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.1002599 | Widespread Site-Dependent Buffering of Human Regulatory Polymorphism | The average individual is expected to harbor thousands of variants within non-coding genomic regions involved in gene regulation. However, it is currently not possible to interpret reliably the functional consequences of genetic variation within any given transcription factor recognition sequence. To address this, we comprehensively analyzed heritable genome-wide binding patterns of a major sequence-specific regulator (CTCF) in relation to genetic variability in binding site sequences across a multi-generational pedigree. We localized and quantified CTCF occupancy by ChIP-seq in 12 related and unrelated individuals spanning three generations, followed by comprehensive targeted resequencing of the entire CTCF–binding landscape across all individuals. We identified hundreds of variants with reproducible quantitative effects on CTCF occupancy (both positive and negative). While these effects paralleled protein–DNA recognition energetics when averaged, they were extensively buffered by striking local context dependencies. In the significant majority of cases buffering was complete, resulting in silent variants spanning every position within the DNA recognition interface irrespective of level of binding energy or evolutionary constraint. The prevalence of complex partial or complete buffering effects severely constrained the ability to predict reliably the impact of variation within any given binding site instance. Surprisingly, 40% of variants that increased CTCF occupancy occurred at positions of human–chimp divergence, challenging the expectation that the vast majority of functional regulatory variants should be deleterious. Our results suggest that, even in the presence of “perfect” genetic information afforded by resequencing and parallel studies in multiple related individuals, genomic site-specific prediction of the consequences of individual variation in regulatory DNA will require systematic coupling with empirical functional genomic measurements.
| A comprehensive understanding of the contribution of individual genome sequences to disease and quantitative traits will require the general ability to predict consequences of genetic variation in non-protein-coding regions, particularly those involved in gene regulation. Here we tested the power to predict such consequences when presented with “complete” information encompassing the genomic DNA binding site patterns of a well-studied regulatory protein across multiple related individuals, coupled with all individual genome sequences at the binding positions. We find that, while there is reasonable ability to predict the average effects of variation within the consensus recognition sequence of a transcriptional regulator, it is not possible to determine reliably the consequences of variation at any given genomic instance. This suggests that the interpretation of individual genome sequences will require comprehensive complementation with functional genomic studies.
| A growing number of studies associate variation within regulatory DNA and risk of human disease [1]–[3]. Variation in regulatory DNA may result in modulation of recognition by sequence-specific transcription factors (TFs), resulting in altered gene expression [4]–[6]. That the vast majority of variants emerging from human resequencing studies lie in non-coding regions creates an urgent need for determining the consequences of variation within regulatory DNA.
Functionally significant variation within the genomic recognition sequences for certain TFs appears to be correlated in aggregate with nucleotide-level evolutionary conservation and/or position-specific information content [7]–[10]. Although surveys have identified sites of allele-specific occupancy of TFs and RNA Polymerase II or allele-specific chromatin states [11]–[15], these studies have not established the distinguishing characteristics of regulatory sequence variation with an experimentally-observed effect on occupancy. As such, it is currently not possible to interpret reliably the functional consequences of variation within any given TF recognition sequence.
To address this, we apply a novel experimental design to identify comprehensively patterns of genetic variation with heritable effects on the occupancy of the major genomic regulator CTCF [16]. Unlike most sequence-specific regulators which rely on cooperative interactions with other factors to bind DNA, CTCF is able to access target DNA within chromatin in a relatively autonomous fashion through its rich binding interface. By combining quantitative genome-wide occupancy analysis by ChIP-seq in a multi-generational pedigree with comprehensive resequencing of the binding site landscape across all individuals, we achieve complete knowledge of variation in both sequence and occupancy, thus creating a benchmark for assessing the characteristics of functional and heritable regulatory sequence variation.
We mapped binding sites for CTCF by ChIP-seq in B-lymphoblastoid cells derived from 12 members of a three-generation pedigree (Figure 1A, 1B). We identified a total of 51,686 binding sites across all individuals at a false discovery rate (FDR) of 1%. To comprehensively identify genetic variation with potential functional consequences for CTCF binding, we performed targeted resequencing by array capture focused on the 134 bp interval surrounding 46,568 CTCF sites (total ∼6 Mbp) in all family members assayed by ChIP-seq. 7,394 of the 35,709 surveyed binding sites (or 21%) overlapped one or more SNPs, some of which had clear associations with occupancy in the direction predicted by the CTCF motif (Figure 1C, 1D). We did not consider other variation such as copy number variants or small indels. In order to minimize reference mapping bias for the ChIP-seq data, we remapped tags to personalized genomes including discovered SNPs [17]. Additionally, to avoid artifacts resulting from uncertain mapping of 36 bp reads to the genome, we simulated all reads including discovered SNPs from ±147 bp centered on the ChIP-seq peak and excluded sites with too many ambiguously mapped tags.
We integrated the genetic and functional data sets to survey genome-wide heritable variation in transcription factor occupancy. We reasoned that the strongest signal of heritable variation would be from segregating variants overlapping the binding site. Thus we performed a linear regression of the ChIP-seq density on the SNP genotype in cis (Figure 1D). Of 5,828 polymorphic sites, this analysis identified 325 (5.6%) sites with a significant association of SNP genotype with occupancy at a false discovery rate (FDR) of 1% (Figure 2, Table S4). We tested whether several confounding factors might be responsible for our results, however sites at which SNPs were significantly associated with changes in occupancy were similar to polymorphic sites without changes in occupancy in terms of GC content (median 53.7% vs. 53.0%, Mann-Whitney p<0.044) and ChIP-seq input signal (3.61 vs. 3.61, Mann-Whitney p<0.84), and distance to the nearest RefSeq TSS (33 kb vs. 36 kb, p<0.96). Significant sites were only slightly weaker in terms of ChIP-seq density (2.73 vs. 2.93, p<2.7*10−3), and DNase I signal (5.47 vs. 6.92, p<7.8*10−8). Thus we conclude that the SNP genotype is associated with differences in occupancy at 325 of 5,828 sites tested.
We used a hypothesis-driven linkage analysis to assess the heritability of the remaining unexplained differential occupancy. First, we identified 1,376 sites of differential occupancy. Of these, 200 (15%) were already associated to an underlying SNP (FDR 1%), 65 (4.7%) had allele-specific occupancy (FDR 0.1%), and 197 (14%) were on chromosome X. To test for heritable inheritance of occupancy not explained by these factors, we performed Haseman-Elston sib-pair linkage analysis in aggregate at sites differentially occupied among the 6 grandchildren (Figure S1A and S1B; see Materials and Methods). The 47 binding sites already significantly associated with SNPs had a regression slope of −2.86 (Figure S1C), confirming substantial heritability (p<8.9*10−7, permutation). The remaining 50 sites without significant associations had a regression slope of −1.01 (Figure S1D), indicating a lower but still significant level of heritable variation (p<3.2*10−5). These results suggest that SNPs directly overlapping the cognate recognition sequence explain most but not all of the heritable variation in this pedigree. Remaining variation in occupancy might be heritable due to sequence variants not considered in our SNP-based analysis or heritable epigenetic variation such as methylation. However, it is quantitatively less significant than the heritability attributable to the direct effect of SNPs on occupancy (Figure S1E).
Understanding the effect of DNA sequence variation on transcription factor occupancy is critical to a mechanistic interpretation of non-coding variation. To interpret the association results in the context of the CTCF motif, we scanned the center of the ChIP-seq peak with the known position weight matrix (PWM) [18], which measures the contribution of each nucleotide in the binding site to the energy of the protein-DNA interaction [19], [20]. 888 binding sites did not contain a motif match (fimo p-value<10−2) and 1,040 binding sites overlapped multiple SNPs within ±180 bp of the ChIP-seq peak. Excluding these sites, we analyzed the 4,428 binding sites with a single SNP and a single motif match. These SNPs were distributed throughout the resequenced region surrounding each CTCF motif (Figure 3A).
In contrast, we expected that SNPs associated with occupancy differences would be concentrated in the 44 bp region of protein-DNA contact [21]. Indeed, despite a slight reduction (1.08-fold) in local sequence diversity, 85% of the SNPs that affected occupancy were within this region (Figure 3B). The allele observed to have higher occupancy matched the energetically more favorable one for 83% of these SNPs. Associated SNPs outside the region of contact had less significant q-values (median q-value of 1.3*10−3 outside the versus 1.3*10−5 inside), consistent with these SNPs being false positives or sites with ambiguity in the true location(s) of protein-DNA interaction. Alternatively, some of these SNPs might affect CTCF occupancy indirectly by perturbing an adjacent co-factor binding site.
We compared our results to an allele-specific occupancy test performed at heterozygous sites (Figure S2). Despite a weaker enrichment of significant sites within the core motif and a less substantial concordance of the higher occupancy allele with the energetically more favorable nucleotide, this allele-specific analysis broadly corresponded to the results of the association analysis. Thus, we interpret the results of our analysis as indicating that we have correctly identified the motifs at most binding sites, and that the significant SNPs directly affect occupancy through modulating the protein-DNA interaction at these sites.
Although differences in occupancy were largely associated with SNPs at positions strongly affecting overall binding energy (Figure 3B), only 13% of SNPs at the interface of protein-DNA interaction affected occupancy (Table S4). Thus although functional SNPs are highly concentrated in the region of protein-DNA contact, the majority of SNPs, even in this region, do not measurably affect occupancy. Since our data set includes multiple sites with the same two alleles at the same position measured relative to the binding motif, we investigated the proportion of sites at which a given change was found to affect occupancy (Figure 4A). Like in Figure 3, the most disruptive changes were observed at positions of high information content in the motif. However, even over the 14 bp core motif, changes affected occupancy at a median of only 36% of the sites where they were observed. We found no changes that uniformly affected occupancy without regard to context. Instead, we observed a strong, progressive depletion in the proportion of changes affecting occupancy at the strongest sites (Figure 4B), and a smaller depletion at the weakest sites. Indeed, simply clustering the ChIP-seq intensities identified three major groups, of low, medium and high occupancy, which were also distinguished by varying proportions of significant SNPs (Figure S6). This result places an upper limit on the accuracy of methods that predict the effects of non-coding SNPs without consideration of their context.
Strength-dependent buffering could be explained by a model where changes in occupancy are observed only when a SNP causes the affinity of a site to cross a threshold for binding. In this case, the strongest and weakest sites will only be affected by the greatest genetic perturbation, while smaller perturbations would affect binding only at sites of intermediate strength. This would create the impression of epistasis between all positions in the cognate recognition site as any affinity-affecting change could potentially buffer another [22], [23]. We thus compared the inherent affinity of the site with the magnitude of the perturbation caused by each SNP. We divided SNPs affecting occupancy into bins based on the strength of their match to the canonical motif. We found that SNPs at sites matching the CTCF motif more strongly in turn exhibited higher log-odds differences (Figure 4C). We observed no such trend at SNPs not associated with occupancy differences (Figure S3). These results are consistent with stronger motifs being buffered against all but the largest perturbations. Although a linear regression identifies a significant effect (p<0.006), an r2 of 0.04 indicates that the strength of the motif match alone can not explain the breadth of buffering observed.
Buffering might also be a consequence of the non-additive effect on binding energy of individual positions in the cognate binding site, as has been observed in vitro [24]. The relevance of non-additive interactions for identifying binding sites has been questioned [25], [26], but the implications for understanding the function of specific variants in vivo have remained unclear. To explore the power of our data set to discover epistatic interactions, we measured the mutual information between the sequence context per-base in the core motif and whether a SNP at each location affects occupancy (Figure 4D). This analysis identifies two positions in the consensus sequence that significantly buffer the effect of a SNP at another position. First, of the 24 SNPs observed at position 1 in the motif, 13/13 that affected occupancy had an adenine at position 5, compared to only 5/11 for those that did not affect occupancy (Figure 4E, above). Interestingly, the second significant buffering interaction is between position 7 and SNPs at the adjacent position 8 (Figure 4E, below), suggesting local compensation for the adjacent SNP. These results indicate that higher-order models may be necessary to fully model the effect of polymorphism on protein-DNA interaction, and are consistent with a model where local factors determine whether polymorphism affects occupancy.
Resequencing and association studies are producing large amounts of data on polymorphism in non-coding regions, yet as we have illustrated, their functional classification is difficult. To investigate the power of existing metrics to predict functional polymorphism in non-coding regions, we used as a reference set the 1,368 sites with SNPs within the 44 bp vicinity of a recognizable CTCF motif. We first assessed the predictive power of evolutionary constraint, which has been used successfully to discover regulatory motifs [27], to highlight functional positions within motifs [9], [10], [28], and to predict the effect of coding variants [29]–[31]. Conservation is a particularly attractive operational metric in genome scans, as it can be applied in an unbiased fashion without directly measuring protein-DNA interaction or modeling context effects. Indeed, CTCF binding sites are clearly marked by increased conservation [32]. Thus, we tested the sensitivity and specificity of per-nucleotide conservation (phyloP 44-way vertebrate alignment, from UCSC browser) to correctly identify the 186 significant SNPs in our reference set. However, despite being applied only across experimentally determined binding sites, conservation had little predictive power on this data set, with an AUC of 0.57 (Figure 5).
Then we measured the improvement from evaluating potentially functional SNPs within the context of protein-DNA binding energetics. Applying such predictor showed a marked improvement over conservation, with an AUC of 0.75 (Figure 5), although positive predictive power was greatest for the most severe perturbations (Figure S4). Nevertheless, these results illustrate the power to be gained from considering non-coding polymorphism within the context of functional genomics data on transcription factor occupancy.
In contrast to the vast diversity of protein function, the elements that regulate gene expression recruit from a shared repertoire of transcription factors, offering the potential for a common regulatory sequence code. The torrent of variants emerging from human resequencing studies – the vast majority of which lie in non-coding regions – coupled with the growing number of common, disease-associated non-coding variants [1]–[3] has created an urgent need for determining the consequences of variation within regulatory DNA. However, the proportion of variants within regulatory DNA that have reproducible functional consequences on regulatory factor binding is currently unknown, and our ability to predict such outcomes from known rules of protein-DNA interaction is uncertain.
We have described a novel, hypothesis-driven genetic method employing targeted capture and genome-wide in vivo occupancy profiling to investigate directly the consequences of heritable variation in regulatory sequence. Our results show that individual transcription factor binding sites are surprisingly robust to genetic variation, even at evolutionarily constrained positions. While previous studies have observed differences in transcription factor occupancy among individuals using occupancy profiling alone [15], genome-wide linkage scans [33], or allele-specific occupancy approaches [11], this work is the first systematic analysis of patterns of functional alteration in TF recognition sequences. This study further advances the characterization of heritable variation in TF binding by using highly accurate sequence information throughout a three-generation pedigree.
Our study has revealed a large degree of context dependence for changes to the CTCF recognition sequence. Indeed, even over the core 14 bp motif, only 36% of SNPs affected occupancy (Figure 4A). Our estimate of the percentage of SNPs that affect occupancy in this 14 bp region ranges between 24% to 42% at FDRs 0.1% and 5%, respectively, indicating that the magnitude of this effect cannot be explained by the choice of significance cutoff. We have suggested that buffering is partly mediated by the strength of the binding site, as well as the sequence context at the local CTCF recognition sequence. In addition, buffering might be facilitated by a feedback process that maintains a constant CTCF occupancy despite alterations to the site's inherent affinity. However, while 21% of the SNPs in the region of protein-DNA contact that were significant in our association analysis also exhibited allele-specific occupancy in heterozygous samples, only 3.7% of the non-significant SNPs did, indicating that buffering is not likely to be the consequence of a feedback process. Alterations in DNA methylation might also mask the effect of otherwise significant genetic changes. However, only 30% of polymorphic CTCF sites contain a CpG at positions 1 or 11 of their recognition sequences. Furthermore, the prevalence of CpGs at these positions is the same at sites where a SNP does and does not affect occupancy, limiting the potential scope of methylation to fully explain the observed buffering. As this study was performed on transformed B-lymphoblastoid cells, it is worth noting that the specific CTCF sites that are buffered may not be extrapolated to primary cell types. However, assuming that EBV transformation does not invoke novel cellular mechanisms to regulate protein-DNA interaction, our primary conclusion stands that TF occupancy is strongly modulated by site-dependent effects.
Our results establish a low level of mutational load directly affecting transcription factor occupancy in the 4 founder genomes. Although variants were found at 21% of surveyed CTCF sites, only 0.9% of binding sites exhibited a difference in occupancy due to polymorphism (Figure 2). Previous studies have identified varying levels of positive and negative selection in transcription factor recognition sequences by estimating changes in binding energy [34]–[36]. However, our results indicate that 87% of polymorphism observed in the region of protein-DNA contact does not affect binding (Table S4). This is a higher proportion of silent variation than predicted by binding energy models (Figure S4), providing evidence that the scope of sequence change consistent with neutral evolution may be larger than previously thought.
Interestingly, we observed that the allele with higher occupancy was the derived allele in 40% of the cases (assuming the chimpanzee allele is ancestral). This indicates that approximately 40% of the functional substitutions in the human lineage increased occupancy, which is surprisingly high given that most mutations might be expected to reduce binding energy.
Previous work studying the power of comparative genomics has predicted a steep increase in the number of sequenced genomes required to obtain nucleotide resolution, particularly in the absence of perfect conservation [37]. While genome-wide phylogenetic footprinting approaches have highlighted substantial conservation of transcription factor sequence specificities [27], [32], [38], [39], functional studies of diverged species have uncovered low conservation in occupancy at orthologous sites [40]–[42]. Any phylogenetic approach is thus a compromise between statistical power gained by sampling more diverged species and the ability to recognize similar functional elements by sequence similarity. The optimum evolutionary distances to sample may be different for assessing functional non-coding elements than for more conserved coding sequence [43], [44]. This tradeoff suggests a potential motivation for broad resequencing of natural populations, though even this approach faces the fundamental limitation of ineffective purifying selection in primates and humans [45].
Gene-based studies have successfully identified causal variants using current methods for prediction of functional non-synonymous protein variants [46]. Coding mutations in the CTCF gene affecting its DNA binding specificity have been identified in cancer samples [47], but lesions in its binding sites are harder to interpret. The link between cognate recognition sequence and cellular consequence is complicated by potential influences from the cell-type specific chromatin landscape [48], maintenance of regulatory function despite sequence rearrangement [49], altered association with protein complexes [50], [51], and the lack of binding specificity and occupancy data for common transcription factors. In spite of these caveats, our results indicate that a motif-based classifier of variation in experimentally-identified CTCF binding sites predicts functional variation with a 59% true positive rate and a false positive rate of 20% (Figure 5). In comparison, current methods for prediction of non-synonymous protein variants achieve a 73% to 92% true positive rate at the same false positive rate [30] – not dramatically greater considering the comparatively greater depth of variant databases used to assess coding variation. Given the encouraging performance of a straightforward functional genomics approach, that the majority of variants presently associated with human physiology and pathology lie in non-coding regions [3] should be grounds for optimism.
In summary, our results indicate the existence of widespread recognition site-dependent buffering of polymorphism within regulatory DNA regions. A major implication of our work is that the potential for accurately predicting the consequences of variation affecting regulatory factor recognition sequences is severely limited by complex context dependencies, necessitating empirical assessment using functional genomic approaches. The feasibility of approaches such as the one we describe here has recently dramatically increased owing to coupled advances in next-generation sequencing technology and molecular biology, and continuation of this trend should in the near future enable further systematic investigations into the effect of polymorphism on protein-DNA interaction on a routine basis.
The B-lymphoblastoid cell lines from CEPH pedigree 1459 were obtained from Coriell and cultured in RPMI1640 medium (Cellgro), supplemented with 15% fetal bovine serum (FBS, Hyclone), 2 mM L-Glutamine, and 25 IU/mL penicillin and 25 µg/mL streptomycin (Cellgro).
B-lymphoblasts (5 million cells) were crosslinked with 1% formaldehyde (Sigma), lysed in lysis buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA pH 8.0, 1% SDS), and sheared by Bioruptor (Diagenode). The supernatant was further diluted 10-fold with dilution buffer (50 mM Tris-HCl pH 8.0, 166 mM NaCl, 1.1% Triton X-100, 0.11% sodium deoxycholate). For each immunoprecipitation, 100 µL Dynabeads (M-280, sheep anti-rabbit IgG, Invitrogen) were incubated with 20 µL CTCF antibody (#2899, Cell Signaling) for at least 6 hours at 4°C. The antibody-conjugated beads were then incubated overnight with sheared chromatin. The complexes were washed with IP wash buffer I (50 mM Tris-HCl pH 8.0, 0.15 M NaCl, 1 mM EDTA pH 8.0, 0.1% SDS, 1% Triton X-100, 0.1% sodium deoxycholate), high salt buffer (50 mM Tris-HCl pH 8.0, 0.5 M NaCl, 1 mM EDTA pH 8.0, 0.1% SDS, 1% Triton X-100, 0.1% sodium deoxycholate), and TE buffer (10 mM Tris-HCl pH 8.0, 0.1 mM EDTA pH 8.0). Crosslinking was then reversed in elution buffer (10 mM Tris-HCl pH 8.0, 0.3 M NaCl, 5 mM EDTA pH 8.0, 0.5% SDS) at 65°C overnight. The DNA was separated from the beads and treated with Proteinase K (Fermentas) and purified by phenol-chloroform extraction and ethanol precipitation.
Sequencing libraries were constructed according to Illumina's genomic prep kit protocol as previously described [28]. Briefly, ChIP DNA was end-repaired using the End-it DNA repair kit (Epicentre). Adenines were added to the 3′ ends of the blunt-ended DNA using Taq DNA polymerase (NEB). PE adapter (1∶20 dilution, 1 µL for 15–50 ng starting ChIP DNA, Illumina) was ligated to the ends of the A-tailed ChIP DNA with T4 DNA ligase (NEB). 1/3-1/4 of the purified ligation product was PCR amplified for 16 cycles with High-fidelity PCR master mix (NEB) and PE primer 1.0/2.0 (Illumina). Libraries were run on 2% agarose gels, size-selected, and purified with QIAquick gel extraction kit (Qiagen). The libraries were sequenced to 36 bp on an Illumina Genome Analyzer by the High-throughput Genomics Unit (University of Washington) according to standard protocol. ChIP-seq experiments were performed on 2–3 independently cultured biological replicates per sample (Table S1, Table S2).
High quality reads were aligned to the reference genome using the Eland aligner. SPP [52] was used to call peaks on total tag data from the 12 samples, resulting in 51,686 peaks at a Poisson-derived FDR of 1%. Using the aggregate of all tags was more conservative (in terms of number of peaks) than taking the union of peak calls on individual samples, but less conservative than taking the intersection. The MTC method (“tag.lwcc”) was used to call point binding positions (Table S5). Motif representations used Weblogo [53].
We designed an Agilent 244k SureSelect microarray for targeted resequencing on the 51,686 ChIP-seq peaks identified in the 12 samples. We used fimo (http://meme.sdsc.edu/meme/) to scan for instances of the core 14 bp of the canonical motif [18] with a p-value of 10−2. We adjusted the target locations to center on matches to the nearest CTCF motif if the motif was within 50 bp, and added flanking targets to capture additional nearby motifs. 5 potential probes were tiled at 15 bp spacing to the 120 bp surrounding each target. We adjusted probe binding energy similarly to Ng et al. [54], adjusting the spacing of probes by up to 5 bp and adjusting the lengths to between 40–60 bp to reach a predicted Tm between 60–72°C. We used the Duke Uniqueness 20 bp track (UCSC genome browser) to filter out 5,828 probes with potential for cross-hybridization. We further excluded 145 probes in satellite repeats (RepeatMasker, UCSC genome browser) or with high blast scores to multiple genomic locations. The final design had 242,380 probes targeting 46,652 CTCF sites.
Genomic DNA was extracted from cultured cells, and targeted capture was performed based on Supplementary Protocol 3 of Mamanova et al. 2010 [55] with modifications. Briefly, 12 µg of genomic DNA was sheared in a Covaris S2 (Covaris, Inc.) using a duty cycle of 10%, intensity 5, cycle/burst 100, time 600 sec. DNA was end-repaired, A-tailed, and ligated to SE adapters (Illumina), and purified with Agencourt AMPure XP beads (Beckman Coulter). A trial PCR amplification using Phusion HF polymerase (NEB) and SE primers SLXA_FOR_AMP and SLXA_REV_AMP [54] (IDT) was performed on a fraction of ligated DNA with an Roche LightCycler 480 to determine the optimal number of cycles. PCR for all ligated DNA was then performed in eight 200 µL tubes using the following program: 98°C for 30 s, then a previously determined number of cycles of 98°C for 10 s, 65°C for 30 s, 72°C for 30 s, followed by 72°C for 5 min. The final DNA library was pooled and concentrated to 5 µL in a SpeedVac. For hybridization, 10 µg of DNA library was combined with formamide (Ambion), blocking oligos (SLXA_FOR_AMP, SLXA_REV_AMP, SLXA_REV_AMP_rev and SLXA_FOR_AMP_rev [54], Human C0t-1 DNA (Invitrogen), 2× Hi-RPM Hybridization Buffer (Agilent) and 10× Blocking Buffer (Agilent). Hybridization was performed using the Maui Hybridization System (BioMicro Systems, Inc.) at 55°C for 48 hours according to “MAUI Mixer LC on Agilent 244K CGH Microarrays” protocol. After hybridization, microarrays were washed with Agilent aCGH Wash Buffers 1 and 2, sealed with Secure-Seal (GRACE Bio-Labs) and placed on heat block (VWR) for elution of DNA. DNA was eluted with 3 mL of 95°C PCR-grade water, concentrated, and amplified with SE primers using the same PCR program as above. The libraries were sequenced to 36 bp on an Illumina Genome Analyzer by the High-throughput Genomics Unit (University of Washington) according to standard protocol.
Reads were aligned to the human genome (hg18) using bwa 0.5.8 [56] using default settings, allowing up to 2 mismatches (Table S3). Some lanes exhibited an excess of mismatches to the reference sequence at the 5′ or 3′ end, so tags were trimmed by up to 9 bp. Reads with identical 5′ ends were presumed to be PCR duplicates and were excluded using Picard v1.22 MarkDuplicates (http://picard.sourceforge.net/). SAMtools v0.1.7 [57] was used to generate a pileup of potential SNPs from uniquely-mapping reads with a mapping quality above 30. SNPs were called (Table S6) as biallelic variants with at least 8× resequencing coverage, a Phred-scaled SNP quality of at least 30, at least 20% of reads matching the allele with lower coverage, and Mendelian segregation according to PLINK v1.07 [58]. The chimpanzee allele was identified using axtNet alignment files for PanTro2 from the UCSC Genome Browser.
We performed two validations of the SNP calls from our targeted resequencing. First, we performed Sanger sequencing on PCR products from genomic DNA (Table S8). We tested 33 SNPs in all 12 samples. 0 of the 388 genotype calls were discordant. Second, we compared genotypes with genotypes available from the HapMap project for 6 of the 12 samples (Table S9). Release 27 genotypes were obtained from http://hapmap.ncbi.nlm.nih.gov, and matched to capture resequencing SNPs by location. 244 of the 27,808 genotype calls in common (0.88%) were discordant.
For each individual, a custom human genome was created from hg18 including autosomes, unscaffolded contigs, mitochondrial DNA, a Y chromosome for males, and the Epstein-Barr virus genome (gi|9625578|ref|NC_001345.1). Each genome was personalized to reflect SNPs identified by resequencing, using IUPAC codes to represent heterozygous position. ChIP-seq data was mapped using MosaikAligner v1.1.0021 (http://code.google.com/p/mosaik-aligner/) with the options “-bw 13 -act 20 -mhp 100 -m unique -mm 4 -minp 1.0”, requiring a unique mapping considering up to 4 mismatches. Reads with more than 2 mismatches were then discarded. Reads with identical 5′ ends were presumed to be PCR duplicates and were excluded using Picard v1.22 MarkDuplicates. Smoothed density tracks were generated using bedmap (http://code.google.com/p/bedops/) to count the number of tags overlapping a sliding 150 bp window, with a step width of 20 bp. Density tracks were normalized for sequencing depth by a global linear scaling to fix an arbitrary value of 25 as the 50th percentile of bins with more than 15 reads. We identified instances of the canonical motif (fimo p-value<10−2) within 15 bp of the center of the resequencing target, keeping the motif with the best p-value. We measured occupancy by the maximum normalized ChIP-seq tag density over the 14 bp motif.
We applied a regression method to measure whether a particular biallelic SNP is associated with occupancy, and if so which allele is associated with higher occupancy. We tested only sites with ≥8× resequencing coverage in ≥6 samples, sufficient mappability, and data for ≥4 data points each for ≥2 genotypes and ≥12 data points overall (Table S7). We further excluded 242 sites overlapping indels in the CEU population identified by the 1000 Genomes Project release 2010_07 [59]. We used a negative binomial generalized linear model (glm.nb in the R package MASS) to measure the significance of the effect of polymorphism on occupancy using an additive effect model, and including a replicate term to account for batch effects. We used two ChIP-seq replicates per sample, except for GM12870, which had one replicate (Table S1).
For sites with more than one SNP, we tested only the SNPs with more data points and those inside the region of protein-DNA contact. If there were still multiple SNPs per window, we chose the SNP with the lowest p-value, though these sites were then excluded from analyses depending on the known position of the SNP relative to the motif. We also tried fitting a dominant effect model where permitted by sample size: we chose between additive and dominant effect models using the Akaike information criterion. We separately fitted a linear model on the same data to estimate the r2.
We used the R package qvalue to estimate q-values (Figure S5) [60], which established a cutoff of p<9.6*10−4 as an FDR of 1% (325 sites). Using a Benjamini-Hochberg FDR strategy confirms a similar cutoff for FDR 1% (4.9*10−4, 293 sites). To independently confirm our FDR methodology, we considered the proportion of significant SNPs within the region of protein-DNA contact (Figure 3), which ranged from 71%, to 85%, to 91% at FDRs 5%, 1%, 0.1%, respectively (Table S4).
We used 31,128 SNPs identified in our resequencing as markers to generate a map of identity-by-descent (95th percentile marker spacing of 0.5 cM). We used recombination rate data [59] to place our SNP coordinates onto a genetic map, choosing SNPs with fewer missing genotypes at duplicate map positions. We then used MERLIN [61] to filter out improbable genotypes and compute IBD at our marker locations (option “–ibd”). We used the nearest marker at each binding site to estimate IBD for the 15 possible sib pairs. IBD values were placed into 3 bins (0, 0.5, and 1.0); values with >0.05 uncertainty were excluded.
We then used the package DESeq [62] to identify differentially occupied regions, both throughout all 12 samples as well as among just the grandchildren, using two replicates per sample (Table S1). We then applied a variance stabilizing transformation, and normalized the occupancy at each site to a mean of 0 and standard deviation of 1. We then averaged the occupancy of the two replicates.
We then performed Haseman-Elston regression at the 97 autosomal sites differentially occupied among the grandchildren, treating separately sites whose differential occupancy was already associated with a SNP or allele-specific binding. We considered all sib pairs and all sites simultaneously. Although regression methods exist with higher power [63] or that use data from all members of the pedigree [64], we applied the original method of regressing squares of trait differences for sib pairs, reasoning that the robustness of a simple method would have more forgiving assumptions. To account for the non-independence of measuring multiple sib pairs from the same family, we assessed the significance of the regression slope by permuting IBD vectors to random sib pair difference vectors for all differentially occupied sites, thus maintaining any correlation structure in the data.
We tested allele-specific occupancy in pooled replicate data for each sample. Given the reliance of allele-specific occupancy tests on high coverage at heterozygous sites, we included data from an additional replicate for 4 samples (Table S1). We had sufficient power to test 2,535 heterozygous binding sites with adequate mappability and 13× ChIP-seq read coverage. We used a chi-squared test against a 50∶50 null expectation to derive a p-value from the counts of the ChIP-seq tags mapping to the two alleles. We used the R package qvalue to estimate an FDR (Figure S5) [60]. In interpreting our FDR threshold, we considered the observed concentration of significant SNPs within the 44 bp region of protein-DNA interaction (Figure S2), which increased at more conservative FDRs: 63%, 66%, 68% at FDRs 1%, 0.5%, 0.1%, respectively (Table S4). At sites for which multiple samples had testable heterozygous sites with the same alleles, the sample with the most total reads was picked as representative for plotting.
We used PWM models [19] to measure the effect of a polymorphism on information content. CTCF binds in a multivalent fashion [21], [65], [66], wherein three modes of binding are distinguished by the presence and position of an upstream motif. At each site we chose the best-matching of the three motif models. To measure information content, we converted frequencies to log-odds, using a pseudo-frequency of 0.01 (the minimum observed frequency).
We calculated the mutual information between whether a given SNP affected occupancy (FDR 5%) and sequence context at 14 positions in the core CTCF motif using the mutualInfo function of the R package bioDist. To estimate the significance of the mutual information values, we applied a bootstrap method for each pair of positions tested. We calculated p-values from 2,000 iterations of resampling per pair of positions (Figure S7). To account for multiple testing across all positions, we used the R package qvalue [60].
We downloaded phyloP based on a 44-way vertebrate alignment from the UCSC Genome Browser. SNPs were ranked in decreasing order of the phyloP score at the location of the SNP, thus ranking sites with the most indication of purifying selection the likely to be disrupted by a SNP. To measure the predictive power of models of protein-DNA binding energy, we used a position weight matrix to compute the difference in log-odds score between the two alleles of each SNP. Sites with the largest difference between alleles at the location of the SNP were ranked as most likely to be disruptive. Plots were generated using the R package ROCR [67].
ChIP-seq data are viewable in the UCSC Genome Browser (http://genome.ucsc.edu, version hg18), and have been deposited in GEO (GSE30263). Resequencing data are available in the SRA (SRP009457), and the capture resequencing array design in GEO (GPL14147).
|
10.1371/journal.pntd.0007345 | A highly expressed intestinal cysteine protease of Ancylostoma ceylanicum protects vaccinated hamsters from hookworm infection | Human hookworms (Necator americanus, Ancylostoma duodenale, and Ancylostoma ceylanicum) are intestinal blood-feeding parasites that infect ~500 million people worldwide and are among the leading causes of iron-deficiency anemia in the developing world. Drugs are useful against hookworm infections, but hookworms rapidly reinfect people, and the parasites can develop drug resistance. Therefore, having a hookworm vaccine would be of tremendous benefit.
We investigated the vaccine efficacy in outbred Syrian hamsters of three A. ceylanicum hookworm antigen candidates from two classes of proteins previously identified as promising vaccine candidates. These include two intestinally-enriched, putatively secreted cathepsin B cysteine proteases (AceyCP1, AceyCPL) and one small Kunitz-type protease inhibitor (AceySKPI3). Recombinant proteins were produced in Pichia pastoris, and adsorbed to Alhydrogel. Recombinant AceyCPL (rAceyCPL)/Alhydrogel and rAceySKPI3/Alhydrogel induced high serum immunoglobulin G (IgG) titers in 8/8 vaccinates, but were not protective. rAceyCP1/Alhydrogel induced intermediate serum IgG titers in ~60% of vaccinates in two different trials. rAceyCP1 serum IgG responders had highly significantly decreased hookworm burdens, fecal egg counts and clinical pathology compared to Alhydrogel controls and nonresponders. Protection was highly correlated with rAceyCP1 serum IgG titer. Antisera from rAceyCP1 serum IgG responders, but not nonresponders or rAceyCPL/Alhydrogel vaccinates, significantly reduced adult A. ceylanicum motility in vitro. Furthermore, rAceyCP1 serum IgG responders had canonical Th2-specific recall responses (IL4, IL5, IL13) in splenocytes stimulated ex vivo.
These findings indicate that rAceyCP1 is a promising vaccine candidate and validates a genomic/transcriptomic approach to human hookworm vaccine discovery.
| Hookworms are voracious, blood-feeding, soil-transmitted nematode parasites. Adult hookworms infect the small intestine, causing iron-deficiency anemia and other complications. Hookworms are among the most disabling parasites of the developing world. Drugs are useful for controlling hookworm disease. However, because people often get reinfected rapidly and parasites can develop drug resistance, a vaccine that provides long-term protection would improve control and help lead to eradication. At present, there is no licensed hookworm vaccine, and progress towards a vaccine has been limited. We identified a cysteine protease in the intestine of the human hookworm Ancylostoma ceylanicum that is among the most strongly expressed genes during blood feeding and that may help digest blood and be essential for hookworm survival. Vaccination of hamsters with this cysteine protease gave high levels of protection when antigen-specific antibodies in the blood were induced. These antigen-specific antibodies also made hookworms less mobile in culture. This cysteine protease is a promising candidate for further investigation as a human hookworm vaccine antigen.
| Human hookworms (Necator americanus, Ancylostoma duodenale, and Ancylostoma ceylanicum) are soil-transmitted nematodes (STNs) that infect the small intestine and feed on blood [1]. Human STNs encompass three phylogenetically distant parasites: hookworms, large roundworms (Ascaris lumbricoides), and whipworms (Trichuris trichiura). Among human STNs, hookworms carry the highest disease burden [2]. In children, infection by hookworms causes significant growth stunting, cognitive deficiencies, malnutrition, iron-deficiency anemia and hypoproteinemia; in adults infection results in adverse birth outcomes (e.g., low birthweight babies) and reduced productivity [3–5]. It is estimated that hookworms infect ~500 million people worldwide [6], and although concentrated in Latin America, sub-Saharan Africa, and Southeast Asia, even people in impoverished regions of the United States (US) still get infected [1,7]. Once at 40% prevalence in the southern US (circa 1911), hookworm infections led to an estimated 43% reduction in future earnings of children infected and were responsible for 22% of the income gap and 50% of the literacy gap between North and South [8]. The elimination of hookworms via treatment campaigns, improved sanitation, education, and economic development undoubtedly had a major impact on the vitalization and success of the South today. Currently, hookworm disease is estimated to cause 4.1 million disability adjusted life years (DALYs) and US$139 billion in indirect economic losses each year [9]. Hookworms are the second most important parasitic cause of global anemia after malaria [10].
Mass drug administration (MDA) of benzimidazoles (albendazole, mebendazole) in school-aged children is the current control measure for hookworms [11,12]. From 1990–2013, MDA reduced hookworm prevalence by only 5.1%, compared to 25.5% reduction for A. lumbricoides [6]. Additionally, poor efficacy of mebendazole against hookworms is well known [13] and poor or reduced efficacy of albendazole against hookworms is being reported in multiple locations around the world (e.g., egg reduction rates as low as 0% in Ghana [14] and cure rates as low as 36% in Lao PDR [15]). Veterinary parasites phylogenetically and biologically similar to hookworms (e.g., the blood-feeding Haemonchus contortus) develop resistance to anthelmintic drugs frequently, rapidly, and broadly [16,17]. Water, sanitation and hygiene (WASH) is being explored as a control strategy to combine with MDA [18]. Improvements in WASH (water, sanitation, and hygiene), although important, are insufficient to tackle the enormous STN problem alone [18–20]. Having a vaccine to prevent infection from occurring in the first place would be of tremendous benefit.
Although it is widely accepted that a hookworm vaccine is needed [21], there is only a single phase 1 clinical trial underway testing two individual recombinant hookworm proteins formulated on the Th2 adjuvant Alhydrogel [22]. There are no other candidates in advanced preclinical development [23]. Because targeting infectious third staged larval (L3i) antigens carries the risk of triggering allergic reactions in previously exposed people [24], efforts are focused on adult stage antigens, namely an aspartic protease (APR1) and a glutathione S-transferase (GST1) [25]. Both APR1 and GST1 localize to the adult canine hookworm Ancylostoma caninum intestine (and non-intestinal tissues); these enzymes are thought to help digest hemoglobin and detoxify heme, respectively [25–28]. In canine and hamster models, recombinant protein immunogens from A. caninum gave 33–53% decreased hookworm burdens [26,27,29]. However, in a phase 1a trial, although rNaGST1 was safe and immunogenic in human volunteers [30], IgG had negligible neutralizing effect on rNaGST1 enzymatic activity, despite the observation that IgG from immunized mice were highly neutralizing [31]; these results suggest a decreased potential for this vaccine in human trials. It remains crucial that further and expanded efforts be undertaken to develop new hookworm vaccines. Development of hookworm vaccines, however, have been limited and lag far behind more concerted efforts, such as against malaria [23,32]. This may be in part because full genomes for human hookworms were formerly unavailable, which prevented large-scale reverse vaccinology [33] against hookworms. This situation has recently been addressed for all three species of human hookworms [34–36].
Genomics and transcriptomics for A. ceylanicum hookworm infections in hamsters is being used to identify new and potent vaccine antigen candidates [35]. Syrian hamsters are the only laboratory rodent permissive for the human hookworm life cycle, and A. ceylanicum infections in hamsters are an excellent model for hookworm infections in humans [37]. We previously identified two classes of A. ceylanicum genes, that are strongly expressed and upregulated during blood feeding, as encoding potential antigen candidates [35]: cathepsin B cysteine proteases (CPs) and small Kunitz-type protease inhibitors (SKPIs). Here, we explore the vaccine efficacy of two CPs and one SKPI (AceyCP1, AceyCPL, AceySKPI3) using the A. ceylanicum hookworm—hamster model system, and investigate functional and immunological aspects of protection with one of these vaccine candidates.
Animal experimentation was carried out under protocols approved by the University of Massachusetts Medical School Institutional Animal Care and Use Committees (IACUC). All housing and care of laboratory animals used in this study conform to the NIH Guide for the Care and Use of Laboratory Animals in Research and all requirements and all regulations issued by the United States Department of Agriculture, including regulations implementing the Animal Welfare Act (P.L. 89–544).
Male Syrian hamsters (strain HsdHan:AURA) were purchased from Envigo at 3 weeks of age and housed 4 hamsters per cage. Hamsters were provided with food and water ad libitum. Male hamsters were used for all studies because female hamsters are ~5-fold less susceptible to A. ceylanicum infection. Using females would require much larger numbers of animals to achieve adequate infection intensity, prevalence, and statistical significance.
To assess intestinal versus non-intestinal gene expression for A. ceylanicum genes (including those encoding vaccine candidates in this study), we used RNA-seq data for A. ceylanicum that we and others had previously generated from whole worms and adult male intestine [35,38]. These data were a mixture of paired- and single-end reads with varying lengths. To make cross-comparisons of these data as unbiased as possible, we trimmed all read sets to have single-end 50-nt reads, using quality_trim_fastq.pl and the arguments "-q 33 -u 50". We quality-filtered RNA-seq reads by running Trimmomatic 0.36 with the following arguments: "java -jar $TRIM/trimmomatic-0.36.jar SE -threads 7 -phred33 [input read FASTQ file] [output read FASTQ file] ILLUMINACLIP:[illumina adaptors sequence FASTA file]:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:50". The sequence file for Illumina adaptors included both sequences and reverse-complemented sequences for all of the following Illumina adaptor sequences from manufacturer's instructions: TruSeq Universal Adapter and TruSeq Adapter Index 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 18, 19, 20, 21, 22, 23, 25, and 27. We assayed expression levels against our previously published protein-coding gene/transcript set for A. ceylanicum, downloaded from the ParaSite database (release 6; ftp://ftp.wormbase.org/pub/wormbase/parasite/releases/WBPS6/species/ancylostoma_ceylanicum/PRJNA231479/ancylostoma_ceylanicum.PRJNA231479.WBPS6.CDS_transcripts.fa.gz) [39]. We quantitated gene expression from all of our quality-filtered A. ceylanicum RNA-seq data sets with Salmon 0.7.2 (https://github.com/COMBINE-lab/salmon/releases/download/v0.7.2/Salmon-0.7.2_linux_x86_64.tar.gz), generating expression values in Transcripts Per Million (TPM) and estimating mapped read counts per gene [40]. For Salmon's index program, we used the arguments "—no-version-check index—kmerLen 31—perfectHash—type quasi—sasamp 1"; for Salmon's quant program, we used the arguments "—libType A seqBias gcBias numBootstraps 100—geneMap [transcript-to-gene table]", with "—unmatedReads" specifying the 50-nt single-end data. For gene annotations, we created new Pfam motif annotations with hmmscan from HMMER version 3.1b2 [41] and the Pfam 31.0 database, using the arguments "—cut_ga -o /dev/null—tblout [table]" which invoked reliably curated domain-specific thresholds; we also generated new InterPro motif annotations with interproscan.sh from InterProScan 5.18–57.0 [42], using the arguments "-dp -hm -iprlookup -goterms". Both Pfam and InterPro motifs were computed solely for the largest isoform of each gene's predicted protein products (downloaded from ParaSite release 6; ftp://ftp.ebi.ac.uk/pub/databases/wormbase/parasite/releases/WBPS6/species/ancylostoma_ceylanicum/PRJNA231479/ancylostoma_ceylanicum.PRJNA231479.WBPS6.protein.fa.gz); these largest isoforms were extracted with get_largest_isoforms.pl using the argument "-t parasite". All other gene annotations were taken from our previous work [35]. The Perl scripts quality_trim_fastq.pl and get_largest_isoforms.pl are available from https://github.com/SchwarzEM/ems_perl.
Adult A. ceylanicum hookworms were collected from the small intestine of a day 22 post-inoculation (PI) hamster into a 1.5 mL microfuge tube, were rinsed three times with Milli-Q water, and then snap-frozen in liquid nitrogen and stored in -80°C. Tissue homogenization was performed in the same 1.5 mL microfuge tube on liquid nitrogen using a pre-chilled tapered flat end weighing spatula followed by a pre-chilled micropestle. Total nucleic acid was isolated with Nucleospin RNA kit (Machery-Nagel) according to the manufacturer’s instructions, except that on-column DNase treatment was omitted. The total nucleic acid was then treated with RNase-free DNase I (NEB) according to the manufacturer’s instructions. RNA was precipitated by addition of 1:10 vol 3 M sodium acetate and 2.5 vol ethanol with O/N storage in -20°C. The RNA pellet was washed twice with ethanol and then dissolved in 50 μL Milli-Q water. cDNA was synthesized with qScript cDNA SuperMix (Quantabio) according to the manufacturer’s instructions. PCR was performed with Platinum Taq DNA Polymerase High Fidelity (Invitrogen) according to the manufacturer’s instructions using the following primers: Aceys0154g3007cdsF and Aceys0154g3007cdsR (AceyCP1), Aceys0532g3038t1cdsF and Aceys0532g3038t1cdsR (AceyCPL), and Aceys0034g2829t1cdsF and Aceys0034g2829t1cdsR (AceySKPI3) (Table S1). The PCR products were purified with Monarch PCR and DNA Cleanup Kit (NEB) and were sequenced at GENEWIZ.
The validated CDS sequences were sent to Genscript for P. pastoris codon-optimized DNA synthesis and the CDSs without native signal peptides (AceyCP1, nt 40–1,032; AceyCPL, nt 46–1,032; and AceySKPI3, nt 52–240) were subcloned into pPICZαA in-frame with yeast α-factor signal. During subcloning polyhistidine tags were added directly 5’ to the CDSs by PCR. The plasmids were linearized with SacI and transformed into P. pastoris X-33. Single transformed colonies (confirmed by colony PCR) were inoculated into 25 mL Buffered Glycerol-complex Medium (BMGY) and grown to an OD600 of 3.0. The 25 mL cultures were used to inoculate 0.5 L BMGY cultures at an OD600 of 1.0, and these cultures were grown to an OD600 of 3.0. The BMGY cultures were centrifuged and each was resuspended in 2 L of Buffered Methanol-complex Medium (BMMY) distributed into four 2 L baffled flasks, and these cultures were grown for 4 days. After 4 days of incubation, the BMMY cultures were centrifuged, and the clarified supernatants were collected.
rAceyCP1, rAceyCPL and rAceySKPI3 were purified from X-33 culture supernatants by immobilized metal affinity chromatography using a Ni resin and column (GenScript). Proteins bound to the resin were washed with Triton X-100 to reduce endotoxin levels to <1 EU/μg. The eluates were buffer exchanged into PBS (pH 7.4) by dialysis, and then filter sterilized with 0.22 μm Millex-GP Syringe Filters. Endotoxin levels were detected by ToxinSensor Gel Clot Endotoxin Assay Kit (GenScript). Protein concentrations were determined by Bradford assay using BSA as standard (GenScript).
For SDS-PAGE, 4 μg of each protein was boiled for ~5 min in Pierce Lane Marker Reducing Sample Buffer (Thermo Fisher) and loaded into a 12% Tris-Glycine mini gel. Electrophoresis was run for ~2 hr at 100V in a Mini-PROTEAN Tetra Cell (Bio-Rad). Proteins were stained with Coomassie Blue, and the gels were imaged with a ChemiDoc XRS+ System with Image Lab Software (Bio-Rad). Molecular weights were estimated in Image Lab from the SDS-PAGE gels.
For Western blots, immediately after electrophoresis proteins were transferred to PVDF membranes using a Trans-Blot Turbo Transfer System with RTA Mini LF PVDF Transfer Kit (Bio-Rad). The membranes were blocked for 1.5 hr in blocking buffer (3% non-fat dry milk prepared in PBST). The blocked membranes were washed for 5 min twice with PBST, and then incubated for 1.5 hr in 6x-His Tag Monoclonal Antibody (HIS.H8) (Invitrogen) diluted 1:2,000 in blocking buffer. The membranes were then washed for 5 min 3 times with PBST, and then incubated for 1 hr in Goat anti-Mouse IgG (H+L) Secondary Antibody, HRP (Invitrogen) diluted 1:3,000 in blocking buffer. The membranes were washed for 5 min 3 times with PBST, and then incubated in the dark for 5 min in SuperSignal West Pico PLUS Chemiluminescent Substrate. The membranes were imaged with a ChemiDoc XRS+ System with Image Lab Software (Bio-Rad). Signal accumulation mode was used first to find the optimum exposure time (60 sec), and then the membranes were washed for 5 min with PBST and then incubated again in substrate. Finally, the membranes were imaged manually with the optimum exposure times.
Vaccines were prepared fresh for each immunization on ice in a total volume of 1.8 mL (rAceyCPL/Alhydrogel, rAceySKPI3/Alhydrogel, rAceyCP1/Alhydrogel trial 1) or 2.6 mL (rAceyCP1/Alhydrogel trial 2) in sterile 5 mL microfuge tubes. Two hundred twenty-five μg (rAceyCPL/Alhydrogel, rAceySKPI3/Alhydrogel, rAceyCP1/Alhydrogel trial 1) or 325 μg (rAceyCP1/Alhydrogel trial 2) of immunogen was diluted up to 1.575 mL (rAceyCPL/Alhydrogel, rAceySKPI3/Alhydrogel, rAceyCP1/Alhydrogel trial 1) or 2.275 mL (rAceyCP1/Alhydrogel trial 2) in PBS (pH 7.4), and 225 μL ((rAceyCPL/Alhydrogel, rAceySKPI3/Alhydrogel, rAceyCP1/Alhydrogel trial 1) or 325 μL (rAceyCP1/Alhydrogel trial 2) of Alhydrogel (InvivoGen) was added. Immunogens were adsorbed to Alhydrogel according to the manufacturer’s instructions. For Alhydrogel control, 225 μL (rAceyCPL/Alhydrogel, rAceySKPI3/Alhydrogel, rAceyCP1/Alhydrogel trial 1) or 325 μL (rAceyCP1/Alhydrogel trial 2) of Alhydrogel was added to 1.575 mL (rAceyCPL/Alhydrogel, rAceySKPI3/Alhydrogel, rAceyCP1/Alhydrogel trial 1) or 2.275 mL (rAceyCP1/Alhydrogel trial 2) of PBS (pH 7.4). Final doses were 25 μg of protein and 250 μg of aluminum content (per dose).
Hamsters were injected subcutaneously (SC) with insulin syringes (BD) three times with two-week intervals in the scruff of the neck with 200 μL of vaccine. One week after the final immunization, hamsters were separated into individual cages, and the individual cages were randomly arranged on the shelves. Twelve days after the final immunization, peripheral blood was collected by saphenous venipuncture using PrecisionGlide 20 G x 1” hypodermic needles (BD) and SAFE-T-FILL Capillary Blood Collection Tubes–Serum (RAM Scientific). Blood was allowed to clot for at least 30 min at room temperature before centrifugation. The collected serum was stored in -20°C until ELISA. Thirteen days after the final immunization, hamsters were weighed and blood was collected again, but into SAFE-T-FILL Capillary Blood Collection Tubes–EDTA (RAM Scientific). Blood hemoglobin concentrations (g/dL) were measured with a STAT-Site M Hgb Hemoglobin Analyzer and Test Cards (Stanbio).
Exactly two weeks after the final immunization hamsters were inoculated with ~150 A. ceylanicum L3i by oral gavage. L3i were obtained by coproculture of feces collected from infected hamsters, and had been stored in the dark at room temperature for <2 weeks in BU buffer (50 mM Na2HPO4, 22 mM KH2PO4, 70 mM NaCl, pH 6.8) plus PSF (100 U/mL penicillin; 100 μg/mL streptomycin; 0.25 μg/mL amphotericin B) before inoculations. This A. ceylanicum line was originally obtained from Dr. John Hawdon at George Washington University. On day 20 PI, hamsters were weighed and blood hemoglobin concentrations measured as before. On day 22 PI, hamsters were placed on fecal collection wires overnight. Two layers of moistened paper towels were placed in the bottoms of the cages underneath the fecal collection wires. On day 23 PI, hamsters were euthanized by CO2 overdose and cervical dislocation (according to IACUC protocol). Small intestines were removed, longitudinally sectioned, and incubated in Hank’s Balanced Salt Solution (HBSS; Thermo Fisher) for 45 min at 37°C, 5% CO2. Hookworm burdens were counted under a stereomicroscope. Fecal pellets were collected from the cage of each hamster, and FECs were measured using a McMaster chamber (Hausser Scientific).
Immunogens were coated overnight at 4°C onto Nunc MaxiSorp flat-bottom 96-well plates at 5 μg/ml in carbonate/bicarbonate (100 mM), pH 9.6 coating buffer (100 μL/well). Wells were washed three times with PBST (200 μL/well), and then blocked for 1.5 hr in blocking buffer (5% non-fat dry milk in PBST) (200 μL/well) at room temperature. Wells were washed two times with PBST (200 μL/well), and then hamster sera serially diluted in blocking buffer was incubated (100 μL/well) for 1.5 hr at room temperature. Wells were washed three times with PBST (200 μL/well), and then Peroxidase AffiniPure Goat Anti-Syrian Hamster IgG (H+L) (Jackson ImmunoResearch) was diluted 1:5,000 in blocking buffer and incubated in the wells (100 μL/well) for 1.5 hr at room temperature. Wells were washed three times with PBST (200 μL/well), and 100 μL of 1-Step Ultra TMB-ELISA Substrate Solution (Thermo Fisher) was incubated in the wells for 30 min. Then 100 μL of sulfuric acid (2 M) was added to the wells, and A450 was measured with a Tecan Safire plate reader.
In multiple pilot experiments, we tested Rabbit anti-Syrian hamster IgM-HRP (Rockland), Goat anti-Mouse IgA-HRP (Thermo Fisher; [43]) and Goat anti-Mouse IgE-HRP (Thermo Fisher) in ELISAs to soluble A. ceylanicum hookworm extract (HEX; [43]) using serum from infected and drug-cleared hamsters. Anti-Syrian hamster IgM-HRP gave extremely high background to HEX with serum from uninfected hamsters. Neither anti-Mouse IgA-HRP or anti-Mouse IgE-HRP reacted to HEX with serum from infected hamsters, while Goat anti-Syrian hamster IgG (mentioned above) reacted strongly to HEX only with serum from infected hamsters. Thus, only Goat anti-Syrian hamster IgG was useful for serum ELISAs.
A. ceylanicum adult hookworms were collected from the small intestines of initially naïve hamsters on day 17 PI. Small intestines were longitudinally sectioned and incubated for 2 hr in HBSS pre-warmed at 37°C in a mini Baermann Funnel apparatus. Every 20–30 min the small intestines were moved around with forceps. The HBSS containing motile hookworms that had migrated through the wire mesh and settled at the bottom of the funnel was poured into a petri dish, and the motile hookworms were hand-picked with a worm picker into Milli-Q water, and rinsed three times. Hookworms were hand-picked into individual wells of a 96-well plate (two hookworms per well) containing 50 μL of HCM with 50% heat-inactivated fetal bovine/calf serum [44] replaced by hamster serum (mHCM: 49.5% RPMI 1640 Medium containing L-glutamine without Phenol Red Indicator [Thermo Fisher]; 49.5% hamster serum; 1% 100X PSG [100 U/mL penicillin; 100 μg/mL streptomycin; 0.292 mg/mL L-glutamine; Thermo Fisher]). Each hamster serum group included three wells (two hookworms/well for a total of six hookworm adults scored per group), and the average motility for the three wells was calculated using a standard 3–0 motility index assay (3 = highly motile; 2 = less motile; 1 = motile only when stimulated by touch; 0 = immotile) [45–47]. Motility was monitored for 76 hr (once per day). To address any concerns about subjectivity, each replicate well for each condition was randomized in the setup so that the investigator was blinded as to which well contained which condition during the scoring process.
Another set of hamsters was vaccinated with rAceyCP1/Alhydrogel and Alhydrogel control exactly as before in the vaccine trials. Serum IgG responses were evaluated by ELISA as before using A450 readings at 1:100 serum dilutions. Two weeks after the final immunization hamsters were euthanized as before. Spleens were removed with ethanol-sterilized forceps and transferred to 5 mL DMEM-10 cell culture medium (89.5% DMEM [Dulbecco's Modification of Eagle's Medium; Mediatech]; 9.5% fetal calf serum [Thermo Fisher]; 1% 100X PSG; sterilized with 0.22 μm filter; stored in 4°C) in 60-mm petri dish on ice. Each spleen was cut into three pieces with ethanol-sterilized surgical scissors. Each spleen piece was smashed individually between two frosted microscope slides and rinsed back into the petri dish, removing any remaining solid tissue left on the slide with forceps. Each spleen material in DMEM-10 was passed through syringe needle series of 18, 22, and 26 G (BD) to prepare the splenocyte suspension that was collected into a 15-mL conical centrifuge tube pre-chilled on ice. Splenocyte suspensions were centrifuged at 1,500 rpm for 5 min at 4°C. The supernatant was discarded and cell pellet resuspended in 1 mL ACK Lysing Buffer (Thermo Fisher) and incubated for 10 min at room temperature. Immediately after, 9 mL of DMEM-10 was added and then centrifuged as before. Each splenocyte suspension was resuspended in 2 mL DMEM-10, and 1 mL was transferred into each of 2 different wells of a 48-well plate. For each splenocyte suspension, 25 μg of rAceyCP1 was added to one well and PBS (pH 7.4) to the other well. Splenocytes were stimulated for 48 hr at 37°C, 5% CO2.
Stimulated splenocytes were transferred to 1.5 mL microfuge tubes and centrifuged as before. Supernatants were discarded, cell pellets were resuspended in lysis buffer and vortexed, and RNA was isolated as before. Either 10 μg (γ-actin, IFN-γ, IL17A, IL10, and TGF-β) or 50 μg (IL4, IL5, IL13, and IL21) of RNA (as determined in primer efficiency tests) was used as template for qRT-PCR using qScript One-Step SYBR Green qRT-PCR Kit (Quantabio) according to the manufacturer’s instructions. Protocol: 49°C for 10 min, 95°C for 5 min, 35 cycles of 95°C for 15 sec and 60°C for 45 sec. Amplification specificities were verified by melting curve analysis. Melting curve protocol: 95°C for 1 min, 55°C for 10 sec and a slow temperature ramp from 55 to 95°C. The primer sets (Table S1; [48]) were pre-validated with mean ± S.D. amplification factors of 2.0 ± 0.3 in primer efficiency tests using the same sample type. qRT-PCR was performed on an Eppendorf Mastercycler RealPlex2. Four technical replicates were included for every primer set on each RNA sample. Minus reverse transcriptase reactions were also run on every RNA sample for every primer set, and this indicated that the RNA samples had undetectable levels of gDNA. γ-actin was used as reference. We had hoped to look at even more cytokines by qRT-PCR, but we did not have RNA for all biological replicates. Data were analyzed using the 2-ΔΔCT method [49].
All data analyses were plotted in Prism 7 (GraphPad Software). For serum IgG endpoint titers, plotted are the inverses of the final serum dilutions for each serum sample from each hamster within each vaccine group that gave an A450 reading that is >3 S.D. from the mean A450 reading for naïve hamster sera (n = 8) at the same dilution. For hookworm burden, mean indicates the mean hookworm burden amongst all hamsters in each vaccine group. For FECs (eggs/g), mean indicates the egg count per group from all cages in the group at a given time point. For Δweight and Δhemoglobin, mean indicates the mean Δweight and Δhemoglobin amongst all hamsters in each vaccine group. For motility index, mean indicates the mean motility amongst all wells (2 hookworms/well, 3 wells/group) in the group at a given time point. For serum IgG A450, plotted are the raw A450 readings for each serum sample from each hamster per vaccine group at 1:100 dilution with average Alhydrogel background subtracted. For relative fold change in mRNA, mean indicates the mean relative fold change in mRNA in splenocytes amongst all hamsters in each vaccine group.
Statistical 2-sample comparisons between experimental and control vaccine groups were carried out by one-tailed Mann-Whitney, with the hypothesis that successful vaccination will result in decreases in infection parameters (worm burdens, fecal egg counts) and improvements in measurements of sequelae (weight, hemoglobin). Nearly identical statistical results were obtained using student’s t test. All comparisons between multiple experimental groups with control (i.e., Alhydrogel alone) group were carried out using one-way ANOVA with Dunnett’s post-hoc test, comparing each group to control. Significant correlations in linear regression analyses (i.e., if slopes are significantly non-zero) were determined by F test.
Transcriptomic analyses revealed that the hookworm A. ceylanicum upregulates proteases and protease inhibitors during infection, which are likely to be important for successful parasitism [35]. Proteases are thought to help hookworms dig through their host's tissues, destroy proteins needed for the host's immune response, and digest proteins in the host's blood [50]. Protease inhibitors are thought to block host proteases needed for the immune response as well [51]. These proteases and protease inhibitors therefore defined an initial set of vaccine candidates and included putatively secreted cathepsin B-like proteases (CPs), which have homologs in other strongylids, and a previously undescribed family of putatively secreted small Kunitz-type protease inhibitors (SKPIs) that are strongly upregulated in adult hookworms [35].
We successfully cloned full-length cDNAs for A. ceylanicum CP1 (rAceyCP1; AceyCP1 [genomic name, Acey_s0154.g3007]), AceyCPL (rAceyCPL [Acey_s0532.g3038]), and AceySKPI3 (rAceySKPI3 [Acey_s0034.g2829]) (Fig 1A). We expressed the proteins in Pichia pastoris and purified them using standard protocols (Fig 1B–1E; see Methods for details). P. pastoris expression was used for previous hamster-hookworm vaccine trials [52,53]. Since the hookworm intestine is a key target of protective vaccine antigens, we examined expression of these genes in published male hookworm intestinal transcriptomic data (Table S2) [38]. AceyCP1 is very strongly expressed in the male intestine at 11,600 transcripts per million (TPM), making it 1.2% of all transcripts in that tissue. AceyCPL and AceySKPI3 are present but considerably more weakly expressed in the male hookworm intestine at 52 TPM and 1.7 TPM respectively. Neither AceyCP1 nor AceyCPL are expressed in L3i (0.15 and 0 TPM respectively). AceySKPI3 has modest expression in L3i (14 TPM), but this is less than or comparable to expression levels for the A. ceylanicum homologs of APR1 and GST1 (AceyAPR1 [Acey_s0242.g3404], 744 TPM; AceyGST1 [Acey_s0110.g143], 44 TPM), which in N. americanus are considered acceptable vaccine candidates.
By SDS-PAGE and Western blot, three bands were observed for rAceyCP1 at ~42, ~44 and ~46 kDa (Fig 1B and 1C), two bands were observed for rAceyCPL at ~40 and ~42 kDa (Fig 1B and 1C), and two bands were observed for rAceySKPI3 at ~12 and ~19 kDa (mostly single band at ~12 kDa; Fig 1D and 1E). AceyCP1 and AceyCPL are proenzymes containing N-terminal pro-regions and a CP domain (Fig 1A). Two N-glycosylation sites were identified in AceyCP1 with NetNGlyc 4.0 (http://www.cbs.dtu.dk/services/NetNGlyc) [54], both of which are located in its CP domain near putative active site residues (Fig 1A). One N-glycosylation site was identified in AceyCPL within its pro-region (Fig 1A). These predicted N-glycosylation sites match up with the observed banding patterns (Fig 1B and 1C). NetNGlyc 4.0 did not identify any potential N-glycosylation sites in AceySKPI3 (Fig 1A). Treatment of rAceyCP1 and rAceyCPL with endoglycosidase (Endo) H resulted in shifts to a single ~42-kDa band for rAceyCP1 and a single ~40-kDa band for rAceyCPL (Fig 1B and 1C). According to these results, rAceyCP1 consists of unglycosylated, monoglycosylated, and biglycosylated forms, while rAceyCPL consists of unglycoslyated and monoglycosylated forms. The weak ~19-kDa upper band in rAceySKPI3 could be a form of dimer or a monomer with other post-translational modification(s). Densitometric analysis determined all three recombinant proteins to be >95% pure.
rAceyCPL was adsorbed to Alhydrogel and injected SC into Syrian hamsters three times at two-week intervals (Fig 2, timeline). Twelve days after the final immunization (Fig 2), rAceyCPL serum IgG titer was determined by indirect enzyme-linked immunosorbent assay (ELISA). All rAceyCPL/Alhydrogel vaccinates (8/8) had serum IgG titers above background in Alhydrogel controls ranging from 100,000 to ≥500,000 (Fig 3A). Hamsters were infected with ~150 A. ceylanicum third stage infective larvae (L3i) exactly two weeks after the final immunization (Fig 2). At day 23 post-inoculation (PI), the number of hookworms in the small intestines (hookworm burden) were counted at necropsy, and feces were collected for fecal egg counts (FECs; hookworm eggs shed per g of feces). Mean hookworm burden and FEC were statistically similar between rAceyCPL/Alhydrogel vaccinates and Alhydrogel controls (Fig 3B and 3C), indicating that parasitism was unaffected by vaccination. As markers for the clinical pathology caused by hookworm infection, changes in weight and hemoglobin (Δweight and Δhemoglobin) were evaluated (on day 20 post-inoculation (PI) prior to necropsy; Fig 2). Consistent with hookworm burden and FEC, mean Δweight and Δhemoglobin were statistically similar between rAceyCPL/Alhydrogel vaccinates and Alhydrogel controls (Fig 3D and 3E), indicating that clinical pathology was also unaffected by vaccination.
All rAceySKPI3/Alhydrogel vaccinates (8/8) had serum IgG titers above background in Alhydrogel controls ranging from 100,000 to, remarkably, ≥10,000,000 (Fig 4A). However, as with rAceyCPL/Alhydrogel, rAceySKPI3/Alhydrogel vaccination did not significantly alter hookworm parasitism (Fig 4B and 4C), or clinical pathology (Fig 4D and 4E) compared to Alhydrogel controls. Thus, these results indicate that although both rAceyCPL/Alhydrogel and rAceySKPI3/Alhydrogel induce high serum IgG titers above background with 100% responder rates, these antigens are not protective.
Two independent vaccine trials were conducted for rAceyCP1, the first with eight animals per group and the second with 12 animals per group. In trial 1, 5/8 vaccinates (62.5%) had rAceyCP1-specific serum IgG titers above background in Alhydrogel controls ranging from 4,000–10,000 (Fig 5). In trial 2, 7/12 vaccinates (58.3%) had rAceyCP1 serum IgG titers above background in Alhydrogel controls ranging from 2,000–20,000 (Fig 5). In trial 1 and trial 2, hookworm burdens were decreased in rAceyCP1/Alhydrogel vaccinates compared to Alhydrogel control vaccinates by a mean of 31% and 19% respectively, achieving statistical significance in trial 1 (Fig 6A) and nearly achieving statistical significance in trial 2 (Fig 6A). Significant decreases in FECs were seen in both trials (Fig 6B). In trial 1 and trial 2, respectively, FECs were significantly decreased in rAceyCP1/Alhydrogel vaccinates compared to Alhydrogel control vaccinates by a mean of 46% and 26%.
Vaccination with rAceyCP1 also generally led to improvements of hookworm sequelae. Δweight was signficantly improved in rAceyCP1/Alhydrogel vaccinates by 7.6 g (Syrian hamsters weigh around 115 g at this age; ~12 weeks) compared to Alhydrogel controls in trial 1 (Fig 6C), and there was a trend of 2.8 g toward improved Δweight in trial 2 (Fig 6C). The Δhemoglobin was statistically similar between rAceyCP1/Alhydrogel vaccinates and Alhydrogel controls in both trials (Fig 6D).
In both vaccine trials, we noticed that rAceyCP1 serum IgG responders were qualitatively more protected than nonresponders (Fig 6; compare blue and orange data points). We hypothesized that the responders were protected from infection and sequelae whereas the nonresponders were not. Thus, we reanalyzed them as separate groups, and compared them to Alhydrogel controls. Strikingly, in trial 1, hookworm burden was dramatically and significantly decreased in responders by 54% compared to Alhydrogel controls (Fig 7A), whereas nonresponders showed no protection compared to Alhydrogel controls (Fig 7A). In trial 2, hookworm burden was dramatically and significantly decreased in responders by 40% compared to Alhydrogel controls (Fig 7A), whereas nonresponders showed no protection compared to Alhydrogel controls (Fig 7A). Protection from infection based on hookworm burdens was mirrored in FECs. In trial 1, FECs were dramatically and significantly decreased in responders by 66% compared to Alhydrogel controls (Fig 7B), whereas nonresponders showed no protection compared to Alhydrogel controls (Fig 7B). Accordingly, in trial 2, FECs were dramatically and significantly decreased in responders by 54% compared to Alhydrogel controls (Fig 7B), whereas nonresponders showed no protection compared to Alhydrogel controls (Fig 7B). Sequelae based on Δweight were also improved. In trial 1, responders had a dramatically and significantly improved Δweight of 9.6 g compared to Alhydrogel controls (Fig 7C), whereas nonresponders showed no improved Δweight compared to Alhydrogel controls (Fig 7C). In trial 2, responders had a dramatically and significantly improved Δweight of 6.0 g compared Alhydrogel controls (Fig 7C), whereas nonresponders showed no improved Δweight compared to Alhydrogel controls (Fig 7C). In trials 1 and 2, responders had improved Δhemoglobin by 2.2 g/dL and 2.0 g/dL respectively, although neither was statistically significant (Fig 7D).
A complete summary of the efficacy results from rAceyCP1/Alhydrogel vaccine trials 1 and 2 is given in Table 1, with the two immunogens currently being tested in phase 1 clinical trials included for comparisons. Linear regression analysis was performed on the data to investigate the correlation between rAceyCP1 serum IgG titer and changes in sequelae and parameters of infection. These analyses determined that, in both vaccine trials, rAceyCP1 serum IgG titer highly significantly correlated with all four measures of protection, including Δhemoglobin (Fig 8). Moreover, serum IgG titers of 10,000–20,000 in the two trials (n = 4) gave a mean of 64.6% decreased hookworm burden and 76.9% decreased FECs compared to Alhydrogel controls. These findings indicate that when rAceyCP1 serum IgG titer is sufficiently induced, rAceyCP1/Alhydrogel is a highly protective hookworm vaccine.
We hypothesized that serum IgG might be at least partly responsible for vaccine-induced protections in rAceyCP1/Alhydrogel serum IgG responders. As a first test for this hypothesis, we incubated adult A. ceylanicum hookworms obtained from naïve hamsters in a modified hookworm culture medium (mHCM) containing 50% antisera from rAceyCP1/Alhydrogel serum IgG responders and nonresponders, as well as rAceyCPL/Alhydrogel vaccinates and Alhydrogel controls. We scored motility over a period of 76 hr; scoring was performed blind relative to treatment condition to prevent bias. Hookworm motility was significantly reduced by 76 hr in rAceyCP1/Alhydrogel serum IgG responder antisera by a mean of 33.3% compared to in Alhydrogel control sera (Fig 9). Conversely, hookworm motilities in serum IgG non-responder antisera and rAceyCPL/Alhydrogel antisera were reduced by just 5.6% and were statistically similar to Alhydrogel control sera (Fig 9), demonstrating specificity of the activity seen above. Thus, the unique toxicity of antisera alone from rAceyCP1/Alhydrogel serum IgG responders is consistent with the observed protections.
Vaccination with rAceyCP1/Alhydrogel induced protection that was highly correlated with serum IgG titer, and found to intoxicate/reduce adult hookworm motility in vitro. In order to gain an understanding of the adaptive, cellular immune responses induced in rAceyCP1/Alhydrogel vaccinates, eight hamsters were vaccinated with rAceyCP1/Alhydrogel exactly as in trials 1 and 2. Peripheral blood was collected one week after the final immunization to measure serum IgG responses. Necropsy was performed exactly two weeks after the final immunization (i.e., at the exact time vaccinates were infected with L3i in trials 1 and 2), and splenocyte suspensions were prepared. rAceyCP1/Alhydrogel and Alhydrogel control splenocytes were stimulated ex vivo with rAceyCP1 or PBS. Total RNA was then isolated and used directly as template for quantitative real-time reverse transcription (qRT)-PCR using pre-validated primer sets [48] for the following cytokines: IFN-γ (Th1); IL4, IL5, IL13 (Th2); IL17A (Th17); IL21 (Th17/Tfh); IL10, TGF-β (Treg) (S1 Fig).
rAceyCP1/Alhydrogel vaccinates (4/8) gave rAceyCP1 serum IgG responses (raw A450 readings) above background in Alhydrogel controls in serum diluted 1:100 (Fig 10A). rAceyCP1-stimulated splenocytes from rAceyCP1/Alhydrogel IgG responders (n = 4) resulted in elevated levels of all three Th2 cytokine mRNAs (IL4, IL5 and IL13) that were significantly greater than in stimulated splenocytes from Alhydrogel controls (n = 5; Fig 10B). No other cytokines were elevated in splenocytes from rAceyCP1/Alhydrogel serum IgG responders at levels that were significantly greater than in Alhydrogel control, indicating a highly specific, canonical Th2 cytokine recall response. Also, all cytokines in stimulated splenocytes from rAceyCP1/Alhydrogel IgG nonresponders were statistically similar to Alhydrogel control (Fig 10B).
We demonstrate here that the highly expressed, substantially intestine-enriched cathepsin B cysteine protease, AceyCP1, is a promising protective antigen candidate for vaccination against A. ceylanicum hookworm infection in Syrian hamsters. Two antigens (AceyCPL, AceySKPI3) with much lower transcript levels in the adult A. ceylanicum intestine compared to AceyCP1 were not protective (Figs 3 and 4). These data are supportive of the value of using intestinal expression to prioritize antigen candidates.
Other intestinal CPs were shown to be protective against other blood-feeding gastrointestinal nematodes (canine hookworm A. caninum, small ruminant parasite H. contortus, N. americanus hookworms) in animal hosts (dogs, sheep, hamsters respectively) [29,55–58], albeit not to the extent that we report here for AceyCP1 (e.g., vaccination with N. americanus CP2 gave 29% reduction in worm burdens in hamsters). Interestingly, intestinal CPs were shown not to be protective against an STN parasite of cattle that does not ingest blood, Ostertagia ostertagi [59,60], whereas CPs in O. ostertagi ES products were shown to be protective [59,61]. Moreover, recently, IgG induced by vaccination with whole worm extracts of Ascaris suum (another non-blood-feeding STN) was determined (in IgG transfer experiments) to be protective against infection [62]. Furthermore, O. ostertagi intestinal CP components in a larger native protein complex did cross-protect against H. contortus in vaccinated sheep [60]. These important CP enzymes are therefore vulnerable antigens that can be accessed in the gut only by immune factors that are ingested in the blood. Previous studies of intestinal CPs of blood-feeding STNs have localized the CPs within the worm intestinal lumen with serum IgG from vaccinated hosts, and serum IgG has been implicated as the effector component that neutralizes CP digestion of the blood meal [55,56,58,63,64]. However, our finding that AceyCPL vaccination gave a strong immune response but no protection confirms that not all cathepsin B cysteine protease antigens are useful for vaccination.
rAceyCP1/Alhydrogel serum IgG responders had dramatically decreased hookworm burdens, FECs, and weight losses in two different vaccine trials compared to Alhydrogel controls, whereas nonresponders did not (Fig 7, Table 1). rAceyCP1/Alhydrogel is a highly protective vaccine, reducing hookworm burdens and FECs by ~50% and ~60%, respectively, in responders (Fig 7, Table 1), and ~65% and ~77%, respectively, when serum IgG titers were ≥10,000. Although there were moderately decreased blood losses in responders compared to Alhydrogel controls (Fig 7, Table 1), these results were not significant. On the other hand, blood loss was highly negatively correlated with rAceyCP1 serum IgG titer (Fig 8).
rAceyCP1-induced protection is among the highest seen to date of current hookworm antigens (Table 1). Consistently, AceyCP1 is expressed 220 and 6,800 times more strongly in the A. ceylanicum intestine compared to AceyAPR1 and AceyGST1 (11,600 TPM versus 52 and 1.7 TPM, respectively) (Table S2). Furthermore, AceyCP1 is almost completely unexpressed in L3i (0.15 TPM), while AceyAPR1 and AceyGST1 have significant expression levels in L3i (744 and 44 TPM, respectively). Thus, AceyCP1 is unlikely to be recognized by IgE and to induce urticarial reactions in previously exposed people from hookworm endemic regions [24], since L3i is the predominant IgE-reactive stage [65]. Although we cannot rule out effects from different adjuvants, hosts, and Ancylostoma species of vaccine studies carried out to date, rAceyCP1 is clearly a positive addition to the hookworm vaccine antigen arsenal.
Antisera from rAceyCP1/Alhydrogel responders, but not from nonresponders or rAceyCPL/Alhydrogel vaccinates, reduced adult A. ceylanicum motility in vitro as early as 24 hr, and was significant by 76 hr (Fig 9). These results are consistent with a model whereby neutralizing serum IgG inhibit AceyCP1 blood digestion within the hookworm gut, thus leading to starvation—comparable to models for other gut protease antigens of blood-feeding STNs [25,27,56,58,63].
We observed Th2-specific cytokine recall responses in stimulated splenocytes from rAceyCP1 responders ex vivo that highly correlated with serum IgG titer (Fig 10). Our findings reinforce the notion that neutralizing IgG in serum (likely assisted by Th2 cytokines) is the protective component to gut antigens in blood-feeding STNs.
Vaccinated hamsters in trials 1 and 2 gave ~60% responder rates (Fig 5). Clearly, a major focus in the future will be to improve responder rates, and rAceyCP1 has the potential to teach us more about how to make a better hookworm vaccine. The ~40% nonresponder rate could be due to a number of factors. First, rAceyCP1 was expressed as a secreted, partially glycosylated protein in P. pastoris yeast consisting of unglycosylated, monoglycosylated and biglycosylated forms with relative abundances as follows: biglycosylated > unglycosylated > monoglycosylated (Fig 1A–1C). Glycosylation of rAceyCP1 may play an important role in the IgG nonresponder rate and/or intermediate titers in responders. It has been hypothesized that proper antigen glycosylation plays a critical role in vaccine efficacy of antigens against other gastrointestinal nematode parasites (reviewed in [66]). Expressing rAceyCP1 in other systems with other glycosylation patterns is therefore an important next step. Another potential contributing factor to the ~40% nonresponder rate and intermediate rAceyCP1 serum IgG titers in responders is major histocompatibility complex (MHC) class II (MHC-II) allelic variation in the outbred hamster colony, which would give variable Th cell responses [67]. Envigo maintains their hamster colony as outbred by non-sib matings and claims that due to high litter average and reproductive vigor, there should be a diversity of MHC alleles segregating in the colony.
In conclusion, rAceyCP1/Alhydrogel is a highly protective vaccine against A. ceylanicum hookworm infection in Syrian hamsters when serum IgG is sufficiently induced, which likely requires help from Th2 cytokines. Serum IgG may target AceyCP1 within the gut, thereby neutralizing hookworm digestion of the blood meal. Efforts are underway to improve rAceyCP1 neutralizing IgG responder rate and titers, efforts that will ultimately improve translation to humans. Hookworm is among the most disabling parasitic diseases of the developing world, and our findings provide important information for advancing hookworm vaccine development.
|
10.1371/journal.pgen.1006446 | Mechanisms for Complex Chromosomal Insertions | Chromosomal insertions are genomic rearrangements with a chromosome segment inserted into a non-homologous chromosome or a non-adjacent locus on the same chromosome or the other homologue, constituting ~2% of nonrecurrent copy-number gains. Little is known about the molecular mechanisms of their formation. We identified 16 individuals with complex insertions among 56,000 individuals tested at Baylor Genetics using clinical array comparative genomic hybridization (aCGH) and fluorescence in situ hybridization (FISH). Custom high-density aCGH was performed on 10 individuals with available DNA, and breakpoint junctions were fine-mapped at nucleotide resolution by long-range PCR and DNA sequencing in 6 individuals to glean insights into potential mechanisms of formation. We observed microhomologies and templated insertions at the breakpoint junctions, resembling the breakpoint junction signatures found in complex genomic rearrangements generated by replication-based mechanism(s) with iterative template switches. In addition, we analyzed 5 families with apparently balanced insertion in one parent detected by FISH analysis and found that 3 parents had additional small copy-number variants (CNVs) at one or both sides of the inserting fragments as well as at the inserted sites. We propose that replicative repair can result in interchromosomal complex insertions generated through chromothripsis-like chromoanasynthesis involving two or three chromosomes, and cause a significant fraction of apparently balanced insertions harboring small flanking CNVs.
| By traditional cytogenetic techniques, the incidence of microscopically visible chromosomal insertions was estimated to be 1 in 80,000 live births. More recently, by aCGH in conjunction with FISH confirmation of the aCGH findings, insertion events were demonstrated to occur much more frequently (1 in ~500 individuals tested). Although frequently detected, little is known about the molecular mechanisms of their formation. In this study, we identified 16 individuals with complex chromosomal insertions among 56,000 individuals tested at Baylor Genetics using clinical microarray analysis (CMA) and FISH. Custom high-density aCGH was performed on 10 individuals with available DNA, and breakpoint junctions were fine-mapped at nucleotide resolution by long-range PCR and DNA sequencing in 6 individuals to glean insights into potential mechanisms of formation. In addition, we analyzed 5 families with apparently balanced insertion in one parent detected by FISH analysis and found that 3 parents had additional small copy-number variants (CNVs) at one or both sides of the inserting fragments as well as at the inserted sites. We propose that replicative repair can result in interchromosomal complex insertions generated through chromothripsis-like chromoanasynthesis involving two or three chromosomes, and cause a significant fraction of apparently balanced insertions harboring small flanking CNVs.
| Chromosomal insertion occurs when a segment of one chromosome is translocated and inserted into an interstitial region of another non-homologous chromosome (interchromosomal insertion), or into a different region of the same chromosome (intrachromosomal insertion). Insertions are considered as complex chromosomal rearrangements (CCRs) since they require at least three chromosome breakage events. [1] Chromosomal insertions are also considered as complex genomic rearrangements (CGRs) as they consist of more than one simple rearrangement, and have two or more DNA breakpoint junctions. [2, 3] By cytogenetic techniques, the incidence of microscopically visible insertions was estimated to be 1 in 80,000 live births.[4] More recently, by array-comparative genomic hybridization (aCGH) in conjunction with fluorescence in situ hybridization (FISH) confirmation of the aCGH findings, insertion events were demonstrated to occur much more frequently, with estimated incidence of 1 in 500[1] or 1 in 563[5] individuals tested. Another study demonstrated that ~2.1% of apparently de novo, interstitial CNVs were actually consequences of imbalances resulted from parents with balanced insertions.[6] These data highlight the importance of identifying such parental genomic information for reproductive counseling and potential recurrence risk estimates.
Phenotypic consequences of insertions vary, depending on the size, gene content and orientation of the inserted fragment, in addition to possible disruption or dysregulation of a gene or topologically associating domain (TAD) at the inserted genomic locus. Complex insertions are defined as insertions generated by more than three DNA breakages and joining events.[1, 7] Usually, additional copy-number gain or loss is observed at the inserted site for these events complicating the interpretation of potential phenotypic consequences observed. Little is known regarding the molecular mechanisms for the formation of insertions; particularly with regards to the mechanism(s) of formation of complex insertions. Thus, we aimed to elucidate the potential underlying mechanisms generating complex insertions. Surprisingly, we observed complex insertions as part of apparent chromothripsis-like, chromoanasynthesis events involving two or three chromosomes.
Chromothripsis was first described as a catastrophic phenomenon in cancer genomes and observed as highly complex somatic rearrangements, with a distinct pattern of frequent oscillations between neutral and deleted copy-number states and seemingly focused on one chromosome.[8] A similar apparent chromosome shattering mechanism has been observed as de novo mutations in individuals with neurodevelopmental abnormalities, and this type of germline chromothripsis involves complex balanced rearrangement among several chromosomes.[9] These events may appear as balanced rearrangements by conventional metaphase chromosome analysis. Both somatic and germline chromothripsis were proposed to be caused by a similar chromosome shattering mechanism that undergoes repair through non-homologous end-joining (NHEJ).[10, 11] A third type of chromothripsis-like event, defined as chromoanasynthesis, was observed as de novo constitutional CGRs involving region-focused copy-number changes including duplications and triplications. These chromoanasynthesis events were proposed to be generated through replication-based mechanisms, such as fork stalling and template switching and/or microhomology-mediated break-induced replication (FoSTeS/MMBIR) with iterative template switching resulting in extensive complexity.[12] The molecular analysis and findings in complex insertions we report here mostly resembled the patterns observed in constitutional genomic chromoanasynthesis events.
In this study, we identified 16 individuals with distinct complex insertions among 56,000 individuals tested at Baylor Genetics (BG) using clinical aCGH and FISH. We fine-mapped DNA breakpoint junctions in 6 complex insertions at nucleotide resolution, and three of them resembled chromoanasynthesis events with multiple chromosomes involved. In addition, we analyzed 5 families with unbalanced insertions detected in probands and inherited from parents with apparently balanced insertion detected by FISH analysis. We found that 3 parents had additional small CNVs at one or both sides of the inserting fragments as well as at the inserted sites likely generated during formation of the structural variant. We propose that these events are due to DNA replicative repair errors generated by replication-based mechanism(s) using iterative template switching.[3]
Previously, we demonstrated that by performing confirmatory FISH of the copy-number gains identified in clinical chromosome microarray analysis (CMA) testing, some duplications were shown not to be represented by tandem duplication events, but were rather translocated and inserted at another locus in the genome.[1] This approach allowed the discovery of 40 individuals with insertions among the 18,000 individuals tested in the CMA laboratory at BG from July 2005 to January 2009. Among these 40 individuals, 8 were found to carry complex insertions (S1 Table, individuals Cplex1–8).[1] In this study, we expanded the cohort to 56,000 individuals tested from July 2005 to November 2014, and identified an additional 76 individuals with chromosome insertions (out of the subsequent 38,000 individuals tested), therefore resulting in the incidence of insertions being consistently about 1 in 500. This incidence is likely underestimated given that some of the insertions are too small to be verified by FISH. Among these latter 76 individuals, we identified another 8 individuals with complex insertions (S1 Table, individuals Cplex9–16).
Among the 16 individuals with complex insertions, 2 are intrachromosomal insertions (Cplex1 and Cplex2), and the remaining 14 are interchromosomal insertions (Cplex3–16) (S1 Table). Cplex2 was previously demonstrated in detail with all proposed breakpoint junctions mapped (BAB3105 from Ref. 12) in the paper that first defined the chromoanasynthesis phenomenon and thus was excluded from the current study. For the remaining 15 individuals, we were able to obtain genomic DNA from 10 individuals (Cplex1, 3, 4, 5, 6, 7, 9, 11, 12, and 16) and repeated the CMA testing using the latest version (Baylor CMA version 10.2 oligo).[13] To map the breakpoint junctions to nucleotide resolution, we further designed high-density custom aCGH specifically targeting the inserted fragment and the potential inserting loci in 8 individuals based on the CMA results (Cplex3, 4, 5, 6, 7, 9, 11 and 12). By long-range PCR with Sanger sequencing, we were able to map (or partially map) the breakpoint junctions to nucleotide resolution in 6 individuals (Cplex4, 5, 6, 9, 11 and 12). For the remaining 4 individuals, we were unable to map breakpoint junctions, probably due to the limitations of the techniques applied in this study and potential further complexity at those breakpoints. Parental samples were not available for these 16 individuals.
Among the 6 individuals with breakpoint junctions mapped, three individuals showed basic complex insertions, with a duplicated fragment translocated and inserted into another genomic locus with a deletion at the inserting position (Fig 1, S1 Table). Cplex4 demonstrated an ~11.8 Mb duplication on chromosome 14 (14q22.3q24.1) and an ~4.4 Mb deletion on chromosome 13 (13q21.31q21.32) revealed by array results (Fig 1A). FISH analysis and breakpoint junctions mapping demonstrated that the third copy of the duplicated segment on chr14 (chromosome 14) was inserted into chr13 at the position of the deletion (Fig 1B, FISH images previously published in a case report).[14] Similarly, in individual Cplex9, array results revealed an ~2.2 Mb duplication on chr9 at band 9q21.31 and an ~8.3 Mb deletion on chr13 at bands 13q12.3q13.3, while in Cplex12, array results revealed an ~0.8 Mb duplication on chr6 at band 6q27 and an ~0.5 Mb deletion on chr5 at band 5p14.3. Both FISH and breakpoint junctions mapping demonstrated that the duplicated fragment was inserted into the locus at which the deletion was observed in these latter two individuals Cplex9 and Cplex12 (Fig 1C and 1D and S1, S2A and S3A Figs). Note that in individual Cplex4, the inserted fragment was in the same orientation as the reference genome, however, in both Cplex9 and Cplex12, the inserted fragments were inverted when insertionally translocated (Fig 1B–1D). The CGRs in all three individuals were proposed to be generated through two breakpoint junctions, with 1 bp microhomology observed at both junctions in Cplex4 (Fig 1A, Table 1), 2 bp and 3 bp microhomologies observed at the junctions in Cplex9 (S2B Fig, Table 1), and 2 bp and 3 bp small insertions at junctions in Cplex12 (S3B Fig, Table 1).
In contrast to the three individuals described above, Cplex5, Cplex6, and Cplex11 showed multiple CNVs in addition to the insertions and were generated through multiple breakpoint junctions (Table 1). Cplex5 exhibited 4 CNVs from the array results on both chr6 and chrX: an ~1.3 Mb duplication at 6q21, an ~0.4 Mb deletion at 6q24.2, an ~8.6 Mb deletion at 6q25.1q25.3 (resulting in an overall duplication-normal-deletion-normal-deletion CGR pattern on chr6), and an ~1.5 Mb duplication at Xq28 (Fig 2A). FISH analysis revealed that the duplicated fragment of Xq28 was inserted and translocated to chr6, potentially at the deleted locus of 6q24.2 (S1 Table). Breakpoint junction mapping confirmed the findings observed by FISH, and the 4 mapped junctions enabled developing a parsimonious model accounting for all available data potentially explaining the rearrangement in this individual (Fig 2B). In brief, the duplicated fragment of Xq28 was inserted into 6q24.2, replacing the deleted region of 6p24.2 (Junction 1 and 2), while a duplicated fragment of 6p21 was inserted into 6q25.1, again replacing the other deleted region of 6q25.1q25.3 (Junction 3 and 4). Breakpoint junction sequencing revealed a 7 bp templated insertion (copied from nearby sequences) at Junction 1, 2 bp microhomology at Junctions 2 and 4, and 5 bp microhomology at Junction 3 (Table 1; S4 Fig).
In individual Cplex6, CMA showed an ~0.58 Mb duplication at 5p15.33, and an ~0.07 Mb duplication at Xq28. High-density aCGH revealed that the duplication on Xq28 actually contained a small triplication (~6 kb) embedded in the duplication (S5A Fig). FISH analysis and breakpoint junction mapping demonstrated that the duplicated fragment of 5p15.33 was inserted in an inverted orientation to Xq28. In addition, the triplication was also embedded in the duplication in an inverted orientation (Fig 2C, S1 Table), revealing a duplication—inverted triplication—duplication; a CGR pattern analogous to that previously observed and designated DUP-TRP/INV-DUP.[15] The proximal side of the duplication at Xq28 was joined to the distal side of the duplication at 5p15.33 (Fig 2C, Junction 1), while the proximal side of the 5p15.33 duplication was joined to the proximal side of the triplication embedded in the Xq28 duplication, leading to the triplication being inverted (Junction 2). We hypothesized a third junction connecting both distal sides of the triplication and the duplication at Xq28 should be present to generate the overall CGR in this individual; however, we were unable to uniquely position and map this breakpoint, possibly due to the presence of a low copy repeat (LCR) (S5A Fig). Sequences of Junction 1 in this individual showed blunt ends, while Junction 2 showed an insertion of 376 bp templated from at least three nearby genomic loci on both chr5 and chrX (Table 1, S5B Fig).
Individual Cplex11 exhibited the most complicated rearrangement in this study. Array results demonstrated a duplication-normal-duplication-normal-deletion pattern at 13q33.2 to 13q34 and a duplication-normal-duplication-triplication-duplication pattern at Xq21.1 (Fig 3A); FISH analysis showed that both of the two duplicated regions on chr13 were inserted into chrX (S6 Fig, S1 Table). Breakpoint mapping further demonstrated that the rearrangement between chr13 and chrX could be potentially generated through 6 junctions (Fig 3B). With the exception of the hypothetical Junction 5, we were able to map the remaining 5 junctions to nucleotide resolution. Based on the information of the five junctions mapped and the CNVs observed, we proposed the existence of Junction 5 to most parsimoniously explain the observed overall rearrangement in this individual (Fig 3B). Upon careful examination of the junctions, we observed that sequences of Junction 2 contained an 8,192 bp insertion from Xq13.2, followed by a 5,167 bp insertion from 4q13.1, leading to the discovery of the involvement of a third chromosome, chromosome 4, in this individual’s CGR (Table 1, S7 Fig). The remaining mapped junctions showed 2 bp microhomology (Junction 6) or blunt ends (Junction 1, 3 and 4).
Previously, we reported a child (BAB1379) with PLP1 deletion that resulted from a maternal balanced insertion (BAB1381) of a segment on chrX containing the entire PLP1 gene translocated and inserted into the telomeric region of the q arm of chr19 (Fig 4A).[16] The PLP1 deletion breakpoint junction was mapped in the previous publication, showing an Alu-Alu mediated rearrangement (Junction 3 in S8 Fig, re-drawn in hg19). This junction was present in the mother (BAB1381) and her affected son with Pelizaeus-Merzbacher disease (BAB1379), but not in the unaffected son (BAB1380). To fine map other breakpoint junctions involving the insertion, we designed high-density aCGH targeting both the regions on chrX containing PLP1, and the potential insertion site at 19qter. Surprisingly, in the mother, we did not see complete copy-number neutral genomic intervals around the PLP1 region as expected for her balanced insertion, but instead observed small CNVs that map at the exact loci of both ends of the deletion position in her affected son (Fig 4B). More specifically, an ~10 kb deletion at the proximal boundary, and an ~22 kb duplication at the distal boundary of the deletion position in her son (Fig 4B). In addition, an ~182 kb duplication was detected at 19q13.4, the potential inserting site, in the mother (Fig 4B). Further breakpoint junction mapping in the mother revealed that the distal side of the duplication on chr19 joined the distal side of the small deletion on chrX (Junction 1 in Fig 4C), while the proximal side of the chr19 duplication joined the distal side of the small duplication on chrX (Junction 2 in Fig 4C). The two small CNVs detected on chrX in the mother were actually due to unbalanced insertion from chrX to chr19, together with a duplication at the inserting site at 19q13.4. Sequences of the junctions showed 3 bp microhomology (Junction 1) and 15 bp templated insertion from nearby sequences at Junction 2 (Table 1, S8 Fig).
Observations in this family intrigued us to consider that the phenomenon may not be unique—CNVs inherited from parents with apparently balanced insertions may not be completely balanced at the molecular level. Given the small size of the potential CNVs, some may evade detection by clinical CMA. Therefore, we searched for similar families in the CMA database at BG, and found 12 families with a proband having a CNV inherited from a parent with apparently balanced insertion (named Family 1 to Family 12). We consented 4 families (Family 3, 4, 7 and 12) for further research studies, and discovered that in 2 families (Family 3 and Family 12), the apparently balanced insertions in the parents were not completely balanced, but actually had additional complexities revealed by molecular analyses.
In Family 3, Proband 3 (P3) showed a ~4.588 Mb deletion at 7p15.2p14.3 from array results; this deletion was further found by FISH analysis to be inherited from Mother 3 (Mat3) with apparently balanced insertion from chr7 into 9p24 (S9A Fig, S1 Table). Upon careful interpretation of high-density aCGH results, a small deletion (~4 kb) was observed in the mother at the exact boundary of the deletion in her child (Fig 5). We were able to fine map the identical deletion breakpoint junction present in both P3 and Mat3. Interestingly, an 815 bp insertion from 9p24 (chr9:5874574–5875388) was found at the chr7 junction sequences (Jct1)–the potential insertion locus observed from FISH in Mat3 (Fig 5). We further performed high-density aCGH in both Mat3 and P3 targeting the entire short arm of chr9. No promising CNVs were identified in either Mat3 or P3, however, three probes covering chr9:5874574–5875388 showed elevated ratio only in P3 but not Mat3 (S9B Fig). Based on this observation, we suspected an exchange of genetic material between chr7 and chr9 in the mother Mat3 –the ~4.588 Mb fragment from 7p15.2p14.3 was inserted to chr9, replaced by a small fragment from chr9p24.1 (815 bp from chr9:5874574–5875388). Note that the large fragment of 7p15.2p14.3 broke and re-joined during the inserting process based on the observation of mapped breakpoint junction 2 (Jct2), and additional junctions(s) should be present that connect the inserted fragments from chr7 to chr9 except for the mapped junction 3 (Jct3, S9C Fig).
Her child P3 inherited the deleted chr7 with the 815 bp insertion from chr9, together with an unaltered paternal chr9. We also suspected that the insertion site on chr9 was around chr9:5874574–5875388, and therefore performed walk-in PCRs and successfully pinpointed the insertion site on chr7 (S9C Fig). Another interesting observation is the presence of human endogenous retroviral elements (HERVs) at the boundaries of both Jct1 and Jct2, which are known to promote genome instability and induce CNV formation.[17]
In Family 12, Mother 12 (Mat12) had two children with CNVs in the long arm of chr19: an ~3.5 Mb duplication at 19q13.33q13.41 in her daughter (P12_dup), and a slightly smaller deletion (~3.352 Mb) at the same locus in her son (P12_del) (S1 Table). FISH analysis demonstrated an apparently balanced insertion of a segment at 19q13.33 into the short arm of chr19 at 19p13 in Mat12; FISH analysis also demonstrated the same insertion in P12_dup, indicating the duplication present in P12_dup was likely a recombination product of intrachromosomal maternal insertion (S10 Fig). The reciprocal deletion present in P12_del was also likely a recombination product (S11B Fig). High-density aCGH revealed that the apparently balanced insertion in Mat12 was not balanced—at both proximal and distal boundaries of the duplication/deletion in her two children, there were two small duplications of ~111 kb and ~77 kb in size, respectively. In addition, a small triplication (~33 kb) was found embedded in the duplication near the proximal side in P12_dup (S11A Fig). These additional complexities were likely accompanying events with the insertion in Mat12 that was subsequently transmitted and inherited by her two children, similar to the rearrangement events involving the PLP1 observed in BAB1381 mentioned above (S11C Fig, refer to S11 Fig for details of proposed rearrangements in Family 12).
