solution, the same as used for immunostaining, and another image was acquired approximately 2 min later.
All morphometric measurements were done using MetaMorph Imaging software (Molecular Devices) unless stated otherwise. Graphs and statistical analysis were done using SigmaPlot software. Statistical significance was determined by the Student t-test.
Quantification of lamellipodia and filopodia were performed as described [20]. Briefly, MetaMorph line function was used to trace and measure the whole-cell perimeter and the cell perimeter with adjacent lamellipodia (an actin-rich fringe with fluorescence intensity gradually declining with the distance from the edge) on fixed phalloidin-stained cells. The fraction of the cell perimeter occupied by lamellipodia was used as a parameter for quantification. For quantification purposes, filopodia were defined as actin-rich finger-like protrusions crossing the cell edge and having fluorescence intensity at least 1.2-fold above background. Quantification was performed using a blind experimental procedure by an uninformed observer on coded samples. Control and experimentally treated cells in mixed populations were measured using phalloidin channel only. After numbers were assigned to individual cells, the phalloidin images were combined with other channels showing siRNA or a rescue construct, the identity of samples was decoded, and numbers were entered into respective columns of a spread sheet for statistical analysis.
For spreading assays, the projected cell area was measured 2 h after plating. For cell migration analysis, cells transfected with mDia2 siRNA (Cy3 labeled) or control siRNA (nonlabeled) were co-cultured in 35-mm dishes and imaged 36–48 h post-transfection (4 h after plating). Phase contrast time-lapse sequences were acquired with 5-min intervals for 6 h using a 4× objective. Fluorescence images to identify siRNA-transfected cells were acquired immediately before and after each movie. Cell positions were recorded every ten frames using Track Object tool in MetaMorph. An average instantaneous rate was calculated for each cell, and then the mean speed of cell migration was determined for each group of cells. To analyze the lamellipodial dynamics, time-lapse sequences were acquired with 3-s intervals for 10 min using a 100× objective. Kymographs were generated along straight lines drawn in the direction of protrusion. Rates of lamellipodia protrusion were determined based on slopes produced by advancing leading edges. Persistence corresponds to time intervals during which individual protrusions occurred. The relative mDia2 levels in control and mDia2 siRNA-treated cells were determined by immunostaining after fixation with paraformaldehyde and permeabilization with Triton X-100. Integrated fluorescence intensity of mDia2 staining for each cell after background subtraction was plotted against the fraction of the cell edge occupied by lamellipodia.
To determine the degree of mutual orientation of actin filaments in lamellipodia of control and ΔGBD-mDia2-expressing cells, square EM images of the lamellipodial network ranging from 0.13 to 0.53 μm2 were thresholded using Adobe Photoshop to maximally highlight the linear features. Matlab software was used to generate two-dimensional Fast Fourier Transform from these images, collect radial intensity line scans, plot intensity as a function of angle, and fit a Gaussian curve to the major peak. Standard deviation of the Gaussian fit was used as a parameter of the orientational order. Matlab code for this analysis was provided by J. Winer, Q. Wen, and P. Janmey (University of Pennsylvania).
For immunoblotting, cells were lysed in buffer containing 10 mM Tris (pH 7.5), 150 mM NaCl, 1% Triton X-100, 10% glycerol, and a protease inhibitor tablet (Roche). Protein concentration of the lysates was determined using Bio-Rad protein assay kit (Bio-Rad). Proteins were separated by SDS-PAGE (7.5%–10% polyacrylamide). Tubulin was used as loading control. Immunoblots were developed using ECF Western blotting kit (Amersham).
Coimmunoprecipitation and in vitro binding assays were performed as described [59]. For coimmunoprecipitation, total cellular lysates (1 mg) of 293T cells cotransfected with GFP-FL-mDia2 and Abi1, alone or in combination, were immunoprecipitated with an Abi1 antibody. For GST pull-down assay, total cellular lysates (1 mg) of 293T cells transfected with GFP-mDia2 were incubated with 1.5 μM of immobilized GST-Abi1, GST-Abi1-SH3 (aa 330–480), or GST, as a control. Lysates (20 μg) and bound proteins in both cases were resolved by SDS-PAGE and immunoblotted. GFP antibody was used to detect GFP-mDia2 constructs.
|
10.1371/journal.pbio.2003714 | Identification of a noncanonical function for ribose-5-phosphate isomerase A promotes colorectal cancer formation by stabilizing and activating β-catenin via a novel C-terminal domain | Altered metabolism is one of the hallmarks of cancers. Deregulation of ribose-5-phosphate isomerase A (RPIA) in the pentose phosphate pathway (PPP) is known to promote tumorigenesis in liver, lung, and breast tissues. Yet, the molecular mechanism of RPIA-mediated colorectal cancer (CRC) is unknown. Our study demonstrates a noncanonical function of RPIA in CRC. Data from the mRNAs of 80 patients’ CRC tissues and paired nontumor tissues and protein levels, as well as a CRC tissue array, indicate RPIA is significantly elevated in CRC. RPIA modulates cell proliferation and oncogenicity via activation of β-catenin in colon cancer cell lines. Unlike its role in PPP in which RPIA functions within the cytosol, RPIA enters the nucleus to form a complex with the adenomatous polyposis coli (APC) and β-catenin. This association protects β-catenin by preventing its phosphorylation, ubiquitination, and subsequent degradation. The C-terminus of RPIA (amino acids 290 to 311), a region distinct from its enzymatic domain, is necessary for RPIA-mediated tumorigenesis. Consistent with results in vitro, RPIA increases the expression of β-catenin and its target genes, and induces tumorigenesis in gut-specific promotor-carrying RPIA transgenic zebrafish. Together, we demonstrate a novel function of RPIA in CRC formation in which RPIA enters the nucleus and stabilizes β-catenin activity and suggests that RPIA might be a biomarker for targeted therapy and prognosis.
| The pentose phosphate pathway generates NADPH, pentose, and ribose-5-phosphate by RPIA for nucleotide synthesis. Deregulation of RPIA is known to promote tumorigenesis in liver, lung, and breast tissues; however, the molecular mechanism of RPIA-mediated CRC is unknown. Here, we demonstrate a role of RPIA in CRC formation distinct from its role in these other tissues. We showed that RPIA is significantly elevated in CRC. RPIA increased cell proliferation and oncogenicity via activation of β-catenin, with RPIA entering the nucleus to form a complex with APC and β-catenin. Further investigation suggested that RPIA protects β-catenin by preventing its phosphorylation, ubiquitination, and subsequent degradation. In addition, the C-terminus of RPIA (amino acids 290 to 311), a portion of the protein not previously characterized, is necessary for RPIA-mediated tumorigenesis. Finally, we observed that transgenic expression of RPIA increases the expression of β-catenin and its target genes and induces tumorigenesis. Our findings suggest that RPIA can enter the nucleus and associate with APC/β-catenin, and suggest precise treatment of human CRC by targeting its nonenzymatic domain.
| Colorectal cancer (CRC) is one of the most common forms of cancers and results in more than 600,000 deaths annually [1–3]. Mutations in adenomatous polyposis coli (APC) and β-catenin, members of the Wnt signaling cascade, are among the major causes of colon tumorigenesis [4–6]. APC acts as a cytoplasmic scaffolding protein and induces the ubiquitin-mediated degradation of β-catenin [7]. In addition to its cytoplasmic activity, APC also modulates nuclear β-catenin levels as a result of its intrinsic nuclear-cytoplasmic shuttling capability [8–11]. Truncation of APC protein results in accumulation of nuclear β-catenin in CRC cells [12–14]. However, existing APC truncation mutants differentially affect the phosphorylation and ubiquitination of β-catenin, suggesting that these functions may be controlled by different APC domains [15,16]. In the nucleus, β-catenin acts as a coactivator with T-cell transcription factor 4 (Tcf-4)/lymphocyte enhancement factor (LEF) to activate the transcription of downstream targets such as Cyclin D1 (CCND1) and Cyclin E2 (CCNE2). Abnormal activation of the β-catenin signaling pathway can lead to increased cell proliferation and immortalization [17–19]. For example, a human CRC cell line expressing wild-type (WT) APC and a mutant version of β-catenin protein (with a single amino acid deletion at residue S45) is sufficient to induce a cancerous phenotype [20]. However, the precise activation process of β-catenin signaling is still largely unknown.
The pentose phosphate pathway (PPP) is critical for cancer cell survival and proliferation [21,22]. Ribose-5-phosphate isomerase A (RPIA) is an important integral member of the PPP and regulates cancer cell growth and tumorigenesis [2,23,24]. In pancreatic ductal adenocarcinoma (PDAC), RPIA expression is required for maintenance of tumor cells overexpressing KRasG12D, an activated form of Ras [23]. Our previous study showed that in hepatocellular carcinoma (HCC), RPIA regulates tumorigenesis via PP2A and extracellular signal-regulated kinase (ERK) signaling [24]. Studies performed in colon tumor tissues expressing microRNA-124 revealed that cells expressing low RPIA levels led to a reduced tumor size, while high RPIA expression was correlated with reduced survival and increased tumor growth [2].
Here, we report that in CRC tissue RPIA is significantly up-regulated, and it is expressed at multiple stages of tumorigenesis, including early stages. It directly interacts with β-catenin and APC to activate target genes downstream of β-catenin that are important for carcinogenesis. High levels of RPIA expression stabilize β-catenin levels by preventing phosphorylation and ubiquitination of β-catenin. Transgenic zebrafish overexpressing RPIA under the control of a gut-specific promoter exhibited enhanced β-catenin expression and elevated mRNA levels of the colon cancer marker gene ccne1. Our work uncovers a new role of RPIA and provides a molecular mechanism of RPIA-mediated β-catenin stabilization and activation necessary for colon cancer formation.
To assess the role of RPIA in the progression of CRC, we measured RPIA protein levels using immunohistochemistry (IHC) with tissue arrays from stage I through IVB and metastatic colon cancer. The RPIA immunoreactive score (IRS) was calculated by multiplying the staining intensity by the proportion of positive cells [25] (S1A Fig). Highly elevated RPIA expression was found in all stages of colon cancer when compared to non-cancerous samples (Fig 1A). A control, non-immune antibody was employed to assess background staining and was stained negative. In Fig 1A, “normal colon” shows dark staining in an epithelial region to the lower left that is of much lower intensity than that of the tumor samples. IRS quantification revealed that RPIA expression is significantly up-regulated in all stages of colon adenocarcinoma and even in metastatic carcinoma (Fig 1B). To examine whether the RPIA mRNA level is also up-regulated in CRC patients, 80 paired tissues, including tumors and the adjacent normal tissues, were analyzed using real-time quantitative PCR (qPCR). In 78% (62 of 80) of the CRC specimens, RPIA mRNA was more than 2-fold overexpressed from stage I to IV and in metastatic carcinoma (Fig 1C). Taken together, we found that RPIA is overexpressed at both the mRNA and protein levels in all stages of colon cancer formation.
To examine the effects of RPIA overexpression on cellular proliferation, two colon cancer cell lines, HCT116 and SW480, were used. SW480 is a human colon cancer cell line with APC C-terminal truncation at 1338, but the β-catenin binding region is retained; HCT116 is a highly metastatic cell line with WT APC and both an S45 mutation and the WT allele for β-catenin, but most of the β-catenin protein comes from the mutant allele. WST-1 assays examine metabolic activity that represent the viability of the cell. We tested three small interfering RNAs (siRNAs) (S1B Fig), which were pooled or treated separately, and found all three siRNAs had similar effects. Therefore, we selected number 3 siRNA for the rest of the experiments. Knock down of RPIA significantly decreased cell proliferation in both cell lines (Fig 2A and S2A Fig). Conversely, overexpression of RPIA increased cell proliferation in both HCT116 and SW480 cells (Fig 2A and S2A Fig). In the knockdown and followed by overexpressing the RPIA for rescue, the results clearly showed that RPIA small interfering RNA (si-RPIA) decreased proliferation and β-catenin protein level can be rescued by overexpression RPIA in both HCT116 and SW480 cell lines (Fig 2A and S2A Fig). Knockdown of RPIA also dramatically decreased the colony formation ability in both cell lines (Fig 2B and S2B Fig), and overexpression of RPIA increased the colony formation ability in both HCT116 and SW480 cells (Fig 2B and S2B Fig). These data suggest that the RPIA expression level is positively correlated with cellular proliferation and colony formation ability in colon cancer cells.
Aberrant β-catenin accumulation is a major cause of uncontrollable proliferation in colon cancer cells [26,27]. β-catenin exerts its proliferation-promoting effects via translocation to the nucleus where it binds to T-cell transcription factor (TCF)/LEF to activate the transcription of downstream β-catenin target genes [28]. Therefore, we were interested in determining whether RPIA affects the β-catenin level in colon cancer cells using HCT116 and SW480 cells.
Our results indicate that knockdown of RPIA decreased the β-catenin protein level in both HCT116 (Fig 2C) and SW480 cells (S2C Fig) without affecting β-catenin (encoded by CTNNB1 gene) mRNA levels. Conversely, RPIA overexpression increased β-catenin protein expression levels in both HCT116 (Fig 2D) and SW480 cell lines (S2D Fig), while β-catenin mRNA levels were unchanged. Using TOPflash/FOPflash luciferase reporter assay, we found that β-catenin activity was significantly attenuated upon RPIA knockdown (Fig 2E and S2E Fig) and dramatically increased when RPIA was overexpressed (Fig 2F and S2F Fig). Using qPCR to detect the expression levels of known downstream targets of β-catenin, including CCND1, CCNE2, and AXIN2, we found that knockdown and overexpression of RPIA reduced (Fig 2G and S2G Fig) and increased (Fig 2H and S2H Fig), respectively, the expression of these target genes. However, the effect of overexpression of RPIA was not as dramatic as that of knock down of RPIA, likely because CRC cell lines already overexpress RPIA. These results suggest that the RPIA expression level is positively correlated with β-catenin protein levels and its transcriptional activity.
Our previous study indicates that ERK signaling participates in RPIA-mediated hepatocarcinogenesis [24]. These observations, in combination with other studies demonstrating crosstalk between β-catenin and ERK in other types of tumors [29,30], led us to investigate whether ERK signaling might also play a role in RPIA-mediated tumorigenesis in colon cancer. Therefore, the effects of both RPIA overexpression and reduction on both β-catenin and ERK protein levels were examined in HCT116 and SW480 cells. Reduction of RPIA by knockdown significantly decreased nuclear β-catenin protein levels in HCT116 (Fig 3A) and both cytoplasmic and nuclear β-catenin protein were decreased in SW480 (S3A Fig), but did not affect the levels of activated, phosphorylated ERK (pERK) (Fig 3B and S3B Fig). Conversely, overexpression of RPIA increased β-catenin protein levels in HCT116 (Fig 3A) and SW480 (S3A Fig) cells without altering both cytoplasmic and nuclear level of pERK and ERK levels in these cell lines (Fig 3B and S3B Fig). Other mechanisms important for intestinal cell proliferation such as epidermal growth factor receptor (EGFR) signaling were also examined. Neither EGFR protein level nor the phosphorylated EGFR (pEGFR) were altered upon overexpression or knockdown RPIA in both cell lines (S3C Fig). Furthermore, IHC staining analyses revealed that both RPIA and β-catenin protein levels were significantly higher in the nuclei of colon cancer tissues than in the nuclei of normal tissues. In addition, a positive correlation was presented between RPIA and β-catenin protein levels in the nuclei of colon cancer tissue (Fig 3C). These data suggest that promotion of β-catenin signaling, but not ERK or EGFR signaling, is involved in transducing the effects of RPIA-mediated colon cancer tumorigenesis.
As only β-catenin protein levels were affected by the levels of RPIA expression, we proposed that RPIA increased β-catenin protein stability in colon cancer cells. To test this hypothesis, the protein synthesis inhibitor cycloheximide (CHX) was used to determine the half-life of β-catenin in cell lines with either reduced or elevated levels of RPIA. Reduction of RPIA by knockdown decreased the β-catenin protein half-life from 3 to 1.1 h in HCT116 cells (Fig 3D) and from 2.7 to 0.9 h in SW480 cells (S3D Fig). Conversely, overexpression of RPIA strongly increased the half-life of β-catenin from 5.6 to 10.4 h in HCT116 cells (Fig 3E) and from 4.9 to 9.9 h in SW480 cells (S3E Fig). Therefore, we conclude that RPIA increases β-catenin protein stability in colon cancer cells.
Because it has been shown that β-catenin protein levels can be controlled by ubiquitination and subsequent proteasome degradation [31], we tested whether RPIA could modulate β-catenin protein levels in colon cancer cell lines by changing the ubiquitination-mediated degradation process. As shown previously, RPIA knockdown resulted in a reduction in β-catenin protein levels (Fig 3F and S3F Fig, left panel). Treatment with 5 μM MG132, a proteasome inhibitor, rescued the reduction in β-catenin protein levels observed in cells expressing RPIA siRNA (Fig 3F and S3F Fig, left panel). Immunoprecipitation (IP) revealed that more ubiquitin was coprecipitated with β-catenin in RPIA-siRNA-treated cells than in negative control siRNA (si-NC)-treated cells (Fig 3F and S3F Fig, right panel). In addition, the phosphorylated, targeted for degradation form of β-catenin (with phosphorylation at residues Ser33/Ser37) was elevated upon RPIA knockdown relative to total β-catenin (Fig 3G and S3G Fig). Moreover, expression of a non-degradable β-catenin mutant (S33Y) rescued the reduction of proliferation upon RPIA knockdown in HCT116 and SW480 cell lines (Fig 3H and S3H Fig). To demonstrate that β-catenin is indeed required downstream of RPIA, the β-catenin inhibitor ICRT14 was applied to the RPIA overexpression cells. The results showed that RPIA-promoted cellular proliferation was attenuated in a dose-dependent manner by ICRT14 (Fig 3I and S3I Fig). This confirms that β-catenin is required for RPIA overexpression-mediated cell proliferation. The levels of an inactive form of GSK3β (with phosphorylation at residue ser9; pGSK3βser9), which does not have the ability to phosphorylate β-catenin, were further examined. Because phosphorylated GSK3βser9 was elevated upon overexpression of RPIA in both HCT116 and SW480 cell lines and there is no difference in the nucleus GSK3β, we suggest the RPIA modulate GSK3β only in the cytoplasm (Fig 3J and S3J Fig). Moreover, treatment with GSK3β inhibitors (lithium chloride [LiCl] or CHIR99021) rescued the reduction of proliferation upon RPIA knockdown, indicating the involvement of GSK3β in this process (Fig 3K and S3K Fig). These data show that RPIA retains a novel function to protect β-catenin from phosphorylation-mediated ubiquitination and degradation via proteasomes.
The cytoplasmic complex that targets β-catenin for degradation includes the scaffolding protein APC [32]. In addition, APC associates with β-catenin in the nucleus and directs the nucleocytoplasmic export of β-catenin [10,11]. We used immunostaining to detect RPIA localization. In the pcDNA3 vector only control (pcDNA), RPIA was expressed in the cytoplasm exclusively. Overexpression of RPIA in both HCT116 and SW480 cells resulted in an increase in nuclear and cytoplasmic RPIA expression (Fig 4A and S4A Fig) with a punctate pattern of RPIA in the nucleus of the DAPI-negative nucleoplasm [33]. IP of different proteins followed by western blotting indicated that RPIA can form a complex with APC and β-catenin in the nucleus in both HCT116 and SW480 cell lines (Fig 4B and 4C and S4B and S4C Fig). Interestingly, we noticed the minor difference between these cells. In HCT116, RPIA interacted strongly with APC and β-catenin, respectively, in the cytoplasm (Fig 4C). In addition, RPIA/APC and RPIA/β-catenin complex levels are induced by RPIA WT (RPIA-WT) (Fig 4B and 4C). In HCT116, the nuclear interaction of RPIA-β-catenin is much weaker than in cytoplasm. We suspect the interaction between RPIA and β-catenin in the nucleus might be indirectly through APC. In SW480, the nuclear RPIA-β-catenin interaction is much stronger than in cytoplasm. However, only the β-catenin bond to RPIA and promoted from RPIA-WT in cytoplasm in SW480 (S4B and S4C Fig) and the nuclear β-catenin-RPIA interaction can be regulated by the RPIA amount. The differences in HCT116 and SW480 might be caused by the truncated APC in SW480, and HCT116 has an S45 mutation from β-catenin.
The RPIA protein sequence is conserved among humans (Homo sapiens), mice (Mus musculus), and zebrafish (Danio rerio) (Fig 5A). To determine which protein domain(s) are important for RPIA-mediated tumor cell proliferation, five RPIA deletion mutants were generated. These include RPIA-ΔA (deletion of the active domain of RPIA), RPIA-ΔB (deletion of the catalytic domain of RPIA), RPIA-Δ(A+B), RPIA-ΔC, and RPIA deletion domain D mutant (RPIA-ΔD) (Fig 5B). The WT and five deletion mutants were transfected into HCT116 and SW480 cells. WST-1 assays were performed to examine metabolic activity, and the RNA and protein levels from the WT and five deletion mutants were verified (S5A and S5B Fig). We noticed that different RNA constructs might have different regulation of RPIA stability. Interestingly, only the expression of RPIA-ΔD failed to enhance cell proliferation, while the other mutants produced no significant changes in proliferation (Fig 4C and S4C Fig). These data suggest that the C-terminal domain D of RPIA (AAs 290 to 311) is essential for RPIA-mediated tumor cell proliferation. Domain D also seems to be necessary for the RPIA-mediated increase in β-catenin protein stability in colon cancer cells because overexpression of RPIA-ΔD did not stabilize β-catenin protein levels like the WT RPIA. Following overexpression of RPIA-ΔD, the half-life of β-catenin was approximately 6.1 and 3.3 h in HCT116 and SW480 cells, respectively, which was similar to that of the pcDNA vector alone (5.6 and 4.9 h in HCT116 and SW480 cells, respectively; Fig 3E and S3E Fig). Moreover, RPIA-ΔD did not interact with APC and β-catenin in either the cytoplasm or nucleus (Fig 4B and S4B Fig). Furthermore, RPIA-ΔD was unable to elevate TCF reporter activity in the colon cancer cells (Fig 5D and S5D Fig). These results demonstrate that domain D of RPIA is essential for the RPIA-mediated increase in β-catenin protein stability, activation of β-catenin target genes, and cell proliferation advantages seen in colon cancer cells. Together, our data also indicated that the RPIA D domain exhibits a novel function in addition to the enzymatic region.
Histopathologically, many of the features found in human colon adenocarcinoma are similar to those in zebrafish, an important vertebrate cancer model system [34,35]. To test the effects of RPIA overexpression on colon cancer formation, we generated transgenic zebrafish that overexpressed RPIA under the control of a gut-specific promoter (ifabp). In particular, we analyzed the histopathology of the intestinal bulb (IB), middle intestine (MI), and posterior intestine (PI) collected from non-transgenic and transgenic (ifabp:RPIA) fish of different ages. Increased nuclear-to-cytoplasmic ratio, nuclear atypia, and moderately differentiated adenocarcinoma were observed in 3- and 5-month-old Tg (ifabp:RPIA) fish relative to the WT controls (Fig 6A and 6B, upper panel). In WT zebrafish, the β-catenin protein is detected at low levels in the intestinal villi at intracellular junctions [36]. Using IHC, we also observed low-level and intracellularly located β-catenin expression in WT fish, while overexpression of RPIA in transgenic fish resulted in increased β-catenin expression and nuclear localization (Fig 6A and 6B, lower panel).
We next explored the β-catenin target genes, including ccne1, ccnd1, cdkn2a/b, myca, mycb, and lef1. A log 10-fold change value was used to show the level of genes [37] in 3-month-old Tg (ifabp:RPIA) fish, and the expression levels of β-catenin target genes were found to be positively correlated with RPIA expression levels (Fig 6C and 6E) and more highly expressed in the PI (Fig 6E) than in the IB (Fig 6C). In addition, as ccne1 is required for cell cycle/proliferation and tumor growth in CRC [29,38], it was used as colon tumorigenesis marker. The mRNA levels of ccne1 were elevated in 3-month-old Tg (ifabp:RPIA) fish, especially in the PI, which had dramatically increased RPIA expression. In addition to 3-month-old fish, we also analyzed β-catenin target genes in 5-month-old fish (S6A–S6C Fig). In accordance with the 3-month-old Tg (ifabp:RPIA) fish, the β-catenin target genes were up-regulated more significantly in the PI than in the IB (S6C Fig). Then, we noticed that RPIA and the expression of most of the β-catenin target genes were slightly decreased in 5-month-old fish compared with 3-month-old fish. Whether this phenomenon was the result of mature recovery in zebrafish remains to be determined [37,39]. A number of physiological changes are associated with cancer in human patients [40,41] including decreased body weight, decreased body width, decreased body length, and reduced intestinal length. With the exception of reduced body length, all these changes were observed in transgenic 1-year-old fish (S6D–S6G Fig). These results demonstrate that in vivo, in an important model system, RPIA increases β-catenin protein levels and induces colon tumorigenesis, resulting in overall weaker and smaller fish.
Recent findings have revealed that the non-oxidative PPP is a critical pathway for tumor formation [21]. Aberrant activation of the canonical Wnt/β-catenin pathway has also been shown to be involved in gastrointestinal cancers [36]. In cancer cells, β-catenin protein has a dual function: at the membrane, β-catenin coordinates adherent junctions for maintenance of epithelial cell barriers, while in the nucleus, β-catenin acts as a transcriptional activator to regulate proliferation genes [42–44]. In this study, we demonstrate that RPIA exhibits a novel role in CRC through association with and activation of β-catenin. High levels of RPIA expression were detected early and throughout multiple stages of 80 paired samples in CRC human patients. These results are consistent with the Human Gene Database and the Human Protein Atlas, which indicates about the 10-fold higher expression of RPIA in CRC patients than in normal tissues. Furthermore, we found RPIA stabilizes and subsequently promotes activation of β-catenin downstream target genes. We suggest that the increased cellular proliferation and oncogenicity are induced by RPIA through β-catenin pathway.
In other cancer types, RPIA promotes tumorigenesis via different mechanisms [2, 23,24]. In pancreatic and hepatic cancers, RPIA expression is required for maintenance of tumor cells overexpressing KRasG12D, while in HCC, RPIA regulates tumorigenesis via PP2A and ERK signaling. Interestingly, the RPIA-mediated CRC tumorigenesis does not involve the activation of ERK and presumably Ras signaling. This observation raises an interesting question: "Is RPIA-mediated stabilization and activation of β-catenin merely in CRC?" If valid, it may influence the decision of choosing different therapeutic targets of molecules and/or signaling pathways for treating different cancer types.
In the canonical β-catenin signaling pathway, APC binds to β-catenin in the cytoplasm in normal cells (Fig 7). This recruits GSK3β phosphorylates β-catenin, resulting in the eventual proteasomal degradation of β-catenin. We propose that in CRC cells, overexpression of RPIA results in the binding of RPIA to β-catenin and protects β-catenin from phosphorylation and subsequent cytoplasmic degradation. Intriguingly, we also found that RPIA interacts with APC and β-catenin in the nucleus. According to current studies, superfluous β-catenin is shuttled from the nucleus to the cytoplasm by APC [18,38]. We propose that RPIA might interrupt the APC-mediated process of exporting β-catenin by forming a complex in the nucleus. Consequently, colon cells developed tumorigenesis. Moreover, we found that C-terminal 22 amino acid of RPIA D domain is required for the RPIA-mediated β-catenin activation, stabilization, and enhanced colon cancer cell proliferation. This region is distinct from its enzymatic domain and a portion of the protein not previously identified as playing a role in CRC. Thus far, nothing is known about the function of the D domain, except this report. Cross species comparison of RPIA protein sequences between human, mouse, and zebrafish reveals that D domain is highly conserved across species. Accordingly, we hypothesize that the RPIA D domain exhibits a novel function in addition to the enzymatic region. It may associate with important partners, such as APC, β-catenin, and other proteins and form multimolecular complexes in both the cytosol and the nucleus. The phenomenon raises the questions such as “Does RPIA act as transcription co-activator?” and “Does RPIA have different protein partners in various cancer types?” We are currently searching for additional proteins that interact with RPIA in cancer cells.
The observation that RPIA is expressed at high levels early and throughout CRC is consistent with a role for RPIA in initiation and maintenance of carcinogenesis. Linking the clinical samples to our in vivo studies in zebrafish, misexpression of RPIA in the intestines of zebrafish is sufficient to induce spontaneous tumor formation in fish as young as 3 month and to cause additional physiological hallmarks of cancer, including reduced body weight, body width, and intestinal length in adult fish. These in vivo results are consistent with the observation that in cancer cell lines, downregulation of RPIA using microRNA reduces cell growth and colony formation ability, while overexpression of RPIA is correlated with enhanced growth and lower survival rates. In addition, the examination of apc/+ zebrafish revealed high levels of β-catenin that are disorganized and accumulate both in the cytoplasm and nucleus [45], and similar to our study, these fish develop spontaneous, intestinal tumors. In human patients, we noticed that RPIA expression was slightly decreased at the metastasis stage (Fig 1B and 1C). It was reported that invasive CRC cells exhibit low levels of proliferation markers [46]. We suggest that RPIA is necessary for primary tumorigenesis and that the RPIA level decreases at the metastasis stage so that tumor cells undergo epithelial-mesenchymal transition (EMT). Taken together, our studies demonstrate that RPIA functions as an activator for β-catenin-mediated colon tumorigenesis at the initiation stage.
One of the important functions of PPP is to generate ribose-5-phosphate for nucleotide synthesis. ATP provides the phosphate group via the salvage pathway [47], and RPIA mediates this enzymatic step in the cytoplasm. It has been proposed that knockdown of RPIA hinders tumor cell proliferation by reducing nucleotide synthesis [2]. However, in colorectal cells, the RPIA enzymatic and catalytic domain deletion mutants still promoted cell proliferation and activated β-catenin downstream target genes, revealing that at least in this type of cancer this is not how RPIA modulates tumor growth. In these cells, the RPIA D domain is necessary for cell proliferation, stabilization of β-catenin, and formation of a β-catenin/APC complex. Accordingly, D domain may be a therapeutic target for inhibiting the oncogenicity ability without affecting RPIA canonical enzymatic function in PPP. We therefore suggest that combination of anti-RPIA D domain therapy with conventional chemotherapy might improve the inhibition of CRC progression.
The mRNA from 80 paired tissues, including CRC and the adjacent normal tissues were obtained from Taipei Veterans General Hospital, procedures were undertaken in accordance with the Institutional Review Board of Taipei Veterans General Hospital, and the IRB number is 2015-04-010-AC. All adult participants provided written informed consent and there were no child participants.
All zebrafish experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of the NHRI and were in accordance with International Association for the Study of Pain guidelines (protocol number: NHRI-IACUC-104157-A). Taiwan Zebrafish Core Facility (TZCF) at NHRI or TZeNH is a government-funded core facility, and since 2015, the TZeNH has been AAALAC accredited.
The CRC cell lines HCT116 and SW480 were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM) (Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS), 100 units/ml of penicillin, and 100 μg/ml of streptomycin and incubated at 37°C and 5% carbon dioxide. DNA typing of the cell lines was verified by Mission Biotech (Taipei, Taiwan) using a Promega GenePrint 10 System.
Transient transfection of siRNA was performed using Lipofectamine RNAiMax (Invitrogen) according to the manufacturer’s manual. Three individual RPIA siRNAs (HSS117931, HSS117932, and HSS117933) and si-NC were purchased from Invitrogen.
pcDNA3.0-RPIA, RPIA-ΔA, RPIA-ΔB, RPIA-Δ(A+B), RPIA-ΔC, and RPIA-ΔD were constructed by subcloning full-length or truncated RPIA cDNAs into a pcDNA 3.0 expression vector. Truncated RPIA products were amplified by specific primer sets: RPIA-ΔA (nt 1–522 deletion) (forward) 5ʹ ATAGAATTCATGGGCGGAGGCTG CCTGAC 3ʹ and (reverse) 5ʹ AGACTCGAGCTTGCAGGGTCAACAGAAAGGCT 3′; RPIA-ΔB (nt 523–567 deletion) (forward) 5ʹ ATAAGTCGCTTCATCGTGATCGCT 3ʹ and (reverse) 5ʹ AGAACCCTTGATGAGATTGAGATCA G 3′; RPIA-Δ(A+B) (nt 1–567 deletion) (forward) 5ʹ ATAGAATTCATGAGTCGCTTCATCGTGATCGCT 3ʹ and (reverse) 5ʹ AGACTCGAGCTTGCA GGGTCAACAGAAAGGCT 3′; RPIA-ΔC (nt 568–867 deletion) (forward) 5ʹ ATAATGGCTG AGAGAGTCTACTTTGGGATG 3ʹ and (reverse) 5ʹ AGAAGCATAGCCAGCCACAATCTTCT 3′; RPIA-ΔD (nt 868–936 deletion) (forward) 5ʹ ATAGAATTCACTTCAGCGGAGGCCGGAG 3ʹ and (reverse) 5ʹ AGACTCGAGGTTGATGAATAGGCCTGTGTCC 3ʹ. Transient transfection of the plasmid DNAs was performed using Lipofectamine 2000 (Invitrogen) according to the manufacturer’s manual.
Colon cancer tissue staining data were obtained by using the tissue array CDA3 from SUPER BIO CHIPS and the tissue array MC5003a from US Biomax. The slides were incubated with mouse anti-RPIA (1:100) or rabbit anti-β-catenin (1:150) primary antibody at 4°C overnight after the dewax, rehydration, and antigen retrieval steps. The HE staining procedure was performed as outlined in our previous report, and tissues were examined with light microscopy [48].
Total protein was extracted from cells using whole-cell extract (WCE) lysis buffer. Lysates were vibrated for 30 min and centrifuged at 13,200 rpm for 20 min at 4°C. Western blotting was performed as outlined in our previous report [24], and the fractionation protocol used has been described previously [49]. Primary antibodies include RPIA (Cat# ab67080; Abcam, Cambridge, United Kingdom), β-actin (Cat# GTX109639; GeneTex, Inc, Irvine, CA), β-catenin (Cat# GTX61089, GeneTex; Cat# ab22656, Abcam), β-catenin (phospho Ser33/Ser37) (Cat# GTX11350 GeneTex), APC (Cat# GTX61328, GeneTex), Ubiquitin (Cat# 3936; Cell Signaling Technology, Danvers, MA), K48-linkage specific polyubiquitin (Cat# 4289, Cell Signaling Tecnology), β-Tubulin (Cat# ab52866, Abcam), and Lamin A/C (Cat# ab108922, Abcam). For IP, 100 μg of protein lysate was incubated with primary antibody overnight and subsequently incubated with protein A/G-Sepharose beads for 1.5 h. The interaction results were assessed with western blotting.
RNA was extracted from paired samples of patient tissues, transgenic zebrafish tissue, or cell lines homogenized in TRIzol. cDNA was reverse transcribed from RNA using a High Capacity RNA-to-cDNA Kit (Cat# 4387406; Applied Biosystems, Foster City, CA). qPCR was performed using an ABI Prism 7500 Sequence Detection System (Power SYBR Master Mix, Cat#4367659, Applied Biosystems). Gene expression was amplified with the primers listed in Supporting information: S1 and S2 Tables.
Cells were transfected with TOPflash (containing a WT TCF binding site) or FOPflash (containing a mutated TCF binding site), which were purchased from Millipore, and Renilla luciferase was used as an internal control. The transfected cells were harvested 48 h post-transfection and lysed by the buffer supplied in the Dual-Glo Luciferase Assay Kit (Cat# E2940, Promega, Madison, WI), and luciferase activity in lysates was measured with a luminometer.
Transfected cells grown on cover slips were fixed in 4% paraformaldehyde for 10 min at room temperature and permeabilized in 0.5% Triton for 10 min. After 1 h of blocking in 2% FBS at room temperature, slides were incubated with anti-RPIA or anti-APC primary antibody at 4°C overnight. Secondary antibodies conjugated with Texas Red or FITC were used, and DAPI was used to stain nuclei. The images were scanned and captured with confocal microscopy.
The coding region of human RPIA (NM_144563.2) was amplified by PCR with the attB1-F-RPIA and attB2-R-RPIA primer pair using cDNA from the HEK293 cell line as a template. PCR was performed using a KOD FX (Toyobo, Osaka, Japan) and a 994-bp amplicon. The following forward primer was used for Gateway cloning: attB1-F-RPIA (Tm:59°C):5ʹGGGGACAAGTTTGTACAAAAAAGCAGGCTATGCAGCGCCCCGGGCC3ʹ, and the following reverse primer was used for Gateway cloning: attB2-R-RPIA (Tm:58°C):5ʹGGGGACCACTTTGTACAAGAAAGCTGGGTTCAACAGAAAGGCTTCTCCCTCATG3ʹ. PCR comprised the following steps: stage I: 94°C for 5 min; stage II (35 cycles): 95°C for 30 sec, 58°C for 30 sec, and 72°C for 2.5 min; stage III: 72°C for 7 min; and stage IV: 4°C.
Gateway cloning was performed to generate the final expression construct, namely, pTol2-ifabp: RPIA; myl7: EGFP, using a MultiSite Gateway Three-Fragment Vector Construction Kit (Invitrogen). The transgenic zebrafish model was established via microinjections of the above constructs, which were performed as described elsewhere, and the transgenic fish were selected as described previously [50]. The F2 generations of Tg (ifabp: RPIA; myl7: EGFP) zebrafish (n = 36) and AB line (WT) zebrafish which served as controls (n = 12) were analyzed in this study.
The zebrafish were maintained at the TZCF under an automated 14:10-h light:dark cycle and a constant temperature of 28°C under continuous flow. MS 222/tricaine methanesulfonate (160 mg/L) were applied for anesthesia.
|
10.1371/journal.pntd.0004450 | Interferon-γ Is a Crucial Activator of Early Host Immune Defense against Mycobacterium ulcerans Infection in Mice | Buruli ulcer (BU), caused by infection with Mycobacterium ulcerans, is a chronic necrotizing human skin disease associated with the production of the cytotoxic macrolide exotoxin mycolactone. Despite extensive research, the type of immune responses elicited against this pathogen and the effector functions conferring protection against BU are not yet fully understood. While histopathological analyses of advanced BU lesions have demonstrated a mainly extracellular localization of the toxin producing acid fast bacilli, there is growing evidence for an early intra-macrophage growth phase of M. ulcerans. This has led us to investigate whether interferon-γ might play an important role in containing M. ulcerans infections. In an experimental Buruli ulcer mouse model we found that interferon-γ is indeed a critical regulator of early host immune defense against M. ulcerans infections. Interferon-γ knockout mice displayed a faster progression of the infection compared to wild-type mice. This accelerated progression was reflected in faster and more extensive tissue necrosis and oedema formation, as well as in a significantly higher bacterial burden after five weeks of infection, indicating that mice lacking interferon-γ have a reduced capacity to kill intracellular bacilli during the early intra-macrophage growth phase of M. ulcerans. This data demonstrates a prominent role of interferon-γ in early defense against M. ulcerans infection and supports the view that concepts for vaccine development against tuberculosis may also be valid for BU.
| Mycobacterium ulcerans is the causative agent of Buruli ulcer (BU), a slow progressing ulcerative skin disease. The mode of transmission of M. ulcerans remains unknown and only little is known about the early stages of the disease and the nature of protective immune responses against this pathogen. Given the increasing evidence for an early intracellular growth phase of M. ulcerans, we aimed at evaluating the impact of cell-mediated immunity for immunological defense against M. ulcerans infections. By comparing wild-type and interferon-γ-deficient mice in a BU mouse model, we could demonstrate that interferon-γ is a critical regulator of early host immune defense against M. ulcerans infections, indicative for an important role of early intracellular multiplication of the pathogen. In mice lacking interferon-γ the bacterial burden increased faster, resulting in accelerated pathogenesis. The observed differences between the two mouse strains were most likely due to differences in the capacity of macrophages to kill intracellular bacilli during the early stages of infection.
| Buruli ulcer (BU), caused by infection with Mycobacterium ulcerans (M. ulcerans), is a progressive disease of the skin and subcutaneous tissue. The disease is primarily affecting West African rural communities, but has also been reported from America, Australia and Asia. The pathogenesis of BU is mainly attributed to mycolactone, a macrolide exotoxin produced by M. ulcerans [1]. Mycolactone is essential for bacterial virulence and is highly cytotoxic for a wide range of mammalian cell types in vitro and in vivo, including fibroblasts, keratinocytes and adipocytes [1–4]. Injection of the toxin induces the formation of necrotic non-inflammatory lesions similar to BU lesions. In addition to the induction of apoptosis, mycolactone possesses immunosuppressive characteristics and has been demonstrated to downregulate local and systemic immune responses [5,6], by interfering with the activation of immune cells such as T-cells, dendritic cells, monocytes and macrophages [7–10]. Furthermore, exposure to mycolactone results in complete inhibition of tumor necrosis factor alpha (TNFα) production by monocytes and macrophages, affects T-cell homing and interferes with the expression of T-cell receptors as well as co-stimulatory molecules including CD40 and CD86 [6–12].
Despite these immunosuppressive features of mycolactone, sera of individuals living in BU endemic regions frequently contain M. ulcerans-specific antibodies, demonstrating that many individuals develop immune responses associated with exposure to M. ulcerans without developing clinical disease [13,14]. Moreover, high mRNA levels for the cytokines interferon-γ (IFNγ), interleukin-1β and TNF-α were found in human BU lesions, indicating that the innate immune system is activated at the site of infection [15]. Reports on spontaneous healing of BU [16,17], and a partial protective effect of Bacille Calmette-Guérin (BCG) vaccination in humans and experimentally infected mice [18–22] are all factors indicating that clearance of the M. ulcerans infection by the immune system is possible, in particular before large clusters of mycolactone producing extracellular bacteria have formed. These clusters are located in necrotic subcutaneous tissue of advanced BU lesions and are no longer reached by infiltrating leukocytes.
Antibodies against surface antigens of M. ulcerans do not seem to have a protective effect [23], indicating that cellular, and in particular type 1 helper (TH1) cell responses [1,24] are more important in immune defense against BU than humoral responses.
IFNγ is critical for host defense against intracellular pathogens. In Mycobacterium tuberculosis (M. tuberculosis) infections, IFNγ produced by TH1 cells, but also CD8 cytotoxic T (Tc) cells and NK cells, renders the macrophage competent to kill intracellular bacteria by overcoming the pathogen-induced block in phagosome-lysosome fusion and by producing microbicidal effectors such as nitric oxide (NO), resulting in host cell apoptosis and clearance of the bacteria [25–28]. During M. ulcerans infection, an early intra-macrophage growth phase seems to play an important role before the formation of extracellular clusters of mycolactone producing bacteria can be observed [6,29–31]. Protection mediated by IFNγ stimulated macrophages seems to be impaired by the suppression of IFNγ production after local build-up of mycolactone [32].
Here we have re-evaluated the role of IFNγ for host immune defense against M. ulcerans by comparing progression of the infection in IFNγ knockout and wild-type mice experimentally challenged with a fully virulent M. ulcerans isolate.
This study was carried out in strict accordance with the Rules and Regulations for the Protection of Animal Rights (Tierschutzgesetz SR455) of the Swiss Federal Food Safety and Veterinary Office. The protocol was granted ethical approval by the Veterinary Office of the county of Vaud, Switzerland (Authorization Number: 2657).
Mice were kept in specific pathogen-free facilities at the Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland. All experiments were performed under BSL-3 conditions either in 8 week old female C57Bl/6 wild-type mice or mice homozygous for the Ifngtm1Ts targeted mutation (IFNγ-/-, B6.129S7-Ifngtm1Ts/J, Jackson Laboratory). In total, 20 wild-type and 20 IFNγ-/- mice were infected and 5 animals per group were euthanized at week 1, 3, 5 and 8 and used for qPCR analysis (3 mice) or histopathology (2 mice). The experiment was performed in two independent biological replicates. Animals were infected with the M. ulcerans strain S1013 isolated in 2010 from the ulcerative lesion of a BU patient from Cameroon [33] which is regularly tested for the production of mycolactone by ASL extraction and subsequent cytotoxicity tests on L929 fibroblasts as well as for the presence of the virulence plasmid pMUM001 by PCR. The bacteria were cultivated from a low passage cell bank for six weeks in Bac/T medium (Biomerieux, 251011), pelleted by centrifugation and diluted in sterile PBS to a stock concentration of 125 mg/ml wet weight. Mice were infected subcutaneously into the hind left foot pad with 30 μl (about 1 x 104 bacilli as determined by qPCR corresponding to 5 x 103 CFUs when plated on 7H9 ager plates) of an appropriate dilution of the stock suspension in sterile PBS. Progression of the infection was followed by weekly measurements of the foot pad thickness using a caliper. At weeks 1, 3, 5 and 8, groups of mice were euthanized and pictures of the feet were taken using a compact camera (WG-20, RICOH). The foot pads were aseptically removed for determination of the bacterial load by quantitative real-time PCR (qPCR) or for histopathological analysis.
Feet designated for the quantification of M. ulcerans were cut into 4 pieces using a scalpel and transferred to hard tissue grinding tubes (MK28-R, Precellys, KT03961-1-008.2). Next, 750 μl sterile PBS was added and feet were homogenized using a Precellys 24-Dual tissue homogenizer (3 x 20 s at 5000 rpm with 30 s break). The lysates were transferred into new tubes and the remaining tissues were homogenized for a second time after adding 750 μl of sterile PBS. The two lysates were pooled and used for DNA isolation. The DNA was extracted from 100 μl of a 1/20 dilution of the foot pad lysates as described by Lavender and Fyfe [34] and the isolated DNA was analyzed for insertion sequence (IS) 2404 by qPCR as previously described [34]. The number of genomes per foot pad was calculated according to the standard curve established by Fyfe et al. [35].
Mouse feet used for histopathological analysis were fixed in 10% neutral-buffered formalin solution (4% formaldehyde, Sigma, HT501128-4L) for 24 hours at room temperature, decalcified in 0.6 M EDTA and 0.25 M citric acid for 14 days at 37°C and transferred to 70% EtOH for storage. After dehydration and embedding in paraffin, 5 μm thin sections were cut. Sections were then deparaffinised, rehydrated, and stained with Haematoxylin/Eosin (HE, Sigma, 51275-500ML, J.T. Baker, 3874) or Ziehl-Neelsen/Methylene blue (ZN, Sigma, 21820-1L and 03978-250ML) to stain for mycobacteria according to WHO standard protocols [36]. Finally, the sections were mounted with Eukitt mounting medium (Fluka, 03989) and pictures were taken with an Aperio scanner or with a Leica DM2500B microscope.
Ten μg of M. ulcerans whole cell lysate was resolved on a 1-well 4–12% gradient gel (NuPAGE Novex 4–12% Bis-Tris Gel, Invitrogen, NP0330BOX) using MES running buffer and transferred to nitrocellulose membranes with the iBlot dry-blotting system (Novex, Life Technologies) according to the manufacturer’s recommendations. The membrane was blocked in 5% skim milk / PBS overnight at 4°C, cut into thin strips and incubated with the indicated sera diluted 1:400 in 1% skim milk / PBS-Tween-20 for 1.5 hours. After washing in 1% skim milk / PBS-Tween-20, the membrane was incubated for 1 hour with HRP-conjugated goat anti-mouse IgG γ-chain secondary antibody (Southern Biotech, 1030–05) diluted 1:4000 in 1% skim milk / PBS-Tween-20. Blots were developed using the ECL Western Blotting Substrate (Pierce, 32106).
A non-parametric Mann-Whitney test (Prism GraphPad) was used for statistical analysis of foot pad thickness measurements. Because of the small sample size for each group at a certain time point, the measurements of the bacterial loads were analyzed using non-parametric regression models according to the Brunner-Langer method [37]. The factor of interest, increase in bacterial burden between week 3 and 5 in the case of (1) and bacterial burden at week 5 in the case of (2) was included in a model to determine its effect on the examined outcome. In (1), because all 4 time points are compared in a second model, results from the regressions were adjusted for multiple comparisons using Dunnett-Hsu’s correction. The global effect of group, time point and of the interaction of group by time point were first tested [38].
In order to evaluate the role of IFNγ in host immune defense against M. ulcerans infections, we infected 8 week old female C57Bl/6 wild-type (WT) mice and mice homozygous for the Ifngtm1Ts targeted mutation (IFNγ-/-) into the left hind foot pad with 1 x 104 M. ulcerans bacilli as determined by qPCR. Progression of the disease was followed by weekly measurements of the foot pad thickness with a caliper. While all of the ten IFNγ-/- mice displayed strong swelling of the infected foot pads after 5 weeks of infection, no swelling was observed for the ten WT animals (Fig 1A). After 6 weeks of infection, one of the remaining five WT animals also started to show swelling of the infected feet. However, there was still a significant difference in foot pad thickness between the two different groups, which only resolved by week 8 (Fig 1A).
Complementary to the determination of the foot pad thickness, we documented the disease progression with pictures of the infected feet at 1, 3, 5 and 8 weeks after infection (Fig 1B). At week 5 infected foot pads of all ten WT mice did not show any macroscopic difference to the non-infected right control foot pads (Fig 1B). In contrast, all infected feet of the ten IFNγ-/- mice were swollen and showed signs of inflammation (Fig 1B). Although the difference in the foot pad thickness resolved after 8 weeks of infection, the infected feet of the IFNγ-/- animals were more inflamed and clearly more ravaged at this time point (Fig 1B).
Histopathological analysis of two representative foot pads was performed to evaluate whether the increased foot pad thickness in IFNγ-/- mice at week 5 was caused by cellular infiltration or mainly by oedema formation. While no changes in tissue integrity were observed after 1 week of infection in both groups (Fig 2A1–2A3 and 2B1–2B3), the two IFNγ-/- mice displayed massive oedema formation and tissue necrosis after 5 weeks of infection (Fig 2B5, 2B4 and 2B6, respectively). Both are typical hallmarks of BU pathogenesis [39,40]. In contrast, the foot pads of the two WT animals were devoid of oedema formation or tissue necrosis at this time point (Fig 2A4–2A6). Eight weeks after infection the foot pads of the IFNγ-/- mice were still more oedematous and necrotic than those of the WT animals and the infection even affected the adjacent joints and legs (Fig 2A7–2A9 and 2B7–2B9).
Next, we assessed whether the more severe course of M. ulcerans infection in the IFNγ-/- mice was associated with a higher bacterial burden in these animals. The bacterial load in footpads of three WT and three mutant mice was determined 1, 3, 5 and 8 weeks after infection by qPCR [23,34,35]. Strikingly, IFNγ-/- mice showed a significantly higher increase in the bacterial load between week 3 and 5 (Fig 3A). Furthermore, we found that the bacterial load in the mice lacking IFNγ was significantly (3.5 fold) higher after 5 weeks of infection than in WT mice (Fig 3A), correlating with the strong foot pad swelling observed at this time point in only the mutant mice (Fig 1A and 1B). As for the foot pad thickness, the differences in the bacterial load had resolved 8 weeks after infection (Fig 3A).
To complement the qPCR results we stained tissue sections of whole foot pads with ZN to detect AFB. After 3 weeks of infection, only few AFB were found which were predominantly intracellular (Fig 3B1 and 3B2). As for the qPCR analysis (Fig 3A), no difference in the total number of AFB was observed between the two groups at this time point. However, a trend to less extracellular bacterial debris and more intact extracellular bacilli was observed for IFNγ-/- foot pads at this time (S1 Fig).
In contrast, more AFB were detected in both IFNγ-/- mice 5 weeks after infection (Fig 3C3 and 3C4), as compared to the two WT controls (Fig 3C1 and 3C2), which again corresponded with the results of the qPCR analysis (Fig 3A). At this time point, AFB were present as a mix of intra- and extracellular bacteria (Fig 3C). Interestingly, while the bacterial load was different for the two groups at this time point, no marked differences in the total cell infiltration was observed (Fig 3C). In line with the findings from the qPCR analysis (Fig 3A), the differences in the bacterial load had resolved 8 weeks after infection (S2 Fig).
To evaluate whether the stronger increase in the bacterial load between weeks 3 and 5 in the IFNγ-/- mice (Fig 3A) was caused by a diminished innate immune response as a result of lack of activating IFNγ or rather by reduced antibody-mediated immune responses against M. ulcerans, we tested the reactivity of sera of infected mice with M. ulcerans whole cell lysates by Western Blot analysis. A complete absence of specific antibodies was observed both for the five WT and five IFNγ-/- mice after 5 and 8 weeks of infection (Fig 4). Together with the observed presence of less extracellular debris in IFNγ-/- mice during the early phase of the infection (S1 Fig), this indicates that CMI is critical for host immunity against M. ulcerans infections.
Evidence for an early intra-macrophage growth phase of M. ulcerans has led to the suggestion that the immune effector mechanisms protecting against M. ulcerans infection are similar to those active against M. tuberculosis [41–43]. However, in contrast to this closely related pathogen, M. ulcerans has the capacity to produce the cytotoxic macrolide mycolactone, which eventually kills the host cells and causes the characteristic necrotizing pathology of BU [1,40]. In the case of M. tuberculosis infection, the host immune response involves cell-mediated immunity (CMI) accompanied by a delayed type hypersensitivity (DTH) reaction [44]. Similarly, several reports showed that CMI and DTH responses are frequently induced in BU patients [43,45–50].
If CMI is required for immunological defense against M. ulcerans infections, IFNγ which is produced primarily by TH1, but also by TC and NK cells, is likely to play a critical role in this process by activating macrophages to kill intracellular bacteria at an early stage of infection. To test this hypothesis, we have used an experimental BU mouse model and compared the disease progression in WT and IFNγ-/- mice during active infection with a highly virulent M. ulcerans strain recently isolated from the lesion of a BU patient [33]. Our study conclusively demonstrates a key role of IFNγ for early immune defense against M. ulcerans infection in vivo, as mice lacking this cytokine suffered from an accelerated and more severe pathology associated with a significantly higher bacterial burden after 5 weeks of infection. These results indicate that CMI and IFNγ-dependent activation of the bactericidal activity of macrophages helps to contain the infection during its largely intracellular early stages. Further support for this hypothesis came from our histopathological analysis, where a trend to lower levels of extracellular acid-fast debris was found in the IFNγ-/- mice at the early intracellular stages of the infection.
Moreover, these findings are in line with the observation of Torrado et al. who have reported that IFNγ-dependent phagosome maturation and NO production are required to control the intracellular proliferation of M. ulcerans in vitro [32]. In the same report, it is described, that IFNγ-deficient mice show increased susceptibility only for mycolactone-negative or intermediate virulent, but not for highly virulent M. ulcerans strains [32]. However, these at first view contradictory results can be explained by the fact that the mice infected by Torrado et al. with a highly virulent M. ulcerans strain were only monitored over a period of 20 days post infection, a time frame that is too narrow to detect the differences between WT and IFNγ-/- mice, as we did not observe them before 5 weeks of infection. In addition, different M. ulcerans strains differing in the geographic origin, the mycolactone variants produced and the pattern of genomic changes associated with evolutionary genome reduction [51] were used by the two groups.
In conclusion, our results indicate that the outcome of an infection with M. ulcerans may depend strongly on cellular immune defense mechanisms. IFNγ is likely to play an important role both as an element of innate immunity in the very early phase of host-pathogen interaction after inoculation and also in the subsequent development of protective adaptive cellular immune responses. Innate and adaptive immune defense mechanisms seem to be strong enough in the majority of exposed individuals living in BU endemic areas to protect them from developing clinical disease [13,14]. However, when the immune response of an individual is too weak to kill the intracellular bacteria, BU disease may develop. In line with this, HIV positive individuals are at higher risk for BU and AIDS-associated immunosuppression has a negative influence on the severity of BU [52–54]. In the case of an insufficient immune response, intracellular multiplication of the bacteria may take place and small accumulations of bacteria found as globus-like structures [30,55] may represent the origin for the formation of large clusters of mycolactone producing M. ulcerans bacteria. As a result of mycolactone-induced host cell apoptosis, necrotic areas are forming around the bacteria. Furthermore, in the advanced BU lesions viable leukocyte infiltrates are no longer found close to the infection foci in the necrotic subcutaneous tissue, indicating that the accumulation of mycolactone is preventing macrophages and other defense cells from reaching the now extracellular pathogens before they are killed. As a result, a chronic M. ulcerans infection may develop, leading to the formation of large BU lesions, often resulting in severe morbidity and disability and requiring long and costly hospitalization [40].
|
10.1371/journal.ppat.1002520 | Characterising the Mucosal and Systemic Immune Responses to Experimental Human Hookworm Infection | The mucosal cytokine response of healthy humans to parasitic helminths has never been reported. We investigated the systemic and mucosal cytokine responses to hookworm infection in experimentally infected, previously hookworm naive individuals from non-endemic areas. We collected both peripheral blood and duodenal biopsies to assess the systemic immune response, as well as the response at the site of adult worm establishment. Our results show that experimental hookworm infection leads to a strong systemic and mucosal Th2 (IL-4, IL-5, IL-9 and IL-13) and regulatory (IL-10 and TGF-β) response, with some evidence of a Th1 (IFN-γ and IL-2) response. Despite upregulation after patency of both IL-15 and ALDH1A2, a known Th17-inducing combination in inflammatory diseases, we saw no evidence of a Th17 (IL-17) response. Moreover, we observed strong suppression of mucosal IL-23 and upregulation of IL-22 during established hookworm infection, suggesting a potential mechanism by which Th17 responses are suppressed, and highlighting the potential that hookworms and their secreted proteins offer as therapeutics for human inflammatory diseases.
| Parasitic worms reside in the gastrointestinal tracts of billions of humans in developing countries. Despite the enormous disease burdens associated with these infections, very little is known about the immune response in the gut tissue of humans to these parasites. We conducted a clinical trial where we obtained gut biopsies from people experimentally infected with hookworms and present here the first report of the immune response by healthy human gut tissue to a parasitic worm. We show that hookworms suppress the production of pro-inflammatory molecules and promote the expression of anti-inflammatory and wound healing molecules in the gut, providing a potential mechanism by which parasitic worms reside for long periods in their human hosts and suppress inflammation associated with auto-immune diseases.
| The hookworms Necator americanus and Ancylostoma duodenale infect an estimated 740 million people, mostly in tropical regions of the world, causing significant burden of disease [1]. As for most Neglected Tropical Diseases (NTDs), there is currently no prophylactic or therapeutic vaccine against hookworm infection, although clinical trials are underway on a number of promising candidate antigens [2]. Despite efforts to eliminate this disease from developing countries, experimentally-induced hookworm infection offers potential as an anti-inflammatory therapy for human autoimmune [3] and allergic [4], [5] diseases. However, despite their importance in regards to disease burden in resource poor countries (especially in children and women of child bearing years) and their potential as an anti-inflammatory therapy for use in industrialized countries, little is known about the mucosal immune responses of humans to hookworm, or indeed any other gastrointestinal (g.i.) helminths parasites.
Unlike many other human g.i. helminths, despite a robust, parasite-specific immune response, naturally acquired protection against hookworm is only partially effective at best; indeed, in endemic areas the oldest people often have the heaviest worm burdens [6], [7]. Nonetheless, previous studies on people naturally infected with hookworm have identified associations between reduced egg counts and Th2 responses. For example, IL-5 production correlates positively with resistance to reinfection after anthelmintic drug cure [8], and levels of IgE reactive against defined larval antigens are negatively associated with hookworm egg counts [9].
A small number of experimental infections in hookworm naive, healthy human volunteers have been conducted, with an exclusive focus on the systemic immune response at both the humoral and cellular levels [10]–[12], and gross observations of the gut via capsule endoscopy [3]. These earlier observations described the onset of eosinophilia, production of parasite specific IgG and IgE, and secretion of both Th1 (IFN-γ and TNF-α) and Th2 (IL-5 and IL-13) cytokines. With the onset of patency, IL-10 was produced and T cell proliferation was blunted and was not restored until long after curative therapy [7], [11].
Most of our understanding of mucosal immunity to g.i. nematodes comes from studies in laboratory mice. Th2 cytokines are required for resistance to many g.i. helminths, as seen in mice that are genetically deficient in Th2 cytokines and associated signalling molecules [13], [14]. In the draining lymph node, Th2 cytokines are responsible for class-switching of B cells to IgG1 and IgE, as well as recruiting and activating innate immune cells and blocking parasite effector molecules [15]. At the site of adult worm residence in mice, the duodenum, Th2 cytokines are responsible for increased mucus and fluid production in the gut and smooth muscle contractility, which increases ejection of parasites [16], [17]. They also lead to recruitment, expansion and differentiation of innate immune cells such as eosinophils, alternatively activated macrophages, mast cells and basophils in the gut which can directly or indirectly lead to ejection of parasites [15]. Thus differentiation of Th2 cells and production of Th2 cytokines, both systemically and in the mucosa, may be important for intestinal parasite clearance in mice.
Th1 (and Th17) responses are also induced during some helminth parasite infections. In the absence of a Th2 response, or where Th1/Th17 responses have been artificially upregulated, an uncontrolled Th1/Th17 response to schistosomes leads to acute pathology and ultimately death in mice [18], [19]. Thus, it has been proposed that the Th2 response generated during schistosomiasis may downregulate Th1/Th17 responses, leading to suppression of immunopathology and survival of the host [15]. Suppression of Th17 responses by Th2 cytokines in the mucosa has also been shown in mice infected with g.i. nematodes [20], prompting the suggestion that nematodes may ameliorate inflammatory gut diseases by dampening pro-inflammatory Th17 responses.
We previously reported a study using human hookworm infection to treat celiac disease [21]. Although no overt suppression of clinical pathology was detected, suppression of gluten-specific inflammatory Th1 and Th17 responses was seen in the mucosa [22]. After established hookworm infection but prior to challenge with gluten, samples were taken from control and hookworm infected individuals, and here we prospectively collected data on the hookworm-specific cytokine responses in the peripheral circulation and, for the first time, the duodenal mucosa, of hookworm naive individuals before and after controlled experimental infection with N. americanus. This is the first description of the mucosal immune response of humans to hookworms; indeed, other than a case study where an individual patient with active ulcerative colitis was treated with whipworm and the mucosal immune response was assessed [23], this is the first report of the mucosal immune response in healthy volunteers in a clinical trial to experimental infection with helminths, and provides valuable information to support the development of both vaccines against hookworm infection and hookworm-derived peptidic therapies for inflammatory diseases.
The Princess Alexandra Hospital, Queensland Institute of Medical Research and Townsville Hospital Human Research Ethics Committees approved the study. Written informed consent was obtained from all subjects.
The methods used for our placebo-controlled, blinded clinical trial using hookworm to treat celiac disease have been described elsewhere [21]. Briefly, twenty confirmed HLA-DQ2+ celiac disease sufferers on a long-term gluten-free diet (and therefore in remission for celiac disease) were recruited, randomised into 2 groups and either infected with 10 infective larvae (L3) of N. americanus (“hookworm” group) or given a placebo of topical chilli (Tabasco sauce) (“control” group). Twelve weeks later, a booster infection of 5 infective larvae (or a placebo infection) was administered. At week 20 post-prime infection, all individuals were given a gluten challenge consisting of four slices of white bread per day for 5 days. This trial will herein be referred to as “Trial 1”. Approximately 6 months after the end of Trial 1 (during which all participants returned to a strict gluten-free diet), seven of the ten control subjects (those who did not receive hookworm in Trial 1) participated in a continuation trial: two could not participate due to other commitments, and one could not participate due to raised tissue transglutaminase antibodies. These 7 participants were infected with N. americanus, boosted and challenged with gluten in an identical manner to that described for Trial 1. This trial will herein be referred to as “Trial 2”.
In both trials hookworm infection was confirmed in all subjects by either fecal egg counts and/or identification of adult parasites in the duodenum during endoscopy [21]. The structure of the trials is summarised in Figure 1.
Peripheral blood mononuclear cells (PBMCs) were isolated from blood drawn into heparinised tubes over a Ficoll-Paque Plus gradient (GE Healthcare) as described in Figure 1. Cells were cultured for 120 h at 37°C, 95% O2/5% CO2 at 2.5×105 cells/well in round-bottom 96-well plates in Tissue Culture Medium (Med: RPMI 1640, 10% fetal bovine serum, 100 U/µl penicillin, 100 µg/ml streptomycin and 2 mM L-glutamine), in the absence (“Med”) or presence of 10 µg/ml N. americanus excretory/secretory proteins (NaES). NaES was prepared as previously described [24], and was depleted of endotoxin using two rounds of phase separation using Triton-X114 [25]. Endotoxin levels in NaES after depletion were assessed using E-toxate (Sigma); levels were below the detection limit of the assay (<0.05–0.1 EU/ml) for stock solution of NaES at a concentration of 1.77 mg/ml. Cell-free supernatants were collected and analysed using a Cytometric Bead Array (CBA; BD Biosciences). Antigen-specific production of cytokines was determined by subtraction of baseline cytokine levels from unstimulated PBMCs (Med) from those stimulated with NaES.
Duodenal biopsies were taken from week 20 post-prime infection (prior to gluten challenge) in both groups from Trial 1, and also from sites adjacent to (within 0.5 cm) a hookworm attachment site where adult hookworms were found in the upper duodenum (5 of 10 hookworm-infected individuals) by endoscopy [21]. In Trial 2, biopsies were taken at week 0 (prior to infection) and at week 20 post-prime infection. Whole biopsies were placed in wells of a 24-well plate containing 500 µl MED alone or MED containing 10 µg/ml NaES, and cultured for 24 h in 95% O2/5% CO2 at 37°C. Cell-free supernatants were taken and analysed using a Cytometric Bead Array (BD Biosciences). Biopsies were then placed into Trizol (Invitrogen) and RNA was purified following the manufacturer's protocols.
For quantitative real-time RT-PCR (qPCR), RNA was prepared from biopsies in Trial 2 by the phenol-chloroform method (Trizol). mRNA quality was tested using a Bioanalyzer (Agilent) or agarose gel electrophoresis prior the reverse-transcription step. cDNA was prepared using Superscript III reverse transcriptase (Invitrogen) according to the manufacturer's protocol. PBMCs from a healthy donor were cultured for 24 h in 95% O2/5% CO2 at 37°C with phytohemagglutinin-A (PHA) and used to create standard curves and positive controls. Levels of transcripts were normalised to the housekeeping gene β-actin and are presented as arbitrary units. SyBr Green mastermix (Qiagen) was used in a Rotor-Gene Q thermal cycler (Qiagen) according the manufacturer's protocol. Primers used for each gene product are listed in Table S1.
All analyses were carried out using Prism 5.0 (Graphpad). Paired data were compared by Wilcoxon matched-pairs signed rank test; 3 or more sets of paired data were compared by the Kruskal-Wallis non-parametric ANOVA. Unless otherwise indicated, differences were not significantly different. N.S. = Not Significant, * = p<0.05, ** = p<0.01, *** = p<0.001. All error bars show the standard error of the mean.
All volunteers infected with hookworm were confirmed to have active infections using a combination of capsule endoscopy (to visualize adult worms in the gut) and/or the presence of eggs in the feces [21]. PBMCs from volunteers infected with 15 third-stage larvae (L3) of N. americanus were restimulated with NaES and showed increased antigen-specific production of the Th2 cytokines IL-4, IL-5 and IL-13, compared with PBMCs from uninfected controls, reaching a peak 12 weeks after infection (Figure 2A–C), although increases in IL-4 levels did not reach statistical significance. These data indicate that, as expected, hookworm infection induces a systemic Th2 response.
To establish whether this Th2 response was present at the site of adult worm residence, duodenal biopsies were taken at week 20 after hookworm infection (immediately prior to gluten challenge). Biopsies were also taken from directly adjacent to the hookworm attachment site at week 21 (after gluten challenge) from 5 of the 10 hookworm-infected individuals where adult worms were observed by endoscopy. All biopsies were cultured without stimulation and supernatants were removed for cytokine analysis. Biopsies from both control and hookworm infected individuals produced similar levels of IL-4 and IL-13 (Figure 3A and C). However, significantly increased levels of IL-5 were produced by biopsy cells from hookworm-infected individuals, especially those biopsies taken adjacent to the hookworm attachment site (Figure 3B). The increased levels of IL-5 at the hookworm attachment sites were not the result of gluten challenge, because at week 21, biopsies from sites distal to the hookworm attachment sites produced decreased levels of IL-5 (12.77 pg/ml +/− 13.90) compared to biopsies from week 20 (23.47 pg/ml +/− 31.88) or the hookworm attachment site (41.18 pg/ml +/− 21.33).
Figures 2 and 3 show that experimental hookworm infection induces a systemic, Necator antigen-specific, Th2 response, and a weak but detectable basal mucosal Th2 response. In order to further characterise this response, duodenal biopsies were taken before (week 0) and after an established hookworm infection (week 20 post-prime infection) from infected individuals in Trial 2. Cytokines were measured in the supernatants of duodenal biopsies cultured for 24 h in medium only. There was no significant difference in the protein levels of IL-2, IFN-γ, TNF-α, IL-17A, IL-4, IL-5, IL-10 and IL-13 when comparing wk 0 (pre-infection) to wk 20 (post-infection) (Fig. S1). RNA transcripts were obtained to assess gene expression levels in the absence of ex vivo stimulation for a range of cytokines and transcription factors associated with different T helper cell phenotypes. Levels of mRNA encoded by the Th2/Th9 genes IL-4, IL-5, IL-13, IL-9 and GATA-3 appeared unaffected by hookworm infection using this technique (Figure 4A–E). Accumulation of mRNA transcribed by the regulatory T cell associated gene Foxp3 (Figure 4F), or the gene encoding the immunosuppressive cytokine TGF-β (Figure 4G), were also unaffected, although levels of Foxp3 mRNA were below the detection limits of the assay in the majority of the samples tested. However, accumulation of mRNA encoded by the ALDH1A2 gene was significantly increased after hookworm infection (Figure 4H). ALDH1A2 encodes retinaldehyde dehydrogenase, an enzyme that is important for production of retinoic acid from vitamin A metabolites. Transcription of the Th1 cytokine gene IFN-γ (Figure 4I) and the T cell proliferative cytokine gene IL-15 (Figure 4J) were also upregulated after hookworm infection. Levels of the Th17-associated genes, IL-17A and RORγt, were both extremely low, close to or below the detection limit of the assay, but nevertheless appeared unchanged after hookworm infection (Figure 4K and M). Accumulation of mRNA transcribed by the Th17 inducing and stabilising cytokine IL-23, however, was strongly down-regulated after hookworm infection (9.6-fold decrease in the mean value) (Figure 4L).
In Trial 1, PBMCs that were restimulated with NaES from hookworm-infected individuals but not uninfected controls produced Th2 cytokines (Figure 2). PBMCs and duodenal biopsies were cultured with NaES or MED alone from all individuals in Trial 2 before and after hookworm infection. Supernatants from both cultures were taken for soluble cytokine analysis, and restimulated biopsies were taken after culture for RNA preparation and qPCR. In pre-infection biopsies (wk0), there was no significant difference in IL-2, IFN-γ, TNF-α, IL-17A, IL-4, IL-5, IL-10 and IL-13 produced after restimulation in culture with NaES compared to medium only (Figure S2). When PBMCs from hookworm-infected participants were restimulated with NaES they produced IL-4, IL-5 and IL-13 (Figure 5A, D, G), as previously shown (Figures 2A–C). We then extended these studies to show that restimulated biopsies also produced these cytokines, both at the levels of secreted protein (Figure 5B, E and H) and RNA transcripts (Figure 5C, F and I), although we did not detect a change for IL-4 transcript levels. PBMCs and biopsies from infected individuals also produced IL-9 and IL-10 in response to NaES (Figure 5J–O), however increased IL-10 production to NaES was not detectable by qPCR.
The qPCR data from unstimulated biopsies taken before and after hookworm infection indicated that infection may induce a Th1 response, while suppressing a Th17 response (Figure 4). We also assessed the levels of inflammatory cytokines produced by PBMCs and biopsy cultures when stimulated with NaES. As shown in Figure 6, restimulation with NaES induced upregulation of the proliferative cytokine IL-2 in both PBMCs and biopsy cultures at the protein level (Figure 6A and B), but corresponding levels of mRNA were too low to detect by qPCR (Figure 6C). Restimulation with NaES also induced production of the Th1 cytokine IFN-γ from PBMCs (Figure 6D), but not from biopsy cultures, either at the protein (Figure 6E) or RNA (Figure 6F) levels. We did not detect upregulation of the Th17 cytokine IL-17A from cultures of PBMCs or biopsies restimulated with NaES (Figure 6G–I). The accumulation of mRNA transcribed by another T cell proliferative cytokine gene, IL-15 (Figure 6J), the immunosuppressive cytokine gene TGF-β (Figure 6K) and the wound healing cytokine IL-22 (Figure 6L), were also increased in NaES restimulated biopsies. Again, biopsies taken prior to hookworm infection did not produce upregulated expression of any of these cytokines upon NaES stimulation (Figure S2).
Here, we present the first description of the mucosal cytokine response of healthy humans to either an experimental or naturally acquired helminth infection. As all individuals in this trial were infected when in established remission of their celiac disease whilst maintaining a strict gluten-free diet, and all analyses presented in this study were performed with blood and tissue collected prior to gluten challenge (excepting data acquired from the adult hookworm attachment site, Figure 3), we regard the subjects as being representative of normal, healthy individuals, and treat our results accordingly.
The polarisation of the T cell response in hookworm infection is of some debate, with some studies showing a mixed Th1/Th2 response, while others report only a polarised Th2 response [12], [26]. These conflicting results might be explained by differences in methods used to assess cytokine levels and antigen preparation [26]. We found a robust Th2 response produced to hookworm antigen as expected, with some evidence of a systemic, but not mucosal hookworm-specific Th1 response. Using qPCR with unstimulated biopsies taken before and after hookworm infection, we showed upregulation of IFN-γ transcripts. However, restimulation of post-infection biopsies with NaES did not result in increased IFN-γ production. Thus, although an innate IFN-γ response develops after hookworm infection, we could not identify antigen-specific memory Th1 cells in the duodenal mucosa.
We did not detect a significant alteration of levels of mRNAs encoded by the Th17 cytokine IL-17A or transcription factor RORγt in PBMCs or biopsies. This is surprising, as we observed increased levels of ALDH1A2 and IL-15 mRNAs in unstimulated mucosa with further enhancement of IL-15 mRNA accumulation following stimulation with NaES. Although retinaldehyde dehydrogenase (encoded by ALDH1A2) produces the normally immunosuppressive retinoic acid, which imprints gut homing on T cells [27] and switches the pro-inflammatory Th17 to a regulatory response [28], in active celiac disease the impact appears quite the opposite whereby elevated retinoic acid and IL-15 promotes Th1/17 responses in the gut mucosa [29]. We did not detect an increase in Treg markers in response to hookworm infection in this study, or when previously measured by FACS and immunohistochemistry [22], despite a wealth of evidence of Treg induction following experimental helminth infections and administration of ES proteins in mouse models [30]–[32], and naturally acquired infections of humans with other helminths [33], [34].
In active celiac disease, IL-15 is considered a crucial cytokine in maintaining autoimmune (Th1/Th17) pathology, a relationship now recognized somewhat incongruously as being dependent on retinoic acid [29]. In isolation, our IL-15 and ALDH1A2 data would argue against the potential for hookworm infection to protect against gluten toxicity in celiac disease, the primary incentive for undertaking this clinical trial [21], [22]. However, in contrast to the increased accumulation of IL-2, IL-15 and ALDH1A2, accumulation of mRNA encoding the innate Th17-inducing and stabilising IL-23 was potently suppressed by hookworm infection, potentially neutralising the impact of these Th17 promoting cytokines. Consistent with suppression of IL-23 mRNA accumulation, Th17 inflammation did not occur. IL-23 is produced by antigen presenting cells under the influence of microbial signals, and is a key cytokine in driving intestinal inflammation [35]. Moreover, IL-23 was recently shown to induce production of pro-inflammatory cytokines by innate lymphoid cells in the gut of patients with Crohn's disease [36]. Thus we hypothesise that hookworm infection suppresses pro-inflammatory cytokine production (such as IL-23) by innate cells in the gut, similarly to that seen in H. polygyrus infection in mice where a suppressive dendritic cell subset is expanded in the mucosa [37], and ES proteins from the parasite suppress activation of these cells and subsequent cytokine production [38].
IL-22, an IL-23 dependent Th17 cytokine, acts via the IL-22R expressed on intestinal epithelial cells, promoting innate immunity against bacteria, cell regeneration and tissue healing. In inflammatory bowel disease, high levels of IL-22 are present in inflamed tissue [39]. Interestingly, Broadhurst et al. described a case study of a patient with ulcerative colitis who deliberately ingested thousands of eggs of the whipworm Trichuris trichuria in which infection ameliorated disease activity, and this effect correlated with increased expression of Th2 cytokines and IL-22 [23]. In our study, biopsies from patients infected with a small number of N. americanus larvae showed upregulation of IL-22 mRNA levels after restimulation with NaES in vitro. It is beyond the scope of this discussion to attempt to define what cytokine milieu determines when and what Th17 complex drives inflammation or regulation. It does seem, however, that helminth-stimulated IL-22, perhaps derived from a non-Th17 source, such as NK cells or CD11c+ cells, contributes to the biological relationship between parasite and host, whilst conditioning and promoting a less inflammatory phenotype [23], [40], [41].
Both the regulatory cytokines, TGF-β and IL-10, were induced by hookworm infection, but this was only evident in mucosa restimulated with NaES. During H. polygyrus infection in mice, Th2 responses are induced in lamina propria T cells, and Th1 responses in these T cells are inhibited by parasite-induced TGF-β- and IL-10-producing T cells [42]. We may have identified a similar regulatory process, possibly adapted to further fine tune Th2 associated damage at the hookworm attachment site. If the rate of progression and the severity of damage to the mucosa accompanying the worm's attachment is central to determining which population of parasites a particular host will sustain, as has been suggested in an earlier endoscopic study of N. americanus survival in experimentally infected humans [3], these inflammation-modifying cytokines almost certainly have a role.
Herein we characterised the systemic and mucosal immune responses to an anthropophilic hookworm infection. As expected, we detected a systemic and mucosal hookworm-specific Th2 response in experimentally infected people. Our data indicate that although an antigen-specific Th1 response was detectable in the blood, no antigen-specific IFN-γ was detectable in the mucosa. Therefore the increased IFN-γ we detected in the mucosa most likely comes from an innate source, possibly NK cells [24]. Despite enhanced production of IL-15 and ALDH1A2, levels of IL-23 were dramatically suppressed after hookworm infection, possibly accounting for the absence of a Th17 response via suppression of antigen presenting cell function.
|
10.1371/journal.pgen.1004299 | Rosa26-GFP Direct Repeat (RaDR-GFP) Mice Reveal Tissue- and Age-Dependence of Homologous Recombination in Mammals In Vivo | Homologous recombination (HR) is critical for the repair of double strand breaks and broken replication forks. Although HR is mostly error free, inherent or environmental conditions that either suppress or induce HR cause genomic instability. Despite its importance in carcinogenesis, due to limitations in our ability to detect HR in vivo, little is known about HR in mammalian tissues. Here, we describe a mouse model in which a direct repeat HR substrate is targeted to the ubiquitously expressed Rosa26 locus. In the Rosa26 Direct Repeat-GFP (RaDR-GFP) mice, HR between two truncated EGFP expression cassettes can yield a fluorescent signal. In-house image analysis software provides a rapid method for quantifying recombination events within intact tissues, and the frequency of recombinant cells can be evaluated by flow cytometry. A comparison among 11 tissues shows that the frequency of recombinant cells varies by more than two orders of magnitude among tissues, wherein HR in the brain is the lowest. Additionally, de novo recombination events accumulate with age in the colon, showing that this mouse model can be used to study the impact of chronic exposures on genomic stability. Exposure to N-methyl-N-nitrosourea, an alkylating agent similar to the cancer chemotherapeutic temozolomide, shows that the colon, liver and pancreas are susceptible to DNA damage-induced HR. Finally, histological analysis of the underlying cell types reveals that pancreatic acinar cells and liver hepatocytes undergo HR and also that HR can be specifically detected in colonic somatic stem cells. Taken together, the RaDR-GFP mouse model provides new understanding of how tissue and age impact susceptibility to HR, and enables future studies of genetic, environmental and physiological factors that modulate HR in mammals.
| Cancer is a disease of the genome, caused by accumulated genetic changes, such as point mutations and large-scale sequence rearrangements. Homologous recombination (HR) is a critical DNA repair pathway. While generally accurate, HR between misaligned sequences or between homologous chromosomes can lead to insertions, deletions, and loss of heterozygosity, all of which are known to promote cancer. Indeed, most cancers harbor sequence changes caused by HR, and genetic and environmental conditions that induce or suppress HR are often carcinogenic. To enable studies of HR in vivo, we created the Rosa26 Direct Repeat-Green Fluorescent Protein (RaDR-GFP) mice that carry an integrated transgenic recombination reporter targeted to the ubiquitously expressed Rosa26 locus. Being able to detect recombinant cells by fluorescence reveals that the frequency of recombination is highly variable among tissues. Furthermore, new recombination events accumulate over time, which contributes to our understanding of why our risk for cancer increases with age. This mouse model provides new understanding of this important DNA repair pathway in vivo, and also enables future studies of genetic, environmental and physiological factors that impact the risk of HR-induced sequence rearrangements in vivo.
| DNA is constantly subjected to endogenous and environmental DNA damaging agents that can lead to toxicity, mutations, and ultimately disease [1]. Maintaining genomic stability in the face of the thousands of DNA lesions that are formed in each cell every day poses a major challenge, especially in the case of double strand breaks (DSBs), which are acutely toxic and can lead to the loss of millions of base pairs if a portion of a chromosome is lost [1], [2]. The two major pathways used by cells to repair DSBs are non-homologous end-joining (NHEJ), which directly rejoins DNA ends, and homologous recombination (HR), which requires a homologous duplex for DSB repair [3]–[8]. The correct balance of NHEJ and HR is essential for preventing genomic instability [4], [9]. If there is a deficiency in HR (e.g., loss of function of BRCA2), cells can suffer misrepair of DSBs, resulting in cytotoxicity and translocations that promote cancer and aging [10]–[12]. Ironically, despite the fact that HR is essential, too much HR can also be detrimental, since HR carries the risk of misalignments that cause insertions, deletions, as well as loss of heterozygosity (LOH) [13], [14]. It is likely that HR events contribute to sequence changes in virtually all cancers, since loss of function of almost all tumor suppressor genes requires LOH, and many, if not most, LOH events are caused by HR [14]–[16]. Further, sequence changes generated by HR have been found in multiple cancers [17]–[22], and many conditions that promote HR also promote cancer (as a few examples, exposure to UV light [23], [24], exposure to benzo[a]pyrene [25], [26] and mutations in BLM [27] and Ku70/80 [28], [29]).
Dozens of genes are either directly involved in HR or modulate HR activity [6], [30]. An essential early step in HR is the resection of double strand ends to create a 3′ single stranded overhang [31], [32]. Subsequently, BRCA2 helps to load RAD51 onto the single stranded DNA to form a nucleoprotein filament that is capable of homology searching [33]–[37]. Strand invasion leads to formation of a D-loop that is then either resolved by synthesis-dependent strand annealing, which is not associated with crossovers, or by second-end capture and formation of a double Holliday junction, which may or may not be associated with a crossover [5], [30], [38]–[40]. Although crossovers during HR are relatively rare [4], [41], HR-associated crossovers have been shown to cause LOH [14]–[16], [20]. In addition to its important role in the repair of two-ended double strand breaks, HR is essential for repair of one-ended double strand breaks that arise as a consequence of replication fork breakdown [5], [30], [42]. In HR deficient cells, such broken ends cannot be faithfully repaired via reinsertion into the sister chromatid, leading to an increase in misrepair via joining to an inappropriate end [4], [9], [30]. Despite HR's critical role in maintaining genomic stability, little or nothing is known about HR activity in most tissues in vivo, due to the lack of effective tools for studying HR in mammals.
Using mouse models that harbor sequences amenable to studies of HR, key insights about HR in vivo have been gleaned for certain cells types and tissues. In pioneering work by the Schiestl laboratory, pun mice, which carry a natural duplication wherein a change in pigmentation indicates an HR event, have been used to study the impact of genes and exposures on HR [43], [44]. Additionally, mice engineered to be heterozygous at the Aprt locus have been used to show that LOH is often driven by HR in vivo [45], [46]. More recently, our laboratory set out to create mouse models in which HR can be detected via direct repeat HR reporters.
Studies in S. cerevisiae first demonstrated that direct repeat substrates are useful for studying HR [47]–[49]. Briefly, two expression cassettes for a selectable marker are integrated into the genome adjacent to each other. Each expression cassette lacks sequences that are essential for expression. If the expression cassettes misalign and undergo homologous recombination, sequence information can be transferred from one cassette to the other, which can reconstitute full-length sequence to enable expression of the selectable marker (e.g., Figure 1A; black bars indicate deleted sequences). Studies exploiting direct repeat HR substrates in mammalian cells have given rise to fundamental information about the mechanism of HR as well as the impact of sequence orientation, distance between repeats, and exposures on HR [50]–[53]. The Nickoloff laboratory incorporated a site for the homing endonuclease I-SceI, which creates a double strand break that induces HR. Controlling the position of the double strand break gave rise to additional insights into the underlying mechanisms of HR [54], [55]. More recently, the Jasin laboratory designed HR substrates wherein a site-specific double strand break induces HR events that can be detected by expression of EGFP [56], and these assays have been used extensively to reveal the genetic underpinnings of HR [4]. We later created a plasmid-based fluorescence recombination assay which was used for studies of the impact of inflammatory chemicals on HR [57]. To move from in vitro studies to in vivo studies, we subsequently used elements of the plasmid assay to create a fluorescence-based direct repeat HR substrate in mice. The fluorescent yellow direct repeat (FYDR) mice carry a direct repeat substrate wherein HR can lead to the reconstitution of the full-length coding sequence of the enhanced yellow fluorescent protein (EYFP) gene [58], [59]. The FYDR mice are the first genetically engineered animal model that specifically detects HR, and the FYDR HR substrate intentionally does not include a site for artificial introduction of a double strand break (e.g., via I-SceI), since our primary objective is to enable studies of environmental, genetic and physiological factors that modulate HR.
The use of fluorescence has proved to be an effective approach for detecting HR in the FYDR mice in vivo [58], [60]–[63]. As expected [50], spontaneous recombination at the HR substrate is rare (the frequency of recombinant cells is ∼1/105) [58], [59]. Nevertheless, the frequency of recombinant cells can be quantified by flow cytometry, and a fluorescent readout makes it possible to identify the cell types that have undergone HR within intact tissue via histological analysis. Furthermore, independent recombination events (as opposed to frequency of cells harboring recombinant DNA) are detectable as fluorescent foci in freshly excised intact tissue by imaging whole organs [60], [61]. To learn more about the factors that impact the frequency of recombinant cells, we also developed a 3D imaging platform for intact tissue, which made it possible to determine how many recombinant cells result from de novo recombination events versus cell division [64]. These studies showed that both de novo recombination and clonal expansion drive the accumulation of recombinant cells with age [61], [64]. Taken together, studies using the FYDR mice show that fluorescence detection of HR in vivo provides valuable insights into genetic, environmental and physiological factors that modulate HR [58]–[60], [62], [63]. Importantly, however, only a limited number of tissues can be studied in the FYDR mice as a consequence of poor expression in some tissues (presumably due to the random locus integration following pronuclear injection) [58], [65]. We therefore set out to generate a recombination reporter mouse with broad reporter expression.
In order to create a mouse model in which HR can be studied in virtually any cell type, we created targeting vectors to enable integration of a direct repeat recombination reporter into the Rosa26 locus [66]. Here we describe the Rosa26 Direct Repeat-Green Fluorescent Protein (RaDR-GFP) mice, which harbor two uniquely truncated EGFP expression cassettes in tandem. HR at the direct repeat can reconstitute full-length EGFP coding sequence, giving rise to fluorescence (Figure 1A). Using this system, we were able to quantify HR in all tissues tested using flow cytometry. Furthermore, we show that several tissues are susceptible to DNA damage-induced HR, and using a novel automated image analysis program for analysis of fluorescence within intact tissue, we show that HR events accumulate in the somatic stem cells of the colon. The RaDR-GFP mice therefore open doors to studies of exposure-induced HR and make it possible to perform an integrated analysis of how cell type, tissue type and age impact HR in vivo. Together with the development of quantitative approaches for assessing HR, the RaDR-GFP mice enable studies of how genetic and environmental factors modulate susceptibility to HR events in cancer-relevant tissues.
To study recombination in vivo, we previously created a direct repeat substrate in which two EGFP expression cassettes are positioned in tandem (Figure 1A) [66]. Essential sequences were deleted from each of the EGFP cassettes to create Δ5egfp, which lacks 15 bp at the 5′ end, and Δ3egfp, which lacks 81 bp at the 3′ end. Recombination between the non-functional expression cassettes can reconstitute full-length coding sequence, which can then be expressed under the CMV enhancer/chicken beta-actin promoter [CAG] (Figure 1B) [66], [67]. The promoter, intron, and polyadenylation signal sequences are the same as for the established FYDR mouse model [58]. In the FYDR model, expression levels were high in some tissues (such as pancreas), but there was almost no expression in other tissues (such as the colon), presumably as a consequence of gene silencing associated with the locus of integration.
To enable broad expression, we targeted the HR reporter to the Rosa26 locus, which was originally identified for its nearly ubiquitous expression [68]. Using a Rosa26 targeting construct (a kind gift from Dr. P. Soriano) [68], we previously created a targeting vector that includes a short arm (SA), a positive selection marker (NeoR), a direct repeat HR substrate, a long arm (LA), and a negative selection cassette (diphtheria toxin fragment A; DTA) (Figure 1C) [66]. The construct design strategy is shown in Figure S1. While our prior studies were focused on HR in ES cells in vitro, here we set out to create a knock-in mouse. The targeting construct was electroporated into mouse 129S4/SvJae (129 background) ES cells. Out of 100 colonies, we identified seven candidates using primers designed to yield a 1.16 kb product from wild type DNA and a 1.24 kb product from the targeted allele (Figure 1C–D). Five out of seven candidates harbored the diagnostic 8.2 and a 2.3 kb HindIII fragments when analyzed by Southern blot (Figure 1C and 1E). Ten to fourteen 129 ES cells were injected into 3.5-day-old C57BL/6 blastocysts, and the resultant chimeric males were bred with 129 females to establish the RaDR-GFP mouse line. While the 129 background was maintained, the transgene was also backcrossed into the C57BL/6 background for 10 generations. The transgene follows Mendelian inheritance with 49.5% of offspring of heterozygous/wild type parents inheriting the transgene (n = 99).
To initiate studies of HR in the RaDR-GFP mice, we first analyzed primary ear fibroblasts. Cells were harvested, expanded in culture, and examined by flow cytometry. Gates defining ‘green fluorescent’ and ‘autofluorescent’ cells were drawn conservatively to prevent autofluorescent from being identified as fluorescent, while capturing the majority of the EGFP expressing cells (Figure 2B).
To formally determine whether or not green fluorescent cells had indeed undergone HR, we isolated fluorescent cells to learn if they harbor full-length EGFP coding sequence. We previously designed PCR primers that specifically amplify Δ3egfp, Δ5egfp, or full-length EGFP (Figure 2A and Table S1). Here, we developed methods to analyze cells for the presence or absence of each cassette using cDNA as a template, rather than genomic DNA as previously described [66]. Our rationale for this approach was that by exploiting the multiple copies of cassette sequences present in mRNA, we would be able to query the presence and absence of cassettes in single cells in future experiments. As a first step, primers were used to analyze cDNA from control ES cell lines that had previously been targeted with each cassette individually, as well as ES cells that harbor both Δ3egfp and Δ5egfp [66]. Conditions were optimized so that both Δ3egfp and Δ5egfp are detectable in a single PCR reaction so that each cassette serves as a positive control for the other. Results show specific detection of each cassette in isolation and together, and full length sequence is only observed in the positive control EGFP expressing cells, as expected (Figure 2C, first five panels). To create the RaDR-GFP mice, we created new early passage clones of ES cells targeted with the recombination substrate. PCR analysis of RaDR-GFP cells that carry the unrecombined substrate reveals both the Δ3egfp and Δ5egfp cassettes, but not the full length EGFP, as expected (Figure 2C, panel six).
Having created RaDR-GFP mice that carry the Rosa26 targeted HR substrate (Figure 1C–E), we next set out to determine whether or not fluorescent cells from these animals indeed harbor the full length EGFP sequence, as anticipated following HR. Fluorescent and autofluorescent control cells were isolated from a single cell suspension of disaggregated RaDR-GFP pancreatic cells using FACS (Figure 2B). Primers that flank the coding sequence were optimized for nested PCR (Table S2), and cDNA was analyzed either by direct PCR or nested PCR, as indicated. Analysis of autofluorescent RaDR-GFP pancreatic cells revealed the presence of Δ3egfp and Δ5egfp, whereas full-length EGFP sequence was not detected (Figure 2D). In contrast, full-length EGFP was readily detected in samples of green fluorescent RaDR-GFP pancreatic cells (Figure 2D). The Δ3egfp and Δ5egfp cassettes were also detected (Figure 2D), which is consistent with their potential retention following HR (Figure 1A). The RaDR-GFP HR substrate is designed so that over a dozen base pairs need to be restored to give rise to a functional full-length EGFP coding sequence [66]. As restoration of a significant number of nucleotides requires HR for alignment and transfer of sequence information, these data show that fluorescence is an indicator of homologous recombination at the RaDR-GFP substrate.
Ultimately, this mouse model can be used to study the underlying molecular changes that caused sequences to be restored to full length. Gene conversions without a crossover can be identified by the presence of one of the two original cassettes, along with full-length sequence. In contrast, replication fork repair or gene conversion with crossover will result in a triplication wherein both of the original cassettes are present along with the full-length sequence (Figure 3). We had previously performed this type of analysis on ES cells that had been clonally expanded in vitro [66]. Here, we set out to develop methods that would enable studies of HR in vivo. Because clonally expanding single cells from mouse tissues is difficult, we set out to develop methods that would enable analysis of single fluorescent cells isolated from mouse tissues using FACS. Initial data indicate that single cell analysis can indeed be used to identify cells with each of the three major recombination classes (Figure S2B).
Previous studies of FYDR positive control mice (which express EYFP from the same promoter and locus as the HR reporter) show that there is little or no expression of EYFP in many tissues (presumably due to silencing), which greatly limits the utility of the FYDR model [65]. While we anticipated that targeting the EGFP direct repeat reporter to a site with ubiquitous expression would overcome this barrier to studies of HR, prior studies of expression at the Rosa26 locus had been done using the Rosa26 promoter [68], whereas the CAG promoter drives the RaDR-GFP transgene. To address the formal possibility that EGFP expression from the RaDR-GFP reporter might not be ubiquitous, we assessed the extent of expression of EGFP from a positive control mouse in which EGFP is expressed specifically from the CAG promoter at the Rosa26 locus (see Materials and Methods for details). Analysis of tissues from the FYDR positive controls showed high expression of EYFP in the pancreas, and low expression in the liver and the colon (Figure 4A, upper row), which is similar to the low expression previously observed in the kidney and lung [65]. In contrast, expression of EGFP in the Rosa26 positive control mice was very strong in all three tissues (Figure 4A, bottom row). By using the same imaging parameters, these data also show that fluorescence from EGFP is significantly brighter than that of EYFP. Analysis by flow cytometry similarly shows that EGFP fluorescence is high not only in pancreas, liver and colon (Figure 4B), but also in eight additional tissues (Table 1). The nearly ubiquitous expression of EGFP in the positive control mice suggests that fluorescent recombinant cells in the RaDR-GFP mice would be detectable in most mouse tissues. Furthermore, the positive control mice are essential for comparisons of HR frequency among tissues, since the frequency of GFP positive cells in the positive control mice provides the required baseline for comparing HR frequencies among tissues in the RaDR-GFP mouse model.
To explore the feasibility of studying HR in multiple tissues (including tissues that had previously been inaccessible to HR analysis), 11 tissues from RaDR-GFP mice were disaggregated and analyzed by flow cytometry, first by gating for live cells, and subsequently by gating for fluorescent cells. Remarkably, fluorescent recombinant cells were present in all tissues (Figure 4C). Recombinant cells were relatively frequent in the pancreas (similar to the FYDR mice) and in the spleen. Recombinant cells were also observed at a significant frequency in the kidney, heart, liver, mammary gland, and colon of the RaDR-GFP mice (all of which had previously been inaccessible for studies of HR within mammalian tissues in vivo). In contrast, very few fluorescent cells were detected in stomach or brain tissue (Figure 4C). The observation that ∼90% of cells from brain tissue of the Rosa26 positive control mice are fluorescent (Table 1) indicates that fluorescent recombinant cells can be detected. These results together therefore show that there are very few recombinant cells in the brain (note that the detection of rare fluorescent cells is limited to ∼1/106). One possible explanation for the low frequency of EGFP positive cells in the brain is the short time period during which HR is active in the developing brain [69], where it plays a critical role in neurogenesis and cancer suppression [70]. It is possible that relatively few recombinant cells accumulate in the RaDR mouse brain compared to other tissues due to the short time during which HR is highly active. Although further studies are needed for a more in depth understanding of HR among tissues, taken together, these studies show for the first time that spontaneous HR is pervasive in adult mammalian tissues.
Our previous studies, as well as results presented here, show that recombinant cells can be detected in situ within intact pancreata of FYDR mice as fluorescent foci (Figure 5A) (see [60], [61]). Importantly, since recombination is a rare event and pancreatic cells do not migrate significantly, independent recombination events can be identified as isolated fluorescent foci. Analysis of recombination events provides greater sensitivity compared to the frequency of recombinant cells as a means for detecting genetic and environmental factors that modulate HR [65].
To explore the efficacy of RaDR-GFP mice for studies of HR events within intact tissue, pancreatic tissue from a RaDR-GFP mouse was stained with DAPI and imaged using fluorescence microscopy at low magnification (×1). Fluorescent foci are readily apparent in the RaDR-GFP pancreatic tissue (Figure 5B). Tissue from 11 RaDR-GFP mice was compressed to 0.5 mm and imaged for manual quantification of foci. Using this approach, we observed that the median frequency of spontaneous recombination events is ∼140/cm2. In addition, unlike the FYDR mice, recombinant foci are also readily detected in both the intact liver and the intact colon (Figure 5C).
Differences in the frequency of foci among tissues reflect both the frequency of HR events as well as the optical properties of each tissue. Therefore, it is difficult to discern tissue-specific differences in HR using this approach (note that flow cytometry of disaggregated tissues overcomes this limitation). Importantly, however, for studies of factors that modulate HR in a specific tissue, analysis of HR events in situ provides a powerful approach both in terms of increased sensitivity [65] and in terms of learning about HR in specific cell types (see below).
Although HR events are rare, it is nonetheless possible to identify fluorescent foci within frozen 5 µm sections using epifluorescence microscopy. After imaging, sections can be stained with hematoxylin and eosin (H&E) to reveal tissue architecture. Image overlays for pancreatic fluorescent foci reveal that for both FYDR and RaDR-GFP, recombination is detected in pancreatic acinar cells (Figure 5A and 5C, right). These observations are consistent with studies of FYDR mice in which analysis of >100 pancreatic foci revealed only acinar cells [61]. In the case of liver and colon, overlay of fluorescent images with H&E images reveals fluorescent hepatocytes in the liver, and fluorescent epithelial cells in the colon (Figure 5C). Pancreatic acinar cells, liver hepatocytes and colonic epithelial cells all give rise to tumors in their respective tissues, raising the possibility that the RaDR-GFP mice can be used to study the etiology of cancer (see Discussion).
Somatic stem cells are of particular interest in cancer research. In the colon, there are only one or a few somatic stem cells at the base of each colonic crypt. Somatic stem cells are defined as being cells that have the ability to give rise to the epithelial layer in that crypt [71]–[73]. Therefore, a single HR event in a colonic somatic stem cell can lead to “crypt conversion” wherein all of the epithelial cells of its crypt share the same genetic change (Figure 5D). Since transit cells are short lived, lasting only a few days before the epithelial layer of the crypt is replaced [73], mutations in transit cells are less likely to contribute to cancer compared to mutations in colonic somatic stem cells, which can persist throughout the lifetime of the animal [73].
Analysis of thin sections via epifluorescence microscopy revealed a cross section of a colonic crypt in which it appears that all of the central epithelial cells are fluorescent (Figure 5C, bottom right), suggesting that a stem cell from this crypt replaced the crypt epithelial cell layer with fluorescent daughter cells (crypt boundaries can be identified by a ring of epithelial cells with higher staining intensity; Figure 5C). To learn more about the possibility of crypt conversion, colonic tissue was processed to gently remove crypts. Intact wholly fluorescent crypts were readily identified among disaggregated crypts from RaDR-GFP mice (e.g., Figure 5E), which is consistent with replacement of crypt epithelial cells by a single somatic stem cell that had undergone HR at the RaDR-GFP substrate. Taken together, the RaDR-GFP mice enable studies of HR in a cell type that is highly relevant to colon cancer.
Aging is a critical risk factor for almost all cancers. To learn about the potential for recombinant cells to accumulate with age in the colon, we imaged and analyzed colonic tissue from young (3–4 months old) and old (9–10 months old) animals. Foci were counted by eye in a blinded fashion, and results indicated that there was no significant difference in the frequency of recombinant cell foci between the young and old animals (Figure 6D, left). Foci in colonic tissue appear both as a consequence of transit cell recombination and somatic stem cell recombination. Given that transit cells are only present for a few days, unless the rate of recombination changes for young and old animals, one would not anticipate an observable increase in the frequency of transit cell foci. In contrast, as described above, somatic stem cells can persist for years [73], which raises the possibility that fluorescent foci that result from recombination events in stem cells would accumulate and be detectable by the presence of whole crypt conversion. In order to favor detection of HR events in somatic stem cells, we therefore set out to create an image analysis program that differentiates large foci (more likely to be due to whole crypt conversion) from small foci (more likely to be the result of HR in transit cells).
We created a foci counting program that favors detection of large foci by using automated quantification techniques that exploit both intensity and morphological features. Classification was enabled using support vector machines. We simulated the data using a noise model, which includes the homogenous noise of the sample as well as the detection noise, to analyze the performance of our algorithms. To avoid false positives, only large foci with a consistent morphology and intensity were counted, and small foci or irregularly shaped foci were excluded (Figure 6A). Although this approach has a potentially high false negative frequency, it is more important to avoid false positives than false negatives. Analysis of the lumen of large samples of colonic tissue shows the clear appearance of bright foci (Figure 6B). Using the automated analysis software, large foci were marked with a dark cross if considered to be positive (Figure 6C). Direct comparison of Figure 6B and Figure 6C shows that the majority of the large foci are identified by the program. We validated this approach by comparing the automated counting results to manual counts. A more detailed description of this software will be published separately.
Using our image analysis software, we reanalyzed the colonic tissue from young and old mice. Remarkably, there is a highly significant (p<0.01) increase in the frequency of larger foci with age (Figure 6D, right). Since the largest foci result from clonal expansion of somatic stem cells, these results indicate that recombination events indeed accumulate in colonic somatic stem cells. It is noteworthy that inclusion of foci from transit cells is anticipated to lead to smaller foci that mask detection of changes in the more rare larger foci, as indicated in Figure 6D (left) such that inclusion of false positives damps the signal from the somatic stem cells. Taken together, these results provide some of the first insights into the relative susceptibility of transit cells and somatic stem cells to recombination with age, and open doors to future studies of conditions that modulate the risk of recombination in cells that have the potential to give rise to cancer.
Alkylating agents are carcinogenic, used for cancer chemotherapy, and have been shown to be recombinogenic in mice [74], [75]. We were therefore interested in the extent to which RaDR-GFP tissues would be susceptible to exposure-induced HR. Here, we focused on methylnitrosourea (MNU), a model SN1 alkylating agent similar to temozolomide, which is used in cancer chemotherapy [74]. In parallel ongoing studies of FYDR mice, we tested multiple exposure conditions for efficacy in inducing HR, and we found that the combination of MNU and thyroid hormone (T3), which impacts pancreas physiology, was the strongest inducer of HR among the conditions that we tested. We therefore asked whether or not the RaDR-GFP model is sensitive to exposure-induced HR by treating animals with combined MNU/T3 (see Materials and Methods). In addition to pancreas, we also evaluated colon and liver (Figure 7A). For all three tissues, MNU/T3 was a strong inducer of HR. For the pancreas, the increase in the frequency of de novo recombination events was most dramatic (Figure 7B), making it infeasible to quantify recombinant foci manually. Automated image analysis using a modified version of our foci analysis program (optimized for the pancreas) enables quantification of small/faint foci that are difficult to quantify by eye (Figure 7C). Furthermore, the automated foci counting program enables future studies of foci characterization based on size and other morphological characteristics. Automated quantification of foci in RaDR-GFP mouse pancreata shows that, on average, exposure to MNU/T3 leads to ∼400 new recombination events per cm2 (Figure 7D). In addition to the pancreas, exposure-induced HR was also observed in the liver and colon of RaDR-GFP mice (Figure 7D). Taken together, these results demonstrate the efficacy of RaDR-GFP mice for studies of exposure-induced HR in multiple tissues.
Although HR is essential [76], [77], its activity must be carefully controlled in order to maintain genomic integrity [30], [70]. Inherent defects that either suppress or induce HR are known to be tumorigenic [11] and exposures that induce HR are often carcinogenic [22], [44]. Despite its importance, progress in our understanding of the role of HR in mammals has been hampered by the lack of effective tools for studying HR in many mammalian tissues. Here, we describe the RaDR-GFP mice, which harbor an integrated direct repeat that causes cells to fluoresce following HR. By targeting the reporter to the Rosa26 locus, expression of the transgene is nearly ubiquitous, thus enabling studies of HR in nearly all major organs, including liver, colon, spleen, heart, lung, kidney, stomach, thymus, brain, breast, and pancreas, many of which have been hitherto inaccessible for analysis.
HR events at the RaDR-GFP substrate can occur via several different mechanisms. Prior studies of ES cells show that most recombinant fluorescent RaDR-GFP cells have undergone gene conversion without crossovers [66], which are thought to result primarily from the synthesis dependent strand annealing pathway (see [5], which includes animations for HR pathways). DSB-induced crossovers between sister chromatids can also be detected by the RaDR-GFP substrate. Importantly, one of the critical roles of HR is to repair one-ended DSBs at broken replication forks, and these events can readily be detected using the RaDR-GFP substrate (Figure 3). One challenge when using the direct repeat approach for studies of HR is that these canonical HR events can be overshadowed by single strand annealing (SSA), a subpathway of HR that is the most frequent spontaneous event at a direct repeat [5], [66]. Specifically, when a DSB is formed between repeats, the ends are resected to reveal 3′ overhangs that can readily anneal to one another. As we are primarily interested in conditions that stimulate problems during replication, we designed the RaDR-GFP substrate so that SSA is not detected (Figure 3 shows that SSA gives rise to an expression cassette that harbors both of the original deletions). This approach enables studies of spontaneous and exposure-induced HR events that are less frequent at a direct repeat, yet biologically important, such as replication fork repair. Taken together, both DSBs and broken replication forks can lead to fluorescence in the RaDR-GFP model, thus providing a window into how mammalian cells respond to a broad range of conditions that impact genomic stability by either suppressing or inducing HR in vivo.
To learn about spontaneous HR in vivo, we quantified recombinant fluorescent cells in 11 different tissues and found that recombinant cells are present in all tissues studied. The frequency of recombinant cells is highly variable among tissues, ranging from very low in the brain and stomach, to very frequent in the pancreas and spleen. The observation that recombinant cells are relatively frequent in the pancreas suggests that HR is highly active in this organ (which is consistent with the studies of aging; see below). Interestingly, mutations in BRCA2, which plays a key role in initiating HR, are known to increase the risk of pancreatic cancer [30], [78]. Thus, for the pancreas, there is a correlation between HR activity and the potential for a defect in HR to contribute to cancer [79]. For some other tissues, the frequency of HR is either unexpectedly high, or unexpectedly low. In the case of the heart, which has a relatively low proliferative index, there are a surprisingly high number of recombinant cells. One possibility is that progenitor cells that gave rise to cardiac tissue underwent HR, leading to the appearance of recombinant fluorescent cells in the adult tissue. One way to differentiate HR during development versus in the adult animal is to monitor tissue during aging to see if HR is active in adult animals. In contrast to cardiac tissue, the stomach had an unexpectedly low frequency of recombinant cells. It is noteworthy that not all of the cells in the disaggregated stomach tissue from the positive control mice were fluorescent (∼75% were positive by flow cytometry). This means that for some cell types, HR will not give rise to fluorescence. Although beyond the scope of this particular study, knowledge about HR in specific cell types can be achieved through a comparison of EGFP expression in RaDR-GFP mice (yielding information about HR) and EGFP expression in the positive control mice (yielding the baseline frequency of cells in which HR can be detected).
As the RaDR-GFP mice age, the frequency of recombinant somatic stem cells increases in the colon. Being able to monitor the burden of recombinant cells is valuable for long-term studies of conditions that impact HR. The burden of cells harboring sequence changes is critical to cancer development since an increase in the frequency of cells harboring a tumorigenic mutation leads to an increased risk of subsequent tumor-promoting mutations. Interestingly, exposure to MNU/T3 induced hundreds of recombination events in the RaDR-GFP mice. In essence, the burden of recombinant cells in young DNA damage-exposed mice is similar to aged mice, calling attention to the burden of mutant cells as a commonality for these two key risk factors for cancer. Being able to monitor HR over time and in response to exposures shows that RaDR-GFP mouse model can be used for studies of long-term exposures and physiological factors that impact the burden of recombinant cells, thus providing insights into fundamental processes that promote cancer.
A key advantage of fluorescence as a marker for HR is that it is possible to reveal the underlying cell types that have undergone HR. Using a fluorescent overlay on H&E images, we observed fluorescent recombinant pancreatic acinar cells, liver hepatocytes and colonic epithelial cells. Knowledge about genomic stability in all three of these cell types is relevant to cancer. Although most pancreatic carcinomas are thought to originate from ductal cells [80], mutation of Kras in acinar cells can lead to neoplasia of the ductal phenotype [81], and furthermore there is evidence that acinar cells can undergo acinar to ductal transdifferentiation [82]. HR is also detectable in hepatocytes, which are precursors to hepatocellular carcinomas. Additionally, being able to study genetic change in vivo in the liver has broad implications, since liver genotoxicity is a major barrier in drug development [83]–[85]. In the colon, we observed HR in colonic epithelial cells. Most epithelial cells are rapidly sloughed off, making these cells unlikely targets for initiating mutations for cancer. In contrast, colonic somatic stem cells persist for years [72], [73]. Our observation that there are crypts in which all cells appear to be fluorescent is consistent with an HR event in a somatic stem cell or early daughter cell of that crypt. Interestingly, methods have previously been developed for visualizing cells that have lost Dlb-1 gene function in colon crypts [86]. In Dlb-1 heterozygous mice, LOH can lead to a positive crypt by any of several different mechanisms (e.g., point mutations, frameshifts, deletion, HR, etc.). An advantage of the RaDR-GFP substrate is that it is designed to specifically detect HR.
To learn about exposure-driven HR, we elected to exploit an alkylating agent that provides insights into the biology of cancer chemotherapeutics. The model agent MNU is an SN1 type methylating agent that generates methylated bases such as 3-methyladenine, 7-methylguanine and O6-methylguanine [74]. Several methylating agents creating these lesions are currently used in cancer chemotherapy including temozolomide, which is used to treat metastatic melanoma and malignant gliomas [87]. Importantly, HR activity contributes to resistance to methylating agents used in the clinic [87]. Furthermore, HR induced in healthy tissues during treatment with chemotherapeutic alkylating agents may be linked to therapy-induced secondary cancers [88]. Because of the broad reporter expression and sensitivity to methylation-induced HR, the RaDR-GFP mice offer a new approach for probing the extent to which treatments impact genomic stability both within the tumor and within healthy tissues, which is relevant to the risk of secondary cancers.
In addition to FYDR and RaDR-GFP mice, several other mouse models that harness fluorescence as a marker for HR have been developed, including the HPRTdupGFP, which is currently in development in the Noda laboratory and promises to offer its own advantages. In addition, the Jasin laboratory extended their studies of DSB-induced HR in vitro to an animal model. The DR-EGFP mice harbor a recombination reporter that carries sequences for site-specific cleavage by I-SceI, and thus enable studies of DSB-induced HR in cells cultured from that mouse [89]. Using this model, it has been shown that a deficiency in Brca1 leads to reduced HR in cultured cells, and that DSB-induced HR can be studied in various cell types in vitro using cells derived from disaggregated tissues of the DR-EGFP mouse. While the use of a homing endonuclease greatly increases the frequency of HR, making it easier to quantify, the endonuclease needs to be introduced in vitro, which is not compatible with studies of HR in vivo. Furthermore, the DR-EGFP reporter is integrated into the Pim-1 locus. In the absence of a positive control, it is not possible to assess the relative frequency of HR among tissues, since a low frequency of fluorescent cells may be due to either a lower rate of HR or suppressed expression of EGFP. In contrast, for the RaDR-GFP mice, it is possible to compare HR among tissues since the number of cells that potentially express EGFP can be deduced using a complementary positive control mouse line with the identical locus and promoter. Unlike the DR-EGFP studies of HR in cells that have been isolated from mice, the mice and the methods described here enable analysis of HR in cells within their normal tissue context in vivo, which enables studies of more complex physiological processes, including cancer development and chronic exposures.
Many mouse models have been developed for studies of point mutations/small deletions in vivo (Pig-a, MutaMouse, Big Blue, Plasmid lac-z, cII, Gpt-Δ [90]–[95]. For each of these mouse models, as well as for the RaDR-GFP mice, susceptibility to sequence changes is being monitored at a specific locus. Although vulnerability to sequence changes is anticipated to be locus dependent, these models nevertheless provide useful tools for assessing the impact of genetic and environmental factors that impinge on genomic stability. Unlike the transgenic models that are used to study point mutations, the RaDR-GFP model exploits fluorescence. The median frequency of fluorescent cells in RaDR-GFP tissues is approximately ∼2/105, whereas the frequency of point mutations is much more rare (∼1/108 per base pair) [1]. Consequently, strategies that exploit fluorescence to detect cells that have undergone a specific point mutation within intact tissue have not yet been described. Success in studies of point mutagenesis has been achieved by isolating DNA from mouse tissues, packaging the DNA into phage particles, and subsequently detecting mutation events via phenotypic change in E. coli [91]–[95]. This process is laborious, expensive, slow, and significant expertise is required in order to obtain reliable data, which together severely limit the utility of these models. In contrast, analysis of recombinant cells within intact RaDR-GFP tissue requires minimal expertise, can be performed with standard fluorescent microscopy, and requires much less time (e.g., processing one RaDR-GFP tissue takes minutes, as opposed the many days that are required for analysis of point mutations). Nevertheless, as the underlying factors that modulate point mutagenesis are very different from those that drive HR, methods that enable studies of point mutations and HR are highly complementary.
Intensive research in the past decade has given rise to sophisticated models for the molecular basis of HR, and has revealed that imbalanced HR contributes to genomic instability and cancer [75], [96], [97]. Here, we describe a novel mouse model that enables studies of HR in at least 11 different tissues. Here we show that HR is pervasive among mammalian tissues, that the frequency of HR is tissue-dependent, and that recombination events accumulate with age. The RaDR-GFP mice open doors to a wide range of studies. Knowledge about the extent to which HR is normally active in different tissue types is relevant to our understanding of how defects in HR lead to cancer in certain tissues. By crossing with genetically engineered mice, it is now possible to establish how specific genes impact HR throughout mammalian tissues, and furthermore how HR capacity impinges on cancer development. For example, the HR capacity of tumors that are anticipated to be HR deficient (e.g., those that arise in a Brca2+/− mouse model) can potentially be formally tested in vivo using the RaDR-GFP model. In terms of exposures, HR can be monitored over time, which makes this model compatible with studies of long-term environmental conditions that are relevant to human cancer risk. Furthermore, this model can serve as a tool in the development of cancer chemotherapeutics by providing a window into tissue specific effects. In particular, the risk of secondary cancers can be reduced by developing approaches that induce HR and associated genotoxicity in the tumor, while suppressing sequence rearrangements in healthy tissues. Additionally, in terms of cancer treatment, the RaDR-GFP mice make it possible to assess the efficacy of pharmaceutical agents that are designed to either suppress or induce HR in a tumor-specific fashion. Taken together, we have demonstrated how key processes, including tissue context, aging and exposure to a DNA damaging agent, impact the risk of HR in vivo. By creating new avenues for studies of HR in multiple tissues, the work described here enables future studies of genetic, environmental, and clinical conditions that impact genomic stability in mammals.
Plasmid construction was described previously [66]. Briefly, truncated EGFP coding sequences (Δ5egfp lacking 15 bases at the 5′ end and Δ3egfp lacking 81 bases at the 3′ end) were amplified by PCR from plasmid pCX-EGFP, using primers that each insert unique sequences. PCR products were cloned in a tandem orientation (Δ5egfp followed by Δ3egfp) into the pCX-NNX backbone to form the direct repeat HR substrate, yielding plasmid pCX-NNX-ΔGF. The HR substrate was then cloned into pBigT-TpA, released together with the neomycin resistance gene and cloned into pRosa26PA [68] (a kind gift from Dr. P. Soriano, Mount Sinai School of Medicine) to yield the targeting plasmid pRosa26-ΔGF (Figure S1).
All animals were housed and handled in Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC)-accredited facilities with diets, experimental methods, and housing as specifically approved by the Institutional Animal Care and Use Committee. The MIT CAC (IACUC) specifically approved the studies as well as the housing and handling of these animals.
The pRosa26-ΔGF targeting plasmid (Figure S1) was linearized by digestion with XhoI (New England Biolabs) and electroporated into mouse 129 embryonic stem (ES) cells. Clones were selected for resistance to G418 by growing in selective media (40% DMEM + glucose, 40% EmbryoMax DMEM, 1% β- mercaptoethanol, 15% FBS, penicillin, streptomycin, glutamine, nonessential amino acids, LIF, G418) and screened for correct targeting by PCR and Southern blot. Cells from clones with correct targeting were injected into the blastocoel of 3.5-day-old C57BL/6 blastocysts, which were implanted into pseudopregnant female mice. All ES cell manipulations and transgenic mouse development were performed by the ES Cell and Transgenics Facility at the Swanson Biotechnology Center of the Koch Institute for Integrative Cancer Research at MIT. All procedures involving mice were approved by the Massachusetts Institute of Technology Committee on Animal Care and in accordance with the National Institutes of Health guidelines for the humane care of animals.
To identify clones with correct targeting of the RaDR-GFP substrate, we used a forward primer annealing 5′ to the targeted locus and a reverse primer landing in the neomycin resistance gene within the construct, yielding a 1.24 kb PCR product (Figure 1C). In the absence of insertion, the forward primer yields a 1.16 kb PCR product with a reverse primer landing within the Rosa26 locus (Figure 1C). All primer sequences and exact PCR amplification conditions can be found in Tables S1, S2, S3. PCR detection of the Δ5egfp, Δ3egfp, and full-length EGFP sequences was performed as described previously [66].
Embryonic stem (ES) cells (104–106) or RaDR-GFP mouse pancreatic cells (∼1000) were lysed with 1 ml TRIzol (Life Technologies) and either stored at −80°C or processed immediately. Total RNA was extracted and column purified using the RNeasy Mini Kit (Qiagen). Briefly, TRIzol-lysed cells were mixed with 200 µl chloroform and centrifuged at 12,000 g for 15 min at 4°C. The aqueous phase was mixed with 500 µl ice-cold isopropanol and applied to an RNeasy column. The column was washed based on the manufacturer's protocol and RNA was eluted with 30 µl RNase-free water. Total RNA (500–2000 ng) was converted to cDNA with the SuperScript III First-Strand Synthesis System for RT-PCR (Life Technologies) with both random hexamers and oligo(dT). The volume was brought to 10 µl with RNase-free water and incubated at 65°C for 5 min before placing on ice for at least 1 min. Reverse transcriptase master mix was added and the reaction was incubated at 25°C for 10 min, 50°C for 50 min and 85°C for 5 min. Finally, E.coli RNase H (1 µl) was added and the reaction was incubated at 37°C for 20 min to remove RNA-cDNA duplexes before proceeding with PCR.
PCR detection of full-length EGFP sequences was performed with primers A FL FOR and C FL REV using Platinum Taq DNA Polymerase (Life Technologies). 5 µl 10× diluted cDNA was used as the template in the presence of 0.2 µM primers and enzyme mix according to the manufacturer's instructions. cDNA was denatured at 94°C for 3 min, and then incubated for 40 cycles at 94°C for 45 s, 56°C for 45 s and 72°C for 1.5 min. Reactions were then incubated at 72°C for 5 min and placed on ice. In order to detect Δ5egfp and Δ3egfp, two primer sets were used in a single reaction. Primers E D5 FOR2 and F D5 INT REV were used to detect Δ5egfp, and primers G D3 INT FOR and H D3 REV2 were used to detect Δ3egfp. Each reaction contained 0.2 µM primers. PCR reactions were incubated at 94°C for 3 min, and then at 94°C for 45 s, 55°C for 30 s and 72°C for 1 min 10 s for 40 cycles. Samples were incubated at 72°C for a final 5 min and placed on ice.
External PCR primers were designed to anneal upstream and downstream of the EGFP coding sequence. Primers (0.2 µM) BPEF3 and NEST Rev were added to Platinum Taq DNA polymerase mix with 5 µl 10× diluted cDNA following the manufacturer's protocol. Reactions were incubated at 94°C for 3 min, and then for 40 cycles at 94°C for 45 s, 58°C for 30 s and 72°C for 1 min 10 s. Reactions were ended with incubation at 72°C for 5 min and then placed on ice. PCR products were purified using the MinElute PCR Purification Kit (Qiagen) and eluted with the same volumes of EB buffer. Purified PCR products (5 µl) were used for subsequent full length EGFP PCR as described above. PCR products were analyzed by 1.5% agarose gel electrophoresis.
Single cells from RaDR mouse spleen were sorted by FACS into 5 µl lysis buffer (400 ng/µl proteinase K and 17 µM SDS in nuclease-free water). As a control, a single colony of RaDR-GFP ES cells was also added to lysis buffer. Cell lysates were freeze-thawed once at −80°C, and added to a total volume of 50 µl Platinum Taq DNA Polymerase (Life Technologies) mix with 0.2 µM primers BPEF3 and NEST Rev (Table S2). External PCR was performed as described above. External PCR products (2–5 µl) were then used for internal PCR as described above.
The EGFP probing sequence was 32P-labeled by random priming (NEBlot, New England Biolabs). Genomic DNA was isolated from candidate clones and digested with HindIII (New England Biolabs). DNA fragments were resolved by electrophoresis and transferred to a nylon membrane (Hybond-XL, GE Healthcare). The blot was incubated at 65°C in ExpressHyb (BD Biosciences/Clontech) with the 32P-labeled EGFP probe. The probed blot was visualized on a Storm 840 PhosphorImager (Molecular Dynamics).
B6.Cg-Gt(ROSA)26Sortm6(CAG-ZsGreen1)Hze/J mice (Jackson Laboratory) carry the green fluorescent protein gene ZsGreen1 at the Rosa26 locus driven by the CAG promoter, with an upstream STOP codon flanked by loxP sites and a downstream WPRE mRNA stabilizer. These mice were crossed with B6.C-Tg(CMV-cre)1Cgn/J mice (Jackson Laboratory) that carry the Cre recombinase gene driven by the CMV promoter, resulting in the deletion of loxP-flanked sequences in all tissues including the germline. Mice positive for both transgenes were then backcrossed to C57BL/6J. The resulting Cre negative progeny expressing ZsGreen1 under the CAG promoter at the Rosa26 locus were used to determine the reporter expression profile. Mice were in the C57BL/6 background, and were bred in house. All animals were housed in pathogen free barrier facilities and treated humanely with regard for alleviation of suffering.
Tissues were kept in 0.01% trypsin inhibitor (Sigma) on ice for up to 16 hours before analysis. Tissues were minced with scalpel blades or with a gentleMACS tissue dissociator (Miltenyi Biotec) and digested with 2 mg/ml collagenase V (Sigma) in HBSS (Invitrogen) at 37°C for 45 min. After digestion, the cell suspension was triturated and filtered through a 70 µm cell strainer (BD Biosciences) into an equal volume of DMEM with 20% FBS on ice. Cells were pelleted at 1500 rpm for 10 minutes, resuspended in OptiMEM (Invitrogen) and passed through a 35 µm cell strainer (BD Biosciences) before flow cytometry. Cells were analyzed with a FACScan flow cytometer (BD Biosciences) or sorted with a MoFlo cell sorter (Cytomation). Live cells were gated using forward and side scatter and then examined for fluorescence (excitation 488 nm, emission 580/30 nm). For RNA extraction from spleen cells, 1000 EGFP positive or 1000 non-EGFP positive cells were sorted into 200 µl TRIzol using a MoFlo (Cytomation) or a FACSAria (BD Biosciences) cell sorter. TRIzol volumes were then made up to 1 ml and cells were stored at −80°C until RNA extraction.
