Sampling, DNA extractions and quantification

Arctic fox muscle tissue samples were obtained from various independent trappers and organizations as part of the University of Alaska Museum of the North tissue collections or as part of previous research [36; S1 Table]. No direct handling/sampling of animals took place for this study. Samples were stored at -80°C until processed. Samples were digested in 200 μL 1X lysis buffer (4 M Urea, 0.2 M NaCl, 0.5% n-lauroyl sarcosine, 10 mM ethylenediaminetetraacetic acid (EDTA), 0.1 M Tris HCl pH 8.0) with the addition of 20 μL proteinase K and incubated at 56°C for two hours. During digestion, samples were vortexed and briefly spun down every 30 minutes. DNA was extracted from the resulting lysate utilizing the DNeasy Blood and Tissue Kit (Qiagen) following manufacturer protocols with the exception that DNA was eluted in a total volume of 60 μL, using two 30 μL aliquots of TE buffer (10 mM Tris, 0.1 mM EDTA). Isolated DNA was quantified using the Quant-iT PicoGreen dsDNA Assay Kit (ThermoFisher Scientific) and quality assessed by ethidium stained 0.8% agarose gel electrophoresis (90 V for 45 minutes) where DNA fragment size was assessed in context of HighRanger 1 kbp DNA ladder (Norgen Biotek). A subset of 96 high molecular weight DNA samples suitable for sequencing were selected from three regions across the arctic fox’s distribution in North America (Fig 1; S1 Table). Samples from Hooper Bay and Chevak are referred to as a single region ‘Southwestern Alaska’ based on their geographic proximity to one another but were left ungrouped for neutral genetic analyses (~ 30 kilometers).

Library preparation, sequence capture and high-throughput sequencing

DNA libraries were prepared using Kapa HyperPlus Kit (Roche) following the SeqCap-EZ HyperCap UGuide V1.0 (Roche) protocol. Seven cycles were implemented as part of the pre-LM PCR as recommended by the manufacturer with the following modifications to the workflow: i) PCR-grade water was used for dilutions and elution, ii) samples were treated with 5 μL of conditioning solution during fragmentation, iii) TruSeq HT Dual-Index Adapters (Integrated DNA Technologies) were used in place of SeqCap Adapter Kits A and B (Roche), and iv) Illumina P5 and P7 primers (Integrated DNA Technologies) were substituted in place of the Pre LM-PCR Oligos 1 & 2 (Roche). At the end of the Pre-capture LM-PCR step, DNA library quality was assessed using ethidium bromide-stained gel electrophoresis as per above.

A 1 μg DNA multiplex was created from equal-molar amounts of each of the 96 libraries. Target enrichment was performed as previously described [35], using the designed SeqCap EZ Developer Library probe. Modifications to the enrichment protocol implemented in this study included: i) replacement of the NimbleGen Multiplex Hybridization Enhancing Oligo Pool (Roche) with 2 μL xGen Universal Blockers–TS Mix (Integrated DNA Technologies), ii) NimbleGen SeqCap EZ Developer Reagent (Roche) was used in place of the NimbleGen COT Human DNA (Roche) during hybridization sample preparation, and iii) hybridization was carried out at 47°C for 20 hours. A final product assessment was conducted with a bioanalyzer on the target-enriched multiplex before sequencing on an Illumina MiSeq V3 run using 2x300 bp reads (Advanced Analysis Centre Genomics Facility, University of Guelph).

Sequence alignment and variant annotation

Utilizing the bwa-mem command in Burrows-Wheeler Aligner v0.7.12 [44], paired-end reads for each of the 96 samples were aligned to the canine reference genome (CanFam3.1; Fig 1; S1 Table). After sequence metrics were obtained using SAMTOOLS v1.5 [45], the Genome Analysis Toolkit (GATK, V4.0.0.0) best practices pipeline and standard hard filtering parameters were used to perform duplicate sequence removal, SNP/INDEL variant annotation, genotyping, and variant recalibration [46–48]. The SelectVariants function was then used to compile a VCF file containing only bi-allelic SNPs.

