The canine XX DSD model pedigree.

This pedigree was established in 1982 by crossing a male (ACS1) from an American cocker spaniel (ACS) family segregating XX DSD [32] to normal beagle (BGL) females obtained from a commercial pedigree in which XX DSD has not been identified [33]. Matings of F1 offspring to each other, and F1 backcrosses to ACS1, produced affected mixed breed (ACS/BGL) dogs [33]. No affected F1 females were identified. Only parents of affected dogs were retained as breeding stock at each generation. Additional normal female beagles were later introduced periodically to reduce inbreeding depression. A second American cocker spaniel sire (ACS2) that had produced affected offspring was introduced into the pedigree in 2005. He was bred to females in the pedigree and beagles, and the same breeding strategy as above was used to maintain the model.

All XX DSD in the model pedigree had a female karyotype (78,XX), were SRY-negative by PCR assay and had a complete uterus [33–35]. Affected phenotypes included testes (T-XX DSD), ovotestes (OT-XX DSD), and those with an ovotestis and female external genitalia (subclinical XX DSD) with or without ovarian dysfunction [28,33,35,36]. Dogs with a normal male karyotype (78,XY) were unaffected, but were presumed proven carriers after producing affected offspring. Matings of proven carrier males to beagles did not produce affected dogs. Inheritance, based on early breeding experiments, was consistent with sex-limited, autosomal recessive transmission [33]. However, prevalence and severity of XX DSD significantly decreased later as BGL background increased in the pedigree (1982–2005, S1 Fig, S1 and S2 Appendices), indicating that simple Mendelian inheritance alone was insufficient to account for XX DSD expression.

GWAS region identified in the XX DSD model pedigree.

Twenty-seven dogs were genotyped. Those from the model pedigree were 17 XX DSD, ACS1, ACS2, and 4 females that had produced affected offspring (presumed proven carriers, Table 1). The range of XX DSD phenotypes were mild (1 ovotestis paired with 1 ovary) to severe (bilateral testes). Twelve of the XX DSD were descendants of ACS1, and 5 were descendants of ACS2. Because dogs with female external genitalia in the model may have subclinical XX DSD, 4 BGL females from the commercial vendor were chosen as normal female controls.

Download:

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

Table 1. Characteristics of dogs included in GWAS (N = 27).

https://doi.org/10.1371/journal.pone.0186331.t001

We performed a genome-wide association study (GWAS) of the 27 dogs genotyped on the Affymetrix version 2 canine SNP array (www.broadinstitute.org/mammals/dog/)). A logistic mixed model [37] was then applied to each of the genotypes, to account for breed structure and relatedness of dogs in the pedigree (Fig 1, S1 Table, S3 Fig, S3–S5 Appendices). We then assessed whether the most significant peak of genotypes in the region of CFA9 (GWAS region:2605238–6874718 CanFam3.1, genome.ucsc.edu), which included uncorrected p-valuesas low asp <0.0001, indicated suggestive genomic variants for further investigation. Given that an experiment with a sample size of twenty-seven is extremely under-powered when applying a standard GWAS multiple test correction, we used a strategy of applying a multiple test correction after heavy filtering of SNPs to lower the test correction burden, by removing SNPS with missing genotypes, low minor allele frequency SNPs (MAF <0.3), and by thinning every 10^th SNP, leaving less than 2,000 SNPs total. After such filtering, we found one SNP in the CFA9 peak region that had an Benjamini-Hochberg (BH) adjusted p-value <0.1, where the next most significant BH adjusted p-value was p-value = 0.297. While SNPs in the CFA9 region were therefore not statistically significant by criteria used in GWAS with samples of thousands of unrelated individuals, given our extremely small sample size, we considered the significance level of SNPs in the CFA9 region to be suggestive enough for further investigation. When considering the 32 SNPs spanning the peak of the CFA9 region, this 4.27MB region overlaps the canine ortholog for the upstream regulatory region of human SOX9 [38,39] and the canine ortholog for human XX DSD (Online Mendelian Inheritance in Man, OMIM 278850, hg38 chr17:69100001–76800000). The GWAS region does not include the canine SOX9 coding region, the canine ortholog of the mouse testis-specific enhancer of Sox9 (TESCO) [40], or the canine ortholog of human RevSex [13]. This region is in LD with SNPs spanning the CNVR1 region that was previously identified in controls and some cases of canine XX DSD [30], but does not overlap this region.