Whole organs were processed for imaging by compressing between coverslips to a thickness of 0.5 mm. The colon was cut lengthwise to expose the lumen. Tissues were imaged with a Nikon 80i microscope (×1 objective) in the FITC channel using a fixed exposure time. Serial images scanning the entire tissue surface were captured using an automated stage. Images were automatically compiled using NIS Elements software (Nikon) or Adobe Photoshop (Adobe Systems). Brightness and contrast of all images were adjusted identically in Adobe Photoshop. Fluorescent foci were either counted manually in a blinded fashion or with an in-house program written in MatLab (MathWorks). Tissue surface area was determined using ImageJ (NIH) by manually tracing the tissue outlines. Frozen sections (5 µm) were imaged with a ×60 objective in the FITC channel, stained with hematoxylin and eosin, and imaged again under visible light. Images were then overlaid manually. For each estimate of the average number of foci per cm2, the entire organ was evaluated in order to suppress the impact of variations in foci number in different regions of each organ.
Colonic crypts were isolated according to [98], with some modifications. Briefly, tissue samples were washed with HBSS to remove any fecal material. Dissected samples (0.5 to 1 cm2) were treated with 1 mM EDTA, 0.05 mM dithiothreitol (Sigma) at 37°C. After incubation for 30 min, tissue samples were gently shaken in the EDTA/DTT solution by inverting the tubes to release epithelial cells. This process was repeated twice. Crypts were stained with 1 µg/ml Hoechst 33342 (Invitrogen) and imaged with an Axio Observer Z1 microscope (Zeiss) at ×10 in the brightfield, FITC, and DAPI channels. Crypt images were captured using Axiovision Rel. 4.8 software (Zeiss) and compiled with Image J 1.46r (NIH).
Images were preprocessed using median filtering, and intensity shoots identified with an extended maxima transform [99] were treated as foci candidates. Candidates were segmented using a local thresholding-based algorithm where the threshold for each focus was adaptively selected by modeling the focus as a two-dimensional Gaussian distribution. Based on intensity and morphological features extracted by preprocessing and segmentation, foci candidates were classified into true foci and noise, and foci were further classified into large bright foci and small irregular foci using a support vector machine (SVM) with a radial basis function (RBF) kernel. The SVM was trained on annotations from an experienced biologist over multiple images.
Five- to seven-week-old heterozygous RaDR-GFP mice (C57BL/6 background) were used. DNA damage was elicited by combined treatment with N-methyl-N-nitrosourea (MNU, Sigma) and thyroid hormone (T3, Sigma). Details will be published separately. Briefly, T3 was administered in the diet (prepared by TestDiet) at 4 ppm according to [100]. MNU was administered at 25 mg/kg as an intraperitoneal injection at the time of peak cell proliferation in the pancreas induced by T3. Control mice were fed an identical diet without T3, and received control PBS injections. Feeding of T3 continued for 2 days after MNU injection. 3.5 weeks after MNU injection, mice were humanely euthanized and organs were harvested for the RaDR-GFP assay.
Recombinant cell frequencies and foci frequencies do not follow a normal distribution and were therefore compared using a two-tailed Mann–Whitney test. A p value of less than 0.05 was considered to be statistically significant.
|
10.1371/journal.pcbi.1005146 | DREISS: Using State-Space Models to Infer the Dynamics of Gene Expression Driven by External and Internal Regulatory Networks | Gene expression is controlled by the combinatorial effects of regulatory factors from different biological subsystems such as general transcription factors (TFs), cellular growth factors and microRNAs. A subsystem’s gene expression may be controlled by its internal regulatory factors, exclusively, or by external subsystems, or by both. It is thus useful to distinguish the degree to which a subsystem is regulated internally or externally–e.g., how non-conserved, species-specific TFs affect the expression of conserved, cross-species genes during evolution. We developed a computational method (DREISS, dreiss.gerteinlab.org) for analyzing the Dynamics of gene expression driven by Regulatory networks, both External and Internal based on State Space models. Given a subsystem, the “state” and “control” in the model refer to its own (internal) and another subsystem’s (external) gene expression levels. The state at a given time is determined by the state and control at a previous time. Because typical time-series data do not have enough samples to fully estimate the model’s parameters, DREISS uses dimensionality reduction, and identifies canonical temporal expression trajectories (e.g., degradation, growth and oscillation) representing the regulatory effects emanating from various subsystems. To demonstrate capabilities of DREISS, we study the regulatory effects of evolutionarily conserved vs. divergent TFs across distant species. In particular, we applied DREISS to the time-series gene expression datasets of C. elegans and D. melanogaster during their embryonic development. We analyzed the expression dynamics of the conserved, orthologous genes (orthologs), seeing the degree to which these can be accounted for by orthologous (internal) versus species-specific (external) TFs. We found that between two species, the orthologs have matched, internally driven expression patterns but very different externally driven ones. This is particularly true for genes with evolutionarily ancient functions (e.g. the ribosomal proteins), in contrast to those with more recently evolved functions (e.g., cell-cell communication). This suggests that despite striking morphological differences, some fundamental embryonic-developmental processes are still controlled by ancient regulatory systems.
| The dynamics of a biological system can be controlled by its own internal mechanisms and external perturbations. To gain intuition on this, we may draw a comparison with a mass hanging from a spring. The mass will move naturally by itself but its dynamics is also affected by one’s pulling it. That is, the dynamics of the mass is governed by the effect of the external perturbations superimposed on the internal mechanism of the spring (i.e. Hooke’s law). Similarly, given a group of genes, their temporal gene expression dynamics can be controlled by both transcription factors inside the group and external regulatory factors. Therefore, it is useful to identify the expression dynamics that are exclusively controlled by internal or external factors and compare them across various systems. While state-space models have been widely used to decouple the internal and external effects in physical systems, such as the mass and spring, typical biological systems do not have enough time samples to infer all the model’s parameters, and applications of state-space models were not very effective in these instances. Hence, we developed a general-purpose computational method by integrating state-space models and dimensionality reduction to identify temporal gene expression patterns driven by internal and external regulatory networks. We applied our method to the embryonic developmental datasets in the worm and fly (and also in a human cancer context). We successfully identified the temporal expression dynamics of cross-species conserved genes that were driven by conserved and species-specific regulatory networks.
| Gene regulatory networks systematically control the gene expression dynamics. These networks are highly modular, and consist of various sub-networks. Each sub-network contains a number of regulatory factors representing a subsystem that drives specific gene regulatory functions [1,2]. The subsystems interact with one another, and work together to carry out the entire gene regulatory function. For example, the gene expression in embryogenesis is controlled by the combinatorial effects of various regulatory subsystems composed of complex evolutionary regulatory networks [3]. These regulatory subsystems drive very diverse developmental programs, from the highly conserved (e.g. DNA replication) to the species-specific (e.g. body segmentation). As such the orthologous genes that are evolutionary conserved genes across species can therefore be regulated by both orthologous and species-specific transcription factors (TFs) [4]. The orthologous TFs form an “internal” regulatory network, while the species-specific TFs form an “external” one. Unfortunately, existing experimental gene expression data cannot decouple the expression components that are driven by the different subsystems. Thus, computational methods are required to assess the contribution from each factor or subsystem from the gene expression data. In this study, we propose a novel computational method, DREISS—dynamics of gene expression driven by external and internal regulatory networks based on state space model. Using DREISS, we are able to identify temporal gene expression dynamic patterns for evolutionarily conserved genes during embryonic development, as driven by conserved and species-specific regulatory subsystems. These results advance our current understanding of gene regulatory networks during evolution, as well as the differentiation during development.
Developmental gene regulatory networks control gene expression during the developmental processes. These particular regulatory networks have evolved, making it difficult to understand their regulatory mechanisms at the system level. Hence, one typically compares developmental gene expression across species to infer biological activities of developmental gene regulatory networks. For example, embryogenesis provides a platform to study the evolution of gene expression between different species. Recent work has showed that significant biological insight can be gained by cross-species comparisons of the expression profiles during embryogenesis for worms [5], flies [6], frogs [7] and several other vertebrates [8]. It was found that the orthologous genes have minimal temporal expression divergence during the phylotypic stage, a middle phase during the embryonic development across species within the same phylum. These patterns are often characterized as “hourglass” [9]. In addition, the conserved hourglass patterns were observed even within a single species while comparing the developmental gene expression data across distant species, such as worm and fly [10]; i.e., the expression divergence among evolutionarily conserved genes become minimal during the phylotypic stage in both worm and fly. However, much less is known about how the orthologous genes in each species eventually contribute to their species-specific phenotypes due to the lack of appropriate computational approaches. Thus, we aim to use DREISS to discover the components of the orthologous gene expression during embryonic development driven by species-specific transcription factors.
The state-space model has been widely used in engineering [11], and also in biology for the analysis of gene expression dynamics [12–14]. It models the dynamical system output as a function of both the current internal system state and the external input signal. A well-known example in engineering is the vehicle cruise control system where the system state can be the vehicle’s speed. Based on the road conditions, the cruise control requires various fuel amounts in order to keep the desired speed level. In biology, we can look at the transcription factors and microRNAs as internal and respectively external regulatory factors of the protein-coding genes expression (See more internal-external examples in S1 Table). Similarly, the state-space model can be applied for studying the expression of orthologous genes at different developmental stages using information regarding their expression (internal) and species-specific regulatory factors (external) at the current known developmental stage. Unlike earlier studies that calculate the expression correlation between individual genes, the state-space model predicts the temporal causal relationships at the system level; i.e., the state at a time is determined by the state and external input at the previous time. The earlier work applied the state-space model to study the gene expression dynamics focusing on small-scale systems, and did not explore the analytic dynamic characteristics of the inferred state-space models. The complex and large-scale biological datasets, especially temporal gene expression data, are very noisy, and high dimensional (i.e., the number of genes is much greater than the number of time samples), thereby preventing an accurate estimation of the state-space model’s parameters. The dimensionality reduction techniques have thus been used to project high-dimensional genes to low-dimensional meta-genes (i.e., the selected features representing de-noised and systematic expression patterns [1,15,16]) as well as the principal dynamic patterns for those meta-genes [17,18]. Using DREISS, we are able to apply the dimensionality reduction to the gene expression data, and develop an effective state-space model for their meta-genes, and then identify a group of canonical temporal expression trajectories representing the dynamic patterns driven by the effective conserved and species-specific meta-gene regulatory networks according to the model’s analytic characteristics. These dynamic patterns reveal temporal gene expression components that are controlled by conserved or species-specific GRNs.
DREISS is a general-purpose tool and can be used to study the gene regulatory effects from any different subsystems for a given group of genes. As an illustration, we applied DREISS to the gene expression data during embryonic development for two model organisms, worm (Caenorhabditis elegans) and fly (Drosophila melanogaster). In both species, we were able to identify the expression patterns of worm-fly orthologs driven by the conserved regulatory network consisting of the worm-fly orthologous TFs (i.e., the conserved regulatory subsystems between two species), as well as the worm/fly-specific regulatory network consisting of non-orthologous TFs (i.e., the species-specific regulatory subsystem). Our results reveal that, in addition to executing conserved developmental functions between worm and fly, their orthologous genes are also regulated by species-specific TFs to involve in species-specific developmental processes. In summary, DRIESS provides a framework to analyze both distantly and closely related species allowing for a better understanding of the gene regulatory mechanisms during development.
A gene regulatory network is made up of various subsystems [1,2]. These subsystems work together to execute regulatory functions. Given a group of N1 genes in a subsystem, defined as the internal gene set, Ω, their gene expression levels are not only controlled by internal interactions among Ω, but also affected by the regulatory factors from other subsystems outside Ω. We define an external gene set, Ψ consisting of those external regulatory factors. For example, we consider the worm-fly orthologous genes as internal set Ω. The worm-fly orthologous TFs from internal set Ω are the internal regulatory factors, and non-orthologous TFs such as worm- or fly- specific TFs are the external regulatory factors. Both the internal and external regulatory factors control gene expressions in dynamic ways (i.e., their regulatory signals at the current time will affect gene expressions at subsequent times). Thus, the regulatory mechanisms for gene expressions form a control system. In this study, we used a state-space model (defined by linear first-order difference equations, Fig 2A) to formulate temporal gene expression dynamics for internal set Ω (comprising N1 genes) with external regulation from external set Ψ (comprising N2 genes) at time points 1, 2, …, T as follows:
Xt+1=AXt+BUt
(1)
, where the vector Xt∈RN1×1, the “state”, includes N1 gene expression levels at time t in Ω, and the vector Ut∈RN2×1, the “input or control”, includes N2 gene expression levels at time t in Ψ. The system matrix A∈RN1×N1 captures internal causal interactions among genes in Ω (i.e., the ith, jth element of A, Aij describes the contribution from the jth gene expression at time t to the ith gene expression at the next time t+1), which instantiates a gene regulatory network. The control matrix B∈RN1×N2 captures external causal regulations from the genes in Ψ to genes in Ω (i.e., the ith, jth element of B, Bij describes the contribution from the jth gene expression in Ψ at time t to the ith gene expression in Ω at the next time t+1). R represents the real number domain. According to the state space model (1), the gene expression dynamics in Ω is determined by the system matrix A and the control matrix B. In particular, based on Eq 1, the state Xt can be expanded as follows:
Xt=AXt−1+BUt−1=A(AXt−2+BUt−2)+BUt−1=A2Xt−2+ABUt−2+BUt−1=A3Xt−3+A2BUt−3+ABUt−2+BUt−1=⋯=At−1X1+At−2BU1+At−3BU2+⋯+ABUt−2+BUt−1=At−1X1⏟XtINT+∑k=1t−2AkBUt−1−k⏟XtINTER+BUt−1⏟XtEXT
(2)
, where XtINT=At−1X1 is defined as the expression vector of the gene components driven only internally by genes in Ω. The rest terms ∑k=1t−2AkBUt−1−k+BUt−1 captures the expression expression vector of the gene components in Ω affected externally by the genes in Ψ. In particular, XtEXT=BUt−1 represents the expression vector of gene components in Ω driven purely by the genes in Ψ since it only involves B and U, and XtINTER=∑k=1t−2AkBUt−1−k captures the expression vector of gene components in Ω driven by the interactions between internal and external groups for involving A, B and U. In this paper, we mainly focus on the purely internal dynamics. As for the external-related terms, we should emphasize that any subdivision of the rest of the terms ∑k=1t−2AkBUt−1−k+BUt−1 is completely arbitrary. That is, although we subdivided it into a purely external term and an interaction term here, one could subdivide it multiple ways. That is, given a particular type of subdivision, each of the subdivided terms sums up a group of terms from AkBUt−1−k, k = 0,1,2,…,t-2. For example, one can look at ∑k=2t−2AkBUt−1−k+(ABUt−2+BUt−1), where ABUt−2 + BUt−1 shows the contribution from the inputs up to two time steps back to Xt.
The temporal gene expression experiments normally have limited time samples (for example, there may only be a dozen time points), which are far less than the time samples needed to estimate the large matrices A and B when internal and external groups, Ω and Ψ are composed of hundreds or thousands of genes. One way to deal with lack of time samples is dimensionality reduction. Thus, we project high dimensional temporal gene expressions to much lower dimensional meta-gene expression levels using a dimensionality reduction technique (Fig 2B). Those meta-gene expression levels should capture original gene expression patterns, such as the ones having the greatest degree of co-variation. We calculate the meta-gene expression levels as follows:
X˜t=WX*Xt;U˜t=WU*Ut
(3)
, where X˜t∈RM1×1, the “meta-gene state” at time t, includes M1 (<< N1 and <T) meta-gene expression levels; i.e., the first M1 elements of the tth row of the matrix whose columns are right-singular vectors of the matrix [X1 X2 ⋯ XT] in Ω by the singular value decomposition (SVD) [19]; the vector U˜t∈RM2×1, the “meta-gene input or control” at time t, includes M2 (<< N2 and <T) meta-gene expression levels; i.e., the first M2 elements of the tth row of the matrix whose columns are right-singular vectors from SVD of the matrix [U1 U2 ⋯ UT] in Ψ; WX∈RN1×M1 is the linear projection matrix of SVD from M1 meta-gene expression space to N1 gene expression space in Ω, WU∈RN2×M2 is the linear projection matrix of SVD from M2 meta-gene expression space to N2 gene expression space in Ψ, and (.)* is a pseudo-inverse operation; i.e., W*W = I, where I is the identity matrix.
Next, we obtain the effective state-space model for meta-genes using linear projections WX and WU between genes and meta-genes as follows (Fig 2C). By replacing (1) using (3), we obtain that
WXX˜t+1=AWXX˜t+BWUU˜t.
(4)
After multiplying the pseudo-inverse of WX, WX*∈RM1×N1 s.t. WX*WX=I where I is an identity matrix, at both sides of (4), we have that
X˜t+1=WX*AWX⏟A˜X˜t+WX*BWU⏟B˜U˜t=A˜X˜t+B˜U˜t
(5)
, where the effective meta-gene system matrix A˜=WX*AWX∈RM1×M1 captures internal causal interactions among meta-genes in Ω (i.e., an element of A˜, A˜ij describes the contribution from the jth meta-gene expression at time t to ith meta-gene expression at time t+1), and the effective control matrix B˜=WX*BWU∈RM1×M2 captures external causal regulations from meta-genes of Ψ to meta-genes of Ω (i.e., the ith, jth element of B˜, B˜ij describes the contribution from the jth meta-gene expression in Ψ at time t to ith meta-gene expression in Ω at time t+1). Eq 5 describes the effective state space model for the meta-genes of Ω, whose expression dynamics is determined by A˜ and B˜. Because the meta-gene dimension, M1 (M2) is less than T, and much less than N1 (N2), we can estimate A˜ and B˜ as follows.
We rewrite Eq 5 as a matrix product on the right side:
X˜t+1=A˜X˜t+B˜U˜t=[A˜B˜][X˜tU˜t].
(6)
By applying Eq 6 to time points, 2,3, …, T, we then obtain that
[X˜2X˜3⋯X˜T]⏟Ζ=[A˜B˜][X˜1X˜2⋯X˜T−1U˜1U˜2⋯U˜T−1]⏟Υ
(7)
, where Ζ∈RM1×(T−1) and Υ∈R(M1+M2)×(T−1).
Because of dimension reduction, Υ has more columns than rows so that it has right pseudo-inverse. Thus, the effective internal system matrix A˜ and external control matrix B˜ can be estimated by:
[A˜B˜]=ΖΥ*
(8)
, where Υ*∈R(T−1)×(M1+M2) is the right pseudo-inverse of Υ; i.e., ΥΥ* = I, with M1<N1, M2<N2, M1+M2<T, t = 1,2,…,T. It is worth noting that if we do not reduce the dimensionality, and obtain Eq 7 from Eq 5, then Υ will have much more rows than columns so that it doesn’t have right pseudo-inverse; i.e., there doesn’t exist a matrix Υ* such that ΥΥ* is a full-rank identify matrix. In addition, the condition of M1+M2<T also makes ΥΥ* be a full-rank identify matrix.
The analytic solution to a general first-order linear matrix difference equation [20], Qt+1 = CQt is
Qt = CtQ0 = (HEH-1) tQ0 = HEtH-1Q0 = HEtS, where the columns of the matrix H are eigenvectors of C, the diagonal elements of the diagonal matrix E are eigenvalues of C such that CH = HE, and the vector
S = H-1Q0. Then, if we rewrite Qt by a linear combination of the time exponential of eigenvalues of C, we have that Qt=HEtS=∑i=1mcαitsiHi=∑i=1mcαitKi, where mc is the total number of eigenvalues of C, αi is the ith eigenvalue of C, si is the ith element of S, Hi is the ith eigenvector of C (i.e., the ith column of H), and Ki = siHi is the coefficient vector of Qt over the tth time exponential of αi.
By Eq 5, the matrix à determines the meta-gene states components whose expression dynamics are internally controlled by the meta-genes of Ω. As Eq 2, we define the expression of the meta-gene components driven only internally by themselves in Ω at time t as X˜tINT, an M1-dimensional vector; i.e., their expression at two adjacent time points have X˜t+1INT=A˜X˜tINT∈RM1×1. According to the above analytic solution, it can be a linear combination of M1 dynamic patterns determined by the eigenvalues of the effective system matrix A˜ as follows:
X˜tINT=∑p=1M1λptK˜p; i.e., the internally driven component of ith meta-gene’s expression across all time points,
[X˜1INT(i)X˜2INT(i)…X˜TINT(i)]=∑p=1M1K˜p(i)[λp1λp2…λpT]⏟pthiPDP
(9)
, where λp and K˜p∈CM1×1 are the pth eigenvalue of A˜ and its coefficient vector from the analytic solution, which determines the pth dynamic pattern driven by effective internal regulations, defined as the pth internal principal dynamic pattern (iPDP) = [λp1λp2…λpT], in which λpt represents the tth power of λp, and Ξ(i) represents ith element of the vector Ξ. C represents the complex number domain. If an eigenvalue λ is complex when A˜ is asymmetric, then its conjugate λ¯ is also an eigenvalue, so we sum its iPDP and its conjugate eigenvalue, λ¯’s iPDP, as a unified iPDP with real elements equal to [λp1+λ¯p1λp2+λ¯p2…λpT+λ¯pT].
The internal principal dynamic patterns (iPDPs) represent canonical temporal expression trajectories, which can be either increasing, or damped oscillation and so on depending on iPDP’s eigenvalues (Fig 3). The iPDPs can be ordered by sorting their eigenvalues.
Also by Eqs 2 and 5, the expression of the meta-gene states components driven purely by the external group Ψ at time t is defined as X˜tEXT, an M1-dimensional vector, and its expression dynamics is determined by the equation X˜t+1EXT=B˜U˜t∈RM1×1; i.e., the externally driven components of meta-gene states at two adjacent time points. In particular, the externally driven component of ith internal meta-gene’s expression across time points:
[X˜2EXT(i)X˜3EXT(i)…X˜TEXT(i)]=∑q=1M2B˜i,q[U˜1(q)U˜2(q)…U˜T−1(q)]⏟qthePDP
(10)
, where X˜tEXT(i) and U˜t(q) are ith and qth elements of X˜tEXT and U˜t, respectively with t = 1,2,…, T, the vector [U˜1(q)U˜2(q)…U˜T−1(q)]⏟ is defined as qth external principal dynamic pattern (ePDP), and B˜i,q is the element of B˜ at ith row and qth column, which is also the coefficient of the externally driven component of ith internal meta-gene’s expression over qth ePDP. Based on Eq 2, the expression of the meta-gene components driven by the interactions between internal and external meta-genes is given by X˜tINTER=∑k=1t−2A˜kB˜U˜t−1−k. In this paper, we focus on the purely driven internal patterns (i.e., iPDPs) and compare them across different biological systems.
Because genes and meta-genes have linear relationships in terms of their expression levels as described in Eq 2, the components of gene expression levels in Ω driven by internal regulations, XtINT∈RN1×1 can be also expressed as linear combinations of M1 iPDPs:
XtINT=WXX˜tINT=∑p=1M1λptWXK˜p⏟Cp=∑p=1M1λptCp;i.e.,
the internally driven component of ith gene’s expression across all time points,
[X1INT(i)X2INT(i)…XTINT(i)]=∑p=1M1Cp(i)[λp1λp2…λpT]⏟pthiPDP
(11)
, where Cp=WXK˜p∈CM1×1 represents the gene coefficient vector for pth iPDP. Similarly, the gene expression components driven by external genes in Ψ, XtEXT∈RN1×1 can be also expressed as linear combinations of M2 ePDPs:
XtEXT=WXX˜tEXT=WXB˜⏟DU˜t=DU˜t;i.e.,
the externally driven component of ith gene’s expression across all time points,
[X2EXT(i)X3EXT(i)…XTEXT(i)]=∑q=1M2Di,q[U˜1(q)U˜2(q)…U˜T−1(q)]⏟qthePDP
(12)
, where XtEXT(i) is ith element of XtEXT with t = 1,2,…, T, and Di,q is the element of D=WXB˜ at ith row and qth column, which is also the coefficient of the externally driven component of ith gene’s expression over qth ePDP.
Gene expression data during embryogenesis provide important information about the dynamics of genomic functions throughout the developmental process, from the conserved functions such as DNA replication to the species-specific functions such as body segmentation, but hardly reveal any data regarding the evolutionary gene regulatory subsystems that drive those developmental functions [3]. Thus, in order to understand the relationships between those subsystems and their driving genomic functions, we apply DREISS to worm and fly gene expression datasets during embryogenesis in modENCODE and we are able to identify various developmental genomic functions of worm-fly orthologous gene pairs driven by two different evolutionary regulatory subsystems, conserved (worm-fly orthologous TFs) and non-conserved (worm/fly specific TFs). As model organisms for developmental biology, both worm and fly have been used previously to study embryogenesis.
DREISS enables us to compare expression dynamic patterns between two or more temporal gene expression datasets even though they have different numbers of samples, as well as differences in the times at which those samples were collected. For example, we can apply DREISS to two different datasets of the same group of genes, and identify both the common (similar) and the specific (different) dynamic patterns driven by internal regulations captured by the eigenvalues of the effective system matrices between the two datasets.
In this paper, we apply DREISS to 3,153 one-to-one orthologous genes between worm (Caenorhabditis elegans) and fly (Drosophila melanogaster) as internal group, Ω to study their expression dynamics during embryonic development [10]. We refer to species-specific TFs as external regulations; i.e., external group Ψ. We found that worm-fly orthologs have similar internal dynamic patterns, which may be mainly driven by conserved TFs, but have very different external dynamic patterns driven by species-specific TFs between worm and fly embryonic developmental stages. The data is summarized as follows.
We define internal group Ω as 3,153 one-to-one orthologous genes between worm and fly during embryonic development, and external group Ψ as all the species-specific TFs (509 worm-specific TFs, 442 fly-specific TFs) [21,22]. We used their temporal gene expression levels (as measured by the RPKM values in RNA-seq) during embryonic development from the modENCODE project [10]. The worm embryonic development dataset includes T = 25 time stages at 0, 0.5, 1, 1.5, …, 12 hours, and the fly dataset includes T = 12 time stages at 0, 2, 4, …, 22 hours, but t = 1,2,..,25 for worm and t = 1,2,…,12 for fly are used in this paper, representing the relative time points for the entire embryonic development processes. Because M1+ M2<T in Eq 8, we choose M1 = M2 = 5 meta-genes for fly (T = 12), and find that five meta-genes of Ω and five meta-genes of Ψ capture ~98% of the co-variation of orthologous gene expressions and fly-specific TF gene expressions, respectively. In order to compare worm and fly, we also choose M1 = M2 = 5 meta-genes for worm, which capture ~98% of the co-variation of orthologous gene expressions and worm-specific TF gene expressions.
We find that the meta-gene canonical temporal expression trajectories driven by conserved regulatory networks (i.e., internal principal dynamic patterns, iPDPs) include four major patterns in both the worm and fly embryonic developmental process by order of eigenvalues: 1) a late highly varied pattern; 2) an early fast decaying pattern; 3) a slowly increasing pattern; and 4) an oscillating pattern (Fig 4A); i.e., the pattern of canonical trajectories (VL, D, I, O) in Fig 3. In contrast to the observed iPDP similarities, we find that worm and fly have very different external principal dynamic patterns (ePDPs) (Fig 4B); i.e., the expression dynamic patterns driven by species-specific TFs. The principal dynamic patterns driven by the worm-specific regulatory network; i.e., worm ePDPs, include a varied pattern that decreases until the middle stage and then increases, an increasing pattern, a varied pattern with a peak entering middle stage, a pattern that varies early and then increases during the embryonic development, and a cosine-like oscillating pattern with roughly two periods during the embryonic development. The fly ePDPs, however, have a varied pattern with low expression at the early stage, a sine-like oscillating pattern with roughly one period during the embryonic development, an increasing pattern, another sine-lie oscillating pattern with roughly two periods during the embryonic development, and a varied pattern that is like damped oscillation. In addition, we checked the sensitivity of iPDPs to small perturbations to internal/external regulatory networks by the leave-one-out method; i.e., we removed one gene in the internal/external group, ran DREISS, and obtained the ordered iPDP eigenvalues for the remaining genes. We repeated the leave-one-out method for all genes, and finally found the ranges in which iPDP eigenvalues vary shown as error bars in S1 Fig. We can see that the iPDP eigenvalues almost stay at the same values (small error bars) for both worm and fly, which implies that the principal dynamic patterns of worm-fly orthologous genes driven by their conserved regulatory network are robust to small changes.
The above results suggest that the conserved regulatory networks from orthologous meta-genes between worm and fly have similar effects to orthologous meta-genes, given their similar iPDPs (i.e., both have four patterns, as described above). The species-specific regulatory networks from species-specific meta-genes (i.e., worm-specific or fly specific TFs) have effects that differ from the orthologous meta-genes for their different ePDPs. In addition, the expression dynamic patterns driven by the interactions between internal orthologous genes and external species-specific TFs are also different between worm and fly (S2 Fig).
In both worm and fly, we observe the similar four types of internally driven canonical temporal expression trajectories; i.e., four matched internal principal dynamic patterns (iPDPs) (Fig 4A). Thus, we are interested in seeing how individual orthologous genes relate to those dynamic patterns. We find that the worm-fly orthologous genes have correlated coefficients over each of the four iPDPs. Based on Eq 10, we can obtain the coefficients of orthologous genes for each iPDP. We find that their coefficients are significantly correlated between worm and fly iPDPs with a similar pattern (Fig 5): r = 0.33 (p<2.2e-16) for the highly varied pattern at late embryonic development stages (first iPDP), r = 0.66 (p<2.2e-16) for the fast decaying pattern at early embryonic development stages (second iPDP), r = 0.67 (p<2.2e-16) for the slowly increasing pattern during embryonic development (third iPDP), and r = 0.73 (p<2.2e-16) for the oscillation pattern during embryonic development (forth iPDP), where r represents Spearman correlation of iPDP coefficients of 3,153 orthologous genes between worm and fly. This implies that, not only do the orthologous meta-genes have similar internal (conserved) regulatory effects (i.e., similar iPDPs), but the worm-fly orthologous genes also have similar internally-driven expression dynamics as resulted from their significantly correlated coefficients for iPDPs. The ePDPs between worm and fly generally do not show a high degree of matching similarity, but the worm ePDP No. 2, and the fly ePDPs No. 3 are roughly representing the growing patterns. We find that orthologous gene correlation coefficients between these ePDP patterns are very small (Spearman correlation r = -0.22 of the orthologous gene coefficients of worm ePDP No.2 and fly ePDP No. 3).
The ribosome produces proteins, which is an ancient process and conserved across worm and fly, organisms separated by almost a billion years of evolution. The ribosomal genes are highly expressed during embryogenesis, since intensive cell division and migration require a large amount of proteins to be synthesized. We collected 195 ribosome-related genes based on the GO annotations. We ranked the coefficients of orthologous genes for each iPDP and ePDP in ascending order, and compared the rank values of iPDP and ePDP coefficients of ribosomal genes. We found that their average ranks of iPDP coefficients are significantly larger than ePDP ones in both worm (t-test p<2.2e-16) and fly (t-test p<2.6e-13) as shown in Fig 6. This means that the ribosomal gene expression is significantly more influenced by the conserved regulatory network than by the species-specific regulatory network, which is consistent with ribosomal genes having conserved functions during embryonic development.
The orthologous genes related to signal transduction for cell-cell communication (a significantly more recent evolutionary adaptation relative to the ribosome) exhibit the opposite trend. We found that 320 signaling genes from GO annotations have significantly larger average rank values of ePDP coefficients than iPDP ones in both worm (t-test p<5.6e-11) and fly (t-test p<8.3e-4), as shown in Fig 6. This result implies that the signaling gene expression is significantly more driven by the species-specific regulatory network than by the conserved regulatory network, which is consistent with the signaling genes being commonly associated with species-specific functions, such as body plan establishment and cell differentiation.
We next turn to the biological meaning of individual canonical temporal expression trajectory for iPDPs and ePDPs. For the fast-decaying pattern (2nd iPDP), we find that the DNA replication is significantly enriched in Top 300 (~10%) orthologous genes that have the most negative coefficients for this pattern, in both worm (p<1.6e-8) and fly (p<4.5e-6). The GO enrichment analysis was performed using DAVID [23]. The very negative coefficients for the fast decaying pattern mean high positive coefficients for a fast-growing pattern (vertically flipped 2nd iPDPs of worm and fly represent a fast-growing pattern), showing a drastic increase at the beginning of embryogenesis, then remain flat during the late embryogenesis (red curves in Fig 7). Most of the cell division of embryogenesis in both worm and fly happens approximately within the first 300 minutes. Then, the cell elongation and migration start to dominate the development [24,25]. The mRNA abundance of the genes involved in DNA replication may change accordingly. This is well reflected by the second iPDP. Interestingly, the original expression patterns of those top orthologous genes actually do not have fast-growing patterns (black curves in Fig 7), probably because of the combined effects of both conserved and species-specific GRN. Maternal mRNAs, which are pre-loaded before fertilization, may also mask the fast growing pattern of DNA replication genes. This pattern could only be observed after we separated the effect of two types of TFs using DREISS. In addition, we did not find any enrichment of DNA replication in top genes of other iPDPs (p>0.05). Therefore, the fast-growing iPDP patterns identified by our method reveal conserved regulation on the elementary cellular process of both species (i.e. DNA replication).
Besides a fast growing pattern driven by conserved worm-fly orthologous TFs, we also identified a fast growing pattern driven by non-conserved TFs for the two species. The Top 300 orthologous genes (~10%) with the fast-growing worm ePDP (ePDP No.2) (i.e., driven by species-specific regulatory networks) are enriched in ‘proteasome’ (p<9.8e-16). Protein degradation is not only a key process in apoptosis, but also throughout the entire course of development [26,27]. For example, eliminating proteins that are no longer needed is a vital process during embryo development; e.g., the maternal proteins need to be cleaned as the embryogenesis proceeds). Previous reports also showed that different species usually have different maternal mRNA in the oocyte, which indicates that species-specific strategies might be utilized to regulate the protein degradation process [28]. In this study, after separating the effect of conserved and non-conserved regulatory networks, we observed that the protein degradation is significantly enriched in the genes majorly driven by species-specific TFs in worms. In contrast, the Top 300 orthologous genes with fast growing fly ePDP3 are enriched in ‘mitotic cell cycle’ (p<3.5e-29), ‘translation’ (p<1e-30) and ‘mitochondrion’ (p<7.7e-20). Those enriched function related to energy generation is probably indicative of the large energy requirement during fly embryogenesis [29], which did not provide the evolutionary conservation of this energy-related gene regulation. Our result reveals that the fly genes associated with respiration are more up-regulated by fly-specific TFs relative to conserved TFs, and that this up-regulation evolved after the separation of worm and fly.
Besides the fast-growing pattern driven by species-specific TFs, we also observed some other interesting patterns. For example, worm ePDP3 displays a dramatic peak about 5 hours after fertilization. Among the Top 300 worm orthologous genes of this pattern, genes involved in synaptic transmission (p<5.6e-9) and cell-cell signaling (p<1e-7) are over-represented, suggesting that they are transiently activated in this stage of embryogenesis by worm-specific TFs. This observation indicates the gene regulatory network for these genes have evolved after the speciation.
We applied DREISS to another example (also see supplement) about cancer. We are also interested to identify the gene expression dynamic patterns driven by conserved and human-specific regulatory networks during breast cancer cell cycle. Thus, we applied DREISS to a time-series gene expression data for human estrogen-responsive breast cancer cell line (ZR-75.1) before and after hormonal stimulation, which has 12 time points covering a complete mitotic cell cycle (0–32 hours) of hormonal stimulated cells [30]. The internal group, Ω is defined as a set of cross-species conserved human genes (i.e., 1132 worm-fly-human orthologs including 150 orthologous TFs), and the external group, Ψ consists of 1870 human-specific TFs. As shown in S3 Fig, the internally driven principal dynamic patterns (iPDPs) of conserved human genes include an oscillation trajectory whose period is roughly equal to a full cell cycle (iPDP No. 4), but the externally driven patterns (ePDPs No. 2–4) oscillates more frequently than internal one, which suggests that though the evolutionarily conserved TFs regulate the normal cell cycle, the human specific TFs potentially drive the abnormal cycling behaviors of conserved gene expression responding to the hormonal stimulation.
In this paper, we presented a novel computational method, DREISS, which decomposes time-series expression data of a group of genes into the components driven by the regulatory network inside the group (internal regulatory subsystem), and the components driven by the external regulatory network consisting of regulators outside the group (external regulatory subsystem). DREISS is a general-purpose tool that can be used to study the gene regulatory effects of any interested biological subsystems such as protein-coding transcription factors, micro-RNAs, epigenetic factors and so on. As an illustration, we applied DREISS to the time-series gene expression datasets for worm and fly embryonic developments from the modENCODE project [10], and compared the worm-fly orthologous gene expression dynamic patterns driven by the conserved regulatory network (i.e., regulation effects from orthologous TFs), with the patterns driven by the species-specific regulatory networks (i.e., regulation effects from worm or fly specific TFs). We found that the conserved TFs drive similar genomic functions, but non-conserved TFs drive species-specific functions of orthologous genes between worm and fly, implying that, in addition to having ancient conserved functions, orthologous genes have been regulated by evolutionarily younger GRNs to execute species-specific functions during the evolution. This work can be easily extended to study the regulatory effects from orthologous TFs and species-specific TFs to species-specific genes. For example, one can find the expression dynamic patterns of worm/fly specific genes driven by specific TFs, and identify the genes with strong patterns associated with worm/fly specific functions, such as body formations. To the best of our knowledge, DREISS is the first method to reveal how the evolution of GRNs affects gene expression during embryogenesis.
We emphasize that DREISS is a general-purpose method (a free downloadable R tool available from github.com/gersteinlab/dreiss). Users can define the internal group (Ω) and external group (Ψ) according to their interests. For example, if users want to identify the protein-coding expression patterns driven by miRNAs, they can define miRNAs as an external group and protein-coding genes as an internal group. Additionally, DREISS can be applied to more than two datasets, such as comparing worm, fly and human embryonic stem cell developmental data, and finding their conserved and specific developmental expression patterns. The expression patterns driven by human-specific regulatory factors will potentially help us understand human-specific developmental processes along with the associated human genes.
Due to the limited time samples in gene expression datasets, DREISS uses the simple linear state space model (i.e. the first order linear invariant difference equation) to model the temporal gene expression dynamics, and identify principal temporal dynamic patterns. This model assumes that the gene regulatory networks controlling temporal gene expression dynamics does not change across the entire biological process such as (A, B) in Eq 1. Thus, based on the analytic analysis, the principal dynamic patterns (PDPs) must follow a small set of canonical temporal trajectories (Fig 3). With the rapidly increasing gene expression data, we can extend DREISS to more advanced models such as switched and hybrid system models, non-linear models [31], which will allow us to study the gene regulatory networks are time varying, and potentially find the more temporal gene expression patterns capturing the more complex gene regulatory activities.
|
10.1371/journal.pgen.1000408 | Pro-Aging Effects of Glucose Signaling through a G Protein-Coupled Glucose Receptor in Fission Yeast | Glucose is the preferred carbon and energy source in prokaryotes, unicellular eukaryotes, and metazoans. However, excess of glucose has been associated with several diseases, including diabetes and the less understood process of aging. On the contrary, limiting glucose (i.e., calorie restriction) slows aging and age-related diseases in most species. Understanding the mechanism by which glucose limits life span is therefore important for any attempt to control aging and age-related diseases. Here, we use the yeast Schizosaccharomyces pombe as a model to study the regulation of chronological life span by glucose. Growth of S. pombe at a reduced concentration of glucose increased life span and oxidative stress resistance as reported before for many other organisms. Surprisingly, loss of the Git3 glucose receptor, a G protein-coupled receptor, also increased life span in conditions where glucose consumption was not affected. These results suggest a role for glucose-signaling pathways in life span regulation. In agreement, constitutive activation of the Gα subunit acting downstream of Git3 accelerated aging in S. pombe and inhibited the effects of calorie restriction. A similar pro-aging effect of glucose was documented in mutants of hexokinase, which cannot metabolize glucose and, therefore, are exposed to constitutive glucose signaling. The pro-aging effect of glucose signaling on life span correlated with an increase in reactive oxygen species and a decrease in oxidative stress resistance and respiration rate. Likewise, the anti-aging effect of both calorie restriction and the Δgit3 mutation was accompanied by increased respiration and lower reactive oxygen species production. Altogether, our data suggest an important role for glucose signaling through the Git3/PKA pathway to regulate S. pombe life span.
| Lowering caloric intake by limiting glucose (the preferred carbon and energy source) increases life span in various species. Excess glucose can have deleterious effects, but it is not clear whether this is due to the caloric contribution of glucose or to some other effect. Glucose sensed by the cells activates signaling pathways that, in yeast, favor the metabolic machinery that makes energy (glycolysis) and cell growth. The sensing of glucose also reduces stress resistance and the ability to live long. Does glucose provoke a pro-aging effect as a result of its metabolic activity or by activating signaling pathways? Here we addressed this question by studying the role of a glucose-signaling pathway in the life span of the fission yeast S. pombe. Genetic inactivation of the glucose-signaling pathway prolonged life span in this yeast, while its constitutive activation shortened it and blocked the longevity effects of calorie restriction. The pro-aging effects of glucose signaling correlated with a decrease in mitochondrial respiration and an increase in reactive oxygen species production. Moreover, a strain without glucose metabolism is still sensitive to detrimental effects of glucose due to signaling. Our work shows that glucose signaling through the glucose receptor GIT3 constitutes the main cause responsible for the pro-aging effects of glucose in fission yeast.
| Glucose is the major carbon source entering the metabolic pathways. Glucose ultimately generates ATP to supply the energy necessary for the cell biosynthetic and functional demands. Substantial evidences support the idea that excess glucose acts as a pro-aging and pathogenic factor [1],[2]. Consistently, lowering glucose intake in a calorie restriction diet increases life span in many species, from yeasts to mammals [3],[4].
Research carried out in Saccharomyces cerevisiae has been fruitful to unravel the role of nutrient sensing in longevity. Mutations blocking the action of genes controlling nutrient- signaling pathways increase replicative life span (RLS), defined as the number of times a mother yeast cell produces a daughter cell [5]–[10]. For instance, genetic deletion of PKA signaling via Gpr1 or Gpa2 genes resulted in the extension of RLS [11]. Likewise, nutrient-signaling shortens chronological life span (CLS), the time a yeast population remains viable in stationary phase [12]–[15]. In other words, nutrient-signaling pathways have a pro-aging effect in budding yeast. So far, mutations found to increase life span in S. cerevisiae map to genes that respond to multiple nutrients, such as the PKA, Sch9 and Tor pathways [16]. Glucose is the major source of calories for yeast. Experimentally, calorie restriction (CR) is achieved by reducing the concentration of glucose in S. cerevisiae cultures. Under these conditions, yeast cells exhibit an increase in both their replicative life span, and their CLS. It is therefore possible that nutrients, and more particularly the glucose-signaling pathway, are major regulators of the effects of calorie restriction on aging.
In yeast, the connection between nutrient sensing and mitochondrial activity has been depicted in different contexts. This regulation of mitochondria allows yeast to adapt its energy metabolism to the available nutrients, and is crucial for the control of longevity [16]. Several genetic studies demonstrate that forcing S. cerevisiae to use respiration instead of fermentation induces a gain in both chronological and replicative life span [17]–[20]. To summarize, the activity of nutrient-signaling pathways seem to promote aging by inhibiting both stress resistance and respiration. However, the predominance and the interdependence of each of these two functions, metabolic changes and signaling in the control of longevity are still nebulous.
Our laboratory introduced Schizosaccharomyces pombe as a model organism for the study of chronological aging [21]. The use of this particular yeast is justified by the differences existing with budding yeast in traits that can potentially affect longevity. Both have been referred as Crabtree-positive yeast because of their capability to repress mitochondrial respiration in favour of glycolysis when glucose is abundantly available [22]. Nevertheless in fission yeast, the Crabtree effect is less pronounced than in S. cerevisiae since the inhibition of oxygen consumption by glucose is smaller; in other words S. pombe maintains a higher respiration rate in the presence of glucose [22]. Consistently, it is hard to isolate respiratory-deficient cells (petite) in S. pombe [23],[24], while these mutants occur spontaneously in S. cerevisiae. Furthermore, S. pombe differs from S. cerevisiae because of its lack of glyoxylate cycle that makes it inefficient in ethanol consumption as carbon source [25],[26]. Finally, fission yeast is also distinguishable in its mitochondrial inheritance which is mediated by microtubules like in higher eukaryotes [27].
In the present study, we wished to determine whether glucose metabolism or extracellular glucose signaling is responsible for the regulation of life span. We found that environmental glucose decreases CLS in S. pombe in a dose-dependent manner, and this effect is mimicked in cells lacking the glucose receptor Git3p, a G protein-coupled receptor (GPCR) which signals the presence of glucose in the medium through a cAMP/PKA pathway [28],[29]. Consistently, the constitutive activation of the Gα subunit of the G protein-coupled to the glucose receptor significantly decreases CLS. Deletion in the Git3/PKA signaling is characterized by higher oxidative stress defense, respiration and mitochondrial membrane potential; the same features observed in CR. Interestingly, CR has no effect either on stress defense or longevity in the strain constitutively activated for Git3/PKA (by mutational activation of the Gα subunit), although it still enhances respiration. Knockout of S. pombe hexokinase genes (hxk1 and hxk2), which are required to channel extracellular glucose into glycolysis, does not extend CLS in S. pombe. On the contrary, these mutant yeast strains accumulated glucose in the medium, exhibited increased glucose signaling and accelerated aging. Reduction of extracellular glucose or mutation of the glucose receptor Git3p rescued their aging phenotype. Altogether, our data suggest that glucose signaling constitutes the main pathway in the pro-aging effect of glucose in fission yeast.
To study the effects of glucose concentration on CLS, wild-type S. pombe cells were cultured in rich medium with different concentrations of glucose. Survival was assessed by counting colony forming units (CFU) as a function of time, after cells entered stationary phase [21]. Decreasing the concentration of glucose from 2% to 0.05% resulted in a dose-dependent extension of chronological life span (Figure 1A). Cultures with higher glucose concentration exhibited a premature appearance of aged-cell phenotype upon entering the stationary phase. This phenotype is characterized by a shrunken shape and oversized vacuoles (Figure 1B). DNA content analysis by flow cytometry revealed that cells cultured in glucose 2% and 0.2% displayed a typical G2 cell-cycle arrest in stationary phase (data not shown). Moreover, the cells had similar doubling times at different glucose concentrations during the exponential growth phase of the culture (Figure S1).
Aging in yeast results in part from cellular damage due to the accumulation of reactive oxygen species (ROS) [30]. In agreement, cells cultured in higher glucose concentrations accumulated more ROS than cells grown at lower glucose concentrations, as shown by staining with dihydrorhodamine 123 and dihydroethidium (Figure 1C and 1D and Figure S2). On the other hand, culturing cells in SMC medium lacking glucose and containing glycerol as carbon source increased chronological longevity up to tenfold longer than in 2% glucose (Figure 1E). Glycerol as sole carbon source forces the cell metabolism toward mitochondrial respiration, as evidenced by the diminished growth rate and the rise in oxygen consumption [31]. Altogether, these results confirm that the relationship between nutrition and longevity in S. pombe is similar to that observed in other model organisms.
A number of mechanisms could account for the pro-aging effects of glucose in S. pombe. For instance, the effect of glucose on aging could be due to extracellular glucose sensing (signaling pathway) or through intracellular glucose effects including glucose metabolism and cytoplasmic glucose sensing. To distinguish between them, we studied a strain deleted for the git3+ cytoplasmic membrane glucose receptor gene. S. pombe cells lacking this receptor (Δgit3) exhibited extension of their CLS (Figure 2A), suggesting that the pro-aging effects of glucose depends, at least in part, on the activation of a signaling pathway initiated by this receptor. To further confirm this idea, we used a constitutively active Gα subunit (Gpa2R176Hp) that acts downstream of Git3p in the glucose-signaling pathway. Gpa2R176Hp constitutively activates the PKA kinase independently of the presence of glucose by promoting the synthesis of a high level of cAMP [28],[32]. As expected, cells expressing this activated Gα protein displayed a significantly reduced CLS (Figure 2A).
To confirm that Δgit3 and gpa2R176H cells have decreased and increased glucose signaling, respectively, we took advantage of the fact that in S. pombe, glucose represses the transcription of the fructose-1,6-bisphosphatase fbp1+ gene via PKA activation [33],[34]. We used an fbp1-driven lacZ reporter integrated in the S. pombe genome to measure fbp1 transcription [35],36. Thus, the β-galactosidase activity inversely reflects the level of PKA activation in this glucose-sensing pathway. As expected, at late logarithmic phase, deletion of git3+ increased expression of this reporter, while the gpa2R176H mutation reduced its expression (Figure 2B). Consistently, culturing WT cells in low glucose conditions also increased fbp1-lacZ expression (Figure 2B).
Since chronological aging in yeast is linked to the accumulation of ROS [21],[30],[37], we next measured the levels of ROS in Δgit3 and gpa2R176H cells. As expected, deletion of the glucose receptor reduced ROS levels, while the constitutive activation of the glucose-signaling pathway increased ROS levels (Figure 2C). Although these results suggest that glucose signaling regulates aging independently of glucose utilization, it is possible that loss of the glucose-signaling pathway reduces glucose metabolism in this mutant. Indeed, the PKA pathway is known to control glucose intake via the regulation of hexose transporters responsible for glucose import in S. cerevisiae [38],[39]. We thus measured glucose consumption and found that mutations affecting the glucose-signaling pathway did not change the rate of glucose consumption (Figure S3). In conclusion, these results suggest that the glucose-signaling pathway controls chronological aging independently of glucose intake and utilization.
Experimentally, the intervention referred as calorie restriction (CR) is achieved by reducing the calorie intake of an organism and represents the most effective way to increase life span [3]. This phenomenon has been verified in almost all species studied, from yeast to mammals [7],[40] including non-human primates [41]. CR improves general health and delays the inception of many late-onset diseases in a variety of organisms [42].
In S. cerevisiae, calorie restriction is implemented by culturing the cells on low glucose concentrations [5],[15],[43]. Above, we showed that culturing S. pombe in low glucose decreases glucose signaling, and demonstrated that mutations affecting this signaling pathway increase the life span of S. pombe when cultured on high glucose concentration.
Increased respiration correlates with longevity in yeast [17],[20], and mammals [4]. In yeast, low glucose availability leads to a switch of the pyruvate metabolism from fermentation toward mitochondrial tricarboxylic acid cycle and respiration [26],[44]. To determine if mutations in the glucose-signaling pathway affect respiration in S. pombe, we measured the oxygen consumption of long-lived Δgit3 and short-lived gpa2R176H cells. In high glucose, we observed that Δgit3 cells display a higher level of oxygen consumption as compared to that of WT (Figure 3A). The effect of respiration on the mitochondrial membrane potential (Δψm) was determined using the DiOC6 dye and showed that the Δψm in stationary phase cells was higher in Δgit3 compared to WT cells (Figure S4). Interestingly, WT cells cultured in 0.2% glucose exhibited a higher level of oxygen consumption than the Δgit3 mutant in 2% glucose (Figure 3A), and an increased mitochondrial membrane potential (Δψm) in early exponential phase (Figure S4). This could explain why CR is slightly more efficient than the git3+ deletion in extending CLS (Figure 2A). Higher respiration was concomitant with a better growth on respiration medium (glycerol 3%) of both, WT cells previously grown on CR conditions and Δgit3 cells grown on either normal or CR conditions (Figure 3B). In addition to glucose repression, the participation of the PKA/cAMP-mediated signaling pathway in mitochondrial functions has been suggested in budding yeast [45],[46]. Our data and the observation that pka1+ deletion increased respiration as well (not shown) support the involvement of Git3/PKA in the regulation of mitochondrial functions in fission yeast.
To investigate if Git3/PKA is the only pathway regulating the metabolic switch toward mitochondrial respiration during CR, we subjected Δgit3 cells to CR and measured respiration. We observed an increase in respiration when Δgit3 mutation was combined to CR compared to Δgit3 cells in 2% glucose (Figure 3A and C). Hence, CR can increase respiration by mechanisms independent of the glucose receptor Git3.
On the other hand, we observed that the activated Gpa2R176Hp prevents the full activation of respiration induced by CR (Figure 3D). These results are supported by the observation that gpa2R176H cells did not grow on respiration medium (glycerol) (Figure 3B). Moreover, as oxygen consumption of WT and Δgit3 cells was 30% higher than gpa2R176H cells in CR, we observed that the Δψm of Δgit3 and WT cells in stationary phase remained higher than gpa2R176H cells (Figure S4). Altogether these data suggest that CR and reduced glucose signaling are not equivalent, and that Git3/PKA is involved in the control of respiration.
Guarente and colleagues proposed that in yeast, CR increases life span by increasing respiration but not oxidative stress resistance [17]. However, another study from Kaeberlein and collaborators contradicted these data. They showed that reducing glucose levels increased replicative life span in respiratory-deficient yeast [47]. We showed above that both CR and the Δgit3 mutation increase respiration, while the gpa2R176H mutation decreased the effect of CR on respiration. Combining CR and the Δgit3 mutation did not increase respiration over the values with CR alone. However, the survival of Δgit3 cells was higher on CR than that of WT cells (Figure 4A). On the other hand, CR did not increase the respiration rate of the strain expressing activated Gpa2R176Hp as this intervention did in WT cells (Figure 3D), and this defect could partially explain its short life span (Figure 4B).
To investigate further whether the additive effect of CR and loss of Git3p signaling involves respiration, cells were cultured in 20 µM of antimycine A, an inhibitor of complex III of the mitochondrial electron transport chain, that creates a leakage of electrons [48] and increases ROS production. Glucose restriction and git3+ deletion together increased longevity in this high-ROS context (Figure S5). This suggests that low glucose signaling cooperates with other effects of CR acting downstream of ROS production, perhaps stimulating ROS defense mechanisms. Together, the data suggests that CR and reduced glucose signaling are not equivalent and these manipulations can actually cooperate to increase life span by a mechanism different than an increase in respiration.
To investigate whether resistance to oxidative stress could explain the longevity effects of CR and git3+ deletion, we next study the effects of several pro-oxidants molecules on WT and mutants S. pombe cells grown at high or low glucose concentrations. First, CR and, to a lesser extent, loss of the Git3p GPCR increased hydrogen peroxide and menadione resistance (Figure 4C). Moreover, CR strengthened the already high stress resistance of Δgit3 cells (Figure 4C). On the other hand, the resistance to both hydrogen peroxide and menadione treatment in gpa2R176H cells was significantly lower than in WT (Figure 4D). This stress sensitivity could also explain the very short CLS of this mutant in both high and low glucose.
We also measured the levels of cytosolic Cu/Zn-superoxide dismutase (SOD1) and mitochondrial Mn-SOD (SOD2) by quantitative PCR (Figure 4E). The importance of these two enzymes for long-term survival was demonstrated in budding yeast cultured in high glucose concentration [49]. No significant differences of expression were seen, neither in SOD2 (Figure 4E) nor in glutathione peroxidase (Gpx1, data not shown) for all the mutants and growth conditions tested. Interestingly, WT cells on CR showed no increased expression of SOD1 or SOD2 despite their very high oxidative stress resistance (Figure 2C). On the other hand, the deletion of glucose receptor increased significantly SOD1 expression. This correlates with the gain of oxidative stress resistance of this strain (Figure 4C). An unexpected three to six time rise of SOD1 transcript was observed in the gpa2R176H mutant, even if this strain displayed a very weak oxidative stress resistance. This could be the consequence of a feedback mechanism attributable to the very high production of ROS in this strain (Figure 2C). Although further studies on the mechanisms of stress resistance will be necessary, our data clearly shows that glucose signaling regulates SOD1 expression in S. pombe. Since SOD1 is not regulated by CR, it may be part of the mechanism by which the git3+ deletion cooperates with CR to increase the resistance to oxidative stress and life span.
In yeast, hexokinase 2 is responsible for channeling glucose into metabolic pathways by catalyzing phosphorylation of this sugar. It also has a function in glucose signaling in S. cerevisiae by promoting the down-regulation of glucose-repressed genes [50],[51]. Mutants of hexokinase do not influence CLS but increase replicative life span in S. cerevisiae [17],[43]. In fission yeast, the glucose phosphorylation activity is provided by two hexokinases (Hxk1p and Hxk2p), but the main enzymatic activity is due to hexokinase 2 [50]. Loss of Hxk1p has no significant phenotype (data not shown) and loss of Hxk2p dramatically decreases the growth rate in glucose [50]. The double knockout of both hxk1+ and hxk2+ is not viable on glucose [50]. To determine if hexokinase affects CLS in S. pombe, we first measured the life span of an S. pombe Δhxk2 deletion strain. Unlike in S. cerevisiae, we observed a significant decrease in CLS in this strain (Figure 5A).
We have shown above that glucose signaling mediates pro-aging effects in S. pombe. Therefore, we reasoned that defective glucose utilization in the Δhxk2 strain could result in an accumulation of intracellular glucose followed by the inhibition of glucose import. Moreover in S. cerevisiae, glucose can be re-exported in the extracellular medium by the hxt hexose transporter [52]. In turn, the high extracellular glucose concentration would lead to increase the duration of glucose signaling. To test this hypothesis, we first measured the glucose concentration in the medium during the growth of both wild-type and Δhxk2 cells. We found that glucose levels remained high in the Δhxk2 culture as compared to WT, and that this strain has a very slow growth rate (Figure 5B). Congruently with this observation, at early stationary phase Δhxk2 cells exhibited increased glucose signaling in comparison with control cells, as represented by the drop of fbp1-lacZ reporter expression (Figure 5C). In budding yeast, hexokinase activity has been involved in glucose-signaling pathways during exponential phase [53]. Our results do not contradict, but support those observations since the Δhxk2 mutant has a defect in glucose signaling in exponential phase when compared to wild type, as indicated by elevated fbp1-lacZ expression (data not shown). However, the Δhxk2 mutant reaches stationary phase with glucose in the medium and its short life span was completely rescued when cultured in 0.2% glucose. As expected, culturing Δhxk2 cells in low glucose resulted in a two-fold increase in β-galactosidase activity indicating an increase in fbp1-lacZ reporter expression. This shows a reduction in signaling through the Git3/PKA pathway (Figure 5A and 5C). Taken together, the results are consistent with the model that an increase in glucose signaling via the Git3/PKA pathway accelerates aging in Δhxk2 mutants.
The strain Δhxk2 still has the hexokinase 1 activity permitting glucose metabolism (Figure 5B) [50]. To confirm the importance of the pro-aging effect of glucose signaling isolated from the effect of glucose utilization as energy source, we constructed a double knockout of both hexokinases in fission yeast (hxk1+ and hxk2+). These two mutations should prevent glucose from entering glycolysis and the pentose phosphate pathway. However, so far attempts to obtain this double mutant has been unsuccessful [50]. It was concluded that Δhxk1 Δhxk2 strain is not viable on glucose. To circumvent this problem, we complemented Δhxk2 with a plasmid expressing Hxk2p (pREP41_hxk2+) and crossed it with a Δhxk1 strain. After sporulation of the diploid, we selected for offspring containing both Δhxk1 and Δhxk2 deletions and the plasmid pREP41_ hxk2+. Then we allowed the strain to lose the hxk2+ plasmid in a medium containing only glycerol as carbon source and picked clones without plasmid. Because we obtained viable double mutants, we concluded that hexokinase activity and possibly glucose phosphorylation was required for sporulation but not for survival in S. pombe. Using the same approach, the triple knockout, Δhxk1 Δhxk2 Δgit3 was created.
These mutants Δhxk1 Δhxk2 and Δhxk1 Δhxk2 Δgit3 could not grow when switched on plates with only glucose as carbon source. After at least ten days of incubation however, in some plates we observed for both strains clones that grew on glucose at a frequency between 10−6 to 10−7 (data not shown). The appearance of such clones was attributed to genetic reversion due to the nature of hexokinase 2 knockout that was created by insertion of a marker rather than complete suppression of the open reading frame [50].
Mutants Δhxk1 Δhxk2 and Δhxk1 Δhxk2 Δgit3 were grown in glycerol as a carbon source for around two divisions with a doubling time of around ten hours. At this point, they were spotted on plates containing glycerol plus glucose (Figure 6A). The double mutant Δhxk1 Δhxk2 did not grow on glycerol with 2% glucose, but it did on glycerol with 0.2% glucose. This result is consistent with the idea that the absence of hexokinase activity leads to a sustained and toxic glucose signaling. In agreement, the impaired growth of the double hexokinase mutant on glycerol plus glucose 2% was completely restored by a deletion in the glucose receptor git3+ (Figure 6A).
To assess whether the increase in glucose signaling of the double hexokinase knockout decreases the viability in stationary phase (chronological aging), we added 2% glucose to liquid cultures at late exponential phase (OD595 5–6). Then, viability was evaluated as a function of time by counting the number of living cells per mL (Figure 6B). After glucose addition, cultures with and without glucose needed two-day incubation to reach saturation corresponding to OD595 13 to 16. The Δhxk1 Δhxk2 double deletion mutant exposed to 2% glucose displayed striking loss of viability 24 hours after glucose addition in comparison to cultures with no added glucose (Figure 6B). This loss of viability was prevented by CR (0.2% glucose) or by deletion of git3+ (Figure 6B). To further characterize the loss of viability induced by glucose in double hexokinase knockout strains, we stained yeast cells with Phloxin B, a dye accumulated by dead cells. We found a high proportion of stained cells (30%) at 18 hours after glucose addition in comparison to 5% in control cells (Figure 6C). Notably, Phloxin B stained cells were longer and displayed oversized vacuoles, a typical phenotype of aging in yeast (Figure 6C). ROS production was evaluated 36 hours after glucose addition by flow cytometry with DHE staining. A considerable number of DHE stained Δhxk1 Δhxk2 cells was observed in the culture with 2% glucose (Figure 6C). Again, Phloxin B staining and the increase in ROS were prevented by CR (glucose 0.2%) or deletion of git3+ (Figures 6B and C).
Our results show that glucose signaling via the Git3p GPCR is required for the pro-aging effects of glucose in S. pombe and is sufficient to mediate detrimental effects even in the absence of glucose consumption.
Excessive glucose signaling has been associated with humans diseases such as diabetes, as well as with the less understood process of aging [54]. Several mechanisms have been proposed for the harmful effects of glucose. Glucose can be directly toxic to cell components because it can promote non-enzymatic glycosylation and the accumulation of advanced glycation end products (AGE) which impair cellular functions [55],[56]. Excess glucose metabolism can also be deleterious because glucose oxidation increases the source of electrons to the mitochondrial respiratory chain in the form of NADH. In cells with a very active glucose metabolism, excess electrons can promote the generation of deleterious ROS if there is no matching increase in the efficiency of electron transport [54],[57]. Glucose and/or nutrient-signaling pathways also control life span in various species including yeast [15]. The data raise the question about the relative contribution of signaling and metabolism to the regulation of life span [58].
Here we examined this question in S. pombe using mutants of the Git3/PKA glucose-signaling pathway. In this pathway, PKA kinase is activated by glucose signaling through the Git3p G protein-coupled receptor (GPCR), which results in the Gα subunit (Gpa2p)-mediated activation of adenylate cyclase [29] as represented in Figure 7. This, in turn, produces a linear increase in cAMP levels. The cAMP is bound by the Cgs1 regulatory subunit of Pka1 kinase, activating PKA. The consequence is a re-localization of PKA to the nucleus followed by the inhibition of the Rst2 transcription factor, an increase in stress sensitivity and a decrease in cell survival [21],[33],[59]. We previously demonstrated the importance of cAMP/PKA pathway in regulating S. pombe aging by showing that knocking out the only catalytic subunit of the PKA complex results in increased chronological life span as well as enhanced stress resistance [21]. However, other nutrient-signaling pathways may activate PKA complex in yeast, so the specific role of glucose signaling in the longevity of S. pombe was unknown.
We show here that low glucose levels increase CLS in S. pombe, a typical CR response. Also, mutants with a defective Git3/PKA pathway have an increased life span, a normal glucose consumption rate, and only a slightly reduced growth rate. The reverse is also true. High glucose concentration, acting through the Git3/PKA pathway, promotes aging and decreases stress defense and respiration. Likewise a constitutively active Gα subunit, normally coupled to the Git3p GPCR, mimics the effects of high glucose even in low glucose.
Further support for a role of glucose signaling in the control of CLS in S. pombe was obtained by studying hexokinase deletion strains. These mutants die prematurely in stationary phase concomitant with prolonged stimulation of Git3/PKA signaling. Since cells without hexokinase cannot metabolize glucose, these results suggest that sustained glucose signaling, caused by the excess of extracellular glucose that remains in the medium of hexokinase mutants, promotes aging in S. pombe. The loss of Git3p GPCR blocks the detrimental effects of glucose in double hexokinase mutant. This suggests that glucose exerts a strong pro-aging effect via the Git3/PKA signaling pathway. Notably, the premature death of double hexokinase mutant due to high glucose is concomitant with an accumulation of ROS.
It is remarkable that the effect of deleting hexokinases differs between S. pombe and S. cerevisiae. In the budding yeast, deletion of all major hexokinases (glucokinase, hexokinase 1 and 2) impairs cAMP production and activation of the PKA pathway [60]. Conflicting with these data, we show that in S. pombe, hexokinase mutants die prematurely due to sustained signaling through this pathway. Careful examination of our results also reveals that hexokinase mutants have a defective PKA pathway during the exponential phase of the cultures. However, these mutants in S. pombe enter stationary phase with high concentrations of glucose in the medium and a continual activity of the Git3/PKA pathway that is responsible for their premature aging. In contrast, hexokinase mutants in S. cerevisiae have longer RLS and a normal CLS [8],[43]. This apparent discrepancy could be the result of particular differences in the glucose-signaling pathways and energy metabolism between S. pombe and S. cerevisiae. For instance, the regulation of glycolysis is different between these two yeasts. S. cerevisiae growth on glucose is sensitive to trehalose biosynthesis whereas S. pombe is not [61].
Despite the very significant role of Git3/PKA pathway in the pro-aging effect observed in the double hexokinase mutant, our work showed that the signal from the Git3p GPCR dependent pathway is not the only regulator of all the effects on aging due to glucose. First, in minimal medium completed (SDC), lowering glucose concentration had no effect on longevity (data not shown). Nevertheless, glucose decreased longevity when S. pombe were grown in synthetic medium based on yeast nitrogen base [26]. Other nutrient limitation is suspected to affect PKA-regulated processes. For instance, conjugation efficiency is controlled in both Git3/PKA cAMP-dependent manner and in a cAMP/PKA independent manner sensitive to medium composition [62]. This PKA-independent nutrient sensing could mimic the effect of glucose restriction in rich medium and may explain why the life span of S. pombe grown in SDC is not affected by glucose. This explanation is consistent with the fact that the respiration rate in 2% glucose is higher in synthetic medium than in rich medium (complex medium) [26]. Another indication that the signal from the Git3/PKA pathway is not the only one to control the rate of aging is provided by the glucose receptor mutant (Δgit3). This deletion strain still responds to CR with a higher oxidative stress resistance, lower ROS levels and an increased survival (Figure 4C, 2C and 4A). In agreement, studies in Caenorhabditis elegans [63], Drosophila [64] and mice show that disabling the insulin/IGF-1 signaling pathway can cooperate with CR to increase longevity [65]–[72]. What are the other possible mechanisms by which glucose could accelerate aging in S. pombe?
In budding yeast, glucose activates the glucose repression pathway, which is regulated by the AMP-activated protein kinase (AMPK) Snf1p complex [53],[73]. So the Git3/PKA-independent effect of glucose could be explained by the activation of AMPK complex which affects aging in yeast and metazoans [74],[75]. Our S. pombe hexokinase deletion mutants are expected to be defective in this pathway [53],[76] but they still age prematurely when grown in high glucose concentrations, suggesting that glucose repression is not involved in the pro-aging effects of glucose. Conversely, we could not discard the possibility that some pro-aging effects of glucose are mediated by the non-enzymatic glycosylation of proteins by glucose. Nevertheless, altogether our data point toward a regulation of longevity primarily via the glucose signaling through Git3/PKA pathway, raising the question about the underlying mechanisms. Although further work is required to discover the mechanisms by which glucose signaling accelerates aging in S. pombe, our current evidence points to the mitochondria as the target of glucose signals. First, CR (low glucose) in wild type yeast enhances respiration and mitochondrial membrane potential, prevents ROS production and improves oxidative stress defense. Second, Δgit3 cells have a similar phenotype and, in addition this strain displays a higher expression of cytosolic superoxide dismutase in stationary phase. These could explain the additional longevity of Δgit3 cultured under CR conditions (Figure 7).
In yeast, the cAMP/PKA glucose sensing pathway possibly represents the ancestor pathway of insulin/IGF-1 signaling in multicellular eukaryotes. This pathway signals the presence of glucose, the preferred energy source. It also controls stress resistance, growth rate and sexual development, modifies mitochondrial metabolism, and ultimately controls life span as we have shown in this study. Similar to our observations for the Git3/PKA pathway in fission yeast, a decrease in the insulin/IGF-1 signal increases longevity as a function of CR in mammals. The extent to which dietary restriction may actually be effective in humans is still unknown. Our results also show that CR and loss of the Git3p GPCR cooperate to increase life span. This suggests that if this pathway is conserved in higher organisms, its inhibition may lead to an anti-aging treatment not relying on strict diets with a limited caloric content as used in animal research. Interestingly, inhibition of cAMP synthesis by the knockout of the type 5 adenylyl cyclase (AC5) gene induced Raf/MEK/ERK-dependent stress resistance and lengthened life span in mice [77]. The effect of reducing glucose signaling in S. pombe also results in a decreased cAMP synthesis in response to glucose, because Git3p, via Gpa2αp, activates adenylate cyclase [32]. Since CR and inhibition of glucose signaling cooperate to extend life span in S. pombe, it would be interesting to combine agents that reduce cAMP synthesis or reduce PKA activity with CR in mammals.
In conclusion, our work with S. pombe highlights the importance of glucose-signaling pathways and oxidative stress resistance in aging. Given the importance of glucose as a central metabolite, it is surprising that the pathway for glucose sensing existing in S. pombe has not been found yet in mammals. Whether a glucose receptor contributes to these signaling pathways in metazoans remains to be demonstrated. Our data together with the interesting phenotype of the AC5 KO mice provide the rationale for further inquiry into glucose sensing pathways in mammals.
This study was conducted according to the principles expressed in the Declaration of Helsinki.
MM refers to Edinburgh Minimal Medium [78] complemented by adenine, uracil, leucine and/or histidine 75 mg.L−1 (A,U,L,H). SMC refers to synthetic medium complemented and is composed of MM plus adenine, uracil, leucine and/or histidine 444 mg L−1 (A,U,L,H). Its composition is described in a previous study [21]. The same medium with glycerol 3% ethanol 0.2% for carbon source was named SMC glycerol. Mating and sporulation were carried out on MEA plates (bacto malt extract 3%, glucose 0.4%, pH 5.5, supplemented by adenine, histidine, uracil, leucine 225 mg.L−1 each). Yeast extract complete medium (YEC), was made of yeast extract 5 g.L−1 (BD, Difco) supplemented with 222 mg.L−1 of adenine, uracil, leucine and histidine, and glucose 2% unless otherwise specified. All cultures were incubated at 30°C in a rotating incubator shaker at 250 rpm (New Brunswick instrument).
Growth curves represent the average of three independent cultures. Morphological analysis of wild type cells in YEC glucose 0.05%, 0.2%, 0.5% or 2% was done in 10 mL cultures in 50 mL conic tubes with air-permeable cap grown overnight. Early log phase refers to OD595 0.5.
The strains used in this work are all described in supplementary Table 1. Wild type refers to strain SP14000, except for Figure 2B, 5C, S4 in which it refers to FWP87. The gpa2R176H (RWP1) [32] Δgit3 deletion (CHP984) [28], Δhxk1 and Δhxk2 deletions (CJM387, CJM389) [50] alleles were previously described.
The double Δhxk1 Δhxk2 mutant was constructed as follows. Δhxk2 (CJM389) was transformed with a plasmid bearing the hxk2+ ORF, previously amplified by PCR and inserted into the SalI site of pREP41 (pREP41_Hxk2). PCR primes sequences will be provided upon request. The Δhxk2 pREP41_Hxk2 strain (SP14405) was mated with Δhxk1 (CJM387). Corresponding diploids were sporulated in MES media and spores hxk1::ura4+ hxk2::his3+ harbouring the pREP41_Hxk2 plasmid were selected on MMA media. Haploids were grown to saturation in liquid SMC supplemented with adenine and leucine 222 mg.L−1 and with glycerol 3%, ethanol 0.2% as carbon sources. Then, they were diluted in the same fresh medium and cultured a second time to saturation in order to force cells to lose pREP41_Hxk2 plasmid. At this point, clones without plasmids were selected on plates SMC AL glycerol. The loss of pREP41_Hxk2 plasmid was validated by verifying that these clones Δhxk1 Δhxk2 (SP14483 and SP14493) cannot grow without leucine, the marker on the pREP41 plasmid. In addition, these clones cannot grow on SMC AL glucose 2%. The triple mutant Δhxk1 Δhxk2 Δgit3 was obtained by first constructing a Δhxk1 Δgit3 double knockout (SP14373) after mating the single mutants Δhxk1 (SP14313) and Δgit3 (SP14105). The resulting Δhxk1 Δgit3 strain was crossed with Δhxk2 pREP41_Hxk2 (SP14405) and the haploid strain Δhxk1 Δhxk2 Δgit3 without plasmid was isolated as described previously for Δhxk1 Δhxk2.
Three independent cultures of each double hexokinase mutant Δhxk1 Δhxk2 and Δhxk1 Δhxk2 Δgit3, were started in 75 mL (250 mL flask) YEC glycerol and incubated 24 hours. At OD595 0.6, 0.1 mL were harvested, washed in sterile water and serial diluted to be plated as drop test on solid YEC glycerol 3% ethanol 0.2% glucose 2% or 0.2% or no glucose. Plates were incubated 8 days at 30°C. The same 75 mL cultures were grew until OD595 5 to 6 and split in 3 times 25 mL cultures, one let with glycerol only, one complemented with glucose 2%, one with glucose 0.2%. These 18 cultures were then studied as described below.
The frequency of cells able to recover the ability to grow on glucose in Δhxk1 Δhxk2 mutants was measured by plating serial dilution of 100 µL of a saturated culture of SP14383 and SP14393 on SMC AL glycerol and on SMC AL glucose and by counting colonies forming units. The average of the ratios of six independent clones of SP14383 and SP14393 was 1.6×10−6. Because of the very low frequency of this event and the long time revertants take to grow and reach a significant number, we consider that these revertants did not influence our data.
The protocol for CLS measurement by CFU counting has been described previously [21] except that the first estimation of the number of living cells was delayed. Cells that reached maximal density were harvested, serial diluted and plated 24 hours and 48 hours after the optical density was stable and maximal; the higher number of living cells from these two samples was considered as the beginning of CLS curve (i.e., survival 100%). Error bars represent standard deviation calculated from four cultures separated from a single initial culture at the end of exponential phase. Each assay was repeated at least three times. All CLS analysis were performed in YEC AULH 222 mg L−1 except in Figure 1E where the medium is SMC AUL 444 mg.L−1 glycerol 3%. For antimycine A treatment, cultures were started at OD595 0.2 with 20 µg.mL−1 antimycine A (solubilized in ethanol 100%) and CLS was measured as described above, except that cells entered stationary phase at a lower OD.
Number of living cells per mL was calculated by plating dilutions of sample of the cultures as described above accepted that solid YEC glycerol was used. The concentration presented (living cells/mL) with standard deviation represents the average of three independent cultures. Survival analysis by Phloxin B staining was done according to a previous publication [21], with the exception that the percentage of stained cell was obtained by counting under microscope after background subtraction. At least 500 cells were counted for each condition.
Epifluorescence microscopy analyses were performed using an inverted Nikon Eclipse E800 microscope equipped with a Nikon_60 DIC H (1.4 NA) lens and a Photometrics CoolSNAP fx CCD camera. Images were acquired using a motion-picture camera CCD CoolSnapFX 12 (Photometrics, Tucson, AZ, USA) bit and analysed with UIC Metamorph software (Molecular Devices Corporation, Downington, PA, USA). The percent of ROS-positive cells was measured with dihydrorodhamine 123 (DHR 123, Sigma) following a previously described protocol [21]. The fluorescence of this dye is activated by peroxynitrite and peroxide in the presence of peroxidase [79]. A total of 500 to 700 cells per culture were counted to determine the percentage of positively stained cells and standard deviations were calculated using three independent experiments. Staining by dihydroethidium (DHE, Sigma) is more specific to superoxide production [80] and was achieved as followed. 1.4×107 cells were collected and resuspended in 0.1 mL 1× PBS with DHE 50 µM and incubated 10 minutes at 30°C. The DHE solution was removed, cells were resuspended in 20 µL 1× PBS and deposited on a microscope slide with a thin layer of agarose 1%. Counting was done the same way than for DHR123 using a Cy3 filter. Flow cytometry analysis was performed following the protocol detailed in [21] excepted for cells grown in glycerol. They were sonicated 15 seconds with a Sonicator Dismembrator Fisher Scientific Model 100 set to 12 watts. FACS analysis was done using FL1 filter for DiOC6 dye and FL3 filter for DHE dye.
Oxygen consumption was measured in cultures grown to cell concentrations between OD595 0.8 and 1.5. Cells were cultured in YEC to a given OD, and then the culture was diluted in its own medium if OD595 was greater than 1.5, or concentrated in its own medium by centrifugation if OD595 was less than 0.8. The goal was to measure the respiration of cultures with similar concentrations and in the exact medium in which the samples were taken. 10 mL of culture, sometimes diluted or concentrated, was incubated with gentle agitation at 30°C and 5 mL was loaded in the measurement chamber at 30°C with agitation. The oxygen consumption was followed with a Clark electrode YSI model 53 oxygen monitor until all oxygen was consumed in the chamber. The calibration of the Clark electrode for the maximum oxygen concentration (100%) was done on the air. The consumption was linear, the measure was recorded with a tracer Linear1100 and the slope was calculated for each sample. The result corresponding to the rate of respiration was normalized on the OD595 of the cells in the chamber and expressed in %O2.min−1.OD−1.
Glucose concentration was measured on the supernatants of cultures at different ODs following the protocol given in Quantichrom™ Glucose Assay Kit from BioAssay Systems®. The results presented are the averages of three independent cultures.
β-galactosidase activity, expressed from the fbp1-lacZ reporter, was determined as previously described [36] except that cultures were grown in YEC to late exponential phase (OD595 9 in glucose 2%, and OD595 2 in glucose 0.2%). CHP1229 was grown to only OD595 5.5 corresponding to the end of its exponential phase.
Cells were cultured in YEC glucose 2% or 0.2% to stationary phase, and harvested 24 hours thereafter. Cultures were diluted to OD595 0.5 to 0.8 in water and submitted to various oxidative shocks at 30°C. These include 1 M H2O2 for 120 minutes; 0.75 M Menadione for 180 min; 0.5 M H2O2 for 30 min or 0.3 M Menadione for 90 minutes. Then, cells were washed twice with 1 mL water and serially diluted tenfold four times. Each dilution was spotted on YEC plates and incubated five days at 30°C. For re-growth on glycerol plates, cells were grown in YEC glucose 2% to OD595 0.5 and washed twice in water. Cells were serially diluted as described above and spotted on SMC AULH glycerol 3%.
Mitochondrial membrane potential (Δψm) was measured with DiOC6 (Molecular Probes). Yeast strains were grown over-night in 10 mL YEC using 50 mL tubes with half screw cap to allow gas exchange. Early exponential phase refers to cells harvested at OD595 from 0.7 to 1. Late exponential phase refers to cultures at OD595 2.1 to 2.4 in glucose 0.2% and OD595 4 to 5 for cultures in 2% glucose. Stationary phase refers to cultures let 24 hours in the incubators after saturation, corresponding to OD595 2.4 to 2.7 in 0.2% glucose and OD595 6 to 7 in 2% glucose. Then, 1.4×107cells were collected, concentrated in 0.1 mL of their own medium and incubated in DiOC6 0.175 µM 15 minutes at 30°C. Next, 50 µL of this volume was diluted in 0.95 mL of 1× PBS and flow cytometry analysis was carried out as described above.
Total RNA were reverse transcribed in a final volume of 100 µL using the High Capacity cDNA Reverse Transcription Kit with random primers (Applied Biosystems, Foster City, CA) as described by the manufacturer. Reverse transcribed samples were stored at −20°C. A reference RNA (Human reference total RNA, Stratagene, Ca) was also transcribed in cDNA. Gene expression level was determined using primer and probe sets provided upon request. PCR reactions for 384 well plate formats were performed using 2 µL of cDNA samples (50 ng), 5 µL of the Express qPCR SuperMix (Invitrogen), 2 µM of each primer and 1 µM of the probe in a total volume of 10 µl. The ABI PRISM® 7900HT Sequence Detection System (Applied Biosystems) was used to detect the amplification level and was programmed FAST with an initial step of 3 minutes at 95°C, followed by 45 cycles of 5 seconds at 95°C and 30 seconds at 60°C. All reactions were run in triplicate and the average values were used for quantification. The relative quantification of target genes was determined using the ▵▵CT method. Briefly, the Ct (threshold cycle) values of target genes were normalized independently to endogenous control genes (▵CT = Ct target−Ct endogenous) and compared with a calibrator (WT 2% glucose sample C): ▵▵CT = ▵Ct Sample−▵Ct Calibrator. Relative expression (RQ) was calculated using the Sequence Detection System (SDS) 2.2.2 software (Applied Biosystems) and the formula is RQ = 2−▵▵CT. All gene expression (RQ) represents the average of three RQ from three independent experiments. Standard deviations were calculated with these three RQ. Two endogenous control genes were used: Top1+ and SPBC887.02; both selected to be highly and constitutively expressed during stationary phase. Similar results were obtained with both of them. Results showed were calculated with Top1+.
5 mL of stationary phase culture (day 1) was resuspended in 300 µL guanidinium isothiocyanate Solution (Guanidinium Isothiocyanate 4 M, Sodium Citrate 25 mM, pH 7.0, β-Mercaptoethanol 1 M) in 2 mL screw cap tubes. 0.3 mL of RNase-free beads was added and vortexed 4 times 30 seconds with Bead Beater. All the homogenate was transferred to a 2 mL Phase Lock Tube (PLG) (Qiagen). 26 µL Sodium Acetate 2 M (pH 4.0) was added to the sample, cap the PLG tube and mix briefly. 260 µL water-saturated phenol was added to the sample, cap the PLG tube, and mix thoroughly. 75 µL Chloroform: Isoamyl Alcohol (49∶1) was added to the sample in the same PLG 2 ml tube and mix thoroughly by repeated gentle inversion. Incubate on ice for 15 minutes, and centrifuge at 13,000 rpm for 5 minutes in a microcentrifuge. The aqueous phase was transferred to a new pre-spin PLG 2 ml tube, 250 µL Phenol-Chloroform-Isoamyl Alcohol (50∶49∶1) was added and mixed thoroughly by repeated gentle inversion and centrifuge. In the same PLG tube, 250 µL Phenol-Chloroform-Isoamyl Alcohol (50∶49∶1) was added, then mixed and centrifuged. The resultant aqueous phase was collected; an equal volume of 100% Isopropanol was added, and mixed by repeated inversion. The solution was centrifuged at 13 000 rpm for 20 min at 4°C. The resultant supernatant was discarded and the pellet was washed 4 times with 200 µL 80% ethanol, using 2 minutes centrifugation to re-pellet the sample if necessary. The final wash was discarded and the pellet dried at room temperature. Finally, the pellet was dissolved in 100 µL RNase-free water and stored at −70°C. RNA integrity was checked on 1.5% agarose gel electrophoresis with RNA loading buffer (Qiagen).
The code in parenthesis refers to pombe genome project nomenclature git3+ (SPCC1753.02c); gpa2+/git8+ (SPAC23H3.13c); hxk1+ (SPAC24H6.04); hxk2+ (SPAC4F8.07c); fbp1+ (SPBC1198.14c); sod1+ (SPAC821.10c); sod2+ (SPAC1486.01); gpx1+ (SPBC32F12.03c); top1+ (SPBC1703.14c); unnamed a chloride channel (SPBC1703.14c)
|
10.1371/journal.pcbi.1003150 | A Mixture of Delta-Rules Approximation to Bayesian Inference in Change-Point Problems | Error-driven learning rules have received considerable attention because of their close relationships to both optimal theory and neurobiological mechanisms. However, basic forms of these rules are effective under only a restricted set of conditions in which the environment is stable. Recent studies have defined optimal solutions to learning problems in more general, potentially unstable, environments, but the relevance of these complex mathematical solutions to how the brain solves these problems remains unclear. Here, we show that one such Bayesian solution can be approximated by a computationally straightforward mixture of simple error-driven ‘Delta’ rules. This simpler model can make effective inferences in a dynamic environment and matches human performance on a predictive-inference task using a mixture of a small number of Delta rules. This model represents an important conceptual advance in our understanding of how the brain can use relatively simple computations to make nearly optimal inferences in a dynamic world.
| The ability to make accurate predictions is important to thrive in a dynamic world. Many predictions, like those made by a stock picker, are based, at least in part, on historical data thought also to reflect future trends. However, when unexpected changes occur, like an abrupt change in the value of a company that affects its stock price, the past can become irrelevant and we must rapidly update our beliefs. Previous research has shown that, under certain conditions, human predictions are similar to those of mathematical, ideal-observer models that make accurate predictions in the presence of change-points. Despite this progress, these models require superhuman feats of memory and computation and thus are unlikely to be implemented directly in the brain. In this work, we address this conundrum by developing an approximation to the ideal-observer model that drastically reduces the computational load with only a minimal cost in performance. We show that this model better explains human behavior than other models, including the optimal model, and suggest it as a biologically plausible model for learning and prediction.
| Decisions are often guided by beliefs about the probability and utility of potential outcomes. These beliefs are learned through past experiences that, in stable environments, can be used to generate accurate predictions. However, in dynamic environments, changes can occur that render past experiences irrelevant for predicting future outcomes. For example, after a change in government, historical tax rates may no longer be a reliable predictor of future tax rates. Thus, an important challenge faced by a decision-maker is to identify and respond to environmental change-points, corresponding to when previous beliefs should be abandoned and new beliefs should be formed.
A toy example of such a situation is shown in figure 1A, where we plot the price of a fictional stock over time. In this example, the stock price on a given day (red dots) is generated by sampling from a Gaussian distribution with variance $1 and a mean (dashed black line) that starts at $10 before changing abruptly to $20 at a change-point, perhaps caused by the favorable resolution of a court case. A trader only sees the stock price and not the underlying mean but has to make predictions about the stock price on the next day.
One common strategy for computing this prediction is based on the Delta rule:(1)According to this rule, an observation, , is used to update an existing prediction, , based on the learning rate, and the prediction error, . Despite its simplicity, this learning rule can provide effective solutions to a wide range of machine-learning problems [1], [2]. In certain forms, it can also account for numerous behavioral findings that are thought to depend on prediction-error signals represented in brainstem dopaminergic neurons, their inputs from the lateral habenula, and their targets in the basal ganglia and the anterior cingulate cortex [3]–[15].
Unfortunately, this rule does not perform particularly well in the presence of change-points. We illustrate this problem with a toy example in figure 1B and C. In panel B, we plot the predictions of this model for the toy data set when is set to 0.2. In this case, the algorithm does an excellent job of computing the mean stock value before the change-point. However, it takes a long time to adjust its predictions after the change-point, undervaluing the stock for several days. In figure 1C, we plot the predictions of the model when . In this case, the model responds rapidly to the change-point but has larger errors during periods of stability.
One way around this problem is to dynamically update the learning rate on a trial-by-trial basis between zero, indicating that no weight is given to the last observed outcome, and one, indicating that the prediction is equal to the last outcome [16], [17]. During periods of stability, a decreasing learning rate can match the current belief to the average outcome. After change-points, a high learning rate shifts beliefs away from historical data and towards more recent, and more relevant, outcomes.
These adaptive dynamics are captured by Bayesian ideal-observer models that determine the rate of learning based on the statistics of change-points and the observed data [18]–[20]. An example of the behavior of the Bayesian model is shown in figure 1D. In this case, the model uses a low learning rate in periods of stability to make predictions that are very close to the mean, then changes to a high learning rate after a change-point to adapt more quickly to the new circumstances.
Recent experimental work has shown that human subjects adaptively adjust learning rates in dynamic environments in a manner that is qualitatively consistent with these algorithms [16], [17], [21]. However, it is unlikely that subjects are basing these adjustments on a direct neural implementation of the Bayesian algorithms, which are complex and computationally demanding. Thus, in this paper we ask two questions: 1) Is there a simpler, general algorithm capable of adaptively adjusting its learning rate in the presence of change-points? And 2) Does the new model better explain human behavioral data than either the full Bayesian model or a simple Delta rule? We address these questions by developing a simple approximation to the full Bayesian model. In contrast to earlier work that used a single Delta rule with an adaptive learning rate [17], [21], our model uses a mixture of biologically plausible Delta rules, each with its own, fixed learning rate, to adapt its behavior in the presence of change-points. We show that the model provides a better match to human performance than the other models. We conclude with a discussion of the biological plausibility of our model, which we propose as a general model of human learning.
Human subject protocols were approved by the University of Pennsylvania internal review board. Informed consent was given by all participants prior to taking part in the study.
To familiarize readers with change-point processes and the Bayesian model, we first review these topics in some detail and then turn our attention to the reduced model.
In this paper we are concerned with data generated from change-point processes. An example of such a process generating Gaussian data is given in figure 2. We start by defining a hazard rate, , that in the general case can be variable over time but for our purposes is assumed to be constant. Change-point locations are then generated by sampling from a Bernoulli distribution with this hazard rate, such that the probability of a change-point occurring at time is (figure 2A). In between change-points, in periods we term ‘epochs,’ the generative parameters of the data are constant. Within each epoch, the values of the generative parameters, , are sampled from a prior distribution , for some hyper-parameters and that will be described in more detail in the following sections. For the Gaussian example, is simply the mean of the Gaussian at each time point. We generate this mean for each epoch (figure 2B) by sampling from the prior distribution shown in figure 2C. Finally, we sample the data points at each time , from the generative distribution (figure 2D and E).
The goal of the full Bayesian model [18], [19] is to make accurate predictions in the presence of change-points. This model infers the predictive distribution, , over the next data point, , given the data observed up to time , .
In the case where the change-point locations are known, computing the predictive distribution is straightforward. In particular, because the parameters of the generative distribution are resampled independently at a change-point (more technically, the change-points separate the data into product partitions [22]) only data seen since the last change-point are relevant for predicting the future. Therefore, if we define the run-length at time , , as the number of time steps since the last change-point, we can write(2)where we have introduced the shorthand to denote the predictive distribution given the last time points. Assuming that our generative distribution is parameterized by parameters , then is straightforward to write down (at least formally) as the marginal over (3)where is the inferred distribution over given the last time points, and is the likelihood of the data given the generative parameters.
When the change-point locations are unknown the situation is more complex. In particular we need to compute a probability distribution over all possible values for the run-length given the observed data. This distribution is called the run-length distribution . Once we have the run-length distribution, we can compute the predictive distribution in the following way. First we compute the expected run-length on the next trial, ; i.e.,(4)where the sum is over all possible values of the run-length at time and is the change-point prior that describes the dynamics of the run-length over time. In particular, because the run-length either increases by one, with probability in between change-points, or decreases to zero, with probability at a change-point, the change-point prior, , takes the following form(5)Given the distribution , we can then compute the predictive distribution of the data on the next trial, in the following manner,(6)where the sum is over all possible values of the run-length at time .
All that then remains is to compute the run-length distribution itself, which can be done recursively using Bayes' rule(7)Substituting in the form of the change-point prior for we get(8)Thus for each value of the run-length, all but two of the of the terms in equation 7 vanish and the algorithm has complexity of computations per timestep. Unfortunately, although this is a substantial improvement compared to complexity of a more naïve change-point model, this computation is still quite demanding. In principle, the total number of run-lengths we must consider is infinite, because we must allow for the possibility that a change-point occurred at any time in the past. In practice, however, it is usual to introduce a maximum run-length, , and define the change-point prior here to be(9)With this procedure, the complexity of the computation is bounded but still can remain dauntingly high.
Despite the elegance of the full Bayesian algorithm, it is complex, requiring a memory of a large number () of different run-lengths, which, in the worst case, is equivalent to keeping track of all the past data. Thus, it seems an unlikely model of human cognition, and a key question is whether comparable predictive performance can be achieved with a simpler, more biologically plausible algorithm. Here we introduce an approximation to the full model that addresses these issues. First we reduce the model's complexity by removing nodes from the update graph (Figure 3). Then we transform the update equation for into a Delta-rule update equation in which the sufficient statistic on each node updates independently of the other nodes. The resulting algorithm is a biologically plausible mixture of Delta-rules that is able to flexibly adapt its overall learning rate in the presence of change-points and whose performance is comparable with that of the full Bayesian model at a fraction of the computational cost. Below we derive new update equations for the sufficient statistics and the weights of each new node for this reduced model.
To more easily distinguish the full and reduced models, we use to denote run-length in the reduced model and to denote run-length in the full model. Thus, the reduced model has nodes, where node has run-length . The set of run-lengths, , are ordered such that . Unlike the full model, the run-lengths in the reduced model can take on non-integer values, which allows greater flexibility.
The first step in our approximation is to remove nodes from the update graph. This step reduces the memory demands of the algorithm but also requires us to change the update rule for the sufficient statistic and the form of the change-point prior.
Consider a node with run-length . In the full Bayesian model, the sufficient statistic for this node would be(20)Note that this form of the update relies on having computed , which is the sufficient statistic at run length . In the full Bayesian model, this procedure is straightforward because all possible run-lengths are represented. In contrast, the reduced model includes only a subset of possible run-lengths, and thus a node with run-length will not exist for some values of . Therefore, the reduced model must include a new method for updating the sufficient statistic and a new form of the change-point prior.
We first note that another way of writing the update for is as(21)This sliding-window update equation depends only on information available at node and thus does not rely on knowing the sufficient statistic at node . However, this update also has a high memory demand because, to update the sliding window, we have to subtract , which we can only do if we keep track of the previous data points on each node.
In our model, we remove the dependence on , and hence the additional memory demands, by taking the average of equation 21. This procedure leads to a memoryless (yet approximate) form of the update equation for each node. In particular, if we take the average of equation 21 with respect to , we have(22)where we have introduced as the Delta-rule's approximation to the mean sufficient statistic and(23)as the mean of the node. Dividing equation 21 by gives us the following form of the update for the mean(24)Note that this equation for the update of is a Delta rule, just like equation 1, with a fixed learning rate, . Thus, the reduced model simply has to keep track of for each node and update it using only the most recent data point. This form of update rule also allows us to interpret non-integer values of the run-length, , in terms of changes in the learning rate of the Delta rule on a continuum. In figure 4 we show the effect of this approximation on the extent to which past data points are used to compute the mean of each node. The sliding window rule computes the average across the last data points, ignoring all previous data. In contrast, the Delta rule computes a weighted average using an exponential that decays over time, which tends to slightly under-emphasize the contributions of recent data and over-emphasize the contributions of distant data relative to the sliding window.
Reducing the number of nodes in the model also requires us to change how we update the weights of each node. In particular the update for the weights, , is given as(25)This equation is similar to equation 7 but differs in the number of run-lengths available. Crucially, this difference requires an adjustment to the change-point prior. The adjusted prior should approximate the full change-point prior (Eq. 5) as closely as possible. Recall that the full prior captures the fact that the run-length either decreases to zero if there is a change-point (with prior probability ) or increases by one if there is no change-point (with prior probability ).
To see how to compute this adjusted prior in the reduced model, we first decompose the change-point prior into two terms corresponding to the possibility that a change-point will occur or not; i.e.,(26)where is the probability that the run-length is given that there was a change-point and that the previous run-length was . Similarly is the probability that the run-length is given that the previous run-length was and there was not a change-point.
The change-point case is straightforward, because a change-point always results in a transition to the shortest run-length; i.e., is zero, except when when it takes value 1.
The no change-point case, however, is more difficult. In the full model the run-length increases by 1 when there is no change-point, thus we would like to have(27)However, because the nodes have variable spacing in the reduced model, this form is not possible as there may be no node with a run-length . We thus seek an approximation such that the prior defines an average increase in run-length of 1 if there is not a change-point. That is, we require(28)For we can match this expectation exactly by setting(29)For we approximate using(30)In this case we do not match the expected increase in run-length. For the final node, , it is impossible to transition to a longer run-length and so we simply have a self transition with probability 1; i.e.,(31)Taken together with equation 26, equations 29, 30 and 31 define the change-point prior in the reduced model.
Like the full Bayesian model, our reduced model also has a graphical interpretation. Again each node, , keeps track of two quantities: 1) the mean , computed according to equation 24, and 2) the weight . As in the full model, the weights are computed by passing messages along the edge of the graph. However, the structure of the graph is slightly different, with no increasing message being sent by node and an extra ‘self’ message from to itself. The increasing message has weight(32)the self message has weight(33)and the change-point message has weight(34)Finally the new weight for each node is computed by summing all of the incoming messages to implement equation 25.
In this section we present the results of simple simulations comparing the reduced and full models, investigate the error between the reduced model's predictions and the ground truth and use our model to fit human behavior on a simple prediction task with change-points.
First we consider the simplest cases of one and two nodes with Gaussian data. These cases have particularly simple update rules, and their output is easy to understand. We then consider the more general case of many nodes to show how the reduced model retains many of the useful properties of the full model, such as keeping track of an approximate run-length distribution and being able to handle different kinds of data.
Here we derive an approximate, but analytic, expression for the average discrepancy between the predictions made by the reduced model and the ground truth generative parameters. We then use this result to compute approximately optimal node arrangements for a variety of conditions and investigate how the error varies as a function of the parameters in the model.
In this section, we ask how well our model describes human behavior by fitting versions of the model to behavioral data from a predictive-inference task [24]. Briefly, in this task, 30 human subjects (19 female, 11 male) were shown a sequence of numbers between 0 and 300 that were generated by a Gaussian change-point process. This process had a mean that was randomly sampled at every change-point and a standard deviation that was constant (set to either 5 or 10) for blocks of 200 trials. Samples were constrained to be between 0 and 300 by keeping the generative means away from these bounds (the generative means were sampled from uniform distribution [from 40 to 260]) and resampling the small fraction of samples outside of this range until they lay within the range. The hazard rate was set at 0.1 except for the first three trials following a change-point, in which case the hazard rate was zero.
The subjects were required to predict the next number in the sequence and obtained more reward the closer their predictions were to the actual outcome. In particular, subjects were required to minimize the mean absolute error between prediction and outcome, which we denote . Because prediction errors depended substantially on the specific sequence of numbers generated for the given session, the exact conversion between error and monetary reward was computed by comparing performance with two benchmarks: a lower benchmark (LB) and an higher benchmark (HB). The LB was computed as the mean absolute difference between sequential generated numbers. The HB was the mean difference between mean of the generative distribution on the previous trial and the generated number. Payout was then computed as follows:(50)
A benefit of this task design is that the effective learning rates used by subjects on a trial-by-trial basis can be computed in terms of their predictions following each observed outcome, using the relationships in equation 1. Our previous studies indicated that these learning rates varied systematically as a function of properties of the generative process, including its standard deviation and the occurrence of change-points [17], [24].
To better understand the computational basis for these behavioral findings, we compared five different inference models: the full Bayesian model (‘full’), the reduced model with 1 to 3 nodes and the approximately Bayesian model of Nassar et al [17]. The Nassar et al model instantiates an alternative hypothesis to the mixture of fixed Delta rules by using a single Delta rule with a single, adaptive learning rate to approximate Bayesian inference.
On each trial, each of these models, , produces a prediction about the location of the next data point. To simulate the effects of decision noise, we assume that the subjects' reported predictions, , are subject to noise, such that(51)where is sampled from a Gaussian distribution with mean 0 and standard deviation that we fit as a free parameter for all models.
In addition to this noise parameter, we fit the following free parameters for each model: The full model and the model of Nassar et al. have a hazard rate as their only other parameter, the one-node model has a single learning rate and the remaining models with nodes () have a hazard rate as well as the learning rates.
Our fits identified the model parameters that maximized the log likelihood of the observed human predictions, , given each of the models, , which is given by(52)We used the maximum likelihood value to approximate the log Bayesian evidence, for each model using the standard Bayesian information criterion (BIC) approximation [25], which takes into account the different numbers of parameters in the different models; i.e.,(53)where is the number of free parameters in model .
Models were then compared at the group level using the Bayesian method of Stephan et al. [26]. Briefly, this method aggregates the evidence from each of the models for each of the subjects to estimate two measures of model fit. The first, which we refer to as the ‘model probability’, is an estimate of how likely it is that a given model generated the data from a randomly chosen subject. The second, termed the ‘exceedance probability’, is the probability that one model is more likely than any of the others to have generated the behavior of all of the subjects.
An important question when interpreting the model fits is the extent to which the different models are identifiable using these analyses. In particular we are interested in the extent to which different models can be separated on the basis of their behavior and the accuracy with which the parameters of each model can be fit.
The question of model identifiability is addressed in figure 10, where we plot two confusion matrices showing the model probability (A) and the exceedance probability (B) for simulated data. These matrices were generated using simulations that matched the human-subjects experiments, with the same values of the observed stimuli, the same number of trials per experiment and the same parameter settings as found by fitting the human data. Ideally, both confusion matrices should be the identity matrix, indicating that data fit to model is always generated by model and never by any other model (e.g., [27]). However, because of noise in the data and the limited number of trials in the experiment, it is often the case that not all of the models are completely separable. In the present case, there is good separation for the Nassar et al., full, 1-node, and 2-node models and reasonable separation between the 3-node model and others. When we extended this analysis to include 4- and 5-node models, we found that they were indistinguishable from the 3-node model. Thus, these models are not included in our analyses, and we consider the ‘3-node model’ to represent a model with 3 or more nodes. Note that the confusion matrix showing the exceedance probability (figure 10B) is closer to diagonal than the model probability confusion matrix (figure 10A). This result reflects the fact that exceedance probability is computed at the group level (i.e., that all the simulated data sets were generated by model M), whereas model probability computes the chance that any given simulation is best by model .
To address the question of parameter estimability, we computed correlations between the simulated parameters and the parameter values recovered by the fitting procedure for each of the models. There was strong correspondence between the simulated and fit parameter values for all of the models and all correlations were significant (see supplementary table S1).
The 3-node model most effectively describes the human data (Figure 11), producing slightly better fits than the model of Nassar et al. at the group level. Figure 11A shows model probability, the estimated probability that any given subject is best fit by each of the models. This measure showed a slight preference for the 3-node model over the model of Nassar et al. Figure 11B shows the exceedance probability for each of the models, the probability that each of the models best fits the data at the group level. Because this measure aggregates across the group it magnifies the differences between the models and showed a clearer preference for the 3-node model. Table 1 reports the means of the corresponding fit parameters for each of the models (see also supplementary figure S1 for plots of the full distributions of the fit parameters). Consistent with the optimal parameters derived in the previous section (figure 9E), for the 2- and 3-node models, the learning rate of the 1st node is close to one (mean ∼0.95).
The world is an ever-changing place. Humans and animals must recognize these changes to make accurate predictions and good decisions. In this paper, we considered dynamic worlds in which periods of stability are interrupted by abrupt change-points that render the past irrelevant for predicting the future. Previous experimental work has shown that humans modulate their behavior in the presence of such change-points in a way that is qualitatively consistent with Bayesian models of change-point detection. However, these models appear to be too computationally demanding to be implemented directly in the brain. Thus we asked two questions: 1) Is there a simple and general algorithm capable of making good predictions in the presence of change-points? And 2) Does this algorithm explain human behavior? In this section we discuss the extent to which we have answered these questions, followed by a discussion of the question that motivated this work: Is this algorithm biologically plausible? Throughout we consider the broader implications of our answers and potential avenues for future research.
To address this question, we derived an approximation to the Bayesian model based on a mixture of Delta rules, each implemented in a separate ‘node’ of a connected graph. In this reduced model, each Delta rule has its own, fixed learning rate. The overall prediction is generated by computing a weighted sum of the predictions from each node. Because only a small number of nodes are required, the model is substantially less complex than the full Bayesian model. Qualitatively, the outputs of the reduced and full Bayesian models share many features, including the ability to quickly increase the learning rate following a change-point and reduce it during periods of stability. These features were apparent for the reduced model even with a small number of (2 or 3) nodes. Thus, effective solutions to change-point problems can be achieved with minimal computational cost.
For future work, it would be interesting to consider other generative distributions, such as a Gaussian with unknown mean and variance or multidimensional data (e.g., multidimensional Gaussians) to better assess the generality of this solution. In principle, these extensions should be straightforward to deal with in the current model, which would simply require the sufficient statistic to be a vector instead of a scalar. Another obvious extension would be to consider generative parameters that drift over time (perhaps in addition to abrupt changes at change-points) or a hazard rate that changes as a function of run-length and/or time.
To address this question, we used a model-based analysis of human behavior on a prediction task with change-points. The reduced model fit the behavioral data better than either the full Bayesian model or a single learning-rate Delta rule. Our fits also suggest that a three-node model can in many cases be sufficient to explain human performance on the task. However, our experiment did not have the power to distinguish models with more that three nodes. Thus, although the results imply that the three-node model is better than the other models we tested, we cannot rule out the possibility that humans use significantly more that three learning rates.
Despite this qualification, it is an intriguing idea that the brain might use just a handful of learning rates. Our theoretical analysis suggests that this scheme would yield only a small cost in performance for the variety of different problems considered here. In this regard, our model can be seen as complementary to recent work showing that in many probabilistic-inference problems faced by humans [28] and pigeons [29], as few as just one sample from the posterior can be enough to generate good solutions.
It is also interesting to note that, for models with more than one node, the fastest learning rate was always close to one. Such a high learning rate corresponds to a Delta rule that does not integrate any information over time and simply uses the last outcome to form a prediction. This qualitative difference in the behavior of the fastest node could indicate a very different underlying process such as working memory for the last trial as is proposed in [30], [31].
One situation in which many nodes would be advantageous is the case in which the hazard rate changes as a function of run-length. In this case, only having a few run-lengths available would be problematic, because the changing hazard rate would be difficult to represent. Experiments designed to measure the effects of variable hazard rates on the ability to make predictions might therefore be able to distinguish whether multiple Delta rules are indeed present.
The question of biological plausibility is always difficult to answer in computational neuroscience. This difficulty is especially true when the focus of the model is at the algorithmic level and is not directly tied to a specific neural architecture, like in this study. Nevertheless, one useful approach to help guide an answer to this question is to associate key components of the algorithm to known neurobiological mechanisms. Here we support the biological plausibility of our reduced model by showing that signatures of all the elements necessary to implement it have been observed in neural data.
In the reduced model, the update of each node uses a simple Delta rule with a fixed learning rate. The ‘Delta’ of such an update rule corresponds to a prediction error, correlates of which have been found throughout the brain, including notably brainstem dopaminergic neurons and their targets, and have been used extensively to model behavioral data [3]–[15].
More recently, several studies have also shown evidence for representations of different learning rates, as required by the model. Human subjects performing a statistical-learning task used a pair of learning rates, one fast and one slow, that were associated with BOLD activity in two different brain areas, with the hippocampus responsible for slow learning and the striatum for fast learning [32]. A related fMRI study showed different temporal integration in one network of brain areas including the amygdala versus another, more sensory network [33]. Complementary work at the neural level found a reservoir of many different learning rates in three brain regions (anterior cingulate cortex, dorsolateral prefrontal cortex, and the lateral intraparietal area) of monkeys performing a competitive game [34]. Likewise, neural correlates of different learning rates have been identified in each of the ventral tegmental area and habenula [35]. Finally, outside of the reward system, other fMRI studies using scrambled movies have found evidence for temporal receptive fields of increasingly long time scales (equivalent to decreasingly small learning rates) up the sensory processing hierarchy [36].
Applied to our model, these results suggest that each node is implemented in a distinct, although not necessarily anatomically separated, population of neurons. For our task and the above-referenced studies, in which trials last on the order of seconds, we speculate that the mean of a node is encoded in persistent firing of neurons. Alternatively, for tasks requiring learning over longer timescales, other mechanisms such as changes in synaptic weights might play key roles in these computations.
Our model also depends on the run-length distribution, . Functionally, this distribution serves as a weighting function, determining how each of the different nodes (corresponding to different run lengths) contributes to the final prediction. In this regard, the run-length distribution can be thought of as an attentional filter, similar to mechanisms of spatial or feature-based attention, evident in multiple brain regions that enhance the output of certain signals and suppress others. For longer timescales, this kind of weighting process might have analogies to certain mechanisms of perceptual decision-making that involve the readout of appropriate sensory neurons [37]. Intriguingly, these readout mechanisms are thought to be shaped by experience – governed by a Delta-rule learning process – to ultimately enhance the most reliable sensory outputs and suppress the others [38], [39]. We speculate that a similar process might help select, from a reservoir of nodes with different learning rates, those that can most effectively solve a particular task.
The brain must also solve another challenge to directly implement the run-length distribution in our model. In particular, the update equation for the weights (Eq. 25) includes a constant of proportionality that serves to normalize the probability distribution. On a computer, ensuring that the run-length distribution is normalized is relatively straightforward: after the update we just divide by the sum of the node weights. In the brain, this procedure requires some kind of global divisive normalization among all areas coding different nodes. While such divisive normalization is thought to occur in the brain [40], it may be more difficult to implement over different brain regions that are far apart.
|
10.1371/journal.ppat.1000480 | Innate Immune Sensing of Modified Vaccinia Virus Ankara (MVA) Is Mediated by TLR2-TLR6, MDA-5 and the NALP3 Inflammasome | Modified vaccinia virus Ankara (MVA) is an attenuated double-stranded DNA poxvirus currently developed as a vaccine vector against HIV/AIDS. Profiling of the innate immune responses induced by MVA is essential for the design of vaccine vectors and for anticipating potential adverse interactions between naturally acquired and vaccine-induced immune responses. Here we report on innate immune sensing of MVA and cytokine responses in human THP-1 cells, primary human macrophages and mouse bone marrow-derived macrophages (BMDMs). The innate immune responses elicited by MVA in human macrophages were characterized by a robust chemokine production and a fairly weak pro-inflammatory cytokine response. Analyses of the cytokine production profile of macrophages isolated from knockout mice deficient in Toll-like receptors (TLRs) or in the adapter molecules MyD88 and TRIF revealed a critical role for TLR2, TLR6 and MyD88 in the production of IFNβ-independent chemokines. MVA induced a marked up-regulation of the expression of RIG-I like receptors (RLR) and the IPS-1 adapter (also known as Cardif, MAVS or VISA). Reduced expression of RIG-I, MDA-5 and IPS-1 by shRNAs indicated that sensing of MVA by RLR and production of IFNβ and IFNβ-dependent chemokines was controlled by the MDA-5 and IPS-1 pathway in the macrophage. Crosstalk between TLR2-MyD88 and the NALP3 inflammasome was essential for expression and processing of IL-1β. Transcription of the Il1b gene was markedly impaired in TLR2−/− and MyD88−/− BMDM, whereas mature and secreted IL-1β was massively reduced in NALP3−/− BMDMs or in human THP-1 macrophages with reduced expression of NALP3, ASC or caspase-1 by shRNAs. Innate immune sensing of MVA and production of chemokines, IFNβ and IL-1β by macrophages is mediated by the TLR2-TLR6-MyD88, MDA-5-IPS-1 and NALP3 inflammasome pathways. Delineation of the host response induced by MVA is critical for improving our understanding of poxvirus antiviral escape mechanisms and for designing new MVA vaccine vectors with improved immunogenicity.
| Modified vaccinia virus Ankara (MVA) is a highly attenuated, replication-deficient, poxvirus currently developed as a vaccine vector against a broad spectrum of infectious diseases including HIV, tuberculosis and malaria. It is well known that robust activation of innate immunity is essential to achieve an efficient vaccine response, and that poxviruses have developed numerous strategies to block the innate immune response. Yet, the precise mechanisms underlying innate immune sensing of MVA are poorly characterized. Toll-like receptors (TLR), RIG-I-like receptors (RLR) and NOD-like receptors (NLR) are families of membrane-bound and cytosolic sensors that detect the presence of microbial products and initiate host innate and adaptive immune responses. Here, we report the first comprehensive study of MVA sensing by innate immune cells, demonstrating that TLR2-TLR6-MyD88, MDA-5-IPS-1 and NALP3 inflammasome pathways play specific and coordinated roles in regulating cytokine, chemokine and interferon response to MVA poxvirus infection. Delineation of the pathways involved in the sensing of MVA by the host could help designing modified vectors with increased immunogenicity, which would be of particular importance since MVA is considered as a leading vaccine for HIV/AIDS vaccine following the recent failure of an adenovirus-mediated HIV vaccine trial.
| Attenuated poxviruses are currently being developed as vaccines vectors against various infectious diseases including HIV, malaria and tuberculosis [1]. Modified vaccinia virus Ankara (MVA) and NYVAC are highly attenuated poxvirus strains due to multiple deletions of viral genes and are replication-deficient in human cells. MVA and NYVAC are immunogenic and safe and have been shown to be excellent vaccine vectors for the expression of foreign antigens. MVA is a leading vaccine candidate for delivery of HIV genes with efficient induction of T-cell mediated immune responses [1]–[3]. Profiling of the immune responses triggered by poxvirus vaccine vectors is critical not only for optimal design of vaccine vectors but also for anticipating potential harmful interactions between naturally acquired or vaccine-induced immune responses against the vaccine target. This is indeed an important lesson learned from the adenovirus type 5 (Ad5) HIV vaccine (MRKAd5) STEP trial. Pre-existing neutralizing antibodies against the Ad5 vaccine vector were found to increase the relative risk of HIV infection [4],[5]. Hence the need for extensive assessments of vaccine-induced innate and adaptive immune responses to prevent unexpected adverse events.
Sensing of invasive pathogens by sentinel innate immune cells is a fundamental feature of the host antimicrobial defense response. Toll-like receptors (TLRs), retinoic acid-inducible gene-I (RIG-I) like receptors (RLRs) and nucleotide-binding and oligomerization domain (NOD)-like receptors (NLRs) have recently emerged as central innate sensors of viruses [6]. Virus sensing by TLR occurs at the cell surface and in the endosomal compartment. At the cell surface, TLR2 or TLR4 recognize either DNA (herpes viruses) or RNA viruses (respiratory syncitial, hepatitis C and measles viruses). In the endosomal compartment, TLR7, TLR3 or TLR9 sense single stranded (vesicular stomatitis virus, Sendai, West Nile and influenza viruses) and double stranded (reovirus) RNA viruses, and DNA viruses (herpes simplex viruses, cytomegalovirus), respectively [7]–[13]. Two members of the cytosolic pattern recognition RLR receptors, RIG-I (also known as DDX58) and melanoma differentiation-associated gene 5 protein (MDA5) (also known as helicard), have been shown to function as sensors of RNA viruses [14]–[19]. RIG-I detects 5′-triphosphate of ssRNAs and short dsRNAs, while MDA5 preferentially recognizes long dsRNAs. NALP3 (NLRP3 also known as cryopyrin) is a member of the NLR family which have been involved in the sensing of both DNA (adenovirus) and RNA (rotavirus, Sendai and influenza viruses) viruses [20],[21]. NALP3, ASC and pro-caspase 1 form a multimeric cytosolic molecular complex known as the NALP3 inflammasome that controls the processing of the IL-1β cytokine precursor pro-IL-1β into IL-1β [22]. Sensing of viruses by TLRs, RLRs and NLRs activates intracellular signalling pathways resulting in the expression of pro-inflammatory cytokines and type I interferons that then act on innate immune cells to limit viral replication and promote the adaptive immune response.
Here we report that the TLR2-TLR6-MyD88, MDA-5-IPS-1 and NALP3 inflammasome pathways are the main innate sensors of MVA in the macrophage and that they induce a cytokine response profile characterized by a vigorous chemokine, IFNβ and IL-1β production. Beyond the dissection of the molecular bases of MVA recognition by the innate immune system the present data are likely to help design MVA vaccine vectors with improved immunogenicity.
The profile of innate immune responses elicited by MVA was first examined by RT-PCR and ELISA in a mouse model of poxvirus infection [23]. MVA infection induced a robust innate immune response in peritoneal cells, peritoneal lavage fluid, splenocytes and splenocyte homogenates characterized by the production of pro-inflammatory cytokines (TNF, IL-1β, IL-6, IL-12p40), chemokines (IP-10/CXCL10, RANTES/CCL5, MCP-5/CCL12, MIP-2/CXCL2) and type I interferon (IFNβ) mRNA and protein (Figure 1A and B and data not shown). Infection of human whole blood with MVA also induced a vigorous innate immune response characterized by an abundant production of chemokines (IL-8/CXCL8, MIP-1α/CCL3 and IP-10) and less abundant production of pro-inflammatory cytokines (TNF, IL-1β, IL-6) (Figure 2). Interestingly, MVA was previously shown to down-regulate IL-8 and IL-1β mRNA expression in human monocyte-derived dendritic cells [24],[25], suggesting that MVA infection may induce the production of various patterns of cytokine depending upon the cell-type studied.
Dissection of the molecular mechanisms of MVA-induced innate immune responses was preformed in PMA-differentiated human THP-1 macrophages and primary human macrophages. Flow cytometry analyses performed with GFP-expressing MVA (MOI 5) indicated that MVA rapidly infected THP-1 cells (Figure 3A and B). More than 60% of cells became GFP positive within 2 h followed by a progressive decline of GFP fluorescence thereafter, which could result either from MVA-induced apoptosis as observed in human HeLa and monocyte-derived dendritic cells [24],[25] or from the shutting down of protein synthesis through activation of the PKR pathway by MVA [26]. Indeed, the number of apoptotic cells increased from 5% at 6 h to 35% at 24 h post-infection as assessed by annexin V and propidium iodine staining (data not shown).
The profile of cytokines and chemokines released by MVA-infected THP-1 cells was analyzed with the Luminex technology. Twenty four h after infection, 12 of the 30 mediators analyzed (see Materials and Methods) were detectable in cell-culture supernatants. Similarly to the results obtained with human whole blood (Figure 2) and in agreement with a recent report by Lehmann et al. [27], MVA induced the production of large quantities of chemokines (IL-8, MIP-1α, MIP-1β/CCL4, MCP-1/CCL2, RANTES and IP-10). MVA also induced large amounts of IFNβ and of IL-1ra, but small amounts of pro-inflammatory cytokines (TNF, IL-1α, IL-1β, IL-6 and IL-12p40) (Figure 3C and D). Kinetics and patterns of chemokines and type I interferon mRNA expression were similar in MVA-stimulated THP-1 cells and primary human macrophages (Figure 3E and F). We then also examined the production of cytokines and chemokines induced by two other vaccinia virus (i.e. the attenuated NYVAC strain and the virulent Western Reserve strain). When compared to MVA, NYVAC induced low levels of IL-8, IL-1β and IFNβ and no TNF, IL-6, MIP-1α, RANTES or IP-10 (Figure S1). The virulent Western Reserve strain of vaccinia virus was observed to also induce low levels of IL-8 and IFNβ in THP-1 cells, but no IL-1β, MIP-1α or IP-10 (Figure S2 and data not shown).
Altogether, these results indicated that the innate immune response induced by MVA in human macrophages was characterized by a powerful chemokine production and a less abundant production of pro-inflammatory cytokines probably related to the attenuation of MVA [28]. In contrast, the NYVAC and Western reserve strains stimulated less powerful chemokine and cytokine responses, that most likely reflect differences in the expression of immunomodulatory genes in the genome of MVA, NYVAC and Western Reserve [24],[25].
TLRs have been shown to play an important role in the sensing of viruses and in the initiation of the anti-viral host defense response [29],[30]. Analyses of the TLR repertoire used by the host for sensing of MVA were conducted in bone marrow-derived macrophages (BMDMs) isolated from TLR1−/−, TLR2−/−, TLR4−/−, TLR6−/−, MyD88−/− and TRIF−/− mice and the read-out was the expression of IFN-independent chemokine MIP-2 and of IFNβ. MVA-induced MIP-2 production by BMDMs was completely abrogated in TLR2−/−, TLR6−/− and MyD88−/− cells but not in TLR1−/−, TLR4−/− and TRIF−/− cells, which produced amounts of MIP-2 similar to that of wild-type cells (Figure 4A). In contrast, the amount of IFNβ produced by TLR2−/−, TLR6−/− and MyD88−/− BMDMs was similar to that of wild-type cells (Figure 4B), a finding consistent with the notion that activation of the TLR2 pathway is not implicated in the production of type I IFNs. Similar results were obtained with THP-1 cells stably transduced with a lentiviral delivery system expressing a short hairpin RNA (shRNA) targeting the expression of the TLR2 gene (Figure S3). All together, these results indicated that the activation of the TLR2-TLR6-MyD88 pathway was required for the induction of IFNβ-independent chemokines in MVA-stimulated macrophages. Experiments conducted with NYVAC and the Western Reserve strain of vaccinia virus confirmed that TLR2 was required for IL-8 production by THP-1 cells (Figure S1 and S2).
Vaccinia virus penetrates into target cells either by endocytosis or by membrane fusion in a low pH-independent manner [31]. To determine the contribution of endocytosis to MVA-induced intracellular signalling and cytokine production, THP-1 cells were treated with cytochalasine D, an actin-depolymerizing drug that blocks the endocytotic trafficking, or with chloroquine, a lysosomotropic weak base to neutralize the acidic environment of endocytic vesicles. IL-1β and to a lesser extend IFNβ production were inhibited by cytochalasine D and chloroquine treatment. The inhibition was not related to drug toxicity because chloroquine did not affect IL-8 production and cell viability (Figure 5 and data not shown). The reason why the inhibition of cytokine production (particularly IFNβ) was only partial after treatment with the inhibitors remains uncertain. The data suggest that additional non-endocytic pathways may play a role in the production of IFNβ. In agreement with a key role for membrane-bound TLR2 for IL-8 induction, the production of IL-8 was not reduced after cytochalasine D or chloroquine treatment (Figure 5). UV treatment of MVA causing a nearly complete (i.e. 90%) inhibition of the expression of the early C6L gene (data not shown) did not affect IL-1β, IL-8 and IFNβ production (Figure 5). Although one cannot completely rule out a contribution of residual viral protein synthesis, these observations support the view that induction of cytokines by MVA is most likely independent of viral gene synthesis [32]–[34]. Overall, endocytosis of MVA was required for IL-1β and IFNβ release suggesting a role for intracellular pattern recognition receptors in the production of these cytokines.
The RLR family of cytosolic pattern recognition receptors has been implicated in the sensing of RNA viruses [35], but very little is known about their role in host response to DNA viruses. Extending the observations by Guerra et al. who noted an increased expression of RIG-I and MDA-5 mRNA in human dendritic cells infected with MVA [24], we observed that MVA caused a time-dependent increase in RIG-I, MDA-5 and IPS-1 mRNA and protein expression in THP-1 cells (Figure 6A and B). RIG-I and MDA-5 mRNAs rose within 3 h of infection and remained elevated for up to 24 h (Figure 6A). In vivo, MVA up-regulated RIG-I and MDA-5 mRNA levels in peritoneal cells and splenocytes (Figure S4). When compared to MVA, NYVAC induced lower levels of MDA-5 and, to a lesser extent, RIG-I and IPS-1 mRNA and protein expressions (Figure S1 and data not shown). Using shRIG-I, shMDA-5 and shIPS-1 THP-1 cells (Figure S5), we then examined whether RIG-I and MDA-5 were involved in MVA-induced IFNβ production. IFNβ and IP-10 mRNA and protein levels were markedly reduced in shMDA-5 and shIPS-1 cells, but not in shRIG-I cells. By contrast, the time-course and magnitude of the IL-8 and IL-1β production was similar in shMDA-5, shIPS-1, shRIG-I and control THP-1 cells (Figure 7A and B). Sensing of MVA by the MDA-5/IPS-1 pathway is therefore critical for the production of IFNβ and IFNβ-dependent chemokines in macrophages. In line with these data, the production of IFNβ, but not of IL-8, was also dependent on the MDA-5/IPS-1 pathway in cells infected with NYVAC and the Western Reserve strain of vaccinia virus (Figure S1 and S2).
IL-1β is a key cytokine of antimicrobial host defenses, whose expression is regulated at a transcriptional and post-transcriptional level [36]. IL-1β is likely to play an important role during poxvirus infection, as suggested by the fact that poxviruses encode for IL-1β decoy receptor and disrupt intracellular IL-1 receptor signalling [37],[38]. We therefore examined whether activation of the TLR2-MyD88 pathway was implicated in the activation of the IL1b gene. As shown in Figure 8A, up-regulation of IL-1β mRNA was markedly impaired in TLR2−/− and MyD88−/− BMDMs infected with MVA, indicating that activation of the TLR2-MyD88 signalling pathway is critical for transcription of the IL1b gene during MVA infection. Secretion of mature IL-1β p17 in response to endogenous and exogenous danger signals requires the cleavage of the inactive pro-IL-1β precursor by the cysteine protease caspase-1. Conversion of pro-caspase-1 into caspase-1 is tightly regulated by the NALP3 inflammasome composed of NALP3, ASC and pro-caspase-1 [22]. To examine the contribution of the NALP3 inflammasome in the production of IL-1β triggered by MVA, we analyzed the expression of pro-IL-1β and IL-1β p17 in THP-1 cells deficient in NALP3, ASC or caspase-1 [39]. Knocking down of either one of the three components of the NALP3 inflammasome (i.e. NALP3, ASC or caspase-1) was associated with a massive reduction of mature and secreted IL-1β (Figure 8B and C). Similar results were obtained in THP-1 cells infected with NYVAC (Figure S1) and in NALP3−/− BMDMs infected with MVA (Figure 8D and E). Of note, in THP-1 cells and in BMDMs the expression of pro-IL-1β was unaffected by the absence of either NALP3, ASC or caspase-1 clearly indicating that NALP3 inflammasome does not itself regulate the transcriptional and translation control of the IL-1β precursor. The NALP3 inflammasome was also dispensable for activation of the IRF3 transcription factor and IFNβ secretion (Figure S6). Altogether, these data demonstrate that IL-1β production after MVA infection requires a crosstalk between TLR2-MyD88 (initiation of the transcription and translational of IL-1β) and the NALP3 inflammasome (processing of pro-IL-1β into mature IL-1β).