The SeqCap EZ Developer Library probe was originally designed from a draft version of the red fox genome [33], thus positions of the probe-baited targets needed to be determined and converted into positions in the canine reference genome (CanFam3.1; Accession: PRJNA12384) using BLASTn (S2 Table). These targeted regions were compiled into a list of on-target intervals. On-target intervals were used to further categorize SNPs as being within coding regions (including 1,500 bp upstream from the start codon) or within intergenic (off-target; outside targeted coding regions and promoters) regions. We attempted to mitigate biases to identify loci under selection using F_ST outlier tests as recommended in the literature [49] by accounting for: linkage disequilibrium within datasets, method variation by implementing several F_ST outlier tests, and filtering variants for minimum allele frequency (MAF).

SNP filtering and analyses

Both sub datasets of SNPs from within coding regions, and those from intergenic regions were filtered using VCFtools v0.1.13 to retain only biallelic variants with a MAF threshold of 2.5%, a maximum missing genotype threshold (per site) of 20% and excluding variants on the X-chromosome. Additionally, both filtered sub-datasets were pruned for linkage disequilibrium as implemented by the SNPRelate package in R v.3.5 [50, 51] and further pruned for physical linkage to only retain SNPs ≥ 100kbp from one another using bcftools v1.9 [52] (S1 File for further details). SNPs from within intergenic regions underwent further filtering using the Ensembl Variant Effect Predictor [53] to retain only those SNPs ≥ 40 kbp from an annotated coding region. Disequilibrium and physical linkage pruning occurred after F_ST outlier tests for the sub-dataset composed of SNPs from within coding regions. Mantel tests (as implemented in R) [50] were performed on both the intergenic SNP-dataset and the SNP-dataset from coding regions to identify patterns of isolation-by-distance.

Raw sequence data

We obtained an average of ~ 563,000 raw reads per library, 99.5% of which mapped to the canine reference genome. After processing the raw data through the GATK SNP calling pipeline, an average of ~ 109,000 (19.35%) reads were filtered from each sample library leaving ~ 452,000 reads per library. An average of 56% reads per library mapped to probe targeted regions with an average of 54 X coverage across all 96 libraries (S3 Table).

SNPs in intergenic (off-target) regions

A dataset of 5,490,704 intergenic SNPs was filtered using Variant Effect Predictor, which was then pruned to minimize linkage disequilibrium. This yielded a dataset of 29 intergenic SNPs presumed to be not under selective pressure with an average depth of coverage of 15 X across all SNPs (average depth of coverage of 5 X when excluding 4 SNPs with coverage exceeding 15X), and a MAF of 2.5% (S4 Table). These data were visualized using PCA, DAPC and STRUCTURE. PCA did not identify genetic structuring as all clusters had extensive overlap with each other (S9 Fig). The DAPC identified K = 6 as the most likely number of clusters; however, STRUCTURE analyses of K = 2–6 showed high levels of admixture across all analyses and indicated an optimal K = 2 clusters (Fig 2). Power analyses of these 29 SNPs indicated a power of ~86% at an expected F_ST of 0.01 and a power of 100% at an expected F_ST of 0.05 or above, indicating a high likelihood that if population differentiation was > 1%, our dataset had the power to detect that difference. Isolation-by-distance was not observed within intergenic SNPs based on Mantel tests. We observed a total of 4 outlier SNPs (13 prior to linkage pruning) within the off-target dataset, all of which were identified by PCAdapt, and visualization of these data demonstrate no genetic clustering (S8 Fig). Overall, these analyses indicated no apparent genetic structure across the sample design for the off-target, and presumed neutral, SNP dataset.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Neutral genetic homogeneity of arctic fox across North America.

Analyses of the 29 presumed neutral SNPs after filtering with Variant Effect Predictor, MAF = 2.5%, and pruning for Linkage disequilibrium A) principle component analysis B) STRUCTURE analysis of K = 2 and 3 (top and bottom respectively), where each bar across horizontal axis indicates an individual, vertical axis depicts cluster assignment, and different colors depict each genetic cluster; Arviat = black square; Chevak = yellow circle; Hooper Bay = blue triangle; Victoria Island = green diamond.

