Samples

Blood samples were collected from foxes maintained at the experimental farm of the Institute of Cytology and Genetics (ICG) in Novosibirsk, Russia. All animal procedures at the ICG complied with standards for humane care and use of laboratory animals by foreign institutions. The study was approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Illinois at Urbana-Champaign. DNA was extracted from blood using Qiagen Maxi Blood kits, as per the manufacturer’s instructions (Qiagen, Valencia, CA).

Three sets of DNA samples we used in this study:

1. Genotyping-by-sequencing (GBS) set. We used samples from 40 individuals, 20 each from the tame and aggressive populations, for the genotyping-by-sequencing (GBS) experiment. Each population sample included equal numbers of males and females, and individuals were not closely related (i.e. did not share parents or, in most cases, grandparents).
2. Validation set. The second sample set was selected to validate the differences in allele frequency of SNPs from the distal part of fox chromosome 3 (VVU3) which were observed between the tame and aggressive populations using GBS data. The validation set included 45 aggressive, 46 tame, and 92 conventional farm-bred foxes (foxes which were not deliberately selected for behavior). Again, animals were, generally, not closely related and animals from both genders were equally represented.
3. Genetic mapping set. Fox F2 pedigrees including parents, grandparents, and 536 offspring were used for construction of a meiotic linkage map of VVU3 and quantitative trait loci (QTL) analysis. The F2 pedigrees were developed in a previous study [12, 25] by crossbreeding tame and aggressive foxes and then breeding F1 individuals to each other. All F1 pedigrees were produced in reciprocal manner with respect to parental gender and population of origin. Behavior of F2 individuals was tested, videotaped and scored from video records with a set of 98 traits [7, 12, 25]. Principal component (PC) analysis was used to define main axes of fox behavior as previously described [12]. Both PC defined phenotypes and scores for individual behavioral traits were used for QTL mapping [12, 25] (Nelson et al., in preparation).

Restriction enzyme (RE) selection and GBS optimization

To determine the best restriction enzyme (RE) system for reducing the complexity of the fox genome, DNA (100 ng) from a single individual was digested, separately, with several REs (ApeKI, EcoT22I, PstI, and EcoT22I/PstI double digest) according to the enzyme manufacturer’s protocol (New England Biolabs, Ipswich, MA). Samples were digested at 37°C for 2 hours and then, incubated at 80°C for 20 minutes to inactivate REs. Digested DNAs were ligated to adapters as previously described [22]. Samples were then purified (QIAquick PCR Purification Kit; Qiagen, Valencia, CA) and 2 μl of each library was amplified in a 50 μl volume containing 1 x Taq master mix (New England Biolabs, Ipswich, MA) and 25 pmol of each of two primers containing complementary sequences to ligated adapters and Illumina solid-phase oligonucleotides bound to the flowcell lane surface. PCR was performed following the previously described protocol [22]. Amplified libraries were purified again, as above, and fragments were visualized using the Experion (Bio-Rad, Hercules, CA) (S1 Fig). DNA profiles for ApeKI and EcoT22I enzymes showed enrichment for fragments of desirable sizes (S1 Fig) and did not contain repeat-associated peaks, therefore libraries were made with both of these enzymes.

GBS library preparation and DNA sequencing

The first run of sequencing was done with separate libraries made with ApeKI or EcoT22I enzymes [22]. Forty-seven fox DNA samples comprising 20 tame, 20 aggressive, and seven duplicated individuals were genotyped. The DNA samples (100 ng) were digested, separately, with ApeKI or EcoT22I at 37°C for 2 hours, and each sample was ligated to equal amounts of a different barcode-containing adapter and the same common adapter. The 48 barcode sequences used for EcoT22I GBS library construction (a negative control was also included) are listed in S1 Table. The barcode sequences used for ApeKI GBS library construction were published previously [22]. Individual ligation reactions (5 μL each) were pooled and DNA was purified (QIAquick PCR Purification Kit; Qiagen, Valencia, CA). A sample from the pooled library (10 μL) was amplified by PCR (50 μL total volume) containing 1 x Taq master mix (New England Biolabs, Ipswich, MA) and 12.5 pmol of each PCR primer (see S1 Table; [22]). PCR products were purified as above and quantified using the Nanodrop 2000 (Thermo Scientific, Wilmington, DE). First, the ApeKI or EcoT22I GBS libraries were sequenced on a HiSeq 2000 (Illumina Inc., San Diego, CA) in the same sequencing flowcell lane (single-end sequencing with 100 bp reads), with approximately half of the sequencing coming from each.