Poxviruses have been reported to activate the NF-κB, ERK1/2 and JNK pathways in epithelial and fibroblastic cell lines [40]–[43] and IRF3 and IRF7 in dendritic cells [24],[25]. Having identified the pathogen recognition receptors implicated in macrophage response to MVA (TLR2-TLR6, MDA-5 and NALP3), we next examined which downstream signalling pathways are activated for the expression of cytokines, chemokines and type I IFNs. Kinetics studies of NF-κB, ERK1/2 and JNK MAP kinases and IRFs activation were performed in THP-1 cells (Figure 9A). Electrophoretic mobility shift assay revealed that NF-κB nuclear content peaked 3 h after MVA infection. Phosphorylation of the ERK1/2 and JNK MAP kinases was between 1 and 6 h after infection. IRF3, which is essential for transcription of the IFNB gene, was detected 3 h after infection, peaked at 6 h and rapidly decreased thereafter. IRF7 was detected 3 h after infection and levels remained unchanged for 24 h. Phosphorylation of signal transducer and activator of transcription 1 (STAT-1), a critical target of IFNβ signalling required for the transcriptional activation of IFNβ-dependent genes, was first detected 3 h post-infection and gradually increased until 24 h (Figure 9A). The functional significance of the increased binding activity of NF-κB and phosphorylation of the IRF3 was confirmed by showing that MVA increased the transcriptional activities of multimeric-κB and IRF3-dependent-IFNβ promoter luciferase reporter vectors in transiently transfected THP-1 cells (Figure 9B and C). Confirming the importance of NF-κB and ERK1/2 in mediating innate immune response to MVA infection, pre-incubation of THP-1 cells with drugs (i.e. NEMO and U0126, see Materials and Methods) selectively inhibiting the NF-κB and ERK-1/2 signalling pathways impaired, albeit to a different extent, IL-1β (70% and 65% inhibition), IL-8 (75% and 72% inhibition) and IFNβ (28% and 42% inhibition) mRNA expression (p<0.05 for all conditions). Therefore, consistent with the fact that several pattern recognition receptors are engaged in the sensing of MVA by the innate immune system, multiple intracellular signalling pathways, including NF-κB, MAP kinases and IRFs were found to be activated upon infection of THP-1 macrophages with MVA. Of note, NYVAC induced very weak induction of intracellular signalling (i.e. NF-κB, ERK-1/2, IRF3 and STAT-1) and low levels of cytokines and IFNβ when compared with MVA (Figure S1) which is likely due to the expression of different patterns of immunomodulatory genes by these two poxviruses [24],[25].
Analyses of pattern recognition receptors engagement by poxviruses are essential for improving our understanding of the pathogenesis of this important class of DNA viruses and for designing new viral vaccine vectors with improved immunogenicity. Dissection of the molecular bases of innate immune responses elicited by the attenuated poxvirus MVA strain in human macrophages revealed a critical role for TLR2-TLR6-MyD88, MDA-5-IPS-1 and NALP3 inflammasome pathways in the production of chemokines, IFNβ and IL-1β. These observations provide novel information on MVA recognition by sentinel innate immune cells and highlight the existence of potential differences between attenuated and non-attenuated poxviruses in the engagement of or recognition by innate sensors.
Up to now the retinoic acid-inducible gene-I-like receptors (RLR) RIG-I and MDA-5 had been viewed as master cytosolic sensors of RNA viruses [29]. However, recent observations suggested a role for the RLR pathway in the recognition of DNA viruses. Mouse embryo fibroblasts deficient in IPS-1 displayed reduced induction of IFNβ in response to MVA lacking the E3 protein [44]. Adenovirus and HSV1 have also been shown to replicate at much higher titers in RIG-I mutant than in RIG-I wild-type human hepatoma cell lines [45]. Moreover, microarray analyses revealed that RIG-I and MDA-5 expression was upregulated in human monocyte-derived dendritic cells infected with MVA [24]. Here we also showed that MVA caused a strong up-regulation of RIG-I, MDA-5 and IPS-1, yet only MDA-5 and IPS-1 were found to mediate MVA-induced IFNβ and IFNβ-dependent chemokine production by macrophages (Figure 10). As anticipated, transcriptional activation of IFNb and IFNb-dependent chemokine genes was associated with the activation of IRF3 and IRF7 and STAT-1. To the best of our knowledge this is the first demonstration of a direct role for MDA-5 in innate sensing of a DNA virus. Moreover, the MDA-5/IPS-1 pathway was also implicated in the production of IFNβ by macrophages infected with the NYVAC and the Western Reserve strains of vaccinia virus (Figure S1 and S2).
RIG-I has been shown to be involved in the induction of TNF and type I IFN by myxoma poxvirus in human macrophages [46]. Yet, silencing of MDA-5 was associated with a small (about 25%) but clear reduction of macrophages response to myxoma virus suggesting that both RIG-I and MDA-5 were implicated, albeit to various degree, in innate immune response to myxoma virus. The nature of the component(s) of DNA viruses activating the RLR pathway remains to be identified. Obvious candidate molecules include, envelope or core proteins, early mRNA and DNA itself. Unless RLR engagement is used primarily to the virus own benefit, it is likely that poxviruses have developed antiviral escape strategies interfering with the host RLR antiviral defense pathway. In line with this assumption, the dsRNA binding protein E3 of vaccinia virus has been reported to inhibit IPS-1 signaling, IRF3 phosphorylation, cytokine and IFNβ production [47]–[49]. Should inhibitors of the RLR pathway be identified in the MVA genome, gene deletion might provide an opportunity to generate new MVA vaccine vectors with increased immunogenicity.
In addition to RLR, profiling of the cytokine response induced by MVA in the macrophage revealed a key role for the heterodimeric TLR2-TLR6 complex and the adapter protein MyD88 in the production of IFNβ-independent chemokines (such as IL-8, MIP-1α, MIP-1β and MIP-2) (Figure 10). Innate immune recognition of the vaccinia virus has also been shown to depend on TLR2 and MyD88 [32]. The present observation is one of the few examples of viral recognition mediated by TLR2 heterodimers. Recognition of human cytomegalovirus has been shown to be mediated by a TLR2-TLR1 heterocomplex and that of hepatitis C virus by either TLR2-TLR1 or TLR2-TLR6 [50],[51]. The facts that TLR2 is expressed at the cell surface and that the inhibition of endocytosis or UV-irradiation of MVA did not affect IL-8 production by macrophage suggest that a component of the MVA envelope or a core protein is responsible for the activation of the TLR2-TLR6-MyD88 pathway. However, the nature of the viral component likely to serve as ligands for these TLR2-TLR1/TLR6 heterodimers has so far remained elusive.
Other TLRs have also been implicated as innate sensors of poxviruses. Ectromelia virus, the causative agent of mousepox, was shown to be recognized by mouse dendritic cells in TLR9 dependent manner [33]. In contrast, responses of dendritic cells to MVA was both TLR9-dependent (up-regulation of CD40) and TLR9-independent (up-regulation of CD69 and production of IFNα and IL-6) [33],[52]. Although we did not perform experiments with TLR9-deficient macrophages in the present study, the data obtained with MyD88 deficient cells clearly rule out the implication of TLR9 in MVA-induced IFNβ and IFNβ-dependent chemokines. However, we cannot exclude the involvement of TLR9 in the production of IFNβ-independent chemokines. Finally, in a mouse model activation of TLR3 contributed to the pathogenesis of Western Reserve vaccinia virus [53]. In contrast, experiments conducted with TRIF-deficient macrophages clearly showed that the production of chemokines and IFNβ induced by MVA was TLR3-independent in the present study. Taken together these observations demonstrate that TLRs may exert a two-sided role in poxvirus infections acting on the one hand as key initiators of the host anti-poxvirus defense response and on the other hand as important mediators of viral pathogenicity and tissue damage.
The other important intracellular innate immune sensor of microbial products and endogenous molecules is the NALP3 inflammasome that controls the processing and maturation of the cytokines IL-1β and IL-18 [22]. Here we show that MVA is a potent activator of the NALP3 inflammasome and of IL-1β release by macrophages. IL-1β and IL-18 are key mediators of the host antimicrobial defense response and several lines of evidence suggest that these cytokines are likely to play an important role in host defenses against poxvirus infections. For example, the B15R gene of the vaccinia virus encodes an IL-1β decoy receptor blocking the activity of IL-1β and IL-18 and inactivation of B15R gene reduces the virulence of the vaccinia virus [38],[54]. Furthermore, poxviruses release IL-18 binding proteins inhibiting IL-18 activity and vaccinia viruses A46R, A52R, N1L and, K1L gene products have been shown to disrupt the IL-1 receptor intracellular signaling pathway at multiple levels [37],[55]. Interestingly, we observed that MVA stimulated the release of large amounts of the IL-1 receptor antagonist by macrophages (Figure 3C) adding further support to the view that IL-1 is an important target of the poxvirus antiviral escape strategy. Finally, consistent with the notion that the NALP3 inflammasome plays an important role in host defenses against poxviruses, several inhibitors of caspase-1 and ASC, like CrmA (cowpox virus), M13L-PYD (myxoma virus) and PYD-only (shope fibroma-virus) have been identified in the genomes of several poxviruses [56]–[58].
Crosstalks between TLRs and NLRs have been demonstrated to occur in the course of bacterial infections, such as between TLR5 and the IPAF inflammasome after exposure to flagellated bacteria or the flagellin protein itself [59]–[61]. To the best of our knowledge, however, the present data provide the first demonstration of a crosstalk between the TLR and NLR pathways in the context of a viral infection (Figure 10). While TLR2 and MyD88 were necessary to induce IL-1β mRNA expression (Figure 8A), the NALP3 inflammasome was absolutely required for the processing of pro-IL-1β and IL-1β secretion (Figure 8B and C). Dual activation pathways coupling MVA recognition to IL-1β may provide the host with an increased capacity of fine tuning of its cytokine response.
In summary, the present data show that the TLR2-TLR6-MyD88, MDA-5-IPS-1 and NALP3 inflammasome pathways exert both specific and coordinated functions in the sensing of MVA infection and in the regulation of cytokine, chemokine and IFNβ responses (Figure 10). After the unfortunate failure of the adenovirus type 5 HIV vaccine STEP trial due to issues related to natural immunity against this virus, the attenuated MVA and NYVAC strains of poxvirus have become attractive vaccine vectors against HIV/AIDS. Arguments supporting the use of MVA and NYVAC as vaccine vectors include excellent immunogenicity and safety profiles and limited pre-existing immunity to poxvirus in the population at risk of HIV infection due to the abandon of vaccine campaigns after the eradication of smallpox in the 1970s. The present findings are therefore likely to provide important information relevant to the study of the pathogenesis of poxvirus infections, the understanding of antiviral escape mechanisms of poxvirus and may help to design new vaccine vectors with increased immunogenicity.
All animal procedures were approved by the Office Vétérinaire du Canton de Vaud (authorizations n° 876.5, 876.6, 877.5 and 877.6) and performed according to our institution guidelines for animal experiments.
Eight to ten-week-old female BALB/c and C57BL/6 mice were purchased from Charles River Laboratories (L'Arbresle, France) and were acclimatized for at least one week before experimentation. MyD88−/−, TRIF−/−, TLR1−/−, TLR2−/−, TLR4−/−, TLR6−/− and NALP3−/− C57BL/6 mice have been described previously [62]–[68]. Mice were bred and housed in specific pathogen free conditions.
The human monocytic THP-1 cell line (American Type Culture Collection, Manassas, VA) was cultured in RPMI 1640 medium containing 2 mM L-glutamine, 50 µM 2-mercaptoethanol, 100 IU/ml of penicillin, 100 µg/ml of streptomycin (all from Invitrogen, San Diego, CA) and 10% heat-inactivated FCS (Sigma-Aldrich, St. Louis, MO). THP-1 cells differentiated into macrophages by treatment with 0.5 mM phorbol 12-myristate 13-acetate (PMA, Sigma-Aldrich) for 24 h were used in all experiments except those for reporter gene analyses. THP-1 cells stably expressing control, NALP3, caspase-1 and ASC shRNA have been described previously [69],[70]. THP-1 cells expressing TLR2, IPS-1, MDA-5 and RIG-I shRNA were generated using lentiviruses expressing hairpins directed against TLR2, IPS-1 and MDA-5 (5 for TLR2, 5 for IPS-1, 2 for MDA-5 and 5 for RIG-I) produced with the second-generation pMD2-VSVG and pCMV-R8.91 packaging plasmids as described previously and cultured in the presence of 5 µg/ml puromycin [71]. The sequence of the hairpins selected that gave the best targeting of TLR2, IPS-1, MDA-5 and RIG-I were AAACCCAGGGCTGCCTTGGAAAAG, CAAGTTGCCAACTAGCTCAAA, CCAACAAAGAAGCAGTGTATA and AAACCCAGGGCTGCCTTGGAAAAG, respectively. Levels of expression of targeted genes were analyzed by real-time PCR using specific oligonucleotides (Table S1) and the most efficiently silenced THP-1 subsets were selected for further studies (i.e. cell lines #1 in Figure S2).
Peripheral blood mononuclear cells from healthy donors (recruited by the Blood Center, Lausanne, Switzerland) were purified by Ficoll-Hypaque density gradient (GE Healthcare, Uppsala, Sweden). Macrophages were obtained by culturing adherent PBMCs cells for 6 days in RPMI 1640 with Glutamax. Bone marrow-derived macrophages (BMDMs) isolated from wild-type, TLR1−/−, TLR2−/−, TLR4−/−, TLR6−/−, MyD88−/− and TRIF−/− mice were cultured for 7 days in IMDM (Invitrogen) containing 50 µM 2-mercaptoethanol and monocyte-colony stimulating factor to obtain BMDMs. All media were supplemented with 10% FCS, 100 IU/ml of penicillin and 100 µg/ml of streptomycin. In selected experiments, cells were stimulated with 100 ng/ml Salmonella minnesota ultra pure LPS (List Biologicals Laboratories, Campbell, CA), 10 µg/ml polyinosine-polycytidylic acid (poly(I∶C), Invivogen, San Diego, CA), 1–10 µg/ml S-[2,3-bis(palmitoyloxy)-(2RS)-propyl]-[R]-cysteinyl-[S]-seryl-[S]-lysyl-[S]-lysyl-[S]-lysyl-[S]-lysine×3 CF3COOH (Pam2CSK4) or N-Palmitoyl-S-[2,3-bis(palmitoyloxy)-(2RS)-propyl]-[R]-cysteinyl-[S]-seryl-[S]-lysyl-[S]-lysyl-[S]-lysyl-[S]-lysine×3 HCl (Pam3CSK4) lipopeptides (EMC microcollections, Tuebingen, Germany), or treated with 50 µg/ml of anti-IFNβ antibodies (BioLegend, San Diego, CA), 2 µM cytochalasine D, 100 µM chloroquine (Sigma-Aldrich), 10 µM SB203580 (p38 inhibitor), 10 µM U0126 (MEK1/2 inhibitor) or 50 µg/ml NEMO-binding domain binding peptide (IkB kinase inhibitor) (Calbiochem-Novabiochem, Nottingham, UK).
MVA and NYVAC were cultured in chicken embryo fibroblasts and WR in HeLa cells. Viruses were purified by two sucrose cushions and titrated on BHK-21 and BSC-40 cells as previously described [24],[72]. Cells were infected with MVA, NYVAC or WR at various multiplicities of infection (MOI 1, 5 or 20 pfu/cell). After 1 h of contact with cells, the virus inoculum was removed and fresh medium added to the cultures. Cell-culture supernatants and cells were collected at different time points after infection and processed for flow cytometry, Luminex technology, ELISA, RNA extraction, and Western blot analyses. In selected experiments, MVA suspension (0.2 ml in 24-well plates laid on ice) was irradiated by a 15-min exposure to a 365-nm UV bulb at a distance of 4 cm. UV-irradiation caused a 90% inhibition of the expression of C6L early gene as determined by RT-PCR using oligonucleotides (5′-3′ sense and antisense at position −19541/−19503 and −19071/−19090 in MVA019L) AACTGCAGAAATGAATGCGTATAATAAAGCCGATTCGTTTTCTTTAGAG and CGGGATCCTTACTTGTCATCGTCGTCGTTCTTGTAGTCCSTGTTTAGGAAAAAAfAAATATC. MVA did not propagate in THP-1 cells as demonstrated by the absence of infective viral particles in cell-culture supernatants collected 24 h after infection (data not shown).
For whole blood assay, 100 µl of heparinized whole blood collected from 3 healthy volunteers were diluted 5-fold in RPMI 1640 medium containing MVA (MOI 1) and incubated for 24 h at 37°C in the presence of 5% CO2. Samples were centrifuged, and cell-free supernatants were stored at −80°C until cytokine measurement. For in vivo studies, 2×107 PFU of MVA in 1 ml phosphate-buffered saline (PBS) were injected intraperitoneally into BALB/c mice. After 12 h, a peritoneal lavage was performed. The supernatant obtained after centrifugation of the lavage fluid was collected for cytokine measurement by ELISA whereas the cell pellet was processed for gene expression analysis by RT-PCR. Spleens were collected from the same animals to quantify cytokine protein and mRNA expression levels.
To follow cell infection, THP-1 cells were infected (MOI 5) with a GFP-expressing mutant MVA, whereas all other experiments used wild-type MVA. The percentage of GFP-positive THP-1 cells was measured 0, 2, 4, 6, 12 and 24 h after infection. MVA-induced cell apoptosis was determined 6 h and 24 h post-infection using the Annexin-V FITC apoptosis detection kit according to manufacturer's recommendations (BD Biosciences, Erembodegem, Belgium). Acquisition and analysis were performed using a FACS Calibur (BD Biosciences) and FlowJo 8.5.3 software (FlowJow, Ashland, OR).
A screening of mediators produced by MVA-infected THP-1 cells was performed with the human cytokine Bioplex assay (Bio-Rad, Hercules, CA) using the Luminex technology (Luminex Corporation, Austin, TX) available at the Cardiomet Mouse Metabolic Evaluation Facility, Center for Integrative Genomics, University of Lausanne, Lausanne, Switzerland. Thirty mediators were tested: TNFα, IL-1α, IL-1ra, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12p40, IL-12p70, IL-13, IL-15, IL-17, IFNγ, RANTES, IP-10, MIP-1α, MIP-1β, MCP-1, eotaxin, fractalkine, TGFα, EGF, VEGF, GM-CSF, G-CSF and sCD40L. The concentrations of human IL-1β (Bender MedSystems, Vienna, Austria), IL-8, (BD Biosciences), IP-10, MIP-1α (R&D) and IFNβ (PBL Biomedical Laboratories, Picataway, NJ) in whole blood assay and cell-culture supernatants were measured by ELISA. TNF and IL-6 concentrations were measured by bioassay as described elsewhere [73]. Mouse IL-1β, MIP-2 (R&D) and IFNβ were quantified by ELISA (Biomedical Laboratories, Picataway, NJ).
Total RNA was isolated from THP-1 cell lines, human monocytes/macrophages, peritoneal cells and splenocytes using the RNeasy kit (Qiagen, Hombrechtikon, Switzerland). Reverse transcription of 1 µg of RNA was performed using the ImProm II RT System kit (Promega, Dübendorf, Switzerland). Quantitative PCR was performed with a 7500 Fast Real-Time PCR System (Applied Biosystems, Rotkreuz, Switzerland) using the Power SYBR Green PCR Master Mix (Applied Biosystems) and primer pairs listed in Table S1. All samples were tested in triplicates. Amplifications consisted of a denaturation step at 95°C for 15 sec and an annealing/extension step at 60°C for 60 sec, with the 9600 Emulation mode. For each measurement, a standard made of successive dilutions of a reference cDNA was processed in parallel. Gene specific expression was expressed relative to the expression of HPRT in arbitrary units (A.U.). Gene specific over HPRT ratios were validated using the house-keeping gene ACTB (human studies) or Gapdh and Actg1 (mouse studies).
THP-1 cells were seeded at 5×104 cells per well in 24-well plates. The following day, cells were transiently transfected with 700 ng of multimeric κB site [73] and IFNβ promoter [74] luciferase reporter vectors together with 70 ng of a Renilla luciferase control vector (Promega) using jetPEI™ transfection reagent (Polyplus-transfection SA, Illkirch, France). Twenty-four h after transfection, cells were infected with MVA. Luciferase and Renilla luciferase activities were measured 24 h latter using the Dual-LuciferaseTM Reporter Assay System (Promega). Results were expressed as relative luciferase activity (the ratio of luciferase to Renilla luciferase activity).
THP-1 cells were washed with ice cold PBS and lysed for 5 min at 4°C with the M-PER Mammalian Protein Extraction Reagent (Pierce Biotechnology Inc, Rockford, IL). Reaction mixtures were centrifuged 5 min at 14'000 rpm. Protein concentration of supernatants was determined using the bicinchoninic acid protein assay (Pierce Biotechnology). Cell-lysates were electrophoresed through 12% (w/v) polyacrylamide gels and transferred onto nitrocellulose membranes (Schleicher & Schuell, Keene, NH). Membranes were incubated with antibodies directed against RIG-1, MDA-5, IPS-1 (Apotech Corporation, Epalinges, Switzerland), cleaved IL-1β, total- and phospho-p44/42 (ERK1/2), and -JNK MAP Kinases, phospho-IRF3 (Cell Signalling Technology, Danvers, TX), caspase 1 (Santa Cruz, Santa Cruz, CA), phospho-STAT-1 (BD Biosciences), IRF7 (Zymed, San Franciso, CA) and tubulin (Sigma). After washing, membranes were incubated with horse radish peroxidase (HRP)-conjugated secondary antibody (Pierce). Signals were revealed using the ECL Western blotting Analysis System (GE Healthcare).
Nuclear extracts were prepared and analyzed by EMSA [73]. Briefly, protein concentration of cell extracts was measured using the Bradford-dye assay (Bio-Rad). Two µg of nuclear extracts were incubated for 15 min at room temperature with a radio-labeled consensus NF-κB probe (Santa Cruz). Reaction mixtures were electrophoresed through 6% non-denaturing polyacrylamide gels. Gels were dried and exposed to X-ray films. Supershift experiments using anti-p65 antibody (sc-109, Santa Cruz) were performed as previously described [75] (data not shown).
Comparisons among treatment groups were performed by two-tailed paired Student's t-test. p values less than 0.05 were considered to indicate statistical significance.
|
10.1371/journal.pcbi.1003006 | Lifespan Differences in Hematopoietic Stem Cells are Due to Imperfect Repair and Unstable Mean-Reversion | The life-long supply of blood cells depends on the long-term function of hematopoietic stem cells (HSCs). HSCs are functionally defined by their multi-potency and self-renewal capacity. Because of their self-renewal capacity, HSCs were thought to have indefinite lifespans. However, there is increasing evidence that genetically identical HSCs differ in lifespan and that the lifespan of a HSC is predetermined and HSC-intrinsic. Lifespan is here defined as the time a HSC gives rise to all mature blood cells. This raises the intriguing question: what controls the lifespan of HSCs within the same animal, exposed to the same environment? We present here a new model based on reliability theory to account for the diversity of lifespans of HSCs. Using clonal repopulation experiments and computational-mathematical modeling, we tested how small-scale, molecular level, failures are dissipated at the HSC population level. We found that the best fit of the experimental data is provided by a model, where the repopulation failure kinetics of each HSC are largely anti-persistent, or mean-reverting, processes. Thus, failure rates repeatedly increase during population-wide division events and are counteracted and decreased by repair processes. In the long-run, a crossover from anti-persistent to persistent behavior occurs. The cross-over is due to a slow increase in the mean failure rate of self-renewal and leads to rapid clonal extinction. This suggests that the repair capacity of HSCs is self-limiting. Furthermore, we show that the lifespan of each HSC depends on the amplitudes and frequencies of fluctuations in the failure rate kinetics. Shorter and longer lived HSCs differ significantly in their pre-programmed ability to dissipate perturbations. A likely interpretation of these findings is that the lifespan of HSCs is determined by preprogrammed differences in repair capacity.
| All hematopoietic stem cells (HSCs) are characterized by the capacities to produce all blood cell-types by differentiation and to maintain their own population through self-renewal divisions. Every individual HSC, therefore, can generate a complete blood system, or clone, conveying oxygenation and immune protection for a limited time. The time for which all mature blood cell-types can be found in a clone is called the lifespan. Interestingly, HSCs with different lifespans co-exist in the same host. We address the unresolved question: what controls the lifespan of HSCs of the same genotype exposed to the same environment? Here, we use a new approach to multi-scale modeling based on reliability theory and non-linear dynamics to address this question. Large-scale fluctuations in the experimental failure rate kinetics of HSC clones are identified to predict small-scale, genome level, events of deep penetrance, or magnitudes that approach population size. We broadly find that one condition explains our experimental data: repair mechanisms are a priori imperfect and do not improve, nor deteriorate, during the lifespan. As a result, progressively “worse-than-old” genome replicates are generated in self-renewal. A likely interpretation of our findings is that the lifespan of adult HSCs is determined by epigenetically pre-programmed differences in repair capacity.
| Adult tissue stem cells, such as hematopoietic stem cells (HSCs), are distinguished from mature cells by the ability to generate all mature cell-types of a particular tissue (multi-potency). To generate mature cells, HSCs differentiate into cells of lower potency. The resulting loss of stem cells must be compensated for by self-renewal, i.e. cell divisions which preserve the multi-potential differentiation capacity of the ancestral HSC. The reliability with which HSCs can transfer their identity and maintain self-renewal upon proliferation has been of keen interest to the field [1], [2]. Important questions are: Are daughter HSCs “as good as old” after self-renewal? How often can individual HSCs self-renew? Do different HSCs have different self-renewal capacities? What controls the fidelity of self-renewal? These questions remain incompletely understood.
Because of their extensive self-renewal capacity, HSCs were initially thought to be immortal. This view was supported by the observation that populations of HSCs could be serially transplanted for a very long period of time - exceeding the normal lifespan of the donor [3], [4]. However, when HSCs were examined on the clonal level, extensive heterogeneity in lifespan was revealed [5]–[7]. A detailed analysis of a large panel of HSCs showed that the lifespan of individual HSCs is mathematically predictable [8]. HSCs with lifespans from 10 to nearly 60 months were found side-by-side in the same donor [8], indicating that the lifespan is pre-determined on the level of each HSC. Because lifespans of single transplanted HSCs are predictable from few initial values of their repopulation kinetic, the lifespan is a programmed HSC-specific property [8]. The population dynamics, therefore, predict that the molecular machinery which preserves self-renewal, will ultimately fail.
Several hypotheses have been developed to identify and explain how HSCs limit their lifespan. The generation-age hypothesis [9] states that for every cell division, an HSC loses some quality that is referred to as “stemness”.
According to Hayflick's hypothesis [10], the probability that somatic cells produce viable daughter cells which can themselves divide, decreases as the number of divisions increases. The decrease might be caused by progressive telomere shortening [10]. Hence, an extension of Hayflick's hypothesis predicts that stem cell self-renewal capacity should be self-limiting at the level of individual HSCs.
Yet, HSCs and other stem cells, express telomerase [11]–[14]. This enzyme repairs telomere damage and, thus, aids in preserving genomic integrity. Thus, telomere shortening alone is unlikely to explain a limited lifespan of HSCs. Indeed, mice that have been homozygously ablated for telomerase activity show only mild effects and need to be severely stressed to reveal deficiencies in the hematopoietic system [15]. Potentially in line with these findings in mice are clinical data. It was suggested that telomerase expression declines in the long-run and may be a cause for late bone marrow transplant failure [16]. Declining telomerase expression may act in conjunction with the high stressor load imposed by the many co-morbidities affecting transplant patients [17].
Another proposal suggested that, in conjunction with oxidative stresses, high levels of reactive oxygen species (ROS) could be a damaging force acting on the long-term repopulating capacity of HSCs [18], [19]. The corresponding restoring force is provided by Forkhead box class O (FoxO) transcription factors. FoxO transcription factors increase the expression of genes whose products blunt the effects of elevated ROS [20]–[22]. That different sources of self-renewal failures could be causally co-dependent is suggested by findings that oxidative stress could shorten telomeres [23].
Along-side genome stability, the preservation of epigenetic patterning is an important prerequisite to reliably produce functional daughter HSCs upon self-renewal. It has been suggested that both maintenance and de novo methylation are needed to maintain epigenetic stability [24]. The expression levels of DNA methyltransferases [25], [26] responsible for maintenance (DNMT1) and de novo methylation (DNMT3a and DNMT3b) could be important for restoring HSC multi-potency [27], [28]. Quantitative work has suggested that small failures may accumulate over time in the DNMT1 pathways leading to the loss of maintenance methylation and, ultimately, epigenetic stability [29]. Yet, neither of these mechanisms and hypotheses explain how HSCs with different lifespans co-exist in a single host.
It was suggested that HSCs could preserve their functional integrity over long periods of time by alternating between two states, called resting or quiescent, and active, respectively [30]–[32]. This idea poses that intermittent transitions to quiescence could provide the time needed to minimize the detrimental effects of repeated DNA replication and other stresses on the HSC population as a whole [33]. Elegant mathematical models of this idea have been formulated [34]–[36]. Surprisingly, quiescence may leave HSCs more vulnerable to mutations following DNA repair [37]. Quiescent and active HSCs may use different DNA repair mechanisms and the restoring pathway used by quiescent HSCs may lead to higher differentiation probabilities following re-activation. Never-the-less, when HSC quiescence was inhibited by the expression of the Wnt inhibitor Dickkopf-1, the HSC pool exhausted prematurely [38]. This suggests that periods of rest in the niche are essential for controlling HSC lifespan - supporting the idea that repair is necessary to maintain HSC lifespan.
Mathematically, the lifespan of populations has been addressed in manufacturing, engineering, actuarial and biological applications of reliability theory [39]. Reliability theory was first developed as a quality control tool to predict the time-to-failure - the manufacturing term for lifespan - of manufactured goods to determine warranty times. When examined as a population, the lifespan of manufactured goods proceeds through well-defined phases (Figure 1). First, a decline in population size is found, which is interpreted as failure due to factory error. Second, there is a period of little change, known as the useful phase of the population of goods. Thereafter, the population size declines again, this time caused by age-related failure of essential machine components (wear-out phase). The second phase can be prolonged, if goods are repaired. If repairs occur repeatedly, the useful life will be extended, yet the population of goods will fail in the end, because of a general deterioration of many essential parts. In biology, failure theory has been applied to develop general laws of aging and longevity [40]–[42], respector-ligand dissociation [43], or genome instability [44]. Here, we show that the principles of reliability and failure can be exploited to craft a new model of HSC self-renewal suggesting that HSCs differ a priori in the number of (self-)repair cycles they can undergo.
We obtained repopulation data experimentally by transplanting single HSCs into ablated mice as described previously [45]–[48]. The donor HSCs and the host type mice differed in the allelic forms of the Cluster of Differentiation 45 (CD45) antigen. The CD45 antigen is expressed in most hematopoietic cells. This allowed us to follow the mature progeny derived from the transplanted HSCs by staining white blood cells for the donor-type marker. Mice were analyzed every other month and the percent of white blood cells that stained for the donor-type marker were recorded (% donor-type cells). Together, all the data points form the repopulation kinetic as a time series (compare Algorithm 1, Input specification). Because all cell populations were derived from a single HSC, the repopulation kinetic of the clone represents the total repopulation capacity of the original HSC.
We previously showed that any HSC will eventually fail to repopulate all mature cell populations [8] (also compare summary Figure 2). The time period until the multi-potential repopulation capacity fails was called the HSC lifespan. Since the lifespan marks the loss of multi-potency, it is the time-to-failure of a stem cell clone as a system [39]. We showed previously that the time-to-failure of individual HSCs is mathematically predictable with great accuracy from few initial measurements of the repopulation kinetic [8]. Hence, the time-to-failure (lifespan) is a deterministic property of individual HSCs, but can be treated as a (Gumbel-distributed) random variable at the level of the HSC compartment.
Statistical analysis of 38 repopulation kinetics ascertains that the time-to-failure estimator is unbiased and almost efficient (compare Text S1). This means that the repopulation kinetic of a clone provides near optimal information about the time-to-failure. Of note, the time-to-failure can be modeled as a power law of the proliferative capacity of the clone (compare Text S1, eq 1). From the traditional interpretation of a power law relationship (for example, see [49]), we can expect that reliability and failure rate analyses of the clonal time-to-failure are unaffected by the time scale on which repopulation data were obtained. Therefore, the systems reliability approach can be applied to repopulation kinetics obtained using different time scales (for example, [50]).
The time-to-failure , or lifespan, of an HSC, is a deterministic quantity that measures the time until at least one mature cell population is no longer repopulated. , therefore, equals the time to the first, and last, failure of multi-potency in all, not just single, clonal HSCs.
In systems theory, reliability is defined as a conditional probability ([39]; also compare Materials & Methods (M&M), section “Reliability Theory”). Specifically, a system is said to operate reliably, if its strength is likely to exceed its load at a future time, given that the system has operated within specifications up to the present time.
The strength of a clonal system lies in the self-renewal and multi-potential differentiation capacities of its HSC population. Here, the term “capacity” can be given the rigorous quantitative meaning of “obtainable work”. Clonal experiments measure the amount of “realized work”, i.e. how much of the strength has actually been transformed into new HSCs (by self-renewal) and mature cells (by differentiation) over time, given load.
Therefore, we could use the repopulation kinetics to identify the rate at which “work” is performed to quantitate clonal reliability. Specifically, we defined the clonal repopulation reliability as the normalized area under the curve of the repopulation rate kinetic (compare Algorithm 1, Line 8; also see Figure 3, A and B). Of note, the clonal reliability can be estimated even if only few initial repopulation data are known, since an HSC's lifespan is predictable from the first few measurements of its repopulation kinetic [8].
Using the repopulation data of 38 HSC clones with lifespans ranging from to nearly 60 months (compare Figure S1 in Text S1), we first determined the repopulation rate kinetics for the whole clone (parameters shown in Table S1 of Text S1) and also for all major subpopulations of T cells, B cells and myeloid cells (representative example shown in Figure 3, A; calculation: Algorithm 1, Line 6). Next, we used the clone data to calculate the respective reliabilities (Figure 3, B; calculation: Algorithm 1, Line 8). An additional curve can be inferred, if the clonal structure is considered in a so-called “common cause” model of clonal reliability. A common cause model poses that, in a multi-component system, the unknown reliability of a central component can be approximately determined by the relationship of all remaining system components to the system's reliability ([39], pp. 217–222).
Since all mature lineages derive from HSCs, a mature cell population and the HSC population can be viewed as serially connected system components. On the other hand, the populations of mature cells appear connected in parallel, since the failure of one such population does not imply the failure of all. The serial connectivity is a mathematical way of representing multi-potency. Hence, an inferred reliability structure model should generate the reliability kinetic of the clonal HSC population.
Considering the HSC population a “common cause” with reliability denoted , we posed that the clonal reliability (denoted ) is connected to the reliabilities of the T cell, B cell and myeloid cell populations (denoted , , , respectively) by:(1)Eq 1 is a balance formula expressing the conservation of system structure over the life of a clone. It states that two parallel structure models of a clone behave similarly for all times (indicated by the notation ). The first model (left-hand side, eq 1) considers the HSC population and the clone as a parallel reliability structure. The second model (right-hand side, eq 1) quantitates the reliability of the major mature cell-types as a parallel structure. We applied eq 1 to calculate the values at discrete time points given by our data. The inferred kinetics are shown as blue curves in Figure 3.
The common cause argument suggests that the HSC population reliability closely resembles the reliability of the clone, modulo a time lag that is small compared to the lifespan (Figure 3, B). This outcome is consistent with our previous conclusions, obtained by different methods that the information contained in the repopulation time series predicts HSC behavior [8].
By definition, the reliability is a forward looking measure that predicts the chances that a system will continue to operate according to its specifications for some time into the future, provided that it has operated reliably in the past. For HSCs, reliable operation means that their main characteristics, i.e. self-renewal and multi-potential differentiation capacities, are preserved when these cells divide. What are the chances of unreliable operation?
In systems theory, unreliability is defined by the conjugate probability of the reliability (for computation compare Algorithm 1, Line 9). Specifically, the probability that the system fails, or failure probability, equals 1 minus the probability that it can continue to operate as before. Hence, the repopulation failure probability, or repopulation “unreliability”, is a cumulative probability function [39] (high/low values indicate high/low likelihood of failure). Its graphical representation is an S-shaped curve whose shape is a horizontally flipped image of the corresponding reliability kinetic (Figure 3).
The failure probability gives rise to the failure density. The latter is a probability density in the usual sense and defined by the rate with which the failure probability changes over time (small values mean little change, high values mean lots of change). We determined the failure probabilities and densities for all clonal populations (Figure 3, C and D; Algorithm 1, Line 10). Because we could predict the reliability kinetics of the HSC population (previous section; also Figure 3, B), we could predict an approximate shape for the repopulation failure density of the HSC population, as well.
Application of the method of symbolic time series comparison in [51] shows that the relationship between the failure densities of the subpopulations of a clone changes over time (Figure 3 C). Specifically, the repopulation failure densities of the HSC population and some mature cell populations becomes more similar (increasingly converge to the same symbol sequence (data not shown)), as the lifespan is approached. By contrast, the densities lack similarity at the beginning of clonal life. To clarify this observation, we conducted a more detailed analysis of the long-range failure dynamics of HSC clones with different lifespans.
In systems theory, the failure rate provides information about how system failure occurs as a function of system load acting against system strength over time [39]. Since the future behavior of HSCs is largely pre-programmed [8], [46], the failure rate kinetic informs about how strength, or capacity, is transformed into new HSCs (by self-renewal) and mature cells (by differentiation) over time.
The failure rate is defined as the ratio of failure density divided by the probability of reliable operation (which equals the negative rate of change over time of the logarithm of the reliability [39]; also compare M&M, “Reliability Theory”). We applied this definition to the reliability kinetics of the clone, each mature subpopulation and the predicted reliability of the HSC pool to obtain the respective failure rate kinetics (Figure 3, D and Algorithm 1, Line 11). The main observation is that, for all HSC clones, the failure rate kinetics of all populations, and the clone as a whole, increase sharply towards the end of clonal life. The overall behavior is valid for HSCs of all lifespans (Figure 4, A, C, E; also compare Theorems 1 and 2, and Lemma 1). The onset of the increase, which we called extinction transition, coincides with the onset of the increase in the predicted failure rate kinetic of the HSC population. However, the HSC failure rate increases more rapidly than those of the mature cell populations (Figure 3, D). This suggests that clonal extinction is due to an event that affects the HSC population as a whole and, likely, synchronously.
To better understand the failure behavior of the clonal HSC population before the extinction transition, we looked at truncated failure rate kinetics. We defined a failure rate kinetic as -truncated, if elements are removed from the beginning and from the end of the time series, respectively. Hence, the full time series is -truncated, while -truncation means removing the last elements of the time series. For all repopulation kinetics in our database (compare Figure 4 for representative examples of different lifespans; also see Figure S1 of Text S1), we analyzed the behavior of the (0,2)-truncated failure rates (Figure 4, B, D, F).
We had previously shown that past values in HSC repopulation kinetics predict future values [8]. In other words, memory of past behaviors influences the long-term behavior of repopulation kinetics. We now asked, if memory effects could be shown in the failure rate kinetics. To find out, we calculated the Hurst exponent [49], [52], [53] of each -truncated failure rate kinetic (Figure 5; compare section “Computation of Hurst Exponent” in M&M; for values compare Table S1 of Text S1). The Hurst exponent method is a statistical approach for finding long-term memory patterns in a time series. Evidence of memory is defined by the inequality , where denotes the Hurst exponent. If , the behavior of a time-series is characterized by a pattern of reversions to a mean, where decreases/increases are followed by increases/decreases. Such a pattern is called “anti-persistent” behavior, which is considered stronger the closer is to 0. When , past time series values influence future values either only in the upward, or downward, direction, i.e. increases/decreases follow increases/decreases. This pattern is aptly called “persistent behavior”. Persistence is considered stronger the closer is to 1. Values close to are interpreted as evidence that no relationship exists between past and future values of a time series. Since we had previously shown that past values of HSC repopulation kinetics predict future values [8], we could hypothesize that we might find values of significantly different from in failure rate kinetics.
To properly conducted Hurst exponent analysis, for example, using a standard approach such as the rescaled range or R/S method, knowledge of the average behavior of all sufficiently large segments of the data is required. Because clonal repopulation kinetics have a deterministic core behavior of ballistic shape [8], we could determine averages based on a deterministic failure rate kinetic for each clone (compare Theorem 1). Plotting the experimental failure rate data together with the deterministic failure rate kinetic showed that the former alternate around the latter (data not shown). Using the additional information available in the framework of HSCs, we could overcome the problem of small time series, similar to approaches in financial market analyses, where standard methods require modifications [54], [55]. Specifically, we used the ballistic trend of repopulation data as a domain-specific mean in the Hurst approach, instead of the uniform mean usually applied. For all clones, the Hurst exponents had median value (compare Figure 5; Wilcoxon test, significant difference of the median to , ). Therefore, the values of indicated that the failure rate kinetics of HSCs show anti-persistent memory behavior.
Anti-persistence describes a long-range memory behavior of a time series [52], [53], where increases in value are followed by decreases and decreases by increases, as opposed to increases/decreases following increases/decreases, as would be the case for persistent behavior. Using the traditional interpretation, the anti-persistent behavior in the failure rate kinetic of an HSC indicates that the “noise” of the failure rate data follows a long-term pattern that is informative about the biology of HSCs. This pattern suggested the presence of a mean-reverting process in the context of failure kinetics. Therefore, we next considered the possibility that mean-reverting behavior of the failure rate could, biologically, indicate the effects of repair mechanisms acting to decrease the failure rate following increases. In this model, fluctuations in the failure rates, as obtained from measurements of clonal repopulation kinetics, are viewed as indicators for the successive interaction of failure generation and repair, a marked anti-persistent behavior. What is missing, is a quantitative rationale for repair - acting at the cell level, but derived from cell population data.
Because we found evidence of mean-reversion, we asked how the -truncated failure rates would fit to realizations of the proto-type mean-reverting process, the Ornstein-Uhlenbeck process [56], [57]. The benefits of linking the HSC failure rates to this process are: (a) A quantitative rationale for repair, in the form of a failure dissipation rate; (b) An iterative model for simulating -truncated failure rate kinetics based on clonal repopulation data.
We showed numerically that the weighted sum of the variance-adjusted rate of change plus the standardized rate of each -truncated failure rate kinetic could be regressed to the rate of change of noise in the data. Mathematically:(2)Both sides of the similarity eq 2 can be determined from the data. , on the right-hand side, represents the discrete rate of change of noise isolated from the data (for normally distributed noise, is called a “Wiener process”). and denote the mean and standard deviation of the -truncated failure rate kinetic, respectively. denotes a positive weight parameter, called the “dissipation rate”. Because the dissipation rate occurs in the context of failure kinetics, we interpreted as a quantitative indicator of repair activity - implying that repair could be modeled mechanistically as a dissipation of failure. Eq 2 is equivalent to:(3)Eq 3 is a discrete form of the Ornstein-Uhlenbeck stochastic differential equation [56], [57]. Its solution, called Ornstein-Uhlenbeck process, is well-known as the prototypical mean-reverting process.
Our numerical analyses showed that had a non-linear fit of the form . The values of the weight , calculated from experimental data, were specific for individual HSCs. Furthermore, we found that in distribution, is similar to , where the second factor denotes the standardized normal distribution with mean 0 and standard deviation 1.
In Theorem 2, we proved that the deterministic repopulation kinetic [8] gives rise to a deterministic differential equation for the failure rate. This equation is formally similar to eq 3, without the noise term. We considered eq 3, therefore, as the approximate Ornstein-Uhlenbeck representation for the -truncated experimental failure rate data.
Together, these findings independently support the prediction, established earlier by Hurst analysis that the truncated failure rate kinetic of an HSC is mean-reverting. We could conclude that the -truncated failure rate kinetics can be simulated by an Ornstein-Uhlenbeck model [58], [59] using the iterative scheme:(4)We now asked, if eq 4 could be a model for the full, not only the -truncated failure rate kinetics. To answer this question in light of the extinction transition seen in the data (sharp increase of failure rates near the end of clonal life), we needed to explain how the mean-reversion property may break down. This required a closer look at the behavior of the parameters , and, in particular, , the quantitative indicator of population level repair activity through dissipation of failure.
We first asked what the properties of are. The Ornstein-Uhlenbeck formulation (eq 3) allowed us to assess HSC repair efficiency from our clonal repopulation data. We broadly defined “repair” as the total of repair mechanisms available to HSCs. The dissipation rate quantitates the strength of the restoring forces as the rate with which the system variable, in our case the failure rate, reverts toward an average behavior . We, therefore, could consider the dissipation of failures as evidence of repair activity.
quantifies how rapidly the clonal system reverts back in the direction of -equilibrium. Analysis of the dissipation rate for our experimental data showed that depends on the lifespan . Specifically, determining each by fitting to a noisy process (eq 2) and plotting the results as points , suggested that the dissipation rate has a non-linearly increasing tendency when increasing, but fixed, lifespans are considered (compare Figure 6, A; green curve). Explicitly, we found:(5)(compare Figure 6, A; green curve; goodness-of-fit: Akaike Information Criterion ; parameter p-values and , respectively). It is important to understand that eq 5 represents a tendency, not a dependency, among the dissipation rates of repopulation kinetics for independent time-to-failure values . However, going back to a remark at the beginning, the time-to-failure is a function of load which, as noted, comprises two components relating to peripheral demand, and demands due to disease or injury, respectively. Biologically, load may affect, in parallel, all accessible HSCs in the HSC compartment of a single host. Therefore, if is parametrized by load, eq 5, and the eq 6 below, may be interpreted as a power law, dependent on load exposure. This suggested that may be dependent on “running” time, as well.
To find out, we used an analytical approach that took advantage of the deterministic behavior of HSC repopulation kinetics. In Theorem 2, we showed that an analytical definition of can be found as a function of and , denoted , based on the ballistic model of repopulation kinetics developed by us previously. To see if behavior similar to eq 5 can be found analytically, we averaged , defined by , where integration extends over bounds that are analogous to -truncation of failure rates ( is explicitly given in Theorem 2). We found that the dissipation rates derived analytically or from data have similar properties (compare Figure 6, A). As before (compare eq 5), we fitted the resulting set of values to a non-linear model:(6)(compare Figure 6, A; blue curve; goodness-of-fit: Akaike Information Criterion ; parameter p-values and , respectively). Both approaches show the same tendency of the dissipation rate to increase as a point-wise function of . Hierarchical cluster analysis showed that the (lifespan, dissipation rate) data separate into three clusters. The dissipation rates of the three cluster centroids , and are significantly different (Wilcoxon Test, Bonferroni corrected p-values: , , ).
Together, our findings suggested that HSCs of different lifespans may differ in their ability to utilize repair mechanisms. We used the formulae developed in the proof of Theorem 1 to compare the dynamics of failure generation and failure dissipation (repair). As stated above, we found that, for the deterministic repopulation kinetics, the dissipation rate is a function of time and the lifespan , i.e. , for . equals the negative logarithmic derivative of the failure density function. As shown in the proof of Theorem 2, for all , with that can be calculated from data. This means that, throughout most of the lifespan period, failures are generated at a higher rate than they are dissipated. Therefore, we could use the inequality to quantitatively define “imperfect repair” and to ascertain that repair activity is largely insufficient to compensate for the rates at which failures are generated. The analytical treatment supported the notion that failures should accumulate in the long-run.
The mathematical analysis also predicted that, during the initial expansion period of a clone, quantitated by , repair capacity exceeds the rate at which failures are generated. Quantitatively, for . In particular, our theory predicts that each HSC starts “its” clone with a non-zero “initial damage load” [40]–[42]. The “initial damage load” can be calculated for each repopulation kinetic. Its precise biological meaning and, particularly, its developmental origins, will need to be determined experimentally. The switch from higher repair capacity to may be attributable to the diluting effects of clone size on the programmed, epigenetically heritable, repair capacity of the founder HSC.
Evidence favoring the hypothesis that shorter lived HSCs may be less efficient in dissipating the effects of failures than longer lived HSCs, can be biologically interpreted by placing the dissipation rate in a time-context. The time-context is given by the half-life of the dissipation rate. The half-life is defined by:(7)
The half-life data (compare Figure 6, B) show a tendency, where failure rate increases dissipate more rapidly in long-term repopulating HSCs with higher lifespans . This holds independently of the manner in which values were obtained, either analytically or from data. Mathematically:(8)(goodness-of-fit: Akaike Information Criterion ; parameter p-values and , respectively). Hierarchical cluster analysis showed that the (lifespan, half-life) data separate into three clusters. The half-lives of the three cluster centroids , , are significantly different (Wilcoxon Test; Bonferroni corrected p-values: , , ). Like eq's 5 and 6, the relationship eq 8 must be interpreted with care as is a time-to-failure variable and should not be confused with a continuum.
The data show larger failure dissipation rate half-lives for HSCs with shorter lifespans, while smaller half-lives associate with longer life. A possible interpretation is that repairs may occur less frequently in shorter lived HSCs than in longer lived HSCs. According to Theorem 1 shorter lived HSCs may have to counteract higher initial damage loads (as suggested by the higher initial values of for smaller lifespan values). Together, these findings may explain the larger and longer-lasting “peaks” and “valleys” seen in the failure rate kinetics of shorter lived HSCs (Figure 4).
An important observation common to all failure rate kinetics is that, near the end of clonal life, the failure rates strongly increase. When we compared actual failure rate kinetics with kinetics generated by simulation using the experimentally derived parameters in eq 4, we noticed that the terminal behavior of the experimental failure rate departed significantly from the mean-reverting characteristic of the simulated rates (for more detailed discussion see below). This suggested that regime-breaking may characterize the extinction transition. “Regimes” is standard terminology to describe disjoint regions in the phase space of a dynamical system such that transitions between the regions are rare (compare Figure 7 for a representative example). We, thus, analyzed the regimes of the phase space of experimental HSC failure rate kinetics.
The phase space trajectory of an HSC's failure rate kinetic starts in the initial engraftment regime (Figure 7, region E). Here, it remains for up to 3 months of the HSC's life. Then, the trajectory escapes to the region where failure rates are governed by mean-reversion (Figure 7, region OU). It remains in this regime for most of the clonal life - first contracting, then slowly expanding. At some point, the phase space trajectory transitions to a third regime (Figure 7, region T), where the clone becomes extinct. Therefore, regime-breaking is associated with clonal extinction.
We next established the conditions under which the mean-reverting regime breaks. As discussed above, comparison of the experimental failure rate kinetics with realizations of the iterated Ornstein-Uhlenbeck process (eq 4) showed that the simulated process diverges from the experimental data in the terminal regime (Figure 8, Part A). The simulated failure rates continue as before, but the experimental failure rates rise sharply. This behavior could be changed, and brought closer to the experimental data, when a further constraint was added.
Specifically, we asked whether the mean of the experimental failure rates is a constant or changes in time. Analysis of the -truncated failure rate kinetics of each clone using moving averages showed that the parameter , the mean failure rate, increases in a well-defined pattern that occurs in all kinetics (compare an example in Figure 8, B). Though this pattern of increasing mean is present in the raw failure rate data, the change there is subtle, but cumulative. It is enhanced and, thus, becomes more recognizable if moving averages are used (for short lifespans only small windows are needed; larger windows (up to 5 months) are required for longer lifespans). Using non-linear fitting, we found that the succession of moving averages, labelled , increases over time as:(9)The critical limit is slightly smaller than the time-to-failure, or lifespan, of the clone. This result obtained from data analysis has an analytical counterpart. In Theorem 1, we showed that the ballistic model of HSC repopulation kinetics [8] leads to a deterministic formula for the failure rate function of any HSC clone, such that, according to Lemma 1, in the limit . Therefore, the analytically derived result matches the behavior extracted from data in eq 9.
We could, thus, conclude that as more self-renewals occur, repair mechanisms must revert to larger failure rates means in daughter HSCs. Biologically, the increasing mean suggests that damage accumulates over successive generations, likely due to imperfect repair.
Clonal hematopoiesis begins with a single HSC and ends with its loss after months to years [4]–[8]. During this period of time - the lifespan -, the genetic/epigenetic program of the original HSC is replicated to many daughter HSCs. In HSC self-renewal, discrepancies between replicates should be vanishingly small. Yet, clones extinguish after a limited time, suggesting that self-renewal may not be perfectly reliable in the long-run.
Working from repopulation kinetics, we needed to develop quantitative measures of reliability, failure and repair that capture events of magnitudes that approach population size. Our data show that increases in failure rates associate with increasing population-wide failure loads, and decreases relate to dissipative effects of the collective repair strength. Together, the fluctuation patterns of the failure rate kinetics characterize the summary dynamics of microscopic failure and repair events at the macroscopic level of clonal HSC populations.
Our approach does not depend on particular failure sources and repair mechanisms. An advantage, therefore, is that our current understanding of stem cell Omics is not limiting. Rather, reliability theory could open new avenues for interpreting longitudinal network data. For example, we make the prediction that repair capability stays approximately constant through the lifespan of an individual HSC. Constancy does not mean that individual repair mechanisms are unaffected by failure inducing processes. Instead, the function of a deteriorating repair mechanism could be taken over by an alternative pathway.
We predict that the repopulation failure rate kinetics stay at low levels for a long time, but will never revert to zero failure rate. This supports the conclusion that failures are continuously generated, but are never completely cleared. Indeed, failure rates increase slowly, indicating that failures accumulate. Evidence of failure accumulation is not only seen in our experimental data. It also followed by mathematical proof (M&M section; Theorem 2) for the deterministic failure rate kinetics derived from our previously developed ballistic model [8].
Daughter HSCs of successive generations may, thus, carry an increasing failure load - consistent with previous experimental findings on the aging of HSCs [60]. As shown here, increasing failure load coupled with constant repair capacity is key to explaining the differences in lifespans of HSCs in a single host in the absence of strong extrinsic perturbations (e.g., expression of the Wnt inhibitor Dickkopf 1 [38]). These predictions could be tested experimentally using longitudinal genome studies comparing genetic networks in myeloid-biased (My-bi) and lymphoid-biased (Ly-bi) HSCs [47]. The former are typically longer lived than the latter. Reliability analysis may aid in predicting universal check-points (such as in Theorem 2) to meaningfully time longitudinal genome studies of My-bi and Ly-bi HSCs. So far, the genome of My-bi HSCs has only been mined at isolated instances [47], [61]. Timing of genomic analyses is experimentally challenging, since HSCs are rare relative to mature blood cell populations.
We showed that the failure rate is stable, but “noisy”. This “noise”, properly classified using rescaled range, or Hurst, analysis [53] shows how stability is generated. The experimental failure rate kinetics alternate irregularly around the deterministic failure rate component, which we used as the mean value in our rescaled range analysis of the data. We found that the Hurst exponent is for the failure rate kinetics of all long-term repopulating HSCs. HSCs with lifespans greater than one mouse life have smaller exponents than for lifespans below one mouse life. Using non-linear time series theory, we could conclude that an anti-persistent, or mean-reverting, regime governs the time period where failure rates are at a minimum. The implication is that the interaction between failure-generating mechanisms, such as HSC self-renewal divisions, and failure-dissipating processes, or repairs, creates a self-organized failure-repair equilibrium. An HSC clone exits the equilibrium state, and becomes extinct shortly there-after, when failure load has accumulated sufficiently to surpass repair capacity.
Our reliability analysis of HSCs has implications on aging in HSCs [2], [45], [62] and in other non-homogenous, hierarchically organized, cell systems. The elegant general theory of aging developed by the Gavrilovs [40]–[42] - tested extensively for populations of organisms but less for those of cells [43], [44] - poses that aging systems have three properties: redundancy, initial damage load, and redundancy depletion. For HSCs, redundancy emerges over time as a function of self-renewals. The declining quality of genome replicates, due to imperfect repair as predicted here, quantifies the rate of redundancy depletion in HSCs. The mathematical results of this paper predict that the failure dissipation rate determined from our data provides a quantitative measure of “progressive damage load” starting from an HSC-specific “initial damage load”. However, what constitutes “initial damage load” in the biology of HSCs, and what its sources are, must first be investigated experimentally - primarily to address the question of how HSCs are programmed during early development.
Freshly explanted BM cells were transplanted in limiting dilution into lethally irradiated CD45 congenic hosts exactly as described [45]–[47], [63]. Each host received on average 0.2–0.5 HSCs together with 2×transplanted BM as a source of radio-protecting cells. Mice were bled in regular intervals and the % myeloid and lymphoid cells amongst the donor-type cells were measured by Flow cytometry. All experiments were approved by the IACUC.
We used Mathematica version 8.0.1 (Wolfram Research, Inc) for numerical mathematics and computer simulations. R version 2.12.2 and Instat version 3 (GraphPad, Inc) were used for all statistical analyses. Figures were generated using Mathematica version 8.0.1 and edited using GIMP version 2.8.3. The manuscript was written in LaTeX2e using GNU Emacs version 23.3.50 (Free Software Foundation).
The subject of reliability theory is to determine the length of time for which a system is capable of bearing “load” given its material “strength”. Mathematically, system reliability versus unreliability at time is quantitated by the respective inequalities between strength and load :(10)Strength and load are non-negative functions of time, i.e. and . A system's time-to-failure is defined as the earliest point in time for which load exceeds strength, i.e.:(11)For repairable systems, multiple occurrences of are possible, only to be reset by the repair process to some level of reliable operations, i.e. . Hence, the operation of a repairable system will generate a sequence of time-to-failure values. By contrast, in an unrepairable system, the first time-to-failure is also the last.
Though load and strength can be deterministic, it is advantageous to consider the general case where the strength-load inequality (compare eq 10) is subject to uncertainty. Hence, system reliability is usually defined by a probability measure for events (colloquially: “time-to-failure not yet reached”):(12)The explicit reference to the strength-load inequality can be suppressed due to the definition of the time-to-failure (eq 11). The definition of probabilities implies that .
The failure probability is defined as the conjugate probability:(13)The rate of change of the failure probability over time is used to define the failure density:(14)This definition requires that the reliability is a differentiable function of time .
The rate of system failure, or failure rate, is determined by the ratio of the failure density to the system's reliability, provided that :(15)Due to the properties of the reliability and the failure density, it follows that for all permissible . Using eqs 14 and 13 in eq 15 shows that the failure rate can also be represented by the derivative of the negative logarithm of the reliability.
In practical applications, a system's reliability is determined based on field measurements of a particular system variable associated with well-defined, system-specific, time-to-failure conditions. From these measurements, one attempts to form a discrete empirical distribution as a time series , where , normalized such that , and (compare Algorithm 1).
Statistical distribution fitting with appropriate goodness-of-fit measures may identify a suitable closed-form model (such as in eq 12), so that . In this case, a system's reliability and failure rate evolution can be described using probability, dynamical systems theory, and stochastic processes. We show in the main narrative that the failure rate of the repopulation kinetics of hematopoietic stem cells follow an Ornstein-Uhlenbeck stochastic process.
Time series representing the evolution of discrete failure probabilities and failure densities can be obtained by point-wise application of eqs 13 and 14 (as used in Algorithm 1). In the case of the failure density, the differences operator is used to approximate differentiation. This operator is defined by(16)
We computed discrete failure rates as the ratio of density to reliability, as introduced in eq 15. This approach is computationally more efficient and more transparent than using the numerical derivative of the negative logarithm of the reliability sequences.
To make predictions about the long-term reliability of water reservoirs, Harold Hurst introduced a new statistic, called the rescaled range. This statistic is determined by forming the ratio of the difference between the water level extremes over a long time period, called the range, to the standard deviation from the mean water inflow over the same time period, but using sub-divisions of the time scale into smaller segments [64].
Because of its generality, the rescaled range, or R/S, statistic can be used in many different contexts - with appropriate reinterpretation of time scales and measured entities. Hurst's original finding, expanded by Mandelbrot and collaborators [49], [52], [53], was that, in the limit, the average rescaled range over time periods of increasing size , behaves like a fixed power of :(17) is called the Hurst exponent. An algorithm for determining the rescaled range statistic and the empirical Hurst estimate is given below (Algorithm 2). Explanations for the notation used above are given there. Implementations of the R/S and other methods for estimating the Hurst exponent are available in the statistical programming language R.
In applications, the benefit of determining the rescaled range sequence is that we can analyze and interpret data that have no characteristic scale. This is sometimes interpreted as lacking bias introduced by specific measurement scales. The Hurst exponent, , measures the smoothness of (self-similar) time series based on the asymptotics of the rescaled range sequence. indicates a persistent, or trend-reinforcing, time series. In this case, increases (decreases) of time series values are followed by increases (decreases). This trending increases as . holds for mean-reverting, or anti-persistent, time series. In this case, deviations from the mean lead to reversal of the time series values towards a long-term mean. The “strength” of the mean reversion increases as . Geometrically, anti-persistent time series will appear more jagged as . The value is considered indicative for lack of correlation in the time series: Any values do not inform about future values. In the case of the hematopoietic system, we showed previously that past values of the clonal repopulation kinetics are predictive for future values [8].
We wished to identify the failure rate kinetics under noise-less conditions. To do this, we could use the deterministic model of clonal repopulation kinetics, denoted below, that we had developed previously [8] based on experimentally obtained clonal repopulation data.
In Theorem 1, we showed that in the time limit approaching the lifespan. The following result details the asymptotic growth properties of the deterministic failure rate kinetic.
We wanted to know the order of magnitude of the “initial damage load” in relation to the lifespan . In particular, is there a lower bound on ?
We wanted to know under which conditions on time the deterministic failure rate kinetic increases and decreases. To find answers, we determined explicitly the rate of change of the noise-less failure rate kinetic derived in Theorem 1. Throughout, we used the simplified notation for the derivative of a differentiable function of variable .
To analyze the (lifespan, dissipation rate) and (lifespan, half-life) data, denoted and , respectively, in the main narrative, we first standardized the data by subtracting the sample mean and dividing by the sample standard deviation column-wise. We used hierarchical cluster analysis on the standardized data to determine the number of clusters and their centroids. Confirmatory analysis was conducted using a standard partitioning approach.
|
10.1371/journal.pbio.1001100 | Defining the Specificity of Cotranslationally Acting Chaperones by Systematic Analysis of mRNAs Associated with Ribosome-Nascent Chain Complexes | Polypeptides exiting the ribosome must fold and assemble in the crowded environment of the cell. Chaperones and other protein homeostasis factors interact with newly translated polypeptides to facilitate their folding and correct localization. Despite the extensive efforts, little is known about the specificity of the chaperones and other factors that bind nascent polypeptides. To address this question we present an approach that systematically identifies cotranslational chaperone substrates through the mRNAs associated with ribosome-nascent chain-chaperone complexes. We here focused on two Saccharomyces cerevisiae chaperones: the Signal Recognition Particle (SRP), which acts cotranslationally to target proteins to the ER, and the Nascent chain Associated Complex (NAC), whose function has been elusive. Our results provide new insights into SRP selectivity and reveal that NAC is a general cotranslational chaperone. We found surprising differential substrate specificity for the three subunits of NAC, which appear to recognize distinct features within nascent chains. Our results also revealed a partial overlap between the sets of nascent polypeptides that interact with NAC and SRP, respectively, and showed that NAC modulates SRP specificity and fidelity in vivo. These findings give us new insight into the dynamic interplay of chaperones acting on nascent chains. The strategy we used should be generally applicable to mapping the specificity, interplay, and dynamics of the cotranslational protein homeostasis network.
| In every cell, ribosomes translate the genetic instructions carried by messenger RNAs into the proteins they encode. Molecular midwives called chaperones often bind to nascent protein chains as they emerge from the ribosome to help them fold. Very little is known about this process. Do all proteins need chaperone assistance as they exit the ribosome? Do different chaperones recognize different polypeptide chains and, if so, how? Answering these questions has been hard because most studies have examined only a handful of model proteins and their interactions with a specific chaperone. Here, we used a systematic approach to investigate the challenging question of chaperone specificity in living cells. We isolated specific chaperones that interact with nascent proteins during translation along with the ribosomes and associated mRNAs encoding the emerging proteins. We then used DNA microarrays to identify the full suite of mRNAs and thus the encoded proteins that interact cotranslationally with each of these factors. We learned from these studies that individual chaperones interact with a specific set of nascent proteins. Furthermore, overlapping specificity enables one chaperone to modulate the specificity and fidelity of another. The picture that emerges suggests that these molecular midwives are an important part of the intricate systems that maintain specificity, precision, and efficiency in expressing the genome's instructions.
| Ribosomes translate the linear genetic code into polypeptide chains that must fold into a specific three-dimensional structure and often assemble with other polypeptides to be born as functional proteins. During this process, as nascent proteins emerge from the ribosome, they lack information to complete their folding and are susceptible to misfolding and aggregation. A diverse set of molecular chaperones act as midwives to stabilize and facilitate the folding of newly translated polypeptides into functional proteins. Among these, Chaperones Linked to Protein Synthesis (CLIPS) [1] interact physically with ribosomes and associate cotranslationally with nascent polypeptides. In addition to folding within the cytosol, many polypeptides must be directed to various membrane-bound organelles, such as the ER and mitochondria. A number of specific targeting factors recognize nascent polypeptides before they have a chance to fold in the cytosol and deliver them to specific cellular membranes. One of the best understood mechanisms involves the cotranslational recognition of characteristic hydrophobic nascent chain segments by the Signal Recognition Particle (SRP), which facilitates proper delivery of the entire ribosome-nascent chain complex (RNC) to the ER membrane for cotranslational translocation.
The multiplicity of fates and possible interactions available to a polypeptide as it emerges from the ribosome in the eukaryotic cytosol raises a number of intriguing questions. Do all nascent chains interact with chaperones? Is there any specificity in the recognition of nascent chains by chaperones? How do cytosolic chaperones and targeting factors such as SRP discriminate among their respective substrates, and how is the fidelity of this process achieved? These questions are fundamental to understanding the mechanisms governing polypeptide fate as it emerges from the ribosome.
Much of our understanding of nascent chain interactions with chaperones or other targeting factors comes from the study of model proteins, chosen for a convenient enzymatic or structural assay for folding or translocation. As a result, the overall logic and organization of the system that mediates the critical events in delivery and birth of a nascent polypeptide as a functional protein is still a mystery. To begin to fill this gap, we developed a systematic approach to define the principles underlying the specificity of cotranslational chaperones. In the present work, we apply it to study the specificity and interplay of two important nascent-chain interacting factors: SRP and the Nascent Chain Associated Complex (NAC).
Eukaryotic SRP was initially identified as a factor for targeting proteins to the ER. SRP is a ribonucleoprotein complex comprising six proteins (in yeast Srp72, Srp68, Srp54, Sec65, Srp21, Srp14) and a non-coding RNA (scR1) [2]. SRP binds weakly to all ribosomes, even those that do not translate ER-destined proteins, by virtue of its contacts with multiple ribosomal sites. One of the contact sites, the ribosomal protein Rpl25, is also a proposed binding site for NAC [3],[4]. SRP recognizes characteristic hydrophobic sequences such as the N-terminal signal sequence (SS) and transmembrane domains (TM) in nascent polypeptides as they emerge from the ribosome. The dual recognition of ribosome and nascent chain by SRP ensures high affinity for cognate RNCs. SRP-bound RNCs are targeted to the membrane through interaction with the membrane bound SRP receptor (SR), where nascent chains bearing SS or TM domains are translocated across the ER membrane by a protein complex called the Sec61 translocon. Interestingly, the Sec61 translocon itself can also interact directly with ribosomes [5], preferentially recognizing RNCs bearing hydrophobic SS or TM regions; this might provide an SRP independent route to the ER or a proofreading mechanism for ER import. SRP-independent co- and post-translational ER targeting pathways also exist, including post-translational chaperone-assisted translocation [6] and direct ER targeting of mRNA through RNA-binding proteins (reviewed in [7]). The respective contributions of the various targeting pathways to ER import in vivo and the determinants that channel an ER-bound protein through either SRP-dependent or SRP-independent pathways are not entirely understood.
Very little is known about the function of the abundant and ubiquitous NAC complex. NAC is phylogenetically conserved across eukaryotes and archaea but is absent from prokaryotes [8]. Structural characterization of archaeal NAC indicates that its subunits must assemble in tightly folded dimers [9]. Most NAC complexes are heterodimers of two subunits, α and β, but homodimers have also been reported [10]. Yeast contains a single alpha subunit gene, EGD2, and two β subunit genes, EGD1 and BTT1. NAC contacts Rpl25 [3] and Rpl31 [11] in close proximity to the ribosomal exit site and can crosslink to very short nascent chains [12], suggesting an early role in the birth of nascent proteins. NAC deletion causes embryonic lethality in mice, flies, and nematodes [13]–[15] but only minor growth defects in yeast [16]. Despite its abundance and conservation, the specificity and function of NAC are obscure and controversial. NAC does not associate with proteins after release from the ribosome and has no apparent chaperone activity. From in vitro experiments, NAC was initially proposed to be essential for faithful SRP-targeting of proteins to the ER [12] and preventing inappropriate association of RNCs lacking SS or TM with the translocon Sec61 [17]. This hypothesis was not supported, however, by subsequent in vitro and in vivo studies, which did not reveal aberrant translocation phenotypes in NAC-deleted strains [16],[18]. A regulatory role for NAC in mitochondrial protein import, suggested by in vitro experiments [19],[20], was not corroborated by in vivo studies [16]. Given the robustness of protein homeostasis pathways, loss of NAC could be compensated by other systems. Indeed, NAC deletions exacerbate the effect of deleting the yeast Hsp70 homolog SSB, leading to higher levels of ribosomal protein aggregation [21].
A number of experimental challenges have hindered progress towards understanding the robust network of chaperones and cofactors acting cotranslationally on nascent chains. Because nascent chains comprise a small, transient, and heterogeneous cellular pool of chaperone substrates, proteomic analyses are currently impractical. The high degree of redundancy within the cellular chaperone network often masks obvious loss-of-function phenotypes. Our understanding of the specificity and mechanism of cotranslationally acting chaperones comes from in vitro translation experiments using individual model proteins, and thus the generality of such experiments is hard to ascertain. To circumvent these difficulties, we developed a sensitive, systematic method for defining the substrate specificity and interplay of cotranslationally acting chaperones and other nascent chain binding and modifying factors (e.g., acetylation enzymes) in vivo. Here we used this approach to characterize the specificity of the interactions of SRP and NAC with nascent polypeptides and how the interplay between these two factors serves to modulate that specificity.
Cotranslationally acting chaperones recognize substrates as they emerge from ribosomes; the identity of the polypeptide substrate is determined by the mRNA programming its translation. We reasoned that we could leverage the specificity, sensitivity, and comprehensiveness of RNA identification to systematically identify the substrates of factors that associate cotranslationally with nascent polypeptides.
Our basic experimental strategy was to isolate specific chaperone-bound ribosome-nascent chain complexes (RNCs) from cells and identify the polypeptide substrates through their encoding mRNAs (Figure 1A). The isolation exploits Tandem-Affinity-Purification (TAP) tagged chaperones expressed from their endogenous chromosomal locations to ensure their expression at physiological levels. Following isolation of a specific tagged chaperone and associated RNCs complexes, we can identify the mRNAs encoding the polypeptide substrates selectively bound by that chaperone, by DNA microarray hybridization (Figure 1A).
In this study, we applied this approach to define the substrate specificity of two ribosome-associated factors from the yeast S. cerevisiae: SRP and NAC. The multicomponent SRP complex was isolated using SRP54-TAP. To purify NAC we used TAP-tagged variants of each of the NAC subunits: Egd1/β, Egd2/α, and Btt1/β'. A similar strategy relying on C-terminal TAP-tags of two different solvent exposed ribosomal proteins, Rpl16 and Rpl17, was used to purify ribosomes directly. Sucrose gradient fractionation confirmed that the TAP-tagged Srp54, all three NAC subunits, and Rpl16 and Rpl17 all associated with polysomes (Figure S1A and Figure S6B). We initially examined the TAP-purified complexes by SDS-PAGE followed by silver staining (Figure 1Bi). Whereas the untagged purification control (Figure 1Bi, lane 1) revealed only background bands corresponding to the TEV protease preparation, all other lanes showed characteristic associated proteins (Figure 1Bi, lanes 2–7). Shared among all purifications were a set of low molecular weight proteins identified as ribosomal proteins by mass spectrometry (Figure 1Bi; unpublished data). The presence of the 40S ribosomal subunits in the TAP immunopurifications (IPs) was confirmed by RT-PCR detection of the 18S rRNA (Figure 1Bii) while immunoblot analysis for ribosomal protein Rpl3 confirmed the presence of the 60S subunit (Figure 1Biii). Importantly, neither 18S rRNA nor Rpl3 were detected in purifications carried out from untagged control cells (Figure 1Bi, lane 1). These results show that the TAP-tag does not disrupt ribosomal binding of either SRP or NAC and that our isolation procedure efficiently recovers their ribosome-associated complexes. We subsequently employed the TAP-tag isolation approach to systematically identify all mRNAs associated with SRP and the three subunits of NAC, as well as those engaged with translating ribosomes in actively growing cells.
The TAP-tags in ribosomal proteins Rpl16 or Rpl17 were used to purify all translating ribosomes irrespective of their association with chaperones (Figure 1Bi, lanes 2,3) and the associated mRNA was analyzed using DNA microarrays with the total mRNA from the same cells serving as a comparative standard (Figure 1C). The experiments were carried out as three independent biological replicates for each ribosomal protein. As shown in the clustering analysis in Figure 1C, the results of these experiments were highly reproducible (Rpl16, r = 0.96; between Rpl16 and Rpl17, r = 0.93). In principle, the relative occupancy of each mRNA with Rpl16 and Rpl17 provides a measure of that mRNA's association with translating ribosomes. At a stringent 1% false discovery rate (FDR) [22], we identified that 1,673 mRNAs are highly enriched in both Rpl16 and Rpl17 datasets. As expected, a disproportionate number of these mRNAs encode ribosomal proteins (GO “ribosome”, n = 212 genes, p<1×10−56), metabolic enzymes (GO “carboxylic acid metabolic process”, n = 199 genes, p = 1×10−10), and mitochondria (GO “mitochondrion”, n = 450, p = 1×10−4), which correspond to the mRNAs with the highest translation rates in actively growing cells. In contrast, the least enriched mRNAs encoded proteins likely not translated at appreciable rates in mid-log phase, including meiosis and transposition. Similar conclusions were obtained when translation was assessed in the same yeast cells by isolation of actively translated mRNAs from the polysome fractions of sucrose gradients (Figure S1); our results are also consistent with previous findings [23],[24].
To identify the cellular substrates of SRP in vivo, we used immunoaffinity isolation of Srp54-TAP along with its cotranslational associated RNC-mRNAs complexes to isolate mRNAs encoding nascent proteins specifically recognized by SRP (Figure 1B, lane 7; Figure 2A). Using DNA microarrays we identified approximately 924 mRNAs reproducibly enriched at a stringent statistical threshold in Srp54 IPs (Figure 2A, note high reproducibility of three independent Srp54 replicates). Disrupting the translating 80S ribosomes with EDTA, which releases the translated mRNAs, prevented the recovery of mRNAs but not the SRP RNA scR1 in the SRP isolations (unpublished data). This indicates that the association of mRNAs with SRP was mediated through translating ribosomes, supporting our premise that analysis of the mRNAs associated with RNC-SRP complexes provides information on the specificity of SRP interaction with the translating polypeptides.
Hierarchical clustering based on quantitative enrichment of mRNAs in association with Rpl16/17 and Srp54 respectively indicated clear selectivity of SRP-associated complexes for a distinct subset of translated mRNAs (Figure 2B), consistent with the SRP specificity for a distinct subset of nascent polypeptides (Figure 2B, yellow highlight). Secretory pathway proteins (Figure 2C, yellow bars) were disproportionately represented among SRP-associated mRNAs, whereas the mRNAs encoding cytosolic and mitochondrial proteins (Figure 2C, cyan bars) were significantly under-represented among SRP-associated mRNAs. The consistency of these results with the known function of SRP suggests that this procedure can selectively identify the mRNAs encoding nascent polypeptides that are in vivo substrates of specific cotranslational chaperones.
The recognition code for SRP derived from in vitro studies provided a basis for several algorithms that predict SS and TM domains from sequence information; these are used to identify putative secretory pathway proteins (reviewed in [25]). The systematic identification of SRP substrates provides an unprecedented opportunity to benchmark these predictive algorithms against the experimentally determined SRP substrates from yeast. We used published algorithms (SignalIP, RPSP, TMHMM, HMMT, and the curated Uniprot database) to identify putative SS or TM regions encoded by mRNAs that were associated with SRP with high confidence (1% FDR) (Figure 2D and Figure S2A) as well as in the mRNAs least enriched in our SRP IPs (herein the “non-SRP interactors”) (Figure S2Bi and S2Bii). All these programs predicted, with good agreement, TM domains in ∼60%–75% of the SRP interactors (Figures 2D and Figure S2Aii, hairline denotes consensus among programs) and an SS in 15%–35% of the SRP interactors (Figure S2Aii). Notably, however, the algorithms found no SS or TM domains in about a quarter of the proteins encoded by SRP-associated mRNAs (Figure 2D, 102 targets when using SignalIP and TMHMM). These could represent bona fide SRP substrates that are recognized by novel, yet-to-be-determined features. Indeed, 12% of these proteins are annotated as membrane or secretory pathways (Figure 2E; Table 1). For instance, Sed4, a known integral ER membrane protein, and Sec20, a v-SNARE membrane glycoprotein involved in Golgi to the ER retrograde transport, are both encoded by mRNAs that we found to be enriched in association with SRP, though both lack predicted TM or SS regions. Despite the overall consistency of our results, some of the apparent interactions might be stochastic or spurious: For instance, 48% of proteins encoded by SRP-associated mRNAs that lack predicted TM or SS domains localize to the nucleus or mitochondrion (Table 1; Figure 2E). Future studies on the mechanistic and physiological significance of these potential non-canonical SRP interactors may reveal novel aspects of SRP function.
As expected, applying the same algorithms to the proteins encoded by mRNAs not associated with SRP yielded few proteins with predicted SS or TM regions (Figure S2B; “non-SRP interactome”). Approximately 6% of these proteins had a predicted SS and ∼6% had a predicted TM domain (Figure S2B, note slight variations among algorithms). Interestingly, some of these proteins are annotated as localizing to the plasma membrane (Fus1p) or ER (Ost4p) and might therefore represent weakly SRP-bound or SRP-independent secretory proteins. Others correspond to mitochondrial proteins, which are generally not recognized by SRP; although dual targeting of some polypeptides to the mitochondria and the ER has been reported [26]. Still others, such as ribosomal protein Rpl45, contain a predicted SS yet are clearly cytoplasmic proteins.
Messenger RNAs encoding proteins with predicted TM regions were generally more highly enriched by our SRP affinity isolation procedure than proteins with predicted SS (Figure 2F), suggesting that the interaction of the correspondent nascent polypeptide with SRP was stronger or more sustained. Since TM regions are generally more hydrophobic than SS, this is consistent with previous biochemical experiments indicating that proteins with more hydrophobic sequences have a higher dependency on SRP for efficient ER translocation [27]. SRP-binding substrates lacking predicted SS or TM domains were typically less enriched than those containing either predicted TM or SS domains, suggesting that their SRP-binding sequences may be weaker and thus not recognized by algorithms designed to find sequences that bind strongly to SRP. While the hydrophobicity of the SS or TM regions is clearly important for SRP interaction, we only found a very weak correlation between this parameter alone and SRP enrichment (Figure S3 and unpublished data).
The presence of canonical SRP-binding, ER-targeting sequences in proteins that did not appear cotranslationally associated with SRP and the apparent enrichment of nascent proteins with no SS or TM regions in association with SRP suggest that our understanding of SRP specificity in vivo is still incomplete and that SRP-binding might be influenced by additional cis- and transacting factors.
A number of important questions surround the mechanisms and functions of mRNA association with membranes (reviewed in [28]). SRP inactivation is not lethal to yeast [29]–[31], indicating that SRP is not the only route to membrane association. mRNA binding proteins known to localize to cellular membranes could provide additional mechanisms for targeting selected mRNAs to the ER [32]. Experimental evidence that many mRNAs encoding cytosolic proteins associate with membranes has led to a suggestion that a substantial fraction of all translation in eukaryotic cells occurs in association with membranes [33].
To examine the contribution of the SRP-mediated route to overall mRNA targeting to membranes, we empirically defined the global complement of mRNAs associated with yeast membranes. We used a previously established differential centrifugation procedure [34],[35] to isolate membrane-associated mRNAs as well as the cytosolic, membrane-free mRNAs (Figure 3A). At a stringent statistical threshold (1% FDR), we identified 1,168 membrane-associated mRNAs (∼45% of the translatome, Figure 3B). Hierarchical clustering of SRP-bound and membrane-associated mRNAs demonstrated extensive overlap, as expected (Figure 3B; r = 0.6). A large fraction of membrane-bound mRNAs encoded proteins localized to ER, Golgi, or plasma membrane (Figure 3C i versus ii; red, pink, and orange, respectively), consistent with previous findings [35]. SRP-associated (Figure 3Ci) and membrane-associated (Figure 3Cii) fractions showed comparable enrichment for mRNAs encoding ER, Golgi, and Plasma membrane proteins. For instance, 60% of all mRNAs annotated as corresponding to ER proteins were enriched in the SRP-associated dataset (log2 ratio greater than 0) (Figure 3Ci, red line) and 70% were enriched in the membrane-bound dataset (log2 ratio greater than 0) (Figure 3Cii, red line). In contrast more than 90% of mRNAs encoding cytosolic proteins were included in neither the SRP-associated nor membrane-associated fractions (Figure 3C green). This result suggests that cellular membranes are not the major site of cytosolic protein synthesis, at least in yeast.
The enrichment for mRNAs encoding mitochondrial proteins was clearly higher in the membrane-associated than in the SRP-associated fractions (Figure 3C ii, blue line; Figure S4A and Table 2). This likely reflects the presence of mitochondria in our membrane preparation and supports the idea that a fraction of mitochondrial proteins are imported cotranslationally into mitochondria (reviewed in [36]).
Joint analysis of the quantitative enrichment of each mRNA in association with SRP and membrane respectively gave further insight into modes of mRNA localization (Figure 3D–F and Figure S4). Comparison of both SRP and membrane-associated RNCs (significantly enriched targets at 1% FDR) reveals that most mRNAs that were both SRP-associated and membrane-associated (SRP+/Mem+) encoded proteins annotated as belonging to the secretory pathway (Figure 3D,E for ER; Figure S4B–D for Plasma membrane and Golgi). Interestingly, 24% of the SRP-associated RNCs in which the nascent polypeptide lacks either predicted SS or TM regions were also membrane-associated (Table S1); thus, these nascent chains are likely bona fide SRP targets despite their lack of a canonical SRP binding site. Virtually no transcripts encoding cytosolic proteins (Figure 3D, green) and few encoding mitochondrial proteins (Figure S3A; Figure S4C) were SRP+/Mem+. As expected, these mRNAs were overwhelmingly SRP−/Mem−. Notably, a number of mRNAs encoding secretory pathway proteins also fell into this class. We reasoned that these might represent proteins imported into the ER post-translationally. Indeed, known substrates of post-translational translocation pathways were SRP−/Mem− (Figure 3E). These include tail-anchored proteins (Figure 3E, TA, highlighted in black), which use the post-translational GET pathway [37] and pre-pro-alpha-factor (Mfa1, Figure 3E), which uses cytosolic chaperones to reach the ER membrane [38],[39]. Further analysis of this dataset may reveal additional substrates of these pathways.
Of particular interest were secretory pathway proteins whose mRNAs were membrane-associated but not SRP-associated (SRP−/Mem+; Figure 3F and Table 2), such as the chaperone EPS1, the plasma membrane protein IST2, and Golgi protease KEX1. These may represent translocation substrates whose mRNAs are targeted to membranes in an SRP-independent mechanism. Interestingly, IST2 mRNA is known to localize to the bud tip by an actomyosin-driven process and is associated with cortical ER via an SRP-independent mechanism [40]. One possible mechanism for this process could be direct localization through specific membrane-associated RNA-binding proteins (RBPs) [32],[41]. However, we could not detect significant overlap between the SRP−/Mem+ mRNAs and the mRNA targets of previously described membrane-associated RBPs (unpublished data). Thus, novel yet-to-be-determined pathways and factors may function to localize these mRNAs to membranes.
To gain insight into the cotranslational specificity of NAC, we systematically identified mRNAs cotranslationally associated with NAC complexes, using DNA microarrays to profile the mRNAs associated with each of the three NAC subunits, Egd2, Egd1, and Btt1 (Figure 4A). Importantly, dissociation of the ribosome-mRNA-nascent chain complexes with EDTA abrogated the association of NAC with mRNAs (unpublished data), suggesting that the mRNAs identified by this assay in association with individual NAC subunits reflect the cotranslational specificity of NAC for the nascent polypeptide.
Each of the TAP-tagged NAC subunits was ribosome associated (Figure 4B, lower panel; Figure 1B). The extent of α/β heterodimer formation for each β subunit was assessed by immunoblot analysis of α/Egd2 enriched by immunoaffinity purification of each of the two β isoforms (Figure 4B, middle panel, lanes 2 and 3). As expected, the EGD1-encoded β subunit was strongly associated with the α subunit, Egd2, consistent with previous reports [16],[42]. On the other hand, little of the Egd2/α subunit copurified with the BTT1 encoded β' subunit (Figure 4B, compare lanes 2 and 3). This is consistent with evidence that Btt1 elutes predominantly at a homodimer molecular weight during size exclusion chromatography of yeast cell extracts [16].
Hierarchical clustering of the mRNAs based on their patterns of enrichment in association with each NAC subunit reveals two striking properties of NAC: First, there were clear differences between NAC subunits, suggesting that the different NAC subunits recognize different subsets of mRNA-RNC complexes. Second, NAC targets include virtually every mRNA associated with Rpl16/17, suggesting that at least one NAC isoform can interact with virtually every nascent polypeptide in the cell. This result is consistent with the estimated stoichiometry of NAC to ribosomes (1.25∶1) together with evidence that most of NAC in the cell are ribosome-bound [43]. Importantly, no mRNA was recovered by NAC complexes from non-ribosome-associated fractions (unpublished data), suggesting that the mRNA association and specificity are mediated through translating ribosomes (unpublished data). Similarly, omission of cycloheximide during cell extract preparation and analysis, which leads to polysome dissociation, dramatically reduced the number of mRNAs associated with Egd2 (Figure S5). Because association of Egd2 with mRNAs is critically dependent on the presence of intact polysomes, we conclude that Egd2 does not interact with mRNAs directly, but rather, through its association with translating nascent chains.
What determines the substrate specificity of different NAC subunits? The nascent proteins associated with different NAC subunits exhibited significant differences in a number of physicochemical properties, most notably length, hydrophobicity and intrinsic disorder, as well as inferred translation rate (Figure 4D–G). Btt1 associates with mRNAs encoding proteins of higher intrinsic disorder and lower hydrophobicity, whereas Egd2 associated with mRNAs encoding proteins with low intrinsic disorder and high hydrophobicity (Figure 4D,E). The length distribution of predicted protein products, which correlates inversely with the overall rate of folding (Figure 4F) [44], as well as translation rate of the mRNAs (Figure 4G) were also significantly different among sets of mRNA respectively associated with different NAC subunits. These differences suggest that each NAC subunit participates in recognizing specific features of the nascent polypeptide; Egd2 may have higher affinity for longer, more slowly folding polypeptides exposing hydrophobic determinants, whereas Btt1 may preferentially recognize more polar, disordered chains.
There were also differences in the distribution of functional roles of nascent chains associated with the different NAC (Figure 5A) subunits. Egd1 and Egd2 targets were enriched for mRNAs encoding metabolic enzymes, whereas the targets of Btt1 were enriched in mitochondrial and ribosomal proteins (Figure 5A). RNCs translating membrane and secretory pathway proteins were also associated with NAC α/Egd2. Preferential NAC association with nascent proteins sharing specific physicochemical properties may have resulted indirectly in the enrichment for specific functional categories. For instance, the preferential interaction with nascent ribosomal proteins with Btt1 may reflect its preference for short, highly disordered polypeptide chains with high translation rates. Alternatively, some features differentially associated with both overall physicochemical properties and functional roles of the translated proteins may underlie the observed differential specificity of NAC subunits.
Our GO analysis also revealed overlaps in specificity among pairs of subunits, most notably for Egd1 and Egd2. mRNA association patterns of these two subunits were similar to each other (r = 0.74 average of three replicates for every subunit) and more distinct from the alternative β subunit, Btt1 (Figure 4C). This is consistent with, and likely reflects, the predominance of the Egd1/Egd2 heterodimer in vivo [16]. NAC subunits appear to exist as a combination of homo- and hetero-dimers in the cell [10],[42], and each complex may have a different set of specificities. To explore this possibility, we extracted those substrate sets shared by a α/β pair: that is, likely Egd1/Egd2 or Btt1/Egd2 substrates, and those associating solely with individual NAC subunits, that is, likely substrates of a NAC homodimer. We thus examined whether specific functional themes were significantly enriched in each category (Figure 5B). Few nascent polypeptides associated with Egd1 alone, suggesting that Egd1 primarily functions in a complex with Egd2. Egd1/Egd2 preferentially associated with RNCs translating proteins that function in carbohydrate metabolism, while the Btt1 and Btt1/Egd2 preferentially associated with RNCs translating mitochondrial and ribosomal proteins (Figure 5B). Some protein classes, including redox and nucleotide metabolism, interacted with all NAC subunits, whereas Egd2 only and to a lesser extent Egd1/Egd2 also associated with RNCs translating secretory proteins; notably, this subset of nascent polypeptides also associated with SRP.
We next examined how this analysis reflected on the physicochemical properties of substrates (Figure 5C–F). Incorporating into our analysis the idea that NAC exists as heterodimers and homodimers exacerbated the differences in the intrinsic properties observed for each subunit set. The binding specificities of Egd2/Egd1 and Egd2/Btt1 appeared to reflect the combined specificity of the subunits in the dimer (Figure 5C–F; green, Egd1/Egd2; orange, Egd2/Btt1; blue, Egd2/Egd2; purple, Btt1/Btt1). In contrast, the nascent polypeptides associated exclusively with Btt1 (Figure 5C,D, purple) comprised proteins with the highest intrinsic disorder and lowest hydrophobicity, whereas the RNCs associated with Egd2 translated the most hydrophobic proteins (Figure 5C, D blue). Importantly, the fact that the interaction specificity of each subunit correlated so strongly with the predicted physical properties of the translated polypeptide is strong evidence that each NAC subunit recognizes determinants in the nascent chain itself. Furthermore, both components of each NAC dimer appear to contribute to nascent chain recognition, expanding both the specificity and number of RNCs recognized by NAC. Although different NAC homo- or heterodimers differentially associated with ribosomes translating different sets of mRNAs, the specificity of the ensemble of NAC complexes appears to encompass virtually every translated polypeptide.
The role of NAC in SRP specificity and substrate selection has been a matter of debate [17],[45],[46]. NAC and SRP both contact the ribosomal protein Rpl25 [3]. NAC was originally proposed to compete with SRP for ribosome binding [17]. However, our global analysis revealed that many nascent secretory pathway proteins can interact with both SRP and the NAC subunits Egd2 and Egd1. We tested whether the specificity overlap might reflect joint binding at the ribosome. We isolated SRP-containing RNCs and tested for the presence of NAC (Figure S6A). Indeed, immunopurification via either Srp54p, Srp68p, or Srp72p revealed the presence of Egd2p in SRP-associated polysomes (Figure S6A), suggesting that NAC and SRP might bind simultaneously to the same RNCs, though this experiment does not exclude that these factors might bind to different ribosomes engaged in translation of the same mRNA. However, NAC does not detectably affect the extent of SRP association with ribosomes, as shown by the similarity of SRP cofractionation with polysomes in WT and Δegd1Δegd2 cells (, see also below Figure 6B).
To further explore the functional interplay between SRP and NAC, we examined whether the absence of NAC affects SRP recognition of nascent proteins, as reflected by its pattern of association with mRNA (Figure 6). To this end, we compared the ribosome-nascent polypeptide interaction specificity of SRP in wild type cells (herein NAC+) with that in cells lacking NAC subunits Egd1 and Egd2 (Δegd1Δegd2, herein ΔNAC) (Figure 6A). Clustering analysis highlights the striking differences between the patterns of SRP-bound mRNAs in NAC+ and ΔNAC cells (Figure 6A; Figure S6A). SRP association with a core set of RNCs encoding secretory proteins was relatively independent of NAC, as these were enriched by SRP affinity isolation in both NAC+ and ΔNAC cells (Figures 6A and S7A, purple). In contrast, the SRP-association with another set of secretory proteins appeared to be NAC-dependent, as it was lost in ΔNAC cells (Figures 6A and S7A, blue). Two SRP-dependent proteins, DPAPp and Kar2p [27], fell into this category (Figure S7A). Strikingly, a third set of mRNA-RNC complexes only associated with SRP in ΔNAC cells; most of these mRNAs encoded cytosolic proteins (Figures 6A and S7A, green; Off-target interactors). The relative depletion from the SRP-associated RNCs of transcripts encoding “bona fide” SRP substrates, that is, secretory pathway proteins and the corresponding increase in mRNAs encoding cytosolic proteins from the SRP-associated RNCs, was also reflected in the GO analysis of the proteins interacting cotranslationally with SRP (Figure S7B). Using a 1% FDR to analyze the SRP interactomes, we find that 70% of SRP-associated RNCs in NAC+ cells contained nascent polypeptides with predicted SS or TM regions (Figure 6B) whereas in ΔNAC cells, only 40% of the SRP-associated mRNAs encoded proteins with SS or TM domains. Of note, the depletion in “bona fide” SRP substrates in ΔNAC cells was independent of the statistical stringency of the analyses (Figure S8).
Interestingly, deletion of the second NAC isoform (Δegd2/Δbtt1) (ΔNAC') had a similar effect on SRP specificity (Figure S8). As observed for ΔNAC cells, the SRP interactome in ΔNAC' cells was also depleted in proteins containing predicted SS or TM regions (Figure S8B, Figure S8C) and mRNAs encoding secretory proteins (Figure S8D). In these cells, SRP also associated with more cytosolic and mitochondrial proteins (Off target, Figure S8A, Figure S8D). Thus, NAC significantly influences the in vivo specificity of SRP interactions.
How does NAC affect SRP specificity? The extent of SRP interaction with ribosomes and the salt-sensitivity of this interaction were not affected by the absence of NAC (Figure 6C). The fact that NAC appears to enhance the association of SRP with some nascent polypeptides (i.e., NAC-dependent) and prevent SRP interactions with others, leaving still others unaffected, suggests a complex mode of regulation. We first hypothesized that less hydrophobic SRP-binding nascent polypeptide sequences might be more easily displaced in the absence of NAC. Our analysis did not support this hypothesis; NAC-dependent and NAC-independent SRP interacting proteins were indistinguishable based on the length and hydrophobicity of their predicted SS or TM domains (Figures 6D, S7C and unpublished data). mRNA abundance and translation rate provided the strongest identifiable differences between NAC-dependent and NAC-independent SRP interactions (Figure 6E, note log scale in 6Ei). NAC-independent SRP substrates were relatively highly translated, abundant proteins; NAC-dependent SRP substrates tended to be much less abundant membrane and secreted proteins (Figure 6E). Because abundance and translation rate appeared key to the NAC-modulation of SRP specificity, we compared the relative enrichment of each class of SRP-associated mRNAs in the presence or absence of NAC (Figure 6F). The abundant, NAC-independent SRP substrates were the most highly enriched SRP interactors even in wild type cells; their level of enrichment was only modestly affected by the absence of NAC. In contrast, the NAC-dependent SRP substrates were less highly enriched in association with SRP, even in wild type cells. Their interaction with SRP was completely undetectable in the absence of NAC. The nascent cytosolic nascent proteins whose latent ability to interact with SRP was apparently blocked by NAC were generally highly abundant cytosolic proteins with high translation rates (Figure 6E, Off-target). These cytosolic proteins do not bind appreciably to SRP in wild type cells, but displayed an enrichment level comparable to bona fide SRP substrates in the absence of NAC, despite their lack of canonical SRP binding sequences.
It is known that SRP can recognize hydrophobic sequences with broad specificity [47]. One of the most striking aspects of SRP function observed here is that, in vivo, in wild type cells, SRP displays exquisite specificity for its cognate substrates independent of their concentration in the cell. In the absence of NAC, however, SRP also interacts with very abundant nascent polypeptides that lack high affinity SS or TM binding sites. These may compete for SRP, effectively lowering its availability to sample less abundant cognate sequences. NAC thus effectively acts as both a positive and negative regulator of SRP interactions with potential binding targets, tuning out the noise and enhancing the specific interactions with low abundance cognate substrates (Figure 6G).
Deletion of NAC has few phenotypic consequences for the cell (unpublished data) [16],[21], while loss of SRP function severely compromises growth. We reasoned that if SRP were to bind inappropriately to nascent cytosolic proteins in the absence of NAC and directs the paused ribosome to the ER, the associated mRNA should inappropriately localize at the ER membrane. We thus investigated whether loss of NAC would also affect global mRNA targeting to the ER (Figure 7A). Strikingly, the distribution of mRNAs between the membrane-associated and soluble fractions was indistinguishable between NAC+ and ΔNAC yeast cells (Figure 7B; r = 0.96). For instance, cytosolic proteins that inappropriately interacted with SRP in ΔNAC cells (Figure 7C, “Off-target”, green) were nevertheless largely absent from the membrane fraction in both wild type and ΔNAC cells (Figure 7C). Conversely, the “NAC-dependent” nascent polypeptides whose association with SRP was impaired in ΔNAC cells still efficiently associated with membranes in ΔNAC cells, despite their diminished association with SRP (Figure 7D, blue, compare with Figure S7B). We conclude that despite the loss in SRP specificity under these conditions, loss of NAC activity has little or no effect on the fidelity of mRNA targeting to membranes, despite the loss in SRP specificity.
We chose two NAC-dependent SRP substrates, Kar2 and DPAP, whose association with SRP but not with membranes was impaired in ΔNAC cells, for further biochemical analysis (Figure 7E). The efficiency of their ER translocation in wild type or ΔNAC cells was evaluated by determining the ratio of processed versus unprocessed protein following a short pulse with 35S-methionine. Defective translocation results in accumulation of precursors of these two proteins, that is, uncleaved Kar2 (pre-Kar2) and non-glycosylated DPAP. The presence or absence of NAC did not affect the speed or efficiency of translocation for either Kar2 or DPAP (Figure 7E). In contrast, impairing SRP function using the temperature-sensitive mutation sec65-2 did reduce translocation of both proteins. Thus, the apparent decrease in SRP binding of these SRP substrates in the absence of NAC did not appreciably impair their translocation to the ER. These experiments highlight the robustness and fidelity of membrane targeting pathways.
To better understand how the cell compensates for the loss of NAC function, we examined the transcriptional response to joint deletion of either Egd1 and Egd2 or Egd2 and Btt1 (Figure 7F). A different complement of genes was induced in response to these two perturbations, but major features of the responses were shared. Transcripts encoding ribosomal proteins, ribosome biogenesis and mitochondrial biogenesis machines, and chaperones and stress response genes were induced in response to both defects (Figure 7F and Tables 3 and 4). Notably, loss of NAC activity did not lead to transcriptional induction of an unfolded protein response (UPR), consistent with the lack of a translocation defect in these cells (Figure 7F and unpublished data). The chaperones induced in response to NAC deletion included stress-inducible chaperones like SSA3 and small HSPs, as well as CLIPS, most notably SSB1/2. The synthetic genetic interaction between SSB1/2 and the NAC complex suggests that induction of SSB1/2 may contribute to functional compensation for the loss of NAC [21]. The induction of ribosomal proteins and ribosome biogenesis genes is in accord with the observation that NAC has a role in ribosome biogenesis ([21]; Figures 4 and 5). Loss of Egd2/Btt1 led to induction of numerous mitochondrial biogenesis factors, including AFG3, SED1, and MIA4, suggesting a role for this NAC complex in mitochondrial biogenesis.
Affinity isolation of cotranslationally acting chaperones from cells under conditions that preserve their interaction with the nascent polypeptide and associated ribosomes and quantitative profiling of the associated mRNAs open a window on the specificity and interplay of chaperones and targeting factors responsible for cotranslational protein homeostasis. This approach should enable us to probe the structure of the CLIPS network and the interplay between different chaperones and targeting systems. Unlike previous studies defining chaperone interactors by proteomic analysis, our approach focuses on cotranslational interactors as potential chaperone substrates. The approach presented here opens a window to understand the pathways and principles of cotranslational chaperone action.
The full complement of nascent chains that interact with SRP in vivo has never been defined. Studies of SRP recognition using model substrates and peptides have shown that SRP recognizes highly hydrophobic signal sequences and transmembrane regions [27] but can also recognize hydrophobic stretches found in cytoplasmic proteins [39],[48]. We find that in vivo SRP displays considerable specificity for previously defined recognition sequences; approximately 80% of the in vivo substrates we identified contained a predicted SS or TM domain (Figure 2D). Our analysis also indicates that additional factors contribute to SRP specificity and affinity in vivo: approximately 20% of SRP interactors lack a discernible SS or TM domain. Since several SRP-associated mRNAs encode secreted or membrane-associated proteins that appear to lack canonical SS or TM domains, these interactions may be functionally relevant. Conversely, a number of proteins with clear SS or TM domains were translated in these cells but were not enriched in association with SRP, suggesting that SRP recognition might be regulated by features or mechanisms beyond its intrinsic affinity for SS or TM regions. Interestingly, it has been reported that, in bacteria, basic residues promote binding of SRP to a subset of signal peptides whose hydrophobicity falls slightly below a critical level [49]. Recent studies also suggest that a hydrophobic stretch can recruit SRP to the ribosome before it emerges from the ribosomal exit tunnel, presumably by changing the conformation of the ribosome [50],[51]. SRP-substrates without canonical SS or TMs may contain sequence elements that similarly enhance binding to SRP by this indirect mechanism or even by recruiting bridging factors.
The SRP-interactome also yielded some surprising observations. Although the glucose metabolism enzyme glyceraldehyde-3-phosphate dehydrogenase (GAPDH, Tdh1-3 in yeast) is reportedly mostly cytoplasmic [52], the mRNAs encoding all Tdh isoforms (i.e., Tdh1-3) were both SRP-associated and enriched in the membrane fraction (see Table 2). Notably, all three isoforms of this enzyme in yeast have predicted signal sequences at the N-terminus and have been detected on the outer surface of the cell wall [53]. Our findings suggest a mechanism by which Tdh can reach the outer cell wall and may also explain the observation that other primarily cytosolic proteins, including glycolytic enzymes, are secreted by yeast spheroplasts and found as integral components of the cell wall [54],. Interestingly, GAPDH mRNA has also been found to associate with membranes in mammalian cells [56].
The variable efficiency of the SRP-export pathways for different mRNAs has been recognized both as a regulatory mechanism and as a source of misfolded proteins [57]. Even a small fraction of untranslocated precursors may represent a substantial burden for the cytosolic protein quality control machinery. Our analyses show quantitative variation in binding of SRP to its targets, which may be related to previous observations that some secretory and membrane proteins are more efficiently translocated than others [58],[59]. Differential translocation efficiency is proposed to underlie disease pathologies, such as prion formation, and to regulate protein flux into the secretory pathway [57],[60]. We find that mRNAs encoding proteins with predicted TM regions are more enriched in association with SRP than either polypeptides with SS domains or those with no detectable SS or TM domains (Figure 2F). Notably, in bacteria, SRP is a main targeting factor for membrane proteins, while secretory proteins follow a different, SecB-dependent, pathway (reviewed in [61]). Although differences among signal sequences have been shown to modulate translocation in yeast for a small number of substrates (reviewed in [57],[62]), we could not identify a clear correlation between SRP enrichment and any defined physicochemical property in the SS or TM domains themselves, such as length, overall or maximal hydrophobicity, or amino acid composition. In bacteria, the codon adaptation index of the SS and the efficiency in translation initiation have been proposed to influence targeting efficiency (reviewed in [63]); we found no significant correlation between SRP enrichment and these parameters (unpublished data). Our data indicate that, in vivo, the features that control SRP recognition of a given nascent polypeptide are more complex than expected. In principle, additional features promoting a pre-recruitment of SRP to translating ribosomes could increase the efficiency of SRP-dependent translocation and enhance physiological robustness. For example, the translation properties of a given mRNA might influence the efficiency with which a potential SS or TM domain or other hydrophobic stretches are recognized. Additional ribosome-associated factors could also modulate the SRP association with nascent chains, as shown here for NAC. What determines the variable efficiency we observe in SRP association with SS and TM containing nascent polypeptides remains an important unanswered question.
The comparative analysis of SRP- and membrane-bound mRNAs provides an overview of overall partitioning of co- and posttranslational events in membrane targeting. Our data indicate that most cytosolic mRNAs are not membrane-associated, suggesting the existence of mechanisms that separate cytosol-bound from membrane-bound mRNAs. SRP appears to be involved in cotranslational targeting of most membrane and secretory proteins to the ER; ∼80% of membrane-associated mRNAs encoding these proteins were also SRP-associated, at a 1% FDR threshold. However, we also found evidence for SRP-independent translocation pathways. A significant minority of mRNAs, roughly 20%, appears to associate with membranes through SRP-independent pathways. We estimate that 25% of the secretory pathway proteins that do not associate cotranslationally with membranes are translocated posttranslationally to the ER; these include tail-anchored proteins, which use the GET pathway [37],[64], and Mfa1, whose translocation is assisted by cytosolic chaperones including Hsp70 and TRiC/CCT [38],[39].
SRP is not essential in yeast [65] and downregulation of SRP in mammalian cells has a mild effect on growth [33],[66], indicating the existence of SRP-independent mechanisms for translocation [27]. The membrane-associated mRNAs that encode membrane or secreted proteins but do not bind SRP (SRP−/Mem+) are candidates for substrates of SRP-independent cotranslational translocation pathways. Little is known about these pathways. They may involve direct recruitment of RNCs to the Sec61 translocon [67], as well as additional factors [31],[68]. Direct, translation-independent targeting of mRNAs to membranes could also involve RNA-binding proteins (RBPs) such as Puf1, Puf2, Pub1, Scp160, Ypl184c, Khd1 [41], and Whi3 [69], all of which bind distinct sets of mRNAs encoding membrane or secreted proteins. Interestingly, while few of these RBPs bind mRNAs in the SRP−/Mem+ set, there is also considerable overlap between their targets and the mRNAs we found enriched in association with SRP (unpublished data), suggesting these RBPs may provide redundancy or another level of control to cotranslational SRP targeting to membranes. Our experiments will provide an opportunity to refine our understanding of the signals and features that direct secretory proteins along these alternative non-SRP pathways.
The functions and localization patterns of mRNAs that were membrane-associated but not SRP-bound suggest several additional roles for SRP-independent membrane sorting of mRNAs. Many of these (150 out of 541) encoded mitochondrial protein precursors, which may be imported cotranslationally into mitochondria [70]. Among the remaining non-mitochondrial mRNAs, there was a paucity of mRNAs encoding ER-localized proteins (Lro1) and proteins with transmembrane domains, but the set included many mRNAs encoding proteins involved in other membrane systems in the cell. Two other She2 targets, Mtl1 and Lsb1, were included in this set and may also be associated with the cortical ER during trafficking to the bud [71]. Most strikingly, there were a number of mRNAs encoding proteins involved in endocytosis and actin patch assembly (Vps35, Aly2, Swh1, Lsb6, and Pan1), clathrin-mediated vesicle transport (Apl4, Apl3, Laa1, and Sec16), bud formation (Sbe2, Ypk1, Lrg1, Prm10, and Bem3), and vacuole function and assembly (Sch9, Vps13, Vac8, Tre2, Tre1, Fab1, and YIR014W). Potentially, these mRNAs are localized to specific membrane compartments to preferentially translate the proteins near their site of action. Alternatively, the nascent polypeptides could associate cotranslationally with membrane-associated interacting partners.
The set of membrane-associated mRNAs included an abundance of regulatory factors, including transcription factors (Stp2, Ino2, and Ppr1), RBPs (Puf2 and Puf3), and signaling molecules (Tor1, Bem3, Fab1, and Sch9). Localization of these mRNAs to appropriate membrane structures may facilitate co-translational association of their products with localized signaling partners or enable locally controlled regulation of their translation by signaling systems linked to these membranes. For instance, Stp2 promotes expression of permease genes and is synthesized as an inactive precursor that associates with the plasma membrane and is cleaved upon sensing of external amino acids [72]; Tor1, a component of the TOR complex, is a peripheral membrane protein that regulates cell growth in response to nutrient availability and stress [73], and Bem3 is a Rho GTPase activating protein specific to Cdc42, which controls establishment and maintenance of cell polarity, including bud-site assembly [74].
The abundant, ubiquitous, and evolutionarily conserved Nascent Chain Associated Complex (NAC) binds ribosomes in close proximity to the nascent chain exit site [75]. Despite its conservation, little is known about its function. Our analysis of the association of the three NAC subunits with nascent polypeptides revealed a surprising and unanticipated division of labor. Considered as a group, the three NAC subunits have translation-dependent interactions with almost every mRNA. Each subunit, however, exhibits distinct specificity for RNCs engaged in translation of mRNAs with different functional themes.
Based on the crystal structure of the archaeal NAC domain, NAC complexes are obligate dimers, where two subunits must complete the folded beta-sheet NAC-domain. Our analysis supports the idea that NAC subunits can function as either homodimers or heterodimers [10],[16]. We found a large overlap between the sets of transcripts associated with Egd1 and Egd2, consistent with the idea that the Egd1/Egd2 complex is the most abundant form. This dimer associated preferentially with nascent metabolic enzymes, including those in carbohydrate metabolism, such as glycolysis. Egd2, either as a homodimer or in a complex with Egd1, was also cotranslationally associated with a large fraction of mRNAs encoding membrane or secreted proteins, many of which also associate with SRP. Btt1, either as a homodimer or in a seemingly minor Btt1/Egd2 complex, associated primarily with RNCs translating ribosomal proteins and nuclear-encoded mitochondrial proteins.
In yeast, the three NAC subunits can be deleted with minimal impact on growth. Deletion of all three NAC subunits leads to enhanced ribosomal protein aggregation in cells also lacking the Hsp70 homologs Ssb1 and Ssb2 [21]. This would suggest that the putative function of NAC is masked by the redundancy of the CLIPS protein homeostasis network. Our analysis of the transcriptional response to NAC deletion (Figure 7F–G) provides insight into how the cellular circuitry compensates for the loss of NAC: A set of chaperones including both stress-inducible chaperones (e.g., Ssa2/4, Hsp42, and Hsp104) and CLIPS (e.g., Ssb1 and Ssb2), as well as several ribosomal proteins and assembly factors (Tables 3 and 4), were induced. This multifaceted response suggests that loss of NAC impairs protein folding and ribosome assembly. NAC has been proposed to have a role in mitochondrial targeting, as shown by a synthetic growth defect by deletion of cells lacking both Egd2 and the mitochondrial targeting factor Mft1 [19]. Our analysis revealed that mRNAs encoding mitochondrial proteins are enriched in association with both Btt1 and Egd2 (Figure 5A,B). Moreover we found that several proteins involved in mitochondrial assembly were induced in cells lacking NAC. Thus, a possible auxiliary role for NAC in cotranslational targeting polypeptides to the mitochondria deserves further investigation.
Is NAC a chaperone? Purified NAC does not prevent protein aggregation and NAC cannot bind directly to nascent chains unless they are ribosome associated [12],[39]. While this is unexpected for a traditional chaperone, NAC may be akin to trigger factor in bacteria, which interacts primarily with nascent chains in the context of the ribosome [76]. The distinct physicochemical properties of the nascent polypeptides associated with different NAC subunits may reflect the direct binding specificity of each individual subunit. A more detailed understanding of NAC substrate specificity must await better structural and biochemical understanding of this complex. The results of our global analysis will open the way for these experiments.
The interplay between SRP and NAC has been controversial [12],[16],[17],[45],[77],[78]. In vitro experiments suggested that SRP can bind to cytosolic non-cognate nascent chains and that NAC and SRP can compete for RNC binding. On the other hand, in vivo analyses did not support the idea that NAC is required for proper SRP function and translocation [16]. Our experiments reconcile these observations and provide an integrated view of the regulation of SRP specificity by NAC. NAC modulates the interaction of SRP with nascent chains in vivo, favoring SRP binding to cognate substrates and disfavoring interactions with non-cognate targets (Figure 6F). Some ER-bound nascent proteins appear to depend on the presence of NAC in the cell for their interaction with SRP (NAC-dependent) while others do not (NAC-independent). Surprisingly, mRNA abundance and translation rate, rather than direct determinants of SRP affinity such as SS or TM hydrophobicity, are the major distinguishing features of the NAC-dependent versus the NAC-independent SRP interactions. This raises the idea that, in vivo, the specificity of the factors that interact with nascent proteins is governed not only by properties of the nascent polypeptide sequence, such as the intrinsic affinity of a given nascent chain for SRP, but also by the competition among cognate and non-cognate nascent polypeptides for these factors and by interactions between factors, exemplified by NAC and SRP.
Our analysis provides insight into the question of how signals such as SS or TM, which are recognized in a variable manner depending on affinity and concentration, can be read in the cell to determine a binary fate such as translocation, that is. proteins do or do not get translocated. Our data show that in wild type cells, SRP does bind with exquisite specificity to cognate substrates spanning a very wide range of cellular mRNA abundances, while disregarding very abundant cytosolic substrates that contain hydrophobic stretches with potential SRP-binding affinity. In ΔNAC cells, however, this specificity is relaxed, so that highly abundant non-cognate substrates bind to SRP and low abundance cognate substrates are lost from SRP. Thus NAC provides an additional level of specificity that fine-tunes SRP interactions to “sharpen” the response.
Our data can be explained in light of previous biochemical and biophysical measurements. NAC and SRP both contact the same ribosomal protein, L25, but have additional non-overlapping binding sites on the ribosome [11]. We observed that SRP-associated polysomes also contain associated NAC (Figure S6A). The interplay between these factors appears to be relevant for SRP specificity. SRP samples most translating ribosomes to encounter RNCs translating cognate polypeptides. Affinity measurements indicate that all translating ribosomes can bind SRP [58]. RNCs translating cytosolic polypeptides have significant affinity for SRP (ca. 8 nM) [58], however this interaction is salt sensitive and likely has a higher dissociation rate [79]. In contrast, SRP binds with extraordinarily high, subnanomolar affinity to RNCs bearing cognate substrates; this interaction is also salt-resistant, perhaps related to its low dissociation rate in vivo [80]. Of note, NAC was shown to reduce association of SRP to non-cognate RNCs. Accordingly, loss of NAC would result in a higher residence time for SRP on ribosomes translating highly abundant non-cognate mRNAs and a lower availability of SRP to bind low abundance cognate mRNAs. Our data suggesting that SRP and NAC overlap in binding to RNCs, much as proposed for trigger factor and SRP in bacteria, open the possibility for them acting in concert on a translating nascent chain. Because it appears that the conformational state of the ribosome contributes to SRP recruitment [43],[51], a more speculative possibility is that NAC exerts its regulatory activity through modulation of the ribosomal cycle.
Despite the relaxed specificity of SRP binding to nascent chains in ΔNAC cells, there was no detectable difference in mRNA targeting to membranes in these cells, and no significant induction of a UPR response (Figure 7), supporting previous findings that NAC has no impact on translocation or the interaction of RNCs with membranes [77],[78]. This is likely the combined result of the redundancy of mRNA targeting pathways, which ensure that secretory proteins reach the membrane, together with proofreading mechanisms that prevent non-cognate SRP-RNC complexes from associating with membranes. For instance, the SRP targeting pathway contains an inbuilt proofreading mechanism at the SRP receptor (SR) level whereby the SRP-SR interaction is enhanced when SRP is bound to a signal sequence [81],[82]. Furthermore, the Sec61 translocon can stringently recognize signal sequence RNCs [5],[47]. These different mechanisms may together provide a robust system that ensures the fidelity of translocation even when the specificity of SRP interactions is impaired.
Strains carrying chromosomally integrated Rpl16-TAP, Rpl17-TAP, Egd1-TAP, Egd2-TAP, Btt1-TAP, Srp68-TAP, and Srp72-TAP were obtained from Open Biosystems, Srp54-TAP from Euroscarf. Δegd1 and Δegd2 yeast strains from the Saccharomyces Genome Deletion Project [83] were used to obtain Δegd1/Δegd2 by mating, sporulation, and tetrad dissection. Sec65-1 strain was a kind gift of Peter Walter. Immunoaffinity purification of specific ribosome-associated factors together with ribosomes and associated RNAs was carried out exploiting the C-terminal TAP-tagged derivative of each selected protein [84]. Briefly, 1 liter cultures were grown to OD 0.7–0.8 in YPD. Following addition of cycloheximide (CHX) (0.1 mg/ml) to stabilize ribosome-nascent chain complexes, cells were harvested by centrifugation, washed twice in buffer A (50 mM Hepes-KOH [pH 7.5], 140 mM KCl, 10 mM MgCl2, 0.1% NP-40, 0.1 mg/ml CHX), resuspended in 2 ml of buffer B (buffer A plus 0.5 mM DTT, 1 mM PMSF, 20 µg/ml pepstatin A, 15 µg/ml leupeptin, 1 mM benzamidine, 10 µg/ml aprotinin, 0.2 mM AEBSF (Sigma), 0.2 mg/ml heparin, 50 U/ml Superasin (Ambion), and 50 U/ml RNAseOUT (Invitrogen)), and dripped into a conical 50 ml Falcon tube filled with and immersed in liquid nitrogen. Frozen cells were pulverized for 1 min at 30 Hz on a Retsch MM301 mixer mill. Pulverized cells were thawed and resuspended in 5 ml of buffer B; cell debris was removed by two sequential centrifugation at 8,000 g for 5 min at 4°C. A 100 µl aliquot (5%) of the supernatant was removed for reference RNA isolation. The remaining lysate was incubated with 6.7×106 beads/µl of IgG-coupled magnetic beads (Dynabeads, Invitrogen) at 4°C for 2 h. Beads were washed once in 5 ml of buffer B for 2 min and 5 times in 1 ml buffer C (50 mM Hepes-KOH [pH 7.5], 140 mM KCl, 10 mM MgCl2, 0.01% NP-40, 10% glycerol, 0.5 mM DTT, 10 U/ml superasin, 10 U/ml RNAseOUT, 0.1 mg/ml CHX) for 1 min, resuspended in 100 µl of buffer C, and incubated for 2 h in 0.3 U/µl TEV protease (Invitrogen) at 16°C. Supernatant was recovered as final pulldown for protein and RNA isolation. Reference RNA was isolated using RNeasy mini kit (Quiagen), while RNA from the eluate was isolated by sequential extraction with Acid Phenol:Chloroform 125∶24∶1 (Ambion), Phenol/Chloroform/Isoamyl Alcohol 25∶24∶1 (Invitrogen), and chloroform followed by isopropanol precipitation with 15 µg of Glycoblue (Ambion) as carrier.
A total of 20 OD254 nm were loaded on a 7%–47% sucrose gradient in buffer B without NP-40. The samples were centrifuged on a Beckman SW-41 rotor for 90 min at 42,000 rpm at 4°C. Gradients were continuously fractionated on an ISCO collector with a flow cell UV detector recording the absorbance at 254 nm. For protein detection by western blotting, fractions were precipitated with trichloroacetic acid, separated by SDS-PAGE, and analyzed by immunoblotting using the indicated antibodies. For RNA isolation and microarrays analysis, fractions corresponding to 60S, 80S, and polysomes were pooled to isolate polysome-associated RNA and the supernatant and low sucrose fractions were pooled to isolate free RNA. RNA was purified with RNeasy Mini Columns kit (Qiagen).
Free cytosolic polysomes and membrane-bound polysomes were fractionated by sedimentation velocity exactly as described [34],[85] starting with 250 ml of exponential growth cultures (of WT or Δegd1/egd2) in YPD. Total RNA from free and membrane-associated polysomes was purified with RNeasy Mini Columns Kit (Qiagen).
50 ml cultures of WT, Δegd1/egd2, or sec65-1 cells were grown in YPD at 30°C or followed by 1 h at 37°C (sec65-1). Cells were starved in SD-Met media for 30 min and labeled with 35S-methionine for 7 min. Endogenously expressed Kar2 and DPAP-B were immunoprecipitated with specific antibodies (a kind gift of Peter Walter and Mark Rose, respectively) and analyzed by SDS-PAGE in 8% acrylamide gels. Translocation defects were measured by comparing the ratio of non-processed precursor versus processed mature protein, namely non-signal sequence-cleaved versus cleaved Kar2 and an unglycosylated precursor versus glycosylated protein for DPAP-B.
3 µg of reference RNA and 50% or up to 3 µg of TAP-tag affinity purified RNA were reverse transcribed with Superscript II (Invitrogen) in the presence of 5-(3-aminoallyl)-dUTP (Ambion) and dNTPs (Invitrogen) with a 1∶1 mixture of N9 and dT20V primers. The resulting cDNA was covalently linked to Cy3 (reference RNA) and Cy5 (purified RNA) NHS-monoesters (GE HealthSciences). Dye-labeled DNA was diluted into 20–40 µl solution containing 3× SSC, 25 mM Hepes-NaOH, pH 7.0, 20 µg poly(A) RNA (Sigma), and 0.3% SDS. The sample was incubated at 95°C for 2 min, spun at 14,000 rpm for 10 min in a microcentrifuge, and hybridized at 65°C for 12–16 h in the MAUI hybridization system (BioMicro). Following hybridization, microarrays were washed in 400 ml of four subsequent wash buffers made of 2×SSC with 0.05% SDS, 2×SSC, 1×SSC, and 0.2×SSC. The first wash was performed at 65°C for 5 min and the following washes for 2 min each at room temperature. Slides were briefly immersed in 95% ethanol and dried by centrifugation in a low-ozone environment to prevent Cy5/3 dyes destruction. Once dry, the microarrays were kept in a low-ozone environment during storage and scanning.
For fractionation experiments, 10 µg of free polysomes-RNA (Cy3) and 3 µg of rER polysomes-RNA were used for reverse transcription.
For analysis of transcriptional levels on mutant strains, 3 µg of reference RNA (wild type strain) (Cy3) and 3 µg of experimental RNA (mutant strain)(Cy5) were used for reverse transcription.
Microarrays were scanned with an Axon Instrument Scanner 4000B (Molecular Devices). PMP levels were adjusted to achieve 0.05% pixel saturation for IP experiments and 0% saturation for analysis of transcriptional levels. Data were collected with the GENEPIX 5.1 (Molecular Devices), and spots with abnormal morphology were excluded from further analysis. Arrays were computer normalized by the Stanford Microarray Database (SMD) [86]. Log2 median ratios were filtered for a regression correlation greater than 0.6 and a signal over background greater than 2.5 to remove low-confidence measurements. Hierachical clustering was performed with Cluster 3.0 [87], and results were visualized with Java TreeView [88].
At least three, usually four, independent biological replicates were employed for each condition. After removing features missing two or more values, we generated a representative dataset by running a one-class t test with 800 (SAM) [22]. Ultimately, substrates were defined as those encoded by mRNAs that were differentially expressed with a false discovery rate (FDR) (q-value) of 1.
Enriched GO terms among the identified targets were retrieved with GO Term Finder [89], which uses the hypergeometric density distribution function to calculate p values and the programs Genetrail [90] and FuncAssociate [91].
The GO database [89] was used to collect a list of GO categories. In these classifications, gene products can be affiliated with one or more GO category assignments.
Lists of proteins with predicted signal peptides and transmembrane regions were downloaded from the Saccharomyces Genome Database (SGD), which uses the prediction programs SignalIP [92] and TMHMM 2.0 [93], respectively.
Intrinsic disorder was predicted from the protein sequences with the Disopred2 software [94] after filtering out coiled-coil and transmembrane regions with the program pfilt (http://bioinf.cs.ucl.ac.uk/downloads/pfilt). Reported is the fraction of the protein sequence that is predicted to be unstructured. Sequence hydrophobicity was approximated by the average of the hydrophobicity profile, computed from the Kyte and Doolittle scale [95] with averaging over sliding windows of size 7. Hydrophobic stretches were defined as 5 or more consecutive amino acids that surpassed a threshold of 1 in the hydrophobicity profiles. Data on translation rate, ribosomal density, and mRNA expression were retrieved from [23] and [96]. Statistical data analysis was performed in R (www.r-project.org). Box plots indicate the data distribution through median, 25%, and 75% quartiles (filled box), as well as the range of non-outlier extremes (dashed lines).
|
10.1371/journal.pntd.0001211 | Characterisation of the Wildlife Reservoir Community for Human and Animal Trypanosomiasis in the Luangwa Valley, Zambia | Animal and human trypanosomiasis are constraints to both animal and human health in Sub-Saharan Africa, but there is little recent evidence as to how these parasites circulate in wild hosts in natural ecosystems. The Luangwa Valley in Zambia supports high densities of tsetse flies (Glossina species) and is recognised as an historical sleeping sickness focus. The objective of this study was to characterise the nature of the reservoir community for trypanosomiasis in the absence of influence from domesticated hosts.