Download:

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

Fig 1. GWAS Manhattan plot for logistic mixed model analysis showing association of XX DSD to CFA9 in the canine model pedigree.

The probability of association [-Log₁₀ (P)] is shown for single nucleotide polymorphisms (SNPs) on chromosomes 1–40. (Quantile-Quantile plot of the p-values is in S3 Fig).

https://doi.org/10.1371/journal.pone.0186331.g001

Post GWAS SNP genotyping and MLPA studies.

To narrow the 4.27MB CFA9 GWAS region to a smaller area around the causative allele, genotypes were assembled by chromosomal locus to identify SNPs common to ACS1 and ACS2, and thus indicate an affected haplotype. Two haplotype blocks were identified by homozygosity mapping in which the alternate SNP alleles were homozygous in ACS1 and ACS2 and the reference SNP alleles were homozygous in 4 female controls. A custom canine genomic tiling array (Roche Diagnostic Corp., see Further associated variant discovery, Methods) was used for SNP genotyping to further narrow down the candidate interval within the combined haplotypes totaling 2MB (2604935–3349954 and 5922063–7019389; CanFam3). Two XX DSD (C3023, C3481), and ACS2 from the model pedigree and 2 female controls (A304, A888, Table 1) were genotyped with this array in the 2MB region, identifying 5 SNP haplotype blocks (Regions 1–5) that were shared by ACS2 and affected dogs, but not by controls. SNPs within these haplotype blocks were validated by Sanger sequencing of PCR products. When the CanFam3 assembly became available post-analysis, we converted the 5 SNP haplotype blocks to CanFam3 coordinates (LiftOver function, UCSC Genome browser), but found no change in their relative positions on CFA9. Further investigation of these blocks by homozygosity mapping was based upon the principle that regions inherited from the ancient domestic dog population tend to be transmitted whole, as they are shorter than intra-breed haplotype blocks that range from 0.5–1 MB [41]. Thus, affected purebred dogs can have alternate alleles that are identical‐by‐descent at markers located near a disease locus [41]. Region 3 had the highest proportion of homozygosity for the alternate, rather than the reference, SNP allele in 28 affected dogs in the study pedigree, and in 26 unrelated affected dogs of 21 breeds. In contrast, homozygosity of alternate alleles was shared only within the study pedigree in Regions 2, 4, 5, and only in the study pedigree and ACS and ECS breeds in Region 1. Two variant SNP alleles (6397782 and 6398082; CanFam3) in Region 3 haplotype block segregated with XX DSD in a total of 20 and 12 breeds respectively, including the Afghan hound, the most ancient breed tested. However, these two SNP alleles were also identified in female controls. Taken together, the data suggested that these two SNPs were breed-associated, but could be markers for a nearby variant in a shared haplotype block from the ancient domestic dog population that contains Region 3.

As CNVs have been associated with human XX DSD, particularly those mapping to the RevSex region within the SOX9 regulatory region [13], we hypothesized that alternate SNPs in the Region 3 haplotype might indicate the presence of a CNV. Therefore, we used Multiplex Ligation-dependent Probe Amplification (MLPA) to screen for CNVs in the Region 3 SNP haplotype block and the canine RevSex ortholog (S2 and S3 Tables, see Methods: MLPA for CNV detection). Because no probes in the canine RevSex ortholog had been included in the GWAS genomic tiling array for SNP genotyping, this region was also SNP genotyped by Sanger sequencing of PCR products (see Methods, SNP genotyping). No variations unique to XX DSD dogs were identified by MLPA or SNP genotyping in the Region 3 haplotype block or canine RevSex ortholog. These results in the canine RevSex ortholog are comparable to another study in which no CNVs uniquely associated with canine XX DSD were identified [30], and differ from studies in which CNVs including SOX9 or upstream of SOX9 were identified in sporadic canine XX DSD cases [29,31].