The average number of reads for the ApeKI library was 1,482,234, with high variation in read numbers among samples (SD± 902,649; median: 1,569,610). Nine animals received so little sequence that they were judged as failed. This was likely due to technical errors during the library preparation. Due to uneven results we did not use sequencing data generated for the ApeKI library in the following analysis.

The average number of reads per sample for the EcoT22I library was 1,181,370 (SD +/-314,990; median: 1,066,520) and no samples failed. Because the EcoT22I library produced good results in the first run we sequenced already created EcoT22I library on an additional flowcell lane to obtain deeper sequencing data for this library. This strategy produced coverage depths equivalent to 32-plex sequencing.

SNP discovery

SNPs were called from raw DNA sequences using the GBS pipeline as implemented in TASSEL 3.0.166 [26]. Raw sequences were converted to tag counts using the FastqToTagCountPlugin (options:-c 1, -s 300000000), where-c 1 is the minimum taxa count within a qseq file for a tag to be output; default: 1;-s 300000000 is the maximum number of tags the TBT (tags-by-taxa) can hold while merging (default: 200000000). The tag counts were merged using the MergeMultipleTagCountPlugin (options:-c 3). A fastq file with unique 64 bp sequence tags was created using the TagCountToFastqPlugin (options:-c 1), and aligned to the dog genome (//hgdownload.soe.ucsc.edu/goldenPath /canFam3/bigZips/canFam3.fa.gz accessed on 31 March 2014 with md5sum 07084e3a9843991825a67891d23e3735). Dog chromosomes were renamed for compatibility with the GBS pipeline (leading ‘chr’ notations were removed, and X and M chromosomes were converted to 39 and 40). The modified dog genome was indexed for use with BWA version 0.7.8-r455 [27] and the fox tags were aligned to it with BWA aln/samse using the default parameters. TBT files were created using the FastqToTBTPlugin (options:—c 1, -y) and merged using the MergeTagsByTaxaFiles plugin (options:-s 300000000). SNPs were called with the tbt2vcfPlugin (options:-ak 3, -mnLCov 0, -mnMAF 0), where-ak 3 is the maximum number of alleles that are kept for each marker across the population; default: 3;-mnLCov 0 is the minimum locus coverage (proportion of taxa with a genotype; default: 0.0;-mnMAF 0 is the minimum minor allele frequency; default: 0.0 = no filter.

Duplicate SNPs were merged with the MergeDuplicateSNP_vcf_Plugin (options:-ak 3). For VCF files, likelihood scores were calculated according to Etter et al., 2011, formula 3.8 [28]. The most likely genotype was assigned, and a genotype quality (GQ) score was calculated according to GATK:

“GQ: The Genotype Quality, or Phred-scaled confidence that the true genotype is the one provided in GT. In the diploid case, if GT is 0/1, then GQ is really L(0/1) / (L(0/0) + L(0/1) + L(1/1)), where L is the likelihood that the sample is 0/0, 0/1/, or 1/1 under the model built for the NGS dataset. The GQ is simply the second most likely PL—the most likely PL. Because the most likely PL is always 0, GQ = second highest PL—0. If the second most likely PL is greater than 99, we still assign a GQ of 99, so the highest value of GQ is 99.”

(http://gatkforums.broadinstitute.org/discussion/1268/how-should-i-interpret-vcf-files-produced-by-the-gatk accessed on 21 November 2014)

TASSEL 4.3.7 was used to merge 7 replicated samples using the plugin MergeIdenticalTaxaPlugin (options:-hetFreq 0.8, -maxAlleleVCF 3) [26].