Previously, we demonstrated that confirmatory and parental studies of CNVs by FISH analysis, especially the copy-number gains identified through CMA testing, led to the discovery of chromosomal insertions at a rate as high as 1 in ~500 individuals tested.[1] A similar high rate of 1 in ~563 individuals was independently reported.[5] Although it is now widely recognized that chromosomal insertions are not rare events,[1, 5, 6] the underlying mechanisms for their formation remain largely unknown. Most of the previous studies on insertions were based on relatively low resolution genome analysis by clinical arrays in combination with molecular cytogenetics, FISH, and chromosome analysis; only a few breakpoint junctions have been mapped to nucleotide resolution.[5, 18, 19]
In this study we focused on a subset of chromosomal insertions—complex insertions with additional copy-number gain or loss at the inserted site. High-density aCGH revealed additional complexities that evaded detection by CMA testing, including small triplications embedded in duplications (in individuals Cplex6 and P12_dup) and small CNVs in individuals with apparently balanced insertions (in individuals BAB1381, Mat3, and Mat12). In addition, breakpoint junction mapping and careful examination of the junction sequences provided insights into the potential mechanisms for formation of these complex insertions, leading to the observation of distinct molecular characteristics of apparently basic complex insertions versus chromothripsis-like, chromoanasynthesis insertions. Of note, only individuals with CNVs large enough to be detected by clinical microarray, and subsequently with copy-number gains large enough to be verified as insertions by FISH, were initially identified and molecularly studied. Therefore, copy-number neutral insertions, and insertions with smaller CNVs that escaped detection by clinical array or FISH validation, were selected against inclusion in this study.
We categorized individuals Cplex4, Cplex9, and Cplex12 as basic complex insertions (S12 Fig) based on the following observations: first, only one duplication was observed in these individuals, in contrast to the multiple copy-number gains observed in other individuals in this study; second, a deletion was always present at the inserting site; third, none of them were de novo events (S1 Table). Breakpoint junctional sequences in these individuals showed 1–3 bp microhomology or 2–3 bp small insertions; these features represent mutational signatures of breakpoint junctions observed in structural variants potentially generated by either non-homologous end-joining (NHEJ), or alternatively, microhomology-mediated end-joining (MMEJ) or FoSTeS/MMBIR with a single template switch.[3, 20–23]
In contrast to the individuals with basic complex insertions that were potentially generated by a number of different mechanisms, individuals Cplex5, Cplex6 and Cplex11 showed multiple CNVs including triplications. In addition, Cplex5 and Cplex11 are de novo events (S1 Table, inheritance mode in Cplex6 is unknown). Breakpoint junctions’ sequences in these individuals showed longer homology (>4 bp) and hundreds to thousands of base pairs of templated insertions. CNVs in these individuals resembled chromoanasynthesis events[12], and their breakpoint junctions features are signature findings observed in structural variants generated through replicative repair based mechanism, e.g. FoSTeS/MMBIR with multiple iterative template switch events.[24, 25] Interestingly, one of the 16 individuals identified with complex insertions initially included in this study, Cplex2, was included and analyzed in detail in the paper that first defined the chromoanasynthesis phenomenon (BAB3105 from Ref. 12). This further strengthens our proposal that complex insertions could be part of a chromoanasynthesis event.
Currently, three similar yet distinct types of chromothripsis, or chromothripsis-like events have been described, together they were referred to as ‘chromoanagenesis’.[26] In somatic changes in the cancer genomes, chromothripsis was shown to be a catastrophic, one-step event leading to a signature pattern of frequent oscillations between unaltered and deleted copy-number states.[8] In cancer chromothripsis, most CNVs observed from genomic sequence analyses are deletions, with much less duplications resolved, and usually involves one chromosome. In contrast to the frequent copy-number loss in cancer chromothripsis, germline chromothripsis observed in individuals with neurodevelopmental abnormalities was shown to be balanced rearrangements—although several chromosomes were shattered and rejoined, the overall complex rearrangement involved almost no copy-number changes (except for deleting or inserting short sequences at breakpoint junctions).[9, 10] A recent study on unbalanced interchromosomal translocations revealed two individuals with de novo chromothripsis translocations generated through at least 18 or 33 breakpoint junctions, respectively, and both individuals only carried two large deletions (from 800 kb to 6.6 Mb) but no copy-number gains.[27] Both somatic and constitutional chromothripsis were proposed to be generated by NHEJ, given that the vast majority of the breakpoint junctions in these events showed blunt ends, 1 or 2 bp microhomology, or small insertions.[10, 11, 28]
In contrast to the balanced germline chromothripsis involving shattering and rejoining of several chromosomes, another type of chromothripsis-like events, observed by high-density aCGH and mechanistically defined as chromoanasynthesis, was shown to involve multiple copy-number changes, particularly multiple gains of copy-number including duplications and triplications.[12] Notably, chromoanasynthesis was considered to be region-focused events.[12, 29, 30] In the original paper that defined the chromoanasynthesis phenomenon, all 17 individuals studied showed CNVs on the same chromosome, more specifically, 15 out 17 individuals showed CNVs confined within the distal half of the involved chromosome arms.[12] It was proposed that co-occurrence of CNVs with substantial interchromosomal exchanges would result in a non-viable offspring.[10] Here, we demonstrated that chromoanasynthesis could involve two or even three chromosomes, as we observed a templated insertion as long as 5,167 bp from a third chromosome in addition to the two chromosomes involved in the rearrangements in Cplex11 (S7 Fig).
We categorized individuals Cplex5, Cplex6, and Cplex11 as chromoanasynthesis events (S12 Fig) based on the observations that included: i) multiple copy-number gains, including triplications, were detected, ii) longer microhomology (>4 bp) observed at breakpoint junctions and iii) long templated insertions from multiple genomic loci also present at breakpoint junctions. Similar to previously reported region-focused chromoanasynthesis events, these features are likely found in CGRs generated by iterative template switching during replicative repair based mechanisms, e.g. FoSTeS/MMBIR.[3, 24, 25] Note that in individuals Cplex6 and Cplex11, some of their breakpoint junctions sequences showed blunt ends (S5B and S7 Figs); it is not uncommon to observe that a portion of the junctions in CGRs potentially generated through replication based mechanisms can show blunt ends, small insertions or short microhomology (1 or 2 bp).[15, 31, 32] In studies conducted in the yeast model organism, FoSTeS/MMBIR has been demonstrated to occur in the absence of microhomology (with 0–6 bp homology at breakpoint junctions). [33] Although breakpoint junctions with short microhomology (1–3 bp) have been observed in rearrangements proposed to be potentially generated through NHEJ, MMEJ and MMBIR in the human genome, iterative template switches are unique to the mechanism of FoSTeS/MMBIR. Therefore, it is important to consider not only microhomology length, but also the occurrence of templated insertions at junctions, and other evidence for potential iterative template switch events, in addition to whether copy-number gains (especially triplications) are present, when postulating potential biological mechanisms responsible for the generation of CGRs.
Recent studies on DNA damage in micronuclei provided a potential further explanation for the chromothripsis and chromoanasynthesis events. Micronuclei are common outcomes of cell division defects; they are structurally similar to intact nuclei, but contain only one or a few chromosomes or chromosomal segments.[34] They could undergo defective and asynchronous DNA replication, resulting in DNA damage and extensive chromosomal fragmentation including catastrophic processes like chromothripsis; most importantly, their damaged and rearranged DNA fragments could be integrated back into the genome.[32, 35] Rearrangements proposed to be generated through NHEJ or MMBIR have been observed in micronuclei DNA, and segments from a single chromosome were observed in the majority of the micronuclei—potentially explaining why most observed chromoanasynthesis events are chromosome or chromosome region-focused. The rare chromoanasynthesis events involving two or three chromosomes we observed in this study are potentially in accordance with the rare observation of micronuclei DNA from two chromosomes undergoing chromothripsis.[35]
In this study, we also discovered that some apparently balanced insertions are actually unbalanced insertions; small deletions and duplications could be generated accompanying the inserting process. From the mechanistic aspect, it is crucial to reveal these small CNVs—a completely balanced insertion could be attributed to mechanisms like NHEJ, however, the additional CNVs, especially the copy-number gains, are more parsimoniously explained by replicative repair based mechanisms. For example, in the family with PLP1 insertion, the most parsimonious explanation for the small CNVs at both the inserting site 19q13.42 and missing proximal/additional distal segments accompanying the inserted fragment from Xq22.2 is FoSTeS/MMBIR. During the replication process, a stalled replication fork at chr19 invaded and annealed to chrX, and after replication of a genomic interval containing the entire PLP1 gene on Xq22.2, the replication fork switched back to chr19q13.42, however, to a more proximal locus, therefore leading to the small duplication on chr19 (Fig 4C). We consider the situation in Family 12 to be similar to the PLP1 family, due to the two duplications on both boundaries of the inserting fragment potentially generated accompanying the chromosomal insertion in the mother Mat12 (S11 Fig). Whereas in Family 3, the situation may be different—unlike in the chromoanasynthesis subjects and in BAB1381, whose CNVs are most parsimoniously explained by template switching during the replication process (copying material from the inserting chromosomes to the inserted loci, always one direction), there was an exchange of genomic segments between chr9 and chr7 in Mat3. In addition, no copy-number gain was observed in this family, and the insertions in Mat3 are mostly balanced except for the 4 kb deletion at chr7. We propose the bi-directional, mostly balanced insertions in Mat3 may result from multiple breakages and re-joining of both chr7 and chr9, therefore may be generated through NHEJ or MMEJ.[9, 27]
LCRs and repetitive elements are known to facilitate genomic rearrangements.[12, 29, 36, 37] Enrichment of breakpoint in these repetitive sequences has been observed in nonrecurrent and complex structural changes at multiple genomic loci. [29, 38, 39] In the current study, we observed involvement of LCRs at breakpoint junctions in individuals Cplex5 and Cplex6 (Table 1), and HERV elements at breakpoint junctions in Family 3 and Cplex6 (S9C Fig). In addition, we observed involvement of other repetitive sequence, e.g. SINEs (short interspersed nuclear elements) at junctions in Cplex9, Cplex6, BAB1381 and in Family 12, also LINEs (long interspersed nuclear elements) at junctions in Cplex4, Cplex12, Cplex5, Cplex11 and BAB1381 (Table 1). These repeat and repetitive sequences may stimulate genomic instability and potentially assist replicative repair catalyzed genomic rearrangements facilitating template switching and the generation of the nonrecurrent and complex insertion events.[3, 29, 40]
In summary, from studies of complex chromosomal insertions, we observed that chromoanasynthesis could occur beyond a confined chromosomal region and involve two or three chromosomes. We observed microhomologies and templated insertions at the breakpoint junctions, resembling the breakpoint junction signatures found in CGRs generated through replication-based mechanism(s) and iterative template switches: FoSTeS/MMBIR.[3] We propose that DNA replicative repair mechanisms can potentially result in interchromosomal complex insertions, and cause a significant fraction of apparently balanced insertions; especially those harboring small flanking CNVs.
Sixteen individuals with complex chromosome insertions were identified in the CMA laboratory at Baylor Genetics among the ~56,000 individuals tested from 2007 to 2014. This study was approved by the Institutional Review Board for Human Subject Research at Baylor College of Medicine (IRB H-25466). Informed consent was obtained prior to collecting identifiable DNA samples (BAB1379, BAB1380, BAB1381, P3, Mat3, P12_del, P12_dup and Mat12). The remaining DNA samples were de-identified for breakpoint and mechanistic studies (named Cplex1, Cplex2, Cplex3, etc).
Custom designed BCM OLIGO V6.5, V7, V8, V9 or V10 oligonucleotide arrays were performed as previously described.[41, 42] Arrays were designed to specifically interrogate clinically significant regions with an average resolution of 30 kb between probes. Interphase and metaphase FISH were performed to confirm the CMA findings and tested using available parental samples.[1]
To further characterize the CNVs identified by CMA and FISH involving complex insertions, we designed several 4X 180K oligonucleotide arrays with ~200 bp per probe spacing from Agilent Technologies (AMADID 073188, 073189, 076797, 079204, 071585, 024241 and 015482). Hybridization controls were gender matched (Individual NA10851 as male control and Individual NA15510 as female control). Scanned array images were processed using Agilent Feature Extraction software (version 10) and extracted files were analyzed using Agilent Genomic Workbench (version 7.0.4.0). Array designs and sequence alignment for breakpoint analysis were based on the February 2009 genome build (GRCh37/hg19 assembly).
To further confirm the CNVs identified by high-density arrays and map the breakpoint junctions, primers flanking the predicted breakpoints were designed and long-range PCRs were conducted using TaKaRa LA Taq according to the manufacturer’s protocol (TaKaRa Bio Company, Cat. No. RR002) as previously described.[29] PCR products were prepared for sequencing using ExoSAP-IT (Affymetrix, Cat. No. 78201) according to the manufacturer’s protocol or gel extracted and purified with the Zymoclean Gel DNA Recovery kit (Zymo Research, Cat. No. D4001). Purified PCR products were then sequenced by Sanger di-deoxynucleotide sequencing (BCM Sequencing Core, Houston, TX, USA). To elucidate the insertion site in individual Mat3, the APAgene GOLD genomic walking kit was used according to the company’s protocol (BIO S&T, Cat. No. BT901-RT). Generally, this kit enables isolation of unknown sequences which flank known sequences. Three rounds of nested PCR with degenerate random tagging primers provided by the kit were performed, and the end PCR products were cloned into a TA vector (pGEM-T Easy Vector Systems, Promega, Cat. No. A1360) and were further subjected to Sanger sequencing.
|
10.1371/journal.pcbi.1000904 | Graph-Based Analysis of the Metabolic Exchanges between Two Co-Resident Intracellular Symbionts, Baumannia cicadellinicola and Sulcia muelleri, with Their Insect Host, Homalodisca coagulata | Endosymbiotic bacteria from different species can live inside cells of the same eukaryotic organism. Metabolic exchanges occur between host and bacteria but also between different endocytobionts. Since a complete genome annotation is available for both, we built the metabolic network of two endosymbiotic bacteria, Sulcia muelleri and Baumannia cicadellinicola, that live inside specific cells of the sharpshooter Homalodisca coagulata and studied the metabolic exchanges involving transfers of carbon atoms between the three. We automatically determined the set of metabolites potentially exogenously acquired (seeds) for both metabolic networks. We show that the number of seeds needed by both bacteria in the carbon metabolism is extremely reduced. Moreover, only three seeds are common to both metabolic networks, indicating that the complementarity of the two metabolisms is not only manifested in the metabolic capabilities of each bacterium, but also by their different use of the same environment. Furthermore, our results show that the carbon metabolism of S. muelleri may be completely independent of the metabolic network of B. cicadellinicola. On the contrary, the carbon metabolism of the latter appears dependent on the metabolism of S. muelleri, at least for two essential amino acids, threonine and lysine. Next, in order to define which subsets of seeds (precursor sets) are sufficient to produce the metabolites involved in a symbiotic function, we used a graph-based method, PITUFO, that we recently developed. Our results highly refine our knowledge about the complementarity between the metabolisms of the two bacteria and their host. We thus indicate seeds that appear obligatory in the synthesis of metabolites are involved in the symbiotic function. Our results suggest both B. cicadellinicola and S. muelleri may be completely independent of the metabolites provided by the co-resident endocytobiont to produce the carbon backbone of the metabolites provided to the symbiotic system (., thr and lys are only exploited by B. cicadellinicola to produce its proteins).
| Some bacteria, called endocytobionts, permanently live inside the cells of a pluricellular organism and often bring an adaptative advantage to their host by providing compounds that the latter cannot produce or find in its diet. The association may involve several species of bacteria within the same host. The sap-feeding insect called glassy-winged sharpshooter thus maintains a permanent metabolic association with two different species of bacteria that it hosts within specialised cells. Complete genome annotations of the two endocytobionts allowed to reconstruction of their metabolism. By manually inspecting those annotations and comparing them to reference metabolic functions, earlier studies revealed a great complementarity between the metabolisms of the two endocytobionts and indicated potential metabolic exchanges between them. However, the metabolism of an organism is complex enough that such an approach could only give a partial description of the metabolic exchanges in the symbiotic system. We therefore determined all the metabolic exchanges in the symbiotic system by a systematic and automatic exploration of the full metabolism of the two endocytobionts in order to detail those leading to the biosynthesis of compounds involved in the symbiotic function of each bacterium. Our results highly refine our knowledge about the complementarity and the connections between the metabolisms of the two bacteria and their host.
| Intracellular symbiosis involves a unicellular organism (the endocytobiont) which durably lives inside the cells of the other partner (the host). In the last century, the crucial role of intracellular symbiosis in the ecology and evolution of many eukaryotes was many times demonstrated [1], [2].
Intracellular mutualism (where the presence of the endocytobiont increases the fitness of both host and endocytobiont) was particularly well described in several associations between insects and bacteria [3], [4]. The association is most often metabolic: each partner provides metabolites that the other one cannot produce nor find in its environment. The complete genome annotation of mutualistic endocytobionts associated with insects revealed for all of them an extreme genome reduction paired with an extreme metabolism reduction [5]. Many metabolic functions of the bacterium are thus provided by the host and, inversely, the metabolism of the endocytobiont appears specialised into functions that are absent in the metabolism of the host. In addition, it often occurs that a host provides a habitat for more than one mutualistic intracellular bacterium species. This is the case for instance of the sharpshooter (Homalodisca coagulata) which hosts two bacteria: the -proteobacterium Baumannia cicadellinicola and the Bacteroidetes Sulcia muelleri. The complete genome annotation of the two endocytobionts revealed that their metabolic capacities are broadly complementary [6], [7]. The metabolism of B. cicadellinicola is globally devoted to cofactor and vitamin biosynthesis whereas the metabolism of S. muelleri is specialised in the essential amino acid biosynthesis that the sharpshooter cannot produce nor find in its diet, the xylem sap. Nevertheless, the partition of these metabolic roles is not so perfect: B. cicadellinicola produces two essential amino acids, methionine and histidine, that S. muelleri cannot produce while the latter appears to be able to synthesise menaquinone, a vitamin. Moreover, the complementarity between the two metabolisms also concerns the biosynthesis of some metabolites not needed by the insect host, such as the fatty acid biosynthesis pathway, supplied by B. cicadellinicola, except for one step which is provided by S. muelleri [7].
However, these previous analyses were essentially manually performed by comparing the lists of annotated metabolic genes using as reference the metabolic pathways available in metabolic databases such as KEGG [8] or MetaCyc [9]. Even when highly reduced, a metabolic network is however complex enough that such an approach based on lists of genes and metabolic pathways could only give a partial description of the metabolic exchanges in the symbiotic system, even for those directly involved in the symbiotic functions of the endocytobionts.
The aim of this study was therefore to determine the possible metabolic exchanges in the symbiotic system by a systematic and automatic exploration of the full metabolic networks of the two endocytobionts in order to detail those leading to the biosynthesis of metabolites involved in the symbiotic function of each bacterium.
Defining in an exhaustive way the metabolic exchanges in a symbiotic system implies to be able to indicate all the metabolites needed by one partner and produced by another partner. Our first task was thus to identify for each endocytobiont the metabolites potentially imported from the host cell (that is, the so-called “seeds”) and produced by another partner.
Focusing on the biosynthesis of the specific compounds that each bacterium produces and provides to the symbiotic system (from now on, we denote such compounds by “targets”) then necessitates to determine which sets of exchanged metabolites lead to their production. Our second task was thus to identify for each endocytobiont the subsets of seeds (from now on, we denote such subsets by “precursor sets”) that are sufficient to produce the targets, and to identify them all, that is all alternative precursor sets for each target.
These two tasks, and particularly the second one, are hardly feasible by just manually inspecting the metabolic pathways inferred from genomic annotations. Such broad and systematic analyses require working with the full metabolic network to consider it in a systemic way.
An intuitive way to define the seeds of a metabolic network is to consider as nutrient a metabolite not produced by any reaction but consumed by one or several ones. However, in particular because of reversible reactions that produce nutrients, this definition is not sufficient. Borenstein et al. extended the seed definition in a metabolic network by decomposing the metabolic graph into strongly connected components and then detecting those without incoming edge [10].
Whatever way is adopted to define the sources of a metabolic network, the next question is to determine which precursor sets are able to produce the targets. Romero et al. (2001) proposed a method returning alternative precursor sets for a set of target compounds [11]. Their algorithm was based on a backtrack traversing of the metabolic graph from the target compounds to the seeds. Unfortunately, how cycles are dealt with during backtracking is not described in the method.
Another method proposed by Handorf et al. to find precursor sets is based on a forward traversing of the metabolic graph [12]. Their algorithm is based on the concept of scope defined as the set of compounds that a set of initial seeds is able to produce [13]. The way to find precursors proposed by Handorf et al. is then to test the reachability of several sets of seeds heuristically defined [12]. However, the method does not take into account cycles that may appear between the seeds (see Methods) and only provides a subpart of the possible precursor sets for a target compound.
Recently, we proposed the first definition of minimal precursor sets that explicitly addresses the problem of cycles and the first exact method to find them [14]. In addition, our method, called PITUFO (for “Precursor Identification To U For Observation”), is able to deal with any definition of seeds.
To explore the metabolic exchanges occurring between B. cicadellinicola and S. muelleri, we first defined the set of seeds for each bacterium thanks to the method developed by Borenstein et al. [10]. By comparing with the compounds produced by each metabolic network, we were able to discriminate between the seeds produced by the co-endocytobionts and those potentially produced by the insect host or found in its diet. We then applied PITUFO to determine which subsets of seeds are involved in the biosynthesis of compounds already known to participate in the symbiotic function of each bacterium. The two steps are summarised in Figure 1.
Our study offers the first detailed and systematic description of the metabolic exchanges occurring in a symbiotic system. Our results also demonstrate the usefulness of graph-based dedicated methods in the metabolic analysis of multi-species systems.
Draft metabolic reconstructions for the bacterial genomes of Baumannia cicadellinicola and Sulcia muelleri were downloaded from the MaGe annotation system [15]. This platform makes available metabolic networks built from re-annotated genomes. Each metabolic network reconstruction is available in the pathway-tools format [16]. We restricted the networks to the small-molecule metabolism, meaning that reactions involving macromolecules such as nucleic acids or proteins were removed from the final metabolic networks.
A first manual curation consisted in removing what may be considered as “fake” reactions. Indeed, an enzyme is potentially able to catalyse several reactions but a limited number of them actually takes place in a given organism. The reactions that clearly do not happen correspond to those that either are disconnected from the network (they use as inputs compounds that are not produced in the organism and produce compounds not used as substrates by other reactions), or that are connected to the network only by cofactors. The topology of the network provides thus a clue to remove 14 reactions from the metabolic network of S. muelleri and 37 reactions from the metabolic network of B. cicadellinicola (see Tables S1 and S2). Such reactions can be automatically detected by a topological analysis but their elimination requires a manual inspection and some biological a priori since some reactions appear disconnected because of a hole in the network, that is, of a single reaction that is missing due to an error or incompleteness in the annotation process. The main clue to fill such holes is to inspect the completeness of the metabolic pathways predicted in the organism. By comparing these predictions with data from the literature, we are able to complete some metabolic pathways, and thus the metabolic network.
In addition, several reactions involve generic compounds (for instance, an aldehyde) in the draft metabolic networks. When the same reaction existed with specific compounds, the generic reaction was simply removed. This is the case of nine reactions in the metabolic network of B. cicadellinicola. If specific reactions do not exist, the reaction has to be duplicated into several reactions so that they involve compounds already existing in the metabolic network. This is the case of the reaction RXN-8972 for which a substrate is “lysine or meso-diaminopimelate”. This reaction was thus splitted into two reactions (RXN-8972BIS and RXN-8972TER in Table S4) that involve lysine and meso-diaminopimelate respectively.
In automatic metabolic reconstructions, several reactions can be assigned to annotated enzymes. These reactions often use the same main substrates but different cofactors or even completely different substrates. When several reactions were assigned to a same enzyme, we removed the reactions for which the substrates required were absent in the metabolic network. This is the case of eight reactions in the metabolic network of S. muelleri and of 19 reactions in the metabolic network of B. cicadellinicola. We removed also six reactions in the metabolic network of S. muelleri and seven reactions in the metabolic network of B. cicadellinicola that were classified into the small molecule metabolism whereas they clearly involve macromolecules (see Tables S1 and S2).
In the end, a total of 16 reactions in the metabolic network of S. muelleri and 58 reactions in the metabolic network of B. cicadellinicola were thus removed, using the clues mentioned above (see Tables S1 and S2).
The gene (epd) that catalyses the production of erythronate-phosphate from erythrose-phosphate appears as absent in the genome of B. cicadellinicola. Wu et al. made the assumption that this role could be carried by glyceraldehyde 3-phosphate dehydrogenase [6]. We thus added this reaction to the metabolic network of B. cicadellinicola (ERYTH4PDEHYDROG-RXN in Table S4).
McCutcheon et al. mentioned that no gene in S. muelleri could be assigned as argE or dapE, potentially coding for an enzyme catalysing one step in either the lysine biosynthesis and the arginine biosynthesis [7]. Interestingly, a gene (SMGWSS-116) was assigned as argE in the genome of S. muelleri by the MaGe annotation system.
We then considered the corresponding reactions in the two metabolic pathways as present (SUCCDIAMINOPIMDESUCC-RXN and ACETYLORNDEACET-RXN in Table S3).
The direction of the reactions was first assigned based on the pathways where they are involved in the MetaCyc database [9]. A reaction is thus assigned as irreversible if it occurs in the same direction in all the MetaCyc pathways. If it was not possible to infer a unique direction, then the reaction remained reversible. The direction of most of the reactions in both metabolic networks were assigned in this way (Tables S3 and S4). Various manual corrections were also performed, essentially based on the constraints brought by the topology of the network and the biology of the organism, exemplified in Figure 2. For instance, the molecular weight of compound in Figure 2 could be too large to allow a transport of the molecule. Furthermore, the classification of the reactions that are in the same metabolic pathway where and classically appear as intermediate metabolites can be an additional clue to assign the direction of . Eight reactions in the metabolic network of S. muelleri and 28 reactions in the metabolic network of B. cicadellinicola were hence assigned as irreversible (Tables S3 and S4).
The whole set of reactions of the metabolic networks of S. muelleri and of B. cicadellinicola are displayed in Tables S3 and S4. The metabolic networks of S. muelleri and of B. cicadellinicola are available in SBML format [17] in Datasets S1 and S2.
We restricted our study to the metabolism involving transfers of carbon atoms between molecules. In each reaction, we thus removed sets of molecules that do not participate into carbon exchanges. We called these metabolites “side compounds”.
First, we established a list of 24 classical transformations between side compounds (e.g. ) present in the metabolic networks of the two bacteria (see Table S5). When one of these transformations is identified in a reaction, the corresponding side compounds were removed from the reaction. Since whether metabolites are side-compounds in a given reaction is not always clear, some reactions were then manually corrected.
The following inorganic compounds were also removed: water, proton, phosphate, diphosphate, ammonia, hydrogen peroxyde, sulfite, sulfate and oxygen. Reactions that do not imply a transfer of carbon atoms are also eliminated. This is for instance the case of the reactions involved in the sulfate reduction.
The filtered metabolites are written in non-bold in Tables S3 and S4. The filtered metabolic networks of S. muelleri and of B. cicadellinicola are available in SBML format [17] in Datasets S3 and S4.
In order to describe the metabolic exchanges between the endocytobionts, the first step consisted in identifying which metabolites each bacterium potentially acquires from its environment. For this, we based ourselves on the definition of Borenstein et al. of the seed set of a network: “the minimal subset of the occurring compounds that cannot be synthesized from other compounds in the network (and hence are exogenously acquired)” [10].
To apply the Borenstein method to identify the seed sets, the metabolic network of each bacterium was modelled as a directed compound graph. In such a graph, nodes represent compounds and there is an arc between two compound nodes if at least one reaction produces one of the compounds (possibly more) from the other (possibly more). A reversible reaction between two metabolites is modelled by two arcs with opposite directions linking them. Since the side compounds were previously filtered (see previous Section), we avoid paths between metabolites that are biologically meaningless for our study.
The seeds identification is based on the detection of the strongly connected components (SCC) in the compound graph. An SCC is a subgraph that contains a maximal set of nodes such that for any pair of nodes and in , there exists a path between and and a path between and . An SCC with no incoming arc is called a source component. Any compound inside a source component is a potential seed or, in our case, just seeds.
This definition of seeds allows to take into account the uncertainty about the direction of some peripheral reactions and about the presence of reactions producing metabolites that are actually exogenously acquired. We invite the reader to refer to the paper of Borenstein et al. for more precisions [10].
For each symbiont, we further inspect the collection of seed sets identified in order to classify these seeds as potentially provided by the insect or by the other co-symbiont. This is done by analysing the feeding source of the host as described in the literature [6], [7] or the metabolic network of the other co-symbiont to check whether these metabolites may be produced, and may therefore be supplied.
We implemented a version of the Borenstein's method using the Igraph package [18] and applied it to the compound graph of each bacterium.
Once the set of seeds was defined for each metabolic network, the next step was to identify which subsets of the seeds, henceforward called “precursor sets”, are sufficient to produce the metabolites known to be involved in the symbiotic metabolic association, that from now on we call the “targets”. Those are metabolites output by one bacterium that may then be used by the other co-symbiont or the host. We first put as targets the metabolites reported as involved in the symbiotic association by McCutcheon et al. [7]. We then added erythrose-4-phosphate, phosphoenolpyruvate, oxaloacetate and ribose-5-phosphate to the list of target compounds for B. cicadellinicola because of their presence both in its metabolic network and in the precursor sets identified for S. muelleri. These additional targets are particularly interesting since they could directly correspond to metabolic pathways shared between the two metabolic networks. For the same reasons, we added homoserine and 2-ketovaline to the list of target compounds for S. muelleri.
To identify the precursor sets, we used the PITUFO method that we recently developed [14]. Given as input a metabolic network, a list of seeds and a set of target metabolites, PITUFO returns the list of all minimal precursor sets for the target metabolites. For the purposes of this paper, we consider single sets of target metabolites, that is sets with only one element. Notice that once we get as result all minimal sets of precursors that are able to produce a target, we are covering all alternative paths that may lead to its production. understand the reasoning.
The strength of PITUFO comes from the fact that it takes into account cycles in the definition of precursor sets in a fully formalised manner. This allows to find paths from the precursor sets to the targets that pass through cycles in the network but are still feasible. Previous methods, such as those that compute the scope of a subset of the seeds as defined by Handorf et al. [13] and were later used to test the reachability of a target compound from a set of seeds [12], fail to link some sets of seeds to a target compound if there is such a cycle in the paths between them. The scope of an initial seed set is itself and then any metabolite that can be produced using only substrates already in and added to it until no new compound can be produced [13]. This iterative process is called forward propagation [11] or network expansion [13].
The strategy of PITUFO to deal with cycles is to allow the use of metabolites involved in cycles if they are also produced (regenerated) in the forward propagation from the seeds to the targets. Indeed, in Figure 3, the scope of the set does not contain but if we allow the use of or in the forward propagation process, then the scope of contains . However, this could lead to clearly unrealistic paths without the constraint of regeneration of the compounds involved in cycles. For instance, in the same figure, if we allow the use of (and possibly also ), the scope of contains also but uses up all of and unless both were in infinite supply, in which case they should be considered as seeds.
The metabolites or , inside a cycle in Figure 3, that may be used and that are regenerated when the network is fired from a subset of the seeds (the set in the figure), are called “self-generating metabolites” [14]. Observe that these are defined in relation to a subset of the seeds. They do not need to be given as input but will be identified by the algorithm together with the sought precursor sets.
The following definitions allowed us then to formally establish what is a precursor set of a given target compound: a subset of the set of seeds is considered as the precursor set of a target compound if there exists a set of metabolites such that the scope of , allowing the use of in the forward propagation process, contains and all the metabolites in , which ensures the regeneration of . The set is considered as a minimal precursor set if there is no set strictly contained in that verifies this property.
In the above definition (and in [14]), a metabolic network is considered as an hypergraph. Nodes are metabolites and there is an hyperarc between two sets of metabolites if there is a reaction that produces one of the sets from the other. Contrary to simple graphs where the nodes represent either compounds or reactions and arcs link individual nodes, the topology of an hypergraph takes into account the need in general for more than one substrate to activate a reaction [19].
The use of hypergraph modelling allows the formalisation of hyperpaths between precursor sets and targets since when a reaction is modelled as an hyperarc, it already explicitly establishes that all of its substrates are needed. On the other hand, an hypergraph could lead to some confusion for the method previously applied to identify the seed sets and proposed by Borenstein et al. [10] since there is no clear definition of strongly connected components of an hypergraph. For instance, isolated vertices may not be considered as SCCs and there is no unique definition of cycles in hypergraphs. For these reasons, the method was applied as in the original work on a compound graph representation of the metabolic network.
Since PITUFO is an exact method, it is enough to describe its input and output without recalling how the second is produced from the first. For those interested in the method itself, the algorithm is described in detail in [14].
The current version of the algorithm was implemented in Java and takes as parameter an SBML file describing a metabolic network [17] and a file containing a list of seeds and one or a list of target compounds.
The reconstructed metabolic networks complete or filtered are available in the Supplementary material. The method used is available at this address: http://sites.google.com/site/pitufosoftware/.
Figures S1 to S20 display the sub-networks linking each target metabolite to their precursor sets. These reconstructions were performed from the PITUFO results using the visualisation software Cytoscape [20].
Table 1 shows the number of reactions and compounds in each metabolic network as indicated in the Reconstruction Section. As mentioned in previous studies, both metabolic networks are extremely reduced. The metabolic network of B. cicadellinicola is less than half the size of the metabolic network of the free bacterium Escherichia coli. The reduction is even more important in S. muelleri since, with only 64 reactions, its metabolic network is less than ten percent the size of the network of E. coli. In both cases, the extensive manual curation allowed to highly reduce the number of reversible reactions as we succeeded to assign a direction to most of the reactions in the two metabolic networks.
Figure 4 displays the set of seeds identified in the metabolic network for each bacterium. Coloured arrows mark those produced in the metabolic network of the co-endocytobiont and those potentially provided by the insect host according to the literature. Seeds that correspond to annotated transport reactions are also tagged.
We recall that the reactions involving big molecules are not taken into account in this analysis. For instance, the reactions charging amino acids onto their corresponding tRNAs do not appear in the metabolic network we built. This means two things. First, an amino acid involved only in the production of proteins does not appear in the seeds that we identified, which explains the absence of some essential amino acids in the set of seeds identified in B. cicadellinicola. Second, this also means that an amino acid identified as a seed is involved in the small molecule metabolism and not only in the production of proteins.
Moreover, we focused on the transfers of carbon atoms. There are then no inorganic metabolites in the sets of seeds. Furthermore, the organic compounds not involved in carbon atom transfers do not appear in this list. This is the case for instance of glutamine that appears as a source of nitrogen but not of carbon in the metabolism of B. cicadellinicola. This metabolite is thus a seed in the original metabolic network of the bacterium but not in the filtered one.
We identified 19 seeds in the metabolic graph of B. cicadellinicola and 10 seeds in the metabolic graph of S. muelleri. Only three seeds are common to the two sets: serine, aspartate and bicarbonate ion.
Whereas none of the seeds in S. muelleri could be linked to a transport reaction, four seeds (lysine, glutamate, aspartate and glucose) correspond to transport reactions annotated in B. cicadellinicola. Furthermore, a general amino acid ABC transporter annotated in its genome enables B. cicadellinicola to import also other amino acids identified as seeds: threonine, glycine, tyrosine and alanine.
Among the 10 seeds identified in S. muelleri, three are amino acids (cysteine, aspartate and serine) and three are sugars: erythrose-4-phosphate and ribose-5-phosphate are classically produced by the pentose phosphate pathway and ribose-5-phosphate by the glycolysis pathway.
Among the 19 seeds identified in B. cicadellinicola, we found 13 amino acids or related metabolites (such as homoserine or 2-ketovaline) and only one sugar, glucose.
In the compound graph of B. cicadellinicola, serine belongs to the same source component (see Methods) as threonine and glycine. In fact, there are two reversible reactions that, respectively, link serine and threonine to glycine (Figure 5). It is thus impossible to distinguish which one(s) actually produces the other(s). All were considered as potential seeds and were taken into account by the PITUFO method for detection of the precursor sets.
There is only one other example of such alternative seeds detected by the method of Borenstein [10]: these are oxaloacetate and aspartate linked by the same reversible reaction in the metabolic network of S. muelleri (Figure 6). All the other seeds found are metabolites that are not produced by any reaction.
Among the 16 seeds specific to B. cicadellinicola, five are produced by S. muelleri and two are certainly not provided by the insect host: threonine and lysine. Five seeds were already mentioned as potentially provided by the sharpshooter: glucose 6-phosphate, tyrosine, glycine, glutamate and alanine. Protoheme and porphobilinogen were mentioned by Wu et al. as needed to be imported by B. cicadellinicola to complete the siroheme biosynthesis pathway [6]. They seem not to be produced by S. muelleri and should be provided instead by the insect.
Five seeds identified in the metabolic graph of S. muelleri are produced by B. cicadellinicola: erythrose-4-phosphate, phosphoenolpyruvate, ribose-5-phosphate, octaprenyl-diphosphate and oxaloacetate, but all could be also available in the insect cell.
Other seeds that do not seem to be produced by the co-endocytobiont were not reported before as potentially provided by the host. We assume that these seeds are produced by the insect or present in its diet. Knowledge of the metabolic network of the sharshooter should confirm or disprove the production of these metabolites by the insect host.
Figures 7 and 8 indicate the precursor sets for the target metabolites selected in the metabolic networks of the two endocytobionts (see Methods).
The parsimony of the metabolic network of both bacteria is reflected in the small number of precursor sets found for the target metabolites to which we applied PITUFO: the maximum number of solutions for a target is only three and the maximum total number of involved precursors is seven.
For S. muelleri, apart from menaquinone, all targets are amino acids, which explains the uniformity of the results. Two seeds are present in all minimal precursor sets computed for these amino acids (except homoserine): erythrose-4-phosphate and phosphoenolpyruvate. Both are potentially provided by B. cicadellinicola. We found oxaloacetate and aspartate as alternative precursors for the synthesis of isoleucine and lysine. Indeed, each one can produce the other by the same reversible reaction in the metabolic network of S. muelleri (Figure 6). This leads to two possible scenarii, depending on which one of them is actually provided. Aspartate is one of the primary components of the xylem sap, it is thus reasonable to think that this compound should be provided by the host. In the metabolic network of S. muelleri, aspartate is involved in other reactions that in particular participate in the synthesis of other amino acids. Oxaloacetate is only involved in the reaction that produces aspartate. Since S. muelleri is able to produce aspartate from oxaloacetate, and since the former is not used in other reactions, the import of oxaloacetate by S. muelleri seems to be more realistic than the import of aspartate. Moreover, B. cicadellinicola could provide oxaloacetate while the bacterium is able to synthesise it from aspartate (see Figure 8).
Some seeds appear as obligatory in the synthesis of several targets in B. cicadellinicola. Glucose and aspartate, reported as provided by the insect cell, thus appear as obligatory for the synthesis of, respectively, twelve and five target compounds.
As mentioned previously, serine, glycine and threonine have been detected as alternative seeds by the Borenstein method [10]. For B. cicadellinicola, they appear in the minimal precursor sets for methionine, coenzyme A, glutathione and thiamine. McCutcheon et al. suggested that homoserine and 2-ketovaline, potentially provided by S. muelleri, could be precursors of metabolites supplied by B. cicadellinicola. Homoserine was reported as a precursor of methionine and 2-ketovaline as a precursor of coenzyme A [7]. Our results confirm these hypotheses. For methionine, our method adds precision by indicating also the alternative precursors serine-glycine-threonine. For coenzyme A, our method further suggests this triplet and also -alanine as obligatory precursors.
Interestingly, we observed that only methionine and coenzyme-A require metabolites provided by S. muelleri. Moreover, the metabolites needed by the other targets could be all potentially acquired from the host cell by B. cicadellinicola.
Graph-based modelling of the metabolic networks of B. cicadellinicola and S. muelleri enabled us to complete and precise the description of the metabolic exchanges between these two endocytobionts and with their host, the sharpshooter Homalodisca coagulata. By automatically computing the set of seeds for each metabolic network, we thus offer the first exhaustive list of metabolites potentially imported by S. muelleri and B. cicadellinicola. By using our method to find precursor sets for given target compounds, we provide a general and detailed view of the metabolic exchanges that potentially lead to the synthesis of metabolites involved in the mutualistic association.
The definition of seeds by Borenstein et al. [10] allowed to indicate alternative ones that could not be found only by defining the seeds as metabolites not produced by any reaction. We found only two instances of such alternative seeds: oxaloacetate-aspartate in the metabolic network of S. muelleri and glycine-serine-threonine in the metabolic network of B. cicadellinicola. The method of Borenstein et al. remains highly suitable to detect seeds in metabolic networks where many reactions cannot be assigned a direction.
From the list of seeds previously defined, the method we developed, PITUFO, was able to find the precursor sets of metabolites reported as involved in the symbiotic association. Contrary to previous methods, PITUFO is an exact algorithm and returns all the precursor sets for a given target compound. In addition, by explicitly taking into account cycles in the definition of precursors and in the algorithm, PITUFO is able to find solutions not reachable by the previous methods. Unfortunately, because the implementation of previous methods is not available or is dataset-dependent, we were not able to compare their performance with the one of PITUFO.
Most of our results could only hardly be found by manual analysis of the metabolic pathways. However, the pertinence of our or of previous analyses is highly linked to the quality of the metabolic network reconstructions. The most time-consuming part in this study was then to refine the metabolic reconstructions available for the two bacteria (see Methods). Interestingly, the methods we used to find seeds and precursor sets also helped us to refine the metabolic reconstructions when some inconsistencies were found. There are several ways to complete this study and to improve the tools that we used. First, PITUFO only returns sets of precursors and not the possible hyperpaths between them and the target compounds. The identification of key metabolites, such as those involved in hyperpath intersections, and compression of the information contained in the metabolic hyperpaths could be a way to provide results easier to interpret for the analyst. When alternative precursor sets are indicated, it would be interesting to point to those that are the most likely to be actually used. Taking into account the stoichiometric coefficients would allow to prune precursor sets not consistent with the stoichiometric constraints. Measuring the production rate of the target compounds would be a way to sort the precursor sets. Finally, PITUFO is restricted to the identification of all the minimal precursor sets leading to the production of the set of targets specified by the user, putting aside the other metabolic functions, even vital for the organism. The identification of all minimal precursor sets leading to the production of both essential metabolites (e.g. those participating to the biomass) and metabolites involved in the mutualistic association was beyond the scope of this study but is certainly of interest and will be developed in the future.
For both bacteria, the number of seeds that we identified is very reduced, even considering that our study is limited to the carbon metabolism of small molecules. This means that the global reduction of the metabolism in the symbionts comes with a reduction in the number of metabolites imported from the host cell. The identification by PITUFO of a unique precursor set for most of the selected target compounds shows that there are almost no alternative sources to produce essential compounds. Indeed, the mutualistic association of the symbionts with their host is very ancient (70 to 100 millions years for B. cicadellinicola and approximatively 280 millions years for S. muelleri) [21]. The stability of their environment, particularly because of their vertical mode of transmission, made their metabolism specialised in the exploitation of a restricted set of substrates.
However, our results showed that the two bacteria use very differently their environment. Indeed, only three seeds common to the two metabolic networks have been identified. Even the common seeds have a completely different fate in the two bacteria as they are not involved in the same metabolic pathways. The complementarity of the two metabolisms is then not only manifested in the metabolic capabilities of each organism but also by their different use of the nutrients available in the host cell.
The set of seeds identified in the metabolic network of B. cicadellinicola is mainly composed of amino acids or related metabolites such as 2-ketovaline or homoserine. Three seeds identified in the metabolic network of S. muelleri are also amino acids. The presence of amino acids in the seeds identified in B. cicadellinicola (glutamate, lysine, alanine, serine, aspartate and glycine) or in S. muelleri (serine, aspartate and cysteine) is interesting in the sense that they are not only provided as essential building blocks of proteins but also as starting points of the biosynthesis of other metabolites. This clarifies the role of the exchanged amino acids as reported in earlier studies [6], [7]. Conversely, the absence of other amino acids in the seeds indicates that they do not participate to the formation of the carbon backbone of other compounds.
Some seeds automatically defined in our study were already mentioned in earlier studies [7]. Aspartate, identified as a common seed in both metabolic networks, is an amino acid indicated to exist in great concentration in the xylem sap that the sharpshooter feeds upon [7]. Its large availability in the direct environment of the two bacteria makes of it an efficient source for the production of other metabolites. Furthermore, a specific aspartate transporter has been annotated in the B. cicadellinicola genome. Other metabolites such as non-essential amino acids and glucose, the only sugar identified as seed in the metabolic network of B. cicadellinicola, were also mentioned as components of the xylem sap and then available for the two symbionts [7].
However, some metabolites mentioned as highly present in the xylem sap were not identified in our set of seeds. For instance, arginine, an amino acid which is abundant in proteins, is absent from the metabolic network of B. cicadellinicola and appears only as output in the metabolic network of S. muelleri. Glutamine in both metabolic networks is only used as a nitrogen source and thus does not appear in the filtered metabolic networks. Malate is completely absent from the metabolic network of S. muelleri. It is produced from fumarate in the metabolic network of B. cicadellinicola and then does not need to be imported.
Three seeds identified in the metabolic network of S. muelleri are glycolytic products. Erythrose-4-phosphate and ribose-5-phosphate are commonly produced in the pentose phosphate pathway. Phosphoenolpyruvate is commonly produced during glycolysis. PITUFO returned erythrose-4-phosphate and phosphoenolpyruvate as obligatory precursors in the synthesis of the metabolites that S. muelleri provides to the symbiotic system, except homoserine and menaquinone. Ribose-5-phosphate was identified as obligatory precursor for tryptophan. These three metabolites, as well as oxaloacetate and octaprenyl diphosphate, are produced by the metabolic network of B. cicadellinicola. However, it is likely that these metabolites could be made available by the insect host. This means that the carbon metabolism of S. muelleri may be completely independent of the metabolic network of B. cicadellinicola. On the contrary, the two essential amino acids (threonine and lysine) identified as seeds for B. cicadellinicola are certainly not produced by the insect host nor present in its diet and must be provided by S. muelleri. The carbon metabolism of B. cicadellinicola therefore appears as dependent on the metabolism of S. muelleri, at least for these two amino acids. This dependence is added to the obligatory supply of other essential amino acids by S. muelleri that are required for protein biosynthesis in B. cicadellinicola [6], [7].
However, among the precursors identified for the synthesis of metabolites that B. cicadellinicola passes on to the symbiotic system, only methionine and coenzyme A need metabolites produced by S. muelleri: homoserine and 2-ketovaline. The first one may be produced by the plant and then be present in the diet of the insect, and the second one may be produced by the insect via the degradation of valine. This suggests that B. cicadellinicola, as well as S. muelleri, may be only dependent on the metabolites obtained from the insect to produce the metabolites provided to the symbiotic system. Indeed, threonine and lysine which are supplied by S. muelleri to B. cicadellinicola, are only exploited by the latter to produce its proteins.
The reconstruction of the metabolic network of the host and a better knowledge about the metabolome of the plants that the insect feeds upon will inform us whether some seeds, such as intermediates in the biosynthesis of essential amino acids, are actually produced by the insect or present in its diet.
One challenging issue concerns how the metabolites are exchanged between the three partners. The annotation of transporters for amino acids and sugars in B. cicadellinicola [6] gives only a partial answer to this question. Indeed, very few transporters were found during the annotation of the genome of S. muelleri, and none corresponds to the seeds that we identified [7]. This means that other scenarii have to be proposed to explain the exchanges of metabolites in the symbiotic system. In particular, the cells of B. cicadellinicola often appear to adhere to the surface of the much larger cells of S. muelleri [6]. This proximity should facilitate the exchanges between the two bacteria. However, the two bacteria seem to be not always in the same cells [6]. This poses the problem of how the essential amino acids are provided to B. cicadellinicola.
One other remaining interesting question is evolutionary: how did the reductions of the metabolism of the two symbionts get organised during evolution to reach their current complementarity? A recent study compared the metabolic gene sets of two pairs of co-resident endocytobionts. One pair was formed by the two endocytobionts of the sharpshooter studied in this paper and the other pair was formed by another strain of S. muelleri (SMDSEM) and by Hodgkinia cicadicola, found in the cells of cicadas [22]. The authors showed that the two strains of S. muelleri exhibit almost identical metabolic capabilities. They suggested also that, although phylogenetically distant, H. cicadicola and B. cicadellinicola have converged on similar metabolic functions, especially those that are complementary to the metabolism preserved in S. muelleri.
The application of such methods as we used in this study to this other pair of co-resident endosymbionts should allow to identify some common patterns in the sharing of a set of nutrients and their use in the metabolic networks of the different partners. The extension to other endosymbiotic systems could provide us with crucial information to understand the establishment of such nutritional associations.
|
10.1371/journal.ppat.1000768 | Direct Presentation Is Sufficient for an Efficient Anti-Viral CD8+ T Cell Response | The extent to which direct- and cross-presentation (DP and CP) contribute to the priming of CD8+ T cell (TCD8+) responses to viruses is unclear mainly because of the difficulty in separating the two processes. Hence, while CP in the absence of DP has been clearly demonstrated, induction of an anti-viral TCD8+ response that excludes CP has never been purposely shown. Using vaccinia virus (VACV), which has been used as the vaccine to rid the world of smallpox and is proposed as a vector for many other vaccines, we show that DP is the main mechanism for the priming of an anti-viral TCD8+ response. These findings provide important insights to our understanding of how one of the most effective anti-viral vaccines induces immunity and should contribute to the development of novel vaccines.
| Professional antigen presenting cells fragment viral proteins and display some of the resulting peptides bound to MHC molecules at the cell surface. When virus-specific CD8+ T cells recognize these viral peptides they become activated, proliferate, and kill virus-infected cells to help rid the body of the virus. Two pathways have been described for the origin of the peptides presented by professional antigen presenting cells. In cross-presentation, the antigen presenting cells acquire the proteins from other cells which, in the case of a viral infection, must be infected. In direct presentation, the antigen presenting cells synthesize the proteins themselves and, therefore, during responses to viruses must be infected. However, the participation of direct presentation in anti-viral responses has never been deliberately demonstrated experimentally. In this paper we demonstrate that direct presentation occurs and is the main pathway to induce CD8+ T cells during infection with vaccinia virus. These findings provide important insights to our understanding of how one of the most effective anti-viral vaccines induces immunity and should contribute to the development of novel vaccines.
| Activated CD8+ T lymphocytes (TCD8+) kill virus infected cells that display virus-derived peptides presented on cell surface MHC I molecules. Hence, TCD8+ play an essential role in the clearance of many primary viral infections. Moreover, the memory TCD8+ that remain after a primary infection or vaccination can also participate in resistance to disease following a secondary infection [1],[2],[3],[4]. While most cells of the body express MHC I and can therefore be targets of TCD8+ killing, their initial activation and expansion (priming) during many viral infections requires antigen presentation by bone marrow-derived (BMD) professional antigen presenting cells (APC) [5],[6],[7]. The two major routes of MHC I antigen presentation are direct- and cross-presentation (DP and CP). In DP the Ag presenting cell synthesizes the Ag. Thus, DP presentation requires the infection of the Ag presenting cell. In CP, uninfected cells acquire the Ags from other infected cells. While most cells can engage in DP, CP is a function of phagocytic BMD APC such as DC and Μφ [8],[9]. Several years ago we showed that when a virus cannot infect BMD APC, CP can still prime anti-viral TCD8+ [6]. Since then, the specific role of CP and DP in priming anti-viral TCD8+ has been a topic of discussion with some arguing that CP is in general important or essential, whereas others propose that it is physiologically irrelevant [8],[10],[11],[12],[13],[14]. The main reason for this ongoing discussion is a dearth of direct data supporting DP or CP as the main mechanism of TCD8+ priming in viral infections [15]. This most likely resulted from the difficulty in establishing appropriate experimental models that can exclude CP during an anti-viral response while maintaining similar levels of peptide-MHC complexes at the cell surface. For example, previous work by us and others has shown that (M)SIINFEKL expressed as a mini-gene during VACV infection is not a substrate for CP [16],[17] and further earlier work by Restifo et al. and Wherry et al. [18],[19] had shown that (M)SIINFEKL can prime TCD8+. Placing both pieces together, it could be argued that DP by VACV-infected cells has already been shown. However, because it does not require processing, VACV-(M)SIINFEKL infected cells express supra-physiologic Kb-SIINFEKL complexes at the surface of infected cells (∼85,000 vs. 3,000 complexes per cell for VACV-full-length OVA [20]), has an extremely short half-life [21], and its ability to stimulate TCD8+ responses does not correlate with the very high levels MHC I-peptide complexes at the cell surface [19]. Furthermore, whether this construct requires BMD APC has not been investigated. Similarly, Norbury et al. has shown that SIINFEKL embedded in a rapidly degraded construct (Ub-R-NP-SIINFEKL-EGFP) is not cross-presented but induces a TCD8+ response [17]. However, while this construct requires processing, it is degraded very fast (10 minutes), resulting in faster Kb-SIINFEKL formation and at least three times more Kb-SIINFEKL complexes at the surface of infected cells as compared with a slowly degraded counterpart NP-SIINFEKL-EGFP [21]. Understanding how TCD8+ are primed, in particular for those viruses that are useful as vaccines, is of major importance as it may directly impinge on vaccine efficacy. Here, we explore the role of DP and CP in the priming of TCD8+ to vaccinia virus (VACV) which was used as the vaccine that eliminated smallpox and is proposed as a vaccine vector for a number of infectious diseases and cancer [22],[23].
Previous work by others showing clustering of TCR transgenic TCD8+ (TCD8+) with VACV infected APC suggested that DP can prime anti-VACV TCD8+ responses [24],[25]. However, this work did not formally prove that this clustering resulted in effective priming or TCD8+ expansion. To directly look into this issue, we made VACV-Kb+46-SIINFEKL-16, a double recombinant VACV co-expressing the MHC I molecule H-2 Kb and 46-SIINFEKL-16, a truncated form of chicken ovalbumin (OVA, 386 amino acids) comprising the Kb-restricted immunodominant determinant SIINFEKL preceded by 46 and followed by 16 amino acids from the natural OVA sequence. Of interest, this construct is a substrate for DP but not CP [26]. As shown in Figure 1A, Kb-negative A9 cells infected with VACV-Kb+46-SIINFEKL-16 induced B3Z T cells, a T cell hybridoma that produces β-galactosidase (β-gal) upon recognition of Kb-SIINFEKL and can be used to compare amounts of Kb-SIINFEKL at the surface of cells [27],[28]. On the other hand, control A9 cells infected with VACV wild type (VACV-WT) or the single recombinants VACV-Kb and VACV-46-SIINFEKL-16, did not induce β-gal in B3Z cells. Additional controls showed that B3Z cells were induced when infecting Kb+ MC57G fibrosarcoma cells with VACV-Kb+46-SIINFEKL-16 or VACV-46-SIINFEKL-16 but not with VACV WT or VACV-Kb (not shown). Thus, virus encoded Kb can directly present virus encoded SIINFEKL in tissue culture. To determine whether virus-encoded Kb results in DP in vivo, B6.C-H2bm1/ByJ mice [bm1 mice; a C57BL/6 (B6) congenic strain carrying a mutant Kb allele (Kbm1)] were adoptively transferred with CFSE labeled splenocytes from OT-I TCR transgenic mice [29] and infected with various viruses. As shown in Figure 1B and C, the OT-I TCD8+ proliferated extensively and significantly increased in proportion relative to the endogenous TCD8+ population when the mice were infected with VACV-Kb+46-SIINFEKL-16 but not when infected with VACV-Kb or VACV-46-SIINFEKL-16. Some loss of CFSE fluorescence in a sizeable number of the OT-I cells in mice infected with VACV-46-SIINFEKL-16 (Figure 1B, center panel) was not reproducible (see the wide SD), and was most likely background because the proportion of OT-I cells did not increase significantly in these mice (Figure 1C). Similar results were obtained in the D-LN of mice inoculated IP and in the spleen and D-LN of mice inoculated SC (Figure S1A). In control experiments, VACV-46-SIINFEK-16 strongly stimulated OT-I cells in B6 mice (Figure 1B, right panel). Of note, the OT-I cells in B6 mice infected with VACV-46-SIINFEKL-16 expanded much more than in bm1 mice infected with VACV-Kb+46-SIINFEKL-16 as indicated by their significantly higher proportional increase (to 42.6±6.3% of total TCD8+, not shown) most likely indicating that Kb expressed by the virus cannot faithfully reproduce endogenous Kb expression. Regardless, because OT-I TCD8+ cells recognize SIINFEKL in the context of Kb but not of Kbm1 [30],[31],[32], the results with bm1 mice strongly suggest that infected cells can directly present antigen to OT-I cells in vivo.
While OT-I cells have been used extensively to detect antigen presentation in vivo, there is the caveat that, because of their high TCR affinity, their priming requirements likely differ from those of a polyclonal naïve repertoire. In fact, their value as a tool in priming and T cell kinetics experiments has been questioned [33]. Thus, to extend our findings to a polyclonal naïve repertoire we determined whether infection with a Kb-expressing virus can induced an endogenous TCD8+ response in bm1 mice. For this purpose, we infected bm1 mice with 106 PFU VACV-46-SIINFEKL-16 or VACV-Kb+46-SIINFEKL-16 and, seven days PI, we determined the endogenous TCD8+ responses to SIINFEKL and also to the dominant Kb-restricted genuine VACV determinant TSYKFESV [34] using appropriate Kb tetramers. We found that VACV-Kb+46-SIINFEKL-16 but not VACV-46-SIINFEKL-16 was able to stimulate significant anti-Kb-SIINFEKL and anti-Kb-TSYKFESV responses in the peritoneal cavity of bm1 mice (Figure 1D and E) demonstrating that VACV infected cells can use DP to expand polyclonal (non-transgenic) TCD8+ to the recombinant determinant SIINFEKL and also to TSYKFESV in bm1 congenic mice. Similar results where obtained for the spleens, peritoneal washes and lymph nodes of mice infected with 105 PFU of the viruses either IP or SC (Figure S1B, showing examples of two individual mice to demonstrate reproducibility). The response was Kb-peptide and not Kb-allo -specific because Kb-SIINFEKL tetramers stained a significant proportion of TCD8+ in bm1 mice infected with VACV-Kb+ 46-SIINFEKL-16 but not with VAC-Kb (Figure S1C and D). The data also indicate that the repertoire of bm1 mice includes at least some TCRs capable of recognizing TSYKFESV and SIINFEKL in the context of Kb. However, the response in bm1 mice, in particular against TSYKFESV, was much smaller than in B6 mice (see, for example, Figure 3D and E). This may be due to different expression of virus-encoded vs. endogenous Kb (as with the OT-I cells) and/or defective positive selection of Kb-restricted T cells in the bm1 thymus as previously reported by Nikolic-Zugic et al. [32]. The fact that we detected Kb-SIINFEKL specific cells in bm1 mice while Nikolic-Zugic did not may be because we used a more potent antigenic stimulus (OVA encoded by VACV vs. OVA-loaded cells) and/or that they detected the responses using the 51Cr release assays while we used tetramer staining.
The previous data strongly suggested that DP can prime anti-VACV TCD8+. However, the experimental system had the caveat of using a semi-allogeneic system and that it does not distinguish between direct priming by infected cells of bone marrow vs. non-bone marrow (parenchymal) origin. We have previously used bone marrow chimeras with deficient expression of MHC I at the cell surface of BMD cells (from TAP1 deficient mice) to show that only BMD APC can prime TCD8+ responses to VACV and other viruses [6]. Thus, to directly address the role of DP by BMD APC in the priming of endogenous TCD8+ responses, we reconstituted lethally irradiated B6 mice with bone marrow from mice deficient in H-2 Kb and Db (MHC I KO). Four months after reconstitution, most cells in the spleen, bone marrow (Figure 2A), and peritoneal wash (not shown) of MHC I KO→B6 mice lacked Kb and Db expression with the exception of a small residual population of cells in the bone marrow (∼2%) and spleen (∼5%), most of which were not “professional” APC because they lacked MHC II expression. Thus, the vast majority of professional APCs in aged MHC I KO →B6 mice lack DP as well as CP abilities due to deficient MHC I expression. However, because these APC lack MHC I but otherwise their antigen presentation machinery is intact, they could regain at least some DP capabilities if infected with a Kb-expressing virus. In addition, the APC in MHC I KO → B6 mice should also be capable of presenting pre-formed Kb-peptide complexes obtained thorugh membrane exchange (ME) with parenchymal cells, a mechanism of antigen presentation that was discovered somewhat recently [35],[36],[37],[38],[39]. Four months after reconstitution, the MHC I KO→B6 and B6→B6 control mice were infected with recombinant VACV-β-gal or with VACV-Kb. Seven days later, the anti-TSYKFESV TCD8+ response was measured in different organs by restimulating lymphocytes for 4 h with APC pulsed with TSYKFESV in the presence of brefeldin A followed by surface (CD8) and intracellular IFN-γ staining (IIS) and FACS analysis. We found that VACV-β-gal infection of MHC I KO→B6 mice resulted in an anti-TSYKFESV response in the spleen that was very reduced as compared to B6→B6 controls (Figure 2B), confirming our previous work [6] demonstrating that BMD APC are essential for the anti-VACV TCD8+. Moreover, this experiment shows that priming by ME (which was unknown at the time of our previous work) from parenchymal cells to APC does not play a dominant role in the anti-VACV TCD8+ response. More important, we found that much of the anti-TSYKFESV response was significantly restored when the MHC I KO→B6 chimeras were infected with VACV-Kb. Furthermore, MHC I KO→B6 mice mounted a significantly stronger response to TSYKFESV in the peritoneal wash when infected with VACV-Kb as compared with VACV-β-gal (Figure 2B). This strongly supports the hypothesis that BMD cells, but not parenchymal cells, infected with VACV can use DP to prime an endogenous polyclonal TCD8+ response to the VACV immunodominant determinant TSYKFESV.
To determine differences in DP by BMD APC vs parenchymal cells, we infected BMD DC (as a model for APC) and MC57G cells (as a model for parenchymal cells) with VACV 46-SIINFEKL-16 and measured the relative amount of Kb-SIINFEKL complex at the cell surface using B3Z cells (which do not require BMD APC for stimulation). We found that the two cell types were quantitatively comparable in their ability to stimulate B3Z cells indicating that they expressed roughly similar amounts of Kb-peptide complexes at the cell surface (Figure 2C). However, while infected WT DC primed an anti-VACV response in vivo, MC57G cells did not (Figure 2D). This strongly suggests that the difference in the ability of BMD APC vs parenchymal cells to prime TCD8+ to an Ag that needs processing is qualitative rather than quantitative and further suggested that DP or ME by parenchymal cells do not play a major role in the anti-VACV TCD8+ response. Of note, the priming following inoculation of infected DC was due to their expression of MHC and not to an adjuvant effect because infected DC deficient in MHC I did not induce an anti-VACV response (not shown).
The data thus far demonstrated that anti-VACV TCD8+ responses can be induced by DP. However, the experiments did not address to what extent DP and CP contribute to priming during VACV infection. Using a transfection/infection model, we have recently shown that during VACV infection, 61-SIINFEKL-121 (a truncated form of OVA comprising SIINFEKL preceded by 61 and followed by 121 AA of the natural OVA sequence) and 46-SIINFEKL-16 are processed for DP with similar efficiency. However, even though both constructs have extended half-lives, only 61-SIINFEKL-121 is processed for CP [26]. Hence, we tested whether the antigenic properties of 61-SIINFEKL-121 and 46-SIINFEKL-16 were maintained when expressed in recombinant viruses. As expected, MC57G cells infected with either virus induced B3Z cells by DP with identical efficiency (Figure 3A). On the other hand, consistent with our results with the transfection/infection system [26], infection of A9 cells with recombinant VACV 61-SIINFEKL-121 but not with VACV 46-SIINFEKL-16 resulted in CP in vitro (Figure 3B) and CP to OT-I cells in vivo (Figure 3C). Next, we infected mice with 106 PFU of VACV 61-SIINFEKL-121 or VACV 46-SIINFEKL-16 IP and, using specific MHC tetramers, we determined the potency of the anti-SIINFEKL TCD8+ response in the peritoneal cavity and in the spleen. The anti-TSYKFESV response served as an internal control and to normalize the anti-SIINFEKL response. We did not find any significant difference between the two viruses (Figure 3D and 3E). Similar results were obtained for mice infected SC and/or with higher or lower viral doses (108 PFU and 104 PFU) determined by either tetramer staining or IIS (not shown). Because 46-SIINFEKL-16 is not cross-presented, these results imply that CP is dispensable for the induction of a maximal TCD8+ response during VACV infection independent of the dose or route.
It has been shown that pre-treatment of mice with the TLR9 ligand CpG induces maturation of DC, blocks CP, and inhibits the TCD8+ response to herpes simplex virus (HSV) and influenza virus in vivo [11]. In our hands, this treatment also inhibited CP because mice treated with CpG had significantly reduced TCD8+ responses to SIINFEKL and TSYKFESV when inoculated IP with L cells that had been infected with VACV 61-SIINFEKL-121 to induce Ag expression, and then treated with UV light and paraformaldehyde to eliminate any traces of live virus (Figure 4A). However, CpG treatment did not significantly reduce priming of anti-SIINFEKL or anti-TSYKFESV TCD8+ in mice that had been infected with 103–106 PFU VACV 61-SIINFEKL-121 (Figure 4B–D) or VACV 46-SIINFEKL-16 (not shown), even though the potency of priming decreased with reduced virus dose. These results further imply that CP is dispensable for the induction of efficient anti-VACV TCD8+ responses following infection with live VACV. In fact, the only significant change that we observed with CpG treatment was an increase in the anti-TSYKFESV response in mice inoculated with 106 PFU. The reason for this increase is unknown but we speculate it may be due to an adjuvant effect of CpG. Why this increase was not observed for other viral doses or for SIINFEKL remains to be explored.
We have previously shown the strict requirement for BMD APC in the priming of TCD8+ responses to VACV and other viruses and that CP can prime TCD8+ when DP by BMD APC is abrogated [6],[7]. However, the extent whereby the DP and CP pathways contribute to an anti-viral response when both mechanisms are possible remained elusive because of the difficulty in ablating CP. Hence, priming of an anti-viral response exclusively by DP has never been demonstrated intentionally. In this paper we developed novel methods to disrupt CP and used them to demonstrate efficient priming of anti-VACV TCD8+ by DP following IP and SC inoculation. Furthermore, we show that when DP is available, CP is dispensable for eliciting a maximal anti-VACV TCD8+ response.
It has previously been shown that some anti-viral TCD8+ responses require or are partially dependent on CP. For instance, Shen et al. showed a decreased TCD8+ response to influenza virus in the absence of Cathepsin S, which is required for the processing of exogenous Ag via the TAP independent pathway [40] while Wilson et al.[11] showed that inhibiting CP by administration of the TLR9 ligand inhibited the TCD8+ response to HSV 1. In the case of VACV, we and others have shown that VACV encoded Ags can indeed be cross-presented [16],[41],[42]. Attempts have also been made to quantify the contribution of CP and DP to the overall anti-VACV response. For instance, Gasteiger et al. has shown that the TCD8+ response to the MVA strain of VACV requires CP [43]. However, the different requirements for this strain for VACV could be due to the fact that MVA is highly deficient in viral replication. Also, Basta et al. and Shen et al. [44],[45] compared the TCD8+ responses to recombinant VACV expressing US2 and/or US11 from human cytomegalovirus (HCMV) US11, or β-gal as a control. Because these viruses induced TCD8+ responses to different degree depending on the route of infection, it was concluded that CP and DP contribute differentially to the anti-VACV TCD8+ response. However, the conclusions assumed that US2 and US11 shut down DP in vivo, which has never been demonstrated. Moreover, the conclusions were based on the presumption that molecules that inhibit the MHC I pathway could not maintain functionality and block CP when transferred from the Ag donor cell to the APC. However, more recent work from the Cresswell laboratory [46] showed that exogenous ICP47 from HSV (another protein that blocks MHC I Ag presentation) can block CP making the supposition doubtful. In addition, while the direct interaction between infected APCs and TCR transgenic cells specific for a virus encoded Ag has been shown [24],[25], a clear demonstration of direct priming of naïve polyclonal anti-viral TCD8+ by infected APC expressing MHC I-peptide at relatively normal levels was still lacking. Here we have used four novel models to demonstrate that in vivo priming of anti-viral TCD8+ by DP occurs and that CP is dispensable to efficiently prime anti VACV TCD8+ in vivo. First, we used a semi-allogeneic model where the restricting MHC I and the Ag were exclusively encoded by VACV. Using this model we showed that following SC or IP infection, DP can stimulate TCR transgenic OT-I T cells and can also prime endogenous polyclonal responses to a recombinant (SIINFEKL) and an authentic (TSYKFESV) VACV determinant. It should be pointed out, however, that the OT-I responses in bm1 mice were not as strong as in B6 mice probably because the expression of endogenous MHC I cannot be faithfully replicated by virus-driven expression and, in the case of the endogenous responses, the repertoire capable of recognizing peptides in the context of Kb may be reduced in bm1 mice. Second, using bone marrow chimeras that lack expression of MHC I on BMD APC and infecting with VACV-Kb or control virus or inoculating with infected cells of bone marrow or parenchymal origin, we also showed priming by DP against TSYKFESV following IP or SC infection or DC inoculation. Further, we ruled out the transfer of preformed peptide MHC I complexes [35],[36],[37],[38],[39] from endogenous or inoculated parenchymal cells as a major mechanism for priming during VACV infection. In addition, these data also confirmed our earlier work that the priming of anti-VACV TCD8+ requires Ag presentation by BMDC [6]. Third, by comparing TCD8+ responses to 46-SIINFEKL-16, a form of OVA that is not cross-presented and 61-SIINFEKL-121, a form of OVA that is cross-presented [26], we showed that CP is not essential for full-fledged TCD8+ responses to VACV independent of the route or dose of infection. Fourth, we showed that in vivo blockade of CP using the TLR9 ligand CpG does not inhibit the anti-VACV TCD8+ response as it did for HSV [11]. Together, our experiments demonstrate that DP is the main mechanism for the priming of anti-VACV TCD8+.
Current models of Ag presentation mostly based on inert Ag suggest that APC acquire Ag in tissues, then mature, and finally migrate to the draining lymph node (D-LN) to prime T cells. While it is straightforward to imagine an uninfected APC loaded with Ag migrating to the D-LN, it is also possible to imagine that an APC infected with a cytopathic virus such as VACV would be migration-impaired. Thus, a remaining important question is to determine whether infected APC are still able to migrate to the D-LN following SC inoculation. Alternatively, free viral particles could reach the D-LN through afferent lymphatic capillaries as was shown with large inoculums of vesicular stomatitis virus [47] infecting D-LN resident APC. The site of priming following IP infection is more obscure and while it is possible that it occurs in the (para-thymic) D-LN, it is tempting to speculate that the peritoneal cavity, which has large nuber of BMD Μφ, could act as a secondary lymphoid organ.
In summary our work demonstrates that DP is the main mechanism responsible for the priming of anti-VACV TCD8+ responses. These results are important for our general understanding of anti-viral TCD8+ immunity and for the use of VACV as a vaccine vector.
All experiments involving mice were performed according to Fox Chase Cancer Center guidelines for the care and use of laboratory animals and all animal studies were approved by the Fox Chase Cancer Center Institutional Animal Care and Use Committee.
All cells were grown at 37°C in an atmosphere of 5% CO2 in RPMI 1640 medium supplemented with 10% FCS, 2 mM L-glutamine, penicillin-streptomycin, 0.01 M HEPES buffer and 5×10−5 M 2-ME (Sigma-Aldrich, St. Louis, MO). As L cells (H-2K) we used its derivative A9 (ATCC no. CCL-1.4). L cells stably expressing Kb (L-Kb) [34], were a gift from Drs. Yewdell and Bennink. MC57G cells (ATCC no. CRL-2295) are a C57BL/6 fibrosarcoma (H-2b). B3Z is a CD8 T cell hybridoma that produces β-gal upon recognition of SIINFEKL in the context of the H-2Kb molecule [48] without the need of costimulation. Hela S3 (CCL –2.2) and BS-C-1 (CCL-26) were used to propagate virus and determine VACV titer. In vitro differentiation of DC and Μφ from bone marrow was as previously described [42].
VACV stocks were prepared as described [49] VACV-46-SIINFEKL-16 and VACV-61-SIINFEKL-121 were previously described [26]. The VACV-Kb in Figure 2 was a gift from Drs. Jonathan Yewdell and Jack Bennink (NIH, Bethesda, Maryland) and co-expresses β-gal and Kb disrupting the TK gene. VACV-β-gal was generated by homologous recombination into the TK gene using the plasmid pSC65 as described [50]. The VACV-kb in Figure 1, VACV-Kb+46-SIINFEKL-16 and VACV 61-SIINFEKL-121 were generated by homologous recombination using appropriate constructs inserted in the plasmid pRB21 and selection of large plaques as described [50]. The correct sequence of the recombinant proteins was verified by sequencing PCR fragments amplified from viral DNA.
C57BL/6 (B6) were from Fox Chase Cancer Center stock. B6.C-H2bm1/ByJ (bm1, stock #001060) B6.PL-Thy1a/CyJ (B6-Thy1.1, stock # 000406), B6.129S7-Rag1tm1Mom/J and (Rag1 KO stock # 002216) were bred at FCCC from mice purchased from Jackson Laboratories (Bar Harbor, Maine). H-2Kbtm1, H-2Dbtm1 (MHC I KO, stock # 004215-MM) were purchased from the Emerging Models Program at Taconic Farms (Germantown, NY) and bred at FCCC. OT-I mice [29], originally a gift from Dr. Stephen Jameson (University of Minnesota, MN), were bred with Rag1 KO and B6-Thy1.1 to homozygosity at FCCC. Bone marrow chimeras were prepared as previously described [6],[7] using 5–7 weeks old mice as donors and recipients. Except for bone marrow chimeras, all experiments used mice between 6–12 weeks of age. Mice were infected or injected with infected cells as indicated. Bone marrow chimeras were prepared as previously described [6],[7]. For CpG treatment, mice were injected intravenously in the tail vein with 20 nM synthetic phosphorothioated CpG1668 (Integrated DNA Technologies Inc, Coralville, IA) [11].
In vitro DP and in vitro and in vivo CP assays were performed as previously described [16],[42] except that for Ag expression we used recombinant viruses rather than plasmid transfection and infection with WT virus. Thus, for in vivo and in vitro CP, the virus was inactivated by UV irradiating the Ag donor cells as described [42] and fixing with 2% paraformaldehyde overnight followed by extensive washing. To determine DP by inoculated cells, DC or MC57G cells were infected with VACV, 10 PFU/cell for 1 h, thoroughly washed, and 106 were inoculated into mice as indicated.
Determination of proliferation and expansion of CFSE labeled OT-I cells was as before [26]. IIS was performed as previously described [3],[4],[51] except that in some cases, instead of infected cells, the virus-specific TCD8+ were restimulated with cells pulsed in complete media with 1 µM synthetic peptides (Genscript corp) for 1 h in CRPMI and thoroughly washed. Kb-tetramers were produced and used exactly as described [52] except that the SIINFEKL or TSYKFESV peptide were used for the refolding reaction.
One- or two-tailed T test analyses were used according to the hypothesis being tested. Tests were performed using the Graph Pad Prism software.
|
10.1371/journal.pntd.0002848 | Assessing the Relationship between Vector Indices and Dengue Transmission: A Systematic Review of the Evidence | Despite doubts about methods used and the association between vector density and dengue transmission, routine sampling of mosquito vector populations is common in dengue-endemic countries worldwide. This study examined the evidence from published studies for the existence of any quantitative relationship between vector indices and dengue cases.
From a total of 1205 papers identified in database searches following Cochrane and PRISMA Group guidelines, 18 were included for review. Eligibility criteria included 3-month study duration and dengue case confirmation by WHO case definition and/or serology.
A range of designs were seen, particularly in spatial sampling and analyses, and all but 3 were classed as weak study designs. Eleven of eighteen studies generated Stegomyia indices from combined larval and pupal data. Adult vector data were reported in only three studies. Of thirteen studies that investigated associations between vector indices and dengue cases, 4 reported positive correlations, 4 found no correlation and 5 reported ambiguous or inconclusive associations. Six out of 7 studies that measured Breteau Indices reported dengue transmission at levels below the currently accepted threshold of 5.
There was little evidence of quantifiable associations between vector indices and dengue transmission that could reliably be used for outbreak prediction. This review highlighted the need for standardized sampling protocols that adequately consider dengue spatial heterogeneity. Recommendations for more appropriately designed studies include: standardized study design to elucidate the relationship between vector abundance and dengue transmission; adult mosquito sampling should be routine; single values of Breteau or other indices are not reliable universal dengue transmission thresholds; better knowledge of vector ecology is required.
| Routine sampling of mosquito vector populations is common in dengue-endemic countries worldwide despite doubts about methods used or the correlation between vector density and dengue transmission. This systematic review examined the published evidence investigating associations between vector indices and dengue cases. From a total of 1205 papers identified in database searches, 18 were included for review. A range of designs were seen, particularly in spatial sampling and analyses, and all but 3 were classed as weak study designs. Thirteen studies investigated associations between vector indices and dengue cases: 4 reported positive correlations, 4 found no correlation and 5 reported ambiguous/unreliable associations. Of 7 studies that measured the Breteau Index, 6 reported dengue transmission at levels below the currently accepted threshold of 5. There was little evidence of quantifiable associations between vector indices and dengue transmission that could reliably be used to predict outbreaks. Furthermore, appropriately designed studies are required to elucidate the relationship between vector abundance and dengue transmission. Recommendations include: standardizing study designs, particularly with respect to spatial heterogeneity; vector surveillance programs should sample adult mosquitoes; global values of the Breteau Index are not reliable universal dengue transmission thresholds; and better knowledge of vector ecology is required.
| Global dengue incidence has increased markedly over the past 50 years to the point where it is now the most widespread mosquito-borne arboviral disease. The World Health Organisation (WHO) has estimated that 50–100 million dengue infections occur annually, while a recent study calculated that the true figure may be closer to 400 million [1]–[3]. Dengue is endemic throughout the tropics, and almost half of the world's population are at risk of infection, 75% of whom live in the Asia-Pacific region [4]. Dengue has been confirmed in 128 countries worldwide [4], [5] with major social and economic consequences [6]–[10].
Dengue is transmitted by Aedes mosquitoes, primarily by the highly urban-adapted vector Aedes aegypti, and a secondary vector Aedes albopictus [11]. Ae. aegypti thrives in the man-made urban environment, particularly in deprived communities where water storage is routine, sanitation is poor and non-biodegradable containers accumulate.
The abundance of dengue vectors species as well as dengue transmission generally show seasonal variation. Depending on the local ecology, these patterns can be in part driven by meteorological parameters such as rainfall and temperature [12], [13]. Vector surveillance is recommended by WHO and is a routine practice in many dengue-endemic countries to provide a quantifiable measure of fluctuations in magnitude and geographical distribution of dengue vector populations, ultimately with the purpose of predicting outbreaks and evaluating control [14]. The standard protocol relies on the Stegomyia indices, which sample the immature mosquito stages (larvae and pupae) alone [15]. This approach was developed over 90 years ago [16] for yellow fever, a markedly different infection (zoonotic in origin though ultimately transmitted between humans by Ae. aegypti) during a very different era (i.e. in terms of urbanization levels and human population densities). Focks (2004) questioned the reliability and sensitivity of the Stegomyia indices because they correlate poorly with abundance of adult mosquitoes, (i.e. the actual vector stage) which should be sampled directly [15]. Focks and others recommended sampling adult mosquitoes directly or indirectly via pupal/demographic surveys (calculating a pupae per person/area index, defined as the number of pupae divided by the number of residents/area surveyed) [15], [17]. Indices based on actual counts of adult female Ae. aegypti infesting houses are likely to be the most accurate, but this is rarely done [15].
The Stegomyia indices remain central to the monitoring of dengue vector populations. The most commonly used indices are the House (or ‘premise’) index (HI - percentage of houses infested with larvae and/or pupae;) the Container index (CI - percentage of water-holding containers infested with larvae and/or pupae) and the Breteau index (BI - number of positive containers per 100 houses inspected) [14]. Variations in sampling protocols are common and can lead to significant variations in indices: e.g. sampling may be carried out indoors or outdoors only, or at both locations; the presence of cryptic breeding sites may lead to under-sampling or complete omission of certain sites; failure to distinguish Aedes aegypti/albopictus from other common mosquito species, or from each other, may lead to overestimates. Little is known about the relationship between differing proportions of the various sampled larval instars and the accuracy of these data as proxy measures of adult mosquito abundance [17]. Finally, although ovitraps (water-filled pots in which Aedes aegypti lay their eggs) are widely used as a simple sampling tool, Focks [15] showed very convincingly that their reliability is limited to indicating vector presence or absence.
Despite these doubts, many dengue control authorities worldwide routinely collect vector population data based on these indices, although the mathematical relationship between any of the indices and dengue transmission is far from clear. Thresholds indicating dengue outbreak risk for House and the Breteau indices (HI = 1%, BI = 5) have been used for many years [18], [19], even though these values were developed for yellow fever many decades earlier. Simple thresholds may be valid in some situations [20], but a universal critical threshold applicable across many contexts, has never been determined for dengue. In pursuing the goal of identifying dengue thresholds, Scott & Morrison [21] defined the fundamental knowledge gaps as: 1) what is an acceptable level of dengue risk?; 2) what are the mosquito densities necessary to achieve that goal?; 3) what is the best way to measure entomological risk?; 4) at what geographic scale are the components of dengue transmission important? While a number of mathematical models have explored the value of thresholds or rates of change in the vector population for the prediction of dengue outbreaks [22], [23], these knowledge gaps remain and continue to hinder progress [24]. For convenience, dengue outbreaks are often defined as periods when dengue incidence is equivalent to the mean plus 2 standard deviations during the same month of the previous year [25].
Effective dengue surveillance and early warning systems, using information from multiple epidemiological sources, are an important goal for numerous countries worldwide. To determine the value of vector surveillance for such systems, the findings of a systematic review examining the evidence for a relationship between mosquito indices and dengue cases are reported here.
The aim of the study was to evaluate the potential value of vector or entomological survey data for dengue surveillance by examining the evidence from studies that investigated quantitatively the relationship between vector indices and dengue cases. The specific objectives were:
A review protocol was established and agreed upon by all authors. Guidelines from the Cochrane Handbook for Systematic Reviews and the PRISMA Group were followed as standard methodologies [26], [27]. The databases WHOLIS, PubMed, EMBASE, LILACS and Web of Science were searched using the Medical Subject Heading (MeSH) “dengue” followed by the Boolean operator “and” combined with one of each of the following ‘free text’ terms in succession: ‘entomological surveillance’, ‘oviposition trap’, ‘house index’, ‘container index’, ‘Breteau index’, ‘pupal index’, ‘pupal survey’, ‘adult collection’, ‘sticky trap’, ‘aspirator collection’, ‘resting collection’, ‘landing collection’, ‘vector density’. The reference list of each of the included studies was also searched, and “grey literature” was sought by communication with authors for cited unpublished documents.
Results were collated in EndNote (EndNote X5, Build 7473) where abstracts were reviewed in accordance with agreed inclusion and exclusion criteria. Full text review was completed using ‘Papers’ (Papers 2, version 2.2.10). No limits were placed on year of publication, language or location.
The criteria for inclusion or exclusion of individual studies were set in advance (Table 1) and were used to assess each abstract and/or the full text.
The following definition was used for the term ‘vector surveillance’: “Any ongoing surveillance of entomological indices, including larval indices (House Index (HI), Container Index (CI), Breteau Index (BI)), pupal indices (Pupal Productivity Index (PPI) and other variations), oviposition trap data and data from adult mosquito collections (methods include sticky, traps, CO2, odor-baited, visual or other traps, resting catches, human landing catches), used in relation to dengue outbreak/control.”
Given the strict nature of the inclusion criteria, study design was assessed at the data extraction stage using the Quality Assessment Tool for Quantitative Studies (QATQS) [28]. QATQS provides a recognized standardized method to assess study quality by assigning scores based on possible selection bias, study design, confounders, data collection methods, intervention integrity and statistical analyses. This ensured each study could be ranked qualitatively. The study design classes were intervention, case-control and longitudinal. If clarification was required, authors were contacted for any missing data or information.
The information extracted included first author, year of publication, year of study, population size, study design, indices and case definitions, study objectives, duration of study, frequency of data collection, results and conclusions (as viewed by all reviewers; Table S1). A table of bias was created to help identify the strengths and weaknesses of each study (Table S2).
No ethical review was required for this systematic literature review.
A total of 1205 potentially relevant studies were identified in the database search. After reviewing abstracts, 102 were selected and retrieved for full text evaluation, of which 18 were considered to have satisfied all inclusion and exclusion criteria and explored in detail (Figure 1) [20], [29]–[45].
Regarding the 84 studies excluded, the most common reasons for exclusion were: study duration less than 3 months (22 studies); absence of a reliable dengue case definition (21 studies); use of datasets that did not correspond temporally or spatially (19 studies). Note that although such dislocated spatial comparisons were not captured by the exclusion criteria originally defined (simply because it had not been expected), exclusion at this point was considered to be valid. Other reasons for exclusion were: measurement of only one outcome (i.e. vector or dengue cases only: 9 studies); opinion or review articles (8 studies); use of incomplete datasets – where only ‘selected’ portions of all of the data available during the study period were used (5 studies). Again, although the latter reason was not captured by the original criteria, exclusion of studies where this occurred was considered to be valid. Full details of the 18 studies reviewed are summarised in the supporting data files (Checklist S1, Table S1, Table S2).
The origin of the data used in analyses differed between studies. Some generated novel data as an integral part of the study, thus ensuring complete or independent control over the quality of the data obtained, while others obtained existing or retrospective data from external sources, including local surveillance data (e.g. local government records, private companies, hospitals or health centers, independent physicians and self-reported data). Twelve studies generated vector data [30]–[32], [34]–[40], [42]–[44], five generated dengue case data [29], [30], [35], [36], [38], four of which generated both vector and dengue case data [30], [35], [36], [38].
Fourteen studies were longitudinal, two were case-control, one was an ecological study (as defined by the unit of analysis) and one was a vector control intervention. Applying QATQS [28], fifteen studies [20], [30]–[33], [35]–[41], [43]–[45] scored 3 (defined as a weak study), two studies [29], [42] scored 2 (a moderate study design) and one study [34] scored 1 (a strong study design)(Annex 2). In the latter study, Chadee and colleagues [34] used controls matched on age and sex from a neighboring community, although the report did not state whether or not this process was randomized.
Details of the sampling protocols used in each study are shown in Table 2. Eleven of eighteen studies generated indices for immature stages of the vector and collected combined larval and pupal numbers to calculate either the CI, HI or BI [20], [29]–[32], [34], [35], [37], [40], [42], [43]. One of these [37] combined Ae. aegypti and Ae. albopictus data. Four studies sampled only larvae [33], [36], [44], [45].
Thirteen studies reported the location of the immature stage mosquito samples: six studies sampled both indoor and outdoor containers [30], [34], [35], [40]–[42], while seven searched indoor containers only [20], [29], [31], [32], [36], [37], [39]. Thus, where reported, all studies included indoor sampling.
Pupal indices were reported in two studies [20], [35]. Adult mosquitoes were sampled in three studies [38], [39], [43].
Thirteen studies examined the association between entomological indices and dengue, using a range of different statistical approaches. Seven studies calculated regression coefficients [36], [37], [39], [40], [43]–[45], two calculated rate ratios [31], [38], one calculated odds ratios [29] and two calculated the G-test for significance [32], [36]. One study used only specificity, sensitivity and positive and negative predictive values [20].
The spatial unit of analysis, an important consideration in dengue epidemiology (see Discussion) varied considerably across studies, with units ranging from individual houses, housing blocks and clusters to neighborhoods and even large municipalities (Table 2).
Four studies reported statistically significant positive relationships between entomological indices and dengue incidence [29]–[31], [39]. Of these, only one sampled adult mosquitoes (33% of those studies that sampled adults) [39] while the remainder sampled immature stage mosquitoes (20% of all those that sampled immatures) [29], [30], [31](Table 2). These are discussed in detail here.
Sanchez (2006) [29] conducted a case control study using two geographical units for analysis, blocks (units of approximately 50 houses) and neighborhoods (each containing approximately 9 blocks). Any block or neighborhood with at least 1 confirmed case was considered positive, while a control was defined as a block or neighborhood without confirmed cases. HI and BI mean values were “consistently, substantially and significantly higher” in blocks with dengue cases compared with control units. An odds ratio (OR) of 3.49 (p<0.05) for dengue transmission was associated with the presence of a single positive container in a block; fifteen of the seventeen dengue cases recorded lived in a neighborhood where at least 1 block had a BI>4.
In Trinidad, Chadee (2009) [30] compared retrospective routine entomological household data with concurrent entomological data taken from confirmed dengue households, using a cardinal points approach (i.e. the ‘index’ house plus the four adjacent houses at its cardinal points). Chadee found that significantly more (P<0.001) immatures were collected during dengue case investigations than during the routine inspection and treatment cycles. The report also stated that pupae per person indices were higher and significantly more adults emerged (as a function of total pupae count collected from household containers) at locations where dengue was confirmed at the index house, compared with routine investigations.
Pham et al. [31], examined monthly dengue case data, vector larval indices and meteorological data from central Vietnam, between 2004 and 2008. They found significant associations between all entomological indices and dengue cases by univariate analysis but only the HI and “household mosquito index” (not defined in the paper), temperature and rainfall were significant after multivariate analysis.
In Venezuela, Rubio-Palis et al. [39] used a simple regression analysis to investigate correlations between vector indices, climatic variables and dengue incidence for the period 1997–2005. Analyses indicated a significant relationship (R2 = 0.9369) between the numbers of dengue cases, Ae. aegypti abundance (both immatures and adults) and rainfall. Acknowledging the retrospective nature of the study, the authors expressed caution in the predictive value of the findings. Moreover, another limitation was that entomological data were derived only from actual homes and neighbouring houses of confirmed dengue cases but no data were collected from ‘control’ houses.
Within these four studies was some additional evidence that observed changes in vector indices might be useful for the prediction of impending dengue transmission or outbreaks. In Cuba, Sanchez (2006) [29] reported that blocks with BImax (defined as the highest or ‘maximum’ block level BI in a neighborhood) values greater than 4 were significantly more likely to record positive cases in the following month, and had a 3–5 times greater dengue risk in comparison with control blocks. The report concluded that BImax>4 and neighborhood BI>1 during the preceding 2 months provided “good predictive discrimination”. In northern Venezuela Rubio-Palis et al. [39] found the most significant correlation between rainfall levels and the appearance of dengue cases two months later, indicating that the magnitude of outbreaks might be predictable to some extent following periods of rainfall. Pham et al. [31] confirmed an association between dengue transmission and periods of higher rainfall and mosquito abundance in the central highlands of Vietnam, but did not indicate whether this could be used in advance of transmission as a predictive tool.
A further five studies [34], reported ambiguous evidence of associations, both positive and negative, between entomological data and dengue cases. In Belo Horizonte, Correa et al. [43] found a 5–7 fold increase in mean monthly dengue incidence where the ‘infestation rate’ (defined as house index) was “between 1.33% and 2.76% and equal to or higher than 2.77% when compared to areas showing 0.45% or less”, although it was unclear whether or not this was statistically significant. They reported a moderate but significant correlation between adult Aedes spp. infestation rates and numbers of dengue cases (R = 0.67) even though HI and dengue cases were only weakly correlated (R = 0.25 at the municipal level; R = 0.21 and R = 0.14 at the district and village level). Sulaiman et al. [37] reported a significant correlation between BI and HI and dengue cases in certain areas of Kuala Lumpur, but not in others. In Trinidad, Chadee et al. [34] found that 75% of DHF cases were located in areas where BI was greater than 10, although BI and dengue infections were rarely correlated. An additional two studies reported either very low correlations between vector indices and dengue [44], or utilized highly variable inter-annual data precluding such analyses [40].
Four studies, from Malaysia [36], Brazil [35] and Colombia [35], [45] found no statistically significant relationships between entomological indices and dengue cases. Foo et al. [36] observed a positive but non-significant association between dengue cases and HI and BI, which they suggested may have been influenced by the small sample size, the presence of Ae. albopictus and socio-demographic factors. Honorio et al. [38] found no significant associations between recent dengue cases and Ae. aegypti densities and proposed that infections received outside the home were responsible. In Colombia, Romero-Vivas and Falconar [35] reported distinct positive temporal correlations between the larval density index and pupal density index (p<0.005) and a negative association between the larval density index and egg density index (p<0.01); however, they found no correlation between any of the larval, pupal or adult indices with either rainfall or dengue-like cases. The spatial model of Arboleda et al. found no indication that the BI was in any way correlated with the dengue cases or those areas predicted as ‘suitable’ [45].
In the remaining studies [20], [32], [33], [41], [42] a variety of mixed, inconclusive or weak associations were reported. Gurtler et al. conducted analyses on the effect of a given intervention on mosquito indices but not on dengue cases [32]. Although Katyal et al. [33] did not present any statistical analysis, they reported the observation that over a five year period, a fall in cases was visually correlated with a fall in indices. However, they conceded that “an increasing trend of cases was observed [in 2001] in spite of a further declining HI trend”, and concluded that HI had no predictive value at the ‘macro’ level. Despite the absence of statistical analysis, Chaikoolvatana et al. [41] reported a suggestive link between dengue haemorrhagic fever (DHF) during peak annual rainfall months and high abundance of mosquitoes. Chadee et al. observed ambiguous associations, with BI partially correlating with dengue fever cases for two out of three years [42]. As in their earlier study at the same Cuban location [29], Sanchez et al [20] reported that while BImax≥4 was a useful predictor for outbreaks at the block level, sensitivity during outbreaks ranged between 62% and 81.8% and specificity between 71.9% and 78.1%.
The Breteau Index (BI) was used as an outcome measure in seven studies [29]–[31], [34], [36]–[38] and BImax threshold was considered in three (Table 2) [20], [29], [40]. Here, BI values ranged from 1 to 66 during periods when dengue transmission was recorded (Figure 2). In other studies, both recent [46] and historic [47], dengue transmission was recorded when BI values were lower than the widely accepted transmission threshold of 5. Notably, in a study in Trinidad, ‘high’ transmission (25–40 cases for 75% of sample ‘cycles’) took place in areas with relatively ‘low’ abundance (∼BI<5) while, conversely, a consistently higher BI of 5.4 in neighbouring areas did not result in dengue cases [34]. In Rio de Janeiro, the BI did not correlate with dengue incidence and transmission occurred in association with a wide range of BI levels (range 3.30–20.51) [38].
With worldwide dengue transmission levels at an all time high, predicting dengue outbreaks in advance of their occurrence or identifying specific locations where outbreak risks are highest is of critical importance. This review considered the evidence that changes in vector populations can be correlated with dengue virus transmission and whether or not monitoring fluctuations in vector indices might be employed to provide reliable advance warning of impending dengue outbreaks.
Eighteen studies that had the potential to provide evidence of any association between vector indices and dengue incidence were identified and examined. Notably, only 4 studies utilized new data on both vector indices and dengue cases collected de novo as an integral part of the study. More common was a reliance on local government-level records for the dengue case data, a practice that potentially introduces error or bias for number of reasons. First, hospital reports are prone to selection bias, as asymptomatic/inapparent infections may not be recorded and the actual number of cases may have been significantly underreported. Second, there can be a considerable delay between the times of onset of infection and reporting which, if the infection date is not calculated, would result in a temporal mismatch of vector and case data. Third, differences between the geographic location of the vector and dengue case data, or between the spatial units from which each was originally calculated, would result in a geographic mismatch or mask potential relationships, respectively.
The latter point is of particular significance not only from the point of view of these studies, but also when considering the design of future investigations. A growing body of evidence indicates that the distribution of dengue cases typically is highly clustered in both time and space. In various studies, post-dating those reviewed, the size of such clusters ranged from 800 m [48] to less than 100 m [49]. The effective area of such key ‘pockets’ or ‘hotspots’ is likely to be determined by dispersal of the vector [49], [50] which itself can vary over time [51], and is influenced by house density [52] and by human movement within and beyond the infection cluster [53]. Consequently, in studies attempting to correlate vector indices with dengue transmission, and where the geographical unit is too large, high vector densities in key dengue hotspots might be diluted by inclusion of neighboring areas with low densities, thus masking any true relationships [see 38].
Indeed, human movement potentially confounds dengue vector data that derive from residential areas alone as increasingly, evidence indicates that only a proportion of dengue infections are transmitted in the individual's own home, with many infections (possibly the majority) resulting from bites by virus-infected mosquitoes at other houses, schools, workplaces or numerous locations remote from the home [53], [54]. Clearly, this presents a serious challenge when considering the use of vector data for surveillance and highlights a need for inclusion of data from public locations [55] in addition to residential areas, in any surveillance program.
Returning to the studies examined in this review, the fact that there was no clear indication of any consistent association between vector indices and dengue cases is not unexpected, given the diverse and mostly weak study designs. One study found there was no apparent increase in vector indices coinciding with what was the largest increase in dengue fever cases of all areas studied [40], while in another, dengue transmission remained low despite exceptionally high vector indices [44]. In studies where correlations were calculated for HI, BI and dengue cases, regression coefficients ranged from weak/moderate non-significant (R = 0.43 and R = 0.35 respectively; p>0.05) [38], to moderate significant associations (R = 0.61 and R = 0.60 respectively, but only in the urban centre; p<0.05) [37].
Only two studies calculated pupal indices, even though fifteen of the eighteen studies reviewed were published more than three years after WHO acknowledged that the traditional Stegomyia indices were inadequate for the measurement of dengue vector abundance [56]. In the two studies included in this review that calculated pupal indices, only one reported increases in the pupal index, but its relationship with dengue cases was not statistically significant, possibly due to the low numbers of pupae recorded [30], [35]. A major problem with pupal surveys is the difficulty in locating breeding sites and the potential existence of important or key but cryptic breeding sites (e.g. overhead tanks on houses or underground water reserves such as sewers or wells) that may harbor significant proportions of the vector population [57], [58].
Clearly, calculation of adult female Aedes aegypti indices is the most direct measure of exposure to dengue transmission [15]. Of the four studies reviewed that reported some correlation between vector indices and dengue cases, two [31], [39] recorded adult vector data. The adult population of Aedes aegypti is rarely sampled, partly due to the erroneous but commonly held belief that carrying out such sampling is time-consuming, difficult or expensive [59].