https://doi.org/10.1371/journal.pone.0258975.g002

SNPs in coding (on-target) regions

We found 9,467 SNPs located within target intervals before filtering. After applying search criteria for 2.5% MAF, a maximum of 20% missing data, and discarding SNPs on the X-chromosome, 2,277 SNPs remained. F_ST outlier analyses on these 2,277 SNPs produced a sub-dataset containing 107 SNPs (S5 Table). After accounting for linkage disequilibrium, a final sub-dataset containing 22 F_ST outlier SNPs remained, several of which were associated with interleukins, Toll-like receptors, and the MHC (S5 Table). Based on Variant Effect Predictor results, four SNPs retained in the final sub-dataset associated with DLA-DQA, NOD1, RAG1 and TLR5 genes, were likely to convey a change in chemical characteristic of the encoded amino acid during translation (e.g., acidic to basic amino acid). A further 9 SNPs, that conveyed such missense changes, were removed from the final sub-dataset when filtering for linkage disequilibrium (S5 Table). DAPC and STRUCTURE analyses of the remaining 22 F_ST outlier SNPs were assessed identifying K = 2 clusters. We found arctic fox sampled from Alaska appeared genetically distinct from foxes from Northern Canada (Fig 3) with F_ST estimates of 0.127. F_ST confidence intervals indicated that F_ST between Southwestern Alaska and the two regions in Canada were significantly different from zero; however, F_ST between the two Canadian arctic fox populations was not significantly different from zero (S6 Table). It is important to note however, that we could not exclude isolation-by-distance as a potential factor contributing to these patterns (could not be calculated being only 2 geographic points–Canadian vs Alaskan samples).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Arctic fox immunogenetic structure differentiates sampled regions in Southwestern Alaska and Northern Canada.

Analyses of 22 protein-coding F_ST outlier SNPs after filtering for MAF = 2.5% and pruning for Linkage disequilibrium A) principle component analysis and B) STRUTURE analysis of K = 2 and 3 (top and bottom respectively), where each bar across horizontal axis indicates an individual, vertical axis depicts cluster assignment, and different colors depict each genetic clusters; Arviat = black square; Chevak = yellow circle; Hooper Bay = blue triangle; Victoria Island = green diamond (note, Chevak and Hooper Bay were pooled for this analysis).

https://doi.org/10.1371/journal.pone.0258975.g003

Estimates of pN/pS were determined for 90 of the initial 116 targeted genes (S7 Table) although for 17 of these genes, pN/pS ratios could not be calculated due to a lack of polymorphic substitutions (either nonsynonymous or synonymous) resulting in pN values equal to zero or division by zero errors (occurring when pS = 0). The pN/pS ratios for most genes were indicative of purifying selection; however, 10 genes (C2, DLA-12, DLA-79, DLA-DQA, DLA-DBQBC1, IL1B, IL23A, MYD88, STAT3, and STAT6) appeared to be under positive selection (pN/pS ratio ≥ 1) in all three arctic fox populations sampled (S7 Table). One gene appeared to be under positive selection only in Southwestern Alaska (CD8A), whereas both CCL2 and CCR8 were under positive selection within foxes sampled from Canada (Arviat and Victoria Island; S7 Table). Based on an assessment of Chi² p-values (S7 Table) only the STAT3 gene had a p value of p < 0.05 (0.000137), indicating a significantly different pN/pS ratio from expected values.

Analyses of intergenic SNPs (off-target)

The generation of secondary (off-target) sequences using targeted GBS approaches is a common feature observed in other studies [77]. These untargeted products are often consistently sequenced allowing these data to be used in downstream analyses if they are of sufficient coverage and quality [77–79]. Studies have noted that simulation-based assessments should be combined with off-target datasets to determine the power of these data to discern differences between populations and therefore form meaningful conclusions [80, 81]. Although the observed neutral genetic structure presented here is devoid of the subtle structure previously found among arctic fox populations in Alaska (F_ST = 0.02 with microsatellites) [31], it remains consistent with coarse geographic scale studies of the species [36–38]. This is unsurprising to a limited extent given foxes from only the northern coast of Alaska were used as representatives for the whole state in these coarse-scale assessments as previously noted by Goldsmith et al. [31]. It is worth noting that the inability to detect similar patterns of neutral population genetic structure in this study were likely due to the limited number of variants passing the filtering parameters, combined with differences in the power of discrimination between microsatellites and biallelic SNPs. Furthermore, when we attempt to identify F_ST outliers within the off-target dataset, there were insufficient numbers of outliers (4 after linkage pruning) to draw any meaningful result. The lack of identified outliers likely results from prominent gene flow documented among arctic fox, and as such, in conjunction with data presented in these previous studies [31, 36–38], we take the presumed neutral data presented herein to provide further evidence of the extensive gene flow among North American arctic fox populations.