SNP filtering

Two sets of SNPs, filtered and stringently filtered sets, were produced using VCFtools [29]. Filtering was initially performed using the following parameters: (1) the SNP was called at least 50% of individuals (--max-missing 0.5), (2) minor allele frequency (MAF) was > 2.5% (--—maf 0.025), (3) only two alleles were present (--min-alleles 2--max-alleles 2), (4) average read depth < 150 (--max-meanDP 150), (5) < 60% heterozygous individuals for autosomal SNPs and the X chromosome pseudoautosomal region (PAR) (0–6.65Mb) [30], or > 30% heterozygous individuals for the remainder of the X chromosome. More stringent filtering was performed by eliminating loci with quality score < 98 (considers sequence depth per locus) (--minGQ 98) and SNPs with missing data (--max-missing 1)

SNP evaluation

Consistency of SNP genotype calling was evaluated by comparing the two replicates from each of the seven duplicated DNA samples. Both filtered and stringently filtered SNP data sets were created for the unmerged data (where replicates of the duplicated samples were treated as separate samples), in the same manner as described above for the merged data set. In the filtered set, two types of discordance between the replicates were observed: 1) when one sample in a pair did not receive a call but the other did (missing data) and 2) SNPs for which both samples in the pair received a genotype call but the genotypes differ between the replicates. In the stringently filtered set, only second parameter was present because all SNPs in this data set had genotyping calls for all samples. The percent of concordant genotypes was calculated for each of the duplicated individuals in both the filtered and stringently filtered sets. The percent of the discordant genotypes due to missing data was also calculated for each duplicated individual but only in the filtered set.

The depth of the read coverage for the concordant and discordant SNPs was determined using the unmerged, filtered set of SNPs. The discordant genotypes due to missing data were not included in this analysis, only those SNPs that have genotypes for both replicates of the duplicated individuals were used. The mean and standard deviation of the depth was calculated for both concordant and discordant SNPs in this data set. The average heterozygosity of SNPs located on the X chromosome outside of PAR (0–6.65Mb) [30] was calculated. This was computed for males and females in each population separately by determining the percent of individuals of the same gender with heterozygous genotypes for each SNP, summing the values and dividing by the number of SNPs with data.

Estimating SNP positions on fox chromosomes

The location of GBS SNPs on the fox chromosomes was estimated by alignment of the fox meiotic linkage map against the dog genome as previously described [11, 13]. Briefly, dog chromosomes were divided into corresponding “fox segments” based on the location of known mapped SSR markers. When dog chromosomes were split into syntenic regions located on two fox chromosomes, the length of the entire region between known dog markers was included on both of the corresponding fox chromosomes. This was done to prevent markers that might be far apart in the fox genome from appearing to be close together on the estimated fox map. SNPs that mapped to unassigned portions of the dog genome (regions with unknown synteny and “Un” chromosomes) were not mapped to fox positions.

Population structure analysis, estimation of LD and effective population size

The stringently filtered SNP set (8,437 SNPs) was used for population structure analysis using Principal Component (PC) analysis and the Bayesian inference program STRUCTURE [31]. PC analysis was performed using PLINK2 (https://www.cog-genomics.org/plink2) [32]. Clustering analysis was performed using STRUCTURE v.2.3.4 [31, 33] at 100,000 iterations of the Gibbs sampler after a burn-in of 100,000 iterations. Each run was repeated eight times at each value of K using correlated allele frequency model with admixture model with default settings. The runs were completed for K from 2 to 5 without the population information.

A subset of stringently filtered SNPs, whose positions were extrapolated on fox chromosomes (8,405 SNPs), was used to estimate linkage disequilibrium (LD) decay in both tame and aggressive populations. Calculations of the squared correlation of the alleles at two loci (r²) were done in PLINK2 [32]. Distances between SNPs were estimated using the inferred SNP positions on fox chromosomes and were based on the corresponding physical distances in the dog genome. Average r² between adjacent SNPs was computed by grouping SNPs by pairwise estimated physical distances into 14 bins ranging from 1–1000 bp to 100–210 Mb (size of the largest approximated fox chromosome is 210 Mb) (S2 Table). LD was also examined between SNPs located on different chromosomes (i.e., r² was computed for all pairs of SNPs in which both SNPs were located on separate fox chromosomes).

The set of stringently filtered mapped SNPs was also used to calculate effective population size (Ne) using the LD method of Waples and Do [34] as implemented in NeEstimator [35].