Sampling adult female Aedes aegypti is a relatively simple task, though it can be limited by the fact that mosquito numbers often remain low during outbreaks [60]. Nonetheless, it is possible to aim to sample adult mosquitoes as a routine procedure with minimal additional training and resources. A number of novel sampling devices [61]–[63] offer the potential to monitor vectors during outbreaks [64] and at the spatial scale required to accurately sample populations of Ae. aegypti [65]. Simple affordable low-tech tools that enable localized sampling of adult Ae. aegypti and other mosquito vectors are available, with initial studies demonstrating their ease and effectiveness in comparison with older methods [66], [67]. In Brazil, routine sampling of Ae. aegypti adults with gravid traps deployed at relatively low densities was used to identify high risk localities which were then targeted for vector control [68], [69]. This ‘Intelligent Dengue Monitoring’ system was reported to have prevented over 27,000 dengue cases over two ‘dengue seasons’ between 2009 and 2011 with considerable reductions in cost burden to the communities where it was deployed [70].
None of the studies reported on viral infection rates in the vector. This perhaps is not surprising given that techniques suitable for application in routine surveillance, such as PCR or NS1, have not been available until recently, that vector infection rates with dengue virus are of the order of 1% even in areas where transmission is ongoing [64], [70]–[72] and the cost of running the large numbers of tests to detect meaningful infection levels could be considered prohibitive for many authorities. Nonetheless, routine screening for dengue virus of trapped adult female Aedes aegypti is possible and has been incorporated into the routine surveillance program in Belo Horizonte, Brazil [73]. The relative low dispersal rates of Ae. aegypti as compared with the high mobility of humans as they commute daily from the home to the workplace, school, etc., means that virus-infection rates in the vector potentially could provide an accurate or epidemiologically valid indicator of dengue risk in any particular locality, thus informing vector control. Clearly, elucidating the relative value of such an index would require substantial research investment, while integrating it into routine surveillance programmes would demand significant sustained investment, but the importance of metrics like the sporozoite or entomological inoculation rates used in malaria epidemiology [59] already indicate the potential.
This review has also demonstrated the unreliability of accepted vector thresholds for dengue transmission. A number of studies reported dengue transmission at BI levels below the currently accepted threshold of 5 (Figure 2) [29], [34], [36]–[38] or when the HI was below 1% [74], [75]. Elsewhere, Focks proposed a pupal productivity index of 0.25 as a threshold for dengue transmission in Honduras [76], yet in Brazil dengue transmission occurred at PPI levels of 0.15 [58]. While the desire for a single globally applicable transmission threshold is understandable, it seems unlikely that such a threshold exists, given the variety and complexity of other parameters that potentially influence the risk of outbreaks today [19], [77], [78]. Chadee concluded in 2009 that dengue transmission occurs, not at a fixed entomologic figure/quantity but rather at a variable level based on numerous factors including seroprevalence, mosquito density and climate [30]. It is becoming increasingly apparent that thresholds differ at different locations and in different contexts, and while they must be calculated independently at each location [19], [79]. Moreover, empirically defining thresholds, which must be expected to be dynamic, rising and falling as the susceptibility of the local population changes, will require comprehensive prospective, longitudinal vector studies [80], with simultaneous monitoring of the relationship between Ae. aegypti population densities and dengue virus transmission in a spatially relevant human cohort.
In spite of reference searches and use of grey literature, publication bias will likely remain given the very nature of a systematic review. However, we also sought to further limit the effect of publication bias by placing no restriction on language, and those languages encountered were: English, French, Portuguese, Spanish and Chinese.
Additionally, one should be cautious when interpreting these data due to the study design of the 18 articles. As defined by QATAS assessment methods, study design was often weak (15 studies), meaning that studies were prone to bias and confounding factors, which may have skewed some of the reported associations. In addition, most (n = 13) studies relied on dengue case data from external sources, rather than obtaining study-generated data. With the exception of vector sampling and generation of vector index, there were few similarities in the approaches across the different studies.
Despite the widespread practice of collecting vector population data, the review has revealed that very few rigorous studies have been undertaken to determine the relationship between vector abundance and dengue transmission; of those that have been published, few provide tangible evidence of such a relationship, and therefore it is not possible to draw a firm conclusion. After decades of vector surveillance in many countries and considering the magnitude of the dengue threat today both in those and other countries that have recently experienced major dengue outbreaks, this is disappointing. Yet it is also indicative of the lack of basic knowledge of dengue epidemiology, in particular with regard to transmission. Clearly, this is a major knowledge gap that requires attention with a degree of urgency and the following research priorities are recommended:
In the absence of definitive evidence that dengue vector surveillance data can contribute to the prediction of dengue outbreaks, it might be tempting to consider abandoning the practice altogether. However, this would be a rash and premature judgment. At the very least, this systematic review has demonstrated that the potential of vector surveillance data has not yet been evaluated. Indeed, its full potential will not be apparent until its contribution to a complete predictive model incorporating all other covariates influencing dengue epidemiology have been considered. That will not be possible until multiple high quality studies investigating the relationship between vector populations and dengue transmission have been carried out.
|
10.1371/journal.pmed.1002373 | Malaria, malnutrition, and birthweight: A meta-analysis using individual participant data | Four studies previously indicated that the effect of malaria infection during pregnancy on the risk of low birthweight (LBW; <2,500 g) may depend upon maternal nutritional status. We investigated this dependence further using a large, diverse study population.
We evaluated the interaction between maternal malaria infection and maternal anthropometric status on the risk of LBW using pooled data from 14,633 pregnancies from 13 studies (6 cohort studies and 7 randomized controlled trials) conducted in Africa and the Western Pacific from 1996–2015. Studies were identified by the Maternal Malaria and Malnutrition (M3) initiative using a convenience sampling approach and were eligible for pooling given adequate ethical approval and availability of essential variables. Study-specific adjusted effect estimates were calculated using inverse probability of treatment-weighted linear and log-binomial regression models and pooled using a random-effects model. The adjusted risk of delivering a baby with LBW was 8.8% among women with malaria infection at antenatal enrollment compared to 7.7% among uninfected women (adjusted risk ratio [aRR] 1.14 [95% confidence interval (CI): 0.91, 1.42]; N = 13,613), 10.5% among women with malaria infection at delivery compared to 7.9% among uninfected women (aRR 1.32 [95% CI: 1.08, 1.62]; N = 11,826), and 15.3% among women with low mid-upper arm circumference (MUAC <23 cm) at enrollment compared to 9.5% among women with MUAC ≥ 23 cm (aRR 1.60 [95% CI: 1.36, 1.87]; N = 9,008). The risk of delivering a baby with LBW was 17.8% among women with both malaria infection and low MUAC at enrollment compared to 8.4% among uninfected women with MUAC ≥ 23 cm (joint aRR 2.13 [95% CI: 1.21, 3.73]; N = 8,152). There was no evidence of synergism (i.e., excess risk due to interaction) between malaria infection and MUAC on the multiplicative (p = 0.5) or additive scale (p = 0.9). Results were similar using body mass index (BMI) as an anthropometric indicator of nutritional status. Meta-regression results indicated that there may be multiplicative interaction between malaria infection at enrollment and low MUAC within studies conducted in Africa; however, this finding was not consistent on the additive scale, when accounting for multiple comparisons, or when using other definitions of malaria and malnutrition. The major limitations of the study included availability of only 2 cross-sectional measurements of malaria and the limited availability of ultrasound-based pregnancy dating to assess impacts on preterm birth and fetal growth in all studies.
Pregnant women with malnutrition and malaria infection are at increased risk of LBW compared to women with only 1 risk factor or none, but malaria and malnutrition do not act synergistically.
| More than 125 million pregnant women are at risk of malaria in pregnancy annually, producing detrimental effects on maternal, newborn, and infant health.
Maternal undernutrition is estimated to be responsible for 800,000 newborn deaths annually.
Prior evidence from 4 small studies indicated that the harmful impact of malaria on fetal growth and birthweight (BW) may depend upon the macronutrient nutritional status of the mother.
If malaria and maternal undernutrition have synergistic negative impacts on pregnancy outcomes, interventions targeted to high-risk women might provide substantial public benefit.
The present study provides a robust assessment of potential malaria–nutrition interactions in pregnancy and overcomes size and methodological limitations of earlier exploratory studies.
We present a large, pooled analysis of individual participant data from 13 studies conducted in sub-Saharan Africa and the Western Pacific investigating the interaction between maternal malaria infection and malnutrition on the risk of low birthweight (LBW) and reduced mean BW.
The findings suggest that women who are both infected with malaria and malnourished are at greater risk of LBW than their uninfected, well-nourished counterparts.
However, the study found no conclusive evidence of interaction between the 2, i.e., the impact of malaria on BW was independent of the macronutrient nutritional status of the mother.
Subgroup analyses did find that studies conducted just in Africa had slight evidence of interaction, but this was not consistent throughout all analyses.
Although there was no overall evidence of malaria–nutrition interactions, more than 1 in 3 pregnant women suffered from malaria and/or undernutrition, emphasizing the importance of joint approaches to decrease maternal malaria and improve nutrition to minimize adverse pregnancy outcomes.
| Annually, over 20 million infants are born low birthweight (LBW; <2,500 g), predominantly in low- and middle-income countries (LMICs) [1]. LBW can have negative impacts on neonatal mortality and childhood neurological, metabolic, and physical development [2]. The World Health Organization (WHO) has set a Global Nutrition Target of 30% reduction in LBW by 2025 [1].
One preventable cause of LBW in LMICs is maternal malaria infection [2,3]. Its prevalence remains high, despite targeted malaria prevention programs [2]. Annually, 125 million pregnant women are at risk for malaria [4]. The predominant species, Plasmodium falciparum, sequesters in the placenta, causing LBW through fetal growth restriction (FGR) and preterm delivery [2]. Prior estimates from Africa suggest that malaria infection doubles the risk of LBW [2,4]. The prevention of malaria infection during pregnancy remains a public health priority.
Another modifiable risk factor for impaired fetal growth is maternal malnutrition, specifically undernutrition [5]. Up to 20% of African women of reproductive age are undernourished [5–7]. Maternal protein-energy-fat (macronutrient) and micronutrient reserves and dietary consumption influence fetal growth. Micronutrient deficiencies are difficult and costly to assess; therefore, anthropometrics are commonly used as sensitive but nonspecific indicators of protein reserves, fat stores, and malnutrition more broadly [7].
Recent evidence indicates that the relationship between malaria infection and LBW may depend upon the mother’s nutritional status [8]. Studies in Papua New Guinea (PNG) and Benin found inconsistent evidence of modification of the malaria infection–LBW relationship by maternal anthropometric status, but studies from Kenya and the Democratic Republic of the Congo (DRC) reported significant modification [9–12]. Notably, in the DRC, the risk of FGR associated with malaria infection was 2 to 8 times higher among malnourished women [11]. Malaria infection and malnutrition may act along similar physiological pathways by affecting placental development and nutrient transfer [2,4,5].
To date, work on this potential interaction has been limited to 4 studies, with only 1,318 pregnant women from Africa and 1,369 pregnant women from PNG. Not only were these studies somewhat inconsistent in their findings, but their interpretation is hindered by relatively small sample sizes, and their findings may not be generalizable to other malaria-endemic countries. The objective of this study was to investigate the putative interaction between maternal malaria infection and malnutrition in relation to birthweight (BW) using a large, pooled dataset of 14,633 live birth pregnancies from women participating in 13 studies conducted in multiple LMICs. We hypothesized that there would be a synergistic interaction, such that the observed joint effect of being both infected with malaria and malnourished would be greater than expected if considering each exposure independently.
We used data from 14,633 singleton live birth pregnancies from women participating in 13 studies conducted from 1996 to 2015 in 8 African countries and the Western Pacific (PNG) as part of the Maternal Malaria and Malnutrition (M3) initiative [9,11,13–24]. The M3 initiative has been described in detail previously [25]. Briefly, the M3 initiative is a collaboration with the Malaria in Pregnancy Consortium (MiPc) and affiliated malaria and nutrition researchers who agreed to pool resources to improve the understanding of malaria–nutrition interactions. A convenience sampling approach was taken to obtain eligible studies identified by researchers within the MiPc, and inclusion of studies for the individual participant data meta-analysis stopped 1 January 2016. Studies were eligible if they were an observational study or randomized controlled trial conducted between 1996 and 2015 enrolling pregnant women during pregnancy with follow-up through delivery and they met the following criteria: ethical approval allowed for secondary analyses and data sharing, malaria was endemic in the area with medium to high transmission, assessment of malariometric indices (light microscopy [LM] and/or rapid diagnostic tests [RDT]) at enrollment/first antenatal care visit (ANC), assessment of anthropometric indicators at enrollment (mid-upper arm circumference [MUAC] and/or body mass index [BMI]), and assessment of infant weight within 24 hours postpartum or within 7 days of birth if timing of weight measurement data was available. Data was shared by each individual study using a standardized data transfer file. Participating studies had been undertaken for a range of objectives, including investigation of the mechanisms leading to LBW as a result of malaria, evaluation of antimalarial interventions during pregnancy such as intermittent preventive therapy during pregnancy (IPTp) or insecticide-treated bed nets (ITN), or the assessment of the potential of nutritional supplementation during pregnancy to improve birth outcomes (S1 Table). All studies received approval by their local ethics board and obtained informed consent from all participants. The prospective protocol for the IPD analysis is included in the supplemental text (S2 Text).
The main outcome measure was BW, analyzed both continuously and dichotomized at 2,500 grams (LBW) [1]. Ten studies used digital scales to weigh newborns, 2 studies used spring or digital scales, and 1 study used a hanging weighing scale (S2 Table). Weights measured after 24 hours (13% of weights) were adjusted using a cubic regression model to account for weight changes in the first week of life [26]. Among 9 studies with ultrasound-dated gestational age, we considered 2 secondary outcomes: small for gestational age (SGA; a BW less than the 10th percentile of the INTERGROWTH-21st reference) and preterm birth (PTB; gestational age less than 37 weeks) [27].
Diagnostics for malaria were collected at study enrollment and at delivery. For the interaction analyses, we chose to focus on malaria infection at enrollment instead of at delivery for 2 reasons. First, from a public health perspective, if there was interaction at the time of study enrollment, this might help inform future interventions that could be implemented during antenatal care. Second, it has been hypothesized that malaria infection and malnutrition may act along similar physiological pathways to alter fetal growth by decreasing maternal–fetal oxygen transfer and reducing uteroplacental blood flow; 2 mechanisms that would be altered earlier in pregnancy versus at delivery. At study enrollment, we defined malaria based on LM examination of a Giemsa-stained peripheral blood smear or a RDT for malaria antigen [28]. At delivery, we defined malaria based on peripheral or placental LM or placental histology (active or past infection). Given the uncertain impact of submicroscopic infections on LBW and the variation in the availability of polymerase chain reaction (PCR) diagnostics across studies, we excluded PCR results [29]. In sensitivity analyses, we explored alternative definitions of malaria, including any PCR results and “any malaria,” defined as a positive LM, RDT, or PCR at enrollment, delivery, or during pregnancy (in 5 studies with repeat malaria diagnostics throughout pregnancy).
The primary measure of maternal malnutrition was low MUAC at enrollment, dichotomized at 23 cm [7]. MUAC changes little over pregnancy, making it a useful measure of malnutrition [7]. Since some studies did not measure MUAC, we used BMI as a secondary measure of malnutrition. According to WHO, a prepregnancy BMI <18.5 kg/m2 is predictive of adverse birth outcomes [30]. BMI at enrollment was used to estimate prepregnancy BMI by adjusting maternal weight measured in the second/third trimesters using a cubic regression model to account for gestational weight gain [30]. Low adjusted-BMI was defined as values under 18.5 kg/m2. As the correlation between BMI and MUAC is not perfect, indicators were analyzed separately [7]. The reason for dichotomizing MUAC and BMI was 2-fold. First, cutoffs are endorsed by WHO, are clinically easier to use, and are commonly used in the current literature to define undernutrition [7]. Second, while continuous exposures can be assessed in interaction models, interpretation is difficult, as the interaction estimates vary according to the levels of the exposures being compared and can vary in directionality as well [31].
We developed a checklist of study characteristics for each of the included individual studies to assess the risk of bias for the main evaluation of the interaction between malaria infection and maternal malnutrition on BW. Criteria were specific to the research question and were informed by the Newcastle-Ottawa Scale, Downs and Black instrument, and the Meta-Analysis of Observational Studies in Epidemiology checklist [32–34]. For each included study, we evaluated the individual study publications or contacted individual study collaborators to identify the following items to categorize studies as being either at lower or higher risk of bias: participant retention rate (<75% versus ≥75%), measurement of important confounders (maternal age, gravidity, rural versus urban residence, HIV infection, and anemia at enrollment), clearly described measurement of malaria parasitemia, measurement of MUAC and/or BMI, >80% of BWs measured using electronic scale with known precision ≤20 g, and >80% BWs measured within 24 hours. Studies were defined as at lower risk of bias if every item was determined to be at a lower risk of bias.
We analyzed maternal malaria infection and malnutrition as coprimary exposures and assessed malnutrition as a modifier of the malaria–LBW relationship. While effect measure modification (EMM) assesses how the effect of 1 exposure varies across strata of another variable, interaction analyses assess the joint effects of 2 exposures [35]. We performed both interaction and EMM analyses; however, in the context of this work, interaction is preferable to EMM because interventions for both malaria infection and malnutrition might prevent LBW.
There are 2 commonly employed approaches for handling individual pooled data, a 1-stage and a 2-stage approach, although there is no consensus as to which approach is preferable [36–38]. We employed a 2-stage approach, as it is generally considered more easily interpretable and allows the investigator to visually present forest plots and quantify statistical heterogeneity [36]. We examined the consistency of results with a 1-stage approach, fitting a generalized mixed model with random intercepts and slopes. Study-specific risk ratios (RRs) and mean BW differences were calculated using linear and log-binomial regression models controlling for confounding using inverse probability of treatment weights (IPTW) truncated at the 1st and 99th percentiles. A minimally sufficient set of confounders was identified using a directed acyclic graph based upon background knowledge of covariate relationships [39]. We identified confounders for both malaria infection and malnutrition relative to LBW since we were analyzing them as coprimary exposures. Confounders for the relationship between malaria infection at enrollment and LBW included maternal age, gravidity, rural versus urban residence, malnutrition (MUAC when available, otherwise BMI), and HIV infection. Because malaria infection is a cause of anemia, the latter was considered a mediator and not a confounder. We explored modification of the effect of malaria infection at enrollment on LBW by maternal gravidity and doses of intermittent preventive therapy (IPTp) received. When assessing malaria infection at delivery, anemia at enrollment and the number of IPTp doses were considered additional confounders. Confounders for the malnutrition–LBW relationship included maternal age, gravidity, rural versus urban residence, anemia at enrollment, and HIV infection. Partially missing data were imputed using multivariate normal multiple imputation (S1 Text) [40]. We calculated interaction estimates using a product term in the multiplicative and additive model for LBW and the additive model for mean BW [35]. These estimates reflect whether the effect of exposure to both malaria infection and malnutrition exceeds the product (or sum) of the effects of each exposure considered separately, defined as synergy. A product term greater than 1 on the multiplicative scale or greater than 0 on the additive scale is indicative of synergistic interaction between malaria infection and malnutrition.
Study-specific estimates were pooled using DerSimonian and Laird restricted maximum likelihood method random-effects models [41]. When τ2, the estimated variance of the random-effects distribution, was greater than 0, we calculated 95% population effects intervals (PEI), which incorporate the estimated variance between studies [41]. If τ2 equaled 0, the random-effects model was interpreted as a fixed-effects model. We decided a priori to evaluate the modification of the results by time period (before versus after 2008) due to changes in antimalarial recommendations, study type (trial/cohort), location (Africa/Western Pacific), and the study-level prevalence of malaria infection at study enrollment and delivery based on the individual study data, using meta-regression. We further decided post hoc to conduct a sensitivity analysis for the interaction analyses restricted to adolescent women.
Using a convenience sample approach, a total of 18 studies were considered for inclusion by the time of our inclusion cutoff date (1 January 2016), of which 13 were included in the pooled analysis (Fig 1). We excluded 5 studies: 2 studies did not assess malaria at antenatal enrollment [42,43], 1 study had data that were not yet available for inclusion [44], 1 recruited women comparatively late in pregnancy [10], and 1 had not directly measured the number of sulfadoxine-pyrimethamine (SP) doses given for IPTp [45]. Following the cutoff date, 5 further studies were identified, of which 4 could be eligible with a collective sample size of 3,528 pregnant women (S3 Table) [46–50].
Twenty-five percent of the pooled dataset comprised adolescent women aged 19 or younger. The trimester at enrollment, anemia prevalence, gravidity distribution, area of residence, and HIV prevalence varied across studies (Tables 1 and 2). The prevalence of malaria infection at enrollment, malaria infection at delivery, low MUAC, and joint malaria infection at enrollment and low MUAC also varied by study (Fig 2 and S5 Fig). Among 8,152 women with both measurements, only 2% had both low MUAC and malaria infection at enrollment. The prevalence of malaria infection among women with low MUAC was 16%, compared to 12% among well-nourished women (p = 0.0005). The prevalence of low BMI varied across studies and was different from, although correlated with, the prevalence of low MUAC (χ2 p < 0.0001; S1 Fig). The joint prevalence of malaria infection at enrollment and low BMI was also 2%. Of all 14,633 women, 35% were infected with malaria at either enrollment or delivery or had low MUAC or BMI. The prevalence of LBW was 9% (range 5% to 15% among studies). Among 9 studies with ultrasound-dated gestational age, the prevalence of SGA was 19% (range 13% to 25%), and the prevalence of PTB was 11% (range 3% to 20%).
Five of the thirteen included studies were judged to be at a lower risk of bias for the assessment of interaction between malaria infection and maternal malnutrition on BW (S4 Table). Among the 8 other studies, 3 had a <75% retention rate for the primary outcome, 5 did not measure at least 80% of BWs with an electronic scale with known precision ≤20 g, and 3 did not measure at least 80% of BWs within 24 hours.
The pooled IPTW-adjusted risk ratio (aRR) for the effect of malaria infection at enrollment on LBW was 1.14 (95% CI: 0.91, 1.42; 95% τ2 = 0.05 [95% CI: 0.00, 0.25]; PEI: 0.72, 1.80), and the mean BW difference was −55 g (95% CI: −79, −30; τ2 = 0 [95% CI: 0.00, 1,610]) (Fig 3a). The effect of malaria infection at delivery was more pronounced: aRR, 1.32 (95% CI: 1.08, 1.62; τ2 = 0.04 [95% CI: 0.00, 0.39]; 95% PEI: 0.91, 1.91) (Fig 3b). When considering SGA and PTB as secondary outcomes, results were similar for malaria infection at enrollment and attenuated for malaria infection at delivery (S5 Table). The effect of malaria infection at enrollment was attenuated among those with more than 1 IPTp dose versus 1 or 0 doses (aRR 0.98 versus 1.22) and was slightly stronger among primi/secundigravid versus multigravida women (aRR 1.19 versus 1.14). A slightly stronger effect of malaria infection was seen among women enrolled in studies conducted prior to 2008, in Africa, or with malaria infection prevalence at or above the median (S2 Fig).
The aRR for the effect of low MUAC on LBW was 1.60 (95% CI: 1.36, 1.87; τ2 = 0 [95% CI: 0.00, 0.05]); the mean BW difference was −142 g (95% CI: −171, −113; τ2 = 0 [95% CI: 0, 100] (Fig 4a). Results were similar for low BMI: aRR, 1.49 (95% CI: 1.26, 1.76; τ2 = 0 [95% CI: 0.00, 0.16]); mean BW difference −133 g (95% CI: −158, −108; τ2 = 0 [95% CI: 0.00, 0.00]) (Fig 4b). There was no modification by study characteristics on the malnutrition–LBW relationship (S3 Fig). Similar but weaker trends were observed when SGA was used as the outcome among the studies with ultrasound data, but low MUAC or low BMI were significantly associated with an increased risk of PTB (S5 Table).
The joint aRR for both malaria infection at enrollment and low MUAC was 2.13 (95% CI: 1.21, 3.73; τ2 = 0.25 [95% CI: 0.00, 1.82]; 95% PEI: 0.80, 5.67), and the mean BW difference was −163 g (95% CI: −253, −75; τ2 = 6,995 [95% CI: 0, 58,414]; 95% PEI: −328, 0). The multiplicative interaction term for LBW was 1.30 (95% CI: 0.62, 2.72; τ2 = 0.37 [95% CI: 0.00, 3.97]; 95% PEI: 0.39, 4.31), the additive interaction term for LBW was −0.01 (95% CI: −0.09, 0.08; τ2 = 0.003 [95% CI: 0.00, 0.04]; 95% PEI: −0.11, 0.09), and the additive interaction term for mean BW difference was 38 g (95% CI: −90, 166; τ2 = 17,198 [95% CI: 0, 120,165]; 95% PEI: −219, 295). Sensitivity analyses that varied the definitions of malaria, malnutrition, outcome, and analytic approach largely did not qualitatively alter the results; however, restriction to adolescent women did suggest potential multiplicative and additive interaction between low MUAC and malaria infection at enrollment among this subgroup (product term 2.49 [95% CI: 0.88, 7.02]; additive interaction term 0.08 [95% CI: −0.07, 0.22]) (S6 Table). Additionally, meta-regression indicated apparent multiplicative interaction and slight additive interaction between MUAC and malaria infection at enrollment among studies conducted in Africa (multiplicative interaction term, 2.47 [95% CI: 1.12, 5.42]; additive interaction contrast, 0.06 [95% CI: −0.05, 0.17] S4 Fig), but this interaction was not seen when assessing malaria infection at delivery or BMI or when accounting for multiple comparisons with a Bonferroni correction (99% CI: 0.88, 6.95). In EMM analyses, the aRR for the effect of malaria infection at enrollment on LBW among low MUAC women was 1.32 (95% CI: 0.66, 2.63; τ2 = 0.43 [95% CI: 0.00, 3.40]; 95% PEI: 0.36, 4.79), compared to 0.98 (95% CI: 0.74, 1.29; τ2 = 0 [95% CI: 0.00, 0.32]) among well-nourished women.
The joint aRR for both malaria infection at delivery and low MUAC was 2.16 (95% CI: 1.25, 3.74; τ2 = 0.23 [95% CI: 0.00, 1.61]; 95% PEI: 0.84, 5.55), and the mean BW difference was −196 g (95% CI: −301, −92; τ2 = 10,904 [95% CI: 0, 86,721]; 95% PEI: −401, 8). The multiplicative interaction term for LBW was 0.82 (95% CI: 0.50, 1.33; τ2 = 0 [95% CI: 0.00, 3.79]), the additive interaction term for LBW was −0.01 (95% CI: −0.10, 0.07; τ2 = 0 [95% CI: 0.00, 0.06]), and the additive interaction term for mean BW difference was −49 g (95% CI: −190, 93; τ2 = 20,087 [95% CI: 0, 154,675]; 95% PEI: −326, 229).
Using the large M3 initiative dataset, we found that pregnant women who were both infected with malaria and malnourished were at greater risk of LBW and reduced mean BW compared to their uninfected, well-nourished counterparts, but there was overall no convincing evidence of synergism, i.e., excess risk due to interaction. This finding was consistent for both time points of malaria diagnosis (at enrollment and delivery) and both definitions of malnutrition (MUAC and BMI). This suggests that malaria infection and malnutrition largely act independently to influence fetal growth and gestational length.
A 2004 review estimated that women infected with placental malaria were twice as likely to have a LBW infant [51]. Our findings are broadly consistent with this review, although with weaker effects on LBW (overall aRR for malaria infection at delivery: 1.32 [95% CI: 1.08, 1.62], aRR restricted to African studies: 1.55 [95% CI: 1.29, 1.85]), possibly reflecting increased access to preventive strategies and fewer chronic infections [3,4]. In support of this hypothesis, the effect of malaria infection on LBW appears lower in women who received more doses of IPTp. The effects of malaria infection at enrollment on LBW were weaker than at delivery, contradicting the theory that malaria infection earlier in pregnancy is more disruptive to placental function [2]. This weaker effect at enrollment could either suggest that antimalarial treatment, provided in most studies, cleared infection and allowed catch-up growth or that infection at delivery represents more severe infections that were not cleared despite medications. Both malaria infection at enrollment and delivery were associated with a reduction in BW of around 55 grams, which has been found in other studies [52].
Our data are consistent with a 2011 meta-analysis, which estimated that underweight women had increased risk of LBW (aRR: 1.64 [95% CI: 1.38, 1.94]), although studies included in that meta-analysis used different definitions for underweight [53]. In our study, using consistent cutoffs of malnutrition across studies, both low MUAC (aRR 1.60 [95% CI: 1.36, 1.87]) and low BMI (aRR 1.49 [95% CI: 1.26, 1.76]) increased the risk of LBW. This information is consistent with other evidence that adequate maternal nutrition is integral for fetal growth [5].
Prior literature on the interaction between malaria infection and malnutrition is sparse. Two studies in the DRC and Kenya showed that the association between malaria infection and reduced fetal growth was greatest among malnourished women [10,11]. In a third study in Benin, the effect of malaria infection on fetal growth velocity was greatest among women with low anthropometric status, but there was no modification by maternal nutrition on the effect of malaria infection on BW z-scores. A fourth study in PNG found that the effect of histology-defined placental malaria infection on LBW was higher among women with a low BMI, but that study found that malnutrition did not modify the association between peripheral blood malaria infection parasitemia and SGA [9]. The Benin, Congo, and PNG studies were included in the present analysis, but our analytic approach differed from the original publications in the assessment of both interaction and modification. Unlike these prior studies, our pooled results suggest that there is a negligible impact of maternal anthropometry on the relationship between malaria infection and LBW and further indicate that there is no evidence of excess risk of LBW due to interaction (i.e., synergism). There was some indication of multiplicative and additive interaction between low MUAC and malaria infection at enrollment among adolescent women; however, these estimates were very imprecise and were only pooled across 4 studies that enrolled enough adolescent women to assess this subgroup. Adolescent women are recognized to be at high risk of adverse pregnancy outcomes [54], and tailored antenatal care programs addressing malaria, nutrition, and other health issues should be considered for this group. In an a priori sensitivity analysis restricted to African studies, there was apparent interaction between malaria infection at enrollment and MUAC, which is consistent with the prior publications. Regional differences could be due to genetics, low MUAC, or anemia prevalence; however, these subregion effects were not statistically significant when properly accounting for multiple comparisons and were absent when using other definitions of malaria (i.e., at delivery) or malnutrition (i.e., BMI). Additionally, the additive interaction, which has been argued to be the more relevant measure for public health impact [55], was only slightly elevated among the African studies. Notably, only 183 women (2%) were jointly infected and malnourished (low MUAC). Thus, even if there is a multiplicative interaction between malaria infection and MUAC among African women or among adolescent women, the proportion of women implicated is small, and does not indicate a large public health burden. However, even in the absence of strong interaction between malaria infection and malnutrition on LBW, we emphasize that interventions on both malaria infection and malnutrition are warranted given their independent effects.
This work had several strengths and limitations. We substantially increased the number of women in whom the hypothesized interaction between malaria infection, malnutrition, and LBW was investigated; notably, the number of pregnant women from Africa was almost 10 times more than all prior studies. Analyzed studies were performed in a variety of settings, increasing the generalizability of these results. Furthermore, availability of individual-level data enabled us to harmonize definitions and minimize heterogeneity. Our work is strengthened by providing results for SGA and PTB as secondary outcomes, which showed findings consistent with LBW. However, we were only able to assess SGA and PTB among a subset of nine of the 13 studies with available ultrasound-dated gestational age. There is no alternative satisfactory dating tool to ultrasound in later pregnancy, thus we used all ultrasound data provided regardless of gestation. Some women were enrolled after 24 weeks gestation (S2 Table), reducing the accuracy of ultrasound among these pregnancies and potentially underestimating gestational age in some SGA babies. Missing data were imputed using multivariate normal multiple imputation, and while not all variables followed a normal distribution (e.g., the binary variable LBW), simulation studies have shown that multivariate normal multiple imputation provides less biased estimates than complete-case analysis even when imputing binary or ordinal variables [56]. We were obliged to pool malaria diagnostics of varying sensitivity and specificity, and we were limited to 2 cross-sectional assessments of malaria infection. Nevertheless, sensitivity analyses that evaluated alternative definitions of malaria, or incorporated repeat diagnostics during pregnancy, were consistent with the main results. Additionally, there may be selection bias due to excluding pregnancy losses. There were only 116 (3%) pregnancy losses in 4 studies (N = 4,571) in the M3 initiative that collected these data, but this is almost certainly an underestimate, since many studies enrolled women after the first trimester. We were obliged to extrapolate prepregnancy BMI using gestational age and BMI at enrollment. Additionally, the M3 initiative represents a convenience sample of available and eligible studies identified through the MiPc and not an exhaustive aggregation of all existing studies available to assess interactions between malaria and malnutrition on LBW. This could potentially lead to selection bias if selection of studies were associated with the effect estimates in that study; however, we did not observe any qualitative differences between studies providing individual participant data and those studies not included in the meta-analysis (S3 Table). Furthermore, women enrolled in studies were likely healthier and received better antenatal care than the general population; the effects of malaria and malnutrition in reality might well be greater than were observed within these research settings. The risk of bias assessment identified 10 studies as being at a higher risk of bias, primarily due to BW not being measured with an electronic scale within 24 hours of delivery. Finally, we cannot discount possible unmeasured confounding, particularly by helminth infections, sexually transmitted infections, environmental pollutants, or micronutrient deficiencies; however, it is important to note that because neither malnutrition nor malaria could be randomized, large-scale, multisite cohort analyses such as this one are necessarily the gold standard for addressing these scientific questions. Future studies may wish to assess joint effects of malaria with other nutritional indicators (e.g., height, obesity, anemia, other micronutrients). Additionally, future studies may wish to further investigate possible interactions between malaria infection and malnutrition on risk of LBW in adolescent mothers.
In summary, our findings suggest that women who are both infected with malaria and malnourished are at greater risk of LBW than their uninfected, well-nourished counterparts but that there is no conclusive evidence of synergistic interaction between the 2. Rather, we propose that malaria infection and malnutrition act independently to disrupt fetal growth and that malnutrition in particular has a strong effect on LBW. Of all 14,633 pregnancies, 35% were affected by malaria infection and/or malnutrition, illustrating the high burden of at-risk pregnancies in LMICs. Malaria infection and malnutrition represent 2 established and modifiable causes of LBW that should both be addressed to optimize pregnancy outcomes in LMIC.
|
10.1371/journal.pcbi.1000403 | Functional States of the Genome-Scale Escherichia Coli Transcriptional Regulatory System | A transcriptional regulatory network (TRN) constitutes the collection of regulatory rules that link environmental cues to the transcription state of a cell's genome. We recently proposed a matrix formalism that quantitatively represents a system of such rules (a transcriptional regulatory system [TRS]) and allows systemic characterization of TRS properties. The matrix formalism not only allows the computation of the transcription state of the genome but also the fundamental characterization of the input-output mapping that it represents. Furthermore, a key advantage of this “pseudo-stoichiometric” matrix formalism is its ability to easily integrate with existing stoichiometric matrix representations of signaling and metabolic networks. Here we demonstrate for the first time how this matrix formalism is extendable to large-scale systems by applying it to the genome-scale Escherichia coli TRS. We analyze the fundamental subspaces of the regulatory network matrix (R) to describe intrinsic properties of the TRS. We further use Monte Carlo sampling to evaluate the E. coli transcription state across a subset of all possible environments, comparing our results to published gene expression data as validation. Finally, we present novel in silico findings for the E. coli TRS, including (1) a gene expression correlation matrix delineating functional motifs; (2) sets of gene ontologies for which regulatory rules governing gene transcription are poorly understood and which may direct further experimental characterization; and (3) the appearance of a distributed TRN structure, which is in stark contrast to the more hierarchical organization of metabolic networks.
| Cells are comprised of genomic information that encodes for proteins, the basic building blocks underlying all biological processes. A transcriptional regulatory system (TRS) connects a cell's environmental cues to its genome and in turn determines which genes are turned “on” in response to these cues. Consequently, TRSs control which proteins of an intracellular biochemical reaction network are present. These systems have been mathematically described, often through Boolean expressions that represent the activation or inhibition of gene transcription in response to various inputs. We recently developed a matrix formalism that extends these approaches and facilitates a quantitative representation of the Boolean logic underlying a TRS. We demonstrated on small-scale TRSs that this matrix representation is advantageous in that it facilitates the calculation of unique properties of a given TRS. Here we apply this matrix formalism to the genome-scale Escherichia coli TRS, demonstrating for the first time the predictive power of the approach at a large scale. We use the matrix-based model of E. coli transcriptional regulation to generate novel findings about the system, including new functional motifs; sets of genes whose regulation is poorly understood; and features of the TRS structure.
| Complex regulatory networks control the transcription state of a genome and consequently the functional activity of a cell [1]. Even relatively simple unicellular organisms have evolved complicated networks of regulatory interactions, termed transcriptional regulatory networks (TRNs), to respond to environmental stimuli [1],[2]. External signals known to impact transcription in microorganisms include carbon source, amino acid, and electron acceptor availability, pH level, and heat and cold stress [2]–[6]. Mapping the links between these environmental growth conditions through signaling networks and ultimately to the resulting transcriptional response is of primary interest in the study of cellular systems [1]. Consequently, reconstructions of the TRNs of model organisms are underway [3].
To effectively describe the interconnected functions of the regulated genes and associated regulatory proteins within a given TRN, we recently developed a formalism involving a regulatory network matrix called R [7]. The R matrix represents the components (extracellular cues, metabolites, genes, and proteins, including regulatory activators and repressors) and reactions (regulatory rules) within a transcriptional regulatory system (TRS). We illustrated how, by using the fundamental properties of linear algebra, this matrix formalism allows characterization of TRS properties and facilitates in silico prediction of the transcription state of the genome under any specified set of environmental conditions.
Importantly, as previously reported (see [7]), the R matrix is distinct from existing approaches that use matrix formalisms and matrix algebra to analyze gene expression data (e.g., see [8]–[12]), as it describes relationships governing gene transcription derived from experiments characterizing how specific inputs regulate the expression of individual genes (e.g., ChIP-chip assays). In this way, the R matrix extends previous approaches for characterizing features of TRNs, including Boolean networks [2], [13]–[16], Bayesian networks [17], and stochastic equations [18] (see [1] for a review of the field). By representing the regulatory rules in matrix form, we can characterize the fundamental subspaces of the matrix (as described below), which in turn uniquely represent properties of the TRS that the R matrix contains. Furthermore, by using a “pseudo-stoichiometric” approach as discussed below, the R matrix representation of a TRN is consistent with, and thus easily integratable with, related approaches using stoichiometric matrices to computationally represent the reactions underlying metabolic and signaling networks [19]–[22].
To date, this approach for representing and analyzing TRSs has only been applied to relatively small systems, including the well-studied four-gene lac operon in Escherichia coli as well as a small 25-gene prototypic TRS [7]. Although these model systems have been useful for prototyping studies of the capabilities and behavior of the R matrix, a key unanswered question is how this approach scales to larger, more complex biological systems. Here we present first steps toward this end by assembling the R matrix for the genome-scale E. coli TRN, for which regulatory relationships have been previously characterized [23] and extensive experimental data (e.g., gene expression datasets) are available [3],[24]. To our knowledge, the work that we present here represents the first R matrix-based model of a genome-scale TRS, and this work has enabled us to gain important insights into the behavior of the R matrix at a larger scale, challenges associated with the scale-up, as well as the underlying biology of E. coli transcriptional regulation.
Specifically, we derived R directly from a previously developed genome-scale model of E. coli in which transcriptional regulatory rules were overlaid on a constraint-based model of metabolism [23]. This integrated transcriptional regulatory-metabolic model is well-suited for these initial genome-scale R matrix efforts as Boolean regulatory relationships are already defined and the behavior of this model has been well-studied using constraint-based analyses [23],[25]. To validate our R matrix analysis, we compared the expression states that we predicted for various environmental growth conditions with available gene expression data (as well as with predictions from the original Boolean model). We also explored the fundamental subspaces of a related matrix R* representing the complete E. coli TRS (to be defined below) to describe key systemic properties, including new hypotheses about network structure. Ultimately, this work yields an understanding of how the E. coli transcriptional regulatory program functions as a whole and demonstrates the utility of the regulatory network matrix formalism in studying transcriptional regulatory systems at the genome scale moving forward.
We formulated a regulatory network matrix R for the genome-scale TRN of E. coli. Here, we summarize how we constructed the R matrix representing the E. coli TRN, sampled the space of possible environments for the TRN, and evaluated the fundamental subspaces of the matrix R* (for the complete E. coli TRS) to describe systemic properties.
Significant efforts have focused on identifying the components and interactions that comprise the E. coli TRN [3]. These efforts have ranged from large-scale experimentation using post-genomic techniques [23],[26] to compiling previously reported regulatory relationships into literature-based representations of the E. coli TRN [5],[23]. Furthermore, several online resources have been developed to integrate both high-throughput as well as low-throughput (i.e., individual regulatory interactions elucidated through targeted experiments) experimental data into comprehensive databases [3],[24]. For example, EcoCyc [24] and RegulonDB [3] are two online resources that provide extensive information regarding transcription factor-target gene (DNA binding site) relationships. RegulonDB also catalogs known promoter sequences, experimentally-defined and computationally-predicted operons, as well as environmental stimulus-transcription factor relationships.
These data formed the basis for a previous integrated regulatory-metabolic network reconstruction called iMC1010v1 [23]. In this model, Boolean rules dictating regulatory interactions were overlaid on a constraint-based model of E. coli metabolism. Here, these Boolean rules were used in the generation of a regulatory network matrix R for the genome-scale E. coli TRN. Three additional regulators (UlaR, MngR, and GntT) and their respective regulatory targets were added to the list of components and interactions, based on recent literature reports. In addition, several regulatory rules were either updated or refined to reflect current data, as measured using ChIP-chip assays and microarray experiments. The Boolean rules governing transcription of 46 new genes were added to the model, and the transcription rules for 11 other genes were modified. The underlying metabolic model was also updated from iJR904 [27] to the recently expanded E. coli model known as iAF1260 [28], including isozyme and multidomain subunit enzymes defined by similar Boolean relationships. The final E. coli TRN reconstruction was comprised of 147 environmental stimuli affecting 125 transcription factors that in turn influence 503 downstream target genes (see Figure 1 and Dataset S1). Ultimately, these target genes give rise to metabolic enzymes and transporters.
Importantly, constructing Boolean rules from experimental findings is not a trivial task. Published experimental data (ranging from high-throughput chip-ChIP assays or expression arrays spanning genome-scale information to “low-throughput” experiments focused on particular genes) are scoured for evidence indicative of a regulatory rule governing gene transcription, i.e., information describing how a transcription factor induces or represses transcription of target genes. As an example, the phrase “Crp induces the expression of sdhC within E. coli” is translated into a Boolean rule indicating that Crp is required for the transcription of the gene sdhC (succinate dehydrogenase subunit C) (i.e., “sdhC: IF (Crp)”). Conversely, a phrase that states “sdhC transcription is inhibited by either ArcA or Fnr” is translated into a Boolean statement “NOT(ArcA OR Fnr).” There are times when conflicts in the literature need to be resolved as well. In these instances, it is important to gauge which dataset appears to make a stronger case about a particular gene and its transcriptional requirements, in terms of the specific experimental conditions that were used and the corresponding likelihood for error. Alternatively, it may be possible to include both rules in the model separately and assess which one results in better model validation. The rules listed in Dataset S1 are accompanied by references.
In order to define the regulatory network matrix R for E. coli, the Boolean rules from the updated integrated transcriptional regulatory-metabolic model for E. coli were translated into pseudo-stoichiometric relationships or “regulatory reactions.” As described in [7], the term “pseudo-stoichoimetric” is intended to indicate that this formalism delineates the relationships between components of the network (i.e., the chemical transformations) while not enforcing that the resultant reactions are mass-balanced in a strictly stoichiometric manner. Thus, effectively, the regulators (i.e., environmental cues and/or transcription factors that serve as inputs to a given gene regulatory rule/reaction) are “consumed” and the gene products or proteins (i.e., outputs of a gene regulatory rule/reaction) are “produced.” Importantly, however, this formalism can account for mass-balanced relationships as they become delineated in TRSs. To automate this conversion from Boolean logic to pseudo-stoichiometric reactions for large-scale systems, an expression parser was developed in Perl. Briefly, for each gene, the parser converted Boolean statements into regulatory reactions (i.e., pseudo-stoichiometric relationships) that could be represented in a R matrix. We used the formalism developed in [7] when implementing the expression parser to perform this conversion. For example, experimental data suggesting that transcription of (Gene 1) is induced if Metabolites A and B are both present within the system was represented in Boolean form, as in(1)
This Boolean rule was then converted by the parser into a reaction form, as in(2)
When a Boolean rule was comprised of several clauses separated by “OR” statements, as in(3)the expression parser generated multiple regulatory reactions for the gene, as in(4)and(5)as satisfying each clause (the presence of Metabolite A or the presence of Metabolite B) can result in protein synthesis independently. (The parser is included as Protocol S1.) Effectively, this parsing recast a gene's regulatory rule in disjunctive normal form (DNF) [29], with each clause of the DNF an independent regulatory reaction describing gene transcription. Importantly, the regulatory reactions distinguished the presence and absence of metabolites and transcription factors, as each of these regulates gene transcription differently. For example, consider a representative regulatory rule for the E. coli gene sdhC, shown at the top of Figure 2. Based on experimental data, sdhC is known to be transcribed if (1) both ArcA and Fnr are absent; (2) Crp is present; or (3) Fis is present. In other words, transcription of sdhC is induced by either Crp or Fis, and it is repressed by ArcA and Fnr in tandem. Consequently, the absence of ArcA (ArcAA in Figure 2, where the subscripts “A” and “P” indicate absence and presence, respectively) as well as the absence of Fnr (FnrA in Figure 2) needs to be incorporated into R. In addition, fully describing the system with a R matrix required the inclusion of reactions governing both activation and repression of gene transcription for each gene as well as exchange reactions balancing the production of proteins (see an example of this for the gene sdhC in Figure 2); these effectively balanced the network so that functional states could be calculated as described below (i.e., an input was “consumed” and a product was “produced” without external manipulation). Reactions governing inactivation of gene transcription were included only for those genes whose protein products repress transcription of downstream genes. The compiled set of regulatory reactions effectively defined the E. coli R matrix, as illustrated in Figure 2. See Dataset S1 for a complete reaction listing. Ultimately, the complete R matrix was comprised of 1009 components (rows) spanning 1685 reactions (columns), including 579 exchange reactions. This study thus constituted the construction of the first genome-scale R matrix for an organism. The R matrix is unique among matrix-based approaches in the field of transcriptional regulation in that it catalogs experimentally-characterized relationships governing gene transcription, thereby facilitating in silico expression state analysis. Other matrix analyses have interrogated experimental gene expression data (see [8]–[12] for examples of these studies) without necessarily having an underlying functional and/or predictive model.
To evaluate the behavior of the genome-scale E. coli TRS in the context of particular environments (i.e., sets of environmental cues defined as present or absent), we further defined environment matrices. Each environment matrix, E, was comprised of the same number of rows as R. The columns of E delineated the availability (i.e., presence or absence) of environmental cues, transcription factors, and proteins with respect to a particular environment. Consequently, in the case of the E. coli TRN, there were 776 different columns in E, one for each unique metabolite, transcription factor, or target gene. For a given environment, E is appended to R to form R*, which captures the complete TRS (see Figure 2 for an example of how a particular gene rule was combined with a representative environment to yield R*, as well as [7] for further details about this process). In this way, multiple environments were simulated by randomly selecting for the availability of environmental cues and other inputs (see below). These environments were used to assess the behavior of the system across a random sampling of all possible environments. See Dataset S2 for a listing of 1000 randomly-sampled environments (as introduced below). In addition, separately, we evaluated two specific environments (anaerobic and aerobic minimal media) for which gene expression data have previously been experimentally characterized, as described below (see Dataset S3 for these environments).
We analyzed the fundamental subspaces of the regulatory network matrix to describe properties of the E. coli TRS. Specifically, a given TRN represented by R responds to environmental signals whose states (i.e., presence or absence) need to be specified [7]. Consequently, the R matrix is further combined with an environment matrix E that characterizes the environment against which a set of regulatory rules is to be evaluated [7]. As the combination of R and E (i.e., the matrix R*) captures the TRS being analyzed, we interrogated the fundamental subspaces of this matrix to describe properties of the E. coli TRS.
Briefly, the four fundamental subspaces of a matrix, namely the column space, left null space, row space, and null space, describe key properties of the matrix and, in turn, the system that the matrix represents [32]. In the case of R*, these fundamental subspaces were previously shown to represent key system properties for a prototypic TRS as well as the E. coli lac operon TRS [7]. As shown in Figure 3B and described in more detail in Text S2, singular value decomposition (SVD) is used to decompose a matrix into three matrices, often named U, Σ, and V (see Figure 3B) [32], and these matrices delineate the four fundamental subspaces of the original matrix (see [7] and Figure 4B). We performed SVD to characterize the fundamental subspaces of multiple R* matrices (describing different randomly-generated environments) for the E. coli TRS, and we summarize the results below. As we describe in our “Results” below, our understanding of the four fundamental subspaces of R*, which we originally proposed in [7] on the basis of our work with two small-scale systems, has been considerably enhanced by the extension of R and R* to the genome-scale E. coli TRS.
Besides the fundamental subspaces, we also computed the angles between columns and rows of R* as these are also informative about the TRS that R* represents. For every pair of column (or row) vectors contained in the matrix, we computed the angle between the vectors by taking the inverse cosine of the dot product between the vectors. The angles between columns of R* indicate the similarity or dissimilarity in the rules governing regulation of the genes. For example, a small angle between a pair of columns suggests that the regulatory rules of the two corresponding genes are relatively similar and affect the state of the TRS in a similar fashion. Likewise, angles between rows of R* indicate the overall similarity or dissimilarity of network component participation in the generation of expression states. For instance, a large angle between a pair of rows (e.g., extracellular cues) suggests that the two network components are relatively dissimilar and affect the transcription of different sets of genes or affect the transcription of the same genes in different ways (e.g., one might be a transcriptional activator while the other is a repressor).
As described above, a parser that converts Boolean logic into regulatory reactions was implemented in Perl. A freely available extreme pathway analysis program (ExPa, University of California, San Diego) [33] was used to convert the regulatory reactions into a regulatory network matrix. Ultimately, this matrix was imported into MATLAB v. 7.6 (part of the R2008a release package, MathWorks, Natick, MA), and code was written to explore the structure of the matrix and to simulate the behavior of the TRN under various environments. The MATLAB representation of the E. coli R matrix and a sample R* matrix is provided in Protocol S2. Maps of the E. coli TRS were constructed using Cytoscape v. 2.6 [34].
Here we present initial steps toward applying the regulatory network matrix formalism to the genome-scale E. coli TRN. In order to facilitate this process, a previously developed model of the E. coli TRN [23] was updated to reflect recently published regulatory interactions as well as an expansion of the underlying metabolic model [28] (see Dataset S1). The resulting updated Boolean rules describing the regulation of the underlying components were then used to generate pseudo-stoichiometric relationships or “regulatory reactions.” The compilation of these reactions represents the scope of the R matrix for E. coli and illustrates the complexity involved when applying this approach to a genome-scale system.
The R matrix of the E. coli TRS is comprised of 1009 components (rows) spanning 1685 reactions (columns), including 579 exchange reactions (see Figure 4A). As illustrated in Figure 1C, the E. coli TRS exhibits a hierarchical structure, as highly connected global regulators act broadly to influence the expression of major and minor regulators and thus directly and indirectly affect the transcription of numerous target genes. Examples of global regulators include traditional regulators such as transcriptional dual regulator Crp, which senses cyclic AMP (cAMP) levels and thus monitors the nutritional status of the cell, and nucleoid binding proteins such as histone-like nucleoid structuring protein (H-NS) and factor for inversion stimulation (Fis), which bind the chromosome and thus influence its topology within the cell in addition to directly impacting gene expression. Alternative sigma factors such as RNA polymerase sigma factor (RpoS) are also found among this class of proteins as they influence the expression of diverse and numerous targets in response to various cellular stresses.
To further evaluate properties of the E. coli TRS, we considered fundamental subspaces of multiple R*, with each R* corresponding to a unique, randomly-generated environment. A representative subset of these randomly-generated environments is presented in Dataset S2. We performed singular value decomposition (SVD) on each R*, as described in [7] and shown in Figure 3B, yielding R* = U•Σ•VT. The diagonal entries of the matrix Σ = diag(σ1, σ2, … , σr), where r is the rank of R* and σ1≥σ2≥…≥σr, indicate the relative contribution of the corresponding left singular vector (a column of U) and right singular vector (a row of VT) in the overall construction of the TRS [36]. Note that an important feature of SVD is that the singular vectors are orthonormal to each other and consequently each principal mode is decoupled from all the others.
Interestingly, across many different randomly-generated environments, the singular value spectra of the matrix R* representing the genome-scale E. coli TRS (i.e., the singular values σ1, σ2, … , σr) were relatively consistent, suggesting that the environment does not contribute significantly to the properties of the TRS. The number of inputs to the system ( = 147 environmental cues) constitutes less than six percent of the columns within R*. Furthermore, although the singular value of the first mode is larger than that of the next closest mode (10.60>7.895), the singular value spectrum of a given R* is rather uniformly distributed, as shown in a representative spectrum in Figure 7A. In other words, the information content of a given R* is evenly distributed throughout the matrix, or, alternatively, there are few components or reactions that dominate the genome-scale E. coli TRS that R* describes. This result is particularly insightful as it contrasts with the structural hierarchy (with few regulators affecting many genes, and many affecting few genes) that is evident by purely inspecting the network map (Figure 1C). Furthermore, whereas in metabolic networks approximately 27 percent of the information content of a stoichiometric matrix is often captured in the first four principal modes (less than one percent of all the principal modes) [36], here, to capture an equivalent information content of a regulatory matrix, the first 150 principal modes (or about 15 percent of all the principal modes) must be recapitulated. Thus, although a simple inspection of the network would suggest that only a small handful of regulators control a large fraction of the network, control of the TRN is significantly more distributed. We discuss this result in detail below (see “Discussion”).
The results presented here represent the first steps toward applying the regulatory network matrix formalism at the genome scale. Specifically, we constructed a regulatory network matrix R for the genome-scale E. coli transcriptional regulatory network, including direct interactions between environmental stimuli, transcription factors, and other downstream target genes. Ultimately, we (1) identified features of the E. coli TRN, including the numbers of components and regulatory relationships; (2) validated our model in the context of available experimental data and illustrated how the R matrix at genome scale affords predictions of expression states for all possible systemic environments; and (3) characterized the fundamental subspaces of the regulatory system matrix R* for the E. coli TRS, noting unique properties about these subspaces of R* not previously observed, including the distributed (and non-hierarchical) nature of the functional states of the genome-scale transcriptional regulatory network which is in contrast to that observed for genome-scale metabolic networks.
As illustrated in Figures 1 and 4, for a system of 776 total environmental stimuli, transcription factors, and target genes, R scales to 1009 components (rows)×1685 regulatory reactions (columns), including 579 exchange reactions. When coupled with an environment matrix E, the size of the eventual R* matrix representing the complete TRS was 1009 rows by 2461 columns. It is reasonable to expect that similar observations will be made for systems that maintain similar distributions of inputs per regulated gene (see Figure 4) as well as multi-subunit complex and isozyme composition for metabolic enzymes and transporters.
Recently, the functional states of a prototypic TRS as well as a small-scale E. coli lac operon TRS, as represented by this pseudo-stoichiometric regulatory network matrix formalism, were characterized. The sheer number of environmental stimuli defined in this system, however, prohibits a comprehensive analysis encompassing all possible combinations as was performed for the prototypic TRN in [7]. Instead, we performed a random sampling of all possible environments to characterize key properties of the functional states of the E. coli TRS. Specifically, we computed the percentage of these randomly-simulated environments in which the 629 regulated genes were expressed, identifying those genes most significant to the E. coli regulatory program as those ubiquitously expressed across the environments. Importantly, this type of in silico expression analysis offers an efficient way to characterize differences in a regulatory program across multiple environments. Indeed, our results for two environments for which microarray profiling has previously been completed exhibited strong concordance with the experimental data.
This work constitutes the first genome-scale analysis of the fundamental subspaces of R*, and understanding of these subspaces has been significantly enhanced with the genome-scale implementation. Specifically, we describe how the column and left null spaces of R* are spanned by components of the system that are either very connected or very disconnected, respectively, among the regulatory relationships. For example, the column space of a representative E. coli R* matrix contained Crp and oxygen, two key systemic regulators. By contrast, the left null space of a representative matrix contained such extracellular cues as D-galactarate, which are minimally involved in the E. coli regulatory program. Thus, the left null space can identify network features that are poorly characterized (and require further experimental interrogation) or network function with minimal effect on phenotype. We further describe how the row and null spaces of R* together describe all possible expression states of the E. coli TRS for a given environment.
Interestingly, the singular value spectrum for the E. coli TRS is uniformly distributed (see Figure 7A), implying that there are few dominating components or reactions within the system. As described previously, this result contrasts with the network topology observed in Figure 1B as well as the connectivity distributions shown in Figure 4, and it implies that the functional states of the genome-scale E. coli TRS are diffuse. Moreover, the result suggests that transcriptional regulation is inherently different from metabolism, in which the first few principal modes sufficiently recapitulate a significant fraction of the underlying stoichiometric network [36]. Although an important caveat to this result is that metabolism is far better studied than regulation, it is noteworthy that control of many complex systems is distributed, including chemical plants, pharmaceutical manufacturing pipelines, electrical power grids, and sensor networks [37]–[39]. A distributed control system (DCS) is one in which there exist multiple controllers, with one or more of these controllers managing each component or “subsystem.” DCSs facilitate cost savings (as there exist fewer input/output connections), improved scalability (as a central node does not become overburdened as additional components are added), and greater redundancy (as no one node serves as a key hub) [37]. These advantages are critically important in biological systems. For example, there is increasing evidence that control of energy balance is distributed through different parts of the brain [40]. As a TRS constitutes the “control system” for a single living cell, a distributed regulatory network seems a likely choice for cells to evolve toward over time. Indeed, a recent study identified a hidden distributed architecture underlying the scale-free TRN of yeast [41]. Similarly, riboswitches, the structured elements found in 5′ untranslated regions of mRNAs that regulate gene expression by binding to small metabolites, have been shown to exhibit distributed functional effects within a genome [42]. Whereas the structure of metabolic pathways remains constant across multiple environments, our findings suggest that there exist many direct and specific (i.e., one-to-one) relationships between a given environment and the sets of genes that are turned “on” (and, in turn, the fluxes through the metabolic pathways). Thus, the uniform distribution of the singular value spectrum that we have observed implies that, in spite of the operon and regulon structure observed in the network topology, there exists a need for a functional analysis rather than a structural one. Further work exploring this type of a relationship in other organisms may provide interesting insights into evolutionary differences in their regulatory programs.
Among the E. coli TRN components, 125 are transcriptional regulators. While a few of these are global regulators like Crp and affect the expression of many genes, the majority of regulators control few targets (see Figures 1 and 4). This architecture of the E. coli TRN [43] has been explored extensively in recent years [44]. A drawback to these structural studies is that their direct relevance to the functional state of the cell is often unclear since they rely on inferential associations, e.g., based on the functional annotation of target genes to assign causal relationships to identified network motifs.
The work presented here focuses on the network for which relationships between environmental stimuli and transcription factors are directly ascribed. In so doing, relationships between the environment and transcriptional state can be mapped. Furthermore, the representation of the E. coli TRS in Figure 1B illustrates the complexity of the network in terms of the numbers of components and interactions. Tracing the edges to determine which genes are transcribed under different environmental conditions would be a difficult process. Instead, representing these interactions in a structured matrix facilitates the use of linear algebra techniques for characterizing emergent properties of a TRS and generating novel hypotheses of the system.
While this work involves an investigation of a previously constructed model of E. coli, this formalism may also prove useful for structuring and analyzing emerging high-throughput data for E. coli. For example, ChIP-chip data analyzing the genome-wide binding profiles for several microbial transcriptional regulators, including Crp, Fnr, and Lrp as well as various nucleoid binding proteins and sigma factors have appeared [45]–[47]. These results have suggested that, in spite of the interactions that have been characterized thus far, there remains considerable complexity within the E. coli TRS that needs to be further elucidated [46]. For instance, as shown in Figure 5C, certain GO categories of genes exhibited much poorer validation than others, suggesting that specific parts of the TRS require further study. The model performed well for aspects of E. coli biology that have been thoroughly studied to date, namely regulation of genes involved central metabolism and carbon uptake [27]. By contrast, energy metabolism and building block biosynthesis-related genes exhibited less than 70 percent validation. These results will likely be similar for other organisms as well, as the initial focus of study for biological systems has primarily been metabolism. Importantly, given the distributed functional nature of the E. coli TRS, the probability of a single incorrect gene expression prediction resulting in a large-scale reduction in accuracy (owing to residual effects upon downstream target genes) is small. Specific genes whose model-predicted expression states did not match with experimental measurements will nevertheless need to be explored. A defined environmental perturbation is critical for proper mapping of regulatory response and interactions with downstream targets. Furthermore, ChIP-chip data in isolation are not sufficient for this methodology to be successful. Corresponding transcriptional profiling data in order to derive directionality of regulation (i.e. up or downregulation of targets) are also important. Relatively conservative criteria should be used in incorporating these data into the R matrix.
Importantly, in spite of the advances using a R matrix formalism, there are certain limitations to the pseudo-stoichiometric approach that we have utilized for representing the E. coli TRS. In particular, as our reconstruction is based on an existing Boolean model, it is binary both at the level of control (i.e., how inputs affect individual genes) and expression (i.e., genes are predicted to be turned “on” or “off” in response to a given environment). In the future, mechanisms for incorporating species concentrations (particularly at the level of inputs) will need to be incorporated. In addition, during our analysis, the dynamics of the TRS are approximated. Although for a given environment we compute the sequence of expression states before a “steady-state” (or oscillation) is observed effectively tracing through the dynamics, we do not incorporate time explicitly.
Nevertheless, the pseudo-stoichiometric approach to defining regulatory interactions is akin to existing stoichiometric strategies for modeling metabolic and signaling systems. Accordingly, future work could directly incorporate the regulatory equations described herein to develop a comprehensive model of the cell [48]. For example, a recent approach characterizing dynamic properties of integrated signaling, metabolic, and regulatory systems is predicated on the availability of stoichiometric or pseudo-stoichiometric matrix representations of these types of systems [49]. Ultimately, in this way, the application of the R* matrix to genome-scale regulatory networks may enable the quantitative investigation of emergent properties of biochemical systems, including whole-cell dynamics.
|
10.1371/journal.pntd.0003923 | Bacterial Infection and Immune Responses in Lutzomyia longipalpis Sand Fly Larvae Midgut | The midgut microbial community in insect vectors of disease is crucial for an effective immune response against infection with various human and animal pathogens. Depending on the aspects of their development, insects can acquire microbes present in soil, water, and plants. Sand flies are major vectors of leishmaniasis, and shown to harbor a wide variety of Gram-negative and Gram-positive bacteria. Sand fly larval stages acquire microorganisms from the soil, and the abundance and distribution of these microorganisms may vary depending on the sand fly species or the breeding site. Here, we assess the distribution of two bacteria commonly found within the gut of sand flies, Pantoea agglomerans and Bacillus subtilis. We demonstrate that these bacteria are able to differentially infect the larval digestive tract, and regulate the immune response in sand fly larvae. Moreover, bacterial distribution, and likely the ability to colonize the gut, is driven, at least in part, by a gradient of pH present in the gut.
| Symbiotic microorganisms influence many aspects of the physiology of their hosts. In insects, symbiotic bacteria are able among other things to modulate the immune response and the development of the insect from larval stages to adult. Many bacteria first gain access to insect tissues, such as the gut, during larval development, and are acquired from the environment. Thus, depending on the insect ecology, aquatic vs. terrestrial, the bacterial gut flora found in insects can vary widely. Little is known about the events that follow bacterial infection in larval guts and the driving forces for colonization of the gut by such bacteria. We investigated the distribution of two bacteria, a Gram-positive (Bacillus subtilis) and a Gram-negative (Pantoea agglomerans) fed to sand fly larvae. Our results indicate that bacteria distribution in the larval gut is driven by their ability to multiply at a given pH, as pH in the gut also varies. Gut distribution by these bacteria lead to an immune response that the sand fly larva is able to modulate according to the bacterial species. Our findings can influence development of paratransgenic approaches that utilize bacterial symbionts to control vector population.
| Bacterial symbionts significantly influence many aspects of the physiology of their host. In insects, both pathogenic and non-pathogenic bacteria have been shown to modulate immune response, homeostasis, development, and overall health of midgut physiology for both larval and adult stages. Both Gram-positive (G+) and Gram-negative (G-) bacteria are commonly associated with the midgut tissue of Diptera, including several disease vectors. Many bacilli and enterobacter such as Lactobacillus and Pantoea have been identified from the midgut and other tissues of the fruit fly Drosophila melanogaster [1]. In field collected Anopheles stephensi, Anopheles gambiae, and Aedes aegypti from laboratory colonies, a number of G+ and G- bacteria were identified [2–5]. Among G-, Pantoea agglomerans was also the most common genus identified from all cultivable bacteria in both male and female Aedes albopictus collected from two out of four sites in Madagascar [6]. In sand flies, several studies have focused on regional and potential species-specific variability in the microbial community of both Phlebotomus and Lutzomyia species. P. agglomerans and Bacillus spp. were commonly found in both natural and laboratory-reared sand fly populations. Bacillus subtilis and non-Pantoea members of the Enterobacter family were shown to be present in populations of Phlebotomus papatasi from, India, Turkey, Tunisia, and Egypt [7,8]. Bacillus spp, Serratia marcescens, and P. agglomerans were identified in natural and laboratory populations of adult Lutzomyia longipalpis [9–12]. In laboratory colonies, Bacilli and Enterobacter were also identified in the larval stages of L. longipalpis [13,14]. However, to date, most studies were focused on microorganisms present in the adult sand fly with little or no attention paid to physiological effects brought about by these microorganisms on developing stages. Here, we investigated the effects of EGFP-expressing B. subtilis (Bs) or GFP-expressing P. agglomerans (Pa) on the midgut innate immunity and epithelial homeostasis of 3rd instar larvae of L. longipalpis. Additionally, we determined that the distribution of these bacteria within the midgut of sand fly larvae is in part driven by the gut pH, and demonstrate a cytotoxic effect on the midgut epithelium by infection with Pa. Strikingly, we show evidence suggesting a differential and suppressive response to infection with respect to G+ and G- bacteria, likely influenced by the gene encoding the negative regulator of immunity, "poor immune response upon knock-in” or Pirk. Up-regulation of Pirk transcript level leads to a significant depletion of the transcripts encoding the antimicrobial peptide Attacin and the immunomodulatory peroxidase IMPer. With respect to the distribution of these microbes across the length and distal spacing of the midgut, G+ Bs were observed throughout the entire alimentary canal in larvae, whereas G- Pa were found primarily in the posterior midgut. Our results strongly suggest that the range of pHs present within the sand fly larval gut likely is a driving force defining the ability of certain bacteria such as P. agglomerans to infect areas of the gut. The results presented here may have implications beyond the sand fly system and may explain how the distribution (and possibly colonization) of bacteria and other microbes may occur within the guts of insects.
L. longipalpis (Jacobina strain–LLJB) were the colony maintained in the Department of Entomology, Kansas State University. Larvae were maintained in 250 or 500 ml plastic jars (Nalgene) with an approximately 2 cm-thick bed made of dental plaster (Schein), and fed on either larval chow (a mixture of 50% rabbit droppings and 50% rabbit food) or on 1.5% agar in LB, with or without the (E)GFP-expressing bacteria (see below).
EGFP-expressing B. subtilis (strain 1012 transformed with plasmid pAD43-25, obtained from the Bacillus Genetic Stock Center) and GFP-expressing P. agglomerans (strain EPA-E325 transformed with plasmid pT-3078-5, a gift from Dr. David Lampe) were cultured at 30°C overnight in LB medium supplemented with 5 μg/ml chloramphenicol (Alfa Aesar, A Jonson Mattey Co.) or with 50 μg/ml Carbenicillin (Teknova), respectively. Bacterial cultures were centrifuged at 2500 rpm for 20 minutes at room temperature and the pellets were washed twice with 1X PBS. Bacteria were then suspended in PBS for a final concentration of 109 bacteria in 50-to-80 µl that was spread on a plate containing a thin layer (2–3 mm thick) of LB-agar (no antibiotics) and grown overnight. The following day, fluorescence of the bacterial lawn was confirmed and the LB-agar was cut into 3-to-5 mm2 pieces to be fed to early L3 L. longipalpis larvae (depicted in S1 Fig), and lawns were replaced every 48 h. Third instar larvae were used to maximize food intake as L3s feed more and for a longer period of time than L4. Prior to feeding on the LB-agar with Bs or Pa, all larvae were starved for 6 to 10 hours to allow for excretion of midgut contents, and rinsed in sterile water. As controls, L3 larvae were fed on LB-agar plus 5 mM Paraquat, an insecticide that strongly induces apoptosis [15], and on LB-agar plus Kanamycin (50 µg/ml) (we were unable to use plain LB-agar due to contamination). Larvae fed ad libitum for up to 48 h at 26°C and 80% humidity, with a 12:12 h light-dark cycle. Groups of (n = 20) larvae were collected at 12, 24, 36, and 48 h post feeding with three biological replicates. Food intake was determined by examining each larva under a dissecting microscope (10X). All larval feedings were done according to feeding groups using 500 ml Nalgene pots with a 2–3 cm layer of dental cement.
Alternatively, larvae were fed for 12 h on bacteria-containing LB-agar and transferred to pots with plain LB-agar (no bacteria and no antibiotics). Larvae in groups of 3 to 5 were collected every 3 h and assessed for GFP signal using a Zeiss confocal LSM microscope. CFUs were also measured for larvae collected at 12 h and 24 h, by surface sterilizing each larva, dissecting and grinding each whole gut using a hand-held homogenizer in 60 µl 1X PBS, and plating the homogenate on selective media (5 µg/ml chloramphenicol for B. subtilis or 50 µg/ml carbenicillin for P. agglomerans) and incubating at 28ºC to 30ºC.
To assess for any effects of diet on midgut development, the length and width of the larval midgut were measured using the LSM 510 using the software ZEISS LSM Image Browser (Zeiss International). Midgut length was determined by measuring from the beginning of the anterior midgut to the posterior region of the midgut. Width measurements were obtained from three regions of each midgut: the anterior (ant), the middle (mid), and the posterior (pos) regions.
Whole guts from L. longipalpis L3 larvae were dissected (n = 3) from three separate treatments of L3 into PBS and fixed for 20 minutes at room temperature with 4% paraformaldehyde in PBS. Tissues were washed 4 times for 30 minutes with PBS containing 0.3% Triton X-100 (PBST), then blocked with PBS containing 1% bovine serum albumin for 30 minutes at room temperature. Tissues were then incubated overnight at 4°C with primary antibodies for rabbit anti cleaved caspase3 (Cell Signaling Technology) diluted 1:500 in PBST. Tissues were washed 3 times for 30 minutes with PBST, and incubated overnight at 4°C with Alexa Fluor® 594 goat anti-rabbit (Invitrogen) diluted 1:1000 in PBST. Tissues were washed 3 times for 30 minutes with PBST, and nuclei were stained for 5 minutes with 10 µg/mL of DAPI (Invitrogen). Samples were mounted in Vectashield® (Vector Laboratories) anti-photo bleaching reagent, and images were obtained with a LSM 510 confocal microscope using the software ZEISS LSM Image Browser. In addition, measurements of larval guts length and width were obtained for each feeding treatment using the LSM510 confocal microscope. For microscopy, entire alimentary canals were used for viewing clarity, whereas for gene expression analyses, all hindguts were removed prior to RNA isolation of larvae.
We assessed the pH within the midgut by feeding L. longipalpis L3 larvae with LB-agar containing 0.4% of the pH indicators Bromothymol blue and Phenol red. LB-agar medium adjusted to pH 7 added prior to sterilization. Each indicator agar medium was fed to a group to 50 L3 larvae following six hours of food deprivation. Feeding of the larvae was performed by placing the larvae and fragments of approximately 3–5 mm2 of the dye-containing agar inside a 500 ml Nalgene pot with a 1 cm layer of plaster, maintained at room temperature and with a relative humidity of 80–85%. Larvae were allowed to feed ad libitum for 20 hours. The pH indicator dyes were visualized through the translucent cuticle of the larvae. The pH inside the larval gut was determined by comparing the color and intensity shown within the gut with those from 0.4% solutions of both dyes made in 8 ml LB medium (with one added drop of chloroform to prevent bacterial growth), and with pH ranging from pH 4 to 10 in 0.5 increments. Larvae were also fed on 0.1 and 0.4% thymol blue.
Overnight cultures of fluorescent Bs and Pa were diluted to OD600 = 0.1 and further diluted 1:104 prior to plating onto LB-agar plates supplemented with either 50 µg/ml of carbenicillin (CAR) or 5 µg/ml of chloramphenicol (CAM) for selection of Pa or Bs, respectively. The pH of plates ranged from pH 6 to 9.5 in 0.5 increments. Plates were incubated overnight at 37°C for Bs, and at room temperature for Pa, and each experiment was performed in triplicate, and repeated twice. The following day bacterial colonies were counted, and colonies growing at each pH were observed under fluorescent microscope. One-way ANOVA with a Tukey test was performed to determine differences between pH.
Midguts from L. longipalpis L3 larvae were dissected under a stereoscope microscope in Hyclone (Thermo Scientific) phosphate buffered saline (PBS) at 12, 24, and 36 h post feeding in either the Bs or Pa. Sterile, 1.5% agar in LB was used as feeding control. Total RNA was isolated from pools of 20 midguts using TRIzol (Invitrogen). For each group of larval midguts, RNA isolation was done in triplicate. RNA quality was assessed by electrophoresis on 1% agarose-5% formaldehyde in 1x MOPS, and stored at -80°C. First strand cDNA synthesis was conducted using Superscript III reverse transcription kit (Invitrogen) as described [16].
mRNA levels were quantified with iQ SYBER Green Supermix (Bio-Rad) using 95°C melting, 57°C annealing, and 72°C extension temperature for 40 cycles using a Realplex4 Master cycler (Eppendorf). Relative fold changes were assessed using the ∆∆Ct method [16,17], and calibrated against the expression observed for same stage larvae fed on the plain LB-agar control. Sequences for Attacin (Att), IMPer, Vein, Domeless, IMD, Pirk, USP36, and Duox were obtained using the tBLASTN algorithm from corresponding annotated sequences found in D. melanogaster blasted against L. longipalpis contigs. Predicted full-length transcripts were made using GENSCAN (http://genes.mit.edu/GENSCAN.html), and primers sequences were generated using Primer3 (http://biotools.umassmed.edu/bioapps/primer3_www.cgi). Primers for RPS6 were previously described in [16]. Primers for Def1 were based on the sequences described in an earlier study [11]. All other primers used in this study were designed from gene sequences in the NCBI database and their accession numbers are as follows: IMPer, AJWK01035414.1; DUOX, AJWK01035414.1; IMD, AJWK01008032.1; Domeless, AJWK01008028.1; Attacin, AJWK01017071.1; Pirk, AJWK01015539.1; USP36, AJWK01027563.1; and Vein, AJWK01005322.1.All primer sequences used in this study are summarized in S1 Table.
A scheme representing the anatomy of L. longipalpis sand fly 3rd instar (L3) gut is depicted in Fig 1A. Following continuous feeding of LB-agar containing EGFP-expressing Bs to larvae, a pervasive signal was found across the entire length of the midgut for infected insects as depicted by the fluorescent signal of full length images of the gut (Fig 1B and S2A Fig and S1 Video). However, when the GFP-expressing Pa was fed to larvae in a similar manner the bacteria were mostly localized to a narrow area of the posterior portion of the midgut (Fig 1C and S2B Fig), and were only found on the apical surface of the midgut lumen (S2 Video). Also observed were areas of higher intensity GFP signal in Pa-infected guts, suggesting the presence of biofilm (Fig 1C). A less pronounced GFP signal in Pa fed was also observed in the proventriculus of the gut (Fig 1C). Infection rates as determined by a qualitative assessment of the GFP signal within the midgut of the larvae following continuous feeding are described in Table 1.
To further investigate what could be driving such a distinct microbial distribution, we assessed the gut pH range within larvae in vivo, and compared it to pH growth assays for the two GFP-expressing bacteria in vitro. LB-agar supplemented with pH indicators Bromothymol blue or Phenol red were fed to larvae, and the color gradient generated was visible through the translucent cuticle of larvae using light microscopy. Intensity of colors varied between larvae due to the initial ingestion time, load size, and bolus movement across the gut. Fifteen larvae of each treatment were compared to the pH references. The Bromothymol blue dye has a range of pH from 6 to 7.6, and Phenol red has a range of 6.8 to 8.4. Previous results on L. longipalpis larval gut pH indicated a basic pH >9 in the anterior portion and an more acidic pH >6.5 in the posterior portion of the midgut [18]. Larvae fed on thymol blue (at concentrations of 0.4% and 0.1%) displayed a green-colored gradient that was not easily distinguishable between pHs 8.5 to 9.5 through the larval cuticle. Two other pH indicators, alizarine yellow and thymolphthalein, were considered during our studies. However, because of our choice of using whole larva, the output colors of these pH indicators were not suitable for visualization through the insect cuticle. The results shown in Fig 2 confirm such a pH gradient in the L. longipalpis larvae, clearly pointing to a basic pH for the anterior part of the midgut, including the proventriculus (PV), and an acidic pH in the posterior part of the midgut.
Both Bs and Pa bacteria were grown on antibiotic supplemented LB-agar plates with pH ranging from 6 to 9.5. CFU counts were obtained in triplicate to assess the viability of the two strains at different pH (Fig 3A). Colonies of EGFP-expressing Bs did not show a significant difference in CFU counts at any given pH. Additionally, the colony size for Bs fed was smaller at low (6–6.5), and high (9.5) pH. However, Pa showed a significant difference in CFU counts at pH ranging from 6–7 with respect to pH 9.5 (Fig 3B). Also, at pH 8–9 colony size and fluorescent intensity began to decrease, and by pH 9.5 there was no visible growth.
Following the continuous feeding experiment described above, we tested whether similar results could be obtained by feeding larvae once with either Bs or Pa. Larvae were fed for 12 h on LB-agar with the respective bacteria and transferred to pots with fresh LB-agar (no antibiotics). Pa bacteria were cleared by 21 h after infection, whereas Bs were cleared by 24 h (Table 2). CFU counts (Table 3) were generally in agreement with the results observed for the GFP signals assessed. No bacteria from pupae were able to be re-cultured. Interestingly, sand fly larvae may become cannibalistic when food is scarce. We also noticed that the Pa-infected larvae appeared to rely on cannibalism more frequently than Bs-infected.
Bs and Pa were assessed for their ability to infect the sand fly L3 larvae midgut following feeding, and their effect on induction of apoptosis. A monoclonal antibody targeting the cleaved caspase3 was used as an immunocytochemical marker to identify epithelial cells undergoing caspase-dependent programmed cell death, and to assess the integrity of the midgut. This antibody was used previously in Drosophila to detect c-Jun N-terminal kinase (JNK) pathway activation in response to infection [19]. In order to test if this was a viable approach, we fed larvae with LB-agar supplemented with the apoptosis inducer Paraquat, and compared its effects to larvae fed on LB-agar alone. The LB-agar fed larvae displayed a well-defined midgut epithelium with little background staining for active caspase3 (S3A Fig). In contrast, larvae fed on LB-agar supplemented with Paraquat showed midgut epithelia with significant loss of integrity, that were also severely flattened after mounting on the slide with reduced luminal space detectable by looking at nuclei alone (S3B Fig). The apoptotic effect of the Paraquat was further confirmed by the presence of a large population of cells showing heavy cytoplasmic specific staining for caspase3 (S3C and S3D Fig). After 12h of infection with Bs an extensive amount of luminal nucleic material is observed using DAPI staining (Fig 4A), and a massive infection can be seen in Fig 4B. Very little background caspase staining is observed in Bs-fed compared to the Paraquat treated controls (Fig 4C and 4D, and S3 Fig). In contrast, when Pa was used for infection, the Gram-negative bacteria induced staining comparable to that of the Paraquat control (Fig 4G and 4H and S3 Fig). However, we did not observe a similar breakdown in midgut superstructure.
Effects of feeding agar to the developing sand fly L3 larvae were assessed by comparing the length of the whole midgut and the width of three areas within the midgut to those obtained from larvae raised on regular larvae food (50% rabbit feces + 50% rabbit food). The same parameters were also measured from guts of Pa and Bs infected larvae. A significant difference was found for midgut length when comparing regular sand fly larval food and the agar fed, except at 48 h. For the three measurements of gut width (anterior, middle and posterior), significant differences were only observed between regular food and agar (S4 Fig).
We compared the mRNA expression profiles of nine genes involved in various physiological processes ranging from innate immunity, homeostasis, and epithelial regeneration in midguts of L. longipalpis L3 larvae fed on Bs and Pa bacteria. Among the transcripts assessed were Att, Def1, Duox, IMPer, Vein, Domeless, IMD, Pirk, and USP36. Results from qRT-PCR indicate that, compared to control fed, larvae fed on agar containing either Bs or Pa showed significant difference in expression for a number of genes analyzed.
At 12 h post infection, transcript levels for Att were downregulated by nearly 75% for both Bs and Pa infected larvae, and IMPer was downregulated by 25% in Pa fed (Fig 5A). Pirk showed a 2.5-fold change (~150% increase) in expression following infection with Pa (Fig 5B).
At 24 h post infection, Domeless and IMD followed a similar profile and were upregulated in both infections, albeit Domeless was not significantly different between the control and Bs fed. Pirk was also upregulated at in both bacterial infections, and with a profile that also was similar to Domeless and IMD (Fig 5E). USP36 was downregulated only in Bs infection, and Def1 was upregulated only in Pa (Fig 5F).
At 36 h post infection, Domeless and IMD again displayed similar profiles, however both were downregulated in comparison to control (Fig 5H). For Pirk, whereas expression if Bs fed returned to control levels, in Pa infection it remained significantly higher (nearly 2-fold) (Fig 5H). Finally, USP36 expression was reduced by roughly 10% in both infections, but for Bs the significance was 0.078 (Fig 5I).
In insects, gut bacteria have been shown to significantly contribute to nutrition, modulation of the immune response, and protection from parasites and other pathogens. The insect gut varies greatly in terms of morphology and physicochemical properties, and these may influence the distribution and structure of the microbial community in the gut. During development, sand fly larvae are exposed to a wide variety of soil bacteria and other microorganisms that are able to colonize the insect gut [7–14,20]). However, as we are aware, no studies have focused on the mechanisms by which bacteria are able to develop within the sand fly gut, or the types of specific responses induced by the colonization. Here, we assessed the ability of two bacteria previously identified from the guts of insects, including sand flies, to infect the gut of sand fly larvae, and investigated the specific responses (innate immunity, epithelia regeneration, homeostasis) induced by these bacteria.
When fed to L. longipalpis larvae, EGFP-expressing Bs bacteria were distributed throughout the entirety of the alimentary canal, mainly within the peritrophic matrix and along the lumen. In contrast, GFP-expressing Pa bacteria were mostly localized to the posterior midgut, and only at the apical surface (although we did observe GFP signal for Pa at the proventriculus of the gut, it is possible that these bacteria had not yet been killed by the alkaline conditions). We speculated that this phenomenon must be driven by pH and/or by specific cell types that line the midgut lumen. A pH driven effect on the distribution of bacteria in the sand fly larval gut was suggested by the use of pH indicators. It has been previously determined in sand fly larvae that the pH of the midgut is highly alkaline at the anterior portion and decreases towards the posterior region [18]. Our approach allowed us to confirm the location within the larval gut wherein the pH ranges between 6 and 7, which also coincides with the infection of the Gram-negative Pa. These results were indirectly confirmed by in vitro growth assays obtained for Pa in which these bacteria clearly favor a pH in the range of 6-to-7. The ability of Bs to sporulate under unfavorable growth conditions may have further contributed to its distribution along the gut of the larvae.
In addition, we observed a marked difference between the rates and the persistence of infections of Pa and Bs in sand fly larvae. Following continuous feeding on bacterial lawn, most of the Pa bacteria are cleared within 24 h whereas infection with Bs remained for up to 48 h. However, if the bacterial lawn is replaced at 48 h during continuous feeding, larvae do re-infect. In contrast, with the non-continuous feeding on the bacterial lawn led to clearing of Pa by 21 h and Bs by 24 h. Hence, the data indicate that the sand fly larvae are able to clear bacterial infection, by the activation of the antimicrobial immune response, if exposure is not maintained. Another possibility is that the loss of Pa and Bs during non-continuous feeding may also be caused by competition with other microorganisms present in the larval gut. And in spite of differences known to exist in the half-lives of GFP and EGFP proteins [21,22], the CFU counts reported support the clearing of Pa from the midgut during loss of GFP signal.
Using a caspase3 antibody [19,23] to detect apoptotic activity in L. longipalpis, we also were able to clearly identify differences between Bs and Pa infection of the sand fly larval gut. The microscopy data strongly suggest that only Pa induces caspase activity within the midgut of the sand fly, while Bs causes little to no staining.
Quantitative RT-PCR analyses were used to assess changes in the expression profiles of nine selected genes chosen based on their roles in insect midgut immunity and homeostasis. Related to midgut immunity, selected genes included those coding for effector molecules such as the antimicrobial peptides (AMPs) attacin (Att) and defensin (Def1), as well as Duox and IMPer. Also included in this category was the immunodeficiency regulatory gene encoding " poor immune response upon knock-in” or Pirk.
Attacin has long been implicated in bacteria killing from a number of studies pertaining to its role in innate immunity [24]. Def1 was shown to be upregulated in adult L. longipalpis after bacterial challenge [11]. It has been shown that Defensin A, acting in concert with Cecropin A, blocks Plasmodium transmission in A. aegypti [25]. The effector molecules Duox and IMPer have been demonstrated to have effects on the midgut peritrophic matrix structure and parasite killing in A. gambiae mosquitoes [26–28]. For Def1, Att, DUOX, and IMPer, there are multiple studies suggesting that these effector molecules are regulated by the immunodeficiency pathway [27,29,30]. Pirk has been previously shown to be a negative regulator of IMD activity [31,32]. While Pirk acts to suppress IMD at the level of signal transduction, Caspar negatively regulates IMD at the level of transcription. Studies in A. gambiae implicate the knockdown of IMD in increased infectivity of mosquitoes [33,34]. In sand flies, Caspar knockdown led to a decrease in Leishmania mexicana load in L. longipalpis [12].
Domeless, Vein, and USP36 were selected based on their roles in pathways related to innate immunity to midgut regeneration. When the innate immune response is activated in the midgut, there are associated energy costs and damage to healthy epithelial cells that can negatively affect the insect. Artificially activating ROS production in A. stephensi led to reduction in infective lifespan, and deleterious effects associated with mitochondria [35]. Domeless is a receptor in the JAK/STAT pathway that is crucial for recognizing damage to healthy epithelial cells. JAK/STAT signaling reaches intestinal stem cells (ISCs) and enteroblasts (EBs) leading to the secretion of an epidermal growth factor (Vein) ending in regeneration of midgut epithelia via proliferation and differentiation of ISCs and EBs [19,36–38]. Additionally, the deubiquitinating enzyme USP36 is a negative regulator of IMD and provides a route of cross-talk between IMD and JAK/STAT pathways [39,40]. USP36 is also involved in controlling selective autophagy [40].
Our results suggest that Pirk may be acting to suppress Att and IMPer activity at 12 h post infection for Pa-infected insects, however, another still unidentified mechanism likely is involved in the reduction of Att levels for Bs-infected. Although Pirk was significantly upregulated during Bs and Pa infection at 24 h, there was also an increase for the immune transcription factor IMD. Such IMD increase can be linked to an associated upregulation of Def1 in Pa infected, but no significant difference was found in Def1 for Bs infected. The upregulation of IMD may be explained by the concomitant down regulation of USP36 in the Bs infected larvae, but not Pa infected. Additionally, the nearly two-fold increase of Domeless in Pa fed larvae suggests the possibility of homeostatic response to damage by the larvae immune response that occurs within the first 12 h of infections as our data have demonstrated.
By 36 h, upregulation of Pirk continued in Pa infected individuals, but no effect was observed for the expression of the effector molecules. With the downregulation of USP36, we expected an upregulation of IMD. However, the opposite was detected: IMD was downregulated. Interestingly, Domeless was also downregulated at 36 h, possibly due to lack (or clearing) of bacteria in the gut as indicated by the non-continuous feeding experiments. It remains to be investigated whether downregulation of IMD, when USP36 levels were also lowered, is associated with increased autophagy during bacterial clearing. The expression analyses data corroborates what was observed with regards to the progress of infection in Bs and Pa. Of significance, our results indicate that sand fly larvae are able to differentially regulate (or suppress, as the case here) their immune response according to the bacterial challenge they are exposed to.
We have previously shown that feeding different bacteria to L. longipalpis larvae affects survival and development [14]. In the current study, we demonstrate a selective distribution of bacteria in the larvae driven by gut pH and downstream effects on the larval gut. This provides a link between the type of bacteria infecting (as the case here) or colonizing the gut, physicochemical aspects of the gut, and overall insect health. With regards to mechanisms driving the localization of bacteria, it is also likely that different cell types lining the gut epithelia are involved. In support of this hypothesis, concentrated pockets of Pa binding to the posterior end of the larval gut were observed, indicating the presence of a biofilm. However, the presence of a preferred cell type or membrane receptor involved in binding of bacteria cannot be discarded. Gram-negative bacteria are known to form biofilm within the gut of vectors [41]. P. agglomerans form intestinal biofilms in the Mediterranean fruit fly Cerititis capitata [42] that resemble what we observed in sand fly larvae. Taken together, these data suggest a pH-dependent localization or growth of bacteria within the insect midgut previously reported to be a random event [42].
With regards to cell type, Fernandes et al [43] reported the presence of different cell types in A. aegypti during development and metamorphosis, and a precedent for favored microbial binding was previously demonstrated for Leishmania major binding to the midgut epithelial cell lining of the sand fly P. papatasi [44]. Thus, it is conceivable that at least one of these events may also be involved in dictating the success of bacterial colonization within the sand fly gut. Nevertheless, mechanisms such as autophagy may also play a role in bacterial removal (reviewed in Huang et al. [45]). Further, both Pa and Bs do not survive metamorphosis.
It is important to note that the agar based feeding system used in our experiments does not replicate the natural conditions faced by sand fly larvae and agar does not provide the necessary nutrition for normal larvae development. As shown by our analyses, the midgut length and width differed significantly between larvae fed on agar versus those fed on regular sand fly larvae chow. However, no differences in such parameters of the midgut morphology were observed between agar fed larvae and the agar plus bacteria fed larvae. Interestingly, differences observed for the midgut parameters tested only lasted until either the agar or the bacterial lawn were replaced. Additionally, larvae were not able to sustain a GFP-positive signal when fed on LB-agar plus bacteria for 12 h and then transferred to plain LB-agar. These data were also supported by CFU counts. Notwithstanding, this method was proven useful for specifically delivering the EGFP- or GFP-expressing bacteria. Similar approaches may be used to deliver selected microbes to sand fly larvae in paratransgenic applications to control sand fly populations [46,47].
In conclusion, this study demonstrates that bacteria selectively infect the sand fly larvae midgut, (possibly) leading to epithelial damage. In addition, the data also point to a modulation of the innate immune response likely controlled by expression of Pirk. We also show for the first time that the insect midgut pH is a factor driving microbial organization of the gut. Our results contribute towards understanding of midgut responses to infections and provide new insights for development of vector control approaches using paratransgenesis.
|
10.1371/journal.pntd.0001383 | Induction of CD4+CD25+FOXP3+ Regulatory T Cells during Human Hookworm Infection Modulates Antigen-Mediated Lymphocyte Proliferation | Hookworm infection is considered one of the most important poverty-promoting neglected tropical diseases, infecting 576 to 740 million people worldwide, especially in the tropics and subtropics. These blood-feeding nematodes have a remarkable ability to downmodulate the host immune response, protecting themselves from elimination and minimizing severe host pathology. While several mechanisms may be involved in the immunomodulation by parasitic infection, experimental evidences have pointed toward the possible involvement of regulatory T cells (Tregs) in downregulating effector T-cell responses upon chronic infection. However, the role of Tregs cells in human hookworm infection is still poorly understood and has not been addressed yet. In the current study we observed an augmentation of circulating CD4+CD25+FOXP3+ regulatory T cells in hookworm-infected individuals compared with healthy non-infected donors. We have also demonstrated that infected individuals present higher levels of circulating Treg cells expressing CTLA-4, GITR, IL-10, TGF-β and IL-17. Moreover, we showed that hookworm crude antigen stimulation reduces the number of CD4+CD25+FOXP3+ T regulatory cells co-expressing IL-17 in infected individuals. Finally, PBMCs from infected individuals pulsed with excreted/secreted products or hookworm crude antigens presented an impaired cellular proliferation, which was partially augmented by the depletion of Treg cells. Our results suggest that Treg cells may play an important role in hookworm-induced immunosuppression, contributing to the longevity of hookworm survival in infected people.
| The hookworm infection is characterized by the long-term survival of the parasite and the concomitant modulation of the host immunity. Among several mechanisms that may account for the suppression of T cell response, we here described the presence and role of T regulatory cells (also known as Tregs) in the human hookworm infection. Tregs are a minor subpopulation of CD4+ T-cells, which also express specific cell markers that allow its further identification (CD25 and FOXP3). Our results showed that hookworm infection induce an augmentation of Tregs in the peripheral blood, followed by the higher levels of circulating Treg cells expressing several markers and cytokines associated with cell regulation (CTLA-4, GITR, IL-10, TGF-β and IL-17). We also demonstrated that in vitro depletion of Tregs partially enhanced the naturally impaired cellular proliferation of lymphocytes from infected individuals after antigenic stimulation. Our results suggest that Treg cells may play an important role in hookworm-induced immunosuppression, contributing to the longevity of hookworm survival in infected people.
| Human hookworm infection is mainly caused by the blood-feeding nematodes Ancylostoma duodenale and Necator americanus, which infects 576 to 740 million people worldwide, especially in the tropics and subtropics [1], [2], [3]. Hookworm is considered one of the thirteen poverty-promoting neglected tropical diseases, and the second most important parasitic infection of humans [4]. This infection takes a particularly devastating toll on the most vulnerable of the world's population, including children, productive men, and women of childbearing age [5], [6], [7]. Persistent blood and serum protein loss attributable to chronic hookworm infection are associated with anemia, malnutrition, and growth/cognitive retardation, resulting in the annually loss of tens of millions of disability adjusted life-years (DALYs) [8]. Despite its overall impact over the global public health system, the immunomodulatory mechanisms associated with hookworm survival on host's intestine in face of an immunologically hostile environment are not yet fully understood.
Human hookworm infection is a longstanding, chronic infection with complex life cycle stages and host-parasite interactions. Although this parasitic infection may seem unnoticed by the immune system, an intense T helper (Th) 2 phenotype immune response is mounted by the host against this helminthic infection [9]. In recent years, evidence has accumulated that the immune response to hookworms may not be a simple polarized and putatively protective Th2 response, but rather a mixture of Th1/Th2 responses, presenting significant levels of interferon gamma and IL-12 production [10]. In fact, hookworms have a remarkable ability to downmodulate the host immune response, protecting themselves from elimination and minimizing severe host pathology. The parasite may promote its survival by excreting/secreting a panel of molecular immunosuppressive agents and, possibly, by stimulating the appearance of regulatory T-cell populations. Among the most striking aspects already described of this downregulation is the ablation of parasite specific T cell proliferative responses (“hyporesponsiveness”) [11]. Indeed, hookworm infection in animal models has been classically associated with impaired lymphocyte proliferation, functional defects in antigen presentation/processing and increased secretion of nitric oxide [12]. Recently, Fujiwara et al. have also shown that human dendritic cells differentiation and maturation may also be downmodulated by these worms, contributing to the T cell hyporesponsive state observed in individuals chronically infected with N. americanus [13].
The discovery of regulatory T lymphocytes (Treg) that are actively involved in maintaining immune tolerance has recently led to new insights into mechanisms of tolerance breakdown and/or immunoregulation in human diseases, including those resulting from allergic, autoimmune, or infectious causes [14]. These cells have been shown to suppress cellular immune responses through direct contact with immune effector cells and by the production of regulatory cytokines, including TGF-β and IL-10 [15], [16]. In fact, Geiger et al. have shown that hookworm infection is accompanied by elevated levels hookworm antigen-specific IL-10 production dependent on parasite stage, as well as significantly higher levels of CD4+/CD25+ T-cells [11]. While studies in experimental models have provided evidence for increased FOXP3+ (forkhead box P3 transcription factor) Treg function during different helminth infections [17], [18], [19], the role of Tregs cells in human hookworm infection is still poorly understood and has not been addressed.
To investigate Treg activity in human hookworm infection, we have evaluated Treg frequencies, function and immune responses to hookworm antigens in N. americanus-infected individuals from a rural area of Minas Gerais state (Brazil). In the present study we described an augmentation of circulating CD4+CD25+FOXP3+ regulatory T cells in hookworm-infected individuals compared with healthy non-infected donors. We have also demonstrated by flow cytometry that infected individuals present higher levels of circulating Treg cells expressing CTLA-4 (cytotoxic T lymphocyte antigen 4), GITR (glucocorticoid-induced tumor necrosis factor receptor), IL-10, TGF-β and IL-17. Additionally, we showed that hookworm crude antigen stimulation reduces the number of CD4+CD25+FOXP3+ T regulatory cells co-expressing IL-17 in infected individuals. Furthermore, PBMCs from infected individuals pulsed with excreted/secreted products or hookworm crude antigens presented an impaired cellular proliferation, which was augmented after the depletion of Treg cells. Taken together, our results suggest that Treg cells may play an important role in hookworm-induced immunosuppression, contributing to the longevity of hookworm survival in infected people.
The study was conducted in areas endemic for N. americanus in the Northeast Minas Gerais State, Brazil. Ten volunteers between the ages of 18 and 76 were recruited over the course of two months (Table 1). These volunteers reside in areas of moderate N. americanus transmission and presented with low to moderate (up to 872 epg) intensity of Necator infection. Individuals were selected on the basis of not having any other helminth infection (mono-infected after analysis of 6 slides of Kato-Katz fecal thick-smear and Baermann-Moraes techniques). The presence of Necator infection was determined by formalin–ether sedimentation and, if positive, two more stool samples were analyzed by the Kato–Katz fecal thick-smear technique and parasite load was expressed as eggs per gram of feces (epg) [20]. Ten hookworm-naive individuals were enrolled as healthy non-infected individuals from Belo Horizonte, Minas Gerais State, Brazil, where no transmission occurs. None of these individuals had a history of Necator infection and all presented with egg-negative stool (6 slides of Kato-Katz fecal thick-smear and Baermann-Moraes techniques) and no specific antibodies to Necator crude antigen extracts. The geographic areas included in this study are not endemic for tissue-dwelling helminth infections. Furthermore, the nutritional status of non-infected volunteers (controls) was similar to those presented by hookworm-infected individuals as determined by anthropometric measurements. The nutritional status of adults was determined using the absolute body mass index and classified as eutrophic (18.5–24.9 kg/m2), underweight (<18.5 kg/m2) or overweight (≥25 kg/m2) [21].