A cross-sectional survey of trypanosome prevalence in wildlife hosts was conducted in the Luangwa Valley from 2005 to 2007. Samples were collected from 418 animals and were examined for the presence of Trypanosoma brucei s.l., T. b. rhodesiense, Trypanosoma congolense and Trypanosoma vivax using molecular diagnostic techniques. The overall prevalence of infection in all species was 13.9% (95% confidence interval [CI]: 10.71–17.57%). Infection was significantly more likely to be detected in waterbuck (Kobus ellipsiprymnus) (Odds ratio [OR] = 10.5, 95% CI: 2.36–46.71), lion (Panthera leo) (OR = 5.3, 95% CI: 1.40–19.69), greater kudu (Tragelaphus strepsiceros) (OR = 4.7, 95% CI: 1.41–15.41) and bushbuck (Tragelaphus scriptus) (OR = 4.5, 95% CI: 1.51–13.56). Bushbucks are important hosts for T. brucei s.l. while the Bovidae appear the most important for T. congolense. The epidemiology of T. vivax was less clear, but parasites were detected most frequently in waterbuck. Human infective T. b. rhodesiense were identified for the first time in African buffalo (Syncerus caffer) and T. brucei s.l. in leopard (Panthera pardus). Variation in infection rates was demonstrated at species level rather than at family or sub-family level. A number of significant risk factors interact to influence infection rates in wildlife including taxonomy, habitat and blood meal preference.
Trypanosoma parasites circulate within a wide and diverse host community in this bio-diverse ecosystem. Consistent land use patterns over the last century have resulted in epidemiological stability, but this may be threatened by the recent influx of people and domesticated livestock into the mid-Luangwa Valley.
| Animal and human trypanosomiasis are constraints to both animal and human health in Sub-Saharan Africa, but there is little recent evidence as to how these parasites circulate in natural hosts in natural ecosystems. A cross-sectional survey of trypanosome prevalence in 418 wildlife hosts was conducted in the Luangwa Valley, Zambia, from 2005 to 2007. The overall prevalence in all species was 13.9%. Infection was significantly more likely to be detected in waterbuck, lion, greater kudu and bushbuck, with a clear pattern apparent of the most important hosts for each trypanosome species. Human infective Trypanosoma brucei rhodesiense parasites were identified for the first time in African buffalo and T. brucei s.l. in leopard. Variation in infection is demonstrated at species level rather than at family or sub-family level. A number of significant risk factors are shown to interact to influence infection rates in wildlife including taxonomy, habitat and blood meal preference. Trypanosoma parasites circulate within a wide and diverse host community in this bio-diverse ecosystem. Consistent land use patterns over the last century have resulted in epidemiological stability, but this may be threatened by the recent influx of people and domesticated livestock into the mid-Luangwa Valley.
| Trypanosomes are true multi-host parasites capable of infecting a wide range of wildlife species that constitute a reservoir of infection for both people and domestic animals. Natural infections of trypanosomes in wildlife were first identified during the Sleeping Sickness Commission in the Luangwa Valley, Zambia [1], which had been set up following the identification of the first case of the ‘rhodesian’ form of human sleeping sickness in 1910 [2]. Since that time, many surveys from across Africa have identified extensive natural infection of wild animal hosts with a variety of trypanosome species [3]–[18]. Although much knowledge about infection rates in wildlife has been generated, there have been very few attempts to conduct epidemiological analyses of data from surveys of wildlife. This is, in part due to difficulties in the collection of sufficiently large sample sizes of a representative nature from wildlife species to enable a thorough statistical analysis [19]–[21]. Some studies have examined the association between wildlife infection rates and factors that might influence them [8], [17], but most have reported host species infection rates without any statistical analysis of the data. Consequently, knowledge of the factors associated with trypanosome infection in wildlife is limited.
In the Luangwa Valley, high tsetse (and trypanosome) challenge coupled with high levels of predation and an arid environment, have meant that livestock keeping has been virtually non-existent in this area. The majority of the land is protected for the conservation of the environment and industries based around both consumptive and non-consumptive tourism form the main sources of revenue for local peoples. Pressure for land is lower than in many wildlife areas of Africa, although there has been an influx of people from Eastern Province into the Central Luangwa Valley in recent years. Three general surveys of trypanosome infections of wildlife have been conducted [1], [7], [13]. All three surveys have utilised parasitological methods of diagnosis, although the study by Keymer, 1969 combined parasitological identification with rodent inoculation and used the blood incubation infectivity test (BIIT) [22] to investigate the presence of human infective T. b. rhodesiense. This human infective, zoonotic subspecies has been identified in bushbuck (Tragelaphus scriptus), duiker (Sylvicapra grimmia), giraffe (Giraffa camelopardalis thornicrofti), impala (Aepyceros melampus), lion (Panthera leo), warthog (Phacocoerus africanus) and waterbuck (Kobus ellipsiprymnus) [23], impala and zebra (Equus quagga boehmi) [24] and warthog [7], [25] in the Luangwa Valley to date.
The investigation of the wildlife reservoir for the human infective subspecies has for a long time been complicated by difficulties in the definitive diagnosis of the parasite. Although the BIIT [22] represented a significant advance in diagnostic capabilities, recent advances in molecular diagnostic methods have resulted in the development of a new, highly specific and robust diagnostic tests for this parasite [26]. Advances in molecular methods of diagnosis have also resulted in the development of a multispecies polymerase chain reaction (PCR) that is capable of differentiating all the major pathogenic trypanosomes of domestic livestock in a single test [27]. This study aimed to apply these novel laboratory techniques to further characterise the wildlife reservoir for trypanosomes in the Luangwa valley. A statistical analysis of infection rates in wildlife was to be used to identify the principle components of the wildlife reservoir community for T. brucei s.l., T. congolense and T. vivax, in the absence of any domesticated animal hosts. Additionally, in order to understand the transmission of trypanosomes better, the data was to be used to identify risk factors for infection.
A cross-sectional survey of the trypanosomiasis prevalence in wildlife hosts was conducted in the Luangwa Valley from 2005 to 2007. The Luangwa Valley is situated in the Eastern and Northern provinces of Zambia and represents an extension of the Great Rift Valley in East Africa. Sample collection took place between May and November of each year as the main roads north and south within the Luangwa Valley are impassable during the rainy season. Due to the inherent difficulty in collecting samples from wildlife in remote areas samples were collected using non-randomised convenience sampling techniques.
Samples were collected using two sampling approaches. Firstly, professional hunters were recruited to provide samples from animals harvested as part of the commercial ‘safari hunting’ system operating in Zambia. These animals are shot under licence in game management areas (GMAs) outside the national parks and hunting licenses are issued under a quota system regulated by the Zambian Wildlife Authority (ZAWA). The study area covered eight GMAs with a widespread distribution across the north and central Luangwa Valley and one private hunting area further south in Luembe Chiefdom (Figure 1). Additional samples were also collected from animals destroyed in GMAs as part of population control measures or problem animal control.
Secondly, samples were collected from animals immobilised or captured as part of routine conservation management activities within the national parks of the Luangwa Valley. The majority of the animals sampled were captured in North Luangwa National Park (NLNP) by a commercial game capture unit as part of a re-stocking programme for the Malawi / Zambia Transfrontier Conservation Area. Additional samples were collected from South Luangwa National Park and Lower Lupande GMA. The overall study area covered represents the largest area covered by any trypanosomiasis survey of wildlife in the Luangwa Valley to date.
This study utilised blood samples collected from wild animals that had already been shot as part of commercial safari hunting activities under a strictly licensed quota system managed by the Zambian Wildlife Authority. These animals were not shot for the purpose of this study. Additional samples were also collected from animals captured and released unharmed as part of a translocation exercise for the Zambia / Malawi Transfrontier Conservation Area. All activities in protected areas were fully approved by the Zambian Wildlife Authority (permit numbers 316295 and 323947). All sampling protocols were approved by the Zambian Wildlife Authority and the Zambian Department of Veterinary and Livestock Development. The relevant export and import licences were obtained for samples from animals on CITES appendices 1 and 2.
Blood samples were collected onto FTA® Cards (Whatman, Maidstone, Kent, UK) and were left to air dry out of direct sunlight. Samples were stored in multi-barrier pouches (Whatman) with desiccant at ambient temperature prior to being processed in the laboratory. Samples collected by hunters were collected either directly from the bullet wound immediately after the animal was shot, or at the time of skinning from heart blood or muscle smeared onto the FTA® card. A small number of samples (26) were collected onto Isocode cards (Schleicher & Schuell, Dassel, Germany), rather than FTA® cards, and were stored in a similar manner. The remaining samples were collected from a superficial ear vein into heparinised capillary tubes and applied to FTA® Cards. Where sufficient amounts of blood could not be obtained from an ear vein, blood samples were collected from larger peripheral veins (jugular, saphenous, cephalic or abdominal) into EDTA coagulated vacutainers and stored at 4° centigrade prior to application to FTA® Cards.
Each sample was assigned a unique identification number. Supplementary data on the species, sex, age, date and location were recorded. Age was recorded using knowledge of breeding seasons and examination of the physical maturity of the animal to assign it to one of three categories: (i) born in the present breeding season or within the last six months (juvenile); (ii) born the previous breeding season, but not yet mature (sub-adult); (iii) physically mature (adult). For the samples collected through the professional hunter survey, Global Positioning System (GPS) coordinates were requested for all samples, but a number of hunters were unable to provide this information. GPS coordinates were recorded for all other samples using a hand-held Garmin GPS device.
All samples collected were screened using PCR methods. Eluted deoxyribonucleic acid (DNA) was used to seed each PCR and was prepared using the following sample preparation protocols. Samples collected onto Whatman FTA® Cards were firstly prepared for elution by punching five 3 mm diameter discs from each card using a Harris Micro-Punch ™ Tool. The use of multiple sample punches from each card increases the likelihood of detection of trypanosomes present at low levels of parasitaemia [28]. The discs were then washed twice in 1000 µl of Whatman Purification Reagent for 15 minutes followed by two washes in 1000 µl of 1× concentrated TE buffer for 15 minutes to remove any residual Whatman purification reagent. The discs were then transferred to 100 µl PCR tubes, with all five discs from each sample placed in one tube, and allowed to dry at room temperature. Once the discs had dried, a Chelex 100® elution protocol was used to elute the DNA [29]. 100 µl of 5% (w/v) Chelex solution (sodium form, 100–200 mesh; Bio-Rad Laboratories, Hemel Hempstead, Hertfordshire, UK) was added to each sample and mixed thoroughly by pipetting. The samples were then heated to 90°C for 30 minutes in a DNA Engine DYAD™ Peltier Thermal Cycler. Eluted DNA was stored at 4°C and used in a PCR within 12 hours of elution. Unused sample DNA was stored in aliquots at −20°C for longer periods.
Samples collected onto Isocode cards, were prepared for elution by punching three 3 mm diameter discs from each sample card and placing them together in a sterile 1.5 ml microcentrifuge tube. 500 µl of sterile de-ionised water (dH2O) was added and the tubes pulse vortexed for five seconds a total of three times. The discs were gently squeezed against the side of the tube and placed in a sterile 0.5 ml microcentrifuge tube. The DNA was eluted using a simple water elution protocol. 100 µl of dH2O was added and the tubes were then heated to 95°C for thirty minutes in a DNA Engine DYAD™ Peltier Thermal Cycler. The tubes were gently tapped twenty times to mix the sample and the discs containing the matrix were removed. The eluate was stored at 4°C and used in a PCR within 12 hours of elution. Excess sample eluate was stored at −20°C in 10 µl aliquots.
Samples were initially screened using a multispecies nested PCR that distinguishes all clinically important African trypanosome species and some sub-species [27]. The PCR targets the Internal Transcribed Spacers (ITS) of the small ribosomal subunit (200 copies per genome), producing different sized products for different trypanosome species. ITS-PCR was performed on each sample in a standard reaction volume of 25 µl using 1 µl of eluate as a template for each reaction, under the reaction conditions described by Cox et al [27].
All samples were also screened using a species-specific PCR for T. brucei s.l. [30]. The TBR-PCR is a species specific PCR for trypanosomes belonging to the Trypanozoon subgenus. The primers are designed to amplify a target region with a copy number of 10,000 per genome making it a highly sensitive test. The PCR was carried out on each sample using 5 µl of eluate in a standard reaction volume of 25 µl under the reaction conditions described by Moser et al [30].
Any samples detected as being positive for T. brucei s.l. using either of the above PCRs were then subjected to a multiplex PCR for the detection of the T. brucei rhodesiense subspecies [26], [31]. The two T. brucei s.l. subspecies, T. brucei brucei and T. brucei rhodesiense are distinguished by the presence of the SRA gene in the latter's genome. The multiplex PCR is designed to amplify the SRA gene, thereby enabling the differentiation of the two subspecies. As the SRA gene is a single copy gene, primers amplifying the single copy GPI-PLC gene are also included within the PCR as a positive control to show that enough genomic material was present for the SRA gene to be detected if present. A failure to detect the GPI-PLC gene in TBR-PCR positive samples might suggest that the prevalence of T. b. rhodesiense was being underestimated. 5 µl of eluate from each positive sample was used in a standard reaction volume of 25 µl under the reaction conditions described by Picozzi et al [26]. In order to improve the sensitivity of the method, the reaction was run three times for each sample using 45 cycles and three times using 50 cycles.
In all PCRs, a positive control (genomic DNA) and two negative controls (blank FTA punch and water) were run with each reaction. A DNA Engine DYAD™ Peltier Thermal Cycler was used to run the reactions and PCR products were separated by electrophoresis in a 1.5% (w/v) agarose gel containing 0.5 µg/ml ethidium bromide. Separated products were then visualised under ultraviolet light in a transilluminator.
Logistic regression models with binomial errors were used for the investigation of trypanosomiasis prevalence. Data was initially entered and evaluated using the Microsoft ®Office Excel 2003 spreadsheet program. All analytical exploration of data was conducted using the statistical software package, R: A language and environment for statistical computing [32]. Additional functions included within the Epicalc 2.7.1.2. package for R (V. Chongsuvivatwong) were also used in the analysis. The likelihood ratio test was used to assess the significance of individual factors in each model. For individual factor categories, the likelihood of infection in comparison to the reference category was presented as the odds ratio (OR). The Walds statistic was used to assess the significance of the OR and is presented as the probability (p) value. Statistical significance was accepted at the 95% confidence level throughout the analysis.
The models were firstly used to investigate the overall prevalence detected of all trypanosome species combined. The analysis was then repeated to investigate the prevalence of T. brucei s.l., T. congolense and T. vivax separately. The effect of host species, age, sex, area, month and year on trypanosome prevalence detected was examined using each factor as an explanatory variable. The potential confounding effect of sample collection method was also investigated with each sample being assigned to a category according to whether it was sampled alive or dead. The data from the samples collected by hunters were also initially analysed as a separate dataset as were the data from the samples collected from national parks before the whole dataset was analysed together. Over-saturation of some FTA cards with blood was observed during laboratory analysis and there was a concern that excess haem might interfere with the PCR for these samples. To investigate this each sample was assigned to a non-oversaturated category if the colour of the eluate was clear and to an over-saturated category if the eluate was discoloured by residual blood pigments.
To further investigate the variation in prevalence between species and to facilitate a multivariable analysis, three methods of grouping the species sampled were compared. Firstly, a grouping based on the taxonomic classification of species at the sub-family level (or family level where no sub-family exists) was used. This followed the standard text by Wilson and Reeder [33]. Secondly, a grouping based on both the predominant vegetation type that the species favoured (open, closed or mixed) and the territorial or spatial movement patterns of the species (sedentary or non-sedentary) was investigated, again using the above text as a reference. Habitat was classified in this way in order to reflect potential association with preferred tsetse habitat. The final grouping method investigated was designed to reflect the blood meal preferences of the three tsetse species found in the Luangwa Valley. Clausen et al's publication on blood meal preferences [34] was used as reference to assign each wild animal species sampled from into one of three levels (low, medium and high) depending on the proportion of blood meals it accounted for. Although this publication contained many samples from Zambia, blood meal preferences were presented only by tsetse species not by geographical region so no regional values could be used. Additionally, many species sampled from were not included in the publication and for these species more general publications on host preferences were used to subjectively assign categories [35], [36]. The groupings and how they were calculated are summarised in Table 1. The data was initially explored at the univariable level and then at the multivariable level. However, multivariable analysis was only possible for overall trypanosome prevalence as there was inadequate data for the individual trypanosome species.
In total 418 samples were collected from 24 species in the survey (Table 2). The majority of these were collected from GMAs through the professional hunter survey in which 331 samples were collected from 22 species. All of these samples were from adult animals and only four were from female animals. A total of 80 samples from five species were collected from the NLNP and these were more representative with 34 samples from male animals and 46 from females. Twenty-four sub-adults were sampled along with two juveniles, the remainder being adults. An additional seven samples were collected from the other management activities.
The cumulative prevalence of all trypanosomes in the dataset was 13.9% (95% CI: 10.71–17.57%). Four mixed infections were detected giving an overall prevalence of mixed infections of 1.0% (95% CI: 0.26–2.43%). The percentage of total infections present as mixed infections was 6.9%. All involved T. brucei s.l., with three occurring concurrently with T. congolense (one each in a bushbuck, warthog and wildebeest (Connochaetes taurinus cooksoni)) and one occurring concurrently with T. vivax (in a waterbuck).
The effect of wild animal species on the overall prevalence of trypanosome infections was highly significant (p<0.001) and several species had a statistically significant increased risk of being infected with trypanosomes (Table 3). Waterbuck were the most likely species to be detected as being infected, with a significant OR of 10.5 (95% CI: 2.36–46.71, p = 0.002). Lion, greater kudu (Tragelaphus strepsiceros) and bushbuck were also significantly more likely to be detected as being infected, with respective ORs of 5.3 (95% CI: 1.40–19.69, p = 0.014), 4.7 (95% CI: 1.41–15.41, p = 0.012) and 4.5 (95% CI: 1.51–13.56, p = 0.007). The prevalence detected in each species with at least one positive sample is shown in Figure 2 (A).
The effect of taxonomy group on the overall trypanosome prevalence was highly significant (p = 0.002). However, no individual taxonomy group had a significantly increased risk of being infected compared with the reference Suidae group. The group with the highest prevalence was the Pantherinae and the odds of this group being detected as infected with trypanosomes approached significance (OR = 2.8, 95% CI: 0.90–8.75, p = 0.077). The prevalence detected in each sub-family is shown in Figure 2 (B). The effect of habitat group was also highly significant as a factor (p<0.001) with the sedentary closed habitat group significantly more likely to be infected than the reference sedentary open habitat group (OR = 6.6, 95% CI: 2.18–20.15, p<0.001) (Figure 2 (C)). Both sedentary and non-sedentary mixed habitat groups had an increased likelihood of being infected, but this was not significant in either group. Using Tukey contrasts as a method of the multiple comparison of means, the sedentary closed habitat group had a significantly higher prevalence of trypanosome infections than both the sedentary mixed and sedentary open groups (p = 0.004 and p = 0.009, respectively).
The effect of blood meal preference group was also significant as a factor (p = 0.018). The high blood meal preference group had an increased risk of being infected (p = 0.013) with an OR of 2.2 (95% CI: 1.18–3.99) compared to the reference low blood meal preference group. Interestingly, the likelihood of the medium blood meal preference group being infected was lower than the reference group (OR = 0.9, 95% CI: 0.37–2.01, p = 0.722), but this difference was not statistically significant. The prevalence detected by blood meal preference group is shown in Figure 2 (D).
There were no significant effects of age on the prevalence of trypanosomes detected in any of the datasets. Sex was found to be a significant factor (p = 0.028), with male animals having higher odds of being infected than females (OR = 3.2, 95% CI: 0.96–10.58, p = 0.058). However, this effect was confounded by species as only one female sample was collected from a species with a high prevalence of trypanosomes. When adjusted for species the effect was no longer significant (p = 0.544) and the adjusted OR was lower (OR = 1.5, 95% CI: 0.38–6.15, p = 0.554). There were no significant effects of area on the prevalence of trypanosome infection and no spatial patterns were apparent in the data (Figure 3). No significant effects of the year of sampling were detected. Although month of sampling had a significant effect (p = 0.028), most of the samples from NLNP were collected in September and the species sampled had a lower prevalence. When the effects of month were adjusted for confounding by area the effect was no longer statistically significant (p = 0.248). Although sample collection method appeared to have a significant effect (p = 0.002), this was no longer the case when adjusted for either species (p = 0.267) or area (p = 0.432). The effect of over-saturating Whatman FTA or Isocode cards with blood was not statistically significant when the whole dataset was considered. However, when the samples collected by hunters were considered alone the effect approached statistical significance (p = 0.059) and samples that were classified as being over-saturated had a reduced likelihood of being detected as infected (OR = 0.6, 95% CI: 0.31–1.03, p = 0.063). When combined with the rest of the data, the effect became statistically insignificant (p = 0.209), but the odds of being detected as infected was still lower for over-saturated samples (OR = 0.7, 95% CI: 0.39–1.24, p = 0.215). The results of the univariable analysis of risk factors for overall infection with trypanosomes are summarised in Table 4.
A multivariable analysis was conducted using the taxonomy grouping of species. This grouping method was selected for the final analysis as it had the lowest residual deviance. However, despite the grouping of species, the nature of the data resulted in large standard errors so a reduced dataset (326 observations) with all species with a sample size less than five or no positive samples removed was used. The final multivariable model included taxonomy grouping and over-saturation of sample cards, with area included as a confounding variable. No factors were significant, but the effect of over-saturation approached significance (p = 0.054) with over-saturated cards less likely to be detected as infected (OR = 0.5, 95% CI: 0.28–1.02, p = 0.058). Infection rates were highest in Pantherinae (OR = 2.0, 95% CI: 0.53–7.27), Bovinae (OR = 1.6, 95% CI: 0.57–4.59) and Reduncinae (OR = 1.2, 95% CI: 0.41–3.71) taxonomy groups.
The overall cumulative prevalence of T. brucei s.l. in all species was 5.7% (95% CI: 3.71–8.42%). The prevalence detected using the individual species PCR for T. brucei s.l. was 5.3% (95% CI: 3.33–7.86%) compared with 0.5% (95% CI: 0.06–1.72%) using the multispecies PCR. Two T. b. rhodesiense infections were detected using the SRA-PCR giving a prevalence of 0.5% (95% CI: 0.06–1.72%). The positive samples came from a male adult bushbuck from Chifunda hunting block in Musalangu GMA and a male adult buffalo (Syncerus caffer) from the Nyamaluma area of Lower Lupande GMA. The proportion of all T. brucei s.l. infections that were identified as T. b. rhodesiense was therefore 0.08, or 8.3%. However, the GPI-PLC gene was not detected in the majority of the T. brucei s.l. positive samples.
Host species was again significant as a factor (p = 0.042) and the bushbuck presented a significantly greater odds of being detected as infected (OR = 7.1, 95% CI: 1.7–29.33, p = 0.007). No other host species had a significantly greater likelihood of being detected as infected when compared with the reference warthog (Table 5). A bar chart of the prevalence detected in all species with at least one positive sample is presented in Figure 4 (A). Oversaturation of Whatman FTA cards also had a significant effect on T. brucei s.l. prevalence both when the complete dataset was analysed (p = 0.024) and when the samples collected by hunters were considered separately (p = 0.010). Over-saturated FTA cards were significantly less likely to be detected as positive with an OR of 0.4 (95% CI: 0.13–0.94, p = 0.038) using the complete dataset and 0.3 (95% CI: 0.11–0.81, p = 0.018) using the hunter dataset. Year also had a significant effect on the prevalence (p = 0.015), with samples collected in 2007 presenting a reduced likelihood of being detected as positive for T. brucei s.l. (OR = 0.2, 95% CI: 0.04–0.57, p = 0.005).
The overall prevalence of T. congolense in all species was 6.0% (95% CI: 3.91–8.70%). Host species had a significant effect on the prevalence (p = 0.001) with greater kudu the species most likely to be detected as infected (OR = 8.7, 95% CI: 2.24–33.58, p = 0.002), followed by lion (OR = 5.2, 95% CI: 1.11–24.31, p = 0.036). No other species had a significantly increased risk of infection compared with the reference warthog species. A summary of the prevalence detected and OR for each species is shown in Table 5 and a bar chart of the prevalence detected for each species with at least one positive sample is shown in Figure 4 (B). No other factors had a significant effect on the T. congolense prevalence using the combined dataset. There was, however, a significantly lower likelihood of detecting T. congolense in the month of September (OR = 0.2, 95% CI: 0.05–0.64, p = 0.008).
The T. vivax prevalence of 3.1% (95% CI: 1.67–5.26%) was lower than that for the two other trypanosome species that were investigated in this study. Host species had a significant effect on this prevalence (p = 0.002) and waterbuck was highly significant as a host with an OR of 55.0 (95% CI: 5.33–567.59, p = <0.001) (Table 5). Although buffalo also had an increased likelihood of being detected as infected, the OR was not significant. No other factors had significant effects on the T. vivax prevalence. Figure 4 (C) shows a bar chart of the prevalence of T. vivax in all wild animal species with at least one positive sample.
The accurate diagnosis of trypanosomes in field surveys of wildlife populations has historically presented many challenges, in particular for T. brucei species. The protocol employed in this survey offered the advantage of an efficient method of sample collection and storage, combined with highly specific molecular techniques for diagnosis. The use of hunter kills as a source of surveillance material enabled a wide range of species to be sampled and increased the sample size obtainable from the resources available. Although the data generated was a convenience sample and is likely to be biased in terms of the sex and age distribution of the population sampled, this is a common problem with surveys of wildlife populations [19]–[21] and is difficult to overcome.
Where resources allow, molecular techniques of diagnosis offer the advantage over more traditional techniques of improved diagnostic specificity and sensitivity. This survey, along with a sister-project in Tanzania [37], represented the first use of the multispecies ITS-PCR [27] on field samples collected from free-ranging wildlife. A recent publication that used very similar protocols reported the specificity for T. brucei s.l. in a cattle population in Kenya to be 0.997 for the ITS-PCR and 0.998 for the TBR-PCR [38]. The sensitivities were not as high, however, with estimates of 0.640 and 0.760 respectively for the two techniques. The lower sensitivities achieved were attributed to the use of filter paper cards for DNA preservation and illustrate the main disadvantage of this technique of sample storage. The figures reported in this paper, therefore, although highly specific, are likely to underestimate the true prevalence of infection in wildlife. Only the data for T. congolense Forest, Kilifi and Savannah sub-species, T. brucei s.l. and T. vivax were used in the data analysis due to difficulties encountered in the accurate differentiation of bands at the sizes expected for T. simiae and T. simiae Tsavo. This limited the ability to detect mixed infections and the level reported (1.0%) might be lower than expected. However, it is possible that the high level of trypanosome challenge experienced by wild hosts in this ecosytem encourages the formation of a cross-immunity, as has been postulated for lions, and this may reduce the prevalence of mixed infections [39].
Host species was consistently identified as the most significant risk factor for infection with trypanosomes throughout the univariable analysis and no other factors had a significant effect after adjustment for confounding. The taxonomy grouping of species (p = 0.002) and habitat grouping (p<0.001) were also highly significant with blood meal preference grouping (p = 0.02) less so. When the residual deviances for each model were compared, the lowest value was obtained when no grouping was used at all (267.04) compared with the models containing the taxonomy grouping (304.03), habitat grouping (306.20) and blood meal preference grouping (328.57). As the residual deviance represents the unexplained deviance in the model, it is clear that much of the variation in infection rates occurs at the species level. However, it is also clear that complex interactions between parasite, host and vector determine the infection status of wildlife hosts.
At the individual trypanosome species level, host species was again the most significant factor explaining the variation in infection rates. Oversaturation of sample cards had a significant effect on T. brucei s.l. prevalence only and this suggests that the TBR-PCR may be more sensitive to inhibition by haem products than the multispecies ITS-PCR. The potential temporal effects detected in this study were most likely induced by the study design and resulted from sampling more species of low prevalence in the NLNP survey in 2007. It was difficult to investigate any potential confounding induced by the different sampling methods used as samples collected using the two methods were not collected from the same areas and species. This illustrates the difficulties with collecting data from wildlife suitable for a robust statistical analysis.
Although no association between the prevalence of trypanosome infection and age was evident in this study, the age distribution of the data limited the ability to investigate this as a risk factor. Very few studies have previously investigated the effect of age on trypanosome prevalence, but of those that have, one reported that the prevalence in buffalo peaked at two and a half years [8] and another found no statistically significant difference between the prevalence in young and old animals [17]. However, a more recent study of 13 lion prides in the Greater Serengeti Ecosystem in Tanzania where the actual age was known to within an accuracy of one month, reported that the prevalence of T. brucei s.l. infections showed a distinct peak and decrease with increasing age [39]. Most infections with T. brucei s.l. were cleared between three and five years of age and no human infective T. b. rhodesiense parasites were detected in lions over six years old. It was postulated that frequent challenge and an exposure dependent cross-immunity following infections with more genetically diverse species such as T. congolense, led to partial protection sufficient to prevent animals from harbouring human infective T. b. rhodesiense. As most samples in the study presented here were collected from older animals, it is possible that the prevalence in lions and potentially other species has been underestimated.
The results of this survey demonstrate the ability of trypanosomes to survive in a very wide variety of wildlife hosts. New identifications of T. b. rhodesiense in African buffalo and T. brucei s.l. in leopard (Panthera pardus) suggest that the reservoir community is even more diverse than previously thought. However, as illustrated by the species prevalence graphs, the majority of infections were concentrated in a smaller number of species. The majority of T. brucei s.l. infections were detected in four species, namely bushbuck, leopard, lion and waterbuck. Of these, the bushbuck was the only species to have a significantly greater likelihood of infection than warthog (OR: 7.1, 95% CI: 1.7–29.33). The bushbuck has previously been identified as an important reservoir host for T. brucei s.l. in the Luangwa Valley [1], [40] as well as in the Lambwe Valley in Kenya [3], [41]. Although overall densities of bushbuck are not high, they are locally abundant within the dense woodland and thicket vegetation common near the Luangwa River and its tributaries [42]. The high proportion of blood meals taken from this species by Glossina pallidipes tsetse [34], despite the relatively low overall population of bushbuck, suggests that there is a close ecological association between the two species. A transmission cycle involving this sedentary host and G. pallidipes is therefore likely to play an important role in maintaining foci of T. brucei in the Luangwa Valley, as has been proposed for the Lambwe Valley in Kenya [3].
The identification of T. brucei s.l. in samples from leopard in this study appears to be the first published record of infection in this species. Compared to many other protected areas in Africa, leopards are relatively common in the Luangwa Valley. It would appear that both lion and leopard are capable of supporting a moderate prevalence of T. brucei s.l. infection and there may be a secondary transmission cycle involving the carnivores. Interestingly it has been postulated that carnivores can become infected from their prey through abrasions in the oral mucosa and this has been demonstrated in artificial experiments [4]. Despite spending much of the day lying in dense thicket, these species also do not account for a high proportion of tsetse blood meals [34]. Trypanosome infections in these species may therefore be an example of bioaccumulation rather than vector transmitted disease, but no conclusions can be drawn on the route transmission from this study. However, considering their relatively low density, their contribution to trypanosome transmission is unlikely to be large whichever transmission route is involved.
The precise contribution of the waterbuck to the transmission of T. brucei s.l. is still unclear, although the high prevalence detected in this study and others [1], [7] suggests that they are highly susceptible to infection. Although they may be locally abundant overall densities are not high and they are rarely fed on by tsetse [34], [42]. They have been reported to produce allomones that repel tsetse and reduce the likelihood of feeding once landed [43]. However, they occupy a niche environment on the fringes of thicket and woodland and are clearly very susceptible to infection with all three trypanosome species. It has been postulated that their high susceptibility to infection has resulted from the fact that they are rarely challenged by infected tsetse bites [35], but the same can be said for many other species in which low infection rates are detected.
The diagnosis of T. b. rhodesiense in a sample from an African buffalo is, as far as the authors are aware, the first identification in this species. Buffalo are abundant in many savannah ecosystems and are capable of acting as a reservoir host for many pathogens of cattle, most probably because of their close phylogenetic relationship to the latter [44]. They have previously been demonstrated to be susceptible to sub-clinical infections with T. brucei s.l. [6], [8] so this finding is not surprising, but has important implications for the control of the disease. Buffalo are not sedentary animals and herds frequently move over large distances with the potential to disseminate infection to other host species. The finding of this parasite in Nyamaluma, not far from Mambwe and Msoro Districts where there have recently been large influxes of cattle and people, also raises concerns about the possibility of infection becoming established in the cattle population of the Luangwa Valley. In other parts of Africa, particularly Uganda, cattle have demonstrated to be effective maintenance hosts for T. b. rhodesiense [45], [46]. Areas with increasing populations of cattle adjacent to wildlife areas have also been identified as being at risk from epidemics of trypanosomiasis [47]. The prevalence of T. brucei s.l. in buffalo was relatively low, a finding that is in keeping with the moderate level of blood meals coming from this species despite their relative abundance [34], [42]. The only other positive identification of T. b. rhodesiense in this study was in a bushbuck in Musalangu GMA. This subspecies has previously been isolated from a bushbuck in the Luangwa Valley [23] as well as in the Lambwe Valley in Kenya [41] and its important role within the community of hosts for T. brucei s.l. has already been outlined.
The overall prevalence of T. b. rhodesiense detected in the study was relatively low at 0.5% (95% CI: 0.06–1.72%) suggesting that approximately 8% of all T. brucei s.l. identifications were T. b. rhodesiense. However, given that the majority of TBR-PCR positive samples did not test positive for PLC gene, it is likely to be an underestimate of the true prevalence. Even allowing for this, it is clear that the human infective subspecies is maintained at low levels by the reservoir community alongside a rich diversity of other trypanosome species. This might seem surprising given that wildlife hosts have long been regarded as the natural host for this parasite. It is in stark contrast to the ecological picture in Uganda where the human infective parasites circulate efficiently between cattle and man against a background of reduced biodiversity [48]. This raises the possibility that maintenance of biodiversity within the Luangwa Valley ecosystem has influenced the limited emergence of this parasite, although the data in this study are insufficient to prove this and many factors have been implicated in determining the patterns of parasite species richness [49], [50]. As spillover from wildlife has often been implicated as a risk factor for human infection [39], [51], the possibility that maintaining biodiversity might, conversely, limit the risk of infection in some situations warrants further investigation.
Of interest from a conservation perspective was the identification during the study of T. brucei s.l. in two black rhinoceros (Diceros bicornis) that had recently been re-introduced into the Luangwa Valley from a tsetse free area of South Africa (and were therefore not included in the data analysis). Histopathology of the brain from one rhinoceros which had died revealed severe meningo-encephalitis that was considered to be consistent with a diagnosis of clinical trypanosomiasis. Laboratory analysis of blood samples provided positive identification of T. brucei s.l. using the TBR-PCR and the GPI-PLC gene was positively identified using the SRA-PCR. Interestingly, the rhinoceros samples both had very strong positive GPI-PLC bands in comparison to other T. brucei s.l. positive samples where the GPI-PLC band was often negative, a finding which is suggestive of a higher level of parasitaemia in this species. The only clinical signs observed in the rhinoceros were depression and poor condition, although it was not examined by a veterinary surgeon. Although the final cause of death may be attributed to trypanosomiasis, it is not clear if it was a primary or secondary problem. Trypanosomiasis, including infection with T. brucei s.l., has been implicated previously in the post-translocation deaths of rhino [52]–[54].
The community of reservoir hosts for T. congolense would appear to be wider than that for the other trypanosome species, with members of the Bovidae family most frequently represented. Again, two separate transmission routes would appear to occur, one involving many of the ungulate species that are regularly fed on by tsetse and a second one involving the carnivores and possible oral transmission. Of the ungulates, a significant prevalence was detected in this study in greater kudu (OR = 8.7, 95% CI: 2.24–33.58), with moderate levels of infection in bushbuck and warthog. Both greater kudu and bushbuck are preferred hosts for G. pallidipes [34] and are sedentary hosts living largely in thicket or dense woodland, which is the prime habitat for this species of tsetse. This contrasts with the situation regarding warthog, where a close ecological association with G. m. morsitans has been described [34], [55]. A significant prevalence of infection was detected in lion in this study (OR = 5.2, 95% CI: 1.11–24.31), but, as with T. brucei s.l, it is doubtful that they form an important component of the community of reservoir hosts in terms of onwards disease transmission.
It is less straightforward to draw conclusions about the epidemiology of T. vivax infections in wildlife as the overall prevalence detected was much lower. Of all species sampled, waterbuck was the only species with a significant likelihood of infection (OR = 55.0, 95% CI: 5.33–567.59). Although this is clearly a significant odds ratio, the precise contribution of waterbuck to the transmission of infection is unclear and the reasons for this are as discussed for T. brucei s.l. A moderate prevalence with T. vivax was also detected in the more abundant buffalo, with occasional infections in other ungulates. Previous surveys have suggested that bushbuck and greater kudu are also capable of supporting T. vivax infections [7], [13] and agree with the high levels of infection detected in waterbuck [1], [7], [13]. Therefore, although the epidemiological picture is less clear for this species of trypanosome, it is likely that a transmission cycle involving the bovinae sub-family is the most important component of the reservoir.
The Luangwa Valley ecosystem is unusual in modern day Africa due to the limited level of contact between domesticated livestock and wildlife. The results of the survey presented here along with historical surveys conducted in the valley [1], [7], [13] suggest that the epidemiology of trypanosomiasis has remained largely unchanged over the last century. This is in keeping with consistent land use patterns with an almost complete absence of livestock and only a modest change in the human population over the same time period. Infection rates in many species in this survey were comparable with previous surveys in the Luangwa Valley (Table 6). However, in recent years an influx of people and livestock into the Msoro and Mambwe Districts of central Luangwa Valley has led to the development of a new wildlife / livestock / human interface. An investigation into the prevalence of trypanosomiasis in domestic livestock at the site of this new interface in Msoro District, revealed infection rates of 33.3% in cattle, 20.9% in pigs, 27.6% in sheep and 10.2% in goats [56]. Although the laboratory protocol differed slightly from that used in this study, the same multispecies ITS-PCR was used. This is a much higher prevalence than that found in the surrounding wildlife population and represents a significant departure from the historical situation, with ramifications both for trypanosomiasis transmission and that of other infectious diseases. New interfaces have been identified as an important factor in disease transmission [19] and areas surrounding these interfaces have been identified as being at risk of epidemics of bovine trypanosomiasis [47]. Prevalence data produced from trypanosome surveys in neighbouring countries are not directly comparable due to the different diagnostic techniques used, but in general the prevalence recorded has been lower than that in the Luangwa Valley [57], [58].
Trypanosoma parasites circulate within a wide and diverse host community in this bio-diverse ecosystem. With the identification of the African buffalo and the leopard as new host species for T. b. rhodesiense and T. brucei s.l. respectively, it is clear that the reservoir community is wider than previously demonstrated. However, although the host range is very wide, the majority of infections are concentrated in a smaller number of species with a clear pattern of species forming the bulk of the reservoir community for each trypanosome species. Host species was the only consistent risk factor for infection identified in this study and, although many factors may interact to influence the trypanosome prevalence in wildlife, most of the variation in infection rates occurs at the species level. The epidemiology of trypanosomiasis in the Luangwa Valley has remained remarkably stable since the first survey in 1913, in keeping with consistent land use patterns despite some changes in the human population over that period. The recent influx of cattle and people from the plateau regions of Eastern Province represents a significant diversion from these land use patterns and will almost certainly result in changes in the epidemiology of trypanosomiasis in the Luangwa Valley, with cattle becoming increasingly important members of the reservoir community.
|
10.1371/journal.pntd.0004978 | Nonstructural Proteins Are Preferential Positive Selection Targets in Zika Virus and Related Flaviviruses | The Flavivirus genus comprises several human pathogens such as dengue virus (DENV), Japanese encephalitis virus (JEV), and Zika virus (ZIKV). Although ZIKV usually causes mild symptoms, growing evidence is linking it to congenital birth defects and to increased risk of Guillain-Barré syndrome. ZIKV encodes a polyprotein that is processed to produce three structural and seven nonstructural (NS) proteins. We investigated the evolution of the viral polyprotein in ZIKV and in related flaviviruses (DENV, Spondweni virus, and Kedougou virus). After accounting for saturation issues, alignment uncertainties, and recombination, we found evidence of episodic positive selection on the branch that separates DENV from the other flaviviruses. NS1 emerged as the major selection target, and selected sites were located in immune epitopes or in functionally important protein regions. Three of these sites are located in an NS1 region that interacts with structural proteins and is essential for virion biogenesis. Analysis of the more recent evolutionary history of ZIKV lineages indicated that positive selection acted on NS5 and NS4B, this latter representing the preferential target. All selected sites were located in the N-terminal portion of NS4B, which inhibits interferon response. One of the positively selected sites (26M/I/T/V) in ZIKV also represents a selection target in sylvatic DENV2 isolates, and a nearby residue evolves adaptively in JEV. Two additional positively selected sites are within a protein region that interacts with host (e.g. STING) and viral (i.e. NS1, NS4A) proteins. Notably, mutations in the NS4B region of other flaviviruses modulate neurovirulence and/or neuroinvasiveness. These results suggest that the positively selected sites we identified modulate viral replication and contribute to immune evasion. These sites should be prioritized in future experimental studies. However, analyses herein detected no selective events associated to the spread of the Asian/American ZIKV lineage.
| Zika virus is mainly transmitted by mosquitoes and is phylogenetically related to other human pathogens (e.g. dengue virus). After the outbreak in South America, the WHO declared that the spread of ZIKV should be regarded as a public health emergency. In fact, growing evidence suggests that ZIKV infection during pregnancy increases the risk of congenital birth defects. Moreover, ZIKV can trigger Guillain- Barré syndrome, a severe neurological disorder characterized by progressive muscle weakness. Evolutionary studies can help identify sites that allow viral adaptation—i.e. that increase viral fitness at least in some conditions. We analyzed the evolution of the polyproteins encoded by ZIKV and by related viruses and identified several sites in nonstructural proteins that were subject to natural selection. Most of these are located in protein regions that mediate interaction with the host immune system or that may regulate viral RNA synthesis. In ZIKV isolates, the NS4B protein was the preferential selection target with three selected residues. Changes at these sites are expected to modulate some aspect of viral fitness, either in mosquitoes or vertebrate hosts. Future studies to clarify the mechanisms of ZIKV pathogenicity should address the role of these sites in the modulation of viral phenotypes.
| The Flavivirus genus (family Flaviviridae) comprises a large number of viral species, many of which are important human pathogens; these include dengue virus (DENV), yellow fever virus (YFV), Japanese encephalitis virus (JEV), West Nile virus (WNV), and the latest emerged pathogen, Zika virus (ZIKV).
ZIKV was first discovered in 1947 in Uganda, in a sentinel rhesus monkey, and subsequently in mosquitoes of the Aedes genus. Between 1947 and 2006, only sporadic human cases were reported in Africa and in Southeast Asia, until multiple outbreaks in the Pacific islands occurred. The first sizable outbreak was reported in the Federated States of Micronesia (Yap Island) in 2007, followed by an outbreak in French Polynesia in 2013. In 2014, the epidemic spread to Cook Islands, New Caledonia and Easter Island, and reached South America in late 2014 –early 2015 [1–3]. As of May 18, 2016, sixty countries/territories have reported ZIKV cases (http://www.who.int/emergencies/zika-virus/situation-report/en/).
Although ZIKV infection is often asymptomatic or causes only mild symptoms, the WHO declared that the spread of ZIKV should be regarded as a public health emergency of international concern. In fact, growing evidence suggests that ZIKV infection during pregnancy increases the risk of microcephaly, brain damage, and congenital abnormalities [4–6]. Also, retrospective studies indicated that ZIKV can trigger Guillain-Barré syndrome (GBS), a severe neurological disorder characterized by progressive muscle weakness [7]. Moreover, even if Aedes mosquitoes species such as Aedes aegypti and Aedes albopticus represent the primary vectors for natural transmission, perinatal and congenital infections, as well as sexual transmission and infection through blood transfusion have been recently documented [1].
ZIKV is a member of the Spondweni (SPOV) serocomplex and, like other members of the Flavivirus genus, it is a single-stranded positive-sense RNA virus. Its genome consists of about 11,000 nucleotides with two flanking non-coding regions and a single long open reading frame. The encoded polyprotein is co- and post-translationally processed by viral and host proteases to produce three structural (capsid, C; pre-membrane, prM; envelope, E) and seven nonstructural (NS) proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, NS5) [8].
Genetic and phylogenetic studies indicated that ZIKV has evolved into 2 major lineages: African and Asian/American, this latter responsible of the recent outbreaks and associated with reports of GBS and fetal malformations [9, 10].
Analysis of ZIKV genomes from microcephaly cases revealed no shared amino acid changes, suggesting that viral genetic features alone are not responsible for fetal abnormalities [2]. Likewise, inspection of amino acid differences between the Asian and African lineages provided no clear indication of viral genetic features that may result in altered virulence or increased pathogenicity, although no functional study of these variants has been performed to date [2]. It was thus proposed that, if the link with GBS and fetal abnormalities is confirmed, factors other than viral genetics, including infection with other viruses or the host genetic background [2], are responsible for these adverse effects. An alternative possibility is that all ZIKV lineages increase the risk of microcephaly and/or GBS, but the association has been previously missed due to the limited size of African outbreaks and to the lack of surveillance programs. Whereas addressing these questions will require extensive epidemiological and clinical surveys, analysis of all ZIKV strains and of their evolution within the wider perspective of closely related flaviviruses can identify positively selected amino acid changes. These latter are expected to entail a functional effect and can therefore be prioritized in further studies of viral pathogenesis. Indeed, evolutionary analyses in WNV have detected positively selected changes that modulate viral phenotypes such as virulence [11] and superinfection exclusion [12].
Herein we investigated the evolution of the viral polyprotein in ZIKV and in related flaviviruses. Results indicate that NS1 was a major selection target during flavivirus speciation and revealed ongoing selection in ZIKV strains in NS4B and NS5.
Coding sequences were retrieved from the NCBI database (http://www.ncbi.nlm.nih.gov/), all flaviviruses analyzed were selected to have full coding sequence information. A list of accession number is reported in S1 Table.
Alignment errors are common when divergent sequences are analyzed and can affect evolutionary inference. Thus, we used PRANK [13] to generate multiple sequence alignments and GUIDANCE2 [14] for filtering unreliably aligned codons (we masked codons with a score <0.90), as suggested [15].
Substitution saturation was checked using the Xia's index implemented in DAMBE [16, 17]; this test compares an entropy-based index of saturation (Iss) with a critical value (Iss.c). If Iss is significantly lower than Iss.c, sequences have not experienced substitution saturation.
The presence of recombination was assessed using two methods, GARD [18] and Recco [19]. Whereas GARD uses phylogenetic incongruence among segments in the alignment to detect the best-fit number and location of recombination breakpoints, Recco is based on cost minimization and dynamic programming. In GARD, the statistical significance of putative breakpoints is evaluated through Kishino-Hasegawa (HK) tests; breakpoints were considered significant if their p value was < 0.05.
The Recco's output includes a p value for the whole dataset that, controlling for false positives, provides an indication as to whether a significant amount of recombination is detectable in the whole alignment. We concluded that no substantial recombination was present when the dataset p value was >0.05. For alignments showing evidence of recombination in Recco (dataset p value <0.05), we considered sequences as recombinants if the number of savings was >20, and the sequence p value was <0.001, as suggested [20]. Recombination breakpoints were defined accordingly.
Phylogenetic trees were reconstructed using the program phyML with a maximum-likelihood approach, a General Time Reversible (GTR) model plus gamma-distributed rates and 4 substitution rate categories. Branch support was evaluated using a non parametric bootstrap analysis (100 replicates) [21].
The nonsynonymous/synonymous rate ratio (dN/dS or ω) is a widely used method to detect positive selection. Positive selection is inferred when the rate of nonsynonymous (dN) substitutions is higher than that of synonymous (dS) substitutions (dN/dS >1).
To test for the action of episodic positive selection in flaviviruses, we applied the branch-site test [22] from the codeml software [23]. The test estimates selective pressure changes among branches and sites in the phylogenetic tree. Two nested models (MA and MA1) are compared: MA allows positive selection on one or more lineages (called foreground lineages), and the MA1 does not allow such positive selection. Twice the difference of likelihood for the two models (ΔlnL) is compared to a χ2 distribution with one degree of freedom [22]. A false discovery rate correction was applied to take into account a multiple hypothesis issue generated by analyzing different branches on the same phylogeny [24].
A Bayes Empirical Bayes (BEB) analysis was used to evaluate the posterior probability that each codon belongs to the site class of positive selection on the foreground branch, only when 2ΔlnL was statistically significant.
BUSTED (branch-site unrestricted statistical test for episodic diversification) [25] is an alternative approach implemented in the HyPhy package [26] designed to describe episodic positive selection acting on specific branches in the phylogenetic tree at a proportion of sites. A model that allows the action on positive selection on foreground branches is compared with a null model that doesn't allow ω >1. Twice the ΔlnL of the two models is then compared to a χ2 distribution (with two degrees of freedom); if the null model is rejected, at least one site is under positive selection on the foreground branches. To detect selection at individual sites, twice the difference of the likelihood for the alternative and the null model at each site is compared to a χ2 distribution (one degree of freedom).
To be conservative, we considered a site under episodic positive selection if it showed both a p value ≤ 0.05 in BUSTED and a BEB posterior probability ≥ 0.90.
To better understand the evolution of ZIKV genomes, we also applied two random site (NSsite) models implemented in codeml: a null model (M7) that assumes that 0<ω<1 and is beta distributed among sites in all branches of the phylogeny, and a positive selection model (M8); this model is the same as M7 but also includes an extra category of sites in the alignment with ω>1. A χ2 distribution is used to assess statistical significance of 2ΔlnL of the two models.
Positively selected sites were identified using the posterior probability (≥ 0.90) from M8 BEB.
Individual sites under diversifying positive selection were also identified using Random effects likelihood (REL) [27] and Fast Unconstrained Bayesian AppRoximation (FUBAR) [28] methods from the HyPhy package.
REL estimates ω at each site by inferring a gene distribution for both synonymous and non-synonymous rate variations and assuming independent draw at each site from this distribution. We considered a site under positive selection if it showed a Bayes Factor > 50.
FUBAR is an approximate hierarchical Bayesian method that uses an unconstrained distribution of selection parameters by averaging over a large number of predefined site classes. Given this distribution, FUBAR estimates the posterior probability of positive diversifying selection at each site in the alignment (with a cutoff ≥ 0.90).
In order to be conservative, we finally considered a site as under diversifying positive selection if it was detected by at least two different methods.
Data on DENV experimentally verified immune epitopes were obtained from the NIAID Virus Pathogen Database and Analysis Resource (ViPR) online (http://www.viprbrc.org) [29]. Human epitopes were searched for by using the gene product name as a query. Linear epitopes with positive results in any assay type category (B cell, T cell and MHC binding) were included. We used ClustalOmega [30] to align epitopes onto the DENV protein sequence and from this onto the ZIKV sequence.
The structure of NS1 of ZIKV was obtained by homology modeling using the West Nile virus NS1 (PDB ID: 4O6C) structure as a template; analysis was performed through the SWISS-MODEL server [31]. The accuracy of the model was assessed with VADAR (Volume, Area, Dihedral Angle Reporter), which uses several algorithms to calculate different parameters for individual residues and for the entire protein [32].
Images were created using PyMOL (The PyMOL Molecular Graphics System, Version 1.5.0.2 Schrödinger, LLC).
The membrane protein topology for the ZIKV NS4B protein was predicted by using TMHMM (http://www.cbs.dtu.dk/services/TMHMM/) [33].
As mentioned above, ZIKV belongs to the Spondweni group of mosquito-borne flaviviruses. In addition to Spondweni virus (SPOV), the viral species more closely related to ZIKV include the Kedougou (KEDV) and dengue (DENV) viruses [10]. To investigate selective events that took place during the speciation of ZIKV and closely related flaviviruses, we obtained complete coding sequence information for 21 ZIKV, 1 SPOV and 1 KEDV, as well as 11 DENV. ZIKV sequences were selected to represent viruses sampled in both African and in non-African countries, in distinct years, and from different hosts (see S1 Table). In the case of KEDV and SPOV, only one complete genome is available for each virus. For DENV, sequences were selected to belong to the four major serotypes (DENV1 to DENV4); for each serotype, sequences representative of the most common genotypes (based on complete E nucleotide sequences in [34]) were included. DENV sequences were also selected to cover different geographic locations and isolation dates.
The structural and nonstructural coding regions were analyzed separately, and the alignments were pruned of unreliably aligned codons using the GUIDANCE utility (see Methods). This procedure resulted in the masking of 8.6% and 9.5% of codons in the structural and nonstructural regions, respectively. A test for substitution saturation was performed using Xia's method [16] and indicated no substantial saturation in either alignment (S2 Table).
We next analyzed the alignments for the presence of recombination using two different methods, GARD (Genetic Algorithm Recombination Detection) [18] and Recco [19]. No evidence of recombination in the nonstructural region was detected using either program, whereas Recco (but not GARD) suggested the presence of a recombination breakpoint around position 1350–1360 (relative to AY632535 coding sequence) in the E region (Fig 1A). The structural region was thus split into two sub-regions for the following analyses, so as to avoid false positive inferences of positive selection as a result of unaccounted recombination [35].
Phylogenetic trees of the three regions were obtained with phyML. Trees were very similar and fully consistent with previously reported phylogenies for flaviviruses [36], with African and non-African ZIKV isolates forming distinct branches [3] (Fig 1B).
We next searched for evidence of episodic positive selection along the internal branches of flavivirus phylogenies using branch-site tests (Fig 1B). Two different methods were applied to ensure consistency: the branch-site unrestricted statistical test for episodic diversification (BUSTED) [25] and the maximum-likelihood models (MA/MA1) implemented in the PAML suite [23]. These two approaches rely on different assumptions of ω (nonsynonymous/synonymous rate ratio) variation among branches. Episodic positive selection at each tested branch was declared when statistically significant support was obtained with both methods. Using this criterion, we found no evidence of positive selection in the structural region. Conversely, both tests detected evidence of positive selection on one branch in the phylogeny of the nonstructural region (Fig 1 and Table 1). Selected sites along this branch were identified using the Bayes empirical Bayes (BEB) procedure from model MA and with BUSTED; again, only sites detected by both methods were considered. A total of 16 positively selected sites were detected (Fig 1A); notably, seven of such sites are located in the NS1 protein. To test whether this number is higher than expected, we performed random sampling across codons in the nonstructural region (i.e. we assumed that all codons that were not masked by GUIDANCE in any sequence have the same probability of being called as positively selected). Results indicated that the likelihood of having 7 selected sites in NS1 amounts to 0.0039; thus, this protein represented a preferential selection target during flavivirus speciation.
We note that the percentage of codons masked by GUIDANCE ranged widely among protein regions, from 3.3% in NS3 to more than 30% in the 2k and NS2A region (S3 Table). Whereas this does not affect the significant enrichment we obtained for NS1 (as we accounted for masked codons), the power to detect selection in extensively masked regions is clearly reduced; in fact, the ultimate result of masking is fewer codons available for analysis or a shallow phylogeny at available codons.
In NS1, the 7 positively selected sites are distributed along the entire protein region (Fig 1C). To gain insight into the role and the spatial localization of these sites, we performed homology modeling of ZIKV NS1 using the West Nile virus protein structure (PDB ID: 4O6C) as a template (Fig 2). We also retrieved the location of experimentally validated immune epitopes in NS1. Selected sites 112 and 114 map on a disordered loop of the wing domain; this loop is exposed, and several DENV immune epitopes were described in this region, most of them covering both positions (Fig 1C). In DENV, an alanine mutation at the 114 site affects virus particle production [37]. This residue was also suggested to have a role in the interaction between NS1 and the envelope glycoprotein [37]. Interestingly, three sites (residues 77, 112, 114) in the “wing” domain localize in close proximity on the protein structure (Fig 2), suggesting that they are involved in formation/stabilization of the same interactions. NS1 position 164 is located in a hydrophobic protruding loop, flanking a smaller loop that is essential for DENV viral replication [38]. Mutations in flanking positions (residues 160 and 162) affect both RNA synthesis and virus viability [38]. Indeed, residues 159–162 of the connector domain together with the β-roll (where the Y22 selected site maps) form a hydrophobic protrusion that faces the membrane (Fig 2). This hydrophobic structure is thought to be involved in the interaction between the NS1 homodimer and the replication complex through the NS4A and NS4B proteins [38]. The β-roll domain is also involved in NS1 dimerization [38]. Site 22 localizes in spatial proximity to the first β-strand of the β-ladder domain, where the G185 positively selected site also maps. Both sites are located at the dimerization interface [38, 39].
As for selected sites in proteins other than NS1, site V15 in NS2A maps to a hydrophobic protein region within the lumen of the endoplasmic reticulum (ER); mutations at nearby residues impair DENV virion assembly [40]. Positively selected sites were also detected in NS3 (M41, P82, T570, P577). Interestingly, site 570 flanks a conserved asparagine that is essential for NS3-NS5 binding in DENV [41]. Finally, the positively selected site in NS4B (I162) is located in a cytoplasmic loop involved in the interaction with NS3 and with host proteins [42].
We next investigated whether positive selection also occurred during the recent evolution of ZIKV. To this purpose, we retrieved coding sequence information for all complete ZIKV genomes (n = 39, as of March 26th, 2016) (S1 Table). Again, the structural and nonstructural regions were analyzed separately. In the structural region, GARD detected no recombination, whereas Recco inferred possible breakpoints at nucleotides 802–838 (relative to AY632535 coding sequence) in the M region. Both Recco and GARD detected a recombination breakpoint in the nonstructural region (position 8994, GARD; position 9040–9054, Recco) within the NS5 region. The portions encompassing the breakpoint positions were thus removed and the alignments were split into two sub-regions. Inspection of the Recco output indicated that in all cases recombination involved sequences from the African isolates. In fact, the phylogenetic trees for all sub-regions showed a clear separation of the African and non-African sequences (S1 Fig).
To obtain an estimate of the degree of constraint along ZIKV genomes, we used FUBAR to identify sites showing significant evidence of negative selection. This analysis indicated an uneven distribution of negatively selected sites, with the weakest selective pressure acting on the structural portion; conversely, more than 80% of sites are negatively selected in the NS1-NS4B region (Fig 1A).
We next tested for positive selection using both the codeml site models (M7 vs M8) and the branch-site models (MA1 vs MA). These latter models were used to test for selection on the branch of the phylogeny that separates the African and non-African sequences. No evidence of positive selection was obtained for the two sub-regions from the structural portion. Conversely, for the nonstructural region covering nucleotides 2371–8994, a model of evolution that allows a class of codons to evolve with ω >1 (NSsite model M8) better fitted the data than the neutral model (NSsite models M7), supporting the action of positive selection (-2ΔlnL = 18.89, degrees of freedom = 2, Bonferroni- corrected p value for two tests = 1.58 X 10−4). Positively selected sites were identified using the BEB procedure from M8 and with two additional methods from the HyPhy suite, REL and FUBAR. Sites were defined as being positively selected if they were detected by at least two different methods. Using this conservative criterion, 5 positively selected sites were detected (Table 2 and Figs 1A and 3). Three of them are located in the relatively short NS4B region (M26, M87, and H88); using the same approach as above, we determined that this number is unlikely to occur by chance (random sampling, p value = 0.007), indicating that NS4B is the preferential positive selection target in ZIKV.
To gain insight into the location of positively selected sites in NS4B, we performed an in silico prediction of transmembrane helices. The resulting topology model was very similar to those previously proposed or determined for other flaviviruses [42] (Fig 3B). Residue M26 maps to the N-terminal region located in the ER lumen. Interestingly, the corresponding position was previously found to be positively selected in sylvatic DENV2 isolates; in JEV, a nearby residue is positively selected, as well [43, 44] (Fig 3B). Residues 87 and 88 are also located in the ER lumen and reside in the second loop (Fig 3B), a region involved in NS4B-NS1 interaction in WNV [45].
Two other positively selected sites are located in NS5 (N287 and V374). Position 374 is part of the nuclear localization signal (NLS) region of NS5. Dengue virus serotypes have different nuclear localization, and these differences are due to changes in their NLS [46]. Analysis of DENV immune epitopes indicate that some of them comprise positions 374 and 287 (Fig 3C).
Herein we provide an analysis of the selective forces acting on ZIKV and related flaviviruses. We show that positive selection contributed to the genetic diversity of these human pathogens and we report ongoing adaptive evolution in ZIKV strains.
The evolutionary analysis of viral genomes poses challenges related to the possible presence of recombination, as well as to the high sequence divergence, with consequent saturation issues and alignment uncertainties. We accounted for all these possible confounding effects, which would otherwise affect inference of positive selection. Indeed, we adopted recommended alignment and filtering criteria to minimize erroneous codon alignments [15], and we tested for substitution saturation. As for recombination, we applied two methods, based on different features of the data, to screen the alignments and to infer the most likely position of breakpoints. These latter were used to split alignments into sub-regions that were separately analyzed. In this respect, it is worth noting that we did not detect recombination breakpoints in the nonstructural region for the extended flavivirus phylogeny, whereas we found evidence of recombination when all ZIKV strains alone were analyzed. The explanation for this apparently contradictory finding is that one single ZIKV African sequence contributed to the recombination events and it was not represented in the flavivirus phylogeny. Moreover, the flavivirus alignment was partially masked to remove unreliably aligned codons. This procedure clearly determines the removal of the most divergent regions, which may derive from recombination events. This most likely accounts for the discrepancy between our results and those from a previous report that indicated recombination between Asian ZIKV strains and SPOV within NS2B [47]. Another previous study analyzed African and non-African ZIKV isolates and reported the presence of four recombination breakpoints in ZIKV genomes [48]. In our analysis we only detected two breakpoints. We believe that the main reason for the discrepancy with this previous analysis derives from the fact that the ArD142623 strain, which contributed most recombination events in Faye's dataset [48], was not included in our study because its genome sequence is not complete and because its polyprotein sequence is annotated as “nonfunctional due to mutation” in GenBank. In this respect, it is worth mentioning that despite our findings and those previously reported by others for ZIKV [47, 48] and DENV [34], experimental data have indicated that flaviviruses have very low propensity for recombination [49, 50]. Moreover, under laboratory conditions specifically devised to detect recombination, extremely rare events were observed that generated aberrant JEV genomes with reduced growth properties [50]. These observations raise the possibility that recombination events identified through analysis of existing sequences in public databases are artifacts of laboratory contamination or incorrectly assembled sequence files. This was previously suggested to be the case for some “recombinant” DENV sequences [34, 51]. All the recombination events we detected involved one or few sequences from African ZIKV isolates. Whereas we cannot control for the accuracy of the deposited sequences, we have to take the possible recombination events into account; failure to do so would affect positive selection inference, irrespective of whether recombination actually occurred. Clarification of these potential caveats, though, is extremely relevant for epidemiological and preventive purposes. Because ZIKV, DENV, and other arboviruses can co-circulate during outbreaks [10], it will be extremely important to assess if and with what efficiency these viruses can recombine.
The branch-site tests we applied to analyze the flavivirus phylogeny were aimed at detecting episodic positive selection—i.e. selective events on one branch of the phylogeny and thus likely to have occurred during or after speciation. Using this approach we were able to show that positive selection acted on the branch that separates DENV from the other analyzed flaviviruses and mainly targeted NS1. It should be noted that the branch-site tests have low false positive rates and are largely insensitive to violations of the assumption of neutral evolution on the background branches [22, 24], but lack power [52]. Thus, selection may act on additional branches than the one we detected and more selected sites may exists. When analysis was performed on ZIKV genomes, which are characterized by much lower divergence compared to the flavivirus sequences in the inter-species analysis, tests to detect episodic and pervasive positive selection were applied. The branch-site test showed no evidence of episodic selection and we consequently identified no selective events leading to the spread of the Asian ZIKV lineage. We mention, however, that the branch-site test may have failed to detect weak selection or selection at a very limited number of codons. We also stress that the lack of selection signatures does not imply that amino acid differences between African and non-African ZIKV lineages are irrelevant or nonfunctional.
Conversely, we identified pervasive selection—i.e. selective events that involve all ZIKV lineages- and, again, selected sites were found to occur in nonstructural portions of the ZIKV genome (NS4B and NS5). These portions also display the strongest level of selective constraint.
Structural proteins (the E protein in particular) might be a priori considered to be preferential selection targets during flavivirus evolution for least two reasons: these proteins [1] mediate the initial and essential steps of host infection via host cell binding and entry and [2] represent major targets for immune responses influencing antigenic selection [53]. Nevertheless, we found no evidence of positive selection in structural regions, either in flaviviruses or in ZIKV isolates. To our knowledge, no study has investigated the occurrence of positive selection in ZIKV or during flavivirus speciation, but efforts at detecting positive selection in DENV strains or isolates were performed. Depending on the serotype analyzed, on the geographic and temporal origin of the viruses, as well as on their transmission cycle (sylvatic or endemic), different genomic regions were found to represent targets of positive selection in DENV [54–59]. These regions were not limited to the structural portion, but also included nonstructural proteins [54–58]. Likewise, analysis of JEV sequences revealed selection in both structural and nonstructural regions [43, 44]. Also, ample evidence indicated that although most neutralizing antibodies are directed against flavivirus E proteins [53], non-neutralizing anti-NS1 antibodies are protective against severe disease during DENV, YFV, and JEV infection [60–64].
Finally, cell-mediated immunity was shown to target both structural and nonstructural DENV proteins, with the vast majority of T-cell epitopes located in nonstructural proteins [65]. In this respect, it is worth noting that several of the positively selected sites we detected in NS1 and in NS5 are located within immune epitopes. This observation suggests that the underlying selective pressure responsible for selection at these sites is exerted by the host adaptive immune system. These data are likely to be relevant for the current efforts to develop a ZIKV vaccine and to assess the possible cross-protection afforded by natural or vaccine-induced immunity against other related viruses.
Nonstructural proteins play different roles in flavivirus life cycles and several of them interact with innate immunity molecules. NS1, the major selection target in the flaviviruses we analyzed herein, is essential for viral RNA replication and is involved in immune system evasion. In particular, secreted hexameric NS1 represents a major antigenic marker of viral infection for all DENV serotypes. Soluble NS1 in the serum of patients has been found to correlate with severe clinical disease [66], suggesting that the NS1 protein also plays an important role in the pathogenesis of dengue. Importantly, the NS1 protein from WNV and DENV2 interacts with multiple components of the complement system (C1S, C4, C4-binding protein, CFH), as well as with toll-like receptors (TLR3, TLR2, and TLR6) [67, 68]. The molecular details of these interactions are presently unknown, but the presence of several positively selected sites in NS1 suggests a possible arms race with the host innate immune system. It will be extremely important to assess whether amino acid differences in flavivirus NS1 proteins affect the interaction with innate immunity components and consequently modulate the host response to ZIKV or DENV.
NS1 is also required for efficient viral genome replication. Recently, it has been proposed that dimeric NS1 plays an organizational role in the formation of the replication complex on the cytoplasmic side of the ER membrane [69], and that this function is mediated by interactions with NS4A and NS4B [45, 70]. Interestingly, the positively selected site in the connector domain (residue 164) is located in the hydrophobic protrusion that may contact NS4A and NS4B [38]. Although mutation of residue T164 to alanine has no effect on RNA replication or on the assembly of DENV particles [37], we cannot exclude its involvement in (de)stabilizing the interaction with the ER membrane; in fact, the introduction of a histidine (Fig 2) at this site might affect the protein function more importantly than the conservative alanine substitution. Scaturro and colleagues [37] also reported that NS1 plays a critical role in the biogenesis of DENV virions, a function that is mediated by interaction with structural proteins. In this context, a key role is played by two residues (114 and 115) in the flexible loop of the NS1 wing domain. Indeed, alanine mutation of residue S114 abrogates DENV2 NS1 binding to E, prM, and C [37]. Notably, we identified residue 114 as positively selected in the flavivirus phylogeny and two additional selected sites were located in close spatial proximity.
Overall, these observations suggest that positive selection at NS1 is acting to optimize viral fitness by modulating viral replication efficiency and/or favoring evasion from the host immune system. Similar considerations may apply to NS4B, which displays 3 positively selected sites in ZIKV isolates. This membrane protein has a role in the formation of the replication complex and in virus pathogenesis [71]. Several mutations in the NS4B region of JEV, YFV, and WNV were shown to modulate neurovirulence and/or neuroinvasiveness [42]. Notably, one of the positively selected sites we identified in NS4B (26M/I/T/V) was previously reported to represent a selection target in sylvatic DENV2 isolates but not in the endemic strains [57]. A nearby residue was also found to evolve adaptively in JEV, both in genotype I and genotype III isolates [43, 44]. However, variation at this site seems not to be associated with host preferences in JEV [44]. Although the functional significance of changes at position 26 in NS4B remains to be clarified, the fact that this residue or a flanking one is targeted by positive selection in three closely related flaviviruses suggests an important role in viral adaptation. Indeed, the N-terminal region of flavivirus NS4B (amino acids 1–125) inhibits interferon (IFN) response by blocking IFN-α/β signaling [72]. This region includes two additional positively selected sites (M87 and H88) and is involved in host protein (e.g. STING) binding. It is interesting to note that YFV NS4B, but not DENV NS4B, can bind STING [73], suggesting that positive selection in this region results from adaptation to the host innate immune system to modulate binding of viral sensors. The region surrounding positions 87 and 88 is also responsible for NS1-NS4B binding and the same study demonstrated the importance of the F86C mutation in WNV NS4B to rescue viral replication in presence of NS1 nonfunctional mutations [45]. Finally, the DENV NS4B region spanning residues 84 to 146 is required for interaction with NS4A, another molecule involved in flavivirus replication [74].
Thus, based on data from other flaviviruses, the three positively selected sites we identified in NS4B of ZIKV are located in a protein region important for interaction with other viral proteins and with host molecules.
The relatively sparse sampling of ZIKV genomes and the paucity of ZIKV sequences isolated from humans in Africa and from mosquitoes in Asia/America, prevents drawing any definite conclusion about the role of selected sites on host preference, pathogenicity, or infectivity. Moreover, as anticipated above, a potential issue associated with viral sequence analysis concerns laboratory contaminations, especially during serial passages in culture. Contaminations were previously suggested to account for discrepancies in DENV phylogenies [34], and a few of the African ZIKV strains we included in the study were passaged several times in suckling mice or cell culture [9]. This process may also introduce variants that are not present in nature, potentially affecting evolutionary inference. These issues are unlikely to affect the analyses we performed on the extended flavivirus phylogeny, as variation on terminal branches has minor effects. However, these caveats should be kept in mind in the analysis of ZIKV strains, especially for positively selected sites showing variability in a minority of sequences. Further evolutionary analysis of ZIKV will greatly benefit from the sequencing and inclusion of additional isolates, not only from the ongoing American epidemic, but also from African countries.
Despite these limitations, we suggest that the positively selected sites we identified should be prioritized in future experimental studies. These amino acids changes are expected to modulate aspects of viral fitness, either in mosquitoes or vertebrate hosts. In this respect, reverse genetic approaches will be instrumental to assess the role of specific changes on different viral phenotypes including transmission by distinct Aedes mosquito vectors or alternative (e.g. human-to-human) transmission modes, increased viremia in humans, and altered tissue tropism.
Finally, we note that NS1 and NS4B are regarded as attractive candidates as direct or indirect targets for antiviral drugs in flavivirus infections [42, 75]. Nonetheless, these proteins are fast evolving in ZIKV and related flaviviruses, and the numerous selected sites are expected to entail functional differences among closely related viruses or even among viruses belonging to the same species. Thus, our data suggest that compounds developed against DENV NS4B [42] or drugs that result in DENV NS1 misfolding [75, 76] may not be active against ZIKV.
|
10.1371/journal.ppat.1005502 | A Functional Bacterium-to-Plant DNA Transfer Machinery of Rhizobium etli | Different strains and species of the soil phytopathogen Agrobacterium possess the ability to transfer and integrate a segment of DNA (T-DNA) into the genome of their eukaryotic hosts, which is mainly mediated by a set of virulence (vir) genes located on the bacterial Ti-plasmid that also contains the T-DNA. To date, Agrobacterium is considered to be unique in its capacity to mediate genetic transformation of eukaryotes. However, close homologs of the vir genes are encoded by the p42a plasmid of Rhizobium etli; this microorganism is related to Agrobacterium, but known only as a symbiotic bacterium that forms nitrogen-fixing nodules in several species of beans. Here, we show that R. etli can mediate functional DNA transfer and stable genetic transformation of plant cells, when provided with a plasmid containing a T-DNA segment. Thus, R. etli represents another bacterial species, besides Agrobacterium, that encodes a protein machinery for DNA transfer to eukaryotic cells and their subsequent genetic modification.
| Since the discovery of gene transfer from Agrobacterium to host plants in the late 1970s, this bacterial pathogen has been widely used in research and biotechnology to generate transgenic plants. Agrobacterium’s infection process relies on a set of virulence proteins that mediate the transfer of a segment of its own DNA (T-DNA) into the host cell genome. To date, Agrobacterium is believed to be the only prokaryote with the capability of cross-kingdoms gene transfer. However, homologs of the Agrobacterium’s virulence proteins are found in some symbiotic plant-associated bacterial species, belonging to the Rhizobium genus. Here we show that one of these species, Rhizobium etli, encodes a complete set of virulence proteins and is able to mediate transfer and integration of DNA into host-plant cell genome, when provided with a T-DNA. This is the first time that a bacterium-to-plant DNA transfer machinery encoded by a non-Agrobacterium species is shown to be functional.
| The Rhizobiales order contains many species of plant-associated bacteria, such as the related genera Agrobacterium and Rhizobium. Phylogenetic analyses based on 16S rDNA sequences led to the idea that Agrobacterium and Rhizobium could be regrouped into one genus [1]. Yet their lifestyles are very different. Agrobacterium comprises species that are often, but not always [2, 3], pathogenic and can genetically transform their host plant cells by transferring a segment of their own plasmid, the T-DNA, and induce neoplastic growths that synthesize small molecules used as nutrients by the bacteria [4–6]. This Agrobacterium capability to modify genetically their host cells is widely used in research and biotechnology for generating transgenic plants [7] as well as fungi [8]. In contrast, Rhizobium belongs to a group of very diverse symbiotic bacteria (collectively termed rhizobia) that form nitrogen-fixing nodules on the roots of legume plants [9–12]. Rhizobium and Agrobacterium species have complex genomes composed of one or two chromosomes and several plasmids [13–16]; the chromosomes are designed as “core” components defining the species as opposed to the “accessory” components that are the plasmids [17]. The outcome of interactions of these bacteria with plants is essentially determined by large specialized plasmids, the tumor inducing (Ti) plasmid for Agrobacterium, and symbiotic (pSym) plasmid for Rhizobium. Indeed, introducing an Agrobacterium Ti plasmid into some rhizobia species resulted in virulent bacteria capable of inducing tumors in host plants [18]. In general, rhizobia species are known to gain T-DNA transfer ability only when provided with the virulence (vir) genes [4, 5] of the Agrobacterium Ti plasmid [19, 20]. Rhizobium, therefore, is thought to possess chromosomal, but not plasmid-based factors required for plant genetic transformation, and because of that lack endogenous DNA transfer capacity.
Intriguingly, however, many Rhizobium species harbor different sets of homologs of the Agrobacterium vir genes; specifically, R. etli carries a complete set of vir genes [15, 21] whereas the closely related R. leguminosarum lacks such degree of homology. Here, we show that R. etli can independently mediate functional DNA transfer and stable genetic transformation of plant cells, when provided with a plasmid containing a T-DNA segment. Thus, R. etli represents another bacterial species, in addition to Agrobacterium, capable of genetic modification of plants.
Sequencing of the R. etli CFN42 genome revealed that it encodes a complete set of virulence (Vir) proteins encoded by the vir genes [15, 22]. Indeed, Fig 1A shows that all the essential Vir proteins encoded by the p42a plasmid of R. etli exhibit a high level of homology with their counterparts from different Agrobacterium Ti plasmids, except for the VirD3 and VirD5 proteins, which are non-essential for DNA transfer. Phylogenetic analysis demonstrated that the Vir proteins of R. etli and Agrobacterium are very close to each other, as exemplified for VirE2 (Fig 1B). In contrast, the putative Vir protein orthologs of R. leguminosarum only share a relatively weak homology, i.e., usually less than 40% identity, with Agrobacterium. Fig 1C shows that, within the p42a plasmid of R. etli, the vir genes are grouped in a cluster, forming a virulence region that is similar in many ways with the vir region of Agrobacterium Ti-plasmids, but it also displays some notable differences. Specifically, the organization of the “core” of the vir region—the virA, virB, virG, virC, virD, and virE operons—is nearly identical, but the order of the virD and virE operons is inverted in R. etli. In addition, in R. etli, the virB2 coding sequence is not part of the virB operon, but is located at a distant locus on the same plasmid, and two virF homologs are present, virF1 and virF2, which are related to the virF genes from octopine (A6) and nopaline (C58) Agrobacterium strains, respectively. The presence of many transposase insertion sequences in the vicinity of the vir cluster of R. etli [15] may explain the rearrangements in the organization of the vir region. In R. leguminosarum, the organization of the vir region located on the pRL7 plasmid appears to be scrambled, with several operons having been duplicated (see the pRL7 map in the KEGG database, http://www.genome.jp/kegg/). Although other Rhizobium species, such as R. mesoamericanum and R. tropici, contain homologs of several vir genes (S1 Table), a high level of homology with all essential vir genes is found only in R. etli. Whereas a complete vir region is present in the R. etli p42a plasmid which is homologous to the Agrobacterium vir genes, we could not detect homologies to any of the Agrobacterium T-DNA sequences; specifically, our search for T-DNA-specific oncogenes and opine synthesis genes and for the T-DNA border sequences did not yield significant homology.
To examine potential functionality of the vir genes of R. etli, we introduced into R. etli cells a plasmid that harbors a T-DNA sequence with reporter genes gfp or gus-int, and selection gene nptII but lacks any vir sequences. This strain was then tested for its ability to promote transient T-DNA expression in plant cells and generate stably transformed transgenic plants, and compared to A. tumefaciens EHA105, one of the standard strains for plant genetic transformation [24]. After infiltration of Nicotiana benthamiana leaves with R. etli, expression of both GFP (Fig 2A and 2B) and β–glucuronidase (GUS) reporters (Fig 2C) was consistently observed in the inoculated plant tissues, although expression levels with R. etli were about ten times lower than those with A. tumefaciens (Fig 2B). Thus, R. etli was able to transfer to plant cells DNA that subsequently could be expressed. In contrast, in similar experiments performed with R. leguminosarum, transient expression of the reporter gfp or gus-int genes was never observed (Fig 2).
That R. leguminosarum—which is very closely related to R. etli, except for the vir region—is unable to effect genetic transformation suggests that it is the vir genes that are required for the T-DNA transfer by R. etli. We tested this notion directly using R. etli carrying p42a with virG or virE2 genes mutated by insertion of a promoterless gusA gene [21]. PCR-based analysis using primers specific for gusA and virG and virE2 showed that R. etli cells with the mutated p42a plasmids, i.e., p42a virGmut and p42a virE2mut, indeed, contained the mutagenic sequences inserted in the sense orientation within the virG and virE2 genes. Specifically, Fig 3A shows that the reverse primer, corresponding to the 3’-end of gusA, and forward primers, corresponding to the 5’-ends of virG and virE2, amplified fragments of ca. 2.3 Kb (lane 1) and 2.8 Kb (lane 5) for the virGmut and virE2mut mutants, respectively, but not for the wild-type genes in the same strains, i.e., for virE2 in the virGmut strain (lane 2) and for virG in the virE2mut strain (lane 4). As expected, no gusA sequences were detected in the wild-type p42a plasmid (Fig 3A, lanes 7, 8) whereas all samples contained bacterial chromosomal DNA (Fig 3A, lanes 3, 6. 9). Neither of these plasmids was able to promote transfer and transient expression of the gfp reporter gene (Fig 3B). In control experiments with R. etli carrying the wild-type p42a, the gfp reporter was transferred to plant cells, resulting in expression of its protein product (Fig 3B, see also Fig 2A). These observations suggest that the T-DNA transfer mediated by R. etli relies on and requires its vir genes.
For stable genetic transformation, tobacco (N. tabacum) leaf discs were inoculated with R. etli or A. tumefaciens harboring a plasmid with the selection gene nptII encoding resistance to kanamycin as well as the gfp marker gene in its T-DNA. Regenerating plantlets were observed after four weeks incubation under kanamycin selection, which indicates stable genetic transformation (Fig 4A and 4B). Consistent with the transient T-DNA expression data, the genetic transformation efficiency mediated by R. etli was much lower than with A. tumefaciens (compare Fig 4A to Fig 4B). Confirming stable transgene expression in the regenerated plants, GFP was observed in a typical nucleocytoplasmic pattern in virtually all cells in leaves of one-month-old transgenic plants generated using R. etli (Fig 4C).
Finally, we confirmed the actual presence of the T-DNA within the genome of these transformed plants. Genomic DNA was isolated from two independent stable transgenic lines, designated TL1 and TL2, and from a wild-type, untransformed plant and analyzed by Southern blot hybridization. Specifically, the DNA samples were digested with EcoRI and hybridized them with a probe corresponding to the T-DNA right border-proximal nos promoter region of the T-DNA of pBin19-RCS1-GFP that has no recognition sites for EcoRI. Fig 5A shows that no hybridization signal was detected in the DNA from the wild-type plant (lane 1) whereas T-DNA-specific signal was present in the DNA of both transgenic TL1 and TL2 lines (lanes 2, 3, asterisks), suggesting a single integration site of the T-DNA within the genome of each of the tested plants. When we similarly digested purified pBin19-RCS1-GFP DNA, which has only one EcoRI recognition site in its entire sequence, a 11.9-kb band, corresponding to the linearized plasmid, was observed (Fig 5B, lane 1). Additional negative controls, which probed EcoRI-digested wild-type and transgenic plant DNA with sequences specific for the p42a plasmid (Fig 5B, lanes 2, 3) or for the R. etli chromosome (Fig 5B, lanes 5, 6) did not yield any signal. As expected, positive controls detected specific signals using undigested p42a DNA hybridized to the p42a-specific probe (Fig 5B, lane 4) and undigested R. etli chromosomal DNA hybridized with the chromosome-specific probe (Fig 5B, lane 7).
Taken together, the stable expression of two marker genes, gfp and nptII, and the physical presence of the transforming DNA in the plant genomic DNA, indicate that the T-DNA was indeed integrated into the plant genome.
Our results demonstrate that R. etli within its p42a plasmid contains a complete and functional vir region, encoding a set of Vir proteins able to mediate functional T-DNA transfer into plant cells. Whereas it has been known that the vir genes from Agrobacterium can function in several rhizobia species [18, 19], this is the first time that an endogenous virulence system encoded by a non-Agrobacterium species is shown to be functional in DNA transfer and stable genetic transformation.
The virE2 and virG mutants, which render R. etli unable to promote genetic transformation, previously have been shown to have no effect on formation of nitrogen-fixing nodules or on nodulation competitiveness [21]. Thus, the vir genes likely fulfill a function unrelated to symbiosis. Two factors might account for the presence of a functional vir region in R. etli. First, the ability to transform host plant cells may have been widespread among bacterial species in the past, and not restricted to the Agrobacterium genus. That we could not identify T-DNA-like sequences in R. etli suggests that Rhizobium-mediated plant transformation does not occur at present, although it cannot be ruled out that other Rhizobium strains, not yet sequenced, harbor a T-DNA. Furthermore, proteins from other rhizobia, such as Mesorhizobium loti R7A (see S1 Table for Vir protein sequence homologies with M. loti R7A), can be recognized by the Agrobacterium VirB/D4 type IV secretion system (T4SS) and exported to plant cells [25], suggesting that T4SS could substitute for the type III secretion system (T3SS) during effector protein translocation in some rhizobia species. Thus, the VirB/D4 T4SS encoded by p42a could also function to translocate protein effectors in R. etli. Second, because the p42a plasmid is transmissible between Rhizobium and Agrobacterium [21], this plasmid may belong to an “interspecies plasmid pool”, and R. etli may function as a “vector” for p42a which is then transferred to Agrobacterium and only then used for plant genetic transformation. It would be interesting to examine whether quorum sensing signals that activate conjugative transfer of plasmids between Agrobacterium cells also induce conjugation between Rhizobium and Agrobacterium. Indeed, in natural Agrobacterium populations, Ti-plasmids are not present in all cells [2, 3], but, in response to bacterial and plant signals via a quorum sensing mechanism, conjugative plasmid transfer can be activated [26].
The need to identify or even generate non-Agrobacterium bacterial species that could be used as a vector for plant genetic transformation has been emphatically articulated [27]. First, a non-Agrobacterium vector might be more efficient in some hosts that are difficult to transform by Agrobacterium. Indeed, although the efficiency of R. etli mediated transformation of Nicotiana species was very low compared to Agrobacterium, R. etli might be more efficient with other plant species, such as its native hosts. Second, several aspects of plant genetic transformation methods are legally limited by existing patents, and using a different bacterial species may help to circumvent these limitation and avoid litigation [28].
In conclusion, we demonstrate that R. etli, a symbiotic Rhizobium species different from the phytopathogenic Agrobacterium, contains the complete molecular machinery able to transfer DNA to the plant genome, which has implications for evolution and origin of the Agrobacterium virulence system as well as for potential utilization in biotechnology.
Protein sequences were compared using the blastp program (PubMed); the percentages of identity of full sequences were calculated as the percentage of identity corrected by the query cover percentage. VirE2 phylogenetic tree was generated using MEGA version 6 [29], via the minimum evolution method. The KEGG database release 71.0 (http://www.genome.jp/kegg/) was used to design schematic maps for the different vir regions.
R. etli CE3, a streptomycin-resistant isolate of the CFN42 strain, and R. leguminosarum bv. viciae strain 3841 (kindly provided by Dr. Russell Carlson, University of Georgia, Athens) were grown in TY medium (5 g.L-1 tryptone, 3 g.L-1 yeast extract, and 10 mM CaCl2). A. tumefaciens strain EHA105, derived from nopaline wild-type strain C58.C1, was grown as described [30]. T-DNA containing plasmids were introduced into these Agrobacterium and Rhizobium strains using the classical CaCl2 protocol, with minor modifications in the case of Rhizobium [31]. R. etli strains, carrying p42a with mutated virG and virE2 genes were described previously [21].
For transient expression of GFP, pCB302T-GFP was obtained by inserting the gfp expression cassette from pSAT1-EGFP-C1 [32] into the AgeI-BglII sites of pCB302T-MCS [33], derived from pCB302 [34]. For transient expression of GUS, pBISN1 [35], carrying an expression cassette for a gus reporter gene with a plant intron sequence (gus-int), was used. For stable transformation, the multiple cloning site of pPZP-RCS1 [36] was first introduced into the EcoRI-HindIII sites of pBin19 [37], forming pBin19-RCS1. Then, the gfp expression cassette from pSAT1-EGFP-C1 was inserted into the AscI site of pBin19-RCS1, resulting in pBin19-RCS1-GFP carrying both nptII and gfp expression cassettes in its T-DNA segment.
Agrobacterium and Rhizobium strains carrying pCB302T-GFP or pBISN1 were grown 24–48 h at 28°C, and infiltrated into intact N. benthamiana leaves as described [38]. The bacterial suspension was first adjusted to OD600nm 0.6 and then diluted 20 or 50 times before infiltration. Reporter gene expression was monitored three days after infiltration. For detection of GUS expression, leaf discs were excised from the infiltrated zone and subjected to the histochemical assay as described [39]. GFP expression was observed under a Zeiss LSM 5 Pascal confocal microscope at low magnification with a 10x objective; the number of GFP-expressing cells per cm2 of infiltrated leaf surface was counted as described [38].
Stable genetic transformation was performed using N. tabacum cv. Turk and Agrobacterium and Rhizobium strains carrying pBin19-RCS1-GFP in the classical leaf disc protocol [40]. Transgenic plantlets were selected on MS regeneration medium (30 g.L-1 sucrose, 8 g.L-1 agar, 10 mg.L-1 BAP, 1 mg.L-1 NAA) supplemented with 50 mg.L-1 timentin and 50 mg.L-1 kanamycin. Images of regenerated transgenic plantlets were recorded after 4 weeks of incubation on the regeneration/selection medium, using a Leica MZ FLIII stereoscope. Regenerated plantlets were then placed on rooting medium (30 g.L-1 sucrose, 8 g.L-1 agar) supplemented with 25 mg.L-1 kanamycin for one month before GFP expression in the leaves was analyzed by confocal microscopy as described above, but with a 40x objective.
Total DNA was extracted from cultures of R. etli harboring p42a, p42a virGmut, or p42a virE2mut [21] and PCR-amplified for 32 cycles using the primer pairs 5’ATGAAAGGTGAACGGTTGAAACAC3’/5’CCGGAATTCTCATTGTTTGCCTCCCTGCTGC3’ specific for virG (RHE_PA00053) and gusA, 5’ATGGATCCGAAAAGCGAAGACAAT3’/5’CCGGAATTCTCATTGTTTGCCTCCCTGCTGC3’ specific for virE2 (RHE_PA00061) and gusA, or 5’CTCCTGCGTGTCCTGATTGGC3’/5’AGCGGCGCGACGAACGTGAC3’ specific for a 320-bp segment of the R. etli chromosome between positions 109,451 and 109,770. Before proceeding with this analysis, we determined the orientation of the mutagenic gusA insertion in the virG and virE2 genes. We showed that, with forward primers corresponding to the 5’-ends of virG and virE2, a PCR product was observed only with the reverse primer corresponding to the 3’-end of gusA, but not with the primer corresponding to its 5’-end (S1 Fig), which reflects the sense orientation of gusA both within virG and virE2.
Total genomic DNA of wild type and transgenic tobacco plants was purified using the DNeasy plant DNA extraction kit (Qiagen) according to the manufacturer’s instructions. The purified DNA (10 μg) was digested with EcoR1 (New England Biolab) overnight. The digested DNA was resolved on a 1.0% agarose gel for 6 hours at 60 V, and DNA was transferred onto a nylon charged membrane with alkali transfer buffer [41]. For the T-DNA-specific probe, we used a 300-bp segment of the nopaline synthase (nos) promoter of the T-DNA region of pBin19-RCS1-GFP amplified using the primer pair 5’CAATATATCCTGTCAAACACTGATAG3’/5’GAAATATTTGCTAGCTGATAGTGAC3’; this probe fragment did not contain recognition sites for EcoRI. For the p42a-specific probe, we used a 240-bp segment of the virB5 gene amplified using the primer pair 5’5’ATGCATGAGCTCATGAAGATGTCGAGACTAGTTAC3’/5’AAAGGATCCCCTCGTGGCGGGATACTGG3’. For the R. etli genomic probe, we used a 320-bp segment of the chromosome between positions 109,451 and 109,770 amplified using the primer pair 5’CTCCTGCGTGTCCTGATTGGC3’/5’AGCGGCGCGACGAACGTGAC3’. For Southern blot analysis of transgenic plants (Fig 5A), agarose gel electrophoresis, blotting, and detection were performed at Lofstrand Labs Ltd. (Gaithersburg, MD) using a 32P labeled the T-DNA-specific probe (3.26 x 106 dpm/ml of hybridization buffer in a total volume of 50 ml). The hybridization was carried out for 3 days at 68°C; after washes, the membrane was autoradiographed for 17 hours with an intensifier screen at -80°C. For control experiments (Fig 5B), biotinylated probes were prepared using the biotin decalabel DNA labeling kit (Thermo Scientific); hybridization and detection were performed using the Phototope Star kit (NEB) according to the manufacturer’s instructions. Based on the 4.5 Gb size of the complex allotetraploid genome of N. tabacum [42], ca. 4 kb size of the T-DNA region of pBin19-RCS1-GFP, the DNA size-to-mass conversion ratio of 978 Mb = 1 pg (http://ebook2.worldlibrary.net/articles/C-value), and at least one T-DNA insertion per genome, we estimated that 10 μg of total transgenic plant DNA would contain ca. 9 pg of T-DNA, which is well within the detection range of the classical Southern blot analysis [43]. For comparable controls, we utilized 100 pg of purified pBin19-RCS1-GFP, 50 pg of p42a DNA, and 1 ng of R. etli chromosomal DNA.
|
10.1371/journal.pgen.1003614 | FGF Signalling Regulates Chromatin Organisation during Neural Differentiation via Mechanisms that Can Be Uncoupled from Transcription | Changes in higher order chromatin organisation have been linked to transcriptional regulation; however, little is known about how such organisation alters during embryonic development or how it is regulated by extrinsic signals. Here we analyse changes in chromatin organisation as neural differentiation progresses, exploiting the clear spatial separation of the temporal events of differentiation along the elongating body axis of the mouse embryo. Combining fluorescence in situ hybridisation with super-resolution structured illumination microscopy, we show that chromatin around key differentiation gene loci Pax6 and Irx3 undergoes both decompaction and displacement towards the nuclear centre coincident with transcriptional onset. Conversely, down-regulation of Fgf8 as neural differentiation commences correlates with a more peripheral nuclear position of this locus. During normal neural differentiation, fibroblast growth factor (FGF) signalling is repressed by retinoic acid, and this vitamin A derivative is further required for transcription of neural genes. We show here that exposure to retinoic acid or inhibition of FGF signalling promotes precocious decompaction and central nuclear positioning of differentiation gene loci. Using the Raldh2 mutant as a model for retinoid deficiency, we further find that such changes in higher order chromatin organisation are dependent on retinoid signalling. In this retinoid deficient condition, FGF signalling persists ectopically in the elongating body, and importantly, we find that inhibiting FGF receptor (FGFR) signalling in Raldh2−/− embryos does not rescue differentiation gene transcription, but does elicit both chromatin decompaction and nuclear position change. These findings demonstrate that regulation of higher order chromatin organisation during differentiation in the embryo can be uncoupled from the machinery that promotes transcription and, for the first time, identify FGF as an extrinsic signal that can direct chromatin compaction and nuclear organisation of gene loci.
| Changes in the position of genes within the nucleus and in their local organisation frequently correlate with whether or not genes are turned on. However, little is known about how such nuclear organisation is controlled and whether this can be separated from the mechanisms that promote transcription. We show here that central nuclear position and chromatin de-compaction correlate with onset of expression at key neural differentiation gene loci in the mouse embryo. Conversely, the locus of a gene that is down-regulated as neural differentiation commences exhibits a shift towards the nuclear periphery as this takes place. Importantly, we show that signalling through the fibroblast growth factor (FGF) pathway regulates changes at this level of nuclear organisation. FGF represses differentiation gene transcription and keeps differentiation gene loci compact and at the nuclear periphery. By blocking FGF signalling in a retinoid deficient embryo in which differentiation genes are not expressed, we further show that control of nuclear organisation by FGF is not just a consequence of gene transcription. These findings are the first to demonstrate that such higher order nuclear organisation is regulated in the developing embryo, that this takes place downstream of FGF signaling, and can be uncoupled from the machinery of gene transcription.
| Differentiation is directed by extrinsic signals that regulate expression of transcription factors, which determine cell fates. A further critical level of regulation is provided by so-called higher order chromatin organisation, which includes changes in local chromatin compaction and nuclear position of gene loci. Such changes have been documented in in vitro differentiation assays, but this level of organisation has not been analysed as extensively during embryonic development and the role of signalling pathways in modulating chromatin and nuclear organisation in the developing embryo remains unexplored.
During vertebrate embryonic development, induction of the future brain is followed by the progressive generation of neural tissue as the body axis elongates and this provides a unique opportunity to investigate steps leading to the onset of neural differentiation. New neural tissue arises from the stem zone/caudal lateral epiblast (adjacent to the primitive streak), which includes resident axial stem cells [1], [2] (Figure 1A). As cells leave this regressing region they either ingress to form paraxial mesoderm or remain in the epiblast and commence neural differentiation. Stem zone cells are highly proliferative and are maintained by FGF and Wnt signalling [3], [4]. This is attenuated by retinoid signals synthesised in the forming somites [3], [5], [6] (Figure 1A). Retinoic acid (RA) promotes neural differentiation in at least two steps; first repressing FGF/Wnt signalling and then promoting expression, in the forming neural tube, of key genes characteristic of neural progenitors, such as Sox1, Sox3 and Pax6 [3], [7]–[9]. Importantly, FGF signalling also counteracts retinoid signalling, repressing expression of Raldh2 which encodes retinaldehyde dehydrogenase 2 - an RA synthesising enzyme - in the presomitic mesoderm and RA receptor beta (RARb) in the forming neuroepithelium as well as promoting expression of Cyp26a1, encoding a RA catabolising enzyme reviewed in [1] (summarised in Figure 1A).
The underlying molecular mechanisms through which this opposing signalling switch controls differentiation onset in the embryonic body axis are not well understood, but might involve changes in gene expression determined by altered chromatin structure around target gene loci. One way in which chromatin compaction is locally regulated is by the action of polycomb repression complexes (PRC); and there is some evidence implicating FGF signalling in the regulation of polycomb complex component genes. Polycomb group (PcG) proteins are key regulators of cell growth and differentiation genes [10]–[12] and are found in two broad classes of complex; PRC2, which mediates the histone modification H3K27me3 associated with transcriptional repression through the activity of the Ezh1/2 histone methyltransferase and PRC1, which mediates local chromatin compaction [13]. In the zebrafish, the epiboly/tailbud phenotype of Ph2α morphants (homologue of the PRC1 component polyhomeotic) is similar to that of Fgf8 morphants, and Ph2α acts downstream of FGF signalling, which is necessary, although not sufficient for Ph2α expression [14]. Mice mutant for Fgf8 or for PcG genes (Eed, Ezh2 or Ring1b) also share a common gastrulation failure phenotype, with some reported proliferation defects [15]–[19], suggesting conservation of a relationship between FGF signalling and polycomb function in the early embryo.
Retinoic acid can signal directly to chromatin via liganded retinoic acid receptor – retinoid X receptor (RAR-RXRs) heterodimers and their sequence specific binding to retinoic acid response elements (RAREs) and this is known to attenuate binding of PRC2 components and to decrease H3K27me3 enrichment at these sites [20]–[22]. These observations suggest that in some contexts FGF may promote, while retinoid signalling represses, the action of polycomb complexes. Furthermore, as activation of polycomb target loci, such as the Hox gene clusters, is accompanied by visible unfolding of the compact state [23], [24], such signals might alter chromatin compaction at differentiation gene loci. Importantly, changes in chromatin compaction and local organisation are not simply a consequence of transcription; experimental translocation of a 3′ Hoxb1 transgene to the 5′ end of the Hoxd cluster elicited such chromatin changes in a cellular context in which Hoxb1 is not transcribed [25]. This phenomenon shows that alteration of chromatin organisation can prefigure gene transcription.
A further important manifestation of higher order chromatin organisation that frequently correlates with transcription is the position of a locus with respect to the nuclear periphery, which can be a repressive environment. Although recent studies have shown that artificial tethering to the nuclear periphery need not necessarily lead to gene silencing [26]–[28], many loci do exhibit a change in distance to the nuclear edge, and association with the nuclear lamina [29] which correlates with their potential for transcription. As extrinsic signals orchestrate development by directing gene transcription, it is likely that this involves regulation of such higher order organisation, however, it is not known whether particular signalling pathways direct such mechanisms nor whether they can elicit changes in chromatin organisation independently of transcriptional regulation.
To assess changes in higher-order chromatin organisation during the progressive generation of neural tissue in the elongating body axis of the mouse embryo, we used fluorescence in situ hybridisation (FISH) combined with super-resolution structured illumination microscopy (SIM). We analysed changes in higher-order chromatin organisation at the loci of exemplar neural progenitor genes Pax6 and Irx3 as differentiation takes place and compared this with the Fgf8 locus, which is transcriptionally downregulated as cells leave the stem zone (Figures 1B–B‴ 1C–C‴). As retinoid signalling is required for transcription of differentiation genes (including that of Pax6) we analysed chromatin organisation around loci in the Raldh2−/− mutant embryo, which is unable to synthesize RA in the elongating embryonic body axis [30]. In retinoid deficient embryos caudal FGF signalling expands rostrally from the stem zone [3], [6] and by blocking FGFR signalling in Raldh2 mutants we dissected the consequences of FGF loss in a context in which many differentiation genes fail to be transcribed. Our data demonstrate, for the first time, that FGF signalling acts upstream of mechanisms that direct higher-order chromatin organisation around differentiation gene loci and further reveal that such mechanisms can be uncoupled from the machinery that mediates transcription of such genes.
To determine if the onset of Pax6 transcription in the E8.5 mouse embryo involves a change in chromatin compaction, fosmid probes separated by 65 kb and specific for sequences flanking Pax6 (Table S1) were used for 3D FISH on wildtype CD1 mouse embryos (Figure 2A). Images were captured using SIM and inter-probe distances, (d) in µm, were measured in transverse sections of the stem zone and preneural tube (which lack Pax6 transcription) and in the neural tube, where Pax6 is now transcribed (excluding Pax6 negative cells in dorsal and ventral most positions) (Figures 1B–B′, 2A, B). Fosmids were also used to measure changes around a control locus, alpha-globin (Hba-a1), which is not transcribed in the embryo at this stage [31] (Figures 2C,D). Chromatin compaction was assessed by a comparison of d2 values for each data set, as this is the value that scales linearly with genomic separation [32] and that has been used previously to identify differences in chromatin compaction between cells at different stages of differentiation [24] and between wildtype and mutant cells [13] (Figure 2E) (see Methods for data set collection). There was no statistical difference between d2 values for stem zone and preneural tube (p>0.05), but a clear increase in inter-probe distances across Pax6 in neural tube in comparison with measurements in either stem zone or preneural tube nuclei (p<0.05, Figures 2B,E, Tables S2, S3). Additionally, in recently formed somites, which lie adjacent to the neural tube and which do not and will not express Pax6, chromatin across the Pax6 locus is as compact as it is in the stem zone and preneural tube, and significantly more compact than in neural tube (Figures 2B, E). In contrast, inter-probe distances around a control gene locus (alpha-globin, Hba-a1) were not significantly different between stem zone, neural tube and somite data sets (p>0.05, Figures 2D, F). This controls for any overall change in chromatin condensation at the onset of differentiation. These data therefore indicate that chromatin is compact around the Pax6 locus in cells that do not and will not express this gene (somites) and those that will later come to express it (stem zone and preneural tube), and that it specifically decompacts coincident with onset of Pax6 transcription in the neural tube. Analysis of genome-wide histone modification data sets in mouse ES cells and derived neural progenitors [11] further reveals that the Pax6 locus is subject to H3K27me3 enrichment in ES cells and is relieved of this mark upon neural differentiation, suggesting that this locus is a target of polycomb mediated repression (Figure S1).
We further assessed chromatin organisation across the locus of the gene Fgf8, which is expressed in the stem zone and downregulated as neural differentiation commences. Fgf8 is marked by H3K27me3 and H3K4me3 in mES cells and upon neural differentiation H3K4me3 is lost and H3K27me3 retained [11] (Figure S1). Fgf8 is a smaller gene than Pax6 and FISH signals from fosmids flanking this locus (100 kb separation) were barely resolved in any tissue assessed (Figure S2). These findings suggest that polycomb group proteins regulate Fgf8 expression, but in a manner that does not involve visibly detectable chromatin compaction. Neighbouring genes Npm3 and Mgea5 show similar patterns of gene expression as Fgf8 (Figure S3), but not of histone modifications (Figure S1). This suggests that PRC-mediated chromatin compaction around the Fgf8 locus may be too subtle, or masked by the chromatin environments of neighbouring genes, to be detected by FISH.
To investigate the potential relationship between the position of a gene within the nucleus and its transcriptional activity along the embryonic body axis, we analysed the proximity of FISH signals for Pax6 and or Fgf8 (Figure 3A) to the nuclear periphery as defined by DAPI staining. The Pax6 locus is closer to the nuclear periphery in the stem zone than in the neural tube (p<0.05; Figures 3B,B′), whereas the converse is the case for Fgf8 (Figures 3C,C′). The relative nuclear position of the control Hba-a1 locus was the same in the stem zone and neural tube (Figures 3D, D′). These data show that nuclear position correlates well with transcription of Pax6 and Fgf8 in the normal embryo.
Pax6 is not expressed in the neural tube of mouse embryos lacking the RA synthesising enzyme retinaldehyde dehydrogenase 2 (Raldh2−/−) [9], [30] (Figures 4A, B). To determine whether this is also accompanied by failure to undergo changes in higher order chromatin organisation, FISH with probe pairs across the Pax6 locus was carried out on E8.5 Raldh2−/− embryos. There was no difference in chromatin compaction (d2) between stem zone and neural tube (p>0.05; Figure 5C) indicating that chromatin decompaction, normally observed across the Pax6 locus in the wildtype neural tube (Figures 2A, B), does not take place in this retinoid deficient condition. Indeed, the distribution of inter-probe distances in Raldh2−/− neural tube nuclei was similar to that found in stem zone of wildtype mice (p>0.05; Figure 4C). The Pax6 locus also remained compact in somites of wildtype and mutant animals (Figures 4 C, D, E).
The absence of retinoid signalling also resulted in a failure of Pax6 to reposition away from the nuclear periphery in the neural tube compared to stem zone (Figure 4F). Moreover, Pax6 is more peripherally located in the Raldh2−/− neural tube than in the wildtype neural tube (p<0.05; Figures 4D, E and F). These data show that, for the Pax6 locus, neither chromatin decompaction nor a shift away from the nuclear periphery take place in the retinoid deficient neural tube in which Pax6 is not transcribed.
To determine whether exposure to retinoic acid leads to decompaction and a more central nuclear position of the Pax6 locus we treated explanted caudal regions with retinoic acid or vehicle DMSO control for 10 h (Figure 5A). Explants were then processed either for in situ hybridisation to monitor Pax6 transcription or for FISH to assess local chromatin organisation (Figure 5B). This confirmed that retinoic acid induces Pax6 expression (Figures 5B, B′) and demonstrated that this correlates with the decompaction and more central nuclear location of this locus (Figures 5C, D; p<0.05 and p<0.05, respectively).
FGF signalling ectopically persists in the preneural tube of retinoid deficient quail embryos [3] and in the neural tube of Raldh2−/− mouse embryos [5], [6]. As FGF signalling represses onset of expression of neural differentiation genes, including Pax6, in the elongating body axis [3], [33], it is possible that failure to express Pax6 in the Raldh2 mutant is due to an excess of FGF signalling.
To determine whether FGF signalling represses differentiation onset via a mechanism that involves regulation of higher order chromatin organisation, FGF signalling was blocked with the FGFR inhibitor PD173074 [34]. Explanted whole E8 wildtype embryos were cultured in vitro exposed to either DMSO vehicle control or PD173074 for 7 h and then processed for FISH, or analysed for expression of the FGFR pathway target Sprouty2 (Spry2) [35] and for Pax6. The repression of Spry2 (DMSO n = 0/5; PD173074 n = 5/5, Figure S4A) confirmed the effective blocking of FGFR signalling (and see [36]). Inhibition of FGFR signalling in the elongating neural axis also leads to precocious onset of Pax6 expression, which is then detected more caudally in the preneural tube in the chick embryo [33]. Consistent with this, Pax6 transcripts were detected in the preneural tube of PD173074 treated mouse embryos (DMSO n = 0/4; PD173074 n = 3/4, Figures 6A, B).
Analysis of chromatin compaction across the Pax6 locus by FISH (Figure 6C) revealed that, unlike the situation in untreated (Figure 2) and control (DMSO treated) embryos where Pax6 chromatin was more compact (smaller inter-probe distances) in stem zone and somites than in neural tube, this difference was abolished in PD173074-treated embryos. In these conditions the chromatin across Pax6 appears to decompact in stem zone and the somites to the level normally seen in the neural tube (p<0.05; Figures 6C, D, E). This indicates that FGFR signalling normally promotes chromatin compaction around Pax6 in caudal regions and somites (see Discussion). Blocking FGFR signalling also promoted a shift in Pax6 localisation towards the nuclear centre in stem zone and somitic nuclei in comparison with DMSO control (p<0.05 and p<0.05 respectively; Figures 5D, E, and F and S7A). No significant change in compaction or nuclear position in control and PD173074 treated embryos was seen at the control Hba-a1 locus, indicating that changes in chromatin organisation around the Pax6 locus do not reflect a general consequence of FGFR inhibition (Figures S4B, C). Together these data indicate that FGF signalling acts upstream of mechanisms that regulate chromatin compaction and nuclear position at Pax6.
To extend this analysis we assessed chromatin organisation around the locus of an additional neural progenitor marker gene, Irx3. Like Pax6, onset of Irx3 transcription takes place in the neural tube of the elongating body and is initially broadly expressed across the dorso-ventral axis [37]. Blocking FGFR signalling led to a caudal expansion of the Irx3 expression domain (n = 0/5 DMSO treated and n = 4/6 PD173074 treated embryos) (Figures 7A, B). Also like Pax6, the Irx3 locus is a PRC target in ES cells [11] (Figure S1). Using fosmids flanking Irx3 (Figure 7C) FISH analysis confirmed that this region of chromatin decompacts and relocates towards the nuclear centre in the neural tube coincident with its transcription (Figures 7D, E). Furthermore, blocking FGFR signalling led to decompaction and a more central nuclear position of the Irx3 locus in stem zone nuclei and also somites (Figures 7D, E, S7B). These data demonstrate that FGF signalling consistently acts upstream of chromatin re-organisation at differentiation gene loci.
To determine whether excess of FGF signalling is responsible for the lack of Pax6 expression in the absence of retinoid signalling, Raldh2 mutant embryos, and their wild type littermates produced from heterozygous Raldh2+/− crosses, were cultured with PD173074 or DMSO. Strikingly, blocking FGFR signalling did not rescue Pax6 expression in the neural tube of Raldh2−/− embryos (Pax6 mRNA was detected in neural tube of 0/4 mutant embryos and 0/4 PD173074 treated mutant embryos; Figures 8A, B). Attenuation of FGFR signalling in this condition is therefore not sufficient for onset of Pax6 expression and is consistent with a further requirement for retinoid signalling to promote neural differentiation [3].
This finding does, however, raise the possibility, that blocking FGFR signalling in retinoid deficient conditions still promotes initial steps in the differentiation process upstream of Pax6 transcription and this perhaps includes chromatin re-organisation. Indeed, FISH revealed that the Pax6 region decompacts in the stem zone and in the neural tube of PD173074 treated Raldh2−/− embryos compared to the control DMSO-treated mutant embryos (p<0.05 for both comparisons; Figures 8C,D, E). Blocking FGFR signalling in wildtype or in Raldh2 mutant embryos also decompacts chromatin in somites, despite the absence of Pax6 expression in this tissue (p<0.05 for both comparisons; Figures 8C,D, E) (see below). Similarly, blocking FGFR signalling in this context also induced a shift towards the nuclear centre of the Pax6 locus in both stem zone and neural tube; and this relative position is similar to that seen in the neural tube of wildtype or DMSO-treated embryos (p>0.05, Figure 8F). In this context, FGFR signalling therefore acts upstream of mechanisms that direct both local chromatin compaction and nuclear position and that can be uncoupled from the activity of retinoid mediated transcription factor complexes that are required to promote expression of neural differentiation genes such as Pax6.
Although we could not use FISH to measure chromatin compaction at the Fgf8 locus, this approach can be used to assess nuclear position. Consistent with rostral expansion of Fgf8 transcription in such mutants [6] (Figures 9A, A″), the Fgf8 locus fails to locate towards the nuclear periphery in the Raldh2 −/− neural tube, (p<0.05 in comparison with wildtype, Figures 9B, S5). Here, nuclear position therefore correlates with changing Fgf8 expression and these findings indicate that retinoid signalling is upstream of mechanism(s) that directs nuclear position of the Fgf8 locus.
It is further possible that the location of the Fgf8 locus is influenced by FGF signalling itself, as transcription of Fgf genes can be maintained by positive auto-regulatory feedback loops e.g. [38], [39]. To address this possibility we blocked FGFR signalling in wildtype and Raldh2−/− mutants (Figures 9A″, A‴, B′, B″). In both conditions this led to a more peripheral localisation of the Fgf8 locus in the stem zone (where this gene is expressed) in comparison with wildtype and Raldh2−/− DMSO controls. Strikingly, in the Raldh2 −/− mutant neural tube blocking FGFR signalling also rescued the failure to shift to the nuclear periphery observed in untreated mutants (Figures 9B–B″, S5). Fgf8 transcripts are still detected in PD173074 exposed embryos (Figures 9A″, A‴′) and this may reflect the known stability of Fgf8 mRNA [40], (although some intronic Fgf8 transcripts were detected in the stem zone of PD173074 treated embryos by whole mount in situ hybridisation (Figure S6), indicating that not all active Fgf8 transcription is lost). Overall, these findings demonstrate that FGFR signalling regulates nuclear position of the Fgf8 locus and that it is responsible for the persistent central location of this gene in the retinoid deficient neural tube.
This study reveals changes in higher-order chromatin organisation during neural differentiation in the mouse embryo and demonstrates, for the first time, that this level of organisation is regulated by key signalling pathways that direct differentiation (summarised in Figures 10A, B). We identify FGF signalling in the caudal region of the embryo as a factor acting upstream of mechanisms promoting chromatin compaction and peripheral nuclear position at neural differentiation gene loci, and further demonstrate that these large-scale changes can be uncoupled from transcription. We additionally show that FGF signalling promotes a central nuclear position for the Fgf8 locus. These data demonstrate that FGF can constrain differentiation via multiple mechanisms that control higher-order chromatin organisation.
We found that chromatin decompaction around Pax6 and Irx3 correlate with their transcriptional activation in the newly generated neural axis. The decompaction observed at these loci is reminiscent of that seen upon activation of Hox loci during embryonic development [23], [24], where chromatin compaction of the silent loci has been shown to be mediated by the PRC1 polycomb complex [13]. Decompaction around Pax6 and Irx3 in the neural tube nuclei corresponds to the first transcription of these genes during development and is consistent with polycomb regulation as indicated by the presence of H3K27me3 [11] and also the PRC1 protein Ring1b at the Pax6 locus in ES cells [41]. H3K27me3 is also associated with the Fgf8 locus in ES cells [11]. As we detect no increase in chromatin compaction when Fgf8 is transcriptionally downregulated, it is possible that in the embryo this does not involve polycomb mediated repression. However, Fgf8, but not neighbouring genes (NPM3 and Mgea5), has associated H3K27me3 in ES cells (Figure S2) and we cannot exclude that compaction local only to Fgf8 is not detected by our FISH assay.
We found that retinoid signalling is sufficient and necessary for chromatin decompaction around the Pax6 locus. Pax6 is not a direct target of the RAR/RXR transcriptional complex, but has an important cross-regulatory relationship with the proneural gene Ngn2 which is a direct target [8], [42]–[44]. However, RA signalling has been linked directly to the loss of PRC2 binding and H3K27me3 due to its promotion of MSK1/2 mediated phosphorylation of H3S28 in an embryonic carcinoma cell assay [45]. This modification is adjacent to H3K27 and correlates with the loss of PcG binding and loss of repression at a subset of PRC target genes. Nevertheless, RA is just one of several extrinsic signals that can promote MSK1/2 activity in vitro and mice null for both MSK1 and 2 are viable and fertile [46], suggesting that regulation of MSK1/2 is not a key endogenous mechanism for removal of polycomb-mediated repression in development. Instead, our data indicate that the requirement for retinoid signalling in the embryo is for removal of FGF signalling, which is in turn responsible for compaction around Pax6 and Irx3.
As well as undergoing chromatin decompaction, Pax6 and Irx3 also relocates to a more central position in the nucleus in the neural tube. For Pax6, this does not occur in the absence of retinoid signalling. Fgf8 shows the converse pattern of nuclear movements, but in the apparent absence of chromatin compaction changes, suggesting that nuclear position and local chromatin organisation are regulated by distinct molecular mechanisms. The relocation of Fgf8 toward the edge of the nucleus in the neural tube is blocked in Raldh2 mutants which correlates with its ectopic transcription [6]. Fgf8 has upstream RAREs indicating that it may be directly repressed by RA signalling [47], [48] and our finding here suggests that this may be linked to nuclear positioning mechanisms.
Importantly, although nuclear position of the Fgf8 locus correlated well with transcription of this gene, we found no cellular context (including wildtype, FGFR or retinoid signalling deficient conditions) in which chromatin decompaction (observed around Pax6 and Irx3) took place without a concomitant shift towards the nuclear centre. This suggests that de-compaction may be contingent upon a more central nuclear position.
Association with nuclear lamins correlates well with location of genomic regions to the nuclear periphery and so-called Lamin associated domains (LADs) generally have low levels of transcription. Genome-wide maps of Lamin B1 association in mouse ES cells show that many neural differentiation genes are located in LADs [29]. However, neither the Pax6 nor Irx3 loci, nor the whole Fgf8 region is a LAD in either ES cells or ES-derived neural progenitors (Figure S8) suggesting that at least in this in vitro context nuclear re-positioning of these loci is unlikely to be mediated by altered Lamin B1 association. However, while Lamin B genes appear not to be required for differentiation in ES cells, Lamin B null mice do exhibit profound neural defects [49], [50]. These include both neural progenitor proliferation and nuclear lamina integrity, and so a role for LAD mediated regulation of nuclear positioning of neural differentiation genes in the embryo cannot be ruled out [49], [50].
Our discovery that blocking FGFR signalling in the Raldh2 mutant, where many differentiation genes fail to be transcribed, restores chromatin decompaction and a more central nuclear position of the Pax6 locus suggests that RA acts first to inhibit FGFR signalling and that FGF is upstream of molecular mechanism(s) that direct higher order chromatin organisation at differentiation gene loci. This result is further supported in wildtype embryos in which FGFR signalling is blocked, as here Pax6 transcription extends caudally, being precociously expressed in the preneural tube, but we detect Pax6 decompaction and its more central nuclear position in stem zone cells, which have yet to express Pax6. Importantly, these experiments uncouple regulation of higher order chromatin organisation from gene expression itself and indicate that these large-scale chromatin changes take place as an initial step in the differentiation process. Although we assess these changes by detailed investigation around two exemplar neural differentiation genes, FGF signalling in this context represses expression of many such genes, including Sox1 and Sox3 (further regulators of the neural progenitor cell state) and prevents the onset of ventral patterning and neuron production in the newly generated spinal cord [3], [7]. It is therefore likely that FGF signalling (directly or indirectly) regulates a general mechanism(s) that determines chromatin organisation at such differentiation genes, many of which are known PRC2 targets in ES cells.
Intriguingly, blocking FGFR signalling also led to decompaction and a more central nuclear position of Pax6 and Irx3 loci in somites, where this gene is never expressed. These somites will have formed (1 somite every 2 hours) during the 7 h period of exposure to FGFR inhibitor, at the start of which these cells would have been experiencing FGF signalling in the presomitic mesoderm. The reorganisation of Pax6 and Irx3 loci in this context may thus reflect the finding that high level FGF signalling is required for mesoderm induction, while reduction elicits neural differentiation, as observed in Fgfr1 mutant mice, reviewed in [1]. Sudden loss of FGFR signalling in the early presomitic mesoderm might therefore elicit initial steps in neural differentiation.
Importantly, we show that blocking FGFR signalling does not lead to global chromatin reorganisation, as inter-probe distances and the fractional radius for control Hba-a1 locus and inter-probe distance for the region of the Fgf8 locus remain unchanged in all tissues examined. The Fgf8 locus does, however, alter its nuclear position in response to changes in FGFR signalling. When FGFR signalling is blocked in either wildtype or Raldh2 mutant embryos the Fgf8 locus remains close to the nuclear periphery in all tissues examined, including the stem zone where this gene is normally expressed and in the Raldh2 −/− mutant, where this inhibition of FGF signalling rescues the ectopic centralised location of Fgf8 locus. Although location at the nuclear periphery generally correlates with gene repression, we do detect some intronic Fgf8 transcripts in PD173074 treated embryos, indicating that in the timeframe of this experiment the peripheralisation of the Fgf8 locus does not simply correlate with loss of transcription. This may reflect an initial heterogeneous response to the loss of FGF signalling across the stem zone cell population, however, active transcription and peripheral locus position it is not incompatible with transcription [27], [28]. Overall then, FGF is upstream of mechanisms in the stem zone that lead to Pax6 and Irx3 compaction and peripheral location, and that promote a central position of Fgf8 within the nucleus. This shows that in this context FGF signalling influences multiple distinct molecular mechanisms, which regulate chromatin compaction and promote movement towards or away from the nuclear centre in a locus specific manner.
Attenuation of FGF signalling in human embryonic stem (hES) cells and mouse epiblast stem cells leads to loss of self-renewal [51]–[53]. Furthermore, as observed in the elongating embryonic neural axis [33] and in mouse ES cells that have experienced a period of endogenous FGF/Erk [7], inhibition of FGF/Erk signalling in hES cells induces rapid expression of Pax6 [53]. The attenuation of FGF signalling in stem cells of epiblast origin and in multipotent epiblast cells located in the stem zone/caudal lateral epiblast therefore serves as a common trigger for onset of differentiation and it is likely that conserved molecular mechanisms that include relief from polycomb mediated repression at differentiation genes underlie this initial step. Key future tasks are to determine how FGF signalling regulates local chromatin compaction and orchestrates nuclear positioning to constrain cell differentiation.
Wildtype CD1 embryos were collected at E8.5, dissected, fixed and processed for in situ hybridization (ISH) or for FISH as described below. Heterozygous Raldh2 mutant CD1 mice [30] were crossed to generate litters at E8-8.5 containing Raldh2−/−, Raldh2+/− and wildtype embryos. These were either dissected, genotyped as described previously [30], fixed and processed for ISH or FISH (see below), or E8 embryos within yolk sacs were collected in warmed (37°C) culture medium (rat serum, tyrode solution; 1∶1) containing control DMSO (0.5 µl/1 ml culture medium) or FGFR inhibitor PD173074 (Calbiochem) at 50 µM. Embryos were then cultured for 7 hours in a water-saturated roller-tube incubator at 37°C in 5% CO2, 20% O2. These were then dissected, genotyped, fixed and processed for FISH. For treatment with retinoic acid wild type CD1 E8-8.5 embryos were dissected to give explants pairs of the caudal embryo (Figure 5A) with one explant treated with 250 nM RA and the other DMSO vehicle control cultured in collagen as previously described [3] for 10 h. Explants were then fixed in 4% PFA and processed for ISH or FISH. For FISH analysis nuclei in sections taken from the central third of each explant were measured (5 explant pairs, >30 nuclei per explant measured) for inter-probe distance and fractional radius. Initial analyses compared differences between treated and untreated explants taken from the same embryo and these were all significantly different (Table S4). We therefore pooled all treated and all untreated explant data (Figures 5C, E).
All procedures using animals were performed in accordance with UK and French legislation and guidance on animal use in bioscience research.
Standard procedures were used to carry out in situ hybridisation in whole embryos to detect mRNAs for Pax6, Irx3, Fgf8, Spry2, Npm3 and Mgea5/OGA (primers used to clone Irx3, Npm3 and Mgea5/OGA can be found in Figure S3). A subset of these were embedded and cryo-sectioned to visualise mRNA localisation at a cellular level. Intronic Fgf8 was detected using a probe for the region between exons 5 and 6 of the mouse Fgf8 gene (a kind gift from Olivier Pourquie, [40]).
Mouse embryos stored in 100% MeOH were cleared in xylene, embedded in wax, sectioned at 7 microns and dried down on thin TESPA-coated 50×22 #1.5 coverslips (Scientific Laboratory Supplies Ltd) suitable for OMX microscopy. The protocol for FISH on mouse tissue sections was then adapted from [54]. Coverslips with sections were heated to 65°C (20 min), washed ×4 in xylene (10 min) and re-hydrated through an ethanol series to dH20. Coverslips were then microwaved for 20 min in 0.1 M citrate buffer, pH6.0, cooled in buffer (20 min) washed and stored in dH20 prior to pre-hybridisation steps and denaturation as previously described [54]. Fosmids pairs separated by inter-genomic distance of 60–120 kb were selected from the WIBR-1 Mouse Fosmid Library (Whitehead Institute/MIT Center for Genomic Research) and sequences confirmed by targeted PCR (Table S1, Figure S9). These were then labelled with either digoxigenin-11-dUTP or biotin-16-dUTP by nick transcription. Approximately 150 ng probe along with 15 µg mouse Cot1 DNA (Invitrogen) and 5 µg sonicated salmon sperm DNA (sssDNA) were used per coverslip, denatured and hybridised to coverslips [54]. After overnight incubation and washing, digoxigenin labelled probes were detected with anti-dig FITC (1∶20, Roche) and amplified with anti-sheep Alexa Fluor 488 (1∶100, Molecular Probes); biotin labelled probes with biotinylated anti-avidin (1∶100) and Alexa streptavidin 594 (1∶500, Molecular Probes). Nuclei were counterstained with DAPI and coverslips mounted onto slides with 25 µl of Slowfade Gold (Molecular Probes).