Estimation of population parameters

The filtered SNP set (48,294 SNPs) was used to calculate population parameters. Observed heterozygosity was computed as the percent of heterozygous individuals (calculated separately for each population) among the total number of individuals that received genotype calls for that SNP. The average observed heterozygosity of the population was calculated as a sum of the SNP heterozygosity in that population divided by the total number of SNPs identified in the population. The mean, median and standard deviation of the minor allele frequency (MAF) were calculated separately for the tame and aggressive populations. Expected heterozygosity was calculated for each SNP by formula 2.16b from Hedrick, 4th Edition, page 93 [36]. The obtained value was divided by the number of individuals in the population with data (N) to calculate the proportion of individuals that would be expected to be heterozygous given the allele frequencies seen (the formula 2.16b calculates the expected number of individuals expected to be heterozygous). To calculate the mean expected heterozygosity, the expected heterozygosity values for each SNP within each population were summed and the obtained numbers were divided by the total number of SNPs.

Scan for SNP allele frequency differences between the fox populations

To identify regions in the fox genome where the two populations differed in allele frequency, 48,042 SNPs from the filtered set that mapped to the fox genome were compared between the tame and aggressive samples using the (--assoc--adjust) option in PLINK2 [32] with Bonferroni correction.

In addition to the above analysis, we also computed fixation index (F_ST) using VCFtools [29]. The value for “weighted_F_ST” is reported. The F_ST was calculated in windows of 1 Mb with the overlap of 500 Kb using the Weir and Cockerham estimator [37] (--weir-fst-pop aggr--weir-fst-pop tame--fst-window-size 1000000--fst-window-step 500000)

Confirmation of allele frequencies and construction of the fox chromosome 3 (VVU3) map

To confirm and further characterize a region of genetic divergence between tame and aggressive populations on fox chromosome 3 (VVU3) and to extend the existing linkage map for this chromosome, four new fox markers (three SSRs and one indel) were developed. The indel was identified through sequencing fox amplicons produced using dog derived primers; two of the SSR markers corresponded to microsatellites identified in the dog genome, and one SSR was selected from the dog meiotic linkage map [38]. Primers were designed with Primer3 [39] using the dog genome sequence. Fluorescent primers (S3 Table) were genotyped following the protocol described in Kukekova et al., 2007 [11]. Three SSR markers (VV0683, CM6.72b, CM6.75) and the indel marker (26749b) (S3 Table) were genotyped in fox F2 pedigrees which were previously genotyped with 18 SSRs assigned to VVU3. A high confidence map was constructed at a confidence level of 1000:1 (LOD ≥3) with crimap V2.504a [40]. Because GridQTL [41], the program used for QTL mapping, requires all markers to be placed at unique locations, the LOD 3.0 map was then saturated with unmapped SSRs regardless of the actual likelihood (LOD 0.0).

The indel marker (26749b) was also genotyped in 183 additional animals from three populations: tame (46), aggressive (45), and conventionally farm-bred foxes (92). The Fisher exact test (http://graphpad.com/quickcalcs/contingency2/) was used to evaluate the significance of the differences between each pair of populations.

QTL analysis

Behavioral traits and phenotypes defined using PC analysis [12, 13, 25] were mapped in the F2 pedigrees using the F2inbred algorithm of GridQTL [41]. Permutation (n = 1,000) was used to establish chromosome-wide and experiment-wide significance thresholds.

Library sequencing

The EcoT22I GBS library sequenced using Illumina technology yielded a total of 35.5 Gbp of sequence data. After quality filtering (sequences without a barcode or restriction site remnant or with “Ns” were discarded) 20.8 Gbp of sequence was retained for analysis. On average, this dataset contained 5,202,895 reads of 100 bp length per sample (SD± 2,019,764; median: 4,517,500).

SNP discovery and filtering

The GBS data were analyzed using the TASSEL-GBS pipeline to produce SNP calls [26]. A total of 2,003,563 tags (unique sequences) were identified and 1,530,295 (76.4%) of these aligned to single locations in the dog genome, 23,319 (1.2%) aligned to multiple positions, and 449,950 (22.5%) could not be aligned. The tags aligned to single locations in the dog genome produced a total of 101,940 SNP loci, of which 99,450 had exactly two alleles. Two sets of SNPs, filtered (48,294 SNPs, S4 Table) and stringently filtered (8,437 SNPs) sets were then produced. The amount of information obtained using GBS in foxes demonstrates the value of this approach for simultaneous SNP identification and genotyping in species without well-developed genomic tools.

SNP evaluation

The SNP call reproducibility was evaluated using duplicated samples. In the stringently filtered set, the genotypes of replicates were in perfect concordance (the two replicates received the exact same calls) for 97.9–99.2% of SNPs among the seven duplicated samples.