Approximately 24 mL of blood was collected in heparinized tubes for separation of peripheral blood mononuclear cells (PBMC) and 4 mL of blood in EDTA tubes for the immunological assays described below.
The study was approved by the Ethical Committee on Research of Universidade Federal de Minas Gerais (COEP) (Protocol #ETIC0449.0.203.000-09). Written consent was obtained from all individuals prior to enrollment in this study. Ancylostoma ceylanicum adult worms were obtained from purpose-bred hamsters maintained at the Universidade Federal de Minas Gerais according to a protocol approved by the Committee for Animal Experimentation of Universidade Federal de Minas Gerais (Protocol# 66/08). All animals procedures were performed under the guidelines from COBEA (Brazilian College of Animal Experimentation) and strictly followed the Brazilian law for “Procedures for Scientific Use of Animals” (11.794/2008).
For preparation of excreted-secreted (ES) antigens, worms were removed from the intestines of euthanized hamsters, washed several times in phosphate-buffered saline (PBS), and then cultured overnight in RPMI 1640 containing 100 U/ml penicillin G sodium, 100 µg/ml streptomycin sulfate, and 0.25 µg/ml amphotericin B (all reagents from Sigma-Aldrich, St. Louis, MO) at 37°C with 5% CO2 in a humidified incubator. The ES products were concentrated using microconcentration filter units with a 10-kDa-cutoff membrane (Millipore, Bedford, MA). Adult worm crude extract was prepared by direct maceration of parasites using a tissue grinder and further rupture using a cell disruptor (Sonifier Cell Disruptor, Branson Sonic Power Co., Danbury, CT, USA) in PBS for 1 min at 40 Watts, in an ice bath. The procedure was repeated five times, with 1 min intervals between disruptions.
All of the antigen preparations used in cell cultures were passed through a 0.22 µm low-protein binding syringe filter (Millipore, USA), and the resulting protein concentration was determined using a BCA protein assay kit (Pierce, USA). Hookworm antigen preparations were tested negative for endotoxin content by the Limulus lysate assay (sensitivity of 0.06 U/ml; Cambrex, USA), and stored in aliquots at −80°C.
Whole blood was collected in Vacutainer tubes containing EDTA (Becton Dickinson, USA) and 100 µL samples were mixed in tubes with 2 µL of undiluted monoclonal antibodies PerCP anti-human CD4 (clone BNI3), FITC anti-human CD25 (clone M-A251) and APC anti-human FOXP3 (clone 236A) (all from BD Pharmingen, USA). After adding the antibodies, the cells were incubated in the dark for 30 minutes at room temperature. Following incubation, erythrocytes were lysed using 2 mL of FACS Lysing Solution (BD Biosciences, USA) and washed twice with 2 mL of phosphate-buffered saline containing 0.01% sodium azide and 0.5% bovine serum albumin (SIGMA, USA). Intracellular staining was performed after cell fixation in formaldehyde (4%) and permeabilization with saponin buffer (0.5%) (Sigma, USA) for 15 minutes. Cells were washed twice with 2 mL of phosphate-buffered saline containing 0.01% sodium azide and 0.5% bovine serum albumin (SIGMA, USA) and incubated for 30 minutes with 2 µL of undiluted monoclonal antibodies PE anti-human IL-10 (clone JES3-9D7), PE anti-human TGF-β (clone TB21), PE anti-human IL-17 (clone 64CAP17), PE anti-human CTLA-4 (clone BNI3) and PE anti-human GITR (clone eBioAITR) (all from BD Pharmingen, USA).
After incubation and washing with PBS with 0.01% sodium azide and 0.5% bovine serum albumin, the cells were the fixed in 200 µL of fixative solution (10 g/L paraformaldehyde, 1% sodium-cacodylate, 6.65 g/L sodium chloride). Phenotypic analyses were performed by flow cytometry with a FACScalibur flow cytometer (BD Biosciences, USA). Data were collected on 30,000 events (gated by forward and side scatter) and analyzed using CellQuest® software (BD Biosciences, USA).
Whole blood was stimulated in vitro with ES and crude antigens (5 µg/well) in RPMI 1640 media supplemented with 1.6% L-glutamine (Sigma, USA), 3% antibiotic-antimycotic (Invitrogen, USA), 5% of heat inactivated AB+ human serum (Sigma, USA), for 24 hours at 37°C with 5% CO2. Unstimulated cultures were used as negative controls. During the last 4 hours of culture, Brefeldin A (Sigma, USA) (10 µg/mL) was added to the cultures.
Phenotypic analyses were performed by flow cytometry after staining using the same antibody panel described for ex vivo immunophenotyping assays. Data were collected on 30,000 events (gated by forward and side scatter) and analyzed using CellQuest® software (BD Biosciences, USA).
Peripheral blood mononuclear cells (PBMC) were isolated by Ficoll-Hypaque (GE Healthcare, USA) using density gradient centrifugation (Sigma, USA). Cells were then washed and resuspended at 5×106 cells/mL in RPMI 1640 medium (Invitrogen, USA), supplemented with 5% heat-inactivated human AB serum (Sigma, USA), 2 mM of L-glutamine (Sigma, USA), 50 U/mL of penicillin, and 50 g/mL of streptomycin (Invitrogen, USA).
CD4+CD25+ T cells were purified from PBMCs using the CD4+CD25+ regulatory T cell isolation kit and a QuadroMACS cell separator (both from Miltenyi Biotec, USA), according to the protocol provided by the manufacturer. In brief, cells were suspended in PBS supplemented with 2 mM EDTA and 0.5% BSA (Sigma, USA) at a density of 107 cells in 90 µL of buffer and 10 µl of biotin-Ab mixture. Cells were incubated at 4°C for 10 minutes. Then, 20 µL of anti-biotin microbeads was added and incubated for 15 min at 4°C. CD4+ cells were eluted and resuspended at a density of 107 cells in 90 µL of buffer and 10 µL of CD25 microbeads. The cells were then incubated at 4°C for 15 min and further separated by a magnetic field. CD4+CD25+ T cell fraction retained in the column was eluted by removing the column from the magnetic field and flushing out the cells with 1 mL elution buffer (PBS with 2 mM EDTA and 0.5% BSA). CD4+CD25+ T cells were washed and used immediately. The purity of CD4+CD25+ T cells after purification reached up to 95% (data not shown) (Supplementary Figure S1). CD4+CD25−/low T cells and CD4− cells together were considered as Treg-depleted PBMCs (dPBMCs) and used in functional assays.
Stimulation assays were performed in duplicates and mitogen and antigens were added at previously determined concentrations known to result in optimal proliferation [22]. The mitogen phytoemagglutinin-l (PHA–L) (Sigma, USA) was used for polyclonal stimulation of peripheral blood mononuclear cells (PBMCs). Crude and excreted-secreted (ES) antigens from adult Ancylostoma ceylanicum were employed for hookworm specific cellular stimulation at a final concentration of 5 µg/well.
For the analysis of the effect of CD4+CD25+ T cells on cellular proliferation, two different experiments were performed. Firstly, PBMCs (106 cells/mL in PBS/1% BSA) were co-labeled with 0.4 mM CFDA-SE (carboxyfluorescein diacetate succinimidyl ester, Vybrant™ CFDA-SE Cell Tracer Kit, Molecular Probes, USA) and PE-conjugated anti-human CD8 (clone UCTH-4) or PE-Cy5 anti-human CD4 (clone RPA-T4) (BD Pharmingen, USA) for 10 minutes at room temperature. The previously purified CD4+CD25+ T cells were incubated at ratio of 1∶10 with autologous CFDA-SE-labeled PBMCs pulsed with hookworm antigens, for 96 hours at 37°C with 5% CO2 atmosphere.
In a second experiment, Treg-depleted PBMCs were co-labeled with CFDA-SE/CD4 or CD8 and pulsed with hookworm antigens, as previously described. The cell proliferative response of both experiments was assessed using a FACScan® cytometer (Becton Dickinson, USA) and CellQuest® software (BD Biosciences, USA). Analysis of CFDA-SE proliferation was performed as previously described [23].
The cytokines IL-2, IFN-γ, IL-10, and IL-5 were detected and quantified in cell supernatants by commercially available sandwich ELISA kits (R&D Systems, USA). Assays were performed according to the manufacturer's instructions. Biotin-labeled detection antibodies were used, revealed with streptavidin-HRP (Amersham Biosciences, USA) and OPD substrate system (Sigma). The colorimetric reaction was read in an automated ELISA microplate reader at 492 nm. Calculations of chemokine/cytokine concentrations from mean optical density values were interpolated from the standard curve using 5-parameter curve fitting software (SOFTmax® Pro 5.3, Molecular Devices, USA). Results were achieved in pg/mL and the detection limits were as follows: 15.6 pg/mL for IL-2; 3.9 pg/mL for IFN-γ; 23.4 pg/mL for IL-10 and IL-5. Samples with values above the top of the standard curve were retested at 1/10 or 1/100 dilutions in RPMI 1640, and the chemokine/cytokine levels were recalculated.
The one-sample Kolmogorov-Smirnoff test was used to determine whether variability followed a normal distribution pattern. The Mann-Whitney U test was used to determine the differences (p value<0.05) of non-parametric variables between Necator-infected individuals and non-infected individuals. The maximum residual test (Grubb's test) was used to detect possible outliers. All statistics were carried out using Prism 5.0 for Windows (GraphPad Software Inc., USA).
Regulatory T cells were identified by flow cytometry as CD4+ T cells expressing both CD25 and FOXP3 marker (Figure 1A) and are reported as frequency (Figure 1B) and absolute numbers of cells per mm3 (Figure 1C). Analysis of PBMCs from Necator-infected individuals showed a significant increase in frequency (p<0.0001, Figure 1C) and absolute numbers (p = 0.0018, Figure 1B) of circulating CD4+CD25+FOXP3+ T cells (21.4±15.4%, 477.4±131.8 cells/mm3) when compared with non-infected naive individuals (2.3±0.6%, 78.9±11.2 cells/mm3).
Once observed the elevated number of Treg cells in the peripheral blood of hookworm infected donors, we further characterized this cell population by evaluating the expression of molecules and cytokines associated with cell modulation. Surface expression of the GITR molecule and intracellular expression of CTLA-4, IL-10, TGF-β and IL-17 cytokines, were assessed by flow cytometry. Infection of N. americanus significantly increased the proportion of cells expressing CTLA-4 (p = 0.0002) and GITR (p<0.0001). Flow cytometric analysis also showed a significant augmentation of CD4+CD25+FOXP3+ cells producing IL-10 (p<0.0001), TGF-β (p<0.0001) and IL-17 (p = 0.0003) in hookworm infected individuals (Figure 2). Similar results were found when frequency of cells were analyzed (Supplementary Figure S2).
The expression of analyzed surface and intracellular markers was determined by median intensity of fluorescence in order to obtain the absolute expression level per cell basis. Conversely to the increase in the absolute numbers of Treg subpopulations, no differences in the expression of IL-10, TGF-β, IL-17, CTLA-4 and GITR, between infected and non-infected individuals were observed (data not shown).
In order to determine the possible effect of hookworm antigens on the expression of cell surface markers (CTLA-4 and GITR) and cytokines (IL-17, TGF-β, and IL-10) by CD4+CD25+FOXP3+ regulatory T cells in N. americanus-infected donors, whole blood cultures were stimulated with either hookworm crude antigen or ES products. When crude antigen was added to the in vitro cultures, the percentage and the absolute counts of CD4+CD25+FOXP3+ regulatory T cells co-expressing IL-17 was significantly reduced (p<0.0030) (Figure 3). Although there was a tendency in increase on the expression of IL-17 in ES stimulated blood it was not statistically significant (p = 0.3562) (Figure 3). No differences in the proportion of cells expressing CTLA-4, GITR, TGF-β, IL-10, (Supplementary Figure S3) and median intensity of fluorescence were seen after hookworm antigen stimulation (p>0.05 for all).
Chronic human N. americanus infection is classically associated with a profound ablation of cell proliferation. In order to determine the possible effect of Treg cells on the immune response during hookworm infection, functional assays were designed to evaluate whether CD4+CD25+FOXP3+ Treg cells could modulate the in vitro cellular proliferation of CD4+ and CD8+ lymphocytes after parasite antigen stimulation. Hookworm antigen-stimulated PBMCs from infected individuals showed a naturally impaired proliferative response, which was not further suppressed by co-incubation with Tregs (data not shown). However, in vitro cultures, where Tregs cells were depleted (dPBMC), showed a significant increase on the CD4+ cell proliferative response induced by crude (p = 0.0039) and ES (p = 0.0012) hookworm antigenic stimulation (Figure 4A). Interestingly, depletion of Treg cells significantly augmented the proliferation of CD8+ PBMCs of infected donors in response to hookworm ES products (p = 0.0039), but not to crude antigen (Figure 4B). The depletion of Tregs elicited the increase of IL-2 and lower levels of IL-10 in supernatants of dPBMCs with or without antigenic stimulation although statistical significance was not achieved (Supplementary Figure S4). No differences were also observed for IFN-γ and IL-5 (Supplementary Figure S4) after CD4+CD25+ depletion. These results suggest that T regulatory cells do have the capacity to modulate the in vitro proliferative response in hookworm-infected patients. Additional experiments using PBMCs from control individuals demonstrated the absence of cell proliferative response after antigenic stimulation, which remains unaltered by the add-back or depletion of Tregs (Supplementary Figure S5). No differences on cell proliferative response of PBMCs from both infected and control individuals were observed in cultures after stimulation with the mitogen PHA–L.
CD4+CD25+FOXP3+ regulatory T cells (Tregs) constitute a minor subpopulation of CD4+ T-cells, which play an important role in controlling the extent of the immune-mediated pathology and maintaining immunological self-tolerance and immune homeostasis [24], [25]. These cells suppress the activation and proliferation of CD4+ and CD8+ T cells by direct contact of with effector T cells or secretion of immunoregulatory cytokines, such as IL-10 and TGF-β [16], [26]. Moreover, the balance of Treg cell–dependent immunomodulation may lead to enhanced pathogen survival and, in some cases, their long-term persistence [27]. In fact, one of the hallmarks of chronic helminth infections is induction of T-cell hyporesponsiveness and bystander suppression [28]. While the mechanisms involved in the immunomodulation by parasitic infection may be multiple, some experimental evidences have pointed toward the possible involvement of natural and inducible Treg in downregulating effector T-cell responses upon chronic infection. Over the past four decades [29], [30], several studies have attempted to describe the role of regulatory T cells in parasitic diseases (reviewed in [16], [27]), including leishmaniasis, schistosomiasis, malaria and lymphatic filariasis. However, a limited number of studies have focused on the currently known Treg dynamics and functional capacity in human helminth infections.
Evidence for Treg activity in human chronic helminth infections has been firstly provided by T cell clones generated from onchocerciasis patients [31] and recently described for geohelminth infection in humans [32]. Several studies in animal models of filariasis [33] and schistosomiasis [34], [35], demonstrated that Treg phenotype populations develop following infection, whilst in infection with the murine gastrointestinal nematode Heligmosomoides polygyrus [36], functional regulation by CD4+CD25+ T cells suppresses the bystander response to an allergic provocation. In the present study we describe the role of CD4+CD25+FOXP3+ T cells in human N. americanus infection. To explore cellular immune mechanisms underlying classic hookworm-induced T cell hyporesponsive state, we have analyzed Treg frequencies in peripheral blood and performed in vitro Treg depletion and add-back experiments with PBMC isolated from N. americanus-infected individuals from a rural area of Minas Gerais State, Brazil.
We initially showed that hookworm-infected individuals present a significant increase of circulating Treg cells in peripheral blood compared to non-infected healthy volunteers, as previously demonstrated in other nematode infections [18], [19], [37], [38], [39]. Of note, while the expansion of CD4+CD25+FOXP3+cells is observed in infected individuals, no differences were observed in the expression of all markers associated with cell suppression, including FOXP3. Such increase in the absolute number of circulating FOXP3+ Treg cells might be driven as a direct consequence of the infection. In fact, the expression of FOXP3+ in naïve T cells can be elicited by excretory-secretory products of nematodes, resulting in induced de novo FOXP3+ expression and active suppressor cells [17]. Interestingly, hookworm-infected individuals also presented with significant lower levels of circulating lymphocytes, which may be partially explained by the inhibition of effector T cells emergence during the inductive phase of the immune response in the secondary lymphoid tissues by IL-10-independent mechanisms [40]. It is possible that increased Treg activity may trigger modulation of host immune response and consequently facilitate hookworm prolonged survival. On the other hand, increased Treg responses might also account for limitation of exacerbated infection-induced tissue pathology, which would be ultimately beneficial to the host. However, it would not limit the blood loss or anemia induced by the infection.
A variety of potential mediators of Treg activity that could contribute to the suppression of the host's immune response have been identified, including GITR [41], [42], CTLA-4 [43], FOXP3 [44], [45], and the anti-inflammatory cytokines IL-10 and TGF-β [46], [47], [48]. In the current study, a significant increase of circulating CD4+CD25+FOXP3+ lymphocytes, co-expressing GITR, CTLA-4, IL-10, TGF-β or IL-17, was demonstrated in N. americanus-infected donors, compared to non-exposed volunteers. Nevertheless, a higher expression of these markers on per cell basis has not been observed in hookworm-infected donors in relation to healthy individuals.
Both GITR and CTLA-4 molecules are constitutively expressed on cell surface of natural Tregs [27] and are regulated by FOXP3 expression [49], [50]. Initial studies related to the effects of GITR signaling on Treg cells indicated that interaction of this receptor with anti-GITR antibody or GITR ligand (GITRL) lead to an apparent abrogation of suppressive activity of Tregs [41], [42], [51]. Indeed, treatment of Trichuris muris infected mice with anti-GITR resulted in an earlier worm expulsion [19]. Although not essential for the T cell suppressor activity [50], the engagement of GITR promotes proliferation of Tregs [50], [52] and potential enhancement of their suppressive function [51]. The significant increase of circulating CD4+CD25+FOXP3+GITR+ and CD4+CD25+FOXP3+CTLA-4+ in Necator-infected donors might partially reflect the concomitant augmentation of Tregs. Similarly, mice experimentally infected H. polygyrus or Litomosoides sigmodontis also presented a prominent increase of GITR and CTLA-4 expression [39], [53]. The inhibitory receptor CTLA-4 presents partial homology to CD28 molecule and interacts to the same ligands, CD80 and CD86, with a much higher affinity [54]. The suppressive effect of CTLA-4 is associated with the reduced IL-2 production and IL-2 receptor expression, and by arresting T cells at the G1 phase of the cell cycle [55], [56]. Moreover, CTLA-4 expressing Treg cells induce the expression of the enzyme indoleamine 2,3-dioxygenase (IDO) by antigen-presenting cells which degrades tryptophan, and the lack of this essential amino acid inhibits T cell activation and promotes T cell apoptosis [57]. In helminth infections, such as lymphatic filariasis, the expansion of CTLA-4+ T cell populations in was associated with suppressed T cell function [58]. The increased number of circulating GITR+ and CTLA-4+ Treg cells in hookworm-infected individuals suggests that these cells might play a suppressive role on host immune regulation.
Although it has long been recognized that IL-10-producing T cells could be generated in vivo during parasitic infection [53], it is only recently that the concept has emerged that specialized subsets of regulatory T cells contribute to this regulatory network [16]. In the current study, hookworm-infected individuals presented a significant augmentation of CD4+CD25+FOXP3+cells producing the anti-inflammatory cytokines IL-10 and TGF-β. In fact, a prominent rise in IL-10 secretion was demonstrated in ES antigen-stimulated PBMC cultures during primary experimental and natural human hookworm infection [22], [59]. It is well known that IL-10 and TGF-β are naturally produced by Tregs [27] and are required to induce FOXP3 expression [60]. Although it is not clear whether or how precisely hookworm infection influences the production these anti-inflammatory cytokines, lower levels of IL-10 were observed in supernatants of cultures after Treg depletion, which might support the possibility that that N. americanus induced-Tregs contribute as an important source of their production.
In this study, a significant increase of circulating CD4+CD25+FOXP3+cells producing the IL-17 was observed in infected donors, corroborating previous studies on other parasitic diseases [15], [17], [60]. Noteworthy, a recent study demonstrated that human hookworm products are able to influence the pro-inflammatory Th17 pathway, promoting a significant decrease in IL-17 production in the mouse infection model [61]. Moreover, it has been suggested that worm infection could block mucosal IL-23 and IL-17 secretion, leading to an important mechanism of control of inflammatory responses [62]. In fact, Ruyssers et al. also observed that helminth antigens could reduce the expression of IL-17 in both colon and mesenteric lymph node T cells [63]. Interestingly, a significant reduction in CD4+CD25+FOXP3+IL-17+ T cells after hookworm crude antigen stimulation was also demonstrated, suggesting the possible secretion of this cytokine or downmodulation of IL-17 expression after cell restimulation. Indeed, Elliott et al. showed that mouse colonization with the helminth H. polygyrus reduces IL-17A mRNA expression by mesenteric lymph node (MLN) cells and inhibits IL-17 production by cultured lamina propria mononuclear cells and MLN cells [64]. Moreover, the co-expression of FOXP3 and IL-17 may indicate the transient status of CD4+ lymphocytes from hookworm-infected individuals between Treg (FOXP3) and Th17 (IL-17) profiles, where the development of either pathway of differentiation is driven by TGF-β and IL-6 [65]. The decrease of IL-17 after antigenic stimulation might imply the differentiation of these transient cells in truly effector Tregs. Nonetheless, our results suggest that hookworm products are able to induce an immunomodulated microenvironment at the site of infection. Although recent evidences demonstrate the local suppression of Th1 and Th17 inflammatory cytokines during hookworm infection [66], the role of IL-17 in hookworm-induced Tregs still remains to be addressed.
Based in our results, we demonstrated that hookworm infected individuals present a significant augmentation of activated Treg cells in the peripheral blood, observed by increased numbers of Treg subpopulations expressing cell surface molecules and mediators associated with suppression of immune responses. These cells might contribute to the suppressive effect of this parasitic infection, leading to reduction of antigen-specific proliferative responses previously demonstrated in human populations and animal models [13], [67]. Indeed, chronic human N. americanus infection has classically been associated with a profound ablation of cell proliferation, which may even extend to other infectious agents and mitogens (“bystander effect”) [22], [68]. In the present study, we have demonstrated by functional assays that both T CD4+ and CD8+ cell proliferative responses to either ES or hookworm crude antigen were increased after Treg depletion, followed by an increase of IL-2 secretion and lower levels of IL-10, although not statistically significant, further implying that these cells may exert a specific immunomodulatory effect during persistent hookworm infection. Recently, Cuéllar et al. showed that coincubation of mouse splenic T cells with dendritic cells pulsed with the hookworm antigen Ac-TMP-1 induced their differentiation into CD4+ and, particularly, CD8+CD25+FOXP3+ T cells that expressed IL-10 [69]. These cells were able to suppress proliferation of naive and activated CD4+ T cells by TGF-β-dependent (CD4+ suppressors) or independent (CD8+ suppressors) mechanisms. However, while we showed that the expansion and suppressive effects of Tregs are prominent during chronic hookworm infection, no changes in the number of T cells nor in the absolute counts of regulatory T cells were observed in the human primo-infection with N. americanus [70]. Nonetheless, the data in the present work suggest that hookworms exploit Treg cells to facilitate its own survival by dampening host immune response. In conclusion, hookworm-induced Treg activity may be able to control and divert selective proliferative and cytokine responses to numerous disorders, such as intestinal inflammation, airways inflammation/hyper-reactivity, diabetes, and multiple sclerosis. While it is known that ES products from nematodes may stimulate T reg cells [17], only few studies have demonstrate the immunomodulatory properties of hookworm-derived antigens [38], [71]. Considering the recent availability of transcriptomic data sets for hookworm species [72], [73], further studies are still required to identify specific antigens directly associated with host's immune suppression, leading to the understanding of mechanisms used by the parasite to skew the immune response in its favour and the possible discovery of several promising candidate vaccine antigens.
While our results shed light on the patent mechanisms of immunosuppression present in hookworm infection, it is important to mention that the sample size and age variation of our studied population might be considered as possible limiting factors of our study. Moreover, although effort was made to match the nutritional status of all participants (endemic and non-endemic areas), unfortunately perfect matching of control individuals and infected patients by age and sex was not always possible. The absence of negative individuals from endemic areas as controls was preferred once it is not possible to guarantee the absence of infection in these donors (due to the limited sensitivity of fecal exams or long pre-patency period). Moreover, it has been previously shown that the immunological status of helminth-infected patients remain unaltered after anthelmintic treatment for several months [71]. Nonetheless, this study was designed to minimize potential confounders, which could mask the immunological assessment of hookworm infection. Therefore, our data should be further validated by large immunoepidemiological surveys to be conducted in endemic areas.
|
10.1371/journal.pntd.0002275 | A Hamster-Derived West Nile Virus Isolate Induces Persistent Renal Infection in Mice | West Nile virus (WNV) can persist long term in the brain and kidney tissues of humans, non-human primates, and hamsters. In this study, mice were infected with WNV strain H8912, previously cultured from the urine of a persistently infected hamster, to determine its pathogenesis in a murine host.
We found that WNV H8912 was highly attenuated for neuroinvasiveness in mice. Following a systemic infection, viral RNA could be detected quickly in blood and spleen and much later in kidneys. WNV H8912 induced constitutive IL-10 production, upregulation of IFN-β and IL-1β expression, and a specific IgM response on day 10 post-infection. WNV H8912 persisted preferentially in kidneys with mild renal inflammation, and less frequently in spleen for up to 2.5 months post infection. This was concurrent with detectable serum WNV-specific IgM and IgG production. There were also significantly fewer WNV- specific T cells and lower inflammatory responses in kidneys than in spleen. Previous studies have shown that systemic wild-type WNV NY99 infection induced virus persistence preferentially in spleen than in mouse kidneys. Here, we noted that splenocytes of WNV H8912-infected mice produced significantly less IL-10 than those of WNV NY99-infected mice. Finally, WNV H8912 was also attenuated in neurovirulence. Following intracranial inoculation, WNV persisted in the brain at a low frequency, concurrent with neither inflammatory responses nor neuronal damage in the brain.
WNV H8912 is highly attenuated in both neuroinvasiveness and neurovirulence in mice. It induces a low and delayed anti-viral response in mice and preferentially persists in the kidneys.
| West Nile virus (WNV) has been reported to persist long term in the brain and kidney tissues of humans, non-human primates, and hamsters. To define a murine model of persistent WNV renal infection, we characterized infection by WNV H8912, an isolate cultured previously from the urine of a persistently infected hamster. Our findings indicate that WNV strain H8912 is highly attenuated in both neuroinvasiveness and neurovirulence for mice. The virus persisted preferentially in kidneys of the mouse, and less frequently in the spleen and the brain. Moreover, mice infected with WNV H8912 had a delayed induction of IFN- β and IL-1β expression and WNV- specific IgM response, but a constitutive production of serum IL-10. There was a lower proinflammatory response in mouse kidneys when compared to equivalent findings in the spleen. This response may lead to a reduced T cell response in kidneys, which could ultimately contribute to renal-specific WNV persistence. Defining a murine model of WNV persistence by using a well-characterized, hamster-derived WNV urine isolate should provide important insights into understanding the mechanisms of WNV persistence.
| West Nile virus (WNV) is a mosquito-borne flavivirus with a positive-sense, single-stranded RNA genome that encodes three structural proteins: the nucleocapsid protein (C), membrane and envelope (E), and seven nonstructural (NS) proteins [1]–[2]. Human infection results from mosquito bites, blood transfusion, organ transplantation, breast feeding and in utero or occupational exposure [3]–[6]. About 80% of human infections with WNV are asymptomatic. Among persons with clinical illness, the features of acute illness range from WN fever, to neuroinvasive conditions, including meningitis, encephalitis, acute flaccid paralysis and death [7]–[8]. There is no specific therapeutic agent for treatment of WNV infection, and an approved vaccine for its prevention in humans is not currently available. About 20–50% of WNV convalescent patients have significant long-term morbidity years after their acute illness; symptoms include muscle weakness and pain, fatigue, memory loss, and ataxia [9]–[13]. Although the cause of the persistent sequelae remains undefined, accumulating evidence suggests that persistence of the virus and chronic infection may play a role. Some WNV convalescent patients have been reported to have detectable serum or cerebrospinal fluid (CSF) WNV-specific IgM and - IgA years after the acute infection, which is suggestive of the existence of viral antigens in the periphery or the central nervous system (CNS) of these individuals 14–16. Indeed, WNV antigen or RNA has been detected in the brain or urine of WNV patients ranging from a few months to several years after the initial acute illness [17]–[18].
Persistent WNV infection has also been reported in non-human primates, hamsters and mice [19]–[22]. The first well-documented WNV persistence was reported by Pogodina [20] in non-human primates in 1983, in which infectious virus was mostly detected in the CNS tissues and kidneys for up to 5 ½ months. Experimental infection of hamsters with the WNV NY99 strain induced chronic renal infection and persistent viruria for up to 8 months post- infection, accompanied by moderate renal histopathologic changes [22]–[23]. Siddharthan et al. [21] also demonstrated an active CNS infection and chronic neuropathological lesions in hamsters for up to 100 days after WNV infection. In comparison, infectious virus was cleared from hamster blood and spleen within the first 2 weeks of inoculation [23]–[24]. Following systemic infection of mice with the wild-type WNV, virus persisted preferentially in skin, spinal cord, brain and lymphoid tissues, but the persistence was found less frequently in kidneys [19]. In a recent report, we have demonstrated that infection by a strain of WNV H8912, cultured from urine of a persistently infected hamster, induced a differential proinflammatory cytokine response in mouse macrophage and kidney epithelial cell lines [25]. In the present study, we attempted to further define a murine model of persistent WNV renal infection, using the hamster-derived WNV urine isolate H8912.
6- to 10-week-old C57BL/6 (B6) mice were purchased from Jackson Laboratory (Bar Harbor, ME). Groups were age- and sex-matched for each experiment and housed under identical conditions. This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animal experiments were carried out under a protocol approved by the Institutional Animal Care and Use Committee at the University of Texas Medical Branch.
Two WNV strains were used: the parental wild-type strain WNV NY385-99 (WNV NY99, [26]), which had been passaged once in African green monkey kidney (Vero) cells and twice in Aedes albopictus (C6/36) cells, and WNV H8912, which was recovered from hamster urine 274 days post-infection after three consecutive passages of a urine isolate from a persistently infected hamster [27]. To evaluate its neuroinvasiveness, mice were inoculated intraperitoneally (i.p.) with 10-fold serial dilutions of WNV H8912, ranging from 102 to 106 PFU. For intracranial (i.c.) infection, anesthetized mice were inoculated with 102, 104, and 106 PFU of WNV H8912 in 50 µl of PBS with 5% gelatin. Infected mice were monitored twice daily for morbidity, including lethargy, anorexia and ataxia.
Spleen, kidney and brain tissues were harvested from the WNV-infected mice or controls following perfusion with PBS. For the WNV persistence study, tissues were homogenized and re-suspended in modified Eagle's medium (MEM, Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS, Sigma, St. Louis, MO). Tissue homogenate was then inoculated into flask cultures of Vero cells three consecutive times (original passage, followed by two blind serial passages harvested at weekly intervals). At the final passage, infected cells were harvested and assayed for WNV RNA by Q-PCR analysis. Non-infected Vero cell samples were used as negative controls. Cells or tissues were resuspended in Trizol (Invitrogen) for RNA extraction. 200 ng RNA was used to synthesize cDNA by reverse transcription (qScript, Quanta Biosciences). The sequences of the primer sets for WNV envelope (WNVE), cytokine genes, and genes for leukocytes, such as CD4+ and CD8+ T cells, neutrophils and monocytes, and PCR reaction conditions were described previously [28]–[31]. The assays were performed in an iCycler (Bio-Rad, Hercules, CA). To normalize the samples, the same amount of cDNA was used in a Q-PCR for β-actin. The ratio of the amount of amplified gene compared with the amount of β-actin cDNA represented the relative levels in each sample. For Il-10, Tgfβ and leukocyte genes, results were calculated based on Ct values by using the formula 2∧ -[Ct(cytokine or leukocyte gene)-Ct(β-actin)] as described in the user manual (SA Biosciences). In the persistence study, samples were considered positive for WNV RNA if the WNVE gene expression level was significantly higher than the negative controls.
Culture supernatant or sera of infected mice were collected for analysis of cytokine production by using a Bio-Plex Pro Mouse Cytokine Assay (Biorad).
Microtiter plates were coated with recombinant WNV-E protein expressed in Drosophila melanogaster S2 cells [32] overnight at 4°C at 100 ng/well in coating buffer [0.015 M Na2CO3, 0.03 M NaHCO3, and 0.003 M NaN3 (pH 9.6)]. Sera from infected mice were diluted 1/40 in PBS with 2% BSA, added to the duplicate wells, and incubated for 1 h at room temperature. Plates were washed with PBS-Tween (PBS-T). Alkaline phosphatase-conjugated goat anti-mouse IgG or IgM (Sigma) at a dilution of 1/1000 in PBS-T was added for 1 h at room temperature. After washing with PBS-T, color was developed with p-nitrophenyl phosphate (Sigma) for 10 min and intensity determined at an absorbance of 405 nm by using a spectrophotometer.
Isolation of leukocytes from kidneys was performed by methods described earlier [33]–[34]. Briefly, mice were perfused with PBS, then the kidneys were removed and tissues were mashed through a 70-µm-pore-size strainer in RPMI with 5% FBS. The resulting cell suspensions were centrifuged, and the pellet was resuspended in a 36% Percoll solution (Sigma) and gently overlaid onto 72% Percoll in RPMI, and centrifuged at 1000× g for 30 min at room temperature with no brakes. Cells were harvested from the Percoll interface and washed extensively in RPMI with 5% FBS.
Freshly isolated splenocytes and kidney leukocytes were stained with PE-conjugated Pro5 MHC pentamers (H-2Db SSVWNATTA and H-2Kb RSYCYLAT, Proimmune, Oxford, UK) for 10 min at 22°C. WNV- infected BM-DCs were harvested at 48 h post-infection and were stained with antibodies for cell surface markers, including CD80, CD86, MHC II (e-Biosciences). After staining, cells were washed and fixed with 0.5% paraformaldehyde in PBS and acquired by using a C6 Flow Cytometer (Accuri cytometers, Ann Arbor, MI). To measure intracellular cytokine production, splenocytes from WNV-infected mice or controls were isolated and stimulated with WNV-specific NS3 and E peptides (RRWCFDGPRTNTILE and PVGRLVTVNPFVSVA, respectively [35]) for CD4 T cells, or WNV-specific NS4B and E peptides (SSVWNATTA and IALTFLAV, respectively [36]–[37]) for CD8 T cells for 5 h at 37°C. Golgi-plug (BD Biosciences) was added at the beginning of stimulation. Cells were harvested, stained with antibodies for CD4 or CD8, fixed in 2% paraformaldehyde, and permeabilized with 0.5% saponin before adding PE-conjugated anti-IFN-γ, or control PE-conjugated rat IgG1 (eBiosciences, San Diego, CA).
Anesthetized mice were perfused with 30 ml of ice cold PBS. Brains and kidneys were removed and fixed in 4% paraformaldehyde. Subsequently, specimens were transferred to 70% ethanol and processed for histopathologic examination. Five-micron paraffin sections were prepared for staining with hematoxylin & eosin, in the case of kidneys, six sections were made for each mouse. Stained sections were examined for lesions by a pathologist, who was blinded to the origin of the samples. As a lesion was defined, its incidence was scored based on presence or absence in a single section, thus allowing for determination of % incidence in the histologic material.
Bone-marrow (BM)- derived dendritic cells (DCs) were generated as described previously [38]. Briefly, BM cells from B6 mice were isolated and cultured for 6 days in RPMI-1640 supplemented with granulocyte-macrophage-colony stimulating factor, and interleukin-4 (Peprotech) to generate myeloid DC. Day 6-cultured DCs were infected with WNV NY99 or WNV H8912 strain at a multiplicity of infection (MOI) of 0.2. Primary cortical cultures from mouse embryos of either sex were prepared as described previously [39]. Briefly, cortices were dissected from B6 mouse embryos (E18) and washed in Hanks' balanced salt solution (HBSS, Invitrogen). Meninges and excess white matter were removed. Cortices were chopped into small tissue blocks (0.5–1 mm3) and transferred to a sterile tube containing HBSS. Cortical tissues were treated with 0.25% trypsin (Sigma) for 8 min at 37°C, and dissociated to single cell suspensions, followed by centrifugation at 120× g for 5 min. The cell pellets were re-suspended in Dulbecco's Modified Eagle Media (DMEM, Invitrogen) containing 10% FBS and 1% penicillin-streptomycin. Cells were inoculated into culture plates coated with 20 µg/ml poly-D-lysine (PDL, Sigma) in 0.1 M borate buffer (PH 8.5, 50 mM H3BO3, 12.5 mM Na2B4O7(10H2O)). Two hours later, DMEM was replaced with Neurobasal Medium (Invitrogen) supplemented with 2% B27 (Invitrogen), 1% penicillin-streptomycin, and 0.5 mM L-glutamine. Cells were cultured at 37°C and one-third of the medium was replenished every 3 days until 10 days in vitro. For WNV infection, 0.3×106 cells were grown on the coverslips in a 24-well tissue culture plate and infected with WNV NY99 or WNV H8912 at a MOI of 0.003. At indicated time points post-infection, supernatants were collected and measured for viral titer by plaque assay. Cells were also harvested and resuspended in Trizol (Invitrogen) for RNA extraction. Viral load was determined by Q-PCR assay.
Vero cells were seeded in 6-well plates in DMEM (Invitrogen) supplemented with 10% FBS 24 h before infection. Serial dilutions of cell culture supernatants of both NY99 and H8912 infected cells were added and incubated for 1 h. Subsequently, DMEM containing 2% FBS and 1% low-melting-point agarose were added and the plates were incubated for 3 days. A second overlay of 4 ml 1% agarose-medium containing 0.01% neutral red (Sigma) was added to visualize plaques. Virus concentrations were determined as PFU/ml.
Data analysis was performed by using Prism software (Graph-Pad) statistical analysis. Values for viral burden, plaque assay, and cytokine production experiments were presented as means ± SEM. P values of these experiments were calculated with a non-paired Student's t test. Statistical significance was accepted at P<0.05.
WNV H8912 was originally isolated from urine of a WNV persistently infected hamster. It has been shown to be highly attenuated in neuroinvasiveness and to induce chronic renal infection in hamsters [27]. To characterize WNV H8912 infection in a murine model, we infected B6 mice i.p. with 10-fold serial dilutions of 102 to 106 PFU of WNV H8912. Infected mice were monitored twice daily for morbidity for over a month. All mice survived infection (Fig. 1A); and less than 20% of the mice infected with the highest dose (106 PFU) of WNV H8912 developed mild disease symptoms (ruffeled fur, irritability, etc). We also measured viral load in blood, spleen and kidneys in B6 mice following an i.p. inoculation of 500 PFU of WNV H8912. At days 1, 3, 6 and 10 post-infection, tissues were collected, and viral load was measured by a Q-PCR assay. As shown in Fig. 1B, viremia increased quickly by day 1 and declined on day 6 post-infection compared to equivalent findings in naïve mice (P<0.05). Viral load in the spleen was also increased on day 1 and continued to be detectable on day 10 post-infection (Fig. 1C, P<0.05 or P<0.01). In contrast, viral load in kidneys was not detectable until on day 10 post-infection (Fig. 1D, P<0.05). These results indicate that WNV H8912 has a highly reduced neuroinvasiveness in mice, and that its replication in blood, spleen and kidneys showed differential kinetics.
Innate cytokine responses, including type 1 interferon (IFN)s, proinflammatory cytokines and regulatory cytokines, play an important role in protection and pathogenesis during wild-type WNV NY99 infection [29], [40]–[42]. To examine cytokine production, we infected mice i.p. with 500 PFU of WNV H8912. Blood was collected at days 1, 3, 6 and 10 post-infection for cytokine measurement by Q-PCR and Bioplex. Compared to the samples from non-infected mice, IFN-α gene expression was not changed at any of the time points examined (Fig. 2A, P>0.05). There was a 3–4 fold induction in IFN-β gene expression on day 10 post infection (Fig. 2B, P<0.01). Higher interleukin (IL)-1β production was also detected in serum on day 10 post infection (Fig. 2C, P<0.01). No changes were noted in the expression of the two genes at earlier time points nor in the production of other proinflammatory cytokines, including IL-6 and tumor necrosis factor (TNF)-α (Figs. 2B–2E, P>0.05). We found that WNV H8912 constitutively induced serum IL-10 levels compared to those in naïve mice (Fig. 2F, P<0.01 or P<0.05). B cell-mediated humoral immune responses are critical for host defense against disseminated infection by WNV [43]–[45]. In particular, induction of a specific, neutralizing IgM response early in the infection limits viremia and dissemination into the CNS and protects the host against lethal wild-type WNV infection [46]. Interestingly, WNV H8912-infected mice had no detectable serum WNV-specific IgM until at day 10 post-infection (Fig. 2G, P<0.01). In comparison, there was a low WNV-specific IgG production at days 1 to 6 post-infection and the titer was further increased at day 10 (Fig. 2H, P<0.01). Together, these results suggest to us that WNV H8912- infected mice had a delayed induction of IFN-β and IL-1β expression and IgM response, but a constitutive production of serum IL-10.
In a previous report, the hamster-passaged WNV urine isolates, including WNV H8912 were shown to induce persistent renal infection in hamsters following a systemic infection [27]. Here, we wondered whether WNV H8912 would induce persistent infection in mice. At days 30, 60 and 84 post-infection, spleens and kidneys were harvested from infected mice, and the tissue homogenate was inoculated into Vero cells consecutively for three times before detection of WNV RNA by Q- PCR assay. WNV RNA was equally detected in both tissues at day 30 but at a higher frequency in kidneys than in spleen tissues at days 60 and 84 post-infection (Table 1). Furthermore, in mice at day 84 post-infection, pathologic examination revealed a sporadic, focal chronic inflammation consisting of clusters of lymphocytes in the intertubular interstitium of all four kidney samples of experimental mice that were detected as positive for WNV RNA (Fig. 3A, right panel). The incidence of the inflammatory lesion varied from 17% to 83% (number of sections with lesion present per 6 total sections/mouse) with a mean observance of 46%±14%. No such inflammatory lesions were observed in equivalent numbers of kidney sections of control mice (Fig. 3A, left panel). Interestingly, we noted that there was a detectable serum WNV-specific IgM response at all these time points, though it dropped at days 30 and 84 post-infection compared to that of day 10 (Fig. 3B, P<0.01). In comparison, WNV-specific IgG production was increased at days 30 and 60 post infection compared to that at day 10 (Fig. 3C, P<0.01). Both WNV-E- and NS4B-specific CD8+ T cells dominated during wild-type WNV NY99 strain infection [36]–[37]. In WNV H8912-infected mice, we found that the number of these two T cell populations in the splenic tissues were either maintained or increased at day 34 post-infection compared to those on day 8 post-infection (Figs. 3D & 3E), whereas the number of both T cell populations was significantly lower in kidney tissues and was even reduced at the late stage of infection (day 34) compared to those at day 8 (Figs. 3D & 3E). To understand the underlying mechanism of WNV persistence, we measured and compared the expression of innate cytokines in both tissues. Our results showed that IL-6 expression levels in the spleen were induced during the early stage of infection (days 1 to 10 post-infection), but were reduced at days 30 and 60. In comparison, no induction of IL-6 expression was observed in the kidneys of WNV H8912-infected mice (Fig. S1A). The TNF-α expression level was slightly increased in spleens throughout the course of infection, whereas it was decreased in kidneys of H8912-infected mice except at day 6 post-infection (Fig. S1B). Further, we did not note any significant differences in the expression of regulatory cytokines, such as TGF-β and IL-10, in these two tissues during WNV H8912 infection (Figs. S1C & S1D). These results suggest that WNV H8912 induces a persistent infection preferentially in mouse kidneys rather than in the spleen, possibly due to the differential proinflammatory cytokine expression and T cell responses in these two tissues.
A previous report shows that systemic wild-type WNV NY99 infection induced virus persistence preferentially in the spleen than in mouse kidneys [19]. We next compared splenic CD4+ and CD8+ T cell responses following WNV H8912 or NY99 infection. We noted CD4+ and CD8+ T cells of WNV H8912 or NY99-infected mice produced similar levels of IFN-γ upon ex vivo stimulation with WNV peptides on days 8 and 38 post-infection (Fig. 4A). Nevertheless, splenocytes of WNV H8912-infected mice induced significantly less IL-10 production compared to those of WNV NY99-infected mice upon ex vivo stimulation with PMA and ionomycin (Fig. 4B, P<0.01). DCs represent the most important antigen- presenting cells exhibiting the unique capacity to initiate primary T cell responses and are permissive to WNV infection. At day 2 post-infection, we noted 18–19% increase on the percentage of CD80 and CD86 expression on WNV H8912-infected BMDCs when compared to WNV NY99-infected cells respectively (Figs. 4C & 4D, P<0.01). Thus, a decreased IL-10 production by WNV H8912-infected splenocytes may contribute to the lower frequency of virus persistence in spleen compared to that of WNV NY99-infected mice.
To determine the neurovirulence of WNV H8912, we infected mice i.c. with 102, 104 and 106 PFU of WNV H8912. Infected mice were monitored twice daily for morbidity and mortality for over a month. All mice inoculated with 104 and 106 PFU of WNV H8912 infection died; while 41% of mice inoculated i.c. with 102 PFU of WNV H8912 survived (Fig. 5A). To further study its neurovirulence, we infected mouse primary cortical neuron/glia cultures with WNV H8912 and its wild-type control WNV NY99. At days 1 and 3 post-infection, the viral load in WNV H8912-infected cells was about 75%–80% lower than that in WNV NY99-infected cells as measured by Q-PCR analysis (Fig. 5B, P<0.01). Plaque assay results also showed a 94% decrease in viral titers of WNV H8912- infected cells compared to that of WNV NY99-infected (Fig. 5C, P<0.05) on day 1 post-infection. We next studied WNV persistence in the brains of mice that survived the 102 PFU of WNV H8912 i. c. infection. At day 45 post-infection, WNV RNA was only detected by Q-PCR assay in one of the five surviving mice (20%, data not shown). Among the surviving mice, 60% had a detectable serum WNV-specific IgM response; whereas 80% of them were positive for WNV-specific IgG (Figs. 6A and 6B). To determine if WNV persistence in the brain was accompanied by inflammatory responses, we studied brain pathology. On day 45 post-infection, we did not detect significant histologic evidence of inflammation or neuronal damage in the cortex and olfactory bulb regions of the brain of surviving mice, when compared to equivalent tissues in non-infected controls (Figs. S2A & S2B). In further phenotype analyses of the brain leukocytes by Q-PCR assay, no significant infiltration of either CD4+ T cells (Fig. 6C), monocytes (CD11b+, Fig. 6E), or neutrophils (Ly6G+, Fig. 6F) was observed in the brains of the surviving mice. Nevertheless, we noted one of the surviving mice- which was also positive for viral RNA, had more CD8+ T cell infiltrates in the brain (Fig. 6D) at day 45 post-infection than non-infected controls. Collectively, these results suggest that WNV H8912 is highly reduced in neurovirulence. It appears to induce WNV persistence much less frequently in the brain than in the kidneys with no histopathologic changes in the CNS.
WNV H8912, a viral isolate recovered from urine of persistently infected hamster, was attenuated in neuroinvasiveness but capable of producing chronic renal infection in hamsters [27]. In this study, we found that WNV H8912 was also highly attenuated in neuroinvasiveness and neurovirulence for mice. Following a systemic infection with WNV H8912, viral RNA was detected more frequently in kidneys than in spleen by a Q-PCR assay. This was concurrent with a focal chronic inflammation response in the inter-tubular interstitium of kidney samples found to be positive for WNV RNA on day 84-post-infection. In comparison, WNV persisted at a lower frequency in the brain following i.c. inoculation with no significant changes in the CNS. In summary, our results suggest that WNV H8912 was highly attenuated and induced a tissue-specific, persistent renal infection in mice.
B cell-mediated humoral immune responses are critical in host defense against disseminated infection by WNV [43]–[45]. Here, we found that WNV H8912 induced a delayed serum WNV-specific IgM and a low WNV-specific IgG production in the early stages of infection, which may contribute to its persistence in the peripheral tissues. Moreover, the WNV-specific IgM response remained detectable in the later stages of infection and decreased only slightly at days 30 and 84 post-infection, compared to that on day 10. WNV-specific IgG production was enhanced at days 30 and 60 post-infection, compared to that in the early stage of infection. The detection of WNV-specific IgM and IgA in sera of WNV convalescent patients years after having the acute infection has been reported and appears to be correlated with the persistence of viral RNA in the kidneys, urine or the CNS of these patients [14]–[16]. Thus, these results further support that WNV H8912 induces persistent infection in mouse peripheral tissues. T cells are important for host survival following wild-type WNV infection and contribute to a long-lasting protective immunity [47]–[48]. Elevated levels of IL-10 during chronic viral infections are known to contribute to diminished T cell activity and the failure to control viral infection [49]–[50]. Interestingly, sera IL-10 levels in WNV H8912-infected mice were constitutively high within the first two weeks of infection. There were also delayed anti-viral responses of IFNs and IL-1β in these mice. Collectively, these factors may potentiate persistent viral infection in mouse peripheral tissues following systemic WNV H8912 infection.
Systemic wild-type WNV NY99 infection induced virus persistence more frequently in spleen than in mouse kidneys [19]. In this study, we found WNV H8912 induced virus persistence preferentially in kidneys following an i.p. infection. Multiple virus and host determinants could contribute to a tissue –specific WNV persistence in mice. First, our recent work showed that WNV H8912 had a significantly reduced replication rate in mouse kidney epithelial cells compared to wild-type WNV strain [25], which indicate a greater tissue tropism is not involved in WNV H8912-induced renal-specific persistence. Differential immune responses have been reported to contribute to tissue-specific viral persistence. For example, a Treg-mediated suppression of CD8+ T cell response occurred specifically in the spleen but not in the liver during chronic Friend virus infection. This resulted in a 10-fold less viral load in the latter tissue [51]. Here, we next measured proinflammatory cytokine responses in spleen and kidney tissues and found there were lower levels of IL-6 and TNF-α expression in the kidneys of WNV H8912-infected mice compared to spleen tissues. This is consistent with our recent findings, in which WNV H8912 infection in mouse kidney epithelial cells did not induce IL-6 and TNF-α expression [25]. DC maturation is an innate response that leads to adaptive immunity to foreign antigens [52]–[53]. Proinflammatory cytokines are known to promote this process [54]–[55]. We found significantly less WNV-specific CD8+ T cells in WNV persistently infected kidneys than in the spleen tissues throughout the infection. Thus, a lower inflammatory response in the kidneys may lead to reduced WNV specific T cell infiltrates and ultimately to a preferential WNV persistence in mouse kidneys. Lastly, our results also showed that splenocytes of WNV H8912-infected mice produced much less IL-10 than those of wild-type strain-infected mice. There were higher levels of expression of DC maturation markers on WNV H8912-infected BMDCs than wild-type WNV –infected cells. In conclusion, differential proinflammatory cytokine and IL-10 responses in spleen and kidney tissues lead to tissue-specific virus persistence during systemic WNV infection. WNV H8912 is highly attenuated in mice. It induced a low frequency of virus persistence in the CNS following i.c. inoculation, concurrent with no significant changes in the cortex and olfactory bulb regions of the brain of surviving mice. Neurons and microglias are the two major cell types permissive to WNV infection [29], [56]. Our results showed that WNV H8912 had a significantly lower titer in mouse primary cortical neuron/glia cultures compared to wild-type WNV strain, which suggests WNV H8912 does not have a tissue tropism in the CNS. While the underlying mechanisms by which WNV H8912-induced a low frequency of viral persistence are under investigation, these results further support that WNV H8912 induces a tissue-specific renal persistence.
WNV encephalitis (neuroinvasive disease) has been a serious public health concern in North America for more than a decade. Neither treatment nor human vaccines are available. In recent years, some WNV convalescent patients have been reported to have persistent sequelae, which occurred 6 to 12 months after the acute infection [9], [11]. Murray et al. [18] detected the presence of WNV RNA in the urine of patients convalescent from WNV neuroinvasive disease for up to 1–7 years after their initial infection. Among these WNV RNA-positive patients, 80% reported chronic neurological symptoms, while 20% had renal failure [18]. These observations raise the possibility that persistent infection is associated with WNV-induced chronic diseases. Persistent WNV infection was detected in the hamster kidneys and brain, accompanied by neurological sequelae [21]–[23]; this association is similar to the clinical findings in some WNV convalescent patients having long-term morbidity. Compared to the parent wild-type NY99 strain, full genome sequencing reveals conserved genetic mutations in the coding and non-coding regions of the viral genome of all WNV isolates recovered from urine of persistently infected hamsters [27], including WNV H8912, indicating an association of genetic mutations with WNV persistence renal tropism. Thus, hamster model would likely serve as a suitable in vivo model to further investigate the underlying immune mechanisms of WNV persistence. Nevertheless, there is an obstacle to fully investigating immunity to chronic WNV infection in the hamster model, due to limited availability of hamster reagents. Mice are easier to work with, amenable to immunological manipulation and are relatively inexpensive. Systemic wild-type WNV NY99 infection in mice induced virus persistence preferentially in spleen tissues than in kidneys [19]. Our study has now shown that the hamster-derived WNV urine isolate H8912 induced persistent infection in mouse kidneys, accompanied by focal renal histological inflammatory changes for up to a few months post infection, which is similar to the observations in WNV convalescent patients having long-term morbidity. Thus, defining a murine model of WNV persistence by using a well-characterized, hamster-derived WNV urine isolate should provide important insights into the mechanisms of WNV CNS persistence and its associated neurological sequelae. This information may enable us to define the elements of the immune response that fail or are insufficient to mediate WNV clearance in the infected animals and ultimately help us to set therapeutic goals to modulate immune functions pharmacologically and create vaccines which could induce robust T cell memory responses.
|
10.1371/journal.ppat.1000996 | Viral Protein Inhibits RISC Activity by Argonaute Binding through Conserved WG/GW Motifs | RNA silencing is an evolutionarily conserved sequence-specific gene-inactivation system that also functions as an antiviral mechanism in higher plants and insects. To overcome antiviral RNA silencing, viruses express silencing-suppressor proteins. These viral proteins can target one or more key points in the silencing machinery. Here we show that in Sweet potato mild mottle virus (SPMMV, type member of the Ipomovirus genus, family Potyviridae), the role of silencing suppressor is played by the P1 protein (the largest serine protease among all known potyvirids) despite the presence in its genome of an HC-Pro protein, which, in potyviruses, acts as the suppressor. Using in vivo studies we have demonstrated that SPMMV P1 inhibits si/miRNA-programmed RISC activity. Inhibition of RISC activity occurs by binding P1 to mature high molecular weight RISC, as we have shown by immunoprecipitation. Our results revealed that P1 targets Argonaute1 (AGO1), the catalytic unit of RISC, and that suppressor/binding activities are localized at the N-terminal half of P1. In this region three WG/GW motifs were found resembling the AGO-binding linear peptide motif conserved in metazoans and plants. Site-directed mutagenesis proved that these three motifs are absolutely required for both binding and suppression of AGO1 function. In contrast to other viral silencing suppressors analyzed so far P1 inhibits both existing and de novo formed AGO1 containing RISC complexes. Thus P1 represents a novel RNA silencing suppressor mechanism. The discovery of the molecular bases of P1 mediated silencing suppression may help to get better insight into the function and assembly of the poorly explored multiprotein containing RISC.
| RNA silencing is an evolutionarily conserved sequence-specific gene-inactivation system that also functions as a major antiviral mechanism in higher plants and insects. Viral RNAs are processed by Dicer-like proteins into small interfering (si) RNAs, which trigger the RNA-induced silencing complex (RISC) assembly. Then siRNA loaded RISC inactivates cognate viral RNA. However, viral silencing suppressors evolved to counteract with RNA silencing targeting one or more key points in the silencing machinery. Here we show that in Sweet potato mild mottle virus, the role of silencing suppressor is played by P1 protein and it works by inhibiting si/miRNA-loaded RISC through targeting Argonaute 1 (AGO1). We confirmed using immunoprecipitation and in vitro binding assays that the interaction between P1 and small RNA loaded AGO1 is specific and direct. The suppression activity mapped to the N-terminal part of P1 containing three WG/GW motifs that resemble the AGO-binding linear peptide motif conserved in metazoans and plants. Site-directed mutagenesis proved that these three motifs are essential for both binding and suppression of AGO1 function. P1 protein is the only silencing suppressor identified so far that inhibits active RISC and this is the first demonstration of a WG/GW protein having negative effect on RNA silencing.
| Most eukaryotes, including plants, make use of a well-conserved RNA silencing mechanism to regulate many essential biological processes, ranging from development and control of physiological activities, to responses to abiotic and biotic stress, in particular antiviral defense [1], [2].
Antiviral defense in plants begins with the activity of RNase III type Dicer-Like (DCL) enzymes, which target viral RNAs [3], [4]. Concerted action of the DCL4, DCL2, DCL3 and occasionally DCL1 enzymes results in the appearance of 21–24 nt small interfering RNAs (siRNAs), the central components of the RNA silencing pathway [4], [5]. These viral siRNAs subsequently loaded to endogenous AGO proteins, which are catalytic component of RNA-induced silencing complex (RISC) [6], [7]. AGO1 and AGO7 are suggested to be involved in antiviral silencing [8], [9], [10] although previous study failed to detect viral siRNAs in tagged AtAGO1 [11]. It has been also shown that AGO7 favors less structured RNA targets, while AGO1 is capable of targeting viral RNAs with more compact structures [9]. AGO proteins are responsible for targeting RISC to viral genomes (either RNA or DNA), and exert their action either through cleavage or inhibition of translation [12]. The RNA-dependent RNA polymerases (RDRs) of the host also play important roles in antiviral RNA silencing, being involved in production of secondary viral siRNA [13], [14], [15], [16], [17], [18].
Viruses have evolved suppressors to counteract the RNA-silencing defense of the host [1], [2], [19]. The more than 35 viral silencing-suppressor families so far identified use different strategies to inhibit RNA silencing [2], [20]. Sequestering siRNAs by siRNA-binding suppressors is a very common way to inhibit RISC assembly [21], [22], but other mechanisms have been described, such as inhibiting the biogenesis of 21 nt siRNA species [4], [20], [23]. Other suppressors inhibit RNA silencing through protein-protein interaction. The 2b protein of CMV strain Fny is suggested to inhibit RISC activity via physical interaction with the PAZ domain of the plant AGO1 protein [10]. Polerovirus P0 suppressor protein has been suggested to target PAZ domain of AGO1 and directing its degradation [24], [25].
The Potyviridae is the largest family of plant RNA viruses; in most members, the single-stranded RNA genome is about 10 kb in size and encodes a single polyprotein that is processed into at least 9 mature proteins [26] (Figure 1). In the genus Potyvirus, the multifunctional HC-Pro (helper component-proteinase) was the first viral product to be recognized as a silencing suppressor [27], [28], [29]. The genome of Cucumber vein yellowing virus (CVYV), genus Ipomovirus, family Potyviridae, lacks HC-Pro but contains two P1-type proteases [30], properties shared by at least one other ipomovirus [31]. In CVYV, the second P1 cistron (P1b) was found to suppress RNA silencing [30] with a mode of action resembling that of the HC-Pro of potyviruses [32]. Interestingly, the type member of the genus Ipomovirus, Sweet potato mild mottle virus (SPMMV), possesses an HC-Pro region and a single large P1 serine protease [33].
The peculiarities of the SPMMV genome that incorporates the largest P1 region among all known members of the family together with a typical HC-Pro region (Figure 1), prompted us to study how this virus might deal with the RNA-silencing machinery in its hosts.
In the present study, we show that the large P1 protein of SPMMV possesses silencing-suppressor activity, while its HC-Pro protein does not, on its own. Using various reporter systems, we show that in vivo P1 inhibits target RNA cleavage mediated by RISC complexes loaded with either endogenous miRNA or with virus-derived siRNA. Moreover, suppression activity mapped to the N-terminal half of P1, a region containing three WG/GW motifs that mimics AGO-binding linear peptide motif conserved both in metazoans and plants [34], [35]. We have also determined that the WG/GW motifs at the very N-terminal end in P1 are required for AGO1 binding and for silencing-suppression, suggesting that P1 may use the conserved WG/GW motif binding surface of Ago proteins to inhibit RISC activity.
A partial sequence of 3633 nucleotides was obtained from a plant infected with SPMMV African isolate 130 after RT-PCR amplification with primers designed to flank the first two cistrons of the viral polyprotein. The N-terminal part of this sequence shares structure with the only SPMMV whole genome sequence available in databanks [33]. It presents a large P1 cistron encoding 743 amino acids (15 residues more than the published sequence, starting at position 362), followed by a 453 amino acid HC-Pro cistron, which is more similar in size to other potyviral HC-Pros. The expected cleavage sites and the corresponding residues for the active sites of P1 and HC-Pro proteases [36] could be recognized in the sequence. However, SPMMV HC-Pro lacks the conserved FRNK box characteristic of potyviral HC-Pros and required for small RNA binding and symptom development [37]. These characteristics make SPMMV unique among the Potyviridae, including other ipomoviruses (Figure 1). Sequences are compared in Figure S1.
To investigate whether P1 and/or HC-Pro serve (s) as RNA silencing suppressor for SPMMV, we used the standard Agrobacterium coinfiltration assay [22]. The complete cistrons for P1 and HC-Pro were cloned into binary vectors and the resulting expression constructs were transferred into A. tumefaciens. Cultures of A. tumefaciens able to express GFP from a 35S-promoter GFP binary plasmid were mixed with cultures transformed with our SPMMV constructs before infiltration into Nicotiana benthamiana leaves. In this assay both fluorescence and RNA analysis identified SPMMV P1, but not SPMMV HC-Pro, as the suppressor of RNA silencing (Figure 2A). Weak fluorescence and low GFP mRNA levels were observed in patches infiltrated with the pBin61 empty vector (negative control), and strong suppressor activity and increased GFP mRNA level were detected in patches infiltrated with a construct expressing the P1b of CVYV (positive control) [32].
Experiments designed to compare in parallel the suppression activity of SPMMV P1 with that of a suppressor from a potyvirus, the HC-Pro protein of Tobacco etch virus (TEV) were performed next. First, we checked in vitro if SPMMV proteins could bind either typical 21 nt ds siRNAs, or longer dsRNAs. Extracts of N. benthamiana leaves infiltrated with Agrobacterium strains expressing different suppressors were tested for siRNA binding with labeled 21 nt ds siRNA, and the complexes were resolved on a native gel. As expected, TEV HC-Pro bound ds siRNA, while SPMMV P1 and HC-Pro did not show any siRNA binding activity (Figure 2B). The same extracts were then incubated with a labeled 49 nt dsRNA, and the putative complexes were analyzed on a native gel. In this case, formation of the expected RNA-protein complex only occurred between the 49 nt dsRNA and the Sigma3 protein of a Reovirus [38] used as positive control, but no complexes were detected in any of the other samples from constructs of P1 and HC-Pro of SPMMV (Figure 2C).
Next, we tested if P1 inhibits small RNA processing in leaves of transgenic N. benthamiana line 16C, expressing a GFP transgene, that were coinfiltrated with an Agrobacterium strain harboring a GFP inverted repeat (GFP-IR) construct. To this end, patches infiltrated with constructs expressing SPMMV P1, SPMMV HC-Pro (individual proteins), or SPMMV P1HC-Pro (a construct containing both proteins in cis), or with TEV HC-Pro, were analyzed after 3 days for the presence of GFP mRNA and siRNAs by Northern blotting. No reductions in siRNA processing from the GFP-IR were observed in all expressed proteins (Figure 2D), in contrast to the complete abolition observed in the positive control, which was the dsRNA-binding Sigma3 protein of Reovirus [38]. The 16c plants agroinfiltrated were also observed under UV light at 8 days after agroinfiltration to monitor the spreading of silencing signal. Importantly we found that the presence of SPMMV P1 did not abolish movement of the signal, and therefore silencing of the transgene around the agroinfiltrated area was observed (Figure 2F).
We also checked the capacity of the different viral proteins to inhibit in vivo 3′ modifications of small RNAs by the HEN1 methyltransferase [39], [40]. We expressed P1 along with different silencing suppressor proteins, and the 3′ end methylation status of the mature and star strands of miR168 were then evaluated by oxidation and beta elimination followed by Northern blotting of total RNA samples. Consistently with our previous results [41], TEV HC-Pro inhibited the 3′ methylation of both strands of the endogenous miR168. However , SPMMV P1 and HC-Pro had no effect on HEN1 mediated 3′ modification. (Figure 2 E).
Our findings showed that in contrast to several other silencing suppressors SPMMV P1 does not interfere with the initial steps of the silencing pathway. Thus we hypothesized that it might compromise assembled RISC activity. Active RISC complexes are known to contain ss siRNA and are licensed to cleave the target RNA in a sequence specific manner [42]. Recently, we developed assays based on the transient expression of sensor constructs to test the effect of RNA-silencing suppressors on miRNA and siRNA loaded active RISC. Using these assays we have previously demonstrated that silencing suppressors with ds siRNA binding capacity such as the HC-Pro of potyviruses does not have any effect on miRNA and siRNA loaded RISCs in planta [22], [43].
To determine whether SPMMV P1 might inhibit miRNA loaded RISC complexes, we agroinfiltrated GFP171.1 and GFP171.2 sensor constructs [44] with or without the viral suppressors. In these sensors a full complementary miR171 binding site was placed downstream of the STOP codon of GFP ORF allowing miR171-mediated silencing of the GFP171.1 mRNA, while GFP171.2 carried a mutant miR171 target site, which is refractory to miR171-driven RNA silencing [44]. In this experiment the control construct used was TEV HC-Pro. At two days postinfiltration, GFP fluorescence was evaluated under UV light, and then the infiltrated patches were used for RNA and protein isolation. Consistent with previous results, miR171-driven RNA silencing downregulated GFP171.1, but not GFP171.2 at both the RNA and protein level. Strikingly, comparable GFP fluorescence and GFP mRNA and protein were detected in samples infiltrated with GFP171.1+P1 and GFP171.2+P1, indicating that SPMMV P1 efficiently inhibited miR171 loaded active RISC complexes (Figure 3 A, B). As expected for the control, TEV HC-Pro did not inhibit miR171 mediated degradation of GFP171.1 mRNA [22].
Next, we investigated if SPMMV P1 inhibits viral siRNA-loaded active RISC complexes. A previously described system which exploits N. benthamiana plants infected with Cymbidium ringspot virus (CymRSV) 19 Stop mutant (Cym19S) was used [45]. Cym19S, not expressing the ds siRNA binding silencing suppressor p19, permits a strong RNA silencing response against the virus to be initiated and maintained by enabling viral siRNAs to be loaded into RISC complexes, leading to the recovery of the initially infected plant [45], [46]. At 14–18 dpi of plants carrying Cym19S, the first systemic leaves showed recovery as a consequence of the remarkable amount of active RISCs loaded with siRNAs derived from the virus [22]. Messenger RNAs expressed from the sensor construct GFP-Cym, in which GFP ORF is fused with a ∼200 bp portion of the CymRSV, could be targeted and cleaved by RISC complexes containing Cym19S-derived siRNAs, while GPF-PoLV, in which GFP fused with a ∼200 bp region of Pothos latent virus, a virus unrelated to CymRSV, cannot be cleaved, and was used as a negative control [43]. Recovering leaves of Cym19S-infected plants were infiltrated with GFP-Cym and GFP-PoLV alone or with the indicated silencing suppressors. At 2 days post-agroinfiltration (dpa), efficiency of RNA silencing was monitored by visual examination followed by Northern and Western blotting of RNA and protein samples isolated from infiltrated patches (Figure 3C,D).
When the agroinfiltration was performed only with sensors, Northern analysis using a GFP probe detected a shorter hybridizing band, diagnostic for RISC cleavage of the mRNA expressed from the GFP-Cym construct mediated by viral siRNA, while the hybridizing band remained intact in the case of the GFP-PoLV sensor. As expected, the control TEV HC-Pro was not competent to inhibit ss viral siRNA-loaded active RISC complexes, so the GFP-Cym sensor RNA was cleaved [22]. Remarkably, the Northern and Western analyses showed that GFP mRNA and protein levels were similar in GFP-Cym+P1 and in GFP-PoLV+P1 infiltrated samples, and no cleavage product of GFP was detected in the GFP-Cym+P1 infiltrated sample, suggesting that SPMMV P1 efficiently inhibited the slicing activity of the viral siRNA loaded RISC complexes (Figure 3 C,D).
The RISC complex is of high molecular weight (>669 kD) in animals [47], [48], contains the catalytic AGO protein, and has intrinsic small-RNA-dependent target cleavage activity. In plants such as N. benthamiana, transiently expressed or endogenous AGO1 protein co-fractionates in extracts with small RNAs, and can be found in at least two distinct complexes of above 669 kD and 158 kD [49]. In addition, it was reported that high molecular weight complexes containing viral siRNAs exhibited nuclease activity in vitro and preferentially targeted homologous viral sequences [50]. Having established that P1 inhibits active RISC, we hypothesized that inhibition of RISC requires physical interaction of P1 with AGO1 and small-RNA-containing complexes. To investigate this, we first tested whether P1 co-fractionates with AGO1 and small RNAs on a gel filtration column. N-terminally HA-tagged SPMMV P1 (HA-P1), 6×myc-tagged AGO1 of Arabidopsis thaliana (myc-AGO1) [10] and GFP-IR were co-expressed in N. benthamiana leaves. At 3 dpi, extracts prepared from infiltrated leaves were fractionated on a Superdex 200HR column. Small RNAs were extracted from each fraction and analyzed by Northern blotting, and the AGO1 and P1 protein contents of fractions were monitored by Western blotting using antibodies raised against the HA and myc tags. Consistent with previous results, GFP siRNAs and miR159 were fractionated in two distinct complexes peaking at >669 and 158 kD, although they appeared in all fractions which also contained AGO1 (Figure 4A). We experienced technical difficulty in separating the protein peaks, and this might reflect the disintegration of the large complexes during chromatography or more likely due to limited availability of the other RISC components. Despite these problems, the infiltrated myc-AGO1 co-fractionated clearly with small RNAs suggesting that GFP siRNAs had been loaded into the myc-AGO1-containing complexes. Interestingly, SPMMV P1 co-fractionated mainly with the 669 kD, but not with the smaller 158 kD myc-AGO1 and small RNA-containing complexes (Figure 4 A).
Next, we investigated if co-fractionation of P1 with myc-AGO1 and small RNAs was due to physical interaction. To test this, we agroinfiltrated HA-P1 with myc-AGO1 and GFP-IR. As a negative control, myc-AGO1 and GFP-IR were agroinfiltrated with HA-UPF1, which is known not to be involved in RISC formation. At 3 dpi, extracts were prepared from infiltrated leaves, and HA-tagged proteins were immunoprecipitated (IP) with an anti-HA antibody. Inputs and eluates of IPs were tested for proteins by Western blotting and for GFP siRNA in Northern blots. The results showed that HA-P1 and HA-UPF1 were expressed at comparable levels and could be efficiently immunoprecipitated from extracts. Importantly, we found that myc-AGO1 coimmunoprecipitated with HA-P1, but not with HA-UPF1, confirming that the interaction between HA-P1 and myc-AGO1 is specific (Figure 4B). Moreover, we found that GFP siRNAs, but not tRNA coimmunoprecipitated exclusively with HA-P1 and myc-AGO1, strongly suggesting that P1 interacts with small RNA-loaded AGO1.
Taken together, we showed that siRNAs derived from GFP-IR became incorporated into myc-AGO1, and that the P1 silencing suppressor specifically interacted with GFP siRNA-loaded myc-AGO1. These results, along with earlier data proving that P1 inhibits si- and miRNA programmed RISC, suggest that the large complex (669 kD) containing AGO1 and small RNAs corresponds to the plant RISC complex.
P1 contains an extended N-terminal region and a protease domain at its C-terminal end, similar to the P1b of other ipomoviruses [30]; Text S1). To determine the minimal region required for silencing suppression, we constructed P1 mutant truncated from the C-terminal end but retaining the first (N-terminal) 383 aa, designated as HA-P11-383 (Figure 5A). To evaluate its silencing-suppressor activity, the mutant was co-expressed with GFP-171.1 in N. benthamiana plants. Visual examination under UV light and analysis of GFP-171.1 sensor RNA and GFP expression showed that HA-P11-383 was an effective silencing suppressor although lacking the entire C-terminal protease domain. Then, we checked the interaction between the deletion mutant of P1 and AtAGO1. To test for interaction, we immunoprecipitated myc-AGO1 from extracts of infiltrated leaves expressing myc-AGO1 and GFP-IR with HA-P1 and HA-P11-383. We used the pBIN61 empty vector as negative control. We found that HA-P11-383 interacted with myc-AGO1 as strongly as wt P1. We concluded that P1 may be composed of two functional domains, the silencing suppressor domain is located at the N-terminal part, and the C-terminal part of P1 contains the protease domain. However, the protease activity of P1 was not analyzed.
Our results showed that the N-terminal end of the P1 is required for Ago binding. Further inspection of the N-terminal end of P1 revealed repeating tryptophan-glycine/glycine-tryptophan residues (WG/GW) (Figure 6A and Figure S2), which are identical to the core amino acids of the WG/GW motifs recently found in Argonaute binding proteins, such as Tas3 and RNA Pol V (El-Shami et al, 2007; Till et al, 2007). Furthermore, analysis of amino acid composition revealed that the regions neighboring the WG/GW residues in P1 are rich in alanine, serine, glutamic acid, asparagine and aspartic acid, providing a context similar to that described for Tas3 and RNA Pol V proteins [35], [51].
This observation prompted us to investigate the significance of the tryptophan residues at the N-terminal end of P1 in silencing suppressor activity. For this, we generated single, double and triple mutants of P1 by replacing tryptophan (W) by alanine (A) residue(s) at positions 15, 101 and 131, individually and in all double (3) and triple (1) combinations, by site-directed mutagenesis. Silencing-suppressor activity of the HA-tagged P1 single mutants was compared to the HA-P1wt, co-infiltrated with the GFP-171.1 sensor construct. Expression analysis of the GFP marker gene reflecting the strength of suppression of RNA silencing showed that suppressor activity of any of the single mutants was not reduced significantly, suggesting that presence of the remaining two tryptophan residues were sufficient to maintain the silencing suppressor activity of P1 (data not shown). In contrast, the suppressor activities of double and triple mutants were greatly reduced. Consistently, wt P1 and the mutants were expressed at comparable level (Figure 6 B,C).
To test whether the WG/GW motifs are required for RNA silencing-suppression because they contribute to AGO1 binding, we tested the interactions between AtAGO1 and P1 double and triple mutants. Myc-AGO1 and GFP-IR were co-infiltrated with double and triple mutants of HA-P1 in line GFP16c/RDR6i N. benthamiana plants. As positive control, we used HA-P1wt and as negative control, myc-AGO1 and GFP-IR were infiltrated without P1 wt. At 3 dpi, extracts of infiltrated leaves were used to immunoprecipitate HA-tagged P1 wt and mutant proteins. Western analysis showed that HA-tagged proteins were expressed at comparable levels and were successfully immunoprecipitated. However, probing Western and Northern blots to detect myc-AGO1 protein and GFP siRNA derived from GFP-IR revealed that myc-AGO1 protein and GFP siRNAs were specifically co-immunoprecipitated with P1 wt, but not with any of the double or triple mutants of P1 (Figure 6D).
These results showed that changing at least two out of three tryptophan residues to alanine in the WG/GW motifs of P1 abolished its silencing suppressor activity. Moreover, our analysis showed that the ability of P1 to bind AGO1 depends on the presence of these motifs, suggesting a correlation between AGO1 binding and its activity as silencing suppressor.
AGO1 of A. thaliana is involved both in the miRNA and the antiviral RNA silencing pathways [10], [11], [52]. Our results showed that P1 interacts with AGO1 to inhibit active miRNA and siRNAs loaded RISC.
To get better insight into the mechanism of inhibition of RISC mediated by P1, we performed immunoprecipitations with small RNA-loaded RISCs against P1 (wild type) and P1mut1-2-3 (triple mutant)-infiltrated leaf extracts. RNA samples from inputs and immunoprecipitates were probed for presence of two endogenous miRNAs (Figure 7A). The results showed that mature miR159 and miR319 specifically co-immunoprecipitated with wt P1, but not with P1mut1-2-3. In addition, we found only mature miRNAs in the eluates of P1 immunoprecipitates, and the star strands for miR159 and mir319 could not be detected in inputs, or in eluates. Thus our results indicated that P1 interacts with endogenous RISC complexes loaded with single-stranded miRNAs (Figure 7A). To test whether P1 interacts directly or indirectly with AGO1 we performed in vitro pull-down assays using recombinant MBP-AGO1366-1048 containing the PAZ-MID-PIWI domains and the 6×His-P11-383 N-terminal fragment of wt P1. The results showed that MBP-AGO1366-1048 binds wt 6×His-P11-383 efficiently, while the triple mutant P11-383 was bound less strongly by AGO1 protein (Figure 7B). This result strongly suggests a direct interaction between P1 and AGO1 proteins in vivo as well.
Available data suggest that virtually all plant viruses encode at least one suppressor and more than 35 individual viral silencing suppressor families have been identified to date. However, the mechanisms of action have been explored only for a few [2], [20]. Among the best-characterized RNA silencing suppressors, the p19 protein of tombusviruses and the HC-Pro protein of potyviruses share the ability to bind and sequester siRNAs, which are the most conserved element of the silencing machinery. SiRNA binding has been postulated as a common and effective strategy to counteract plant defenses against viruses [22]. However, further studies have shown that many viruses use other strategies and can adapt unrelated proteins to target and interfere with different steps in the silencing pathway. For instance, inhibition of siRNA generation by TCV p38 has been described [4], as has the targeting of AGO proteins, the conserved catalytic components of RISC, by viral proteins through protein-protein interactions [10], [25], [53].
In this work we explored the mechanism of silencing suppression mediated by the SPMMV P1 silencing suppressor protein, which interferes with miRNA and siRNA driven RISC activity by binding to the AGO1 subunit of RISC complexes through its WG/GW motifs conserved also in Argonaute binding cellular proteins.
SPMMV is the only ipomovirus that has a typical potyvirid genome structure with P1 and HC-Pro regions in the C-terminal part of the polyprotein [33]. Other ipomoviruses with available complete sequences do not possess HC-Pro regions [31], [54], [55]. Despite the presence of an HC-Pro in SPMMV, we have found that the role of RNA-silencing suppressor is played by P1.
To better understand the molecular basis of P1-mediated silencing suppression, we analyzed the effect of P1 on different steps of the RNA silencing pathway. We found that P1 does not seem to interfere with the biogenesis of either transgene-derived siRNAs or endogenous miRNAs, since we observed that the accumulation of GFP-IR-derived siRNAs was mostly unaltered in the presence or absence of P1 (Figure 2D). Similarly, accumulation of endogenous miRNAs and their 3′ methylation status were not influenced by the expression of P1, in contrast to the well known TEV HC-Pro suppressor, which binds ds siRNA and miRNA intermediates and partially inhibits their 3′ methylation [22], [41] (Figure 2E). We also showed that P1 failed to bind short and long ds RNAs, in contrast to potyviral HC-Pro and the reovirus sigma3, which efficiently bind ds siRNAs and long dsRNAs, respectively (Figure 2B,C). Thus, this mechanism of silencing suppression seems to be unique among virus-encoded silencing suppressors identified so far.
We tested the effect of P1 on miRNA- and viral siRNA-activated RISC complexes using GFP sensor constructs (Figure 3) and it turned out that P1 efficiently inhibited both types of activated RISC in vivo. Moreover, we showed that transiently expressed AGO1 protein was found in a large (>667kD) and an approximately 158 kD protein complexes in size. Interestingly, P1 was found co-fractionating only with the large AGO1 containing complex with GFP siRNAs. This results distinguishes P1 from previously studied silencing suppressors, because it does not inhibit RISC assembly, like small RNA binding suppressors, nor inhibits RISC assembly by promoting degradation of AGO proteins, as it was found in the case of P0 protein of poleoviruses [25], [49], [53]. The 2b protein of CMV FNY strain was also shown to interact with AGO1 in vivo and in vitro and to inhibit RISC activity in vitro [10]. However, it is not known whether 2b prevents RISC assembly or inhibits siRNA loaded RISC by AGO1 binding. In addition, a recent report revealed that 2b proteins of Tomato aspermy virus (TAV) and the FNY strain of CMV bind 21nt ds small RNAs [56], [57], so it is not clear, whether 2b protein of cucumoviruses inhibits RNA silencing through siRNA binding, interacting with AGO1 or both. In contrast, P1 binds RISC by interacting the AGO1 subunit of RISC loaded with si- or miRNAs, as shown by our immunoprecipitation studies (Figure 4 and 7). Importantly, P1 interacted with AGO1 containing mature miRNAs but not their star strand; this adds support to our hypothesis that P1 interacts with the AGO1 component of active RISC complexes, and is in line with the efficient inhibition of GFP-sensor silencing by P1. Our cumulative evidence strongly suggests that the P1 interaction with AGO1 is a direct physical interaction. Finally, using mutant P1 proteins with their silencing suppressor activity compromised/abolished, we obtained evidence that silencing suppression and AGO1 binding are linked.
The WG/GW motifs located at N-terminal part of P1 strongly resemble the evolutionarily conserved GW linear peptide motifs shared by different silencing-related proteins used as “Ago hooks” to interact with Argonaute proteins [35], [58]. Such WG/GW motifs have been described in proteins from different organisms, such as in the largest subunit NRPD1b of the RNA polymerase V in plants [51], the P body-localized human protein GW182 [59], [60], [61], and the Tas3 homologue of the GW182 RITS complex component in yeast [62], [63]. All these proteins can interact with Argonaute proteins [35]. Recently, an RdDM effector KTF1 containing abundant WG/GW motifs and SPT5-like domains has also been identified as an AGO4 binding element [64], [65]. Similarly to cellular WG/GW proteins, our analysis of P1 mutants indicates that the tryptophan residues are essential for interaction with AGO1 and are strictly required for silencing suppressor function (Figure 6).
Recent results showed that the AGO-binding domains and the effector domain of GW182 paralogs map in different parts of the proteins [66]. Thus, the modular architecture of the WG/GW proteins that allowed the evolution of Ago-binding elements with positive effects on different RNA silencing pathways, like RNA Pol V, Tas3, GW182 and KTF1 [35], [51], [64], [65], [66], could have been mimicked by a viral protein, although in the case of P1 the effect is negative/suppressive. An attractive possibility to explain the negative effect exerted by P1 on the silencing machinery could be its capacity to outcompete essential AGO1 interacting components of RISC, although in plants these hypothetical AGO1 interactors have not been identified yet. Further experiments will be required to test this possibility and to identify which endogenous elements might be displaced by P1.
We can also postulate alternative explanations for the action of P1. Since small RNA-dependent target cleavage by RISC requires base-pairing between the small RNA and the target RNA, the presence of P1 as an AGO1 interactor might result in covering the small RNA-binding groove of AGO1, thus interfering with base-pairing between the small RNA and the target RNA. This latter possibility is really plausible, because precluding base-pairing between the target RNA and the small RNA would inhibit translation as well, and indeed the importance of translation inhibition in plants has recently been highlighted [12]. In agreement with this, our results with viral siRNA-loaded RISC complexes (Figure 3C and D) show that target cleavage activity did not always correlate with GFP expression (compare lanes 4 and 8 in Figure 3 D); this may indirectly indicate that translational inhibition is hampered by P1 in our system as well.
The efficient binding of AGO1, and inhibition of its function by P1, shown by our experiments suggest that this suppressor might have evolved to bind AGO1 protein with high affinity to inhibit its function. Independently of its final mode of action during suppression, P1 is another example of the extraordinary adaptation of viruses, which are able to target highly conserved key elements of the antiviral silencing response, to be able to complete their infectious cycle. In our study we analyzed only AGO1 as P1 interactor, and we don't know whether P1 is able to interact with other plant AGO proteins. Interestingly, P1 failed to show any effect (Peter Moffett, personal communication) when assayed in an R gene-induced anti-viral response test that is dependent on AGO4-like but not AGO1-like activity [67], suggesting that P1 is not able to target all AGOs.
Our results can also help to explain the pathology of SPMMV, either alone or in synergism with other viruses. Generally speaking, defeat of the host RNA silencing response by a virus equipped with a silencing suppressor requires a high concentration of the suppressor in infected cells, above both the dissociation constant of the suppressor with its target and the intracellular concentration of the target molecule [68]. In SPMMV-infected cells, both existing and de novo assembled RISCs, including miRNA- and viral siRNA-loaded RISCs, should be considered as potential targets for P1. At early stages of infection, the concentration of existing active RISC might be much greater than that of the de novo viral siRNA-loaded active RISC, and this would lead to the sequestration of P1 mainly by existing active RISC complexes. Consequently, we may hypothesize that the newly formed viral siRNA-loaded RISC could escape from suppression, resulting in low SPMMV titre, mild transient symptoms, and recovery of the plant. However, in cases where plants have additionally been infected with other viruses such as SPCSV, the initial antiviral response might be suppressed by the two-component silencing suppressor system of SPCSV [69], [70] resulting in a much higher titre of SPMMV, which in turn might allow a high concentration of P1 in infected cells. P1 might then efficiently suppress RISCs loaded with both endogenous small RNAs and antiviral RNAs, which would then lead to the synergistic sweet potato disease. High accumulation of SPMMV has indeed been observed in mixed infection with SPCSV, resulting in a severe disease [71]. It is likely that symptom aggravation comes from the fact that both pathogens encode suppressors with complementary effects.
The African isolate SPMMV-130 was kindly provided by Jari Valkonen (University of Helsinki, Finland) in a sweet potato plant, and maintained in N. tabacum Xanthi plants. The complete P1 and HC-Pro regions of the virus were RT-PCR amplified from total nucleic acid extracts using primers 5′CCTCTAGAATGGGGAAATCCAAACTCACTTAC3′ and 5′GTCCCGGGTCAATAGAATTGTATCTGTTTAAGTTTACTAG3′ for P1, and 5′CCTCTAGAATGGCAAGTTCTGTTGTACCCAATTTC3′ and 5′GTCCCGGGTCAACCAACCTTATAGGTTAACATCTCAC3′ for HC-Pro, and cloned, using the restriction sites highlighted in bold, into competent plasmids for sequencing. The two viral genes were cloned into pBIN-derived constructs for transient expression in N. benthamiana leaves. Variants incorporating the tagging element HA were also prepared. Plants were kept in a greenhouse at 22°C under a photoperiod of 12 h/12 h light/dark. Infiltration assays were performed on expanded N. benthamiana leaves of plants about 21 days old.
N. benthamiana leaves were infiltrated essentially as previously described [46]. Agrobacterium strains harboring 35S-GFP, GPF-IR, GFP171.1, GFP171.2, GFP-Cym, GFP-PoLV and miR171c precursor were infiltrated with OD600 = 0.1. SPMMV 35S-HC-Pro, HA-UPF1 [72], 6×myc-AtAGO1 [10] and RNA-silencing suppressors such as 35S-P1, 35S-P1HC-Pro, HA-P1 were infiltrated with OD600 = 0.2–0.3. Infiltrated patches were used for total RNA and protein isolation, then analyzed by Northern and western blotting.
Total RNA was isolated using TRIZOL reagent. RNA was analyzed on 37% formaldehyde containing agarose gels as described [46]. Small RNAs were analyzed on 12% arcylamide 8M urea gels. RNA isolation from column fractions also was described earlier [73]. Briefly, equal volume of 2×PK buffer was added to each fractions and Proteinase K at final concentration of 80ng/µl. Samples were incubated at 55°C for 15 min. Then, RNA was extracted by phenol-chlorophorm and precipitated with 2,5 volumes of ethanol. After recovering, RNA was resuspended in 50% formamide containing buffer and loaded on 12% arcylamide and 8M urea gels. The gels were blotted and hybridized with riboprobes to detect small RNAs or random primed DNA probes for conventional Northern blots.
Total RNA samples were oxidized, ß-eliminated and detected as described in [41]. Briefly, a total of 10 µg total RNA was dissolved in 17.5 ml borax buffer, pH 8.6, 50 mM boric acid and 2.5 ml 0.2 M sodium periodate was added. The reaction mixture was incubated for 10 min at room temperature in the dark and, after addition of 2µl of glycerol, incubation was repeated. The mixture was lyophilized, dissolved in 50 ml borax buffer, pH 9.5 (33.75 mM borax, 50 mM boric acid, pH adjusted by NaOH) and incubated for 90 min at 45°C. RNA species were then separated on 12% denaturing PAGE blotted and hybridized using 32P labeled LNA oligonucleitide probes [74] as described above.
Extracts for immunoprecipitation were prepared in IP buffer containing 30mM TRIS (pH 7.5), 150mM NaCl, 5 mM MgCl2, 5mM DTT and 10% glycerol, then incubated for 1 hour at 4°C with beads containing anti-HA (Roche) or anti-myc antibody (Sigma). The beads were then washed with IP buffer. Half of the eluates were used for RNA isolation as described [73]. Commercially available antibodies were used for detecting GFP (Roche), HA-tag (Roche), myc-tag (Sigma), His-tag (Amersham Biosciences) MBP-tag (Sigma). For AGO1 detection we used previously described anti-peptide antibody against N. benthamiana AGO1 [49].
Plant extracts for gel filtration were prepared in IP buffer and fractionation was carried out similarly as described earlier [73]. Briefly, 200 µl of extracts were loaded on the Superdex 200HR column and washed with IP buffer with 0.5 ml/min. 25 fractions were collected and after vortexing them for equilibration, each fraction were divided into two for RNA and protein isolation. RNA was isolated, as described above. Proteins were precipitated with 4 volumes of acetone and collected by centrifugation, then solubilized in 2×Laemmli buffer. Proteins were detected by western blotting.
Site-directed mutagenesis was performed using the Quickchange site-directed mutagenesis kit (Stratagene) according the manufacturer's instructions to generate single, double and triple mutants with oligonucleotides listed in Text S1. Deletion mutants were prepared by PCR using oligonucleotides P1-5′ 5′GGGGATCCCTAGAATGGGGAAATCCAAACTC3′, and P1-383 5′GCGGATCCTCAATCATCAACTTGTGCGTTTAGGGA3′. All mutants were verified by sequencing.
GenBank accession numbers for new viral nucleotide sequence: GQ353374 and for complete SPMMV sequence: NC_003797.
|
10.1371/journal.pntd.0004858 | Leptospira Seroprevalence and Risk Factors in Health Centre Patients in Hoima District, Western Uganda | The burden of human leptospirosis in Uganda is unknown. We estimated the seroprevalence of Leptospira antibodies, probable acute/recent leptospirosis, and risk factors for seropositivity in humans in rural Western Uganda.
359 non-pregnant adults visiting the Kikuube and Kigorobya Health Centers were sequentially recruited during March and April 2014. A health history survey and serum were collected from consented participants. Overall, 69% reported having fever in the past year, with 49% reporting malaria, 14% malaria relapse, 6% typhoid fever, 3% brucellosis, and 0% leptospirosis. We tested sera by microscopic agglutination test (MAT) against eight Leptospira serovars representing seven serogroups. Leptospira seroprevalence was 35% (126/359; 95%CI 30.2–40.3%) defined as MAT titer ≥ 1:100 for any serovar. The highest prevalence was against L. borgpetersenii Nigeria (serogroup Pyrogenes) at 19.8% (71/359; 95%CI 15.9–24.4%). The prevalence of probable recent leptospirosis (MAT titer ≥1:800) was 1.9% (95%CI 0.9–4.2%) and uniquely related to serovar Nigeria (serogroup Pyrogenes). Probable recent leptospirosis was associated with having self-reported malaria within the past year (p = 0.048). Higher risk activities included skinning cattle (n = 6) with 12.3 higher odds (95%CI 1.4–108.6; p = 0.024) of Leptospira seropositivity compared with those who had not. Participants living in close proximity to monkeys (n = 229) had 1.92 higher odds (95%CI 1.2–3.1; p = 0.009) of seropositivity compared with participants without monkeys nearby.
The 35% prevalence of Leptospira antibodies suggests that exposure to leptospirosis is common in rural Uganda, in particular the Nigeria serovar (Pyrogenes serogroup). Leptospirosis should be a diagnostic consideration in febrile illness and “smear-negative malaria” in rural East Africa.
| Leptospirosis is an emerging zoonotic disease caused by bacteria of the genus Leptospira. Despite its relatively common frequency, mild/moderate leptospirosis often goes unrecognized, due to its usually non-specific symptoms of fever, vomiting, and malaise. Knowledge of leptospirosis in Uganda is limited, and the disease may often be misdiagnosed as malaria. This study sought to define the percentage of healthcare seeking Ugandans in rural Western Uganda who have antibodies to Leptospira in their blood, suggesting prior exposure. We found 35% of study participants had antibodies to at least one Leptospira serovar, predominantly L. borgpetersenii sv Nigeria representing the Pyrogenes serogroup (20% of all participants). Individuals with increased odds of having antibodies to leptospires included participants who skinned cattle and those who reported monkeys near their home. Individuals who self-reported recent diagnosis of malaria were more likely to have leptospirosis antibodies. Antibodies to leptospirosis are not lifelong, typically lasting a few years. The high 35% seroprevalence suggests there is ongoing exposure. Further studies are needed to understand the burden of leptospirosis in rural Africa, the risk factors associated with exposure, and the public health opportunities to mitigate leptospirosis.
| Leptospirosis is a zoonotic bacterial disease with a worldwide distribution that is endemic in subtropical and tropical countries. Transmission occurs through exposure to urine or aborted tissues of infected animals, either through direct contact with carrier animals or contact with contaminated water or soil [1]. Leptospira is a Genus of spirochetes that comprise 20 species and almost 300 different serovars (sv). They have a large range of mammalian hosts which carry specifically adapted serovars in their renal tubules and excrete them in the environment for months or years. Humans are considered accidental hosts who most likely do not transmit the bacteria [1–3]. Infection patterns vary depending on the climate and rainfall, and the socio-economic, occupational, or recreational activities which bring a susceptible host into contact with infected water or animals. In tropical regions, leptospirosis outbreaks in animals and humans often occur after flooding [4].
Human incidence of leptospirosis is commonly underreported due to low awareness of the disease, lack of surveillance systems in place, nonspecific clinical symptoms, and the complexity of diagnosis [4,5]. The World Health Organization (WHO) Leptospirosis Epidemiology Reference Group (LERG) estimated in a 2015 systematic literature review that the annual incidence was 1.03 million cases (95%CI, 305,000–1,750,000) and 58,900 deaths (95%CI, 23,800–95,800) due to leptospirosis worldwide [6]. East Africa had an estimated annual incidence of 25.6 (95%CI 9.3–43.3) per 100,000 population [6].
Studies on leptospirosis in Uganda have only been published in animals. In 2011, Millan et al. [7] sampled 105 dogs around three national parks using the Microscopic Agglutination test (MAT) and found 27% (95%CI, 19–36) seropositivity; across six serovars, most frequently Icterohaemorragiae and Canicola [7]. In 2014, Atherstone found a 29% seropositivity in cattle (n = 92) and 42% seropositivity in buffaloes (n = 92) for Leptospira Hardjo serovar when using an ELISA [8].
Elsewhere in East Africa, a few studies on Leptospira prevalence in humans and animals have been published. A recently published systematic literature review on leptospirosis in Africa reported a prevalence of acute human leptospirosis ranging from 2.3% to 19.8% in 11 studies in hospital patients with acute febrile illness and compatible symptoms [9]. A study conducted in Tanzania in 2013 estimated 75–102 clinical leptospirosis cases per 100,000 population [10]. In two studies of Tanzanian cattle in 2011 and 2014, a 30% seroprevalence was found (n = 1758) using the MAT [11,12]. Assenga et al. reported a seroprevalence of 29% in buffaloes (n = 38), 20% in rodents (n = 207), and 30% in humans (n = 267) [12]. The most prevalent serovar was Hardjo in humans (16%) and cattle (18%), but in rodents was Australis (19%). These results point to a human-cattle transmission pathway, which is plausible in the agro-pastoral environment of Tanzania [12].
Although cross-sectional seroprevalence studies mainly indicate past exposure to leptospires, persons with high titers may have recent acute disease or recent recurrent exposure. Sero-conversion may serve as an imperfect, proxy measure of incidence. Biggs et al. demonstrated 33% of 870 acute febrile illness patients had antibodies against Leptospira serovar in Moshi, Tanzania in 2007–2008 [13]. Further, 8.8% of 870 had seroconverted with a greater than four-fold increase in serum MAT titer on convalescent testing, consistent with “confirmed leptospirosis”. An additional 3.6% had titers of ≥1:800 in one tested sample, and therefore suited the case definition of “probable recent leptospirosis” [13]. Among their laboratory confirmed cases of leptospirosis, 44% had previously received the clinical diagnosis of malaria.
Given these foundational studies and the similar eco-systems and agro-pastoral activities between Uganda and Tanzania, leptospirosis may present an unrecognized disease burden in Uganda, particularly when misdiagnosed as malaria or another febrile illness. This study’s objectives were to estimate the seroprevalence of Leptospira antibodies and probable recent leptospirosis in humans in rural western Uganda. Secondary objectives included analysis of risk factors for seropositivity against specific serovars and for probable recent leptospirosis.
Hoima District in western Uganda presents unique challenges regarding diagnosis and management of leptospirosis. The district lies on the eastern coast of Lake Albert, the northwestern most of the African Great Lakes. Hoima contains a widely varied ecology, with protected conservation areas, mixed-use pastoral farm lands, coastal fishing villages, and urbanized towns. The climate is tropical, with average rainfall of 1270 mm/year, average temperatures above 21°C and two rainy seasons from March-May and August-November which can bring intermittent flooding [14]. Malaria is holoendemic and most commonly diagnosed by microscopy, with greater than 80% of children under the age of ten infected [15]. As a result of these high levels of parasitemia, febrile illnesses may be routinely diagnosed as malaria, obscuring the true burden of other infectious pathogens like leptospirosis. Over a quarter of the district’s 550,000 inhabitants live in poverty [14]. Occupations focus predominantly around subsistence agriculture characterized by mixed farming and pastoralism (i.e. livestock). Novel trends show increasing land area devoted to rice farming that requires periodic flooding and recent oil and gas extraction [16,17]. The district also contains the UNHCR refugee settlement at Kyangwali sub-county, where >20,000 refugees from Democratic Republic of Congo are located [16]. These mixed communities, land use patterns, and ecologies create a complex milieu allowing zoonotic disease transmission. Hoima’s community Health Centres serve this heterogeneous patient population.