While we acknowledge that filtering parameters employed herein were rigorous and likely removed informative SNPs form the final subsets used in analyses, determining neutral genetic structure of arctic fox across North America was not a direct objective of this study having been investigated thoroughly elsewhere [31, 36–38]. Additionally, recent research suggests recombination distances vary across chromosomes in red foxes (ranging from 0.07 cM– 5 cM between pericentromeric regions and chromosome ends) [82], meaning that the filtering step to ensure these SNPs were not in proximity (≥ 40 kbp) of any annotated coding regions provides only a coarse approximation of neutrality at these sites. With these potential limitations acknowledged, we used off-target data to provide a contextual baseline for the on-target data.

Analysis of SNPs in protein-coding regions

Multiple F_ST outlier identification programs were implemented to identify SNPs within genomic regions of interest in order to mitigate discordance between methods [62]. Of the 107 identified F_ST outlier SNPs within protein-coding regions, only 32 were identified by multiple programs, and only two were identified by all four methods used. The number of identified F_ST outliers between programs ranged from 9–61, where Bayescan identified the least and PCAdapt identified the largest number of F_ST outliers. Most identified F_ST outlier SNPs resulted in synonymous changes at their respective positions in the genome and thus unlikely to affect function of synthesized proteins. In contrast, 13 identified F_ST outlier SNPs conveyed a missense mutation that also changed chemical characteristics of translated amino acids, increasing the likelihood that these mutations could affect subsequent protein function [83]. These 13 missense F_ST outlier SNPs were associated with gene sequences of DLA-DQA, DLA-DQBC1, NOD1, RAG1 and TLR5, although only SNPs associated with the latter four genes were retained in the final filtered, linkage pruned, sub-dataset of 22 F_ST outlier SNPs. Both DLA-DQA and DLA-DQBC1 are class II components of the dog leukocyte antigen, responsible for initiating the immune response through antigen presentation and recognition [84]. NOD1 recognizes gram-negative bacteria and initiates a pro-inflammatory response [85], and RAG1 is a factor initiating immunoglobulin V(D)J recombination [86]. Finally, TLR5 detects bacteria with flagellin and induces a pro-inflammatory response [87]. Of interest are the large number of F_ST outlier SNP associations to interleukin and Toll-like receptor gene families, especially in context of AR, as members of these gene families are implicated in mediating a response to rabies [70–73]. Further, three missense F_ST outlier SNPs were associated with TLR5 demonstrating that this gene may play a prominent role in mediating responses to AR variants, although specific mechanisms of this response require further investigation.

Analyses of signatures of selection through assessments of pN/pS ratios demonstrate several genes that may be related to the spatially distinct distributions of AR variants. The CD8A gene is uniquely under positive selection in Southwestern Alaska, where AR variant 4 circulates. However, genes CCL2 and CCR8 appear to be under positive selection in both sampled regions in Canada where different AR variant 3 circulates. These data suggest that these three genes may be under positive selection, indicative of differential selection, relative to the AR variants in those locations and may play a role in the mechanisms that maintain unique distributions of AR variants. Of the ten genes under positive selection in all three populations, a large portion are involved in the MHC, which is expected given this gene family’s involvement in antigen binding and overall health of organisms [15–18]. Furthermore, two genes under selection in all three populations are associated with interleukins, a gene family already implicated in mediating a response to rabies [70, 71]. It is important to note that estimates of pN/pS are subject to potential biases in systems where there is prominent gene flow and migration [67], as is the case of arctic fox across their range, and by genes where there are few polymorphisms identified [69, 76]. To mitigate these potential biases, some researchers implement a threshold of pN/pS ratios > 2 to be indicative of positive selection [67]. In this context, only the STAT3 gene would appear to be under positive selection across all three of our sampled populations (pN/pS ratio of ~3.4). This gene was also the only gene to have a statistically significant pN/pS ratio with a p value of p < 0.05 (0.000137). However, in combination with the potentially inflated ratio of STAT3 due to those biases mentioned above, there were no outlier SNPs detected within the STAT3 gene. As such, although this gene may be under directional selection, it is unlikely to be contributing to the patterns of genetic structure observed within the on-target SNP-dataset. This is contrasted by outlier SNPs with the potential to change protein function identified within the genes DLQ-DQA and DLQ-DQBC1, however, these genes lacked significance in their pN/pS ratios that were suggestive of positive selection (pN/pS > 1; S7 Table). Given these results and potential biases, we present these pN/pS ratios only as an initial estimate requiring further testing and that these pN/pS results should be interpreted with caution.