In the filtered data set, the genotypes were in concordance 66.5–73.8% of the time. The disagreement was due to one sample not receiving a genotype call 55.6–71.7% of the time (i.e. missing data). The average depth for the discordant SNPs was 14.0 ± 25.4 reads, while for concordant SNPs the average depth was 34.0 ± 41.6 reads, thus the average depth for the concordant SNPs was 2.4 times greater than for the discordant ones (S5 Table). We hypothesized that many of the discordant SNPs are heterozygous SNPs where only one allele was captured. The analysis of the combined samples showed that 77.7–80.4% of discordant SNPs received heterozygous genotype calls in combined samples confirming that the greater depth increases the likelihood that both alleles are captured in heterozygous individuals. The analysis of the heterozygosity of SNPs located on X chromosome outside of PAR was consistent with these observations. The X chromosome SNPs from the filtered set showed increased heterozygosity in males (in average, 5.4% of males were heterozygous in each population) and decreased heterozygosity in females (in average, 12.4% of females in tame and 13.9% of females in aggressive population were heterozygous) (S5 Table). In contrast, the X chromosome SNPs from the stringently filtered set showed low level of heterozygosity in males (1.2% in tame and 1.0% in aggressive population) and relatively high heterozygosity in females (25.2% in tame and 29.4% in aggressive population) (S5 Table). We expect that increased heterozygosity in males largely reflects genotyping errors and low sequencing depth, while decreased heterozygosity in females reflects heterozygous under calling also due to not sufficient sequencing data. The high concordance and low level of heterozygosity of SNPs from X chromosome in males in the stringently filtered set are indicative of a low error rate among those SNPs.

Assignment of SNP positions on fox chromosomes

The dog is the fox's closest relative with a sequenced genome. We used the comparative dog/fox map [13] and the known positions of fox SNPs in the dog genome to infer SNP positions on the fox chromosomes. Positions of 48,042 SNPs from the filtered SNP set (99.5% from the total number of SNPs in this set) and positions of 8,405 SNPs from the stringently filtered set (99.6% from the total number of SNPs in this set) were inferred in the fox.

Comparative analysis of tame and aggressive fox populations using SNPs

Principal component analysis identified genetic divergence between the two populations.

The stringently filtered set of 8,437 SNPs with no missing data for any of the 40 individuals was subjected to Principal Component (PC) analysis. The PC analysis revealed genetic differences between the two strains (Fig 1). PC1 separated the samples into tame and aggressive groups, in agreement with the fact that the two populations have been maintained separately for more than 40 generations. PC2 identified more diversity in the aggressive strain than in the tame one, which may be attributable to the differences in history of the two populations. The tame population was maintained as a closed line through the breeding program, while the aggressive population experienced an introduction of foxes from conventionally farm-bred stock in the 1990s.

Download:

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

Fig 1. Principal component analysis.

Principal component analysis of SNP data for 20 tame and 20 aggressive foxes. 8,437 SNPs with genotypes available for all individuals were used in this analysis. Aggressive individuals are represented by red dots, tame individuals are represented by green triangles. PC1 is plotted on the x-axis, PC2 is plotted on the y-axis.

https://doi.org/10.1371/journal.pone.0127013.g001

Estimation of the genetic structure using STRUCTURE.

The genetic structure analysis at K = 2 clearly separated the tame and aggressive individuals into two clusters (Fig 2). At K = 3 the stratification of the aggressive population became apparent. The further segmentation of the aggressive population was apparent at K = 4 and 5. Very little stratification was observed within the tame population at all K tested. The split of the aggressive population into subpopulation at K = 3 and higher Ks is consistent with the results of the PC analysis. In fact, the aggressive individuals which showed highest and lowest amount of segmentation based on the assignment into inferred clusters at K = 3, 4, and 5 had highest and lowest PC2 values, respectively. Both the PC analysis and the population structure analysis revealed the genetic diversification within the aggressive population which was likely caused by the admixture of the aggressive population with conventional population in 1990s.

Download:

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

Fig 2. Estimation of population structure using STRUCTURE.