Samples were imaged on a Deltavision 3D OMX Structured Illumination Microscope (Applied Precision) using a protocol adapted after [55]. Regions of interest (ROIs) were identified using a Deltavision microscope, mapped using Softworx (Applied Precision) and acquired with a UPlanSApochromat 100× 1.4 NA oil-immersion objective lens (Olympus) and back-illuminated Cascade II 512×512 EMCCD camera (Photometrics) on the OMX version 2 system (Applied Precision) equipped with 405, 488, and 593 solid-state lasers. Samples were illuminated by a coherent scrambled laser light source that had passed through a diffraction grating to generate the structured illumination. Potential photo-bleaching was minimised by using lowest possible laser power and exposure times (50 and 250 ms). Raw images were processed and reconstructed using the Softworx structured illumination reconstruction tool (Applied Precision) [56]. The 405, 488 and 593 channels were then aligned in x and y, using predetermined shifts which were measured using a target lens and 100-nm Tetraspeck fluorescent beads (Invitrogen) in the Softworx alignment tool (Applied Precision).
For analysis of chromatin compaction and nuclear position, measurements were made in images of >50 nuclei per region in each of 3 different embryos per condition.
Stem zone was defined as epiblast cells adjacent and just caudal to the node (∼5 sections per embryo), preneural tube as neuroepithelium rostral to the node underlain by notochord and presomitic mesoderm, neural tube as neuroepithelium flanked by 2 or 3 most recently formed somites, and these adjacent somites were also used to represent somitic tissue. As nuclei in tissues are not as spherical as in cultured cells it was not possible to apply standard nuclear segmentation tools to define nuclear position. Instead sections in which a fosmid signal and nuclear edge were in sharp focus were used to measure the shortest distance from the probe centre to the periphery. The broadest distance across the nucleus was also measured as an indication of nuclear diameter and this was halved and data presented as a proportion the nuclear radius (fractional radius). Super resolution images were uploaded into an OMERO server (Open Microscopy Environment) and ROIs containing hybridisation signals for both dig and biotin-labelled probes were identified by manual inspection in OMERO-insight. ROIs typically extended over several z-sections to accommodate the whole volume of the signals. These ROIs were analysed by a custom script developed in MATLAB (Michael Porter, University of Dundee). This script first segments the objects defined by each probe from the background using Otsu thresholding and then calculates the xyz coordinates the centroid in each object. The centroids of these two objects and the distance between them, d (µm), were then output to a spread-sheet. The inter-probe distance was then squared because in interphase nuclei the mean physical distance squared between two points is linearly related to the known genomic distance [32]. Within each nucleus, the line measurement tool was used to determine the distance of the edge of the nucleus from the hybridisation signal of the biotin-labelled probe in sections in which it was in sharp focus and this was then averaged. The radius of the nucleus, also measured with the line measurement tool, was then divided by this distance. This gave the distance of the gene locus from the nuclear periphery as a proportion of nuclear size.
Box plots in figures show distribution of data. Top and bottom whiskers show highest and lowest data points respectively. Top and bottom lines of box represent 3rd and 1st inter-quartiles and the middle line represents the median. Non-parametric Mann-Whitney U test was used for analyses as data were not normally distributed. For comparison between explant pairs, a paired-sample Wilcoxon signed-rank test was used (Table S4).
|
10.1371/journal.ppat.1003655 | The Suf Iron-Sulfur Cluster Synthesis Pathway Is Required for Apicoplast Maintenance in Malaria Parasites | The apicoplast organelle of the malaria parasite Plasmodium falciparum contains metabolic pathways critical for liver-stage and blood-stage development. During the blood stages, parasites lacking an apicoplast can grow in the presence of isopentenyl pyrophosphate (IPP), demonstrating that isoprenoids are the only metabolites produced in the apicoplast which are needed outside of the organelle. Two of the isoprenoid biosynthesis enzymes are predicted to rely on iron-sulfur (FeS) cluster cofactors, however, little is known about FeS cluster synthesis in the parasite or the roles that FeS cluster proteins play in parasite biology. We investigated two putative FeS cluster synthesis pathways (Isc and Suf) focusing on the initial step of sulfur acquisition. In other eukaryotes, these proteins can be located in multiple subcellular compartments, raising the possibility of cross-talk between the pathways or redundant functions. In P. falciparum, SufS and its partner SufE were found exclusively the apicoplast and SufS was shown to have cysteine desulfurase activity in a complementation assay. IscS and its effector Isd11 were solely mitochondrial, suggesting that the Isc pathway cannot contribute to apicoplast FeS cluster synthesis. The Suf pathway was disrupted with a dominant negative mutant resulting in parasites that were only viable when supplemented with IPP. These parasites lacked the apicoplast organelle and its organellar genome – a phenotype not observed when isoprenoid biosynthesis was specifically inhibited with fosmidomycin. Taken together, these results demonstrate that the Suf pathway is essential for parasite survival and has a fundamental role in maintaining the apicoplast organelle in addition to any role in isoprenoid biosynthesis.
| Iron is essential for the survival of blood stage P. falciparum and is used primarily in the synthesis of iron-sulfur (FeS) cluster cofactors. We investigated the role that (FeS) clusters play in malaria parasites. We demonstrated that the synthesis of FeS clusters is partitioned between two organelles: the Isc pathway is mitochondrial while the Suf pathway is found exclusively in the apicoplast organelle. Attempts to interfere with the Suf pathway through a dominant negative approach were only successful when parasite cultures were supplemented with an isoprenoid product. This result demonstrates that isoprenoid biosynthesis depends on a functional Suf pathway. Unexpectedly, we also observed the complete loss of the apicoplast organelle when we disrupted the Suf pathway. This phenotype does not result from inhibition of isoprenoid biosynthesis; we treated parasites with high levels of the isoprenoid inhibitor fosmidomycin without any loss of the apicoplast organelle. These results demonstrate that the Suf pathway has a fundamental role in maintaining the apicoplast organelle in addition to any role in isoprenoid biosynthesis. Inhibition of the Suf pathway, which is not found in humans, will block the growth of malaria parasites.
| Iron-sulfur (FeS) clusters are ancient protein cofactors found in most organisms. These cofactors have a variety of roles including the transfer of single electrons, donation of sulfur atoms, initiation of free radical chemistry, oxygen sensing, and purely structural roles [1], [2]. FeS clusters are found in a variety of forms, but the most common are cubane 4Fe-4S, cuboidal 3Fe-4S, and binuclear 2Fe-2S clusters [3]. Proteins typically bind these clusters through cysteine residues, although other amino acids have been shown to be involved in coordinating the cofactor [1].
Proteins containing FeS clusters are typically sensitive to oxygen and the clusters rapidly degrade in extracellular environments. Thus, clusters are synthesized de novo by one of three known FeS biosynthetic pathways. The Nif pathway, the first synthesis pathway described, is primarily found in nitrogen-fixing bacteria [4]. The Isc and Suf pathways are the dominant FeS cluster synthesis pathways found in eukaryotes, and are also present in bacteria and archaea [5], [6]. In eukaryotes, the Isc pathway is mitochondrial [5] while the Suf pathway has thus far been found in species harboring a plastid organelle and has been localized to the chloroplast in Arabidopsis thaliana [7], [8]. The protozoan parasite Blastocystis, which lacks a plastid, contains components of the Suf pathway in the cytosol [9]. While the protein components of the Isc and Suf machinery are quite different, both pathways follow the same basic steps of sulfur mobilization, cluster assembly, and cluster transfer (Figure 1).
The Isc and Suf systems both depend on a cysteine desulfurase to mobilize sulfur from L-cysteine. The cysteine desulfurases of the eukaryotic Isc pathway (IscS) and of the Suf pathway (SufS) are only active when in complex with a partner protein (Figure 1). Isd11, a component of eukaryotic Isc pathways, is essential for mitochondrial FeS cluster synthesis in Saccharomyces cerevisiae and Trypanosoma brucei [10], [11], [12] but is not present in prokaryotes [13]. In the absence of Isd11, yeast IscS is prone to aggregation [10], [11]. Isd11 has a conserved LYK/R motif that is essential for its ability to activate IscS cysteine desulfurase activity [14]. SufS is activated by SufE, an accessory protein which is found in both prokaryotic and eukaryotic Suf pathways. Unlike Isd11, SufE forms a persulfide bond with the mobilized sulfur atom and acts to transfer the persulfide sulfur to the SufBCD assembly machinery [15]. Bacterial SufE has been shown to accelerate the cysteine desulfurase activity of SufS [16], [17]. In the presence of SufE, the Vmax of Escherichia coli SufS is increased eight fold and an additional rate enhancement of 32 fold is observed when the assembly machinery (SufBCD complex) is present to accept the sulfur from SufE [16]. In E. coli, SufE does not interact with the Isc cysteine desulfurase [16] while in A. thaliana SufE has been shown to localize to both mitochondria as well as chloroplasts and serves to activate both cysteine desulfurases [18].
Malaria parasites harbor a plastid organelle called the apicoplast that is thought to have arisen from two sequential endosymbiotic events [19]. The apicoplast harbors biochemical pathways of prokaryotic origin such as type II fatty acid synthesis (FASII), lipoate synthesis, tRNA modification, and 2-C-methyl-D-erythritol 4-phosphate (MEP) isoprenoid biosynthesis [20]. Enzymes in these pathways are predicted to require FeS cluster cofactors. In prokaryotes, lipoate synthase (LipA), the tRNA modification enzyme MiaB, as well as the MEP enzymes IspG and IspH contain 4Fe-4S clusters [21], [22], [23], [24], [25]. The activity of these FeS proteins is in turn thought to be dependent on the 2Fe-2S electron transfer protein ferredoxin (Fd) [26], [27]. In malaria parasites, only Fd and IspH have thus far been shown to contain FeS clusters [27].
The MEP isoprenoid biosynthesis pathway, the target of the antimalarial fosmidomycin, was recently shown to be essential for the survival of erythrocytic stage malaria parasites [28], [29]. Parasites cultured in the presence of the MEP pathway product IPP (isopentenyl pyrophosphate) were no longer sensitive to fosmidomycin. Additionally, supplementation with IPP allowed parasites to survive without the apicoplast organelle, demonstrating that isoprenoids are the only metabolites produced in the apicoplast that are needed outside the organelle [28]. FeS cluster proteins are likely required for the production of essential isoprenoids. However, the synthesis of FeS clusters themselves has not been well characterized in malaria parasites. Only the P. falciparum SufC protein, part of the SufBCD assembly complex, has been studied to date, and was demonstrated to be an active ATPase localized to the apicoplast [30].
In this report, we investigated two putative FeS cluster synthesis pathways (Isc and Suf), focusing on the initial step of sulfur acquisition. In P. falciparum, SufS and its partner SufE were found exclusively in the apicoplast and SufS was shown to have cysteine desulfurase activity in a complementation assay. IscS and its effector Isd11 were solely mitochondrial, suggesting that the Isc pathway does not contribute to apicoplast FeS cluster synthesis. We disrupted the Suf pathway using a dominant negative mutant of SufC and showed that these parasites only survive when cultured in the presence of IPP. Furthermore, these parasites lack the apicoplast organelle and its organellar genome – a phenotype not observed when isoprenoid biosynthesis was specifically inhibited with fosmidomycin. Taken together, these results demonstrate that the Suf pathway has a fundamental role in maintaining the apicoplast organelle in addition to any role in isoprenoid biosynthesis.
Bioinformatic studies suggest that the genomes of Plasmodium spp. encode both Isc and Suf proteins, including candidate cysteine desulfurases [20], [31], [32], [33], [34]. In most eukaryotes the cysteine desulfurases of the Isc and Suf pathways act in complex with the effector proteins Isd11 and SufE, respectively. SufE is essential for Suf FeS cluster synthesis in E. coli [15], [16], [35], but was originally thought to be absent from malaria parasites [32], [34]. More recent bioinformatic studies identified a potential sufE gene [30], [36] and a candidate isd11 gene [33], [37]. We used the PATS [38], PlasmoAP [39], and PlasMit [40] algorithms to predict the subcellular localization of P. falciparum Suf and Isc pathway proteins (Table 1). Most of the Suf pathway proteins were predicted to be apicoplast localized while the Isc proteins were predicted to be mitochondrial. In other systems, however, there is precedence for dual localization and crosstalk between components of Isc and Suf pathways. In Arabidopsis, SufE is dually localized to chloroplasts and mitochondria and activates both the Isc and Suf cysteine desulfurases [18]. In E. coli, SufE serves only the Suf pathway; however, the cluster transfer proteins are interchangeable between the pathways. SufA can rescue an IscA knockout, demonstrating that SufA can interact with the rest of the Isc pathway proteins; likewise, IscA can interact with the Suf machinery [35]. In S. cerevisiae, IscS has been localized to the mitochondria as well as the nucleus where it has a poorly defined but essential role [41]. In order to understand how the P. falciparum Suf and Isc pathways are partitioned in the parasite, we localized the IscS and SufS cysteine desulfurases and their effector proteins Isd11 and SufE in blood stage parasites.
We localized the SufS and SufE proteins in P. falciparum by expressing protein constructs fused to a C-terminal green fluorescent protein (GFP) tag. For SufS, the leader peptide (SufSlp) consisting of the first 59 amino acids was appended to GFP, since this region was predicted by the PATS algorithm [38] to contain the organellar targeting peptide. The mycobacteriophage Bxb1 integrase method was used to generate parasite strains with a single copy of SufSlp-GFP integrated into a specific recombination site in the P. falciparum genome [42], [43]. Live fluorescence microscopy demonstrated the presence of GFP fluorescence in an elongated organelle distinct from the parasite mitochondrion, which is typical of apicoplast morphology (Figure 2A). To verify localization to the apicoplast, we performed immunofluorescence analysis using antibodies against the apicoplast marker acyl carrier protein (ACP) (Figures 2B and S1). We also visualized the processing of this fusion protein upon import into the apicoplast by western blot using an antibody against GFP (Figure 2C). There was a small amount of unprocessed SufSlp-GFP while the majority of the fusion protein ran as a smaller processed species consistent with a cleavage event that occurs upon import into the apicoplast [44].
Full length SufE (SufEfl-GFP) could not be expressed in P. falciparum when driven by the strong calmodulin (CaM) promoter. Therefore, we used the lower strength ribosomal L2 protein (RL2) promoter [45]. SufEfl-GFP parasites displayed the same ramified pattern as SufS expressing parasites by live microscopy (Figure 3A). Detection by immunofluorescence demonstrated co-localization of SufEfl-GFP with the ACP apicoplast marker (Figures 3B and S2). As observed for SufS, SufEfl-GFP also appears to be processed, consistent with import into the apicoplast (Figure 3C). These results demonstrate that SufS and SufE are localized to the apicoplast of erythrocytic stage P. falciparum and that SufE does not appear to be dually localized as observed in A. thaliana [18].
In E. coli, the Isc and Suf pathways are partially redundant; deletions of essential elements of either pathway result in conditional lethality while deletion of both pathways is lethal [35]. E. coli deficient in the Suf pathway are more sensitive to iron starvation and oxidative stress than wild type or Isc deficient strains [35]. We used the iron starvation phenotype to test the cysteine desulfurase activity of SufS in E. coli. ΔsufS E. coli transformed with the mature (processed) form of SufS (pGEXT-SufS60) were able to grow in the presence of an iron chelator (2,2′-dipyridyl) while ΔsufS E. coli transformed with empty vector (pGEXT) were unable to grow (Figure 4). Thus, SufS can complement the loss of EcSufS, demonstrating that the parasite protein has cysteine desulfurase activity. This result also demonstrates that SufS is able to participate in an active E. coli Suf complex, even though mature SufS is only 30% identical to EcSufS.
We next wanted to know whether SufS is the only cysteine desulfurase that functions in the apicoplast. We localized IscS, the only other candidate cysteine desulfurase in malaria parasites, and its effector protein Isd11, using the same strategy described above for the Suf proteins. A full-length IscS construct (IscSfl) fused to GFP co-localized with mitotracker in live fluorescence microscopy (Figure 5A). Additionally, the 35 amino acid leader peptide of IscS (IscSlp, as predicted by PlasMit [40]) is sufficient to target GFP to the mitochondrion (Figure S3). We used the same integration strategy to localize a full-length construct of Isd11 (Isd11fl), which in yeast is necessary to activate IscS. Live fluorescence showed complete co-localization with mitotracker indicating the exclusive presence of Isd11 in the mitochondrion (Figure 5B). Thus, both IscS and Isd11 are mitochondrial and there is no evidence of additional nuclear localization of IscS as reported for S. cerevisiae IscS [41]. Taken together, these results suggest that SufS and SufE are solely responsible for sulfur acquisition for FeS synthesis in the apicoplast and we next attempted to determine whether this activity is essential in blood stage malaria parasites.
A conserved cysteine (at residue 51) in E. coli SufE is required for rapid transfer of sulfur from SufS to the SufBCD complex (Figure 1), and mutant SufE (C51S) binds to SufS and the SufBCD complex in a nonproductive manner [15], [16]. We attempted to interfere with iron-sulfur cluster synthesis and downstream metabolic pathways by generating an overexpression construct of P. falciparum SufE with the equivalent cysteine substituted with serine, SufE(C154S)-HA. We were able to select parasites expressing SufE(C154S)-HA (Figure S4A) in the presence of 200 µM IPP, however this parasite line was not dependent on IPP for growth (Figure S4B). Western blot analysis identified two protein bands for SufE(C154S)-HA, consistent with processing of the apicoplast leader peptide, and immunofluorescence showed that the protein co-localized with the apicoplast marker ACP (Figure S4C). Thus, SufE(C154S)-HA was expressed and properly trafficked to the apicoplast organelle, but ultimately failed to interfere with apicoplast metabolism enough to make these parasites dependent on IPP supplementation. Although SufE(C154S)-HA failed to act as a dominant negative mutant, this construct helps to confirm the apicoplast localization observed with SufE-GFP in Figure 3.
We designed another dominant negative mutant based on the recent finding that an active site lysine in E. coli SufC is required for ATPase activity and for accumulation of iron on the SufBCD assembly complex [46]. We generated a construct of P. falciparum SufC driven by the calmodulin promoter (CaM) with the active site lysine substituted with alanine, SufC(K140A)-HACaM. We were unable to select parasites expressing the SufC(K140A)-HACaM construct, suggesting that expression of this construct is toxic. To bypass this toxicity, we then transfected parasites with the SufC(K140A)-HACaM construct in the presence of 200 µM IPP and were able to select transgenic parasites. Western blot analysis showed a single band consistent with the expected molecular weight of our dominant negative construct (Figure 6A). However, unlike the apicoplast proteins shown in Figures 2C, 3C and S4A, there was no indication of apicoplast leader peptide processing with the SufC(K140A)-HACaM construct. Presumably, this construct was initially targeted to the apicoplast, but over time its expression led to apicoplast dysfunction and a loss of apicoplast leader peptide processing. Consistent with the effects of a dominant negative likely disrupting isoprenoid biosynthesis, these parasites were only able to grow when supplied with IPP (Figure 6B).
We next examined the condition of the apicoplast in dominant negative parasites. We generated a control parasite line expressing the apicoplast targeting peptide of the acyl carrier protein (ACP) fused to the red fluorescent protein mCherry. Parasites expressing ACP-mCherry (ACP-mCh) were cultured for six days with IPP in the presence or absence of 100 nM azithromycin (1× IC50). Azithromycin treatment is known to result in loss of the apicoplast organelle [28], [47]. Parasites treated with azithromycin (1× Az) were compared to untreated parasites and dominant negative parasites using a PCR assay. We amplified genes from the apicoplast genome (sufB), the mitochondrial genome (cox1) and the nuclear genome (sufS). All parasite lines maintained their mitochondrial and nuclear genomes. However, in contrast to the wild type parasites, the azithromycin-treated parasites and the SufC(K140A)-HACaM dominant negative parasites no longer contained sufB, indicating that both strains had lost the apicoplast genome (Figure 6C). The localization of apicoplast marker ACP in dominant negative parasites, and the SufC(K140A)-HA protein itself, were examined by immunofluorescence and found to be present in multiple foci spread throughout the cell rather than in a single apicoplast organelle (Figures 6D and S5). The same phenotype was observed when the apicoplast was chemically disrupted [28], confirming that the apicoplast had been similarly disrupted in the dominant negative parasites.
It is possible that high level expression of SufC(K140A)-HACaM could interfere with secretory pathway function, leading to general toxicity and loss of the apicoplast. To test this, we generated a parasite line expressing SufC(K140A)-HA from the weaker strength RL2 (ribosomal protein L2) promoter [45], SufC(K140A)-HARL2. In contrast to the CaM driven construct presented in Figure 6A, we were unable to detect the protein by western blot, even when the blot was loaded with ten-fold more SufC(K140A)-HARL2 parasite material. Despite the lower expression level, this mutant line was also dependent on continuous supplementation with IPP for growth (Figure 7A) and PCR analysis indicated that these parasites no longer contained the apicoplast sufB gene (Figure 7B). Furthermore, immunofluorescence analysis of this line showed localization of ACP to multiple puncta spread throughout the cell similar to that seen with the strong promoter (Figure 7C). Thus, even when expressed at a lower level, the SufC(K140A)-HARL2 dominant negative construct still causes the loss of the apicoplast organelle.
We next generated a parasite line expressing wild type SufC-HA driven by the CaM promoter, SufC-HACaM, to test whether overexpression of this construct would lead to loss of the apicoplast. Unlike the SufC(K140A)-HACaM line shown in Figure 6, the SufC-HACaM line is not dependent on IPP for growth, has not lost the sufB gene, and appears to contain a single intact apicoplast organelle (Figure 7). The SufC-HACaM construct is expressed in this parasite line and co-localizes with ACP in the apicoplast organelle (Figure S6). Notably, the SufC-HACaM construct is processed in a manner consistent with apicoplast import (Figure S6A) where as the SufC(K140A)-HACaM construct is not (Figure 6A). Similarly, endogenous ACP protein is processed in the SufC-HACaM line, but not in the dominant negative line (Figure S7). Taken together, these data demonstrate that the toxicity of the dominant negative construct is not due to the expression level or the presence of the HA tag, but rather depends solely on the K140A mutation.
Isoprenoids produced by the MEP pathway are the only metabolites produced in the apicoplast that are required outside of the organelle during the erythrocytic stages [28]. It is not known, however, whether the MEP pathway is required for maintenance of the apicoplast itself. To test this, we treated parasites with either azithromycin, to target all apicoplast functions, or fosmidomycin, which specifically targets the MEP pathway. We inhibited the MEP pathway in the presence of IPP by treating ACP-mCh parasites with 50 µM fosmidomycin (100× IC50), 100 µM fosmidomycin (200× IC50), 100 nM azithromycin (1× IC50), or no drug for six days. In subsequent growth experiments, only the azithromycin-treated parasites were dependent on IPP for growth (Figure 8A). Consistent with this growth phenotype, these parasites also lacked the apicoplast gene sufB (Figure 8B). These results indicate that treatment with azithromycin leads to loss of the apicoplast organelle while treatment with fosmidomycin does not. The ACP-mCherry produced by fosmidomycin-treated parasites and untreated control parasites is trafficked to a single branched organelle consistent with normal apicoplast morphology (Figure 8C). By contrast, parasites treated in parallel with azithromycin contain ACP-mCherry in multiple foci throughout the cell (Figure 8C). Thus, inhibition of the MEP pathway with fosmidomycin does not lead to loss of the apicoplast organelle. The dominant negative disruption of the Suf pathway results in similar molecular and cellular phenotypes as the general disruption of the apicoplast by azithromycin and not the specific inhibition of the MEP pathway by fosmidomycin (Figures 6 and 7). These results suggest that in addition to providing FeS clusters for isoprenoid biosynthesis enzymes, the Suf pathway also plays a role in the maintenance of the apicoplast organelle.
In the apicoplast of Plasmodium falciparum there are several pathways that are predicted to rely on FeS cluster cofactors (Figure 9), and one of these pathways is known to be essential for erythrocytic stage growth. An early step in MEP isoprenoid synthesis is the target for the antimalarial fosmidomycin [29] which is currently being evaluated in human trials as a partner drug with piperaquine. Recently, it was shown that supplementing parasites with isopentenyl pyrophosphate (IPP, one of the two final products of the MEP pathway) rescues sensitivity to antibiotics targeting apicoplast maintenance (e.g. chloramphenicol, clindamycin, doxycycline), demonstrating that isoprenoid synthesis is essential for blood stage parasite growth [28]. Antibiotic-treated parasites no longer contain an intact apicoplast or the organellar genome, however, these abnormalities should not affect the expression of the MEP pathway proteins. All of the enzymes in the MEP pathway are nuclear encoded and should still be produced under conditions in which the apicoplast is disrupted by antibiotic treatment. This is certainly true for the nuclear encoded apicoplast protein ACP, which is still produced regardless of whether the apicoplast is disrupted (Figures 6, 7 and S7). Unlike ACP, the enzymes that catalyze the penultimate and final steps of isoprenoid synthesis (IspG and IspH, respectively) should both contain FeS clusters [22], [27]. As described below, these clusters should not be available in parasites that lack an apicoplast.
SufB is one of the few non-housekeeping genes encoded in the apicoplast genome. In other systems, SufB plays an essential role in FeS cluster assembly and is the scaffold on which the clusters are built [35], [46]. When apicoplast maintenance is disrupted, SufB, and thereby FeS cluster synthesis, should be lost; this would then lead to disruption of the MEP pathway. Consistent with these expectations, we found that disruption of the Suf pathway with the SufC(K140A)-HACaM dominant negative mutant was toxic to blood stage malaria parasites. Parasites were only viable if supplemented with IPP, indicating that disruption of the Suf pathway ultimately leads to loss of the MEP isoprenoid biosynthesis pathway (Figure 6). Thus, the Suf pathway supports the MEP pathway and is essential for the survival of blood stage malaria parasites.
In addition to the MEP pathway, the apicoplast of malaria parasites harbors a type II fatty acid synthesis (FASII) pathway which is essential for liver stage development [48], [49]. The FASII pathway consumes acetyl-CoA [50] which is produced by the apicoplast-localized pyruvate dehydrogenase (PDH) enzyme complex [51]. Like the FASII pathway, a complete PDH complex (composed of four proteins) is essential during liver stage parasite development [52]. PDH is modified with the protein cofactor lipoate [53] which should be required for enzymatic activity. The synthesis of lipoate in the apicoplast is catalyzed by lipoate synthase (LipA), which we have shown contains 4Fe-4S clusters (Figure S8). These FeS clusters not only need to be synthesized, but they probably also need to be continuously repaired. One of the FeS clusters in E. coli LipA is destroyed every time lipoate is formed, making turnover of LipA dependent on replacing this FeS cluster [54]. Thus, FeS cluster synthesis in the apicoplast should ultimately be required for lipoate synthesis, PDH activity, and the function of the FASII pathway known to be critical for liver stage development in rodent and human malaria parasite species.
In organisms expressing both an Isc and a Suf pathway, such as E. coli, the Isc pathway acts as the default FeS synthesis pathway while the Suf pathway is expressed under conditions of prolonged oxidative stress and iron starvation [35]. It has been suggested that the Suf pathway is more efficient than the Isc pathway under conditions of oxidative stress [55]; this would be an attractive characteristic of the FeS cluster synthesis pathway expressed in oxygen producing compartments such as the plant chloroplast or the ancestral photosynthetic apicoplast [6], [56]. However, the modern apicoplast appears to maintain a reducing environment and is highly resistant to oxidative stress [57]. This protective environment may enhance the activity of enzymes sensitive to oxygen, such as LipA (Figure S8), but it is not clear whether the parasite Suf pathway retains the tolerance to oxidative stress conditions displayed by its orthologs in plants and bacteria.
FeS clusters are synthesized by ancient, highly conserved pathways, at least one of which is found in all organisms [58]. We have confirmed the presence of the Suf pathway in the P. falciparum apicoplast (Figures 2, 3, S1 and S2) and demonstrated the activity of the cysteine desulfurase SufS, the first enzyme in the pathway (Figure 4). In 2003, another group localized IscS as a test of a transfection method [59]. They fused GFP to what was at the time predicted to be the first 135 amino acids of IscS, however, the amino-terminus of the current gene model differs from the sequence used in that study. We repeated the localization using the current gene model which aligns more closely with eukaryotic IscS sequences. P. falciparum IscS and Isd11 both localized exclusively to the mitochondrion (Figures 5 and S3). The subcellular partitioning of the Isc and Suf pathways demonstrates that they function independently of each other, and are likely both essential for erythrocytic stage parasite growth. The same general pattern of organellar partitioning of the Isc and Suf pathways is observed in the only other plastid-containing organism in which both pathway components have been localized, A. thaliana [6]. P. falciparum appears to differ from A. thaliana, however, in that we observe SufE solely in the apicoplast while one of the Arabidopsis SufE homologs appears to be dually localized between the chloroplasts and the mitochondria and has been shown to activate mitochondrial AtIscS [18]. AtIsd11 is only 18% identical to Isd11 from P. falciparum and AtIscS lacks the extended amino terminus of IscS present in P. falciparum and S. cerevisiae. There may be functional differences between these IscS homologs that affect their ability to be stimulated by effector proteins.
FeS cluster modified proteins in the P. falciparum mitochondrion are involved in redox regulation, metabolism, and participate in the electron transport chain. Complex III (cytochrome bc1) is the target of the antimalarial atovaquone, which prevents binding of reduced ubiquinone and also blocks electron transfer from the Rieske type 2Fe-2S cluster, implying that the Isc pathway is essential for blood stage parasite growth [60]. This makes the Isc pathway an attractive drug target, however it is closely related to the host Isc pathway. Closer study of the Isc pathway found in parasites may identify exploitable differences between mitochondrial FeS cluster synthesis in the parasite and in the human host.
The Suf pathway is not found in humans, and the work presented here shows that it is required for the maintenance of the apicoplast organelle. If the Suf pathway was only needed to activate certain MEP enzymes, we would expect disruption of the Suf pathway to have similar effects as inhibition of the MEP pathway. This, however, was not the case. As shown in Figure 8, inhibition of the MEP pathway by the specific inhibitor fosmidomycin does not lead to dependence on IPP for growth, loss of the apicoplast gene sufB, or observable changes in organelle morphology. In contrast to fosmidomycin treatment, disruption of the Suf pathway with the dominant negative mutant SufC(K140A)-HACaM results in loss of the apicoplast organelle. Dominant negative parasites depend on IPP for growth, have lost the sufB gene, and no longer contain an intact apicoplast organelle (Figures 6).
One possible explanation for this broader phenotype is that high level expression of SufC(K140A)-HACaM interfered with secretory pathway function, leading to general toxicity. This seems unlikely, however, since these parasites still traffic ACP into punctate foci in the cell (Figures 6C and S5), consistent with the membrane-bound secretory vesicles observed by Yeh and coworkers [28]. To address this issue, we generated two additional parasite lines. The first expressed the same dominant negative mutant driven by the lower strength RL2 promoter. This parasite line, SufC(K140A)-HARL2, displayed the same loss of apicoplast phenotype, demonstrating the potency of the dominant negative SufC(K140A) mutation (Figure 7). We also generated a parasite line expressing a wild type construct of SufC driven by the strong calmodulin promoter. This SufC-HACaM construct differs from the toxic dominant negative SufC(K140A)-HACaM construct by a single amino acid, yet had none of the molecular and cellular phenotypes associated with loss of the apicoplast organelle (Figures 7 and S6). Thus, the K140A point mutation is solely responsible for disrupting apicoplast metabolism leading to loss of the organelle.
How does the dominant negative mutant interfere with apicoplast metabolism? SufC is known to bind to SufB [30] and presumably forms the SufBCD iron-sulfur cluster assembly complex observed in other organisms (Figure 1). The SufC(K140A) mutant was designed to form a nonproductive complex with endogenous SufB and SufD, thereby limiting the availability of these proteins for cluster assembly. The dominant negative mutant should decrease cluster synthesis, but it could also affect iron homeostasis in the apicoplast, a phenomenon that is difficult to study since organellar iron import and storage mechanisms are not known. We attempted to interfere with the Suf pathway at an earlier step (sulfur acquisition) with the SufE(C154S) mutant, but this construct did not have a dominant negative phenotype, even when overexpressed with the strong calmodulin promoter (Figure S4). It may be that sulfur acquisition is not the rate limiting step in the parasite Suf pathway or that the SufE mutant does not interact with other Suf proteins as observed in the E. coli system.
The most likely effect of the SufC(K140A) dominant negative mutant is inactivation of apicoplast FeS proteins. Known and predicted FeS proteins are shown in Figure 9, including four FeS enzymes (LipA, IspH, IspG and MiaB) and ferredoxin (Fd). Are any of these proteins likely to be required for apicoplast maintenance during blood stage parasite growth? As described above, LipA is responsible for lipoylating the PDH and ultimately supporting fatty acid biosynthesis in the apicoplast. Since components of the FASII pathway and subunits of the PDH complex (albeit not the lipoylated E2 subunit) have been successfully deleted in blood stage malaria parasites [48], [49], [52], LipA is presumed to be similarly dispensable and not required for apicoplast maintenance. Although IspH and IspG should be essential for isoprenoid biosynthesis in blood stage parasites, loss of these enzymes should have the same effect as inhibition with fosmidomycin. As shown in Figure 8, inhibition of isoprenoid biosynthesis does not result in loss of the apicoplast organelle.
This result also suggests that the final FeS enzyme, MiaB, is not required for apicoplast maintenance. MiaB presumably functions in conjunction with an upstream enzyme, MiaA, in the maturation of tRNAs. MiaA has not been studied in malaria parasites, but in most eukaryotes and bacteria this enzyme transfers isopentenyl groups to a specific adenosine base in the anticodon loop of certain tRNAs [61], [62]. MiaB is a methylthiolase that further modifies the isopentenyladenosine tRNA base with a CH3S group [63], [64]. If P. falciparum MiaB functions in an analogous way, then its activity depends on isoprenoid biosynthesis, a pathway that we have shown is not required for apicoplast maintenance. Importantly, MiaA enzymes use the MEP pathway product DMAPP (dimethylallyl pyrophosphate) as the source of isopentenyl groups and would not be able to use the IPP that we supply in our parasite culture conditions unless there is an IPP/DMAPP isomerase present. Thus, based on their predicted activities (these enzymes could have additional noncanonical activities), these four FeS enzymes do not appear to be good candidates to explain why disrupting FeS cluster synthesis leads to loss of the apicoplast organelle.
Among the predicted apicoplast FeS proteins in Figure 9, ferredoxin stands out as the most integral to apicoplast function. P. falciparum Fd contains a 2Fe-2S cluster and has been shown to act as an electron donor to IspH [27]. Other apicoplast pathways may also depend on Fd, since it is predicted to be the preferred electron transfer partner for the other apicoplast FeS enzymes (LipA, IspG and MiaB) and may be required to provide reducing equivalents during certain steps of FeS cluster biosynthesis [36], [65]. Because of its role in FeS synthesis, reduced Fd metalation could have an exaggerated effect by further limiting the production of its own FeS clusters. Even if Fd is required for FeS synthesis, it still does not provide an explanation for how the apicoplast is lost since the downstream FeS enzymes do not have obvious roles in apicoplast maintenance. Loss of the organelle may instead be linked to how redox balance is maintained in the apicoplast. Fd in conjunction with its associated reductase, ferredoxin-NADP+-reductase (FNR), is the only known redox system in the apicoplast [66]. Perturbation of the Fd/FNR system could lead to increased sensitivity to oxidative stress, as observed in other systems [67]. Since the apicoplast is known to be a highly reducing environment [57], failure of this protective system could lead to oxidative damage, particularly of the organellar DNA, and subsequent loss of the organelle. Regardless of the mechanism, it is clear that Suf pathway dysfunction results in a disruption of apicoplast maintenance. Since the enzymes which comprise the Suf pathway are distinct from anything found in the human host, they are attractive targets for inhibition. The Suf pathway appears to lie at the root of apicoplast metabolic function and inhibition of the pathway should block the growth of blood stage and liver stage malaria parasites.
The genes in this study were amplified from gDNA or cDNA prepared from blood stage P. falciparum Dd2 strain parasites and inserted into the pLN-GFP transfection plasmid described by Nkrumah and coworkers [42]. In some cases, the calmodulin (CaM) promoter of pLN-GFP was substituted with the weaker strength ribosomal L2 protein (RL2) promoter [45], and in other cases the GFP tag was removed or replaced with mCherry (mCh) or a hemagglutinin tag (HA).
The iscS gene (PF3D7_0727200) was amplified from gDNA with primers IscS.AvrII.F and IscS.fl.BsiWI.R and inserted into pRL2-GFP, generating plasmid pRL2-IscSfl-GFP (see Table S1 for primer sequences). Nucleotides encoding the 35 amino acid IscS leader peptide (IscSlp) were amplified from pRL2-IscSfl-GFP vector using primers IscS.AvrII.F and IscS.35.BsiWI.R and ligated into pLN-GFP to generate pLN-IscSlp-GFP. The in-frame intron in isd11 (PF3D7_1311000) was confirmed by amplifying this gene from cDNA with primers Isd11.AvrII.F and Isd11.BsiWI.R and inserting into pRL2-GFP, generating plasmid pRL2-Isd11fl-GFP. The sufS gene (PF3D7_0716600) was amplified from cDNA using the primers SufS.TOPO.F and SufS.TOPO.R and ligated into cloning vector pET100/D-TOPO (Invitrogen). Nucleotides encoding the leader peptide of SufS were amplified using the primers SufS.AvrII.F and SufS.59.BsiWI.R and ligated into pLN-GFP, generating plasmid pLN-SufSlp-GFP. Amplification of sufE (PF3D7_0206100) from cDNA confirmed the four exon gene model, but consistently resulted in a frame-shifted amplicon. Gene synthesis (GeneArt) was used to generate the sufE gene flanked by AvrII and BsiWI endonuclease sites which were used to subclone into the pRL2-GFP transfection plasmid generating pRL2-SufEfl-GFP.
A transfection vector was created to express mCherry red fluorescent protein in the apicoplast organelle. The gene encoding mCherry was amplified with primers mCh.BsiWI.F and mCh.AflII.R and inserted into the pLN-TP-ACP-GFP vector described by Gallagher et al. [68]. The resulting transfection vector, pLN-TP-ACP-mCh, encodes mCherry instead of GFP. Constucts SufE(C154S), SufC (PF3D7_1413500) and SufC(K140A) were synthezised (GeneArt) with flanking AvrII and BsiWI sites and inserted into a pLN plasmid modified to have a carboxy-terminal single HA tag, generating pLN-SufE(C154S)-HA, pLN-SufC-HA and pLN-SufC(K140A)-HA. The SufC(K140A)-HA coding region was digested from this plasmid with AvrII and BsiWI and inserted into pRL2 to generate pRL2-SufC(K140A)-HA.
P. falciparum transfections were performed using the Bxb1 mycobacteriophage integrase system in Dd2 strain parasites containing the attB recombination site [42] in combination with a red blood cell (RBC) preloading technique [43]. Infected red blood cells (iRBC) were first observed between 11 and 27 days after beginning selection with 2.5 µg/mL blasticidin. Insertion of the transgene at the attB site was confirmed by PCR using the primers P1, P2, P3, and P4 (Table S1) as described by Spalding et al. [43]. Genomic DNA from each integrated parasite line was purified and used to verify the transgene sequence with primers GFP.R or pLN.790.R and either RL2.F or CaM.F, as appropriate (Table S1).
Parasites were maintained in human red blood cell culture at 2% hematocrit using the general method described by Trager and Jensen [69]. Briefly, blood stage parasites were cultured in RPMI 1640 supplemented with 10% human serum, 28 mM NaCO3H, 25 mM HEPES, and 0.09 mM hypoxanthine. Cultures were gassed with 92% N2, 3% O2, 5% CO2 and incubated in sealed 75 cm2 flasks at 37°C. For the chemical bypass experiments, 0.5 ml or 1 ml parasite cultures were maintained in 24 or 48 well plates and supplemented during daily feedings with 200 µM isopentenyl pyrophosphate (Sigma).
Parasite cultures with a parasitemia between 2% and 15% were incubated for 30 minutes at 37°C with 12.5 nM mitotracker CMX-Ros (Invitrogen) and 1 µg/mL 4′, 6-diamidino-2-phenylindole (DAPI). Cells were washed three times for 5 minutes at 37°C with RPMI or PBS and then sealed on a slide for observation on a Nikon Eclipse 90i equipped with an automated z-stage. A series of images spanning 4 µm were acquired with 0.2 µm spacing and images were deconvolved with VOLOCITY software (PerkinElmer) to report a single combined z -stack image.
Parasites were fixed and permeabilized for immunofluorescence studies. Live parasites were mixed with 4% paraformaldehyde and 0.0075% glutaraldehyde in PBS and placed on poly-lysine coated glass slides for 30 minutes at room temperature. The slides were then incubated with 1% Triton X-100 in PBS for 10 minutes and then washed three times for 5 minutes with PBS. Sodium borohydride (0.1 g/L in PBS) was used to reduce any remaining unreacted aldehydes followed by three more 5 minutes washes in PBS. The slides were then blocked with 3% bovine serum albumin for an hour and then probed with the appropriate primary antibodies [1∶500 rabbit or rat αACP [57], 1∶50 Living Colors mouse αGFP JL-8 (Clontech), or 1∶50 rat αHA mAb 3F10 (Roche)]. Slides were washed three times for 5 minutes with PBS, and then incubated with the appropriate secondary antibodies [1∶3,000 goat αRabbit IgG Alexa Fluor 594, 1∶1,000 goat αMouse IgG Alexa Fluor 488 (Invitrogen), or 1∶1,000 goat αRat IgG Alexa Fluor 488 (Invitrogen)] for one hour at room temperature. The slides were washed three times for five minutes with PBS, then mounted with Prolong Gold antifade reagent with DAPI (Invitrogen).
Expression of SufS and SufE in transgenic parasites was verified by western blot. Host RBCs from 5 mL cultures at 5–15% parasitemia were permeabilized with 0.2% saponin in PBS for 5 min on ice and then washed repeatedly in PBS until the supernatant was clear. Purified parasites were then lysed in gel loading buffer and parasite proteins were resolved on a NuPage 4–12% Bis-Tris reducing gel (Invitrogen) and transferred onto nitrocellulose. The nitrocellulose membrane was blocked for at least one hour with 5% milk in PBS and probed overnight at 4°C with 1∶5,000 Living Colors mouse αGFP JL-8 (Clonetech) in 1% milk. The membrane was washed with PBS three times and probed with 1∶20,000 sheep αMouse IgG horseradish peroxidase (HRP) secondary antibody (GE Healthcare) for at least one hour at room temperature. After three additional washes, the blot was visualized with SuperSignal West Pico detection solution (Thermo Scientific) and exposed to film.
All constructs expressed in E. coli were cloned into the pGEXT vector which expresses the parasite proteins fused to a cleavable amino terminal glutathione-s-transferase (GST) tag [70]. Mature lipA (encoding residues 89 to 415 of PF3D7_1344600) was amplified from cDNA using Pfu polymerase and primers LipA.EcoRI.F and LipA.PstI.R (Table S1). This amplicon was digested with PstI and EcoRI and ligated into vector pMALcHT [71]. Primers LipA.BamHI.F and LipA.EcoRI.R (Table S1) were used to subclone LipA, generating plasmid pGEXT-LipA89. A construct of mature sufS (encoding residues 60 to 546 of PF3D7_0716600) was amplified from vector pET100/D-TOPO (described above) using primers SufS.BamHI.F and SufS.EcoRI.R (Table S1), generating plasmid pGEXT-SufS60.
E. coli containing a deletion of sufS (ΔsufS, Keio collection JW1670) were transformed with either empty vector, pGEXT, or pGEXT-SufS60. Each strain was grown overnight at 37°C in MinE medium as modified by Allary et al. [53]. The overnight culture was used to plate 1 µL of 1.0 OD600 on MinE agar plates containing 100 µM 2,2′-dipyridyl. The plates were incubated for 48 hrs at 30°C and inspected for bacterial growth.
BL21 Star (DE3) E. coli containing the pLysE plasmid were transformed with pGEXT-LipA89 construct produced above. In order to culture the protein in conditions of minimal oxygen, E. coli were grown in flat bottom flasks filled three quarters full with LB medium. When cells reached an OD600 of 0.6 they were induced with 0.4 mM IPTG for 10 hours at 20°C. Cells were harvested by centrifugation, flash frozen in liquid nitrogen, and stored under the liquid layer. The cell pellet was transferred to a Bactron IV (Shell Labs) anaerobic chamber flooded with 5% hydrogen, 5% carbon dioxide, and 90% nitrogen. A palladium catalyst was used to maintain <30 ppm oxygen. Cells were resuspended in anaerobic lysis buffer (20 mM Na/K phosphate [pH 7.5], 200 mM NaCl, 2 g/L lysozyme, and 1 mM phenylmethylsulfonyl fluoride [PMSF]) and incubated at room temperature until cell lysis was apparent. After lysis, 2.5 µg/mL DNase I was added and incubated for 30 minutes at room temperature. The lysate was transferred to an air tight container and centrifuged to separate the soluble and insoluble fractions. GST-LipA89 was purified using a 5 mL GST-Trap HP column (GE Healthcare) connected to a peristaltic pump in the anaerobic chamber.
Plasmids pLN-SufE(C154S)-HA, pLN-SufC(K140A)-HA, pRL2-SufC(K140A)-HA and pLN-SufC-HA were used to generate transgenic parasite lines. As described above, these parasite lines were maintained in the presence of 200 µM IPP. Protein expression was confirmed by western blot using the methods described above with 1∶1,000 rat αHA mAb 3F10 (Roche) and 1∶20,000 goat αRat IgG horseradish peroxidase (HRP) secondary antibody (GE Healthcare) secondary antibody. Growth assays were conducted in triplicate using 24 well culture plates and initiated at a parasitemia of 0.5%. Over a six day period, parasitemia was assessed by flow cytometry using a FACSCalibur cell sorting machine (Becton Dickinson). Samples of 10 µl from each well were incubated with 10 µl of 5 µM dihydroethidium for 15 minutes at 37°C in the dark. Results were analyzed by FlowJo software (Tree Star Inc., Ashland, OR). Whole cell PCR was used to amplify representative genes from the nuclear (sufS), apicoplast (sufB), and mitochondrial (cox1) genomes (Table S1). Phusion High-Fidelity DNA Polymerase (New England BioLabs) was used in accordance with the manufacturer's directions in 25 µL reactions containing 1 µL of parasite culture. The processing of endogenous ACP was visualized by western blot using the 4% formaldehyde 0.1% glutaraldehyde fixation conditions previously described [57] to prevent ACP from diffusing out of the blot membrane. The blot was probed with 1∶5,000 rabbit αACP [57] primary and 1∶3,500 donkey αRabbit IgG horseradish peroxidase (HRP) secondary antibody (GE Healthcare).
Parasites transfected with the pLN-TA-ACP-mCherry vector were supplemented with 200 µM IPP and treated with 100 nM azithromycin, 50 µM fosmidomycin, 100 µM fosmidomycin, or no drug for 6 days. All four ACP-mCherry (ACP-mCh) lines were then tested for IPP dependence and analyzed by live epifluorescence microscopy and whole cell PCR as descibed above.
|
10.1371/journal.pcbi.1002421 | Cytoskeletal Signaling: Is Memory Encoded in Microtubule Lattices by CaMKII Phosphorylation? | Memory is attributed to strengthened synaptic connections among particular brain neurons, yet synaptic membrane components are transient, whereas memories can endure. This suggests synaptic information is encoded and ‘hard-wired’ elsewhere, e.g. at molecular levels within the post-synaptic neuron. In long-term potentiation (LTP), a cellular and molecular model for memory, post-synaptic calcium ion (Ca2+) flux activates the hexagonal Ca2+-calmodulin dependent kinase II (CaMKII), a dodacameric holoenzyme containing 2 hexagonal sets of 6 kinase domains. Each kinase domain can either phosphorylate substrate proteins, or not (i.e. encoding one bit). Thus each set of extended CaMKII kinases can potentially encode synaptic Ca2+ information via phosphorylation as ordered arrays of binary ‘bits’. Candidate sites for CaMKII phosphorylation-encoded molecular memory include microtubules (MTs), cylindrical organelles whose surfaces represent a regular lattice with a pattern of hexagonal polymers of the protein tubulin. Using molecular mechanics modeling and electrostatic profiling, we find that spatial dimensions and geometry of the extended CaMKII kinase domains precisely match those of MT hexagonal lattices. This suggests sets of six CaMKII kinase domains phosphorylate hexagonal MT lattice neighborhoods collectively, e.g. conveying synaptic information as ordered arrays of six “bits”, and thus “bytes”, with 64 to 5,281 possible bit states per CaMKII-MT byte. Signaling and encoding in MTs and other cytoskeletal structures offer rapid, robust solid-state information processing which may reflect a general code for MT-based memory and information processing within neurons and other eukaryotic cells.
| Memory is understood as strengthened synaptic connections among neurons. Paradoxically components of synaptic membranes are relatively short-lived and frequently re-cycled while memories can last a lifetime. This suggests synaptic information is encoded at a deeper, finer-grained scale of molecular information within post-synaptic neurons. Long-term memory requires genetic expression, protein synthesis, and delivery of new synaptic components. How are these changes guided on the molecular level? The calcium-calmodulin dependent protein kinase II (CaMKII) has been heavily implicated in the strengthening of active neural connections. CaMKII interacts with various substrates including microtubules (MTs). MTs maintain cellular structure, and facilitate cellular cargo transport, effectively controlling neural architecture. Memory formation requires reorientation of this network. Could CaMKII-MT interactions be the molecular level encoding required to orchestrate neural plasticity? Using molecular modeling and electrostatic profiling, we show a precise matching between the spatial dimensions, geometry and electrostatics of CaMKII and MTs, and calculate the potential information capacity and bio-energetic parameters of such interactions. Results suggest signaling and encoding in MTs offers rapid, robust information processing with a large potential for memory storage, reflecting a general code for MT-based memory in neurons and other eukaryotic cells.
| The brain's ability to learn and store memory is understood in terms of changes in synaptic connections between neurons: ‘synaptic plasticity’ [1]. This is supported by the paradigm of ‘long-term potentiation’ (LTP) in which repetitive pre-synaptic stimulation increases post-synaptic sensitivity and strengthens synapses (e.g. the adage “neurons that fire together, wire together”). LTP is supported experimentally in vitro [2], [3], and may occur over many brain regions [4] as a common feature of excitatory synapses [5].
Below the level of the synapse, LTP involves complex gene expression, protein synthesis and recruitment of new receptors or even synapses [6]. Generally, however, synaptic plasticity is viewed in terms of changes in function, location and/or number of post-synaptic receptors and ion channels. However, synaptic receptors and channel proteins are transient, being synthesized and degraded in the protein lifecycle, and yet memories can last lifetimes. Therefore, information pertinent to memory must be stored elsewhere, yet remain able to regulate synaptic plasticity.
LTP involves the neurotransmitter glutamate binding to post-synaptic receptors, opening calcium channels to allow influx of calcium ions (Ca2+) to dendritic spines [7], shafts and neuronal cell body. Within dendritic spines, inflow of Ca2+ results in activation of multiple enzymes including protein kinase A [8], protein kinase C [9] and Ca2+-calmodulin kinase II (CaMKII) [10]. These enzymes, in turn, interact with (e.g. by phosphorylating) various intra-neuronal molecules, presumably for storage and processing of synaptic information. These CaMKII-phosphorylated protein structures must then, in some as-yet-unknown way, encode memory and regulate synaptic plasticity.
CaMKII has 12 kinase domains (6 on one side, 6 on the other), normally folded tightly in an inactive state to an association domain. Ca2+ influx to post-synaptic neurons binds calmodulin (CaM) to form Ca2+/CaM complexes, versatile regulators of multiple proteins and enzymes. Ca2+/CaM complexes activate CaMKII kinase regions allowing them to extend from the association domain and enable phosphorylation of a substrate protein, i.e. transfer of a phosphate group (through ATP hydrolysis) to a serine or threonine residue on a protein.
As Ca2+ influx can activate any, or all, of the 12 kinases, CaMKII has been suggested to record synaptic activity, retaining a ‘memory’ of past Ca2+ influx events in terms of activated phosphorylation states [11]. CaMKII may also auto-phosphorylate, in which one kinase phosphorylates an adjacent kinase on a single holoenzyme, resulting in prolonged CaMKII activation and possible information processing after initial Ca2+ influx. CaMKII is effectively a Ca2+ triggered synaptic memory device [12]. How and where information is encoded by CaMKII to be stored, and utilized to regulate synaptic plasticity remain open questions.
Following synaptic activity, CaMKII is rapidly distributed throughout post-synaptic neuronal dendrites and cell bodies [13]. The active sites of substrate phosphorylation on CaMKII kinase regions are referred to as ‘S’ sites and ‘T’ sites, apparently related to shorter and longer-term phosphorylation, respectively [14]. Single point mutation of amino acid threonine 286 in CaMKII results in impaired Ca2+ dependent alterations in synaptic plasticity, as well as in learning and memory in mouse models [10]. CaMKII phosphorylation of post-synaptic protein substrates is a likely mechanism for memory encoding/storage and regulation of synaptic plasticity.
Synaptic plasticity involves neuronal differentiation, movement, synaptogenesis and up- and down-regulation, all requiring, in one way or another, MTs, major structural components of the cytoskeleton. MTs are composed of tubulin, 110 kD protein hetero-dimers which self-assemble into hollow cylinders 25 nm in outer and 15 nm in inner diameter. MT walls have been crystallographically characterized as forming two types of hexagonal lattices (A-lattice and B-lattice) with helical winding patterns, including those following Fibonacci geometry [15]. Each tubulin dimer in MT lattices may occupy different states, and interact with neighbor tubulin dimer states, suggesting to a number of authors [16]–[20] that MTs process information in terms of tubulin states, and function as computational devices, e.g. molecular automata (‘microtubule automata’).
MTs in neuronal axons are arranged in continuous, uninterrupted parallel bundles. However, MTs in dendrites and neuronal cell bodies are uniquely arranged in mixed polarity, anti-parallel arrays of interrupted MTs, interconnected by MT-associated proteins (“MAPs”) including MAP2, whose activities are implicated in learning ([21] and references therein).
Various types of MAPs include those which cross-bridge MTs into scaffold-like networks defining neuronal architecture, and motor proteins (e.g. dynein, kinesin) which transport molecular cargo along MTs for delivery to specific synaptic locations. In most cells, MTs are labile and tend to undergo cycles of assembly and disassembly. However in neuronal dendrites and cell bodies, MTs are capped and stabilized by specialized MAPs including MAP-2 [22], [23].
MTs in brain neurons may also be stabilized by post-translational modifications. α-tubulin isotypes have either tyrosine, or glutamate at their C-terminal ends. Generally, MTs with abundant glutamate-terminating tubulins are stable, while MTs with many tyrosine-terminating tubulins are labile and short-lived [24], [25]. Terminal tyrosine may be removed (“detyrosination”) via tubulin carboxypeptidase, yielding glutamate-terminating tubulin and giving rise to disassembly-resistant MTs [26]. Stable MTs composed of glutamate-terminating α-tubulin appear to interact via MAPs with another, extremely stable cytoskeletal component, namely intermediate filaments [27]. Thus short-term memory stored in MTs may be further encoded and hard-wired in neurofilaments for long-term memory.
Evidence for MT-based information processing includes correlations between tubulin production and visual processing [28], MT signaling between membrane proteins and cell nucleus [29], [30], and MT-MAP alterations correlating with memory, novel learning and cytoskeletal remodeling [31]. Deficits in cytoskeletal function affect learning and memory [20], with the best-known correlation being disruption of MTs and resulting neurofibrillary tangles in brains of Alzheimer's disease patients.
In this study we performed molecular mechanics modeling and electrostatic profiling of binding and phosphorylation between activated CaMKII holoenzyme kinase domains and tubulin lattices in MTs, and analyzed their information processing capabilities.
CaMKII is a holoenzyme with twelve kinase regions in two hexagons sandwiched around a central association domain (see Figure 1 A). Activated by Ca2+/CaM, the kinases extend outward from the association domain, remaining tethered by linker regions (see Figure 1 B).
The phosphorylation process takes place on the inner surface of the kinase domains, in ‘hydrophobic groove’ regions characterized by valine 208 and tryptophan 237 (and, in the inactive state, occupied by threonine 286 on the autoregulatory region). The S-T hydrophobic phosphorylation site is exemplified by valine 208. Symmetric arrangement of the kinase regions provides a unique electrostatic profile in which each kinase region presents a positive potential region surrounded by negatively charged surfaces (see Figure 1 C). Substrates with reciprocal patterns of negatively charged regions on a positive background are excellent candidates for attractive electrostatic interactions.
As previously described, MTs are cylindrical polymers of hetero-dimer tubulin proteins, each composed of α- and β-tubulin monomers (see Figure 2 A). Tubulin dimers self-assemble into MTs, which are hollow cylinders of 13 linear tubulin chains known as ‘protofilaments’ (see Figure 2B). Side-to-side electrostatic interactions between tubulins in parallel protofilaments comprising the cylinder result in two types of skewed hexagonal lattices and helical winding pathways, the so-called A-lattice and B-lattice [32] (see Figure 2 C & D).
Overall, tubulin has a large net negative surface charge with almost half of the negatively charged residues located on the C-terminal tail of each monomer, which extend from the surface into the cytoplasm (see Figure 2A). This net electrostatic charge is, obviously, neutralized by counter-ions in the cytoplasm. Surface charges differ between α- and β monomers resulting in a net dipole moment pointing from the α- towards the β-monomer. Thus, MTs have net dipole moments, reflecting alignment of tubulin dipoles, and a unique skewed hexagonal electrostatic pattern of highly negatively charged regions surrounded by a positive background, dependent on the MT lattice type (see Figure 2 C & D).
Generally, CaMKII shows a preference for phosphorylating protein substrates at the amino acid sequence arginine-X-X-serine/threonine, where X can be any amino acid (although exceptions occur [33]). Wandosell et al. [34] showed that free, unpolymerized α- and β-tubulin are phosphorylated by activated CaMKII on or near the C-terminal region (beyond residue 306), resulting in tubulin conformational change, inhibition of assembly and inability to bind MT-associated protein 2 (MAP2). Multiple serine and threonine residues capable of being phosphorylated exist in this region (see Figure 2) however, the exact CaMKII phosphorylation site, or sites, on tubulin are unknown. Several sites on α- and β-tubulin follow this consensus sequence, however only one falls within the C-terminal region [35]. That site involves threonine 312 located not on the surface, but buried in the cleft between the α- and β-monomers. Another potential site is serine 444 on the βIII-tubulin C-terminal tail. Experiments suggest that phosphorylation of mammalian β-tubulin is restricted to the βIII isotype [27], which is found predominately in neurons. While alternate sites on βIII-tubulin have been argued for [36], it is clear that phosphorylation of tubulin by CaMKII holoenzymes occurs, though the precise sites remain unclear.
Size and geometry of the activated hexagonal CaMKII holoenzyme and the two types of hexagonal lattices (A and B) in MTs are identical (see Figure 3). The CaMKII hexagon is approximately 20 nm in breadth (see Figure 3 A), and the maximum distances between individual tubulin monomer in both the A lattice and B lattice MT neighborhoods are the same (see Figure 3 B). With minimal realignment in the linkers between association and kinase domains to account for MT curvature and lattice asymmetry, CaMKII precisely matches the MT A-lattice geometry, i.e. 6 extended kinases can interface collectively with 6 tubulins (see Figure 3 C – upper panel). Overlaying CaMKII with the 9-tubulin neighborhood of the MT B lattice requires slightly greater flexibility to precisely match the geometry (see Figure 3 C – middle and lower panels). This geometric matching permits activated CaMKII holoenzymes to bind to MT surfaces (see Figure 4 A & B).
The electrostatic pattern formed by a neighborhood of tubulin dimers on a MT surface (see Figure 2 C & D) shows highly negative charged regions surrounded by a less pronounced positive background, dependent on the MT lattice type (see Figure 2 C & D). These electrostatic fingerprints are complementary to those formed by the 6 CaMKII holoenzyme kinase domains making the two natural substrates for interaction. Alignment of the CaMKII holoenzyme with tubulin dimers in the A-lattice MT arrangement yields converging electric field lines indicating a mutually attractive interaction (see Figure 4 C & D). Considering the positive potential region of an individual kinase domain to be 1 kT/e and the negative surface of an individual tubulin monomer to be −10 e gives an electrostatic attraction of 10 kT, or 6 kcal/mol at 310 K, for each kinase-monomer interaction, the lower bound of binding. This attraction reaches a maximum bound of 60 kT (36 kcal/mol) when all 6 kinase regions are involved. This range of attraction is much stronger than thermal vibrational energy and indicates a significant association.
If each extended kinase can either phosphorylate at the S-T site on a tubulin substrate, or not, the process effectively conveys one bit of information (e.g. no phosphorylation = 0, phosphorylation = 1). Each set of six extended kinases on either side of a CaMKII holoenzyme can thus act collectively as 6 bits of information. Ordered arrays of bits are termed ‘bytes’.
Three hypothetical scenarios for CaMKII information encoding in MT lattice are considered (see Figure 5). In the first situation only β-tubulins in the 7-tubulin neighborhood patch of an A-lattice MT may be phosphorylated, giving two possible states – no phosphorylation (0), and phosphorylation (1) (see Figure 5 A). The scenario would be identical if only α-tubulin phosphorylation was considered. The central dimer is not considered for phosphorylation, so 6 dimers are available. Thus, there are 26 possible encoding states for a single CaMKII-MT interaction resulting in the storage of 64 bits of information. This case, however, only accounts for either α- or β-tubulin phosphorylation, not both. In the second scenario each tubulin dimer is considered to have three possible states – no phosphorylation (0), β-tubulin phosphorylation (1), or α-tubulin phosphorylation (2) (see Figure 5 B). These are ternary states, or ‘trits’ (rather than bits). Six possible sites on the A-lattice yield 36 = 729 possible states. The third scenario considers the 9-tubulin B-lattice neighborhood with ternary states. As in the previous scenarios the central dimer is not considered available for phosphorylation. In this case, 6 tubulin dimers out of 8 may be phosphorylated in three possible ways. The total number of possible states for the B lattice neighborhood is thus 36–28−8(27) = 5281 unique states.
Each phosphorylation event requires hydrolysis of a single ATP molecule, releasing ∼20 kT, or 12 kcal/mol of free energy at physiological temperature for a single kinase-monomer interaction, suggesting a robust molecular mechanism for information encoding. The daily energy usage of the human body requires the production of more than 2×1026 molecules of ATP [37]. Brain processes consume approximately one fifth this amount using 5×1020 ATP molecules per second. During a calcium influx there are ∼100 CaMKII subunits activated/phosphorylated per synaptic dendrite [38], each using a single ATP molecule. Taking calcium influx signals at a frequency of 100 Hz, activating 100 CaMKII subunits per synapse, and ∼104 synapses per neuron, one neuron would require 108 ATP/second for maximal encoding. With 1011 neurons in the brain this amounts to 1019 ATP/second, equivalent to 1019 bits, at a cost of approximately 2% of the brain's total energy consumption.
Metabolic energy for the proposed memory encoding by CaMKII phosphorylation of MTs should be distinguished from significantly higher metabolic costs of more coarse-grained forms of neuronal information processing. For example, hydrolysis of 104 ATP molecules is required to transmit one bit of information at a chemical synapse, and hydrolysis of 106 to 107 ATPs is needed for graded signals or spike coding [39].
We presume memory encoded by CaMKII phosphorylation alters, programs and provides a background for ongoing MT-based information processing e.g. microtubule automata, finer-grained processes underlying membrane potentials and synaptic activities. Information processing in post-synaptic dendritic and somatic MTs can thus: 1) integrate inputs to firing threshold at the proximal axon, and 2) regulate synaptic plasticity. In the following, we describe 6 possible mechanisms by which CaMKII-encoded tubulin phosphorylation in MT lattices could regulate cytoplasmic, membrane and neuronal structure and function leading to cognition and behavior.
Each tubulin dimer has two C-terminal ‘tails’, one from each monomer (see Figure 6). Modeling has shown that C-terminal tails can exist in multiple states, and that the tails may dynamically oscillate between extended conformations, involving interaction with water and ions (the “up-state”) (see Figure 6A), and folded conformations, involving interactions with the tubulin body (the “down-state”) (see Figure 6B) [40]. Thus, under proper conditions the negatively charged tails extend outward into the cytoplasm, able to alter local electric fields, as well as ionic and chemical states. CaMKII-induced phosphorylation of particular tubulins may cause dynamical (e.g. hexagonal) patterns of C-terminal extension, able to precipitate reaction-diffusion waves. Reaction-diffusion patterns known as Turing patterns [41], [42] are thought to function in intra-cellular information processing.
MAPs bind on the surface of individual tubulins in the region where phosphorylation occurs [27], [39], and removal of phosphate groups on tubulin results in decreased MT assembly in the presence of MAP2 [43]. These findings suggest that tubulin phosphorylation promotes MAP-MT interaction [27]. CaMKII induced phosphorylation patterns on MTs would thus both promote MAP-MT binding in general, and also serve as templates for attachment of MAPs at specific sites. MAPs attached in this way can form bridges between MTs, and thus cytoskeletal scaffolding determining MT spacing, neuronal architecture and synaptic location (Figure 7 A & B).
Figure 7 C shows pathways of phosphorylated tubulins following MT lattice geometry, intersecting at various tubulin sites. In Figure 7 D the intersection points are sites for MAP attachment. Patterns of phosphorylated tubulins can determine cytoskeletal morphology and cellular function.
In the absence of CaMKII, MAPs bind to MTs in vitro in super-helical lattice geometric patterns [44]–[46]. These may reflect resonance nodes of MT lattice vibrations with which CaMKII phosphorylation and other information modes interact.
Motor proteins dynein and kinesin move (in opposite directions) along MTs (using ATP as fuel) to transport and deliver components and precursors to specific synaptic locations. While MTs are assumed to function as passive guides, like railroad tracks, for motor proteins, the guidance mechanism by which they follow specific paths, e.g. through branching dendrites and interrupted MTs is unknown. Phosphorylation patterns on MT lattices could guide motor proteins to specific locations (see Figure 8 A–C).
Motor protein transport along MTs also depends on other types of MAPs, e.g. optimally spaced between ∼15 and 25 nm (equivalent to the length of two to three dimers) apart along the protofilament length [47], [48]. MAP tau (whose hyperphosphorylation and detachment from MTs correlates with Alzheimer's disease) appears to signal motor proteins precisely where and when to disengage from MTs and deliver their cargo [49] (see Figure 8 D). Consequently, patterns of CaMKII-induced tubulin phosphorylation in MT lattices could regulate motor protein transport along MTs directly, and/or by MAP attachment locations.
Cytoskeletal structures including MTs were first suggested to process information and regulate cellular function by Sir Charles Sherrington [16]. Computational interaction among discrete tubulin states [18], [50], and signals propagating along MT protofilaments and/or helical winding pathways have been proposed [15] (e.g. in dendritic and somatic integration in ‘integrate-and-fire’ brain neurons, regulating axonal firings and synaptic plasticity [51]).
Both MTs and MAPs can support solitary waves of ionic transport flow, e.g. via C-terminal tails on MT exteriors [52]. Signals may also propagate through MTs by intra-tubulin conduction pathways defined by electron resonance rings of aromatic amino acids [15]. Evidence suggests highly conductive electron and/or soliton conductance through MT lattices, including via helical pathways [53]. Signaling has also been proposed to occur in MT hollow cores [54].
MTs appear to have intrinsic electro-mechanical vibrations, e.g. coherent oscillations in the low megahertz range [55]. MT assembly is enhanced by 6 orders of magnitude when exposed to an applied 1 to 3 megahertz (radio frequency) alternating current, emits laser-like 3.1 to 3.8 megahertz radiation, and exhibits some form of electron condensation [53]. Signal propagation will be influenced by, and depend on MT lattice vibrations and related effects.
Dynamic interplay among CaMKII phosphorylation, communicative signaling/vibration, and MAP attachment sites are rich opportunities for regulation of synapses and intra-neuronal processes. Analogies to musical instruments have been suggested, e.g. MAP attachments acting as ‘frets’, as in a guitar [56], or the “guitar string hypothesis” [57] altering wavelengths and changing MT resonant frequency with consequences for functional processes.
In LTP, high frequency inputs (e.g. 50 to 100 Hz) are required for prolonged post-synaptic response. Kumar et al [58] showed that memory formation in dendrites depends on synchronized inputs, with an optimal frequency near 50 Hz. Density and patterns of CaMKII-induced tubulin phosphorylation in post-synaptic MT lattices would depend on frequency and synchrony of inputs.
Clusters of phosphorylated tubulin, and/or MAP attachment may serve as logic gates for propagating information. Figures 9 and 10 demonstrate two types of Boolean logic gates, an AND gate and an exclusive OR gate (XOR) in which MAPs convey inputs, with output along tubulin pathways. Figures 11 and 12 show AND and XOR gates in which MAPs convey output of inputs and processes in tubulins within the MT. The combination of XOR and AND logic gates forms a universal set for computation in which all other logic gates (NOT, OR etc.) can be conceived. Signals propagating through MT-MAP logic circuits may extend throughout cytoskeletal networks, regulating synaptic function, cognition and behavior.
‘Encoding’ is conversion of information from one form to another, each form requiring a geometry in which information is represented, e.g. as ‘bits’. Computer pioneer Alan Turing described a one-dimensional geometry, a string of information on a linear tape. Another computer pioneer John Von Neumann and others [59] developed two-dimensional cellular automata, lattice surfaces of interactive ‘cells’, or units. Simple rule-based interactions among neighbor cell states updating at discrete time steps can lead to complex patterns, computation and self-organization in cellular automata. Recently, three-dimensional assemblies of duroquinone molecules have been shown to function as ‘molecular’ automata (‘nanobrains’) [60].
Biology uses dynamical three-dimensional information via membrane dynamics, concentration gradients of various cytoplasmic components (e.g. reaction-diffusion) and cytoskeletal processes. How information is shared or encoded among these levels in a common framework is largely unknown.
We address information processing in cytoskeletal MTs, polymers of the protein ‘tubulin’, in both A-lattice and B-lattice configurations. Under proper conditions tubulin self-assembles into MTs, cylindrical hexagonal lattices, which directly participate in cell organization. Sherrington in 1951 [16] first proposed the cytoskeleton might serve as a cellular ‘nervous system’, and others have suggested microtubule-based information processing, e.g. with individual tubulins representing bit-like information states [18], [19], [56]. For example MTs have been modeled as von Neumann-type cellular/molecular automata (‘microtubule automata’) in which tubulin subunits interact with neighbor tubulin states by rules based on dipole coupling strengths [50], [61]. In such proposed microtubule automata, tubulin states interact and update coherently at discrete time steps attributed to theorized [62] or experimentally-observed coherent MT resonances, e.g. in the low megahertz range [53], potentially resulting in millions of synchronized updates per second [63].
To simulate microtubule automata, two-dimensional hexagonal MT lattices are slightly skewed according to lattice geometry and wrapped into a three-dimensional cylinder. Simulations show oscillating wedge-shaped, triangular and hexagonal patterns of tubulin states, which evolve, compute and can learn [50]. Because of cylindrical MT lattice geometry, such patterns reverberate, interfere and change nonlinearly. Thus microtubule automata are in principle capable of information processing. How could they interact with membranes and beyond, with extra-cellular processes? How could microtubule automata receive inputs and express outputs?
In this paper we evaluated possible information inputs to microtubules in the context of brain neuronal memory encoding and long-term potentiation (LTP). A key intermediary in LTP involves the hexagonal holoenzyme calcium-calmodulin kinase II. When activated by synaptic calcium influx, the snowflake-shaped CaMKII extends sets of 6 foot-like kinase domains outward, each domain able to phosphorylate a substrate or not (thus convey 1 bit of information). As CaMKII activation represents synaptic information, subsequent phosphorylation by CaMKII of a particular substrate may encode memory, e.g. as ordered arrays of 6 bits (one ‘byte’). We used molecular modeling to examine feasibility of collective phosphorylation (and thus memory encoding) by CaMKII kinase domains of tubulins in a microtubule lattice.
We show, first, complementary electrostatics and mutual attraction between individual CaMKII kinase domains and tubulin surfaces. We also demonstrate two plausible sites for direct phosphorylation of tubulin by a CaMKII kinase domain, and calculate binding energies in the range of 6 to 36 kcal/mol per CaMKII-tubulin phosphorylation event. This indicates encoding which is robust against degradation, yet inexpensive, requiring on the order of 2% of overall brain metabolism for maximal encoding in all 1011 neurons.
We then compare size and hexagonal configuration of the six extended foot-like kinase domains of activated CaMKII with hexagonal lattices of tubulin proteins in MTs. We find that CaMKII size and geometry of 6 extended kinase domains precisely match hexagonal arrays of tubulin in both A-lattice and B-lattices.
We quantified the potential information capacity for CaMKII hexagonal encoding of MT lattice regions, specifically in lattice ‘patches’ of 7 (A lattice) or 9 (B lattice) tubulin protein dimers. The simplest case was taken as CaMKII phosphorylation of an A lattice patch of 7 tubulin dimers (central ‘address’ dimer unavailable for phosphorylation, 6 available dimers). Each kinase domain can phosphorylate one tubulin dimer, either its α- monomer or its β- monomer equivalently. For either dimer, phosphorylation = 1, no phosphorylation = 0. Sets of 6 CaMKII kinase domains interacting with 6 tubulin dimers can then provide 6 binary bits (64 possible states), comprising one byte.
On an A lattice we also consider each tubulin dimer being phosphorylated either on its α-tubulin monomer (0), α-tubulin monomer (1), or neither (2), resulting in 6 ternary states (‘trits’) or 729 possible encoding states (per ‘tryte’).
We also considered ternary states in a B lattice with 9 tubulin dimers in a patch. With the central ‘address’ dimer unavailable for phosphorylation, sets of 6 CaMKII kinase domains can choose and phosphorylate α- (0), β- (1), or neither (2) tubulin monomer in any 6 of 8 available dimers. This yields 5,281 possible information states per neighborhood patch of 9 tubulin dimers. There are approximately 1019 tubulins in the brain. Consequently, potential information capacity for CaMKII encoding of hexagonal MT lattices is enormous.
Assuming input information may be encoded and processed in hexagonal microtubule lattices, how could such information be expressed as output to regulate cytoplasmic and membrane activities? We list six possible mechanisms including a potential connection to reaction-diffusion systems in cytoplasm.
Concentration gradients of various molecules in cytoplasm regulate cellular activities. In 1953 Alan Turing [64] showed that stable patterns of molecular concentrations could emerge through reaction, diffusion and inhibition. Reaction-diffusion systems include periodic waves and hexagonal patterns, which oscillate, undergo phase transitions and are capable of information processing [41]. Turing patterns have been suggested in biological systems across scale, e.g. from intra-cellular processes [42] to neuronal network organization. Cortical neurons representation maps are organized hexagonally [65], and ‘grid cells’ in entorhinal cortex record spatial location as apparent hexagonal maps, with different spatial scales at different layers of entorhinal cortex [66], [67].
‘Scale-free’ implies self-similar information patterns repeating at different spatial and temporal scales (following 1/frequency power laws). Similar to ‘fractals’ and holograms, scale-free structures and processes arise commonly in nature and technology, and are inherently robust and resistant to disruption. In the brain, evidence suggests neuronal network structure, temporal dynamics, and representation of mental states are all ‘scale-free’, with self-similar patterns repeating at various temporal and spatial scales and locations [68]–[71]. Interference patterns of periodic, coherent reaction-diffusion waves in cytoplasm and larger spatial scales could account for scale-free information patterns regulating biological systems including the brain. Microtubules can generate three-dimensional reaction-diffusion patterns [70], and we suggest such patterns operate at multiple time scales to regulate biological systems.
We demonstrate a feasible and robust mechanism for encoding synaptic information into structural and energetic changes of microtubule (MT) lattices by calcium-activated CaMKII phosphorylation. We suggest such encoded information engages in ongoing MT information processes supporting cognition and behavior, possibly by generating scale-free interference patterns via reaction-diffusion or other mechanisms. As MTs and CaMKII are widely distributed in eukaryotic cells, the hexagonal bytes and trytes suggested here may reflect a real-time biomolecular information code akin to the genetic code.
Sequences of human αCaMKII kinase, autoregulatory, linker, and association domains, as defined by Dosemici et al. [72] and Tombes et al. [73] were used to build homology models. Crystal structures 1HKX [74] and 2VZ6, found in the Protein Data Bank (PDB) [75], were used as templates to build basic homology models of the association, kinase and autoregulatory domains, respectively, using MODELLER 9V6 [76]. The CaMKII holoenzyme structure was built with PYMOL 0.99rc6 [77] using the geometry described in Rosenberg et al. [78] with the linker region constructed as a linear chain of residues joining the autoregulatory and association domains.
PDB tubulin protein structure 1JFF [79] was repaired by adding missing residues from 1TUB [80]. The repaired 1JFF dimer was solvated, neutralized and energy-minimized using NAMD [81]. The minimized and repaired 1JFF structure was used as a template to build basic homolgy models of TUBA1A and TUBB3 using MODELLER 9V6 [82]. Using this dimer, MT A and B lattice structures were built with PYMOL 0.99rc6 [77] using MT geometry described in Li et al. [82] and Sept et al. [32]. PYMOL 0.99rc6 [77] was used to model and illustrate the positional geometry changes of CaMKII-tubulin/MT lattice interactions.
To analyze electrostatic matching, hydrogens were added, and protonation states set at pH 7 with PROPKA [83], via PDB2PQR [84], [85] for both CaMKII and MT lattice structures. The Poisson-Boltzmann equation was solved for the structures in given arrangement with the Adaptive Poisson-Boltzmann Solver (APBS) [86] at a grid spacing of less than 1 Å. Isosurfaces were generated at +0.5 kT/e and −0.5 kT/e, and field lines were drawn with gradient magnitude 3. Electrostatic and field line images were generated in PYMOL 0.99rc6 [77] and VMD 1.8.7 [87], respectively.
|
10.1371/journal.pcbi.1006816 | Inferring neural circuit structure from datasets of heterogeneous tuning curves | Tuning curves characterizing the response selectivities of biological neurons can exhibit large degrees of irregularity and diversity across neurons. Theoretical network models that feature heterogeneous cell populations or partially random connectivity also give rise to diverse tuning curves. Empirical tuning curve distributions can thus be utilized to make model-based inferences about the statistics of single-cell parameters and network connectivity. However, a general framework for such an inference or fitting procedure is lacking. We address this problem by proposing to view mechanistic network models as implicit generative models whose parameters can be optimized to fit the distribution of experimentally measured tuning curves. A major obstacle for fitting such models is that their likelihood function is not explicitly available or is highly intractable. Recent advances in machine learning provide ways for fitting implicit generative models without the need to evaluate the likelihood and its gradient. Generative Adversarial Networks (GANs) provide one such framework which has been successful in traditional machine learning tasks. We apply this approach in two separate experiments, showing how GANs can be used to fit commonly used mechanistic circuit models in theoretical neuroscience to datasets of tuning curves. This fitting procedure avoids the computationally expensive step of inferring latent variables, such as the biophysical parameters of, or synaptic connections between, particular recorded cells. Instead, it directly learns generalizable model parameters characterizing the network’s statistical structure such as the statistics of strength and spatial range of connections between different cell types. Another strength of this approach is that it fits the joint high-dimensional distribution of tuning curves, instead of matching a few summary statistics picked a priori by the user, resulting in a more accurate inference of circuit properties. More generally, this framework opens the door to direct model-based inference of circuit structure from data beyond single-cell tuning curves, such as simultaneous population recordings.
| Neurons in the brain respond selectively to some stimuli or for some motor outputs, but not others. Even within a local brain network, neurons exhibit great diversity in their selectivity patterns. Recently, theorists have highlighted the computational importance of diverse neural selectivity. While many mechanistic circuit models are highly stylized and do not capture such diversity, models that feature biologically realistic heterogeneity in their structure do generate responses with diverse selectivities. However, traditionally only the average pattern of selectivity is matched between model and experimental data, and the distribution around the mean is ignored. Here, we provide a hitherto lacking methodology that exploits the full empirical and model distributions of response selectivites, in order to infer various structural circuit properties, such as the statistics of strength and spatial range of connections between different cell types. By applying this method to fit two circuit models from theoretical neuroscience to experimental or simulated data, we show that the proposed method can accurately and robustly infer circuit structure, and optimize a model to match the full range of observed response selectivities. Beyond neuroscience applications, the proposed framework can potentially serve to infer the structure of other biological networks from empirical functional data.
| Neural responses in many brain areas are tuned to external parameters such as stimulus- or movement-related features. Tuning curves characterize the dependence of neural responses on such parameters, and are a key descriptive tool in neuroscience. Experimentally measured tuning curves often exhibit a rich and bewildering diversity across neurons in the same brain area, which complicates simple understanding (e.g., see [1–4]). This complexity has given rise to a tendency towards biased selections of minorities of cells which exhibit pure selectivites, and have orderly and easily interpretable tuning curves. As a result the biological richness and diversity of tuning curves in the full neural population is often artificially reduced or ignored. On the theoretical side too, many network models feature homogeneous populations of cells with the same cellular parameters and with regular synaptic connectivity patterns. Neural tuning curves in such models will naturally be regular and have identical shapes.
New theoretical advances, however, have highlighted the computational importance of diverse tuning and mixed selectivity, as observed in biological systems [3, 5]. Furthermore, diversity and heterogeneity can be produced in mechanistic network models which either include cell populations with heterogeneous single-cell parameters (see e.g., Ref. [2]), or connectivity that is partly random and irregular despite having statistical structure and regularity (see, e.g., Ref. [4–10]). However, a general effective methodology for fitting such models to experimental data, such as heterogeneous samples of biological tuning curves is lacking.
A related central problem in neural data analysis is that of inferring functional and synaptic connectivity from neural responses and correlations. A rich literature has addressed this problem [11–16]. However, we see two shortcomings in previous approaches. First, most methods are based on forward models originally developed in statistics that are primarily inspired by their ease of optimization and fitting to data, rather than purely by theoretical or biological principles. Second, in the vast majority of approaches, the outcome is the estimate of the particular connectivity matrix between the particular subset of neurons sampled and simultaneously recorded in a specific animal [11–16]. Post-hoc analyses may then be applied to such estimates to characterize various statistical properties and regularities of connectivity [12, 16]. However, such statistical properties are, in most cases, the object of scientific interest, as they generalize beyond the specific recorded sample. Examples of such statistical properties are the dependence of connection probability between neurons on their physical distance [17] or preferred stimulus features [18]. Another example is the degree to which neuron pairs tend to be connected bidirectionally beyond chance [19]. A methodology for model-based inference of such circuit properties directly from simultaneously or non-simultaneously recorded neural responses is lacking.
Here we propose a methodology that is able to fit theoretically motivated circuit models to recorded neural responses, and infer model parameters that characterize the statistics of connectivity or of single-cell properties. Conceptually, we propose to view network models with heterogeneity and random connectivity as generative models for the observed neural data, e.g., a model that generates diverse tuning curves and hence implicitly models their (high-dimensional) distribution.
The generative model is determined by a set of network parameters which specify the distribution of structural circuit variables like individual synaptic connections or single-cell biophysical properties. In this picture, the particular realization of the connectivity matrix or of biological properties of particular neurons are viewed as latent variables. Traditional, likelihood-based approaches such as expectation-maximization or related approaches need to fit or marginalize out (e.g., using variational or Monte Carlo sampling methods) such latent variables, conditioned on the particular observed data sample. Such high-dimensional optimizations or integrations are computationally very expensive and often intractable.
Alternatively, one could fit theoretical circuit models by approaches similar to moment matching, or its Bayesian counterpart, Approximate Bayesian Computation [20, 21]. In such approaches, one a priori comes up with a few summary statistics, perhaps motivated on theoretical grounds, which characterize the data objects (e.g., tuning curves). Then one tunes (or in the Bayesian case, samples) the model parameters (but not latent variables) so that the few selected summary statistics are approximately matched between generated tuning curve samples and experimental ones [4]. This approach will, however, generally be biased by the a priori choice of the fit summary statistics, and does not exploit all the information available in the data for inferring circuit properties.
A suite of new methods have recently been developed in machine learning for fitting implicit generative models [22–24], i.e., generative models for which a closed or tractable expression for the likelihood or its gradient is not available. Here, we will demonstrate that a specific class of such methods, namely Generative Adversarial Networks (GANs) [23, 25, 26], can address the above problems. In particular, compared to methods such as moment matching, our proposed approach fits the entire high-dimensional data distribution in a much more unbiased and data-driven manner and without the need to choose a few summary statistics a priori. As we will show, this results in a more accurate and robust inference of circuit properties. In addition to inferring circuit parameters, this approach also allows for a more unbiased model comparison: one can simply simulate the competing circuit models, after fitting them to training data, and compare their goodness of fit, possibly to unseen data including new stimulus conditions not covered in the tuning curves used for training.
The rest of this article is organized as follows. We start the Results section by introducing the conceptual view of circuit models as implicit generative models for tuning curves. We then introduce the GAN framework. Next, we present the results of applying GANs to fit and infer the parameters of two recent influential circuit models from theoretical neuroscience. In Experiment 1 we fit a feedforward model of motor cortex to an experimental dataset of hand-position tuning curves, and compare the match between the distributions of empirical tuning curves and those generated by the models fit using our proposed method and traditional moment matching. In Experiments 2 and 3 we apply the method to fit a recurrent network model of visual cortex to a simulated dataset of stimulus-size tuning curves generated by a ground-truth circuit model. In Experiment 2 we show that the fit model captures the statistics of observed tuning curves very accurately. In Experiment 3, we assess the accuracy of circuit parameter identification using our proposed method, and discuss factors (such as the kind of tuning curves used for inference) that affect it. In the Discussion we conclude by discussing areas for extension and improvement of our proposed methodology, and broader potential applications of it. The details of our experiments, algorithms, and the models are given in Materials and methods; the source code for all implemented examples is available from https://github.com/ahmadianlab/tc-gan under the MIT license.
We consider mechanistic network models of the type developed in theoretical neuroscience, informed by knowledge of biological mechanisms and network anatomy, or by computational principles. We limit ourselves to networks evolving according to feedforward or recurrent firing rate equations (examples include Ref. [2, 4–10, 27, 28]), although the methodology is extendable to spiking networks as well (possibly with slight modifications to ensure differentiability). In the general presentation of this subsection we focus on recurrent rate networks. Abstractly, the neural responses in such networks evolve according to a dynamical system that has the following general structure (for a concrete example see the model in Experiment 2 below, which is governed by Eq (5)):
d v t d t = F ( v t ; I , W , γ ) . (1)
Here, vt is the vector of the state variables of the network’s neurons at time t (components of vt may, e.g., include the firing rate, membrane voltage and other “fast” state variables of individual neurons or synapses) and F is a vector field on state space which is differentiable with respect to its arguments. The matrix W is the partially disordered synaptic connectivity matrix (which can include both the recurrent, as well as feedforward external connections), and γ is the vector of possibly heterogeneous single-cell biophysical constants (e.g., membrane time-constant, spiking threshold potential, parameters of input-output nonlinearity, etc.). Finally, the vector I is the external input to the network which can represent stimuli or state-dependent modulators; we let I depend on a discrete index variable s denoting the stimulus or experimental condition (or discretized parameter). I(s) can in general be time-dependent, but here we assume it is stationary for simplicity. We also assume that I(s) is deterministic, and that all quenched randomness in network structure is captured by W and γ.
In order to exploit tuning curve datasets to constrain mechanistic network models, and in the process make model-based inferences about circuit properties, we propose to view network models of the above type as generative models for tuning curves. This view, which we will expound below, is graphically summarized in Fig 1. (Even though here we take trial-averaged tuning curves as the ultimate functional output of the model, this is not central to our proposed view; as discussed at the end of the Discussion, other quantities characterizing neural activity such as pairwise correlations or higher-order population statistics can augment or replace tuning curves in general applications.)
Given a choice of the non-dynamical variables W, γ and I(s), and the initial condition v(t = 0) (t = 0 can, e.g., be the stimulus onset time), the network dynamics can be simulated to compute the full temporal trajectory of all neural state variables. From this simulated trajectory, the average response, r ¯ ( s ), of a designated “probe” neuron in the network during a given “response interval” can be calculated in each condition s (this is the time-average of a certain component of vt for t in the response interval). The vector x ≡ ( r ¯ ( 1 ) , ⋯ , r ¯ ( S ) ) is then the tuning curve of that neuron, containing its responses in S different stimulus conditions which we assume are present in the training data. (Note that once a network model is trained, it can be applied to stimuli other than those used for training.) Thus, for networks with deterministic dynamics considered here, there is a deterministic mapping between the tuple of network’s structural variables, (W, γ), and the tuning curve of a given network neuron (we assume fixed initial conditions, vt=0, or otherwise ignore the dependence of output on them). We call this mapping f (see Fig 1).
Note that γ and W typically have very large dimensions; for a network of N neurons, γ and W have on the order of N and N2 components, respectively. In the proposed methodology, these large sets of structural and physiological network constants should be viewed as latent variables rather than model parameters that are fit to data. We are interested in cases in which these heterogeneous high-dimensional vectors are sampled from statistical ensembles (distributions) that capture the structured regularities as well as disordered randomness in single-cell properties and network connectivity. Consider a statistical ensemble described by a parameterized distribution P θ ( W , γ ). Through the deterministic map, f, between (W, γ) and x, this distribution in turn induces a distribution, P θ ( x ), over the tuning curve x, which is also parameterized by θ. The model’s parameter vector, θ, which is typically low-dimensional (see the examples in Experiment 1–3 subsections below), determines the network’s statistical structure and constitutes the parameters that we would like to fit to data. Components of θ can control, e.g., the average strength or spatial range of synaptic connections, or the mean and dispersion of biophysical single-cell properties.
Traditional likelihood-based methods infer the circuit properties, encapsulated in θ, by maximizing the likelihood function P θ ( ( x i ) i = 1 N ) ≈ ∏ i P θ ( x i ) given a dataset of tuning curves ( x i ) i = 1 N. However, for cases of interest, the mapping f is typically very complex and practically cannot be inverted (even though it can be relatively cheaply simulated in the forward direction); therefore in practice P θ ( x ) cannot be computed explicitly. Moreover, most likelihood-based methods are based on expectation-maximization-like algorithms; in the expectation step, such algorithms have to infer the high-dimensional latent variables (W, γ), which is a highly expensive computation.
In the next subsection we discuss how recently developed methods in machine learning, in particular generative adversarial networks, can be used to fit generative models of the above type, for which the parametrized data distribution, P θ ( x ), is only implicitly defined. All that is required in those approaches is the generative process that, given a random seed, generates a tuning curve (or a set of tuning curves), via a function that is differentiable with respect to θ. More formally, in such frameworks the generative model, or the “generator”, is characterized by a parametrized function Gθ which, given an input vector, z, of random noise variables that have a fixed or standard distribution, outputs a tuning curve x = Gθ (z). This is almost identical to the case of mechanistic circuit models described above, with two technical differences. First, the generative network process, as described above and in Fig 1, is captured by the function f which does not directly depend on the model parameters, θ. Instead, it is the inputs to this function, namely (W, γ), which implicitly depend on θ, as they are sampled from P θ ( w , γ ). Thus the second difference is that the inputs to f, i.e., the network’s structural variables (W, γ), do not have a fixed standard distribution, but rather a distribution dependent on θ.
However, we can use the so-called “reparametrization trick” [22], to remedy this mismatch and cast the circuit-model-based generative processes of interest to us in the required form. To this end, we will formulate the sampling of (W, γ) from their statistical ensemble via a deterministic function or mapping, gθ, parametrized by θ, that receives the fixed-distribution noise variables z as input. For example, a synpatic weight, w, with a gaussian distribution parametrized by its mean, μ, and variance, σ2, can be generated by the function gθ (z) ≡ μ + σz where z is sampled from the standard (zero-mean, unit-variance) normal distribution, and θ = (μ, σ) in this case. We provide biologically relevant examples of gθ in Materials and methods (see Eqs (17)–(19), (24) and (26)). Note that in the typical application of interest to us, while θ is low-dimensional, z has high dimensionality, on the order of the dimensions of (W, γ). The full generator function Gθ is then simply the composition of f and gθ: Gθ (z) ≡ f (gθ (z)) (see Fig 1B). In other words, first the standard noise variables z and network parameters θ together determine the full particular realization of network structure. The network is then simulated and a tuning curve (or a set of tuning curves) is generated. The function f is typically differentiable, and for many statistical ensembles of interest the function gθ is (or can be closely approximated by a) function that is differentiable with respect to θ. Then Gθ will also be differentiable in θ. As we describe in the rest of the article, with this formulation, we can use methods like generative adversarial networks to fit mechanistic neuronal circuit models to datasets of tuning curves.
Generative Adversarial Networks (GANs) are a framework for training generative models developed by the deep learning community [23, 29, 30]. The GAN approach is powerful because it is applicable to implicit generative models, e.g., the mechanistic networks discussed in the previous subsection, for which evaluating the likelihood function or its gradient is intractable. Another advantage of GANs (in the context of the previous subsection) is that, unlike typical likelihood-based methods, they fit the model parameters θ directly, skipping the computationally costly step of inferring the high-dimensional latent variables, namely the particular realization of network connectivity matrix W or single-cell constants γ. Note that unlike the particular realization of the connectivity matrix between the experimentally sampled cells, the model parameters θ, which characterize the statistics of connectivity and single-cell properties, are generalizable and of direct scientific interest.
All that is required in the GAN approach is a generative process that, given a random seed, generates a sample data object, via a function that is differentiable with respect to θ. While in machine learning applications the generated data object is often an image (and the generative model is a model of natural images), in our case it will be a tuning curve, formalized in the Introduction as the vector x ∈ R S containing the trial-averaged responses of a probed network neuron in S different experimental conditions (e.g., S different values of a stimulus parameter).
In a GAN there are two networks: a “generator” and a “discriminator”. The generator implements the generative model and generates sample data objects, while the discriminator (which can be a classifier) distinguishes true empirical data objects from “fake” ones generated by the generative model. Conceptually, the discriminator and generator compete: the discriminator is trained to better distinguish real from fake data, and the generator is trained to fool the discriminator.
More formally, the generator is characterized by a parametrized function Gθ which, given an input vector of random noise variables z that have a fixed distribution, outputs a sample data object x = Gθ (z). As we saw in the previous subsection, the process of generating a tuning curve by mechanistic neuronal circuit models can also be formulated in this manner (see Fig 1). In that case, the vector z provides the random seed for the generation of the full network structure (i.e., all synaptic weights and single-cell biophysical constants), given which the network is simulated to generated an output tuning curve x. We note that while in our applications the generator parameters θ usually correspond to physiological and anatomical parameters with clear mechanistic interpretations, in typical machine learning usage the structure of the GAN generator (e.g., a deconvolutional deep feedforward network, generating natural images) and its parameters may have no direct mechanistic interpretation (see Materials and methods, under “Differences with common machine learning applications”, for further discussion of this point).
The second network in a GAN is the discriminator. Mathematically, the discriminator is a function D w, parametrized by w, that receives a data object (in our case, a tuning curve) as input, and outputs a scalar. D w is trained so that its output maximally discriminates between the real data samples and those generated by Gθ. The generator is in turn trained to fool the discriminator. If the discriminator network is sufficiently flexible, the only way for the generator to fool it is to generate samples that effectively have the same joint distribution as real data. When D w is differentiable in its input and parameters and Gθ is differentiable in its parameters, training can be done using gradient-based algorithms. GANs can nevertheless be difficult to train and many techniques have been developed to improve their training. In this work we employ the Wasserstein GAN (WGAN) approach which has shown promise in overcoming some of the shortcomings of the traditional GAN approach [25, 26]. (We note, however, that traditional GANs could also be used for the types of application we have in mind.) The WGAN approach is mathematically motivated by minimizing the Wasserstein or earth mover’s distance between the empirical data distribution and the distribution of the generator output [25]. The Wasserstein distance provides a measure of similarity between distributions which (unlike e.g., the Kullback-Leibler divergence, which is the distance minimized in maximum-likelihood fitting) exploits the metric or similarity structure in the data space R S.
In the context of WGANs, the discriminator can be viewed as yielding a scalar measure or “summary statistic” for a (typically high-dimensional) data object or tuning curve. A single, fixed summary statistic D ( x ) can be used to measure the divergence between two distributions as the difference between the expectation of D ( x ) under the two distributions. However, two distributions can lead to the same average D ( x ) and yet be completely different in other respects. For example, consider the case of orientation tuning curves for visual cortical neurons. An example D is the function that receives an orientation tuning curve as input and outputs the half-width of that tuning curve (which measures the strength of orientation tuning). A generative model may produce orientation tuning curves that on average are narrower (or broader) than the average empirical tuning curve. However, even when a model matches the data distribution of tuning curve widths, its generated tuning curves may look very different from true ones along other dimensions (e.g., along the average height or maximum response dimension, or in the variance of tuning widths). One interpretation of the WGAN methodology is that instead of looking at data objects (tuning curves) along a fixed dimension using a fixed scalar measure, it optimizes that measure or probe to maximally distinguish between model-generated vs. empirical data objects. (In other flavors of GAN, the discriminator has other useful interpretations and can provide an estimate of the density of generator output in data space relative to the true data distribution; see Materials and methods, under “Alternatives to WGAN, and alternative views of GANs”.)
Let S be the class of all possible “smooth summary statistics” that can be used to characterize tuning curves; more technically, S is taken to be the set of all scalar functions of tuning curves, D : R S → R, that are Lipshitz-continuous with a Lipshitz constant less than one (if D is differentiable, the latter condition is equivalent to constraining the gradient of D to have norm less than one everywhere). Interestingly, by the Kantorovich-Rubinstein duality [25, 31], the earth mover’s distance, d(ρ, ν), between two distributions, ρ(x) and ν(x), can be expressed as the difference between the expectations of a maximally discriminating smooth summary statistic under the two distributions:
d ( ρ , ν ) = max D ∈ S | E x ∼ ρ [ D ( x ) ] - E x ∼ ν [ D ( x ) ] | . (2)
In our applications, one distribution (say ν) is the data distribution, and the other (say ρ) is the output distribution of Gθ which we would like to move closer to the data distribution. This can thus be done by iterating between improving D to maximize | E x ∼ ρ [ D ( x ) ] - E x ∼ ν [ D ( x ) ] |, and improving Gθ to minimize it.
To obtain a practical algorithm that in this manner approximately minimizes the earth mover’s distance, Eq (2), the expectations over the two distributions are approximated by mini-batch sample averages, and the discriminator class S is approximated by single-output feedforward neural network (parametrized by weight vector w) with input-output gradients of norm less than one (such networks form a proper subset of S; in practice, the gradient norm is only forced to be close to one). This gives rise to an adversarial algorithm in which the discriminator D w and the generator Gθ are trained by iteratively alternating between minimizing the following two loss functions
Loss D ( w , θ ) = E z [ D w ( G θ ( z ) ) ] - E x [ D w ( x ) ] + ( Gradient Penalty ) (3)
Loss G ( w , θ ) = - E z [ D w ( G θ ( z ) ) ] , (4)
where Eq 3 is minimized with respect to the discriminator’s parameters (w), and Eq 4 is minimized with respect to the generator’s parameters (θ). Here E x and E z denote averages over a batch of empirical data samples and samples from the standard noise distribution, respectively (thus E z [ D w ( G θ ( z ) ) ] is the same as the average of D w ( x ) when x is sampled from the generator output, instead of the empirical data distribution). The “Gradient Penalty” term forces the gradient of D w to be close to one (see Materials and methods, under “Conditional Generative Adversarial Networks”, for details). Finally, a penalty term PenaltyG(θ) can be added to the generator loss, Eq (4), as a regularization term for generator parameters.
In this subsection and the next two, we illustrate the GAN-based model fitting approach by applying it to two previously published mechanistic circuit models developed in theoretical neuroscience to explain various nonlinear features of cortical responses.
As our first example we take a feedforward model of primary motor cortex (M1) tuning curves proposed by Ref. [4]. The tuning curves describe the response tuning of M1 neurons as a function of 3D hand position and posture (supination or pronation). The model was proposed as a simple plausible mechanism for generating the significant complex nonlinear deviations of observed monkey M1 tuning curves from the classical linear model [34].
We used the “extended” model of Ref. [4] with small modifications (see Materials and methods subsection “Feedforward Model of M1”). In particular, we did not model response tuning with respect to hand posture (we did this in order to increase the size of the dataset by combining hand-position tuning curves from both posture conditions to allow for proper cross-validated testing of model fits). The model is a two-layer feedforward network, with the input layer putatively corresponding to the parietal reach area or to premotor cortex, and the output layer modeling M1 (see Fig 2). The input layer neurons’ activations are given by 3D Gaussian tuning curves defined on the hand position space. The receptive field centers formed a fine regular grid, but their widths varied randomly across the input layer, and were sampled independently from the uniform distribution on the range [σl, σl + δσ] (see Fig 2A). The feedforward connections from the input to output layer are random and sparse, with a connection probability of 0.01. In our implementation of this model, the strength of the nonzero connections were sampled independently from the uniform distribution on the range [0, J]. The response of an output layer neuron is given by a rectified linear response function with a threshold. The thresholds were allowed to be heterogeneous across the M1 layer, and were sampled independently from the uniform distribution on the range [ϕl, ϕl + δϕ]. Thus in total the model has five trainable parameters θ = (σl, δσ, J, ϕl, δϕ).
Using the WGAN framework, we fit the five model parameters to a dataset of experimentally measured monkey M1 tuning curves recorded by Ref. [4] and available online. With hand-posture conditions ignored, the tuning curves in this dataset describe the trial-averaged responses of a given M1 neuron in 27 experimental conditions corresponding to the monkey holding its hand in one location out of a 3 × 3 × 3 cubic grid of 3D spatial locations. (We ignored the hand-position label by blindly mixing hand-position tuning curves across pronation and supination conditions, as if they belonged to different neurons.) We randomly selected half of the hand-position tuning curves (n = 257) to be used for training the model, and used the other half (n = 258) as held-out data to evaluate the goodness of model fits.
The results showing the performance of the fit model are summarized in Fig 3. The figure shows the data and trained model histograms for four test statistics or measures characterizing the tuning curves or responses: firing rate (across the 27 conditions), coding level, i.e., the fraction of conditions with rate significantly greater than zero (which we took to mean larger than 5 Hz), the R2 or coefficient of determination of the optimal linear fit to the tuning curve, and the complexity score. Ref. [4] defined the complexity score of a tuning curve to be the standard deviation of the absolute response differences between nearest neighbor locations on the 3 × 3 × 3 lattice of hand positions (the tuning curve was first normalized so that its responses ranged from −1 to 1).
The last two statistics, the R2 and the complexity score, are of particular theoretical interest, as they provide two measures for the degree of irregular nonlinearity in the tuning curve and thus deviation from the classical linear model of M1 tuning curves [34]. Therefore, Ref. [4] fit their model by matching the mean and standard deviation of these two statistics between model and M1 data. By contrast, as described in the previous subsection, the WGAN, employed here, optimizes the discriminatory statistic, D, and (for complex enough discriminator networks) seeks to fit the entire joint multi-dimensional (in this case 27-dimensional) tuning curve distribution.
We measured the mismatch between the model and data distributions for each of the four test statistics using the Kolmogorov-Smirnov (KS) distance. As shown in Fig 3B, the goodness of fit improves fast during training. The goodness-of-the-fit of the distributions of complexity score and R2 (Fig 3C and 3D) are comparable to those obtained in Ref. [4] by grid search. It is also notable that the trained model fits the distribution of firing rates and coding levels (Fig 3E and 3F) quite well. In Ref. [4], the authors chose to individually normalize the model and data tuning curves so that their shape, but not their overall firing rate scale, was fit to data. We did not normalize the tuning curves, but as described above added a tunable parameter, J, to the model that scales the feedforward connection strengths. We found that with this addition and without normalization, the model is actually capable of accounting for the variability of firing rates across neurons and conditions as well. In the original model of Ref. [4] the thresholds were chosen separately for each output layer neuron such that the coding level of its response matched that of a randomly selected recorded M1 neuron. By contrast, in order to have all structural variability in the model in a differentiable form, we did not fit individual thresholds, but allowed them to be randomly distributed and only fit the two parameters of their distribution, ϕl and δϕl. Even though we did not match coding levels neuron-by-neuron by adjusting individual neural thresholds, the model was able to match the distribution of neural coding levels in the dataset, without the need for tuning individual thresholds.
The second network model we consider is a recurrent model of local cortical circuitry, the Stabilized Supralinear Network (SSN), which has found broad success in mechanistically explaining a range of nonlinear modulations of neural responses and variability by sensory context or attention, in multiple cortical areas [8, 35, 36]. The SSN is a recurrent rate network of excitatory (E) and inhibitory (I) neurons which have a supralinear rectified power-law input-output function, f ( u ) = k [ u ] + n (where u is the total input to a cell, k > 0 and n > 1 are constants, and [u]+ = max(0, u) denotes rectification). The dynamical state of the network of N neurons is the vector of firing rates r ∈ R N which is governed by the differential equation
τ d r d t = - r + k [ W r + F I ( s ) ] + n , (5)
where W and F denote the recurrent and feedforward weight matrices (with structure described below), the diagonal matrix τ = Diag ( ( τ i ) i = 1 N ) contains the neural relaxation time constants, τi, and I(s) denotes the stimulus input in condition s ∈ {1, …, S}.
We consider a topographically organized version of SSN with a one-dimensional topographic map which could correspond, e.g., to a one-dimensional reduction of the retinotopic map in primary visual cortex (V1). We adopt this interpretation here, and simulate the fitting of an SSN model of V1 to a dataset of V1 grating-size tuning curves, which are commonly used to characterize surround suppression [8, 37, 38]. The model has a neuron of each type, E and I, at each topographic spatial location. For the i-th model neuron, we denote its type by α(i) ∈ {E, I} and its topographic location by xi (which ranged from −0.5 to 0.5 on a regular grid).
In many cortical areas the statistics of local recurrent connectivity, such as connection probability and average strength, systematically depend on several factors, including the types of the pre- and post-synaptic neurons, the physical distance between them, or the difference between their preferred stimulus features [17, 18]. We made a choice for the distribution of Wij (the connection strength from neuron j to neuron i) that accounts for such dependencies in our topographic network, and also respects Dale’s principle [39, 40]. In order to make the GAN method applicable, another criteria was that Wij are differentiable with respect to the parameters of their distribution. The most relevant aspects of the distribution of Wij for the network dynamics are its first two moments (equivalently, mean and variance). We assumed that Wij’s are independent with uniform distributions on the range [〈Wij〉 − δWij/2, 〈Wij〉 + δWij/2], with a mean, 〈Wij〉, and width, δWij, that depend on the pre- and post-synaptic types and fall off with the distance between the pre- and post-synaptic neurons over characteristic length scales. More precisely we chose the fall off to be Gaussian, and set
⟨ W i j ⟩ = ± J ¯ a b exp ( - ( x i - x j ) 2 2 σ a b 2 ) (6)
δ W i j = δ J a b exp ( - ( x i - x j ) 2 2 σ a b 2 ) (7)
where a = α(i) and b = α(j) are the E/I types of the neurons i and j, respectively. The 2 × 2 matrices J ¯ a b, δJab, and σab constitute 12 trainable model parameters; they control the average strength, the degree of random heterogeneity, and the spatial range of the recurrent horizontal connections between different E/I cell types, respectively. All parameters are constrained to be non-negative, and we also enforced the constraint J ¯ a b ≥ δ J a b / 2; the sign on the right side of Eq (6), which is positive or negative, when the presynaptic cell type b is E or I, respectively, enforces the correct sign for Wij according to Dale’s principle.
The feedforward input, FI(s), to the SSN is composed of the stimulus-independent feedforward connectivity F and the topographically structured visual stimulus I(s). We chose the matrix F to be square and diagonal and hence I(s) to be N-dimensional. To model size tuning curves, we let the visual input I(s) in condition s only target neurons in a central band of the topographic grid roughly extending from location −bs/2 to bs/2 (see Fig 2B), as a simple model of visual input from a grating with diameter bs. We included quenched random heterogeneity in the diagonal feedforward weights by sampling Fii independently from the uniform two-point distribution on [1 − V, 1 + V]; thus V controls the degree of disorder in the feedforward inputs to network cells.
We modeled size tuning curves based on the sustained response (defined as the time averaged firing rate during the “sustained response” period; see Materials and methods) of a subset of SSN neurons driven by stimuli of different diameters or sizes. In experiments, typically the grating stimulus used to measure a neuron’s size tuning curve is centered on the neuron’s receptive field. To model this stimulus centering, we let the model output, r ¯ ( s ), in condition s be the sustained response of the excitatory neuron at the center of the topographic grid (which is also the center of the stimulus with size bs). Note that in general, the tuning curves in random SSN’s show variability across neurons as well as across different realizations of the network for a fixed output neuron. Furthermore, when N is large, the variability across the tuning curves of neurons with topographic locations sufficiently near the center often approximates variability of the center neuron across different network realizations with different z. Although we do not do so here, this self-averaging property may in principle be exploited for more efficient training of the model.
We used the GAN framework to fit the 12 recurrent connectivity parameters and one feedforward parameter of the SSN model, θ = ( J ¯ a b , δ J a b , σ a b , V ), to simulated data. The latter consisted of 2048 size tuning curves generated from a “ground truth” SSN with the connectivity parameters listed in Table 1. (We used a slight reparametrization of θ in the actual WGAN implementation; see Eq (28).) All other model parameters were the same between the ground truth and trained SSN models (see Materials and methods for details).
To quantify the goodness of fit between the data and trained model distributions of tuning curves, as in the previous section, we compare the distributions of four test statistics or measures characterizing the size tuning curves: preferred stimulus size, maximum firing rate, the suppression index, and normalized participation ratio. The preferred stimulus size is the stimulus size eliciting the largest response or maximum firing rate. The suppression index for a neuron (or tuning curve) measures how much the response of that neuron is suppressed at large stimulus sizes relative to response to its preferred size. Finally, the normalized participation ratio measures the fraction of tested stimulus sizes that elicit a response close to the maximum response. (See Materials and methods for the precise definition). Note that while to fit the model we used size tuning curves containing responses to stimuli with S = 8 different sizes, for testing purposes, we generated tuning curves from the trained and ground-truth SSN’s using a larger set of stimulus sizes. In particular, the tuning curves constituting the “true data” used for the tests in Fig 4 were not part of the training dataset, but were newly generated from the ground-truth model, and should thus be considered held-out test data. Moreover, they included responses to stimuli of sizes that were not covered in the tuning curves used in training.
Fig 4C and 4D provide comparisons of the distributions of these tuning curve attributes under the trained and ground truth SSN models. As in Experiment 1, we measured the mismatch of these distributions using the Kolmogorov-Smirnov (KS) distance. The KS distance for all distributions becomes very small as a result of learning (Fig 4B), reflecting the close fit of all test statistics from the trained model with the data (Fig 4C and 4D). (Note that since the generator was not directly trained to minimize any of the above KS distances, there is no reason why they should decrease monotonically during learning.)
Although moment matching can capture the overall shape of the one-dimensional (marginal) distributions of some of the test statistics (e.g., see Fig 4D for preferred size), it fails to capture some details in higher-dimensional joint distributions. For example, moment matching underestimates the density of the joint distribution of the suppression index and peak rate between the two peaks. By contrast, the WGAN faithfully captures this joint distribution (Fig 4C). We emphasize again that such summary statistics (and their distributions) did not directly play a role in the WGAN fit; instead, as discussed in subsection “Generative Adversarial Networks” above, the WGAN’s discriminator network automatically discovers the relevant optimal statistic, and in this way fits the full high-dimensional tuning curve distribution, and in particular the low-dimensional distributions plotted in Fig 4C and 4D.
In the point of view adopted in this paper, the distribution of neural tuning curves serves as a probe into the network structure; the richer the tuning curve dataset, the stronger the probe. Richness of tuning curve data can correspond to at least two different factors. One factor is the richness of stimuli, or the breadth and dimensionality of the region of stimulus parameter space covered in the tuning curves. A second factor is the degree to which each neuron’s tuning curve is associated with other functional or anatomical information such as cell type, preferred stimulus, topographic location, etc.
When the probe is not sufficiently strong, it may not serve to fully uncover the network structure as encoded in model parameters. When a model class is at all capable of capturing the observed tuning curve distribution, it may be the case that models within that class but with widely different parameters are equally capable of fitting the tuning curve distribution. In such a scenario the model parameters will not be uniquely identifiable using the available tuning curve data. For example, in Experiment 2 we were able to train an SSN to accurately capture the size tuning curve distribution. However, in many runs the parameters of the trained SSN failed to match the parameters of the ground-truth SSN that had generated the training dataset. This failure is, however, not a failure of the WGAN training algorithm, as the training was consistently successful in capturing the size tuning curve distribution very well. The failure is partly due to the relative poverty of the tuning curve data used. In that example, the dataset contained only the size tuning curves of excitatory and centered neurons (neurons with receptive field or topographic location at the center of the stimulus). Moreover, stimulus parameters other than size, such as contrast, were not varied in the tuning curves. We thus set out to investigate whether enlarging the tuning curve data along some of the mentioned lines can allow for accurate identification of the network parameters using the GAN methodology.
We experimented with including identified inhibitory cell tuning curves in the training data, as well as size tuning curves for offset cells with topographic location not at stimulus center. However, we found that only including offset tuning curves was sufficient to enable parameter identification. We will report the results of this experiment here; in this training dataset we included the size tuning curves of neurons with the following possible offsets from stimulus center: ( x i p ) p = 1 O = ( 0 , 1 / 16 , 1 / 8 , 1 / 4 , 3 / 8 ).
To fit the SSN to the enriched dataset, here we employed the conditional WGAN (cWGAN) algorithm. As explained in the subsection Conditional GANs above, in cWGAN the discriminator and generator depend on a set of “condition” variables in addition to their inputs in the original WGAN. Here, we provided the topographic offset, x i p, of the cells as the conditional input. The cWGAN formalism allows for using tuning curves of neurons for which only some of the offsets are measured. In particular, it allows for exploiting off-center size tuning curves (which are commonly discarded) towards system identification.
We compare the quality of fits using the cWGAN and moment matching respectively (Fig 5). Although moment matching produces a reasonable fit to the distribution of tuning curves for the individual features we explored (Fig 4B), the GAN approach significantly outperforms moment matching at parameter identification (Fig 5). This is summarized in the relative error plots in Fig 5A and 5D; relative error was measured using the symmetric mean absolute percentage error (sMAPE), defined in Eq (44) of Materials and methods. In particular, moment matching severely misestimates the δJab’s (see Fig 5E), which control the heterogeneity in recurrent horizontal connections. On the other hand, cWGAN was successful at identifying parameters with less than 10% error, and the fit of the summary statistic distributions was excellent.
The results demonstrated in Fig 5 are robust. Fig 6 shows a histogram of percent error (quantified by sMAPE, Eq (44)) for multiple moment matching and cWGAN fits, performed using a wide range of hyperparameters (see Materials and methods, under “Hyperparameters for parameter identification experiment”), including different learning rates for the generator, and for the discriminator in the cWGAN case. To provide a fair comparison of performance between the two methods, in Fig 6 we used an impartial termination criterion which is agnostic to the hyperparameters (see Materials and methods, under “Stopping criterion and performance metric”), for both cWGAN and moment matching. In all cases the cWGAN outperforms moment matching.
We observed that successful cWGAN fits required particularly small generator learning rates. By contrast, in other examples (not reported) in which the tuning curve dataset was richer, and inhibitory neuron tuning curves were also observed by the discriminator, trainings with ten times larger generator learning rates successfully identified the parameters. We also observed that moment matching was able to identify the parameters in those easier cases. More generally, we speculate that harder problems (i.e., those with poorer training data), such as in Fig 6, may require a smaller WGAN generator learning rate for accurate parameter identification.
Developing biologically grounded mechanistic models that can capture the diversity and heterogeneity of neural response properties and selectivities is an important aim of theoretical neuroscience. Methods that give us the ability to quantitatively constrain such models directly using neural data, such as datasets of tuning curves from a given brain area, can greatly facilitate this pursuit. Such methods allow inferring the structure of biological circuits of interest from observations of neural responses, as encoded, e.g., by tuning curve distributions.
The statistical fitting of most interesting mechanistic models of brain circuitry using classical likelihood-based approaches is often intractable. A practical solution used in the past has been to instead use models that are designed primarily based on the tractability of their likelihood functions and ease of fitting, as opposed to biological realism or purely theoretical criteria. Therefore the elements and parameters of such models often may not have a direct mechanistic, biological interpretation. Another approach to the challenge of fitting theoretically grounded mechanistic models has been to forego full probabilistic inference using likelihood-based approaches, and use moment matching to only match a few summary statistics (characterizing tuning curves and thus neural responses) between model and data. The drawback of this approach is that it is not information theoretically efficient and does not exploit all the available information in the dataset for the purpose of inferring network properties.
Here, we demonstrated that Generative Adversarial Networks (GANs) enable the fitting of mechanistic models developed in theoretical neuroscience to the full joint distribution of neural response data. Conceptually, we proposed to view mechanistic network models with randomness in their connectivity structure or other biophysical parameters as generative models for response tuning curves. Given this formulation, one can exploit a whole suite of recently developed methods in machine learning (of which GANs are an example) for fitting the full output distribution of implicit generative models (i.e., generative models for which the likelihood function is not available or is highly intractable).
In this paper we specifically focused on using the Wasserstein GAN (WGAN) [25] approach for this purpose. In subsection “Generative Adversarial Networks” of Results we reviewed the basic GAN setup and the WGAN algorithm, and conceptually contrasted it with moment matching. In Experiment 1 and 2 we used this method to successfully fit two representative example models from theoretical neuroscience to real or simulated tuning curve data, demonstrating that this technique is applicable to inference based on a wide range of network models, including feedforward and recurrent models of cortical networks. Furthermore, in Experiment 3 we demonstrated that our method is able to consistently and accurately identify the true parameters of a cortical network model using tuning curve data. This experiment moreover provides an example application in which the WGAN approach, which exploits the full tuning curve distribution, is superior to moment matching which only considers a few moments of that distribution. However, we note that even though moment matching is information-theoretically weaker than GANs, in practice it may be advantageous for fitting simple circuit models as it can be trained faster.
In the rest of the Discussion, we review some of the potentials and pitfalls of our approach and some differences with other usages of GANs, in an attempt to make the path clearer for other future applications of and improvement to this approach.
In our experiments we used both conditional and non-conditional GANs. When using a non-conditional GAN as in Experiments 1 and 2, the discriminator only receives as input the discretized tuning curve in the form x ≡ ( r ¯ ( s ) ) s = 1 S. Here, the index s denotes the stimulus condition, corresponding to combinations of different stimulus parameter values. In this formulation of the tuning curve, the relationship between stimulus parameters and the index s is thus lost to the discriminator. In particular, the discriminator is blind to the metric or similarity structure in the stimulus parameter space (e.g., it does not explicitly know whether two different components of x encode responses to very similar or widely different stimuli), and therefore cannot directly exploit that structure in discriminating between true and generated tuning curves. Another drawback of this formulation is that all neurons in the dataset must have been recorded in all stimulus conditions; when there are several stimulus parameters, however, recording each neuron in all conditions for many trials becomes experimentally prohibitive. On the other hand, non-conditional GANs are advantageous in allowing the generator to learn the joint distribution of single-neuron responses across the entire stimulus parameter space; in particular, it enables to fit the marginal distribution of “global” tuning curve features that depend jointly on responses at different values of multiple stimulus parameters.
The conditional GAN approach provides a complementary scheme for describing the tuning curve, i.e., the relationship between neuronal responses and stimulus parameters. In this case, different values of a subset of stimulus parameters are implicitly represented by the different components of x. However, x (and its component responses) depend also on another subset of stimulus parameters that are provided explicitly as inputs to the discriminator (as well as the generator), in the form of conditional GAN’s condition variables c (more generally, c need not be limited to subsets of stimulus parameters, and can, e.g., also denote a neuron’s cell type or preferred stimulus parameters). Since the value of these parameters is directly provided to the discriminator, the latter is not entirely blind to stimulus similarity structure. On the other hand, the conditional GAN framework only fits the conditional distributions of x at different values of c. This is beneficial in that we now do not need to record each neuron across all values of c. However, by the same virtue, the framework is blind to correlations of single-cell responses at different values of c across neurons, and may not fit the distribution of single-neuron tuning curve features that depend jointly on responses to stimuli with different c-values. By allowing a trade-off between capturing the joint distribution of single-neuron responses across the entire parameter spaces vs. handling a heterogeneous dataset with missing data, the conditional GAN provides additional flexibility in fitting theoretical models to diverse neuronal data.
As with any gradient-based training method, it is possible for a GAN to become stuck in suboptimal local minima for the generator (or the discriminator). It is further an open question whether GAN training will always converge [29, 41, 42]. As research in GANs and non-convex optimization advances this issue will be improved. For now avoiding this pitfall will be a matter of the user judging the quality of fit after the fit has reasonably converged. Starting the gradient descent algorithm with several different initial conditions for generator parameters can also help, with some initializations leading to better final fits.
Apart from the above general problems, when the generator is a recurrent neural network (RNN), other problems may arise within each step of gradient descent. When the generator output is based on a steady-state (fixed point) of the RNN, as was the case in our SSN experiment, a potential issue is lack of convergence to a stable fixed point for some choices of recurrent network parameters. In our experiment with SSN, we initialized the generator network in such a way that it initially had a stable fixed point for almost all realizations of z. For the SSN this would generically be the case when recurrent excitation (which has destabilizing effects) is sufficiently weak. Hence initializations with small JEE and δJEE are good choices. In addition, a relatively large SSN size N improves the stability issue because random quenched fluctuations in total incoming synaptic weights are relatively small when the number of presynaptic neurons is large. Thus, for large networks, for a given choice of network parameters, θ, either the network converges to a stable fixed point for almost all z, or almost never does. To avoid entering parameter regions leading to instability during training, we added the additional regularizing term Eq (32) to the generator loss. We found that the addition of this term is crucial for the success of the algorithm.
An additional problem particular to optimizing GANs is mode collapse (also known as mode dropping), in which some modes of a multi-modal dataset are not represented (or in the worst case only one mode is represented) in the generative model output [29, 43, 44]. Mode collapse is an example of underfitting. The work presented here did not suffer from mode collapse, likely because of the highly structured models employed. Nevertheless, other applications may suffer from the problem of mode collapse. Many approaches have been explored to prevent mode collapse, and we do not give a comprehensive review, but instead cite a selection of interesting approaches. The WGAN itself, which is employed here, is believed to alleviate mode collapse to some degree [25]. Other more sophisticated approaches exist, including the addition of a mutual information maximizing regularizer between model output and the latent variables [45]. One particularly elegant and effective approach is the PacGAN which provides two or more independent, concatenated generator or data samples to the discriminator so that a model suffering from mode collapse will be recognizable from its lack of diversity [46].
Another possible problem is that of overfitting. In the context of generative model training, extreme overfitting corresponds to the generative model approximately memorizing the individual samples in the training data. In our experiments, this would have corresponded to the trained circuit model only generating a finite set of possible tuning curves, namely the samples seen during training. GANs are prone to sample memorization when their generator and discriminator have high complexity or expressivity. For typical circuit models in neuroscience, however, such as the examples we considered here, the output (e.g., tuning curve) distribution is expected to have a smooth density (in contrast to a discrete distribution with support on or near the training-set tuning curves) almost everywhere in parameter space; such a generator can never memorize a finite sample of tuning curves.
In fitting generative models it is possible for models with widely different parameters to result in nearly identical output distributions. In our case, this corresponds to cases in which networks with widely divergent connectivity or single-cell parameters nevertheless generate very similar tuning curve distributions. In such cases it would be impossible to make precise inferences about network parameters (e.g., connectivity statistics) using tuning curve data. This problem is exacerbated for the moment matching method, which discards information in the data by reducing the tuning curve distribution to a few moments. In comparison, the problem should generally be less severe for the GAN method which tries to fit the entire distribution. Irrespective of the fitting method, however, there is no general reason why the distribution of a relatively low-dimensional output of the model, such as the tuning curve with respect to one stimulus parameter, would provide sufficient information for constraining all circuit parameters. Fortunately there is nothing in our approach that prevents one from applying it to datasets of tuning curves with respect to several stimulus parameters, or tuning curves of multiple cell types. The general expectation is that the higher the dimension of the stimulus parameter space underlying the tuning curves in the training data, the more identifiable the network parameters become.
For example, in the case of our SSN experiments, we first trained the generative model using only tuning curves with respect to stimulus size. In the parameter identification experiment (Experiment 3), we enriched the size tuning curve dataset by adding center-offset neurons, and additional sampling conditions by adding inhibitory cell tuning curves. The result was an improvement in the robustness and accuracy of model parameter identification. With datasets of sufficiently rich tuning curves, the GAN-based method provides a promising way to infer biophysical networks parameters, such as those governing connectivity statistics. This also has deep implications for experimental design: the standard approach of using optimal stimuli (in the case of our SSN example, gratings with no center offset), or only focusing on excitatory neurons may produce datasets that are insufficiently rich to allow for model-based inference of all circuit parameters of interest. In particular, a framework like GANs can in principle be used to design experiments, i.e., optimally choose the stimulus conditions and quantities to be recorded, to maximize the identifiability of the parameters of a given model.
The current work can be extended along several directions. In addition to the GAN framework, a suite of other methods have also been developed recently in machine learning for fitting generative models [23]. Examples include variational autoencoders [22, 47], and hierarchical and deep implicit models [24]. These methods can also be fruitful for fitting circuit models from neuroscience and inferring circuit parameters. Recent progress in unifying these approaches [48–50] can further inform their future applications. Of special note is the Bayesian methodology of Ref. [24] which in addition to point estimates for parameters, also yields estimates of their posterior uncertainty (or more generally, their approximate posterior distribution).