Arctic rabies variant distributions and differential selection

Analyses of F_ST outlier SNPs demonstrate that genetic differentiation between arctic fox populations inhabiting an AR variant region were not significantly different from zero, however, genetic differentiation between arctic fox populations inhabiting regions where a different AR variant circulates were significantly different from zero. Given that arctic fox populations have been exposed to AR over a large time frame [88], there remains the potential for coevolutionary forces to have shaped patterns of differential selection between AR variants where analyses of pN/pS ratios and identified F_ST outlier SNPs within immunogenetically relevant genes tentatively support these observations.

Previous data [31, 35–38], combined with those presented herein, demonstrate that arctic fox populations are largely panmictic with prominent gene flow, yet still appear to be locally adapted to the different AR variants. It had been anticipated that elevated levels of gene flow among arctic foxes [31, 35–38] would homogenize AR variants across the landscape, precluding local adaptation. In addition, rabies incubation periods range from several days to several months [85], which would prevent movement of AR variants, and further undermine the spatially distinct AR distributions observed. Some authors have theorized that dispersal capabilities of rabid foxes are reduced [31], and thus maintain AR variants spatial distributions where populations then undergo differential selection and subsequent local adaptation. However, there are no data to suggest phylogenetically distinct AR variants elicit differential responses, making it unclear how the weak genetic structure of immunogenic variants observed elicit differential selection. It is of interest that despite differences in biogeography and geographic distances between arctic fox populations from Arviat and Victoria Island, where AR variant 3 circulates, these two populations of arctic fox show no genetic structure of either presumed neutral or functional markers associated with an immunogenetic response. Further, in arctic fox from Southwest Alaska, where AR variant 4 circulates, there appears to be no neutral genetic structure in context of the Arviat or Victoria Island populations. This pattern contrasts the genetic clustering of functional, immunogenetically relevant, markers distinguishing between those arctic fox populations from Canada relative to those in Alaska; suggestive of differential selection between arctic rabies variants 3 and 4. Overall, we take these data to suggest that AR variants may impose differential selective pressures on populations despite the impressive dispersal capabilities and gene flow within the primary host, arctic fox.

There remains potential that the observed patterns indicative of local adaptation between arctic fox and AR variants to have arisen from purifying selection or genetic drift, rather than directional selection as interpreted here [89], with drift being unlikely in a panmictic system with extensive gene flow. We attempted to account for purifying selection biasing our interpretation by providing pN/pS ratios for genes where calculations were possible. Given the pitfalls of targeted sequencing approaches, such as the difficulty of interpreting signatures of selection from genomic data and successfully targeting the most informative loci [89], we present only candidate genes suggestive of patterns of differential selection. Continued research, such as whole genome analyses or sequencing of identified candidate genes presented herein, will be required to provide more support for the identified patterns of local adaptation. Whole genome analyses would benefit these data, as it would facilitate the testing of whether the observed patterns of genetic structure documented herein are consistent among other regions of the genome or an artifact of sequencing a handful of genes from the genome. Further, future research that aims to investigate the interrelationship between arctic fox and AR should sample from more populations of arctic fox from each AR variant distribution as well as include multiple populations for each AR variant region where possible. Additionally, the inclusion of foxes that have survived exposure to each AR variant, as well as those succumbing to the disease would greatly enhance the study by allowing for more in-depth analyses between AR variant regions, and direct consequences/benefits of specific SNPs. Lastly, there remains the potential for unknown factors, beyond the scope of the objectives and methods implemented here to have identified them, that may better explain the observed patterns of genetic structure.