Cluster analysis of fox genotypes was performed at four values of K (2, 3, 4, and 5) without population information. The numbers of assumed clusters are indicated on the y-axis. The population origin of individuals is indicated on x-axis. On each graph the individuals are listed in the order obtained at K = 3. Each individual is represented by a bar that is segmented into colors based on the assignment into inferred clusters given the assumption of K populations. The length of the colored segment is the estimated proportion of the individual’s genome belonging to that cluster. The analysis was run in 8 replicates for each K, the replicate with the highest likelihood is shown. The genetic structure analysis clearly differentiated the tame population from the aggressive one and did not reveal significant population stratification within the tame population at every K tested. In contrast, the population stratification within the aggressive population became apparent at K = 3.

https://doi.org/10.1371/journal.pone.0127013.g002

Estimation of linkage disequilibrium and effective population size.

The relative order of and distances between SNPs in the fox genome were established by extrapolation from the dog genome. An average r²≥ 0.5 was observed for SNP pairs with inter-marker distance less than 1 Kb in both tame and aggressive populations. An average r²≥ 0.2 was observed for SNPs located within 100–500 Kb in tame and 50–100 Kb in the aggressive population (Fig 3, S2 Table). In general, the r² values for SNPs in bins ranging from 1–1000 bp to 1Mb-5Mb were lower in the aggressive population than in the tame indicating a faster decay of LD in the aggressive population. The number of SNP pairs with r² = 1 was slightly higher in tame than in the aggressive population (S2 Table). Less than 10% of SNP pairs with r² = 1 was observed for SNPs located over 10 kb apart in the aggressive population and over 50 kb apart in the tame population (S2 Table).

Download:

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

Fig 3. Estimation of linkage disequilibrium (r²) in tame and aggressive fox populations.

Distributions of r² values between pairs of SNPs separated by different distances are compared between tame (green) and aggressive (red) populations. SNP pairs were divided into 14 sets (bins) using the estimated distances between SNPs in the fox genome (S2 Table). Each bin is represented by a doubled bar (green and red) on the graph. The range of distances between SNPs in each bin is indicated on the x-axis. The width of the bar represents the relative number of SNP pairs in that bin for that population after a log transformation (wider bars have more pairs of SNPs). Exact numbers of SNP pairs in each bin are presented in S2 Table. The y-axis indicates r² values for pairs of SNPs. The yellow diamonds correspond to the mean r² for all SNPs in that bin in the population. The white circles correspond to the median values. The thin black line within each bar represents r² values in that bin in the population in the interval from the 25th to 75th percentile. The horizontal line corresponds r² = 0.2.

https://doi.org/10.1371/journal.pone.0127013.g003

To test that our data do not show significant spurious associations between unlinked SNPs the r² was calculated for SNPs assigned to different fox chromosomes. In total, 22,184,914 pairs of SNPs in which each SNP in a pair was located on a different fox chromosome were analyzed in the tame and 27,725,667 pairs in the aggressive population. No substantial LD between unlinked SNPs was observed. Average r² between SNPs located on different fox chromosomes was 0.055 (±0.079) in tame and 0.058 (±0.079) in aggressive populations.

Although estimation of the LD in fox strains could be influenced by the small sample size used in this study, the small number of SNPs, and imperfect calculation of intermarker distances based on fox/dog synteny, it still provides a useful indication of the extent of LD in these populations. The length of LD observed in the tame fox population is comparable with average LD length in dog breeds (r² ≥ 0.2 was reported for SNPs located at the distance 500 Kb for several dog breeds [1]) and pigs (r² = 0.15 and 0.2 were reported in two pig breeds for SNPs located 1 Mb apart [42]); and it is moderately higher than LD reported for two cow breeds (r² = 0.3 was reported for SNPs located less than 100 kb in Holstein cattle and r² = 0.25 for SNPs less than 20 kb in Fleckvieh breed [43–45]). The LD decay in cows is comparable with the LD profile in the aggressive fox strain. The length of LD in both fox strains is shorter than the length of LD in outbred mouse stock (r² = 0.5 was reported for SNPs located 2 Mb apart [46]). The LD estimations calculated using sequence data has been reported to be lower than those calculated using SNP chip generated data, due to ascertainment bias of SNPs on the SNP chip [44]. The different technologies and MAF cut-offs between the studies makes the comparison between studies and species an imperfect estimation. The differences between our two populations, which were calculated from the same data, likely reflect the existing differences between the two strains.