Study participants were recruited at the Kikuube and Kigorobya Health Center IV’s within Hoima District during March and April 2014. The Health Centre IV’s provide service to ~100,000 Ugandans each. Every non-pregnant adult aged ≥ 18 years who presented to the health center, either as a patient or as a caregiver, was invited to participate. Common reasons for seeking health care at these centers included: obstetric complications, occupational trauma, and acute febrile illnesses. Demographic interviews were conducted by clinical officers, and serum samples were collected from both field sites on a daily basis. Samples were centrifuged and then stored in -20°C freezers at the Hoima Regional Referral Hospital. Sera were transported on ice to the Ministry of Agriculture, Animal Industry, and Fisheries in Entebbe, and stored at -80°C prior to testing.
Survey data were collected in four major areas (S1 Text). First, information on demographics including age, gender, education, religion, profession, sub-county of residence, and duration of residence in Hoima was collected. Second, past medical history assessed for prior diagnosis with febrile diseases such as malaria, typhoid, brucellosis, leptospirosis, and hemorrhagic fevers. Additional questions assessed recent history of fever within the past year, whether the participant had seen a physician and the outcome of any treatment for febrile illness. Third, animal contact was evaluated using quantity of animals owned, animal product consumption during the last month, involvement in cattle or beef processing within the past two weeks, and the presence or absence of wildlife, such as rats or monkeys, near the home. Finally, questions regarding the participant’s domestic environment included the fabrication material used to construct their home and their major sources of drinking water (Table 1). Data on acute clinical status, such as fever of the patient at the Health Centre, was not collected.
The MAT is the reference test for distinguishing among leptospirosis serovars and giving valuable information on past exposures [2]. Due to the absence of MAT laboratory capacity in Uganda and the lack of isolated and cultured local strains, Leptospira serovars and corresponding antisera were imported from the OIE Reference Laboratory Royal Tropical Institute (KIT), Holland. A panel of eleven serovars from ten different serogroups (sg) was chosen based on the recommendations of experts from KIT, represented in Table 2. These serovars were selected to represent a wide variety of serogroups with minimal crossreactivity, most of which had been isolated from other parts of Africa. Leptospira Kirschneri sv Soikoine, Leptospira borgpetersenii sv Kenya and Leptospira borgpetersenii sv Nona were not tested, as these antigens did not grow well during the testing phase or were contaminated and not regarded fit for MAT testing. MATs were performed in the framework of a capacity building course run by the authors Pearson and Dreyfus in conjunction with authors Alinaitwe and Kakooza. In an effort to partner equally with our Ugandan colleagues, development of MAT capacity in country was considered an essential component of the project.
The serum aliquots were transferred to microtiter plates and stored at -80°C. The MAT measured serum antibodies at doubling two-fold dilutions starting at 1:25 dilution up through 1:3200, as described previously [2]. Given the absence of information on prevalent serovars in Uganda, serovars were considered representative of their serogroup, and within serogroup cross-reactivity was not excluded.
A seropositive case was defined as a MAT titer of ≥1:100 against any serovar. A probable recent leptospirosis case was defined as a MAT titer of ≥1:800 [13,18]. The outcomes of interest were the prevalence of study participants who were seropositive against each specific serovar, seropositive against any serovar, or seropositive with probable recent leptospirosis.
Past medical history was based on self-reported survey information. No biometric or serological data were available to confirm self-reported data. Questionnaire information and serologic test results were entered into Microsoft Access and analyzed using Excel and Stata 10 (StataCorp, College Station, TX, USA). Data deposited in the Dryad repository: http://dx.doi.org/10.5061/dryad.6ns6p [19]. Exploratory data analysis was conducted to evaluate crude associations using 2x2 tables, histograms and summary measures. To improve power in the risk factor analysis, two composite variables were created to identify groups with persons with intensive contact with livestock and potential exposure to livestock associated Leptospira serovars. The first composite variable included persons involved in slaughtering, skinning and butchering cattle, while the second also included exposure to cattle milking and birthing.
A sample size of 278 enabled a 95% confidence interval (95% CI) within a precision of ±5% under the assumption that the true prevalence was 20% [20]. To detect an odds ratio of 2.5 with 80% power, a type I error of 0.05, prevalence of 9% in the exposed group, and an exposed to non-exposed ratio of one-third, the required sample size was 280 study participants.
The association between the outcomes and exposure variables listed in Table 1 was analyzed in two steps. Initial analysis included bivariable comparison of individual exposure variables with outcomes by chi-square tests or logistic regression, followed by a multivariable logistic regression. A manual forward and backward selection method was used to evaluate the association between exposure and confounding variables with the outcome. Exposure variables were entered in the model if the bivariable p-value was ≤0.2 or if they represented biologically plausible risk or confounding factors for the outcome and were kept in the model if the Likelihood Ratio Test was statistically significant (p≤0.05).
All procedures involving human subjects were approved by the institutional review boards of the University of Minnesota, USA and Joint Clinical Research Centre, Uganda and the Uganda National Council of Science and Technology (UNCST). Potential participants were informed of the study during their healthcare encounter. After their routine healthcare visit, each person was given the opportunity to provide written informed consent.
A total of 359 participants provided informed consent, were interviewed, and had serum collected. Of those approached for participation, approximately 50% consented to participate. Participants originated from eight of 13 sub-counties in Hoima District of western Uganda, although 89% (320/359) came from the subcounties immediately surrounding each Health Centre. 71% of participants were women, and 69% (246/259) were <40 years of age. 81% participated in farming as either a primary or secondary occupation, with a minority involved with livestock on a daily basis, such as milking (7%), birthing (2.5%), slaughtering (1.7%), skinning (1.7%) and in butchering (1%) (Table 1).
Over their lifetime, 87% (n = 312) of 359 study participants had self-reported diagnoses of malaria, 16% (n = 56) typhoid, 4% (n = 16), brucellosis, and 1% (n = 4) reported a diagnosis of tuberculosis. No participants reported a diagnosis of leptospirosis. For illnesses within the past one year, 49% of participants reported a malaria diagnosis over the past year. Of participants with malaria diagnoses, 49 participants (20%) noted having had a relapse of fever after initial malaria treatment. Overall, 69% (n = 249) reported a history of fever within the past year, and of those who reported a fever, 70% (175/249) received a clinical diagnosis of malaria. For other illnesses, 8% (n = 20) self-reported typhoid enteric fever, and 5% (n = 12) self-reported a brucellosis diagnosis in the past year.
Regarding other zoonotic infections within the communities, 16% (n = 58) of participants reported to have known someone affected by brucellosis and 15% (n = 54) by rabies. No one reported leptospirosis in the community.
126 study participants (35.0%, 95%CI, 30.2–40.3%) were seropositive against any of the eight serovars. The diversity of responses against the different serovars, representing different serogroups is listed in Table 2. The most frequent reactivity to a serovar was to sv. Nigeria sg Pyrogenes at 19.8% (95% CI 15.9–24.4%, n = 71) which was statistically significantly higher than the prevalence of other serovars (p<0.0001) (Table 2). The second highest prevalence was caused by the sg Sejroe represented by sv Hardjobovis and sv Wolffii with 5.6% (95%CI 3.5–8.6) and 5.3% (95%CI 3.3–8.3), respectively. A very low or nonexistent prevalence was found against sv Bratislava sg Australis with 1.9% (95%CI 0.9–4.2), against sv & sg Grippotyphosa with 0.1% (95%CI 0.0–0.1), and sv & sg Icterohemorrhagiae and sv Patoc sg Semaranga with both 0.0% (95%CI 0.0–1.0).
The seroprevalence did not differ between the sub-counties, but the individual sub-county samples sizes were very small (S1 Fig). Furthermore, the seroprevalence did not differ by age group with 35% (54/155) of those aged 18–29 years, 29% (26/91) aged 30–39 years, 32% (19/59) aged 40–49, and 34% (16/47) aged ≥50 years being seropositive (P = 0.78) (Table 1).
Antibody titers against any serovar ranged from zero to 1:3200. The seven participants with titers ≥1:800 had solely antibodies against sv Nigeria (Fig 1). Therefore, the prevalence of probable recent leptospirosis was 1.9% (95% CI 0.9–4.2%) and was uniquely related to sv Nigeria. Eight (6.4%) of 125 seropositive participants had either been exposed to multiple serovars and/or their sera cross-reacted in the MAT. There were double exposures (or cross-reactions) between Nigeria and Wolffi (n = 1), Nigeria and Butembo (n = 1), Nigeria and Bratislava (n = 2), Nigeria and Hardjobovis (n = 2). Further, between Wolffi and Hardjo (n = 1) and Bratislava and Butembo (n = 1).
All exposure variables (“risk factors”) listed in Table 1, including demographic, past medical history, and occupational/behavioral exposures were tested for a statistically significant association (p≤0.05) with the outcome “probable recent leptospirosis” by bivariable and multivariable analysis. Having been diagnosed with malaria in the past year was statistically significantly associated with being a case of probable recent leptospirosis (P = 0.048) in the chi-square test. However, in multivariable analysis, the association lost its statistical significance (OR 6.5, 95% CI 0.8–5, P = 0.085). All other health history risk factors, including having had fever, a fever that recurred after treatment, brucellosis, or typhoid in the past year, were not associated with probable recent leptospirosis in either analysis.
We further assessed risk factors listed in Table 1 for seropositivity against sv Nigeria, sv Hardjobovis, sv Wolffi, sv Butembo and sv Bratislava and for seropositivity against any of the eight serovars, by bivariable (Table 1) and multivariable analysis (Table 3). In the bivariable analysis, six participants who reported having skinned cattle in the two weeks prior to their blood sample had an odds ratio (OR) of 10.6 (95% CI 1.2–91.9, P = 0.032) for being seropositive against any serovar compared to participants who had not skinned cattle. People who reported monkeys living near their home had 2.1 (95% CI 1.2–3.7 P = 0.018) and 1.8 (95% CI 1.1–3.0 P = 0.013) times the odds of seropositivity against sv Nigeria and any serovar, respectively, compared to persons without monkeys nearby.
The statistically significant risk factors in the multivariable analysis are listed in Table 3. Persons involved in skinning cattle had 9.8 (95% CI 1.0–96.0 P = 0.048) times the odds of seropositivity against sv Bratislava and 12.28 (95% CI 1.39–108.58 P = 0.024) times the odds of seropositivity against any serovar, once adjusted for the effect of having monkeys near the home. Individuals with monkeys near their home had 2.05 (95% CI 1.1–3.7 P = 0.018) and 1.92 (95% CI 1.2–3.1 P = 0.009) times the odds of seropositivity against sv Nigeria and any serovar respectively, once adjusted for the effect of skinning. Other variables tested, including the “livestock-contact” composite variables, ownership of livestock, consumption of animal products, wildlife exposures other than monkeys around the home, and frequency of forest visits were not significantly associated with seropositivity at the individual serovar or all serovar levels in the bivariable and multivariable analysis. Similarly, no association was found between housing materials or drinking water origin and Leptospira seropositivity.
We found 35% prevalence of serum antibodies against eight Leptospira serovars in humans in rural western Uganda. We further detected a seroprevalence of 20% against sv Nigeria sg Pyrogenes, which was the most frequent serovar or serogroup for exposure. Historically, sv Nigeria was isolated from bovine kidneys in Nigeria [21]. The high prevalence of antibodies against sv Nigeria raises concern for a bovine-human transmission pathway in western rural Uganda that deserves further examination. Despite their lower seroprevalence, the presence of antibodies against cattle-associated serovars Wolffi (5%) and Hardjobovis (6%) of the sg Sejroe also strengthen the cattle transmission hypothesis. [22–24].
In addition to the high overall seroprevalence, the prevalence of probable recent leptospirosis was 1.9%. The 1.9% of individuals with titers reflective of probable recent leptospirosis further emphasizes the potential public health relevance of serogroup Pyrogenes in Uganda. If the MAT cut-off for probable leptospirosis was lowered to a titer of ≥1:400 [25], which has been used in other studies, 5.6% (95% CI, 3.5–8.6%) of participants would fit the case definition of probable recent leptospirosis. In the bivariable analysis, probable recent leptospirosis was statistically significantly associated with the person having had malaria in the past year. One possible explanation for this association is the oft-lamented concern that leptospirosis may frequently be misdiagnosed as malaria. However this association should be considered cautiously, as fever and past medical history were purely self-reported variables, and may have significant inherent error and recall bias.
Had all participants specifically come to the health centers due to acute febrile illness, the prevalence of probable recent leptospirosis might have been higher, given that Biggs et al. [13] found that 9% of patients had confirmed leptospirosis who were admitted for acute febrile illness at two health centers in northern Tanzania. Although such clinical inclusion criteria were not used in our study due to ethical concerns, our estimates were more representative of the general population instead of a selected sub-sample of persons presenting only with febrile illness. However, some serovars of the species L. borgpetersenii have been shown to cause primarily asymptomatic infection in humans, making further characterization of sv Nigeria essential to understanding its importance in this community.
Of similar importance is the 0% seroprevalence to L. interrogans sv. Icterohemorrhagiae, traditionally associated with severe clinical infection. The ecology of different serovars has not been described for the study region, but possible explanations include that this serovar may not be present in this region, or that it is less likely to be found in relatively healthy, outpatient clinic visitors. This emphasizes the importance of further study of leptospirosis in clinically ill patients presenting with undifferentiated febrile disease.
Skinning of cattle was associated with seropositivity. Skinning is a biologically plausible risk factor for Leptospira seropositivity, as contact with contaminated urine is possible during the slaughter process. Why skinning and not the actual slaughter of cattle was associated is unclear, but may be due to a higher duration of exposure during the more intricate skinning process as opposed to the relatively brief exposure when cattle are killed. Similarly, skinning has been found to be the strongest risk factor for seroconversion in sheep slaughtering abattoirs in New Zealand [26]. The low sample size (n = 6) led to limited power to analyze risk factors and may be responsible for either artificially exaggerating the association with skinning or minimizing the association with other high risk behaviors. Given the consistency of these results with other studies, overall the risk factor analysis supports the bovine-human transmission hypothesis.
The association between having monkeys living near the home and seropositivity for leptospirosis is more ambiguous. These individuals may live in more remote areas with greater wildlife contact in general, thus giving them more exposure to wildlife classically associated with leptospirosis. The possibility of a more direct role for monkeys in the transmission of leptospires to humans in Uganda is not established here but could also be further explored. Leptospirosis is commonly associated with contact to rodents, however the ubiquitous exposure to rats limited the ability of the study to assess the role of rodent exposures.
The study design and the sampling approach have limitations which may affect generalizability. First, the participants of this cross-sectional study were a convenience sample of those who attended two health clinics for a variety of health problems. Although each Health Centre IV has a theoretical catchment area that covers half of Hoima District, in practice the large majority of patients came from the subcounties immediately surrounding each Health Centre. As a result, the more rural parts of Hoima, including the Kyangwali refugee camp, were highly underrepresented. Furthermore, the study population included only those ≥18 years old and women were overrepresented (71%), whereas in general the population of Hoima is 50% female. In rural areas, women were expected to have a lower Leptospira seroprevalence than men, due to differences in occupational exposures (i.e. high-risk livestock activities), but the overall seroprevalences in women (32%) and men (34%) were similar. Persons working in livestock were also under-represented, and more systematic population sampling may have revealed an even higher seroprevalence.
Alternatively, the seroprevalence may have been overestimated, as this was a predominantly outpatient clinic population and patients were more likely to be ill compared to the population at large. Since many of the participants had fever symptoms in the past year (70%), it is possible Leptospira seroprevalence was higher in the study population than in the general population. However, fever is a very common symptom, especially in rural Africa; hence it is plausible that 70% of the general population will have experienced a fever episode within one year.
The limited sampling period may also have affected the seroprevalence. March-May is traditionally a rainy season within Hoima, which could potentially increase the rates of acute leptospirosis. However, the season was drier than usual, and there were no episodes of flooding in the town center or health centers. Furthermore, while such immediate weather conditions might be reasonably blamed for acute, high titer cases; overall seropositivity for exposure should be less affected. Hence, there were many potential reasons for the study to have underestimated or overestimated community-level seroprevalence.
The definition of the variables involving contact with cattle (including skinning) were not ideal, as the exposure time was set for the last two weeks prior to the interview. The time span may have been too short for some participants to seroconvert in cases of exposure to Leptospira through animal contact. However, most people who endorsed having had contact with cattle within the two weeks prior to the interview will most likely also have had similar prior contact, as skinning or butchering are usually regular activities and antibodies may persist between several months and years [2,27]. Others may have had more remote exposures to these activities that were not captured by the survey. Finally, some of the analyzed exposures occurred rarely, leading to low power for the risk factor analysis (Table 1).
Since laboratory capacity for the MAT did not exist in Uganda, antigens and antisera were imported from the WHO reference laboratory in Holland, and the size of the serovar panel was limited. The chosen panel was not large enough to cover all the common serogroups, and may not encompass all local strains. Hence, the overall prevalence may have been underestimated in this study. Since testing was targeted towards past exposure to leptospires and not acute disease, a MAT sensitivity of 88% and specificity of 98% can be assumed [25]. Therefore, the tested prevalence is an “apparent seroprevalence” and will most likely be slightly underestimated.
An undisputed limitation is the lack of clinical data during the sampling phase. It would have been highly informative to analyze the association between febrile illness and high leptospirosis titers. Due to ethical concerns with collecting clinical data without the ability to immediately diagnose (diagnostic capacity for MAT or PCR was unavailable in Uganda during the sampling period) or treat individual patients, data on clinical status such as the presence of fever, subsequent symptoms or symptom duration was not collected. However, as a result of the capacity building component of this study, laboratory diagnosis of leptospirosis will be available to support future clinical studies.
This initial study on leptospirosis in Uganda raises several research questions of interest for future studies. We recommend further exploration of the “bovine-human transmission pathway” by testing sera (MAT) and urine/kidneys (PCR) of bovines and sera of humans working in their proximity, either in farming/pastoral communities or in abattoirs. Additional testing of rat kidneys for the presence of leptospires may help evaluate the significance of the bovine-human transmission pathway relative to classically described murine-human transmission. In order to estimate the burden of leptospirosis in Uganda and the clinical importance of prevalent serovars, a study in patients with acute febrile illness would also be useful. A case-control study within this study population could assess risk factors in their community, such as location, flooding, contact with different animal species and occupational activities. Yet, before launching a large surveillance study of specific serovars for a new region/country, one consideration might be first sampling abattoirs as likely hotspots of exposure. In testing abattoir workers’ serum for an extensive range of serovars, one might determine which serovars are circulating in a community and then perform targeted testing of acute febrile illness patients using a less expansive, less costly panel. Working in such hotspots may also facilitate isolation of actual pathogenic leptospires.
This is to our knowledge the first article reporting on the prevalence of antibodies against Leptospira serovars in humans in Uganda. The 35% prevalence of antibodies to Leptospira suggests frequent exposure to this pathogen, in particular the Nigeria serovar of the Pyrogenes serogroup. Given this exposure, leptospirosis may have a greater impact on the health of this population than previously recognized. Further research is needed to understand the public health impact of leptospirosis in Uganda.
|
10.1371/journal.pcbi.1007023 | Energetic costs of cellular and therapeutic control of stochastic mitochondrial DNA populations | The dynamics of the cellular proportion of mutant mtDNA molecules is crucial for mitochondrial diseases. Cellular populations of mitochondria are under homeostatic control, but the details of the control mechanisms involved remain elusive. Here, we use stochastic modelling to derive general results for the impact of cellular control on mtDNA populations, the cost to the cell of different mtDNA states, and the optimisation of therapeutic control of mtDNA populations. This formalism yields a wealth of biological results, including that an increasing mtDNA variance can increase the energetic cost of maintaining a tissue, that intermediate levels of heteroplasmy can be more detrimental than homoplasmy even for a dysfunctional mutant, that heteroplasmy distribution (not mean alone) is crucial for the success of gene therapies, and that long-term rather than short intense gene therapies are more likely to beneficially impact mtDNA populations.
| Mitochondria, best known for their role in energy production, are crucial to the survival of most of our cells. To respond to energetic demands and mitigate against mutational damage, cells control the mitochondrial populations within them. However, the character of these control mechanisms remains open. As experimental elucidation of these mechanisms is challenging, theoretical approaches can help us understand the general principles of cellular control of mitochondria in physiology and disease. Here, we use stochastic modelling to compare control strategies by studying their impact on the dynamics of mitochondrial DNA (mtDNA) populations as well as their energetic burden to the cell. We identify optimal strategies for the cell to control against mtDNA damage and preserve energy production and use this theory to explore the action of recently developed mitochondrial gene therapies, which reduce the fraction of mutant mtDNA molecules inside cells. We show how treatment efficiency may depend on pre-treatment distributions of mutant and wildtype mtDNA molecules: treatments are less effective for tissues consisting of cells with highly varying mutant levels, and long-term, rather than short intense, gene therapies should be favoured.
| Most human cells contain 100-10,000 copies of mitochondrial DNA (mtDNA) which are situated inside the mitochondria. The proteins encoded by mtDNA are crucial for mitochondrial functionality, and mutations in mtDNA can cause devastating diseases [1–6]. Heteroplasmy, the proportion of mutant mtDNA molecules in a cell, typically has to pass a certain threshold (∼ 60-95%) before any biochemical defects can be observed [7–14]. The existence of thresholds at which mutant loads begin to have an effect has profound implications for our understanding of disease onset, drawing attention to the variance dynamics of the mutant fraction in cellular populations. As this variance increases more cells can be above threshold, and thus show pathology, even if average mutant load is unchanged.
Mitochondrial biogenesis and maintenance require cellular resources, and mitochondria are key sources of ATP and play other important metabolic roles. The particular ‘effective cost’ that cellular control of mitochondria acts to minimise remains poorly understood: for example, both decreases [15] and increases [15, 16] in wildtype copy numbers have been observed for different mutations as the mutant load increases. Some studies suggest that mtDNA density is controlled [17–19], others that total mtDNA mass [20, 21], or mtDNA transcription rate [22] is controlled. Understanding mtDNA population dynamics inside cells, and how these populations react to clinical interventions, is crucial in understanding diseases [23, 24]. However, experimental tracking of mtDNA populations over time is challenging, necessitating predictive mathematical modelling to provide a quantitative understanding.
In parallel with efforts to elucidate cell physiological control, protein engineering methods to artificially control mtDNA heteroplasmy are making fast progress. Two recently developed methods for cleaving DNA at specific sites involve zinc finger nucleases (ZFNs) and transcription activator-like effector nucleases (TALENs) [25–31], which have been re-engineered to specifically cleave mutant mtDNA [32–36]. MitoTALENs have been successfully used to reduce mutant loads in cells containing disease-related mutations, but elimination of the target mutant mtDNA was not complete [32, 37]. Similarly, treating cells multiple times with mtZFNs led to near-complete elimination of mutant mtDNAs [35, 36]. Quantitative theory for these therapeutic technologies has not yet been developed, leaving open questions about how these tools can be optimally deployed.
In this paper, we develop theory from bottom-up bioenergetic principles which allows us to study the effects of distinct cellular mtDNA control strategies, to analyse the bioenergetic cost of different mtDNA states, and to combine mtDNA control and energy-based cost to identify optimal control strategies for the cell. Finally, we construct a model for therapeutic mtDNA control using recent experimental data [36] and highlight challenges linked to heteroplasmy variance.
We employ a linear form of mtDNA feedback control and assume each mtDNA molecule replicates and degrades according to Poisson processes with rates λ and μ, respectively. Because control of biogenesis or autophagy yield similar behaviours [38], we assume that the degradation rate μ is constant and that feedback control is manifest through the replication rate λ(w, m), where w and m denote the number of mutant and wildtype mtDNA molecules in the cell. To connect with experiments, we use μ ≈ 0.07 day−1 corresponding to a half-life of about 10 days [39]. We only model post-mitotic cells, though our analysis can be extended to include cell divisions.
Specifically, we use a birth rate of the form:
λ ( w , m ) = μ + c 1 ( w o p t - ( w + δ m ) ) (1)
where c1 > 0, wopt > 0 and δ are constants, with wopt denoting the steady state value towards which the effective population, here defined as w + δm, is controlled. The magnitude of c1 determines how tightly the population is controlled. We use the term ‘mitochondrial sensing’ to describe how the cell might sense the mitochondrial population that is present. ‘Mutant sensing’ then refers to how strongly mutants are sensed relatively to wildtypes, which is encoded in the parameter δ. When steady state is reached (i.e. w + δm = wopt), replication and degradation rates are equal. In the absence of mutants, the resulting wildtype steady state is assumed to be optimal. We note that assuming the existence of wopt does not imply a control based on copy number. Other quantities related to mitochondria may be controlled instead, such as total mitochondrial mass or ATP production, their desired values being reached at an effective population size of wopt. Thus, we define ‘mitochondrial sensing’ to refer to a wide range of mechanisms available to the cell to infer properties of its mitochondrial population, which can then be used to decide on a control action.
The deterministic dynamics resulting from this control are described in Eq (4). We do not include the possibility of de novo mutations but our approach can straightforwardly describe the subsequent behaviour if new mutations arise. Our linear model shares features with the ‘relaxed replication model’ [40, 41] (Eq (5)), though is written in a simpler form. The relaxed replication model has been used in a variety of other models [42, 43] and has obtained experimental support [15].
We will first investigate properties of more general control strategies, after which we return to our linear control and discuss parameterisations that optimise the energy status of the cell. Finally, we use the linear control to fit recent experimental data involving treatment of heteroplasmic cells with mtZFNs.
Next, to find general quantitative principles underlying mitochondrial energy budgets, we build a cost function that assigns a cost to any given mtDNA state (w, m) and allows a general quantitative investigation of the tradeoffs in maintaining cellular mtDNA populations. The ‘true’ energy budget of a cell with a given mitochondrial population is highly complex, involving many different metabolic processes in which mitochondria are involved [44–46]. We provide a simpler description, focussing on ATP production as a central mitochondrial function, and removing kinetic details in favour of a coarse-grained representation, to provide qualitative rather than quantitative results.
In this work, we have built a quantitative theory bridging stochastic optimal control, costs of mtDNA populations, and gene therapies. Our results contribute to a growing body of evidence [63–66] that the variance of mtDNA populations has important physiological and therapeutic implications independently of mean heteroplasmy, and underline that stochastic theory is required to understand this biologically and medically important quantity.
Key findings of our model (Table 1) include (I) the identification of tradeoffs in the control of one or the other mtDNA species; (II) the observation that increasing mtDNA variance can lead to increased energetic costs over time and ageing even when means and demands are preserved; (III) intermediate heteroplasmy states can be more expensive than states homoplasmic in either mutant or wildtype; (IV) mutant sensing can be required to avoid an exponentially increasing cost; (V) sensing of cellular energetic status can be more effective than other targets like mitochondrial mass; (VI) reduction of mutant mtDNA alone is not always the optimal control strategy; (VII) high heteroplasmy variance challenges gene therapy treatments; and (VIII) weak, long gene therapy trajectories are more effective than short, intense ones.
Our findings hold qualitatively under the range of conditions we discuss above. The aim of our manuscript is not to make detailed quantitative predictions and conclusions based on complex models, nor do we intend to imply that our models are the only possible models one could construct. Rather, we aim to provide general biologically plausible models to gain qualitative insights and to comment on large-scale behaviours. To this end, our cost function, used to illustrate some of our results, is phenomenological and contains several parameters. Most of these are biologically interpretable, meaning their values can be obtained or estimated from the literature. The main elements in our cost function are quite general: terms involving supply, demand, and resource.
To test the qualitative shape of our cost function, one could sort cells based on mitochondrial copy number and heteroplasmy to obtain samples at different points in (w, m) space. Measurements of e.g. cell proliferation, ROS or apoptosis rates allow for the evaluation of an effective cost at each of these points. By measuring the relative consumption rates of NADH and succinate, as well as the amount of ATP produced per glucose consumed, in identical cells exposed to different energy demands, the saturating output model may be probed.
If the parameter δ is low, i.e. mutants are sensed less, mutant copy numbers at high heteroplasmies will be higher than wildtype copy numbers at low heteroplasmies. Experimentally, it has been observed that heteroplasmic cells can have total mtDNA copy number values that are 5-17-fold higher compared to cells homoplasmic in wildtype [67–70]. The cell has somehow allowed these mutants to expand, which may mean that they are less tightly controlled; controls based on total energy output or mtDNA mass (which can result in δ < 1) may lead to such behaviours. A control on mtDNA mass could explain why deletion mutants are often seen to expand [71, 72] and would also predict normal copy number levels in cells harbouring mtDNA point mutations. Recently, it was found that samples with mtDNA indels had very high mtDNA copy number levels, but single nucleotide variants did not [73].
We showed that heteroplasmy distributions in cell populations can provide important information about the possibility of successfully treating these cells with endonucleases. A tissue may be harder to treat if its high mean heteroplasmy level is caused by a small percentage of dysfunctional cells. Experimental values of mean homogenate heteroplasmy in heart tissue of patients with the 3243A>G mutation are roughly around 0.8 (though ranges can be large [74–77]) and muscle tissue often shows mosaic structures, with deficient patches of cells adjacent to healthy cells. These examples show that it may be that, at least in some cases, high mean levels are indeed caused by a relatively low percentage of cells, meaning that there are still challenges ahead for efficiently treating these tissues.
One of the features of our cost function is that resource limitations play an important role in shaping the cost landscape. There are indications that cellular levels of NAD (a coenzyme involved in oxidative phosphorylation) are limiting, and that a sufficient supply of NAD to mitochondria becomes critical [78–81]. An increase of intracellular NAD can lead to an increase in oxygen consumption and ATP production [81] indicating that resource limitation may, at least in some cases, be a genuine constraint. Adding various kinds of resources can significantly change mitochondrial basal respiration rate [82–84].
Like any other model, our models have a defined range of applicability. A key baseline assumption was using identical replication and degradation rates for mutants and wildtypes. Various possibilities of distinct rates have been offered in the literature, including faster mutant replication rates [22, 68, 85–88], lower mutant degradation rates [89], and higher mutant degradation rates [90, 91]. Including such differences, and other features such as de novo mutations, degradation control, and cell divisions [38, 64, 92, 93], constitute natural extensions to our theory.
Wildtype and mutant mtDNA copy numbers are considered to have birth rate λ(w, m) = μ + c1(wopt − (w + δm)) and death rate μ, leading to the following evolution equations:
d w d t = w ( λ ( w , m ) - μ ) d m d t = m ( λ ( w , m ) - μ ) (4)
The corresponding stochastic system, required to e.g. describe fixation, does not have an explicit solution due to nonlinearities. The deterministic steady state solution of Eq (4) is given by (wss + δmss) = wopt and represents a straight line in (w, m)-space (S1A Fig), whose slope depends on the value of δ. Stochastic dynamics will fluctuate around the steady state line, causing heteroplasmy to change over time until fixation of either species occurs. This means that, over long times, a cell will reach either h = 0 or h = 1 (in the absence of mutations). When mutations do occur, a cell will always reach a state with h = 1 (though many different mutant species may be present).
The relaxed replication model assumes a constant death rate μ and a birth rate of the form
λ ( w , m ) = μ w + m ( α R [ w o p t - ( w + η m ) ] + w + η m ) (5)
with αR > 1 and η constants [40, 41]. We have renamed the parameters of the original model for convenience. Note that both αR and η influence the mutant contribution to λ(w, m) (rather than the single parameter δ in our linear model).
Let the cost per unit time of state (w, m) be denoted by C, and the cost corresponding to the steady state (wss, mss) by C ¯. Even if steady state copy numbers are constant over time (i.e. the mean values of w and m are always equal to wss and mss) the mean cost per unit time is generally not equal to C ¯. By performing a Taylor expansion, the mean cost per unit time can be written as follows:
E [ C ] ( t ) ≈ C ¯+ 1 2 ( var ( w ( t ) ) ∂ 2 C ∂ w 2 + var ( m ( t ) ) ∂ 2 C ∂ m 2 + 2 cov ( w ( t ) , m ( t ) ) ∂ 2 C ∂ w ∂ m ) (6)
where E[C](t) is the expected cost per unit time given that the trajectory starts in state (wss, mss), and all partial derivatives are evaluated at steady state. These findings imply the following: suppose all cells in a population of cells are initialised in a state with minimum cost (corresponding to some specific number of mutant and wildtype mtDNA molecules). At some later time, the mtDNA populations in the different cells will have drifted apart and even if mean copy numbers (averaged over all cells) of w and m are identical to their initial values, the increase in variance between cells means that the overall mean cost (averaged over all cells) is higher than it was initially.
We assume that the net energy supply per unit time in a state (w, m), called S(w, m), involves the following four terms: (i) the energy output per unit time (si) produced by the mitochondria; (ii) a maintenance cost per unit time (ρ1) to maintain the mitochondria, as their presence imposes some energetic cost (e.g. mRNA and protein synthesis); (iii) a building cost (ρ2) for the biogenesis of new mitochondria; and (iv) a degradation cost (ρ3) to degrade mitochondria. We will assume that every mtDNA molecule is associated to a particular amount of mitochondrial volume which we refer to as a ‘mitochondrion’ (section 4 in S1 File).
At any time, mitochondria experience a certain energy demand and to meet this demand they need to have a certain resource consumption rate ri (where i = w, m refers to wildtype or mutant). Here we use the term ‘resource’ as an amalgamation of the substrates used for the oxidation system. We need to specify the relationship between the power supply (s) and the rate of resources consumed (ri) by mitochondria. We use two different models s(ri) which are discussed further in section 3 in S1 File s ( r w ) = ϕ ( r w - β ) s ( r w ) = 2 s m a x 1 + e - k r w - 1 . 1 s m a x (7)
where ϕ, β, k and smax are constants respectively describing the mitochondrial efficiency, a basal proton leak-like term, the saturation rate of the efficiency, and the maximum power supply (section 4 in S1 File).
We assume that pathological mutants can have a deficient electron transport chain (which may support a smaller flux leading to a lower resource consumption rate for mutants and therefore a lower ATP production rate) and a lower energy production efficiency, leading to the following mutant energy output: ϵ2s(ϵ1rw). Here, ϵ1, ϵ2 ∈ [0, 1] describe the mutant resource uptake rate and the mutant energy production efficiency relative to that of a wildtype, respectively. In the main text we set ϵ2 = 1; other values of ϵ2 are discussed in section 4.7 in S1 File.
The mitochondrial maintenance cost is denoted by ρ1 and corresponds to the energetic cost required to maintain the mitochondrion that contains the mtDNA. This energetic costs involves factors like the synthesis and degradation of mitochondrial proteins and enzymes. We assume the maintenance cost is the same for wildtype and mutant mitochondria (though for some mutations this is quite possibly not the case). The net energy supply per unit time, S(w, m), then follows as Eq 3.
To determine the value of rw for a given state (w, m), we first check whether the demand D (which we assume is a constant) can be satisfied by supply S(w, m). If it can, we set Eq (3) equal to D and solve for rw, i.e. we assume that if possible, the mitochondria will exactly satisfy demand. It may, however, not be possible to satisfy demand, which can be because of two reasons: i) there are not enough mitochondria present to produce enough energy, or ii) the resource supply rate, R (a constant), is not enough to meet demand. In the former case, we set rw = rmax (a specified maximum resource consumption rate per mitochondrion): the mitochondria work as hard as possible to keep their energy output closest to demand. In the latter case, we assume that the total available resource supply is shared equally between the mitochondria: r w = R w + ϵ 1 m. Further details of the cost function are given in sections 3–5 in S1 File.
The parameters used in our cost function are summarised in S2 Table and motivated in section 4 in S1 File. Despite our model being simple, most parameters are biologically interpretable.
Experimentally, cells are transfected with two mtZFN monomers: one which binds selectively to mutant mtDNAs, and one that binds mutants and wildtypes with equal strength [62]. We simplify this picture by assuming an ‘effective’ mtZFN pool and use [ZFN] to denote its concentration. The increase in mtDNA degradation rate caused by the mtZFNs is then assumed to be proportional to [ZFN].
Nucleases are imported into the cell and then degrade over time, meaning that their concentration in the cell (and in the mitochondria) may be approximated by an immigration-death model:
d [ Z F N ] ( t ) d t = I ( t ) - μ z [ Z F N ] ( t ) (8)
where I(t) and μZ are the immigration and death rates of the effective mtZFN pool, respectively. In recent experiments [36], nucleases are expressed for short times meaning that the immigration rate will increase sharply at the start of the treatment after which it decreases over time: we chose to model I(t) as an exponentially decaying function, I(t) = I0e−bt, where I0 denotes the initial rate directly after the treatment is initiated and b is a constant describing the duration of the treatment. The mtZFN concentration now becomes
[ Z F N ] ( t ) = I 0 μ z - b ( e - b t - e - μ z t ) (9)
which is shown for various parameter values in S8A Fig. The data we use to fit our models concerns heteroplasmy and total copy number measurements over four rounds of treatment, each treatment consisting of mtZFN transfection followed by a 28-day recovery period. During this recovery period, total copy numbers recover their initial values due to cellular feedback control. The increase in mtDNA death rate due to the presence of the mtZFNs, μZFN, is given by
μ Z F N ( 28 · i < t < 28 · ( i + 1 ) ) = μ + ∑ j = 0 i [ Z F N ] ( t - 28 · j ) (10)
where i = 0, 1, 2, 3 indicates the treatment round. This equation is simply stating that new mtZFNs are added every 28 days. Death rates for m and w are now assumed to be
μ ( t ) w = μ + ξ · μ Z F N ( t ) μ ( t ) m = μ + μ Z F N ( t ) (11)
where μ denotes the baseline degradation rate and ξ represents treatment selectivity (e.g. when ξ = 0 there is no off-target cleavage).
To fit our nuclease model to recently obtained experimental data [36], we use Eq (4) with μ replaced by μ(t)w or μ(t)m and λ(w, m) given by Eq (1):
d w d t = w [ c 1 ( w o p t - ( w + δ m ) ) - ξ · μ Z F N ( t ) ] d m d t = m [ c 1 ( w o p t - ( w + δ m ) ) - μ Z F N ( t ) ] (12)
Total mtDNA copy numbers in pre-treatment 80% heteroplasmy cells were measured using quantitative PCR (section 6.4 in S1 File) and were found to be 889 ± 214 (S.E., n = 3). We therefore assume an initial total copy number of 900, meaning w and m were initialized at 0.2 ⋅ 900 = 180 and 0.8 ⋅ 900 = 720, respectively. These evolution equations incorporate cellular feedback control as well as the nuclease treatment which occurs in cycles of 28 days. The mtZFN degradation rate was assumed to be μz = ln(2) day−1, corresponding to a half-life of 1 day. This is in accord with the experimental observation that almost no mtZFN was present 4 days post-transfection (with a half-life of 1 day, only 6% of initial copy numbers remain after 4 days).
MCMC inference was performed using the Python package Pymc3, a package designed for Bayesian statistical modelling and probabilistic machine learning [94]. A Gaussian error model was assumed, i.e. the observed heteroplasmy y i ( h ) and total copy number y i ( T ) data are given by
y i ( h ) = y ^ i ( h ) + N ( 0 , σ h 2 ) y i ( T ) = y ^ i ( T ) + N ( 0 , σ T 2 ) (13)
where y ^ i ( h ) and y ^ i ( T ) denote our predicted heteroplasmy and copy number values obtained by numerically solving Eq (12), and we allow for different noise variances for h and T (in general, different experimental errors are expected as different methods are used to measure h and T). A metropolis sampler is used for parameter estimation. Maximum a posteriori (MAP) values were found to be ( I 0 , b , c 1 , ξ , δ , σ h 2 , σ T 2 ) M A P ≈ ( 122 . 82 , 46 . 68 , 1 . 90 × 10 - 4 , 0 . 72 , 1 . 26 , 0 . 061 , 0 . 10 ). Due to a degeneracy in our mtZFN dynamics model (section 6.5 in S1 File) the MAP values of I0 and b are not necessarily unique at large b (details in section 6.5 in S1 File).
We explore the ability of our model to account for additional data from Ref. [36] (Fig 5C and 5D) which was not included in our inference. Using the MAP values for parameters I0, b, c1, δ, σ h 2 and σ T 2 (based on the data shown in Fig 5A and 5B), the maximum likelihood estimate of ξ is obtained based on the additional data, using a Gaussian error model similar to Eq (13). This maximum likelihood value is ξ ≈ 0.15.
|
10.1371/journal.pbio.1000597 | EphrinB/EphB Signaling Controls Embryonic Germ Layer Separation by Contact-Induced Cell Detachment | The primordial organization of the metazoan body is achieved during gastrulation by the establishment of the germ layers. Adhesion differences between ectoderm, mesoderm, and endoderm cells could in principle be sufficient to maintain germ layer integrity and prevent intermixing. However, in organisms as diverse as fly, fish, or amphibian, the ectoderm-mesoderm boundary not only keeps these germ layers separated, but the ectoderm also serves as substratum for mesoderm migration, and the boundary must be compatible with repeated cell attachment and detachment.
We show that localized detachment resulting from contact-induced signals at the boundary is at the core of ectoderm-mesoderm segregation. Cells alternate between adhesion and detachment, and detachment requires ephrinB/EphB signaling. Multiple ephrinB ligands and EphB receptors are expressed on each side of the boundary, and tissue separation depends on forward signaling across the boundary in both directions, involving partially redundant ligands and receptors and activation of Rac and RhoA.
This mechanism differs from a simple differential adhesion process of germ layer formation. Instead, it involves localized responses to signals exchanged at the tissue boundary and an attachment/detachment cycle which allows for cell migration across a cellular substratum.
| The formation and maintenance of tissue boundaries is an essential feature of multicellular animals, including humans. Using the frog embryo as a model system, we describe a mechanism of tissue separation that involves repeated cycles of cell attachment and detachment at the boundary between two adjacent tissues. Molecularly, this mechanism is based on a signal exchange across the boundary by a system of receptors and ligands—EphB receptors and ephrinB ligands—that are both integral cell membrane proteins, and thus require direct cell contact for signaling. In this way, cell attachment-dependent contact induces signaling which leads to a temporary detachment, followed by reattachment and a next round of signaling. Such an attachment-detachment mechanism allows for cell migration along the boundary, while at the same time preventing invasion of the stationary tissue by the migrating one.
| When Townes and Holtfreter [1] observed the sorting of mixed embryonic ectoderm, mesoderm, and endoderm cells, they proposed that this segregation of germ layers and the consequent self-assembly of the basic body structure was based on mutual “tissue affinities.” This concept was later refined into Steinberg's [2] Differential Adhesion Hypothesis, which posited that simple adhesion differences between cell populations are sufficient for their separation and their positioning relative to each other, explaining, for example, the arrangement of germ layers in amphibian embryos [3],[4]. However, if the boundary between two germ layers also serves for tissue translocation, more specialized tissue separation mechanisms may be required.
In the Xenopus gastrula, mesoderm translocates across the ectodermal blastocoel roof (BCR), and the boundary between these two germ layers, Brachet's cleft, must permit this movement yet prevent invasion of the BCR by the migratory mesoderm (Figure 1A). Interaction with a sparse network of fibronectin fibrils controls the motility of mesoderm cells, but their adhesion to the BCR is fibronectin-independent [5]. In fact, BCR and mesoderm cells are in direct contact [6], and the same adhesion molecules, C- and XB/U-cadherin, are expressed in both tissues (reviewed in [7]).
This ectoderm-mesoderm boundary has been established as a model for tissue separation. Properties of the mesoderm and the BCR that underlie their separation can be studied in an in vitro assay (Figure 1A) [5],[8]. The molecular control of separation behavior is partly known. Non-canonical Wnt signaling downstream of the Wnt receptor Xfz7 [9], interaction of Xfz7 with paraxial protocadherin (PAPC) and with the ankyrin repeat domain protein 5 (xANR5), and activation of RhoA and Rho kinase [10]–[12] are involved in the implementation of a proximal yet still unknown cellular mechanism that actually generates the boundary.
A candidate for this proximal mechanism is ephrin/Eph signaling. Eph receptor tyrosine kinases are subdivided into EphA and EphB subclasses and their membrane-linked ephrin ligands correspondingly into ephrinAs and ephrinBs. Within subclasses, binding appears promiscuous, although some Eph receptors have a higher affinity for specific ephrins. Receptor ligation and clustering initiates “forward signaling,” but receptor-ligand interaction can also stimulate “reverse signaling” downstream of the ephrin ligand. Ephrin/Eph signaling has been implicated in boundary formation under conditions where receptor and ligand are expressed in complementary patterns [13]–[18]. For example, a model was proposed for rhombomere-boundary formation based on repulsive Eph-ephrin signaling, acting in parallel with an Eph-dependent regulation of adhesion within the rhombomeres [19],[20]. However, direct observation of repulsive behavior at the boundary, similar to what is classically seen during neuron guidance, has not been attempted. In the early Xenopus embryo, expression of several Eph receptors and ephrins has been reported (reviewed in [21]). Their in vivo functions during gastrulation have not yet been established, although gain-of-function experiments indicate that they can regulate cell adhesion, migration, and sorting [22]–[24]. Thus, ephrin signaling is a prime candidate to mediate tissue separation in the early Xenopus embryo.
To examine cell contact dynamics at the boundary in living tissues, membrane-labeled mesoderm explants and ectodermal BCRs were combined, and the reconstituted boundary was observed by confocal microscopy (Figure 1 and Movies S1–S10). Contacts within each tissue appeared tight, as cells remained apposed and moved in concert (Figure 1B–D, arrowheads, Movies S1 and S2). The same behavior was seen at the boundary between BCR and ectoderm explants: cells re-established contacts within minutes and remained stably apposed throughout the experiments (up to 2 h) (Figure 1D, G, arrows, Movies S2 and S6). In contrast, contacts between mesoderm and BCR cells appeared dynamic (Figure 1C arrow, E, Movies S1 and S3) with frequent detachments followed by re-establishment of contacts (Figure 1G).
The frequency and length of detachment phases was variable. Also, separation was not uniform over the length of the boundary (see, e.g., Figure 1B,E, Movies S1 and S3). While in close contact, mesodermal cells appeared to adhere to the BCR. They moved in concert with BCR cells, and during subsequent detachment, taut retraction fibers often spanned the gaps between cells (Figure 1C, Frame Insert 34a; Figure 1F, Frame 2a; Movies S1 and S3). A detachment event typically spread along the boundary, where existing contacts seemed to resist retraction (see, e.g., Movie S1, and Figure 1C and E). Detached cells from either side of the boundary can emit protrusions which probe the cleft (Figure 1C, Frame 34; Figure 1E, Insert Frame 9; Movies S1 and S3). While repeatedly retracting and attaching, mesoderm cells can move along the BCR into gaps produced by other retracting mesoderm cells (Movie S3). This suggests a mechanism for the normal collective migration of mesoderm on the BCR substratum and is consistent with the co-existence of close adhesive contacts and gaps between mesoderm cells and the BCR at the ultrastructural level [6]. Altogether, it appears that contact-induced detachment between ectodermal and mesodermal cells is at the base of a tissue separation mechanism supportive of cell movement at the interface.
We tested the role of ephrins and Ephs on tissue separation in an in vitro assay (Figures 1A, 2B). Test explants were placed on a dissected BCR, and the percentage of explants that remain separate after 1 h was determined. In this assay, wild type ectoderm explants all sink into the BCR, while the majority of wild type mesoderm explants remain distinctly separated, i.e. on the surface of the BCR. Note that after 45–60 min, the reaction is complete: explants either have fully integrated (usually within 15 min for wild type ectoderm) or will remain definitively separated, implying that for individual explants the response is all or none. However, when the percentage of explants remaining separated in a given experiment is counted, the overall outcome is graded, which can be shown by increasing concentrations of interfering reagents (e.g., Figure S3) and/or increasing incubation times with activating Fc fragments (e.g., Figure 2A).
We first performed gain-of-function experiments on ectoderm tissues by treating them with activating, pre-clustered extracellular ephrin-Fc fusion polypeptides. While untreated ectodermal BCR explants readily reintegrated into BCRs, test explants incubated with ephrinB2-Fc tended to remain separate (Figure 2A). The same result was obtained by treating the substrate BCR instead of the test explants. Separation increased with incubation time, but could be detected as soon as 2 min after treatment (Figure 2A), indicating a rapid and direct cellular response. Activation was reversible, as a 30 min wash significantly decreased separation behavior (not shown). Separation was strongest when both the substrate BCR and BCR test explants were treated with ephrinB-Fc, over the whole duration of the assay, suggesting that the effect is not due to the establishment of a difference between two cell population but rather to a change in cell contact behavior. The response is specific for ephrinBs; ephrinA-Fc had very little effect (Figure 2A). These results imply that an EphB receptor is present on ectodermal cells and that its stimulation by ephrinB can immediately induce tissue separation. Consequently, we determined the expression pattern of all known ephrinB and EphB receptor isoforms in the gastrula by RT-PCR (Figure S1): all isoforms were expressed in all germ layers, although at different levels; as an exception, ephrinB3 was restricted to the ectoderm. Eph receptor-ephrin ligand binding is largely promiscuous, providing ample opportunity for redundancy. Moreover, Eph/ephrin signaling has previously been implicated in boundary formation in situations where receptor and ligand are expressed in complementary patterns in vivo (e.g., [25]–[27]) or under experimental conditions (e.g., [24],[28]). With co-expression of receptors and ligands in each of two adjacent tissues, the question arises whether Eph/ephrin signaling can nevertheless be employed for their separation.
To address this issue, we performed loss-of-function experiments using antisense morpholino oligonucleotides (MOs) against ephrinBs and EphBs to knock down putative crucial factors for boundary formation. Interfering with ephrin/Eph signaling led to severe developmental defects (Figure S2D–F). Most strikingly, analysis of early gastrulae revealed a strong reduction of Brachet's cleft (Figure S2A–C). While disruption of the cleft alone will affect gastrulation movements and produce shorter tadpoles (Figure S2E), ephrin/Eph signaling appears to disturb additional processes (EphB depleted embryos die before hatching, Figure S2F). By using the in vitro assay (Figure 1A) and by directly examining cell behavior, we isolated its specific function in tissue separation. Ephrin B1 and B2 MOs indeed inhibited tissue separation when injected either in the mesoderm or in the BCR (Figure 2B,C; Figure S3A,B; unpublished data), showing that both ephrins are required on both sides of the boundary. EphrinB1 MO was slightly more efficient in the BCR (unpublished data) and ephrinB2 MO in the mesoderm (Figure S3A,B), consistent with relative expression levels of the two ligands (Figure S1). Inhibition of separation shows dose-dependence before reaching a plateau at around 30%–40% (Figure S3B). Knockdown of ectoderm specific ephrinB3 caused a similar partial inhibition (Figure S4A). In all cases, separation was rescued by co-injection of corresponding wild type ephrinB mRNA (Figure 2C and Figure S4A). Inhibition was not significantly increased in double knockdown of ephrinB1 and B2 in the mesoderm (Figure S3A,B), and the same was true for triple knockdown of all ephrinB1–3 in the BCR (Figure S4A). We conclude that all ephrinBs expressed in a given tissue are required to a degree related to their relative expression levels, but simultaneous depletion is not sufficient for complete inhibition of separation.
We next tested whether ephrinB2, which is enriched in the mesoderm, could induce separation when overexpressed in the ectoderm. We found that a significant number (∼40%) of ectoderm explants now remained separated from wild type BCR (Figure S4B). However, overexpression of ephrinB1, which is already abundant in the ectoderm, had no effect (Figure S4B). Both constructs are strongly expressed (unpublished data) and effectively rescue normal separation behavior (Figure 2C). These results show that increasing ephrinB2 levels is sufficient to trigger separation and suggests that formation of the ectoderm-mesoderm boundary relies at least partly on the differences in ephrin composition observed between these two tissues (Figure S1).
Since ephrinB2 is required in the mesoderm, overexpression in the BCR of a cytoplasmically truncated form of the cognate receptor EphB4 (ΔC-EphB4) should compete for ephrin B2 binding with all endogenous Eph receptors and inhibit forward signaling in this tissue. Expression of ΔC-EphB4 in the BCR did indeed diminish separation, to a degree very similar to that obtained by ephrinB loss-of-function in the mesoderm (Figure 2C). EphB4 MO mimicked this effect, both in the BCR and in the mesoderm (Figure 2C), demonstrating that EphB4 is required. We conclude that the B family ligands and receptors are required in both ectoderm and mesoderm for tissue separation.
In the Xenopus gastrula, PDGF-A is expressed in the ectoderm and its receptor, PDGFR-α, in a complementary pattern the mesoderm. However, inhibition of PDGF signaling does not affect formation of the boundary (Damm and Winklbauer, submitted manuscript) or separation behavior in the BCR assay (Figure S4C). Likewise, interaction with the fibronectin-rich matrix on the BCR, which together with PDFG-A signaling determines the direction of mesoderm cell movement across the BCR [29], is not required for tissue separation [30], emphasizing the dominant role of EphB/ephrinB signaling in this process.
Our results suggest that ephrinB/EphB signaling occurs within each tissue on either side of the boundary, or that two anti-parallel pathways signal across the boundary. To distinguish between these alternatives, we performed a series of systematic ephrinB/EphB double knockdown/inhibition/rescue experiments. We first established that we maximally inhibited ligand or receptor activities: coinjection of ephrinB1 and B2 MOs indicated saturation of ephrinB inhibition (Figure S3A,B). Moreover, the degree of inhibition upon expression of ΔC-EphB could not be increased by increasing mRNA levels (Figure S3C). Thus, even when a pathway was maximally inhibited on one side, separation was only partially impaired. However, separation could be further reduced by interfering with ΔC-EphB4 on both sides of the boundary (Figure 3A and Figure S3C). Increased inhibition was also obtained by downregulating ephrinB and EphB activity simultaneously in the mesoderm (Figure S3B) or in the BCR (unpublished data), but not by downregulating ephrinBs in one tissue and EphB in the other (Figure 2C). Since these double inhibition experiments were preformed under conditions of maximal EphB interference (Figure S3C), ephrinB inhibition should not have had an effect if in the same EphB pathway. Thus, the simplest interpretation of our results was that separation is controlled by the additive activity of two antiparallel pathways signaling across the boundary. Each pathway involves ephrin ligands on one side and Eph receptors on the other side.
We tested this hypothesis more directly using ephrinB-Fc fragments (Figure 3B,C), which allowed us to stimulate these pathways specifically at the tissue surface, i.e. at the boundary. The rationale was to deplete ephrins in one of the interacting tissues, and then restore the forward signal by treating the other tissue with ephrin-Fc fragments to trigger activation of Eph receptors at its surface (Figure 3B). When we activated EphB signaling in mesoderm explants by treatment with ephrinB2-Fc (which binds to all EphB receptors) and placed them on ephrinB1/B2 depleted BCR, robust separation occurred (Figure 3C). Thus, direct activation of EphB receptors at the mesoderm surface can rescue separation from ephrin-depleted ectoderm, implicating forward signaling from the ectoderm to the mesoderm in tissue separation. The complementary experiment—ephrinB2 depletion in the mesoderm and EphB activation at the surface of the BCR—also resulted in a rescue of separation (Figure 3C), indicating that forward signaling from the mesoderm to the ectoderm is similarly active during tissue separation. The experiments show that separation can be restored by activating forward signaling directly at the surface of adjacent tissues. Thus, the separation phenotype induced by ephrinB loss-of-function can be fully accounted for by an inhibition of signaling across the boundary. Altogether, full tissue separation requires two forward signals, one triggered by mesodermal ephrins binding to the EphB receptors at the surface of the ectoderm and a second one depending on ectodermal ephrins interacting with Eph receptors of the mesoderm.
EphrinB or EphB knockdown impedes cell detachment at the boundary (Figure 4). Compared to controls (Figure 4A), ephrinB2 MOs in the mesoderm (Figure 4B) dramatically decreased the frequency of attachment/detachment events (Figure 4G). Often, mesoderm cells remained apposed to BCR cells for the whole duration of the recording (Figure 4B, Movie S4), similar to ectoderm aggregates (Figure 4E, Movie S6). Mesoderm and BCR cells moved in concert, indicating stable contacts (Movie S4). Detachment was similarly inhibited when ephrinB1 was depleted from the BCR, or EphB4 from the mesoderm (Figure 4C,D,G, Movie S5). Thus, signaling in both tissues is required for cell detachment at the interface. Attachment/detachment cycles were rescued after ephrinB1 depletion in the BCR by incubating wild type mesoderm test explants with increasing doses of soluble ephrinB1-Fc fragments (Figure 4F,G), demonstrating that this behavior is an immediate reaction to ephrin-Eph signaling at the boundary.
The experiments above show that direct ephrin activation at the tissue interface accounts for repulsion between ectoderm and mesoderm. In our time lapse recordings, repulsion is never observed within the tissues, but ephrin/Eph signals may nevertheless affect cell-cell adhesion [18],[19],[20],[22]. We evaluated the effect of Eph/ephrin loss- and gain-of-function on adhesion of ectoderm and mesoderm cells, using a classical reaggregation assay. Aggregation of dissociated cells started within minutes (unpublished data), and mesoderm had formed smaller aggregates than ectoderm after 1 h (Figure S5), consistent with mesoderm cells being less adhesive [31],[32]. EphrinB2 or EphB4 depletion had no effect on the rate of mesoderm cell aggregation (Figure S5), suggesting that ephrinB/EphB activity does not contribute to mesoderm tissue cohesion. In the ectoderm, however, corresponding ephrinB1/EphB4 depletion diminished aggregation. Overexpression of ephrinB2 had a similar, although more variable, effect (Figure S5). Thus, while ephrinB2 overexpression induces tissue separation and ephrin/Eph depletion inhibits separation, both treatments reduce cohesion within the ectodermal tissue. Altogether, putative effects of ephrins and Eph receptors on cell-cell adhesion within tissues are not correlated with their roles in cell detachment at the tissue boundary.
RhoA and Rac are well-established downstream effectors of Ephrin/Eph signaling that modulate cytoskeletal dynamics. RhoA activity in the mesoderm had been implicated in separation behavior [11], but the role of RhoA in the BCR and of Rac in both tissues has not yet been addressed. We systematically tested the effects of manipulating RhoA or Rac function. Dominant negative N19RhoA and N17Rac both inhibited separation when expressed in either of the two tissues (Figure 5A,B). Because expression of constructs interfering with RhoA and Rac function may also have long-term indirect effects, we complemented these data with experiments using specific soluble inhibitors of Rac and of Rho-associated kinase, a direct target of RhoA (Figure 5C). A short incubation with these inhibitors was sufficient to cause mesoderm test explants to integrate into the BCR, thus mimicking the effect of the dominant negative GTPases. This immediate response to the inhibitors suggests that RhoA and Rac activities are directly required during the separation process.
We next asked if RhoA/Rac activation could rescue separation when ephrin/Eph signaling is impaired. Since we had demonstrated forward signaling in both directions, we inhibited signaling on the Eph receptor side, by injection of EphB4 MO in the mesoderm and ΔC-EphB4 in the BCR, and tested the effect of constitutively active forms of RhoA and Rac (V14RhoA,V12Rac), expressed at low levels in the same tissue. In both cases, separation could be efficiently rescued by both V14RhoA and V12Rac (Figure 5D,E). A weak rescue was also observed upon overexpression of wild type RhoA and Rac (Figure S6B). Constitutively active Cdc42 was unable to rescue separation (Figure 5D). We conclude that RhoA and Rac, but not Cdc42, function downstream of Eph signaling on both sides of Brachet's cleft. Since RhoA has also been proposed to act downstream of Xfz7/PAPC/xANR5 in the mesoderm to regulate tissue separation [11], we asked whether co-expression of Xfz7 and PAPC could rescue separation in ephrinB2+EphB4 MO-injected mesoderm. We did not observe any rescue (Figure S4C), indicating that ephrinB/EphB signaling acts downstream of or in parallel to Xfz7/PAPC.
We next examined the effect of RhoGTPases activity on repulsion at the cleft by live confocal microscopy (Figure 6). When dominant negative forms of RhoA or Rac were expressed in the BCR or in the mesoderm, the frequency of attachment/detachment was strongly reduced in all cases (Figure 6H). Most cells established stable contacts between ectoderm and mesoderm that were indistinguishable from contacts within tissues (Figure 6E,F). Conversely, when activated forms of RhoA or Rac were co-expressed with dominant negative EphB receptor, the rescue of tissue separation (Figure 5D,E) was paralleled by a rescue of detachment at the boundary (Figure 6A–D,G, Movies S7 and S8). However, compared to controls, repulsion at the boundary appeared more transient. Although the membranes of abutting cells often remained close to each other for prolonged periods, they were clearly detached, and cells were able to slide along the boundary (Figure 6B), indicating that they had failed to re-establish contacts. It appeared as if cells were locked in a detached state by active RhoA or Rac, but could not fully retract under these conditions.
Our results imply that RhoA and Rac should be locally activated at ectoderm-mesoderm contacts in an Eph signaling-dependent manner. We investigated the subcellular localization of active, GTP-bound GTPases in the mesoderm by expressing the GTPase-binding domains (GBD) of N-Wasp (a target of Cdc42/Rac) or Rhothekin (a target of RhoA) fused to GFP. These mesodermal test explants were combined with BCRs expressing mCherry and examined by spinning disk confocal microscopy. While these constructs can act as inhibitors of their respective GTPases, at low expression levels they are expected to accumulate at sites of high concentrations of active GTPases [33]–[35].
We observed accumulation of both GBDs at the ectoderm-mesoderm boundary (Figure 7A,B, arrows). Signal intensity was generally higher there than at cell contacts within the explants. We quantified the GBD distribution in all cells which were in contact with BCR cells (Figure 7D). While in controls, BCR-BCR contacts showed accumulation in less than 20% of the cells, about 60%–80% of mesoderm cells contacting the BCR scored positive for a high fluorescent signal, for both Wasp and Rhothekin GDBs. This was a significant increase (p<0.001) compared to GFP alone. EphB4 MO seemed to inhibit GBD accumulation at boundary contacts (arrowheads): the frequency of accumulation decreased to 30%–40% (p<0.001). A similar frequency was obtained for mesoderm cells expressing dominant negative GTPases (Figure 7D). We also observed a dramatic decrease in Wasp/Rhothekin-GBD boundary localization when ephrin B1 was depleted in the adjacent BCR (frequency around 20%, p<0.0001 compared to control MO) (Figure 7A,B,D). We have not yet been successful in performing the reciprocal experiment, i.e. imaging GBDs in the BCR. This tissue appears to tolerate expression of GBDs less. However, the data from mesoderm explants demonstrate an Eph signaling-dependent activation of GTPases at the boundary.
Time lapse microscopy of mesoderm cells expressing Wasp/Rhothekin-GBDs revealed that accumulation at the boundary was highly dynamic. As expected from the large variation in contact time and area (see above), and from the superposition of detachment and protrusion formation, we observed fluctuations that spanned a large range of intensities and were highly variable in frequency (Movies S9 and S10). Drastic changes often occurred between two frames, i.e. in less than 2 min, in agreement with a fast dynamics of GTPase signaling. Despite this variation, a correlation could nevertheless be seen between retraction and signal decay. We analyzed cells for which we could unambiguously determine a transition from intimate contact with the BCR to detachment. In almost all cases (14/16 cells for WaspGBD and 13/14 cells for RtknGBD, from four independent experiments) the membranes of mesoderm cells contacting the BCR showed significant accumulation of the GFP construct, which decreased once the cell had detached (Figure 8). Generally, retractions and GFP decay both happened from one frame to the next (e.g., Figure 8A). In some cases, retractions spanned several frames, which then correlated with a slower decay of the GFP signal (e.g., Figure 8B). Some of the changes in signal intensity that were observed in non-retracting cells may be due to small detachments not detectable at the resolution of the GFP-GBD signal. Also, GTPase activation events may simply not be successful in triggering detachment. Note that the fluctuations occurring asynchronously in neighboring areas (Figure 8A, small arrows). The signal from mesoderm-mesoderm (Figure 8B, thin arrows) contacts served an internal control for the absence of photobleaching in these recordings.
As for putative downstream targets of the GTPases, we visualized cytoskeletal F-actin, phospho-nonmuscle-myosinII, and microtubules. We did not detect any significant accumulations or depletions at the boundary for microtubules and P-myosin (unpublished data). F-actin was enriched at the boundary, and this was dependent on ephrin-signaling (Figure S7). However, the pattern did not fully correlate with separation behavior, as a significant decrease was observed in the presence of N17Rac but not of N19Rho. These data indicate that the cytoskeleton is indeed modulated at the boundary, although tissue separation must be mediated by processes that cannot be distinguished at the level of global F-actin distribution. They also suggest that Rac and RhoA have distinct effects downstream of ephrin signaling, although we did not detect evidence for synergy between RhoA and Rac at the level of separation behavior (Figure S6).
Previous attempts to explain the separation of ectoderm and mesoderm [3],[4],[36] have been based on a thermodynamic model involving the minimization of tissue surface free energy. It assumes that two respective cell types are intrinsically different in terms of cell adhesiveness or cortical tension, and that this difference can drive cell sorting, boundary formation, and tissue positioning, analogous to the phase separation of immiscible fluids [2],[36]–[38]. Here we have shown evidence for a different model, in which signaling across the ectoderm-mesoderm boundary is crucial to locally regulate cell detachment and eventually tissue separation. Thus, although all cells of the two populations have the potential to form a cleft-like boundary if juxtaposed, acute separation behavior is not based on permanent adhesion differences between cells, but on transient contact-dependent reactions. The resulting, alternating phases of adhesion and repulsion appear to be part of a self-regulating loop (Figure 9). The spreading of an adhesion zone brings ephrins and Ephs in contact, inducing a repulsion signal. Once cells are apart, the signal decays, cells start to explore the intercellular space created at the boundary, and eventually re-establish adhesion. This mechanism prevents mesoderm cells from intruding into the BCR while providing necessary substratum contacts for migration.
Cell detachment is restricted to the tissue interface, and although various ephrins and their cognate receptors are coexpressed within each tissue, they do not lead to overall mutual cell repulsion and hence tissue dissociation. One possible explanation is suggested by our reaggregation experiments which show that EphB and ephrinB are required for normal ectoderm cohesion. This result is consistent with the fact that Eph/ephrin interaction between cells can mediate not only repulsion but also adhesion. The switch between these opposite responses depends on ephrin/Eph density, with adhesion being generally favored at low levels [39],[40], reviewed by Poliakov et al. [41]. In addition to expression levels, the different affinities for each other of the various Eph and ephrin subtypes [16] are likely to influence the strength of the response. Coactivation of EphB and ephrinB within the same cell can also induce an adhesive instead of a repulsive response [42], a mechanism that may likewise vary for different subtypes. In fact, our observation that overexpression of the mesodermally enriched ephrinB2 in the ectoderm is sufficient to trigger separation, but that ephrinB1 is unable to do so, suggests a surprising degree of specificity. This leads us to propose that the subtypes and levels of ephrins and Eph receptors expressed in each of the two tissues may be the main determinant for tissue separation. Since the ectoderm and the mesoderm express similar yet distinctly different mixtures of receptors and ligands, it is conceivable that the sum of interactions within each tissue is adhesive, while interactions across the boundary result in repulsion. However, in contrast to the ectoderm, no clear effect of Eph/ephrin function on tissue cohesion was observed for the mesoderm in our assay; possibly, the tissue is close to the transition point between adhesion and repulsion. Parallel mechanisms could further attenuate EphB/ephrinB signals within the mesoderm or enhance repulsion at the boundary. A candidate would be Xfz7/PAPC signaling, which is sufficient to induce separation in the ectoderm [11], and functions upstream or in parallel to EphB.
Importantly, regulation of tissue cohesion by EphB/ephrinB function seems to be independent of the control of tissue separation. In the ectoderm, ephrinB2 overexpression which triggers separation and ephrin/Eph depletion which inhibits it both decreased reaggregation, and in the mesoderm, inhibition of separation behavior is not associated with a change in overall cell-cell adhesion. This is consistent with other results indicating that changes in global adhesive strength do not necessarily affect tissue separation [43],[44]. The results indicate also that ectoderm-mesoderm separation is sensitive to the detailed expression pattern of Eph receptors and ligands, which is more complex than the classical complementary pattern, but may determine boundary formation in an essentially similar manner.
Ephrin/Eph-mediated boundary formation often involves the complementary expression of receptors and ligands in adjacent tissues and bidirectional signaling (e.g., [28]). This leads to rapid, large-scale changes in downstream pathways which differ in forward- and reverse-signaling cells [45]. Nevertheless, unidirectional signaling can be sufficient for cell segregation. Thus, in the zebrafish embryo, formation of the gap between adjacent somites depends on Eph forward signaling, while the ephrin reverse signal is dispensable [46]. Surprisingly, the same process using the same receptor and ligand isoforms requires ephrin reverse signaling, but not an Eph forward signal, in the chick embryo [47]. We found that similar to zebrafish somite segmentation, forward signaling is essential for ectoderm-mesoderm separation, and that two antiparallel forward signals are sufficient for complete repulsion at the boundary.
The demonstrated requirement for both RhoA and Rac in tissue separation is consistent with their known functions in Eph/ephrin mediated repulsion. Thus, the termination of EphB-ephrinB interaction, required for repulsion, involves endocytosis of receptor and ligand [48], which in turn depends on Rac function downstream of EphB [49]. RhoA and Rho kinase activation downstream of an EphB receptor can be mediated by the adaptor protein, Dishevelled, and underlies experimentally induced cell sorting of Xenopus gastrula ectoderm cells [24]. The local activation of Rac and RhoA at the ectoderm-mesoderm interface agrees with the notion that Eph/ephrin signaling is activated strongly at the boundary only, despite the widespread co-expression of receptors and ligands.
It is worth noting that despite the very low levels used in our experiments, each of the constitutively active RhoA and Rac construct could efficiently rescue separation. While this may be due in part to the potency of these forms, the cellular phenotypes produced by their overexpression appeared relatively mild. This was illustrated by the fact that even though the boundaries often appeared locked in a separated state upon rescue, attachment/detachment cycles were still observed occasionally. We hypothesize that in our inhibition experiments, ephrin/Eph signaling is decreased below the threshold required to maintain separation, but not completely abolished. Low amounts of active GTPases are then sufficient to boost the process and restore repulsion, while still allowing subsequent re-attachment. Alternatively, low levels of residual ephrin/Eph signaling may be sufficient to deliver exogenous active RhoA and Rac to their respective targets, though not to activate the endogenous GTPases sufficiently. The weak rescuing effects of wild type Rac and RhoA are consistent with both possibilities. In any case, the process of tissue separation appears particularly sensitive to small positive or negative changes in RhoGTPase activity, while larger alterations are required to affect other processes known to depend on such activities, in particular cell-cell adhesion.
Cleft-like boundaries based on attachment/repulsion cycles could generally be a requirement for the migration of cells across the surface of an adjacent tissue. A cleft-like ectoderm-mesoderm boundary is also seen in the zebrafish gastrula [50], and migration of mesoderm cells across the ectodermal layer occurs in Drosophila gastrulation [51]. It would be interesting to see whether the ephrinB/EphB mechanism of tissue separation is conserved in these cases. In contrast to the sparse, cell-permeable network of fibronectin fibrils that covers the ectodermal BCR of Xenopus, mouse or chicken gastrulae possess a well-developed basal lamina which physically separates ectoderm from mesoderm, potentially rendering germ layer separation independent of a cell repulsion mechanism [52]. Indeed, mouse ephrinB1, ephrinB2, and EphB4 null mutants apparently gastrulate normally [53]–[56]. After gastrulation, numerous mass cell migration events take place, for example neural crest migration, which in principle could also employ the mechanism presented here. Finally, deposition of an extracellular matrix seems to require free tissue surface[57],[58], and initially separating rhombomeres or somites by a repulsion cleft could be the first step in the boundary maturation process which ends with a matrix-filled space between tissue blocks.
Plasmids and morpholino oligonucleotides (Genetools) are listed in Supporting Information section.
Recombinant mouse EphrinB2/FC chimera (R & D systems) comprising the extracellular domain of mouse ephrinB2 fused to C-terminal 6X histidin tagged Fc region of human IgG were pre-clustered with anti-human Fc antibody (Jackson ImmunoResearch Laboratories) at a 1∶2 ratio in MBSH [5] and incubated 1 h before application.
mRNAs were synthesized according to manufacturer instructions (mMessage mMachine kit, Ambion). MOs and mRNAs were injected animally in the two blastomeres of 2-cell stage embryos for ectoderm/BCR tissues, and equatorially in the two dorsal blastomeres of 4-cell stage embryos for mesoderm explants, at amounts listed in Text S1.
The assay was performed as described [21]. Mesoderm explants were dissected from the lower lip region before the start of involution, as described [9]. For in vitro activation of Eph receptors, explants or BCRs were pre-incubated with preclustered ephrinB2-Fc fragments (40 nM in MBSH) for 15 min at room temperature. For the statistical analysis, results were compared using the paired sample Student's t test, taking each assay as an experimental unit. Thus each experiment was scored based on the percentage of test explants remaining separate (i.e., it is a graded response).
Explants from embryos injected respectively with 2×200–400 pg mGFP, mYFP, or mCherry mRNA were mounted in a dish with a bottom cover slip. For Figure 1A–C and Movies S1 and S2, time lapse recordings were acquired using a Zeiss LSM510 with a 40× Neofluar NA = 1.3 oil objective. GFP and YFP were excited with the 477 and 514 argon laser lines. Dicroïc and emission filters were HFT477, BP500/20 for GFP, and HFT514, LP530 for YFP. In other experiments, a Quorum technologies WaveFX spinning disc confocal mounted on an automated DMI6000B Leica microscope was used, with a 40× HCX PL APO CS, NA = 1.25 oil objective. GFP and Cherry were excited with 491 and 561 nm diode lasers. Images were collected with EM CCD 512X512 BT camera and controlled with Improvision Volocity 3DM software. Image processing was performed with Metamorph (Universal Imaging Corporation) and Adobe Photoshop7 software. For phalloidin staining, explants were fixed in 4% formaldehyde in MBSX for 10 min, followed by permeabilization (1% formaldehyde, 0.1% TritonX100), 1 h incubation with blocking buffer (10% sheep serum), and overnight incubation with 2 U/ml Alexa488-phalloidin (Invitrogen) in 10% sheep serum. After three washes in PBS and addition of antifade reagent (Slowfade Gold, Invitrogen), samples were examined using the spinning disc microscope.
Mesoderm and inner layer ectoderm were dissociated in alkaline calcium-free buffer (88 mM NaCl, 1 mM KCl, 10 mM NaHCO3, pH 9.3). Dissociated cells were transferred to agarose-coated Petri-dishes in normal MBSH and incubated for 1 h under mild rotation (10 rpm) on an orbital shaker. Images of the whole area where aggregation occurred, i.e. including all single cells and all aggregates, were taken under a dissecting microscope at a 12.5× magnification using a Micropublished RTV3.3 camera (Qimaging) and were analyzed for object size using ImageJ software. Two parameters were measured, average object area and area/perimeter ratio. Results of five independent experiments were normalized using wild type ectoderm as reference (1.0).
|
10.1371/journal.pntd.0003493 | Automated High-Content Assay for Compounds Selectively Toxic to Trypanosoma cruzi in a Myoblastic Cell Line | Chagas disease, caused by the protozoan parasite Trypanosoma cruzi, represents a very important public health problem in Latin America where it is endemic. Although mostly asymptomatic at its initial stage, after the disease becomes chronic, about a third of the infected patients progress to a potentially fatal outcome due to severe damage of heart and gut tissues. There is an urgent need for new drugs against Chagas disease since there are only two drugs available, benznidazole and nifurtimox, and both show toxic side effects and variable efficacy against the chronic stage of the disease.
Genetically engineered parasitic strains are used for high throughput screening (HTS) of large chemical collections in the search for new anti-parasitic compounds. These assays, although successful, are limited to reporter transgenic parasites and do not cover the wide T. cruzi genetic background. With the aim to contribute to the early drug discovery process against Chagas disease we have developed an automated image-based 384-well plate HTS assay for T. cruzi amastigote replication in a rat myoblast host cell line. An image analysis script was designed to inform on three outputs: total number of host cells, ratio of T. cruzi amastigotes per cell and percentage of infected cells, which respectively provides one host cell toxicity and two T. cruzi toxicity readouts. The assay was statistically robust (Z´ values >0.6) and was validated against a series of known anti-trypanosomatid drugs.
We have established a highly reproducible, high content HTS assay for screening of chemical compounds against T. cruzi infection of myoblasts that is amenable for use with any T. cruzi strain capable of in vitro infection. Our visual assay informs on both anti-parasitic and host cell toxicity readouts in a single experiment, allowing the direct identification of compounds selectively targeted to the parasite.
| Chagas disease is a zoonosis caused by the protozoan parasite Trypanosoma cruzi. Endemic to Central and South America, it affects over 10 million people and many more live in risk transmission areas. Although mostly asymptomatic at its initial acute stage in humans, damages can occur over the years in many tissues such as T. cruzi-hosting heart and digestive track. For over 40 years the chemotherapy of Chagas disease has relied on the use of two drugs, benznidazole and nifurtimox, though both are known to lead to severe side effects which often prompt to the discontinuation of the treatments. Despite having good efficacy against the acute stage of infection, its efficacy against the chronic stage is under question. Therefore, there is an urgent unmet need of new anti-T. cruzi drugs. Several efforts have been made in the last years to establish reliable high throughput cell based in vitro assays to be used for drug discovery against Chagas. With the aim to contribute to this field here we describe the development of a new automated image-based assay to identify new compounds against T. cruzi that has been set up using the myoblastic rat cell line H9c2 as cell-cycling amastigotes hosting cells.
| Chagas disease, classified by the WHO as a neglected tropical disease, is a zoonosis caused by the Kineoplastid protozoan parasite Trypanosoma cruzi. It is endemic to Central and South America where it stands as a major public health problem [1]. Migratory population movements between endemic and non-endemic regions during the past decades have globalized Chagas impact, which nowadays represents a public health issue in several non-endemic countries [1, 2]. It is estimated that 12,000 people die annually of Chagas, there are 10 million people currently infected, and 100 million live in risk-transmission areas [3].
The disease progresses over three phases. First, an early acute stage barely spanning over a few weeks during which parasitemia is detectable. Though mostly asymptomatic, it can cause the death of children and immune-suppressed patients. A second indeterminate stage follows that can last over several years (10 to 20), where the parasite presence is hardly detected and no symptoms are observed. Finally, in about 30% of the chronically infected patients, the infection can lead to severe failures of heart and gastrointestinal tract functions that ultimately cause death [4].
There is no effective human vaccine against Chagas disease currently available and its development may entail potentially insurmountable hurdles [2, 5]. The two available drugs against the disease are benznidazole (BNZ; Abarax, laboratorio ELEA) and nifurtimox (NFX; Lampit, Bayer Healthcare), which although have a good efficacy against the short-lasting, mostly asymptomatic acute phase of the disease, its activity against the life-threatening chronic phase is still under study [2, 6]. Both BNZ and NFX have toxic side-effects that often cause premature discontinuation of the treatments. An important difficulty in Chagas drug development is that success in treatment is difficult to evaluate since this is a silent disease that shows its symptomatology decades after the infection and the presence of parasites in the blood is at the limit of detection during the chronic stage [4]. Recently, the promising results achieved with the repurposed anti-fungal azoles (posaconazole and the ravuconazole derivative E1224; [7, 8]) took them into clinical trials, but unfortunately the outcome of both has been disappointing [9–10]. With such outlook, high expectations are put in the discovery of less toxic drug entities that are active against the chronic stage of the infection. As it happens with many other neglected tropical diseases, there is an urgent unmet requirement of new chemotherapies that should widely improve the performance of current treatments [6, 11, 12].
In the field of drug discovery against infectious diseases, high-content screening approaches are being increasingly used since they allow the acquisition and analysis of highly informative data from cell-culture based events [13–14]. Several assays suitable for high throughput screening of chemical collections, including image-based assays and an algorithm to assess anti-T. cruzi drug inhibition, have been described for early drug discovery in Chagas disease [11, 15–18]. With the aim to contribute to the drug search process against Chagas, here we describe the development of a phenotypic assay to identify anti-T. cruzi compounds, that allows high throughput with excellent reproducibility. The assay has been set up on a 384 well plate format with an Opera high-content microscope (Perkin-Elmer), which can be automated to sustain a throughput adequate for both primary screening of compounds or secondary hit qualifier. Based in microscopic image analysis, the developed assay can be used to test any parasite strain that is adapted for infection in vitro (including non-engineered strains and clinically relevant specimens), and since host-cell and parasite drug effects are simultaneously assessed, it provides specific anti-parasitic readouts and host cell toxicity information in a single experiment. In order to achieve a biologically relevant environment, the assay was set up using the rat heart derived cell line H9c2 as host cells.
LLC-MK2 (green monkey kidney epithelial cells) and H9c2 (rat cardiomyocytes) cell lines were cultivated in DMEM (Life-Technologies) supplemented with 10% FBS (Biowest, USA), 100 U/ml penicillin (Sigma-Aldrich), 100 μg/ml streptomycin (Sigma-Aldrich), and 4 mM or 2 mM L-glutamine (Sigma-Aldrich), respectively. Both cell lines were purchased at the European Collection of Cell Cultures (ECACC, Salisbury, UK) and were grown at 37°C, 5% CO2 and >95% humidity. H9c2 cells were cultured in roller flasks (800 cm2 growing surfaces; Corning Inc., NY, USA). A single roller supplied cells to seed at least seven T225 flasks at the assay day. The DMEM formulation for the assay lacked phenol red (Life-Technologies reference 31053) and was supplemented with 2% FBS, 100 U/ml penicillin, 100 μg/ml streptomycine, 2 mM L-Glutamine, 1 mM sodium-pyruvate (Life-Technologies), and 25 mM HEPES (Life-Technologies) [15].
T. cruzi Tulahuen strain parasites expressing β-galactosidase were kindly provided by Dr. Buckner (University of Washington, Seattle, USA; [19]) and maintained in culture by weekly infection of LLC-MK2 cells in the same DMEM formulation used for cell growth, but supplemented with 2% FBS instead of the 10% FBS added to the cell lines maintenance medium. Trypomastigote forms were obtained from the supernatants of LLC-MK2 infected cultures harvested between days 5 and 8 of infection as described elsewhere [15]. They were used to maintain the cycle and to infect H9c2 monolayers.
The compounds used to set up the assay were selected upon literature searches based in their previously described anti-trypanosomatid activity (see Table 1 for details) or their presence in current clinical trials against T. cruzi [12]. Those compounds not available in GSK chemical collection were purchased from Sigma-Aldrich except the following: amiodarone (Pfizer), cloroxylenol derivative CX1 (Chembridge), dihydroergocristine mesylate (Tocris Bioscience), hydrazide derivative PCH1 (InterBioScreen), LP10 (ChemDiv), loperamide (Enamine), posaconazole (Sheckchem.com), pubchem 1473168 and pubchem 3812524 (Bionet), and terconazole (AKSCI-USA). Compounds were pre-dispensed into the plates with an Echo 555 instrument (Labcyte; 250 nl per well) in a 3-fold dilution row-pattern to get eleven concentration points for each compound. Compound concentrations in the assay ranged from 102 to 1.6×10-3 μM except posaconazole that ranged from 1 to 1.6×10-5 μM. Upon preparation, the plates were stored at -20°C until being assayed. Whole ranges were used to determine host-cell toxicity (TC50) and anti-parasitic efficacy (IC50) values.
Control wells used to determine a 100% parasitic growth (full 6th column of each plate) were left untreated, whereas not infected cells were added in the control wells used as 0% parasite growth (or 100% parasite growth inhibition; full 18th column of each plate).
H9c2 cells were seeded in T-225 flasks (5 × 106 cells/flask, 225 cm2 culture surface; Corning Inc., NY, USA) in DMEM-10% FBS for 4 h to allow attachment. Cells were then washed once with PBS before infection. T. cruzi trypomastigotes, collected at days 5 to 8 after infection, from LLC-MK2 parasite infected cultures, were allowed to swim out for 4 hours at 37°C from a centrifuged pellet (2,500 rpm/10 min/RT; [15]). Trypomastigotes were then collected and counted in a CASY Cell Counter (Roche-Applied-Science) using the 60 μm capillar. Trypomastigotes, in supplemented assay DMEM, were added to H9c2 cultures in a multiplicity of infection (MOI) = 1 and incubated for 18 hours. Cells were washed once with PBS before incubation of the infected H9c2 monolayer with trypsin (Life-Technologies) to detach cells from the flask. Cells were counted in a CASY Cell Counter (Roche-Applied-Science) using the 150 μm capillar and their density set at 5×104 cells per ml in supplemented assay DMEM.
Infected H9c2 were dispensed into 384 well plates (μClear bottom, Poly-Lysine CellCoat® treated surfaces; Greiner Bio-One, Germany) at 50 μl per well using a Multidrop (Perkin-Elmer) liquid handling device. Control wells indicating 100% parasitic growth were left untreated, whereas control wells defining 0% parasitic growth just contained H9c2 uninfected cells. The rest of the plate contained the compounds dispensed according to the dose regimen explained in the previous section. DMSO concentration never exceeded 0.5% in all plate wells, which had no effect on the host cells viability, neither in the parasite replicative growth. DMSO tolerability limit of the assay was found to be 2%. After seeding them, the plates were incubated at 37 °C in a humidified 5% CO2 atmosphere for 72 h. Cultures were then fixed and stained by addition of 50 μl of a solution containing 8% formaldehyde and 4 μM Draq5 DNA dye (BioStatus, UK) per well. Plates were kept light-protected and imaged 1 h later (longer times of incubation did not affect the quality of the assay).
Plates were imaged in a Perkin-Elmer Opera confocal microscope using a 20X air objective (NA 0.4) and the next acquisition set: a 635 nm laser excitation line and a 690/50 emission detection filter for Draq5 detection (bandwidth 650–700 nm). Five images were collected per each well for reliable statistical analysis. No significant differences were observed in cell or amastigote numbers among images in different locations within wells. The Opera microscope in our facility is coupled to a robotic arm that automatically loads and unloads the plates to be imaged, contributing to an increased throughput when required.
To define presence or absence of T. cruzi amastigotes within delimited cell boundaries (cells cytoplasms) the Acapella software (Perkin-Elmer) was used to develop a very selective script based in the algorithms provided by the software´s ‘building blocks’ approach. The script included the following steps (Fig. 1):
s
d
(
x
)
=
(
e
σ
2
−
1
)
e
2
μ
+
σ
2
, over the expected value E ( x ) = e μ + σ 2 / 2
(The method to determine these thresholds is explained in “Image analysis method” section of Results). The spots eventually counted as true amastigotes are shown in green. It is observed that there are remaining putative spots which contribute to a certain ‘noisy’ background in non-infected cells and BNZ treated ones (see corresponding panels in Fig. 1e).
Three outputs (per well) were provided by the script: (1) number of host cells nuclei (‘Cells’) to determine drug-related cytotoxicity; (2) number of amastigotes per cell (‘Am/Cell’) as infection level measurement; and (3) percentage of infected cells per well (‘%Infected’) as a second infection level output. The assay window (signal to background; S/B) and the Z´ parameter assessment were performed as previously described [29]. To qualify a plate as pass, Z´ values had to be >0.5. The three selected outputs per plate were exported to ActivityBase software (IDBS, USA) for data normalization and fitting to inhibition equation.
The TC50 value was provided by the ‘Cells’ output and two IC50 values were respectively provided by the ‘Am/Cell’ and ‘%Infected’ outputs. All curves visualization and fitting was carried out with ActivityBase, following the equation by Hill’s 4 parameters being the maximum asymptote between 80 and 100% for all the compounds presented in the article. The selectivity window was calculated for each compound in the plate taking into account their TC50 and ‘Am/Cell’ IC50 values. Results summarized in Table 1 are referred to at least three independent experiments and in all cases are shown as the mean value ± its standard deviation (± sd).
Our goal was to develop an assay specifically aimed at the parasitic amastigote stage, since this is the replicative form in the mammalian host. A rat myoblastic cell line (H9c2) was chosen as host cells, since T. cruzi pathology is frequently related to its presence in muscle tissue during the chronic stage of the disease [2, 4] and these cells are highly susceptible to infection in vitro [30-, 31]. Monolayers of H9c2 myocytes were infected with T. cruzi trypomastigotes for 18 h to allow for development of the intracellular amastigote stage. Infection was performed in T-flasks followed by washing to remove non-invasive trypomastigotes present in the culture medium and trypsinization of the cell monolayers. Several times and multiplicities of infection in the flasks, and host cell numbers per well after transfer to plates, were tested in order to define the best possible assay. We observed that 5×104 cells/ml (2,500 cells per well) provides sufficient cell events counts per view field for an statistically robust measurement (Z’ was used as the discriminatory parameter) without compromising the imaging and image analysis times. A MOI = 1 was considered optimal since it provided a ~50% infection level after 18 h incubation in the flask. The total assay timespan that includes 18 h in the T-flasks and 72 h in the plates (~90 h) permits various parasite replication rounds with negligible host cells bursting due to over-infection (cell integrity was monitored for every condition tested using the parameter ‘Cells’ as described below). Cells were then dispensed into 384 well plates (2,500 cells/well) coated with PolyLys (CellCoat® by Greiner-BioOne). At 72 h incubation, confluency of monolayers was found to be 52.15% of the well surface (average of six independent control wells). Cultures were then stained with a DNA dye, Draq5 (Fig. 1). Similarly to previously reported [32], no major differences were seen between visual or automated counting of amastigotes, with a quantification correlation>90% in four independent experiments.
When images were acquired at 40X, it was observed that, in addition to host cell nuclei, both the nuclear DNA (nDNA) and mitochondrial kinetoplast DNA (kDNA) were stained (Fig. 2). Nevertheless, at the magnification rate used for the automated assay (20X) it was not possible to discern between parasitic nDNA or kDNA and a single spot per amastigote was observed (Fig. 1). Draq5 slightly stains the host cells cytoplasm too, which helped to define the surface of the host cell and differentiate it from the highly stained host cell nuclei with the algorithms from the Acapella software (Fig. 1b). The size differences between the host cells and the T. cruzi amastigotes nuclei permitted the use of a single fluorescence channel to detect both cell types (Ex/Em = 635/690 nm). For accurate quantification of amastigote spots per cell, two parameters were evaluated by the script: ‘Spot area’ to differentiate between host cells and parasite nuclei with a cut off for parasite spot detection at ~35.5 μm2 (~55 px2) and a ‘Corrected intensity’ filter with two thresholds to distinguish true parasitic spots, that have higher intensity, from non-parasitic (‘noisy’) cytoplasmic spots per cell (Fig. 1c; see below for details). These thresholds varied slightly per run of plates and were therefore defined considering the control wells of each one prior to the performance of analysis.
A high number of non-parasitic stained cytoplasmic spots were detected as a consequence of non-specific dye staining (see Fig. 1b and [18, 33]). In order to find a method to best discriminate true amastigotes from other non-parasitic cytoplasmic spots, all fields from control infected and control non-infected cells within six independent plates were taken. Those images were processed using Acapella software with the sensitivity for spot detection set at maximum. The spots retrieved this way (1.18×106) were classified based on their ‘Texture’, ‘Corrected intensity’ and ‘Morphology’ properties into nine classes by an initial clustering method based on the Hartigan-Wong algorithm [34] and processed with the statistical analysis software ‘R’ [35] (Fig. 3). This analysis revealed that ~95% of the spots population in control non-infected wells were accumulated in two of the nine classes (classes 3 and 7 in Fig. 3). A further graphical analysis of the spots populations according to their ‘Corrected intensity’ parameter revealed three discriminating regions for that property and it was observed that those spots that fell into classes 3 and 7 (note they meant a ~95% of the spots retrieved from non-infected controls) belonged to the low ‘Corrected intensity’ region, which therefore determined the spots non-specifically stained by Draq5. In contrast, a significant percentage of the spots found in T. cruzi-infected control wells populated the intermediate ‘Corrected intensity’ region identifying this as the true amastigotes region. The third high intensity region included scarce artifacts of the staining procedure with very high corrected intensity values (Fig. 3). Thus, based on this analysis the ‘Corrected intensity’ condition from the Acapella script was selected to discriminate true from false amastigotes. In order to determine the limits of that medium ‘Corrected intensity’ region that contained the true amastigotes a lower and an upper threshold were set. The lower cut off limit was set by adjusting all non-infected control spots population ‘Corrected intensity’ (this population mainly fell into the low ‘Corrected intensity’ region) to a lognormal distribution and establishing the threshold at 6 standard deviations (sd; Fig. 4a). Then, the upper threshold was determined by adjusting to a lognormal distribution all the spots in infected control wells whose ‘Corrected intensity’ was above the low limit calculated before, plus 3 sd taken over the expected value (see equation in Image analysis section of Methods; Fig. 4a). Lastly, to further contribute to best discriminate true from false amastigotes, an upper threshold for the ‘Spot area’ parameter was also set (Fig. 4b). To define it, all infected control spots that fell between the lower and upper ‘Corrected intensity’ thresholds explained above (i.e. true amastigotes) were adjusted to a lognormal distribution and their upper ‘Spot area’ threshold was set at 3 sd over the expected value, thus excluding from a further analysis those spots above such threshold (Fig. 4b). The final established criteria for automated identification of true amastigotes selected spots consisted on intermediate ‘Corrected intensity’ between the pre-set thresholds (>48 and <274 a.u.) and a ‘Spot area’ size below a cut off of 51 px2 (Fig. 1e and Fig. 4). These values, assigned during the statistical analysis procedure that was followed to build the script were empirically customized for each corresponding run (or assay day) by statistical analysis of the spots from all control wells in the run.
With the algorithms available for the script design, the outputs provided a mean readout per well. To test the reliability of the mean readout and to inform on the infection homogeneity achieved, a cell-by-cell spots population analytic approach was run on both controls (infected and non-infected) in six assay plates made in different days. The analysis showed a very homogenous performance independently of the plate, where the median of amastigotes per cell in control infected was ~10 amastigotes per host cell in all cases, in contrast with the very seldom control uninfected cells that contained >1 putative parasites (population outliers in Fig. 5a). The number of amastigotes per cell (median) in control uninfected cells was very similar to that found in BNZ treated infected cells (Fig. 5a). The analysis was used to categorize the distribution of the number of amastigotes occurring per individual cell. This showed that a very large percentage of the ~50% of infected cells contained over 8 parasites; whereas in BNZ-treated infected cells the few hosts that contained amastigotes were below that number (Fig. 5b).
Compound inhibition of T. cruzi replication was primarily assessed as the reduction of the mean amastigote number per cell (‘Am/Cell’), as well as by a decrease of the percentage of infected cells per well (‘%Infected’). Since parasite number reduction could potentially be due to a drug-related cytotoxic effect on the host cells, the mean host cells nuclei number per well (‘Cells’) was considered in the analysis procedure. The reference value that indicated a 100% of live host cells was always established upon non-treated control infected wells. Five image fields were taken per well, which provided a statistically significant Z’ values for the assay. In all cases, at the assay readout end point (~90 h post-infection; 72 h after addition of compounds) similar host cells number was seen per well in infected control (mean ± sd; 353.92 ± 24.22) and non-infected control wells (mean ± sd; 334.13 ± 23.35), which indicated that host cell lysis was not significantly occurring in the 90 h infection time span.
A reduction in the ‘Am/Cell’ and in the percentage of infected cells per well or ‘%Infected’ measurement in compound-exposed wells, in comparison with the values obtained for the non-treated infected control wells, both yielded a quantitative reading of T. cruzi growth inhibition after 72 h of drug exposure. In untreated infected control wells, the average number of amastigotes per cell was 10.19 ± 1.22, which contrasted with the 0.39 ± 0.08 in non-infected control wells. We established non-infected control wells as 100% of parasite growth inhibition, which defined a 25X signal-to-background window. Under these conditions, the ‘Am/Cell’ output presented acceptable reproducibility with a Z´ of 0.6 (calculated from 16 control wells of three different plates from three independent experiments). The third output, ‘% Infected’, provided a complementary reading of the anti-T. cruzi activity of the compounds and served as well as an assay performance inner control. In all cases, the ‘% Infected’ was above 50% (mean ± sd; 52.44 ± 3.03) in the non-treated infected control (Fig. 1d), and contrasted with that of non-infected control wells (8.72 ±.88). These data offered a signal to background window (S/B) ~6X, which contributed to a robust Z´ parameter also for this output, that was calculated to be 0.66. As quality control for every assay, a requirement for a Z’>0.5 in each of the three main parameters (‘Cell’, ‘Am/cell’ and ‘% Infected’) was considered.
At present, 162 replicate plates of the assay have been performed over a period of 10 months with remarkable reproducibility. The Z’ values of the parameter ‘Am/Cell’ remains consistently >0.5 with an average value of 0.58.
To validate the assay a list of compounds with known anti-trypanosomatid activity was assayed at 11 serially diluted concentrations at least in three independent experiments (Table 1). Most of the compounds in the list had been previously tested against a range of T. cruzi strains either by image-based [17–18] or other in vitro assays [15–16, 19, 26–27], including the primary assay of the recent anti-T. cruzi high throughput screening campaign (HTS) performed in GSK Tres Cantos facility (Spain; Peña et al., manuscript in preparation). Our assay succeeded in identifying all the anti-T. cruzi compounds, and the dose-response analysis made assigned TC50 and IC50 values that were in concordance with those previously reported [15–17, 19, 26–27]. Agreement was best with those previously reported results where compounds were tested against a Tulahuen parasite strain [15, 17, 19, 27], particularly when looking at the Am/Cell assay output. A comparison between our retrieved dataset and the previously one reported by Engel et al. anti-T. cruzi image-based assay [17], also showed a good agreement despite the use of different parasite strains and host cell lines in both assays. Looking at Chagas disease standards of care drugs, BNZ and NFX, the results obtained here are very similar to those previously published for the first, but a certain disagreement was observed for NFX IC50 [25], which is probably due to the use of epimasgote forms to assess anti-parasitic drug performance [25].
The response curves of two current drugs used in the clinic for treatment of Chagas disease (BNZ and NFX), posaconazole (recently tested in clinical trials with negative results) and cycloheximide (second compound in potency against T. cruzi amongst the list tested are shown to illustrate the versatility of the image-based approach that provides the toxicity to the host cells and two anti-parasitic response outputs, amastigotes per cell and percentage of infected cells, in a single experiment (Fig. 6).
There is a pressing need for new compounds to treat Chagas disease. With the perspective to accelerate pre-clinical screening of putative anti-Chagas compounds, we have developed an image-based phenotypic whole cell assay that identifies anti-T. cruzi hits specifically targeted against the intracellular amastigote stage. This is the replicative form in the mammalian host and therefore is considered the major target for anti-T. cruzi future drugs. Imaging assays anti-T. cruzi and an imaging-based screening have been reported before with successful results [11, 17]. Our assay provides novel features that constitute important advantages for the performance of HTS campaigns. In addition to the parasite growth inhibition readout, this imaging-based assay provides a cytotoxic readout, therefore providing in a single experiment the drugs specific activity against the parasite versus the host (selectivity index). High statistically robust discriminating power (better Z’ values) and strong assay reproducibility, in addition to the highly reproducible image analysis method together with the robotic automation of the plates’ loading-unloading process guarantee medium-to-high throughput capabilities, therefore providing a versatile assay that could either be used as a primary screening tool (single dose) as well as a secondary hit qualifier assay (dose response).
Along the development process, a fundamental aspect was the election of the host cell line to support parasite replication. We chose the myoblastic H9c2, which had previously been used to characterize the T. cruzi infectious process [30–31]. H9c2 cells provided percentages of infection above 50% in untreated control wells at MOI of 1 (Fig. 1d). These conditions resulted in homogeneous infections where the rate of infected to non-infected cells was similar in all areas of the well. This factor was crucial for the reproducibility of the parameters determined in the assay, ratio of amastigotes per cell and percentage infected cells. Previous attempts to use NIH3T3 fibroblasts as host cell resulted in non-homogenous infections that provided low reproducibility in the parameters determined.