Fine mapping by analysis of WGS variants in XX DSD and control purebred pet dogs identified a CFA9 insertion associated with the condition.

Based upon a hypothesis that XX DSD prevalence in several dog breeds indicates most breeds inherited the same risk factors from an ancestral domestic dog population [41], we predicted that the major causative genetic defect and associated variants are located within such a canine ancestral segment that, characteristically, has short linkage disequilibrium (LD), and due to small size (~100-500kb), this segment would tend to be inherited intact. We used a custom designed canine SNP array with a maximal limit of 84 allele-specific probes. Variants were selected from the 244 loci on CFA9 identified by WGS. High priority was assigned to unannotated variants with allele frequencies less than 8% (AF<0.08) in a canine WGS database [43], which excluded approximately half having an alternate AF >0.2. Additional priority was assigned to those in which the reference allele was conserved, with lower priority assigned to variants in repeats annotated in the UCSC Genome browser (CanFam3). The final selection included 75 test loci from the WGS region (CFA9:5360823–7243520, CanFam3.1, Fig 2) and 9 published SNPs for quality control (CFA9:5369117–6097313, listed in S5 Table).

The array was used to genotype DNA from 3 groups, totaling 189 purebred pet dogs unrelated to the XX DSD model pedigree (Table 2). The first group, Affected Pets, was expected to be enriched for the causal variant found in the model pedigree because XX DSD in these dogs was confirmed by the PI (M-W), using the same diagnostic criteria as in the model pedigree. The Control Pets 1 group contained purebred dogs having normal female external genitalia that were breed-matched to the Affected Pets. These were expected to be normal females, unidentified carriers, or subclinical XX DSD dogs. The latter two were expected to have the causative allele. The Control Pets 2 group contained purebred dogs having normal female external genitalia from breeds in which XX DSD has not been reported. The majority of these were expected to be normal females, with a minority of unidentified carriers or subclinical XX DSD dogs having the causative allele. Based upon our hypothesis of a common causative variant for XX DSD in purebred dogs, we predicted the allele frequency of an XX DSD associated variant common to several breeds would be highest in the Affected Pets group, lower in the Control Pets 1 group, and lowest in the Control Pets 2 (AF Affected Pets group > AF Control Pets group 1 > AF Control Pets group 2). The results were compatible with this hypothesis (Table 2). Genotypes were obtained from 188 samples tested in the custom designed canine SNP array. After filtering 8 probes with low call rate (<70%), the remaining 76 loci had an average call rate of 96%, and 100% concordance in duplicate samples. Variant genotypes sorted by identification number, breed and group were tested for XX DSD association. One variant, a guanine insertion (G+) at CFA9: 6048201–6048202 (Table 2, Fig 2, rs852549625) was identified as significantly associated with XX DSD (Chi-squared test, P = 1.1e-6 using dominant model of association, Bonferroni threshold 6.6e-4). This variant was not annotated in the canine WGS database [43] or canine genome (CanFam3) at the time it was selected for screening by the array.

Download:

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

Table 2. Purebred pet dog groups genotyped with the custom array of selected WGS variants to identify those that are common to affected dogs of several breeds.

https://doi.org/10.1371/journal.pone.0186331.t002

Genotyping at the insertion locus (Fig 2) was repeated by Sanger sequencing of PCR products from all dogs tested in the custom SNP array. Electropherogram analysis for each dog (S6 Table) confirmed the array results, indicating that the XX DSD associated allele was a single guanine insertion at this locus, and that no flanking alleles associated with XX DSD were present. No deletions or other insertions uniquely associated with XX DSD in the model pedigree were identified by Sanger sequencing of contiguous PCR products within CFA9:6046484–6048494. Furthermore, no CNVs were identified by Droplet Digital PCR (ddPCR) near the 6048201 locus, or in the SOX9 coding region in XX DSD from the model pedigree (probes listed in S7 Table). Taken together, results indicate that the 6048201guanine insertion is a marker for inherited canine XX DSD.