In the current study, we took the output of the mechanistic circuit model to be tuning curves composed of single-cell trial-averaged sustained responses. But the conceptual framework explained in the beginning of Results and the GAN methodology can also be used to constrain mechanistic network models using data featuring the temporal dynamics of neural activity and higher-order statistics of trial-to-trial variability. For example, the network model can be fit not only to match the distribution of tuning curves encoding trial-averaged single-cell sustained responses, but rather the entire peristimulus time histogram or also the distribution of noise correlations between cell pairs in some cortical area, extracted from simultaneously recorded neural activity.
Lastly, although we focus here on applications to neuroscience and neuronal networks, the proposed framework can potentially serve to fit mechanistic models from other corners of biology, in order to infer the structure of other kinds of biological networks from functional data. For example, the framework can potentially be used to infer the structure of gene regulatory networks from data on the expressions of one or a few genes in different environments.
Here we provide the complete expressions for the loss functions used in the conditional WGAN (cWGAN) method and a pseudocode for this algorithm. The non-conditional WGAN setup was described in subsection Generative Adversarial Networks of Results. cWGAN’s are similarly composed of a generator and a discriminator, but now both the descriminator D w and generator Gθ depend on a “condition” argument or input, c, in addition to their primary inputs [33]. The condition variable c can be discrete or continuous and can range over a set of possibilities C. When there is only one possibility for c (in which case this argument can be dropped) we recover the original (non-conditional) WGAN.
The discriminator and generator loss functions for cWGAN are given by
Loss D ( w , θ ) = E z , c [ D w ( G θ ( z ; c ) ; c ) ] - E x , c [ D w ( x ; c ) ] + ( Gradient Penalty ) (8)
Loss G ( w , θ ) = - E z , c [ D w ( G θ ( z ; c ) ; c ) ] + Penalty G ( θ ) , (9)
respectively. Here E x , c denote the average over a batch of data samples, x, together with the conditions, c, at which they were recorded. Similarly, E z , c denotes averaging over a batch of noise variables z, sampled from their fixed distribution, and conditions c, sampled from their empirical distribution. Note that z and c are treated as independent random variables. The “Gradient Penalty” term forces the gradient of D w (with respect to its first argument) to be close to one; following the recipe of Ref. [26] we set it to
Gradient Penalty = λ E z , x , c , ϵ [ ( ‖ ∇ D w ( ϵ x + ( 1 - ϵ ) G θ ( z ; c ) ; c ) ‖ 2 - 1 ) 2 ] (10)
where ϵ is a random variable with uniform distribution on [0, 1], and the gradient ∇ is taken with respect to the first argument of D w, and not the condition c or parameters w. Finally, the term PenaltyG(θ) in the generator loss denotes possible generator model dependent regularization terms. We did not include any such term for the feedforward network example presented in Experiment 1. The PenaltyG(θ) for the recurrent SSN example is described in Eq (32) of Materials and methods.
A stochastic gradient descent algorithm for cWGAN based on these loss functions is shown in Algorithm 1.
Algorithm 1: Improved cWGAN algorithm based on Ref. [26]. For all models, we use n D = 5. For update method, we either use Adam with β1 = 0.5, β2 = 0.9, ϵ = 10−8 [26] or RMSProp with ρ = 0.9, ϵ = 10−6. For the feedforward model, we use α D = α G = 0 . 001, m = 30, γ = 0, PenaltyG(θ) = 0, and Update D = Update G = Adam. The generating processes are considered normal always and the test† always passes. For the SSN α D = 0 . 02, αG = 10−4, m = 128, γ = 0.001, PenaltyG(θ) = Eq (32). The discriminator is updated only if generating processes (Gθ(z)) pass a test† for “normality” (see Eq (33)). We use always RMSProp for the generator in the SSN experiments. We use RMSProp for the discriminator in Fig 4, Adam in Fig 5, and aggregate the results with both RMSProp and Adam in Fig 6.
Input: data distribution P r, the gradient penalty coefficient λ, the number of discriminator iterations per generator iteration n D, the batch size m, update methods Update D ( · ), UpdateG(·), learning rates α D, αG, weight decay hyperparameter γ, and the initial discriminator w0 and generator θ0 parameters.
θ ← θ0;
w ← w0;
while θ has not converged do
repeat
for i = 1, …, m do
Sample real data ( x , c ) ∼ P r, latent variable z ∼ p(z), a random number ϵ ∼ U[0, 1].;
x
˜
←
G
θ
(
z
;
c
)
;
x
^
←
ϵ
x
+
(
1
-
ϵ
)
x
˜
;
Loss
D
(
i
)
←
D
w
(
x
˜
;
c
)
-
D
w
(
x
;
c
)
+
λ
(
‖
∇
x
^
D
w
(
x
^
;
c
)
‖
2
-
1
)
2
;
end
if generating processes are normal† then
w
←
Update
D
(
∇
w
1
m
∑
i
=
1
m
Loss
D
(
i
)
,
w
,
α
D
)
-
γ
w
;
end
until
n
D
updates are tried;
Sample latent variables ( z ( i ) ) i = 1 m ∼ p ( z ) for a mini-batch.;
Sample conditions from real data ( · , c ( i ) ) i = 1 m ∼ P r for a mini-batch.;
θ
←
Update
G
(
-
∇
θ
(
1
m
∑
i
=
1
m
D
w
(
G
θ
(
z
(
i
)
;
c
(
i
)
)
)
+
Penalty
G
(
θ
)
)
,
θ
,
α
G
)
;
end
In this subsection, we provide a quick review of some of the alternatives to the WGAN loss functions, Eqs (3) and (4), developed in the GAN literature, which can be useful in computational biology applications (for a more comprehensive review of GANs see [29]). We also point out an alternative view of GANs inspired by the energy-based framework for unsupervised learning [51, 52].
The original GAN developed in [23] was framed as a minimax or zero-sum game in which the generator and discriminator competed by respectively minimizing and maximizing the same loss:
Loss G ( w , θ ) = - Loss D ( w , θ ) = E x [ log D w ( x ) ] + E z [ log ( 1 - D w ( G θ ( z ) ) ) ] . (11)
Given this form for the loss, the discriminator, D w ( x ), can be thought of as a binary classifier that estimates the conditional probability of the true or empirical category (vs. “fake” or G-generated category) given the observation x. In this case, for a fixed generator whose output has distribution Pθ(x), the theoretically optimal discriminator is given by D * ( x ) = P true ( x ) P true ( x ) + P θ ( x ), where Ptrue(x) is the true data distribution or density. In that sense, a sufficiently expressive and well-trained discriminator (using Eq (11)) learns the ratio of the model likelihood, Pθ(x), to the true data density.
Moreover, for this optimal discriminator solution the loss Eq (11) reduces to the Jensen-Shannon (JS) divergence (up to additive and multiplicative constants) between Ptrue(x) and Pθ(x) [29]. Thus the generator is theoretically trained to minimize the JS distance. By comparison, as noted in subsection “Generative Adversarial Networks” of Results, WGANs theoretically minimize the Wasserstein or earth-mover’s distance between Ptrue(x) and Pθ(x).
Both of these divergence or distance measures are in contrast to the Kullback-Leibler (KL) divergence DKL(Ptrue‖Pθ) which is effectively minimized in classical maximum-likelihood estimation of the generator (or equivalently, of its parameters θ). Correspondingly, others have modified the generator-loss so that, given the theoretically optimal discriminator, it reduces to the KL divergence, or other divergences [53–56]. For example, to minimize the KL divergence, [56] used the same discriminator loss as in Eq (11), in conjunction with the modified generator loss
Loss G ( w , θ ) = E z [ f ( D w ( G θ ( z ) ) ) ] (12)
where f ( u ) = u 1 - u exp ( u 1 - u ).
As we mentioned in the Discussion, Ref. [24] developed a GAN-like framework for variational Bayesian estimation of implicit generative models, in which a discriminator-like network was trained to estimate the generator’s likelihood function. However, this Bayesian framework goes beyond maximum-likelihood estimation: Ref. [24] developed a three-network framework in which, in addition to the (implicit) generative model and discriminator (which estimates the generator’s likelihood), a third network is trained to provide an approximation to the Bayesian posterior over θ (as in more general variational Bayesian approaches [22, 57]).
Finally, we mention energy-based GANs (EBGANs), which are inspired by the energy-based framework for learning [51, 52]. EBGANs use the following loss functions for the generator and the discriminator:
Loss D EBGAN ( w , θ ) = E x [ D w ( x ) ] + E z [ ( m - D w ( G θ ( z ) ) ) + ] (13)
Loss G EBGAN ( w , θ ) = E z [ D w ( G θ ( z ) ) ] (14)
where (x)+ = max(0, x) denotes rectification, m > 0 is a positive margin parameter, and D w is constrained to be non-negative (see [58]).
EBGANs are in part motivated by an alternative view of the role of the discriminator in GANs. In the viewpoint expressed in subsection “Generative Adversarial Networks” of Results, the discriminator is thought of as a flexible and trainable objective function that is used for training the generator. In this interpretation, the generator is key and the discriminator is auxiliary. However, an alternative viewpoint is also possible in which the discriminator is key and the generator is auxiliary (see appendix B of [58]). In this view, which is suggested by the energy-based framework for learning [51, 52], the discriminator learns the relative density of the data distribution. More precisely, it is thought of as an energy function that is shaped during training to be low in regions of high data density, and high elsewhere. This is achieved by minimizing a loss functional during learning; Eq (13) is an example of such a loss functional. Imagine for simplicity that true data lie on a subspace or manifold in the data space. The loss functional is designed such that true data samples serve to “push down” the energy function (i.e., increase the probability density) on the data manifold, while another mechanism is used to “pull-up” the energy function (i.e., reduce the probability density) outside that manifold. In simple unsupervised learning methods such as principle component analysis the pull-up of energy is implicitly achieved due to the rigidity of the energy function itself (see [52]). But when the energy function is sufficiently flexible, an explicit term in the loss functional is needed to pull it up outside the true data manifold. Fake or simulated data points, referred to as contrastive samples, can be used for energy pull-up. In Eq (13), the first and second term serve to push up and pull down the energy function, D w, respectively. Accordingly, in the alternative viewpoint of GANs, the generator is viewed as merely providing such contrastive samples to the discriminator.
As explained below, in contrast to many applications of unsupervised learning, the generator is indeed central in our intended applications, and we therefore focused on the first interpretation of GANs in subsection “Generative Adversarial Networks” of Results. Nevertheless, unsupervised learning of generative models is often carried out with the end-goal of learning a useful representation of observed data, which can, e.g., serve to compress or reduce the dimensionality of data. In our case this corresponds to dimensionality reduction of tuning curves. Estimating the density of observed data is a related end-goal of unsupervised learning, which can, e.g., serve data restoration (e.g., image denoising) applications. Even though these were not the applications motivating the current study (see the next subsection), they do constitute potential applications of GANs in computational biology. The alternative view of the role of discriminator can be more advantageous in such settings.
In most applications of GANs in machine learning and artificial intelligence, the generative model is an artificial neural network (e.g., a deconvolutional deep feedforward network), and that network’s individual connection weights constitute the generator’s trainable parameters, θ. In most such applications, these parameters are not objects of interest on their own and may not be mechanistically meaningful or interpretable. Similarly, the development of generative models in such domains is not necessarily concerned with capturing the true, physical mechanisms underlying the modeled data. There and in other unsupervised learning settings, the end goal is to achieve a generator that produces realistic data objects (e.g., images), or to accurately estimate the data density or its support (which can serve for dimensionality reduction or denoising of data). In such domains, the discriminator itself may be of central importance as it can potentially estimate the relative density of data objects (see the previous subsection for a discussion of this viewpoint).
By contrast, in applications of primary interest to us, it is the generator that is of primary interest. The core of the generator is a circuit model, developed independently of the method used to fit it (in our case GANs), with the scientific goal of uncovering the circuit mechanisms in a neural or biological system. In particular, in our applications, the generator parameters θ typically correspond to physiological and anatomical parameters with clear mechanistic interpretations. Correspondingly, the circuit model is highly structured, with that structure strongly informed a priori by biological domain knowledge and the scientific desire for parsimony. This allows for post-training tests of the model that go beyond testing the fit to held-out data of the same type as training data, and may not be conceivable in many machine learning applications. For example, one can imagine training an SSN circuit model on size-tuning curves (as we did in Experiment 1–3) but test it using stimuli with varying strengths or contrasts and compare the generated distribution of contrast-response function against data. Alternatively, one can feed the trained SSN dynamical noise and then compare the statistics of temporal neural variability (such as pairwise noise correlations) against empirical data. In such tests of generalization, the data-space itself changes (and not just the data points in it) between training and testing, and therefore a discriminator trained on one space simply cannot generalize to the other; the link between the two data-spaces is solely provided by the generative mechanistic circuit model. The promise of a faithful scientific model of a brain network is, in principle, to capture all such neural data at least approximately; the above scenarios can thus be used as strong tests for such models. To perform such tests faithfully and quantitatively, a strong fitting procedure is necessary.
The model of primary motor cortex (M1) tuning curves proposed by Ref. [4] is a two-layer feedforward network, with an input layer, putatively corresponding to the parietal reach area or to premotor cortex, and an output layer corresponding to M1 (see Fig 2). Ref. [4] introduced their model in two versions, a basic one, and an “extended” version. We have used their enhanced version with small modifications noted in Experiment 1 and below which allow our approach to be used. In particular, we did not model response selectivity to hand posture (supination or pronation), and ignored that label in the dataset; i.e., we blindly mixed hand-position tuning curves across pronation and supination conditions, as if they belonged to different neurons. We further simplified the dataset by removing spatial scale information in the positions of target hand locations, which varied slightly between experimental sessions, by rescaling the distance between adjacent hand position to be 1. We randomly selected half of the hand-position tuning curves to be our training dataset, and used the other half as held-out data to evaluate the goodness of model fits (presented in Fig 3).
The input to the feedforward network is the 3D hand position xs, with s ∈ {1, ⋯, 27} indexing the 3 × 3 × 3 grid of possible target locations. The input layer neurons have Gaussian receptive fields defined on the 3D hand position space. The activation, hi(s) of neuron i in the input layer with receptive field centered at xi is thus h i ( s ) ∝ exp ( - 1 2 σ i 2 ‖ x s - x ¯ i ‖ 2 ) in condition s (when the hand is at xs). Across the input layer, the Gaussian centers xi form a fine cubic grid that interpolates and extends (by 3 times) the 3 × 3 × 3 stimulus grid along each dimension. Whereas Lalazar et al. [4] used a grid with 100 points along each axis, we reduced this to 40 to allow faster computations (we checked that changing this resolution beyond 40 only weakly affects the results). Across the input layer, the receptive field widths, σi, were randomly and independently sampled from the uniform distribution on the range [σl, σl + δσ]. This can be expressed by writing σ i = σ l + z i σ δ σ where z i σ are uniformly distributed on [0, 1] and independent for different i’s across the input layer.
The feedforward connections from the input to output layer are sparse and random, with a connection probability of 0.01. In our implementation of this model, the strength of the nonzero connections were sampled independently from the uniform distribution on the range [0, J]. Since output layer neurons are independent, it suffices to describe the model with a single output neuron. If we denote the connections received by this neuron from the i-th input layer neuron by J i, we can thus write: J i = J M i z i J where z i J’s are sampled independently from the standard uniform distribution on [0, 1], and Mi is a binary 0/1 mask that is nonzero with probability 0.01.
The response of the output layer neuron is given by a rectified linear response function with threshold ϕ. The threshold ϕ was sampled uniformly from the range [ϕl, ϕl + δϕ]. Equivalently, ϕ = ϕl + zϕ δϕ with zϕ a standard uniform random variable.
The model thus has five trainable parameters θ = (σl, δσ, J, ϕl, δϕ) (listed in Table 2). For a choice of θ, the collection of quenched noise variables, z = ( z ϕ , ( z i σ ) i = 1 40 3 , ( z i J ) i = 1 40 3 , ( M i ) i = 1 40 3 ), fully determine the network structure: all input layer receptive field sizes, individual feedforward connection strengths, and output layer neural thresholds for a particular network realization. The response r ¯ ( s ; z , θ ) of an output neuron in condition s (for s ∈ {1, ⋯, 27}) is thus given by
r ¯ ( s ; z , θ ) = [ ∑ i = 1 40 3 J i h i ( s ) - ϕ i ] + (15)
where
h i ( s ) = 1 Z ( s ) exp ( - 1 2 σ i 2 ‖ x s - x ¯ i ‖ 2 ) (16)
J i = J M i z i J (17)
ϕ = ϕ l + z ϕ δ ϕ , (18)
σ i = σ l + z i σ δ σ , (19)
z ϕ , z i J , z i σ ∼ i i d U [ 0 , 1 ] (20)
M i ∼ i i d Bern ( 0 . 01 ) (21)
where [u]+ = max(0, u) denotes rectification, and Z(s) is a normalizing factor such that ∑i hi(s) = 1. Crucially, the network’s output, G θ ( z ) ≡ ( r ¯ ( s ; z , θ ) ) s = 1 27, is differentiable with respect to each component of θ, we can thus use the output gradient with respect to model parameters to optimize the latter using any variant of the stochastic gradient descent algorithm. Note that Eqs (17)–(19) constitute an example of the “sampler function” gθ(z) introduced in the subsection “Mechanistic network models as implicit generative models” of Results (here the vector of synaptic weights J i corresponds to W, while the vector (ϕ, σ) capturing single-cell properties corresponds to γ).
Here we provide the technical details of the simulations, fit and analysis of the Stabilized Supralinear Network (SSN) model of the experiments in Experiment 1–3. The SSN is a recurrent network of excitatory (E) and inhibitory (I) neurons. The dynamical state of the network is the vector of firing rates r(t) of the N network neurons. The rate vector is governed by the differential equation
τ d r d t = - r + f ( W r + F I ( s ) ) , (22)
where W and f denote the recurrent and feedforward weight matrices (with structure described below), the diagonal matrix τ = Diag ( ( τ i ) i = 1 N ) contains the neural relaxation time constants, τi, and I(s) denotes the stimulus input in condition s, with s ∈ {1, …, S}. The key feature of the SSN is the input-output nonlinearity of its neurons, which in the original model is a supralinear rectified power-law function: f ( u ) = k [ u ] + n ([u]+ = max(0, u) and n > 1 and k > 0 are constants).
During the training of the model inside the fitting algorithm, however, the model may explore non-biological regions in parameter space that may lead to divergence of model firing rates. To tame such divergences and enforce numerical stability during training, we modified the neural input-output nonlinearity in the model as follows. We let f(u) be a rectified power-law in the biologically relevant range, but smoothly connected it to a saturating branch at very high rates. More precisely we took
f ( u ) = { k [ u ] + n if u < u 0 r 0 + ( r 1 - r 0 ) tanh ( n r 0 r 1 - r 0 u - u 0 u 0 ) otherwise (23)
where k = 0.01, n = 2.2, r0 = 200 Hz, r1 = 1000 Hz, and u0 = (r0/k)1/n.
In our SSN examples, we chose to have all random structural variability (which is the source of heterogeneity manifesting in tuning curve shapes) occur in the connectivity matrices W and f. As described in Experiment 2 (see Fig 2), we experimented with SSN models with one-dimensional topographic structure, on which the structure of W and f depend. The model has a neuron of each type, E and I, at each topographic spatial location; for M topographic locations, the network thus contains N = 2M neurons. Below, for the i-th neuron, we denote its type by α(i) ∈ {E, I} and its topographic location by xi. We let xi’s range from −0.5 to 0.5 on a regular grid.
The statistical ensemble for W was described in Experiment 2: the random variability of the matrix elements Wij’s was taken to be independent with uniform distribution, with mean and range that depend on the pre- and post-synaptic cell types and topographic distances. More precisely, for each instance of the model we generated W via
W i j = ς b ( J a b < + z i j δ J a b ) exp ( - ( x i - x j ) 2 2 σ a b 2 ) , a = α ( i ) , b = α ( j ) (24)
z i j ∼ i i d U [ 0 , 1 ] (25)
where ςb = 1 or −1 if b = E or I, respectively, and U[0, 1] denotes the uniform distribution on the interval [0, 1]. Thus the average weight is 〈 W i j 〉 = J ¯ a bexp ( - ( x i - x j ) 2 / ( 2 σ a b 2 ) ), where we defined J ¯ a b = J a b < + δ J a b / 2, while the standard deviation SD [ W i j ] = 1 2 3 δ J a b exp ( - ( x i - x j ) 2 / ( 2 σ a b 2 ) ). All parameters, J a b <, δJab, σab were constrained to be non-negative (which is equivalent to the constraints J ¯ a b ≥ δ J a b / 2 ≥ 0 and σab ≥ 0, for the alternative parameterization using J ¯ a b, δJab, σab). The first two constraints (together with the sign variable ςb in Eq (24)) ensure that any realization of Wij satisfies Dale’s principle [39, 40].
We chose the feedforward weight matrix, f, to be diagonal with weights having independent random heterogeneity across network neurons. More precisely, for a network of N neurons, f was an N × N diagonal matrix generated via
F = Diag ( ( 1 + z i F V ) i = 1 N ) (26)
z i F ∼ i i d U ( [ - 1 , 1 ] ) (27)
where the binary random variables z i F are sampled independently and uniformly from [−1, 1].
Our recurrent and feedforward connectivity ensemble is thus characterized by 13 non-negative parameters (enumerated in Table 3): the parameter V that controls the degree of disordered heterogeneity in feedforward weights, as well as the elements of the three 2 × 2 matrices J a b <, δJab, σab (where a, b ∈ {E, I}), which control the average strength, disordered heterogenetity, and spatial range of recurrent horizontal connections. These constituted the parameters
θ = ( J a b < , δ J a b , σ a b , V ) a , b ∈ { E , I } , (28)
that were fit using the WGAN (or moment matching). (Because enforcing the non-negativity constraints on the set ( J a b < , δ J a b , σ a b , V ) a , b ∈ { E , I } is easier than enforcing the constraints on the set ( J ¯ a b , δ J a b , σ a b , V ) a , b ∈ { E , I } introduced in Experiment 2, we used the parametrization of Eq (28) in our WGAN implementation.) While these parameters described the statistics of connectivity, a specific realization of the network is determined by the high-dimensional fixed-distribution random variables of the GAN formalism, z, in addition to θ. The former is composed of the N2 independent, standard uniform random variables ( z i j ) i , j = 1 N, and the N independent random variables ( z i F ) i = 1 N which are sampled uniformly from [−1, 1]. Also note that Eqs (24) and (26) constitute an example of the sampler function, gθ (z), introduced in the subsection “Mechanistic network models as implicit generative models” of Results.
In our example experiments, we simulated the fitting of an SSN model of V1 to datasets of stimulus size tuning curves of V1 neurons [8]. As a simple model of the visual input to V1 evoked by a grating of diameter b, the stimulus input to neuron i was modeled as
I i ( b ) = A σ ( l - 1 ( b / 2 + x i ) ) σ ( l - 1 ( b / 2 - x i ) ) , (29)
where σ(u) = (1 + exp(−u))−1 is the logistic function, A denotes the stimulus strength or contrast. Thus, the stimulus targets a central band of width b centered on the middle of the topographic grid (see Fig 2B). The parameter l determines the smoothing of the edges of the stimulated region. For training the model, we chose the sizes from a set of S = 8 different sizes (0, 1/16, 1/8, 3/16, 1/4, 1/2, 3/4, 1) (measured in units of the total length of the network). Letting bs denote the size in stimulus condition s (s ∈ {1, ⋯, S}), the I(s) of Eq (22) and Experiment 2 is given by I(s) = I(bs), with a slight abuse of notation.
The output of the SSN, considered as a generative model for tuning curves, are the size tuning curves of a subset of network neurons which we call “probe” neurons. We define the tuning curve of these neurons in terms of their sustained responses evoked by different stimuli. Thus given a specific realization of the SSN, for each stimulus s ∈ {1, …, S}, we first calculate the sustained network response vector by the temporal average between t1 and t2 r ¯ ( s ) = 1 T ∑ k = 1 T r ( t 1 + k Δ t ) (30)
where Δt is the Euler integration step and T = (t2 − t1)/Δt. We choose t1/Δt = 200 and t2/Δt = 240 to balance the computational cost and the accuracy for approximating the true steady-state. Given the sustained network response r ¯ ( s ) and an a priori selected set of O probe neurons with indices i = (i1, i2, …, iO) (the probe neurons can equivalently be defined by their types and topographic locations), we define the output of the SSN generative model (the GAN generator) to be the vector
G θ ( z ; c = x i p ) = ( r ¯ i p ( s ) ) s = 1 S . (31)
Here, x i p denotes the topographic location of the p-th probe neuron, and we have now made the dependency of the output on the quenched noise variables, z, and model parameters explicit. We treated the probe neuron’s topographic location, x i p, as the condition c in conditional WGAN (cWGAN). In this paper, we only probed excitatory neurons, i.e., α(ip) = E.
In experimental recordings, typically the grating stimulus used to measure a neuron’s size tuning curve is centered on the neuron’s receptive field. To model this stimulus centering, we always set the first probe neuron i1 to be at the center of the topographic grid (i.e., x i 1 = 0), which was the center of the stimulus. In Experiment 2 we only fit the model to the size tuning curves of “centered” excitatory neurons. Since in that case there was only one probe neuron (or cWGAN condition), we denoted the model output more succinctly by Gθ(z), dropping its second argument. By contrast, in Experiment 3 we included tuning curves of neurons with offset receptive fields (topographic locations) in the training dataset and employed a conditional WGAN.
Note that tuning curves for networks such as the SSN described here which have partially random connectivity, show variability across neurons as well as across different realizations of the network for a fixed probe neuron. When the network size N is large, typically a local self-averaging or “ergodicity” property is expected to emerge: the empirical distribution, in a single network realization, across the tuning curves of neurons of the same type and with nearby topographic locations should approximate the distribution across different z for a neuron of pre-assigned index (i.e., type and location). Although we did not do so in our experiments, one may exploit this ergodicity for more efficient training and testing of the model by sampling multiple nearby sites from each connectivity matrix realization.
Given a dataset of size tuning curves, we would like to find model parameters, θ = ( J a b < , δ J a b , σ a b , V ) a , b ∈ { E , I }, that produce a matching model distribution of tuning curves. In this paper we constructed a simulated training dataset of size tuning curves using a “ground truth” SSN model, with parameters θtruth given in Table 1. All other model parameters were the same between the ground truth and trained SSN models, and had the values: N = 402, k = 0.01, n = 2.2, τE/Δt = 20, τI/τE = 1/2, A = 20, l = 2−5.
During training, according to Algorithm 1, every time Gθ (z; x) was evaluated, we simulated the trained SSN using the forward Euler method for t2/Δt = 240 steps (see also Eq (30)). The gradients of the generator output or the generator loss with respect to parameters θ were calculated by standard back-propagation through time (BPTT). To avoid numerical instability, the parameters J a b < , δ J a b , σ a b were clipped at 10−3 during training. To exclude extremely large (non-biological) values, we also clipped them below 10. We also clipped V to bound it within the interval [0, 1]. (These upper bounds can be thought of as imposed Bayesian priors on these parameters.)
Moreover, during training, the SSN may be pushed to parameter regions in which, for some realizations of the quenched noise variables z, the network does not converge to a stable fixed point. Since an implicit model assumption of the SSN is to model sustained responses by stable fixed points of the model with rates in the biological range, dynamical non-convergennce and very high rates can be (strongly) penalized. We encouraged the firing rate of the all SSN neurons to uniformly remain below a permissive threshold of 200 Hz, by adding the following penalty term to the generator loss
Penalty G ( θ ) = η N T m ∑ j = 1 m ∑ k = 1 T ‖ [ r ( t 1 + k Δ t ; z ( j ) ) - 200 ] + ‖ 1 (32)
where m is the size of the mini-batch in the gradient descent, T = (t2 − t1)/Δt is the time window for calculating the sustained response Eq (30), and η = 100 is the weight of the penalty relative to the WGAN generator loss. This penalty, together with the modified neural input-output nonlinearity descrbied in (Eq (23)), ameliorated diverging solutions of SSN dynamics and the resulting extremely large or small generator gradients.
Once the SSN network produces large output rates, it disrupts the learning in the discriminator. Furthermore, Eq (32) alone can fix such behavior of SSN without relying on the discriminator to learn to adapt to new extremely strong inputs (which are the SSN’s output rates). Thus, we find that it is better to skip such generator output samples for stabilizing learning through the GAN framework. Namely, we update the discriminator parameters for a mini-batch only if
Penalty G ( θ ) < 1 (33)
where the bound 1 is rather arbitrary. We observed that for many successful trainings, such large firing rates never occur.
In order to encourage convergence to a fixed point, we also tried penalizing large absolute values of the time derivative dr/dt in the window [t1, t2]. However, we found empirically that the penalty Eq (32) was sufficient to allow the training algorithms to find parameters for which the network converged with high probability to a fixed point.
We looked at the goodness of fit of model outputs by comparing the distributions of several scalar functions or “summary statistics” of tuning curves. We give the precise definitions of these statistics here.
For the feedforward network model of M1 tuning curves, in Experiment 1 we compared the data and model histograms of four test statistics or measures characterizing the hand-location tuning curves, defined as follows. Let r ¯ ( s ) denote the tuning curve, i.e., the trial average firing rate in condition s, with s ∈ {1, ⋯, 27} indexing the hand location from among the 3 × 3 × 3 cubic grid of target locations in the experiment of Ref. [4]). The average firing rate was simply 1 27 ∑ s = 1 27 r ¯ ( s ). The coding level of a tuning curve was defined as
Coding Level = 1 27 ∑ s = 1 27 Θ ( r ¯ ( s ) - 5 Hz ) (34) i.e., the fraction of conditions with rate larger than 5 Hz (Θ(⋅) denotes the Heaviside function). The R2 denoted the coefficient of determination of the optimal linear fit to the tuning curve, i.e.,
R 2 = 1 - ∑ s = 1 27 ( r ¯ ( s ) - L ( s ) ) 2 ∑ s = 1 27 r ¯ ( s ) 2 (35)
where L(s) = r0 + m ⋅ xs is the optimal linear approximation to r ¯ ( s ) (where r0 and m are the coefficients of the linear regression). Finally, following Ref. [4], the complexity score of a tuning curve was defined as in Eq (36).
Complexity Score = SD [ | r ¯ ( s ) - r ¯ ( s ′ ) | max s ( r ¯ ( s ) ) - min s ( r ¯ ( s ) ) | | x s - x s ′ | = 1 ] (36)
where SD denotes standard deviations.
To quantify the goodness of fit between the outputs of the ground truth and trained SSN models in Experiment 2, we compared the distributions of four test statistics characterizing the size tuning curves: preferred stimulus size, maximum firing rate, the suppression index, and the normalized participation ration, as defined below. While to fit the model we used size tuning curves containing responses to stimuli with S = 8 different sizes, bs, in the set (0, 1/16, 1/8, 3/16, 1/4, 1/2, 3/4, 1), for testing purposes and to evaluate the above measures, we generated tuning curves from the trained SSN using a larger set of stimulus sizes (denoted by b below). Letting r ¯ ( b ) denote the size tuning curve (i.e., r ¯ ( b ) is the sustained response of the center excitatory neuron to the stimulus with size b), the maximum firing rate is maxb r ¯ ( b ), and the preferred size is arg maxb r ¯ ( b ). The suppression index is defined by
Suppression Index = 1 - r ¯ ( max ( b ) ) max b ( r ¯ ( b ) ) ,
and measures the strength of surround suppression. Finally, the normalized participation ratio (related to the inverse participation ratio [59]) is defined by
Normalized Participation Ratio = 1 n b ( ∑ b r ¯ ( b ) ) 2 ∑ b r ¯ ( b ) 2 (37)
and measures the fraction of all tested sizes that elicited responses comparable to the maximum response.
The most studied application of GANs is in producing highly structured, high-dimensional output such as images, videos, and audio. In those applications, mathematical structures such as translational symmetry in the data space (for images) is exploited to design complex and structured discriminators such as deep convolutional networks. It has also been noted that the discriminator network should be sufficiently powerful so that it is capable of fully capturing the data and model distributions [29, 43, 44]. In our application, the outputs of the generator are comparatively lower-dimensional objects, with less complex distributions. Furthermore, developing a new discriminator architecture exploiting mathematical structure in the tuning curve space such as metric and ordering in the stimulus space is beyond the scope of this paper. In this work we used relatively simple discriminator networks. Nevertheless care is needed in designing discriminators; a function D that is too simple can preclude the fitting of important aspects of the distribution. For example, if a linear function D were used in the WGAN approach it would result in a fit that matches only the average tuning curve between model and data, and ignores tuning curve variability altogether.
For the M1 feedforward model, we used a dense feedforward neural network as the discriminator D, with four hidden layers of 128 rectified linear units and a single linear readout unit in the final layer. The discriminator network weights were initialized to uniformly random weights with Glorot normalization [60]. We do not use any kind of normalization or parameter regularization other than the WGAN penalty term in Eq (4) (i.e. we set PenaltyG(θ) to zero in this example).
For the SSN recurrent model, we used dense feedforward neural networks with four hidden layers and with layer normalization [61] as recommended for WGAN in Ref. [26]. The discriminator network used in the experiments of Figs 4 and 5 had 128 and 64 neurons in each hidden layer, respectively. In the training experiments underlying Fig 6, we used networks with 32 to 128 neurons in each layer; as indicated by the WGAN histogram, all choices consistently performed well. We note that simple dense feedforward neural networks without normalization do not work well for SSN due to numerical instability in long-running training. It was important, however, not to apply the layer normalization in the first (input) layer, as the mean and variance across stimulus parameters of the turning curves are valuable information for the discriminator which would be discarded by such a normalization. We also used weight decay [62] for all parameters to stabilize the learning. We initialized all neural biases to 0 and initialized all weights as independent standard normal random variables, except in the input layer. For the input layers, we used the same initialization as in the M1 feedforward model’s discriminator.
To provide a benchmark for our proposed GAN-based method, we also fit the SSN using moment matching [63]. We define a generic moment matching loss as
L 0 ( θ ) = 1 D ∑ d = 1 D [ w 1 , d ( m d ( θ ) - μ d ) 2 + w 2 , d ( s d ( θ ) - σ d ) 2 ] .
Here, D = S × O, where S is the total number of stimulus conditions and O the number of neurons whose firing rates are probed in the SSN, and d indexes the combination of stimulus condition and probe neuron. md(θ) and sd(θ) are the mean and variance of response in combination d, across a mini-batch of 32 model-generated sample tuning curves, respectively, while μd and σd are the empirical mean and variance of this response, across the full training dataset of 2048 size tuning curves. The wi,d are the weights given to each moment, and are hyperparameters of the moment matching method. We tried the following options
uniform scaling : w 1 , d = ( 1 D ∑ c = 1 D μ c ) - 2 ,w 2 , d = ℓ w 1 , d 2 , (38)
element-wise scaling : w 1 , d = ( μ d + ε ) - 2 , w 2 , d = ℓ w 1 , d 2 , (39)
relative scaling : w 1 , d = ( μ d + ε ) - 2 , w 2 , d = ℓ ( σ d + ε ) - 2 , (40)
where ℓ controls weight of the variance with respect to the mean and ε = 10−3 is a regularization constant to avoid division by zero. In our preliminary experiments, we found that element-wise Eq (39) and relative Eq (40) scalings are better than uniform scaling Eq (38). Thus, we only used element-wise Eq (39) and relative Eq (40) scalings in results presented here. For fitting the SSN, the same reasons for encouraging dynamical stability as in the WGAN case hold. Thus, we added the penalty term as defined in Eq (32) and minimized the loss
L ( θ ) = L 0 ( θ ) + Penalty G ( θ ) . (41)
For most of the trainings, large firing rates yielding PenaltyG(θ) > 0 rarely occured. The generator parameters are updated using Adam (a variant of stochastic gradient descent) with the hyperparameters β1 = 0.5, β2 = 0.9, ϵ = 10−8 as in Algorithm 1. We use the learning rate 0.001 unless specified.
To compare learned results between the GAN and moment matching and across different hyperparameters on an equal footing, we defined a condition for terminating learning as follows. First, we stop training at the first generator update step, n0, in which the speed-of-change of the generator parameters, as evaluated by
‖ 1 ν ∑ n = n 0 - ν + 1 n 0 θ ( n ) - θ ( n - κ ) κ ‖ 1 (42)
becomes smaller than a tolerance threshold of 0.01. Here θ(n) is the vector of generator parameters at the nth generator update, κ controls the timescale at which the speed is computed, and ν is the size of the moving average window. ‖x‖1 denotes the L1-norm of the vector x, i.e., the sum of absolute values of its components; thus the condition Eq (42) ensures that the speed-of-change of all model parameters are small, i.e., they have approximately converged. To obtain the final result (estimated parameters) of learning, we then compute the average of the generator parameter θ(n) in the ω steps leading to step n0 θ ^ = 1 ω ∑ n = n 0 - ω + 1 n 0 θ ( n ) . (43)
For comparing results across different generator learning rates, αG, we used κ = ν = ω = α G - 1.
As the metric of performance, we use the so-called symmetric mean average percent error (sMAPE) of the estimated generator parameters θ ^ relative to the ground truth parameters θtruth which were used to generate the training dataset. That is we let
sMAPE ( θ ) = 100 % 1 13 ∑ i = 1 13 | θ ^ i - θ i truth | | θ ^ i + θ i truth | / 2 (44)
where 13 is the number of parameters. The performance in any learning run is then computed by the sMAPE ( θ ^ ) of the final result, Eq (43), of that run obtained using the above termination procedure Eq (42). Note that this termination criterion is used only for Fig 6 and not in Fig 5, where we plotted the learning curve over a broader range for demonstration purposes.
For the WGAN-based fits shown in Fig 6 we picked combinations of hyperparameters from the following choices: the generator learning rate was αG = 10−4, 2 × 10−4, or 4 × 10−4, the number of neurons in each discriminator hidden layer was 32, 64 or 128, discriminator update rule was Adam or RMSprop. For the moment matching fits the hyperparameter choices were: the learning rate was 10−4, 2 × 10−4, 4 × 10−4, 10−3, 2 × 10−3, or 4 × 10−3, the weight of variance λ = 0.01, 0.1, 1, and the moment scaling was element-wise Eq (39) or relative Eq (40).
We implemented our GAN-based method and moment matching in Python using Theano [64] and Lasagne [65]. Our implementation is available from https://github.com/ahmadianlab/tc-gan under the MIT license.
|
10.1371/journal.pgen.1004855 | Systematic Cell-Based Phenotyping of Missense Alleles Empowers Rare Variant Association Studies: A Case for LDLR and Myocardial Infarction | A fundamental challenge to contemporary genetics is to distinguish rare missense alleles that disrupt protein functions from the majority of alleles neutral on protein activities. High-throughput experimental tools to securely discriminate between disruptive and non-disruptive missense alleles are currently missing. Here we establish a scalable cell-based strategy to profile the biological effects and likely disease relevance of rare missense variants in vitro. We apply this strategy to systematically characterize missense alleles in the low-density lipoprotein receptor (LDLR) gene identified through exome sequencing of 3,235 individuals and exome-chip profiling of 39,186 individuals. Our strategy reliably identifies disruptive missense alleles, and disruptive-allele carriers have higher plasma LDL-cholesterol (LDL-C). Importantly, considering experimental data refined the risk of rare LDLR allele carriers from 4.5- to 25.3-fold for high LDL-C, and from 2.1- to 20-fold for early-onset myocardial infarction. Our study generates proof-of-concept that systematic functional variant profiling may empower rare variant-association studies by orders of magnitude.
| Exome sequencing has proven powerful to identify protein-coding variation across the human genome, unravel the basis of monogenic diseases and discover rare alleles that confer risk for complex disease. Nevertheless, two key challenges limit its application to complex phenotypes: first, most alleles identified in a population are extremely rare; and second, most alleles are neutral on protein activities. Consequently, association tests that rely on enumerating rare alleles in cases and controls (termed rare variant association studies, RVAS) are typically underpowered, as the many neutral alleles dampen signals that arise from the few alleles that disrupt protein functions. Strategies to securely discriminate disruptive from neutral variants are immature, in particular for missense variants. Here we show that the statistical power of RVAS improves dramatically if variants are stratified according to their in vitro ascertained functions. We establish scalable technology to objectively profile the biological effects of exome-identified missense variants in the low-density lipoprotein receptor (LDLR) through systematic overexpression and complementation experiments in cells. We demonstrate that carriers of LDLR alleles, which our experiments identify as “disruptive-missense”, have higher plasma LDL-C, and that considering in vitro data may make it possible to reduce RVAS sample sizes by more than 2-fold.
| The rate by which sequencing studies in humans are unraveling genetic variants far outweighs our ability to accurately evaluate which of these variants are of the highest relevance to human health and disease [1]. This interpretative gap is considered a key impediment for the wider use of genetics in clinical medicine [2–4], as it challenges sequencing-based diagnoses [5–7] and risks misguiding medical interventions or reproductive decisions [8]. It further limits the statistical power of sequencing studies in families or populations that aim to identify novel disease genes [9, 10].
The vast majority of rare protein-coding alleles are considered to be neutral, i.e., they have no or little impact on disease liabilities. Importantly, this overabundance of neutral compared with damaging alleles creates a tremendous signal-to-noise problem for rare-variant association studies (RVAS) [10] that rely on the aggregation of all or distinct classes of rare variants at the gene level [11]. RVAS have recently allowed us to identify rare variation in the low-density lipoprotein receptor (LDLR) as associated with early-onset myocardial infarction (MI) in the population [12]. Importantly, however, association signals were driven by loss-of-function (LoF) alleles that based on sequence could be unambiguously interpreted as protein-inactivating, including nonsense, splice-site or indel frameshift alleles. Carriers of LoF alleles in LDLR showed an 18.1-fold increased MI-risk as opposed to an only 1.7-fold increased risk in carriers of missense alleles. As missense variants by far outnumber LoF variants across human genes [12–14], it has been hypothesized that including disruptive-missense (i.e., missense variants that disrupt protein functions in the range of LoF variants, “missense LoF”), while ignoring neutral alleles should considerably enhance the association signal and reduce the necessary samples sizes needed to demonstrate association, by on average 2.5-fold [10]. However, missense variants are the most difficult class of variants to adequately predict a biological function [15], particularly in genes under selective pressure like LDLR where the rate of neutral relative to disruptive-missense alleles is expected to be high [10].
Deleterious variation in LDLR is kept at low frequency as heterozygote carriers of mutant alleles show familial hypercholesterolemia (FH), characterized by a 2–3 fold elevation of plasma low-density lipoprotein cholesterol (LDL-C) and premature coronary artery disease [16]. Among Europeans, 4–5% of individuals who suffer from MI before the age of 60 are FH heterozygotes [17]. LDLR is also paradigmatic for a dose-response relationship between gene and function as homozygotes are more severely affected than heterozygotes, and mutations that impair, but not completely abolish receptor activity tend to result in more moderately increased LDL-C, later onset MI and better response to therapies [16]. Mutations can impact different activities of the LDLR protein, including its biosynthesis, subcellular trafficking and capacity to bind and internalize LDL [18], yet biochemical tests to characterize FH mutants are low-throughput and not applied routinely in clinical care [19]. Importantly, LDLR is one of 56 genes in which the incidental detection of known or novel variants is recommended for subsequent medical clarification [20].
Here we establish an experimental strategy to systematically characterize the biological functions of missense alleles identified through exome analysis of large clinical cohorts. We demonstrate at the case of LDLR and MI that a combination of sequencing with systematic variant-profiling in vitro markedly improves the statistical power of RVAS.
With the aim to identify rare missense alleles in LDLR that increase the risk for premature MI, we leveraged the exomes of 1,716 cases with MI prior to age of 46 and 1,519 MI-free controls [12] (see Fig. 1 for workflow of this study). Overall, 194 subjects carried rare LDLR alleles that distributed on 12 clear LoF and 70 missense variants (S1 Table, Methods and S1 Spreadsheet). The burden of LoF alleles associated rare variation in LDLR with LDL-C and MI-risk at genome-wide significance (p<1×10-8) [12]. However, the more abundant missense alleles alone or in combination with LoF variants considerably deflated association signals (e.g., for LDL-C from odds ratio (OR)=34.4 to 3.2 and 4.5, respectively) (Table 1, Tables S2–3). This is consistent with a scenario where the signal of alleles that disrupt LDLR activity—LoF alleles together with missense alleles of a similar impact as LoF alleles (termed “disruptive-missense” alleles)—is swamped by the noise of neutral alleles. A-priori information to separate between these two groups is scarce as an overlap of four frequently used computational prediction tools assign equal proportions of LDLR missense alleles as damaging (51%) and likely benign (49%), respectively (S1 Table). Moreover, the rate of unique alleles (61%) in the studied at-risk cohort matches that of non-MI reference cohorts (S1 Fig.), which further complicated identification of disruptive-missense alleles from sequence data alone.
In order to distinguish disruptive from non-disruptive LDLR missense alleles, we established a workflow to profile the function of missense alleles in an unbiased, quantitative and high-throughput manner in vitro. For this, we applied two complementary experimental strategies: first, an “overexpression” approach where wildtype or mutated LDLR-GFP was transiently expressed in cultured cells; and second a “complementation” approach where the endogenous receptor was silenced with LDLR-siRNA, but receptor activities were reconstituted by co-expressing siRNA-resistant wildtype or mutated LDLR-GFP (S2 Fig. and Methods). Since we assumed that complementation might have the potential to unmask effects that fail to be identified by testing overexpression alone, both approaches were applied in parallel. The efficiency of LDL-uptake into GFP-positive and GFP-negative cells was quantified by multiparametric analyses from images acquired using high-content automated microscopy as described [21, 22] (S3 Fig. and Methods). Expectedly, wildtype LDLR stimulated LDL-uptake, as evidenced by an increased internalization of fluorescent-labeled LDL into endosome-like compartments (Fig. 2A). This effect vanished when LDLR carried the transport–deficient FH-mutation p.G549D [18] that mislocalized the receptor to endoplasmatic reticulum (ER)-like membranes, or the internalization-deficient “JD”-mutant p.Y828C [23] that arrested both, ligand and receptor at the plasma membrane. Multiparametric analysis of the phenotypes obtained from a large number of cells (Fig. 2B,C) demonstrated that our approach could identify and correctly describe functions of previously known LDLR missense variants causing FH.
We applied this workflow to systematically test which of the rare LDLR missense alleles revealed by exome sequencing of our large population cohort disrupted LDLR function. Systematic experimental analyses of LDL-uptake into cells assigned each missense variant a distinct phenotypic profile that enabled conclusions on its mechanisms (Fig. 3A; S4 Fig. and S1 Spreadsheet). Results from overexpression and complementation correlated well (for instance, r2 = 0.56 for parameter “total LDL signal”; Fig. 3B; S4 Table), thus validating most of each other’s findings. Overall, 14 missense variants strongly inhibited LDLR function, typically by reducing LDL-uptake to 6–31% of the wildtype receptor, and were classified as “disruptive-missense”. As an independent validation, we measured whether these variants also impacted total cellular levels of free cholesterol, another phenotype that we have previously shown to vary dependent on LDLR activity [22]. Indeed, all but one disruptive-missense variant not only reduced LDL-uptake, but also free cholesterol levels to less than 50% of controls (Fig. 3C; S5 Table). The only non-validated disruptive-missense variant p.D472Y, as well as two transport-inhibiting ER-associated mutants (p.N316S; p.P526S) reduced LDLR’-GFP protein expression, which indicated an impact on either LDLR biosynthesis or turnover. Like most known FH mutants [18] the majority of disruptive-missense variants clustered in the apoB-ligand binding domain of LDLR and was completely or partially retained in ER-like membranes (Fig. 3D; S5 Fig.). Another 10 variants were defined as of “unclear” functional significance, as they met some, but not all required significance criteria (see Methods). The remaining 46 variants were classified as “non-disruptive”.
We next compared our in vitro results to plasma LDL-C levels available for 2,152 of the individuals in our studied cohort. For 20 variants previously listed in four LDLR locus-specific databases as either causing FH or neutral, experimental data matched with clinical interpretation in 95% of cases (S6 Table and Methods). Importantly, plasma LDL-C was significantly higher in disruptive-missense (221mg/dl) than in non-disruptive (154mg/dl; p<1.36×10-5) and intermediary to LoF LDLR allele carriers (275mg/dl) (Fig. 4A; relative to 135mg/dl in individuals with two wild-type LDLR alleles [12]). As discussed further below, only few carriers of a respective variant class showed LDL-C levels outside the expected range.
These results demonstrated that our strategy efficiently enriched for FH alleles and suggested that considering experimental data might also enhance rare-variant association testing. For this, disruptive-missense alleles were enumerated in cases and controls across the entire cohort (Fig. 4B,C) and tested for association with LDL-C and MI. Indeed, collapsing only disruptive-missense (instead of all LDLR missense) alleles strongly increased odds ratios from 3.2 to 18.6 for association with LDL-C, and from 1.9 to 12.1 for association with MI-risk (Table 1, Tables S2–3). Enumerating disruptive-missense together with LoF variants firmly established rare variation in LDLR as associated with plasma LDL-C (p<6×10-19; OR = 25.3) and MI-risk (p<2×10-10; OR = 20.0) on the population level. Consistent with a theoretically predicted 2.2- to 3-fold reduction in the number of samples needed to be sequenced [10], power simulations suggested that through integration of experimental data sequencing of only 1,200–1,400 (instead of 3,000–4,000) cases and controls would be sufficient to associate rare variation in LDLR with MI-risk at genome-wide significance (Fig. 4D). Notably, experimental data empowered RVAS considerably more than functional prediction tools that correctly evaluated all 14 disruptive-missense variants as damaging, yet consistent with previous observations [24] showed higher type-I-error rates (Table 1; S1 Table; S7 Table and Methods).
Most missense alleles identified in sequencing studies are rare. At limited sample sizes RVAS thus typically fall short on clarifying by how much any individual rare variant contributes to a complex trait [10]. Conversely, one advantage of in vitro studies is that once a variant has been observed in a population, variant frequencies do not matter. We aimed to test whether experimental data could support genetics also for single variant association analyses. In order to increase the number of observations per variant, we analyzed the function of 16 LDLR missense alleles that are represented on the Illumina HumanExome v1.0 SNP array (“exome-chip”) and that were genotyped in 39,186 individuals characterized for LDL-C (Fig. 5; S7 Table and S1 Spreadsheet) [25]. Overall, effect sizes between genotyping and in vitro experiments correlated well (r2 = 0.45). Importantly, the variants with the highest beta (p.E101K, p.P685L) most pronouncedly inhibited LDL-uptake in cells, supporting our hypothesis that systematic experimental data will not only be informative for gene-burden analyses, but also in clarifying by how much individual rare and low-frequency variants contribute to genetic etiologies.
Our study demonstrates that distinguishing disruptive from non-disruptive missense alleles in a well-described disease gene (LDLR) through systematic functional characterization in vitro can further our understanding how rare, potentially damaging genetic variation contributes to common, complex (hypercholesterolemia; MI) as well as Mendelian disease (FH). Thus far, the role of cell-based experiments in human genetics has either been to validate assumed associations between one to few variants and disease, or to better comprehend the mechanisms why variants firmly identified through genetics are pathogenic [2]. Conversely, our study, together with few previous studies [24, 26, 27], predicts that soon unbiased experiments will attain a much more central role in human genetics that could extend to the very core of disease gene discovery.
Optimizing RVAS by stratifying missense alleles according to their in vitro ascertained functions may prove especially powerful to identify and validate genes under a high selective pressure where disruptive-missense are swamped by neutral alleles and sample sizes needed for association become enormous [10]. For LDLR, as a gene with an average endogenous mutation rate, Zuk et al. [10] estimated 17% of missense variants as being disruptive, which is well in line with the 20% we identified experimentally. On the other hand, our strategy may be less amenable to very essential genes where modulation of cellular levels by overexpression or knockdown is less well tolerated. Also, sensitivity of our approach may be limited for genes where the correlation between measured phenotype and gene function is less direct than between LDLR levels and LDL-uptake, or where the odds ratios of even disruptive alleles are small.
For LDLR, our binary classification of alleles as either disruptive or non-disruptive simplifies the range of functional consequences that missense variants can exert on receptor activities [16, 18]. For instance, the inclusion of only disruptive variants for association testing neglects hypomorphic variants that reduce LDLR activity by only few percent. In our study, this is evidenced by slightly elevated odds ratios also in non-disruptive allele carriers. It thus can be expected that through segregation analyses in families, or through more sensitive in vitro readouts, several such alleles will be identified as FH mutants in the future. Although the individual effect of hypomorphic alleles on LDLR activity may be small and, consistent with previous assumptions [10], they in sum add only little power to association tests, future RVAS may profit from counting in also hypomorphic alleles in form of adjusted functional weights.
An intriguing hypothesis is that in addition to rare variation in LDLR, further genetic or environmental factors contribute to increase LDL-C in some carriers of alleles that in our experiments scored as non-disruptive. However, a thorough analysis of known common and rare genetic risk factors from the exomes of 23 individuals with plasma LDL-C levels that did not match expectations from our in vitro analyses did not reveal clear evidence for epistatic effects (see paragraph Search for reasons of aberrant LDL-C in LDLR missense allele carriers in Methods). More carriers of the identical rare alleles, or an even stronger relationship between genetic variant, intermediate and clinical phenotype than between LDLR, LDL-C and MI are needed to exploit the full spectrum of information available from large-scale sequencing studies. Moreover, relationships between in vitro ascertained function and in vivo phenotypes are likely to improve further when the analyzed cohorts can be stratified for important confounders, here, for instance, intake of LDL-lowering medications [28], which was unavailable for this study.
For Mendelian genetics it is worthwhile to note that seven of the variants analyzed here have recently been observed incidentally through clinical exome sequencing of individuals [29, 30] and are listed as potentially requiring medical intervention [20]. Interestingly, however, based on our in vitro studies none of these variants is a strong candidate for causing FH. A more comprehensive annotation of important disease genes through studies like ours together with family-based segregation analyses may help to considerably precise health risks in the future. Through generating scalable cell-based assays for relevant intermediate phenotypes and statistical tools that better incorporate genetic with heterogeneous functional datasets, we expect that composite sequencing-biological studies will become invaluable to human genetics in order to face the flood of novel variants from the ever increasing number of sequenced genomes.
Study cohorts. The Italian Genetic Study of Early-onset Myocardial Infarction (ATVB) is a European case-control collection designed to study the genetics of MI-susceptibility [12, 31, 32]. Exome-sequenced MI cases (n = 1,716) include survivors of a first acute myocardial infarction (defined as more than 30min resting chest pain accompanied by typical ECG and serum abnormalities) at an age of less than 46 years with angiographically documented coronary artery disease. Exome-sequenced MI controls (n = 1,519) were matched for age, sex, and geographical origin and assessed for further MI-risk factors (S10 Table). Principle component analyses did not indicate selection bias between cases and controls (S7 Fig.). For 2,152 subjects (66.5%), plasma low-density lipoprotein cholesterol (LDL-C) at enrollment was available, among them 1,184 MI cases and 968 MI controls. Overall, 251 subjects showed hypercholesterolemia defined as LDL-C above 190mg/dl (4.91mmol/l) (LDL cases) and according to Simon Broome criteria [19, 33] a high likelihood for FH. For 1,901 subjects LDL-C was in the normal range or only moderately elevated (<190mg/dl; LDL controls). As expected, high LDL-C was strongly associated with increased MI-risk in this cohort [12].
Genotype data were obtained from a meta-analysis of 39,186 independent samples characterized with the Illumina HumanExome v1.0 SNP array (“exome-chip”). Samples were from individuals of European ancestry derived from 25 studies on the impact of rare and low-frequency coding variation on plasma lipids [25].
Ethics statement. All analyses in this study conformed to the ethical guidelines of the 1975 Declaration of Helsinki in its crespective latest version. The study has been approved by an IRB from the Broad Institute under protocol number 2013P001840.
Exome sequencing and exome-chip genotyping. Exome sequencing was performed at the Broad Institute Genomics Platform as described previously [34]. Details on all specific steps for reliable variant calling from raw sequence or exome-chip data, as well as performed quality controls for the cohorts used in our study are provided in Do et al. [12] and Peloso et al. [25].
LDLR gene variant selection. LDLR nomenclature throughout the manuscript relates to Homo sapiens low density lipoprotein receptor (LDLR) transcript variant 1 (NM_000527.4; ENST00000558518/ Ensembl73) encoding a protein of 860 amino acids. Overall, 79 DNA sequence variants in LDLR were functionally characterized in this study (S7 Table and S1 Spreadsheet) out of which 78 were identified through exome sequencing and/or exome-chip profiling and one (p.Y828C) was selected from the literature. Based on available biochemical and clinical information, two FH-mutants with firmly established pathogenic mechanisms were chosen as controls, p.G549D [FH Genoa] as example for a transport-inhibiting (class-2) mutant [18] and p.Y828C [FH JD-Bari] that prevents association of LDLR with clathrin-coated pits and its internalization into the endosomal system (class-4) [18, 35]. Exome sequencing of the ATVB cohort [12] identified a total of 82 rare coding variants in LDLR, distributing on 194 alleles. Of these variants, 12 were clear loss-of-function (LoF), causing in 8 cases introduction of a preterm stop codon (p.Q33*; p.Q102*; p.E140*; p.C155*; p.R350*; p.Y419*; p.W533*; p.Q770*) and in 4 cases disruption of splice-donor sites (19:11213463_G/A; 19:11224126_G/A; 19:11224439_G/A; 19:11227676_T/C; NCBI37). Consistent with markedly reduced LDLR activity, LoF variants strongly associated with plasma LDL-C (Table 1; Fig. 4A; S3 Table and [12]) and were omitted from cell-based studies. All 70 ATVB LDLR missense variants were selected for in vitro functional characterization, and 69 comprehensively profiled as described below (with the exception of p.V859M that due to its localization at the LDLR carboxy-terminus failed repetitive cloning attempts). Forty-three (61%) of these missense variants were present only once among the 6,470 ATVB chromosomes, corresponding to a minor allele frequency (MAF) of 1.5×10-4. Twenty-five variants occurred in 2–7 study participants, and two variants in 19 and 40 subjects, respectively (S1A Fig.). Apart from p.T726I with a MAF of 0.00618, all variants fulfilled our definition of being rare by showing a MAF of less than 0.005, corresponding to one heterozygote carrier per 100 study participants. LDLR variants identified in the ATVB cohort were complemented by 16 variants represented on the Illumina HumanExome vs1.0 SNP array that were identified by genotyping 39,186 European subjects from diverse studies characterized for plasma LDL-C [25]. Seven variants (p.R237H; p.G269D; p.E277K; p.G592E; p.E626K; p.P685L; p.R744Q) overlapped between both studies. Frequency distributions of LDLR coding variants among participants of the NHLBI exome sequencing project (ESP) (6,823 individuals; 13,646 chromosomes) (S1B Fig.) were downloaded from the Exome Variant Server (http://evs.gs.washington.edu/EVS/; accessed October 2014).
Locus-specific a priori information. For all 79 variants that underwent functional characterization in this study we systematically searched for availability of a priori clinical or functional information. For this, four public databases retaining locus-specific information on variation in LDLR were queried: the Universal LDLR mutation database (http://www.umd.be/LDLR/) [36]; the LDLR LOVD database at University College London (http://www.ucl.ac.uk/ldlr/) [37]; the NCBI ClinVar database (http://www.ncbi.nlm.nih.gov/clinvar) [38]; and the Human Gene Mutation Database (professional version) (www.hgmd.org) [39]. Information from 111 publications that these databases referred to (S6 Table and Supplemental References) allowed us to classify 19 LDLR variants as either previously validated FH mutant (n = 7), likely benign (n = 5), or of unclear disease relevance (n = 7; including variants identified in compound-heterozygous individuals in combination with a clear FH mutation). All but one FH mutant (p.V523M [FH-Kuwait] that in homozygous fibroblasts was reported as associated with 12–25% residual LDLR activity [40]) met our criteria for being “disruptive-missense” (see below). Except for one variant (p.D118Y) for which disease relevance also after in vitro functional testing remained unclear, all other previously observed variants were classified as non-disruptive. Of four additional variants that in the LDLR LOVD database were listed as FH, but that had not previously been validated in vitro, only one variant (p.C222Y) met our criteria as disruptive-missense. Of 56 variants that were listed in HGMD with the phenotype hypercholesterolemia, yet without functional evidence for this, our analyses classified 13 as disruptive-missense.
Comparison to bioinformatics prediction tools. For each missense variant we determined its likelihood to interfere with LDLR protein activity by applying four commonly used in silico functional predicition tools under default settings: PolyPhen-2 (http://genetics.bwh.harvard.edu/pph2) [41], SIFT (http://sift.jcvi.org) [42], MutationAssessor (http://mutationassessor.org) [43] and MutationTaster (http://www.mutationtaster.org) [44] (S1 Table). Different result categories of each algorithm were assigned distinct numerical values (PolyPhen-2: damaging/probably damaging,-1; possibly damaging,0; benign,+1; SIFT: damaging,-1; tolerated,+1; MutationAssesor: high/medium:,-1; neutral/low,+1; MutationTaster: disease-causing,-1; polymorphism,+1). A summed composite score was calculated for each variant from the overlap of all four prediction tools. A composite score of more than 1 was considered as likely benign, of 0 as unclear and of less than-1 as likely FH. Overall, bioinformatics prediction tools classified 40 of the 79 studied LDLR missense variants (51%) as FH-like, 7 (9%) as of unclear disease relevance and 32 (40%) as likely benign (S7 Table).
Association testing. Rare variant association tests were performed by enumerating all rare LDLR alleles of a distinct class (clear LoF; all missense; bioinformatically predicted as damaging; disruptive-missense; non-disruptive; and unclear) and by calculating association of the burden of variants in cases and controls with plasma LDL-C and MI using Fisher’s exact test (see also [12]) (Table 1, S2 Table). To estimate effect sizes (beta) for continuous levels of LDL-C in the ATVB cohort (S3 Table), linear regression analysis was performed with LDL-C (in mg/dl) as outcome variable, carrier status as independent variable, and sex and age as covariates.
Power calculations for LDLR rare variant association with MI. Based on sequence data from 3,325 ATVB participants, we performed sample size extrapolations for association signals driven by the burden of rare LDLR variants of either LoF variant carriers alone, or LoF variant carriers combined with carriers of variants identified as disruptive-missense. The relative risk of a mutation carrier was assumed to be 5.0. Prevalence of MI was assumed as 0.05. Case:Control ratio was assumed as 1. The number of rare variants was extrapolated into 500,000 individuals. One thousand simulations were performed at a given sample size with intervals of 200 samples (from n = 0–2,000), 400 samples (from n = 2,000–4,000) and 2,000 samples (from n = 4,000–20,000). Power reflects the percentage of simulations that reached genome-wide significance (set at 2.5×10-6 to account for testing of ~20,000 genes) at a given number of samples.
Cells and reagents. HeLa-Kyoto cells and their suitability for measuring the dynamics of LDL-uptake and cellular levels and distribution of free cholesterol (FC) were described in our previous studies [21, 22]. DiI-LDL (Life Technologies), Filipin III (Sigma), Draq5 (Biostatus), Dapi (Hoechst), 2-hydroxy-propyl-beta-cyclodextrin (HPCD) (Sigma), Lipofectamine 2000 (Invitrogen) and Oligofectamine (Invitrogen) were purchased from the respective suppliers.
cDNA cloning, siRNAs and site-directed mutagenesis. A sequence-verified cDNA-clone encoding full-length human LDLR carboxy-terminally linked to EGFP was described previously to adequately reflect activities of the wild-type receptor [22]. To guarantee knock-down of the mRNA encoding the endogenous receptor, but not the heterologously expressed LDLR-GFP cDNA during complementation experiments, three silent mutations (c.A1053G, c.C1056T and c.A1059G) were introduced at Wobble-bases within the 19-nucleotide consensus sequence (CAGCGAAGATGCGAAGATA) of LDLR-siRNA s224006 (Applied Biosciences) by site-directed mutagenesis (see below) using the following primer sequences: 5'-ctggtggcccagcgaaggtgtgaggatatcgatgagtgtca-3' (forward) and 5'-tgacactcatcgatatcctcacaccttcgctgggccaccag-3' (reverse). LDLR-siRNA efficiently reduced levels of the endogenous LDLR mRNA by ~30% and of the endogenous protein by ~75%, respectively, significantly reduced cellular LDL-uptake [22] and abrogated expression of LDLR-GFP. In contrast, levels of the siRNA-resistant LDLR-GFP construct (termed LDLR’-GFP) were unaffected by siRNA-treatment (S2A Fig. and [22]). Subcellular distribution and effect upon overexpression and complementation on DiI-LDL uptake were indistinguishable between LDLR-GFP and LDLR’-GFP (Fig. 2A; S2B Fig. and [22]). LDLR’-GFP served as a template for introduction of studied missense variants using QuikChange Lightning Site-directed mutagenesis kit (Agilent) according to the manufacturer’s instructions. Oligonucleotides for generating distinct LDLR variants were designed using the QuikChange Primer design tool (Agilent), ordered from Metabion (Martinsried, Germany) and are listed in S11 Table. During complementation experiments, siRNA s229174 (Silencer Select, Applied Biosystems) served as a non-silencing control siRNA.
Overexpression, complementation and biological assays. For overexpression analyses, cells were seeded on glass coverslips in 12-well plates (Corning) at a density of 4×104 cells/well, cultured in DMEM (PAA)/2mM L-glutamine/10% FBS (Biochrom) for 24h at 37°C/5% CO2, and fluid-phase transfected with 2μg cDNA/well using Lipofectamine2000 (Invitrogen) according to manufacturer’s instructions. Assays to monitor cellular uptake of fluorescently-labelled LDL (DiI-LDL) were performed as described in more detail in a previous publication [21]. In brief, cells cultured in serum-free medium and exposed to 1% 2-hydroxy-propyl-beta-cyclodextrin for 45min were labelled with 50μg/ml DiI-LDL (Invitrogen) for 30min at 4°C. DiI-LDL uptake was stimulated for 20min at 37.5°C before washing off non-internalized dye for 1min in acidic (pH 3.5) medium at 4°C, fixation, and counterstaining for nuclei (Dapi, Draq5) and cell outlines (Draq5). For quantification of cellular cholesterol, cells were stained with 50μg/ml Filipin III in PBS (from a stock-solution of 1mg/ml in di-methyl-formamide), fixed, and counterstained with cell and nuclear marker Draq5. For complementation experiments, cells were seeded at an identical density, cultured in DMEM (PAA)/2mM L-glutamine/10% FBS (Biochrom) for 24h at 37°C/5% CO2, and fluid-phase transfected with 0.5μl/well of 30μM LDLR-siRNA (s224006) or non-silencing control siRNA (s229174) for 24h using Oligofectamine according to manufacturer’s instructions. One day after siRNA transfection, cells were co-transfected with GFP-cDNAs using Lipofectamine2000 as described above, and cultured for another 24h before biological assays were performed and samples were prepared for microscopic analysis. Overexpression experiments were performed in 3–5, rescue experiments in 1–6 biological replicates per variant. Images were acquired automatically with identical baseline settings from 30 different positions/sample on an Olympus IX81 automated microscope using an UPlanApo 20×0,7NA objective and ScanR software vs. 2.1.0.15 (Olympus Biosciences).
Image data analysis. All images were visually quality controlled using Image J 1.46r (Wayne Rasband, National Institutes of Health, Bethesda) in order to exclude pictures of insufficient technical or biological quality (e.g., due to image acquisition out of focus or aberrant cell density). Biological replicates for each variant analyzed were compared to several controls present during each individual experiment. Each overexpression experiment included wild-type LDLR’-GFP as a positive control as well as two negative controls, i.) a sample where cells expressed a construct encoding only EGFP without the receptor protein (“GFP-control”) and ii.) a sample where cells were exposed only to transfection reagents, but not cDNA (“transfection-control”). Each complementation experiment included four controls: cells transfected either with i.) LDLR siRNA or ii.) negative control siRNA, but no cDNA, as well as two samples where LDLR siRNA-treated cells were co-transfected with either iii.) LDLR’-GFP or iv.) GFP-control cDNAs. Images were analyzed with customized pipelines based on Cellprofiler 2.0 software (http://www.cellprofiler.org) [45]. Analysis strategy was adjusted from [22] and is outlined in S3 Fig. In brief, outlines of individual cells were approximated by stepwise dilation of masks generated from images of Draq5 and/or Dapi (for LDL-uptake) stained cell nuclei. Mean cellular GFP signal (“GFP-expression”) was quantified from background-subtracted images within areas defined as cells. Filipin (for FC) or DiI-signal (for LDL-uptake) was quantified from background-subtracted images within masks that reflected distinct intracellular compartments resembling endosomes (for LDL-uptake) or lysosomes (for FC: see also [22]) as identified by local adaptive thresholding. When cells or compartments exceeded a range of pre-defined parameters (such as signal intensity or shape, minimal/maximal diameter, minimum allowed distance to neighbouring mask or edge of the image) they were omitted from further analysis to exclude for instance dividing or apoptotic cells. Mean cellular background intensity in the GFP channel was determined from the transfection-control sample of each experiment. Tabulated numeric results from image analyses were further processed with customized R-pipelines (R-Studio Inc. vs 0.97.336). Cells with GFP-intensities beneath the 97 percentile of this transfection control sample were defined as “GFP-negative”, and this threshold was applied to determine GFP-negative cells also from the other samples of a respective experiment. Conversely, cells were defined as GFP-expressing (“GFP-positive”) if cellular GFP-signals exceeded this GFP-negative threshold by at least two-fold. Complementation experiments were performed under a “rescue”, but not overexpression setting. Specifically, an upper threshold was introduced for the Cy3 (DiI)-channel, and DiI-LDL uptake was quantified only from the fraction of GFP-positive cells that showed less than 1.25-fold the mean “total LDL signal” (see below) of cells in the transfection-control sample, or less than 5 times the mean “total LDL signal” of cells treated with LDLR siRNA without concomitant cDNA transfection, or cells co-transfected with LDLR-siRNA and GFP-control plasmid, respectively. A justification for this upper threshold is provided by complementation experiments shown in S2B Fig. that demonstrate that reduced DiI-LDL uptake in response to LDLR knockdown can be fully complemented by co-expressing wild-type LDLR’-GFP at only 10–20% of its maximal expression level. For LDL uptake experiments five parameters were quantified per cell: (i) total DiI signal intensity within intracellular endosome-like segments (“total LDL signal”), (ii) mean DiI signal intensity within segments per cell (“LDL concentration”), (iii) number of individual segments within cell masks (“seg. number”), (iv) summed area of all segments within cell masks (“seg. area”), and (v) mean cellular GFP signal intensity (“GFP-expression”).
Statistical analysis of imaging data. For each parameter, means were calculated from all cells per image, and cells were classified as either GFP-positive or GFP-negative. Results from different images of the same biological replicate were averaged, and the ratios of GFP-positive relative to GFP-negative cells were determined. A minimum of 25 GFP-positive cells per variant was required to be considered as independent experimental replicate. Results from different biological replicates were then averaged and compared to outcomes for LDLR’-GFP. Impact of a variant on a distinct parameter was considered as significantly different from wildtype LDLR’-GFP when a paired, two-tailed Student’s t-test resulted in p-values of less than 0.05 and a “deviation value” (a z-score-like measure described in detail in [22]) for parameter total LDL-signal was larger than 1. A variant was categorized as “disruptive-missense” (i.e., severely disrupting LDLR activity as would be expected from an LoF-mutant) if under the overexpression setting “total LDL signal” as well as at least two other parameters reached significance. Under the complementation setting, significance in the parameter “total LDL signal” was regarded as sufficient to validate a variant identified as “disruptive-missense” under the overexpression setting. In order to be classified as “non-disruptive”, none of the eight DiI-LDL parameters quantified from overexpression and complementation settings was allowed to reach significance. A variant was classified as of “unclear” functional significance if it met neither criteria for category “disruptive-missense” nor “non-disruptive”. To test for possible interdependence of measured four DiI-LDL parameters, pairwise Pearson’s correlation values were calculated across the entire dataset (comprising 79 different variants plus wildtype LDLR’-GFP; S7 Table). Consistent with our expectations and the literature (see also [22]), parameters “total LDL signal”, “LDL concentration”, “seg. number” and “seg. area” correlated well, both among each other as well as between overexpression and complementation settings, reflecting a high reproducibility of individual results (S4 Table).
For measuring the impact of disruptive-missense variants on free cholesterol (FC) levels, total filipin signal intensities from lysosome-like intracellular areas were quantified as described [22] from cells cultured and analysed in 96-well plates. Variants that significantly affected cellular FC were determined from the ratio of signal intensities in GFP-positive relative to GFP-negative cells according to identical significance criteria as described above (apart from p.N316S for which no significance could be determined as it reached the minimal number of required GFP-positive cells in only one out of four biological replicates).
Determination of LDLR protein levels. For quantification of LDLR protein levels by Western Blot (Fig. 3D, S2 and S5 Figs.), HeLa-Kyoto cells co-transfected with cDNAs and siRNAs as described above were lysed in 40μl SDS-loading buffer and subjected to immunoblotting with anti-LDLR (Cayman Chemicals), anti-GFP (Roche) and anti-actin (Sigma). Signal intensities of lanes representing 120kDa and 160kDa isoforms of LDLR protein were quantified from background subtracted images using Image J 1.46r (Wayne Rasband, National Institutes of Health, Bethesda) and normalized to levels of beta-actin.
Determination of ΔLDLR’-GFP subcellular localization. Subcellular localization of LDLR’-GFP variants identified as disruptive-missense were re-analyzed at higher resolution using a Zeiss LSM780 laser-scanning confocal microscope using a 63x objective. Assignment of individual variants to different FH-mutant classes was based on i.) phenotypic effects on DiI-LDL uptake, ii.) GFP expression level; and iii.) degree of localization to endoplasmatic reticulum-like relative to endosome-like structures or the plasma membrane as determined visually.
Twenty-three LDLR missense allele carriers from the exome-sequenced cohort (Fig. 1) showed plasma LDL-C levels that did not match expectations from in vitro analyses. For instance, in five carriers of disruptive-missense alleles that all showed early-onset MI, LDL-C was below 190mg/dl. Besides the unlikely possibility for reduced penetrance of heterozygous FH [46] and MI for other causes, a reasonable explanation for this could be that these individuals received LDL-lowering therapies (e.g., statins) at study inclusion. As this information was unavailable to us, precision of the type I error rate for our cell-based analyses is difficult, although it can be assumed as likely small. Of higher relevance is why some carriers of LDLR alleles classified as non-disruptive still showed elevated plasma LDL-C and/or MI, although this is in part this justified by the use of strict sensitivity thresholds that excluded potentially hypomorphic variants from association testing (see Discussion).
It is tempting to speculate that additional genetic variants could have their share in increasing LDL-C in some non-disruptive LDLR allele carriers. One reason for this could be compound-heterozygosity for more than one rare variant at the LDLR locus. For instance, we identified one carrier of the most likely neutral variant p.G20R as also carrying the FH mutant p.G549D, and the latter variant is much more likely to explain that individual’s plasma LDL-C of 218mg/dl. Likewise, compound-heterozygosity for two hypomorphic variants could impair receptor activities in the range of a classic FH-mutant. This is best exemplified by another ATVB individual compound-heterozygous for neutral variants p.L432V and p.Y465N and LDL-C of 309.4mg/dl.
Also, increasing evidence supports a di- or polygenic contribution to the regulation of plasma lipid levels and MI-risk [47–49], and alterations in other genes might explain elevated LDL-C in non-disruptive allele carriers, or unexpectedly low LDL-C in disruptive allele carriers. To test the hypothesis that common risk variants might modify LDL-C levels in these individuals, we calculated polygenic risk scores for variation in LDL-C according to [48] based on 20 lead SNPs from genome-wide association studies for plasma lipids [47] that were represented on the exome chip (S8 Table). Exome chip genotypes were available for 2,433 ATVB study participants. Risk scores relative to plasma LDL-C for all participants are plotted in S6 Fig. In the 23 individuals with unexpectedly low or high LDL-C we did not observe a major contribution of 20 common risk variants when this subcohort was compared to the rest of the ATVB cohort.
We also analyzed these 23 individuals for the presence of rare coding variation in 12 further genes linked to Mendelian causes of abnormal plasma LDL-C (ABCG5, ABCG8, ANGPTL3, APOA5, APOB, APOC3, APOE, LDLRAP1, LIPA, MTTP, NPC1L1 and PCSK9). This produced a total of 21 rare and low-frequency protein-sequence altering variants that distributed over 10 genes (S9 Table). Clinical significance of these variants was evaluated based on information from locus-specific FH databases (for ABCG5, ABCG8, APOB, LDLRAP1 and PCSK9), the Exome Variant Server, ClinVar and HGMD. Only a single variant (p.R238W in LDLRAP1) present in a heterozygous state in two of the 23 individuals had previously been reported from patients with autosomal-recessive FH. However, based on an allele frequency of 0.048 in Europeans and because association of this variant with LDL-C across the ATVB cohort, although indicative, does not yet reach genome-wide significance (p<0.00037; Fisher’s exact test), the contribution of this variant to LDL-C levels in the two LDLR variant carriers that also carry this LDLRAP1 variant remains unclear. One rare variant in APOE (p.G145D) is described as benign polymorphism. No database or literature data is available on the other 19 variants identified, and none has yet been characterized in vitro.
Supplemental Data contains eleven Supplemental Tables, seven Supplemental Figures, one Supplemental Spreadsheet, and Supplemental References.
Exome Variant Server, http://evs.gs.washington.edu/EVS/; Human Gene Mutation Database, http://www.hgmd.org; LDLR UCL LOVD database, http://www.ucl.ac.uk/ldlr/; MutationAssessor, http://mutationassessor.org; MutationTaster, http://www.mutationtaster.org; NCBI ClinVar database, http://www.ncbi.nlm.nih.gov/clinvar; PolyPhen-2, http://genetics.bwh.harvard.edu/pph2/; SIFT, http://sift.jcvi.org; Universal LDLR mutation database, http://www.umd.be/LDLR/
Data, including LDLR sequence data and functional annotations, will be available for download from the NCBI ClinVar database (http://www.ncbi.nlm.nih.gov/clinvar/) under accession numbers SCV000189524—SCV000189592 and SCV000189619—SCV000189628.
|
10.1371/journal.pcbi.1003427 | Communication Efficiency and Congestion of Signal Traffic in Large-Scale Brain Networks | The complex connectivity of the cerebral cortex suggests that inter-regional communication is a primary function. Using computational modeling, we show that anatomical connectivity may be a major determinant for global information flow in brain networks. A macaque brain network was implemented as a communication network in which signal units flowed between grey matter nodes along white matter paths. Compared to degree-matched surrogate networks, information flow on the macaque brain network was characterized by higher loss rates, faster transit times and lower throughput, suggesting that neural connectivity may be optimized for speed rather than fidelity. Much of global communication was mediated by a “rich club” of hub regions: a sub-graph comprised of high-degree nodes that are more densely interconnected with each other than predicted by chance. First, macaque communication patterns most closely resembled those observed for a synthetic rich club network, but were less similar to those seen in a synthetic small world network, suggesting that the former is a more fundamental feature of brain network topology. Second, rich club regions attracted the most signal traffic and likewise, connections between rich club regions carried more traffic than connections between non-rich club regions. Third, a number of rich club regions were significantly under-congested, suggesting that macaque connectivity actively shapes information flow, funneling traffic towards some nodes and away from others. Together, our results indicate a critical role of the rich club of hub nodes in dynamic aspects of global brain communication.
| A fundamental question in systems neuroscience is how the structural connectivity of the cerebral cortex shapes global communication. Here, using computational modeling in conjunction with an anatomically realistic structural network, we show that cortico-cortical communication is constrained by high-level features of brain network topology. We find that neural network topology is configured in a way that prioritizes speed of information flow over reliability and total throughput. The defining characteristic of the information processing architecture of the network is a densely interconnected rich club of hub nodes. Namely, rich club nodes and connections between rich club nodes absorb the greatest proportion of total signal traffic. In addition, rich club connectivity appears to actively shape information flow, whereby signal traffic is biased towards some nodes and away from others. Finally, synthetic networks containing a rich club could almost perfectly reproduce the information flow patterns of the real anatomical network. Altogether, our data demonstrate that a central collective of highly interconnected hubs serves to facilitate cortico-cortical communication. By simulating communication on a static structural network we have revealed a dynamic aspect of the global information processing architecture and the critical role played by the rich club of hub nodes.
| Constrained by finite resources, such as metabolism and physical space, which place severe limits on the number and density of synaptic connections, brain networks are an example of how optimized topology may facilitate information flow. The structural topology of cortical networks can be represented and formally studied using the graph model [1]–[3], whereby the brain is spatially parcellated into a set of grey matter nodes interconnected by a set of white matter edges [4], [5]. This approach has revealed several aspects of network organization that theoretically confer an increased capacity for information processing, including small-world connectivity [6]–[8], the presence of hubs [9] and cores [10], cost-efficient spatial embedding [11], [12] and the coexistence of local segregation and global integration [13].
Recent studies have also uncovered a “rich club” of hub nodes that are more densely interconnected with each other than predicted by chance [14], [15] and that participate in a disproportionately high number of shortest paths in the network [16], [17]. The rich club is hypothesized to act as a central backbone for signal traffic, allowing for rapid integration and dissemination of signal traffic [16].
While this graph theoretic approach can articulate the diverse properties of static neural connectivity, it does not take into account the dynamics of information flow on that connectivity. If information flow is introduced into the network, how does neural connectivity influence the efficacy and speed of communication? In other words, how does network topology enable and constrain the capacity of brain networks to globally integrate information? For instance, while certain areas may bridge distant communities and potentially function as hubs by virtue of their connectivity, other areas may be ill-suited as conduits for information transfer because of their position in the network. Under conditions of elevated network traffic such regions could become bottlenecks, imposing limits on the relay of information [18].
To determine the effect of topology on inter-regional communication, we implemented a macaque anatomical brain network as a communication system in which units of information flow between grey matter nodes along existing anatomical paths (Fig. 1). This allowed us to estimate several metrics of information flow in the network, including the proportion of time a given brain region is in use (utilization), the load on a given brain region (node contents), the time it takes for a unit of information to travel from its source region to its target region (transit time) and the probability of losing information (blocking). The goal of the present study was to use these performance metrics to address the following questions about communication in brain networks. First, does the unique topology of brain networks offer any particular advantage in terms of information processing, and how do brain networks compare to other networks with the same number of nodes and edges, but different topologies? Second, which features of brain network organization contribute most to its capacity for efficient communication? Third, which anatomical regions and pathways are most important for communication?
Information flow on the network was simulated by generating signal units with specific, randomly-selected source and destination nodes. Each signal unit diffused to one of the neighbouring nodes, until it reached its destination. If a signal unit arrived at a node that was occupied, a queue was formed. A maximum buffer size was imposed, such that a signal unit arriving at a full buffer caused the oldest signal unit in the queue to be removed from the system. A signal unit was removed from the network once it reached its destination node. Simulation intensity was controlled parametrically to investigate the effect of increasing load on communication efficiency.
Measures of information flow on the macaque network were compared against a spectrum of degree-matched control networks with an equal number of nodes and edges, but systematically altered topology. One set was comprised of randomized networks, while the other set was comprised of latticized networks (see Materials and Methods for more information on how surrogate networks are generated). Under conditions of increasing load (see SI Fig. S1), all networks experienced increased blocking and utilization, as well as decreased throughput, thus exhibiting signs of congestion. Mean transit times for signal units that reached their destination also decreased with increasing load, but this counterintuitive observation is the result of decreased throughput. At lower network load, more signals reach their destination but some may take a long time to do so, which increases the mean transit time. As the network becomes congested, many such signals may get dropped at over-utilized nodes and never reach their destination, and thus cannot influence the mean transit time.
The macaque network was intermediate on all information flow statistics compared to the randomized and latticized networks. This is consistent with the fact that the randomized and latticized networks represent two extreme and diametrically opposite network configurations and suggests that the organization of the macaque network serves to strike a balance between speed, reliability, utilization and total throughput. Compared to its randomized control network, information flow on the macaque network was characterized by significantly higher loss rates, faster transit times and lower throughput (Fig. 2 top, for all measures, Tables S2,3,4), suggesting that neural connectivity may be optimized for speed rather than fidelity. In general however, the two networks performed similarly, while the latticized control network performed much differently, with significantly lower utilization and throughput, shorter transit times and near total loss of information ( for all measures).
We next sought to determine which feature of network topology most contributed to this pattern of results. To investigate the degree to which the existence of a rich club influences information flow, a synthetic network containing a rich club was created, as well as a set of degree-matched randomized and latticized surrogate networks (Text S1, Section 7, Fig. S7). Information flow on these networks was contrasted with a canonical small world network, which is a ubiquitous and well-studied model for many different kinds of information-processing networks, including neural networks [19]. The pattern of results produced by the synthetic small world and rich club networks and their respective randomized and latticized surrogate networks were considerably different (Fig. 2, middle and bottom). Importantly, the system statistics associated with the rich club network were nearly identical to the macaque network (see Text S1, Section 4, Table S1). The results observed for the macaque network were also similar to the canonical Watts-Strogatz small world network [19], but to a significantly lesser extent (Text S1, Section 4, Table S1). Overall, this suggests that the rich club is an important topological feature for information flow in the brain, as defined by these four statistics.
We now examine regional contributions of the CoCoMac network in detail. To study the individual relevance of nodes for information flow, three complementary node-level metrics of congestion were used: utilization, blocking and mean node contents. Information flow was highly heterogeneous across the network, with some nodes vulnerable to overwhelming influx, while others experienced only occasional traffic (Fig. 3). To a large extent congestion at a given node was predicted by the number of afferent projections to that node (in-degree, and for utilization, blocking and node contents, respectively) and this is expected given the fact that in the present model information flow is implemented as an interactive random walk [2], [20], [21]. With the exception of CA1, all nodes with the largest average contents were previously identified as part of the rich club [17](Fig. S2), indicating that membership in this densely inter-connected subgraph entails a heavy workload. Much of the congestion appears to be concentrated at three distinct sites, mainly along the medial surface, roughly corresponding to medial prefrontal cortex, medial/inferior temporal cortex and precuneus/posterior cingulate cortex.
To determine the extent to which these congestion metrics depend on topology rather than degree sequence, we statistically compared them to a set of metrics from simulations run on a population of randomized control networks for which the topology had been altered while preserving degree sequence [22]. Fig. 4A shows the “raw” mean differences between the two networks for the contents of each node, while Fig. 4B shows the spatial distribution of these differences. Due to the high level of consistency between the three metrics of congestion (utilization, blocking and contents), only the results for node contents are shown. Nodes with contents that are significantly different (, controlled for multiple comparisons using false-discovery rate correction) for the two sets of the networks are labeled. Interestingly, while the majority of these nodes are part of the rich club, nearly half experience greater congestion in the macaque network, while half experience greater congestion in the randomized networks. This suggests that macaque cortical connectivity imposes a characteristic set of traffic patterns, such that signal traffic is directed towards some nodes and away from others, in contrast to what would be expected based only on the degree of these nodes.
We next consider information flow with respect to specific edges. Given the relevance of the rich club in the macaque network [17], we classified edges according to whether they connect rich club nodes [16]. Edges connecting two non-rich club nodes were classified as (L)ocal, those connecting a non-rich club and a rich club node as (F)eeder and those connecting two rich club nodes as (R)ich club. Moreover, these classifications were made with respect to two rich club levels, RC1 and RC2, which represent a more conservative and a more liberal definition of the rich club [17]. An initial observation is that this stratification of edges closely resembles the patterns of edge throughputs. In particular, projections with greater throughput appear more likely to be those connected to at least one rich club node, i.e. Rich Club or Feeder. Despite the fact that the vast majority of edges in the macaque network are Local, followed by Feeder and then Rich Club [17], the mean throughput per edge is greatest for Rich Club edges, followed by Feeder and Local (Fig. 5B). In other words, traffic tends to concentrate not just at rich club nodes, but also at the edges around them, effectively encompassing their local neighborhoods.
Finally, we investigate information flow with respect to every possible pair of source and target nodes. For each pair, all completed trajectories are compiled in order to compute the total number of deliveries (throughput), as well as their mean transit time or delay. For both the throughput and transit time statistics, taking the mean across sources results in greater variance than taking the mean across targets (Fig. 6A,B). For the target nodes, both statistics showed substantial association with in-degree (, for transit time and throughput, respectively). In other words, the mean throughput and transit time were much more dependent on the destination, rather than the source, indicating that some nodes in the network are intrinsically easy to reach, while others are intrinsically difficult.
A non-monotonic relationship emerges when comparing the mean throughput and the mean transit time across target nodes (Fig. 6C, dark grey). When the total throughput is low, any increase in throughput results in slower transit times. However, for a subset of nodes with a high throughput this relationship does not hold and these nodes tend to receive information much faster than would be expected. Most of these nodes belong to the rich club (RC2), indicating that rich club nodes receive more information than other nodes in the network, and do so with a disproportionately faster latency. A similar relationship is observed for degree-preserving randomized controls (Fig. 6C, light grey), indicating that the effect is largely due to the high degrees of rich club nodes. Rich club connectivity enhanced the effect. As expected from Fig. 2, the randomized controls generally have slightly higher throughput, but also longer transit times. This was particularly true for rich club nodes, which received significantly fewer signal units when embedded in the macaque than in randomized networks ( for RC1 and RC2), but did so with significantly faster transit times ( for RC1 and RC2).
The complex anatomical connectivity of the central nervous system suggests that inter-regional communication is important for the functioning of the brain, and in the present report we systematically investigated the effect of network topology on communication. Utilizing a modeling paradigm from telecommunications and statistical physics [2], [21], we superimposed a communication system on an empirically derived network describing macaque cerebral cortex. Our results highlight multiple ways in which structural connectivity has the potential to exert considerable influence on information flow in brain networks.
In terms of global information flow statistics, the macaque network was found to be intermediate to its latticized and randomized reference networks, mirroring the notion that the complex topology of structural networks represents a trade-off between wiring cost and communication efficiency for integrative processing [11], [12], [23]. In particular, the macaque network exhibited an economic balance between speed, fidelity, utilization and sheer volume of transmission. Compared to degree-matched random networks, the macaque network appeared to prioritize speed of transmission over throughput and reliability.
Although many studies have reported evidence of small world organization in structural [6]–[8] and functional networks [24], [25], a canonical small world model by itself could not account for the information processing characteristics observed in the macaque network. However, the added presence of a rich club largely replicated the information flow signature of the macaque network. Therefore, our data suggest that the small world property (in the Watts-Strogatz sense [19]), together with the rich club, is necessary to produce the macaque-like pattern, but by itself is not sufficient. Several recent studies have postulated that a densely interconnected rich club has the potential to facilitate global integration by providing an easily accessible high-capacity backbone that serves to attract and disseminate interregional signal traffic [14]–[17], [26]. By demonstrating that the rich club is a principal topological feature with respect to communication dynamics, our model lends further support to this notion.
Forming a triangle spanning frontal cortex, posterior cingulate cortex/precuneus and medial temporal cortex, the rich club subgraph proved to be a prominent axis in the information processing architecture of the network. Rich club nodes, as well as connections involving rich club nodes, absorbed the greatest signal traffic, indicating that the rich club of densely connected hubs supports the capacity to efficiently centralize and, presumably, integrate information. The fact that rich club nodes are more likely to exhibit signs of congestion warrants further investigation into their potential role as bottlenecks in information processing. Several empirical studies have reported evidence of bottlenecks limiting processing capacity in attention [18] and response selection [27], and the areas they implicate show considerable correspondence with rich club regions (including medial prefrontal cortex and precuneus), although a direct comparison between the macaque network and human fMRI studies is difficult.
Despite the fact that rich club nodes were among the most congested, comparisons with surrogate networks revealed that rich club connectivity may serve to shape information flow, whereby signal traffic is biased towards some nodes and away from others. While a number of rich club nodes consistently experienced heavier traffic than would be expected on the basis of their degrees, others consistently experienced lighter traffic than would be expected on the basis of theirs. Why macaque network topology, and the rich club in particular, shapes cortico-cortical communication in a way that imposes this specific pattern of information flow, remains unclear. It is noteworthy that the under-congested nodes are areas associated with making eye movements, tracking and acting towards objects in space and fusing visual and proprioceptive information. Many of these areas are part of the dorsal attention sub-network [28]–[30], which presumably must be continually responsive and capable of rapidly integrating and communicating information. We therefore speculate that the topology of the global network is configured in a way that relieves congestion at these dorsal attention areas to facilitate fast and efficient interaction with the external environment.
Previous analyses have shown that, despite constituting only a small part of total network density, rich club connections participate in the greatest number of shortest paths in the network [16], [17]. This has led to the hypothesis that if rich club topology is configured in a way that facilitates cortico-cortical communication via shortest paths, there may exist a set of routing or navigation strategies to take advantage of this feature [16], such as “greedy” routing [31]. However, our data demonstrate that the rich club is central to global communication even if information flow is governed by simple diffusion rather than shortest path communication, potentially eschewing the need for a more complex routing mechanism. To characterize the organizational principles of global information flow, further investigation is necessary to determine which types of routing strategies can best take advantage of the unique connectivity of brain networks, as well as which types of routing strategies best replicate empirical functional data.
The modeling paradigm employed in the present study entails a number of features and simplifying assumptions. Therefore, it is important to consider what the biological correlates of these features are and the extent to which they limit the utility of the model. Central to our approach are discrete signal units. At the large spatial scale, it is unlikely that neural communication takes place via discrete signal units. Rather, information flow between large scale neuronal ensembles is likely to be based on spike trains or coordinated volleys of spike trains. It is also possible that information is transferred as an ensemble of signals from multiple neurons. In our model signal units represent the ability of brain regions to influence one another. This simplifying assumption allows us to trace the trajectory of each signal unit as it propagates in the network, and hence to calculate various metrics about the potential for communication that is afforded by the anatomical connectivity.
External arrivals represent the assumption that new information is continuously generated and communicated in the network. The source of this information may be either stimulation exogenous to the nervous system, or some endogenous process. Poisson arrivals were chosen because at the level of individual neurons, inter-spike intervals (ISIs) are found to be exponentially distributed [32] and likewise, in psychophysics and signal detection, the Poisson process is often used to model stimulus fluctuations and other statistical properties of the sensory environment [33]. Queues and finite buffers are constructs that allow us to model how network topology constrains information flow. Queueing is a mechanism by which signal units are made to interact as they flow through the network, modeling the interplay between multiple information flows on top of the structural network [34]. Finite buffers allow for the possibility of signal loss, modeling the poor fidelity of neural transmission [35].
Note also that the present model does not take into account the intrinsic hierarchy of sub-domains of brain networks. For instance, one may expect primary sensory areas to send more information than they receive. Likewise, one may expect higher order, multimodal areas to receive more information than they send. In our model, source and destination nodes for each signal unit are chosen randomly, irrespective of their function, and we largely ignore this aspect of cortical organization. Future studies should investigate this fundamental feature of cortical networks.
The strength of the modeling approach pursued here is that it allows one to generate relative metrics about network communication. The approach is complementary to other, more physiologically realistic paradigms for modeling global system dynamics [36]–[41], which do not model information transmission directly. These models suggest that three key ingredients are needed to generate realistic brain dynamics: empirically derived patterns of structural connectivity, time-delayed transmission and noise [40], [41]. Indeed, a queueing network model has the potential to incorporate all three, and the present implementation includes both empirically derived connectivity and stochastic dynamics.
A fundamental aspect of networked communication is the switching architecture: the manner in which information is directed and transported across the network. In the present study, we utilized a message-switched architecture, wherein an entire “message” is contained in a single discrete signal unit. Our study represents the first attempt to characterize communication dynamics in brain networks, so this type of switching architecture, together with diffusive, stochastic routing, was particularly advantageous because this type of model does not assume that signal units have any knowledge of the global topology or traffic conditions [42]–[44]. A physiologically plausible alternative would be a packet-switched architecture, wherein a message is broken up into packets, which individually take the most efficient path to the destination, where they are re-assembled [45]. This type of architecture has many potential advantages for systems that rely on temporally sparse “bursts” of communication, including lower transit times [45]. Thus, our results and conclusions are strongly tied to the diffusion-based, non-hierarchical, message-switched architecture we used, and may not hold for other switching architectures.
Altogether, our results reveal a dynamic aspect of the global information processing architecture and the critical role played by the so-called “rich club” of hub nodes. Our work lays the foundation for further systematic study of organizational principles for communication in large-scale brain networks, including routing strategies and resource allocation.
The anatomical connectivity data set used in the present study was derived from the online Collation of Connectivity data on the Macaque brain (CoCoMac) database, comprised of data from 413 tract tracing studies of the macaque [46], [47]. The database was originally queried by [48] and further condensed by [17]. To facilitate comparison with previous reports, only cortical nodes were included. The final directed network was comprised of 242 nodes and 4090 edges and was fully connected, such that each node maintained at least one incoming and one outgoing edge.
Two populations of surrogate networks - one randomized and one latticized - were generated to explore the extent to which the topology of the macaque connectivity matrix influenced the simulation results. Randomized networks were generated using a Markov switching algorithm that randomly swapped pairs of edges [22]. Latticized networks were generated using a modified version of the same algorithm, whereby the edges were swapped only if they moved closer to the main diagonal as a result [7]. By randomly re-ordering edges and forcing them closer to the diagonal, the topology of the original network is destroyed and replaced by one where neighbouring nodes are more likely to be connected, as in a ring lattice. Both sets of surrogate networks were degree-matched in the sense that the in-degree and out-degree of each node was preserved. Statistical assessment was performed by comparing 100 simulations on the CoCoMac network with 100 simulations on a randomized null network, for 100 null network realizations. Comparisons between node-specific metrics were made using Welch's t-test for samples with unequal variances [49], and evaluated with respect to degrees of freedom determined using the Satterthwaite approximation [50]. To control the false discovery rate, -values were corrected following the procedure outlined by [51].
A similar procedure was performed for synthetic small-world [19] and rich club [14], [15], [52] networks and their respective null models. A network containing a rich-club was created from a random network by endowing a sub-set of the nodes (the rich club) with greater connection density than the rest of the network, and an even greater connection density amongst each other. The randomized and latticized controls were then created as described above. For the small world scenario, the starting point was a ring lattice. A small world network was generated by randomly permuting 10% of the edges, while a completely randomized network was generated by further permuting each edge 100 times.
Our results have considerable implication for the rich club feature of brain network topology and so for completeness we briefly rehearse the procedure for detecting and defining rich clubs. Fuller descriptions of the rich club phenomenon can be found elsewhere, for brain networks in general [15], [16], as well as for this particular network [17].
For a given graph, a rich club is defined as a set of high-degree nodes (a subgraph) that are more densely connected amongst each other than would be expected on the basis of degree alone [52]. Rich club classification is made with respect to a range of node degrees. For a given degree , all nodes with degree are stripped from the network. A rich club coefficient is calculated as the ratio of remaining connections to all possible connections. Thus, can be thought of as the density of the subgraph. For the same set of nodes, the ratio is also computed with respect to 10,000 degree-matched randomized networks. The normalized rich club coefficient, , measures the density of the subgraph relative to the null model where the global topology has been destroyed. These steps are repeated for a range of , from the lowest to the second-highest degree in the macaque network (2 to 121). A consistently greater than 1 for a range of suggests the existence of rich club organization.
Therefore, across the range of it is possible to define unique sets of rich club nodes corresponding to different values of . These nodes can then be positioned in a nested hierarchy of rich club “levels”, ranging from those containing nodes with the highest degree to those containing nodes with the lowest degree. In the present study, we follow the classification made by [17], whereby two rich clubs were singled out. The first, RC1, was more densely interconnected and comprised of fewer nodes, with greater minimum degree. The second, RC2, was less densely interconnected and contained more nodes, with smaller minimum degree. RC2 is a subset of RC1, and by examining these two levels of the rich club, it is possible to identify robust relationships between rich-club organization and information flow as estimated by our model.
Once nodes have been classified as either rich club or non-rich club, it becomes possible to classify edges as well. Namely, edges that connect non-rich club nodes to non-rich club nodes are classified as “local”, those connecting non-rich club nodes to rich club nodes as “feeder” and those connecting rich club nodes to other rich club nodes as “rich club”.
Signal units were generated and introduced in the network according to a Poisson process with rate , i.e. with exponentially distributed inter-arrival times. For each signal unit, a source node and destination node were randomly selected. To reach its destination node, the signal propagated to one of the neighbouring nodes, with equal probability for each. The time spent at each node (service time) was exponentially distributed with rate . If a signal unit arrived at a node that was occupied, a queue was formed. Units entered the node on a last-come-first-served basis, also known as last-in-first-out (LIFO) queueing [53]–[55]. A maximum buffer size was imposed (), such that a signal unit arriving at a full buffer caused the oldest signal unit in the queue to be ejected and removed from the system. Upon reaching the destination node, the unit was removed from the network. The purpose of queueing is simply to ensure that information flow is interactive, while a finite buffer size allowed us to model imperfect signal transmission [35]. Buffer capacity is not a critical parameter, in the sense that it cannot induce a phase transition in the system. Changes in buffer capacity will produce quantitative, but not qualitative, changes in system behavior (SI Section 3, Fig. S5).
This type of model has two characteristic modes of operation. At low intensities (external arrival rates), the total number of signal units in the network fluctuates around some finite value and the system is said to be in a steady-state. As the intensity is increased, there is a qualitative change in the system dynamics, characterized by a monotonic increase in the number of signal units in the network until all buffers are filled, leading to “jamming” [2], [20]. The key variable is the ratio between the arrival rate and service rate at each node. Therefore, we fixed the service rate () and varied the rate of external arrivals (). The focus of the present study was on the steady-state behavior of the network, and the range of external arrival rates () was chosen to sustain stationary flow, prior to the phase transition.
All simulations were run for 2 million dimensionless time units. Due to the presence of stochastic time variables in the simulation (inter-arrival times and service times), the state of the system was updated at non-uniform time points. Upon completion, the time series of system states were linearly interpolated to produce uniformly sampled time series (Text S1, Section 6, Fig. S6). An initial transient of 40,000 time units, during which the system state had not yet stabilized (determined via the ensemble average method [54]), was discarded from further analysis to avoid transitory effects. The Mersenne Twister [56] was used to generate a uniform distribution, which was then used to generate exponentially distributed random numbers (inter-arrival times and service times) using the standard inverse transform method. All simulations were implemented in Matlab (Mathworks Inc., Natick, MA) and independently verified in Artifex (RSoft Design Group Inc., Ossining, NY), as well as analytically (Text S1 Section 2, Figs. S3,4).
All signal units were uniquely identified, allowing for their position and complete trajectory in the network to be traced across the simulation. These trajectories were then analyzed to compile a set of node-, edge- and network-level statistics. For each node, we calculated the mean proportion of time the node was busy (utilization), the probability of signal loss (blocking) and the mean system contents. For each edge, we calculated the mean throughput of signal units. For each network, we calculated the mean utilization and blocking rates across nodes, as well as the total number of signal units successfully transmitted from source to destination (throughput) and the mean latency of those transmissions (transit time).
More formally, simulation variables were defined as follows. A node at time has two components: the server contents , which describes the number of signal units currently in service, and the queue length , which describes the number of signal units waiting in the buffer. The node contents were thus defined as(1)
Likewise, the contents at any existing channel from node to node was . The total network load is then the sum of all node and channel contents:(2)
The utilization of node is the proportion of simulation time during which . The blocking probability at node was calculated as the number of signal units ejected from divided by the total number of signal units arriving at .
The total time a signal unit spends at a single node, , is the sum of the waiting time in the queue and the service time in the node (3)
Both and are stochastic processes, with determined by the the topology and dynamics on the network, while is drawn from an exponential distribution with rate . For any signal unit, the transit time is the sum of waiting and service times across all nodes traversed from source to destination. Transit time statistics are calculated only for signals that successfully reached their destination.
|
10.1371/journal.pntd.0002479 | CSF CXCL10, CXCL9, and Neopterin as Candidate Prognostic Biomarkers for HTLV-1-Associated Myelopathy/Tropical Spastic Paraparesis | Human T-lymphotropic virus type 1 (HTLV-1) -associated myelopathy/tropical spastic paraparesis (HAM/TSP) is a rare chronic neuroinflammatory disease. Since the disease course of HAM/TSP varies among patients, there is a dire need for biomarkers capable of predicting the rate of disease progression. However, there have been no studies to date that have compared the prognostic values of multiple potential biomarkers for HAM/TSP.
Peripheral blood and cerebrospinal fluid (CSF) samples from HAM/TSP patients and HTLV-1-infected control subjects were obtained and tested retrospectively for several potential biomarkers, including chemokines and other cytokines, and nine optimal candidates were selected based on receiver operating characteristic (ROC) analysis. Next, we evaluated the relationship between these candidates and the rate of disease progression in HAM/TSP patients, beginning with a first cohort of 30 patients (Training Set) and proceeding to a second cohort of 23 patients (Test Set). We defined “deteriorating HAM/TSP” as distinctly worsening function (≥3 grades on Osame's Motor Disability Score (OMDS)) over four years and “stable HAM/TSP” as unchanged or only slightly worsened function (1 grade on OMDS) over four years, and we compared the levels of the candidate biomarkers in patients divided into these two groups. The CSF levels of chemokine (C-X-C motif) ligand 10 (CXCL10), CXCL9, and neopterin were well-correlated with disease progression, better even than HTLV-1 proviral load in PBMCs. Importantly, these results were validated using the Test Set.
As the CSF levels of CXCL10, CXCL9, and neopterin were the most strongly correlated with rate of disease progression, they represent the most viable candidates for HAM/TSP prognostic biomarkers. The identification of effective prognostic biomarkers could lead to earlier detection of high-risk patients, more patient-specific treatment options, and more productive clinical trials.
| HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) is a rare neurodegenerative disease caused by infection with human T-lymphotropic virus type 1 (HTLV-1). HTLV-1 infects 10–20 million people worldwide, and, depending on the region, 0.25–3.8% of infected individuals develop HAM/TSP. As the disease progresses, chronic inflammation damages the spinal cord and lower limb and bladder function gradually decline. In the worst cases, even middle-aged patients can become perpetually bedridden. Today, there are treatments that may alleviate the symptoms to a certain degree, but there is no cure that can halt disease progression, and there are no known biomarkers to indicate the level and speed of disease progression. In this study, we successfully identified three promising candidate biomarkers. We believe that the use of these biomarkers could lead to more accurate prognoses and more prudent, patient-specific treatment plans. We not only hope that these biomarkers are sensitive enough to use as selection criteria for clinical trials, but also that measurements of these biomarkers can be used to accurately evaluate drug effectiveness. In short, the biomarkers we identified have the potential to help more effectively treat current HAM/TSP patients and to pave the way for new drugs to potentially cure future HAM/TSP patients.
| Human T-lymphotropic virus type 1 (HTLV-1) is a human retrovirus associated with persistent infection of T-cells [1]. While the majority of HTLV-1-infected individuals remain asymptomatic, approximately 2.5–5% develop an aggressive T-cell malignancy, termed adult T-cell leukemia (ATL) [2], [3] and 0.3–3.8% develop a serious chronic neuroinflammatory disease, termed HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) [4]–[6]. Aside from Japan, endemic areas for this virus and the associated disorders are mostly located in developing countries in the Caribbean, South America, Africa, the Middle East, and Melanesia [7], [8], which may explain why these conditions have remained ill-defined and virtually untreatable for so long [9].
HAM/TSP is characterized by unremitting myelopathic symptoms such as spastic paraparesis, lower limb sensory disturbance, and bladder/bowel dysfunction [10], [11]. Although the symptoms of HAM/TSP have been well documented for quite some time, the rate at which these symptoms progress has only recently become a point of interest. The clinical course of HAM/TSP has classically been described very simply as insidious onset and continuous progression [12], but recent reports have hinted at a more complex, heterogeneous pool of patients with differing clinical needs. Recent studies have shown that although HAM/TSP usually progresses slowly and without remission as per the classical description, there is a subgroup of patients whose conditions decline unusually quickly and who may be unable to walk within two years of onset and another subgroup whose conditions decline unusually slowly and who may only display very mild symptoms [13]–[15]. It is only logical that these patients should receive treatments tailored to suit their individual needs rather than identically aggressive treatments. Unfortunately, clinicians are currently only able to distinguish between these different groups by observing the way a patient's disease progresses over time, usually years; clinicians often decide to treat the patients immediately and identically rather than wait and allow the disease to progress further. Therein lies the dire need for biomarkers with the power to forecast the rate and extent of disease progression and enable clinicians to make more accurate prognoses and prescribe the most appropriate and effective treatments in a timely manner.
Several candidate prognostic biomarkers with elevated levels in HAM/TSP patients have already been identified in the peripheral blood and cerebrospinal fluid (CSF). In the peripheral blood, such candidates include the HTLV-1 proviral load in peripheral blood mononuclear cells (PBMCs) and serum levels of the soluble IL-2 receptor (sIL-2R) [16], [17]. The level of neopterin in the CSF has been reported to be a useful parameter for detecting cell-mediated immune responses in the spinal cord of HAM/TSP patients and the CSF anti-HTLV-1 antibody titer has been shown to be associated both with CSF neopterin levels and the severity of clinical symptoms [18]–[20]. In addition, several cytokines have been detected in the CSF and/or spinal cord of HAM/TSP patients, including interleukin (IL)-1β, granulocyte-macrophage colony-stimulating factor (GM-CSF), interferon (IFN)-γ, and tumor necrosis factor (TNF)-α [21]–[24]. Some chemokines, such as chemokine (C-X-C motif) ligand (CXCL) 9, CXCL10, and chemokine (C-C motif) ligand (CCL) 5, have been shown to be substantially elevated in both the blood and the CSF with respect to asymptomatic carriers (ACs) or patients with other neurological diseases such as multiple sclerosis [25]–[28]. This is the first study to compare the adequacies of several of these candidate biomarkers for forecasting the rate of disease progression.
We hypothesized the existence of biomarkers capable of differentiating stable and deteriorating HAM/TSP patients. In this retrospective study, a preliminary experiment was first conducted to select the most promising candidate biomarkers by comparing blood and CSF levels in HAM/TSP patients and control subjects (Figure S1). Four candidate blood markers (sIL-2R, CXCL9, CXCL10, and proviral load) and five candidate CSF markers (CXCL9, CXCL10, neopterin, cell count, and anti-HTLV-1 antibody titer) were selected. To evaluate the relative effectiveness of these candidate biomarkers for predicting rate of disease progression, a classification system was created and HAM/TSP patients were designated as either deteriorating or relatively stable. The levels of candidate biomarkers were then compared between the two patient groups. In the current study, we identified three viable candidates for HAM/TSP prognostic biomarkers that could lead to more accurate prognoses and more prudent, patient-specific treatment plans.
The study was designed and conducted in accordance with the tenets of the Declaration of Helsinki. The protocol in this study was approved by the Ethics Review Committee of St. Marianna University School of Medicine (No. 1646). Prior to the collection of blood or CSF samples, all subjects gave written informed consent permitting the analysis of their samples for research purposes as part of their clinical care.
Between April 2007 and February 2013, we enrolled 53 HAM/TSP patients according to the inclusion and exclusion criteria shown in Table 1, and divided them into two cohorts based on the chronological order of their doctor's visits: a 30-patient Training set and a 23-patient Test set. Demographics and clinical characteristics of the Training set and Test set are shown in Table 2 and Table 3, respectively. Between April 2007 and December 2009, we enrolled 22 HTLV-1-infected ACs as control subjects for blood analysis and eight HTLV-1-infected subjects (seven ACs, one patient with smoldering ATL) as control subjects for CSF analysis according to the inclusion and exclusion criteria shown in Table 1. These two groups were not mutually exclusive; some ACs donated both blood and CSF to this study. Demographics of control subjects as compared to the HAM/TSP patients are shown in Table S1.
Blood and/or CSF samples were obtained within a one-hour window for each subject. Peripheral blood samples were collected in heparin-containing blood collection tubes and serum-separating tubes. Plasma and PBMCs were obtained from the former tubes and serum was obtained from the latter. PBMCs were isolated with standard procedures using Pancoll® density gradient centrifugation (density 1.077 g/mL; PAN-Biotech GmbH, Aidenbach, Germany). Plasma and serum samples were stored at −80°C until use. CSF was collected in polypropylene tubes. A small amount of CSF was used for routine laboratory tests, which included total protein, cell count, and IgG level. The remaining CSF was aliquoted into cryotubes and stored at −80°C until undergoing further analysis. All tests in this study were performed on samples from these frozen stocks.
The serum concentration of sIL-2R was determined using an ELISA (Cell Free N IL-2R; Kyowa Medex Ltd., Tokyo, Japan). HTLV-1 proviral load was measured using real-time PCR, following DNA extraction from PBMCs, as previously described [29]–[31]. Plasma levels of IL-1β, TNF-α, and IFN-γ were measured using a cytometric bead array (CBA) (BD Biosciences, Franklin Lakes, NJ USA), which was used according to the manufacturer's instructions. Plasma concentrations of CXCL9, CXCL10, CXCL11, and CCL5 were also measured using a CBA (BD Biosciences).
CSF cell count was determined using the Fuchs–Rosenthal chamber (Hausser Scientific Company, Horsham PA USA). Total protein and IgG levels in the CSF were measured using a pyrogallol red assay and a turbidimetric immunoassay, respectively. The anti-HTLV-1 antibody titer was determined using the gelatin particle agglutination test (Serodia-HTLV-1; Fujirebio, Tokyo, Japan). CSF concentration of sIL-2R was determined using an ELISA (Cell Free N IL-2R; Kyowa Medex). CSF neopterin level was measured using high-performance liquid chromatography. IFN-γ and six chemokines (CXCL9, CXCL10, CXCL11, CCL3, CCL4, and CCL5) were measured using a CBA (BD Biosciences). The CSF concentrations of three chemokines (CCL17, CCL20, and CCL22) and IL-17A were measured using commercially available ELISA kits (CCL17, CCL20, and CCL22: TECHNE/R&D Systems, Minneapolis, MN USA; IL-17A: Gen-Probe, San Diego, CA USA). All assays were conducted according to the respective manufacturers' instructions.
The 53 total HAM/TSP patients without any history of HAM/TSP-targeting treatments were interviewed using a questionnaire (Figure S2) to determine the changes in Osame's Motor Disability Score (OMDS) over time (Figure S3). OMDS is a standardized neurological rating scale as a measure of disability [10] (Figure S1). Based on the changes in OMDS, “deteriorating cases” and “stable cases” were identified in both the Training set and Test set patient cohorts. Patients with deteriorating HAM/TSP were defined as those whose OMDS worsened ≥3 grades over four years and patients with stable HAM/TSP were defined as those whose OMDS remained unchanged or worsened 1 grade over four years. Patients whose OMDS worsened 2 grades over four years were excluded from the patient cohort in order to create a larger gap between the deteriorating and stable patient groups.
GraphPad Prism 5 (GraphPad Software, Inc., La Jolla, CA USA) was used to plot graphs and perform statistical analyses. Differences between the two subject groups were tested using the Mann-Whitney U-test. Receiver operating characteristic (ROC) analysis was performed to examine the sensitivity and specificity of individual biomarkers. For the ROC analyses, an area under the ROC curve (AUC) of 1.0 was used to represent a perfect test with 100% sensitivity and 100% specificity, whereas an area of 0.5 was used to represent random discrimination. Spearman's rank correlation test was employed to investigate the correlation between the four CSF markers (CXCL10, CXCL9, neopterin, and cell count) and the proviral load in PBMCs. To compare the four CSF markers between three groups (HTLV-1-infected control, n = 8; stable HAM/TSP, n = 25; and deteriorating HAM/TSP, n = 20), we used the Kruskal–Wallis test followed by Dunn's post-hoc tests. P-values<0.05 were considered statistically significant.
In order to identify candidate blood markers for HAM/TSP, the concentrations of IL-1β, TNF-α, and IFN-γ were measured in plasma samples from four ACs and four HAM/TSP patients. Plasma levels of IL-1β and TNFα were below the detection limits (<2.3 pg/mL and <1.2 pg/mL, respectively) except in one patient with HAM/TSP. Plasma IFN-γ levels showed no significant differences between ACs and HAM/TSP patients (median 10.4 pg/mL and 13.9 pg/mL, respectively). Therefore, these quantities were not measured in additional samples (Figure S1). The proviral DNA load in PBMCs, serum sIL-2R, and plasma levels of the chemokines CXCL9, CXCL10, CXCL11, and CCL5 were also measured in 22 ACs and 30 HAM/TSP patients without any history of immunomodulating treatments, including corticosteroids, IFN-α, and immunosuppressive drugs. The results revealed that serum levels of sIL-2R, plasma levels of CXCL10 and CXCL9, and proviral DNA load in PBMCs were markedly higher in HAM/TSP patients compared to ACs (p≤0.0001, Figure 1A). These quantities were then compared using ROC analysis to determine which parameters were superior markers for HAM/TSP. From the results of the ROC analysis, we determined that serum sIL-2R and plasma CXCL10 had the highest potential for distinguishing HAM/TSP patients from ACs with high sensitivity and specificity (area under the ROC curve [AUC]>0.9), followed by plasma CXCL9 and HTLV-1 proviral load in PBMCs (0.8<AUC<0.9) (Figure 1B). Thus, four candidate blood biomarkers were selected for further investigation: serum sIL-2R, plasma CXCL10, plasma CXCL9, and HTLV-1 proviral load in PBMCs.
In order to identify candidate CSF markers for HAM/TSP, elevated levels of various potential markers were screened for in CSF samples from HAM/TSP patients. CSF IL-17A was detectable (>3.0 pg/mL) in only one of eight HAM/TSP patients screened (including six deteriorating-type patients), and the level in this one patient (deteriorating-type) was negligible (4.0 pg/mL). CSF IFN-γ was detectable (>1.8 pg/mL) in only 3 of 10 HAM/TSP patients screened (six deteriorating patients), and the levels in all three were negligible (range 3.3–4.2 pg/mL). Therefore, these cytokines were not measured in additional patients. Total protein, cell count, IgG, neopterin, sIL-2R, and nine chemokines (CXCR3 ligands: CXCL9, CXCL10, and CXCL11; CCR5 ligands: CCL3, CCL4, and CCL5; CCR4 ligands: CCL17 and CCL22; CCR6 ligand: CCL20) were also measured in the CSF of 30 untreated HAM/TSP patients and in eight HTLV-1-infected control subjects (seven ACs and one patient with smoldering ATL). The results indicated that CSF levels of CXCL10, neopterin, and CXCL9 were remarkably higher in HAM/TSP patients compared to control subjects (p<0.0001 overall, Figures 2A and S4) and that CSF levels of cell count and CCL5 were less so but still significantly higher (p = 0.0019 and p = 0.0119, respectively; Figure 2A). By contrast, there were no differences in the CSF levels of IgG and total protein between HAM/TSP patients and control subjects, and CSF sIL-2R levels were only detectable in a single HAM/TSP patient (data not shown). ROC analysis showed that the CSF levels of CXCL10, neopterin, CXCL9, and CSF cell count could be used to relatively accurately distinguish HAM/TSP patients from control subjects (AUC>0.8) (Figure 2B). Therefore, these four CSF markers were selected as candidates for further investigation. It should be noted that the sensitivity of CSF cell count was very low (36.7%) when compared to the other three: CXCL10 (83.3%), CXCL9 (86.7%), and neopterin (76.7%) (Figure S5).
In short, we selected nine markers: eight markers chosen based on the analyses described above and CSF anti-HTLV-1 antibody titer, which is a known diagnostic marker for HAM/TSP. To determine which biomarkers were associated with HAM/TSP disease progression, the levels of these nine markers were compared between the deteriorating and stable HAM/TSP patient groups (see Methods for definitions of deteriorating and stable). The results revealed that all five CSF markers were significantly higher in the deteriorating group compared to the stable group (Figure 3A), but that none of the four blood markers, including proviral load, were significantly different between the two groups. The deteriorating group included three patients with particularly rapidly progressive HAM/TSP, defined as those who had been confined to wheelchairs (OMDS: ≥ grade 6) within two years after the onset of symptoms [13], [14] (black circles in Figures 3A and S3B). These rapid progressors exhibited high levels of the CSF markers and high proviral loads. ROC analysis revealed that the levels of the CSF markers (CXCL10, CXCL9, neopterin, and cell count), but not anti-HTLV-1 antibody titer, distinguished clearly between patients with deteriorating HAM/TSP and stable HAM/TSP (AUC>0.8, Figure 3B).
To validate the results obtained using the Training Set, the same nine markers were compared between deteriorating and stable patients using the Test Set (a second cohort of 23 HAM/TSP patients that had not undergone HAM/TSP-targeting treatment). As shown in Figure 4A, the results indicated that the levels of five CSF markers, proviral load in PBMCs, and serum sIL-2R were significantly higher in deteriorating cases than in stable cases. Among them, CSF levels of CXCL10, CXCL9, neopterin, and CSF cell count exhibited particularly high sensitivities and specificities for detecting the deteriorating HAM/TSP cases in the Test set as well as Training set (AUC>0.8, Figures 4B and S1).
The demographics of the HAM/TSP patients versus the control subjects for both the blood tests and CSF analyses were compared and evaluated for statistical significance (Table S1). There were no significant differences in age or gender distribution between the HAM/TSP patients and either control subject group.
Similarly, the demographic and clinical characteristics of stable versus deteriorating HAM/TSP subjects in both the Training and Test sets are shown in Tables 2 and 3, respectively. There were no significant differences in age or gender distribution among either set, but deteriorating patients in both sets were significantly older at disease onset and had been living with the disease for shorter periods of time. Deteriorating patients in the Training set scored higher OMDS values than their stable counterparts (p<0.01), but there was no such significant difference in the Test set.
To investigate the potential influence of disease duration as a secondary variable, a new test group was created containing only those patients for whom the disease onset date was 7–13 years prior to the sample collection day. Patients fitting this criterion were selected from the 53 total available from both the Training and Test sets: eight stable patients and ten deteriorating patients; we confirmed that there was no significant difference in disease duration between these two groups. The results remained consistent with our previous findings: CSF CXCL10, CXCL9, and neopterin were all elevated in deteriorating patients with respect to stable patients (p<0.01, Figure 5).
Four stable HAM/TSP patients were left completely untreated and followed for a period of three to five years. Within this time, one patient rose one grade on the OMDS scale, and the other three experienced no change in OMDS grade at all. The levels of CSF CXCL10 and neopterin remained consistently low over time (Figure S6).
To date, there have been few well-designed studies that have evaluated the relationship between biomarkers and HAM/TSP disease progression. In a previous retrospective study with 100 untreated HAM/TSP patients, a significant association was demonstrated to exist between higher HTLV-1 proviral load in PBMCs and poor long-term prognosis; however, the predictive value of high proviral load appeared to be too low to qualify it as a marker for disease progression in clinical practice [32]. Here we conducted a retrospective study to compare for the first time the relationships of PBMC proviral load and several inflammatory biomarker candidates to disease progression in untreated HAM/TSP patients.
In this study, elevated CSF cell count, neopterin concentration, and CSF levels of CXCL9 and CXCL10 were well-correlated with disease progression over the four year period under study, better even than HTLV-1 proviral load in PBMCs (Figures 3 and 4). As CSF pleocytosis, CSF CXCL10, CSF CXCL9, and CSF neopterin are known indicators of inflammation in the central nervous system [33], [34], our findings indicate that the rate of HAM/TSP progression is more closely reflected by the amount of inflammatory activity in the spinal cord than by the PBMC proviral load. However, we also found a significant correlation between PBMC proviral load and the levels of the CSF markers identified in this study (Figure S7), indicating that a higher PBMC proviral load does indeed suggest more inflammation in the spinal cord and therefore a poorer long-term prognosis. These findings are consistent with the theory that HAM/TSP is the result of an excess of inflammatory mediators caused by the presence of HTLV-1-infected T-cells [35]–[37].
The HTLV-1 proviral load in the CSF as well as the ratio of the proviral load in the CSF to that in PBMCs have been reported to be effective for discriminating HAM/TSP patients from ACs or multiple sclerosis patients infected with HTLV-1 [38], [39]. Some researchers have suggested that these values might be associated with the rate of disease progression, but there has been only one small cohort study and one case report investigating this point, and so the significance of this experimental evidence is still questionable [40], [41]. In addition to statistical validation with multiple, larger cohorts, it would also be beneficial to use precise definitions for progressive versus stable patients, as we have done in this study. Although the volume of CSF available per sample was too limited to measure CSF proviral load in the present study, we plan to incorporate CSF proviral load in a future prospective study and compare its usefulness to that of other biomarker candidates.
From our results, we concluded that of the potential biomarkers under study, CXCL10, CXCL9, and neopterin are the most fit for determining the level of spinal cord inflammation, and thus the most fit for predicting disease progression in HAM/TSP patients. Although the CSF cell count is an easily measurable inflammatory marker, it is not sensitive enough to reliably detect the level of spinal cord inflammation. Numerous patients with CSF cell counts within the normal range exhibited high levels of other inflammatory markers, such as neopterin and CXCL10 (Figure S5). In fact, it has been reported that CSF pleocytosis is present in only approximately 30% of HAM/TSP patients [42]. Furthermore, in our study, there was no significant difference in CSF cell count between the control subjects and the stable HAM/TSP patients (Figure S8).
We also explored the possibility of combining multiple biomarkers via multiple logistic regression to form a combination more sensitive and specific than individual markers, but the results indicated that there is not much to be gained from combinations (data not shown).