While H9c2 are easy to culture and allowed very reproducible experiments and a robust assay performance, they have a large size (~100 μm), which accounted for a low production rate. To overcome this problem, H9c2 cells were cultured in roller, which resulted in a highly increased yield. A single roller supplied cells to seed at least seven T225 flasks at the assay day. Given that host cells for three 384 well plates are obtained per each infected T225, the achieved throughput may reach >20 plates per roller. This translates into 7,000 single dose or 640 dose-response evaluations plus their corresponding controls per roller, conditions that allow for high throughput screening using this method. In our facility, four roller flasks per week were typically used, allowing testing of approximately 35,000 compounds and the processing of 80–100 plates.
It was our objective to develop a reliable and easy to standardize assay. With that in mind, the pre-infection of the cells was made in flasks instead of in plates allowing homogenous infected cells batches and simplified the programmed washing steps. The use of a single stain, Draq5, for host and parasite cells also simplified the procedure. DAPI has previously been used as a single stain for host cells and T. cruzi [17], however, Draq5 dye provided a faint stain of the host cell’s cytoplasmic area, which facilitated the detection of this region in the analysis [18, 33]. Furthermore, as Draq5 can be added together with the formaldehyde fixing solution and no washing steps are required thereafter, the fixation and staining of the assay plates was made in a single step, saving time and avoiding washing steps that may have affected the monolayers. Although the use of a single dye carried the disadvantage of certain unspecific staining of non-parasitic cytoplasmic spots, that was circumvented with a very selective script analysis. Avoiding the use of two different dyes also improved the speed of image acquisition and analysis. An additional advantage of the use of a DNA stain is the possibility to use any strain or clinical isolate of T. cruzi that can be cultured in vitro, upon assay adaptation to its particular characteristics, since it does not require transgenic lines of parasite expressing reporters.
The assay was validated against a series of compounds that had been previously reported to have anti-trypanosomatid activity [15, 17, 19–28]. Despite the expected variability in the results due to the different methods and strains used in different laboratories, our retrieved results mostly agree with the previously available data, being the correlation particularly good with those references that had used the Tulahuen strain [15, 19, 27].
High-content screening is nowadays considered a key technology in the drug development process by the pharmaceutical industry. Nonetheless, several aspects mainly linked to the large volume information generated and the complications associated to data integrity, storage and processing still need to be improved [13]. Our assay has been designed to allow high level data integrity and data tracking featured upon barcode identified plates. It is therefore essential that the readouts can be followed by data management software like the ActivityBase suite (IDBS, USA). With that in mind, three simple outputs per well were selected to provide the needed information to determine whether a compound was or not toxic to the host cells; whether it actively inhibited the parasite growth; and to complement such information, whether the infectivity ratio was constant intra- and inter-plates, which also served as an assay quality control readout. Another important aspect in the development of the assay was the processing time required for both imaging and image-analysis, as these constitute potential throughput bottlenecks. The use of a single laser channel to detect host cells and parasites (Draq5 nucleic acid dye) contributed to reduce the imaging time, performing just one exposure over the five fields acquired per well. Once the plate was inside the Opera microscope, it took 1 h to perform the image acquisition and analysis. But if required, the Perkin-Elmer Opera device and its associated Acapella software allow independent image acquisition and analysis, so the software can be set to run offline upon pre-taken stored images, circumventing the time-consuming computing issue.
The great versatility and reproducibility afforded by this image-based approach may provide the basis for further advancements in the near future, like expanding the panel of T. cruzi strains for testing to allow a wider coverage of the wide phylogenetic spectrum of the parasite. Other potential upgrade chances may lie on the development of an automated assay to discern between cidal and static compounds [15], which would provide further tools for the selection of early hits. Imaging assays rely on the fast advancing optical and computing technologies and definitely stand as a very powerful and adaptable tool in the drug discovery process.
|
10.1371/journal.pgen.1004703 | The DAF-16 FOXO Transcription Factor Regulates natc-1 to Modulate Stress Resistance in Caenorhabditis elegans, Linking Insulin/IGF-1 Signaling to Protein N-Terminal Acetylation | The insulin/IGF-1 signaling pathway plays a critical role in stress resistance and longevity, but the mechanisms are not fully characterized. To identify genes that mediate stress resistance, we screened for C. elegans mutants that can tolerate high levels of dietary zinc. We identified natc-1, which encodes an evolutionarily conserved subunit of the N-terminal acetyltransferase C (NAT) complex. N-terminal acetylation is a widespread modification of eukaryotic proteins; however, relatively little is known about the biological functions of NATs. We demonstrated that loss-of-function mutations in natc-1 cause resistance to a broad-spectrum of physiologic stressors, including multiple metals, heat, and oxidation. The C. elegans FOXO transcription factor DAF-16 is a critical target of the insulin/IGF-1 signaling pathway that mediates stress resistance, and DAF-16 is predicted to directly bind the natc-1 promoter. To characterize the regulation of natc-1 by DAF-16 and the function of natc-1 in insulin/IGF-1 signaling, we analyzed molecular and genetic interactions with key components of the insulin/IGF-1 pathway. natc-1 mRNA levels were repressed by DAF-16 activity, indicating natc-1 is a physiological target of DAF-16. Genetic studies suggested that natc-1 functions downstream of daf-16 to mediate stress resistance and dauer formation. Based on these findings, we hypothesize that natc-1 is directly regulated by the DAF-16 transcription factor, and natc-1 is a physiologically significant effector of the insulin/IGF-1 signaling pathway that mediates stress resistance and dauer formation. These studies identify a novel biological function for natc-1 as a modulator of stress resistance and dauer formation and define a functionally significant downstream effector of the insulin/IGF-1 signaling pathway. Protein N-terminal acetylation mediated by the NatC complex may play an evolutionarily conserved role in regulating stress resistance.
| What are the mechanisms used by animals to cope with stressful environments that inflict damage or restrict essential processes such as growth, development, and reproduction? One strategy is changes in physiology that increase stress resistance, and an extreme version of this strategy is diapause, an alternative developmental state that is enduring and stress resistant. In the nematode C. elegans, stress tolerance and entry into a diapause state called dauer larvae are mediated by the conserved insulin/IGF-1 pathway. Specifically, the FOXO transcription factor DAF-16 promotes stress tolerance and dauer larvae development. However, the targets of DAF-16 that mediate these processes remain largely elusive. Using an unbiased forward genetic screen to discover new mediators of stress tolerance, we identified natc-1, a novel target of DAF-16 and the insulin/IGF-1 pathway. natc-1 encodes a conserved subunit of the N-terminal acetyltransferase C (NAT) complex. The NatC complex modifies target proteins by acetylating the N-terminus. We demonstrated that natc-1 mediates diapause entry and stress tolerance. Furthermore, we elucidated regulation of NatC by demonstrating that natc-1 is a direct transcriptional target that is repressed by DAF-16. These findings may be relevant to other animals because both the insulin/IGF-1 signaling pathway and the NAT system are conserved during evolution.
| The ability to cope with fluctuating environmental stresses is critical for animal survival. Environmental stresses include a wide range of factors such as extremes in temperature, oxidation, and metal availability. A stress response might promote tolerance against a specific challenge or provide broad-spectrum resistance, and a critical question in this field is how specific stress responses mediate resistance to one or more forms of environmental challenge?
The nematode Caenorhabditis elegans is an important model system for studies of stress resistance. In response to stresses such as high temperature, low nutrient availability, and high population density, developing larvae will adopt an alternative L3 stage called dauer that is stress resistant [1]. Studies of dauer formation led to the discovery of an insulin/IGF-1 signaling pathway as a critical regulator of this stress response [2]. Loss-of-function mutations in daf-2 and age-1 cause dauer constitutive phenotypes, whereas loss-of-function mutations in daf-16 cause dauer defective phenotypes [1]. daf-2 encodes an insulin/IGF-1 receptor homolog, and age-1 encodes a phosphatidylinositol-3-OH kinase catalytic subunit homolog [3], [4]. In addition to mediating a developmental switch in larvae, this pathway functions throughout the life of the animal to mediate stress resistance, since daf-2 loss-of-function mutations cause increased tolerance to multiple stresses and an extended lifespan [3], [5], [6]. These daf-2 mutant phenotypes are suppressed by mutations in daf-16, indicating that daf-16 is a major downstream effector of the insulin/IGF-1 signaling pathway that is negatively regulated by daf-2 activity. daf-16 encodes a FOXO transcription factor [7], [8]. Because DAF-16 plays a central role in promoting longevity and stress tolerance, a major goal has been to identify and characterize DAF-16 transcriptional targets [9]–[17]. Although several genes have been demonstrated to be directly regulated by DAF-16, the understanding of how DAF-16 target genes mediate stress resistance and longevity remains fragmentary. The insulin/IGF-1 signaling pathway regulates stress tolerance, metabolism, and longevity in multiple species including mammals [18]–[22]. Thus, the identification of physiologically significant DAF-16 targets may suggest strategies for promoting longevity and stress tolerance in C. elegans and higher eukaryotes.
The metal zinc is a nutrient that is essential for all organisms and plays many roles in biological systems. Zinc functions in signal transduction pathways and contributes to protein structure and activity [23]–[25]. Zinc deficiency and excess both cause a wide spectrum of defects, demonstrating the importance of zinc homeostasis [26]–[31]. Zinc deficiency appears to be deleterious due to the reduced function of many zinc-requiring proteins and signaling events [25]. The mechanisms underlying excess zinc toxicity are not well characterized; excess zinc may displace other physiological metals or bind to low-affinity sites, leading to altered or decreased protein function [32]. In addition to zinc, several other metals are toxic in excess, including cadmium, nickel, and copper [33]–[35]. To characterize mechanisms of metal stress resistance, we performed a forward genetic screen for mutations that caused resistance to high levels of dietary zinc [36]. We reasoned that tolerance to high zinc could be caused by two general mechanisms: (1) mutations might affect zinc metabolism and reduce the accumulation of toxic zinc, or (2) mutations might cause alterations that promote growth and survival in the presence of high levels of zinc. We identified haly-1, which encodes the enzyme histidine ammonia lyase that metabolizes histidine [37]. Loss-of-function mutations in haly-1 do not affect zinc accumulation but rather cause an increase in histidine levels resulting in increased chelation of zinc and nickel, and chelation by histidine reduces the toxicity of these metals [37]. Notably, haly-1 mutations have not been demonstrated to cause resistance to other stressors. By contrast, mutations in daf-2 and age-1, members of the insulin/IGF-1 pathway, cause resistance to a broad-spectrum of stressors including the metals cadmium and copper, oxidative stress, and heat stress [5], [35], [38].
Here we describe natc-1, a new gene that was discovered in the screen for worms that are resistant to high zinc toxicity, which encodes the C. elegans N-α-acetyltransferase 35. N-α-acetyltransferases (NATs) are highly conserved among eukaryotes and function as protein complexes to transfer the acetyl group of acetyl coenzyme A to the α-amino group of the first amino acid of a target protein. natc-1 encodes an auxiliary subunit of the NatC complex which acetylates proteins that begin with the amino acids Met-Leu, Met-Phe, Met-Ile, and Met-Trp [39], [40]. natc-1 mutations caused resistance to multiple metals, oxidation, and heat, indicating that natc-1 modulates broad-spectrum stress resistance, similar to mutations in insulin/IGF-1 signaling genes. Interestingly, the natc-1 promoter contains an evolutionarily conserved DAF-16 binding site, and DAF-16 binds natc-1 in vivo [11], [41], [42]. We demonstrated that natc-1 is transcriptionally repressed by DAF-16 activity and that natc-1 interacts with genes in the insulin/IGF-1 signaling pathway to mediate stress resistance and dauer formation. These results indicate that natc-1 is directly regulated by DAF-16 and functions as a downstream effector of the insulin/IGF-1 signaling pathway. Supporting this model, mutations in natc-1 that increase stress resistance are epistatic to daf-16 mutations. These results provide novel insights into the transcriptional regulation of natc-1 by the insulin/IGF-1 signaling pathway and the biological function of protein N-terminal acetylation in mediating stress resistance. Furthermore, our data elucidate a new mechanism used by the insulin/IGF-1 signaling pathway to mediate stress tolerance and dauer formation.
We performed a forward genetic screen to identify mutant strains that are resistant to the growth arrest and lethality caused by high levels of dietary zinc [36]. Nineteen mutations were identified and positioned in the genome using a genome-wide map of single nucleotide polymorphism (SNP) markers [36]. Here we focus on two of these mutations, am134 and am138, that caused significant resistance to dietary zinc toxicity (Figure 1A,B). Both mutations displayed tightest linkage to the same SNP, pkP5513, positioned at +0.1 map units on chromosome V (Figure 1C). Three factor mapping experiments indicated that am138 is positioned between dpy-11 and unc-42, a 325 kilobase pair interval that contains pkP5513 (Figure 1C) [36].
To identify the gene affected by these mutations, we performed whole genome sequencing using DNA from the am134 and am138 mutant strains. Candidate mutations in the mapping interval were identified by comparing the am134 and am138 DNA sequence to the wild-type DNA sequence. The am134 and am138 strains both contained candidate mutations in the predicted open reading frame T23B12.4, suggesting that these mutations caused resistance to zinc toxicity. The predicted T23B12.4 protein is homologous to human N-α-acetyltransferase 35, an auxiliary subunit of the NatC complex. Thus, we named this gene natc-1. The mutation in the am134 mutant strain, which was induced with the mutagen ENU, is a C to T substitution that changes codon 691 from arginine (CGA) to stop (TGA). This nonsense mutation is predicted to truncate 110 amino acids from the NATC-1 protein (Figure 2A,B). The mutation in the am138 mutant strain, which was induced by ENU mutagenesis, is a 186 base pair deletion that eliminates portions of exons 1 and 2 and all of intron one (Figure 2A,B). This deletion eliminates the codons for amino acids 13–59, and the mutated open reading frame is predicted to have a frame shift and encounter a stop at the new codon 15. The loss of coding sequences and early truncation suggest natc-1(am138) is likely to be a strong loss-of-function or null allele.
To test the hypothesis that mutations in natc-1 cause resistance to zinc toxicity, we analyzed the natc-1(ok2062) deletion mutation that was generated by the C. elegans knock out consortium [43]. natc-1(ok2062) is a 1,108 base pair deletion that eliminates a portion of exon 3, intron 3, exon 4, intron 4, and a portion of exon 5 (Figure 2A,B). natc-1(ok2062) mutant animals displayed significant resistance to zinc toxicity, similar to natc-1(am134) and natc-1(am138) mutant animals (Figure 1B). This result supports the hypothesis that mutations in natc-1 cause resistance to zinc toxicity.
To independently test the hypothesis that the am134 mutation in natc-1 causes resistance to zinc toxicity, we determined whether a wild-type version of natc-1 could rescue this phenotype. We generated five independently derived transgenic strains containing extrachromosomal arrays with wild-type copies of natc-1 in the background of natc-1(am134). All the transgenic strains displayed a significant decrease in zinc resistance when compared to their non-transgenic siblings. This is indicative of a more wild-type phenotype and rescue activity (Figure 1C, Figure S1A). To determine if an intact natc-1 open reading frame is necessary for rescue activity, we generated transgenic animals containing a natc-1 locus that encodes a mutant protein with a 112 amino acid deletion in the background of natc-1(am134). These transgenic animals did not display rescue of the mutant phenotype, indicating that the rescue activity of the natc-1 locus requires an intact open reading frame (Figure 1C, Figure S1B). Together, these results demonstrated that natc-1 is the gene affected by am134 and am138 (reference allele), and that mutations in natc-1 caused resistance to zinc toxicity.
To characterize the products generated from the natc-1 locus, we analyzed natc-1 mRNA. The C. elegans expressed sequence tag (EST) project isolated multiple cDNAs corresponding to natc-1, and we determined the DNA sequence of three independently derived cDNAs. The cDNA sequences were used to infer the mRNA sequence from exon 3 to the 3′ end, including the position of the polyA tail 330 nucleotides downstream of the TGA stop codon. To characterize the 5′ end of the transcript, we conducted a 5′ RACE experiment that showed the natc-1 mRNA contains a 22 nucleotide splice leader 1 (SL1) sequence that begins 14 base pairs upstream of the start codon. Together, the analysis of cDNAs and 5′ RACE indicated that the natc-1 mRNA contains 8 exons and defined the complete predicted open reading frame (Figure 2A).
The predicted NATC-1 protein contains 799 amino acids. To determine the expression pattern and sub-cellular localization of NATC-1, we generated transgenic natc-1(am138) animals expressing NATC-1 protein fused to green fluorescent protein (GFP) under the control of the native natc-1 promoter. Live animals were imaged with confocal fluorescence microscopy, and NATC-1::GFP was detected in a wide range of cells and tissues in a pattern that suggests cytoplasmic localization (Figure 3). NATC-1::GFP was detected throughout development from early larval stages through late adulthood. To confirm that the expression pattern of NATC-1::GFP is representative of the expression pattern of endogenous NATC-1, we demonstrated that the extrachromosomal array expressing NATC-1::GFP rescued the natc-1(am138) zinc-resistance phenotype (Figure 1C, Figure S1C).
Comparison of NATC-1 protein sequence to databases using the method of BLAST revealed that NATC-1 is most similar to N-α-acetyltransferase 35 proteins, which are auxiliary subunits of the NatC complex [44]. Figure 2B shows an alignment of C. elegans NATC-1 with Drosophila melanogaster and human proteins; C. elegans NATC-1 is 24% identical to human NAA35, suggesting that it may have similar biochemical functions. NATC-1 is an auxiliary subunit of the NatC complex, and C. elegans B0238.10 (NATC-2) is the predicted catalytic subunit [45]. The NatC complex catalyzes the acetylation of the N-termini of translating proteins (Figure 2C). The NatC complex specifically acetylates translating proteins that begin with Met-Ile, Met-Leu, Met-Trp, or Met-Phe [39], [40]. To identify predicted NatC target proteins, we conducted a bioinformatic analysis using the fully sequenced C. elegans genome. Approximately 4,300 proteins have Ile, Leu, Trp, or Phe in amino acid position two. These proteins represent ∼17% of the C. elegans proteome and are candidates to be acetylated by the NatC complex.
Zinc resistance displayed by natc-1 mutant animals could be explained by two general models: (1) natc-1 mutant animals have decreased levels of zinc, perhaps as a result of reduced zinc uptake or increased zinc excretion and (2) natc-1 mutant animals have normal levels of zinc but increased tolerance to high zinc toxicity. To distinguish between these possibilities, we used inductively coupled plasma mass spectrometry (ICP-MS) to measure total animal zinc content. Synchronized populations of animals were cultured with NAMM, harvested, and analyzed for zinc content. The total animal zinc content of natc-1(am134), natc-1(am138), and natc-1(ok2062) mutant animals was not consistently different from wild-type animals when cultured with or without supplemental zinc (Figure S2). These results indicate that mutations in natc-1 cause resistance to zinc toxicity by increasing the ability of the animal to tolerate excess zinc that results from a high zinc diet rather then by reducing zinc accumulation.
Mutations in haly-1 that cause resistance to high zinc toxicity were identified in the same genetic screen as mutations in natc-1 [37]. The mechanism of action of mutations in haly-1 appears to be accumulation of histidine, which is hypothesized to reduce high zinc toxicity by chelation of the ion. To determine if mutations in natc-1 may cause resistance to high zinc toxicity by a similar mechanism, we analyzed natc-1(am138);haly-1(am132) double mutant animals. If natc-1(am138) and haly-1(am132) cause zinc resistance by affecting the same pathway or process, then the resistance to high zinc toxicity phenotypes might not be additive. Interestingly, natc-1(am138);haly-1(am132) double mutant animals displayed enhanced resistance to high zinc toxicity compared to natc-1(am138) or haly-1(am132) single mutant animals (Figure S3). This result suggests that resistance to high zinc toxicity caused by natc-1 mutations may be mechanistically distinct from that caused by haly-1 mutations.
To determine if natc-1 causes resistance to additional metals, we cultured wild-type animals and natc-1(am138) animals on NAMM plates supplemented with cadmium, nickel, or copper. Concentrations that caused ∼50% sterility of wild-type animals were chosen for each metal to maximize the sensitivity of the assay. natc-1 mutant animals displayed improved growth and development compared to wild-type animals when cultured with 200 µM zinc, 20 µM cadmium, 50 µM nickel, or 300 µM copper (Figure 4A–D). These data suggest that mutations in natc-1 cause resistance to toxicity induced by both physiological and non-physiological metal ions.
To determine if mutations in natc-1 cause resistance to stressors in addition to metal ions, we analyzed heat stress. Wild-type and natc-1 mutant animals were cultured at 35°C and survival times were monitored. natc-1(am138) animals displayed a significant 19% extension of survival compared to wild-type animals (Figure 5A, Table 1). These data suggest that mutations in natc-1 cause resistance to heat toxicity. To analyze oxidative stress, we cultured animals with 40 mM paraquat and monitored survival time. natc-1(am138) animals displayed a significant 40% extension of survival compared to wild-type animals (Figure 5B, Table 1). Taken together, these data suggest that mutations in natc-1 cause resistance to a broad-spectrum of stressors, including high levels of multiple metal ions, oxidative damage, and high heat.
One hypothesis that might explain the stress resistance phenotype is that natc-1(lf) mutations stimulate the unfolded protein response. To investigate this hypothesis, we used the method of qRT-PCR to analyze the mRNA levels of the stress-induced genes hsp-4, hsp-6, and hsp-16.2. Wild-type and natc-1(am138) animals did not display statistically significant differences in mRNA levels for these genes (p>0.05), suggesting that loss of natc-1 activity does not stimulate the unfolded protein response. Furthermore, gst-4 mRNA levels, which are induced by oxidative stress and proteosomal dysfunction [46] were not significantly altered in natc-1(am138) mutants compared to wild type (p>0.05).
To determine how natc-1 activity affects longevity, we analyzed the lifespan of wild-type and natc-1(am138) mutant animals. natc-1(am138) animals displayed a significant 31% reduction in mean lifespan compared to wild-type animals (Figure 5C, Table 1). These data suggest that natc-1 is a lifespan assurance gene when animals are cultured at an optimal temperature with abundant food, conditions that minimize stress.
To analyze the regulation of natc-1, we used qRT-PCR to monitor the level of natc-1 mRNA. Because natc-1(lf) mutations cause zinc resistance, we examined the transcriptional response to high dietary zinc. natc-1 mRNA levels were not affected by 200 µM supplemental dietary zinc (p>0.05), suggesting that natc-1 transcription is not regulated to promote zinc tolerance. To further analyze regulation, we examined the insulin/IGF-1 signaling pathway because it plays a pivotal role in stress resistance in C. elegans [2]. daf-2 encodes the insulin receptor that functions to inhibit dauer formation and stress resistance; daf-2(e1370) is a partial loss-of-function mutation that causes a temperature-sensitive dauer constitutive (Daf-c) phenotype in larvae and an increased stress resistance phenotype in adults [6], [35], [38]. daf-16 encodes a FOXO transcription factor that is a crucial downstream target that is negatively regulated by the DAF-2 pathway; daf-16(mu86) is a null mutation that causes a dauer defective (Daf-d) phenotype in larvae and a reduced stress resistance phenotype in adults [7]. Interestingly, Lee et al. (2003) used bioinformatic techniques to identify a putative DAF-16 binding site (TTGTTTAC) positioned 90 base pairs upstream of the predicted start codon of the natc-1 locus [11] (Figure 6A). This predicted DAF-16 binding site is evolutionarily conserved in natc-1 homologues in Caenorhabditis briggsae and Drosophila melanogaster, suggesting it is functionally important [11]. We analyzed the genomic locus of human NAA35, the homolog of C. elegans NATC-1, and identified four predicted DAF-16 binding sites, consistent with the model that these binding sites might be conserved during evolution (Figure S4). Furthermore, Gerstein et al. (2010) and Riedel et al. (2013) used the method of chromatin immunoprecipitation followed by massively parallel DNA sequencing to demonstrate that DAF-16 protein interacts with the natc-1 locus in vivo [41], [42] (G. Ruvkun and C. Riedel, personal communication) (Figure 6A).
Based on these observations, we hypothesized that natc-1 transcription is directly regulated by binding of the DAF-16 transcription factor. According to this model, altering DAF-16 activity is predicted to alter natc-1 mRNA levels. When cultured in standard laboratory conditions, worms display a low level of DAF-16 activity because the insulin/IGF-1 pathway is strongly activated [47]. Therefore, comparing daf-16(lf) animals to control animals would not be highly informative. To test our hypothesis, we analyzed natc-1 mRNA levels in daf-2(e1370) mutant animals, since daf-2 mutant animals have been demonstrated to have increased DAF-16 nuclear localization and activity [48]. These daf-2(lf) phenotypes are similar to the consequences of food deprivation or stress, suggesting daf-2(lf) mutant worms have initiated a starvation or stress response [1]. daf-2 mutant animals displayed a ∼2-fold decrease in natc-1 mRNA levels compared to wild-type animals (Figure 6B). To confirm that this change was statistically significant, we analyzed six independent biological replicates of both wild-type and daf-2 RNA. These data are consistent with the model that DAF-16 represses transcription of natc-1, since DAF-16 activity is increased in daf-2 mutant animals. To directly test the function of daf-16, we analyzed daf-16;daf-2 double mutant animals; the decrease in natc-1 mRNA levels was abrogated in these animals, demonstrating that daf-16 is necessary for the regulation of natc-1 (Figure 6B). To confirm that daf-16 mutant animals do not contain daf-16 activity, we analyzed daf-16 transcript levels by qRT-PCR; daf-16 transcripts were detected in wild-type animals but were undetectable in the daf-16;daf-2 double mutant animals. These data suggest that DAF-16 is a transcriptional repressor of natc-1 and natc-1 is an effector of the insulin/IGF-1 signaling pathway that functions downstream of DAF-16.
The insulin/IGF-1 signaling pathway mediates entry into an alternative third larval stage called dauer that has distinctive metabolic and developmental features that promote longevity and stress resistance [3], [8], [49]. To test the function of natc-1 in insulin/IGF-1 signaling, we analyzed dauer larvae formation in natc-1(am138) animals. A single mutation in natc-1 did not cause a Daf-c phenotype (Figure 6C). However, natc-1(am138) strongly enhanced dauer formation in the daf-2(e1370) background, compared to the daf-2(e1370) single mutant animals (Figure 6C). To determine if the daf-2(e1370);natc-1(am138) Daf-c phenotype was daf-16 dependent, we analyzed daf-16(mu86);daf-2(e1370);natc-1(am138) triple mutant animals for dauer formation at 25°C. None of the daf-16;daf-2;natc-1 triple mutant animals displayed dauer formation (N = 111), indicating that daf-16 is required for this Daf-c phenotype. Together, these data demonstrated that natc-1 was necessary to inhibit dauer formation, although the effect was only observed in a sensitive genetic background.
We hypothesized that the enhancement of the Daf-c phenotype caused by a mutation in the natc-1 auxiliary subunit reflects the reduction or loss of the acetylation activity of the NatC complex. To test this hypothesis, we analyzed the function of the predicted catalytic subunit of the NatC complex, B0238.10, which we named natc-2. We used the method of feeding RNAi to reduce natc-1 and natc-2 activity in a daf-2(e1370) mutant background. A significant increase in dauer formation was observed compared to control RNAi (Figure 6D). These data indicate that the catalytic subunit encoded by natc-2 is necessary to inhibit dauer formation, suggesting that the acetylation activity of the NatC complex mediates insulin/IGF-1 signaling.
Mutations in the insulin/IGF-1 receptor daf-2 cause increased longevity, while the natc-1(am138) mutation causes a shortened lifespan [50]. To further characterize the genetic interaction between natc-1 and daf-2, we analyzed the lifespan of wild-type, natc-1(am138), daf-2(e1370), and daf-2(e1370);natc-1(am138) animals. While natc-1(am138) shortens wild-type lifespan, natc-1(am138) had no effect on the daf-2(e1370) longevity phenotype (Figure S5, Table 1). This result suggests that daf-2 activity is necessary for the natc-1(am138) mutation to cause a reduction of lifespan.
To characterize the role of natc-1 in stress resistance mediated by the insulin/IGF-1 signaling pathway, we analyzed interactions between natc-1, daf-16, and daf-2 in response to heat and high zinc stress. Single mutations of natc-1 and daf-2 cause resistance to heat stress, and daf-2(e1370);natc-1(am138) double mutant animals displayed enhanced stress resistance compared to either single mutant animal (Figure 7A). One interpretation of this additivity is that natc-1 and daf-2 function in the same pathway, but neither single mutation maximizes the potential of the pathway to increase stress resistance; this is consistent with the fact that the daf-2(e1370) allele causes a partial loss-of-function. The alternative interpretation is that natc-1 and daf-2 function in parallel to mediate stress resistance.
Compared to wild-type animals, daf-16(mu86) animals displayed a mild sensitivity to heat stress (Figure 7B, Table 1). daf-16(mu86);natc-1(am138) double mutant animals displayed heat stress resistance similar to natc-1 single mutant animals. (Figure 7B, Table 1). These data indicate that natc-1 is epistatic to daf-16 with respect to heat stress resistance, consistent with the model that natc-1 is a downstream effector that is negatively regulated by daf-16.
To further analyze this pathway, we determined if daf-16 was necessary for natc-1(am138) to cause resistance to zinc toxicity. Attempts to analyze daf-2 and daf-2;natc-1 mutant animals for resistance to high zinc toxicity were not successful, since supplemental zinc caused a high rate of dauer formation in these mutant animals, precluding an analysis of growth rates (Figure S6). natc-1(am138) caused similar resistance to high zinc toxicity in both a wild-type and daf-16(mu86) background (Figure 7C), indicating that daf-16 function is not necessary for the natc-1(am138) zinc-resistance phenotype. These data support the model that natc-1 functions downstream of daf-16 to mediate zinc resistance, and together these results suggest that natc-1 is acting as a key downstream effector of the C. elegans insulin/IGF-1 signaling pathway (Figure 8A,B).
The DAF-16 FOXO transcription factor is the key downstream target of the insulin/IGF-1 signaling pathway that promotes stress resistance and longevity [5], [35], [50]–. A critical goal in this field is to identify the functionally significant targets of DAF-16, since these genes are hypothesized to mediate stress resistance, nutrient utilization, and aging. A variety of approaches have been used to identify DAF-16 target genes, including bioinformatic, genomic, and reverse genetic techniques [9]–[17]. Lee et al. (2003) used the biochemically demonstrated DNA binding sequence of DAF-16 to bioinformatically identify predicted binding sites in the C. elegans genome. One of these sites is positioned 90 base pairs upstream of the natc-1 start codon, suggesting that DAF-16 binds the natc-1 promoter. A similar FOXO transcription factor binding site is present in the promoters of genes homologous to natc-1 in Caenorhabditis briggsae, Drosophila melanogaster, and humans, suggesting that this transcription factor binding site has been conserved during evolution and is functionally significant [11]. Gerstein et al. (2010) and Riedel et al. (2013) used the technique of chromatin immunoprecipitation followed by DNA sequencing to identify in vivo binding sites for DAF-16 (G. Ruvkun and C. Riedel, personal communication) [41], [42]. These studies independently demonstrated that DAF-16 binding is significantly enriched at the natc-1 locus, consistent with the hypothesis that DAF-16 occupies the predicted binding site.
Here we demonstrated that natc-1 is transcriptionally repressed by DAF-16. Worms cultured in standard laboratory conditions with abundant food have high activity of the DAF-2 insulin/IGF-1 receptor and low activity of DAF-16. By contrast, daf-2 loss-of-function mutant animals have high activity of DAF-16 and display enhanced stress resistance and extended longevity. daf-2(lf) mutant animals displayed reduced levels of natc-1 transcripts, and natc-1 transcripts were restored to WT levels in daf-16(lf);daf-2(lf) double mutant animals. These results suggest that daf-16 activity reduces the level of natc-1 transcripts, and daf-2 activity increases the level of natc-1 transcripts by negatively regulating daf-16. Lee et al. (2003) used the technique of Northern blotting to examine natc-1 transcript levels and did not detect a substantial difference between wild-type and daf-2(e1370) mutant animals [11]. By contrast, Riedel et al. (2013) used the technique of high-throughput sequencing of mRNA (mRNA-Seq) [42] and detected a significant decrease (∼2-fold) in natc-1 mRNA levels in daf-2(e1370) animals compared to wild-type animals (G. Ruvkun and C. Riedel, personal communication). A possible explanation of these findings is that the techniques qRT-PCR and RNA-Seq, which gave similar results, are more quantitative than Northern blotting and were able to detect a small but statistically significant difference in transcript levels. Based on the presence of an evolutionarily conserved DAF-16 binding site in the natc-1 promoter, the in vivo binding of DAF-16 to the natc-1 locus, and our observed alterations in natc-1 transcript levels caused by mutations in daf-2 and daf-16, we propose the model that DAF-16 is a direct transcriptional repressor of natc-1 (Figure 8B).
Our genetic analysis showed that natc-1 inhibits stress resistance and dauer formation and genetically interacts with the insulin/IGF-1 signaling pathway, suggesting that natc-1 is a physiologically relevant target of DAF-16. These results do not exclude the possibility that natc-1 might be regulated by an additional pathway(s) in parallel to the insulin/IGF-1 pathway; indeed, the natc-1 promoter has been reported to interact with multiple transcription factors such as PQM-1, SKN-1, and PHA-4, suggesting there are additional regulatory control mechanisms [41], [53]. natc-1 mutant animals were discovered in an unbiased forward genetic screen for resistance to the stress of high dietary zinc. Further analysis revealed that natc-1(lf) mutations cause resistance to a wide range of stressors, such as heat, oxidation, and multiple metals, similar to daf-2(lf) mutations. daf-2 functions upstream of daf-16 in a signaling pathway, and daf-16(lf) mutations are epistatic to daf-2(lf) mutations. By contrast, our model that natc-1 functions downstream of daf-16 predicts that the stress resistance caused by natc-1(lf) mutations is epistatic to daf-16(lf) mutations. Indeed, we observed that the resistance to heat and high levels of dietary zinc caused by natc-1(lf) mutations was epistatic to daf-16(lf) mutations. Thus, both molecular and genetic studies support the model that natc-1 is directly targeted by DAF-16 to promote stress resistance. Consistent with this model, we demonstrated that natc-1 inhibits dauer formation, since natc-1(lf) mutations enhanced the dauer constitutive phenotype of daf-2(lf) mutations.
Although a variety of DAF-16 target genes have been demonstrated to interact genetically with the insulin/IGF-1 signaling pathway, these target genes are primarily activated by daf-16, and loss-of-function of these target genes impairs the stress resistance mediated by daf-16 activity. For example DAF-16 promotes transcription of the superoxide dismutase sod-3 [5]. sod-3 functions to promote resistance to oxidative stress [54]. Thus, sod-3 represents a class of DAF-16 targets that are transcriptionally activated by DAF-16 and promote stress resistance (Figure 8A). To our knowledge, our results with natc-1 are the first demonstration of a direct DAF-16 target gene that is repressed by DAF-16 to promote stress resistance (Figure 8A). Thus, these studies make a novel contribution to the understanding of how the activity of the insulin/IGF-1 signaling pathway mediates stress resistance.
In addition to stress resistance, DAF-16 also promotes longevity [55]. Because mutations in natc-1 reduce longevity, DAF-16 likely regulates lifespan through a natc-1-independent mechanism.
N-terminal acetyltransferases (NATs) are multi-subunit enzymes that catalyze the transfer of the acetyl group of acetyl coenzyme A to the α-amino group of the first amino acid of a target protein. N-terminal acetylation is a widespread modification that affects the majority of eukaryotic proteins; for example, ∼80–85% of human proteins are N-terminally acetylated [56]. Eukaryotes possess multiple NAT complexes (NatA-NatE) that vary in subunit composition and substrate specificity [57]. NAT activity is relevant to human health, since mutations in a human NAT gene are associated with Ogden syndrome, an X-linked disorder that is lethal in infancy [58]. Whereas the biochemical activity of NAT enzymes is well characterized, the functional roles of these enzymes are only beginning to be explored.
The NatC complex is comprised of the catalytic subunit Naa30 and the auxiliary subunits Naa35 and Naa38, and it acetylates proteins with the N-terminal sequences Met-Leu, Met-Phe, Met-Ile, and Met-Trp [39], [40], [44], [59], [60]. Genetic studies in yeast revealed NatC complex subunits are necessary for dsRNA virus particle assembly and WT growth rate in media lacking a fermentable carbon source [61], [62]. The Arabidopsis thaliana Naa30 homolog was identified in a screen for photosynthesis altered mutants, and Naa30 mutants display decreased chloroplast density and slow growth [63]. The zebrafish Naa35 homolog is the embryonic growth-associated protein (EGAP), and knock down studies with morpholinos suggest that it is necessary for WT growth and development [64]. The rat Naa35 homolog was discovered based on increased expression in healing corneal epithelium, and it is highly expressed in developing rat cornea and skin [65]. In human cells, reducing the levels of NatC complex subunits causes reduced cell viability and p53-dependent apoptosis [66]. These results suggest that the NatC complex functions to promote growth and development in a wide range of organisms.
Our studies of C. elegans natc-1 make a unique contribution to understanding the biological function of NatC, since these results are the first molecular and genetic characterization of an animal with a NatC subunit mutation. Genetic studies demonstrated that mutations in natc-1 increased resistance to multiple environmental stressors including excess metal, heat, and oxidation. These results suggest that natc-1 activity reduces stress resistance. Environments with low food, high population density, and high temperatures promote formation of dauer larvae. Dauers are a stress resistant form that allows animals in unfavorable environmental conditions to suspend development, resist environmental stresses, and be prepared to resume reproductive development when conditions improve [1]. Dauer formation is mediated by the insulin/IGF-1 signaling pathway [1]. The capacity to form dauer larvae highlights the importance of maintaining a balance between promoting growth and reproduction and surviving environmental stressors. In a sensitized genetic background, natc-1 activity inhibited dauer formation. Because a developmental switch mediates the decision between dauer larvae and larvae destined for reproductive development, these findings indicate that natc-1 activity promotes a development fate characterized by growth and reproduction. These findings identify novel phenotypes associated with a subunit of the NatC complex.
To determine how the NatC complex might execute these newly discovered functionalities, we bioinformatically identified predicted NatC target proteins in C. elegans. NatC might modulate processes such as stress resistance and dauer formation by acetylating groups of proteins with similar functionalities. To investigate this hypothesis, we performed a gene ontology (GO) analysis. Predicted NatC target proteins in C. elegans were enriched for 20 functional classes. To determine if NatC has an evolutionarily conserved preference for these functional classes, we used bioinformatics to predict NatC target proteins in humans. GO analyses of human NatC target proteins identified 11 functional classes that were enriched. Electron carrier and oxidoreductase activity were enriched in C. elegans and human GO analyses suggesting that the NatC complex may regulate a subset of proteins associated with these functionalities (Table S1). We further determined if predicted NatC target proteins were enriched for a specific cellular localization. GO analysis of NatC target proteins identified 4 cellular component terms enriched in C. elegans and 18 enriched in humans. NatC target proteins were enriched for mitochondrial localization in both C. elegans and humans (Table S2). These conserved functionalities and cellular components may inform future experimental efforts aimed at understanding how the NatC complex regulates physiology such as dauer entry and stress tolerance.
To identify protein targets of the NatC complex that might contribute to the mutant phenotype by having altered acetylation in natc-1(lf) mutant animals, we identified predicted protein targets that are implicated in zinc metabolism and stress resistance. Published genes implicated in C. elegans zinc metabolism include the cation diffusion facilitator (CDF) zinc transporters cdf-1, sur-7, cdf-2, and ttm-1 [67]–[70], the metallothioneins mtl-1 and mtl-2 [71], and haly-1 [37]. Of these seven genes, only cdf-2 is a putative target of the NatC complex. Mutations in cdf-2 affect zinc accumulation [69], whereas natc-1(lf) mutations do not alter zinc accumulation, suggesting that cdf-2 is not a critical target of the NatC complex in zinc resistance. Rather, we hypothesize that natc-1 mutations cause high zinc resistance by triggering mechanisms that allow animals to ameliorate the toxicity of high levels of zinc. This hypothesis is supported by the observation that natc-1 mutations cause resistance to a wide variety of stressors. Genes implicated in stress resistance include those encoding proteins involved in reactive oxygen species metabolism and dauer formation. The NatC complex is predicted to target several proteins that met these criteria; sod-1, sod-2, and sod-3 encode superoxide dismutases that increase oxidative stress resistance [54], mev-1 encodes cytochrome b, a subunit of the mitochondrial respiratory chain complex II, and frh-1 encodes a frataxin ortholog that promotes the oxidative stress response [72], [73]. Several predicted protein targets are encoded by genes that influence dauer formation; tph-1 encodes a tryptophan hydroxylase, daf-36 encodes an oxygenase, and cyp-35a3 encodes one of 42 cytochrome P450 proteins predicted to be targeted by the NatC complex [17], [74], [75]. Previous studies of stress responses and dauer formation have largely focused on the importance of transcriptional regulation. Our work suggests posttranslational modifications like N-terminal acetylation might also play an important role, and future proteomic analyses may help identify key effectors of stress tolerance and dauer formation.
Our findings are relevant to a general principle that organisms must balance growth and reproduction, which is facilitated by nutrient-rich environments, with stress resistance and quiescence, which are adaptive in nutrient poor and/or high stress environments. Plants that evolved in low-resource, stressful environments share a common set of traits, including relatively low rates of growth, photosynthesis, tissue turnover, and nutrient absorption [76]–[78]. This has been named the “stress resistance syndrome” (SRS), since these plants are resistant to a wide spectrum of physiologic stressors [79]. SRS may represent an adaptive strategy for coping with harsh environmental conditions [76], and this process has interesting analogies to dauer formation in C. elegans. Both dauer formation and SRS represent organisms balancing growth and stress tolerance to promote survival. The NatC complex may have an evolutionarily conserved role in mediating this balance between growth and stress tolerance. Consistent with this hypothesis, loss-of-function of the catalytic subunit of the Arabidopsis NatC complex causes decreased photosynthetic activity [63], an SRS trait. Additionally, our genetic studies demonstrate that loss-of-function of natc-1 promotes dauer formation and inhibits reproductive development, a trait analogous to SRS. Therefore, we propose that the NatC complex may have evolutionarily conserved functions in maintaining the balance between promoting growth and reproduction and resisting stressful environmental conditions.
Given that NatC functions to mediate the balance between growth and stress resistance, which is finely tuned by environmental conditions, it is important to determine how the activity of NatC is regulated. However, little is known about the regulation of these enzymes. In yeast, the natc-1 homolog (MAK10) protein levels are glucose repressible, but it was not established whether regulation occurs at the level of RNA or protein [61]. Here we demonstrated that natc-1 is negatively regulated at the level of transcription by DAF-16. The insulin/IGF-1 signaling pathway responds to environmental cues, such as nutrient availability, temperature, and dauer pheromone, by regulating the activity of DAF-16. Therefore, our findings establish a direct link between environmental sensing mediated by the insulin/IGF-1 signaling pathway and protein N-terminal acetylation mediated by NatC. These results make a new contribution to understanding NatC regulation in several ways. First, NAT subunits have not previously been reported to be regulated at the level of transcription. Second, this is a novel demonstration that a NAT complex is regulated by the insulin/IGF-1 signaling pathway, linking an environmental sensing pathway to the regulation of protein N-terminal acetylation (Figure 8A,B). NatC complexes appear to have an evolutionarily conserved role in modulating growth and stress resistance, and our findings suggest that they may also have a conserved role in responding to the insulin/IGF-1 signaling pathway.
We have molecularly characterized two genes identified by screening for mutant animals that display increased tolerance to excess dietary zinc [36]: natc-1 and haly-1 [37]. These are the only genes that have been demonstrated to cause resistance to excess dietary zinc in an animal, and mutations in these two genes appear to act by very different mechanisms. First, these genes encode proteins with distinct functions. haly-1 encodes an enzyme that metabolizes histidine, and haly-1 mutant animals display increased levels of histidine. natc-1 encodes a subunit of a protein N-terminal acetylation complex, suggesting that protein N-terminal acetylation of many proteins is altered in these mutant animals, although this prediction has not been biochemically tested. Second, the spectrum of stress resistance caused by mutations in these two genes is distinct. Increased histidine appears to chelate and detoxify excess zinc and nickel, but the effect is quite specific since haly-1 mutant animals are not resistant to the toxicity caused by other metals [37]. By contrast, here we demonstrated that natc-1 mutations cause broad-spectrum stress resistance, including resistance to multiple metals, heat, and oxidation. Consistent with the model that these genes function by distinct mechanisms, the resistance to excess zinc toxicity caused by mutations of natc-1 and haly-1 was additive.
These findings raise a general question about stress resistance; how does a mutation in a single gene such as natc-1 promote resistance to a broad-spectrum of stresses? (1) One possibility is that the single-gene mutation results in a cascade of events that changes the activity of many proteins in the cell. In this model, each specific change in activity might mediate resistance to only one or a small number of stressors. For example, changes in haly-1 activity only mediate resistance to zinc and nickel. However, the combination of many different changes in activity could mediate broad-spectrum resistance. This is an attractive model for daf-2 mutant animals, which display broad-spectrum stress resistance and are documented to have changes in the expression of many genes as a result of the regulation of the DAF-16 transcription factor. This model is also attractive for natc-1, since this enzyme is predicted to mediate the N-terminal acetylation of many different proteins. (2) An alternative model is that diverse environmental stresses converge on a single type of important molecular damage. For example, heat, oxidation, and excess metals may all cause toxicity as a result of similar damage, such as protein unfolding. In this model, changing the activity of a single gene might confer broad-spectrum stress resistance by enhancing tolerance to the major form of cellular damage. These two basic models represent extremes of a continuum, and are not mutually exclusive. Our results document that resistance to high zinc toxicity can be increased by mutations that cause specific or broad-spectrum stress resistance, contributing to a conceptual framework for understanding stress resistance.
C. elegans strains were cultured at 20°C on nematode growth medium (NGM) seeded with E. coli OP50 unless otherwise noted [80]. The wild-type C. elegans strain and parent of all mutant strains was Bristol N2. The following mutations were used: daf-16(mu86) [7] is a strong loss-of-function or null mutation of the DAF-16 forkhead transcription factor; daf-2(e1370) [6] is a partial loss-of-function mutation of the DAF-2 insulin/IGF-1 receptor; haly-1(am132) [37] is a strong loss-of-function or null mutation of the HALY-1 histidine ammonia lyase. natc-1(am134) and natc-1(am138) were identified in a genetic screen for resistance to high zinc toxicity [36], backcrossed four times to wild type, and are described here; natc-1(ok2062) was obtained from the C. elegans knockout consortium [43] and backcrossed four times to wild type. The back crossing procedure replaced ∼94% of the genome of mutant strains with wild-type DNA that has not been exposed to chemical mutagenesis, minimizing background mutations. Double mutant animals were generated by standard methods, and genotypes were confirmed by PCR or DNA sequencing.
Hermaphrodites were cultured on NGM, and one embryo was transferred to a 35×10 mm Petri dish containing NAMM [36] supplemented with zinc sulfate (ZnSO4), cadmium chloride (CdCl2), nickel chloride (NiCl2), or copper chloride (CuCl2) and 5× concentrated E. coli OP50 as a food source. Dishes were analyzed daily until progeny were observed or the animal died, except Figure S3 where dishes were analyzed only until day 6. Animals that generated one or more live progeny were scored as “fertile adults.” To determine the metal concentration for these assays, we generated dose response curves of fertility for each metal using wild-type animals and selected the concentration that caused ∼50% of wild-type animals to fail to display fertility (Figure 4).
Large populations of animals were cultured on NAMM supplemented with 0 or 200 µM zinc sulfate (ZnSO4). The animals were desiccated to determine dry weight, and total zinc content was determined by ICP-MS as described by Murphy et al. (2011) [37].
Live transgenic animals were immobilized using levamisole in phosphate buffered saline (PBS) and mounted onto a thin pad of ∼7.5% agarose. More than 100 transgenic animals were analyzed, and representative images are presented. All images were captured on a PerkinElmer spinning disk confocal microscope utilizing Volocity imaging software.
Gravid adult hermaphrodites were bleached to obtain embryos. Embryos were allowed to hatch in M9 buffer to synchronize at the L1 stage and cultured at 15°C on NGM. For heat stress assays, animals were shifted to 35°C on day 1 of adulthood. Animals were analyzed hourly for spontaneous or provoked motility and pharyngeal pumping; animals displaying none of these traits were scored as dead. Animals were briefly exposed to room temperature (24–25°C) for scoring. For the experiment shown in Figure 7A, animals were cultured continually at 35°C until hourly scoring began at 12 hours; summary statistics were not calculated in this case because some data points were not collected.
For oxidative stress assays, day 3 adults were transferred to NGM dishes supplemented with 40 mM methyl viologen dichloride hydrate (paraquat), fed E. coli OP50 and cultured at 20°C. We analyzed day 3 adults to avoid the high frequency of matricidal hatching in response to oxidative stress displayed by younger adults. Animals were scored every 12 hours for survival.
For lifespan assays, L4 animals were cultured on NGM at 20°C (defined as day 0) and fed E. coli OP50. Adult hermaphrodites were transferred to fresh Petri dishes every day until the cessation of progeny production and analyzed every day for survival.
In heat stress, oxidative stress and lifespan assays, animals that displayed matricidal hatching or a vulval-bursting phenotype were omitted from the analysis.
To analyze dauer formation, we transferred embryos to NGM with E. coli OP50 at 15–25°C until adult animals began to lay embryos, approximately 3–5 days depending on the temperature. Animals were scored as dauer or non-dauer using a dissecting microscope based on the visible radial constriction phenotype [81]. To determine the effect of zinc, we conducted this assay on NAMM supplemented with zinc sulfate (ZnSO4).
To analyze genetic regulation of dauer formation, we performed feeding RNAi as described by Kamath et al. (2001) with minor modifications [82]. Briefly, daf-2(e1370) P0 hermaphrodites and F1 progeny were incubated at 20°C and continuously fed RNAi expressing bacteria. F1 progeny were scored as dauer or non-dauer after approximately 4 days. We used the empty vector control (L4440) and clones encoding dsRNA corresponding to T23B12.4 (natc-1) and B0238.10 (natc-2) from the Ahringer RNAi Library [83]. The DNA sequence of each clone was confirmed by standard methods.
Plasmid pJM5 is pBlueScript SK+ (Stratagene) containing a 3,356 base pair fragment of C. elegans genomic DNA from fosmid WRM067bF02 that extends 139 base pairs upstream of the predicted natc-1 start codon and 395 base pairs downstream of the predicted stop codon. To generate pJM6, we modified pJM5 by digestion with BstEII (New England Biolabs) and religation resulting in a 382 base pair deletion that removes parts of exons 2 and 3 and all of intron 2. The resulting pJM6 mutant open reading frame is predicted to truncate at amino acid 33 in a premature stop codon (TAG). To generate the Pnatc-1::NATC-1::GFP::unc-54 3′UTR translational fusion protein construct (pJM8), we inserted the natc-1 genomic locus (without the stop codon) into pBlueScript SK+ with the GFP coding sequence and the unc-54 3′ UTR. The DNA sequence of each plasmid was confirmed by standard methods.
Transgenic animals were generated by injecting natc-1(am134) hermaphrodites with pJM5 or pJM6, and natc-1(am138) hermaphrodites with pJM8. All injections were done with the dominant Rol marker pRF4 [84]. We selected independently derived Rol self progeny that transmitted the Rol phenotype. These transgenes formed extrachromosomal arrays, since the Rol phenotype was transmitted to only a sub-set of the self-progeny. To analyze transgenic rescue of the natc-1(am134) or natc-1(am138) resistance to high zinc toxicity phenotype, we calculated the fraction of transgenic animals on baseline and high zinc concentrations. nonRol animals were presumed to lack the extrachromosomal array and were thus non-transgenic. We defined rescue as a significant decrease in percentage of transgenic animals able to survive (and thus be quantified) on toxic zinc conditions (300 µM supplemental zinc) compared to the baseline zinc concentration (0 µM supplemental zinc), as described by Murphy et al. (2011) [37].
To identify protein targets of the NatC complex, we wrote a custom Perl script that computationally identified predicted NatC targets in C. elegans and human protein databases from the National Center for Biotechnology Information (NCBI). Gene ontology (GO) analysis was performed using GOrilla [85] by comparing predicted NatC targets to the entire proteome for each species. We report significant GO functional terms (p<0.001) according to Eden et al. (2009) [85].
Three natc-1 cDNA clones (EST) were obtained from the National Institutes of Genetics, Japan (yk194g4, yk262c3, and yk420a1). We determined the complete sequence of these cDNAs using standard techniques. These data were used to infer the mRNA sequence from exon 3 to the polyA tail attached 330 nucleotides downstream of the TGA stop codon. To experimentally define the 5′ end of the natc-1 mRNA, we used 5′ RACE System V2.0 (Invitrogen) according to the manufacturer's instructions. These data were used to infer the natc-1 mRNA sequence from the 22 nucleotide splice leader 1 (SL1) sequence that begins 14 base pairs upstream of the start codon to exon 3.
To generate synchronous populations of worms for RNA extraction, we treated gravid adult hermaphrodite animals with a mixture of bleach and 4M sodium hydroxide (NaOH) and cultured embryos overnight in M9 solution at 20°C, resulting in L1 stage arrest. L1 larvae were transferred to NGM plates at 20°C, fed E. coli OP50, and allowed to develop to the L4 stage (approximately 2 days). RNA isolation and cDNA synthesis were performed as described by Davis et al. (2009) [86]. Quantitative, real-time PCR was performed using an Applied Biosystems 7900HT Fast Real-Time PCR system and the Applied Biosystems SYBR Green Master Mix. mRNA fold change was calculated using the comparative CT method [87]. Forward and reverse amplification primers were: rps-23 5′- aaggctcacattggaactcg and 5′- aggctgcttagcttcgacac; mtl-1 5′-ggcttgcaagtgtgactgc and 5′-cctcacagcagtacttctcac; natc-1 5′-tcagctttacgggtccaatg and 5′-ccgaaaatgctctgtggttac; daf-16 5′- gacggaaggcttaaactcaatg and 5′- gagacagattgtgacggatcg.
All data were analyzed utilizing the two-tailed students t-test of samples with unequal variance unless otherwise specified. For binary data such as dauer entry and fertility, the Chi-squared test was utilized. P-values less than 0.05 were considered statistically significant.
|
10.1371/journal.pntd.0001453 | The Genetic and Molecular Basis of O-Antigenic Diversity in Burkholderia pseudomallei Lipopolysaccharide | Lipopolysaccharide (LPS) is one of the most important virulence and antigenic components of Burkholderia pseudomallei, the causative agent of melioidosis. LPS diversity in B. pseudomallei has been described as typical, atypical or rough, based upon banding patterns on SDS-PAGE. Here, we studied the genetic and molecular basis of these phenotypic differences. Bioinformatics was used to determine the diversity of genes known or predicted to be involved in biosynthesis of the O-antigenic moiety of LPS in B. pseudomallei and its near-relative species. Multiplex-PCR assays were developed to target diversity of the O-antigen biosynthesis gene patterns or LPS genotypes in B. pseudomallei populations. We found that the typical LPS genotype (LPS genotype A) was highly prevalent in strains from Thailand and other countries in Southeast Asia, whereas the atypical LPS genotype (LPS genotype B) was most often detected in Australian strains (∼13.8%). In addition, we report a novel LPS ladder pattern, a derivative of the atypical LPS phenotype, associated with an uncommon O-antigen biosynthesis gene cluster that is found in only a small B. pseudomallei sub-population. This new LPS group was designated as genotype B2. We also report natural mutations in the O-antigen biosynthesis genes that potentially cause the rough LPS phenotype. We postulate that the diversity of LPS may correlate with differential immunopathogenicity and virulence among B. pseudomallei strains.
| Burkholderia pseudomallei is an environmental Gram-negative bacterium and the cause of melioidosis, an often life-threatening disease affecting people in Southeast Asia and northern Australia. Melioidosis is usually contracted by bacterial inoculation, ingestion or inhalation. Effective vaccines for melioidosis are currently unavailable. This organism contains a large genome, which varies greatly among strains due to a high frequency of genetic recombination. We report here on diversity of lipopolysaccharides (LPS) in this species, a major component of the bacterial outer membrane and a known immunogenic virulence factor. We developed LPS genotyping techniques to study frequency of two major LPS types, known as typical and atypical LPS, in B. pseudomallei strains collected from two endemic regions: Southeast Asia and Northern Australia. LPS genotype variation differed among B. pseudomallei populations. During the investigation, we discovered a new LPS genotype in a sub-population group of B. pseudomallei in Australia. We postulate that such differences are likely to be associated with variable immunopathogenicity and clinical presentation in the human host.
| Lipopolysaccharide (LPS) is a major component of the outer membrane of Gram-negative bacteria, playing an important role in cell integrity and in signaling host innate immune response [1]. Structurally, LPS is composed of three major components: lipid A, the bacterial endotoxin that is embedded in the phospholipid bilayer of the outer membrane; core-oligosaccharide; and O-antigen. These three components are linked together as a part of the bacterial outer membrane. In a highly pathogenic bacterial species, such as Burkholderia pseudomallei, LPS has a major role in stimulating host innate immune response during infection [2]. B. pseudomallei LPS has been classified as a type II O-polysaccharide (O-PS) and is one of 4 different surface polysaccharides produced by this pathogen [3]. Previous studies have shown that B. pseudomallei LPS is required for serum resistance and virulence [4]. It has been well established in many bacterial diseases that overstimulation of the host cells by LPS can lead to the features of septic shock [5]. Likewise for B. pseudomallei, septicemia is a major cause of death. In Northeast Thailand especially in Ubon Ratchathani Province where melioidosis is highly endemic, the average incidence rate of melioidosis is 12.7 cases per 100,000 people per year with the average of 42.6% of mortality rate [6]. Cellular recognition of LPS by the innate immune system triggers the proinflammatory cytokines by host cells, which aids in the clearance of the pathogen. Previous studies have supported a potential role for B. pseudomallei LPS in protective immunity, with high concentrations of antibodies to LPS associated with survival in severe melioidosis [7], [8]. As a result, LPS has been used in vaccine development and provided protective immunity in a murine model of melioidosis [2]. In addition, it was demonstrated that LPS had an important role in bacterial virulence because the LPS mutant B. pseudomallei strain SRM117, which lacked the O-antigenic polysaccharide moiety was more susceptible to macrophage killing during the early phase of infection than its parental strain 1026b [9].
A previous study [10] identified LPS diversity based upon electrophoretic mobility with SDS-PAGE and detection using immunoblot analysis. This diversity included two serotypes (A and B) possessing different electrophoretic ladder profiles and a rough type that did not contain the ladder patterns; all were antigenically distinct [10]. Molecular structure of O-antigen serotype A or typical type has been described as the unbranched heteropolymers consisting of disaccharides repeats of -3)-β-D-glucopyranose-(1-3)-6-deoxy-α-L-talopyranose-(1- in which approx. 33% of the L-6dTalp residues bear 2-O-methyl and 4-O-acetyl substituents whereas the other L-6dTalp residues carry only 2-O-acetyl substituents [11]. We note that the structures are not known for any of the other B. pseudomallei O-antigens. B. thailandensis, a genetically related non-pathogenic species, has LPS that is cross-reactive to sera obtained from B. pseudomallei and B. mallei infections, and this has led to the development of a vaccine for melioidosis using LPS from B. thailandensis [12]. B. mallei, the causative agent of glanders, has O-antigen structure similar to those found in B. pseudomallei and B. thailandensis, except that it has different side-group modifications at the L-6dTalp residues which lack the acetylation at the O-4 position [13]. These structural differences are associated with the absence of oacA gene in B. mallei. oacA encodes for O-antigen acetylase A in B. thailandensis and its homolog in B. pseudomallei K96243 is identified as BPSL1936 [14].
B. pseudomallei genomes are very diverse due to horizontal gene transfer events [15], [16] and dynamic changes in repeated sequences [17]. This results in diverse phenotypic characteristics such as bacterial colony morphotypes [18], and importantly, may be implicated in the diverse clinical manifestations observed among melioidosis patients. The latter range from asymptomatic cases, to localized infections, to whole body sepsis, along with differential seroreactivities [19], all of which may be correlated with the great genomic diversity in this species [15], [17]. Nevertheless, the specific roles of genetic diversity in B. pseudomallei in differential clinical presentations of melioidosis requires further analysis, as clinical studies suggest host risk factors are the major determinant of disease severity [20]. Because LPS phenotypic diversity is important for serology and diagnostics, we investigated the genetic and molecular basis of differential LPS phenotypes in a large B. pseudomallei population. Bioinformatics, phenotypic characterization, as well as, population genetics approach were used in this study to better understand this important trait.
Artemis and Artemis Comparison Tool (ACT) software [21] was used to display and compare multiple B. pseudomallei genomes. Genomes and nucleotide sequences used in this study are listed in Table 1. Mutations in O-antigen biosynthesis genes were identified using basic homologous gene based alignments.
Multiplex-SYBR-Green PCR assays were designed to target 3 different LPS genotypes. Gene wbiE of B. pseudomallei K96243, gene BUC_3396 of B. pseudomallei 576, and gene BURP840_LPSb16 of B. pseudomallei MSHR840 were used as the PCR markers to investigate frequency of LPS genotypes A, B, and B2, respectively (Figures 1&2). PCR primers used in this study are as follows: wbiE_F, 5′-TCAAACCTATCCGCGTGTCGAAGT-3′; wbiE_R, 5′-TCGTCGTCAAGAAATCCCAGCCAT-3′; BUC3396_ F, 5′-AATCTTTTTCTGATTCCGTCC-3′; BUC3396_R, 5′ -ACCAGAAGACAAGGAGAAAGGCCA-3′; BURP840_LPSb16_F, 5′-AACCGGGTAGTTCGCGATTAC-3′; and BURP840_LPSb16_R, 5′-ATACGCCGGTGTAGAACAGTA-3′. The PCR assay was conducted in 10-µL reaction mixtures containing 1× SYBR-Green master mix (Applied Biosystems, USA), 0.3 µM of each PCR primer, and 0.1 to 1.0 ng of DNA template. Most tested DNA samples were made in collaborative laboratories in Thailand and Australia using various DNA extraction techniques. The reactions were performed on an ABI 7900HT Sequence Detection System (Applied Biosystems, USA) utilizing 40 cycles. Each cycle contained two steps: denaturation at 95°C for 15 s and annealing at 60°C for 30 s. The PCR products were further analyzed by melting them continuously from 60°C to 95°C to generate a dissociation curve. The melting temperatures of PCR amplicons for genes wbiE, BUC_3396, and BURP840_LPSb16 were constant at 87.0°C, 83°C, and 88.5°C, respectively. We used this assay to analyze DNA templates from 999 diverse B. pseudomallei strains isolated from clinical, animal, and environmental samples from Australia (n = 600), Thailand (n = 349), Malaysia (n = 27), Vietnam (n = 7), Papua New Guinea (n = 2), and unknown countries in Southeast Asia (n = 14), as well as 77 B. thailandensis strains, 2 B. thailandensis-like spp. strains, and 37 strains of unknown soil bacteria.
Whole genome sequencing was performed using 454 sequencing technology (Roche, USA) by US Army Edgewood Chemical Biological Center (ECBC), MD, USA. Artemis –based analysis and BLAST were used to annotate the O-antigen biosynthesis genes of B. pseudomallei strains MSHR840, MSHR139, and MSHR1950. DNA sequencing for wbiI and oacA genes was performed using ABI 3130×l Genetic Analyzer (Applied Biosystems, USA).
LPS identification and characterization: Techniques for LPS extraction and SDS-PAGE analysis followed a previous study [10]. Immunoblot analysis was performed using sera from melioidosis patients with known infection with B. pseudomallei LPS genotype A or B strains as the primary antibodies. Horse radish peroxidase (HRP) – conjugated anti-human IgG was used as the secondary antibody in a standard immunoblot analysis. Monoclonal antibody 3D11, the B. mallei LPS-specific mAb (Research Diagnostics Inc., USA), was used as a primary antibody in the immunoblot analysis of the oacA mutant strains.
Select B. pseudomallei strains were tested for growth, multiplication, and survival in the presence of 30% normal human serum (NHS) as previously described [4] with some modifications. Briefly, each B. pseudomallei strain was inoculated in a 2 mL of TSBDC media and incubated overnight at 37°C and 250 rpm in an orbital shaker. The overnight culture (100 µL) was used to inoculate 3 mL of TSBDC media and then incubated at the same conditions for 4 hr to reach mid exponential growth phase. Serum susceptibility tests were performed in 1.5 mL microfuge tubes containing 100 µL of bacterial culture, 300 µL of NHS (Lonza Inc., USA), and 600 µL PBS. The mixture was incubated at 37°C for 2 hr, and then the number of viable bacterial cells was determined using plate counting. B. pseudomallei 1026b and E. coli HB101 were used as positive and negative controls in this study, respectively.
Nucleotide sequences and annotations of the O-antigen biosynthesis genes in B. pseudomallei strains MSHR840, MSHR139, and MSHR1950LPS were submitted to GenBank under accession nos. GU574442, HM852063, and HM852062, respectively.
To better understand the diversity of genes responsible for the biosynthesis of O-antigen moiety of the LPS in B. pseudomallei, we first used a comparative analysis of all publicly available B. pseudomallei genomes to identify differences within LPS biosynthetic genes. Three different O-antigen biosynthesis gene categories, or genotypes, were identified. Secondly, we examined the genotype frequencies in B. pseudomallei populations using PCR assays targeting each of these genetic types. Thirdly, we correlated LPS genotypes with their differential phenotypes (serotypes). This led to our discovery of a natural mutation in an O-antigen biosynthesis gene in a clonal panel of B. pseudomallei strains isolated from a single human host. The adaptability of B. pseudomallei strains through LPS variation, even within a single human host, represents an important aspect of pathogen biology and a complication for melioidosis host response.
We compared 27 B. pseudomallei, 10 B. mallei, 3 B. thailandensis, and 2 B. oklahomensis genomes (Table 1) to identify the LPS O-antigen biosynthesis genes. Assuming synteny and common genomic locations, along with known or predicted function, B. pseudomallei O-antigen biosynthesis genes were assigned to two major groups. Group A (LPS genotype A) was identical or very similar to the O-antigen biosynthesis operon observed in B. pseudomallei 1026b [4], whereas group B (LPS genotype B) was found in an atypical LPS strain 576 and also in the species type strain, NCTC13177. LPS genotype A was found in most B. pseudomallei and all B. mallei and B. thailandensis genomes examined. Surprisingly, the more distantly related B. oklahomensis strain EO147 also had LPS genotype A, which was different from the predicted O-antigen biosynthesis gene cluster in other B. oklahomensis strains (C6786, C7532, and C7533; data not shown). This may represent a lateral gene transfer event into EO147 and is deserving of additional study. Furthermore, regions within the two clusters had different levels of sequence conservation. Genes located at the ends of these two clusters (e.g., wbiGHI, and rmlBAC; Figure 1) had higher sequence similarity than most of the genes in the core of the clusters. Indeed, many of the cluster cores contain distinct gene composition. The conserved genes include those important for oligosaccharide synthesis and O-antigen biosynthesis [4].
LPS genotype frequencies were analyzed across a large strain collection using PCR-based assays. Multiplex-SYBR-Green PCR assays were designed to target a specific gene unique for each genotype. Gene wbiE (BPSL2676) of B. pseudomallei strain K96243 and gene BUC_3396 of strain 576 were used to represent the presence of LPS genotypes A and B, respectively (Figures 1&2). A total of 999 B. pseudomallei strains from different geographic locations and epidemiological origins (e.g., clinical, animal, and environmental strains) were tested for their LPS genotypes. We noted that 23 B. pseudomallei strains were collected from one melioidosis patient. We found that LPS genotype A was the most common genotype in both Australian and Southeast Asian strain populations (Figure 2). LPS genotype B was relatively rare in Southeast Asian strains (∼2.3%), but was found in 13.8% of Australian strains. Five strains from Australia and two strains from Papua New Guinea were non-typeable using these two PCR gene markers. Three of these strains, MSHR840, MSHR1950, and MSHR139 were further analyzed for O-antigen biosynthesis gene identification using whole genome sequencing. The O-antigen biosynthesis gene clusters from these strains were identified and annotated (GenBank accession nos. GU574442, HM852062, HM852063). Comparative genomics demonstrated that many genes in this new cluster were similar to those of the LPS genotype B genes of B. pseudomallei 576 and were distinct from the K96243 LPS genotype A genes. Hence, these newly identified O-antigen biosynthesis gene clusters represent a variant of the LPS genotype B and, consequentially, were designated as LPS genotype B2 (Table 1). Figure 1 shows the genomic comparison of these three different O-antigen biosynthesis gene clusters: A, B, and B2 (from B. pseudomallei strains K96243, 576, and MSHR840, respectively). We note that %G+C content of the core of these 3 different clusters is relatively low (∼59–60%) compared to the conserved parts of the O-antigen biosynthesis operon (∼68%). This supports the hypothesis that these genomic differences are due to genetic recombination e.g., horizontal gene transfer, which is common in B. pseudomallei [15], [16]. Comparative genomics of these three different clusters using homologous-based alignment are summarized in Table S1. Again, we note that genes wbiGHI, and rmlBAC are conserved among these three different clusters. Furthermore, gene BURP840_LPSb16 from strain MSHR840 was selected for use as a PCR marker to represent the LPS genotype B2. PCR genotype analysis (Figure 2) revealed that all seven of the previously non-typeable strains were positive for the LPS genotype B2. The LPS B2 genotype was found only in strains from Australia and Papua New Guinea. It is important to note that there is no known clonal relationship among these seven strains. The LPS B2 genotype genes were also found in a B. thailandensis-like spp. strain MSMB121, which was isolated in Australia (unpublished data). Complete LPS genotypic data are reported in Table S2.
LPS genotyping results were further examined by direct comparison to LPS electrophoretic phenotypes [10]. Due to the difficulty of international Select Agent transfer and BSL3 handling, we phenotyped only ∼ 24% of the isolates that were genotyped. We note that this is a limitation of our study. That said, all examined LPS A or B phenotypes were perfectly matched with their LPS A or B genotypes. In addition, 22 strains producing the rough LPS phenotype were all identified as LPS genotype A (Table S2). The genetic basis of the rough phenotype and its derivation from the A phenotype is known for only 16 of these strains (see below). SDS-PAGE revealed that LPS genotype B2 strains produced a distinct ladder pattern, though they were all detectable with type B sera using immunoblot hybridization. The B2 phenotype had a wider range of molecular weights (40–120 kDa) than the LPS types A and B. In total, three LPS banding patterns plus the rough LPS type (no ladder) can be detected (Figure 3).
A frame-shift mutation observed in the O-antigen biosynthesis wbiI gene of B. pseudomallei strain MSHR1655 was correlated with its rough phenotype. This is one of nearly 100 strains that were isolated over 8 years from a patient with severe bronchiectasis associated with melioidosis. The mutation was an extra guanine inserted after nucleotide position 815 of the wbiI gene (Figure 4). The wbiI gene encodes an oligosaccharide epimerase/dehydratase and is conserved in all O-antigen biosynthesis gene clusters of B. pseudomallei. A mutation in this gene probably impacts on the synthesis of the O-antigen in this bacterial strain. There were 23 serial B. pseudomallei isolates observed from the chronically infected patient and the wbiI gene sequences were determined in all of them to detect frame shift mutations. The frame-shift mutation occurred in 16 isolates, all of which were collected on or after day 550 of the infection. The wild type sequence was present in the other seven isolates from earlier in the infection (Figure 4). Moreover, phenotypic characterization revealed that LPS samples extracted from the 16 wbiI mutated strains did not have the O-antigen ladder pattern (i.e the rough phenotype) based upon SDS-PAGE and silver straining (Figure 5A). Thus, it seems likely the frame-shift mutation in the wbiI gene blocks synthesis of the O-antigen. A recent study has reported that oacA gene, known to be involved in the acetylation at the O-4 position of the L-6dTalp residues of B. thailandensis O-antigen [14], is mutated in B. pseudomallei MSHR1655. Since MSHR1655 was isolated from the same patient above, we then sequenced the oacA gene in all of these clonal strains. We found that the oacA mutation occurred in the same 16 strains that had the wbiI mutation (Figure 4C). Additional study of the oacA gene in other whole genome sequenced strains determined that B. pseudomallei 112 and B. thailandensis TXDOH also had point mutation in their oacA genes (Table 1; Figure S1). To determine if the oacA gene plays only a single role in the side group modification of the L-6dTalp residues, or a dual role in combination with the synthesis of the O-antigen, both strains were tested for O-antigen production and immunogenic specificity. We found that B. pseudomallei 112 and B. thailandensis TXDOH expressed O-antigen type A ladder pattern and their O-antigen bands were strongly positive with the B. mallei LPS-specific mAb 3D11 (Figure 6) that recognized the lack of 4-O acetylation of the L-6dTalp residues [14]. This suggests the oacA gene in B. pseudomallei and B. thailandensis has a role in the acetylation at the O-4 position of the O-antigen L-6dTalp residues but is not involved in the synthesis of the O-antigen. Thus, we determined that the rough LPS phenotype observed in the 16 clonal chronic lung strains was due to the mutation of their wbiI gene, but not from the effect of the oacA mutation. In this study, we also identified six other independent rough LPS strains, but mutations did not occur in their wbiI or oacA genes. Searching for mutations in other genes of these strains warrants a follow up study to understand alternate mechanisms that generate the rough phenotype.
Because LPS is essential for outer membrane integrity and serum resistance, four B. pseudomallei strains from this chronic lung patient were further tested in serum bactericidal assays. Two of the wbiI mutant strains that expressed the rough LPS phenotype (MSHR1655 and MSHR3042) were unable to grow in the presence of 30% normal human serum (NHS). In contrast, two early infection isolates from the same patient expressing the typical LPS A phenotype (MSHR1043 and MSHR1048) were able to resist the inhibitory human serum effect and grow (Figure 5B). Furthermore, we also confirmed that the LPS genotype B2 strains were killed in growth media containing 30% NHS, whereas the LPS genotype B strains were resistant (Figure S2). We believe that this finding of serum susceptibility in LPS genotype B2 is important and deserves further investigation.
Despite the fact that genes responsible for the O-antigen biosynthesis in B. pseudomallei 1026b were identified many years ago [4], diversity of these genes across multiple B. pseudomallei strains has not been well studied until now. Advances in genome sequencing and comparative genomics have provided insights into the complexity and diversity of B. pseudomallei genomes. B. pseudomallei genomic studies can now strive for correlations between genomic diversity and differential phenotypes; perhaps the clinical outcomes of individual strains of B. pseudomallei may be predicted using basic genomic analysis. In our current study, we were able to establish a correlation between differential LPS phenotypes and diversity of O-antigen biosynthesis genes or known as LPS genotypes. Three different major LPS genotypes have been identified so far. LPS genotype A was designated to the strains that contained the O-antigen biosynthesis genes that were identical or very similar to those found in a reference strain 1026b [4], whereas the LPS genotype B category is represented by the atypical LPS strain 576. Finally, LPS B2 genotype was identified as a variant of the LPS genotype B because many of its O-antigen biosynthesis genes were similar to those of LPS genotype B, and both groups were serotype B positive. LPS genotype A was the most common genotype in both geographic locations: Southeast Asia and Australia where it accounted for 97.7% and 85.3% of the populations, respectively. Interestingly, the frequency of LPS genotype B was relatively high (approx. 13.8%) in Australian strains, whereas they accounted for only 2.3% of the strains from Southeast Asia. LPS genotype B2 was found in only 7 strains, 5 of which were from Australia, and the other 2 strains were from Papua New Guinea. In addition, LPS genotype B2 was also found in a member of B. thailandensis-like species which was recently discovered in Australia [22]. This would suggest that the LPS genotype B2 genes in B. pseudomallei may be acquired by horizontal gene transfer from a common soil bacterial species in Australia, or vice versa. Comparative genomics and phenotypic characterization of this LPS genotype B2 in B. pseudomallei and its near-relative species warrants further investigation.
Because the LPS genotypes B and B2 were frequently found in Australia but not in Southeast Asia, it is possible that this finding may be due to different therapies used for clinical cases in these 2 endemic locations. We have investigated this and found that the majority of these isolates were obtained before any exposure to antibiotics or treatment therapy. In addition, some of the LPS genotype B strains were collected from soil in Australia, and 2 strains of the LPS genotype B2 were found in animal cases. This confirms that the occurrence of LPS types B and B2 in Australia is not associated with the exposure to antibiotics or treatment therapy. Although, we phenotyped only 24% of the isolates that were genotyped, most tested strains were perfectly matched between their genotypes and phenotypes, except those 16 rough LPS genotype A strains from a single chronic case that had mutations in their wbiI genes (Figure 4). In this current study, we were unable to identify the genetic basis or mutations in 6 independent LPS genotype A strains that did not produce the O-antigen (Table S2).
Because the typical LPS was also found in B. thailandensis, the use of anti-LPS antibody based latex agglutination for the identification of B. pseudomallei in environmental specimens was not successful in an early study [23]. B. thailandensis LPS has also been shown to cross-react with rabbit and mouse sera obtained from inoculation with B. pseudomallei or B. mallei suggesting that LPS molecules from B. thailandensis, a non-pathogenic bacterium, may be useful in ongoing efforts to develop novel vaccines and/or diagnostic reagents [24]. This has brought to our attention whether low-grade B. thailandensis infections might naturally provide protection against melioidosis. Although the O-antigen biosynthesis genes in B. pseudomallei and B. thailandensis are similar, a recent study by a Singaporean group has revealed that lipid A components of the LPS from both B. pseudomallei and B. thailandensis must be different; the murine and human macrophages produced lower levels of tumor necrosis factor alpha, interleukin-6 (IL-6), and IL-10 in response to B. pseudomallei LPS than in response to B. thailandensis LPS in vitro [25]. In our current study, the typical LPS was also found in B. oklahomensis strain EO147, formerly known as an American B. pseudomallei strain [26], suggesting that the typical LPS is widely spread in multiple Burkholderia species. This group includes highly pathogenic species such as B. pseudomallei and B. mallei, but also non-pathogenic species: B. thailandensis, B. thailandensis-like species, and B. oklahomensis. The evolution of LPS diversity across these closely related species is likely a function of differential selection and horizontal transfer of genetic elements. This diversity could play a role in frequency and distribution of disease in humans. However, without understanding molecular structures of these O-antigen types, it is difficult to access the phenotypic effects of this genetic diversity. Structural analysis of the O-antigen types B and B2 deserves further investigations. In addition, we have found that the LPS genotype B2 strains were sensitive to 30% NHS, whereas the LPS type B strains were resistant (Figure S2). This finding demonstrates a level of phenotypic differences between these two serologically related groups. We believe that the consequences for case treatment associated with these differential serum susceptibilities also warrant further investigations.
A previous study has shown that the two less common LPS phenotypes (smooth type B and rough type) were more prevalent in clinical than environmental isolates and more prevalent in Australian isolates than Thai isolates [10]. In our current study, LPS genotype B was found in both clinical and environmental strains from Australia, whereas the rough LPS was still found only in clinical strains. Based on our description of the molecular basis for LPS phenotypes, it is unlikely that B. pseudomallei will readily switch its LPS phenotype from A to B, or vice versa, as has been suggested previously [10]. The gene compositions of LPS genotypes A and B are very different and a simple switching mechanism is difficult to envision. In addition, we have found that at least some rough LPS strains have mutations in their O-antigen biosynthesis genes. These include 16 clonally related isolates from a single chronic lung infected patient (Table S2). All of these strains were identified as LPS genotype A with mutations in their O-antigen biosynthesis genes. Using Tn5-OT182 mutagenesis, DeShazer and colleagues identified at least seven genes in the O-antigen biosynthesis operon of B. pseudomallei 1026b that were responsible for O-antigen biosynthesis and serum resistance; these included rmlB, rmlD, wbiA, wbiC, wbiE, wbiG, and wbiI [4]. In our current study, we found point mutations in wbiI and oacA genes of B. pseudomallei isolates that were collected from a chronic lung patient (Figure 4). We hypothesize that the frame-shift mutation in the wbiI genes blocks O-antigen biosynthesis in all mutant strains, but not from the effect of the oacA mutation. This is because we observed the oacA mutations in B. pseudomallei 112 and B. thailandensis TXDOH that had normal O-antigen biosynthesis gene cluster (Table 1 and Figure S1). Our study has demonstrated that these two oacA mutant strains expressed O-antigens identical to those found in B. mallei due to lack of the 4-O acetylation of the L-6dTalp residues of the O-antigen. The lack of the 4-O acetylation of the L-6dTalp residues has recently been described in the oacA knock-out mutant B. thailandensis ZT0715 and a wild-type B. mallei ATCC23344 [14].
We have demonstrated that these wbiI mutant strains produced rough LPS and were sensitive to normal human serum suggesting that the wbiI gene encoding for epimerase, or dehydratase, was essential for the biosynthesis of B. pseudomallei O-antigen. Although loss of the O-antigen might compromise serum survival it might also be adaptive in particular niches. B. pseudomallei survival or persistence in the host might be enhanced without the surface presentation of the O-antigenic moiety of the LPS, as it would not be recognized by host immune systems and would, therefore, avoid being killed by antibodies. The O-antigenic polysaccharide of B. pseudomallei modulates the host cell response, which in turn controls the intracellular fate of B. pseudomallei inside macrophage. This was concluded from the observation that the O-antigen mutant B. pseudomallei strain SRM117 was more susceptible to macrophage killing during the early phase of infection than the parental wild-type strain 1026b [27]. This was also confirmed by the same group when they demonstrated the importance of intracellular killing by the human polymorphonuclear cells (PMNs), macrophages (Mφs), and susceptibility to killing by 30% normal human serum [28].
LPS and CPS (capsular polysaccharide) have been used as subunits in immunizing BALB/c mice against B. pseudomallei infection [2]. Mice vaccinated with LPS developed predominantly IgM and IgG3 responses, whereas the mice vaccinated with the CPS developed a predominantly IgG2b response. Furthermore, immunization with the LPS provided an optimal protective response, and the immunized mice challenged by the aerosol route showed a small increase in the mean time to death compared with the unvaccinated controls [2]. Previously, it was shown that B. pseudomallei LPS from strain 1026b signaled through Toll-like receptor (TLR) 2 and not through TLR4 [29]. This was observed in the TLR2 knock-out mutant mice that displayed a markedly improved host defense, but it was not observed in TLR4 knock-out mice [29]. In contrast, a study in HEK293 cells demonstrated that heat-killed B. pseudomallei strains K96243 or BP-1 activated TLR2 and TLR4, and in the presence of MD-2, LPS and lipid A from BP-1 are TLR4 ligands [30]. We note that B. pseudomallei 1026b and K96243 expressed the typical O-antigen type A, but the O-antigen type of BP-1 was not reported in that study. Although there was no report of association between the LPS types and disease severity (e.g., fatal versus non-fatal, and septicemia versus localized), clinical manifestations (neurologic versus non-neurologic), or underlying risk factors (diabetic versus non-diabetic) observed in a previous study [10], full phenotypic characterization including virulence in animal models, innate immune response, etc of these different LPS types warrants further investigations given the LPS diversity that we have described.
|
10.1371/journal.pcbi.1005779 | Correlated receptor transport processes buffer single-cell heterogeneity | Cells typically vary in their response to extracellular ligands. Receptor transport processes modulate ligand-receptor induced signal transduction and impact the variability in cellular responses. Here, we quantitatively characterized cellular variability in erythropoietin receptor (EpoR) trafficking at the single-cell level based on live-cell imaging and mathematical modeling. Using ensembles of single-cell mathematical models reduced parameter uncertainties and showed that rapid EpoR turnover, transport of internalized EpoR back to the plasma membrane, and degradation of Epo-EpoR complexes were essential for receptor trafficking. EpoR trafficking dynamics in adherent H838 lung cancer cells closely resembled the dynamics previously characterized by mathematical modeling in suspension cells, indicating that dynamic properties of the EpoR system are widely conserved. Receptor transport processes differed by one order of magnitude between individual cells. However, the concentration of activated Epo-EpoR complexes was less variable due to the correlated kinetics of opposing transport processes acting as a buffering system.
| Cell surface receptors translate extracellular ligand concentrations to intracellular responses. Receptor transport between the plasma membrane and other cellular compartments regulates the number of accessible receptors at the plasma membrane that determines the strength of downstream pathway activation at a given ligand concentration. In cell populations, pathway activation strength and cellular responses vary between cells. Understanding origins of cell-to-cell variability is highly relevant for cancer research, motivated by the problem of fractional killing by chemotherapies and development of resistance in subpopulations of tumor cells. The erythropoietin receptor (EpoR) is a characteristic example of a receptor system that strongly depends on receptor transport processes. It is involved in several cellular processes, such as differentiation or proliferation, regulates the renewal of erythrocytes, and is expressed in several tumors. To investigate the involvement of receptor transport processes in cell-to-cell variability, we quantitatively characterized trafficking of EpoR in individual cells by combining live-cell imaging with mathematical modeling. Thereby, we found that EpoR dynamics was strongly dependent on rapid receptor transport and turnover. Interestingly, although transport processes largely differed between individual cells, receptor concentrations in cellular compartments were robust to variability in trafficking processes due to the correlated kinetics of opposing transport processes.
| In cells external signals from ligands are transmitted by receptors to intracellular signaling cascades. Receptor signaling is regulated by receptor transport processes between the plasma membrane and other cellular compartments that are subsumed under the term receptor trafficking [1]. In absence of ligand, receptors are transported to the plasma membrane and are taken up again by the cell. After ligand binding, activated receptors at the plasma membrane can be internalized. To shut down signal transduction, endosomal acidification induces ligand dissociation from the receptor. Subsequently, the receptor is either degraded or transported back to the plasma membrane. These transport processes therefore strongly influence the ability of cells to integrate signals from external ligands and thereby the translation into cellular responses.
In a variety of receptor systems, receptor trafficking was quantitatively studied by a combination of experiments and ODE models based on population average data [2–4]. For example, endocytosis, degradation and receptor recycling were quantitatively studied in the epidermal growth factor receptor (EGFR) [5–10], the erythropoietin (Epo) receptor [11,12], the insulin receptor [13,14], chemotactic peptide receptors on neutrophils [15–17], the transferrin receptor (TfR) [18,19], the low density lipoprotein receptor (LDLR) [20,21], interferon-α and tumor necrosis factor receptors [22,23]. These studies established a canonical receptor trafficking model that accounts for exchange of free receptors between the plasma membrane compartment and an intracellular receptor pool, internalization of ligand-bound receptors, degradation, and receptor recycling [2–4,24]. Quantifying receptor trafficking processes helped to characterize physiologically relevant differences between receptor systems. In particular, kinetic parameters for ligand binding, the internalization of free or ligand-bound receptors and for synthesis and degradation of receptors showed large differences between receptor systems, and could be used to categorize receptors according to functional roles in cells [2,4,24]. Growth factor receptors such as the EGFR are characterized by a high membrane abundance and a strongly accelerated internalization of ligand-bound compared to free receptors at the plasma membrane, a phenomenon denoted as ligand-induced receptor downregulation [5,15,25]. Due to an accelerated internalization upon ligand binding, short reaction times of receptor signaling to changes in ligand concentrations are facilitated [24]. From a systems perspective, this increases the accuracy of signal transduction within involved signaling pathways [4,24]. On the contrary, transport receptors as the TfR or the LDLR typically do not exhibit an accelerated internalization upon ligand binding but show a high rate of receptor internalization compared to the rate of ligand unbinding [24,26–28]. Cytokine receptors, as the EpoR or the interleukin 3 receptor, are characterized by a low membrane abundance and an efficient clearance of ligand from the medium and rapid recovery of receptor levels at the plasma membrane [4,12].
The last four decades contributed to a broad understanding of dynamic properties of receptor systems but most studies described receptor trafficking based on measurements of cell population averages. Because trafficking processes depend on a multitude of biochemical processes including for example vesicle formation and cytoskeleton-dependent transport [1,29], heterogeneous expression of involved proteins can give rise to cell-to-cell variability [30]. In this context, an open question is whether cellular heterogeneity in different receptor trafficking processes can dissolve borders between categories of receptor systems, potentially leading to subpopulations of cells showing features as endocytic downregulation, fast replenishment or an efficient receptor recycling. As a result, cell-to-cell variability in receptor trafficking might cause a diverging behavior of cells in response to an external stimulus. For this reason, it is an important question whether receptor systems exhibit robustness to cellular variability in trafficking processes. A prime example for the importance of receptor transport processes in regulating systems properties is the receptor for the hormone erythropoietin (Epo) [11]. Ligand-induced signal transduction through this cytokine receptor, the EpoR, comprises primarily activation of JAK2/STAT5, PI3K/AKT and MAPK pathways, and is absolutely essential for differentiation, proliferation and cell survival of erythroid progenitor cells to ensure renewal of mature erythrocytes [31,32]. Transport processes regulating EpoR induced signal transduction are (i) receptor internalization and inactivation followed by subsequent degradation, and (ii) receptor recycling encompassing ligand-induced receptor endocytosis and subsequent transport back to the plasma membrane [12,33]. It was reported that the activation of kinases and phosphatases [34], ubiquitination of the receptor [35], and cargo protein and cytoskeleton dependent processes such as assembly of actin oligomers [36] modulate transport of the EpoR.
A characteristic property of the EpoR system is that only a small fraction of the total receptor amount is present at the cell surface [37,38]. By dynamic pathway modeling in combination with binding studies utilizing radioactively labeled Epo we recently showed that extremely rapid receptor turn-over ensures responsiveness of the system for a very broad ligand-concentration range as it is for example observed during continuous erythrocyte renewal and accelerated production in response to severe blood loss [11,12,39]. Further, data-based mathematical models revealed that (1) Epo-induced activation of the JAK2-STAT5 signaling cascade occurs in cycles continuously monitoring the activation status of the receptor [11,12,39] and (2) the two induced negative regulators bind to the receptor and divide the labor to control signaling for a wide range of Epo concentrations [31,32]. The so far established mathematical models were calibrated based on cell population data obtained for suspension cells. The kinetics at the level of single cells is smoothed and underlying biochemical signaling networks might be misinterpreted due to averaging population heterogeneities [40–42]. Furthermore, since the EpoR is also expressed on some tumor cells such as non-small cell lung carcinoma cell lines [43], it of much interest to investigate to which degree principles learned in suspension cells can be transferred to adherent cancer cells.
Here, we developed an approach based on live cell imaging, image segmentation of subcellular compartments, and cell ensemble models to investigate the extent of variability in receptor trafficking and interrelations between the dynamics of transport processes. Single-cell measurements of EpoR concentrations in different cellular compartments were used to estimate kinetic parameters of receptor trafficking processes for individual cells. By model discrimination we determined which receptor transport processes essentially contributed to receptor trafficking of EpoR. Calibrating cell ensemble models with a combination of single-cell datasets improved the identifiability of single-cell kinetic parameters, which was a prerequisite for analyzing correlations between kinetic parameters of receptor transport processes. Despite the large variability in the EpoR trafficking reactions we observed that the correlation between the kinetics of different transport processes had a buffering effect on the concentration of Epo-EpoR complexes at the plasma membrane and in the endosomal compartment. This correlation of the kinetics of different processes involved in the same cellular signaling system might represent a general motif of biological systems to confine cell-to-cell variability.
The EpoR is transported to the plasma membrane, can bind Epo, and is subjected to endocytosis, degradation and transport back to the plasma membrane [12,33]. To quantitatively study these processes at the single-cell level, we developed an approach employing an EpoR-GFP fusion protein (EpoR-GFP) and Epo labeled with the organic dye Cy5.5 (Epo-Cy5.5). The EpoR-GFP fusion protein was stably expressed in the NSCLC cell line H838 and a fluorescent membrane marker, mCherry fused to a myristoylation-palmitoylation (MyrPalm) domain (MyrPalm-mCherry) accumulating at the plasma membrane was co-expressed (Fig 1A) [44]. After recording the first image stack, cells were exposed to Epo-Cy5.5 at a concentration of 4.2nM corresponding to a biological activity of 10U/ml Epo [45]. Subsequently, Epo internalization was studied for at least five hours by recording three-dimensional stacks of confocal microscope images. Analyzing Epo-Cy5.5 in combination with EpoR-GFP and the membrane marker MyrPalm-mCherry enabled simultaneous recording of complementary information on Epo-uptake, EpoR-internalization and EpoR-degradation essential for studying protein turnover by kinetic modeling. While the GFP signal indicated the amount of EpoR-GFP and was affected by EpoR-GFP degradation, the Cy5.5 signal represented the sum of intact and degraded proteins since the dye molecule Cy5.5 is not targeted by protein degradation mechanisms.
Intensities for membrane and cytosolic compartments were extracted from microscopic data to obtain time-resolved measurements, which were proportional to local concentrations of EpoR-GFP and Epo-Cy5.5, and were used for model fitting. For this purpose, we developed a segmentation software to semi-automatically define three-dimensional regions of interest (ROIs) for the plasma membrane using the MyrPalm-mCherry signal, and for EpoR-GFP/Epo-Cy5.5 containing vesicles the EpoR-GFP and Epo-Cy5.5 signals (Fig 1A to 1C, S1 Fig and S1 Movie; for details, see S1 Text). In the ROI for the plasma membrane, Epo-Cy5.5 intensities were associated with the amount of Epo-EpoR complexes, while EpoR-GFP intensities were associated with the total amount of EpoR (Epo-ligated plus free receptors). Further, based on an intensity threshold for Cy5.5, we distinguished between EpoR in voxels containing only EpoR-GFP or Epo-Cy5.5/EpoR-GFP (S1 Text). We extracted the Cy5.5 fluorescence signal in the cytosolic compartment to obtain a quantitative measure of the amount of internalized Epo-Cy5.5. Epo that was bound to internalized EpoR can be either secreted from the cell or degraded. Because Cy5.5 that was coupled to Epo is not proteolytically degraded, the intracellular Cy5.5 signal was assumed to reflect the amount of intact and degraded Epo. To obtain quantities for model fitting that were proportional to EpoR and Epo concentrations, intensities were normalized by cellular volumes, which were defined by the volumes enclosed by outer borders of membrane ROIs.
The described procedure was applied to analyze for example 16 single Epo-treated H838 cells. As shown in Fig 1C for a representative single cell, we observed the strongest signal changes within the first hour after addition of Epo-Cy5.5, indicating fast binding and internalization. The membrane EpoR-GFP fraction and the signal from EpoR-GFP vesicles showed in the exemplary cell only a slight increase, implying that Epo did not have a large influence on the total amount of the EpoR. On the contrary, the intensity from Epo-Cy5.5-containing vesicles continuously increased. While the Cy5.5 intensity at the plasma membrane reached a steady state after about ten minutes, intracellular Cy5.5 intensity showed a prolonged increase suggesting a slow decay of internalized Epo-Cy5.5.
We asked whether initial conditions such as EpoR concentrations in cellular compartments were predictive for EpoR trafficking in the presence of Epo, and evaluated associations between characteristic measures of single-cell trajectories before and after adding Epo-Cy5.5. In particular, we examined which experimental quantities were predictive for the amount of membrane bound Epo-Cy5.5, which can be assumed to reflect the amount of active EpoR [11,12]. For all Epo-treated cells, characteristic parameters were extracted from segmented imaging data, resulting in total EpoR concentrations or EpoR numbers in arbitrary units. Absolute numbers of EpoR-GFP or Epo-Cy5.5 in cellular compartments were estimated by summing up fluorescence intensities in segmented compartment ROIs, while cellular concentrations were estimated by dividing fluorescence intensity sums in cellular compartment ROIs by cell volumes. For scale-free comparisons, single-cell measures were divided by the means of all cells to obtain fold changes relative to single-cell averages (Fig 1E–1G, S2 Fig). Among all cells, the membrane EpoR-GFP fraction contained on average 7.6% (SD: 2.1%) of the total cellular amount of EpoR-GFP. Interestingly, EpoR concentrations in the membrane ROI ([EpoR-GFPmem]) were significantly correlated with EpoR concentrations in intracellular vesicles ([EpoR-GFPves]; Fig 1E; p = 0.0066 for Pearson correlation coefficients). This implies that the kinetics of EpoR transport from the cytosol to the plasma membrane was correlated with kinetics of EpoR transport from the plasma membrane back to the cytosol. The observation of correlated trafficking parameters will be further addressed below. Furthermore, while there was no significant correlation between the total cellular concentrations of EpoR-GFP ([EpoR-GFPtot]) and the concentration of Epo-Cy5.5 in the plasma membrane ROI ([Epo-Cy5.5mem]) at the end of the experiment after 5 hours (Fig 1F; p = 0.25), absolute amounts of cellular EpoR-GFP (NEpoR-GFP,tot) were significantly correlated with the amounts of membrane Epo-Cy5.5 (NEpo-Cy5.5,mem) at 5h (Fig 1G; p = 0.0083). While total amounts of EpoR-GFP and internalized Epo-Cy5.5 were significantly correlated, there were no significant correlations between EpoR-GFP and Epo-Cy5.5 concentrations in different cell compartments (S2 Fig), which indicates that the EpoR transport kinetics strongly varied between cells.
Taken together, we established an experimental setup to quantitatively study the dynamics of the EpoR and the internalization of Epo by live-cell microscopy.
To mechanistically study cell-to-cell variability in EpoR transport processes, we developed different mathematical models (EpoR model) based on ordinary differential equations (ODE) and estimated the model parameters by model fitting to single-cell measurements. The EpoR model variants, consisting of a basic model and variable extensions, described the two observed species, free EpoR and EpoR bound to Epo, in different cellular compartments or at the plasma membrane.
The basic EpoR model describes reversible binding of Epo to the EpoR at the plasma membrane (EpoRm) and formation of active EpoR (EpoRm*) (black arrows in Fig 2A). EpoR permanently cycle between the plasma membrane (EpoRm) and the intracellular compartment (EpoRi). The intracellular pool of the EpoR is subject to degradation and refilled by synthesis. Active EpoR at the membrane EpoRm* are internalized to the endocytic recycling compartment (EpoRRE*). In the model reaction describing EpoR binding to free Epo, Epo is not consumed because the amount of Epo in the medium largely exceeds the total amount of EpoR, as described in the methods section, and can therefore be assumed to remain constant. The basic model was extended by variable parts A to D, which described different possible ways for EpoR transport back to the plasma membrane or degradation. By appending variable combinations of parts A to D to the basic model, 16 possible model variants were formulated to systematically test the contribution of different processes to EpoR trafficking in our cellular system. Since receptor recycling and degradation of ligand-bound receptors were described for several receptor systems as the EGFR, IL3R or TfR [2,12,46–49], we explored their role in EpoR trafficking in our cellular system, and whether their contribution was essential or could be neglected, which was not examined in previous modeling studies on EpoR trafficking. In model variants, internalized Epo is either released back into the extracellular space (parts A and C) or degraded (Epodeg,i) and accumulates inside the cell (parts B and D, Fig 2A) [11,12]. After internalization of Epo-EpoR complexes, receptors recycle back to the plasma membrane (A and B) or are degraded (C and D, Fig 2A) [12,23]. From the endocytic recycling compartment, receptors are recycled via path A directly to the membrane EpoRm or via B to the intracellular pool (EpoRi). All model variants were fitted to data from our single-cell experiments.