The guanine insertion is located in BTBD17 intron 1, which is orthologous to human chr17:74360240–74361728 (hg38,—strand, LiftOver function, genome.ucsc.edu). The guanine insertion also lies within a CpG island (CpG26, CanFam3.1, genome.ucsc.edu ). KIF19 and an unannotated protein (F1PLR5) are nearby. Many potential transcription factor binding sites are present within 125 bp of CFA9:6048201. The guanine insertion introduces potential PAX sites, whereas none are here in the reference genome, and adds potential C2H2 ZF and bHLH sites to those already present (listed in S8 Table).

Genotype results at the 6048201 locus indicate that breed background affects penetrance and support a complex mode of inheritance for canine XX DSD.

In contrast to XX DSD purebred pet dogs, in which most were heterozygous for the insertion (75% G+/-, Table 2), most XX DSD in the model pedigree were homozygous for the insertion (83% G+G+, Table 3). Both founder sires (ACS1, ACS2) were G+G+ and all BGL females bred into the model pedigree were wild type (-/-). The full range of phenotypes was found in dogs with G+G+ genotype in the model pedigree, which had varying proportions of ACS and BGL background. Interestingly, 5 ACS/BGL breeding stock with female external genitalia that had produced affected offspring were G+G+ genotype (Table 3). Gonads from these dogs were removed at 2.5–9 years of age and no seminiferous tubules were identified by histology. Presently there is not a method to definitively determine whether these were unaffected females or the mildest XX DSD phenotype (subclinical XX DSD) that was not ascertained by gonadal histology.

Download:

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

Table 3. Most XX DSD in the model pedigree (83%) were homozygous for the guanine insertion (G+G+) at CFA9: 6048201.

https://doi.org/10.1371/journal.pone.0186331.t003

The homozygous insertion is associated with embryonic lethality in combination with German shorthaired pointer (GSHP) genetic background.

Because genotyping results at the 6048201 locus indicated that the G+G+ genotype was uncommon in XX DSD purebred pet dogs (3% G+G+, Table 2), but common in mixed breed XX DSD in the model pedigree (83% G+G+, Table 3), we hypothesized this genotype was deleterious in purebred dogs. Therefore, we genotyped pedigrees from purebred and crossbred breeding experiments in which the probability of G+G+ offspring varied from 0.25–0.5 (Fig 3). All offspring were born alive and birth weights within the same litter were similar. None of the offspring were G+G+ genotype, including crossbred offspring sired by males from the model pedigree that had previously produced G+G+ and G+/- XX DSD offspring (sires C734 and C752, Table 3). These results indicated that embryonic lethality was associated with the G+G+ genotype in the GSHP genetic background (Fig 3).

Download:

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

Fig 3. Embryonic lethality in association with the homozygous insertion at CFA9:6048201 was identified in breeding experiments.

At left are crosses between V4, a fertile GSHP XX DSD and GSHP males (PM6578, V31). At right are crosses between V4 and males from the XX DSD model pedigree (C752, C734), which vary in proportion of American cocker spaniel and beagle genetic background (ACS/BGL). Genotypes at the insertion locus (CFA9:6048201) are homozygous (G+G+), heterozygous (G+/-), or wild type (-/-). Females are indicated by open circles, XX DSD by filled circles, and XY males by squares. Each symbol represents one dog, except those containing a number, which indicates the number of dogs represented by that symbol. None of the offspring had the G+G+ genotype. Based on simple Mendelian inheritance, the expected number of -/-, G+/- and G+G+ offspring from the G+/- sires and dams is 4.25, 8.5, and 4.25, respectively (N = 17). This is significantly different from the numbers observed (x² = 8.1776, df = 2, Chi-squared test is significant at p<0.025 since 8.1776>7.38). Similarly, the expected number of G+/- and G+G+ offspring from a G+/- dam and G+G+ sire (V4 x C734) is 11 and 11, respectively (N = 22). This is a highly significant difference from the numbers observed (x² = 20.04, calculations include continuity correction for df = 1 [45]; Chi-squared test is significant at p<0.005 since 20.04>7.88). Since all offspring of C734 (G+G+) must have received the G+ allele from him, only those receiving the wild type allele from the GSHP V4 dam (G+/-) were born. Together, these results indicated that the G+G+ genotype, combined with ≥50% GSHP background, is associated with embryonic lethality.

https://doi.org/10.1371/journal.pone.0186331.g003

RNA-seq identified RSPO1 downregulation but not SOX9 upregulation in gonads from the XX DSD model.