While there were no significant demographic differences between subject groups, the clinical characteristics of stable versus deteriorating HAM/TSP patients of course differed widely (Tables 2, 3, and S2). We confirmed the already well-reported statistic that deteriorating patients experience HAM/TSP onset relatively late in life [12], [14], [20] ; our data also reflected the short disease duration expected of deteriorating patients, who by definition progress through the disease more rapidly than their stable counterparts. As patients in all groups were of similar age at sample collection, the significant difference in age of onset should not have any impact on our findings. However, it was necessary to consider the possibility that those patients in a later stage of the disease (i.e. those listed with longer disease durations) might possess elevated or diminished biomarker levels regardless of rate of disease progression. We confirmed that this difference in disease duration was not a confounding factor in our selection of candidate biomarkers by comparing stable and deteriorating HAM/TSP patients with similar disease durations (7–13 years), and we were able to obtain results consistent with our earlier findings (Figure 5). Finally, the OMDS values for the stable and deteriorating patient groups in the Test set were perfectly identical, eliminating the need to consider the possibility that the biomarkers could have been elevated according to disease severity regardless of rate of progression.
The main limitation of our retrospective study is that our samples were collected from patients at the end of the four year period during which the extent of progression was analyzed as opposed to the beginning of the four year period, which would have been optimal for directly measuring their prognostic powers. Of course, the patients with severe HAM/TSP symptoms began undergoing treatment soon after sample collection, rendering any observations on disease course after sample collection un-useable for analysis in this study. While this situation is non-ideal, we hypothesize that biomarker levels in a given patient do not substantially change over a few years' time. We were actually able to monitor the biomarker levels of four untreated HAM/TSP patients over 3–5 years, and the levels remained relatively stable in all four subjects over time (Figure S6), supporting our hypothesis. However, these were all stable HAM/TSP patients (hence the lack of treatment), and so we cannot rule out the possibility that biomarker levels in untreated deteriorating patients may dramatically rise, fall, or fluctuate. The results of the analysis of patients with similar disease durations (Figure 5) also support our hypothesis that disease duration is not an important determinant of biomarker levels, but it is of course not conclusive. We expect that a prospective study in the future will reveal the answer to this question.
The results of this study indicate that CXCL9 and/or CXCL10 may play a key role in the pathogenesis of HAM/TSP by recruiting more inflammatory cells to the spinal cord lesions. In this study, we measured the levels of the chemokines in the CSF that might play a part in inducing the migration of T-helper (Th) cells. CD4+ Th cells differentiate from naïve T-cells to members of the Th subset (e.g., Th1, Th2, Th17, or Treg cells), and each one expresses its own characteristic chemokine receptors [43]. Usually, Th1 cell express CCR5/CXCR3 receptors, Th2 and Treg cells express CCR4, and Th17 express CCR6. Interestingly, CCR4 ligands (CCL17 and, CCL22) and the CCR6 ligand (CCL20) were not detected in the CSF of HAM/TSP patients. Moreover, of the CCR5 ligands, only CCL5 was elevated, but only slightly, and there was no association with rate of disease progression. Of the CXCR3 ligands, only CXCL9 and CXCL10 were correlated with the rate of disease progression. These results show that the pathology of HAM/TSP is unique among immune disorders in that, unlike other inflammatory disorders such as multiple sclerosis or rheumatoid arthritis that exhibit Th17 as well as Th1 involvement, the chemokine involvement in HAM/TSP is Th1-dominant. In a previous study, cytokines produced by HTLV-1-infected T-cells in HAM/TSP patients were analyzed, and the results showed that IFN-γ was elevated and IL-17 reduced [43], [44]. Taken together, the results of these studies indicate that the characteristics of HTLV-1-infected T-cells themselves may be responsible for the Th1-dominant chemokine production observed in HAM/TSP. Also, these results suggest that the CXCR3-ligand (CXCL9 and CXCL10) interactions play an important role in the pathophysiology of HAM/TSP. Recently it was established that these CXCR3-ligand interactions are extremely important for the pathogenesis of several neurological disorders [33]. Therefore, future research on the significance of these interactions in the pathogenic process of HAM/TSP will be important for clarifying the suitability of CXCL9 and CXCL10 as biomarkers or therapeutic targets.
In conclusion, in this retrospective study, we have demonstrated that CSF levels of CXCL10, CXCL9, and neopterin are promising candidate prognostic biomarkers for HAM/TSP. These biomarkers may provide a means for the early identification of patients at increased risk of debilitating disease progression, those that may need anti-inflammatory therapies to limit or prevent this, and for evaluating the efficacy of such therapies. This initial identification of prognostic biomarkers for HAM/TSP should be followed by a future multicenter prospective clinical study.
|
10.1371/journal.pntd.0004097 | Urban Market Gardening and Rodent-Borne Pathogenic Leptospira in Arid Zones: A Case Study in Niamey, Niger | Leptospirosis essentially affects human following contact with rodent urine-contaminated water. As such, it was mainly found associated with rice culture, recreational activities and flooding. This is also the reason why it has mainly been investigated in temperate as well as warm and humid regions, while arid zones have been only very occasionally monitored for this disease. In particular, data for West African countries are extremely scarce. Here, we took advantage of an extensive survey of urban rodents in Niamey, Niger, in order to look for rodent-borne pathogenic Leptospira species presence and distribution across the city. To do so, we used high throughput bacterial 16S-based metabarcoding, lipL32 gene-targeting RT-PCR, rrs gene sequencing and VNTR typing as well as GIS-based multivariate spatial analysis. Our results show that leptospires seem absent from the core city where usual Leptospira reservoir rodent species (namely R. rattus and M. natalensis) are yet abundant. On the contrary, L. kirschneri was detected in Arvicanthis niloticus and Cricetomys gambianus, two rodent species that are restricted to irrigated cultures within the city. Moreover, the VNTR profiles showed that rodent-borne leptospires in Niamey belong to previously undescribed serovars. Altogether, our study points towards the importance of market gardening in maintain and circulation of leptospirosis within Sahelian cities. In Africa, irrigated urban agriculture constitutes a pivotal source of food supply, especially in the context of the ongoing extensive urbanization of the continent. With this in mind, we speculate that leptospirosis may represent a zoonotic disease of concern also in arid regions that would deserve to be more rigorously surveyed, especially in urban agricultural settings.
| We surveyed rodent-borne Leptospira in rodents from Niamey, the capital town of Niger, using bacterial metabarcoding, RT-PCR, sequencing, VNTR typing and GIS-based geostatistics. Two new serovars of Leptospira kirschneri were identified in Arvicanthis niloticus and Cricetomys gambianus, two species that inhabit exclusively urban irrigated gardens. Since no rodent-borne leptospires could be found in the core city, our results point towards the importance of urban agriculture in the maintaining and the circulation of these bacteria in cities from semi-arid regions where they are usually poorly documented and even hardly looked for. Accordingly, this is one of the very rare mentions of these zoonotic agents in Sahel, and the first one in Niger. Keeping in mind the critical role of urban gardening for food security in extensively growing West African cities, we believe that leptospirosis should be more closely scrutinized in Sahelian countries where numerous cases of human fevers are of unknown origin.
| Leptospira is a genus of spirochetes which comprises three lineages, one of which grouping pathogenic species for both animal and human [1]. Leptospirosis is a major zoonotic disease that may affect at least 500,000 and potentially up to 1 million persons, and kill ~60,000 ones per year worldwide [2–5]. Its incidence remains poorly documented because leptospirosis leads to clinical signs that are difficult to distinguish from other widespread endemic pathologies such as dengue, malaria, influenza, etc. [6]. In addition, many countries where it has an obvious burden lack appropriate diagnostic facilities, thus strongly suggesting that cases may be massively underreported [2, 4].
Among other mammals, rodents, especially rats, constitute major reservoirs of Leptospira spp.: the bacterium resides in the host renal tubules and is then excreted into the environment through its urine. Leptospirosis is thought to be essentially associated with water where humans get contaminated following contact with the pathogen through skin abrasions or mucous membranes (reviews in [4, 7]). In particular, rice culture, recreational water activities and flooding have been massively linked to leptospirosis. This is the reason why the disease was essentially looked for, and found in temperate as well as warm and humid tropical regions (reviewed in [8]). Surveys in arid zones are rare, although some mentions exist from desert to sub-desert areas (e.g., Somalia: [9]; Arizona: [10]; Mexico: [11]; Brazil: [12]), thus suggesting that Leptospira may be much more widespread than currently thought and could also extend to dry regions. As an example, prevalence in wild Malagasy mammals was found higher in Northern areas of the island where rainfalls are weaker [13].
Mentions of Leptospira in Africa (review in [14–16]) are quite scattered, and even very rare for some particular regions (see Fig. 2 in [14]). For instance, in the West African Sahel zone, some data are available for Senegal (two investigations in both humans and cattle in the 1970s), Chad (one report in a dog in 2008) and Mali (one human case report in the 1990s, and one investigation in cattle in the early 1970s) while no monitoring has ever been conducted in Burkina-Faso or Niger [14]. Yet, reported sporadic epidemics in various parts of the continent reflect a lack of knowledge of the disease rather than a truly narrow distribution of Leptospira [8]. This suggests that further investigations in Africa in general, and in arid zones in particular are required. Moreover, leptospirosis is often associated with disadvantaged urban areas where poor sanitation together with elevated rodent-human interactions increase the risk of rodent-to-human transmission (e.g., [17–19]; reviewed in [4]). Taking into account the impressive growth of African cities [20], there is little doubt that leptospirosis will be a major (re)emerging disease on the continent [14].
Niger, focus of the present study, ranks last of the World for the Human Development Index (187 out of 187; [21]). The capital city, Niamey, lies on the Niger River in the western part of the country, and is located in the typical Sahelian bioclimatic zone. As such, it is characterized by high temperatures (monthly average temperatures between 22–36°C) and low rainfalls (~540 mm per year) with a single rainy season between May and September. It was created ex nihilo at the very end of the nineteenth century by French colonizers (reviewed in [22]). During the last decades, the city has been experiencing an explosive spatial and demographic growth with its population increasing from >30,000 in the late 1950s, to 707,000 in 2001, and currently reaching more than 1,000,000 inhabitants [22–24]. As often in such cases, this rapid urbanization is characterized by many informal settlements and insufficient sanitary services. Accordingly, data hence knowledge about zoonotic pathogens that may circulate in Niger are extremely scarce, potentially explaining why so many fevers are misdiagnosed as malaria (i.e., more than 55% in the rainy season, and up to 95% during the dry season; [25]). Expectedly, Leptospira appears among the top candidate pathogens that may explain these so many fevers of unknown origin [26].
These are the reasons why we took advantage of a monitoring of urban rodents conducted in Niamey, the main town of Niger [27], to perform the first survey of rodent-borne Leptospira in this very poor country where zoonoses are dramatically under-documented.
The whole rodent trapping campaign was validated by national and local authorities (scientific partnership agreement number 301027/00 between IRD and the Republic of Niger). At the French level, all sampling procedures were approved by the “Comité d’Ethique pour l’Expérimentation Animale—Languedoc Roussillon” (agreement number C34-169-1, valid until 25th July 2017) and were conducted by biologists from the CBGP holding certificates to carry out experiments on live animals (agreement number C34-488). None of the rodent species investigated in the present study has protected status (see UICN and CITES lists). All animals were treated in a humane manner in accordance with guidelines of the American Society of Mammalogists. All rodents were euthanized through cervical dislocation. Permit to enter and work within private properties were systematically obtained through oral but explicit agreement from adequate institutional (research agreement quoted above; mayor) and traditional authorities (both neighborhood and family chiefs).
From October 2009 to February 2011, an extensive survey of urban rodent assemblages was conducted in 52 localities of Niamey, Niger, thus allowing the exploration of more than 215 trapping sites with an effort of >14,500 night-traps (see details in [27]). Among the 987 rodents captured, 578 were included in the present screening of Leptospira. They consisted in 66 Arvicanthis niloticus, 12 Cricetomys gambianus, 350 Mastomys natalensis, 50 Mus musculus and 100 Rattus rattus originating from 49 localities sites within the city (Table 1 and Fig 1).
African rodent species identification may sometimes be difficult due to the frequent co-occurrence of sibling taxa, notably in the genera Rattus [28], Arvicanthis and Mastomys [29]. This is the reason why a special attention was paid to taxonomic diagnosis which relied on karyotyping (for Arvicanthis, Mus and Mastomys), cytochrome b gene sequencing (for Arvicanthis and Rattus), PCR and species-specific RFLP (for Mastomys) and genotyping (for Mastomys and Rattus). All these procedures have been described in details elsewhere (see [27], and references therein).
Individual genomic DNA was extracted from ethanol-preserved kidney tissue using the Qiagen DNeasy Blood and Tissue Kit, and was quantified using Nanodrop technology (Thermoscientific). Kidney DNA samples were then prepared in equimolar concentration. Pools grouping 50 rodent individual DNA samples each were then arranged by species as follows: (i) one pool made of 50 A. niloticus from 7 localities, (ii) one pool of 50 M. musculus from 3 localities, (iii) two pools with 50 black rats each from 11 localities, respectively, and (iv) seven pools of 50 M. natalensis each and representing 32 localities. Samples were chosen in order to cover most (when not all) localities where each species had been found during a recent broader survey of urban rodents of Niamey (Table 1; see [27]). The eleven pools of DNA were then screened for the presence of bacteria using universal PCR primers targeting the hypervariable region V4 of the 16S rRNA gene (251bp) via Illumina MiSeq (Illumina) high throughput sequencing. The V4 region has been proven to offer excellent taxonomic resolution for bacteria at the genus level [30]. A multiplexing strategy enabled the identification of bacterial genera in each pool sample. We followed the method detailed in Kozich et al. [31] for PCR amplification, indexing, pooling of PCR products and de-multiplexing. Bacteria taxonomic identifications at the generic level were performed using the Silva SSU Ref NR 119 database (http://www.arb-silva.de/projects/ssu-ref-nr/) as a reference [32]. Each DNA pool was analyzed in triplicate using three independent PCRs and three amplicon libraries in the same next generation sequencing (NGS) run using a MiSeq sequencer (Illumina).
Rodents that belonged to metabarcoding Leptospira-positive pools as well as 16 A. niloticus and 12 C. gambianus which had not been included in the latter NGS-based survey were all individually screened for pathogenic Leptospira species using a dedicated Real Time PCR-based test.
To do so, sequences of lipL32 gene from Leptospira kirschneri (AF121192), L. interrogans (AF181553, AF245281, AF366366, LIU89708), L. borgpetersenii (AF181554), L. santarosai (AF181555) and L. noguchii (AF181556) were aligned, and a consensus sequence was determined using BioEdit v.7.1.9. New forward (LIP32BF: 5’-AGC TCT TTT GTT CTG AGC GA-3’) and reverse (LIP32BR: 5’-TAC GAA CTC CCA TTT CAG CGA TTA-3’) primers were designed from this consensus sequence using the Light Cycler Probe design software v.2.0 (Roche). This new set of primers was proved to detect most known pathogenic Leptospira species (namely L. interrogans, L. borgpetersenii and L. kirschneri in ‘wet lab’, as well as L. santarosai and L. noguchii in silico) with lower Ct values than the primers used in recent lipL32 RT-PCR-based survey (e.g., [33, 34]). We used the TaqMan probe (FAM-5′-AAA GCC AGG ACA AGC GCC G-3′-BHQ1) previously described in Stoddard et al. [33], thus allowing us to amplify a 199 pb-long fragment of the leptospiral lipL32 gene.
RT-PCR reactions were performed using a LightCycler 480 (Roche) in 96-well microtitre plates with 10μL as final volume for each reaction. Optimal amplification conditions were obtained with 0.5μM of each primer, 0.2μM of probe, 2X of Probe Master buffer (Roche), 0.5U of Fast start Taq DNA polymerase (Roche) and 2μL of sample DNA. RT-PCR program consisted in an initial denaturing step at 95°C for 10 min, followed by 50 cycles of 95°C for 15s, 60°C for 30s and 72°C for 1s, and a final cooling step to 40°C. All samples were investigated in independent duplicates. Genomic DNA isolated from L. interrogans serovar Canicola and L. borgpetersenii serovar Tarassovi were used as positive controls. The Beta Actin gene was amplified from all samples as an internal RT-PCR control in order to detect false negative results [35].
A 330 pb-long fragment of the rrs gene was amplified from genomic DNA of the RT-PCR-positive rodents. Primers A and B were used for a first amplification; when the PCR was negative, a nested PCR was performed with primers C and D [36]. PCR products were sequenced in both directions at Eurofins Genomics. Species-specific identification was performed through Blastn (option Megablast for highly similar sequences) procedure under NCBI database.
Identification at the subspecies level was performed by multiple-locus variable-number tandem repeat analysis (MLVA) using the loci VNTR4, VNTR7 and VNTR10 as previously described [37] with the following modifications. MLVA was performed on DNA extracts using 70 cycles of amplification with a higher concentration of Taq polymerase (GE Healthcare). The sizes of the amplified products were then analysed using a 1% agarose gel electrophoresis, and the profiles were compared with the database of the National Reference Center for Leptospirosis (Institut Pasteur, Paris, France).
Our purpose here was to map the most suitable within-city areas for Leptospira-carrying rodent species, as identified by molecular methods. Since reservoir rodents in Niamey were all found to belong to rural-like species (i.e. A. niloticus and C. gambianus; see below) and since these latter species strictly segregate spatially from true commensal ones throughout the town [27], we chose to focus on rural-like species only.
For such a purpose, a Geographic Information System (GIS) of Niamey was implemented from a SPOT satellite image (CNES 2008) using the following seven land cover categories (LCC): Niger river, ponds, bare soils, tarred areas, trees, other greenings and sheet steel-made roofs.
The local urban landscape was described in the vicinity of each of the 11 sampling points where C. gambianus and/or A. niloticus specimens were caught (Table 1). To do so, circular buffers of 30 m radius were centered upon each sampling location, and the corresponding landscape was extracted using the R software [38] and the package “raster” [39]. Each circular landscape was described using the percentage of landscape (PLAND) composition metric computed for each LCC [40] using the R package SDMTools [41]. This led to a set of 7 PLAND values (one for each LCC) for each sampling location.
The second step of the analysis consisted in processing these compositional data through a Principal Component Analysis [42] using the R package “ade4” [43]. The first Principal Component (PC1) partly, but highly significantly separated the locations with and without trapped rodents (Monte-Carlo test of between-group inertia, 999 replicates, p = 0.009; [44]). The locations with rodents were associated to high values of PLAND for trees and greenings and low proportion of bare soil and sheet steel-made roofs.
As a third step, we rasterized the GIS of Niamey into 60x60 meters cells (N = 67,077) within which the percentage of each PLAND was computed. These pixels were then projected onto PC1 as supplementary rows [42]. Their coordinates onto PC1 thus represented their relative position with regards to the gradient of habitat suitability for Leptospira-carrying rodent species. The pixel coordinate comprised within the range of the coordinates of locations where Leptospira-carrying rodent species were standardized to range between 0 and 1 and subsequently mapped (Fig 2). As such, this map depicts the city-wide spatial variation of the similarity between local habitat and average landscape composition of locations where Leptospira-carrying rodent species were caught. In other words, it shows the distribution of suitable habitats for rodent-borne Leptospira within Niamey. Expectedly, most of the build-up areas of the city were retrieved as unsuitable for rural-like hence Leptospira-carrying rodent species (Fig 2).
In total, 578 rodents from 49 localities (Fig 1) and five main categories of habitats (i.e., households, markets, coach station, gardens and factories, the latter including a slaughter house, a husking rice industry and an industrial storeroom; Table 1) were investigated for the presence of Leptospira using one to four complementary molecular approaches (i.e., metabarcoding, RT-PCR, sequencing, VNTR profiles; Table 1).
First, 550 individuals were screened in triplicated species-specific pools through bacterial 16S metabarcoding (Table 1). A total amount of 287,057 16S sequences was obtained. Among them (which include some bacterial genera of potential medical interest such as Helicobacter, Orientia, Mycoplasma, Streptococcus, Ignatzschineria), the three replicates of the A. niloticus-specific pool were found positive for Leptospira (3385, 3144 and 3050 Leptospira sequences, respectively) while no Leptospira sequence were retrieved for the other pools, with the only exception of one Mastomys-specific pool in which 37, 58 and 64 Leptospira sequences were found. Such a low amount of sequences was intriguing and, after close verifications, we found that that one Leptospira-positive Arvicanthis individual had been added by error to this slightly positive Mastomys-pool (NB: this had been noted on the bench book but this pipetting mistake was then omitted). In order to unambiguously confirm that this lab error was responsible for the few Leptospira sequences retrieved within this particular pool, each Mastomys individual was screened using the lipL32 RT-PCR-based procedure: no positive Mastomys could be found.
Second, the 50 Arvicanthis niloticus specimens from the NGS positive pool as well as 16 additional A. niloticus and 12 C. gambianus individuals (that had not been included in the metabarcoding survey) were investigated individually using duplicated lipL32-targeting qPCR (Table 1). These 78 animals originated from sites J-LMO1, J-LMO2, J-NOG, J-CYA, J-DAR, J-GAM, J-KIR1 and CRA-3 (Table 1). Among them, seven animals trapped in J-GAM, J-KIR1 and J-LMO2 appeared Leptospira-positive twice (with Ct ≤ 31), while one from J-GAM was found positive in only one of the two duplicates (Ct = 38.2). In addition, 12 Cricetomys gambianus from sites J-LMO1, J-KIR2, CRA-1 and CRA-2 were also investigated through RT-PCR: one of them (from CRA-2) was found twice Leptospira-positive (Ct = 20.3 and 20.5).
Third, the DNA of the seven A. niloticus and the C. gambianus qPCR-positive individuals were successfully amplified and sequenced for the Leptospira rrs gene (only the A. niloticus that was qPCR-positive in one of the two duplicates could not be amplified). All eight sequences (Genbank accession numbers KT583752 to KT593759) were found strictly identical (whatever the rodent host species) and, following a Blastn procedure, strictly identical to L. kirschneri sequences (100% identity; 100% sequence cover; E value = 4.e-136; the subsequent most similar sequences belonged to L. interrogans with 99% identity, 100% sequence cover and E value = 1.e-134).
MLVA is a simple and rapid PCR-based method for the identification of most of the serovars of L. interrogans and L. kirschneri [37]. MLVA of the VNTR-4, VNTR-7, and VNTR-10 loci were performed in all nine RT-PCR-positive individuals. No PCR product was obtained for the sample that had been found positive in only one out of the two RT-PCR duplicated screenings while two different patterns were retrieved for the remaining ones. First, all Arvicanthis samples belonged to genotype I (i.e., VNTR4: 450bp, VNTR7: 320bp, VNTR10: 350bp). Second, genotype II was only represented in the single Leptospira-positive C. gambianus specimen (i.e., VNTR4: 370bp, VNTR7: 320bp, VNTR10: no amplified product). None of these genotypes I and II have been described previously.
All the Leptospira-carrying rodents identified in Niamey were trapped in February, October and November. These months all correspond to the dry and cool season. Nevertheless, our sampling did not allow us to investigate seasonality in a satisfying manner, especially within the urban gardens where most rodents were caught in February, October and November, except for one individual trapped in July and two specimens trapped in March.
Our study allows us to highlight for the first time the presence of pathogenic leptospires in Niger. At a wider scale, our data also add to the very rare mentions of Leptospira spp. in the Sahel [14], thus confirming that these bacteria do circulate in Sub-Saharan Africa more extensively than currently thought. Moreover, our molecular investigations showed that rodent-borne Leptospira in Niamey belonged to L. kirschneri and to a genotype that had never been identified previously. Its biological features and medical impact, including its virulence in human, remain to be studied in details.
Leptospirosis is one of the most widespread zoonotic diseases around the World. In tropical areas, contact with contaminated water following heavy rainfall and flooding episodes is thought to be a major risk of exposure to pathogenic Leptospira spp. [45]. In temperate regions, infection mode is less clear, with recreational water activities and animal caretaking potentially also being of epidemiological importance [4]. In developing countries, high infection rates were also found in cities, essentially within disadvantaged urban areas that usually show poor sanitation and where rodents are numerous (e.g., [17–19, 46, 47]). Here, we point towards a potential other major context of Leptospira infection risk in the tropics, namely the market garden areas that surround most cities in developing countries, including those that lie within semi-arid regions.
Indeed, rats are usually considered as the major rodent reservoirs for leptospires worldwide [48]. In Eastern Africa, Mastomys natalensis is thought to be the principal source of human infection [49]. Rattus rattus and M. natalensis are from far the most abundant species that were found within Niamey [27]. Yet, out of the 450 specimens of these two species that were tested here, none could be found Leptospira-positive. On the contrary, only Arvicanthis niloticus and Cricetomys gambianus specimens, all trapped within urban market gardens, were detected as carrying Leptospira. This strongly suggests that Leptospira spp. circulate mostly, if not only in these particular habitats. This is tempting to speculate that irrigated gardens and rice fields along the Niger River provide the warm and moist environmental conditions that favor the bacterium circulation with both the presence of mammalian hosts such as rodents, human-maintained humidity of soils and free water. The absence of rodent-borne leptospires elsewhere in town despite the abundance of potential competent hosts (especially Rattus rattus and Mastomys natalensis; [27]) as well as poor sanitation conditions would be explained by long-term aridity, thus strongly contrasting with the situation observed in other wetter tropical cities.
The importance of environmental factors in the epidemiology of pathogenic Leptospira species has already been suggested in Thailand where the commensal species Rattus exulans was found infected much less frequently than other rural / wild species [34]. Ganoza and colleagues [46] further suggested that anthropogenic modification of the urban habitat was a major driver of leptospiral transmission to human. With this in mind, our study emphases the potentially highly critical role of urban market gardening in leptospirosis epidemiology since horticulture rapidly extends within and around towns of most developing countries. In sub-Saharan Africa, these so-called green cities are considered as a trump card to reach the “zero hunger” challenge [50]. For instance, urban and peri-urban horticulture produces most of all leafy vegetables that are consumed in Accra (Ghana), Dakar (Senegal), Bangui (Central African Republic), Brazzaville (Congo), Ibadan (Nigeria), Kinshasa (Democratic Republic of Congo) and Yaoundé (Cameroon), which represent a total population of 22.5 million inhabitants [50]. Yet, the setup of agricultural spaces in close proximity to, when not inside cities or villages raise public health issues since they may favor the maintaining of some pathogenic agents and eventually their vectors or reservoirs, hence potentially increasing the risk of human exposure to the associated diseases, such as malaria (e.g., in Benin: [51, 52]; in Ghana: [53]), various gastro-intestinal infections (e.g., in Benin: [54]) schistosomiasis (e.g., in Ivory Coast: [55, 56] in Niger: [57]), leptospirosis (this study) or potentially toxoplasmosis (e.g., in Niamey, Niger: [58]).
Fine-scale studies show that the impact of these infectious agents may vary at very local scale, depending on the habitat structure and use (e.g., [55, 56]). In the same manner, in Brazilian slums, human cases of leptospirosis seems to aggregate at the very local scale of some households [59], thus suggesting that city-scale studies are inadequate to fully understand the disease epidemiology [48]. These findings, together with our first description of rodent-borne pathogenic Leptospira within urban market gardens of Niamey, suggest that investigations are now required in order to (i) provide a more precise picture of Leptospira circulation within the urban farming zones of this Sahelian city, and (ii) to look whether human transmission evidence indeed exists in Niger. If this was to be the case, leptospirosis may well represent an important amount of the numerous cases of “fever of unknown origin” that mimic malaria in this semi-arid area. Our GIS-based inferences of suitable areas for Leptospira-carrying rodent species in Niamey clearly correspond to intra-city agricultural zones, especially those along the Niger River and the Gountou Yéna wadi (Fig 2). This suggests that human populations at higher risk may well be urban farmers as well as all people that are in close contact with the river waters for their everyday activities (e.g., fishing, clothes and dish washing, bathing, etc). This is the reason why we recommend that investigations about human prevalence are conducted in these areas where leptospires may represent a very impacting though under-diagnosed health issue. Finally, climatic change together with human-mediated modifications of land use accentuates Niger River-associated flooding events (see, for instance, the dramatic episodes that occurred in Niamey in 2010, 2012 and 2013; [60, 61]). From there, we anticipate an increase of leptospirosis’ impact on human health in Niamey in a near future.
|
10.1371/journal.pcbi.1004075 | Proportionality: A Valid Alternative to Correlation for Relative Data | In the life sciences, many measurement methods yield only the relative abundances of different components in a sample. With such relative—or compositional—data, differential expression needs careful interpretation, and correlation—a statistical workhorse for analyzing pairwise relationships—is an inappropriate measure of association. Using yeast gene expression data we show how correlation can be misleading and present proportionality as a valid alternative for relative data. We show how the strength of proportionality between two variables can be meaningfully and interpretably described by a new statistic ϕ which can be used instead of correlation as the basis of familiar analyses and visualisation methods, including co-expression networks and clustered heatmaps. While the main aim of this study is to present proportionality as a means to analyse relative data, it also raises intriguing questions about the molecular mechanisms underlying the proportional regulation of a range of yeast genes.
| Relative abundance data is common in the life sciences, but appreciation that it needs special analysis and interpretation is scarce. Correlation is popular as a statistical measure of pairwise association but should not be used on data that carry only relative information. Using timecourse yeast gene expression data, we show how correlation of relative abundances can lead to conclusions opposite to those drawn from absolute abundances, and that its value changes when different components are included in the analysis. Once all absolute information has been removed, only a subset of those associations will reliably endure in the remaining relative data, specifically, associations where pairs of values behave proportionally across observations. We propose a new statistic ϕ to describe the strength of proportionality between two variables and demonstrate how it can be straightforwardly used instead of correlation as the basis of familiar analyses and visualization methods.
| Relative abundance measurements are common in molecular biology: nucleic acids typically have to be provided at a set concentration for sequencing or microarray analysis; sequencing methods report a large but finite total of reads, of which any particular sequence is a proportion. Sometimes, researchers are interested in the relative abundance of different components. Other times, they have to make do with relative abundance to gain insight into the system under study. Whatever the case, data that carry only relative information need special treatment.
Awareness is growing [1, 2, 3] but it is not yet widely appreciated that common analysis methods—including correlation—can be very misleading for data carrying only relative information. Compositional data analysis [4] (CoDA) is a valid alternative that harks back to Pearson’s observation [5] of ‘spurious correlation’, i.e., while statistically independent variables X, Y, and Z are not correlated, their ratios X/Z and Y/Z must be, because of their common divisor. (Note: this differs from the logical fallacy that “correlation implies causation”.)
Proportions, percentages and parts per million are familiar examples of compositional data; the fact that the representation of their components is constrained to sum to a constant (i.e., 1, 100, 106) emphasizes that the data carry only relative information. Note that compositional data do not necessarily have to sum to a constant; what is essential is that only the ratios of the different components are regarded as informative.
Correlation—Pearson, Spearman or other—leads to meaningless conclusions if applied to compositional data because its value depends on which components are analyzed [4]. Problems with correlation can also be demonstrated geometrically (Fig. 1): the bivariate joint distribution of relative abundances says nothing about the distribution of absolute abundances that gave rise to them. Thus, relative data is also problematic for mutual information and other distributional measures of association. To further illustrate how correlation can be misleading we applied it to absolute and relative gene expression data in fission yeast cells deprived of a key nutrient [6].
How then can we make sound inferences from relative data? We show how proportionality provides a valid alternative to correlation and can be used as the basis of familiar analyses and visualizations. We conclude by putting this analysis strategy in perspective, discussing challenges, caveats and issues for further work, as well as the biological questions raised in this study.
Our results are based on data from Marguerat et al. [6] on the absolute levels of gene expression (i.e., mRNA copies per cell) in fission yeast after cells were deprived of a key nutrient (Fig. 2). Unlike many experiments where researchers ensure (or assume) cells produce similar amounts of mRNA across conditions [7], this experiment ensured cells produced very different amounts so as to illustrate the merits of absolute quantification (S1 Fig.). Total abundance may vary dramatically in other experimental settings—such as in comparing diseased and normal tissues, tissues at different stages of development, or microbial communities in different environments.
To illustrate the key points of this paper, we worked with positive data only (i.e., we excluded records with any zero or NA values): measurements of 3031 components (i.e., mRNAs) at 16 time points. Furthermore, we applied analysis methods (specifically, correlation) to the absolute abundance data without transformation (e.g., taking logarithms) because we believe this approach yields useful insights and simplifies the presentation of the central ideas of this paper (see [8] and S1 Supporting Information).
Before looking at issues with pairs of components, it is important to note that interpreting differences in the relative abundance of a single component can be challenging.
Tests for differential expression are popular for analyzing relative data in bioscience. Much attention has been given to dealing with small numbers of observations and large numbers of tests, but comparatively little to “…the commonly believed, though rarely stated, assumption that the absolute amount of total mRNA in each cell is similar across different cell types or experimental perturbations” [7].
The relationship between the relative and absolute abundance of a component can be understood in terms of fold change over time. When total absolute abundance of mRNA stays constant, fold changes in both absolute and relative abundance of each mRNA are equal. When total absolute abundance varies, fold changes in absolute and relative abundances of each mRNA are no longer equal and can change in different directions. Between 0 and 3 hours there were 1399 yeast mRNAs whose absolute abundance decreased, and whose relative abundance increased. Clearly, mRNAs are being expressed differently, but to describe them as “under- or over-expressed” is too simplistic—here lies the interpretation challenge (see S1 Supporting Information).
While “differential expression” of relative abundances is challenging to interpret, in the absence of any other information or assumptions, correlation of relative abundances is just wrong. We stress in the absence of any other information or assumptions to highlight the common assumption of constant absolute abundance of total mRNA across all experimental conditions. If this assumption holds, and all the mRNAs comprising that total are considered, the relative abundance of each kind of mRNA will be proportional to its absolute abundance, and analyses of correlation or “differential expression” of the relative values will have clear interpretations. The revisitation of this assumption [7] should raise alarm bells about the inferences drawn from many gene expression studies.
Fig. 1(a) shows why correlation between relative abundances tells us nothing about the relationship between the absolute abundances that gave rise to them: the perfectly correlated relative abundances could come from any set of absolute abundance pairs that lie on the rays from the origin. This many-to-one mapping means that other measures of statistical association (e.g., rank correlations or mutual information) will not tell us anything either when applied to purely relative data.
But is this problem just a theoretical construct? A rare issue? Consider the red mRNA pair in Fig. 2: while their absolute abundances over time are strongly positively correlated, if someone (inappropriately) used correlation to measure the association between the relative abundances of these two mRNAs they would form the opposite view (Fig. 3(a)); correlation between the blue mRNA pair in Fig. 2 is similarly misleading (S2 Fig.). What of the other 4.5 million pairs of mRNAs? Fig. 3(b) summarizes all discrepancies between correlations of absolute abundance, and correlations of relative abundance, showing clearly that the apparent correlations of relative abundances tell a very different story from those of the absolute data. So how should we go about analyzing these relative data?
CoDA theory provides three principles [4, 9]:
Scale invariance: analyses must treat vectors with proportional positive components as representing the same composition (e.g., (2, 3, 4) is equivalent to (20, 30, 40))
Subcompositional coherence: inferences about subcompositions (subsets of components) should be consistent, regardless of whether the inference is based on the subcomposition or the full composition.
Permutation invariance: the conclusions of analyses must not depend on the order of the components.
Correlation is not subcompositionally coherent: its value depends on which components are considered in the analysis, e.g., if you deplete the most abundant RNAs from a sample [10] and use correlation to measure association between relative abundances, you get different correlations to the undepleted sample (S3 Fig.).
Proportionality obeys all three principles for analyzing relative data. If relative abundances x and y are proportional across experimental conditions i, their absolute abundances must be in proportion:
x
i
t
i
∝
y
i
t
i
⇒
x
i
∝
y
i
where ti is the total abundance in condition i (Fig. 1(b)).
We proposed a “goodness-of-fit to proportionality” statistic ϕ to assess the extent to which a pair of random variables (x, y) are proportional [11]. ϕ is related to logratio variance [4], var(log(x/y)), and is zero when x and y behave perfectly proportionally. However, when x and y are not proportional, ϕ has both a clear geometric interpretation and a meaningful scale, addressing concerns raised about logratio variance [3]: the closer ϕ is to zero, the stronger the proportionality. We consider “strength” of proportionality (goodness-of-fit) rather than testing the hypothesis of proportionality because it allows us to compare relationships between different pairs of mRNAs (S1 Supporting Information).
We calculated ϕ for the relative abundances of all pairs of mRNAs and compared it to the correlations between their absolute abundances (S4 Fig.): clearly, the absolute abundances of most mRNA pairs are strongly positively correlated; far fewer are also strongly proportional. Focusing on these strongly proportional mRNAs, we extracted the 424 pairs with ϕ < 0.05. We graphed the network of relationships between these mRNAs (S5 Fig.), an approach similar to gene co-expression network [12] or weighted gene co-expression analysis [13] but founded on proportionality and therefore valid for relative data. The network revealed one cluster of 96, and many other smaller clusters of mRNAs behaving proportionally across conditions. Using ϕ as a dissimilarity measure, we formed heatmaps of the three largest clusters (S6 and S7 Figs.) similar to the method of Eisen et al. [14] but, again, using proportionality not correlation.
This paper does not deny pairwise statistical associations between absolute abundances. What it does say is that once all the absolute information has been removed, only a subset of those associations will reliably endure in the remaining relative data, specifically, associations where values behave proportionally across observations.
Other researchers have recognized the compositional nature of molecular bioscience data, including [15] as discussed in [16]. Strategies have been proposed to ameliorate spurious correlation in the analysis of relative abundances [2, 3]. We contend that there is no way to salvage a coherent interpretation of correlations from relative abundances without additional information or assumptions; our argument is based on Fig. 1.
ReBoot [2] attempts to establish a null distribution of correlations against which bootstrapped estimates of correlations can be compared. Aitchison articulates problems with this approach [4, p.56–58]. SparCC [3] injects additional information by assuming the number of different components is large and the true correlation network is sparse. This equates to assuming “that the average correlations [between absolute abundances] are small, rather than requiring that any particular correlation be small” [3, Eq.14]. This means the expected value of the total absolute abundance will be constant (as the sum of many independently distributed amounts). We are concerned with situations where that assumption cannot be made, or where the aim is to describe associations between relative amounts.
We are also keen to raise awareness that correlation (and other statistical methods that assume measurements come from real coordinate space) should not be applied to relative abundances. This is highly relevant to gene coexpression networks [12]. Correlation is at the heart of methods like Weighted Gene Co-expression Network Analysis [13] and heatmap visualization [14]. These methods are potentially misleading if applied to relative data. This concern extends to methods based on mutual information (e.g., relevance networks [17]) since, as Fig. 1 shows, the bivariate joint distribution of relative abundances (from which mutual information is estimated) can be quite different from the bivariate joint distribution of the absolute abundances that gave rise to them.
Measures of association produce results regardless of the data they are applied to—it is up to the analyst to ensure that the measures are appropriate to the data. Currently, there are many gene co-expression databases available that provide correlation coefficients for the relative expression levels of different genes, generally from multiple experiments with different experimental conditions (see e.g., [18]). As far as we are aware, none of the database providers explicitly address whether absolute levels of gene expression were constant across experimental conditions. If the answer to this question is “no”, we would not recommend these correlations be used for the reasons demonstrated in this paper. If the answer is “yes” we still advocate caution in applying correlation to absolute abundances for reasons discussed in S1 Supporting Information.
While the main aim of this study is to present and illustrate principles for analyzing relative abundances, it has also uncovered intriguing biological insight with respect to gene regulation.
The largest cluster of proportionally regulated mRNAs (96 genes, S1 Supporting Information) was highly enriched for mRNAs down-regulated as part of the core environmental stress response [19], including 66 mRNAs that encode ribosomal proteins, and the remaining mRNAs also associated with roles in protein translation, such as ribosome biogenesis, rRNA processing, tRNA methyltransferases and translation elongation factors. The absolute levels of these mRNAs decrease after removal of nitrogen [6]. The notable coherence in biological function among the mRNAs in this cluster is higher than typically seen when correlative similarity metrics for clustering are applied (e.g., [19]). These 96 mRNAs show remarkable proportionality to each other over the entire timecourse (S8 Fig.), and maintain near constant ratios across all conditions (S9 Fig.). Given the huge energy invested by yeast cells for protein translation (most notably ribosome biogenesis [20, 21], it certainly makes sense for cells to synchronize the expression of relevant genes such that translation is finely tuned to nutritional conditions.
Evidently, numerous ribosomal proteins and RNAs function together in the ribosome, demanding their coordinated expression; more surprisingly, multiple other genes, with diverse functions in translation, show equally pronounced proportional regulation across the timecourse. These findings raise intriguing questions as to the molecular mechanisms underlying this proportional regulation, suggesting sophisticated, coordinated control of numerous mRNAs at both transcriptional and post-transcriptional levels of gene expression.
While proportionality and the ϕ-statistic provide a valid alternative to correlation for relative data, there are still some challenges in their application. First is the treatment of zeroes, for which there is currently no simple general remedy [22]. Second, and related, is the fact that “many things that we measure and treat as if they are continuous are really discrete count data, even if only at the molecular extremes” [23] and count data is not purely relative—the count pair (1, 2) carries different information than counts of (1000, 2000) even though the relative amounts of the two components are the same. Correspondence analysis [24], or methods based on count distributions (e.g., logistic regression and other generalized linear models) may provide ways forwards.
All data and code [25] needed to reproduce the analyses and visualizations set out in this paper are contained in the Supporting Information, along with additional illustrations and detailed explanations.
The “goodness-of-fit to proportionality” statistic ϕ can be used to assess the extent to which a pair of random variables (x, y) are proportional [11]. Aitchison [4] proposed logratio variance, var(log(x/y)), as a measure of association for variables that carry only relative information. When x and y are exactly proportional var(log(x/y)) = 0, but when x and y are not exactly proportional, “it is hard to interpret as it lacks a scale. That is, it is unclear what constitutes a large or small value… (does a value of 0.1 indicate strong dependence, weak dependence, or no dependence?)” [3]. Logratio variance can be factored into two more interpretable terms:
var
(
log
(
x
/
y
)
)
=
var
(
log
x
−
log
y
)
=
var
(
log
x
)
+
var
(
log
y
)
−
2
cov
(
log
x
,
log
y
)
(1)
=
var
(
log
x
)
⋅
(
1
+
var
(
log
y
)
var
(
log
x
)
−
2
var
(
log
y
)
var
(
log
x
)
cov
(
log
x
,
log
y
)
var
(
log
x
)
var
(
log
y
)
)
=
var
(
log
x
)
⋅
(
1
+
β
2
−
2
β
|
r
|
)
≜
var
(
log
x
)
⋅
ϕ
(
log
x
,
log
y
)
(2)
where β is the standardized major axis estimate [26] of slope of random variables log y on log x, and r the correlation between those variables. The first term in Equation 2, var(log x), is solely about the magnitude of variation at play and has nothing to do with y. The second term, ϕ, describes the degree of proportionality between x and y, and forms the basis of our analysis of the relationships between relative values. Other non-negative functions of β and r that are zero when x and y are perfectly proportional could be formed; this is described in more detail in S1 Supporting Information, as well as why ϕ is preferable to an hypothesis testing approach. There is no need to calculate β or r to assess strength of proportionality; they simply provide a clear geometric interpretation of ϕ; in practice, one can use the relationship ϕ(log x, log y) = var(log(x/y))/var(log x).
The ϕ statistic is a measure of goodness-of-fit to proportionality that combines two quantities of interest: β, the slope of the line best describing the relationship between random variables log x and log y; and r, whose magnitude estimates the strength of the linear relationship between log x and log y. “Goodness-of-fit” describes how well a statistical model fits a set of observations and is a familiar concept in regression, including linear and generalised linear models, but note that ϕ—specifically the slope (β) of the standardized major axis—is motivated by allometry rather than regression modeling. We are interested in assessing whether two variables are directly proportional, rather than predicting one from the other: “use of regression would often lead to an incorrect conclusion about whether two variables are isometric or not” [26, p.265]. Note also that ordinary least squares regression fits are not symmetric: in general, the slope of y regressed on x is different to the slope of x regressed on y [27].
While goodness-of-fit measures for regression may not generally be appropriate for assessing proportionality, Zheng [28] explores the concordance correlation coefficient ρc [29] which could be modified to provide an alternative measure of proportionality defined as
ρ
p
(
log
x
,
log
y
)
≜
2
cov
(
log
x
,
log
y
)
var
(
log
x
)
+
var
(
log
y
)
and related to var(log(x/y)) by the terms in Equation 1. This “proportionality correlation coefficient” ranges from −1 (perfect reciprocality) to +1 (perfect proportionality) and lacks the clear geometric interpretation of ϕ.
We have used ϕ(log x, log y) to emphasize the relationship between ϕ and logratio variance. However to ensure that the ϕ values for component pair (i, j) are on the same scale (i.e., comparable to) the ϕ values for component pair (m, n), it is necessary to use the centered logratio (clr) transformation instead of just the logarithm (S1 Supporting Information). The clr representation of composition x = (x1, …, xi, …, xD) is the logarithm of the components after dividing by the geometric mean of x:
clr
(
x
)
=
log
x
1
g
m
(
x
)
,
⋯
,
log
x
i
g
m
(
x
)
,
⋯
,
log
x
D
g
m
(
x
)
,
ensuring that the sum of the elements of clr(x) is zero. Note that dividing all components in a composition by a constant (i.e., the geometric mean gm(x)) does not alter the ratios of components.
Gene co-expression networks [12, 13] are generally based on a pairwise distance or dissimilarity matrix which is often a function of correlation and thus not appropriate for relative data. Proportionality is appropriate, but ϕ does not satisfy the properties of a distance—most obviously, it is not symmetric unless β = 1:
ϕ
(
log
x
,
log
y
)
=
1
+
β
2
-
2
β
|
r
|
ϕ
(
log
y
,
log
x
)
=
1
+
1
β
2
-
2
1
β
|
r
|
.
We are most interested in pairs of variables where β and r are near 1 and want to preserve the link between ϕ(log x, log y), β and r. Hence, our approach to forming a dissimilarity matrix is simply to work with ϕ(log xi, log xj) where i < j, in effect, the lower triangle of the matrix of ϕ values between all pairs of components. This symmetrised form of ϕ was then used to lay out a network of the 145 mRNAs that were involved in 424 pairwise relationships with ϕ < 0.05. We used the symmetrised form of ϕ as the basis of the cluster analysis and heatmap expression pattern display (e.g., S10 Fig.) described by Eisen et al. [14].
|
10.1371/journal.ppat.1002946 | Global Assessment of Genomic Regions Required for Growth in Mycobacterium tuberculosis | Identifying genomic elements required for viability is central to our understanding of the basic physiology of bacterial pathogens. Recently, the combination of high-density mutagenesis and deep sequencing has allowed for the identification of required and conditionally required genes in many bacteria. Genes, however, make up only a part of the complex genomes of important bacterial pathogens. Here, we use an unbiased analysis to comprehensively identify genomic regions, including genes, domains, and intergenic elements, required for the optimal growth of Mycobacterium tuberculosis, a major global health pathogen. We found that several proteins jointly contain both domains required for optimal growth and domains that are dispensable. In addition, many non-coding regions, including regulatory elements and non-coding RNAs, are critical for mycobacterial growth. Our analysis shows that the genetic requirements for growth are more complex than can be appreciated using gene-centric analysis.
| The significant rise in drug resistant strains of Mycobacterium tuberculosis has highlighted the need for new drug targets. Here, we present a novel method of defining genetic elements required for optimal growth, a key first step for identifying potential drug targets. Similar strategies in other bacterial pathogens have traditionally defined a set of essential protein-coding genes. Bacterial genomes, however, contain many other genetic elements, such as small RNAs and non-coding regulatory sequences. Protein-coding genes themselves also often encode more than one functional element, as in the case of multi-domain genes. Therefore, instead of assessing the quantitative requirement of whole genes, we parsed the genome into comprehensive sets of overlapping windows, unbiased by annotation, and scanned the entire genome for regions required for optimal growth. These required regions include whole genes, as expected; but we also discovered genes that contained both required and non-required domains, as well as non protein-coding RNAs required for optimal growth. By expanding our search for required genetic elements, we show that Mycobacterium tuberculosis has a complex genome and discover potential drug targets beyond the more limited set of essential genes.
| Mutagenesis has long been a powerful tool for understanding the roles of genes and other chromosomal elements. Recently, high-density transposon insertion mutagenesis coupled with deep sequencing has enabled comprehensive identification of the required genes in many important bacterial pathogens [1]–[6]. Defining the protein-coding genes required for bacterial growth identifies both key biological processes and potential targets for drug development. However, protein-coding genes are not the only genetic elements that code for required functions. In densely packed bacterial genomes, many regulatory regions are required for appropriate expression of genes [7]. Moreover, all organisms produce large numbers of non-coding RNAs that can be important under a variety of growth conditions [8]–[10]. Gene-oriented analyses also look past cases wherein a single gene encodes several differentially important protein domains.
Here, rather than focusing on genes, we take an unbiased approach to create a comprehensive understanding of genomic requirement in Mycobacterium tuberculosis (Mtb). We model the Mtb genome as made up of “functional units”, a term that encompasses both genes and other genetic elements, many of which have yet to be annotated. By not limiting our analysis to whole-gene regions, we can find otherwise unidentified functional units while also gaining a more nuanced view of the genes required for mycobacterial growth, including critical domains within proteins and non-protein-coding regions that play important roles.
We find approximately 300 protein-coding genes wherein only portions of the coding sequence are required. These include genes, such as ppm1 and fhaA, where we demonstrate that one domain is required for optimal growth whereas other domains are not. Our unbiased analysis also revealed required genomic elements in regions sitting between protein-coding genes. These include two RNAs, the tmRNA and the RNA component of RNaseP, which are required for optimal growth. In addition, we find a number of other regions that influence viability by uncharacterized mechanisms, but whose effects have previously been overlooked by gene-centric analyses.
To perform a comprehensive assessment of Mtb's genetic requirements for growth, we used two ∼100,000-clone Mtb libraries generated through high-density transposon mutagenesis of the H37Rv strain [11]. We generated a library of single-insertion mutants by phage delivery of the Himar1 transposon, which randomly inserts into the genome at sites recognized by the TA dinucleotide (Figure 1A). We then created transposon-mapping probes by selectively amplifying and sequencing transposon-genome junctions using an Illumina Genome Analyzer 2. Using genome sequences adjacent to the transposon genomic sequences, we were able to map the insertion site of mutants in the library (Figure 1B) and count the reads mapped to each insertion site (insertion count, Table S1).
We reasoned that the insertion count should reflect the number of corresponding mutants in the library. To demonstrate this, we picked twelve individual transposon mutants and added each at a known quantity to a manually constructed library. Insertions were again mapped and counted by deep sequencing, and the insertion counts for each site was compared to the known relative quantities of each mutant in the pool (Figure 1C). Insertion counts were highly correlated with the known relative amount of each mutant (Pearson R = 0.880, p-value<0.0001, n = 14). Additionally, we confirmed that insertion counts accurately reflected the library's genome composition by counting the genome-transposon templates represented in our Illumina reads. Since random shearing events create the genome-transposon templates for amplification, the distance between the transposon and the sheared end represents a unique identifier for each template. We assessed the relationship between estimates of unique template molecules for each TA site and the read count for that site (Figure 1D), revealing excellent correlation (Pearson R = 0.945 p-value<0.0001 n = 36,488).
In our Mtb library, transposon insertions occurred at 36,488 of the 72,927 possible insertion sites (TA dinucleotides). Each library generated an average of 2.3 million reads, resulting in a mean insertion count of 64 per hit-site. We counted the number of sequencing reads from each site in the two libraries and compiled the counts correcting for each library's total insertion count.
Having demonstrated that insertion counts faithfully represent mutant numbers (Figure 1C), we used insertion counts to comprehensively assess the relative importance of selected genomic regions. We defined a region as required for optimal growth if total regional insertions were statistically underrepresented compared to genomic controls (Figure 2A). Required regions, therefore, are those in which mutations result in a statistically validated growth defect. We employed a non-parametric test to assess statistical underrepresentation and regions with a p-value of less than 0.01 and a false discovery rate (Benjamini-Hochberg) of less than 0.1 were defined as required for optimal growth in vitro.
Instead of assessing the requirement for growth of genomic regions based on predetermined gene coordinates, we divided the genome into contiguous overlapping windows to assess a comprehensive set of potential functional units. Our non-parametric test was powered to find significant regions containing at least 7 TA sites (6 or fewer precluded confident rejection of a null hypothesis of variation by chance alone). Thus, we focused on regions of sizes likely to contain 7 or more TA sites. The mean number of TA sites in windows of 400, 500, and 600 bp was 6.75, 8.45 and 10.12, respectively, and were thus used for our sliding window analysis of functional requirement for growth. Intergenic (IG) regions are relatively AT-rich in the Mtb genome, allowing us to add a 250 bp sliding window (mean of 6.20 TA sites in IG regions) to the analysis of IG regions. To lower the computational demands of this analysis, we chose to analyze every tenth window, reasoning also that functional units were unlikely to be smaller than 10 base pairs. Thus, we assessed the requirement for growth of every tenth 400, 500, and 600 bp window in the genome, along with every tenth 250 bp window in regions between protein-coding genes (Figure 2B, Table S4).
We overlaid the coordinates of known genes on the generated results to find those that contained regions required for optimal growth. Of the 3,989 annotated genes, 742 contained required functional units, 3,089 contained no required functional units, while 158 did not sustain insertions but also did not contain enough TAs to meet statistical requirements (Figure 3A, Table S2). As a screen for genes with multiple functional units of varying requirement, we searched for genes that contained both required and non-required regions. A total of 317 genes met these criteria (Figure 3A, Table S2).
Our finding that many genes contain both required and non-required regions suggested that using only whole genes for analysis could misrepresent their importance for growth. Either the entire gene could appear required for optimal growth or the entire gene would be considered dispensable, leaving no room for the possibility that only a segment of the gene might be required. To determine how our results would compare to a gene-centric analysis, we calculated the requirement for growth of each gene by applying the non-parametric test to the gene as a whole. As expected, genes with required segments had a wide range of p-values when assessed using annotated boundaries instead of unbiased overlapping windows (Figure 3B, dark blue bars). A total of 170 out of the 317 (53.6%) had p-values above 0.01, demonstrating that our sliding window strategy accounted for a significant number of required functional units that would be ignored by gene-only strategies (Figure 3B).
A transposon insertion into the 5′ end of a gene will often block production of the encoded protein, either by attenuating transcription or disrupting ribosome binding sites and initiation codons. Thus, we expected that regions required for optimal growth would tend to be found at the 5′ ends of predicted genes. Surprisingly, a plot of the likelihood of discovering an required functional region as a function of the intragenic location revealed a symmetric curve, demonstrating that the required regions discovered have an equal likelihood of residing on either end of the gene (Figure 3C). We hypothesized that this may be because the transposon contains a promoter that can direct downstream transcription. To test this, we took two strains that contained transposon insertions and measured mRNA expression upstream and downstream of the transposon (Figure S2A). In both cases, expression upstream of the transposon did not significantly change, while downstream expression increased (Figure S2B). This is consistent with the observation that downstream genes can be transcriptionally activated by transposon insertions [12]. In addition, mycobacteria are able to use several initiation codons thus making it more likely that truncated but functional proteins can be produced from internal start sites.
While we were able to use this analysis to make many novel observations, we also found that our results supported previous findings. The majority of genes (63%) described as fully required for growth were similarly required in microarray-based studies using transposon site hybridization (TraSH) (Figure S1A, Table S2) [13]. The increased resolution from deep sequencing demonstrated that genes with fewer than 7 TAs resulting in an undersampling that prevented statistically confident requirement assessments (a separate category for genes with 6 or fewer TAs that did not contain insertions is noted in Figure 3A and Table S2). Since this was not known previously, we predicted that the microarray-determined set of required genes would be biased towards small genes. This proved to be true. In genes predicted to be required by TraSH but not in this study, the average number of TAs was 9.90 (Figure S1B). In contrast, the average number of TAs in fully required genes from this study was 19.84, a fair representation of the average of all genes assessed (19.47). In fact, of the genes only determined to be required in TraSH and not in this study, 43% had 7 or fewer TAs, accounting for much of the discordance between the two methods.
A more nuanced analysis of Mtb transposon insertion maps defined essential genes as those that contained “gaps,” any statistically significant runs of potential insertion sites lacking transposon insertions [3]. As expected, genes found in our sliding window analysis to have both required and non-required regions were more concordant with essential genes found by sequencing using this gap analysis than with microarray approaches or whole-gene analyses of insertion counts. Of genes described in our approach as fully required, 97.1% were described as “essential” by Griffin et al (Figure S1A), a remarkable level of agreement given the differences in growth media between the two studies. The increased concordance extended to genes containing both required and non-required regions. Griffin et al. described 151 of these genes as essential, while microarray methods only deemed 81 to be essential. The search for required regions within genes, a feature of both analyses, allowed for the discovery of these regions in longer genes, as evidenced by the increase in average number of TAs within these genes (Figure S1B).
We find that, in some genes, encoded domains have different effects on growth, accounting for the varying degrees of requirement across the gene's open reading frame. For example, the gene encoding Ppm1, an enzyme in the lipoarabinomannan (LAM) synthesis pathway, encodes a protein with two distinct domains. The region encoding the carbon-nitrogen hydrolase domain of Ppm1 sustained many insertions, while the region encoding the C-terminal glycosyl transferase was required for optimal growth (Figure 4A). While the specific requirement of the glycosyl transferase is a novel finding, it resonates with a previous report that only the glycosyl transferase was required for the synthesis of LAM, thought to be an essential cell wall component [14]. Another study revealed that Ppm1 has N-acyltransferase activity, which could be the non-required function of this two-domain protein [15].
To confirm that the lack of insertions in this domain was due to a functional requirement and not to insertional bias or the generation of toxic fusions or truncations, we created transposon libraries in the presence of a second copy of ppm1. We reasoned that a second copy would render the endogenous gene non-required and thus permissive for transposon insertion. We designed footprinting PCR primers upstream of the original ppm1 to specifically generate amplicons containing transposon insertions into the endogenous copy (Figure 4B). Footprinting of the original library confirmed our sequencing results, as no insertions were found in the region encoding the glycosyl transferase. However, in the complemented library, that region did contain insertions, suggesting the glycosyl transferase is functionally required for growth. We further reasoned that only sense insertions—that is, insertions wherein the transposon's internal promoter is oriented in the same direction as the disrupted gene—would be tolerated in the 5′ end of ppm1 to allow for the expression of the C-terminal required domain. To assess this, we used primers specifically designed to amplify sense and anti-sense insertions, and noted only sense insertions in the 5′ end (Figure S2C). In addition, we confirmed that many in-frame internal start sites exist between 5′ transposon insertion sites and the beginning of the 3′ domain.
A recent report showed that FhaA was required for optimal growth of Mycobacterium smegmatis and postulated that the importance of the interaction of FhaA with the essential protein MviN for appropriate regulation of growth and peptidoglycan synthesis [16]. These processes are essential for mycobacterial cell division and cell wall biosynthesis. This work further demonstrated the C-terminal forkhead associated (FHA) domain of FhaA was required for MviN-binding, while an N-terminal domain of unknown function was dispensable for this interaction. In agreement with these findings, we show here that the region of fhaA encoding the FHA domain cannot sustain insertions, while the remainder of the gene is dispensable (Figure 4C). We used insertion footprinting to confirm these results, and found that the C-terminal insertion mutants were rescued for growth in the presence of a second copy of fhaA (Figure 4D).
Notably, both ppm1 and fhaA, which we predict to be required for optimal growth based on the presence of a required region within these genes, were classified as non-essential in a previous microarray-based screen [13]. In fact, 247 of the 328 genes containing both required and non-required regions were not previously described as necessary for growth, likely due to the decreased spatial resolution of microarray-based methods. Microarrays limited the resolution of requirement testing to genes, and each gene received a single metric describing its requirement for growth. In addition, as our approach is not confined to gene boundaries, we have the additional resolution to identify domains within genes, as exemplified by ppm1 and fhaA.
Because we are not limited to annotated regions we were also able to probe the importance of intergenic regions. By scanning the genome for required 250, 400 and 500 bp regions, we found 25 intergenic regions required for optimal growth (Figure 2B and Table S3). These required intergenic regions contained many components of known essential cellular functions to be required for in vitro growth. These included 10 tRNAs as well as the RNA catalytic unit of RNaseP, which has been shown to be required for tRNA processing in other bacteria (Figure 5A). Additionally, one required intergenic region contained the tmRNA, a molecule required to release stalled ribosomes and to tag polypeptides for proteolytic degradation through an essential protease (Figure 5B) [17]–[18]. Of the intergenic segments containing functionally required regions, 11 had annotated functions and an additional 6 were adjacent to genes assessed as required for growth and, therefore, might contain promoters or other transcriptional regulatory elements. The remaining 19 required segments are situated between two non-required genes and, as yet, have no ascribed function.
Finding genetic loci that are required for optimal growth under specific conditions helps inform the basic understanding of bacterial physiology and efforts to develop new therapeutics for pathogens. Previously, we and others have used transposon mutagenesis to infer the requirement for genes under different growth conditions by utilizing the information provided by genome annotations [1]–[6]. Deep sequencing, which allows us to map precisely the insertion site of every mutant, affords a higher resolution assessment of genetic requirement, beyond just genes. Here, we demonstrate that an unbiased sliding window approach harnesses the full potential of this increased resolution. This approach identified not only whole genes required for optimal growth but also other required elements, such as non-protein coding RNAs and protein domains within insertion-containing genes, which would otherwise obscured by gene-centric analysis. An alternative analysis that uses significant gaps in insertion—rather than quantitative insertion counts—was also able to assess the requirement of protein domains (DeJesus et al., unpublished data, submitted). This analysis likely identifies regions absolutely essential for viability rather than all regions required for optimal growth.
We found that many genes contain elements that are important for growth even though other regions are not required. In at least two cases, ppm1 and fhaA, published data have shown that the required regions encode specific protein domains. However, in other cases, these might represent non-protein-coding RNAs or cis regulatory elements. Bacteria encode many small RNAs many of which could be required for optimal growth and some of which are embedded within genes [8]–[10]. In addition, most genes have been annotated computationally, an uncertain pursuit that clearly can lead to misannotated start sites [19]. Genes with only 5′ insertions could fall into this category.
Similarly, important non-protein-coding regions could have multiple roles. In some cases, we found that known RNAs, such as rnpB, the catalytic RNA component of RNase P, and the tmRNA were required for optimal growth, supporting previous speculation [20]–[21]. Again, some other required regions might encode as yet unidentified non-coding RNA molecules. Still others might be promoters or other regulatory regions.
In this study, our resolution was limited by the specific properties of the Himar1 transposon in mycobacteria. Our previous studies have shown that insertions are randomly distributed apart from the desired selection against insertion in essential regions [11], [22]. Despite this, we cannot assume that all sites lacking insertions represent required regions since unknown insertional biases of the transposon may exist. Thus, we defined a required region as one with a statistically underrepresented insertion count using a non-parametric test to account for such potentially unique biases within these data (Figure 2A). This allowed us to exclude, for example, windows with 6 or fewer TA sites, which demonstrably lacked power to distinguish a region as essential for growth relative to background variation. In GC-rich protein-coding regions, this limited our scope to windows of greater than 400 bp; less GC-rich intergenic regions allowed the assessment of windows greater than 250 bp. Thus, while we were able to identify many required protein domains and RNAs, it is certainly possible that smaller elements required for growth were missed due to these size constraints. This is a particular problem for non-coding RNAs that are often very small. For example, while we found 10 tRNAs required for growth, the remaining tRNAs reside in non-coding regions that did not have the requisite number of TA sites to determine requirement. Using the Himar1 transposon in organisms with less of a GC bias, or in organisms in which a less restricted transposon exists, should result in increased resolution [4].
The analysis we used provides a powerful tool to perform functional genome analysis. Importantly, this type of approach is useful not only for single conditions, as we described but can also be used to identify elements critical under one growth condition but not another [23]–[25]. This is particularly important in organisms like Mtb, an obligate pathogen that never grows under conditions precisely comparable to those we use in vitro. Coupling high-density insertion libraries with deep sequencing and analytic methods such as that described here provides a powerful experimental tool for functional genome annotation.
Two independent libraries of 100,000 mutants were generated in the Mtb strain H37Rv as previously described on 7H10 agar [11]. Independent libraries were also generated in Mtb strains overexpressing ppm1 and fhaA. Genomic DNA was isolated from each library and randomly fragmented to 400–600 bp pieces by sonication with a Covaris E220. Nicked ends were repaired (Epicentre end repaired kit), and A-tails were added with Taq polymerase to allow the ligation of T-tailed adapters. Transposon-junctions were amplified for 30 cycles (94 degrees, 30 seconds; 58 degrees, 30 seconds; 72 degrees, 30 seconds) using a primer recognizing the transposon end (5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCCGGGGACTTATCAGCCAACC-3′) and one recognizing the adapter (5′-CAAGCAGAAGACGGCATACGAGATCGGTCTCGGCATTCCTGCTGAACCGCTCTTCCGATCGTCCAGTCTCGCAGATGATAAGG-3′). Primers used during amplification contained all the requisite sequence for binding to the Illumina sequencing platform. A 250–400 bp fragment of the amplicon was isolated from a gel and sequenced on an Illumina GA2 instrument with a custom sequencing primer (5′-TTCCGATCCGGGGACTTATCAGCCAACC-3′).
Reads from the Illumina sequencing run were first screened for the presence of sequence from the end of the Himar1 transposon. The following 35 bases were mapped to the Mtb genome, allowing for 2 mismatches. Reads that mapped to the genome at a TA site were designated as mapped insertions. Reads that mapped to multiple sites were randomly assigned to one of the mapped sites. For each library, the number of reads mapping to each site (insertion counts) was counted. Insertion counts were plotted on IGV and CGViewer [26]–[27].
For every possible region size containing x potential insertion sites, a null distribution of mean read counts was generated by calculating the mean read counts from a set of 10,000 randomly selected sets of x sites. The 10,000 randomly generated means were sorted and the rank of the test region's mean insertion count within the ordered null distribution was determined. The p-value was calculated as the rank of the test mean divided by the size the null distribution (10,000). Multiple test correction was performed by calculating the Benjamini-Hochberg false discovery rate over all regions tested. Regions containing 7 TA sites with no insertions had a p-value of 0.008 and an FDR of 0.06, while regions containing 6 TA sites with no insertions had a p-value of 0.018 and an FDR of 0.12. In order to power our study to detect required regions containing at least 7 TAs, we determined a region to be required for optimal growth if it had a p-value less than 0.01 and an FDR less than 0.1.
Footprinting of transposon insertion sites was performed by PCR using a primer recognizing the Himar1 ITR sequence (5′-CCCGAAAAGTGCCACCTAAATTGTAAGCG-3′) and primers recognizing a genomic segment just upstream of the gene of interest. For directional footprinting, we used one primer to amplify sense insertions (5′-TTTTCTGGATTCATCGACTGTGGC-3′)—where the kanamycin resistance gene on the transposon was oriented in the same direction as the disrupted gene—and another for antisense insertions (5′-CAGCTCATTTTTTAACCAATAGGCCG-3′). Standard PCR conditions were used for long amplification with Phusion polymerase (94 degrees, 15 seconds; primer-dependent annealing temperature, 30 seconds; 72 degrees, 2 minutes).
|
10.1371/journal.pgen.1000109 | Calibrating the Performance of SNP Arrays for Whole-Genome Association Studies | To facilitate whole-genome association studies (WGAS), several high-density SNP genotyping arrays have been developed. Genetic coverage and statistical power are the primary benchmark metrics in evaluating the performance of SNP arrays. Ideally, such evaluations would be done on a SNP set and a cohort of individuals that are both independently sampled from the original SNPs and individuals used in developing the arrays. Without utilization of an independent test set, previous estimates of genetic coverage and statistical power may be subject to an overfitting bias. Additionally, the SNP arrays' statistical power in WGAS has not been systematically assessed on real traits. One robust setting for doing so is to evaluate statistical power on thousands of traits measured from a single set of individuals. In this study, 359 newly sampled Americans of European descent were genotyped using both Affymetrix 500K (Affx500K) and Illumina 650Y (Ilmn650K) SNP arrays. From these data, we were able to obtain estimates of genetic coverage, which are robust to overfitting, by constructing an independent test set from among these genotypes and individuals. Furthermore, we collected liver tissue RNA from the participants and profiled these samples on a comprehensive gene expression microarray. The RNA levels were used as a large-scale set of quantitative traits to calibrate the relative statistical power of the commercial arrays. Our genetic coverage estimates are lower than previous reports, providing evidence that previous estimates may be inflated due to overfitting. The Ilmn650K platform showed reasonable power (50% or greater) to detect SNPs associated with quantitative traits when the signal-to-noise ratio (SNR) is greater than or equal to 0.5 and the causal SNP's minor allele frequency (MAF) is greater than or equal to 20% (N = 359). In testing each of the more than 40,000 gene expression traits for association to each of the SNPs on the Ilmn650K and Affx500K arrays, we found that the Ilmn650K yielded 15% times more discoveries than the Affx500K at the same false discovery rate (FDR) level.
| Advances in SNP genotyping array technologies have made whole-genome association studies (WGAS) a readily available approach. Genetic coverage and the statistical power are two key properties to evaluate on the arrays. In this study, 359 newly sampled individuals were genotyped using Affymetrix 500K and Illumina 650Y SNP arrays. From these data, we obtained new estimates of genetic coverage by constructing a test set from among these genotypes and individuals that is independent from the SNPs and individuals used to construct the arrays. These estimates are notably smaller than previous ones, which we argue is due to an overfitting bias in previous studies. We also collected liver tissue RNA from the participants and profiled these samples on a comprehensive gene expression microarray. The RNA levels were used as a large-scale set of quantitative traits to calibrate the relative statistical power of the commercial arrays. Through this dataset and simulations, we find that the SNP arrays provide adequate power to detect quantitative trait loci when the causal SNP's minor allele frequency is greater than 20%, but low power is less than 10%. Importantly, we provide evidence that sample size has a greater impact on the power of WGAS than SNP density or genetic coverage.
| It has been estimated that the human genome contains more than 5 million common single nucleotide polymorphisms (SNPs) with minor allele frequencies (MAF) ≥10% [1]–[3], and 7.5 million common SNPs with MAF ≥5% [4]. These SNPs may account for the genetic risk of many common human disorders. Recently, high-density SNP arrays have been introduced to allow researchers to conduct whole-genome association studies (WGAS). These SNP array platforms are often benchmarked by their genetic coverage and statistical power [4],[5]. Here, genetic coverage of an array platform is defined as the fraction of common SNPs (MAF≥5%) exceeding a predefined correlation threshold with at least one SNP typed by the array. Statistical power in this setting measures the likelihood to detect a statistically significant association between a truly associated SNP marker and a trait.
There are two strategies to building whole-genome SNP arrays. One is to randomly select SNPs that are relatively evenly spaced across the genome, not taking into account the inter-SNP linkage disequilibrium (LD) patterns, such as the Affymetrix 100K and 500K SNP arrays (denoted as Affx100K and Affx500K in this article, respectively) [4],[5]. The other is to select “tag SNPs” based on measures of LD chosen to maximize genetic coverage, such as Illumina HumanHap-300 -550K and -650Y arrays (denoted as Ilmn300K, Ilmn550K and Ilmn650K, respectively) [4],[5]. These tag SNP microarrays were developed using the public HapMap dataset [2],[3],[6].
The identification of tag SNPs is essentially a feature selection problem. It has been well established in the machine learning field that using an independently sampled test dataset is necessary to guarantee an unbiased assessment of the selected features' operating characteristics. It has also been shown that if the evaluation takes place on the training dataset itself, then the quality of the features' performance is often anti-conservatively over-estimated, commonly referred to as the overfitting problem [7],[8]. This problem exists in the context of identifying tag SNPs in two ways: (i) SNP overfitting, where the same set of SNPs are employed in both the training and evaluation steps; and (ii) sample overfitting, where the same set of subjects are used in both the training and evaluation steps.
The key operating characteristics of several whole-genome SNP arrays have been evaluated recently on HapMap data [4],[5],[9]. These studies may have been susceptible to both types of overfitting because the HapMap subjects were used to select tag SNPs; these same subjects and tag SNPs were then used in estimating the genetic coverage. Additionally, the small sample size of the HapMap data may limit the accuracy of estimates of statistical power, an operating characteristic that is critical for WGAS. Here, we present a study with the following characteristics to overcome these potential limitations: (i) the study subjects have been newly and independently sampled, and thus represent an independent sample from HapMap individuals, (ii) we have available a new set of genotyped SNPs which were sampled independently from HapMap data, and (iii) the sample size is relatively larger.
We utilized two commercially available high-density SNP arrays on an American Caucasian cohort to obtain estimates of genetic coverage for these different SNP panels that are robust to overfitting. The estimates we obtain in this cohort are lower than previous reports. In addition, liver RNAs were extracted and profiled on a comprehensive gene expression microarray. By simultaneously utilizing these thousands of gene expression traits scored on a fixed set of genetic backgrounds, we obtain estimates of the relative power of the different SNP genotyping arrays to detect quantitative trait SNPs of varying effect sizes [10]. We also directly quantify the impact of genetic coverage, SNP tagging, and sample size on the power of WGAS.
Human liver tissue samples were collected as described in a companion article [11]. In total, 359 American Caucasian subjects with known gender (heretofore called the “Liver Study subjects”) were successfully SNP genotyped and mRNA profiled.
The HapMap data are comprised of 270 individuals from four ethnic groups: (i) 30 trios from the Yoruba group in Ibadan, Nigeria (YRI); (ii) 30 trios from the CEPH collection, which are Utah residents with Northern and Western Europe ancestry (CEU); (iii) 45 unrelated Han Chinese individuals from Beijing, China (CHB); and (iv) 45 unrelated individuals from Tokyo, Japan (JPT). The CHB and JPT samples are often considered as a single East Asian sample [3]. The HapMap project has genotyped more than 4 million SNPs, among which ∼2.2 million SNPs are common in CEU (MAF ≥5%) [4],[9]. Additionally, Affymetrix Inc. has genotyped these 270 individuals using Affx500K, and made these results publicly available.
Recall that the Affx500K platform harbors 90K common SNPs that were not utilized in the HapMap project (referred to here as Affx NonHapMap SNPs). The genotypes from the Affx500K platform measured on the 359 Liver Study subjects therefore provide two key sources of independent data: (i) genotypes of SNPs identified independently from the HapMap project (the Affx NonHapMap SNPs) and (ii) individuals sampled independently from the HapMap project. This allowed us to study SNP selection overfitting and sample overfitting, respectively, in calculating genetic coverage.
Given that SNPs on the Affx500K were randomly chosen, the 90K Affx NonHapMap SNPs can be considered as a random sample of the entire 5.3M NonHapMap SNPs. We therefore utilized these Affx NonHapMap SNPs as a set of SNPs identified independently from the HapMap data. Also note that the complementary Affx HapMap SNPs (SNPs on Affx500K that are included in HapMap data) represent a random subsample of the 2.2M HapMap SNPs. We utilized these two classes of SNPs genotyped in our independently sampled human cohort scored for thousands of gene expression traits to estimate the true genetic coverage of the different SNP panels as well as assess the power these panels afford to detect associations between SNPs and traits (details below).
Genetic coverage was calculated as the fraction of SNPs exceeding a pre-defined r2 cutoff (r2cutoff) with at least one SNP typed by the microarray [5]. Herein, we employed two widely used values of r2cutoff, 0.8 and 0.9. Results were qualitatively equivalent in the 0.7 to 0.9 range for r2cutoff.
Overfitting of genetic coverage estimates was assessed separately for SNP overfitting and sample overfitting.
The statistical power of the SNP array platforms for WGAS were first estimated from simulation studies. First, we randomly selected a SNP from Affx500K and used its genotype to simulate trait value. Over the range of simulations, SNP genotypes from both Affx HapMap SNPs and Affx NonHapMap SNPs were utilized. We assumed the trait followed a Normal distribution N(μ,σ2), where σ2 was constantly set to 1 and μ varied among genotypes. We set (μAA, μAa, μaa) = (−0.5, 0, 0.5), (−0.25, 0, 0.25), or (−1, 0, 1) to investigate a range of signal strengths. Second, we conducted single-marker tests, which examined association between each SNP and each simulated trait. We surveyed three choices of α level (10−5, 10−6 and 10−7) that are reasonable for WGAS. Kruskal-Wallis and Spearman rank correlation tests were employed because these non-parametric methods were robust to the underlying genetic model and trait distribution, thereby allowing our simulation to be useful for non-normal traits and non-additive models. We defined a “true discovery” to be any association detected within 200 kb of the causal SNP. We calculated statistical power (defined as the probability of calling any SNP within 200kb of the causal SNP significant) and the average number of true discoveries (NTD) over the set of simulated datasets. Two million simulation runs were conducted for each parameter setting.
Again, we firstly simulated a binary phenotype using Affx HapMap SNPs and Affx NonHapMap SNPs, respectively. The genetic model was specified as disease prevalence = 25% and relative risk = 3. Once the phenotypes were generated, we randomly picked 75 cases and 75 controls from the 359 subjects to construct a balanced case-control study. These simulation parameters were chosen to ensure statistical power in a range easy to compare. Since the sample size was relatively small, Fisher's exact test was applied. Two million simulation loops were run for each scenario.
Using the same procedure as above, single-marker association tests were conducted to detect the expression quantitative trait loci (eQTL) for each of the ∼40,000 gene expression traits measured. Furthermore, we repeated the tests on three permuted gene expression datasets and calculated the false discovery rate (FDR). In each permutation run, we first randomized the patient IDs in the expression file, breaking any association between expression traits and genotypes while leaving the respective correlation structures among gene expression traits and SNP genotypes intact. Second, we repeated the association tests for every expression trait and genotype pair, resulting a set of null statistics for each permutation. A standard FDR estimator was then applied to the resulting association statistics, as previously carried out on observed and permutation null statistics [12]. Because the entire set of null statistics were used to calculate the q-value for each test, we were able to use only three permutations and still retain stable significance results.
Based on the Liver Study subjects, we obtained new estimates of genetic coverage for the Illumina and Affymetrix SNP genotyping arrays (Materials and Methods), which are robust to overfitting. Evidence of SNP overfitting in previous estimates can be seen in Figure 1 by comparing the genetic coverage of Affx HapMap SNPs to that of Affx NonHapMap SNPs based on the genotypes from the Ilmn650K array. (Materials and Methods). As can be seen, SNP overfitting is present in genetic coverage estimates derived from both HapMap CEU individuals and Liver Study individuals. Interestingly, the magnitude of SNP overfitting was similar when repeating the analysis on the Ilmn550K and Ilmn300K arrays (Table S1).
By comparing estimates of genetic coverage derived from HapMap CEU to those from Liver Study subjects (Materials and Methods), we also found evidence for the existence of sample overfitting (Figure 1 and Table S1). For example, the Ilmn300K platform's genetic coverage was reported to be 9% higher in CEU individuals than when we make the sample calculation on the Liver Study subjects. In contrast, the magnitude of sample overfitting was smaller in the Ilmn550K and Ilmn650K sets. This phenomenon could be explained by the degree of redundancy in the tag SNP sets. The first genome-wide tag SNP array, Ilmn300K, harbors a “lean” set of 317K tag SNPs optimized only in CEU. As a drawback, these 317K SNPs contained less redundancy and exhibited less transferability. The Ilmn550K and Ilmn650K were developed on multiple ethnic groups [9], and their tag SNP sets had higher degree of redundancy, resulting in better transferability.
By comparing genetic coverage of the Affx HapMap SNPs in CEU to Affx NonHapMap SNPs in Liver Study subjects, we measured the combined size of the two types of overfitting to be ∼18%. Furthermore, we formed weighted estimates by taking the weighted average of the coverage on HapMap SNPs (w = 2.2/7.5) and that on NonHapMap SNPs (w = 5.3/7.5). Among the Liver Study Caucasian subjects, the Ilmn300K and Ilmn650K's weighted genetic coverage was 64% and 76% at r2cutoff = 0.8, which is lower than previous reports (79% and 90%, respectively; http://www.cidr.jhmi.edu/human_gwa.html). Furthermore, we found that the tagSNP arrays cover low MAF SNPs (e.g. MAF<15%) worse than the high MAF ones (e.g., MAF≥15%), and importantly, the overfitting bias appears to be more severe for the low MAF range (Figure S1 and Table S1).
A WGAS requires genotyping thousands of subjects, which is expensive at current genotyping costs [13],[14]. To conserve resources, many WGAS are adopting a two-stage design in which a small sample of subjects (e.g., a few hundred) are genotyped on all markers in stage 1, and a proportion of these markers are genotyped on a much larger sample in stage 2 [13],[14]. Studies have shown that this strategy may preserve much of the power of the corresponding one-stage design and minimizes the genotyping cost [14]. In our study, N = 359 is a reasonable sample size for the stage 1 WGAS [14]–[16]. The SNP arrays showed reasonable power to detect SNPs associated with quantitative traits when the SNR >0.5, but limited power when SNR ≤0.25 (Table S2).
In simulating SNPs causal for a quantitative trait, we assumed μAA<μAa<μaa, so that the Spearman rank correlation test gave higher statistical power and larger NTD than the Kruskal-Wallis rank-sum test. In a separate set of simulations (Figure S2), we found the Kruskal-Wallis test was conservative in the small p-value range (e.g. p-value<0.05). Focusing on SNR = 0.5, we found that the statistical power was highly related to the causal SNP's MAF (Figure 2). For example, when MAF≥20% the Ilmn650K had over 50% power to detect associations with quantitative traits (Spearman rank correlation test). When the MAF ≤10%, little power could be achieved (Figure 2).
Not surprisingly, we observed the Affx500K to exhibit less power than the Ilmn650K, which could be explained the fact that the Affx500K platform contains fewer SNPs and/or that the Ilmn650K platform has higher genetic coverage. Because the Ilmn550K and Ilmn650K platforms had similar genetic coverage in Caucasians, they showed nearly the same statistical power (Figures 2 and 3). In contrast, the Ilmn650K platform offered a larger number of true discoveries (NTD), indicating more significant SNPs were detected around the true causal SNP.
Interestingly, the power of Illumina arrays differed when identifying associations with quantitative traits simulated using Affx HapMap SNPs and Affx NonHapMap SNPs which was essentially the overfitting effect. For example, using Kruskal-Wallis test and a p-value threshold of α = 10−6, Ilmn550K showed 37% power in detecting NonHapMap causal SNPs and 43% power in detecting HapMap causal SNPs (Table S2B), which translated into a difference of 6% in power, likely due to an overfitting bias. We also surveyed other genetic models as well as significance thresholds, and observed considerable SNP overfitting effects (Table S2A, S2C, and S2D).
In a WGAS, a large number of hypothesis tests are conducted, so that statistical significance measures such as the FDR need to be carefully assessed. In our simulation, the true causal SNPs were known. When significant associations were detected >1Mb away from the causal SNP or on a different chromosome, we regarded them as false discoveries. The number of false discoveries (NFD) was proportional to the number of SNPs employed in the WGAS, with NFDAffx500K<NFDIlmn300K<NFDIlmn550K<NFDIlmn650K. Comparing Table S2B and Table S3, we found the FDR was in a manageable range. For example, at a P<10−5, Ilmn550K gave an average of 2.51 NFD and 1.69 NTD using the Kruskal-Wallis test. At a P<10−6, 0.20 NFD and 1.06 NTD were observed; and at a more stringent P<10−7, the FDR was even smaller, suggesting 10−6 or 10−7 as an appropriate significant cutoff in WGAS or initial screening (Figure S3).
Finally, we conducted association tests on actual traits, namely RNA expression levels or “expression traits.” It has been shown that comparing the ND from many related traits, all conditional on the same set of individuals (i.e., genetic backgrounds), can lead to meaningful empirical power comparisons, where simple models often used for simulations do not have to be assumed [10]. Specifically, by considering the relative ND among different approaches at the same error rate, we are able to estimate the relative levels of power, without having to specifically identify which among the discoveries are true discoveries [17].
The number of discoveries (ND) obtained on both observed data and permuted data followed the same pattern: Ilmn650K>Ilmn550K>Ilmn300K>Affx500K. Because the true and false discoveries were no longer distinguishable, we could not directly infer the SNP arrays' statistical power using ND. Instead, we compared the relative power using ND conditioning on FDR (Figures 4 and S3). The Ilmn650K slightly outperformed the Ilmn550K, indicating the “100K YRI SNPs” on Ilmn650K [9] benefited Caucasian studies although they were selected on HapMap YRI. Compared to Affx500K, Ilmn650K discovered 15% more genes that were associated with at least one SNP (FDR = 10%).
After filtering SNPs based on MAF, call rate, and HWE p-values (Materials and Methods), a similar number of SNPs on Affx500K and Ilmn300K (286K and 296K SNPs, respectively) were included in the analysis, which provided an opportunity for a head-to-head comparison between random SNPs and tag SNPs on these expression traits. Unexpectedly, the Affx500K outperformed Ilmn300K in term of ND (Figure S4, upper panels), indicating random SNPs detected more significant associations than the same number of tag SNPs at the same FDR levels. However, the Ilmn300K captured more quantitative traits (Figures 4 and S4, lower panels). One explanation could be that the Affx500K SNPs, clustered on Nsp and Sty restriction fragments rather than strategically spread on human genome, tended to capture certain signals repeatedly but missed other associations.
As a novel feature of our study, we were also able to investigate the power of combining Affx500K and Ilmn650K arrays for a single analysis (Figure 4). Since the location of each expressions trait is known, this allows us to focus on the cis-associations. In brief, for a given expression trait, only SNPs within ±1 Mb of the corresponding gene are tested. By taking these steps, the number of tests is substantially reduced and the statistical power increased, illustrated in more cis-association discoveries in Figure 4 right panel comparing to the left panel. The numbers of cis-association genes also reflect the relative power. Consistent with Figure S4, the Illumina tag SNP arrays are more powerful than Affx500K. For example, using Affx500K as the reference, 650K panels' relative power is 110%, in detecting cis-association. Surprisingly, Affx500K+Ilmn650K (relative power = 115%) only slightly outperforms Ilmn650K, indicating the limited return of adding additional SNPs on top of Ilmn650K.
We collected 68 additional liver samples from Caucasian donors. We performed RNA expression profiling as before and obtained SNP genotypes using the Affy500K array only. We then we pooled the sample (re-normalizing for gender, age, and medical center, and batch) and reran the association tests on the Affx500K SNPs. Surprisingly, this increase in sample size (19%) results in 31% and 33% more cis-association discoveries (at 5% and 10% FDR, respectively), implying a respective 31% and 33% boost in relative power. In contrast, conditioning on the same sample size, Ilmn650K yields about 10% more cis-associations than Affy500K. This is potentially an important observation that sample size has a more profound impact on statistical power than the difference in genetic coverage among current SNP arrays. Since arrays vary greatly in price, and argument has been raised whether to choose high genetic coverage arrays or cheaper ones and run more subjects.
Whole-genome association studies using high-density SNP arrays are viewed as a powerful approach to elucidating the genetic bases of common human diseases. We provided a novel investigation of two key properties for determining the performance of SNP array genotyping platforms in WGAS: genetic coverage and statistical power. The availability of (i) 90K genotyped SNPs which were identified independently from the HapMap, (ii) a new, independently sampled cohort of subjects, and (iii) thousands of related “expression traits” measured simultaneously on these subjects, yielded the opportunity to provide new insights into genetic coverage and statistical power, and make comparisons to previous results.
Two strategies are usually taken in selecting SNPs and constructing genome wide arrays: random SNPs and tag SNPs. These strategies might result in different levels of performance in terms of genetic coverage and statistical power [4]. Regardless of the selection algorithms used, the performance of tag SNPs is most accurately assessed by using a validation dataset independent of the training set [18]. In this article, we systematically investigated two sources of overfitting (SNP overfitting and sample overfitting, respectively) and derived new genetic coverage estimates robust to these two types of overfitting. As a caveat, Affymetrix developed Affx500K array by screening the dbSNP database, which tends to document frequent SNPs rather than rare SNPs. As the result, the Affx SNPs have higher MAF than the totality of SNPs in the human genome, and our weighted genetic coverage estimates may be somewhat upwardly biased. However, this bias is an issue distinct from bias due to overfitting.
Since there are a limited number of common SNPs in the human genome, tag SNPs selected from the complete set (e.g., the 7.5 million common SNPs) would be robust to SNP overfitting in assessing genetic coverage. At the current stage, tag SNPs are usually selected from an incomplete initial SNP set (i.e., HapMap SNPs), and the remaining SNPs (i.e., 5.3M NonHapMap SNPs) would be “hidden” from the training procedure. Previous simulation studies showed that 26% of the common ENCODE SNPs in CEU had no good proxies (r2≥0.8) among the “pseudo” HapMap I SNPs [19]. This implies that these 26% SNPs would have an extremely low likelihood of being captured by tag SNPs (e.g. Ilmn300K) selected using HapMap I. Using empirical datasets, researches studied the SNP overfitting problem in a diverse set of ethnic groups around the world [3],[5],[20],[21]. However, these studies faced the limitation of small chromosomal regions and they didn't consider overfitting in the context of statistical power.
Our study employed 359 individuals, which provided adequate levels statistical power for moderate genetic effects (e.g., SNR = 0.5). Certainly, larger sample size is necessary to detect weaker effects (e.g., SNR = 0.25). Illumina tag SNP arrays were designed to capture common HapMap SNPs. Therefore, most of these tag SNPs have MAF ≥5%. In contrast, Affx500K harbors over 100K rare SNPs (MAF ≤5%). Because we used common SNPs to simulate quantitative traits, the Affx500K rare SNPs provided little statistical power and were therefore excluded from the analysis. Rare SNPs might be useful when the disease-causing polymorphism was also rare. However, Figure 2 showed WGAS may have limited power in such scenarios.
Our SNP filtering resulted in similar number of SNPs on Affx500K and Ilmn300K, which enabled a fair comparison between random SNPs and tag SNPs. We found random SNPs actually achieved larger ND (Figure S4), but many of which were redundant. In another words, Affx500K tended to capture the same signal repeatedly. In contrast, more traits were in association with at least one SNP on Ilmn300K, indicating tag SNPs were more efficient in WGAS. Such a finding was consistent to our simulation study, shown Figures 2 and 3, where Ilmn300K outperformed Affx500K in terms of statistical power.
Affymetrix and Illumina recently released 900K and 1M SNP arrays, respectively. These new products will further enhance WGAS. In evaluating their performance, we recommend utilizing independent test sets as we have done here. Given that we did not utilize these new arrays, we were not able to calibrate the genetic coverage and statistical power for the million-SNP arrays. However, there are a number of reasons why a major performance leap may not be expected. First, it has already been reported that the gain in coverage achieved by increasing the number of tag SNPs follows a pattern of diminishing returns [4],[19]. Second, the current tag SNP selection is still limited to occur within the HapMap dataset. As shown in Figure 1, this strategy results in an approximately 12% genetic coverage loss when applying to NonHapMap SNPs. In the ENCODE regions, ∼10% of the common SNPs had no good proxies (r2≥0.8) among the simulated HapMap II datasets, and those SNPs could only be adequately captured by searching beyond the HapMap data. In another words, the HapMap-derived tag SNP struggles to reach the 90% genetic coverage level. Through simulations, we were able to directly test the causal SNP, allowing us to calibrate the upper bound of a SNP array's performance in WGAS. At SNR = 0.5 (Figure 3), “direct genotyping” provided a gain of 8% more power than Ilmn650K, indicating current arrays are already capable of extracting a substantial level of genetic information. “Direct genotyping” provided a greater increase in power at SNR = 1 (Figure S5), but no extra power at SNR = 0.25.
It is important to continue to systematically quantify the trade-offs among genetic coverage, genotyping cost, and statistical power for WGAS. Based on our results, some conclusions are possible (Figure 5). For example, according to our results a study employing N = 300 subjects and the Affy500K platform offers higher power than a study employs N = 250 subjects and the Ilmn650K platform. This 20% increase in sample size (N = 300 vs. N = 250) provides more power than the 90% increase in the number of SNPs genotyped (286K SNPs vs. 545K SNPs). In scenarios where funding becomes the constraining factor, our results suggest that genotyping larger sample size with cheaper SNP arrays might achieve better statistical power. On the other hand, if the constraining factor is the number of subjects, then it appears that SNP arrays offering the largest genetic coverage should be employed. |
10.1371/journal.pgen.1006498 | Genetic Variation in the Social Environment Contributes to Health and Disease | Assessing the impact of the social environment on health and disease is challenging. As social effects are in part determined by the genetic makeup of social partners, they can be studied from associations between genotypes of one individual and phenotype of another (social genetic effects, SGE, also called indirect genetic effects). For the first time we quantified the contribution of SGE to more than 100 organismal phenotypes and genome-wide gene expression measured in laboratory mice. We find that genetic variation in cage mates (i.e. SGE) contributes to variation in organismal and molecular measures related to anxiety, wound healing, immune function, and body weight. Social genetic effects explained up to 29% of phenotypic variance, and for several traits their contribution exceeded that of direct genetic effects (effects of an individual’s genotypes on its own phenotype). Importantly, we show that ignoring SGE can severely bias estimates of direct genetic effects (heritability). Thus SGE may be an important source of “missing heritability” in studies of complex traits in human populations. In summary, our study uncovers an important contribution of the social environment to phenotypic variation, sets the basis for using SGE to dissect social effects, and identifies an opportunity to improve studies of direct genetic effects.
| Daily interactions between individuals can influence their health both in positive and negative ways. Often the mechanisms mediating social effects are unknown, so current approaches to study social effects are limited to a few phenotypes for which the mediating mechanisms are known a priori or suspected. Here we propose to leverage the fact that most traits are genetically controlled to investigate the influence of the social environment. To do so, we study associations between genotypes of one individual and phenotype of another individual (social genetic effects, SGE, also called indirect genetic effects). Importantly, SGE can be studied even when the traits that mediate the influence of the social environment are not known. For the first time we quantified the contribution of SGE to more than 100 organismal phenotypes and genome-wide gene expression measured in laboratory mice. We find that genetic variation in cage mates (i.e. SGE) explains up to 29% of the variation in anxiety, wound healing, immune function, and body weight. Hence our study uncovers an unexpectedly large influence of the social environment. Additionally, we show that ignoring SGE can severely bias estimates of direct genetic effects (effects of an individual’s genotypes on its own phenotype), which has important implications for the study of the genetic basis of complex traits.
| Social interactions contribute to health and disease (e.g. peer smoking increases one’s risk of taking up smoking). So far, quantifying social effects has required a clear hypothesis about the mechanisms mediating the influence of the social environment (in the example above peer smoking is the trait that mediates the social influence). For many phenotypes however, such hypotheses do not exist. Therefore we propose an alternative strategy to study social effects: we investigate effects on an individual's phenotype that arise from genotypes of social partners (social genetic effects, SGE, also called indirect genetic effects[1, 2]). SGE constitute the genetic basis of social effects and can be detected without prior knowledge of the phenotypes through which the social influence is exerted. SGE have been reported for interactions between mothers and offspring (maternal genotypes indirectly affect offspring phenotypes)[3–8] and more recently for interactions between adult individuals, in livestock and wild animals[2, 9–17] For example, growth rate in farm pigs has been found to be in part determined by the genetic makeup of the other pigs in the pen[2]. However, the extent to which SGE explain variation in biomedical traits is largely unknown.
If SGE do contribute to such traits, they are a promising approach to quantify effects of the social environment. Additionally, they provide an anchor to investigate causal paths and dissect the mechanisms underlying social effects. Finally, in studies of direct genetic effects carried out by the broad community, SGE may be used to account for social environmental effects.
Our study aimed at quantifying the contribution of SGE to multiple biomedical traits. We uncover unexpectedly large social genetic effects on multiple organismal and molecular phenotypes.
To investigate whether SGE explain variation in biomedical traits, we considered two experiments in laboratory mice involving complementary genetic designs, and assessed both organismal and gene expression traits.
We first carried out an experiment with two inbred strains. We chose C57BL/6J (B6) and DBA/2J (D2), the progenitor strains of the largest mouse reference population, the BxD recombinant inbred panel[18]. We co-housed 86 mice at weaning as B6/B6, B6/D2 or D2/D2 pairs. After a period of six weeks during which the mice interacted undisturbed in their home cages, we collected 50 organismal phenotypes relevant to unconditioned anxiety, helplessness (a measure of depressed mood), general locomotor activity, stress, social dominance, wound healing and body weight (Fig 1A, S1 Table). We also profiled genome-wide gene expression in the prefrontal cortex (PFC). The PFC was selected because it is involved in coordinating behavioural responses based on sensorimotor information, motivation and affect, all of which may be affected by the social environment; as a result individual differences in PFC expression levels may reflect behavioural responses to the social environment [19–21].
In this design with two inbred strains there are three potential genetic sources of phenotypic variation: differences in the strain of the focal (i.e. phenotyped) mice (DGE), differences in the strain of their cage mates (SGE) and an interaction between the two, whereby the effect of the strain of the cage mate depends on the strain of the focal mouse (Fig 1B, S1 Fig and Methods). Variance partitioning and model selection (see Methods) provided evidence that interactions between DGE and SGE are common across traits (S2 Fig), with SGE typically affecting a specific trait in one strain (B6 or D2) but not the other (e.g. Fig 1B, S3 Fig). Thus, for each trait, we modeled the measurements collected in B6 mice and D2 focal mice separately (S1 Fig).
For eleven out of 50 organismal phenotypes we found significant SGE in either B6 or D2 mice (P < 0.05, FDR < 43%—see Methods; Table 1, S2 and S3 Tables). Strain-specific differences existed as different measures were affected by SGE in B6 and D2 mice. Of these eleven phenotypes, two were measures of stress and six were measures of anxiety, providing evidence that variation in the genetic makeup of cage mates causes variation in stress-related phenotypes. The direction of effect was consistent across the six measures of anxiety (S3 Fig). We also detected strong SGE on the rate of wound healing (measured from an ear punch, P = 6.6 10−3, Q = 0.36), showing that SGE are not limited to behaviors. (Table 1). Importantly, SGE explained a considerable proportion of phenotypic variance (up to 18%), showing that social effects of genetic origin are important contributors to phenotypic variation in these two strains.
We also found evidence that SGE affect gene expression in the PFC. A gene set enrichment analysis based on the contribution of SGE to gene expression levels (see Methods) revealed “integrin-mediated signaling pathway”(P = 2.5 10−6, Q = 0.024) and “regulation of dopamine metabolic process” (P = 7.0 10−4, Q = 1) as the most significant Gene Ontology (GO) terms in D2 and B6 mice respectively (Table 2). The latter enrichment, although statistically less robust, is strikingly consistent with our finding that SGE affect dopamine levels measured by HPLC in B6 mice (Table 1). Our results therefore converge to show that variation in the genetic makeup of cage mates causes variation in behavioral, biochemical, and gene expression traits relevant to stress. Although the large number of tests limits the statistical power to detect SGE on individual genes (12,898 genes tested), we identified three genes with significant effects (Srsf2, Phlpp2, and Ppid; FDR < 33%; S4 Table).
We next investigated SGE in a dataset from outbred (Heterogeneous Stock) mice[22–24]. This design better represents traditional mouse housing conditions (groups of four, five and six mice mostly rather than pairs; S4 Fig) and genetic variation in natural populations (high genetic diversity and genetically unique individuals rather than two inbred strains). The dataset comprises more than 100 organismal phenotypes measured in 2,448 mice (S5 Table), and gene expression in hippocampus for a subset of 457 mice. To accommodate the genetic design of this study, we fitted random effects models with variance components for DGE, SGE and their covariance. Our models are inspired from models published in the literature[2, 25] yet differ in some important ways, which we explain and justify in S1 Note. Because genome-wide genotypes were not available for a subset of mice (526 out of 2,448), we used pedigree information to estimate pairwise genetic covariance for the mice with no genotype information (see Methods). We found that the results obtained using both pedigree and genotype data for estimation of genetic similarity were in agreement with those obtained using genotype data only from the subset of mice that were in cages where all mice had been genotyped (S5 Fig). All models were fitted using LIMIX[26, 27].
Simulations showed that DGE and SGE were unbiased (S6 Fig and S7 Fig). Of the 117 organismal phenotypes available in this dataset, 43 were significantly affected by SGE (P < 0.05, FDR < 5.7%; Table 3, S5 Table). SGE explained up to 29% of phenotypic variation and an average of 8.9% across the 43 significantly affected traits. Importantly, the estimated contribution of SGE was greater than that of DGE for 8 of the 43 traits.
Among the organismal phenotypes most significantly and strongly affected by SGE in this dataset were measures of lymphocyte activation (in particular size and number of CD4+ T cells and B cells, collected by fluorescence-activated cell sorting and full blood count, Table 3). In contrast, measures related to other leucocytes (neutrophils, basophils and monocytes) and natural killer T cells. Altogether, these results indicate that genotypes of cage mates influence humoral immunity, although this is unlikely that this represents spread of a disease as the mice were kept in a clean mouse facility.
In addition, rate of wound healing, the measure most significantly and strongly associated with SGE in the experiment with two inbred strains, was also significantly affected by SGE in the outbred dataset (P = 1.6 10−3, Q = 3.7 10−3).
Three measures of body weight (collected on weeks 6 and 7) also figured among the traits significantly affected by SGE. Other measures of body weight, collected at weeks 9 and 10, showed no SGE (S5 Table). The two sets of measures likely reflect a different phenotype as the normal physiology of the mice was disrupted between weeks 7 and 9 by aggressive phenotyping (tests of conditioned anxiety involving foot shocks, airway sensitization by allergen, intraperitoneal injection of glucose). Another potential but unlikely explanation for the higher contribution of SGE on earlier body weight measures is a sharp decrease in social effects between weeks 7 and 9.
Finally, a subset of measures of anxiety, blood biochemistry, and lung function were also affected by the genotypes of cage mates.
In contrast, there was no statistical evidence for SGE on gene expression levels in the hippocampus (smallest nominal P value 4.1 10−5 corresponding to a Q value of 64%, S6 Table). This is most likely the result of a much smaller sample size (457 or 5% of mice) for expression traits, and is consistent with published power analyses[28].
We next explored whether studies focused on DGE can safely ignore SGE, focusing on the estimation of the collective effect of additive DGE on phenotypic variation (narrow-sense heritability). In the outbred dataset, cage mates are genetically more similar to each other than average (S8 Fig). As a result, DGE and SGE are correlated (an example of gene-environment correlation). Thus, we hypothesized that failing to account for SGE would bias estimates of DGE. We also hypothesized that fitting cage effects might be sufficient to eliminate this bias. To investigate both hypotheses, we compared DGE estimates obtained using a linear mixed model for DGE that does not account for cage effects nor SGE, a model that accounts for DGE and cage effects, and one that accounts for DGE, SGE, corresponding social environmental effects, and cage effects (“full model”). Models that did not account for SGE led to substantially larger DGE estimates, and estimates from the model with DGE and cage effects were intermediate between those from the model with DGE only and those from the full model. In simulated traits based on the real genotypes and generated from DGE, SGE and cage effects (see Methods), we found that models that did not account for SGE yielded inflated DGE estimates, whereas joint modeling of DGE, SGE and cage effects resulted in unbiased estimates (Fig 2C and S6 Fig). Importantly, fitting cage effects but no SGE did not eliminate the bias. The simulation results strongly suggest that, in the real data, the estimates obtained from the full model are most accurate, and that models that ignore SGE overestimate heritability. This problem is particularly acute when direct and social random genetic effects are positively correlated (i.e. σADS>0, see Methods; Fig 2A and 2B). The problem we highlight here is general and likely to affect other studies in which social partners are related, including twin and family studies used to estimate heritability in humans (see Discussion).
Using two complementary genetic designs–one using two mouse inbred strains and one using outbred mice—we estimated the contribution of social genetic effects to a variety of organismal phenotypes and gene expression traits. The experiment with two inbred strains was designed to investigate SGE and focused on behaviours (anxiety and helplessness), as there is strong evidence that behaviours are socially affected[29–33]. To test whether SGE can be detected in outbred populations and survey a broader range of phenotypes, we re-analysed a large dataset from outbred mice and quantified the contribution of SGE to more than 100 phenotypes. The design of our study raises important questions: are positive results (i.e. evidence of SGE) in the experiment with two inbred strains expected to replicate in the outbred dataset? Are some phenotypes expected to be affected by SGE and some not? We now discuss these points.
Some phenotypes were measured in both experiments with similar protocols. Wound area, the measure of wound healing, was collected in both experiments using the exact same protocol. It is significantly affected by SGE in both experiments. The protocols for measuring body and adrenal gland weight are fairly simple thus reducing technical variation between experiments, and body weight was measured at about the same time point (around 50 days of age) in both experiments. There was strong evidence for SGE on body weight in the outbred dataset but no evidence in the experiment with two inbred strains. No SGE on adrenal gland weight were detected in either dataset. Finally, a partially overlapping set of measures of unconditioned anxiety was significantly affected by SGE in both experiments. While reviewing results from the two experiments in parallel is informative, positive results in one experiment are not strictly expected to replicate in the other. Indeed, although the variants that give rise to SGE in the experiment with two inbred strains also segregate in the outbred population (the two strains used in our experiment were among the eight founders of the outbred population), they have recombined with many additional variants from the six other founders. Moreover, the housing conditions were very different in the experiment with inbred strains and outbred experiment (group size of 2 vs. 2 to 7 respectively, and unfamiliar mice vs. familiar mice housed together). Therefore, one should not expect the overall contribution of SGE be the same in the two experiments. Rather, combining the two experiments provides a first hint at the generalizability of our results. Published studies of SGE provide additional information on this matter, and suggest that SGE may contribute to variation in body weight across species[2].
That social genetic effects contribute to variation in anxiety probably does not come as a surprise but their contribution to wound healing maybe more so. This result is however supported by significant p-values in both experiments of our study and large effect sizes (18 and 6%). When interpreting this result, it is important to bear in mind that social effects on wound healing can (and will, necessarily) be mediated by traits of cage mates that are different from wound healing. For example, social effects on wound healing could be mediated by social grooming, which could either mechanically disrupt the healing process or chemically enhance it[34]. Any traits of cage mates that may induce a systemic stress response in the focal animal could also mediate social effects on wound healing[35, 36]. Thus, social effects on wound healing are not unlikely, and, similarly, social effects may affect any phenotype (e.g. by through the induction of a systemic stress response,).
Because any phenotype may a priori be affected by social effects and the mechanisms at play are rarely known, SGE offer an attractive alternative to investigate social effects. First, as we have shown, they can be used to quantify social effects, effectively providing a lower bound estimate of social effects (as only the genetic component is captured). Second, SGE can be used to test whether a particular trait of social partners has an effect on a phenotype of interest. Establishing a causal relationship between two phenotypes is always difficult because of the risk of reverse causation and independent action of hidden confounders on both traits; SGE provide an anchor to test causality.
Independent of their relevance for studying social effects, we show that ignoring SGE can lead to biased estimates of heritability (i.e. the collective effect of DGE). In our study (outbred dataset), DGE and SGE are correlated by design (mice that share a cage are more genetically similar than average), and we show that this correlation leads to biased estimates of heritability if unaccounted for. Fitting cage effects, which has the primary goal of accounting for environmental effects shared by cage mates (e.g. noise levels), does not eliminate the bias. Our results are of interest to the broad genetics community as DGE and SGE are correlated in most if not all experimental designs traditionally used to estimate heritability in humans and model organisms, and SGE may thus have caused widespread bias. For example, in twin designs, MZ twins not only share 100% of their genotypes but they also share 100% of the genotypes of their sibling; DZ twins in comparison share both 50% of their genotypes and 50% of the genotypes of their sibling. Thus, SGE can contribute to increased concordance between MZ twins compared to DZ twins. If SGE are not modelled, heritability may be overestimated and appear “missing” when compared to genome-wide association results obtained from unrelated individuals[37]. Note that when the covariance σADS between direct and social random genetic effects is negative (competition effects), ignoring SGE may lead to underestimating heritability. SGE in humans were considered once before (“sibling effects” [38]) but were never, to the best of our knowledge, modelled in heritability studies. Because we found that fitting cage effects was not sufficient to eliminate the bias due to SGE, we suspect that accounting for a “common environment” shared by family members, as is commonly done in human studies[39–41], will not eliminate SGE-induced bias.
It is not the first time that unaccounted for gene-environment correlations are put forward as potential causes of bias (e.g. Conley et al. investigated the correlation between genetics and urban setting [42]). However, the impact of the correlation between DGE and SGE is likely to be particularly severe as we have shown that SGE affect a wide range of phenotypes and DGE and SGE are correlated in most experimental designs used to estimate heritability.
Our study sheds light on an important component of the genetic architecture of complex traits, one that lies outside the individual, in social partners. Social genetic effects have already been shown to play an important role in artificial selection of livestock[43] and have important evolutionary consequences[44, 45]. Our results provide evidence that SGE are also an important component of health and disease.
Gene expression data from the experiment with inbred strains are available from ArrayExpress E-MTAB-5276. Phenotype data for the same experiment are provided as S7 Table.
|
10.1371/journal.pbio.3000038 | Structure of the Macrobrachium rosenbergii nodavirus: A new genus within the Nodaviridae? | Macrobrachium rosenbergii nodavirus (MrNV) is a pathogen of freshwater prawns that poses a threat to food security and causes significant economic losses in the aquaculture industries of many developing nations. A detailed understanding of the MrNV virion structure will inform the development of strategies to control outbreaks. The MrNV capsid has also been engineered to display heterologous antigens, and thus knowledge of its atomic resolution structure will benefit efforts to develop tools based on this platform. Here, we present an atomic-resolution model of the MrNV capsid protein (CP), calculated by cryogenic electron microscopy (cryoEM) of MrNV virus-like particles (VLPs) produced in insect cells, and three-dimensional (3D) image reconstruction at 3.3 Å resolution. CryoEM of MrNV virions purified from infected freshwater prawn post-larvae yielded a 6.6 Å resolution structure, confirming the biological relevance of the VLP structure. Our data revealed that unlike other known nodavirus structures, which have been shown to assemble capsids having trimeric spikes, MrNV assembles a T = 3 capsid with dimeric spikes. We also found a number of surprising similarities between the MrNV capsid structure and that of the Tombusviridae: 1) an extensive network of N-terminal arms (NTAs) lines the capsid interior, forming long-range interactions to lace together asymmetric units; 2) the capsid shell is stabilised by 3 pairs of Ca2+ ions in each asymmetric unit; 3) the protruding spike domain exhibits a very similar fold to that seen in the spikes of the tombusviruses. These structural similarities raise questions concerning the taxonomic classification of MrNV.
| The freshwater prawn Macrobrachium rosenbergii is widely cultivated for food. Production is threatened by Macrobrachium rosenbergii nodavirus (MrNV), the causative agent of white-tail disease. Outbreaks in hatcheries often result in mortality rates of up to 100% in larvae and post-larvae, leading to devastating economic losses and threatening food security. We describe the atomic structure of the MrNV capsid, solved by cryogenic electron microscopy and three-dimensional image reconstruction. Our analysis revealed surprising differences between the structure of MrNV and that of other known nodaviruses. Moreover, we observed several features in the MrNV capsid that have been previously described in virion structures of the plant-infecting tombusvirus family. Most notably, the MrNV capsid exhibits pronounced dimeric spikes on its surface, the topology of this region closely resembling tombusvirus capsid spikes. Known nodavirus structures have trimeric spikes and do not display the same protein fold. The MrNV capsid is stabilised by divalent cations and laced together by a network of N-terminal arms that line the interior of the virion. Our analysis raises questions about the taxonomic classification of MrNV as well as revealing the structure of the capsid of this important pathogen. These data have the potential to inform the development of future interventions to prevent white-tail disease.
| The giant freshwater prawn M. rosenbergii is widely cultivated in tropical and subtropical areas for food. The global production of this species has increased dramatically from about 3,000 tons in 1980 to more than 220,000 tons in 2014 [1]. However, productivity is threatened by white-tail disease (WTD), which is caused by M. rosenbergii nodavirus (MrNV). This often leads to 100% mortality rates in larvae and post-larvae of M. rosenbergii [2]. The first MrNV outbreak was reported in Pointe Noire, Guadeloupe in 1997, followed by China [3], India [4], Taiwan [5], Thailand [6], Malaysia [7], Australia [8], and recently Indonesia [9]. To date, neither a vaccine nor effective treatment is available to prevent or manage MrNV outbreaks.
MrNV has been classified within the Nodaviridae family of viruses. These nonenveloped viruses have bipartite, positive-sense, single-stranded RNA genomes that are packaged within T = 3 icosahedral capsids. Presently, this family has 2 established genera: Alphanodavirus and Betanodavirus. The former consists of insect-infecting nodaviruses such as Flock House virus (FHV), Pariacoto virus (PaV), black beetle virus (BBV), Nodamura virus (NoV), and Boolarra virus (BoV), while the latter contains fish-infecting nodaviruses such as Malabaricus grouper nervous necrosis virus (MGNNV), grouper nervous necrosis virus (GNNV), and striped jack nervous necrosis virus (SJNNV). Although MrNV is classified within the Nodaviridae, amino acid sequence comparison revealed that its capsid protein (CP) shares low similarity (less than 20%) with other nodaviruses in the 2 established genera. Thus, it is ambiguous whether MrNV should be grouped in either one of these genera. Conversely, the amino acid sequence of MrNV CP shares approximately 80% similarity with that of Penaeus vannamei nodavirus (PvNV). These 2 crustacean nodaviruses have therefore been proposed to be grouped into a new genus: Gammanodavirus [10].
Nodaviruses are very simple, with their 2 short-genomic RNAs encoding 3 gene products. RNA 1 (3.2 kb) encodes the RNA-dependent RNA polymerase (RdRp) and the nonstructural B2-like protein, while RNA 2 (1.2 kb) encodes the viral CP. The full-length MrNV CP is a polypeptide of 371 amino acids. The N-terminal arginine-rich region interacts with the RNA genome [11], while the C-terminal domain plays crucial roles in host cell attachment and internalisation [12]. A nuclear localisation signal (NLS) has also been identified at the N-terminus (N-ter) (amino acids 20–29) of CP. This has been shown to target the viral capsid to the nucleus of insect cells [13]. Further functional regions of CP have yet to be defined.
Several three-dimensional (3D) structures of alpha- and betanodaviruses have been determined using both X-ray crystallography and cryogenic electron microscopy (cryoEM) [14–18]. These analyses have revealed several common features. Nodaviruses assemble T = 3 icosahedral capsids; 180 CP protomers assemble such that the asymmetric unit comprises 3 identical capsid subunits in 3 quasiequivalent positions termed A, B, and C (here termed CPA, CPB, and CPC). To date, all alpha- and betanodaviruses have been found to have capsomeres that present a trimeric spike. The arginine-rich N-terminal region of the viral CP interacts with the viral RNA segments (exemplified by PaV, Protein Data Bank [PDB]: 1F8V [15]), leading to the formation of a dodecahedral RNA cage at the virion interior.
We have previously shown that recombinant CP of MrNV produced using baculovirus expression in Spodoptera frugiperda (Sf9) cells assembles into virus-like particles (VLPs) with a diameter of approximately 40 nm [19]. We determined the intermediate resolution structure of the MrNV capsid using cryoEM and image reconstruction [20]. At this resolution, our reconstruction revealed distinct dimer-clustering of capsomeres in the T = 3 MrNV icosahedral capsid. Capsomeres were seen to form square, thin, and blade-like spikes on the virion surface. All other nodaviruses have been shown to assemble with trimeric capsomers. Our structure therefore revealed a strikingly divergent morphology for MrNV, lending weight to the proposed classification of MrNV within a new genus of nodaviruses [20].
Here, we present a high-resolution 3D reconstruction of the MrNV VLP, solved at 3.3 Å resolution. From these data, we have constructed an atomic model of the MrNV CP. We show that the shell (S) domain of MrNV CP possesses the canonical 8-stranded β-barrel structure common to all nodaviruses. There are, however, striking structural similarities between the MrNV capsid and those of members of the Tombusviridae. The protruding (P) domain exhibits a similar fold to that which has been previously shown for tomato bushy stunt virus (TBSV), although the spikes are narrower and are oriented quite differently in the 2 dimeric forms (AB and CC). CP–CP interactions in the S domain are stabilised by coordinated Ca2+ ions. Protomers forming CC dimers, located at the icosahedral 2-fold symmetry axes of the capsid, possess an ordered N-terminal arm (NTA). This passes along the capsid interior, forming intermolecular interactions with neighbouring protomers to stabilise the capsid. Unlike the NTA of TBSV, however, which folds back to form an additional β-strand in CPC before going on to form a structure known as a β-annulus at the adjacent icosahedral 3-fold axis, the NTA of the MrNV CPC crosses the icosahedral 2-fold symmetry axis and inserts into a symmetry-related CPB. The NTA then forms a β-annulus at the next nearest 3-fold symmetry axis before continuing to pass along the capsid interior, donating a strand to a β-sheet in a second neighbouring CPB molecule.
We also present an intermediate-resolution structure of the authentic MrNV virion, purified from infected larvae of M. rosenbergii, suggesting that the infectious virus exhibits an identical capsid structure to the one that we have determined for the VLP. Our data present a detailed structural view of this economically important pathogen and raise questions concerning the taxonomic classification of both MrNV and the related PvNV.
To calculate an atomic model of the MrNV CP, we sought to determine a high-resolution 3D reconstruction of the capsid. Frozen hydrated preparations of MrNV VLPs were imaged in a Thermo-Fisher Titan Krios at the United Kingdom Electron Bio-Imaging Centre (eBIC) (Fig 1A). A total of 40,883 particle images were used to calculate a reconstruction with an overall resolution of 3.3 Å (Fig 1B and 1C and S1 Fig, S1 Movie). The reconstructed density map closely matched our previously published 7 Å structure of the MrNV VLP produced in Sf9 cells, showing pronounced blade-shaped dimeric spikes on the capsid exterior and a dodecahedral cage of RNA density within the particle. A cross section through the reconstructed density revealed that the S domains of the VLP were sharply resolved, while the P domains were less well defined, having weaker, fuzzier density (Fig 1B). Local resolution analysis confirmed this, revealing that much of the S domain was solved to 3.2 Å resolution, while the tips of the dimeric capsomeres were poorer than 4 Å resolution (Fig 1D). Local resolution filtering and sharpening was applied with a B-factor of −140 Å2 to generate a density map that was suited for high-resolution model building.
The asymmetric unit of the T = 3 MrNV capsid comprises 3 copies of MrNV CP; CPA, CPB, and CPC. We have previously shown that the P domains assemble to form dimeric spikes, with AB dimers arranged about the 5-fold symmetry axes and the CC dimers located at the 2-fold symmetry axes. We set out to build the sequence of the MrNV CP (S2 Fig) into our density map to produce an atomic model of the MrNV CP for each quasiequivalent position. As a starting point, we docked a homology model into our density map [12, 20]. Overall, this model fitted poorly within the reconstructed density map, with the exception of 2 regions between amino acid residues 104–135 and 232–243. A model for the S domain of CPA was therefore manually built and refined from this starting point. This partial model was then docked to CPB and CPC and further edited and refined, leading to reliable models for the S domains at each quasiequivalent position. Interestingly, our density map presented density consistent with the presence of 2 metal ions per CP. Based on the surrounding residues and distances to coordinating atoms, we have modelled these as calcium ions (discussed below). In our 3D reconstruction, density for the S domain was very well resolved. This allowed us to model this region to a high degree of confidence and with relative ease.
The P domains were, however, rather less well defined and presented a more challenging task, particularly at the distal tips of the dimeric spikes. CPB was found to be the best resolved as judged by continuity of density, while CPC appeared the least well resolved. Throughout the amino acid sequence of the P domain, there are bulky amino acid side chains that gave confidence in our interpretation of the map. Following several rounds of manual editing and refinement, a model was achieved for the full asymmetric unit that matched our density map and had reasonable geometry (Fig 2, S2 Movie, S1 Table).
The N-terminal regions of CP that include the RNA binding sites (amino acid residues 21–29) were not resolved for any of the chains (CPA, CPB, or CPC) in our density map. For CPA, we have successfully modelled amino acid residues 56–371, while for CPB, we were able to build amino acid residues 55–371. CPC has a well-resolved NTA that allowed modelling from amino acid 31. Interestingly, the CPC NTA was found to form extensive contacts with symmetry-related CPB and CPC molecules, forming a network that crosses the capsid interior and is reminiscent of the NTAs previously described for several tombusviruses (Fig 3, S3 Movie). The CC dimer interface lies at the icosahedral 2-fold axis. Each CPC NTA emerges from the S domain close to this symmetry axis and interacts with 2 CPB protomers, donating β-strands to β-sheets within the CPB S domains. The CPC NTA extends across the CC dimer (and icosahedral) 2-fold symmetry axis and inserts into the first CPB subunit, which lies adjacent to the symmetry-related CPC subunit (Fig 3C and 3D). Moving from the C-terminus (C-ter) to the N-ter, the NTA then crosses the adjacent icosahedral 3-fold axis and inserts into the next nearest CPB subunit, donating a second β-strand to the β-sheet comprising that molecule and a symmetry-related CPC NTA (Fig 3E and 3F). This interdigitated arrangement of NTAs extending from CPs at the C-position was first described for TBSV [21] and termed a β-annulus owing to the manner in which CPC NTAs wrap around each other at the icosahedral 3-fold axes. Nevertheless, in the case of MrNV, the lacing together of CPC molecules is more extensive because the NTA does not fold back on the CPC to emerge from the S domain at the nearest 3-fold axis and there form the β-annulus structure (as it does in TBSV). Rather, it crosses the 2-fold axis of the CC dimer and then inserts into 2 CPB molecules arranged about the opposite 3-fold axis, where the β-annulus is formed. The last resolved N-terminal residue (Pro31) lies under the next neighbouring 2-fold axis. This is related to the originating 2-fold by a counterclockwise rotation of 120o about the 3-fold axis of the β-annulus (viewed from the capsid exterior). Although it is not resolved, the arginine-rich putative RNA binding site (amino acids 21–29) must therefore be located proximal to the 2-fold symmetry axes for the CPC chains.
The MrNV CP S domain comprises residues 62–242 and forms the contiguous shell of the capsid. The T = 3 assembly is made up of 180 copies of the canonical 8-stranded antiparallel β-barrel fold, known as the β-jelly roll. This is commonly seen in positive-sense RNA-containing viruses, including both the nodaviruses and tombusviruses. Another interesting parallel between the structure of MrNV and the tombusviruses is the presence of coordinated metal ions at the interface between CP subunits (Fig 4, S4 Movie). X-ray crystallographic difference mapping of TBSV following EDTA treatment identified 2 divalent-cation–binding sites at the interface between CPs within the asymmetric unit, which were modelled as Ca2+ [22]. Chelation followed by a rise in pH (>7.0) has been shown to cause a structural transition to a ‘swollen’ state in these virions, indicating that these divalent cations play a role in virion stabilisation or possibly control of uncoating. Based on the striking similarity in the locations of these putative metal ions in our data compared to those previously published for TBSV (PDB: 2TBV), and the surrounding residues, we have modelled these putative metal ions as calcium (Fig 4B).
We have previously noted the striking differences in the orientations of the P-domain dimer spikes, relative to the underlying capsid shell, between AB and CC dimers. The CC-dimer spike is rotated approximately 85° counterclockwise relative to that of the AB dimer (viewed from the capsid exterior). Moreover, CC-dimer P domains are raised from the capsid surface upon legs of density, whereas the AB P domains sit closer to the capsid shell and are tilted towards their nearest 2-fold symmetry axis. Our atomic model of the MrNV VLP reveals the reason for the substantial differences in pose of these 2 capsomere forms. There is a large linker region between the S and P domains at amino acid residues 241–258. In the CC-dimer, this linker emerges from the S-domain β-jelly roll and forms a straight leg that is normal to the capsid surface (Fig 2E). The interdomain linker in CPA and CPB, on the other hand, has 2 bends: one between residues Pro247 and Pro249, which causes the linker to make a right-angled turn, and another at Ile252–Gln254, which likewise causes a right-angled turn, restoring the path of the linker to its original radial orientation (Fig 2C). The twist in the linkers at the AB dimer induced by these turns therefore accounts for the major differences in the orientations of the 2 types of spike. Interestingly, although we previously noted that the CPB P domain was more closely apposed to the S domain than CPA, our model does not show any contacts. The AB P domain’s orientation is defined by interactions within the AB linker region and with the CPC P domain (S5 Movie).
CC-dimer spikes are less well resolved in our cryoEM map than those of the AB dimers. This is to be expected given the manner in which the AB-dimer spike is stabilised through interactions in the AB linker. In contrast, the P domains of CPC stand on extended polypeptide legs that may not offer the same support. The CC spike is instead stabilised by contacts between the P domains of CPB and CPC. AB spikes act as buttresses to the CC capsomere through polar interactions between amino acid residues 270–276 of CPB and 307–317 of CPC (Fig 5, S5 Movie). This gives rise to the formation of a blade-like superstructure that lays across the 2-fold symmetry axis and comprises 1 CC dimer and 2 AB dimers.
It is noteworthy that the previously described homology model for the MrNV CP structure [12] was based on the CP of a tombusvirus, cucumber necrosis virus (CNV; PDB: 4LLF [23]), rather than other known nodavirus structures. Although the homology model was a poor match for our cryoEM density map, our analysis has confirmed the hypothesised dimer-clustered T = 3 icosahedral capsid structure. Close inspection of the fold of the MrNV P domain also reveals an unexpected structural homology between this nodavirus and the tombusviruses (Fig 6). Tombusvirus P domains have been shown to comprise a 10-stranded antiparallel β-barrel made up of 2 β-sheets annotated as BAJEHG and CDIF (Fig 6A and 6C). Secondary structure assignment in the P domains of the MrNV structure was challenging owing to the poorer resolution in this region. Density in the P domain of CPB was found to be the most clearly resolved, allowing us to build a model in which we have identified a similar β-barrel motif composed of 9 strands arranged into 2 β-sheets (S3 Fig). Based on a 3D pairwise alignment of the P domains for CPB of CNV and MrNV, we have annotated this fold as AJEH2 and DIFGH1 (Fig 6B and 6D, S4 Fig).