To enrich our experimental dataset by kinetic data on EpoR synthesis and degradation, we performed two auxiliary experiments. First, Epo-GFP expressing H838 cells were bleached by applying a short laser pulse. Thereafter replenishment due to EpoR-GFP synthesis was followed in ten treated H838 cells by recording the increase of the GFP signal. Furthermore, EpoR-GFP degradation was studied in seven single H838 cells treated with cycloheximide (CHX) at a concentration of 5μg/ml to inhibit protein translation and by recording the subsequent decrease of the GFP signal. Inhibition of translation by CHX was similarly used in previous systems biological studies to quantitatively study protein degradation [50–52].
The rationale for doing these additional experiments on EpoR synthesis and degradation was that the trafficking dynamics in unperturbed experiments are likely to be a complex superposition of EpoR endocytosis, recycling, synthesis and degradation effects. Therefore, we assumed that a combination with EpoR synthesis and degradation experiments were required to make kinetic parameters for EpoR turnover identifiable. In general, combining experiments on receptor trafficking with experiments on receptor turnover is reasonable because time scales of these processes might be different.
For all model variants, cell ensemble models were constructed [40]. In the cell ensemble models, each single cell of a heterogeneous cell population was described by the same set of ODEs, and cell-to-cell variability was introduced by allowing receptor trafficking parameters and initial EpoR concentrations to be different between cells (as further described below). Cell ensemble models comprised single-cell models for Epo internalizing cells, and simplified models for photobleached and CHX treated cells, in which reactions for Epo uptake were excluded. One single-cell model describing an Epo treated cell contained 6 ODEs and between 7 and 11 parameters (S1–S3 Tables; for details, see S2 Text). Models of photobleached cells contained a reduced set of reactions describing only synthesis, degradation, transport of the EpoR between the plasma membrane and the intracellular pool, and an additional reaction describing removal of detectable EpoR species by photobleaching. Trajectories of CHX treated cells, in which synthesis was inhibited, were described by ODE models describing EpoR degradation and transport between the plasma membrane and the intracellular pool of EpoR (S4 and S5 Tables). Models of photobleached cells consisted of 3 ODEs with 5 kinetic parameters while models of CHX treated cells contained 3 ODEs with 3 kinetic parameters.
The parameters for Epo binding and unbinding, kon,Epo and koff,Epo were defined as being equal for each single-cell model, whereas all other kinetic parameters were allowed to vary between cells. This assumption was made, because kon and koff are biophysical constants, whereas receptor trafficking parameters describe lumped reactions that are controlled by concentrations of various intracellular regulatory proteins. Hence, in line with previous studies, we assumed in our model that cell-to-cell variability arises from heterogeneous expression of cellular proteins [40,53].
An ensemble model describing the complete available dataset of 16 Epo treated, 10 photobleached, and 7 CHX treated cells comprised between 156 and 220 kinetic parameters. Experimental single-cell datasets for GFP and mCherry fluorescence were linked via scaling factors to model variables in absolute concentration units. Taking together kinetic parameters, scaling factors, and initial concentrations [EpoRm](t0) and [EpoRi](t0) resulted in a total number of 230 to 294 parameters for different model variants, which were estimated by model fits of a total of 3996 data points. To estimate the scaling factor between normalized GFP fluorescence intensities in cellular compartment ROIs and absolute receptor amounts, average total cellular EpoR-GFP levels were determined by quantitative immunoblotting (S3 Fig). Immunoblotting and image stack segmentations showed that each cell contained on average 142.000 receptors and had a mean volume of about 5.47pl, which resulted in an average cellular concentration of [EpoR]tot = 43.1nM.
Fitting cell ensemble models to sets of single cells treated under different conditions, i. e., by adding Epo-Cy5.5, CHX or bleaching, can in principle lead to systematic differences between sets of estimated kinetic parameters. However, this is unlikely because the same cell line was used in all conditions. Therefore, kinetic parameters of cells treated under different conditions should follow the same probability distribution [40]. Because kinetic parameters of single cells implicitly depend on concentrations of regulatory proteins that are typically log-normally distributed in cell populations [54,55], we assume log-normal distributions of single-cell parameters for EpoR trafficking processes, EpoR synthesis and degradation. To minimize differences between parameter distributions for the three experimental data sets generated by adding Epo-Cy5.5, CHX or bleaching, we added constraint terms to the likelihood function used for parameter estimations, which penalized for differences in parameter means and variances between experimental sets (for details, see S2 Text). Restricting parameter estimations by these constraint terms was advantageous with regard to model discrimination and parameter identifiability, as described below.
We found that the model variant “ACD”, with parts for direct EpoR recycling to the plasma membrane (part A) and EpoR degradation with either exocytosis (part C) or intracellular accumulation of consumed Epo (part D), could significantly better explain the set of experimental data than the other variants (Fig 2B). This was indicated by the smallest values for the corrected Akaike information criterion (AICcorr), which finds the most parsimonious model by weighing the number of parameters with goodness of fit and experimental noise, thereby preventing overfitting.
Next, we compared the model selection results for different sets of experimental data. Thereby, we assessed to which degree cell ensemble models including constraint improved the model discrimination. Already the comparison between cell ensemble models calibrated solely with data from Epo-treated cells showed that the variant “ACD” performed significantly better than the other variants. Including data for bleached and CHX treated cells further increased the AICcorr difference to other variants and allowed more distinct model discrimination. In contrast, fitting model variants to data from only a single cell, instead of fitting cell ensemble models to data from several cells simultaneously, was not sufficient to determine an optimal model variant (Fig 2C), a situation comparable to conventional ODE models calibrated only with population average data, which ignore cell-to-cell variability. The optimal model variant ACD is visualized in Fig 2D. The complete set of single-cell data for Epo internalizing, bleached or CHX treated cells is shown together with the best-fit ACD model trajectories in Fig 3. In addition, scatter plots of experimental data plotted against corresponding model simulations are shown in S4 Fig, and residuals as well as residual distributions are shown in S5 Fig. Overall, it can be concluded that, our set of single cell data could be well explained by the model. The kinetic parameters associated with the reactions (grey text in Fig 2D) are further analyzed below.
We hypothesized that cell ensemble models improved parameter estimations by combining complementary experimental datasets. To test this, we analyzed parameter identifiability for different combinations of datasets in cell ensemble models in comparison to individual single-cell models. Fig 4A visualizes relative confidence interval sizes, confidence intervals divided by parameter values, obtained from profile likelihood estimation (PLE) for parameters of four exemplary cells and different experimental datasets in a color-coded manner, Fig 4B for an exemplary parameter as error bars. Essentially, confidence interval sizes decreased significantly when using cell ensemble models instead of models fitted to data from one cell at a time, and for fitting cell ensemble models to data from all three experimental conditions instead of only Epo internalizing cells. For all parameters estimated in cell ensemble models, upper confidence intervals were defined by PLE. Only for few parameters, lower confidence intervals included zero indicating that those parameters were not identifiable and that involved reactions might be eliminated in these cells. Similarly, standard deviations from the best 0.5% of 1000 fits, ordered according to their squared sum of residuals, for all model parameters showed that combining datasets for Epo-internalizing H838 cells, bleached H838 cells and CHX treated H838 cells significantly improved the accuracy of single-cell parameter estimations (S6–S8 Figs). In absence of constraint terms (S8 Fig), single-cell estimates of EpoR transport parameters were of similar magnitude as in presence of constraint terms (S6 Fig and S7 Fig) which indicates that including constraint terms improved the identifiability of single-cell parameters but did not affect the variabilities of single-cell parameters. The globally defined parameters for Epo binding and unbinding were not identifiable, which were, however, not in the focus of this study. All scaling factors were identifiable with small confidence intervals (S6 Table and S6 Fig).
In summary, we found that the EpoR model variant ACD was optimal, which is consistent with EpoR trafficking reactions described in the model by Becker et al. that was developed based on cell population average data [12]. In comparison to the model by Becker et al., our model additionally accounts for the intracellular pool of free EpoR, synthesis and degradation of the EpoR. We observed that our EpoR model could not be further reduced but that all components were required to explain the experimental data. Using cell ensemble models allowed clear discrimination between model variants and improved parameter identifiability. Improving the identifiability of single-cell parameters was necessary to analyze correlations between kinetic parameters within a population of cells, which will be further described below.
After determining an optimal model variant, we asked how sub-compartment receptor pools remained largely unchanged in the presence of Epo and why intracellular ligand accumulation was slow. We investigated how EpoR trafficking reactions effectively contributed to these experimental observations. To this end, we extracted the concentrations of EpoR species from the model and analyzed fluxes (concentration changes per minute) through each of the reactions for each cell and at different time points.
Model predictions of single-cell concentrations of EpoRm, EpoRi, EpoR*m and EpoR*RE, and reaction fluxes for all EpoR reactions are shown in Fig 5A and 5B. We superposed means and standard deviations for the best 0.5% of 1000 fits for single cells and average fluxes (Fig 5B). After adding Epo, the largest fraction of the EpoR at the plasma membrane is quickly bound to Epo. The transport from the intracellular pool of EpoR (EpoRi) to the plasma membrane compensates for the internalization of Epo-bound EpoR (EpoRm*) resulting in EpoR concentrations, which are, in agreement with characteristics observed in single-cell trajectories (Fig 1), almost at steady state. Fluxes for EpoR recycling (FEpoR*,REtoM) reach similar magnitudes as fluxes of unoccupied EpoR from the intracellular pool to the plasma membrane (FItoM). Reaction fluxes in different cells varied approximately by a factor of ten implying that EpoR transport dynamics and the consumption of Epo strongly diverge between cells, an observation, which is further analyzed below. Average fluxes at the end of the experiment (t = 300’), when fluxes were close to steady states, are illustrated in Fig 5C. Analysis of fluxes showed that a large fraction of internalized EpoR was recycled to the plasma membrane (FEpoR*,REtoM), while a smaller receptor fraction was degraded, mostly with exocytosis of Epo. Notably, about one percent of the total amount of free EpoR cycles per minute between the plasma membrane and the intracellular compartment (FItoM, FMtoI).
To conclude, similar to previous studies [11,12] we observed an important contribution of receptor recycling and the fast transport of the receptor between the plasma membrane and the cytosol, and showed that the reaction fluxes varied approximately up to an order of magnitude between different cells.
Next, we addressed how the observed strong variability in reaction fluxes affects signal transduction. Specifically, we asked how the concentration of Epo-EpoR complexes at the plasma membrane indicative for the fraction of activated receptors [11,12], and the concentration of internalized Epo-EpoR complexes were dependent on EpoR transport processes.
First, we compared our single-cell parameter estimates with the corresponding kinetic parameters from the mathematical model by Becker et al. [12]. Interestingly, although Becker et al. had used a different cellular system, the murine suspension cell line BaF3 stably expressing the EpoR instead of the human adherent NSCLC cell line H838 stably expressing the EpoR-GFP, all parameters from their population average data model were inside ranges of the single-cell parameters in our model (Fig 6A), and were significantly correlated with single-cell parameter means (ρ = 0.92, p = 0.01). As observed in the study by Becker et al., the kinetic parameters for internalization of Epo-bound EpoR (kEpoR*,MtoRE) were in the range of the parameters for internalization of free EpoR at the plasma membrane (kEpoR,MtoI), indicating that ligand binding did not substantially accelerate internalization.
To further study cell-to-cell variability, we calculated the coefficients of variation (CV), which equal standard deviations divided by means, for single-cell parameters and the concentration of Epo-EpoR complexes at the cell membrane after 5 hours of Epo-stimulation, [EpoR*m](5h), and of internalized Epo-EpoR complexes [EpoR*RE](5h), when reactions were close to a steady state.
Of note, we analyzed the variability of kinetic parameters between cells, which should not be confused with analyzing parameter variances in one single-cell model to assess whether single cell parameters can be uniquely estimated. Here, identifiability of single-cell parameters and small parameter confidence intervals were prerequisites for analyzing the variabilities of parameters in a heterogeneous population of cells.
For kinetic parameters, we observed large CVs of above one besides slightly smaller CVs of about 0.7 for the parameters for EpoR synthesis (ksyn) and for degradation of Epo-bound EpoR with exocytosis of consumed Epo (kEpoR*,deg,REtoEx) (Fig 6B). However, for initial concentration estimates of EpoR, and of Epo-EpoR complexes after 5 hours, CVs had substantially smaller values between 0.2 and 0.5. To analyze this divergence in variabilities, we determined concentration control coefficients for [EpoR*m](5h) and [EpoR*RE](5h). Concentration control coefficients r were calculated as normalized derivatives of parameters k as r = k/[EpoR*m](5h)∂[EpoR*m](5h)/∂k or r = k/[EpoR*RE](5h)∂[EpoR*RE](5h)/∂k, and were expected to have values above one in case of strong sensitivity towards changes of a parameter and below one in case of weak sensitivity [56,57]. All control coefficients were smaller than one, indicating robustness of the system towards parameter changes (Fig 6B).
Importantly, the strong divergence between large CVs for kinetic parameters and a small CV for the concentration of Epo-EpoR complexes after 5 hours, [EpoR*m](5h) and [EpoR*RE](5h), could be explained by positive correlations between kinetic parameters (Fig 7 and S9 Fig). In particular, the parameters kEpoR,MtoI, kEpoR,ItoM, kEpoR*,MtoRE, and kEpoR*REtoM, which described EpoR transport reactions, were positively correlated with high significance (Fig 7A and 7B, S9 Fig). The positive correlation of the parameters kEpoR,MtoI and kEpoR,ItoM was in line with the experimental observation that EpoR concentrations at the plasma membrane were correlated with EpoR concentrations in intracellular vesicles (Fig 1E). Further, the kinetics of processes involved in increasing and decreasing Epo-EpoR complexes at the cell membrane [EpoR*m] or internalized Epo-EpoR complexes [EpoR*RE] were positively correlated, and therefore, variabilities canceled out. Intuitively, this positive correlation between opposing processes is biochemically reasonable because different transport processes depend on the same molecular key components, such as motor proteins or constituents of the cytoskeleton [29], which will be discussed further below.
Simulating the case, in which positive correlations between kinetic parameters were removed, could further illustrate to which degree positive correlation between EpoR trafficking processes reduced noise. To this end, we derived a multivariate log-normal parameter distribution from estimates of single-cell parameters. First, we sampled vectors of single-cell parameters from the derived multivariate distribution using the complete covariance matrix, and simulated values for Epo-EpoR complexes, [EpoR*m] and [EpoR*RE], after 5 hours for each parameter vector. Then, we set covariances for parameters describing the transport of EpoR* (kEpoR*,MtoRE, kEpoR*,REtoM, kEpoR*,deg,REtoEx, kEpoR*,deg,REtoI) to zero, and again sampled parameter vectors from the modified multivariate distribution to simulate values for Epo-EpoR complexes [EpoR*m] and [EpoR*RE] after 5 hours. As expected, reducing parameter covariances resulted in a clear increase of the CV for [EpoR*m](5h) and [EpoR*RE](5h), whereas sampling from the complete covariance matrix resulted in a CV similar to the value obtained from parameter estimates after model fitting (Fig 6B). We concluded that positive correlations between single-cell parameters for intracellular EpoR transport processes reduced variability of the concentration of Epo-EpoR complexes at the plasma membrane, which implies that inter-relations between trafficking processes effectively dampened variability in the output of the system.
Next, we explored which cell-to-cell differences were essential to describe the data. We tested, whether in any of the reactions, global parameter values could be used to describe the same reactions in different cells and allow further model simplification. Single-cell parameters in the optimal model variant ACD, which were estimated individually for each cell, were sequentially defined as global parameters that were equal for all cells. Only ksyn was allowed to be variable in every case to account for different EpoR concentrations in individual cells. After fitting restricted model versions to the experimental dataset, differences in AICcorr to the unrestricted model, in which all parameters could vary between cells, were calculated (Fig 8A and 8B). Subsequent fixing of additional parameters causing the smallest increase in AICcorr showed that fixing the parameters for EpoR degradation (kEpoR*,deg,REtoEx, kEpoR,deg, kEpoR*,deg,REtoI) resulted only in subtle AICcorr increases (Fig 8A, left panel; Fig 8B, lower trajectory) suggesting that variability of these parameters was least important. On the contrary, sequential fixing of additional parameters causing the largest increase in AICcorr showed that variability of the parameters for EpoR transport to the plasma membrane (kEpoR,ItoM) and for EpoR internalization (kEpoR,MtoI) was most consequential (Fig 8A, right panel; Fig 8B, upper trajectory). Apart from the distinct impact of parameter variabilities, AICcorr suggested that all variabilities were essential to fully describe the data indicating that the model could not be further simplified by assuming equal kinetic parameters for the same receptor transport processes in different cells.
To conclude, using cell ensemble models instead of separate single-cell models, and including datasets for bleached and CHX treated cells in ensemble models improved the identifiability of single-cell parameters. The dynamics of EpoR transport processes was similar in adherent H838 cells as previously described for BaF3 suspension cells. Interestingly, a positive correlation between parameter describing opposing receptor trafficking processes provides an explanation for the observed moderate cell-to-cell variability of Epo-EpoR complex concentrations at the plasma membrane despite the large variability in the kinetics of EpoR transport processes in individual cells.
An interesting finding of this study was that single-cell parameter estimates indicated large cell-to-cell variability in EpoR transport processes, whereas the concentration of Epo-EpoR complexes at the plasma membrane representing the activated EpoR was much less variable. Model analysis showed that the positive correlations between kinetic parameters describing opposing EpoR transport processes effectively canceled out parameter variabilities and were responsible for the dampening of cellular heterogeneity in Epo-EpoR complexes at the cell membrane and in the intracellular compartment. Receptor trafficking parameters can be assumed to result from molecular properties of receptors and on the process of vesicle trafficking. Therefore, from the perspective of cellular physiology, two explanations can be considered to explain kinetic parameter correlations. First, properties of receptor molecules, their posttranslational modifications and effects due to receptor signaling might take influence on different trafficking processes in the same manner. Second, vesicle trafficking processes that are responsible for receptor transport to the plasma membrane, internalization of ligand-bound or free receptors might be co-regulated. This appears likely because vesicle trafficking reactions share key components involved in vesicle trafficking as microtubules, myosin or actin filaments that define common paths for vesicles. In general, vesicle trafficking typically requires a small number of different motor proteins, while adaptors bound to transport protein complexes, as Rab proteins that differentially regulate transport of different cargos, are more diverse [29,58]. Transport proteins are in some aspects co-regulated [29,59], which might support synchronization of different trafficking processes. It was shown that the velocity of transport mediated by dyneins, myosins and kinesins is regulated by the concentration of ATP [60–63], and that vesicle trafficking is slowed down after loss of ATP [64,65]. Therefore, the metabolic status of the cell might determine the kinetics of different vesicle transport processes and contribute to synchronized dynamics of transport processes. Moreover, overall correlations were observed for all cellular proteins, especially for proteins involved in the same biological pathways [66,67]. For this reason, also the kinetics of more specific trafficking mechanisms might be correlated, which are dependent on classes of regulatory proteins as kinesins or Rab GTPases [68,69].
Previous studies used trafficking parameters observed at the cell population level to categorize different receptors [2,4,24,70]. Accounting for variability in receptor trafficking changes this picture because features of different functional categories of receptors might coexist in cell populations. Therefore, to fully characterize the functional properties of receptor systems in cell populations, variances and covariances of single-cell kinetic parameters have to be taken into account.
In ODE models describing comprehensively characterized cellular signal transduction networks, ODEs can reflect biochemical reactions in detail rather than summarizing several biochemical processes in single reactions. In this case, the same kinetic parameters can be assumed for different cells, and cell-to-cell variability can be inferred by different initial concentrations of involved signal transduction proteins [40,71,72]. As a consequence, correlations between signaling species become important for quantitative predictions. In a previous modeling study of programmed cell death, it was shown that for describing experimental data from a heterogeneous population of cells undergoing apoptosis, correlations between initial protein concentrations had to be taken into account to obtain realistic model predictions [53]. Furthermore, comparable to our study, correlations between initial concentrations of opposing signaling species, which were either anti- or pro-apoptotic, buffered variability of cell death times.
The motif of limiting variability by correlated kinetics of opposing reactions can be seen in the context of other mechanisms, which limit variability in biological systems such as negative feedback, for example due to ligand-dependent receptor internalization or inhibition of upstream kinases by downstream kinases, or incoherent feed-forward loops [70,73–79]. Dampening of cell-to-cell variability by co-regulation of different trafficking processes would, however, not be regarded as a direct regulatory mechanism that results from the structure of a specific signal transduction network as it is the case for negative feedback loops. In-depth analysis of how different receptor transport processes are mechanistically inter-regulated and depend on the cellular population context that was shown to be relevant for explaining cell-to-cell variability in endosomal trafficking [30], will be an important topic of future work.
In several cellular systems, single-cell dynamics significantly deviate from the behavior observed at the level of cell populations [40,42,80]. On the contrary, we observed for the EpoR that the model by Becker et al. [12], which was based on cell population average data, corresponded to the model variant that explained single-cell data best. In addition to the model by Becker et al., our model accounts for the intracellular pool of free EpoR, for EpoR synthesis and degradation. Although, that study had used a different EpoR-expressing suspension cell line, in this study, we obtained similar kinetic parameters for EpoR trafficking in adherent EpoR-GFP expressing H838 cells. This finding suggests that dynamic properties of the EpoR system are conserved between different types of cells.
We observed that internalization of Epo-EpoR complexes was not substantially accelerated compared to free EpoR, in contrast to other receptor systems, which is consistent with the finding that EpoR is internalized in a ligand-independent manner [81], and was similarly observed in BaF3 cells at the cell population level [12]. Several other receptors as the epidermal growth factor receptor (EGFR), the insulin receptor, the growth hormone receptor or the leukemia inhibitory factor receptor show substantially accelerated internalization of activated receptors [14,47,82–84], which facilitates a high temporal resolution in receptor signaling [1,3,24]. Contrarily, EpoR signaling rather depends on fast transport of EpoR between membrane and cytosolic compartments and rapid ligand depletion [11,12].
Confocal microscopy combined with 3D image segmentation was the method of choice for the time-resolved quantification of fluorescently labeled proteins in cellular compartments but offered lower throughput compared to other experimental methods for studying cellular heterogeneity as fluorescence-activated cell sorting (FACS). Nevertheless, significant correlations between single-cell parameters could be identified with the given set of single-cell data.
An essential aspect of our study was the refinement of the cell ensemble modeling approach. These advances comprised the implementation of constraint terms [40] that minimized the deviations of kinetic parameter distributions in sets of single cells treated under different experimental conditions and that were added to the log-likelihood function for parameter estimations. The approach of merging single-cell trajectories from qualitatively different experiments is widely applicable and can be transferred to various other models of cellular signaling pathways.
Taken together, we could show by combining quantitative live-cell imaging of erythropoietin receptor trafficking with mathematical modeling that receptor transport processes largely differed between individual cells. Receptor concentrations in cellular compartments were nevertheless robust to variability in trafficking processes due to the correlated kinetics of opposing transport processes.
Stable cell lines were generated from the human NSCLC cell line H838 (ATCC CRL-5844) that was purchased from American Type Culture Collection (ATCC, Manassas, VA, USA). From wild-type H838 cells, EpoR-GFP expressing cell lines were selected with 2.0 μg/ml puromycin (Sigma-Aldrich, Taufkirchen, Germany), and MyrPalm-mCherry expressing cell lines were selected with 0.8 mg/ml G418 (Sigma-Aldrich). Cell lines were maintained in DMEM (Invitrogen, Darmstadt, Germany) containing 10% fetal calf serum (Biochrom AG, Berlin, Germany), 100 μg/ml penicillin and streptomycin (Invitrogen). Medium for stably transfected cell lines additionally contained 0.2 mg/ml G418 or 0.2 μg/ml puromycin. For microscopy, cells were maintained in 8-well Lab-Tek chambers (Thermo Scientific, Asheville, NC, USA) with a density of 40.000/well. Before experiments, cells were washed and maintained in DMEM without growth factors for 3 hours to prevent basal phosphorylation of EpoR.
We used the murine EpoR, which was well characterized in previous studies and is functionally equivalent to the human EpoR [12]. To express the fluorescently labeled EpoR, we utilized the retroviral expression vector pMOWS-puro encoding the murine EpoR C-terminally fused to GFP that was previously described in [85] and results in a protein of approximately 90kDa. For retroviral transduction of H838 cells, Phoenix ampho cells were transfected by the calcium phosphate precipitation method. Transducing supernatants were generated 24 hours after transfection by passing through a 0.45 μm filter (Millipore, Billerica, MA, USA). H838 cells were treated with 1ml of supernatant supplemented with supplemented with 8 μg/ml polybrene (Sigma-Aldrich) on a 6-well plate at a density of 2·105 cells per well and spin-infected for 3h at 340g. Stably transduced H838 cells expressing EpoR-GFP were selected in the presence of 1.5μg/ml puromycin (Sigma-Aldrich) 24 hours after infection. The myristoylation-palmitoylation (MyrPalm) fusion construct with mCherry was a kind gift of Joel Beaudouin. It was constructed as described in [44]. To generate H838 cells stably expressing MyrPalm-mCherry and EpoR-GFP, EpoR-GFP expressing H838 cells were transfected with X-tremeGENE 9 (Roche Pharma, Basel, Switzerland) and selected with 2mg/ml G418. Cells were treated with Epo-Cy5.5 (Roche Diagnostics, Penzberg, Germany), which is a fully bioactive EpoR ligand [45]. Cy5.5 fluorescence was shown to be not pH dependent in the physiologic pH range, compared to fluorescein, because of its missing 3’-hydroxyl substituent [86].
Immunoblot samples were lysed with lysis buffer (20 mM Tris/HCl, pH 7.5, 150 mM NaCl, 1 mM phenylmethylsulfonyl fluoride (Sigma-Aldrich), protease inhibitor cocktail, 1% Triton X-100 (Serva, Mannheim, Germany), and 10% glycerol). Cell lysates were analyzed using SDS PAGE gels (Invitrogen). Proteins were transferred to PVDF membrane (Merck Millipore) using wet blotting. Detection was performed using the Pico Chemiluminescent Substrate from Thermo Scientific and a CCD camera (Intas, Göttingen, Germany). EpoR-GFP concentrations in EpoR-GFP-expressing H838 cells were quantified utilizing recombinant eGFP (BioVision, Mountain View, CA, USA). Cell lysates were combined with different amounts of GFP ranging from 0.2 to 10 ng and then loaded onto gels (S3 Fig). To detect EpoR-GFP and GFP in immunoblots, we used an antibody recognizing GFP (clones 7.1 and 13.1) from Roche (Basel, Switzerland). Horseradish peroxidase-conjugated secondary antibodies (Southern Biotech, Birmingham, AL, USA) were used for detection.
Live-cell experiments were performed in a 37°C, 5% CO2 incubation chamber on a CSU-22 Yokogawa spinning disk confocal (Yokogawa Electric Corporation, Tokyo, Japan) on a Nikon Ti inverted microscope equipped with 60x Plan Apo NA 1.4 objective lens (Nikon, Tokio, Japan), a Hamamatsu C9100-02 EMCCD camera (Hamamatsu Photonics, Hamamatsu, Japan) and a PerkinElmer Photokinesis bleaching/photoactivation unit (PerkinElmer, Waltham, MA, USA), using Volocity software (PerkinElmer). GFP (EpoR-GFP) fluorescence was excited at 488 nm and collected with a 527/55 emission filter (Chroma Technology Corp, Bellows Falls, VT, USA) and an exposure time of 200 ms. For bleaching, we used the FRAP module of Volocity software. Cherry (MyrPalm-mCherry) fluorescence was excited at 561 nm and collected with a 615/70 emission filter (Chroma Technology Corp) and an exposure time of 300 ms. Cy5.5 (Epo-Cy5.5) fluorescence was excited at 640 nm and collected with a 705/90 emission filter (Chroma Technology Corp) and an exposure time of 200 ms. Laser intensity was kept at a low level, at which no effect of bleaching was observed. A binning of 2x2 pixels was used. At each time point, z-stacks with 26 slides at 0.7μm step size were recorded.
In live-cell imaging experiments, cells were treated with Epo-Cy5.5 at a concentration of 4.2 nM in a total volume of 400 μl. To facilitate an even distribution of the ligand, cells were kept in 200 μl medium while recording the first image stack, before adding 200 μl Epo-Cy5.5 at a concentration of 8.4 nM to obtain the desired concentration of 4.2 nM. Within the first 30 minutes, we recorded at a time interval of 5 minutes, afterwards at a time interval of 10 minutes to obtain more densely sampled measurements at the beginning of the experiment where the signal changes were strongest.
Given an average flux FEpoR*,ItoM of Epo-EpoR complexes from the plasma membrane to the cytosol of 0.8 nM/min (Fig 5B), and the average cell volume of 5.47 pl obtained from stack segmentations, the average number of Epo molecules internalized by a single cell will be about 440 molecules per minute. Therefore, within the experimental duration of 300 minutes, given the amount of 40.000 cells per well, about 3% of the total amount of Epo-Cy5.5 will be internalized in cells. It was shown that a fraction of the amount of Epo, which was secreted after internalization by cells, was still intact and could stimulate other cells [12]. Therefore, the fraction of Epo removed from the medium will be effectively less than 3%. This justifies the model assumption of constant Epo-Cy5.5 concentrations in the medium.
We developed custom graphical user interface-based software in MATLAB (The Mathworks, Natick, MA, USA) for segmentation of cellular compartments from image stacks (S1 Fig; for details, see S1 Text). MyrPalm-mCherry signals were used to segment the plasma membrane region of interest (ROI), EpoR-GFP and Epo-Cy5.5 signals were used to define EpoR or EpoR-Epo vesicles. To obtain observables that were proportional to variable concentrations, fluorescence intensities were normalized by cell volumes (for details, see S2 Text). Absolute volumes were calculated by multiplying voxel numbers and voxel volumes of 0.29x0.29x0.7μm3.
All ODE models were implemented with the MATLAB toolbox PottersWheel that was used for parameter calibrations (http://www.potterswheel.de) [87]. Model analysis and simulations were performed with custom MATLAB scripts. As a measure for the goodness of fit, we used the corrected Akaike information criterion (AICcorr). Model equations can be found in S1, S2, S4 and S5 Tables, and parameter estimates in S6 and S7 Tables (for details, see S2 Text). To test for linear correlation, we calculated Pearson correlation coefficients.
|
10.1371/journal.pbio.2004037 | The neural system of metacognition accompanying decision-making in the prefrontal cortex | Decision-making is usually accompanied by metacognition, through which a decision maker monitors uncertainty regarding a decision and may then consequently revise the decision. These metacognitive processes can occur prior to or in the absence of feedback. However, the neural mechanisms of metacognition remain controversial. One theory proposes an independent neural system for metacognition in the prefrontal cortex (PFC); the other, that metacognitive processes coincide and overlap with the systems used for the decision-making process per se. In this study, we devised a novel “decision–redecision” paradigm to investigate the neural metacognitive processes involved in redecision as compared to the initial decision-making process. The participants underwent a perceptual decision-making task and a rule-based decision-making task during functional magnetic resonance imaging (fMRI). We found that the anterior PFC, including the dorsal anterior cingulate cortex (dACC) and lateral frontopolar cortex (lFPC), were more extensively activated after the initial decision. The dACC activity in redecision positively scaled with decision uncertainty and correlated with individual metacognitive uncertainty monitoring abilities—commonly occurring in both tasks—indicating that the dACC was specifically involved in decision uncertainty monitoring. In contrast, the lFPC activity seen in redecision processing was scaled with decision uncertainty reduction and correlated with individual accuracy changes—positively in the rule-based decision-making task and negatively in the perceptual decision-making task. Our results show that the lFPC was specifically involved in metacognitive control of decision adjustment and was subject to different control demands of the tasks. Therefore, our findings support that a separate neural system in the PFC is essentially involved in metacognition and further, that functions of the PFC in metacognition are dissociable.
| Decision-making is often accompanied by a sense of uncertainty regarding the outcome. In many situations, there is no explicit feedback or cue to indicate whether the decision is correct or not. Fortunately, our brain can evaluate decision uncertainty using the internal signals and subsequently make appropriate adjustments to initial decisions. The process of considering the outcome of a decision and whether a decision should be adjusted is called metacognition, and it tends to be automatically induced. Thus, decision-making is usually accompanied by metacognition, and the two processes are inevitably coupled. However, the neural systems supporting metacognitive processing remain unclear and have often been misattributed to the neural system of the decision-making process per se. Here, we have analyzed this process in several volunteers by imaging the brain activity in specific regions while they performed Sudoku and random-dot motion (RDM) tasks. Our results suggest the existence of a neural system located in the prefrontal cortex (PFC) mainly involved in metacognition and independent from the neural system of decision-making.
| Decision-making is a process of evidence accumulation. That evidence may come from sensory signals of external stimuli or from mental representations of internal cognitive operations. Variations in evidence can create uncertainty in the person rendering a decision. The decision maker is normally explicitly or implicitly aware of uncertainties about a decision and consequently confirms or revises a decision even prior to, or in the absence of, external feedback. In the framework of cognitive control, the processes of decision uncertainty monitoring—and consequent decision adjustments—are termed metacognition, that is, ‘cognition about cognition’ [1–4]. Although metacognition generally accompanies decision-making with uncertainty, the underlying neural system of the metacognitive processes in decision uncertainty monitoring and consequent decision adjustments remains less clear than that of the decision-making process per se [5, 6].
Much of the work on the neural bases of metacognition in humans has focused on metacognitive monitoring of internal states (i.e., confidence or uncertainty) with regard to the cognitive processes such as episodic memory [7, 8] and sensory perception [9, 10]. Behaviorally, confidence ratings, which reflect subjective accuracy beliefs regarding decisions, have often been found to deviate from the accuracy of an actual decision [11–13]. These observations have suggested the existence of a separate neural processing system (meta-level) in the generation of decision confidence or uncertainty, independent of the decision-making process per se (object-level). We hereafter refer to this description of metacognition as separable from decision-making as “Theory 1” [11–18]. The prefrontal cortex (PFC) has been proposed to play a critical role in metacognition [14], and it has been demonstrated that interference with or lesions in PFC regions may impair metacognitive monitoring of perceptual decisions, but not decisions per se [15–18, but see also 19].
A contrary theory, which we will refer to as “Theory 2,” suggests that metacognition may be merely dependent on the decision-making process and therefore exclusively reliant on accumulated evidence [20–24]. Specifically, this theory, based on bounded accumulation models, has interpreted divergence between decision accuracy and confidence reports as being caused by the accumulation of postdecisional evidence during the interval between decision-making and confidence reporting [20–24]. Furthermore, it implies that decision adjustment naturally occurs as a part of this continuous postdecisional evidence accumulation and therefore is an integrated part of the initial decision-making process [21, 24]. Some proponents of this theory have argued that a separate neural system for metacognition to monitor and control decision-making should not be necessary because the processes are interdependent [24]; however, not all work supporting this theory insists on this notion [20].
Thus, one of the crucial issues in the debate between the two theories is whether a separate neural system for metacognition exists. Single-decision paradigms (depicted in Fig 1A and 1C) are not sufficient to determine the existence or nonexistence of separable systems because the decision-making process and the metacognitive process are inevitably coupled in such tasks. The purpose of retrospective metacognition is to confirm or revise foregone decisions. Given an opportunity to make a decision on the same situation again (i.e., make a “redecision”) (as depicted in tasks shown in Fig 1B and 1D), a decision maker may revise an initial decision as well as confidence in the decision once s/he detects uncertainty regarding the initial decision [25]. Thus, if a separate neural system for metacognition exists as proposed by Theory 1, the metacognitive processes—in particular metacognitive control—should be more extensively involved in redecision, especially if the initial decision must be made quickly. On the contrary, the neural system involved in redecision should be the same as those involved in an initial decision if there is not a separate neural system for metacognition (as proposed by Theory 2). If a separate neural system for metacognition exists, the activity of this system should be manifest after an initial decision is reached, whereas Theory 2 suggests that they share the same underlying neural systems and that neural activity following either a single decision or a redecision should be the same.
Therefore, comparing behavioral and neural differences between the two phases of initial decision and redecision may allow us to test which theory better accounts for the neural processing of metacognition. A specific perspective of metacognition derived from Theory 1 implies that decision uncertainty, rather than decision confidence, should be the key signal for metacognition. If there is no uncertainty regarding a decision, it should not evoke the processes of metacognitive monitoring and control. Therefore, the critical aim of this study was to elicit and analyze neural activity positively correlated with decision uncertainty, rather than that positively correlated with decision confidence.
In the present study, we employed a novel “decision–redecision” experimental paradigm to investigate neural activity associated with metacognition. The participants were asked to make two consecutive decisions on the same situation using a perceptual decision-making task and a rule-based decision-making task (Fig 2A). We combined this novel paradigm with the functional magnetic resonance imaging (fMRI) technique to formally test the two theories and systematically investigate the underlying neural substrates of metacognitive processes accompanying decision-making. Based on our previous study [25], we expected that the frontoparietal control network would be associated with metacognitive processing. In the current study, we focused on specific functions of the regions in the network believed to be involved in metacognition. We found that dorsal anterior cingulate cortex (dACC) activity significantly correlated with metacognitive monitoring of decision uncertainty and that lateral frontopolar cortex (lFPC) activation correlated instead with metacognitive control. These findings provide evidence for distinct neural processes involved in metacognition and decision-making.
We developed a novel decision–redecision paradigm for this study (Fig 2A). The participants made an initial decision (decision phase), immediately followed by another decision on the same situation (redecision phase). This allowed the participant the opportunity to revise the initial decision and update their confidence rating, even without feedback. The internal states of uncertainty regarding initial and final decisions were separately evaluated by confidence ratings. Confidence was rated on a scale of 1 to 4, immediately following the corresponding decisions. Decision uncertainty was then the inverse of the confidence rating (i.e., a confidence rating of 4 corresponded to an uncertainty rating of 1). Critically, our task differed from previous paradigms used to analyze ‘change of mind’ [21, 24]. Previous task paradigms were only able to analyze the small portion of trials in which a participant happened to change their mind, while our paradigm allowed analysis of each trial.
We used two different types of decision-making tasks in the present study: one was a rule-based decision-making task (Sudoku), the other a perceptual decision-making task (random-dot motion [RDM]), which has commonly been used to investigate the neural process of decision-making [5] and more recently, metacognition [21, 22, 24, 26]. The decisions in the Sudoku task rely on internal informational operations, but decisions in the RDM task should be more dependent on accumulation of external information. It is possible to continue accumulating evidence from external stimuli that may affect decision-making in the RDM task, but that is less likely in the Sudoku task because it is rule based. For this reason, the two tasks should result in differential processing in metacognitive control to adjust initial decisions. The sequences of both tasks were identical (Fig 2A, illustrated for the main fMRI experiment [fMRI1]). After a Sudoku problem or RDM stimulus was presented for 2 s, the participant made a choice from 4 possible solutions within 2 s and then reported their confidence rating on that decision within 2 s. A critical feature of our paradigm was that the same Sudoku problem or the same RDM stimulus was immediately repeated for 4 s, and the participant again made a choice within 2 s and again reported their confidence rating within 2 s. To better distinguish the metacognitive process from the decision-making process, we intentionally set a short initial decision phase (2 s), to minimize metacognition during the initial decision-making phase, but set a longer duration in the redecision phase (4 s) to allow enough time for metacognitive processing in redecision. There was no explicit feedback or cue to indicate whether the decision was correct after either the initial decision or the redecision. For both tasks, the task difficulty of each trial (Fig 2B) was adaptively adjusted by a staircase procedure [9, 27] so that the average accuracy for the initial decision was converged to approximately 50% (chance level was 25%). For the control condition, the participant was shown a digital number in the target grid in the Sudoku task, and for the RDM task, s/he was shown an RDM stimulus with 100% coherence. For the former, the participant only needed to press the button matching the number, and for the latter, the participant indicated the unambiguous RDM direction. Prior to the experimental testing, the participant was trained to attain high-level proficiency in Sudoku problem-solving.
The current study was composed of 4 fMRI experiments:
Twenty-one participants took part in fMRI1 (see Materials and methods). In both the Sudoku and RDM tasks, decision uncertainty levels were largely consistent with the percentage of incorrect initial decisions (Fig 2C; Pearson’s r = 0.76 ± 0.12 [mean ± SD], one-tailed t test, t21 = 7.3, P = 1.7 × 10−7 in the Sudoku task; r = 0.71 ± 0.14, t21 = 6.8, P = 5.0 × 10−7 in the RDM task). To examine the trial-by-trial consistency between objective erroneous decisions and subjective decision uncertainty levels in individual participants, a nonparametric approach was employed to construct the receiver operating characteristic (ROC) curve by using the decision uncertainty levels as thresholds to characterize the likelihood of erroneous decisions. The area under curve (AROC) was then calculated to represent the individual uncertainty sensitivity, indicating how sensitive the participant was to the decision uncertainty [9]. As observed in the previous studies, the uncertainty sensitivity of individual participants markedly deviated from decision accuracy in both tasks, which were controlled around 50% (Fig 2D). The response times (RTs) of option choices in the initial decision were positively correlated with the decision uncertainty levels (Fig 2E; t21 = 6.9, P = 4.0 × 10−7 in the Sudoku task; t21 = 4.3, P = 1.6 × 10−4 in the RDM task). The correlation coefficient between RT of option choices and the decision uncertainty level (rRT-uncertainty) in the initial decision was highly correlated with the uncertainty sensitivity (AROC1) across the participants (Pearson’s r = 0.61, z test, z = 3.4, P = 4.0 × 10−4 in the Sudoku task; r = 0.48, z = 2.4, P = 0.0085 in the RDM task). Thus, the RT–uncertainty correlation also reflected individual uncertainty sensitivity.
The level of decision uncertainty was reduced by redecision. The extent of decision uncertainty reduction via redecision was highly correlated with the decision uncertainty level in the initial decision phase (Fig 2F; Goodman and Kruskal’s γ = 0.82 ± 0.11, t21 = 8.8, P = 2.1 × 10−8 in the Sudoku task; γ = 0.78 ± 0.14, t21 = 7.7, P = 8.2 × 10−8 in the RDM task). Accordingly, accuracy also improved along with uncertainty reduction (Fig 2G; Pearson’s r = 0.54 ± 0.13, t21 = 4.2, P = 2.3 × 10−4 in the Sudoku task; r = 0.39 ± 0.14, t21 = 2.8, P = 5.6 × 10−3 in the RDM task). One could suspect that the improvement of uncertainty reduction and the change in accuracy in the redecision phase were caused by regression towards mean in the two separate decisions: higher uncertainty at the first measurement by chance would increase improvement at the second measurement. However, the decision accuracy and decision uncertainty levels for the final decision-making phase remained significantly differential (Pearson’s r = 0.35 ± 0.15, t21 = 2.1, P = 0.032 in the Sudoku task; r = 0.36 ± 0.14, t21 = 2.6, P = 8.9 × 10−3 in the RDM task in Fig 2C; Pearson’s r = 0.32 ± 0.14, t21 = 2.0, P = 0.042 in the Sudoku task; r = 0.32 ± 0.15, t21 = 2.2, P = 0.028 in the RDM task in Fig 2G), indicating that the participants’ performance in redecision reflected metacognitive processing ability rather than chance. Despite the fact that both decision accuracy and decision uncertainty levels were improved in the redecision phase, the divergence between uncertainty sensitivity and decision accuracy remained significant (Fig 2H). Indeed, neither the individual uncertainty sensitivities nor those of individual differences were altered by redecision (Fig 2I; t21 = 0.82, P = 0.21 in the Sudoku task; t21 = 1.0, P = 0.15 in the RDM task). Similarly, neither the individual RT–uncertainty correlation coefficients nor those of individual differences were altered by redecision (Fig 2E; t21 = −0.77, P = 0.22 in the Sudoku task; t21 = 0.35, P = 0.36 in the RDM task). These results show that individual uncertainty sensitivity was stable, was intrinsic to individual metacognitive ability, and was independent of the accumulated evidence and the type of decision-making required.
Commonly across both tasks, brain activation during the initial decision phase was mainly restricted to brain areas posterior to the PFC, in particular the posterior portion of the PFC, the inferior frontal junction (IFJ) (S1A Fig; Fig 3A). In the redecision phase, a frontoparietal control network—consisting of the lFPC, dACC, anterior insular cortex (AIC), middle dorsolateral PFC (mDLPFC), and anterior inferior parietal lobule (aIPL)—was more extensively recruited (Fig 3B; S1B and S2 Figs; S1 Table). In contrast, the lFPC and mDLPFC regions of the frontoparietal control network were not activated when a new Sudoku problem or a new RDM stimulus was presented for the first time during the redecision phase, preceded by the control stimuli in the initial phase (fMRI2, n = 17; S1C and S3 Figs), while the dACC activity during the same phase became much weaker, and its response onset was much delayed from the onset of the stimulus presentation (delay offset >3 s; S3 Fig). Thus, the frontoparietal control network, in particular the regions of the lFPC, mDLPFC, and dACC in the anterior FPC, were more extensively involved in redecision than in the initial decision phase.
Trial-by-trial activity in the regions of the frontoparietal control network in redecision was positively correlated with decision uncertainty level for the initial decision (Fig 3C and S2 Table). Critically, these correlations remained significant even for the correct trials (S1E Fig), indicating that these regions were encoding the decision uncertainty signal rather than the error signal. Furthermore, task difficulty or RT could not explain their association with the decision uncertainty in these regions. The residual fMRI signal changes after the components associated with the task difficulty and RT were regressed out remained highly correlated with decision uncertainty level, but residual fMRI signal changes after the components of the decision uncertainty level were regressed out were not further correlated with the task difficulty and RT. Although the dACC and AIC regions were also partially activated during the initial decision phase (S1A Fig and Fig 3A), this activity—as well as in other regions activated during the same phase—was neither positively nor negatively correlated with the decision uncertainty level (S1D Fig). Activity in the ventromedial PFC (VMPFC) and posterior cingulate cortex (PCC) regions of the default-mode network in redecision were negatively correlated with decision uncertainty level or positively correlated with its inverse—decision confidence (S1F Fig). Thus, the regional activity seen in the frontoparietal control network involved processes intrinsic to redecision but not the activity involved in decision-making for the initial phase.
In the third fMRI experiment (fMRI3, n = 25), we confirmed that the strength of activity in the frontoparietal control network depended critically on whether redecision was required after the initial decision phase. When decision uncertainty levels for initial decisions were matched in the two conditions (two-tailed paired t test, t25 = 0.62, P = 0.27), activity in the frontoparietal control network was much stronger when redecision was required (‘redecision condition’), in comparison with those when redecision was not required (“non-redecision condition”) (Fig 3D), despite the fact that activation of the frontoparietal control network in the ‘non-redecision condition’ was also significant (S1G Fig) and was correlated with decision uncertainty level as well [25]. Thus, the frontoparietal control network, more strongly activated in redecision, should not only be involved in metacognitive monitoring of decision uncertainty of the initial decision but also in metacognitive control in redecision (Fig 1D). We then putatively defined this frontoparietal control network as the metacognition network.
Because the duration of the redecision phase in fMRI1 was longer (4 s) than that of the initial decision phase (2 s), it raised the question of whether the fMRI activity predominately observed during the redecision phase was induced by the longer exposure, specifically in the trials with more difficult decisions. To address this, we scanned an independent group of participants (fMRI4, n = 20) while they underwent the same RDM task as fMRI1 except that the duration of redecision was set to 2 s. The same behavioral and neural results were replicated as in fMRI1 (S4 Fig).
Just as the extent of uncertainty reduction by redecision was found to be highly correlated with the decision uncertainty level of the initial decision (Fig 2F), activity in the regions of the metacognition network were also found to be positively correlated with the extent of uncertainty reduction (S1H Fig). However, the strength of the correlations decreased somewhat after the components associated with the decision uncertainty level were regressed out (S1I Fig). Conversely, correlations with decision uncertainty level in the metacognition network remained significant after the components associated with the extent of uncertainty reduction were regressed out (S1J Fig). These partial correlations complementarily confirmed that the metacognition network in redecision was involved in both metacognitive monitoring and metacognitive control, indicating that the two processes interacted in redecision processing.
The two processes, although interactive, can be dissociated. In the region involved in uncertainty monitoring, activity strength should dynamically represent decision uncertainty level. As decision uncertainty levels were reduced by redecision, the strength of its activity should accordingly be reduced. Therefore, the neural activity change should be negatively correlated with the extent of decision uncertainty reduction. Alternatively, in the region that was critically involved in metacognitive control, its activity should become positively correlated with the extent of decision uncertainty reduction, representing the outcome or the extent of metacognitive control. We found that the activity in the dACC and AIC regions at the late phase of redecision did in fact negatively correlate with the extent of decision uncertainty reduction after the components associated with the decision uncertainty level of the initial decision were regressed out (Fig 4A, S1K and S1L Fig). Conversely, the lFPC activity in the Sudoku task was positively correlated with the extent of decision uncertainty reduction after components associated with the decision uncertainty level were regressed out (Fig 4B), but negatively in the RDM task (Fig 4B and S1I Fig). In addition, VMPFC activity was also positively correlated with the extent of decision uncertainty reduction in both tasks (S1I Fig). The regional activity in the default-mode network appeared intrinsically anticorrelated with the regional activity in the metacognition network (further detail regarding activity in the default-mode network associated with metacognition will be discussed in another study). Thus, the dACC and AIC regions were specifically involved in metacognitive monitoring. In contrast, the lFPC was specifically involved in metacognitive control in redecision, particularly in the Sudoku task. Therefore, their functional roles in metacognition appear to dissociate in redecision processing.
In the Sudoku task, whether the problem would be better solved should be conditioned to individual intrinsic motivation to engage metacognitive control because metacognitive control was effortful. The ventral striatum (VS) activity during the redecision phase was positively correlated with the extent of decision uncertainty reduction in the Sudoku task, but not in the RDM task (Fig 4C). VS might encode the intrinsic motivation or the internal reward on reduction in uncertainty during the redecision phase in the Sudoku task. Critically, the lFPC activity was significantly coupled with the interaction between the VS activity and the decision uncertainty level of the initial decision (Fig 4D; see psycho–physiological interaction [PPI] analysis in Materials and methods). Furthermore, the accuracy change of each participant by redecision was positively correlated with the coupling strength in the Sudoku task (Fig 4E). These results implied that the efficiency of lFPC involvement in metacognitive control in rule-based decision-making tasks (i.e., Sudoku) might be facilitated by the VS activity.
The abilities of metacognitive monitoring and control are behaviorally embodied in two components: uncertainty sensitivity and accuracy change, respectively. Throughout all sessions, including fMRI1 and the other repeated behavioral experiments, the individual uncertainty sensitivity was highly consistent across different sessions of the Sudoku task (Cronbach’s α = 0.91; Fig 5A, left column, upper panel) and the RDM task (α = 0.89, Fig 5A, left column, middle panel), as well as across the two tasks (α = 0.85; Fig 5A, left column, lower panel). In contrast, the individual accuracy change in redecision was not consistent across the two tasks (α = 0.03; Fig 5A, right column, lower panel), although it was consistent between different sessions of the Sudoku task (α = 0.80; Fig 5A, right column, upper panel) and the RDM task (α = 0.76; Fig 5A, right column, middle panel). Thus, individual metacognitive abilities of uncertainty monitoring were reliably consistent, but individual metacognitive control was dissociable in the two tasks.
Accordingly, the individual uncertainty sensitivity (AROC) was positively correlated with the uncertainty-level regression β value of the fMRI signal changes (i.e., neural uncertainty sensitivity), primarily in the dACC and AIC regions (Fig 5B, P < 0.001, cluster size = 20; and Fig 5C upper, one-tailed t test, Pearson’s r = 0.79, t19 = 5.6, P = 6.0 × 10−6 in the Sudoku task; r = 0.55, t19 = 2.9, P = 0.0049 in the RDM task; S3 Table), but not in the lFPC region (Fig 5B and 5C bottom; Pearson’s r = 0.17, t19 = 0.8, P = 0.22 in the Sudoku task; r = 0.21, t19 = 1.0, P = 0.17 in the RDM task), commonly in both tasks. The differences of correlations were significant between the two regions (t19 = 3.8, P = 5.6 × 10−4 in the Sudoku task; t19 = 2.3, P = 0.016 in the RDM task). In contrast, the individual accuracy change was significantly correlated with the mean activity in the lFPC region (Fig 5D, P < 0.001, cluster size = 20; and Fig 5E bottom; Pearson’s r = 0.69, t19 = 4.2, P = 2.2 × 10−4 in the Sudoku task; r = −0.39, t19 = 1.9, P = 0.041 in the RDM task), but not in the dACC region (Fig 5D and 5E upper; Pearson’s r = 0.18, t19 = 0.8, P = 0.21 in the Sudoku task; r = −0.02, t19 = 0.09, P = 0.47 in the RDM task). When the lFPC activity was stronger, the accuracy change was more in the Sudoku task but became less in the RDM task (Fig 5E). The differences of correlations were significant between the two regions (t19 = 2.7, P = 0.007 in the Sudoku task; t19 = 1.8, P = 0.045 in the RDM task). Thus, the dACC activity (AIC as well) commonly represented individual metacognitive abilities in monitoring of decision uncertainty, whereas the lFPC differentially modulated individual metacognitive abilities in control of decision adjustment—in both the Sudoku and RDM tasks—consistent with their dissociated functional roles in metacognitive monitoring and metacognitive control, respectively.
The regions of the metacognition network were also activated in the trials of both tasks with confidence level 4 in comparison with their respective control conditions (Fig 6B and S2 Fig). These activity differences might be partially caused by differentially subjective uncertain states of the two conditions that were not reflected by the four-scale confidence ratings (i.e., the ceiling effect). The averaged accuracy was about 80% in the certain trials of the tasks (Fig 2C), but it was about 95% in the control conditions. Nevertheless, the task baseline activity in the certain trials of the tasks could also predict the individual uncertainty monitoring bias and potential abilities of efficient metacognitive control of decision adjustment. Individual uncertainty monitoring bias—as estimated by averaging the decision uncertainty levels of all trials in each session of the tasks, representing the individual’s overconfident or underconfident tendency—was consistent between different sessions in the Sudoku task (α = 0.95; Fig 6A, left panel) and in the RDM task (α = 0.94, Fig 6A, middle panel), as well as across the two tasks (α = 0.91; Fig 6A, right panel). Accordingly, individual uncertainty monitoring bias was positively correlated with the mean task baseline activity in the dACC region (Fig 6C P < 0.001, cluster size = 20; and Fig 6F left, Pearson’s r = 0.50, t19 = 2.5, P = 0.0096 in the Sudoku task; r = 0.44, t19 = 2.1, P = 0.022 in the RDM task) but not in the lFPC region (Fig 6C and 6F right; Pearson’s r = 0.18, t19 = 0.80, P = 0.22 in the Sudoku task; r = −0.04, t19 = 0.17, P = 0.43 in the RDM task), commonly in both tasks. The differences of correlations were significant between the two regions (t19 = 2.1, P = 0.026 in the Sudoku task; t19 = 1.8, P = 0.042 in the RDM task). Meanwhile, the individual accuracy change in the Sudoku task was positively correlated with the mean task baseline activity in the lFPC region (Fig 6D and 6G right; Pearson’s r = 0.45, t19 = 2.2, P = 0.020) but not with that in the dACC region (Fig 6G left; one tailed t test, r = 0.14, t19 = 0.62, P = 0.27). In contrast, the individual accuracy change in the RDM task was negatively correlated with the mean task baseline activity in the lFPC region (Fig 6E and 6G right; Pearson’s r = −0.40, t19 = 1.9, P = 0.035) but not with that in the dACC region (Fig 6G left; r = −0.13, t19 = 0.57, P = 0.29). The differences of correlations were significant between the two regions (t19 = 1.9, P = 0.039 in the Sudoku task; t19 = 1.8, P = 0.046 in the RDM task). Furthermore, the differences of correlations with the individual uncertainty monitoring bias and the individual accuracy change in the dACC (t19 = 2.2, P = 0.020 in the Sudoku task; t19 = 2.8, P = 0.0055 in the RDM task), as well as in the lFPC (t19 = 1.8, P = 0.046 in the Sudoku task; t19 = 2.0, P = 0.030 in the RDM task), were significant. Thus, the task baseline activity in the dACC region commonly reflected the individual uncertainty monitoring bias in both tasks, whereas that in the lFPC region could predict the individually differential potential abilities of metacognitive control for decision adjustment in both tasks.
Thus far, we have shown that the neural system of metacognition could be dissociated into at least two subsystems: the dACC and AIC regions involved in metacognitive monitoring of decision uncertainty, and the lFPC region involved in metacognitive control of decision adjustment. To further elaborate the subsystems of the metacognition network, we performed analyses of interregional functional connectivity in the metacognition network. By regressing out the mean activity and the modulations by the decision uncertainty level, the RT and the extent of uncertainty reduction, as well as their interactions, we calculated trial-by-trial correlations between each pair of regions in the metacognition network (see Materials and methods). The interregional functional connectivity patterns in both the task condition (Fig 7A) and the control condition (Fig 7B) were almost identical across the two types of tasks and were also similar to those at the resting state (Fig 7C). The interregional functional connectivity patterns consistently showed that the metacognition network might be divided into three subsystems: the lFPC region; the dACC and AIC regions; and the DLPFC and aIPL regions. The interregional functional connectivity within each of the subsystems was systematically stronger than that across the subsystems (paired t test, P < 0.05 in all comparisons). So far, the functional roles of the subsystem consisting of the DLPFC and aIPL regions in metacognition remain unclear. It is worth noting that the functional connectivity between the dACC and the regions of the other two subsystems in the task conditions was numerically stronger than the corresponding one at the resting state but was not statistically significant.
In the present study, we utilized a novel decision–redecision paradigm to examine the behavioral and neural associations of metacognitive processing in redecision, as compared to the processing in an initial decision. The robust findings from our study showed that individual uncertainty sensitivity (both AROC and rRT-uncertainty) remained markedly stable over two consecutive decisions on the same situational task, between different sessions of the same tasks, and across the different tasks. This indicates that individual uncertainty sensitivity was independent of evidence accumulation or the form of the decision-making process. These findings provide evidence to contradict the theoretical prediction of Theory 2. If the processes of metacognitive processing and decision-making were integrated in one network, it should follow that, as more evidence is accumulated after redecision, the uncertainty sensitivity (i.e., AROC) should be also improved [28]. Our study did not support that but rather led us to suggest the existence of an additional neural process in the brain to nonuniformly transform evidence accumulated in the decision-making processes to neural signals encoding decision confidence/uncertainty. The decision confidence/uncertainty should be much constrained by this neural process, as proposed by Theory 1. Using fMRI, we identified patterns of neural activity in the frontoparietal control network that were more extensively involved in redecision than with initial decisions. Furthermore, activity in the regions of this network was positively scaled with decision uncertainty and became stronger in the condition requiring redecision than that in the condition not requiring redecision after the initial decision. These findings suggest that this network is involved both in metacognitive monitoring of decision uncertainty and metacognitive control of decision adjustment. Taken together, the evidence supports the theoretical proposal that metacognition utilizes a separate neural system to monitor and control decision-making (i.e., Theory 1). We have putatively referred to the network revealed by our experiments as the metacognition network. We further propose that this network could be segregated into three subsystems (as shown in Fig 7).
The subsystem consisting of the dACC and AIC regions was involved in metacognitive monitoring of decision uncertainty, common in the two tasks. The neural uncertainty sensitivity (the uncertainty-level regression β value) in the two regions was highly correlated with the behavioral uncertainty sensitivity. Furthermore, their task baseline activity could predict the individual uncertainty bias. Thus, the decision uncertainty signal could be finally represented by the dACC and AIC activity, which might be the outcome of transforming the uncertainty information from the decision-making process [29]. We thus inferred that uncertainty monitoring might indeed consist of two-order processes. We suggest the possibility that the first-order process coincides with the decision-making process that simultaneously generates the uncertainty information, implicitly associated with decision uncertainty, as proposed by Theory 2, and that the second-order process then transforms this uncertainty information from different decision-making processes into common decision uncertainty scales, which are encoded in the dACC and AIC regions, as activity observed in this study would support. This hypothesis then integrates the two theories together and consistently accounts for observed evidence from both sides. It is worth noting that our results differed from previous neuroanatomical studies showing that the lFPC region was associated with individual behavioral uncertainty sensitivity [9, 18].
The dACC and AIC regions have been well recognized for their involvement in conflict and error monitoring of the preceding cognitive processes to signal the need for more control [30–32]. Our results suggest that it is decision uncertainty, rather than decision error or conflict information, that serves as the primary signal to evoke monitoring [25]. Our findings are also profoundly different from previously reported accounts of dACC function in performance monitoring. First, there was no explicit feedback or cue to indicate whether the decision was correct or incorrect. The participants evaluated decision uncertainty via individual internal signals rather than external cues. Secondly, the task difficulty and RT does not explain the dACC and AIC activity in association with decision uncertainty. The dACC and AIC regions have been shown to broadly monitor subjective feelings such as pain, emotion, and others [33]. Critically, the salient information that elicits conscious monitoring in these regions is not necessarily from the somatosensory stimulation [34]. Similarly, the prospective monitoring of uncertainty in judgments of learning (JOL) and feeling-of-knowing (FOK) has also been shown to activate these regions, prior to execution of the decision-making tasks [35]. Therefore, decision uncertainty monitoring in the dACC and AIC regions should be domain general, independent of the sources of uncertainty information and the forms of decision-making tasks. In short, the individual uncertainty sensitivity is a unique and core trait of each individual decision maker, presumably dependent on the circuit of the dACC and AIC regions [33].
Decision uncertainty monitoring could be a bottom-up process. It occurred automatically without any explicit requirement of redecision (fMRI3, S1G Fig) [25]. However, the subsequent metacognitive control of decision adjustment should require top-down cognitive control. In the Sudoku task, lFPC activity was positively correlated with the extent of decision uncertainty reduction within individual participants and the accuracy change by redecision across the participants, suggesting critical involvement of the lFPC region in metacognitive control. Uncertainty-driven exploration could be a critical process in metacognitive control [25, 36–39]. Revising foregone decisions usually requires an exploration of alternative-solution approaches because the previously used solution approach would likely lead to the same unsatisfactory solution. Through exploration of alternative solutions, a more satisfactory option could be found by which the decision uncertainty would therefore be reduced. During this process of exploration, strategy management could be a key function of lFPC involvement in metacognitive control. This top-down strategic signal might regulate the activity in other frontal cortical areas and posterior parietal cortex, to execute the processes of altering the previous uncertain choice [25, 36, 39] or to explore a non-default option [37, 38].
This would lead to the expectation that the lFPC would not be involved in metacognitive control in the RDM task because revising the preceding perceptual decision might simply require more attention to the stimulus in redecision to continue evidence accumulation, not necessarily exploration. However, lFPC activity remained activated as well and was negatively correlated with the extent of decision uncertainty reduction and accuracy change. It is possible that the process of exploration in the lFPC might be competitive with the simultaneous process of exploitation in the posterior brain areas when these two-level systems are not well coordinated [40, 41]. Indeed, an FPC lesion in nonhuman primates enhanced the animals’ performance of a well-learned decision-making task [42]. However, it remains enigmatic why the lFPC was kept activated when it was not necessary and would not facilitate the engaging task. Presumably, the signal for increased control derived from the dACC region that is sensitive to decision uncertainty might nonselectively activate the lFPC because lFPC activity was also conditioned by decision uncertainty. The automaticity of eliciting lFPC involvement in metacognitive control may facilitate uncertainty resolution in the majority of difficult real-world situations, to relieve effort for engagement in metacognitive control, but failure of disengagement could impair the performance adjustment in simple tasks. Instead, intrinsic motivation might boost metacognitive control of decision adjustment in demanding tasks through VS activity.
Metacognitive control is a form of cognitive control; however, not all forms of cognitive control are metacognitive. Although the Sudoku task and the RDM task appeared very different, to our surprise, the fMRI activation patterns associated with the decision-making process in the initial decisions were quite similar between the two tasks. Critically, the IFJ at the posterior PFC was commonly activated. IFJ is ubiquitously engaged in online task execution, in involving cognitive control [43, 44], and attention [45]. Thus, IFJ might play a critical role in object-level cognitive control, generally in decision-making tasks [25, 46, 47]. The segregation of the meta-level cognitive control in the anterior PFC and the object-level cognitive control in the posterior PFC is aligned with the hypothesis of the rostrocaudal functional division of the PFC in cognitive control [25, 48, 49]. However, the PFC functional division proposed in the current paper is subject to the strategy of task implementation [25] rather than to level of task complexity [48, 49]. The initial decision-making merely recruits the posterior PFC to implement the default strategy of exploiting routine processes, whereas metacognition is evoked when the initial decisions are uncertain, recruiting the frontoparietal control network, including the anterior PFC, to control exploration of alternative processes [50]. Therefore, the metacognition process in redecision is not incorporating prior information acquired in the initial decision, but rather it is prone to altering the initial decision.
There were some potential pitfalls for the fMRI data analyses in the current study. Because the metacognition process should automatically accompany the decision-making process with uncertainty, it excludes the possibility of inserting time jitters between the initial decision phase and the redecision phase, as conventionally used in fMRI paradigms. Thus, the two events of the decision-making process and the metacognition process in the general linear models (GLMs) could be collinear and result in inflation of standard errors of the estimated parameters for the regions involved in both processes. Fortunately, the activation of the regions of interest (ROIs) predominately involved in metacognition appeared in the redecision phase. Of note, the variance inflation factor (VIF) was approximately 2.4, which suggests that the collinearity of the GLMs used in the current study was not severe.
In summary, we have constructed and proposed the extent and generality of the functional architecture of the metacognition neural system, which is separate from the decision-making neural system (Fig 8). The metacognition neural system is composed of the metacognitive monitoring system and the metacognitive control system. The metacognitive monitoring system, consisting of the dACC and AIC regions, is domain general. It reads out the uncertainty information from the decision-making process and quantitatively encodes the decision uncertainty states. The metacognitive control system of the lFPC region implements high-level cognitive control (e.g., strategy), dominant in rule-based and abstract inference tasks (e.g., the Sudoku task), and may compete with low-level cognitive control (e.g., attention), dominant in perceptual tasks (e.g., the RDM task). The high-level cognitive control by the lFPC region could be modulated by intrinsic motivational signals from the VS region. These two subsystems separately monitor and control the decision-making system, in which the IFJ region is critically involved. Thus, the decision-making neural system and the metacognition neural system form a closed-loop system to control and adapt our behavior towards desired goals.
All participants were university students, who were recruited through a campus bulletin board system (BBS). Informed consent was obtained from each individual participant in accordance with a protocol approved by Beijing Normal University Research Ethics Committee.
Twenty-one participants (19–33 y old, 12 female) took part in the main fMRI experiment (fMRI1) and the resting fMRI experiment. Out of them, 16 participants (19–33 y old, 9 female) took part in all sessions of the repeated behavioral experiments. In addition, 17 participants (19–25 y old, 10 female) took part in the second fMRI experiment (fMRI2), 25 participants (19–27 y old, 14 female) took part in the third fMRI experiment (fMRI3), and 20 participants (19–26 y old, 11 female) took part in the fourth fMRI experiment (fMRI4).
In an aperture with a radius of 3 degrees (visual angle), about 300 white dots (radius: 0.08 degrees; density: 2.0%) on a black background that were moving in different directions at a speed of 8.0 degrees/s were displayed. The time span of the movement of each dot lasted for 3 frames. A portion of dots was moving in the same direction (one of the four directions: left, right, up, and down), while the others were moving in different, random directions. The participant was required to discriminate the net motion direction. According to the proportion of coherently moving dots, discrimination difficulty was classified into 10 levels (Fig 2B), for which the movement coherence varied from 1.6% to 51.2%; coherence of moving dots in the control condition was 100%.
To fill a 4 × 4 grid matrix, each digital number from 1 to 4 should be filled in once and only once in each column, each row, and each corner with 4 grids. The task used in the present study required participants to fill in a target grid with a digital number from 1 to 4 in a partially completed Sudoku problem. Each problem had a unique solution. A Sudoku generator (custom codes) was used to create thousands of different Sudoku problems. Problem difficulty was classified into 10 levels according to the minimum number of logic operation steps required to arrive at the solution; this classification scheme largely matched with participants’ reported subjective difficulty level (Fig 2B). In the control condition, the presented problem was made up of symbols (“#”) replacing the digital numbers except for the space in the target grid where the digital number was illustrated. Thus, the participant only needed to press the corresponding button.
Participants were trained in the cognitive skills used to solve the 4 × 4 Sudoku problems under experimenters’ guidance for at least two h per d over a continuous span of 4 d. The participants practiced solving problems with no time constraints first in two to four runs of 40 problems at a set difficulty level. Once the average accuracy of a session crossed 90%, s/he then practiced solving problems at the same level within 2 s. Once the average accuracy of the run in a 2-s time frame reached 70%, the participant repeated these steps at the next level of difficulty. After the 4-d intensive training program, each participant had attained high-level proficiency in solving 4 × 4 Sudoku problems in 2 s, as the mean task difficulty finally approached the fifth level.
The sequences of both the Sudoku and RDM tasks were identical. In fMRI1, each trial started with a green-cross cue to indicate that the task stimulus would be presented 1 s later. The stimulus was presented for 2 s, then four options were presented, and the participant made a choice within 2 s. After a choice was made, four confidence level ratings from 1 (lowest) to 4 (highest) were presented, and the participant reported their confidence rating within 2 s. The same stimulus was immediately presented again for 4 s, and the participant again selected a choice and again reported their confidence rating. Each trial lasted for 15 s. The control trials were intermingled with task trials. The sequence of the control trials was identical to that of the task trials. In each task, there were 4 runs, and each run consisted of 30 task trials and 10 control trials. The task difficulty of each trial was adjusted by a staircase procedure through which one level was upgraded after two consecutive correct trials, was downgraded by one level after two consecutive erroneous trials, and was kept at the same level otherwise, so that the mean accuracy converged at about 50%. Prior to each experiment, two runs were carried out to allow each participant to practice and stabilize performance. The Sudoku problems used in the learning and practice sessions were different from those used in the fMRI and behavioral experiments. In addition, a 10-min resting fMRI experiment was conducted when the participant was in a resting state with eyes opened.
The second fMRI experiment (fMRI2, Fig 4 and S1C Fig) was carried out to examine whether the metacognition network would also be essentially involved in the decision-making process in the initial decision if a new Sudoku problem or new RDM stimulus was presented during the redecision phase, following a control stimulus presentation in the initial decision phase, as used in fMRI1. In fMRI2, a randomized selection of a control stimulus, a new Sudoku problem, or a new RDM stimulus was presented in the redecision phase. The appearance of a new stimulus in the redecision phase occurred in half of the trials. The Sudoku problems or RDM stimuli used in this experiment were selected from those at the middle level of task difficulty. This design was used to reduce the participant’s decision uncertainty in this experiment by controlling the difficulty of the task. In all other ways, the task sequence was the same as that used in fMRI1. In total, there were 120 trials across two runs in each task.
The third fMRI experiment (fMRI3, Fig 3D and S1E Fig) was carried out to compare brain activity in a redecision-task condition (like the task in fMRI1) to activity in a task that did not require redecision. In non-redecision trials, a control stimulus was presented in the redecision phase, therefore no redecision was required. The task sequence was the same as the design used in fMRI1 with the exception that the presentation time of the stimulus was 3 s during the redecision phase. In each task, the ‘redecision’ condition and the ‘non-redecision’ condition were randomized, and each consisted of 60 trials across 3 runs.
The fourth fMRI experiment (fMRI4, S4 Fig) was carried out to confirm that the engagement of metacognition was independent of the duration of redecision. The task sequence was exactly the same as was used in fMRI1, with the exception that the redecision phase was 2 s instead of 4 s. Only the RDM task was used in this experiment. In total, there were 120 task trials and 40 control trials across 4 runs.
In the fMRI experiments, the participants viewed images of the stimuli on a rear-projection screen through a mirror (resolution, 1,024 × 768 pixels; refresh rate, 60 Hz). Normal or corrected-to-normal vision was achieved for each participant. All images were restricted to 3 degrees surrounding the fixation cross.
All fMRI experiments were conducted using a 3 T Siemens Trio MRI system with a 12-channel head coil (Siemens, Germany) after the 4-d Sudoku training. Functional images were acquired with a single-shot gradient echo T2* echo-planar imaging (EPI) sequence with volume repetition time (TR) of 2 s, echo time (TE) of 30 ms, slice thickness of 3.0 mm, and in-plane resolution of 3.0 × 3.0 mm2 (field of view [FOV]: 19.2 × 19.2 cm2; flip angle [FA]: 90 degrees). Thirty-eight axial slices were taken, with interleaved acquisition, parallel to the anterior commissure–posterior commissure (AC–PC) line.
To test the reliability of the participants’ metacognitive abilities, behavioral experiments were carried out using the same paradigms as the Sudoku and RDM tasks. Each of the participants completed 6 sessions of behavioral experiments on consecutive days. Each session was composed of 4 runs of the Sudoku task and 4 runs of the RDM task, the same as was used in fMRI1.
A nonparametric approach was employed to assess each participant’s uncertainty sensitivity. The ROC curve was constructed by characterizing the incorrect probabilities with different uncertainty levels for initial decisions as thresholds. The area under curve (AUC) was calculated to represent how well the participant was at detecting and rating their decision uncertainty [9]. The individual uncertainty bias was estimated by the mean uncertainty level of each session, regressed out the factor of Aroc. The accuracy change was the change in mean accuracy from the first decision to the second decision. The individual uncertainty sensitivity and uncertainty bias, as well as accuracy change, were calculated for each session of the fMRI and behavioral experiments.
The analysis was conducted with FMRIB’s Software Library (FSL) [51]. To correct for the rigid head motion, all EPI images were realigned to the first volume of the first scan. Data sets in which the translation motions were larger than 2.0 mm or the rotation motions were larger than 1.0 degree were discarded. It turned out that no data had to be discarded in the fMRI experiments. The EPI images were first aligned to individual high-resolution structural images and were then transformed to the Montreal Neurological Institute space by using affine registration with 6 degrees of freedom and resampling the data with a resolution of 2 × 2 × 2 mm3. A spatial smoothing with a 4-mm Gaussian kernel (full-width at half-maximum) and a high-pass temporal filtering with a cutoff of 0.005 Hz were applied to all fMRI data.
Each trial in fMRI1 was modeled with three regressors. The first regressor represented the decision-making process in the initial decision, which was time-locked to the onset of the first stimuli presentation, with summation of the presentation time (2 s) and the differential RT from the mean RT of control trials as the event duration. The second regressor represented the neural process following the initial decision, including the metacognition process and the decision-making process in the redecision, and was time-locked to the onset of the first confidence judgment—with summation of the confidence report, the second presentation time (4 s) of the stimuli, and the differential RT from the mean RT of control trials as the event duration. The third regressor represented the baseline during the intertrial intervals (ITIs), time-locked to the onset of ITI, with the ITI duration as the event duration. The uncertainty level of the initial decision, the RT, and the level of uncertainty reduction (differences in the uncertainty level between the final decision and the initial decision) were implemented as modulators of the second regressor by demeaning the variances of the uncertainty level (Fig 3C) and consequently orthogonalizing the RT and the level of uncertainty reduction with each other (Fig 3A–3C and S1I Fig), or reversing the orthogonalization order (S1J Fig). It should be noted that the orthogonalization processes were equal to stepwise regression analyses on these covariates. The same analyses were applied to the fMRI data of fMRI2 and fMRI3.
For group-level analysis, we used FMRIB’s local analysis of mixed effects (FLAME), which model both “fixed effects” of within-participant variance and ‘random effects’ of between-participant variance using Gaussian random-field theory. Statistical parametric maps were generated by a threshold with P < 0.05 with false discovery rate (FDR) correction, unless noted otherwise. The regressions of the individual uncertainty sensitivity (AROC), the individual RT–uncertainty correlation coefficient, the individual mean uncertainty level, and the individual accuracy change with the β weights of uncertainty levels (Figs 5B, 5D and 6C)—or with the task baseline activity (Fig 6C–6E)—were calculated at the third level of group analyses. For these analyses, statistical parametric maps were generated by a threshold of P < 0.001 with the cluster-size threshold as 20 (family-wise error correction).
The ROIs of the metacognition network were defined by the voxels that were significantly activated during the redecision phase in the task trials compared to those during the same phase in the control trials across both tasks using conjunction analysis (P < 0.001, cluster-wise correction; green areas in statistical parametric maps). ROI analyses were obtained from both hemispheres of the same region. The VS ROI was anatomically defined by the striatum atlas of FSL templates [52]. The time courses were derived from the ROIs, calculating a mean time course within an ROI in each participant individually. We then averaged the time courses of the same condition across the participants (S2 and S3 Figs); or, we oversampled the time course by 10 and created epochs from the beginning of an event onward, then applied the corresponding GLM to every pseudo-sampled time point separately. By averaging the β weights across participants, we created the time courses shown in Fig 4. SEMs were calculated between participants.
The PPI analysis (Fig 4D) was conducted with the demeaned VS time courses after removing the mean activity and the component correlated with the uncertainty level as the physiological factor, and the uncertainty level convolved with the canonical hemodynamic response function (HRF) during the redecision phase as the psychological factor. The two factors per se, and the interaction between the two factors as confound regressors, were put together into a new GLM analysis across the whole brain.
Functional connectivity analyses were independently conducted for the task and resting fMRI data. For the task fMRI data in each ROI, the mean activity and the components associated with the uncertainty level, RT, level of uncertainty reduction, and their interactions were regressed out; the residual time courses were then averaged across the voxels of the region and segmented into the individual trials of the task and control conditions in the Sudoku and RDM task, respectively. The segmented data of each trial were then modeled using a single regressor during the redecision phase convolved with the canonical HRF, and then a regression value was obtained for each trial. The correlation coefficient of the regression values between each pair of the ROIs in the metacognition network was calculated across the trials of the task or control condition in each participant. Finally, the averaged correlation coefficients were shown (Fig 7A and 7B). For the resting fMRI data, the standard processing was carried out [53], and the averaged correlation coefficients were shown (Fig 7C).
|
10.1371/journal.pgen.1007353 | Patterning mechanisms diversify neuroepithelial domains in the Drosophila optic placode | The central nervous system develops from monolayered neuroepithelial sheets. In a first step patterning mechanisms subdivide the seemingly uniform epithelia into domains allowing an increase of neuronal diversity in a tightly controlled spatial and temporal manner. In Drosophila, neuroepithelial patterning of the embryonic optic placode gives rise to the larval eye primordium, consisting of two photoreceptor (PR) precursor types (primary and secondary), as well as the optic lobe primordium, which during larval and pupal stages develops into the prominent optic ganglia. Here, we characterize a genetic network that regulates the balance between larval eye and optic lobe precursors, as well as between primary and secondary PR precursors. In a first step the proneural factor Atonal (Ato) specifies larval eye precursors, while the orphan nuclear receptor Tailless (Tll) is crucial for the specification of optic lobe precursors. The Hedgehog and Notch signaling pathways act upstream of Ato and Tll to coordinate neural precursor specification in a timely manner. The correct spatial placement of the boundary between Ato and Tll in turn is required to control the precise number of primary and secondary PR precursors. In a second step, Notch signaling also controls a binary cell fate decision, thus, acts at the top of a cascade of transcription factor interactions to define PR subtype identity. Our model serves as an example of how combinatorial action of cell extrinsic and cell intrinsic factors control neural tissue patterning.
| Genetic mechanisms patterning neuroepithelial domains are a critical first step to diversify cellular identity of the developing nervous system. How many cells develop from distinct neuroepithelial domains depends on overall size and growth but also on positioning of boundaries. Using the embryonic optic placode of the fruit fly Drosophila melanogaster as a model, we identify basic genetic mechanisms of how distinct domains with different fates emerge from an early, seemingly uniform, neurogenic region. We show that the boundary between two transcription factors is critical to determine how many cells are incorporated in either domain. This is achieved by coordinated interaction of Hedgehog and Notch signaling, which control proliferation and regulate domain-specific transcription factors. The mechanisms employed here in an epithelial placode to determine photoreceptor precursors display similarities with the ones previously identified in the adult compound eye, further supporting the notion of a common developmental program for the larval eye and adult compound eye.
| In the fruit fly Drosophila melanogaster, all parts of the visual system develop from an optic placode, which forms in the dorsolateral region of the embryonic head ectoderm [1–3]. During embryogenesis, neuroepithelial cells of the optic placode are patterned to form two subdomains. The ventroposterior domain gives rise to the primordium of the larval eye and consists of two photoreceptor (PR) precursor types (primary and secondary precursors), whereas the dorsal domain harbors neuroepithelial precursors that generate the optic lobe of the adult visual system [4–6]. The basic helix-loop-helix transcription factor Atonal (Ato) promotes PR precursor cell fate in the larval eye primordium [4,7]. The orphan nuclear receptor Tailless (Tll) is confined to the optic lobe primordium and maintains non-PR cell fate [4]. Hedgehog (Hh) and Notch (N) signaling are critical during the early phase of optic lobe patterning. The secreted Hh protein is required for the specification of various neuronal and non-neuronal cell types, while Notch acts as neurogenic factor preventing ectodermal cells from becoming neuronal precursors by a process termed lateral inhibition [8,9]. In the optic placode Ato expression is promoted by Hh and the retinal determination genes sine oculis (so) and eyes absent (eya). Notch delimits the number of PR precursors and maintains a pool of non-PR precursors [10]. Ato is initially expressed in all PR precursors in the placode and its expression gets progressively restricted to primary precursors [7,11]. In a second step, primary precursors recruit secondary precursors via EGFR signaling: primary precursors express the EGFR ligand Spitz, which is required in secondary precursors to promote their survival. After this initial specification of primary and secondary PR precursors, the transcription factors Senseless (Sens), Spalt (Sal), Seven-up (Svp) and Orthodenticle (Otd) coordinate PR subtype specification. Sens and Spalt are expressed in primary PR precursors, while Svp contributes to the differentiation of secondary PR precursors [12,13]. By the end of embryogenesis, primary PR precursors have fully differentiated into blue-tuned Rhodopsin5 PRs (Rh5), while secondary PR precursors have differentiated into green-tuned Rhodopsin6 PRs (Rh6) [12,13]. While the functional genetic interactions of transcription factors controlling PR subtype specification has been thoroughly studied, it remains unknown how the placode is initially patterned by the interplay of Hh and Notch signaling pathways. Similarly, the mechanisms of how ato and tll-expressing domains are set up to ensure the correct number of primary and secondary PR precursors as well as non-PR precursors of the optic lobe primordium remain unknown.
Here we describe the genetic mechanism of neuroepithelial patterning and acquisition of PR versus non-PR cell fate in the embryonic optic placode and provide the link to subsequent PR subtype identity specification. The non-overlapping expression patterns of ato and tll in the optic placode specifically mark domains giving rise to the larval eye precursors (marked by Ato) and the optic lobe primordium (marked by Tll). ato expression in the larval eye primordium is temporally dynamic and can be subdivided into an early ato expression domain, including all presumptive PR precursors and a late ato domain, restricted to presumptive primary PR precursors. The ato expression domain directly forms a boundary adjacent to tll expressing precursors of the optic lobe primordium. We show that tll is both necessary and sufficient to delimit primary PR precursors by regulating ato expression. Hh signaling regulates the cell number in the optic placode and controls PR subtype specification in an ato- and sens-dependent manner. Finally, we also show that Notch has two temporally distinct roles in larval eye development. Initially, Notch represses ato expression by promoting tll expression and later, Notch controls the binary cell fate decision of primary versus secondary PR precursors by repressing sens expression. In summary, we identify a network of genetic interactions between cell-intrinsic and cell-extrinsic developmental cues patterning neuroepithelial cells of the optic placode and ensuring the timely specification of neuronal subtypes during development.
During embryonic development, the Drosophila optic placode produces both the larval eye PRs and the precursors of the optic lobe [14]. To document how the boundary between these two groups of cells is established, we mapped the expression patterns of a subset of proteins that are expressed in different subregions within the optic placode.
The optic placode is first detected on the surface of embryos at stage 10, located in the posterior procephalic region. During stage 10, the transcription co-activator Eya starts being expressed in a crescent-shaped domain, overlapping with the ventral-most region of the optic placode (Fig 1A and 1B) [11]. At this stage, virtually all Eya-positive cells within the optic placode display co-staining with an antibody against phospho-Histone H3 (pH3), a mitotic marker, indicating that these cells divide during stage 10 (S1 Fig). At stage 11, a boundary appears within the optic placode that subdivides the Eya-expressing domain. Cells that are located anteriorly start expressing the nuclear receptor Tll, and cells located posteriorly start expressing the basic helix-loop-helix transcription factor Ato. Ato is required for the development of the Bolwig's organ, whereas Tll is important for the formation of the optic lobe. Tll expression is maintained later in development, during stages 12 and 13, whereas Ato expression is lost progressively during these stages (Fig 1C–1F). Both Eya and Sine oculis (So), which are components of the retinal determination network and, therefore, important for adult eye development [15], are co-expressed in the optic placode (Fig 1G and 1H).
It has been suggested that the pool of late Ato-expressing cells will give rise to the primary PR precursors [11,16]. Tll represses Ato in the optic placode, and the number of Bolwig's organ PRs is increased in tll mutants. To see if Ato is responsible for repressing tll, we looked at the expression of the Tll::GFP reporter in ato mutants and found no spreading of Tll::GFP expression into the posterior Eya-positive cells (Fig 1I). Hence, Ato is not required to restrict tll expression.
As the expression and function of Ato and Tll may be linked, we next focused on the role of Tll in larval optic placode patterning. tll is specifically expressed in non-PR precursors in the optic placode [4]. It was previously proposed that Tll counteracts the EGF signal and prevents placode cells from developing as secondary PR precursors [4]. To understand how tll regulates subtype specification of larval PR precursors in the optic placode, we first analyzed ato expression in tll mutants. We found that in tll mutants the optic placode possesses about twice as many Eya-positive cells as compared to control embryos, and this increase is paralleled by an increase in the number of Ato-positive cells (Fig 2A and 2B). Interestingly, in tll mutants the expression of Ato is not limited to the postero-ventral corner but rather, ato expression is expanded and occupies the whole posterior margin of the Eya-positive domain of the optic placode (Fig 2B). Later in development, tll mutants form about twice as many PR precursors as control embryos (an average of 24 cells), which is proportional to the increase in placode size. These ectopic PRs express the transcription factor Hazy (S2 Fig), a transcription factor that activates the expression of phototransduction proteins [17, 18], indicating that these cells are correctly specified. We further analyzed the determination of PR-identity by assessing the PR subtype specific markers Spalt (Sal) and Seven-up (Svp). We found an increase in total PR cell number including Sal-expressing primary PR precursors as well as Svp-expressing secondary PR precursors in tll mutants (Fig 2C–2G). However, the relative fraction of Sal- or Svp-positive PRs remained unchanged. Thus, the ratio between both PR subtypes in tll mutants was comparable to wildtype control (Fig 2H).
To determine if Tll is also sufficient to genetically repress the PR precursor cell fate, we ectopically expressed Tll under the control of so-Gal4. Ectopic expression of Tll does not change the total number of cells in the optic placode, but it does reduce the total number of PRs in the developing larval eye. However, the ratio of Sal-positive and Svp-positive cells remains comparable to that of control embryos (Figs 3A–3G and S3).
These findings suggest that in tll mutants, Ato expression expands, and thus, increases the pool of primary PR precursors. As a result, more secondary PR precursors are incorporated into the larval eye (Fig 3E). However, the finding that the ratio of primary and secondary PR precursors remains the same as in wildtype (Fig 2H) suggests that the recruitment of secondary PR precursors by primary PR precursors in tll mutants occurs in a normal manner. Taken together, our findings support that Tll acts as a critical cell fate determinant by repressing larval eye precursors (Fig 3E).
Hedgehog (Hh) signaling is required for the formation of photoreceptor precursors in the Drosophila compound eye, in the ocelli, and in the larval eye. Hh regulates the onset of PR formation in an Ato-dependent manner [6,19, 20]. The canonical Hh signaling pathway includes two transmembrane proteins Patched (Ptc) and Smoothened (Smo) [21]. In the absence of Hh, Ptc represses Smo preventing signal transduction. However, binding of Hh to Ptc eliminates Ptc-dependent repression of Smo activating a downstream signaling cascade [22]. Therefore, ptc and smo mutant phenotypes correlate with Hh gain- and loss-of-function mutants, respectively. Loss of Hh results in a complete loss of ato expression and thus PR precursor formation, while increased Hh signaling in ptc mutants shows an increase in PR cell numbers [6]. Importantly, in ptc mutants, these additional PRs express the PR marker Hazy, and, by the end of their development, at stage 17, they further terminally differentiate correctly to express Rh5 or Rh6 (S2 and S4 Figs).
Since it remains unknown how Hh signaling affects PR subtype identity, we analyzed the expression of the early PR precursor markers Ato and Sens in ptc mutants (Fig 4A, 4B, 4C and 4D). In agreement with previous observations, we found an increased number of Ato-expressing cells (Fig 4A and 4B) [6]. As expected from the expansion of Ato expression, we also observed an increased number of Sens expressing cells in ptc mutants, suggesting an increase of primary PR precursors (Fig 4C and 4D) [6,20]. By comparing the expression of the primary PR precursor marker Sal and the secondary PR precursor marker Svp in ptc mutants, we observed a significant increase of both precursor types (Figs 4E–4I and S4). The increase of PR precursors in ptc mutants is presumably due to an increase in cell number within the entire optic placode at stage 11. As a consequence more mature larval PRs were observed at the end of embryogenesis (Figs 4I, 4J and S3). Intriguingly, the ratio of Sal- and Svp-positive PRs in ptc mutants remained the same as in wildtype (Fig 4J).
Since Ato-dependent primary PR precursor specification is regulated both by Tll and Hh, we investigated whether Hh regulates Tll to control the formation of primary PR precursors. For this, we analyzed tll expression in the optic placode of ptc mutants and found that it is comparable to control embryos. tll is neither expressed in wildtype nor in ptc mutant PR precursors (Fig 5A and 5B), while it is expressed in a subset of wildtype or ptc mutant optic lobe precursors (Fig 5A’ and 5B’). Therefore, Tll and Hh might act in parallel to regulate the specification of primary PR precursors in the embryo.
Since Hh signaling controls cellular growth by regulating cell cycle specific genes during Drosophila compound eye development [23], we next investigated if the supernumerary PR precursors found in ptc mutants are a result of a general increase in cell number of the optic placode. To label proliferating cells in the optic placode, we marked mitotically active cells with the phospho-Histone H3 (pH3) antibody (Fig 5C and 5D) [24, 25]. However, we found no statistically significant change in the numbers of pH3-positive cells between wildtype and ptc mutants at stage 11 (Fig 5E).
Although the number of pH3-positive cells was not significantly increased at stage 11, the total number of cells in the optic placode of ptc mutants is significantly higher than in wildtype (Figs 4B, 4I and S3). These findings suggest that Hh signaling is required to control cell number in the optic placode and the larval eye probably by controlling proliferation or apoptosis (Fig 5F). Comparable to tll mutants, the ratio of primary to secondary PR precursors remains unchanged suggesting that the recruitment process of secondary PR precursors via EGFR signaling remains unaffected.
During optic placode development Notch has been implicated in maintaining neuroepithelial identity by suppressing the PR cell fate [10,26, 27]. Since the Notch mutant phenotype shows an increase in larval PR precursors [10], we first investigated how PR subtype specification is affected. When analyzing the expression of the Notch activity reporter E(spl)mγ-HLH::GFP in the optic placode we observed that at early stage 11, GFP expression was broadly distributed in the Eya-positive domain of the optic placode (Fig 6A and 6B). However, later, E(spl)mγ-HLH::GFP expression was excluded from the larval eye primordium, marked by Ato (Fig 6C–6E). To address how Notch may be involved in placode patterning, we next analyzed Ato expression in Notch loss-of-function mutants. We observed an expansion of the Ato-expressing domain as compared to wildtype embryos (Fig 6F and 6G). This increase in Ato-expressing cells, which give rise to PR precursors, indeed correlates with the supernumerary PRs found during later embryonic stages. Moreover, these supernumerary PRs expressed the transcription factors Hazy and Kr [17, 18], indicating that these cells are correctly specified (S2 Fig). Interestingly, at stage 11 the Notch mutant optic placode has as many cells as that of wildtype (an average of 46 cells), but it produces about twice the number of PRs (an average of 33) (Figs 6K, 6L and S3). To investigate PR subtype identity we analyzed Sal and Svp expression in Notch mutants. In agreement with previous observations we found a large increase in the total number of larval PR precursors (Fig 6H and 6I) [10]. This increase is based on the expansion in the number of Sal-positive cells (Fig 6H' and 6I') whereas no or a very reduced number of Svp-expressing cells was found in Notch mutants (Fig 6H'' and 6I''). This finding suggests that during PR development Notch regulates binary cell fate decision to promote secondary PR precursor identity while repressing primary PR precursor cell fate. Quantification of the number of Sal- and Svp-positive PRs in the Notch mutants also shows an increase of primary PR precursor and a decrease in secondary PR precursor cell numbers (Fig 6K and 6L).
Since Notch regulates larval PR subtype specification, we next investigated whether Notch is also sufficient to repress primary PR precursor cell fate in the embryo. We therefore ectopically expressed the intracellular domain of Notch (Nintra; active form), derived by proteolytic cleavage of Notch protein [28], under the control of so-Gal4. We indeed found that ectopic expression of Nintra leads to a complete loss of Sal expression while all cells express the secondary PR precursor marker Svp (Fig 6J).
We have previously shown that Sens is an additional marker for primary PR precursors [13]. We next investigated whether Notch regulates Sens activity and thereby controls PR subtype specification in the embryo. For this, we overexpressed a dominant-negative form of Kuzbanian (Kuz) (Kuz encodes a putative zinc metalloprotease responsible for proteolytic cleavage and release of the active form of Notch) [29] in the optic placode by using so-Gal4. Overexpression of kuzDN results in an increase of Sens-positive cells (Fig 7C, 7C’, 7D and 7D’), suggesting that Notch acts upstream of Sens and regulates PR subtype specification by repressing Sens activity. To further support this notion, we next overexpressed the active form of Notch (Nintra) in the optic placode by using so-Gal4. We indeed found that embryos having ectopic Nintra show a complete loss of Sens expression (Fig 7E and 7E’).
Since both Tll activity and Notch signaling regulate primary PR precursor specification, we next investigated whether Notch is required for tll expression. Indeed, we found that tll expression is completely lost in the optic placode in Notch mutant embryos (Fig 7A and 7B), suggesting that Notch acts genetically upstream of tll. We therefore attempted to rescue the Notch mutant phenotype by inducing expression of UAS-tll with so-Gal4 while concomitantly blocking Notch signaling using UAS-kuzDN. Interestingly, in so-Gal4/UAS-kuzDN,UAS-tll the larval eye does not develop suggesting that pan-placodal expression of Tll is sufficient to repress photoreceptor precursor induction and further supporting the notion that Tll is acting downstream of Notch signaling (S5 Fig). In order to test a possible genetic interaction between the Notch and Hh signaling pathways we analyzed the expression of a ptc-lacZ reporter in the optic placode of Notch mutants. In this case, we still detected β-Gal expression in the respective neureopithelial domain, suggesting that these pathways may act in parallel (S6 Fig).
Our findings support a model in which Notch has two temporally distinct roles during larval eye formation. First, Notch signaling maintains neuroepithelial cells of the presumptive optic lobe by repressing ato expression in a Tll-dependent fashion (Fig 7F). Second, Notch represses the Sens mediated binary cell fate decision in the presumptive larval eye where it promotes secondary PR cell fate specification (Fig 7F).
Neurogenic placodes are transient structures that are formed by epithelial thickenings of the embryonic ectoderm and give rise to most neurons and other components of the sensory nervous system [30–34]. In vertebrates, cranial placodes form essential components of the sensory organs and generate neuronal diversity in the peripheral nervous system [35–37]. How neuronal diversity is generated varies from system to system, and different gene regulatory networks have been proposed for each particular type of neuron [17, 18, 38–40]. Interestingly, some transcription factors, like Atonal, play an evolutionary conserved role during neurogenesis both in Drosophila and in vertebrates [41].
Neuroepithelial patterning of the Drosophila optic placode exhibits unique segregation of larval eye and optic lobe precursors during embryogenesis [4,6,20]. We identified genetic mechanisms that control early and late steps in specifying PR versus non-PR cell fate that ensure the expression of precursor cell fate determinants. During germband extension at stage 10, transcriptional regulators (so, eya, ato and tll) show complex and partially overlapping expression patterns in the optic placode. Their interactions with the Notch and Hh signaling pathways define distinct PR and non-PR domains of the larval eye and optic lobe primordium. Intriguingly, our results show a spatial organization of distinct precursor domains, supporting a new model of how the subdivision of precursor domains emerges. In agreement with previous studies initially the entire posterior ventral tip expresses Ato, defining the population of cells that give rise to PR precursors, while neuroepithelial precursors for the presumptive optic lobe are defined by Tll-expression in the anterior domain of the optic placode [4]. Subsequently, Ato expression ceases in the ventral most cells and thus gets restricted to about four primary PR precursors that are located directly adjacent to the Tll expression domain. Hence, a few cell rows are between the primary PR precursors and the ventral most edge of the optic placode. This is in agreement with a recent observation on the transcriptional regulation of ato during larval eye formation [42]. Thus, primary PR precursors are directly adjacent to the Tll-expressing cells while the Ato and Tll negative domain of secondary PR precursors is located at the posterior ventral most tip of the optic placode. Setting the Tll-Ato boundary is critical to define the number of putative secondary PR precursors, which can be recruited into the larval eye, probably via EGFR signaling [43]. We propose a model during which coordinated action of Hh, Notch and Tll restricts the initially broad expression of Ato to primary PR precursors (Fig 8). Lack of Tll results in a de-repression of Ato and results in an increased number of primary PR precursors, which in turn recruit secondary PR precursors. Interestingly, while tll mutants show an increase in both primary and secondary PR precursors, the ratio between both subtypes is maintained. This notion further displays similarities of ommatidal formation in the adult eye-antennal imaginal disc, where Ato expressing R8-precursors recruit R1-R6. In the eye-antennal disc, specification of R8-precursors determines the total number of ommatida and therefore also the total number of PRs, the ratio of R8 to outer PRs however always remains the same. Thus, the initial specification of primary PR precursors defines the total number of PRs in the larval eye similarly to R8 PRs, and the ratio of founder versus recruited cells remains constant. Interestingly, the maintenance of primary versus secondary PR precursor ratio is also maintained in ptc mutants further supporting this model.
During photoreceptor development in the eye-antennal imaginal disc hh is expressed in the posterior margin and is required for the initiation and progression of the morphogenetic furrow as well as the regulation of ato expression [19, 44, 45]. During embryogenesis the loss of hh results in a complete loss of the larval eye, while increasing Hh signaling (by means of mutating ptc) generates supernumerary PRs in the larval eye [6,20]. During early stages, we found an increase of Ato expression in ptc mutants suggesting that similarly to the eye-antennal disc Hh positively regulates ato expression. The observed increase of Ato-expressing cells is not due to a reduction of Tll but is likely due to increased cell proliferation in ptc mutants. Hh also controls proliferation during the formation of the Drosophila compound eye [46, 47].