Effective population sizes of the aggressive and tame populations, respectively, were 38.0 (95% CI: 37.8–38.2) and 81.5 (95% CI: 80.5–82.5). Although the aggressive population has a smaller number of breeding animals and lower effective population size, more genetic variation has been preserved in this population. These findings may reflect the fact the aggressive population had an inclusion of outside individuals in the recent past in comparison to the tame population, which experienced a longer period of closed breeding.

Comparison of SNP allele frequencies between tame and aggressive fox populations

Analysis of the allele frequency differences between the two populations.

Overall, a relatively high level of heterozygosity was observed in both populations (Table 1), clearly demonstrating that efforts to limit inbreeding in these strains has been effective. A slightly lower level of heterozygosity was observed in the tame population than in the aggressive one (mean expected heterozygosity in tame population was 0.22 versus 0.24 in the aggressive population). The average SNP minor allele frequency was slightly lower in the tame strain than in the aggressive strain and a higher percent of SNPs was monomorphic in the tame than in the aggressive strain (Table 1). Comparison of allele frequencies between the two populations (Fig 4a, Table 2) identified 68 SNPs on 12 fox chromosomes that showed significant differences in allele frequency (Bonferroni correction of association test; p<10^-7). These included 19 “isolated” SNPs (> 2Mb from another significant SNP), and 49 SNPs located within 11 clusters on seven fox chromosomes. Clusters comprised at least two SNPs located within a 2 Mb interval (Table 2). To further characterize these genomic intervals, genes located inside or within 50,000 bp from start and end of these regions were identified based on dog genome annotations (Table 3). Several of these intervals were found to include genes involved in neurological functioning and behavior (Table 3).

Download:

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

Table 1. SNP minor allele frequency and heterozygosity in tame and aggressive populations.

https://doi.org/10.1371/journal.pone.0127013.t001

Download:

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

Table 2. SNPs with significant allele frequency differences between tame and aggressive populations.

https://doi.org/10.1371/journal.pone.0127013.t002

Download:

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

Table 3. Genes located inside or within 50,000 bp from start and end of the multi SNP clusters in the dog genome.

https://doi.org/10.1371/journal.pone.0127013.t003

Download:

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

Fig 4. Allele frequency differences of genome-wide and VVU14 SNPs.

The black dots within the gray bar on the top of each figure panel indicate regions with F_ST≥0.5. Dots of different colors are the Bonferroni corrected significance of the allele frequency differences between the tame and aggressive populations calculated by PLINK2. The y-axis indicates-log10 (Bonferroni corrected p-value) for SNP allele frequency differences. The black horizontal line corresponds to a significance cutoff of (-log10 = 6). Genomic position in fox, as extrapolated from syntenic regions in the dog genome, is plotted on the x-axis. 4A. Differences in allele frequency of SNPs genome wide. Colors indicate the different fox chromosomes. The x-axis indicates the cumulative estimated positions in the fox genome in megabases. 4B. Differences in allele frequency of SNPs on VVU14. The horizontal colored bars on the x-axis correspond to syntenic dog chromosomes. 4C. Multi SNPs region on VVU14 with significant allele frequency differences between the two populations. CFA3:42,000,000–46,000,000 corresponding to the cluster on VVU14 with multiple SNPs that are highly significant. The genes in the region are indicated just above the x-axis. The gene NR2F2 (COUP-TFII) is the only gene located in the significant region.

https://doi.org/10.1371/journal.pone.0127013.g004

Analysis of genomic regions on VVU3, VVU8 and VVU14 in fox crossbred pedigrees

Genomic regions with increased divergence between the tame and aggressive populations could represent regions influenced by either behavioral selection or random fixation (genetic drift). To distinguish between these processes, we hypothesized that regions representing true selection targets would be identified by both selective sweep mapping in fox populations and QTL mapping of behavioral traits in fox experimental pedigrees. Regions influenced by random fixation, on the other hand, would be identified only by homozygosity mapping, but would not co-segregate with behavioral phenotypes in informative pedigrees. To test this hypothesis, three genomic regions on fox chromosomes 3 (VVU3), 8 (VVU8) and 14 (VVU14), which were among the regions which showed the greatest allele frequency differences between the two populations (Table 2; Fig 4; S2 Fig, S6 Table; S3 Fig), were examined in more detail.