To ensure that our atomic resolution model of the MrNV capsid is an accurate description of the authentic virion, we determined the structure of purified virions at intermediate resolution. Virions were purified from homogenised, MrNV-infected post-larvae and prepared for cryoEM. 3,931 particle images of frozen hydrated MrNV virions were processed to produce a reconstruction at 6.6 Å resolution (Fig 7, S5 Fig). At this resolution, the map appears identical to that of the MrNV VLP in all respects (compare Figs 7A and 1C). Furthermore, the packaged RNA shows a very similar, albeit noisier, structure to the previously described dodecahedral cage. Thus, we conclude that our high-resolution model is an accurate representation of the structure of the authentic MrNV virion.
We have previously described the intermediate resolution structure of VLPs generated following recombinant expression of the MrNV CP. This revealed a surprising divergence from known nodavirus structures [20]. The MrNV T = 3 icosahedral capsid was seen to assemble with dimeric rather than the usual trimeric capsomeres. Consistent with previously published nodavirus structures, we found that the MrNV VLP exhibited density suggestive of packaging of nucleic acids, most likely the cognate mRNA. We also observed a surprising difference in the orientations of the P-domain spikes between the 2 classes of dimer (AB and CC).
Here, we have extended this study, using cryoEM to calculate a 3D reconstruction of the MrNV VLP at near-atomic resolution. This has allowed us to build an atomic model of the capsid’s asymmetric unit. Our model reveals that MrNV, like most small icosahedral positive-sense RNA viruses, adopts the β-jelly–roll fold in the S domain. We identified major differences between AB and CC dimers in the linker region that connects the S and P domains, accounting for the radically different orientations of their respective P domains.
Beyond providing a detailed description of the structure of MrNV, our data revealed startling similarities between the MrNV capsid structure and those of tombusviruses. We found that the MrNV capsid’s asymmetric unit is stabilised by 6 Ca2+ ions in a manner highly reminiscent of that seen in TBSV. Furthermore, the CPC NTA was found to form an extensive network at the capsid interior that involved an interdigitated structure known as a β-annulus. This motif is also a feature of the tombusviruses. Finally, we found that the fold of the P domain consisted of a β-barrel that also bore a close resemblance to the P-domain structure of the tombusviruses.
Capsid stabilisation by binding of divalent cations is well documented in both nodaviruses and tombusviruses (Fig 8). The alphanodavirus FHV and the betanodavirus GNNV have both been shown to bind a single Ca2+ ion at the interface between each CP subunit in the asymmetric unit (Fig 8E–8H) [17, 24]. FHV also binds a single Ca2+ ion at the quasi-3–fold (Q3) symmetry axis lying at the centre of the asymmetric unit (Fig 8E). Metal binding in MrNV more closely resembles that seen in TBSV, however, in which 2 Ca2+ ions are bound by a DxDxxD motif [25].
There is no significant sequence homology between the CPs of alpha- and betanodaviruses, and comparison of known structures for these genera reveals substantial differences [17]. While both genera assemble capsids that have trimeric spikes, the FHV spike is small, comprising a single β-hairpin motif contributed by each CP subunit, whereas the GNNV CP has a distinct P domain that forms a more substantial capsomere. This feature was previously described as ‘TBSV like’ [26]. Alphanodavirus CPs fold such that both termini are located at the capsid interior; they also undergo proteolytic maturation to produce the γ-peptide, which is required for entry. Betanodaviruses and MrNV both have their C-termini at the capsid exterior and do not encode a γ-peptide.
Like MrNV, GNNV exhibits an extended CPC NTA that crosses the icosahedral 2-fold symmetry axis [17]. The CPC NTA also interacts with symmetry-related CPC NTAs at the adjacent 3-fold axis and was described as forming a β-annulus; however, the CPC NTA is not as extensive as that of MrNV and importantly does not form the interdigitated β-strands that are characteristic of the β-annuli of the tombusviruses (and MrNV). Rather, the network is stabilised by limited hydrogen bonding between CPC NTAs at the icosahedral 3-fold axes.
Close comparison of the structures of the MrNV capsid with those encoded by both genera of the Nodaviridae and by the Tombusviridae (Fig 8) led us to conclude that the structure of MrNV more closely resembles that of the tombusviruses than either the alpha- or betanodaviruses. The divergence of amino acid sequence and distinct structural features of MrNV compared with other nodaviruses supports the assertion that MrNV, along with the related PvNV, might be classified into a new genus, Gammanodavirus [10, 20]. Nodaviruses are characterised as small positive-sense RNA-containing viruses having bipartite genomes that infect fish and invertebrates. Tombusviruses are plant viruses that are classified on the nature of their RNA polymerases but are also seen to have consistent capsid structures. Thus, MrNV having features of both virus families poses a conundrum with respect to its taxonomic status.
It has been demonstrated that the C-terminal domain of the MrNV CP is important for virus attachment and entry. In particular, deletion of the last 26 amino acid residues substantially reduced infectivity. Inspection of the P-domain structure, however, strongly suggests that deletion of this region is liable to significantly disrupt the β-barrel, as it would remove 2 β-strands from the structure, one from the centre of each β-sheet (S6 Fig). Thus, while it seems likely that the receptor binding site is within the P domain, it may not be limited to the last 26 amino acid residues (345–371).
We have previously demonstrated insertion of heterologous epitopes into the MrNV capsid structure at the C-ters, such as the ‘a’ determinant of hepatitis B virus surface antigen (HBsAg) [27] and the ectodomain of matrix 2 protein (M2e) of influenza A virus [28]. Both epitopes were shown to be displayed on the surface of VLPs. Thus, MrNV VLPs present an attractive platform for antigen display. Our structure of MrNV CP now allows us to refine the placement of foreign epitopes. Indeed, each of the 4 loops on the outer surface of the P domain (amino acid residues 268–275, 296–303, 322–326, and 350–355) represents potentially improved targets for further insertions. Combined with the capacity to package nucleic acids, MrNV VLPS may therefore prove to be a useful tool for both vaccine and DNA/RNA delivery.
MrNV threatens livelihoods and food security in developing nations. Our atomic-resolution model of the MrNV capsid provides insights into the fundamental biology of this important pathogen, highlighting features that may prove important in our understanding of virus assembly or entry, such as the presence of metal ions that stabilise the asymmetric unit and the structure of the receptor-binding P domain. Such detailed understanding of the capsid structure provides a platform for the development of interventions to control or prevent disease outbreaks in the future.
The gene encoding MrNV CP was amplified from plasmid pTrcHis2-TARNA2 [29]. The forward and reverse primers used to amplify the coding region were 5′-ATG GCC CTT AAC ATC ACC ATG GCT AGA GGT AAA CA-3′ (NcoI restriction site is underlined) and 5′-CTA TCG TCG GCA ATA ATT AAG GCG AAT TCG AAG CTT ACG T-3′ (EcoRI restriction site is underlined), respectively. PCR profile was denaturation at 95°C for 3 min, followed by 35 cycles of i) denaturation at 95°C for 30 s, ii) annealing at 59°C for 30 s, and iii) extension at 72°C for 1 min. The final extension was performed at 72°C for 10 min. The PCR products were excised and purified using the QIAquick Gel Extraction Kit (Qiagen, Hilden, Germany). The purified DNA was ligated with the linearised pGEM-T vector (Promega, Madison, WI, United States of America) and introduced into competent Escherichia coli DH5α cells. The transformants were plated on Luria Bertani (LB) agar plates containing ampicillin (100 μg/ml). Following an overnight incubation at 37°C, single bacterial colonies were picked and cultured in LB broth. The orientation and nucleotide sequence of the DNA insert were confirmed by DNA sequencing.
To produce the MrNV-CP without a His tag, the QuickChange II site-directed mutagenesis kit (Agilent Technologies, Santa Clara, CA, USA) was used to create an NcoI restriction-nuclease–cutting site in the pFastBac-HTC plasmid (Invitrogen, Carlsbad, CA, USA). The primers used for mutagenesis were 5′-CGG GCG CGG ATC TCG GTC CGA AAC CAT GGC GTA CTA CCA TCA CC-3′ and 5′-GGT GAT GGT AGT ACG CCA TGG TTT CGG ACC GAG ATC CGC GCC CG-3′, where the NcoI restriction-cutting site is underlined.
The pGEM-T TARNA2 plasmid and the mutated pFastBacHT-C plasmid were digested with EcoRI and NcoI, respectively. The digested products were purified using the QIAquick Gel Extraction Kit (Qiagen, Hilden, Germany) and ligated together to produce the pFastBacHTC-TARNA2. This was introduced into competent E. coli DH10Bac cells (Invitrogen, Carlsbad, CA, USA) and plated on LB agar plates containing kanamycin (50 μg/ml), gentamicin (7 μg/ml), and tetracycline (10 μg/ml). White bacterial colonies containing the recombinant plasmid were selected and cultured in LB broth. The recombinant bacmid DNA was extracted, and the presence of DNA insert was confirmed by PCR. The primers used in the PCR were pUC/M13 forward 5′-CCC AGT CAC GAC GTT GTA AAA CG-3′ and pUC/M13 reverse 5′-AGC GGA TAA CAA TTT CAC ACA GG-3′.
Sf9 cells (8 × 105 cells/well) in a 6-well plate were transfected with the recombinant bacmid pFastBacHTC-TARNA2 using Cellfectin II reagent. The transfected cells were incubated at 27°C for 72 h. The cell culture medium was harvested by centrifugation at 500 × g for 5 min at 4°C. The P1 baculovirus stock was amplified by infecting the Sf9 cells (2 × 106 cells/mL) in serum-free Sf-900 III SFM medium (Gibco, Gaithersburg, MD, USA). The infected cells were incubated at 27°C for 72 h. The P2 baculovirus stock was harvested by centrifugation at 500 × g for 5 min at 4°C and stored at 4°C for subsequent experiments.
Sf9 cells were cultured as suspension cells at 27°C in a serum-free Sf-900 III SFM medium to reach a cell density of 2 × 106 cells/ml. Recombinant baculovirus stock (10% [v/v]) was added into the culture, which was further incubated for 4 d at 27°C. The MrNV capsid and the Sf9 cells were separated by centrifugation at 500 × g for 5 min at 4°C. The MrNV capsid was precipitated at 60% (w/v) ammonium sulphate saturation for 2 h at 4°C. The proteins were pelleted by centrifugation at 18,000 × g for 20 min at 4°C. The pellet was resuspended in HEPES buffer A (20 mM HEPES, 100 mM NaCl; pH 7.4) and dialysed in the same buffer overnight. The dialysed sample was purified by size-exclusion chromatography (SEC) using a HiPrep 16/60 Sephacryl S-500 HR column (GE Healthcare, Chicago, IL, USA), which was attached to a fast protein liquid chromatography (FPLC) system (Akta Purifier; GE Healthcare, Chicago, IL, USA). The purified protein was concentrated with a 100 kDa molecular cutoff centrifugal concentrator (Pall, USA), and the protein concentration was determined with the Bradford assay [30].
Lysate of MrNV-infected post-larvae was prepared according to published methods [31] with some modifications. Briefly, the infected post-larvae were homogenised in HEPES buffer B (25 mM HEPES, 150 mM NaCl; pH 7.4), and the homogenate was centrifuged at 6,000 × g for 10 min at 4°C to remove large debris. The supernatant was further clarified by centrifugation at 12,100 × g for 30 min at 4°C. The clarified supernatant was loaded onto a sucrose gradient (8–50% [w/v]) and centrifuged at 210,000 × g for 4.5 h at 4°C. Fractions (500 μl) were collected and analysed by SDS-PAGE and Western blotting. Fractions containing MrNV were pooled and dialysed in HEPES buffer B. The purified MrNV was concentrated by centrifugation using a centrifugal concentrator (molecular weight cutoff 10 kDa, Vivaspin Turbo 15, Sartorius, Göttingen, Germany). The final concentration of purified MrNV was determined using the Bradford assay [30].
Purified MrNV VLPs (at approximately 0.2 mg/ml) or virions (at approximately 0.1 mg/ml) were prepared for cryogenic transmission electron microscopy using a Thermo-Fisher Vitrobot Mk IV (Thermo Fisher Scientific, Waltham, MA, USA). Particles were imaged on thin-continuous carbon films that had been applied to C-flat holey carbon support films (R1.2/1.3; Protochips, Morrisville, NC, USA). Four μl of VLP or virion preparation was loaded onto a grid for 1 min, blotted for 4 s, and plunged into liquid ethane. Vitrified samples were imaged at low-temperature (around 95 K) and under low-electron–dose conditions. To collect high-resolution data on MrNV VLPs, grids were imaged at the eBIC, Diamond Light Source (UK) using a Thermo-Fisher Titan Krios (Thermo Fisher Scientific, Waltham, MA, USA) operated at 47,170× magnification. A total of 2,459 cryomicrograph movies were recorded on a Gatan K2 BioQuantum energy-filtered direct detector camera (Gatan, Pleasanton, CA, USA) operated in zero-loss imaging mode with a slit width of 20 eV. Five-s exposures were recorded in electron counting mode at a frame-rate of 4 frames per s and a dose rate of 1.8 electrons/pixel/frame. The pixel size was 1.06 Å/pixel. MrNV virions were imaged using a JEOL 2200 FS cryo-microscope (JEOL, Tokyo, Japan) operated at a nominal magnification of 50,000× and an accelerating voltage of 200 kV. Frozen grids were held in a Gatan 626 cryo-stage (Gatan, Pleasanton, CA, USA). 263 cryomicrograph movies were recorded on a Direct Electron DE20 camera (San Diego, CA, USA) as 2-s exposures at 20 frames per s and approximately 1.5 electrons/pixel/frame. The pixel size was 1.11 Å/pixel.
All image processing was performed using Relion 2.1 [32]. Image stacks of movie frames were motion-corrected using motioncor2 [33]. Defocus estimation was performed using GCTF [34]. For each dataset, a small subset of particle images was manually picked and subjected to 2D classification to prepare a template for automated particle picking. Thereafter, particles were automatically picked for all motion-corrected micrographs. Individual particle images were extracted in 5122-pixel boxes. For MrNV VLPs, a total of 60,939 putative particles were extracted from motion-corrected micrographs and subjected to 2D classification. Class averages showing particle images with well-resolved structure were selected, reducing the dataset to 56,762 particles. 3D classification was then used to select the best particles for inclusion in the final reconstruction, further reducing the dataset to 40,883 particles. This dataset was then refined, leading to the calculation of a reconstruction with an overall resolution of 3.3 Å. The MrNV virion reconstruction was calculated following an identical workflow in which 7,236 putative particle images were analysed, leading to the definition of a final dataset comprising 3,931 particle images. These data were reconstructed at a resolution of 6.6 Å. Reconstructions were evaluated to determine global and local resolution as well as estimated B-factors by postprocessing of maps calculated from randomised half sets of data, using the Relion postprocessing routine (S2 Table). In our study of MrNV VLPs, we used our previously calculated 3D reconstruction as a template for starting the classification. For authentic virions, we performed the first 3D classification analysis using a Gaussian sphere as the starting model to prevent model bias, as previously described [20].
Atomic models were built from the high-resolution density maps using the CCP-EM suite of programmes [35], in particular COOT [36]. The model was refined using REFMAC [37] and PHENIX [38]. Validation was performed using MOLPROBITY [39]. Secondary structure assignment was performed using STRIDE (http://webclu.bio.wzw.tum.de/cgi-bin/stride/stridecgi.py) [40]. Density maps and atomic resolution models were visualised using UCSF Chimera [41]. Validation of metal ion assignment was performed using the ‘checkmymetal’ server (https://csgid.org/metal_sites) [42]. Contact interface analysis was performed using the PISA server (http://www.ebi.ac.uk/msd-srv/prot_int/cgi-bin/piserver) [43]. A 3D pairwise alignment of the MrNV CP P-domain structure and that of CNV (PDB: 4LLF) was performed using the FatCat server (http://fatcat.burnham.org) [44]. Protein topology diagrams were generated using Pro-Origami (http://munk.csse.unimelb.edu.au/pro-origami/porun.shtml) [45] and edited using Inkscape (https://inkscape.org/en/).
The cryoEM map of the MrNV VLP was deposited in the Electron Microscopy Data Bank with accession number EMD-0129. The cryoEM map of the MrNV virion was deposited in the Electron Microscopy Data Bank with accession number EMD-0130. The atomic coordinates for the asymmetric unit of the MrNV VLP were deposited in the PDB with accession number PDB: 6H2B. The cryoEM image data for EMD-0129 are deposited in EMPIAR as motion-corrected single-frame micrographs with accession number EMPIAR-10203.
|
10.1371/journal.ppat.1005621 | Functional and Structural Mimicry of Cellular Protein Kinase A Anchoring Proteins by a Viral Oncoprotein | The oncoproteins of the small DNA tumor viruses interact with a plethora of cellular regulators to commandeer control of the infected cell. During infection, adenovirus E1A deregulates cAMP signalling and repurposes it for activation of viral gene expression. We show that E1A structurally and functionally mimics a cellular A-kinase anchoring protein (AKAP). E1A interacts with and relocalizes protein kinase A (PKA) to the nucleus, likely to virus replication centres, via an interaction with the regulatory subunits of PKA. Binding to PKA requires the N-terminus of E1A, which bears striking similarity to the amphipathic α-helical domain present in cellular AKAPs. E1A also targets the same docking-dimerization domain of PKA normally bound by cellular AKAPs. In addition, the AKAP like motif within E1A could restore PKA interaction to a cellular AKAP in which its normal interaction motif was deleted. During infection, E1A successfully competes with endogenous cellular AKAPs for PKA interaction. E1A’s role as a viral AKAP contributes to viral transcription, protein expression and progeny production. These data establish HAdV E1A as the first known viral AKAP. This represents a unique example of viral subversion of a crucial cellular regulatory pathway via structural mimicry of the PKA interaction domain of cellular AKAPs.
| Studies of human adenovirus (HAdV), a small DNA tumor virus, illustrate the profound impact of viral proteins on multiple host functions. The multifunctional E1A proteins of HAdV are particularly adept at targeting key cellular regulators. Mechanistically, E1A alters or inhibits the normal function of the cellular proteins that it targets, and also establishes new connections in the cellular protein interaction network. Through these interactions, E1A creates a cellular milieu more conducive for replication. Here we show that HAdV E1A mimics cellular A-kinase anchoring proteins (AKAPs) in both appearance and function. We found that the protein kinase A (PKA) regulatory subunits are conserved targets of most HAdV E1A species. Structural modeling and a docking analysis predict a remarkable similarity between the binding of E1A and cellular AKAPs to PKA, which was confirmed experimentally. In addition, we observed E1A-mediated relocalization of PKA subunits and competition between E1A and cellular AKAPs during infection that contribute to HAdV gene expression and overall viral replication. Together, our studies identify E1A as the first known viral AKAP, and reveal a unique example of viral subversion of the PKA pathway via structural mimicry.
| As obligate intracellular parasites, all viruses are critically dependent upon the host cell. Intensive selective pressure, rapid replicative cycle times and severe restrictions on viral genome size combine to drive virus evolution. As a consequence, viral regulatory proteins have been relentlessly forged into exquisitely sophisticated instruments that functionally reprogram the infected cell [1]. Studies of human adenovirus (HAdV), a small DNA tumor virus, illustrate the profound impact of viral proteins on multiple host functions to maximize viral propagation [2–7].
The multifunctional E1A proteins of HAdV are particularly adept at targeting key cellular regulators. Through these interactions, E1A creates a cellular milieu more conducive for replication. Indeed, E1A enhances cell cycle entry, subverts innate immunity and intensively reprograms the cellular gene expression program [5,6,8]. The modular E1A proteins are dense with short linear sequence motifs that bind to and alter the activity of dozens of critical cellular proteins [9,10]. Many of the interaction motifs in E1A are functional mimics of highly similar sequences present in cellular regulatory proteins. Thus, viral evolution has converged to generate specific high affinity protein interaction surfaces that perturb cell regulation by competing with endogenous targets.
Cellular compartmentalisation of proteins is a widespread cellular mechanism that ensures the interaction of signalling molecules with a localized subset of appropriate effector proteins. As one well studied example, the activation of protein kinase A (PKA) signalling by the second messenger cyclic AMP (cAMP) is precisely restricted to discrete subcellular regions [11]. This is primarily achieved by a diverse set of cytoplasmic scaffolds collectively known as A-kinase anchoring proteins (AKAPs). AKAPs bind to PKA regulatory subunits via a well characterized amphipathic α-helix, localizing them to distinct cellular loci near PKA’s substrates [12]. Compartmentalization of PKA allows its enzymatic activity to be directed in a spatially defined and temporally specified manner and disregulation of this compartmentalization has pathophysiological consequences [13].
Although the E1A proteins from multiple HAdVs can synergize with cAMP to alter viral and cellular gene expression [14–18] the exact mechanism remains unclear. Interestingly, HAdV-12 E1A binds directly to the regulatory subunits of PKA, resulting in the relocalization of one isoform from the cytoplasm to the nucleus [19,20]. These results suggest that E1A may function as a ‘viral AKAP’ by redirecting the subcellular localization of PKA to alter transcription.
Here we show that HAdV E1A mimics cellular AKAPs in both appearance and function. We found that the PKA RIα and RIIα subunits are conserved targets of most HAdV E1A species. Structural modeling and a docking analysis predict a remarkable similarity between the binding of E1A and cellular AKAPs to PKA, which was confirmed experimentally. In addition, we observed E1A-mediated relocalization of PKA subunits and competition between E1A and cellular AKAPs during infection that contribute to HAdV gene expression and overall viral replication. Together, our studies identify E1A as the first known viral AKAP, and reveal a unique example of viral subversion of the PKA pathway via structural mimicry.
The E1A proteins from multiple HAdVs synergize with cAMP to alter viral and cellular gene expression. A direct interaction between HAdV-12 E1A and the type I and type II regulatory subunits of PKA (RIα and RIIα) was previously reported, but has not been investigated further [19]. It was also not known if this interaction was specific to HAdV-12 E1A. To further explore the E1A-PKA interaction, A549 lung adenocarcinoma cells were infected with wildtype (WT) HAdV-5 or a ΔE1A virus and co-immunoprecipitations were performed (Fig 1A). Similarly to HAdV-12 E1A, HAdV-5 E1A interacted with endogenous PKA regulatory subunits RIα and RIIα. Interestingly, we also found a previously unknown interaction between HAdV-5 E1A and the endogenous PKA catalytic subunit Cα. siRNA-mediated downregulation of specific PKA subunits demonstrated that E1A’s association with Cα required expression of RIα and RIIα (Fig 1B). This suggests that the interaction with the Cα subunit may be indirect and that E1A binds the entire PKA holoenzyme.
To determine if the interactions between E1A and the PKA subunits are evolutionarily conserved across the different HAdV species, HT1080 fibrosarcoma cells were transfected with vectors expressing the PKA subunits and the largest E1A isoform from six different HAdV species. Co-immunoprecipitation analysis revealed that RIα, RIIα, and Cα all interacted with each of the E1A proteins tested, with the exception of HAdV-4 (Fig 1C). The conservation of the E1A-PKA interaction across most HAdV species suggests that targeting of PKA is an important evolutionarily conserved function of E1A.
E1A is comprised of a series of protein interaction modules that often can function independently [8]. To grossly define which portion of E1A is required for PKA interaction, lysates from HT1080 cells expressing the PKA subunits and the indicated large fragments of HAdV-5 E1A expressed as EGFP-fusions, were subjected to Co-IP. The N-terminal 82 residues of HAdV-5 E1A were sufficient for association with PKA (Fig 1D). Interestingly, this region of E1A has been previously shown to be involved in alterations in cAMP signalling [21]. In addition, the interaction of HAdV-12 E1A with PKA similarly mapped to residues 1–79 in a yeast interaction assay [19]. As can be seen from the amino acid sequence alignment, there are several areas of high sequence similarity in this region in the E1A proteins from various HAdV species (Fig 1E).
To determine the minimal region of HAdV-5 E1A necessary and sufficient for PKA interaction, we carried out a detailed mutational analysis of the N-terminus of E1A. Cells were co-transfected with vectors expressing PKA subunits and the indicated E1A mutants, each expressed in the context of full-length HAdV-5 E1A and containing a small in-frame deletion in the N-terminus. As expected, deletion of residues 1–82 abrogated interaction with PKA, confirming that the E1A N-terminus as necessary and sufficient for binding PKA (Fig 2A). Several smaller, overlapping deletions also had similar defects for PKA-binding, specifically Δ1–29, Δ4–25, Δ16–28, and Δ26–35. However, adjacent deletion mutants Δ1–14 and Δ30–49, or more distant deletions retained interaction. This suggests that a region spanning residues 14–29 of HAdV-5 E1A is necessary for PKA binding.
We next co-transfected cells with PKA and small E1A fragments expressed as EFGP fusions. Co-immunoprecipitation on lysates of these cells demonstrates that the 14–28 region of E1A was sufficient to confer an interaction with PKA (Fig 2B). This region is similar in the E1A proteins from most HAdV species (Fig 1E) and also has noticeable sequence similarity to the PKA-binding regions of a number of cellular AKAPs (Fig 2C). Interestingly, AKAPs bind PKA regulatory subunits via an amphipathic α-helix secondary structure motif [22,23], and modeling of the N-terminus of HAdV-5 E1A predicts it also forms an amphipathic α-helix (Fig 3). Furthermore, the E1A proteins from all HAdV species are strongly predicted to form an α-helix in this region [8,21], with the exception of HAdV-4 E1A, which is predicted to form a lower-confidence helix (S1A Fig) and does not bind PKA efficiently (Fig 1C). Taken together, this suggests that E1A binds PKA by structurally mimicking the AKAPs’ amphipathic α-helix motif.
We performed in silico molecular modeling to predict the docking of the N-terminus of E1A with PKA. Docking simulations performed using the crystal structure of a dual-specificity cellular AKAP in complex with the RIα homodimer of PKA suggest that the interaction of E1A with RIα is virtually equivalent to that of the cellular AKAP (Fig 3A–3D and 3F). This model predicts a number of distinct interactions between E1A and RIα (Fig 3E and 3G), which were experimentally tested (Fig 3H and 3I). E1A mutants D21K, E26K, V27K and E26A/V27A, reduced the interaction with RIα as predicted, whereas substitution of E25 with K, which is not predicted to alter binding, had no effect (Fig 3H). Similarly, RIα mutants Q28E, L31A K32A and I35A/V36A displayed a reduced ability to bind E1A as predicted by the model (Fig 3I). These results indicate that the docking model can correctly predict key residues necessary for binding, which further suggests that E1A structurally mimics a cellular AKAP in order to bind PKA.
Cellular AKAPs bind to the docking/dimerization (D/D) domain located at the N-terminus of the PKA regulatory subunits RIα and RIIα [22,24]. Given the sequence and predicted structural similarity between E1A and cellular AKAPs, we tested if the D/D domain was necessary for the interaction with E1A. Transfected HAdV-5 E1A was unable to co-immunoprecipitate RIα or RIIα lacking their D/D domain (Δ1–63 and Δ1–45, respectively, Fig 4A and 4B). In addition, when the D/D domains of RIα and RIIα were expressed as fusions to EGFP, they alone were sufficient to co-immunoprecipitate E1A (Fig 4C). Thus, the N-terminus of E1A not only resembles an AKAP based on sequence, but also binds to the same site on the PKA regulatory subunits targeted by cellular AKAPs.
We next determined if the structural similarity between E1A and cellular AKAPs extended to functional similarity. We tested whether E1A could compete with endogenous AKAPs for PKA-binding during infection. A549 cells were infected with WT HAdV-5, a ΔE1A virus, or a virus expressing an E1A mutant unable to bind PKA (Δ4–25). Cell lysates were prepared 18 hours post-infection, subjected to immunoprecipitation with an anti-AKAP7 antibody and any co-precipitating PKA subunits were detected via western blot with specific antibodies for each target. AKAP7 is a dual-specificity AKAP [25], which binds both RIα and RIIα directly, and indirectly binds Cα. Infection with HAdV-5 did not alter the expression of AKAP7 or the various PKA subunits. However, infection disrupted the endogenous interactions between AKAP7 and PKA. Disruption of the AKAP7-PKA interaction during infection required E1A and was dependent on the AKAP like domain in E1A (Fig 5A). These data establish that the AKAP like region in E1A competes with endogenous AKAPs for PKA interaction during infection. These results also suggests that E1A can out-compete at least some cellular AKAPs for binding to PKA, which likely contributes to previously observed perturbation of cellular cAMP signalling by HAdV infection [14,16,21].
We next tested whether in silico-designed peptide inhibitors, which block AKAP-PKA interactions by binding PKA regulatory subunits with sub-nanomolar affinities, could affect E1A’s interaction with PKA. These well characterized inhibitors are short peptides expressed as EGFP-fusions which specifically block binding to RIα (RIAD) or RIIα (sAKAPis) [26,27]. HT1080 cells were co-transfected with vectors expressing the PKA subunits, WT E1A, and each of the inhibitors. Lysates were subjected to immunoprecipitation with an anti-E1A antibody and interacting PKA subunits were detected by western blot. As expected based on their high affinity, both RIAD and sAKAPis competitively reduced E1A’s interaction with PKA in a subunit-specific manner, reinforcing E1A’s role as a dual-specificity viral AKAP (Fig 5B).
Using an expression construct for a known cellular dual-specificity AKAP (AKAP1) [27], we next tested E1A’s ability to rescue the PKA-binding function of this AKAP when its PKA-binding domain was deleted. HT1080 cells were co-transfected with PKA subunits and EGFP-fusions of WT AKAP1, an AKAP1 mutant lacking its PKA-binding domain (AKAP1Δ), or an AKAP1 construct with E1A residues 14–28 cloned in lieu of the deletion (AKAP1-E1A). Lysates were subjected to immunoprecipitation with an anti-EGFP antibody and co-precipitating PKA was detected via western blot. As expected, the AKAP1Δ mutant lost the ability to bind PKA. However, incorporation of the E1A AKAP-like sequence into this mutant rescued PKA-binding to WT levels (Fig 5C). Together, these results strongly suggest that the AKAP-like motif in E1A is functionally indistinguishable from that found in an authentic cellular AKAP.
Transfection of cells with HAdV-12 E1A induces a relocalization of the RIIα subunit of PKA from the cytoplasm to the nucleus [19]. We tested E1A’s ability to alter PKA’s subcellular localization in vivo during a HAdV-5 infection (Figs 6 and S2A). A549 cells were infected with WT virus (dl309), a ΔE1A virus (dl312), or the Δ4–25 E1A deletion mutant virus (dl1101) that does not bind PKA. At 18 hours post-infection, cells were subjected to immunofluorescence staining and biochemical fractionation to determine the subcellular localization of PKA. In WT-infected cells, endogenous RIα was rerouted from the cytoplasm into the nucleus. Additionally, in infected cells, RIα appeared to overlap with the HAdV-5 encoded DNA-binding protein (DBP), suggesting possible co-localization with viral replication centres during infection (S3 Fig). In contrast, the distribution of PKA subunits in cells infected with either the ΔE1A or Δ4–25 virus resembled uninfected cells. Thus, the relocalization of RIα is E1A-dependent and requires the AKAP motif. Subcellular localization of RIIα appeared to be unaffected by the presence of E1A and Cα retained its nuclear/cytoplasmic phenotype in both uninfected and infected cells, thereby rendering any conclusions regarding its relocalization difficult (S2A Fig).
Interestingly, RIα, but not RIIα, is similarly trafficked into the nucleus of HEK293 cells, which stably express HAdV-5 E1A. Knockdown of E1A in HEK293 cells reduces the amount of RIα in the nucleus, further suggesting that E1A is functioning as an AKAP in these cells to redistribute PKA (S2B Fig). Additionally, A549 cells transiently transfected with HAdV-5 E1A conferred a similar result, whereas cells transfected with HAdV-4 E1A (which does not bind PKA via Co-IP [Fig 1]) did not affect PKA localization (S4 Fig). These results demonstrate that the AKAP function of HAdV-5 E1A can alter the localization of PKA whereas E1A from a HAdV species that does not bind PKA lacks this biological function. Interestingly, HAdV-5 E1A appears to primarily affect type-I PKA, whereas the previously reported effect of HAdV-12 E1A was restricted to type-II PKA.
Previous studies indicated that E1A and cAMP synergize to activate viral gene expression [14–16,18,21,28]. To determine if the E1A-PKA interaction contributes to HAdV early gene transcription, A549 cells were first treated with control siRNA or siRNA specific for each PKA subunit and then infected with WT (dl309), ΔE1A (dl312), or Δ4–25 (dl1101) HAdV-5. Cells were harvested 20 hours post-infection, cDNA was prepared and the expression of a panel of HAdV early genes known to be activated by E1A was determined by quantitative real-time PCR. Knockdown of RIα, RIIα, or Cα did not affect expression of the E1A (Fig 7A) or E1B (Fig 7B) transcription units for any of the tested viruses. However, mRNA levels were significantly reduced for both the E3 (Fig 7D) and E4 (Fig 7E) transcription units in WT virus infected cells treated with siRNA for each of the PKA subunits, demonstrating that PKA plays a role in the regulation of these transcription units. Importantly, cells infected with the Δ4–25 virus also showed decreased expression of E3 and E4 as compared to WT infection, and this was not further reduced by knockdown of any PKA subunit (Fig 7D and 7E). This is fully consistent with the inability of this mutant E1A protein to bind PKA and relocalize it to the nucleus. Mechanistially, Chromatin immunoprecipitation (ChIP) experiments showed that PKA’s catalytic subunit (Cα) was recruited to the HAdV E3 and E4 promoter regions in an E1A-dependent manner (S6 Fig). In contrast, E1A did not specifically recruit Cα to the E1B or GAPDH promoters, whose transcription was unaffected by the E1A-PKA interaction (Fig 7). These results strongly support a mechanism of early gene activation that relies on the AKAP function of E1A.
Although knockdown of PKA regulatory subunits had no statistically significant effect on E2 transcripts, knockdown of the catalytic subunit reduced E2 expression for both WT and Δ4–25 virus (Fig 7C). This suggests an independent effect for PKA on this transcription unit that does not rely on the AKAP motif.
To extend the observations that PKA plays a role in HAdV gene expression, we further examined PKA’s role in HAdV-5 protein production (S5 Fig). A549 cells were treated with control siRNA or siRNA specific for each PKA subunits and infected with WT HAdV-5. Cell lysates were collected at 12, 24, and 36 hours post-infection. Viral protein production was assayed by western blot using antibodies against an array of HAdV-5 proteins representing both early and late transcription units. Compared to control-treated cells, knockdown of PKA subunits had no effect on the production of HAdV-5 E1A proteins. In contrast, knockdown of the individual PKA subunits caused a notable reduction in several early proteins. These included a reduction in E3-19K at each time point examined, a reduced level of E4orf6 expression at 24 hours post-infection and a delay in expression of the E2-encoded DBP. E1B-55K was also reduced, most notably in the RIα knockdown. Interestingly, many of the late proteins also exhibited lower expressions levels in PKA-knockdown cells, including hexon, penton, protein V, and protein VII. This confirms a role for PKA in regulating HAdV-5 gene expression.
To establish the biological significance of E1A’s role as a viral AKAP, we also assessed the effect of the E1A-PKA interaction on viral replication (Fig 8). A549 cells were treated with either control siRNA or siRNA specific for each PKA subunit and infected with either WT or Δ4–25 HAdV-5. Production of infectious virus progeny was assayed at various time points over 72 hours by plaque assay. The production of WT virus was reduced by knockdown of each PKA subunit when compared to control-treated cells. Although the production of the Δ4–25 virus was reduced as compared with WT infection, it was not further reduced by knockdown of either RIα or RIIα. This again suggests that the lack of PKA-binding by this E1A mutant is functionally equivalent to PKA knockdown. These results indicate that HAdV replication requires PKA activity and that E1A’s interaction with PKA’s regulatory subunits is required for WT-levels of replication. Interestingly, we observed a reduction in progeny production for both WT and Δ4–25 virus in cells treated with Cα-specific siRNA. However, the observed reduction compared to control-siRNA treated cells was more severe in the WT infection, suggesting an additional role for PKA in HAdV-5 infection that is E1A-independent and specific for PKA’s catalytic subunit. Altogether, these results confirm that the targeting of PKA by the AKAP motif in E1A is a critical aspect in the HAdV-5 replicative cycle.
Cellular AKAPs function as scaffolds that target PKA and other signaling enzymes to specified subcellular locations. These multivalent anchoring proteins serve as important focal points for the processing and integration of intracellular signalling [29,30]. We report here that the adenovirus E1A oncoproteins function as the first known viral AKAPs. Intriguingly, E1A interacts with the with both the RIα and RIIα subunits of PKA in a way that precisely mimics that of cellular dual-specificity AKAPs. Specifically, we found that E1A bound to the N-terminal D/D domain of the regulatory subunit dimer of PKA, which is the same exact domain targeted by cellular AKAPs [11,12]. We identified a short conserved sequence in HAdV-5 E1A spanning residues 14–28 that was necessary and sufficient for interaction with either RIα or RIIα. Like the PKA interaction domains of cellular AKAPs, this region of E1A is predicted to form an amphipathic α-helix. This apparent structural mimicry allows E1A to bind PKA with an affinity comparable to cellular AKAPs, such that E1A can successfully compete with endogenous cellular AKAPs for PKA interaction during infection (Fig 9).
In support of our in vivo and in vitro results, molecular modeling based on a known structure of the AKAP/PKA interaction predicts that E1A binds the exact same surface of the PKA regulatory subunit in a fashion virtually identical to that determined for cellular AKAPs (Fig 3). Substitution of specific residues predicted by this model to make contacts reduced the interaction in vivo, supporting the validity of this structural model of molecular mimicry.
Functionally, as observed for cellular AKAPs, E1A relocalizes PKA to target sites of action. In the case of E1A, the interaction with PKA induces a specific relocalization to the nucleus, which contributes to viral gene expression and efficient virus propagation during infection. Competition by E1A with cellular AKAPs for PKA interaction may also influence cellular gene expression, which may provide some insight into the previous observations that E1A influences cAMP signalling [14–16,21,28,31]. The E1A region mapped as necessary and sufficient for PKA-binding also overlaps with regions previously implicated in E1A’s ability to act as a transforming oncoprotein [32]. Whether PKA contributes to the transforming ability of E1A remains unknown, though PKA itself has been investigated in a variety of cancer-related functions [13,33,34]
Our results also demonstrate that PKA is a conserved target of the E1A proteins from multiple HAdV species, suggesting that this interaction is functionally important for the virus. The E1A proteins from all HAdV types tested bound PKA strongly, with the exception of HAdV-4 E1A which also failed to relocalize PKA (S4 Fig). Modeling of an interaction between HAdV-4 E1A and PKA predicts that key electrostatic and hydrophobic contacts are absent, which are necessary for the HAdV-5 E1A PKA interaction (Figs S1 and 3H). Interestingly, HAdV-4 is unique as it is the sole member of species E HAdV and arose from an interspecies recombination event between chimpanzee and human adenovirus [35].
As mentioned above, during HAdV-5 infection, E1A was able to out-compete endogenous cellular AKAP7 for PKA interaction; however, there exist a plethora of other, diverse AKAPs with varying affinities for PKA. For example, the in silico-designed ‘super AKAPs’ RIAD and sAKAPis [26,27] blocked the binding of E1A to the PKA RIα and RIIα subunits, respectively. Thus, the affinity of the E1A/PKA interaction is not high enough to compete with synthetic AKAPs with sub-nanomolar affinities for PKA. Consequently, these inhibitors are potential tools for further study of E1A function in the context of its role as a viral AKAP.
During HAdV-5 infection, a substantial fraction of the RIα subunit was trafficked from the cytoplasm into the nucleus in an E1A-dependent manner. We also observed signal overlap between RIα and HAdV DBP (S3 Fig), suggesting co-localization with viral replication centres. Interestingly, the HAdV-5 E1A-mediated shift in RIα localization is the opposite finding reported for E1A from HAdV-12, which relocalized RIIα only [19]. While both E1As bound to both type-I and–II PKA in Co-IP assays, our studies suggest that in biologically-relevant conditions they each may exhibit higher affinity or preference for one PKA flavour over another, a property shared by many cellular AKAPs [11,12,27]. The binding affinities and potential preferences of E1A proteins from the other HAdV species during infection remains to be fully explored. It also remains to be determined if type-I and type-II PKA are completely interchangeable, or if there are functional consequences driving the preference of each virus for each regulatory subunit type.
Interestingly, nuclear localization of the PKA holoenzyme is considered relatively unusual, but has been studied in detail in HEK-293 cells [36]. We confirmed nuclear localization of RIα in these cells, which constitutively express HAdV-5 E1A [37]. Our results suggest that nuclear localization of PKA in HEK-293 cells is a likely consequence of the AKAP function of E1A. Furthermore, our data suggests that the results of studies of PKA function in these cells may be confounded by the impact of viral manipulation of this pathway.
The targeting of PKA by E1A is required for maximal expression of the HAdV-5 E3 and E4 transcription units. It appears that E1A is using the regulatory subunits of PKA as a bridge to bind Cα, redistributing it to associate with other E1A binding partners at preferred sites within the nucleus, such as the HAdV early gene promoters (S6 Fig). This could establish new localized connections with cAMP-regulated transcriptional machinery, such as CREB or ATF, at viral or cellular loci. This may help explain the previously-observed ability of E1A to cooperate with cAMP in transcriptional activation [14–16].
The importance of PKA during a productive infection is further underscored by our observation that siRNA-mediated downregulation of PKA subunits reduces progeny production by WT HAdV-5. It is likely that the observed defect in the virus’ ability to express numerous crucial transcripts and proteins in the absence of PKA (or the AKAP function of E1A) contributes greatly to this. It is also possible that the E1A-PKA interaction affects cellular tasks that influence HAdV replication, given that PKA and cAMP have been previously shown to extensively modulate cellular transcription, protein expression, and cell signalling [38–41]. As expected, growth of a virus expressing an E1A mutant unable to bind PKA (Δ4–25) was reduced relative to WT. Importantly, knockdown of regulatory subunits RIα and RIIα did not further reduce the overall replication of this mutant, confirming that the lack of the E1A-PKA interaction contributes to its growth defect. Interestingly, loss of Cα expression negatively affected overall viral replication for both WT and Δ4–25 viruses, suggesting an E1A-independent effect of Cα on the HAdV life cycle. This may be related to reports that PKA activity is involved in dynein-mediated transport of species C HAdV virions to the nucleus during the establishment of infection [42,43].
Although E1A is presently unique in its ability to function as a viral AKAP, the important role of PKA in cellular homeostasis makes it an attractive target for modulation during infection by other viruses. For example, the Herpes simplex virus-1 US3 kinase interacts with and activates PKA to block apoptosis [44]. Varicella-zoster virus also upregulates PKA expression and modulates phosphorylation of PKA substrates to improve replication [45]. More typically, PKA is recruited to phosphorylate viral proteins, altering their stability, folding or ability to interact with other targets [46–49]. As one well characterized example, the E6 oncoprotein from human papillomavirus (HPV) is phosphorylated by PKA during infection, allowing it to interact with numerous cellular proteins [50,51]. While E1A does not appear to be a substrate for PKA, its unique mechanism of commandeering this enzyme via mimicry highlights the diverse ways in which viruses can repurpose the same cellular factors. It is also interesting that rather than encoding an entire PKA ortholog or an entire viral protein to subvert PKA function, HAdV uses a short 15 amino acid fragment of the versatile E1A protein to retask PKA for the benefit of the virus. The fact that the AKAP mimic motif in E1A also overlaps regions required for targeting other cellular regulatory proteins [7,52,53] further demonstrates the incredible effect of selective pressure on maximizing the impact of the relatively limited coding capacity of HAdV.
In summary, we conclusively identify E1A as the first known viral AKAP. We demonstrate that the N-terminus of E1A has evolved to mimic the appearance, structure and function of the PKA interaction domain of cellular AKAPs. Furthermore, we have established that the AKAP function of E1A plays a biologically significant role in redirecting PKA to the nucleus during infection, where it is repurposed to enhance HAdV early gene expression and viral progeny production.
Human A549 (provided by Russ Wheeler, Molecular Pathology/Genetics London Health Sciences Centre), HT1080 (purchased from the American Type Culture Collection), and HEK293 cells [37] were grown at 37°C with 5% CO2 in DMEM (Multicell) supplemented with 10% fetal bovine serum (Gibco). Plasmids were transfected into A549 and HT1080 cells using XtremegeneHP (Roche) following the manufacturer’s recommendation. After 24 hours of incubation, transfected cells were used for downstream experiments.
All viruses were derived from the HAdV-5 dl309 background and express the 289R and 243R E1A proteins [54,55]. A549 cells were infected with WT (dl309) or HAdV containing the indicated E1A mutant: ΔE1A (dl312), Δ4–25 (dl1101). Cells were infected at a multiplicity of infection (MOI) of 5 pfu/cell. Cell cultures were infected at 50% confluence and subconfluent cells were collected at indicated time points for downstream experiments.
Downregulation of PKA subunits RIα, RIIα, and Cα was performed using Silencer Select siRNA (Thermo). Four hours after seeding, siRNA was delivered to A549 cells via transfection with Silentfect (BioRad) according to the manufacturer’s instructions. A scrambled siRNA was used as a negative control. Treated cells were used for experiments 48 hours post-transfection. Downregulation of E1A in HEK293 cells was performed using a cocktail of E1A-specific siRNAs generated by Thermo Fisher’s custom siRNA design platform. All siRNAs used can be found in S1 Table.
All constructs were expressed in vectors under control of the CMV promoter. WT RIα, RIIα, and Cα were PCR amplified (from Addgene 23741, 23789 and 23495) and cloned into pcDNA4-HA and pCANmyc. RIα Δ1–63 and RIIα Δ1–45 were similarly derived and expressed in pCANmyc. D/D fragments of RIα and RIIα were both expressed as EGFP fusions from pEGFP-N1. E1A fragments were expressed as fusions to EGFP and either described previously (1–82, 93–139, 139–204, 187–289) [56] or derived via PCR and cloned into pEGFP-C2 (1–29, 1–14, 14–28, 16–28, 29–49). WT HAdV-5 E1A and its associated deletion mutants were all expressed in pcDNA3. These constructs were previously described (Δ4–25, Δ26–35, Δ30–49, Δ48–60, Δ61–69, Δ70–81) [56] or generated via PCR (Δ1–82, Δ1–14, Δ1–29, Δ16–28). Point of mutants of E1A (D21K, E6K, V27K, E26A V27A, E5K) and RIα (Q28E, L31A K32A, I35A V36A) were generated by PCR and expressed in pcDNA3 and pcDNA-HA respectively. The largest E1A isoform from the six HAdV species were cloned as EGFP fusions. D-AKAP1 mutants were generated via PCR of a construct kindly provided by Thomas Kuntziger (Oslo) and expressed in pEGFP-C2. RIAD-EGFP and sAKAPis-EGFP were provided by Alan Howe (Vermont).
Cells were lysed in NP40 lysis buffer (150mM NaCl, 50mM Tris-HCL pH 7.5, 0.1% NP-40) with protease inhibitor cocktail. Protein concentrations were determined using BioRad protein assay reagent using BSA as a standard. Immunoprecipitations were carried out at 4°C for 4 hours, or overnight for endogenous interactions. 2% of sample was kept as input control. After washing with NP40 buffer, complexes were boiled in 25 µL of LDS sample buffer for 5 minutes. Samples were separated on NuPage 4–12% Bis-Tris gradient gels (Life Technologies) and transferred onto a PVDF membrane (Amersham). Membranes were blocked in 5% skim milk constituted in TBS with 0.1% Tween-20. All antibodies used can be found in S2 table. Horseradish peroxidise-conjugated secondary antibody was detected using Luminata Forte or Crescendo substrate (Millipore). For biochemical fractionation of infected A549 cells, nuclear and cytoplasmic extracts were acquired using an NE-PER kit from Thermo-Fisher.
Cells were fixed in 3.7% paraformaldehyde, permeabilized on ice using 0.2% Triton X-100, and blocked using 3% BSA in phosphate-buffered saline (PBS). Samples were incubated in the indicated primary antibody for 1 hour at room temperature or 4°C overnight and another hour at room temperature with secondary antibodies (Alexa Fluor 594 α-rabbit, Alexa Fluor 488 α-mouse) (Life Technologies). Samples were mounted with Prolong Gold reagent containing DAPI (Life Technologies). Confocal images were acquired using a Fluoview 1000 laser scanning confocal microscope (Olympus Corp). Non-confocal images were acquired using an Eclipse Ti-U inverted laser microscope (Nikon). Quantification of total cellular signal and nuclear signal was conducted using ImageJ. Cells were normalized for both cytoplasmic and nuclear size and %nuclear signal was determined as previously described [57].
Total RNA was prepared with Trizol extraction (Life Technologies). A total of 1 μg of RNA was reverse transcribed into cDNA by random priming using the qScript cDNA supermix (Quanta Biosciences) following the manufacturer’s instructions. Quantification of cDNA was done using Power SYBR-Green mastermix (Applied Biosystems) with oligonucleotide sequences that specifically recognize the indicated target. GAPDH was used as a control for total CDNA along with a no-RT negative control. Results were normalized to the GAPDH and uninfected samples and calculated using the ΔΔCt method [58]. Primers used can be found in S3 Table.
Approximately 107 cells per sample were cross-linked in 2mM ethylene glycol bis(succinimidyl succinate) (EGS) for 1 hour followed by 1% formaldehyde for 15 minutes at room temperature. Reactions were quenched with 0.125M glycine and washed twice with cold PBS. Cell pellets were processed in ChIP buffer 1 (10mM HEPES [pH 6.5], 10mM EDTA, 0.5mM EGTA, 0.25% Triton X-100), ChIP buffer 2 (10mM HEPES [pH 6.5], 1mM EDTA, 0.5 mM EGTA, 200mM NaCl), and ChIP buffer 3 (50mM Tris-HCl [pH 8], 10mM EDTA, 0.5% Triton X-100, 1% SDS, and protease inhibitors). Lysates were sonicated in an ultrasonic bioruptor bath (Diogenode) to yield DNA fragments between 200–500 basepairs. 80 μg of chromatin supernatant was used for ChIP, 1% of this was kept for input controls. Samples were diluted 10-fold in ChIP dilution buffer (50mM Tric-HCl [pH 8], 10mM EDTA, 150mM NaCl, 0.1% Triton X-100, protease inhibitors) and precleared with 30μL of Protein G Dynabeads (Invitrogen) for 1 hour at 4°C. Immunoprecipitations were performed overnight at 4°C using 5μg of the indicated antibody in S2 Table. The next morning, 30μL of Dynabeads were incubated with each sample for 2 hours. Beads were then washed with twice each with wash buffer 1 (20mM Tris-HCl [pH 8], 2mM EDTA, 150mM NaCl, 1% Triton X-100, 0.1% SDS), wash buffer 2 (20mM Tris-HCl [pH 8], 2mM EDTA, 500mM NaCl, 1% Triton X-100, 0.1% SDS), and wash buffer 3 (10mM Tris-HCl [pH 8], 1mM EDTA). Immunocomplexes were extracted twice with 150μL of elution buffer (0.1M NaHCO3, 1% SDS). 25μL of 2.5M NaCl was added to the 300μL pooled elutions and incubated overnight at 65°C to de-crosslink the complexes. DNA was purified using a PCR purification kit (Thermo). qPCR using SYBR-Green was performed as described previously using 80nM oligos and 0.5μL of ChIP DNA per 15μL reaction.
All experiments were carried out with three biological replicates performed in duplicate. Graphs represent mean and standard error of the mean (S.E.M.) of all biological replicates. For western blots a representative image was selected. Statistical significance of numerical differences was calculated using one-way ANOVA and Holm-Sidak post-hoc comparison between experimental conditions.
To model the interaction between PKA and E1A, we first performed a structural prediction of the amino terminus of E1A by submitting the primary sequence to Phyre 2 [59]. The predicted structure of E1A was subsequently docked onto PKA (PDB ID: 3IM4) using the standard settings profile of ClusPro2.0 [60]. Residues forming an E1A-PKA binding interface within 4 Angstroms were selected for further experimental analysis. All images were generated in the PyMOL Molecular Graphics System, Version 1.8 Schrödinger, LLC. Additional in silico comparisons of HAdV-5 and HAdV-4 E1A were conducted using Clustal Omega [61] and the UCL Department of Computer Science’s PSI-PRED protein sequence analysis workbench [62].
|
10.1371/journal.pgen.1006048 | The AP-2 Transcription Factor APTF-2 Is Required for Neuroblast and Epidermal Morphogenesis in Caenorhabditis elegans Embryogenesis | The evolutionarily conserved family of AP-2 transcription factors (TF) regulates proliferation, differentiation, and apoptosis. Mutations in human AP-2 TF have been linked with bronchio-occular-facial syndrome and Char Syndrome, congenital birth defects characterized by craniofacial deformities and patent ductus arteriosus, respectively. How mutations in AP-2 TF cause the disease phenotypes is not well understood. Here, we characterize the aptf-2(qm27) allele in Caenorhabditis elegans, which carries a point mutation in the conserved DNA binding region of AP-2 TF. We show that compromised APTF-2 activity leads to defects in dorsal intercalation, aberrant ventral enclosure and elongation defects, ultimately culminating in the formation of morphologically deformed larvae or complete arrest during epidermal morphogenesis. Using cell lineaging, we demonstrate that APTF-2 regulates the timing of cell division, primarily in ABarp, D and C cell lineages to control the number of neuroblasts, muscle and epidermal cells. Live imaging revealed nuclear enrichment of APTF-2 in lineages affected by the qm27 mutation preceding the relevant morphogenetic events. Finally, we found that another AP-2 TF, APTF-4, is also essential for epidermal morphogenesis, in a similar yet independent manner. Thus, our study provides novel insight on the cellular-level functions of an AP-2 transcription factor in development.
| Mutations in the evolutionarily conserved family of AP-2 transcription factors are associated with multiple birth defects in Char syndrome and Brancio-oculo-facial syndrome. These DNA-binding proteins are known to regulate the proliferation, differentiation and death of specific cells during embryonic development but how point mutations in the AP-2 DNA-binding domain lead to these diseases during development is currently unknown. We have identified a mutation in one of the AP-2 orthologs of the nematode Caenorhabditis elegans, APTF-2, which falls in the same mutation hotspot as in human Char syndrome and Brancio-oculo-facial syndrome patients. Compromised APTF-2 activity in C. elegans results in embryonic lethality and embryos that survive to hatching displays body morphological defects, reminiscent of the aforementioned human diseases. Using time-lapse microscopy, we found that misregulation of cell division in the skin, muscle and neuronal cell lineages is the primary cause of developmental arrest. Our study provides insight into the regulation of cell division timing by AP-2 transcription factors and provides a model to study human diseases associated with AP-2 mutations.
| The AP-2 family of transcription factors is associated with proper development of mammals by maintaining a balance between cell proliferation and cell death [1, 2]. Five members of the AP-2 family have been identified in vertebrates: AP-2α, AP-2β, AP-2γ AP-2δ and AP-2ε [3–8]. All AP-2 transcription factors have a central basic region followed by a highly conserved helix-span-helix (HSH) motif at the carboxyl terminus [9]. The HSH is essential for dimerization and together with the adjacent basic region achieves a sequence-specific DNA binding function [10]. The less conserved proline- and glutamine-rich region at the amino terminus is required for transcription activation [11]. AP-2 transcription factors bind primarily to the palindromic core sequence 5’-GCCN3GGC-3’ and serve a dual role as transcriptional activators or repressors [1].
AP-2 knockout mice display a wide spectrum of anomalies in early development such as craniofacial, neural tube and body wall defects, and polycystic kidney disease associated with uncontrolled apoptosis [12–14]. The phenotypic defects correspond to the diverse and overlapping expression patterns of murine AP-2 family genes in the neural crest cells, forebrain, facial and limb mesenchyme, and various types of epithelial cells [4, 5, 15, 16]. In humans, mutations in TF AP-2-alpha (TFAP2A) have been associated with branchio-oculo-facial syndrome (BOFS), a congenital birth defect characterized by craniofacial abnormalities, skin and eye defects as well as hearing problems [17]. Char Syndrome, a congenital disease characterized by patent ductus arteriosus and facial and hand anomalies, was linked to mutations in TF AP-2-beta (TFAP2B) [18]. Multiple point mutations and deletions in BOFS and Char Syndrome patients have been mapped to the conserved basic region of the DNA binding domain in AP-2α and AP-2β [17–20]. However, the molecular mechanisms by which these mutations manifest in the disease symptoms are not well understood.
In C. elegans, there are four AP-2 TF family members: APTF-1, APTF-2, APTF-3 and APTF-4. APTF-1 functions in the GABAergic neuron RIS to induce sleep-like quiescence in C. elegans [21]. Other APTF members have not yet been studied. Using whole genome sequencing, we identified a mutant allele which gave rise to a single amino acid change in the basic region of APTF-2. Here, we describe the role of APTF-2 during C. elegans embryonic development, specifically during epidermal morphogenesis that involves the formation of a single epithelial layer that envelops the animal. We found APTF-2 is important for epithelial dorsal intercalation and ventral enclosure and mutation of aptf-2 results in larva with body morphology defects as well as embryonic lethality. Cell lineaging revealed misregulation of cell division timing, possibly leading to the phenotypic defects. Thus, C. elegans could serve as a model to study molecular and cellular consequences of mutations in the family of AP-2 TF analogous to those mutations in human AP-2 TF underlying BOFS and Char syndrome diseases.
In a genetic screen for maternal-effect mutations that have an impact on C. elegans development Hekimi et. al. isolated mal-1(qm27) as a mutation that causes extensive embryonic and larval lethality, with surviving homozygous mutants displaying morphological defects characterized by dorsal protrusions on the head and/or shortened body length [22] (Figs 1A and S1A and Tables 1 and S1). Genetic mapping predicted the approximate location of mal-1(qm27) on chromosome IV [22], but the molecular identity of the mal-1 gene has remained unknown. Whole genome sequencing of a mal-1(qm27) strain identified a missense mutation in aptf-2, one of four AP-2-like transcription factors in C. elegans. This mutation changes a highly conserved glutamic acid residue within the basic region of the DNA binding domain into a lysine residue (Fig 1B and 1C). Previous findings have indicated the basic region as a mutation hotspot for BOFS and Char Syndrome [17–20]. We also analysed gk902, a deletion allele of aptf-2 generated by the International C. elegans Gene Knockout Consortium. Similar to qm27, gk902 worms also displayed maternal effect embryonic lethality with 99 ± 0.5% of embryos not hatching and the few hatching larva displaying head and/or tail morphological defects and arresting as larva (Table 1, S1 Table). The gk902 and qm27 alleles failed to complement each other, as the progeny of trans-heterozygote aptf-2(gk902)/aptf-2(qm27) had a level of embryonic lethality in between homozygote aptf-2(qm27) and homozygote aptf-2(gk902), consistent with them being mutations in the same gene (Fig 1D and S2 Table). Moreover, expression of APTF-2::GFP from an integrated array driven by the aptf-2 promoter, completely rescued the embryonic lethality in both aptf-2(gk902) and aptf-2(qm27) strains (Fig 1D, S2 Table, S1 Movie), confirming the embryonic lethality in these strains is due to the mutations in aptf-2. Consistent with Hekimi et. al. [22] we found that qm27 homozygous progeny of +/qm27 worms are phenotypically normal, indicating maternal rescue.
To characterize the developmental defects in aptf-2(qm27) embryos leading to their lethality we used 4D differential interference contrast (DIC) microscopy to follow isolated embryos positioned with either their dorsal or ventral side facing the microscope objective. We identified three major defects, all related to epidermal morphogenesis: failure in dorsal epidermal cell intercalation, failure of ventral epidermal cell enclosure, and arrest during elongation (Table 2). A small percentage of embryos also exhibited leakage of cells out of the body of the embryo during elongation (Table 2). The exact cause for elongation arrest is not easily discerned, but we noted that one third of the ventrally-oriented embryos that arrested during elongation had previous ventral enclosure defects and nearly all of the dorsally-oriented embryos that arrested in elongation displayed earlier defects in dorsal intercalation.
We confirmed the phenotypes observed in DIC microscopy by imaging aptf-2(qm27) embryos expressing fluorescently-tagged cell-cell junction markers E-cadherin/HMR-1 and alpha-catenin/HMP-1. As shown in Fig 2, S2 and S3 Movies, these markers confirmed the failure of epidermal cells to dorsally intercalate (Fig 2A), ventrally migrate (Fig 2B), and elongate the embryo (Fig 2C). Previous studies have shown that ventral enclosure defects are often preceded by failure of ventral neuroblasts to seal the cleft at the end of gastrulation. We imaged gastrulation cleft closure in wild-type and aptf-2(qm27) embryos by DIC and by expression of the neuroblast marker KAL-1::GFP and found that the ventral cleft in the mutant embryos was larger to begin with, took up to four times the amount of time to close and in some cases did not completely close before the onset of epidermal ventral enclosure (S1 Fig).
We next examined the embryonic phenotypes of the null mutant aptf-2(gk902) by DIC microscopy. We found 60% of the embryos died prior to epidermal morphogenesis, and approximately half of these early embryonic deaths were associated with the appearance of many ectopic apoptotic cells (Fig 3A and Table 3). The remaining 40% of embryos that made it to epidermal morphogenesis all exhibited defects in dorsal intercalation, a quarter of them had ventral enclosure defects, and they all arrested during elongation (Fig 3B, 3C and 3D and Table 3). The massive apoptosis phenotype was completely rescued by the expression of APTF-2::GFP (S2 Table), suggesting that it is a result of the complete loss of APTF-2 function. However, this phenotype was never observed in the partial loss of function allele qm27. Neither was it observed in aptf-2(RNAi) nor following injection of aptf-2 dsRNA into aptf-2(qm27).
Using TargetOrtho [23], a phylogenetic footprinting tool to identify transcription factor targets, we identified within the C. elegans genome 1631 putative AP-2 TF binding sites in the 3KB upstream promoter region of 872 genes (S1 Text). Protein domain analysis of these genes revealed enrichment in F-box, Homeobox, EF-hand, SET and CUB domain proteins, as well as others, and gene onthology analysis of biological processes showed enrichment in genes associated with embryonic development, tissue morphogenesis, locomotion, regulation of growth rate, and reproduction, among others (see S1 Text for full list). Among the putative AP-2 TF regulated genes classified as associated with epithelium development our attention was caught by die-1. The zinc finger transcription regulator DIE-1 is autonomously required in the posterior dorsal hypodermis for intercalation, for morphogenesis in other embryonic tissues, and for normal postembryonic growth and vulval development [24, 25]. Given the defects we observed in epidermal morphogenesis we tested whether the expression of die-1 is altered in aptf-2 mutants. Indeed, we found that two out of seven aptf-2(qm27) embryos showed aberrant localization of DIE-1::GFP. Furthermore, we measured a 22.5% reduction in mean intensity of DIE-1::GFP in the nucleus of mutant embryos with proper nuclear localization (2340 a.u. ± 75.93, n = 5) compared to wild type (3018 a.u. ± 63.68, n = 4) (Fig 4).
Analyzing the DIC movies of embryonic development we found that in addition to the various defects in epidermal morphogenesis the aptf-2(qm27) embryos developed more slowly than wild-type embryos at the same temperature. To quantify the delay and find out whether there is a particular stage in development that is slower or if all of embryogenesis is inherently slower we chose easy-to-recognize developmental milestones in dorsally or ventrally oriented embryos and measured the time it took for an embryo to progress from one milestone to the next (S3A and S3B Table). We also measured the same developmental times in aptf-2(qm27) embryos stably expressing wild-type APTF-2::GFP. The results, graphically presented in Fig 5, show that all stages of development are slower, to varying degrees, in aptf-2(qm27) embryos, and the developmental timing is mostly rescued in embryos ectopically expressing APTF-2::GFP. Specifically, ventral cleft closure is three times slower and elongation to 2 fold stage is one and a half times slower, while early development until Ea/Ep ingression is only slightly slower.
To better understand the developmental defects in aptf-2(qm27) embryos we performed cell lineage analysis by following a nuclear marker, HIS-72::GFP, using 4D fluorescence microscopy. The cell division patterns in wild-type and aptf-2(qm27) embryos were captured, then analysed and edited using StarryNite and AceTree, respectively (n = 2 for wild-type and n = 6 for aptf-2(qm27) embryos). Cell division defects were consistently detected in three lineages: ABarp, C and D (Fig 6). The color markings drawn on the wild-type lineage trees illustrate the frequency of defects that occurred in the six aptf-2(qm27) mutant embryos analysed. Strikingly, failure in aptf-2(qm27) cell division occurs mostly in three lineages: ABarp, C and D with the Caaaa division absent in all six aptf-2(qm27) embryos analysed. The missing divisions resulted in the absence of epidermal seam cells and neuroblasts in the AB lineage and the absence of epidermal cells from the main body syncytium (hyp7), body wall muscle cells in the C and most of the D lineage (Fig 6 and S4 Table). In other cell lineages cell divisions appeared to be normal, except for an occasional division absent in the ABala or MSa lineages (S2–S14 Figs).
We used embryos co-expressing HIS::mCherry and the translational fusion of APTF-2::GFP driven by the aptf-2 promoter to follow the subcellular localization of APTF-2 in specific cells during embryogenesis (Fig 6A). We found that in most cells APTF-2 is found uniformly in the nucleus and the cytoplasm. However, in certain cells at specific times during development, APTF-2 was enriched within the nucleus. Based on the lineaging of two embryos for 210 minutes we found significant nuclear enrichment of the APTF-2::GFP signal in neuroblasts and epidermal cells in AB lineage during ventral cleft closure and in epidermal cells in C lineage preceding dorsal intercalation (Fig 7B and S15 Fig). However, there does not appear to be a strong correlation between nuclear enrichment of APTF-2 and defects in cell division. While a high degree of nuclear enrichment was found in the C and ABarp lineages, in which the absence of cell division in aptf-2(qm27) embryos occured in 6/6 embryos, a high degree of nuclear enrichment was also found in ABpra and ABpla lineages that did not experience any defects in cell division. Similarly, in the D lineage, which did not show much nuclear enrichment, the failure in cell division was frequently observed.
In light of the specific nuclear enrichment of APTF-2 in the cell lineages where we observed defects in cell division timing in the aptf-2(qm27) hypomorph, we wondered whether the mutant protein has a defect in nuclear enrichment. To test this possibility we introduced into the APTF-2::GFP construct the same point mutation present in the qm27 allele. As shown in Fig 8A, the mutant protein had no problem in becoming enriched in neuroblast nuclei during ventral cleft closure. To the contrary, once the mutant APTF-2 entered the nucleus, it appeared to remain enriched in the nucleus for longer than the wild-type protein. This raised the question whether abnormal nuclear retention of APTF-2 could explain the defects in aptf-2(qm27). To address this question we engineered an APTF-2::GFP flanked by two nuclear localization signals from SV40 and EGL-13 and expressed it in aptf-2(qm27) and aptf-2(gk902) embryos. In contrast with wild-type APTF-2::GFP, APTF-2::NLS::GFP was continuously and exclusively nuclear in all cells in which it was expressed (Fig 8B). Importantly, expression of the constitutively nuclear APTF-2 was able to significantly rescue embryonic lethality of aptf-2(qm27) and aptf-2(gk902) (Fig 8C and S5 Table). These findings suggest that the aberrant nuclear localization of mutated APTF-2 does not explain its functional defects.
The worm genome encodes for four AP2-like transcription factors (S16 Fig). APTF-1 is expressed in only five head interneurons and is required for a sleep-active neuron to induce lethargus in molting larvae [21]. To test whether APTF-3 and/or APTF-4 may play a role in embryonic development we depleted zygotic and maternal products of the genes by RNAi and tested for embryonic lethality in the progeny. Knockdown of aptf-3 did not result in any embryonic lethality. In contrast, knockdown of aptf-4 resulted in 26 ± 3% embryonic lethality. Moreover, hatched aptf-4(RNAi) larvae often exhibited body morphology defects reminiscent of the defects observed in aptf-2 mutants (Fig 9A). The deletion allele aptf-4(gk582) resulted in 100% larval arrest of homozygous worms, precluding analysis of embryonic phenotypes. Closer examination of embryonic development by 4D DIC and fluorescence microscopy revealed defects in dorsal intercalation, ventral cleft closure, and elongation (Fig 9B–9D, S4 Movie). To test whether APTF-2 and APTF-4 work independently or cooperatively in the regulation of epidermal morphogenesis we tested the combined effect of aptf-4 KD in the background of aptf-2(qm27). We found the embryonic lethality upon co-depletion of aptf-2 and aptf-4 to be higher than the sum of the lethality of single depletions, suggesting synergy between aptf-2 and aptf-4 (Fig 8E and S6 Table). As AP-2 transcription factors are believed to function as heterodimers in some cases [26], one possibility is that aptf-2 and aptf-4 work cooperatively. 4D DIC movie analysis revealed that 100% of the dorsally oriented dual-depleted embryos had dorsal intercalation defects and arrested during elongation and 57% of the ventrally oriented dual-depleted embryos displayed ventral cleft closure defects and 100% of them arrested in elongation (S7 Table). We used expression data for APTF-4 from the EPIC dataset (http://epic.gs.washington.edu/) to compare the nuclear expression pattern between APTF-2::GFP and APTF-4::GFP (S17 Fig). Both APTF-2::GFP and APTF-4::GFP showed similar nuclear enrichment in the AB and C lineages, consistent with their cooperativity in embryogenesis.
Vertebrates and C. elegans AP-2 TF genes share high sequence similarities in their functional domains, although the duplications leading to four family members appear to have occurred independently in C. elegans and in vertebrates (S16 Fig). In this study, we report that partial loss of aptf-2 or aptf-4 resulted in body morphological defects. Patients with BOFS suffer from skin defects while complications associated with Char Syndrome result from derangement of neural-crest-cell derivatives [17, 18]. Our findings from the characterization of aptf-2(qm27) share similarity with the pathological manifestation of BOFS and Char Syndrome patients in epidermal and neuronal tissues. The mutation in the aptf-2(qm27) allele lies in the basic region of the DNA binding domain, a region that was defined as a mutation hotspot for BOFS and Char Syndorme in the human TFAP2A and TFAP2B genes [17–20]. At least 24 mutations in the basic region have been identified for BOFS and five for Char Syndrome [18, 20, 27]. It is challenging to determine the genotype-phenotype relationship in BOFS and Char Syndrome patients due to the small sample size and the large spectrum of mutations affecting TFAP2A and TFAP2B. With recent advances in site-targeted mutagenesis in the C. elegans genome, it is an exciting possibility to generate worm strains carrying mutations of conserved residues in BOFS and Char Syndrome.
The aptf-2(gk902) allele results in a frame shift, generating a null allele. The massive apoptotic phenotype observed following a complete loss of APTF-2 in aptf-2(gk902) embryos is drastically different from the epidermal morphogenesis defects observed when APTF-2 activity is partially compromised as with the aptf-2(qm27) allele. This suggests different thresholds of AP-2 transcriptional activity are required for different cellular functions. Interestingly, in Char Syndrome patients, hypomorphic mutations in TFAP2B result in congenital heart defect, whereas a complete deletion of the mouse ortholog, AP-2β, leads to polycystic kidney disease due to excessive apoptosis of renal epithelial cells [14, 18].
In murine models, depletion of AP-2γ resulted in defective epidermal development due to delayed expression of epidermal differentiation genes [28]. This is consistent with our observation that aptf-2 mutants showed epidermal morphogenesis defects. Neural crest defects in mouse, zebrafish and Xenopus embryos have been attributed to loss of AP-2 transcription factors [1, 29, 30], parallel to the neuroblast migration defect we observed in the C. elegans embryo. Earlier expression studies of AP-2 transcription factors were largely conducted in mice, Drosophila and Xenopus by observing in-situ hybridization and staining patterns [5, 16, 31–33]. Our work in the live C. elegans embryo provided spatio-temporal information at a resolution not described previously. We observed APTF-2::GFP to be enriched in the nuclei of neuroblasts and epidermal cells during ventral enclosure and dorsal intercalation respectively, lack of which (in the case of the mutant) resulted in aberrant cell division in the epidermal and neuroblast lineages. Thus, our work identified lineage-specific regulation of cell division timing by APTF-2. Similar mechanisms could be at play in mammals. Interestingly, we observed that nuclear enrichment of APTF-2 does not always correlate with regulation of cell division, as in the case of D, suggesting that a lower level of nuclear APTF-2 may be required for the division in this lineage. In contrast, nuclear APTF-2 enrichment was observed in ABpra and ABpla and yet an absence of cell division was not been observed in these lineages in aptf-2(qm27) embryos, indicating that either a stronger APTF-2 depletion is required to see cell division defects or APTF-2 plays a different role in these two lineages.
Although various members of the vertebrate AP-2 transcription family have been shown to have overlapping expression patterns, knockout studies in mice revealed specific and localized phenotypic defects. For example, Moser et. al. showed that the AP-2α and AP-2β expression in mouse embryos overlap significantly, [16], but the single knockout models of each gene did not share any phenotypic defects, suggesting non-redundant roles of the two genes [14]. In contrast to the vertebrate system, our results showed both similar phenotypes and similar expression pattern, mostly in AB and C lineages of aptf-2 and aptf-4 in the worm. The fact that their effect is synergistic suggests they may partially function through the same pathway.
For wild-type APTF-2::GFP, expression in the majority of cells was evenly distributed between the nucleus and cytoplasm and was enriched in the nucleus of neuroblasts during ventral cleft closure and in epidermal cells preceding dorsal intercalation. It is possible that APTF-2 functions to regulate gene expression at a basal level, while enrichment in the nucleus of specified cells during epidermal morphogenesis upregulates genes required for proliferation of the neuroblasts and epidermal cells. This would be consistent with observations in Drosophila, where different levels of AP-2 have been shown to result in a variety of morphological defects [32].
AP-2 transcription factors are known to play a dual role as transcription activators and repressors [33]. Pfisterer et. al. identified multiple genes repressed by AP-2α known to induce apoptosis and retards proliferation [34]. There has also been evidence in Xenopus epidermal development regarding the importance of AP-2 TF in promoting the expression of epidermal specific genes [31]. We used TargetOrtho to identify putative APTF-2 targets. Among the candidates, we tested die-1, a well known regulator of epidermal dorsal intercalation, and observed the reduction of DIE-1 nuclear signal in aptf-2(qm27) embryos, suggesting that DIE-1 is likely a target of APTF-2. Future work must determine APTF-2 target genes in neuroblasts and epidermal cells in order to further elucidate its function during morphogenesis.
In conclusion, we have characterized a hypomorphic mutant of C. elegans APTF-2 and have shown it to share genetic and anatomical similarities with human Char Syndrome and Bronchio-occular-facial Syndrome. We propose mutations in C. elegans AP-2 TF genes can serve as disease models to study the cellular mechanisms and tissue dynamics that lead from mutant genotype to disease phenotype.
Strains were maintained at 20°C under standard conditions [35]. Wild-type Bristol strain N2 was used as a control. The aptf-2(qm27) IV line was retrieved in an EMS screen conducted by Hekimi et al. [22] and aptf-2(gk902) was generated by the C. elegans Reverse Genetics Core Facility at the University of British Columbia and was maintained as heterozygotes using the nT1[qIs51] (IV;V) balancer. For analysis using GFP reporters, F2 progeny exhibiting aptf-2 phenotypes and carrying the markers were selected from crosses between aptf-2(qm27) and the following strains: FT250 xnIs96 [pJN455(hmr-1p::hmr-1::GFP::unc-54 3'UTR) + unc-119(+)] [36], SU265 jcIs17[hmp-1p::hmp-1::gfp, dlg-1p::dlg-1::dsRed, rol-6p::rol-6(su1006)] [37], OH904 otIs33[kal-1p::gfp] [38], RW10029 zuIs178 [his-72(1kb 5' UTR)::his-72::SRPVAT::GFP::his-72 (1KB 3' UTR) + 5.7 kb XbaI—HindIII unc-119(+)]. stIs10024 [pie-1::H2B::GFP::pie-1 3' UTR + unc-119(+)] (a gift from Waterston lab) and JIM119 zuIs178 [his-72(1kb 5' UTR)::his-72::SRPVAT::mCherry::his-72 (1KB 3' UTR) + 5.7 kb XbaI—HindIII unc-119(+)]. stIs10024 [pie-1::H2B::mCherry::pie-1 3' UTR + unc-119(+)] (a gift from Waterston lab). die-1::gfp reporter strain was a gift from Hardin lab [25].
To construct plasmids containing wild-type or mutated aptf-2, the aptf-2 promoter (2 kb sequence upstream of aptf-2 start codon) followed by the aptf-2 coding sequence were amplified from N2 and aptf-2(qm27) animals, respectively and inserted into XbaI and AgeI sites upstream of gfp in the original pPD95.75 vector. The wild-type aptf-2-containing plasmid was injected into the gonad of aptf-2(qm27) hermaphrodite animals to examine its potency in rescuing aptf-2(qm27) phenotypes, whereas the plasmid containing mutated aptf-2 was injected into N2. This resulted in the following transgenes: msnEx15 [aptf-2p::aptf-2::gfp; rol-6(su1006)]; aptf-2(qm27) and msnEx239 [aptf-2p::mutated aptf-2::gfp; rol-6(su1006)]. Ten L4 larvae expressing wild-type aptf-2 were subjected to a UV source (BioRad) for 15 seconds to integrate the extrachromosomal array into the genome. Three hundred F2 worms were then singled and incubated for three weeks and subsequently examined for expression and embryonic lethality. Those expressing the transgene and giving rise to 100% viable progeny were selected and outcrossed. The resulting strain, RZB104 (aptf-2(qm27); msnIn104[aptf-2p::aptf-2::gfp; rol-6(su1006)]), was used throughout this study.
To construct aptf-2 tagged with a nuclear localization signal (NLS), the amplified 4.3 kb genomic sequence containing the aptf-2 promoter and the coding region was inserted into XbaI and XmaI sites in pNL74.4 [39], a modified pPD95.75 containing SV40 and EGL-13 NLS flanking the N and the C terminal of the gfp sequence, respectively. The plasmid was injected into the gonad of N2 hermaphrodites and resulted in transgene msnEx103 [aptf-2p::aptf-2-NLS::gfp; rol-6(su1006)]). The transgenic animals were then crossed with aptf-2(qm27) or aptf-2(gk902) to assess the ability of NLS-tagged APTF-2 to rescue the aptf-2 mutants.
Microinjection was performed as described by Mello and Fire [40]. Injection mix included 100 μg/μl salmon sperm DNA digested with PvuII, 20 μg/μl rol-6(su1006) digested with SbfI and 5–10 μg/μl each construct digested with SbfI.
Genomic DNA was extracted from mal-1(qm27) mutant worms using standard method and subjected to whole genome sequencing using Illumina platform and annotated using MAQGene [41]. The whole genome sequencing and its annotation were performed by Hobert lab (Columbia University). Candidate genes altered in mal-1(qm27) were narrowed down using genetic mapping results done by Hekimi et al. [22]. Point mutation in aptf-2 gene was confirmed by amplification of aptf-2 gene in aptf-2(qm27) mutant worms, subcloning into pJET vector (Thermo Scientific) and followed by conventional sequencing (First Base).
For complementation assay, aptf-2(gk902)/nT1[qIs51] males was crossed with aptf-2(qm27) hermaphrodites. Non-GFP F1 animals were singled and incubated to lay embryos for 24 hours. The F1 animals were genotyped for the gk902 deletion and only the cross progeny between qm27 and gk902 alleles was scored for embryonic lethality of their F2s.
For brood size analysis, ten L4 larvae of wild-type, aptf-2(qm27) and aptf-2(gk902) were singled and incubated for 24 hours. Each animal was shifted to a new plate every day for 5 consecutive days to the point that no more embryos were laid. The total number of embryos laid was determined as the brood size. The number of hatched animals was calculated and used to determine the percentage of embryonic lethality. Larvae that did not grow into adult in 48–92 hours after hatching were considered as being arrested. aptf-2(qm27) and aptf-2(gk902) larvae of any stage were subjected to phenotypic analysis to determine the presence and the position of the morphological defects.
Besides wild-type, aptf-2(qm27) and aptf-2(gk902) animals whose embryonic lethality was determined as described above, the embryonic lethality of the remaining strains were determined as follows: ten to fifteen gravid hermaphrodites were placed on the plate and incubated at 20°C for several hours to lay more than 100 embryos. Hermaphrodites were then removed and the number of embryos laid was counted. Twenty-four hours later, the number of larvae hatched was determined. Each experiment was repeated at least five times.
Larvae or embryos collected from gravid hermaphrodite were mounted onto 3% agarose-padded glass slide, closed with a coverslip and sealed with wax. DIC images shown in Figs 1A, 3A, 3B, 3C, 8A, 8C, 8E and S1A were captured using a Nikon Ti Eclipse widefield microscope equipped with DIC 1.40NA oil condenser and a charged-coupled device camera Cool Snap HQ2 (Photometrics). All other images and movies were acquired using a spinning disk confocal system composed of a Nikon Ti Eclipse microscope with a CSU-X1 spinning disk confocal head (Yokogawa), DPSS-Laser (Roper Scientific) at 491 and 568 nm excitation wavelengths and an Evolve Rapid-Cal electron multiplying charged-coupled device camera (Photometrics). For both microscopes, Metamorph software (Molecular Devices) was used to control acquisition. Projected images were created using Fiji. All imaging was done at 20°C in an environmental chamber encompassing the microscope stage heated by a JCS temperature controller (Shinko Technos Co, Japan) within a microscope room kept at 18°C by a CITEC precision air conditioning unit.
aptf-4 dsRNA was synthesized as described [42] and injected into the gonad of twenty wild-type or aptf-2(qm27) L4 larvae. Each animal was singled into a separate plate and its embryonic lethality was examined 24 hours post injection.
Protein sequence of the AP-2 transcription factor family members in the following metazoan species were aligned using Constraint-based Multiple Protein Alignment Tool (COBALT) [43]: A. queenslandica (sponge), T. adhaerens (Placozoa), C. elegans (nematode), N. vectensis (sea anemone), D. melanogaster (fruit fly), S. purpuratus (sea urchin), C. intestinalis (tunicate), B. floridae (lancelete), D. rerio (fish), X. tropicalis (frog), G. gallus (chicken), H. sapiens (human). The resulting alignment was used to build and visualize a phylogenetic tree (neighbor-joining method) using Geneious (Biomatters Ltd.). Illustration of the gene and protein architecture was drawn using Illustrator for Biological Sequences [44].
AP-2 has been shown to bind to the palindromic consensus sequence 5'-GCCN3GGC-3', as well as the binding motif 5'-GCCN3/4GGG-3' [2]. We used either the 9bp or 10bp motif as an input for TargetOrtho [23]. From the program output we selected only putative targets that are conserved in at least 4 Caenoharbditis species, and are located within the 3 kb region upstream of the start codon. Functional annotation was performed using DAVID Bioinformatics Resources 6.7 [45, 46] and the threshold we used for enrichment was an EASE score equal or smaller than 0.05.
For cell lineaging, six aptf-2(qm27) embryos expressing nuclear signal of GFP::HIS-72 and two embryos co-expressing APTF-2::GFP and mCherry::HIS-72 were analysed for at least 270 minutes according to the protocol described in [47–49]. The lineage tree was built using AceTree [50] and compared to that of wild-type. To visualize the temporal enrichment of the nuclear APTF-2::GFP signal during embryogenesis, the minimum/ maximum threshold values were set to display the 75% highest signal. All movies used for lineaging in this paper can be downloaded from http://epic2.gs.washington.edu/Epic2.
Statistical analyses were done using Prism 6 (GraphPad Software, La Jolla, CA). Two-tailed Student’s t-test was applied to compare the values.
|
10.1371/journal.pgen.1006262 | dBRWD3 Regulates Tissue Overgrowth and Ectopic Gene Expression Caused by Polycomb Group Mutations | To maintain a particular cell fate, a unique set of genes should be expressed while another set is repressed. One way to repress gene expression is through Polycomb group (PcG) proteins that compact chromatin into a silent configuration. In addition to cell fate maintenance, PcG proteins also maintain normal cell physiology, for example cell cycle. In the absence of PcG, ectopic activation of the PcG-repressed genes leads to developmental defects and malignant tumors. Little is known about the molecular nature of ectopic gene expression; especially what differentiates expression of a given gene in the orthotopic tissue (orthotopic expression) and the ectopic expression of the same gene due to PcG mutations. Here we present that ectopic gene expression in PcG mutant cells specifically requires dBRWD3, a negative regulator of HIRA/Yemanuclein (YEM)-mediated histone variant H3.3 deposition. dBRWD3 mutations suppress both the ectopic gene expression and aberrant tissue overgrowth in PcG mutants through a YEM-dependent mechanism. Our findings identified dBRWD3 as a critical regulator that is uniquely required for ectopic gene expression and aberrant tissue overgrowth caused by PcG mutations.
| Genetic information is stored in our genomic DNA, and different cells retrieve distinct sets of information from our genome. While it is important to activate genomic regions encoding proteins that are essential for a given cell type, it is equally important to silence genomic regions encoding proteins that are potentially harmful to this type of cells. One of the gene silencing mechanisms frequently used during and after development is mediated by the Polycomb group (PcG) proteins. If this guardian function does not perform correctly due to PcG mutations, genes that are normally silenced—such as oncogenes—are expressed aberrantly. Due to the activation of oncogenes and the loss of other PcG functions, PcG mutant cells often begin to display hallmarks of cancer, such as proliferating beyond control, acquiring stem-cell-like properties, and migrating to distant sites. If the transcriptional mechanisms underlying aberrant gene expression in PcG-mutant cancer cells differ from gene expression in normal cells, we may be able to selectively inhibit the growth of cancer cells without affecting their normal counterparts. Here we show that the difference between these two types of gene expression resides in their sensitivity to dBRWD3, a negative regulator of the deposition of histone H3 variant H3.3. Our results indicate that the inactivation of dBRWD3 or promotion of H3.3 deposition may selectively suppress ectopic gene expression and tumorigenesis driven by mutations in PcG.
| The eukaryotic genome is packaged in a macromolecular complex termed chromatin. Chromatin is composed of DNA, RNA, histones, and non-histone proteins. The nucleosome, the basic unit of chromatin, consists of a histone octamer containing two copies of histones (H3, H2A, H2B, and H4) and 147 base pairs of DNA wrapped around the octamer [1]. Variants of H2A, H2B, and H3 differ from the canonical histones by a few amino acids [2]. Moreover, canonical histones are encoded by multiple repeated sequence arrays and expressed during the S-phase, while the variants are encoded by single-copy genes and expressed in the interphase [2,3]. Once the histone variants are incorporated into nucleosomes, they confer distinct physical and biochemical properties to DNA templates and thus regulate DNA replication, repair and gene transcription [3,4]. The deposition of histone variants is mediated by specific chaperone complexes. For example, H3.3 deposition, which often occurs in actively transcribed regions, is mediated by a histone chaperone named histone repressor A (HIRA) and its associated chaperone Yemanuclein (YEM) [5–10]. We previously showed that HIRA/YEM activity is negatively regulated by dBRWD3 (Bromodomain and WD repeat-containing protein 3) [11], which adds a second layer of complex regulation to H3.3 deposition. Dendritic arborization of peripheral neurons and photoreceptor development are disrupted in dBRWD3 mutants. These phenotypes are effectively suppressed by mutations in yem or H3.3, indicating that dBRWD3 functions largely through restricting YEM-dependent H3.3 deposition [11]. However, it remains unknown where in the genome this regulation of H3.3 deposition takes place and how it affects transcription.
Distinct patterns of transcriptional activation and inactivation of the genome contribute to the diversity of cell types in multicellular organisms. To inactivate transcription, Polycomb group (PcG) proteins bind to specific genomic regions and modify histones posttranslationally [12,13]. PcG proteins are grouped into two evolutionarily conserved complexes, PRC1 and PRC2. In Drosophila, the PRC1 complex consists of Polycomb, Posterior sex combs, Sex combs extra (Sce, the Drosophila homolog of human RING1), Polyhomeotic proximal, Polyhomeotic distal with an accessory molecule, and Sex comb on midleg (Scm) [14]. The PRC2 complex is composed of Enhancer of zeste (E(z)), Suppressor of zeste 12, and extra sex combs [13]. Functionally, PRC1 adds a monoubiquitin moiety onto histone H2AK119 (H2AK118 in Drosophila), whereas PRC2 catalyzes the trimethylation of H3K27 (H3K27me3). The combined activities of PRC1 and PRC2 repress transcription by compacting chromatin [15]. PcG proteins may also silence gene expression through a compaction-independent mechanism, such as by blocking transcription initiation [16–18].
Misregulation of transcription within typically inactive genomic regions leads to the disorganization of tissues and organisms [19]. For instance, loss of PcG function causes the ectopic expression of Homeotic (Hox) genes specific to posterior segments and thus disrupts the anterior-to-posterior body plan in embryos [20,21]. In Drosophila, loss of PRC1 leads to ectopic expression of Unpaired 1–3, driving aberrant cell proliferation and tissue overgrowth by activating the JAK-STAT pathway [22]. In humans, loss of PRC1 function has been shown to promote tumorigenesis [23,24]. Reduced expression of the PRC1 subunit CBX7 has been implicated in bladder, breast, colon, glioma, lung, pancreatic, and thyroid carcinomas [25–31], and CBX7 knockout mice develop lung and liver carcinoma [29]. Loss of PRC2 function also causes tumor formation. For example, the tumor-driving H3.3K27M mutation in pediatric diffuse intrinsic pontine gliomas (DIPGs) results in the inactivation of the PRC2 complex, causing ectopic expression of LIN28B, PLAG1, and PLAGL1, and leading to the de-differentiation and hyperproliferation of tumor cells [32–34]. A second mutation in the PRC2 complex genes in patients with neurofibromatosis increases the likelihood of developing malignant peripheral nerve sheath tumors [35,36].
Currently, no therapeutic strategies have been developed for tumorigenesis caused by ectopic gene expression. This is mainly because little is known about how ectopic gene expression is initiated within de-repressed genomic regions, and how it differs from conventional transcription initiation. Here we show that dBRWD3 is specifically required for ectopic gene expression and tissue overgrowth caused by PcG mutations. dBRWD3 sustains PcG mutation-induced ectopic gene transcription by regulating H3.3 deposition, which in turn affects the way RNA polymerase II occupies transcription start sites. Thus, our results suggest that human BRWD3 could be a potential therapeutic target for PcG mutation-driven tumors.
In the process of investigating how dBRWD3 might affect gene expression, we unexpectedly found that the dBRWD3 mutations suppress the lethality of Scm mutants. Similar to other PcG mosaic mutants [22], ScmD1 mosaic mutant flies died in the pupal stage. Interestingly, a significant portion of the mosaic ScmD1, dBRWD3s5349 double mutants survived to the adult stage, so did the mosaic ScmD1, dBRWD3PX2 double mutants (Table 1). To explore the relationship between dBRWD3 and PcG genes, we examined the genetic interaction between dBRWD3 and Posterior sex comb (Psc), another PcG gene. We found that knockdown of Psc was semi-lethal, whereas simultaneous knockdown of Psc and dBRWD3 was fully viable (Table 2). Taken together, these results suggest a role for dBRWD3 as a suppressor of PcG genes.
Since ectopic gene expression underlies many phenotypes of PcG mutations, we then investigated whether the dBRWD3 mutations also suppresses ectopic gene expression. While the second thoracic segment-specific Hox gene, antennapedia (Antp), was repressed in wild-type eye clones (Fig 1A and 1B, arrow), it was ectopically expressed in Scm mutant eye clones located in the posterior region (Fig 1C, arrow). This ectopic Antp expression was dramatically reduced in ScmD1, dBRWD3s5349 or ScmD1, dBRWD3PX2 double-mutant eye clones (Figs 1D, S1A, S1B and S1C). Similarly, Antp is ectopically expressed in Sce1 mutant eye clones (Fig 1E, arrows) but not in dBRWD3s5349, Sce1 double-mutant clones (Figs 1F and S1D). Interestingly, dBRWD3 is dispensable for the orthotopic expression of Antp in wings (Figs 1G, 1H, 1I and S1E). Overall, these observations reveal that dBRWD3 is involved in the ectopic expression of Antp caused by PcG mutations.
To determine whether dBRWD3 suppresses ectopic gene expression other than Antp in the eyes, we knocked down Pc in the central nervous system by Elav-GAL4, reducing the level of Pc mRNA to 5% (S2A Fig). Ubx is ectopically expressed in the Pc-depleted brains but not in the control (Fig 2A, 2B, 2C, 2D, 2F and 2O). In addition, the Pc depleted ventral nerve cord was thinner and more elongated compared to the control (Fig 2E, bracket). We found that both ectopic expression of Ubx and elongation of ventral nerve cords were suppressed by depletion of dBRWD3 (Fig 2G and 2H). On the other hand, orthotopic expression of Ubx in the ventral nerve cord was not affected in the dBRWD3, Pc double knockdown (Fig 2G) or in dBRWD3 knockdown animals (Figs 2I, S2B, S2C and S2D, arrowhead). Thus, ectopic expression of Ubx depends on dBRWD3 whereas orthotopic expression of Ubx does not.
Depleting E(z), which encodes the H3K27 methyltransferase in PRC2, caused ectopic expression of Ubx in brains (Figs 2J, 2L and S2E) and condensation failure in ventral nerve cords (Fig 2K, bracket). The ectopic expression of Ubx and condensation failure of ventral nerve cords were also suppressed by knockdown of dBRWD3 (Fig 2M, 2N and 2P). By contrast, orthotopic expression of Ubx was not affected in dBRWD3, E(z)-doubly depleted ventral nerve cords (S2F, S2G and S2H Fig). Taken together, our data indicates that ectopic Hox gene expression depends on dBRWD3 whereas orthotopic Hox gene expression does not.
In addition to ectopic expression of Hox genes, loss of Ph, Psc, or Pc induces ectopic expression of unpaired (upd) 1–3, and therefore activation of the JAK-STAT pathway that leads to overgrowth of tumor-like tissues [22,37–39]. By RT-qPCR, we detected mild increases of upd1 and upd2 mRNAs (Fig 3A and 3B) and a strong induction of upd3 mRNA (Fig 3C) in the mosaic ScmD1 mutant brain-eye complex. This upregulation of upd1-3 was prevented or significantly weakened in the mosaic ScmD1, dBRWD3s5349 double mutants compared with mosaic ScmD1 mutants (Fig 3A, 3B and 3C). Consistently, our immunofluorescent micrographs showed that Upd3 accumulated in ScmD1 mutant clones adjacent to the morphogenetic furrow (Figs 3D, 3E and S3A, arrows), but not in ScmD1, dBRWD3s5349 double-mutant clones (Figs 3F and S3A) or wild-type clones (Fig 3G). Similarly, we detected accumulation of Upd3 in Sce1 (Figs 3H and S3B, arrows) and SceKO (S4 Fig) mutant clones, but not in the dBRWD3s5349, Sce1 double-mutant clones (Figs 3I and S3B, arrow). On the other hand, we found that orthotopic upd3 expression in the posterior end of the 2nd instar eye disc was not altered in dBRWD3s5349 mutant clones (Fig 3J and 3K, arrows), indicating that the regulation of upd3 by dBRWD3 is specific to ectopic expression.
We also used the 10XSTAT-GFP reporter to determine whether the JAK-STAT pathway, which is activated by Upd1-3, is affected by dBRWD3 in PcG mutant cells [40,41]. In contrast to the weak and uniform expression observed in wild-type antennal discs (Fig 3L and 3M, arrows), the GFP signal was much higher in ScmD1 mutant clones (Fig 3N), likely due to up-regulation of upd1-3. It remained unchanged in ScmD1, dBRWD3s5349 double mutants (Fig 3O, arrow). The STAT activity in the antennal disc was not affected in dBRWD3s5349 single mutants (Fig 3P, arrow). Given these results, we propose that dBRWD3s5349 suppresses ectopic activation of the JAK-STAT pathway caused by ScmD1 or Sce1 mutations.
By ectopically expressing Upd1, Upd2, and Upd3, mutations in PRC1 and PRC2 complexes cause cell autonomous and non-autonomous proliferations [22,37]. Consistently, we found that ScmD1 (Fig 4A) or Sce1 (Fig 4B) mutant clones were larger than wild-type clones (Fig 4C, 4D and 4E). Since Upd3 is a diffusible ligand stimulating non-cell-autonomous proliferation, we found the non-clonal area of mosaic ScmD1 (Fig 4D) or Sce1 (Fig 4E) eye antennal discs were also larger. Overall, mosaic ScmD1 or Sce1 mutant discs were 1.5- (Fig 4D) or 1.8-fold in size (Fig 4E) compared to wild-type respectively. This tissue overgrowth could be suppressed by dBRWD3s5349 (Fig 4D, 4E, 4F and 4G). As a control, the mosaic dBRWD3s5349 alone did not reduce the disc size (Fig 4H and 4I).
Quantitatively, the numbers of clonal, non-clonal, and total mitotic cells marked by phosphorylation of H3S10 (H3S10ph) were increased in mosaic ScmD1 mutant eye-antennal disc (Fig 4J). It was reduced to a wild-type level in mosaic ScmD1, dBRWD3s5349 mutant eye-antennal discs (Fig 4K, 4L and 4M). A similar suppression of proliferation was found in mosaic dBRWD3s5349, Sce1 mutant eye-antennal discs (Fig 4N) as opposes to Sce1 mutants (Fig 4O and 4P). The elongated Pc and E(z) depleted ventral nerve cords and control ventral nerve cord had comparable mitotic indices and undetectable expression of upd1, upd2 and upd3, indicating that the elongation of the ventral nerve cord was not caused by excessive proliferation.
To determine whether the dBRWD3 mutation also suppresses other types of oncogenic tissue overgrowth, we sampled tissue overgrowth caused by the warts (wts) mutation that activates the hippo pathway [42,43]. We found that dBRWD3s5349 did not suppress the expression of the hippo pathway target gene, expanded (ex) (S5A and S5B Fig) and tissue overgrowth (Fig 4Q and 4R). Thus, the dBRWD3 mutation appears to suppress the oncogenic tissue overgrowth specifically related to PcG mutations. To examine the growth-inhibition effect of the dBRWD3 mutation beyond the developmental stage, we generated overgrown eyes and surrounding tissues by knockdown of ph-p (Fig 4S and 4T). In dBRWD3 and ph-p double-knockdown eyes, the tissue overgrowth phenotype was suppressed (Fig 4U and 4V). From these data, we infer that the tissue overgrowth induced by PcG gene depletion requires dBRWD3.
dBRWD3 contains bromodomain I and II (BRDI and BRDII) that were predicted to be acetylated histone-binding domains (S6A Fig) [44]. To investigate the function of these bromodomains, we complemented ScmD1, dBRWD3s5349 double-mutant cells with wild-type dBRWD3, dBRWD3-N1287A, and dBRWD3-N1451A, in which the conserved asparagines in the BC loops of BRDI and BRDII were mutated into alanines. The wild-type dBRWD3-RFP, dBRWD3-N1287A-RFP, and dBRWD3-N1451A-RFP could restore the ectopic upd3 expression in ScmD1, dBRWD3s5349 double-mutant cells (Figs 5A, S6B and S6C). However, when both the BRDI and BRDII were disrupted, the dBRWD3-N1287A, N1451A-RFP (designated as the dBRWD3-2BC-RFP) could not restore the ectopic expression of upd3 (Fig 5B). This failure to complement is not related to expression levels because dBRWD3-2BC-RFP was expressed more than wild-type dBRWD3-RFP (S7 Fig). Therefore, the BRDI and BRDII of dBRWD3 are functionally redundant in supporting ectopic gene expression. We also complemented ScmD1, dBRWD3s5349 double-mutant cells with a HLH motif-deleted, ΔN-dBRWD3, which no longer interacts with the DNA damage binding protein 1 (DDB1) and cannot be recruited to cullin4/DDB1 organized E3 ligase [11]. The ΔN-dBRWD3-RFP also failed to restore the ectopic upd3 expression in ScmD1, dBRWD3s5349 double-mutant cells (Fig 5C). Together, these results suggest that the activities of dBRWD3 binding to acetylated histones and cullin 4/DDB1 organized E3 ligase are essential for maintaining ectopic gene expression.
Previously, we demonstrated that dBRWD3 limits HIRA/YEM-mediated H3.3 deposition [11]. However, it is not clear whether the BRDI, BRDII, and HLH motif of dBRWD3 are important for dBRWD3 regulation of H3.3. When we complemented dBRWD3s5349 mutant cells with ΔN-dBRWD3-RFP or dBRWD3-2BC-RFP, the H3.3 levels in the dBRWD3s5349 mutant cells remained higher than those in wild-type cells (S8A and S8B Fig, arrows). By contrast, dBRWD3-N1287A and dBRWD3-N1451A reduced the H3.3-dendra2 to a normal level (S8C and S8D Fig, arrows), indicating a negative correlation between accumulation of H3.3 and ectopic gene expression. Indeed, the negative correlation was also observed in the dBRWD3 knockdown brains, where the endogenous H3.3 levels were higher than in control brains (S8E Fig). When we co-immuno-stained the mosaic discs with ant-Antp antibody, the ectopic anti-Antp signals were strongly reduced along with accumulated H3.3 in ScmD1, dBRWD3s5349 double-mutant clones (S9B Fig, arrows).
We next examined whether the increased H3.3 deposition suppresses ectopic gene expression. To this end, we introduced a yem mutation to reduce the dBRWD3 mutation-induced H3.3 deposition (S9C Fig) [11]. In ScmD1, dBRWD3s5349, yemGS21861 triple-mutant clones, the ectopic expression of Antp was restored and coincided with the reduction of H3.3 (Figs 5D, S9C and S10A), suggesting that the dBRWD3 mutation suppresses the Scm mutation through a YEM-dependent mechanism. Moreover, upd3 was ectopically expressed in this triple mutant clone (Figs 5E and S10B). Consistently, the size of the ScmD1, dBRWD3s5349, yemGS21861 triple-mutant eye-antennal disc was larger than that in the ScmD1, dBRWD3s5349 double-mutant (Fig 5F). To further substantiate the role for H3.3 in ectopic gene expression, we investigated whether ectopic gene expression could be suppressed by YEM-induced H3.3 deposition (S11A, S11B and S11C Fig) without any mutation in dBRWD3. YEM over-expression effectively suppressed the ectopic expression of Antp (Figs 5G and S10C). Taken together, these data indicate that dBRWD3 supports ectopic gene expression and tissue overgrowth mediated by PcG mutations by limiting HIRA/YEM-mediated H3.3 deposition.
To understand how the dBRWD3 mutation suppresses ectopic gene expression, we investigated whether dBRWD3 is required for the removal of pre-existing H3K27me3 and H2A118ub at Antp and Ubx loci upon depletion of E(z) and Pc. The ChIP-qPCR analysis revealed a reduction of H3K27me3 in the distal region of Antp and at Ubx when E(z) was depleted (Fig 6A and 6B). Similarly, H2A118ub levels at Antp and Ubx were also decreased in the Pc depleted brains (Fig 6C and 6D). When dBRWD3 was depleted by RNAi together with E(z) or Pc, the H3K27me3 or H2A118ub levels at Antp and Ubx loci remained low or became lower (Fig 6A, 6B, 6C and 6D), indicating that knockdown of dBRWD3 promotes or does not affect the removal of the pre-existing H3K27me3 and H2A118ub. Moreover, dBRWD3 is not required for the removal of pre-existing H3K27me3 in the E(z) depleted wings at the global level, as revealed by the equally reduced H3K27me3 immunostaining signals in the E(z) knockdown, and E(z), dBRWD3 double-knockdown wings (S12A, S12B and S12C Fig). H3K27me3 levels were not changed in the Pc knockdown, and Pc, dBRWD3 double-knockdown wings compared with the control (S12D, S12E, S12F Fig). Similarly, dBRWD3 was not required for the removal of pre-existing H2AK118ub in the Sce mutant clones at the global level (S12G and S12H Fig). trithorax (trx) encodes an H3K4 monomethyltransferase [45] and antagonizes PcG activity by binding to PRE sites, the enhancer cis-elements targeted by PcG proteins. In different cellular contexts, ectopic gene expression might or might not depend on trx [46,47]. To investigate the requirement of trx in ectopic expression of Antp in eyes, we generated ScmD1, trxE2 double-mutant eye clones and found that the ectopic expression of Antp was suppressed (S13 Fig), indicating that trx, like dBRWD3, is required for ectopic Antp expression. We next investigated whether dBRWD3 and trx function in a linear pathway or in parallel. We found that over-expression of trx in the eye disc proper, peripodial epithelium of the eye disc, and wing disc was sufficient to induce ectopic expression of Antp or Abd-B (Figs 6E, S14A, S14B and S14D), but not in a dBRWD3 knockdown background (Figs 6F, S14C and S14E). In addition, Trx-induced Ubx ectopic expression in wing discs was strongly suppressed by knockdown of dBRWD3, albeit incompletely (S14F and S14G Fig). It seems that H3.3 deposition underlies the suppression of this ectopic gene expression, since Trx-induced Abd-B ectopic expression was also completely suppressed by YEM over-expression (Fig 6G).
To understand how dBRWD3 affects Trx-induced ectopic gene expression, we used ChIP to determine the levels of H3K4me1 and PolII over Ubx and Abd-B loci. Compared to the control, Trx increased H3K4me1 levels at the enhancer regions of Ubx and Abd-B irrespective of dBRWD3 depletion (Fig 6H), which was later found to have no effects on Trx-induced monomethylation of H3K4 on a global scale (S15A, S15B, S15C, S15D, and S15E Fig). These data indicate that dBRWD3 is epistatic to trx with respect to ectopic gene expression. By contrast, Trx-induced PolII levels at the transcription start sites and 5' ends of Ubx and Abd-B were significantly reduced when dBRWD3 was depleted by RNAi (Fig 7A and 7B), while the PolII levels for orthotopically expressing Antp were not affected (Fig 7C). When trx is overexpressed in the wing imaginal discs, RNA PolII increased on the Antp promoter, which is likely contributed by the purely orthotopic Antp expression and the Trx-induced ectopic Antp expression. The additional knockdown of dBRWD3 restored the PolII occupancy to a level similar to the orthotopic Antp expression control (S16 Fig), indicating that knockdown of dBRWD3 suppressed only the Trx-induced increase of PolII occupancy but not the PolII occupancy of orthotopically expressing Antp. Similarly, PolII phospho-CTD Ser5 levels around the transcription start sites of Ubx and Abd-B were reduced upon knockdown of dBRWD3 (Fig 7D and 7E). We also detected higher levels of H3K4me3, a marker for active chromatins, at the transcription start sites and 5' ends of Ubx and Abd-B upon trx over-expression in a dBRWD3-dependent manner (Fig 7F and 7G). Nevertheless, the levels of H3K4me3 at the orthotopically expressed Antp were not sensitive to dBRWD3 depletion (Fig 7H). These observations suggest that, in ectopic gene expression, dBRWD3 is required for the activation of chromatin specifically at transcription start sites but not in enhancer regions.
To examine whether the reduction of PolII and H3K4me3 are directly caused by H3.3 deposition, we examined dBRWD3 and H3.3 levels across Ubx and Abd-B loci. ChIP revealed that dBRWD3 was present at these regions and Trx over-expression moderately reduced the occupancy of dBRWD3 (Fig 8A and 8B). Although dBRWD3 was recruited to the promoters, 5' and 3' regions of these ectopically expressed loci, it predominantly reduced H3.3 levels at the transcription start sites (Fig 8C and 8D). These results imply that dBRWD3 maintains PolII levels of ectopically expressed genes by limiting H3.3 deposition at the transcription start sites. Moreover, dBRWD3 was also present at Antp locus (Fig 8E) and limited H3.3 levels at the promoters and the transcription start site of Antp (Fig 8F). Nevertheless, PolII and H3K4me3 levels at Antp were not affected by knockdown of dBRWD3 (Fig 7C and 7H), indicating that the sensitivity toward H3.3 but not H3.3 levels at the promoter and transcription start site per se distinguishes ectopic gene expression from orthotopic gene expression.
Since in ectopic gene expression PolII occupancy appears to be more sensitive to H3.3 deposition at transcription start sites, we speculated that the initiation of ectopic transcription is more vulnerable to perturbation. To test this hypothesis, we reduced the activities of the general transcriptional factor TFII-D by knocking down various TATA box-binding protein (TBP)-associated factors (TAF). Although TAFs are essential factors, animals with 65% reduction of Taf5 or 40% reduction of Taf7 in the central nervous system can grow to the adult stage without discernible defects (S17A and S17B Fig). Ubx-expressing neurons in Pc, Taf5 or Pc, Taf7 double-knockdown brains exhibited 88% or 90% reduction in number relative to Pc depleted brains, respectively (Fig 9A). Next, we investigated whether ectopic expression is more sensitive to general transcriptional factor TFII-H subunits, Cdk7 and CycH, which phosphorylate PolII CTD serine 5. Similarly, Pc, Cdk7 or Pc, CycH double depletion by RNAi significantly reduced the number of neurons ectopically expressing Ubx in brains (Figs 9A, S17C and S17D). By contrast, orthotopic expression of Ubx in ventral nerve cords was not affected by partial depletion of Taf5, Taf7, Cdk7 or CycH (Fig 9B and 9C). The Trx-induced ectopic expression of Abd-B was also sensitive to knockdown of CycH (Figs 9D and S18A), whereas the orthotopic expression of Antp was not (Figs 9E, 9F and S18B). Interestingly, the ectopic expression domain of Abd-B extended to the ventral compartment by over-expressing CycH (Figs 9G and S18A). As a control, over-expression of CycH alone did not induce ectopic expression of Abd-B (Figs 9H and S18A) or affect the orthotopic express of Antp (Figs 9I and S18C). Collectively, we propose that ectopic gene expression involves more sensitive coordination between H3.3 deposition, TFII-D, and TFII-H activities than is required for orthotopic gene expression.
In this study, we provide evidence that loss of dBRWD3 suppresses ectopic gene expression and the tissue proliferation caused by the loss of PcG function, but not orthotopic gene expression. Loss of dBRWD3 also suppresses the ectopic gene expression induced by Trx. This suppression is related to enhanced H3.3 deposition at transcription start sites and reduced H3K4me3, activated PolII, and total PolII levels around the 5' regions of the ectopically expressed genes.
A genome-wide H3.3 ChIP study revealed that H3.3 is enriched in enhancers, promoters, and gene bodies of actively transcribed genes [48]. This correlation suggests that H3.3 deposition may promote gene transcription, a concept that has been supported by the fact that H3.3s are more likely to possess marks associated with active gene expression, including trimethylation at lysine 4 (H3.3K4me3) and acetylation at several lysine residues [49,50]. However, an elegant study demonstrated that gene transcription remains normal when H3.3 is replaced by H3.3K4A, casting doubt regarding the importance of H3.3K4me3 [51]. Moreover, the extent to which H3.3 deposition truly promotes gene transcription is difficult to determine from genetic studies because knockout of H3.3 concurrently leads to up-regulation of one set of genes and down-regulation of another[52]. In the case of trans-retinoid acid induced expression of Cyp26A1 in embryonic stem cells, H3.3 is actively deposited to the enhancer before induction. Upon induction, H3.3 is depleted from the enhancer but deposited into the promoter. Knockdown of H3.3 reduces the binding of RAR and Tip60 to the enhancer region, indicating that deposition of H3.3 at enhancer regions facilitates the activation of inducible genes [53]. However, the role of H3.3 at promoter and gene body in transcription remain unclear.
It is even less clear how H3.3 affects ectopic gene expression, a pathological condition frequently associated with various cancers in humans [54,55]. In this study, we provide several lines of evidence that show that the regulation of H3.3 by dRBWD3 is required for the ectopic gene expression observed in PcG mutants or upon over-expression of trx. Firstly, only dBRWD3 transgenes that are able to reduce H3.3 levels in dBRWD3 mutant cells are capable of restoring ectopic expression of upd3 in Scm, dBRWD3 double-mutant cells. Secondly, the loss of yem, which prevents the aberrant incorporation of H3.3, also restores the ectopic expression of upd3 in Scm, dBRWD3, yem triple-mutant clones. Conversely, the over-deposited H3.3 induced by YEM is sufficient to suppress the ectopic Antp expression and Trx-induced ectopic Abd-B expression.
In ectopic gene expression, H3.3 deposition at the enhancers is variably regulated by dBRWD3. Nevertheless, through a not entirely clear mechanism, the H3K4me1 induced by Trx at enhancers is insensitive to dBRWD3. By contrast, the dBRWD3 depletion-enhanced H3.3 deposition at transcription start sites interferes with H3K4me3 and PolII enrichment at the same regions as well as the 5' ends of the gene bodies. Taken together, these observations suggest that H3.3 deposition at these regions disrupts transcription by interfering with trimethylation of H3K4 as well as PolII engagement or activation. This notion is supported by a recent finding that ectopic gene expression persists longer in Hira mutants, in which H3.3 deposition is reduced at the promoter and 5' regions [56]. Different from the known role of H3K4me3 on the promoters, our data shows that H3K4me3 is enriched on the gene bodies of ectopically or orthotopically expressed genes rather than the promoters. This is most likely due to the bias associated with the selected PCR amplicons.
dBRWD3 regulates the deposition of H3.3 more prominently at the promoters and transcription start sites in both ectopic and orthotopic gene expression. However, dBRWD3 regulates the PolII occupancy and H3K4me3 levels only in ectopic gene expression. Due to a not-yet-defined “robustness” of transcription, orthotopic gene expression is rendered insensitive to the increase of H3.3. Our data suggest that the same robustness of orthotopic gene expression closely cooperates with TFII-D and TFII-H since orthotopic gene expression remains intact under the suboptimal TFII-D or TFII-H activities. Further studies are needed to understand molecular nature of the robustness.
In addition to their negative effects on ectopic gene expression, H3.3 deposition and dBRWD3 may also interact with PcG in different contexts. For example, it has been reported that H3.3 deposition directs PRC2 to bivalent promoters in ES cells [57]. In addition, loss of dBRWD3 up-regulates Pc, pho, and tna but down-regulates phol, Jarid2 and the trithorax group genes ash1 and Iswi [11]. The regulation of ash1, which encodes H3K4 monomethylase, is particularly interesting because an independent transcriptome analysis of adult heads in dBRWD3 hypomorphic mutants also confirmed a lower level of ash1 mRNA. However, functional studies revealed that ash1 depletion by RNAi rescues rather than exacerbates the rough eye phenotype caused by dBRWD3 depletion. In addition, the H3K4me1 levels in dBRWD3 mutant cells are similar to those in wild-type cells. Further studies are needed to determine the significance of dBRWD3 regulation of ash1 mRNA. Finally, knockdown of dBRWD3 causes further reduction of H3K27me3 or H2AK118ub in the E(z) or Pc depleted brains, suggesting that H3.3 deposition may accelerate the removal of these repressive marks. Further investigations are warranted to understand whether the accelerated removal of repressive marks is mainly contributed by the nucleosome turnover associated with H3.3 deposition or involves activation of demethylases and deubiquitylase.
The H3K4me3 levels at the promoter and 5' ends of genes correlate well with active transcription. In fact, it is both a cause and a consequence of active transcription. As a cause, H3K4me3 recruits the TFII-D subunit TAF3, a general transcription factor involved in PolII engagement and transcription initiation [58]. Based on such a scenario, we propose that dBRWD3 increases H3K4me3 levels to the extent required for promoters to recruit TFII-D in ectopic gene expression. Consistently, partial knockdown of Taf5 or Taf7 affects ectopic but not orthotopic gene expression. In other words, ectopic and orthotopic gene expression may require different levels of TFII-D activities.
During active transcription, H3K4me3 is established by hSet1A/B, which is recruited to actively transcribed gene regions by CTD Ser5-phosphorylated PolII [59]. The phosphorylation of PolII’s CTD Ser5 is mediated by CDK7/CycH and is preferentially required for ectopic gene expression. Consistent with this idea, we demonstrated that partial knockdown of Cdk7 or CycH indeed affected ectopic gene expression without a discernible effect on orthotopic gene expression, perhaps because it was more dependent on the phosphorylation of PolII CTD Ser5. A complement study in RING1A, RING1B double knockout embryonic stem cells revealed that de-repressed loci displaying higher levels of PolII phospho-CTD Ser5 are ectopically expressed at higher levels [60], supporting our findings that phosphorylation of PolII CTD Ser5 plays an unique role in ectopic gene expression that is not shared with orthotopic gene expression. Based in part on these findings, we propose that dBRWD3 could play a preferential role in ectopic gene expression by facilitating the phosphorylation of PolII CTD Ser5.
PcG mutations and reduced expression of PcG proteins contribute to tumorigenesis in several human malignancies [25–31,34,35,61–64]. Hence, understanding the regulation of ectopic gene expression will have important medical implications. It has been shown that the ectopic expression of upd1, upd2, and upd3 also underlies tissue overgrowth in Drosophila PcG mutants, suggesting an evolutionarily conserved role for PcG in tumor suppression from insects to humans. Based on our results, we speculate that inhibition of the BRWD3, TFII-D, and TFII-H complex, for example by the CDK7 inhibitor THZ1 [65–67], might preferentially suppress a broad spectrum of tumors driven by PcG mutations.
In summary, we found that ectopic gene expression differs from orthotopic gene expression in their sensitivities to dBRWD3. Inactivation of dBRWD3 selectively suppresses ectopic gene expression and tissue overgrowth induced by loss of PcG function.
p-ENTR-dBRWD3-N1287A-3XFlag, p-ENTR-dBRWD3-N1451A-3XFlag, p-ENTR-dBRWD3-N1287A, N1451A-3XFlag (p-ENTR-dBRWD3-2BC-3XFlag) were generated with the Thermo Scientific Phusion Site-Directed Mutagenesis kit using the previously described p-ENTR-dBRWD3-3XFlag as a template. p-ENTR-HA-yem was generated by PCR from the cDNA clone RE33235, Drosophila Genetic Resource Center. p-ENTR-dBRWD3-N1287A-3XFlag, p-ENTR-dBRWD3-N1451A-3XFlag, and pENTR- dBRWD3-2BC-3XFlag were recombined into the pUWR vector (DGRC Gateway collection) to generate pUWR-dBRWD3-N1287A-3XFlag-RFP, pUWR-dBRWD3-N1451A-3XFlag-RFP, pUWR-dBRWD3-2BC-3XFlag-RFP. p-ENTR-HA-yem was recombined into the pTWF vector (DGRC Gateway collection) to generate pTWF-HA-yem.
Flies were raised in standard conditions at 25°C except as otherwise mentioned. The dBRWD3s5349, and yemGS21861 were described earlier [11,68]. SceKO and ScmD1, trxE2 were kindly provided by Dr. Muller [39,47]. hs-H3.3-GFP was a gift from Dr. Kami Ahmad [69]. GMR-GAL4 (stock number 9146), Elav-GAL4 (stock number 458), ScmD1 (stock number 24158), Sce1 (stock number 24618), wtsx1 (stock number 44251), ex-LacZ (stock number 11067), UAS-mCD8-GFP (stock number 5146), UAS-Taf5-shRNA (stock number 35367), and UAS-Taf7-shRNA (stock number 55216) were obtained from the Bloomington stock center. UAS-trx (stock number 12194) and OK107-GAL4, were obtained from the Drosophila Genetic Resource Center, Kyoto. The ScmD1, dBRWD3s5349 double mutant, ScmD1, dBRWD3PX2 double mutant, dBRWD3s5349, Sce1 double mutant, ScmD1, dBRWD3s5349, yemGS21861 triple mutant, and wtsx1, dBRWD3s5349 double mutant were generated by recombination. UAS-Psc-dsRNA (NIG3886R-4) UAS-ph-p-dsRNA (NIG18412R-1), UAS-E(z)-dsRNA (NIG6502R-3), UAS-Pc-dsRNA (NIG32443R-1), UAS-CycH-dsRNA (NIG7405R-1), and UAS-Cdk7-dsRNA (NIG3319R-1) were obtained from the fly stocks of the National Institute of Genetics, Kyoto, Japan (NIG-FLY). UAS-dBRWD3-dsRNA (VDRC40209) was obtained from the Vienna Drosophila RNAi Center (VDRC). The transgenic flies ubi-dBRWD3-N1287A-3XFlag-RFP, ubi-dBRWD3-N1451A-3XFlag-RFP, ubi-dBRWD3-N1287A, N1451A-3XFlag-RFP (ubi-dBRWD3-2BC-3XFlag-RFP), and pTWF-HA-yem were generated by microinjection for germ-line transformation. The transgenic flies ubi-H3.3-dendra2, ubi-dBRWD3-3XFlag-RFP, ubi-delta-N-dBRWD3-3XFlag-RFP, and 10XSTAT-nlsGFP were described previously [11,41].
Genotypes for mosaic mutant clones in eyes:
Total RNA was isolated from instar larval mosaic eye brain complexes using TRIzol reagent (Invitrogen). Following the manufacturer’s protocol, cDNA was synthesized using oligo(dT) and SuperScript reverse transcriptase (Invitrogen). OmicsGreen qPCR 5X Master Mix (Omics Bio) was used for real-time quantitative PCR on a CFX96 connect Real-Time PCR System (Bio-Rad). RPL32 was used as an endogenous loading control.
3rd instar larval eye imaginal discs were dissected in PBS and fixed for 17 minutes in 4% formaldehyde, followed by three 10-min washes in PBS supplemented with 0.3% Triton-X-100 (PBT) and 30-min blocking in PBT containing 5% normal donkey serum (NDS). After blocking, discs were incubated with primary antibody either overnight at 4°C or 2 hours at room temperature in PBT containing 5% NDS. After incubation with primary antibody, discs were washed three times in PBT before incubating with secondary antibody in PBT containing 5% NDS for one hour at room temperature. After three subsequent washes, discs were mounted with glycerol. Primary antibodies used in this study include mouse anti-H2AK118ub (1:100, Millipore, E6C5), rabbit anti-H3K27me3 (1:100, Millipore), rabbit anti-H3K4me1 (1:100, Active Motif), rabbit anti-H3S10ph (1:500, Millipore), mouse anti-Ubx (1:20, DSHB, Ubx), mouse anti-Antp (1:20, DSHB, 8C11), rabbit anti-upd3 (1:750), and mouse anti-β-Galactosidase (1:1000, Sigma, GAL-50). Secondary antibodies include goat anti-mouse Cy3 (1:1000, Jackson ImmunoResearch), goat anti-mouse Cy5 (1:1000, Jackson ImmunoResearch), and goat anti-rabbit Cy3 (1:1000, Jackson ImmunoResearch).
Chromatin immunoprecipitation was done with a ChIP-IT High Sensitivity (HS) Kit (Active Motif) following the instructions provided by the manufacturer. Briefly, 300 pairs of brain lobes (leaving out the attached ventral nerve cords) or wing imaginal discs of 3rd instar larvae were collected. The collected tissues were fixed with complete tissue fixation solution (28μl 37% formaldehyde in 970μl PBS) at room temperature for 15 minutes. Fixation was stopped with the stop solution at room temperature for 5 minutes. The fixed tissues were washed with ice-old PBS wash buffer and then immersed in the tissues with the chromatin prep buffer. The fixed tissues were sonicated using the UP50H Ultrasonic Processor (Hielscher-Ultrasound Technology), with 30% amplitude and 20 pulse cycles of 30 seconds on followed by 30 seconds off. 6 μg sheared chromatin was incubated with 1μg antibodies overnight at 4°C. 30 μl Protein G agarose beads were added to each IP reaction. The mixture was rotated at 4°C for 3 hours. The ChIP reactions were loaded into columns and washed. The ChIP DNA was obtained by eluting the columns with elution buffer AM4. The elute was treated with Protease K at 55°C for 30 minutes, 80°C for two hours, followed by column clean-up. OmicsGreen qPCR 5X Master Mix (Omics Bio) was used for real-time quantitative PCR on a CFX96 connect Real-Time PCR System (Bio-Rad) to measure the amount of ChIP DNA and input DNA containing indicated sequences from enhancers, promoters, and 5' transcription regions of Antp, Ubx, and Abd-B. Primary antibodies used include rabbit anti-H2AK118ub (Cell signaling, D27C4), mouse anti-H3K27me3 (Abcam, mAbcam 6002), rabbit anti-H3K4me1 (Abcam, ab8895), rabbit anti-H3K4me3 (Abcam, ab8580), mouse anti-PolII (Abcam, 4H8), and rabbit anti-PolII phospho-CTD Ser5 (Abcam, ab5131), rabbit anti-HA (Cell Signaling, C29F4) and mouse anti-Flag (Sigma, M2)
All confocal images were obtained by LSM 700 laser scanning confocal microscope (Carl Zeiss). For quantitative analysis of protein levels, the antibody staining conditions, laser power, and pinhole sizes were kept identical among groups. Pixel number, pixel intensity, and area were provided by the built-in software in LSM 700. The areas of clones (marked by the absence of GFP) and non-clones (marked by GFP) were calculated by the total GFP positive and GFP negative areas respectively. Antp-positive regions in ventral nerve cords were manually marked. The Antp expression areas and lengths were calculated by the built-in software according to the marked regions. H3S10ph-positive mitotic cells and Antp-positive brain cells were manually counted.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.