The second purpose of this study was to identify gene expression alterations in embryonic gonads from the model pedigree that could be informative to XX DSD causation. As these are unique and rare samples, we chose RNA-seq to measure all expressed genes, as well as known genes in the testis and ovary pathways. Because embryonic gonad pairs at the ages tested had insufficient mass for both RNA-seq and histology, the guanine insertion at CFA9:6048201 was used as a marker in genomic DNA to identify embryos at risk of XX DSD from the model pedigree (Table 4).

Download:

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

Table 4. Embryonic gonad pairs for RNA-seq were collected at d34-44 from embryos at risk of XX DSD, XY littermate controls, and XX and XY age-matched controls.

https://doi.org/10.1371/journal.pone.0186331.t004

Gene expression was compared between XX DSD gonads at risk and XX age-matched controls for both the d37-39 and d42-44 age groups. The list of genes with significant expression differences found by cuffdiff v2 was further filtered for minimum expression level in at least one condition and minimum fold-change between conditions to establish a ‘stringent’ set of differentially expressed genes (stringent-DE genes) consisting of 271 genes at d37-39 and 625 genes at d42-44 (combined 754 genes; see Methods). A majority of stringent-DE genes (539) were downregulated in XX DSD gonads in comparison to XX controls. Of these, over 500 were located on chromosomes 1, 5, 6, 9, 13, 20 and 26 (data in S4 Fig). For all chromosomes, the region with the greatest number of downregulated genes was CFA20:50-60M, which contains transcription factor E-2 alpha (TCF3), follistatin-related protein3 (FSTL3), and Anti-Mullerian hormone (AMH), also known as Mullerian Inhibiting Substance. On CFA9, the greatest number of downregulated genes (N = 13) was located between 1-10MB, which includes the GWAS and WGS regions. The stringent-DE gene list contained a total of 38 annotated genes on CFA9 (CanFam3), primarily downregulated in XX DSD gonads at risk in comparison to XX age-matched controls. Only four of the stringent-DE genes on CFA9 were significantly upregulated in XX DSD gonads at d42-44, but these have no reported role in gonadal sex determination (ENSCAFG00000004922, ENSCAFG00000017563, ENSCAFG00000017668/TUBD1, ENSCAFG00000025142). Expression of 3 annotated genes nearest the insertion (BTBD17, KIF19 and F1PLR5) was very low, with Fragments Per Kilobase of transcript per Million mapped reads (FPKM) being <<1 in all groups. No expression in this region was identified for polyA transcripts expressed in adult ovary (CUFF.38126.1, CUFF.38126.2, CanFam3, genome.ucsc.edu ). To investigate novel splice forms and transcripts within and near the GWAS and WGS regions (CFA9:5-9MB), a similar RNA-seq analysis was run that was guided by, but not limited to, the CanFam3 Ensembl transcriptome. No novel transcripts with differential expression between XX DSD and XX controls were found in this interval.

In general, ovary pathway genes were expressed at lower levels in XX DSD gonads at risk in comparison to XX controls in both age groups (Table 5). In particular, RSPO1 and WNT4 expression were lower in XX DSD gonads compared to XX controls (Fig 4). Interestingly, expression was lower by >2 fold for RNF43 at d42-44 (Table 5), a negative regulator of the WNT/CTNNB1 pathway [46]. The major finding in ovary pathway genes was that RSPO1 is significantly downregulated at d42-44 using the stringent criteria of >2 fold difference. While a general downregulation of other genes in the ovary pathway is present, these did not meet the stringent criterion of >2 fold difference. Similarly, testis pathway gene expression was higher in XX DSD than XX control gonads, but none met the stringent criterion of >2 fold difference (Table 5). Notably, SOX9 expression in XX DSD gonads at risk was considerably lower than in XY controls (Fig 5), and no difference in DMRT1, SOX3, SOX8, or SOX10 expression was detected between XX DSD gonads at risk and female controls. In summary, the major difference between XX DSD gonads at risk and XX gonads was RSPO1 downregulation at d42-44, without upregulation of known testis pathway genes.