During embryonic nervous system development Notch dependent lateral inhibition selects individual neuroectodermal cells to become neuroblasts. Notch represses neuroblast cell fate and promotes ectodermal cell fate [48–51]. During compound eye development, Notch regulates Ato expression and acts through lateral inhibition to select Ato expressing R8 PR precursors [52]. Similarly, during Drosophila larval eye development, Notch is required for regulating PR cell number by maintaining epithelial cell fate of the optic lobe primordium [20]. Inhibiting Notch signaling leads to a complete transformation of the optic placode to PRs of the larval eye [20]. In the absence of Notch signaling, Ato expression is expanded in the optic placode and as a result the total number of PRs is increased. Despite the increase of the overall PR-number the number of secondary PR precursors is significantly decreased or lost in the absence of Notch activity. In the compound eye Notch promotes R7 cell fate by repressing the R8-specific transcription factor Sens [53]. It was also proposed that genetic interaction between Notch and Sens is required for sensory organ precursor (SOP) selection in the proneural field in a spatio-temporal manner [54]. We found that during PR subtype specification Notch represses Sens expression, thereby controlling the binary cell fate decision of primary versus secondary PR precursors. Therefore, in the absence of Notch signaling, Sens expression represses the secondary PR precursor fate. As a result, all PR precursors are transformed and acquire primary PR precursor identity. In conclusion, we observed that Notch is essential for two aspects during optic placode patterning. First, Notch activity is critical for balancing neuroepithelial versus PR cell fate mediated through Tll-regulated Ato expression. Second, Notch regulates the binary cell fate decision of primary versus secondary PR precursor cell fate through the regulation of Sens expression.
Flies were grown on standard food medium at 25°C and 12hr/12hr light dark cycle. Wildtype Canton S was used as controls in all cases. The following mutants and transgenic fly lines were used: tll49l [55], ptc9 [56] and N55e11 [57] mutant lines were used to analyze Tll, Hh and Notch function, respectively. tll::GFP [58] and E(spl)mγ-HLH::GFP transgenic lines were used to analyze Tll expression and Notch activity pattern. UAS-tll (gifts from M. Kurusu) [59], UAS-Nintra [60] and UAS-kuzDN [61] transgenic lines were used to overexpress Tll, the Notch intracellular domain and the Kuzbanian dominant negative form. so-Gal4 [62] was used to overexpress genes early in the optic placode and UAS-mCD8::GFP [63] was used to mark Gal4 expression pattern in the embryo. We also used atow [64] (kindly provided by B. Hassan) and ptc-lacZ (gift of K. Basler).
For So antibody production, the complete so coding sequence from the cDNA clone FI01103 was PCR amplified and cloned into pGEX-6P-1 in frame with the GST-Tag. So protein was expressed in BL-21 cells and purified with Glutathione-sepharose beads (GE Healthcare) cleaving it from the GST-tag at the HRV 3C site (THERMO Scientific). The purified protein was sent to Eurogentec for immunization of two guinea pigs.
For immunohistochemistry, Drosophila embryos were dechorionated, fixed and immunostained as described in [13]. Briefly, embryos were collected after keeping flies on apple juice plates for overnight at 25°C. Embryos were dechorionated by using 50% bleach solution for 7 minutes. Embryos were fixed for 25–30 minutes by using equal volumes of n-heptane and 4% formaldehyde (made in 1XPBS; Phosphate buffered saline) solutions. They were devitellinized and stored in methanol at -20°C. Immunostaining of embryos were initiated by washing them in 1xPBS containing 0.3% Triton X-100 (PBST) three times for 30 minutes each. Primary antibody dilutions were made in PBST and embryos were incubated in the primary antibody for overnight at 4°C. On the next day, three washes were performed using PBST for 30 minutes each. Secondary antibodies were diluted in PBST and embryos were incubated with the secondary antibody solution for two hours at room temperature or overnight at 4°C. Washing was performed again and samples were mounted by using Vectashield H-1000 (Vector laboratories). The following primary antibodies were used: mouse anti-Eya and mouse anti-FasII (both at 1:20; Developmental Studies Hybridoma Bank), guinea pig anti-So (1:100; this work), rabbit anti Hazy (1:500) [65] rabbit anti-Ato (1:5000) [66], chicken anti-GFP (1:2000; Life technologies), guinea pig anti-Kr (1:200) and guinea pig anti-Tll (1:100; gifts from J. Jaeger), rabbit anti-Rhodopsin 6 (1:10000) [67], rabbit anti-Rhodopsin 5 (1:40) [68], rabbit anti-Sal (1:300) [69], mouse anti-Svp (1:100) [70], guinea pig anti-Sens (1:800) [71], rabbit anti-Hazy (1:500) [65] and rabbit anti-pH3 (1:200) (Upstate Biotechnology). The following secondary antibodies were used: Alexa488, Alexa568 and Alexa647 (1:200; Molecular Probes).
For cell quantifications in wildtype and mutants, we calculated the mean for each category and determined the standard deviation. Circles represent numbers or percentages of individual samples. Standard functions in MATLAB (MathWorks) (“ttest2”) and in RStudio (RStudio, Inc.) (“aov” and “glht(multcomp)”) were applied for statistical analysis.
A one-way ANOVA followed by Dunnett’s test was applied to test the number of all PRs, the number of Sal-positive PRs, the number of Svp-positive PRs and the number of all optic placode cells of the different genotypes respectively against the particular numbers of the wildtype control.
A one-way ANOVA followed by Dunnett’s test was performed to test the percentage of Sal-positive PRs and the percentage of Svp-positive PRs of the different genotypes respectively against the specific values of the wildtype control.
A two-tailed two sample t test was applied to test the number of pH3-positive cells of wildtype control and ptc9 mutants against each other.
Rejection of the null hypothesis that the numbers or ratios of cells of data sets are the same: (***) means p-value<0.001, (**) means p-value<0.01 and (*) means p-value<0.05.
Confocal stacks were collected by using a 40x (NA 1.3) oil or a 63x (NA 1.3) glycerol immersion objective on a TCS Leica SP5 confocal microscope. Acquired image resolution was 1024x1024 pixels and optical sections were 1 to 1.5μm. Fiji/ImageJ was used for image analysis and Adobe Photoshop CS6 software was used for brightness/contrast adjustment and background subtraction. Figures were assembled in Adobe Photoshop CS6.
|
10.1371/journal.pntd.0006815 | Intrinsic activation of the vitamin D antimicrobial pathway by M. leprae infection is inhibited by type I IFN | Following infection, virulent mycobacteria persist and grow within the macrophage, suggesting that the intrinsic activation of an innate antimicrobial response is subverted by the intracellular pathogen. For Mycobacterium leprae, the intracellular bacterium that causes leprosy, the addition of exogenous innate or adaptive immune ligands to the infected monocytes/macrophages was required to detect a vitamin D-dependent antimicrobial activity. We investigated whether there is an intrinsic immune response to M. leprae in macrophages that is inhibited by the pathogen. Upon infection of monocytes with M. leprae, there was no upregulation of CYP27B1 nor its enzymatic activity converting the inactive prohormone form of vitamin D (25-hydroxyvitamin D) to the bioactive form (1,25α-dihydroxyvitamin D). Given that M. leprae-induced type I interferon (IFN) inhibited monocyte activation, we blocked the type I IFN receptor (IFNAR), revealing the intrinsic capacity of monocytes to recognize M. leprae and upregulate CYP27B1. Consistent with these in vitro studies, an inverse relationship between expression of CYP27B1 vs. type I IFN downstream gene OAS1 was detected in leprosy patient lesions, leading us to study cytokine-derived macrophages (MΦ) to model cellular responses at the site of disease. Infection of IL-15-derived MΦ, similar to MΦ in lesions from the self-limited form of leprosy, with M. leprae did not inhibit induction of the vitamin D antimicrobial pathway. In contrast, infection of IL-10-derived MΦ, similar to MΦ in lesions from patients with the progressive form of leprosy, resulted in induction of type I IFN and suppression of the vitamin D directed pathway. Importantly, blockade of the type I IFN response in infected IL-10 MΦ decreased M. leprae viability. These results indicate that M. leprae evades the intrinsic capacity of human monocytes/MΦ to activate the vitamin D-mediated antimicrobial pathway via the induction of type I IFN.
| Our macrophages are equipped with the ability to detect and kill invading pathogens, and yet, these cells of the innate immune system are still subject to infection by intracellular bacterium. In particular, mycobacterium, the type of intracellular bacteria responsible for diseases such as tuberculosis and leprosy, are very successful at establishing infection within macrophages. By studying Mycobacterium leprae, the etiological agent of leprosy, we describe an immune evasion mechanism whereby this bacterial pathogen utilizes our own antiviral immune response against the macrophage. Type I interferons (IFN) are a major part of our immune response to viral infections; however, this response will also suppress our ability to fight opportunistic bacterial infection. During infection of our macrophages, M. leprae induces an aberrant type I IFN response that subsequently suppresses our macrophage’s ability to activate the vitamin D-mediated antimicrobial pathway, a critical antimicrobial response for containment of mycobacterium. Thus, understanding how these pathogens can evade our immune response will be important for the development of new therapies against these chronic infections.
| The ability of macrophages (MΦ) to kill intracellular pathogens is critical to the outcome of infection. Addition of exogenous ligands derived from the pathogen such as a Toll-like receptor 2 ligand (TLR2L) or from human immune cells, such as IFN-γ, provides an extrinsic signal to activate a vitamin D-dependent antimicrobial activity against mycobacteria in infected MΦ [1, 2]. In the absence of these exogenous stimuli, there is evidence for intrinsic antimicrobial mechanisms, although these are not clearly defined nor sustained [3]. In this regard, microbial pathogens have evolved mechanisms to evade these host defense pathways, establishing infection that results in clinical disease. Here we investigated whether in the absence of exogenous triggers such as TLR2L or IFN-γ, there is intrinsic activation of a vitamin D-dependent antimicrobial pathway in infected human MΦ, as well as the presence of bacterial mechanisms of escape, by studying leprosy as a disease model.
Leprosy, caused by the intracellular bacterium Mycobacterium leprae, presents as a spectrum of disease in which the clinical manifestation correlates with the immunological state of the patient. At one end is the self-limiting tuberculoid form (T-lep) and the other end is the disseminated lepromatous form (L-lep) [4], thus comparisons of each form of leprosy affords the opportunity to uncover the critical intrinsic immune mechanisms needed to contain the infection. Each pole of the leprosy disease spectrum is accompanied by a distinct immunological profile defined by specific T cell subsets, MΦ populations and cytokine patterns [4]. Gene expression profiling of skin lesions from patients with the different forms of leprosy suggested a correlation between activation of the vitamin D pathway and favorable disease outcomes [5].
For over a century, the potential use of vitamin D as a treatment against pathogenic infections has been investigated [6–11]; however, only in recent years has the mechanisms by which the immune system utilizes vitamin D to mount an antimicrobial response been described. Human monocytes and MΦ can synthesize the active vitamin D hormone 1α,25-dihydroxyvitamin D (1,25D) from the inactive prohormone substrate 25-hydroxyvitamin D (25D) upon stimulation via innate or adaptive immune signals resulting in an antimicrobial response to infection [1, 2, 12]. Conversion of 25D into 1,25D is mediated through the enzymatic actions of 25-hydroxyvitamin D 1α-hydroxylase (CYP27B1). Studies performed in vitro with human MΦ have demonstrated that this vitamin D metabolic system plays a role in human MΦ antimicrobial activity against M. tuberculosis infection as well as in patients with tuberculosis [1, 2, 5, 12, 13].
As the role of vitamin D in the human immune response against M. tuberculosis infection becomes increasingly established, studies have also emerged linking these mechanistic findings to other human mycobacterial diseases, in particular, leprosy [5, 14–17]. Expression of the vitamin D pathway during disease may be an important facet to disease pathogenesis; therefore, it is important to understand the factors that regulate the expression and function of this antimicrobial pathway. By comparing the gene expression profiles of lesions derived from T-lep vs. L-lep patients, we were previously able to determine that M. leprae infection of human MΦ resulted in the induction of type I IFN, which directly inhibited activation of the vitamin D-dependent antimicrobial response triggered by exogenous addition of type II IFN (IFN-γ) [17]. Here, we investigated whether the vitamin D antimicrobial pathway is intrinsically activated, in the absence of TLR2 and IFN-γ by M. leprae infection of MΦ, Furthermore, we evaluated whether the ability of M. leprae to induce type I IFN blocks the intrinsic activation of the vitamin D pathway, representing an escape mechanism by which the bacterium evades the host response.
This study was conducted according to the principles expressed in the Declaration of Helsinki, and was approved by the Institutional Review Board of the University of California at Los Angeles. All donors provided written informed consent for the collection of peripheral blood and subsequent analysis.
Experiments comparing two conditions were analyzed using two tailed Student’s t-test. Multiple condition experiments were analyzed using one-way ANOVA with pairwise analysis using Newman-Keuls test. Other tests used are indicated in text. All data used to generate the figures presented are available as S1 Data.
Skin biopsy specimens were collected from untreated patients at the Hansen’s Disease Clinic at Los Angeles Country and University of Southern California Medical Center as well as the Leprosy Clinic at the Oswaldo Cruz Foundation in Brazil. The diagnosis and classification of patients were determined based on clinical and histopathological criteria of Ridley and Jopling [4, 18]. Cryosections of skin lesions from T-lep, and L-lep patients were labeled for CYP27b1 (Santa Cruz Biotechnology, Dallas, TX). Sections were incubated with normal horse serum followed by incubation with CYP27b1 antibody or matched isotype control for 60 minutes. Following primary antibody incubation, sections were incubated with biotinylated horse anti-mouse IgG and visualized by ABC Elite system (Vector Laboratories) and AEC Peroxidase Substrate Kit (Vector Laboratories). The section was then counterstained with hematoxylin and mounted with crystal mounting medium (Biomeda). Skin sections were examined using a Leica microscope (Leica 2500). Expression of CYP27b1 in photomicrographs was quantified using ImmunoRatio (http://153.1.200.58:8080/immunoratio/), an automated image analysis application that calculates the percent diaminobenzidine (DAB)-stained nuclear area per total area [17].
Peripheral blood mononuclear cells (PBMC) were isolated using Ficoll (GE Healthcare, Chicago, IL) density gradient from whole blood obtained from healthy donors with informed consent. Monocytes were enriched using plastic adherence in which PBMCs were cultured for two hours in RPMI 1640 media (Invitrogen, Carlsbad, CA) with 1% FBS (Omega Scientific, Tarzana, CA) followed by washing [2, 12, 14]. Monocytes were cultured in RPMI media with 10% FBS and treated with 100ng/mL Pam3CSK4 (TLR2L), or M. leprae (MOI 10, gift from J.L. Krahenbuhl) immediately following adherence. For blocking type I IFN signaling, monocytes or macrophages were treated with 10μg/mL anti-IFNAR antibody (PBL, Piscataway, NJ).
Following treatments as indicated in the text, monocytes or MΦ were incubated with radiolabeled 3[H]-25D3 (PerkinElmer, Waltham, MA) substrate for 5 hours in serum-free RPMI media. [3H]-metabolites were purified using a C18 column and separated using a Zorbax-sil column (Agilent, Santa Clara, CA). Radioactivity was measured in each sample by scintillation counting. The amount of each metabolite present was quantified from counts per minute plotted against elution profiles of standards for 25D3, 1,25D3, and 24,25D3 [12].
Total RNA was harvested using TRIzol (Thermo Fisher Scientific, Waltham, MA) and cDNA synthesized using SuperScript III Reverse Transcriptase (Thermo Fisher Scientific). mRNA expression levels for CYP27B1, CYP24A1, VDR, CAMP, and IFNB were assessed using the Taqman system and pre-verified primer and probe sets (Thermo Fisher Scientific). Relative expression was quantified by comparing to the housekeeping gene 18S rRNA and using the ΔΔCt method [12, 14]. OAS1 mRNA levels were assessed using the SYBR Green system (Bio-Rad Laboratories, Irvine, CA) compared to the housekeeping gene 36B4 using the ΔΔCt method. 36B4 primers were previously published [14], and OAS1 primers are as follows: forward 5’-GAG ACC CAA AGG GTT GGA GG-3’, and reverse 5’- GGA TCG TCG GTC TCA TCG TC-3’.
Secreted IFN-β protein in the supernatant were measured using VeriKine Human Interferon-Beta ELISA Kit (PBL Interferon Source) following the manufacturer’s protocols.
The microarray experiment was previously published [17]. Gene expression profiles of mRNAs derived from skin biopsy specimens of 16 leprosy patients (T-lep, n = 10; L-lep, n = 6) were determined using Affymetrix Human U133 Plus 2.0 microarrays and analyzed as previously described. The raw gene expression data analyzed in this study are available online through the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) accession number GSE17763.
MΦ were differentiated using our previously described rapid protocols; briefly, monocytes were directly treated with 103 U/mL IL-4 (Peprotech, Rocky Hill, NJ), 10μg/mL IL-10 (R&D Systems, Minneapolis, MN), or 200ng/mL IL-15 (R&D Systems) for 48 hours immediately following plastic adherence [5, 12, 13]. Completion of MΦ differentiation was confirmed histologically. Vitamin D supplementation experiments were carried out in vitro by the addition of 25-hydroxyvitamin D3 (Enzo Life Sciences, Farmingdale, NY) to the extracellular medium of MΦ cultures.
MΦ were infected with single cell suspensions of M. leprae at MOI of 10 in antibiotic free RPMI1640 supplemented with 10% FBS for 16 hours; then washed vigorously to remove extracellular bacterium and cultured in RPMI 1640 supplemented with 10% FBS, penicillin and streptomycin.
MΦ were infected with M. leprae as indicated above. Following washing the infected MΦ were cultured with 10% 25D sufficient human serum. Total RNA and DNA were harvested from M. leprae-infected macrophages as indicated in the text and cDNA was synthesized from the RNA fraction. Bacterial 16S rRNA and M. leprae repetitive genomic element DNA (RLEP) were measured using quantitative PCR. Bacterial viability was calculated by comparing 16S rRNA to RLEP DNA as previously published [14].
Our previous study indicated that components of the vitamin D-mediated antimicrobial pathway, including CYP27B1 mRNA, were more highly expressed in T-lep lesions as compared to L-lep lesions [5]. Examining the distribution of CYP27B1 protein in leprosy lesions by immunohistochemistry, we observed greater expression of the enzyme throughout the granulomas in the T-lep vs. L-lep form (Fig 1A), correlating with mRNA levels [5]. Quantification of protein expression levels in the leprosy lesions indicated that there were 3-fold higher number of CYP27B1 positive cells in the T-lep vs. L-lep lesions (Fig 1B).
Given that M. leprae bacilli are more abundant in L-lep lesions, we determined the effect of M. leprae on CYP27B1 expression and function. Cultures of primary human monocytes were infected overnight with live M. leprae at a multiplicity of infection (MOI) of 10. Following infection of monocytes, total RNA was isolated, then CYP27B1 mRNA expression was assessed by qPCR. M. leprae infection did not significantly induce CYP27B1 mRNA expression in monocytes, whereas stimulation with exogenous Toll-like receptor 2/1 ligand (TLR2L) did (Fig 1C). Given that the function of CYP27B1 is to metabolize 25D into 1,25D, we cultured TLR2L activated or M. leprae-infected monocytes with 25D3 for 6 hours, and the levels of 25D3, 1,25D3 and 24,25D3 were measured using high performance liquid chromatography (HPLC). While monocytes treated with the exogenous TLR2L were able to significantly convert 25D3 into 1,25D3, consistent with previous studies [2], M. leprae-infected monocytes did not (Fig 1D). Conversely, gene expression of CYP24A1 mRNA, the enzyme that catalyzes 25D into the non-VDR-interacting 25D catabolite, 24,25D, was significantly reduced by M. leprae infection and TLR2L (Fig 1E). The reduction in CYP24A1 mRNA expression did not result in a detectable change in enzymatic activity (Fig 1F). These results suggest that although M. leprae expresses a cell wall associated lipoprotein capable of activating TLR2 [19], it does not activate a monocyte intrinsic pathway leading to the induction of CYP27B1 without exogenous TLR2L or IFN-γ.
Given our previous studies demonstrating that M. leprae induces monocytes to release a broad array of cytokines [14, 17], we hypothesized that induction of specific cytokines downregulates CYP27B1 expression. Of the known cytokines triggered in monocytes and MΦ during M. leprae infection, IFN-β [17] has been shown to downregulate host immune responses [17]. In particular, IFN-β inhibits IFN-γ induced CYP27B1 expression in M. leprae infected cells [20, 21]. We therefore hypothesized that M. leprae induction of type I IFN inhibits the intrinsic upregulation of CYP27B1 in the infected cells. Mining unpublished data from our previous study [17], M. leprae infection significantly induced mRNA expression of IFN-β and IFN-β protein; however, TLR2L had no effect (Fig 2A). In new experiments, M. leprae-infected monocytes also significantly expressed type I IFN downstream gene product, OAS1 by qPCR, indicating the presence of a type I IFN response (Fig 2B). Addition of a neutralizing monoclonal antibody against the type I IFN receptor (IFNAR) to the monocyte cultures prior to M. leprae infection inhibited OAS1 expression (Fig 2C), confirming activation of the type I IFN pathway during M. leprae infection. In the same experiments, neutralization of IFNAR during M. leprae infection resulted in a significant induction of CYP27B1 and VDR gene expression (Fig 2D), suggesting that M. leprae induction of type I IFN blocks activation of an intrinsic response leading to upregulation of CYP27B1.
Monocytes are initially recruited to the site of infection where they encounter microbial pathogens, but over time develop into tissue MΦ that contribute to outcome of the infection. In leprosy, T-lep lesions are characterized by M1-like MΦ and a vitamin D antimicrobial gene program; whereas, in L-lep lesions, M2-like MΦ predominate. To further explore the relationship between MΦ, the vitamin D antimicrobial and type I IFN pathways, we interrogated our previously published microarray study [17]. CYP27B1 expression was significantly greater in T-lep vs. L-lep lesions, and in 9/10 T-lep lesions greater than any of the six L-lep lesions (Fig 3A). Conversely, CD163, a marker of the M2-like MΦ in L-lep lesions, was significantly greater in L-lep vs. T-lep lesions, with the expression in 5/6 L-lep lesions greater than any of the T-lep lesions (Fig 3A). Given that IFN-β, a type I IFN, protein and mRNA are differentially expressed in L-lep vs. T-lep lesions, and impacts MΦ function [17], we examined mRNA levels for OAS1, which was significantly higher in L-lep lesions (Fig 3B). Linear regression analysis indicates that CD163 and OAS1 expression are coordinately expressed (P = 0.0004, R2 = 0.6019) and two way ANOVA analysis confirms that leprosy type of the lesion is significantly (P = 0.0003) associated with differential gene expression of CYP27B1, CD163 and OAS1 (Fig 3C).
To model the MΦ subtypes in leprosy lesions, we studied MΦ subsets derived in vitro with IL-15- vs. IL-10-treated monocytes yielding M1-like vs. M2-like MΦ, respectively. We previously described the IL-15- vs. IL-10-derived MΦ, to be phenotypically similar and functionally consistent with MΦ subtypes found in leprosy lesions [5, 13, 22]. We differentiated MΦ by treating monocytes with IL-10 or IL-15 for 48 hours, then determined their ability to metabolize vitamin D using HPLC as we have described above. Consistent with previous findings [5], IL-15 MΦ, but not IL-10 MΦ converted 25D3 to 1,25D3 (Fig 3D), which correlates with our in situ microarray and protein studies. To ascertain the effects of M. leprae infection on the vitamin D metabolic system in MΦ, we infected IL-10 MΦ and IL-15 MΦ with M. leprae, and measured CYP27B1and OAS1 mRNA expression by qPCR. M. leprae infection of IL-10 MΦ did not alter CYP27B1 mRNA expression, although exogenous TLR2 stimulation induced a significant upregulation (Fig 4A). On the other hand, CYP27B1 gene expression in IL-15 MΦ, which have high CYP27B1 gene expression at baseline, was not significantly affected as a result of either infection or stimulation with exogenous TLR2L (Fig 4A). In the same experiments, M. leprae infection of IL-10 MΦ led to the significant induction of OAS1 gene expression, but not stimulation with exogenous TLR2L (Fig 4B). In contrast, neither M. leprae infection not TLR2/1 activation of IL-15 MΦ induced upregulation OAS1 as a result of infection or following TLR2/1 stimulation. These findings suggest that in distinct MΦ subsets differentially regulate the vitamin D and type I IFN pathways.
To determine whether the vitamin D antimicrobial pathway was functional in M. leprae infected IL-15 MΦ, we added 25D3 to infected cells and measured the vitamin D-dependent antimicrobial peptide, cathelicidin (CAMP). Indeed, when exogenous 25D3 was added to M. leprae infected IL-15 MΦ at various doses, CAMP gene expression was induced in a dose-dependent fashion (Fig 4C), suggesting that these M1-like MΦ retain the intrinsic ability to engage the vitamin D-dependent antimicrobial pathway during M. leprae infection in the presence of sufficient 25D3 prohormone.
The induction of the type I IFN response by M. leprae in infected IL-10 MΦ but not IL-15 MΦ, led us to explore the functional consequences of this cytokine response on the intrinsic induction of the vitamin D antimicrobial pathway. IL-10 MΦ were treated with a neutralizing monoclonal antibody specific for IFNAR and then infected with live M. leprae. The addition of neutralizing IFNAR antibody restored an intrinsic activation pathway leading to upregulation of CYP27B1 and VDR mRNA (Fig 5A), but conversely, inhibited expression of the type I IFN downstream gene OAS1 (Fig 5B). These data indicate that intrinsic activation of the vitamin D-dependent antimicrobial pathway during M. leprae infection is inhibited by the ability of the bacteria to induce type I IFN. Finally, we addressed whether the induction of type I IFN by M. leprae regulated an intrinsic antimicrobial response against M. leprae. In M. leprae-infected IL-10 MΦ cultured with 25D-sufficient human serum (44 ng/ml of 25D), treatment with the IFNAR neutralizing monoclonal antibody significantly reduced M. leprae viability as compared to the isotype control (Fig 5C). These results suggest that M. leprae, upon infection of M2-like MΦ, evades activation of the intrinsic antimicrobial response by invoking the type I IFN pathway. In summary, the regulation of the vitamin D antimicrobial pathway in M1-like vs. M2-like MΦ is distinct, and contributes to the outcome of the innate immune response to M. leprae.
Cells of the monocyte/MΦ lineage mount an innate antimicrobial response to defend against infection by an intracellular pathogen, yet the microbe has evolved to counter this intrinsic capacity to kill the foreign invader. Although the exogenous addition of purified ligands (TLR2L) derived from the pathogen provides an extrinsic mechanism to activate a vitamin D-dependent antimicrobial response in infected MΦ, the intrinsic activation of innate antimicrobial responses by the pathogen is not sustained [3]. Here we studied leprosy as a model to ascertain whether MΦ are capable of an intrinsic antimicrobial response to infection by M. leprae. We provide evidence here that infection of human MΦ by M. leprae intrinsically activates the vitamin D antimicrobial pathway as part of the innate immune response, but the bacterium blocks this response via the induction of type I IFN.
There is evidence for the existence of intrinsic anti-mycobacterial responses, although these are not usually sustained, and escaping the initial antimicrobial response is key for the pathogen to establish long term infection [3]. The identification of distinct MΦ gene expression profiles induced according to virulence of the infecting strain of M. tuberculosis suggests that subversion of the initial host response is critical to establishing infection [23]. Therefore, the ability of the invading pathogen to modulate the intrinsic macrophage response, such as antagonizing TLR2 [24] or suppressing the vitamin D-dependent antimicrobial response, will lead to immunopathology. Here, we demonstrate that monocytes and MΦ have the capacity to trigger an intrinsic antimicrobial response during M. leprae infection in the absence of exogenous triggers (TLR2L or IFN-γ), which is inhibited by the aberrant infection-induced expression of type I IFN. M. leprae may activate the intrinsic immune response in MΦ through a variety of pattern recognition receptors including TLR4, TLR9 and NOD2, but most importantly, TLR2 via the 19kDa [19], 33kDa [19, 25] and mLEP major membrane protein-II lipoproteins [26], since TLR2 activation leads to induction of CYP27B1 expression [2].
The intrinsic activation of the vitamin D antimicrobial pathway requires the metabolic conversion by the CYP27B1 enzyme of 25D, the inactive circulating prohormone (25D) and requires into the bioactive form (1,25D3) to transactivate its cognate receptor, the vitamin D receptor (VDR). Activation of human monocytes and MΦ by exogenous innate and adaptive immune signals (such as TLR2/1 and IFN-γ, respectively) have been shown to induce expression and function of CYP27B1 [1, 2]. Our data demonstrate that in contrast to activation with exogenously added TLR2 ligand or IFN-γ, monocytes infected with live M. leprae showed little intrinsic induction of CYP27B1 expression or enzyme activity. When the type I IFN receptor (IFNAR) was neutralized during M. leprae infection, the intrinsic induction of CYP27B1 during M. leprae infection was uncovered. This suppression of CYP27B1 has in vivo relevance, as we found that CYP27B1 is more highly expressed in T-lep vs. L-lep lesions, the self-limited vs. progressive forms of leprosy, respectively, and was inversely correlated with type I IFN signaling. Using our previously characterized and described in vitro models of the L-lep and T-lep resident MΦ (M2-like IL-10 MΦ and M1-like IL-15 MΦ) [5], we demonstrated that M. leprae infection of IL-10 MΦ induces a type I IFN response that inhibits CYP27B1 expression, similar to observations in monocytes. In contrast, IL-15 MΦ maintain their CYP27B1 expression and function in the presence of M. leprae infection. These data indicate that the regulation of CYP27B1 in infected MΦ at the site of disease is critical to activation of the intrinsic antimicrobial response.
The induction of type I IFN is a well-studied host defense mechanism against viral infection; however, their role in the immune response against intracellular infection with mycobacteria and bacteria is less defined [27]. While type I IFN are critical to clearance of viral infections, the same type I IFN response mediates suppression of antibacterial responses, leading to secondary bacterial infections, such as Streptococcus pneumonia [28]. The fact that the robust expression of type I IFN and downstream genes along with low CYP27B1 expression is characteristic of L-lep lesions and vice versa in T-lep lesions [17], suggests that the ability of type I IFN to inhibit CYP27B1 contributes to the outcome of the host response against mycobacteria in leprosy. Given that the neutralization of type I IFN uncovers the intrinsic induction of CYP27B1 expression following M. leprae infection in vitro, the induction of type I IFN provides an escape mechanism by which the bacterium subverts the vitamin D-mediated antibacterial response. The ability of type I IFN to inhibit CYP27B1 expression is likely related to the production of the immunosuppressive cytokine IL-10, which has been shown to be induced by type I IFN during M. leprae infection and inhibit CYP27B1 expression [14, 17]. Alternatively, the CYP27B1 promoter region contains an IRF8 (-543) response element as determined by MotifMap (http://motifmap.ics.uci.edu/), which is a type I IFN inducible transcription factor with known gene suppression functions [29]. Other studies have suggested additional mechanisms by which M. leprae blocks host defenses, some mediated through the mycobacterial cell wall component phenolic glycolipid 1 (PGL-1). Several studies have shown that PGL-1 manipulates host defense mechanisms such as complement activation, phagocytosis as well as cytokine release to inhibit maturation of dendritic cells and modulate T cell responses [30–33]; all of which enables survival of the bacteria. With regard to vitamin D-mediated antibacterial responses, our previous study showed that microRNA-21 (hsa-miR-21) was highly expressed in L-lep lesions vs. T-lep lesions and inhibited CYP27B1 gene expression and function [14]. Similar to the results presented here, hsa-miR-21 induction was exclusive to live M. leprae infection and not induced by purified TLR2 ligands [14]. Importantly, neutralization of the type I IFN pathway in M. leprae infected IL-10 MΦ resulted in decreased bacterial viability by uncovering the intrinsic vitamin D-mediated antimicrobial response. Indeed, further understanding of the pathways by which M. leprae initiates immune inhibitory mechanisms such as induction of the type I IFN pathway will provide novel therapeutic targets for mycobacterial diseases.
There is a well characterized genetic and functional association of the vitamin D pathway with leprosy. Single nucleotide polymorphisms in the VDR gene are associated with the different forms of leprosy [15, 34], as is expression levels of the protein itself [35]. However, the use of vitamin D to treat mycobacterial disease has been studied in clinical trials, which have shown inconsistent benefits [6, 10, 11, 36–39]. Our findings suggest a possible explanation for the varied outcomes. The efficacy of elevated systemic 25D levels in affecting local antimicrobial responses at the site of infection is predicated on the ability of the innate immune cells to convert the circulating 25D to 1,25D at the site of infection. Thus, if the pathogen bearing MΦ, such as those found in L-lep lesions, were unable to convert 25D, it would not be surprising to see minimal therapeutic benefit following vitamin D supplementation. More broadly, our findings suggest that the clinical management of mycobacterial disease using vitamin D supplementation will require simultaneous management of the vitamin D metabolic system to achieve therapeutic benefit. In conclusion, our results demonstrate that the intrinsic capacity of cells to activate antimicrobial defense mechanisms as part of the innate response, versus the ability of the pathogen to mask these responses, is a critical determinant of the outcome of infection.
|
10.1371/journal.pntd.0000324 | Controlling Tungiasis in an Impoverished Community: An Intervention Study | In Brazil, tungiasis is endemic in some resource-poor communities where various domestic and sylvatic animals act as reservoirs for this zoonosis. To determine the effect of control measures on the prevalence and intensity of infestation of human and animal tungiasis, a repeated cross-sectional survey with intervention was carried out.
In a traditional fishing community in Northeast Brazil, humans and reservoir animals were treated, and premise-spraying using an insecticide was done, while a second fishing community served as a control. Both communities were followed up 10 times during a 12-month period. At baseline, prevalence of tungiasis was 43% (95% confidence interval [CI]: 35%–51%) and 37% (95% CI: 31%–43%) in control and intervention villages, respectively. During the study, prevalence of tungiasis dropped to 10% (95% CI: 8%–13%; p<0.001) in the intervention village, while the prevalence remained at a high level in the control village. However, after one year, at the end of the study, in both communities the prevalence of the infestation had reached pre-intervention levels. Whereas the intensity of infestation was significantly reduced in the intervention community (p<0.001), and remained low at the end of the study (p<0.001), it did not change in the control village.
Our study shows that a reduction of prevalence and intensity of infestation is possible, but in impoverished communities a long-lasting reduction of disease occurrence can only be achieved by the regular treatment of infested humans, the elimination of animal reservoirs, and, likely, through environmental changes.
Controlled-Trials.com ISRCTN27670575
| Tungiasis is a disease caused by the sand flea Tunga penetrans, a parasite prevalent in many impoverished communities in developing countries. The female sand flea penetrates into the skin of animals and humans where it grows rapidly in size, feeds on the host's blood, produces eggs which are expelled into the environment, and eventually dies in situ. The lesions become frequently superinfected and the infestation is associated with considerable morbidity. Clearly, tungiasis is a neglected disease of neglected populations. We investigated the impact of a package of intervention measures targeted against on-host and off-host stages of T. penetrans in a fishing community in Northeast Brazil. These measures decreased disease occurrence only temporarily, but had a sustained effect on the intensity of the infestation. Since infestation intensity and morbidity are correlated, presumably the intervention also lowered tungiasis-associated morbidity. Control measures similar to the ones used in this study may help to effectively control tungiasis in impoverished communities.
| Tungiasis is a parasitic disease caused by the sand flea Tunga penetrans. Female fleas penetrate into the epidermis where they undergo a process of so-called neosomy and expel several hundred eggs into the environment. After a period of six weeks, the parasite dies in situ and is sloughed off the epidermis by tissue repair mechanisms [1].
During the last decades growing urbanization and improved housing has resulted in a reduction of prevalence. Today, the occurrence of tungiasis is confined to resource-poor communities located at the coast or in the rural hinterland, and to slums of rapidly growing urban agglomerations in Latin America, the Caribbean and Sub-Saharan Africa [2],[3]. In these settings prevalences range between 15% and 51% [2], [4]–[6]. Since prevalence, intensity of infestation, and morbidity are positively related [7], debilitating and disfiguring sequels are common in resource-poor rural and urban communities. Tungiasis is clearly a neglected disease of marginalized populations [8],[9].
Although by its nature a self-limiting disease, tungiasis causes considerable morbidity [10],[11]. Fissures, ulcers, gangrene, lymphedema, deformation and loss of nails and auto-amputation of digits are known sequels [10]. In non-immune individuals tungiasis is a risk factor for tetanus [12],[13]. Superinfection of the lesions is virtually constant [14],[15], and a variety of aerobic and anaerobic bacteria have been isolated from embedded sand fleas [16],[17].
Beside humans, T. penetrans parasitizes a range of domestic animals, such as dogs, cats, pigs and rodents [18],[19]. In Brazil, dogs and cats act as important reservoirs for the intra- and peridomiciliary transmission of sand fleas [18],[20]. When humans live in close contact with infested animals, the risk of infestation is high and the intensity of infestation is high [20].
The control of tungiasis in a resource-poor population with interventions targeted at the human and the animal population has never been described. Here we present the results of an intervention performed in collaboration with public health services in an endemic community in northeast Brazil. The results show that combining treatment of humans with treatment of animals and focal spaying of an insecticide reduced prevalence and intensity of infestation.
To evaluate the effect of multiple interventions on the prevalence of tungiasis, two endemic communities were selected in Ceará State, northeast Brazil (Balbino and Pedro de Souza). Both communities are situated in Cascavél Municipality about 60 km south of Fortaleza, the state capital. The fishing communities are separated about 6 km and located within sand dunes near the Atlantic Ocean, showing little fluctuation of their population. The two communities are very similar with regard to demographic, crowding, social and economic characteristics. Houses are located on rather large compounds surrounded by fences and built on sandy soil. The quality of housing is poor and streets are not paved. Kitchens are indoors or consist of open stalls on the compound. Both communities are integrated in the national Family Health Program (“Programa da Saúde da Família”) and served by community health care workers (“agentes comunitários de saúde”). During the study, the closest primary health care center was located in another community, some 10 km away.
In October 2002, Balbino was inhabited by 148 families with a total population of 630. The community of Pedro de Souza comprised 251 individuals in 60 families. Inhabitants of all age groups were eligible for the study, provided they had spent at least four days per week in the village during the last three months. Dogs and cats were included if they were born in the villages or had lived there for at least two months. Pigs, goats, sheep, cows and horses, other species of domestic animals occurring in the villages were previously excluded to be animal reservoirs of T. penetrans in this setting [20].
The study was conceived as a repeated cross-sectional survey with intervention. To assess the impact of interventions on the prevalence and intensity of infestation, two complete villages were compared. In the community Balbino various interventions were implemented while the Pedro de Souza community served as a control. Interventions were coordinated and implemented by the Mandacaru Foundation (Fortaleza, Brazil) in collaboration with the Health Secretariat of Cascavel Municipality. Between November 2002 and November 2003, a total of 10 surveys were planned and carried out in each community. The villages were visited during identical periods according to pre-defined dates with a maximum delay of 10 days between intervention and control community.
The study started in November 2002, in the middle of the dry season when the prevalence of tungiasis peaks [21]. During the preparatory phase, contact was made with community leaders, and the objectives of the study were explained. Censuses of the human and the animal population were performed and all houses mapped using a global positioning system (GPS). All data were collected by door-to-door surveys. The primary outcome was the prevalence of tungiasis (dichotomous), and the secondary outcome intensity of infestation (continuous).
All surveys were carried out by the same investigators (S.S., L.W.), accompanied by community health agents. During the surveys the human and the animal population were carefully examined for the presence of embedded sand fleas, according to previously established guidelines. In humans the entire body was examined (except the genitals) to identify any ectopic lesions [22]. In dogs and cats examination focused on paws, abdomen and muzzle, the topographic areas most commonly affected [18]. The following findings were considered to be diagnostic for human as well as animal tungiasis [1]: a red-brownish spot with a diameter of 1–3 mm with visible posterior segments of the penetrated flea (early stage); a circular whitish lesions with a diameter of 4–10 mm with a central black dot (mature stage), round black crust surrounded by necrotic tissue (late stage with dead parasite). Typical residuals in the epidermis, lesions altered through manipulation (such as partially or totally removed fleas leaving a characteristic crater-like sore in the skin), and suppurative lesions (caused by the use of nonsterile instruments), were recorded as well. Lesions were differentiated into viable, dead and manipulated lesions. At each assessment, participants were also asked about any adverse events that may have occurred in consequence of the intervention.
Clinical experience shows that infestation with more than ten sand fleas oftentimes result in considerable morbidity. We therefore considered infestation with up to five embedded fleas as low, between six and as 10 moderate and >10 lesions as a high intensity of infestation, in analogy to a previously used classification [23]. In animals the stratification was <10, 11–20 and >20, respectively.
After baseline examination in November 2002, the control measures were carried out in the intervention village (Balbino). From November 2002 through January 2003, from all infested individuals embedded sand fleas were extracted every two to three weeks by experienced health care professionals under sterile conditions. The remaining sore was treated with an antibiotic ointment. During the same period all cats and dogs were treated with trichlorphone 97% in oily solution (Neguvon, Bayer do Brasil, São Paulo, Brazil) or neck collars impregnated with propoxur and flumethrin (Kiltix, Bayer Bayer do Brasil, São Paulo, Brazil). In case of loss of neck collars these were substituted at the next survey. In February 2003, deltamethrin was used for focal premise treatment. Focal spraying was performed by trained personnel of the Health Secretariat of Cascavel Municipality. The insecticide was sprayed on the ground next to the houses targeting areas in which off-host development of T. penetrans was suspected to occur, such as preferred whereabouts of dogs and cats, and shady places under trees [24], or inside houses in the case of a sandy floor. Focal premise treatment using insecticides was repeated twice during a period of six weeks. Table 1 summarizes the type and the period of interventions.
The study protocol was approved by the Ethical Review Board of the Federal University of Ceará Fortaleza, Brazil (Protocol no. 195/02). In addition, an ad hoc ethical committee, consisting of physicians, community members and professionals of the Health Secretariat of Cascavel Municipality approved the study protocol. Informed written consent was obtained from all study participants. In the case of minors, written consent was obtained from the minors and their carers. Written consent was also obtained from pet owners. At the end of the study surgical treatment of tungiasis was offered to affected individuals in both communities.
Data were entered in Epi Info version 6.04d (CDC, Atlanta, USA), checked for entry errors and transferred to SPSS 11.04 for Macintosh (SPSS Inc., Chicago, IL, USA) for analysis. The χ2 test was employed to determine the significance of difference of proportions between population groups, and between the intervention and control communities. For the comparison of point prevalences within one community, the McNemar test was used. Infestation intensities were compared by the Mann-Whitney test. The prevalence ratio [PR] for tungiasis and the significance of differences in the relative distribution of tungiasis were calculated in contingency tables. To calculate the relative prevalence reduction Balbino and Pedro de Souza were regarded as one study population, and belonging to either community was considered an exposure variable. The relative prevalence reduction was calculated as follows: prevalence of tungiasis in Pedro de Souza–prevalence of tungiasis in Balbino/prevalence of tungiasis in Pedro de Souza.
The intervention community was considerable bigger, but socio-demographic characteristics were similar in both communities. Table 2 summarizes baseline characteristics of the study populations.
At baseline (November 2002) the prevalence of tungiasis in the human population was slightly higher in Pedro de Souza, the control village (43%, 95% confidence interval [CI]: 35–51%), than in Balbino, the intervention village (37%, 95% CI: 31–43%;). However, the difference was not significant (Figure 1A; p = 0.11).
At the first follow-up in December 2002 –after completion of the first cycle of intervention measures 1 and 2 (see Table 1)– the prevalence of tungiasis dropped to 25% (95% CI: 35–51%) in Balbino, while it remained unchanged in Pedro de Souza (pre-intervention versus one month post-intervention p = 0.05 and p = 0.47, respectively; Figure 1A). In the intervention village prevalence continued to decrease to 18% (95% CI: 15–22%) in January 2003 (p = 0.001), 15% (95% CI: 12–18%) in February (p = 0.17), and 10% (95% CI: 8–13%) in March (p = 0.001; all p compared to the preceding survey). During the same period no significant reduction in prevalence of tungiasis was noted in Pedro de Souza (Figure 1A). Here, prevalence in March 2003 remained as high as 36% (95% CI: 29–44%). During the rainy season a strong reduction in prevalence was observed in the control community, too. Prevalence decreased significantly from 36% (95% CI: 29–44%) in March to 9% (95% CI: 4–13%) in May (p<0.001).
With the beginning of the dry season (June–July 2003) prevalence started to rise in both communities, and prevalence curves were almost parallel. By November 2003 one year after beginning of the study, prevalences in both communities had reached the pre-intervention level.
During the 12-month study period, the prevalence curves of animal tungiasis showed a similar pattern as compared to human tungiasis, with higher baseline values in Pedro de Souza (86%, 95% CI: 71–100%) than in Balbino (64%, 95% CI: 52–75%; p = 0.02) and an impressive decrease in prevalence during the rainy season (Figure 1B). However, the variation of measurements was considerably higher than in the human population.
The PR of tungiasis showed no significant association at the beginning of the intervention in November 2002 (PR = 0.81, 95% CI: 0.65–1.03; p = 0.11). After cessation of the intervention (March 2003) the PR had decreased significantly in Balbino, as compared to Pedro de Souza (PR = 0.28, 95% CI: 0.2–0.39; p<0.001). At the end of the study in November 2003 the PR remained slightly but significantly lower in the intervention village (PR = 0.69, 95% CI: 0.52–0.93; p = 0.017). The prevalence reduction in Balbino was 55% and 19% in March and November 2003, respectively.
In the intervention population, the prevalence of individuals with a high and a moderate intensity of infestation decreased significantly from pre-intervention (November 2002) to four months after intervention (March 2003): prevalence of individuals with high intensity 8.4% (95% CI: 6–11%) versus 0.8% (95% CI: 0.02–1.5%; p<0.001) and with moderate intensity 3.3% (95% CI: 1.6–5%) versus 0.6% (95% CI: 0.001–1.2%), p = 0.002 (Figure 2A). In contrast, no significant reduction was observed in the control village: 3% (95% CI: 0.4–6%) versus 2% (95% CI: 0–4%; p = 0.99) and 6.8% (95% CI: 2–11%) versus 3.4% (95% CI: 0.4–6%), respectively (p = 0.25; Figure 2A).
At the end of the study (November 2003) the prevalence of heavy and moderate infested individuals remained significantly lower in Balbino as compared to the pre-intervention level: 8.4% (95% CI: 6–11%) versus 3% (95% CI: 1–5%) after intervention (p = 0.001), and 3.3% (95% CI: 1.6–5%) versus 1.6% (95% CI: 0.3–2.8%; p = 0.013), respectively. In contrast, at the end of the study in Pedro de Souza the proportion of heavily and moderate infested individuals was even higher than at baseline: 3% (95% CI: 0.4–6%) versus 4% (95% CI: 0.5–7%; p = 0.25) and 6.8% (95% CI: 2–11%) versus 6.7% (95% CI: 2–11; p = 0.99), respectively (Figure 2A). Similar, the total number of embedded sand fleas per person was significantly reduced in the intervention village, from a median of 18 in November 2002 to a median number of one in March 2003 (p<0.001) and remained significantly lower in November 2003 (median of five lesions; p<0.001 compared to baseline).
There were no significant differences in the reduction of intensity of infestation in the animal population (Figure 2B).
No adverse events related to the intervention were recorded. However, as our therapy did not differ from the treatment done by most community members themselves, we assume that people did not report any minor adverse events associated with surgical extraction (such as pain and superficial bleeding).
The high prevalence of tungiasis in endemic areas and the important morbidity associated with this parasitic skin disease call for the implementation of control measures. As a first step to prove that successful intervention is possible, we determined the impact of repeated rounds of surgical extraction of embedded sand fleas in humans in combination with on-host treatment of dogs and cats, and focal spraying of an insecticide on the premises. Our study shows that the interventions were effective to control tungiasis in the short-term but failed to show an impact on the long run. This is reflected by a minuscule reduction of the prevalence ratio at the end of the study, i.e. 12 months after start of interventions.
Several factors seem to be responsible for the re-increase of prevalence to baseline level in the intervention village at the end of the study. Firstly, the backbone of the intervention, the surgical extraction of embedded sand fleas, obviously has several shortcomings. As inhabitants of endemic areas rarely remove embedded sand fleas in a systematic manner–it is time-consuming, painful and often results in superinfection [25]– individuals could have abstained from the treatment. As a consequence, the human reservoir of T. penetrans may not have diminished as intended [23].
Presumably the inflammatory reaction of the skin at sites of embedded sand fleas facilitates penetration [25]. Since the barrier function of the epidermis is not immediately reconstituted after sand fleas have been taken out (the sore produced by the surgical manipulation even might temporarily increase the surface of the skin particular susceptible to penetration), the idea that i) the reduction of the parasite burden would prevent re-infestation, and that ii) by consequence, the number of eggs expelled into the environment would reduce transmission, might be an invalid assumption. Actually, people at risk for immediate re-infestation could be more susceptible to the infestation with T. penetrans after extraction and may only profit from the surgical extraction after complete healing of the skin.
We suggest that the prevention of infestation, rather than the surgical extraction of already embedded sand fleas, may interrupt transmission more effectively. Zanzarin, a plant-based repellent, has been shown to effectively prevent the infestation with T. penetrans in areas with high attack rates [26]. This compound would be an ideal candidate for prophylaxis but was not available in Brazil when the study was designed.
Secondly, many cats and dogs remained infested despite the on-host treatment (Figure 1B). By consequence these animals continued spreading T. penetrans and contributed to ongoing transmission in the community [27]. After on-host intervention had been stopped, more and more animals became re-infested, with an even higher prevalence at the end of the study as compared to baseline data (Figure 1B). Assumably, these animals carried sand fleas to the compounds of their owners where they fuelled peri- and intradomiciliary transmission [20]. Actually, in both communities the strongest increase in prevalence of tungiasis in humans followed a strong increase in prevalence in animals (Figures 1 A, B). Recently developed on-host products, such as a combination of imidacloprid and permethrin (Advantix), effectively prevented infestation with T. penetrans in animals and lowered parasite burden [27]. Again, this product was not available when the study was conceived.
Finally, focused premise treatment with deltamethrin aimed to interrupt the off-host cycle of T. penetrans [18],[28]. For an optimal efficacy, focal spraying has to be applied at all sites where off-host development takes part in the soil, which means, that those sites have to be identified first. Due to delivery problems and constraints in qualified personnel, focal premise treatment only started in January, i.e., very late in the seasonal cycle of T. penetrans. In addition, there was no expertise to examine soil samples for developmental stages of sand fleas. Obviously, spraying of breeding sites is better done before the parasite population has expanded, i.e., at the beginning of the dry season [21]. However, the significant reduction in prevalence after the implementation of this intervention seems to have booster the protective effect of the preceding interventions.
When transmission of T. penetrans is altered–e.g. through an intervention–the intensity of infestation reflects the infestation rates individuals have experienced during the last months [7]. Thus, intensity of infestation is a better outcome measure to evaluate the impact of an intervention than assessment of the prevalence. In addition, it is a better proxy of morbidity reduction, as morbidity is significantly correlated to intensity of infestation [7].
In contrast, prevalence merely measures absence or presence of infestation at a given point of time and will therefore rapidly increase if the attack rate is high, such as at the beginning of the dry season. Prevalence does not allow inferring on decreasing attack rates prior to the assessment since embedded sand fleas and their remains are visible for up to three months [1].
Our study shows that in the intervention community a prolonged effect on the intensity of infestation of the human population occurred. This means, that despite a persistent high prevalence of tungiasis, attack rates were reduced. However, this effect was not observed in the animal population (Figure 2B). This indicates that the intervention measures applied were not very effective in reducing the parasite burden in dogs and cats.
Any assessment of intervention methods is hampered by the characteristic seasonal variation of tungiasis. Disease occurrence decreases as soon as the rainy season starts and re-increases with the beginning of the dry season [21]. We therefore opted to include a control community where no intervention was done. Although both fishing communities were selected by their similarity with respect to demographic, physical and socio-cultural characteristics, one cannot exclude that they have differed in an unknown factor of epidemiological relevance, an additional, hitherto neglected animal reservoir. Such a factor could be responsible for different dynamics in prevalence between Balbino and Pedro de Souza during the year. However, different dynamics in the same epidemiological setting have never been reported and the similar development of prevalence towards the end of the study renders such an explanations unlikely.
One major problem in the interpretation of data arises from the differences in population size between the two communities. This difference bares the possibility that significances in the changes of prevalence and/or intensity of infestation in the control village during the study period were not detected due to the smaller sample size.
Non-participation may have biased the assessment of point-prevalences during the observation period. For surveys with a low participation, it is conceivable that over- or under-representation of infested individuals has skewed the results towards higher or lower prevalence. Fluctuation of participation was particularly high in animals, indicating that point prevalences in the animal population were likely to be biased.
Another shortcoming of the study is that the study design does not allow identifying the relative effective of the three interventions applied in Balbino village. To address such an issue, various communities would have to be included, in each of which a particular intervention has to be performed. For operational reasons and financial constraints this could not be done.
Due to the operational nature of the study, we did not include any multivariate analysis. The study was not designed as a typical trial and thus did not include detailed collection of data on possible confounders. We aimed to describe whether an intervention has any effect rather than identifying independent factors for success or failure.
Off-host stages of T. penetrans develop best in dry soil or in dusty soil containing organic material [9],[21],[29]. Measures aiming to interrupt the off-host development should therefore focus on physically changing the environment in which eggs, pupae, and larva develop. This can be done through paving streets, cementing floors, and eliminating uncontrolled disposal of waste in public areas and private compounds [30],[31]. However, theses interventions require substantial funds and are beyond the economic capabilities of most communities where tungiasis is endemic.
Although the present study was conducted in a rural fishing community, it is conceivable that interventions are also applicable to urban settings where the animal reservoirs are similar [18].
Our study is a first step in the exploration of possible control measures against T. penetrans. Further studies are needed to assess the full potential of putative interventions. To do so, a cluster-randomized phased implementation study with various communities phasing-in specific interventions would be an ideal approach [32].
The Brazilian Unified Health System (SUS) with a network of primary health care clinics, many located in endemic communities, would offer a convenient way to coordinate such a study. Randomized clinics could implement intervention measures, such as prevention of infestation by application of Zanzarin, surgical extraction of embedded fleas, distribution of animal treatment, and coordination of premise spraying with insecticides. This would allow identifying the most effect measure over a prolonged period of time and controlling for seasonal variation [32].
|
10.1371/journal.ppat.1007681 | The requirement for co-germinants during Clostridium difficile spore germination is influenced by mutations in yabG and cspA | Clostridium difficile spore germination is critical for the transmission of disease. C. difficile spores germinate in response to cholic acid derivatives, such as taurocholate (TA), and amino acids, such as glycine or alanine. Although the receptor with which bile acids are recognized (germinant receptor) is known, the amino acid co-germinant receptor has remained elusive. Here, we used EMS mutagenesis to generate mutants with altered requirements for the amino acid co-germinant, similar to the strategy we used previously to identify the bile acid germinant receptor, CspC. Surprisingly, we identified strains that do not require co-germinants, and the mutant spores germinated in response to TA alone. Upon sequencing these mutants, we identified different mutations in yabG. In C. difficile, yabG expression is required for the processing of key germination components to their mature forms (e.g., CspBA to CspB and CspA). A defined yabG mutant exacerbated the EMS mutant phenotype. Building upon this work, we found that small deletions in cspA resulted in spores that germinated in the presence of TA alone without the requirement of a co-germinant. cspA encodes a pseudoprotease that was previously shown to be important for incorporation of the CspC germinant receptor. Herein, our study builds upon the role of CspA during C. difficile spore germination by providing evidence that CspA is important for recognition of co-germinants during C. difficile spore germination. Our work suggests that two pseudoproteases (CspC and CspA) likely function as the C. difficile germinant receptors.
| Germination by C. difficile spores is one of the very first steps in the pathogenesis of this organism. The transition from the metabolically dormant spore form to the actively-growing, toxin-producing vegetative form is initiated by certain host-derived bile acids and amino acid signals. Despite near universal conservation in endospore-forming bacteria of the Ger-type germinant receptors, C. difficile and related organisms do not encode these proteins. In prior work, we identified the C. difficile bile acid germinant receptor as the CspC pseudoprotease. In this manuscript, we implicate the CspA pseudoprotease as the C. difficile co-germinant receptor. C. difficile cspA is encoded as a translational fusion to cspB. The resulting CspBA protein is processed post-translationally by the YabG protease. Inactivation of yabG resulted in strains whose spores no longer responded to amino acids or divalent cations as co-germinants and germinated in response to bile acid alone. Building upon this, we found that small deletions in the cspA portion of cspBA resulted in spores that could germinate in response to bile acids alone. Our results suggest that two pseudoproteases regulate C. difficile spore germination and provide further evidence that C. difficile spore germination proceeds through a novel spore germination pathway.
| Clostridioides difficile (formerly Clostridium difficile) [1–3] is a Gram-positive, spore-forming pathogenic bacterium, and has become a leading cause of nosocomial diarrhea in the United States [4, 5]. C. difficile infection (CDI) is commonly the result of disruption to the gut microflora caused by antibiotic use [5–7]. Due to the broad-spectrum nature of many antibiotics, alterations to the ecology of the colonic microbiome results in the loss of the colonization resistance that is provided by the microbiota. Subsequently, patients are treated with other, broad-spectrum, antibiotics (e.g., vancomycin or fidaxomicin) which treat the actively growing, toxin-producing, vegetative cells [8]. Although these antibiotics alleviate the primary symptoms of disease, the continued disruption to the colonic microbiome results in frequent CDI recurrence. The symptoms of CDI are caused by the actions of two secreted toxins. TcdA (an enterotoxin) and TcdB (a cytotoxin) are endocytosed by the colonic epithelium and inactivate the Rho-family of small GTPases leading to loss of barrier function and inflammation of the colonic epithelium [7].
Though C. difficile vegetative cells produce the toxins that cause CDI, they are strictly anaerobic and only survive short periods of time outside the anaerobic colonic environment [9]. However, the spores that are produced by the vegetative form are critical for transmission between hosts because of their resistance to environmental factors such as heat, UV, chemicals and, importantly, oxygen [10–14]. The overall architecture of spores is conserved among all endospore-forming bacteria, C. difficile included. The centrally-located core is composed of DNA, RNA, ribosomes and proteins necessary for the outgrowth of a vegetative cell, post germination [11, 14]. The DNA in the core is protected from UV damage by small acid soluble proteins (SASPs) and much of the water in the core is replaced by pyridine-2, 6-dicarboxylic acid (dipicolinic acid; DPA), chelated with calcium (CaDPA), which provides heat resistance to the spores [11, 14, 15]. The core is surrounded by an inner membrane composed of phospholipids with minimal permeability to small molecules, including water [15]. A thin germ-cell-wall layer surrounds the inner membrane and becomes the cell wall of the vegetative cell upon outgrowth. A thick layer of specialized peptidoglycan (cortex) surrounds the germ cell wall and helps constrain the core against osmolysis [15]. Finally, surrounding the cortex is the outer membrane which, initially, serves as an organization structure / point for the coat layers but may be lost later during spore germination [11, 16–20]. All these features of endospores contribute to ensuring that the spores remain metabolically dormant.
Though dormant, spores still sense their environment for species-specific germination-inducing small molecules and, when appropriate germinants are present, initiate the process of spore germination. Much of our knowledge of spore germination comes from studies in Bacillus subtilis (a model organism for studying spore formation and germination). In B. subtilis, and most other endospore-forming bacteria, germination is activated upon binding of the germinants to the Ger-type germinant receptors that are deposited in or on the inner spore membrane [21, 22]. In B. subtilis, this event triggers an irreversible process whereby CaDPA is released through a channel composed of the SpoVA proteins [23–29]. The release of CaDPA is an essential step for the germination process because it results in the rehydration of the spore core and permits the eventual resumption of metabolic activity. In B. subtilis, the cortex peptidoglycan layer is then degraded, and this event can be activated by the CaDPA that is released from the core [30].
In contrast to B. subtilis, C. difficile does not encode orthologues of the Ger-type germinant receptors suggesting that the initiation of C. difficile spore germination is fundamentally different than what occurs in most other studied organisms [10]. Germination by C. difficile spores is triggered in response to certain bile acids in combination with certain amino acids [10, 31, 32]. In all identified C. difficile isolates, the cholic acid-derivative, taurocholate (TA), is the most efficient bile acid at promoting spore germination, and glycine is the most efficient amino acid co-germinant [33–35]. Recently, calcium was identified as an important contributor to spore germination and may synergize with other co-germinants to enhance C. difficile spore germination [36, 37].
In a screen to identify ethyl methane sulfonate (EMS)-generated mutants that do not respond to TA as a spore germinant, we identified the germination-specific, subtilisin-like pseudoprotease, CspC, as the C. difficile bile acid germinant receptor [38]. Prior to work performed in C. difficile, the Csp locus was best studied in Clostridium perfringens [39–42]. C. perfringens encodes three Csp proteases, CspB, CspA, and CspC, that are predicted to cleave the inhibitory pro-peptide from proSleC, a spore cortex hydrolase that degrades the cortex peptidoglycan, thereby activating the protein [39–42]. In C. difficile, CspB and CspA are encoded by one ORF, cspBA, and the resulting protein is post-translationally autoprocessed by CspB into CspB and CspA and then further processed by the sporulation specific protease, YabG (CspA and CspC do not undergo autoprocessing) [14, 43–45]. CspC is encoded downstream of cspBA and is part of the same transcriptional unit. Interestingly, the catalytic residues in CspA and CspC are mutated, rendering these proteins catalytically inactive suggesting that only the CspB protein can process proSleC to its active, cortex degrading form [38, 45]. Although present in the spore, and essential for C. difficile spore germination, the CspA pseudoprotease has only been shown to regulate the incorporation of CspC into the spore [43, 44].
In our working model for spore germination, activation of CspC by TA leads to the activation of the CspB protein which cleaves proSleC into its active form. Activated SleC degrades cortex and the core releases CaDPA in exchange for water by a mechanosensing mechanism [46]. Because the receptor with which amino acids interact is unknown, we sought to screen for chemically-generated mutants that have altered amino acid requirements during spore germination (similar to the strategy used to identify the bile acid germinant receptor [38]). Here, we report that a mutation in the yabG gene results in strains whose spores no longer respond to or require co-germinants but respond to TA alone as a spore germinant. We hypothesize that the misprocessing of CspBA in the yabG mutant leads to this phenotype and provide evidence that short deletions in CspA alter the requirements for a co-germinant during spore germination. Our results implicate CspA in the recognition of amino acid and divalent cation co-germinants.
C. difficile strains were grown on BHIS agar medium (Brain heart infusion supplemented with 5 g / L yeast extract and 1 g / L L-cysteine) in an anaerobic chamber (Model B, Coy Laboratories Grass Lake, MI) at 37°C (85% N2, 10% H2, and 5% CO2). Antibiotics were added as needed (15 μg / mL of thiamphenicol, 10 μg / mL lincomycin, and 20 μg / mL uracil). Deletion mutants were selected on C. difficile minimal medium (CDMM) supplemented with 5 μg / mL 5-fluoroorotic acid (FOA) and 20 μg / mL uracil [47]. Bacillus subtilis was used as a conjugal donor strain to transfer plasmids into C. difficile and was grown on LB medium with 5 μg / mL of tetracycline and 2.5 μg / mL of chloramphenicol. Conjugation was performed on TY medium (3% Bacto Tryptone, 2% yeast extract, and 0.1% thioglycolate) with or without uracil. E. coli DH5α, E. coli BL21(DE3) and E. coli MB3436 were grown on LB medium supplemented with 20 μg / mL chloramphenicol and / or 100 μg / mL ampicillin at 37°C. All strains are listed in Table 1.
C. difficile spores were purified as described previously [32, 33, 46, 48]. Briefly, the strains without plasmids were grown on BHIS agar medium while strains with plasmid were grown on BHIS agar medium supplemented with 5 μg / mL thiamphenicol and allowed to grow for 4 days. Cells from each plate were scraped into 1 mL sterile water and incubated at 4°C overnight. Next, the cells were washed five times with water and combined into 2 mL total volume. The washed spores were layered on top of 8 mL of 60% sucrose and centrifuged at 4,000 x g for 20 minutes. The supernatant was discarded, and the spores were washed five more times with water and incubated at 4°C until use.
An overnight culture of wild type C. difficile UK1 vegetative cells was diluted to an OD600 of 0.05 into two, separate 15 mL falcon tubes containing BHIS liquid and grown for 3–4 hrs. To one of the cultures, ethyl methane sulfonate (EMS) was added to a final concentration of 50 μg / mL; the other culture was untreated for use as a negative control. The cultures were grown for 3 hours and then centrifuged at 3,000 x g for 10 min. The supernatants were discarded, and pellets were washed two more times with BHIS. After the final wash, the pellets were suspended in 40 mL BHIS medium and allowed to recover overnight. The recovered EMS-treated cells were then plated onto 10–12 BHIS plates (25 μL on each plate) to produce spores. Spores were purified from the EMS-treated strain as described above. Purified spores were heat activated at 65°C for 30 minutes before enrichment as shown in Fig 1A. The EMS-treated spores were treated for 15 minutes with HEPES buffer (50 mM HEPES, 100 mM NaCl, pH 7.5) supplemented with 10 mM TA and 10 mM betaine and then washed twice with buffer. Germinated, washed spores were plated onto BHIS agar medium for spore formation. This procedure was repeated iteratively for 4–5 times before isolating candidate strains for phenotypic screening.
The mutant spores were initially characterized by measuring the CaDPA release from germinating spores. Spores were heat activated at 65°C for 30 minutes and suspended in water at an OD600 = 50. The spores were then added to final OD of 0.5 in 100 μL final volume of HEPES buffer containing 250 μM Tb3+, 10 mM TA and / or 10 mM betaine in a 96 well plate. The CaDPA release was measured using a SpectraMax M3 plate reader for 30 minutes at 37°C with excitation at 270 nM and emission at 545 nM with a 420 nM cutoff.
The phenotype of the yabG mutant (RS08) and cspBA deletion mutants were determined by measuring changes to the OD600 and release of DPA. OD600 was monitored at 37°C for 1–2 hrs. using a plate reader. The spores were added to a final OD of 0.5 in HEPES buffer supplemented with 30 mM glycine alone or 10 mM TA alone or 10 mM TA and 30 mM glycine in 100 μL final volume. CaDPA release was measured as described above with a spore solution at a final OD600 of 0.25.
High-quality genomic DNA was purified from logarithmically growing C. difficile cells [49]. Genomic DNA from the EMS-mutant strains and the wild type parent was sent for paired end sequencing at Tufts University Genomics Core. Reads were aligned to the R20291 genome using DNASTAR software and SNPs determined (DNASTAR, Madison, WI). RAW sequence (fastq) reads were uploaded to the NCBI Sequence Read Archive as follows (Table 2): C. difficile UK1 parent (SRS3677310), Mutant 20C (SRS3677309), Mutant 27E (SRS3677307), Mutant 30A (SRS3677308), Mutant 30C (SRS3677311), Mutant 31D (SRS3677312).
The oligonucleotides used for making the strains and plasmids used in this study are listed in S1 Table. yabG and sleC mutants were created in the C. difficile R20291 background by using the TargeTron mutagenesis system. The potential insertion sites for targeting the group II intron were found using the Targetronics algorithm (Targetronics, LLC.) and a gBlock (Integrated DNA Technologies, San Jose, CA) of the group II intron targeted to yabG at the 279th nucleotide, relative to the start codon, was ordered (S1 Table) and cloned into pJS107 at the HinDIII and BsrGI restriction sites using Gibson assembly. The ligation was then transformed into E. coli DH5α. The site to engineer a TargeTron insertion in sleC (pCA6) was previously described for C. difficile UK1 [48] and the same protocol was used to engineer the insertion into R20291. The TargeTron insertion plasmids were isolated from E. coli DH5α and transformed into E. coli MB3436 (a recA+ strain) and then into B. subtilis BS49. Conjugation between B. subtilis and C. difficile R20291 was performed on TY-agar medium for 24 hrs. before plating onto selection plates. Once the plasmid was inserted into C. difficile, the colonies were screened for tetracycline-sensitive and thiamphenicol-resistant colonies and confirmed with PCR. yabG or sleC TargeTron mutants were selected by plating the colonies onto BHIS supplemented with lincomycin. Lincomycin-resistant colonies were tested by PCR and confirmed by sequencing the mutation. yabG and sleC TargeTron mutants were renamed as RS08 and RS10, respectively (Table 1).
A yabG complementing plasmid (pRS97) was created using primers 5' XbaI_Prom_YabG and 3' YabG_XhoI to amplify the yabG promoter and yabG coding regions and cloned into pJS116. The complementing plasmid was then inserted into C. difficile RS08 by conjugation with B. subtilis, as described above.
All recombinant proteins were expressed from the pET22b plasmid. For expression of the indicated yabG and sleC alleles, the indicated alleles were amplified from the C. difficile R20291 background. Plasmids were created by Gibson assembly of the amplified fragments into pET22b using either NdeI and BamHI (pAC28, pAC35, pAC43, and pAC42) or NdeI and XhoI (pKS08 and pAC41) restriction sites, then transformed into E.coli DH5ɑ for isolation and subsequently into E.coli Rosetta BL21 pLysSRare for protein expression. pAC28 and pAC35 were constructed using oligonucleotides pet22b_YabG_Fp and pet22b_YabG_Rp and used either C. difficile R20291 or Mutant 1 (Table 2) as template. pAC41 was constructed using oligonucleotides 5’pET_SleC and 3’pET_SleC using C. difficile RS31 as template. pAC43 was constructed using oligonucleotides pet22b_SleC_Fp with Pet22b_SleC_QSELI DEL_Rp and Pet22b_SleC_QSELI DEL_Fp with pet22b_yabg_SleC6his_Rp. Finally, pAC42 was constructed with oligonucleotides pet22b_YabG_Fp with YABG_slec_Rp and yabg_SLEC_Fp with pet22b_yabg_SleC6his_Rp.
To engineer the required site-specific deletions, the pyrE-mediated allelic exchange strategy was used with the C. difficile CRG2359 strain (R20291 ΔpyrE) [47]. Briefly, 1 kb upstream and 1 kb downstream fragments that surround the desired mutation in cspBA or preprosleC were cloned into pJS165 using primers listed in S1 Table. The plasmids were inserted into C. difficile CRG2359 strain using B. subtilis conjugation as described above. The strains containing the plasmids were then passaged several times to encourage the formation of single recombinants before passing onto CDMM-FOA-uracil medium. Thiamphenicol-sensitive candidate strains were tested by PCR for the desired mutations and confirmed by sequencing for the mutagenized regions. Where indicated, pyrE was restored to wild type using pRS107 (Table 3).
Samples were prepared for CspB, CspC, and SleC western blot by extracting soluble proteins from 2 x 109 / mL spores [R20291, yabG::ermB, yabG::ermB pRS97 (pyabG)]. For the protein standard, recombinant CspB, SleC and CspC proteins were purified using a previously described protocol [33]. Standard amount of protein or number of spores were solubilized in NuPAGE sample buffer (Life Technologies) and heated at 95°C for 20 minutes. Equal volume of spore extracts and recombinant CspB, CspC or SleC standard proteins were separated by SDS-PAGE. Proteins were then transferred onto low-fluorescence polyvinylidene difluoride membrane (PVDF) at 30V for 16 hours. The membranes were then blocked in 10% skimmed milk in TBS (Tris-buffered saline) and washed thrice with TBS containing 0.1% (vol / vol) Tween-20 (TSBT) for 20 minutes each at room temperature. The membranes were then incubated with anti-CspB, anti-CspC or anti-SleC antibodies for 2 hours and washed with TSBT thrice. For the secondary antibody, AlexaFluor 555-labeled donkey anti-rabbit antibody was used to label the membranes for 2 hours, in the dark. The membranes were washed again, thrice, with TBST, in the dark, and scanned with GE Typhoon Scanner using Cy3 setting, an appropriate wavelength for the Alexa Flour 555 fluorophore. The fluorescent bands were quantified using ImageQuant TL 7.0 image analysis software. Intensity of the extracted protein in each blot was compared to the standard curve that was generated from the recombinant protein included on each blot.
To analyze SleC activation, equal number of spores were suspended in HEPES buffer supplemented with 30 mM glycine or 10 mM TA or 10 mM TA and 30 mM glycine and incubated at 37°C for 1 hr. to 2 hrs. (aerobically). The samples were then centrifuged at 15,000 X g for 1 minute and pellets were suspended in NuPAGE sample buffer and heated for 20 min at 95°C. The suspension was centrifuged at max rpm for 10 min. The supernatant was separated and transferred into new tubes. The samples were stored at -20°C until use. For CspC, CspA and CspB western blots, an equal number of spores were suspended in HEPES buffer and boiled in NuPAGE buffer to extract the protein and loaded in 10% SDS PAGE gel. The spore extracts were then transferred onto nitrocellulose membrane for western blot analysis.
Plasmids (pAC41, pAC42, pAC43) were transformed into Rosetta strain E.coli [pKS08 was transformed into BL21(DE3)] and incubated overnight at 37°C on LB agar plates containing chloramphenicol and ampicillin. Plates were scraped into 1 mL LB and used to inoculate 1 L of 2XTY medium supplemented with chloramphenicol and ampicillin in baffled flasks, such that the starting culture OD600 = 0.01. Cultures were incubated at 37°C until the OD600 measured between 0.6 and 0.7, at which point they were induced with 250 μL of 1 M IPTG then returned to the incubator for an additional overnight (12–16 hours) incubation at 16°C. Cultures were pelleted by centrifugation at 4°C for 30 min at ~6,000 x g. The supernatant was discarded and the culture pellets were frozen at -80°C until use. The cells were lysed in 300 mM NaCl, 50 mM Tris-HCl, pH 7.5, 15 mM imidazole, and 10% glycerol. One liter of culture pellet was resuspended in 25 mL of lysis buffer supplemented with PMSF, lysozyme and DNase I and rocked on ice for 30 min prior to sonication. Samples were then centrifuged for 30 min at 6,000 x g, 4°C and the supernatant was combined with 1 mL of Ni-NTA beads. Samples were rocked overnight at 4°C. Beads were washed twice with 300 mM NaCl, 50 mM Tris-HCl, pH 7.5, 30 mM imidazole, and 10% glycerol and then eluted in the same buffer but supplemented with 500 mM imidazole. Samples were concentrated to <1 mL using a 10K MWCO centrifugal device and the recombinant protein further purified by FPLC, after which they were again concentrated.
Plasmids containing wildtype (pAC28) or mutant (pAC35) yabG were expressed slightly differently for the SleC-YabG incubations. Plasmids were also transformed into the Rosetta E. coli strain, and then cultured at a starting OD600 = 0.01 in a 50 ml volume of LB-CM/AMP until OD600 reached between 0.6 to 0.7 at which point they were induced with 12.5 μL of 1 M IPTG then returned to the incubator for an additional hour at 37°C. YabG cultures were then pelleted by centrifugation at 6,000 x g at 4°C for 30 min. Supernatant was discarded and pellets were resuspended in 4 mL lysis buffer and sonicated. This bacterial lysate was immediately used in the incubation with the indicated, recombinantly expressed and purified, SleC alleles.
All germination assays were performed in technical triplicate of biological duplicates and data points represent the averages from these data sets. Error bars represent the standard error of the mean. A 1-way ANOVA with Tukey’s multiple comparisons test was used to compare the quantified protein amounts. For quantification of proteins, each blot was loaded with 5 standard proteins and three spore samples.
In order to identify the receptor with which amino acid co-germinants interact, we used a strategy that was previously used to identify the bile acid germinant receptor (CspC) [38]. Although other strategies, such as Tn-seq, could be used to generate random mutations, most of these will result in germination null phenotypes and do not permit the screening of subtler phenotypes. As shown in Fig 1A, wild type C. difficile UK1 vegetative cells were exposed to EMS and recovered. Purified spores derived from the mutagenized bacteria were then germinated in buffer supplemented with 10 mM TA and 10 mM betaine. The structural difference between glycine and betaine is the presence of three methyl groups attached to the N-terminus rather than two hydrogen atoms. Because betaine is a glycine analog and does not stimulate spore germination when added with TA [50], we hypothesized that we could isolate change-of-function mutants that recognize betaine as a germinant or those that no longer require glycine as a co-germinant. Spores incubated in the presence of buffered TA and betaine were then plated and allowed to form spores. Potential mutants were enriched with this strategy 4–5 times before isolating colonies and screening for phenotypes. Across several mutagenesis experiments, the most commonly-observed phenotypes were strains that did not require the co-germinant glycine to germinate and germinated in response to taurocholate only (TA-only). As shown in Fig 1B, wild type C. difficile UK1 spores required both TA and glycine to stimulate the release of CaDPA from the core. However, spores purified from isolates derived from separate EMS mutageneses released CaDPA in the presence of TA only (Fig 1C, 1D and 1E). Importantly, though, these mutants still responded to glycine, the germination efficiency / rate increased upon glycine addition. However, none of the mutants recognized betaine as a co-germinant. To identify the mutation(s) that caused this phenotype, five different mutant strains (isolated from 4 independent EMS mutageneses) and a wild-type control were sent for genome re-sequencing. Surprisingly, when the sequences of the mutant strains were compared to the sequence of the wild-type parent, we identified SNPs common to all 5 mutants in the coding region or the promoter region of yabG, coding for a sporulation-specific protease (Table 2).
In the screen to generate the EMS mutants, the C. difficile UK1 strain generates more spores than our C. difficile R20291 isolate, but it is more difficult to genetically manipulate. Thus, to create the yabG mutations, we switched to the closely-related C. difficile R20291 strain. To confirm that the TA-only phenotype is caused by a mutation in yabG, we inserted a group II intron into the yabG gene of C. difficile R20291 using TargeTron technology [38, 51, 52]. Germination of the C. difficile yabG::ermB mutant (RS08) spores was compared to wild type using both OD600 and CaDPA release assays (Fig 2). As shown in Fig 2A, wild-type C. difficile R20291 spores required both TA and glycine in order to germinate. Interestingly, though the mutant strain germinates slower, the yabG::ermB mutant spores germinated in response to TA-only and this germination phenotype was not enhanced by the addition of glycine (in contrast to the phenotype of the EMS-mutant spores) (Fig 2B). We next tested if spores derived from the yabG::ermB mutant responded to other co-germinants or enhancers of germination. C. difficile R20291 spores initiated germination in response to TA and L-alanine or TA and CaCl2. Significantly, because spores derived from the yabG::ermB mutant do not respond to L-alanine or to CaCl2, this suggested a complete loss of co-germinant recognition. When the mutation was complemented in trans by expression of yabG from a plasmid (pRS97), the spores again recognized glycine or L-alanine or calcium as a co-germinant (Fig 2C). The TA-only phenotype in the mutant spores was also confirmed by CaDPA release and compared to spores from both wild type and the complemented strain (Fig 2D, 2E and 2F). These results indicate that the yabG::ermB mutant spores do not respond to co-germinants and germinate in response to TA alone.
In prior work, a yabG mutant strain accumulated unprocessed CspBA into spores [43]. To confirm that the generated yabG mutant results in the accumulation of CspBA, we purified spores derived from R20291, yabG::ermB and the complemented strain and separated the extracted protein by SDS-PAGE. The separated protein was then detected using immunoblotting with CspB-specific antisera (S2 Fig). Although the blot showed that little CspBA is still remaining in both the wild type and complemented strains, the yabG mutant had mostly the unprocessed, CspBA form. The CspA western blot showed that the yabG mutant had only the CspBA form. The mutation in yabG did not appear to affect CspC incorporation into the spores although these results are not quantitative.
YabG also was shown to be required for the processing of preproSleC into proSleC [43]. Indeed, whereas the wild-type and complemented strains incorporated into spores the processed, proSleC, form, only preproSleC was incorporated into the yabG::ermB strain (Fig 2G). When tested for the processing of SleC during germination, the wild-type and the complemented strains required both TA and glycine or TA and L-alanine or TA and calcium to activate SleC (Fig 2G). However, C. difficile yabG::ermB activated SleC in response to TA alone. These results confirm the TA-only phenotype observed in Fig 2B and 2E and suggest that a protein that is not processed in the yabG mutant strain is involved in germinant recognition or regulating germinant specificity.
Because YabG is a sporulation-specific protease, it is possible that the deletion of this protease might alter the amount of germination-related proteins (e.g., CspB, CspC, CspA or SleC) that are incorporated in the spore thereby providing the observed phenotype (i.e., increasing the abundance of the germinant receptors could lead to an increase in germinant sensitivity and loss of regulation). Using the previously described method to quantify the protein levels in C. difficile spores [33], we quantified the abundance of CspB (and CspBA), CspC and SleC in C. difficile yabG::ermB and compared them with the abundances in the wild-type and complemented strains (Fig 3). Briefly, equal amounts of spores were boiled in SDS sample buffer to extract the soluble proteins and loaded in the SDS-PAGE gel with recombinantly expressed and purified protein as standards. The separated samples were transferred to low fluorescence-PVDF membrane for quantitative western blot. Membranes were blocked and proteins labeled with antisera specific for the indicated protein. Primary antibodies were detected by incubating with AlexaFluor 555 labeled secondary antibody in the dark. The fluorescently labeled antibodies were detected by scanning on a Typhoon scanner. The fluorescent signal was quantified and compared to the standard curve generated from the recombinant standards included on each gel.
Introducing the yabG complementing plasmid resulted in significantly increased abundance of proSleC and CspC into mature spores compared to the yabG::ermB strain (Fig 3A; only the preproSleC form is incorporated into the yabG::ermB strain) or the wild-type strain (Fig 3B), respectively. There were no statistical differences in the abundance of SleC or CspC between the wild-type and mutant strains; despite the apparent difference between the WT and yabG::ermB strains, the CspC abundance did not meet statistical significance. Importantly, there was a statistical difference in the abundance of CspB in all pair-wise comparisons of the wild type and complemented strain (there was no quantifiable CspB protein in spores derived from the yabG::ermB strain; Fig 3C). Finally, spores derived from the yabG::ermB strain had significantly more CspBA protein than did the wild-type or the complemented strains (Fig 3D). Because spores derived from the complemented strain had increased abundances of proSleC (Fig 3A), CspC (Fig 3B), and CspB (Fig 3C), but did not produce a TA-only phenotype, this suggests that increased abundance of these proteins is not the reason for the observed TA-only phenotype in the yabG mutant strain. Importantly, though, the yabG mutant accumulated much more CspBA into spores than did the wild-type or the complemented mutant strains (Fig 3D) and only accumulated the preproSleC form (Fig 2G). Therefore, we hypothesized that the presence of full-length CspBA and / or preproSleC could contribute to the observed TA-only phenotype.
Because CspB and CspC are already known to be involved in regulating C. difficile spore germination [38, 43–45, 48], we chose to first focus on the potential processing of CspBA by YabG. In prior work by Kevorkian et al. [43], the CspBA processing site was hypothesized to occur at or near amino acid 548. To test this hypothesis, we deleted from the C. difficile CRG2359 (C. difficile R20291 ΔpyrE) chromosome 12 aa between CspB and CspA (cspBAΔ548–560) using pyrE-mediated allelic exchange. After confirmation of the engineered mutation in the CRG2359 genome, the pyrE gene was restored, and germination of the resulting strain was compared to pyrE-restored CRG2359 strain (C. difficile RS19; S3A Fig. We found that the CspBAΔ548–560 allele did not have any effect on germination (S3B Fig).
Next, to determine if deletions in the coding region in or between cspB and cspA affect spore germination, we deleted various regions within the cspBA gene in the CRG2359 strain (Fig 4A). The results of the germination phenotypes of the various deletions are shown in Fig 4. Deletion of 26 codons from the C-terminus of cspB (RS20; CspBAΔ522–548) did not affect spore germination (Fig 4C) when compared to the wild-type CRG2359 strain (Fig 4B). These results suggest that the C-terminus of CspB is not involved in generating the TA-only phenotype. Interestingly, deletion of 26 codons at the N-terminus of cspA (RS21; CspBAΔ560–586) resulted in spores that germinated in response to TA-only, after 60 minutes of incubation in the germination solution (Fig 4D). Though deletion of 25 codons in between the cspB and cspA coding sequences did not result in a TA-only phenotype (RS23; CspBAΔ542–567) (Fig 4E), extending the deleted region by another 9 codons into the surrounding region (RS24; CspBAΔ537–571) resulted in spores that germinated in response to TA-only, again after 60 minutes of incubation (Fig 4F). Importantly, the spores derived from the RS21 and RS24 strains still respond to glycine as co-germinant, despite also germinating in response to TA-only. Because the TA-only phenotype appeared to be enhanced as more cspA was deleted (RS21 to the RS24 strain), we predicted that a larger deletion might result in a phenotype similar to that observed in the yabG mutant (which does not recognize co-germinants). We found that when 50 codons were deleted from the N-terminus of CspA (RS27) the spores no longer germinated (Fig 4G). These results were confirmed by analyzing the release of CaDPA from the germinating spores (S4 Fig).
During construction of the deletion strains, we encountered significant difficulties in restoring the pyrE allele to wild type. To circumvent this obstacle, and to understand if C. difficile spore germination is affected by the pyrE deletion, we restored pyrE in the RS21 strain (CspBAΔ560–586, pyrE+; RS26) and compared with the RS19 strain (CRG2359 with restored pyrE). As shown in S5A Fig, C. difficile RS19 required both TA and glycine to germinate but the RS26 strain germinated in response to TA-only (S5B Fig). These results were confirmed by analyzing the release of CaDPA from the spore (S5C and S5D Fig). Finally, we analyzed the activation of proSleC to SleC in response to TA alone. Only the RS26 strain cleaved proSleC in response to TA alone (S5E Fig). These observations are identical to the observations made for the parental RS21 strain (Fig 4 and S4 Fig) and indicate that the pyrE allele does not influence C. difficile spore germination in the context of these studies.
To confirm our observations that the strains germinated in response to TA alone, we analyzed by western blot the activation of SleC (Fig 5A). SleC activation in response to 10 mM TA only occurred in CspAΔ560–586 and CspAΔ537–571 while CspAΔ560–610 did not germinate in response to TA and glycine. The CspB western blot showed that CspBA was processed to CspB and CspA in all of the mutants, compared to wild type (Fig 5B). The CspC western blot did not reveal any differences between the wild type and mutant strains, except for the RS27 (CspAΔ560–610) strain where no CspC was detected.
Previous work has shown that CspA has a role in either incorporating CspC or stabilizing CspC in mature spores [43]. The data presented above suggests that the N-terminus of CspA is important for regulating spore germination. We had hypothesized that this region is processed by YabG but all the generated cspA alleles generated a protein that was processed into the CspB and CspA forms (Fig 5). One way to identify the YabG processing site in CspA is by pulldown of CspA from the spore extract and sequencing the CspA protein using mass spectrometry or Edman sequencing. Unfortunately, the CspA antibody was unable to immunoprecipitate CspA from the spores due to the quality of the antibody. However, we predicted that the YabG processing site in CspBA might be conserved in preproSleC. Instead of immunoprecipitating CspA, we immunoprecipitated proSleC from spore extracts derived from wild-type spores and sleC mutant spores (as a negative control) (Fig 6A). Using the sample from the proSleC pull down, we identified fragments by mass spectrometry of trypsin-digested proSleC (Fig 6B). In this experiment, the most N-terminal fragment identified began with glutamine120 followed by serine121 (an SRQS sequence), suggesting that YabG cleaved after R119. However, because the protein was digested with trypsin, which cleaves after arginine and lysine residues, this N-terminal glutamine could be the result of trypsin cleavage and not YabG processing of preproSleC during spore development. Therefore, we recombinantly expressed and purified preproSleC from E. coli. This recombinant protein was incubated with an E. coli lysate that express yabG from an IPTG-inducible promoter for 30 minutes, 1 hour, 3 hours or in buffer alone for 3 hours (Fig 6C). This lysate efficiently processed preproSleC to a form consistent with the removal of the pre sequence and this did not occur in buffer alone. Interestingly, when preproSleC is incubated in E. coli lysate that expressed one of the yabG alleles identified in the EMS-mutagenesis (yabGA46D), preproSleC was not processed efficiently. This indicates that the EMS-generated mutant expresses a less active YabG protein and that the in vitro processing of the preproSleC is due to YabG and not some other E. coli protease. Using the protein processed by wildtype YabG, we submitted the processed protein for Edman sequencing (Fig 6D).
Edman sequencing of the YabG-processed preproSleC protein revealed that the first five amino acids were QSELI (Fig 6D). This sequence matched the mass spectrometry data and confirms that YabG processes preproSleC after R119. These results are similar to what is observed in C. perfringens preproSleC cleavage (C. perfringens preproSleC is processed after R114) [53]. Next, we tested the impact of deletion of the SRQS sequence or the QSELI sequence from preproSleC. These protein alleles were recombinantly expressed and purified and then incubated with E. coli lysate that expressed yabG. The resulting protein was submitted for Edman sequencing. Surprisingly, we found that deletion of the SRQS sequence resulted in YabG cleaving after R115 and the N-terminus of proSleC being SFELI and deletion of QSELI also resulted in YabG cleaving after R115 and the N-terminus being SFSRF. These results indicate that deletion of the YabG cleavage site, or surrounding region, results in YabG processing the protein near the same site. These results suggest that YabG may recognize the structure of its substrates and not a specific amino acid sequence.
When we compared this identified SRQS sequence to the protein sequence in CspA, we found that within the N-terminus of CspA there was a SRQS amino acid sequence that was encompassed within the deletions found in the RS21 strain (CspAΔ560–586; a strain that generated a TA-only phenotype). We deleted the nucleic acid sequence that encodes this SRQS motif in sleC, cspBA, or in both genes (Fig 7A). Surprisingly, the deletion of the SRQS site in CspA resulted in spores that germinated in response to TA-only within 40 minutes after germinant addition (Fig 7B). However, deletion of the SRQS motif within preproSleC did not affect spore germination (Fig 7C). When these two deletions were combined in the same strain, the spores had a TA-only phenotype similar to that of spores with a deletion in CspA alone (Fig 7D and 7A). Next, we confirmed these phenotypes by analyzing the release of CaDPA from the spore (S6 Fig). Again, only when the cspAΔSRQS allele was incorporated into spores did the resulting strains release CaDPA in response to TA-only (S6A and S6C Fig); sleCΔSRQS spores required both TA and glycine to release CaDPA (S6B Fig).
We next analyzed the activation of SleC for these deletions. proSleC was activated in response to TA-only in the RS29 (cspAΔSRQS) and the RS31 (cspAΔSRQS + sleCΔSRQS) strains but not in the RS30 (sleCΔSRQS) strain (Fig 7E). There was no difference in the CspB or CspC incorporation in these SRQS deletion mutants compared to wild type (CRS2359; Fig 7F). However, as predicted by the processing of preproSleC by YabG (Fig 6), deletion of the SRQS sequence in CspA resulted in processing of CspBA to CspB and CspA (Fig 7F). Our results indicate that due to mis-processing of CspBA, or alterations within the cspA sequence, spores lose the requirement for co-germinants during spore germination.
Previous studies on C. difficile spore germination have hypothesized the presence of an amino acid co-germinant receptor [13, 31, 32, 50]. We used EMS mutagenesis to screen for C. difficile mutants whose spores recognize betaine as a germinant. Excitingly, we isolated mutants that did not require an amino acid as a co-germinant, and these EMS-mutants germinated in response to TA-only. However, these mutants still recognized co-germinants; the rate of germination increased upon the addition of co-germinant to the germination solution. We traced the SNP to the yabG locus and confirmed that the TA-only phenotype observed in the EMS-generated mutants was due to the yabG allele. In contrast to the EMS-generated mutants, the yabG mutant spores did not recognize glycine as a co-germinant, and germinated in response to TA-only (Fig 2B and 2C). Oddly, we noticed that spores derived from the complemented strain released some CaDPA in response to TA alone, but this was not observed when germination was measured at OD600 (Fig 2F vs. 2E). We further analyzed this observation and found that the supposed CaDPA release observed in the complemented strain in the presence of TA alone, was due the presence of Tb3+ in the assay (Tb3+ is absent from the germination solution in the OD600 assay). We have submitted a separate manuscript to report this in detail.
To understand if spores derived from the yabG mutant recognized other amino acids as co-germinants, we tested L-alanine and calcium. L-alanine is the second-best amino acid co-germinant during C. difficile spore germination with an EC50 value of 5 mM [32, 50]. Because C. difficile RS08 (yabG::ermB) spores did not respond to glycine or L-alanine or CaCl2 (Fig 2), these results suggest that the protein(s) YabG processes is / are responsible for recognition of the various co-germinant(s).
YabG has been mostly studied in B. subtilis. In B. subtilis YabG is a sporulation-specific protease, and a yabG mutation causes alterations in the coat proteins of B. subtilis spores. The orthologues of most B. subtilis YabG target proteins are absent in C. difficile (e.g., CotT, YeeK, YxeE, CotF, YrbA) [54–56]. In a previous report, the processing of CspBA and preproSleC during C. difficile sporulation was shown to be YabG-dependent [43]. However, germination was not significantly altered in the mutant spores compared to wild type spores (germination efficiency decreased from 1 to 0.8 in mutants, when analyzing CFU counts) [43]. Importantly, though, germination efficiency was only tested on BHIS-TA agar medium and not in the presence of TA alone [43].
In prior work by Kevorkian et al. [43], the authors suggested that the CspBA processing site is near-amino acid 548 and is encoded by a linker DNA sequence between cspB and cspA. To test this, we deleted the codons encoding amino acids 542–567 regions in cspBA (RS23). When the germination phenotype of the RS23 strain was compared with CRG2359, we did not observe a TA-only phenotype (Fig 4B and 4E) and, importantly, CspBA was still efficiently processed (Fig 5). These results suggest that the predicted 548–560 region as the YabG-dependent processing site is not accurate, or that when the YabG processing site is removed, YabG cleaves the CspBA protein at an alternate site. In support of this hypothesis, we found that deletion of the YabG processing site in preproSleC resulted in YabG cleaving positionally within the protein. This is likely to be the case for CspBA processing as well. Further work is needed to understand how a protein that is found in all spore-forming bacteria, YabG, recognizes its substrates, which are diverse.
When the identified SRQS sequence was deleted from CspA (cspBAΔ580–584), we observed a TA-only phenotype in the spores suggesting that the SRQS region might be important for CspA activity. Both RS21 (cspBAΔ560–586) and RS24 (cspBAΔ537–571) strains generated spores with TA-only phenotypes (Fig 4) similar to the SRQS deletion in CspA (Fig 7). However, only the RS21 strain has the deletion of SRQS region, while the SRQS motif is still present in the RS24 strain.
In a recent review on C. difficile spore germination [13], the authors build upon a hypothesized model for spore germination [10] and propose a new, “lock and key” model for C. difficile spore germination. In the “lock and key” model, CspA and CspC are localized in the coat layer, where TA can bind to CspC. Activation of CspC leads to the transport of glycine and Ca2+ through the outer membrane, by an unknown protein, to the cortex where CspB is held inactive in a complex with GerS and proSleC. In the absence of calcium, glycine is transported to the inner membrane where it activates the release of Ca2+ from the core through another unknown process. The released Ca2+ traffics to CspB to activate its protease activity. In the presence of calcium and glycine, both are transported in and Ca2+ activates CspB. When glycine and calcium bind to CspB, CspB can activate proSleC to degrade the cortex [13]. In the first, favored, model, the germinosome complex composed of CspC, CspA, CspB, and proSleC are anchored to the outer spore membrane by GerS (a lipoprotein that is required for spore germination [57]—see below) [10, 13]. Upon binding of a germinant (TA with glycine or Ca2+), CspC and CspA are released from this germinosome complex and CspB becomes free to activate proSleC to degrade the cortex. Subsequently, CaDPA is released from the spore core by a mechanosensing mechanism [26, 46].
Our results support the first model [10] whereby CspB, CspA, and CspC are in a germinosome complex–similar to the germinosome complex found in B. subtilis [58]. Recent studies on GerS have shown that GerS likely does not form a complex with other germination proteins [57, 59]. Rather, it is required to generate cortex-specific modifications in the spore and thus, important for germination of the spore because the SleC cortex lytic enzyme depends on cortex-specific modifications to degrade the cortex layer efficiently [59]. Based upon prior work from our lab, proSleC is unlikely to be part of this complex because SleC is three to four times more abundant than CspB or CspC, depending upon the strain analyzed [33].
The findings by Kochan and colleagues that calcium synergizes with other co-germinants during C. difficile spore germination are interesting [37]. It is possible, however, that what the authors are observing is that the population of spores is germinating in response to TA and amino acids or in response to TA and calcium and that individual spores within that population are responding to either glycine or to calcium as co-germinants. Because spores derived from the yabG::ermB strain lose all requirements for co-germinant (either amino acids or calcium), we hypothesize that CspA is recognizing all co-germinants (amino acids or divalent cations) for C. difficile spores. Though the evidence for this hypothesis is entirely genetic, this hypothesis provides a model that can be tested biochemically (once reagents are built for such experiments).
The processing of CspBA depends on the YabG protease and yabG-mutant spores incorporate mostly the full length CspBA protein. These spores do not recognize amino acids as co-germinants and germinate in response to TA-only (Fig 2B). We hypothesize that CspC (functioning as the bile acid germinant receptor) and CspA (functioning as the amino acid co-germinant receptor) inhibit CspB activity within dormant spores (Fig 8A). These two pseudoproteases would regulate the activity of CspB so that it does not prematurely activate proSleC, potentially similar to how other pseudoproteases / pseudokinases regulate activity of their cognate proteins [60–64]. Should proSleC become activated prematurely, the C. difficile spore could germinate in an environment that may not support growth (e.g., absence of glycine, or other amino acids, or absence of divalent cations).
Pseudoenzymes have gained attention, recently, for regulating biological processes. For example, pseudokinases have been studied in both prokaryotes [e.g., Caulobacter crescentus DivL [65] or Streptococcus pyogenes RocA [66]] and eukaryotes [60–63] where they often regulate their cognate protease / kinase. Pseudoproteases, however, have mostly been studied in eukaryotes [64]. Prior work by Hershey and colleagues [67] has revealed that Magnetospirillum magneticum uses a pseudoprotease, MamO, to bind metals for magnetosome development. Combined with this study, our work provides growing evidence that pseudoproteases also regulate biological processes in prokaryotes.
Although some aberrantly processed CspB is detectable in the yabG mutant spore, the spore packages mostly full-length CspBA, where CspA is tethered to CspB. We hypothesize that in this scenario, CspC alone prevents CspB from cleaving proSleC into its active form, and TA might dislodge CspC from CspB (this is consistent with our prior publication that indicates that CspC may have an inhibitory activity during spore germination [33]) (Fig 8B). Moreover, our data suggest that the N-terminus of CspA might be important for formation of this hypothesized complex with CspC and / or CspB. When portions of the CspA N-terminus are deleted, the binding of CspA to the complex might become unstable causing CspA to randomly disassociate from the complex. This would result in TA alone stimulating germination by disassociating CspC from CspB (Fig 8B). Once CspB is free from the complex, it could then activate many proSleC proteins to maximize the germination process. Further work is needed to test the biochemical implications of this hypothesis.
|
Subsets and Splits