Download:

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

Fig 4. RSPO1 and WNT4 expression in canine embryonic gonads was measured by RNA-seq.

(A) RSP01 and (B) WNT4 expression were lower in XX DSD gonads at risk (filled triangles) compared to those of XX age-matched controls (open triangles) at d37-39 and d42-44. Neither RSP01 nor WNT4 expression in XX DSD gonads at risk or those of XX age-matched controls was as low as those of XY littermate controls (filled inverted triangles) or XY age-matched controls (open inverted triangles). Each open circle represents RNA from one embryonic gonad pair, except for d34, which represents RNA from a gonad pool. Differential expression was statistically tested for XX DSD gonads compared to XX controls at d37-39 and d42-44 (* p-value<0.005; ** p-value ≤ 0.00005). FPKM is Fragments Per Kilobase of transcript per Million mapped reads.

https://doi.org/10.1371/journal.pone.0186331.g004

Download:

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

Fig 5. SOX9 expression in canine embryonic gonads was measured by RNA-seq.

(A) At d37-39, SOX9 expression was significantly greater in XX DSD gonads at risk (filled triangles) compared to XX age-matched control gonads (open triangles), but <2 fold and primarily due to one sample (C3512). Each open circle represents RNA from one embryonic gonad pair, except for d34, which represents RNA from a gonad pool. Differential expression was statistically tested for XX DSD gonads compared to XX age-matched controls at d37-39 and d42-44 (** p-value ≤ 0.00005). (B) At all ages tested, SOX9 expression in XY littermate (filled inverted triangles) and XY age-matched control gonads (open inverted triangles) was much greater than in XX DSD gonads at risk (filled triangles) and XX age-matched controls (open triangles). Each open circle represents RNA from one embryonic gonad pair, except for d34, which represents RNA from a gonad pool. FPKM is Fragments Per Kilobase of transcript per Million mapped reads.

https://doi.org/10.1371/journal.pone.0186331.g005

Download:

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

Table 5. Ovary and testis pathway gene expression was measured by RNA-seq.

https://doi.org/10.1371/journal.pone.0186331.t005

PRO-seq confirmed RNA-seq results.

PRO-seq provides insights into gene regulation by measuring the location and density of active RNA polymerase genome-wide. It produces a quantitative picture of gene transcription at a single time point [47], and identifies the direction of gene transcription (+/- strand). Therefore, we used PRO-seq as a second assessment of embryonic gonadal gene expression. Embryos (d39) were produced from the model pedigree and a beagle control litter (Table 6). Results confirmed those of RNA-seq at d39 (Table 5). High signal intensity near RSPO1 was found in XX age-matched control gonads, lower signals were found in the XX DSD gonads at risk and lowest signals in XY littermate controls. However, there were insufficient sample numbers to statistically test these differences. High signal intensity near SOX9 was found in XY littermate control gonads, with low signals in XX DSD gonads at risk and XX age-matched controls (Fig 6). Signals were not detected near BTBD17, KIF19 and F1PLR5.

Download:

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

Fig 6. Genome browser view of the raw PRO-seq signal near the SOX9 gene in d39 gonads.

SOX9 expression was not significantly elevated in d39 XX DSD gonads at risk compared to those of XX age-matched controls. The SOX9 signal is highest in XY littermate control gonads (C3640, C3639, C3557) and much lower in the age-matched XX control (A1114) and XX DSD gonads at risk (C3641, C3682, C3556, C3681). For each sample, histograms show reads aligned to the plus strand (top, red) and minus strand (bottom, blue). Listed at left are embryo ID number, group (XX/XY) and genotype at the insertion locus: wild type (-/-), homozygous (G+G+), and heterozygous (G+/-) for the guanine insertion associated with XX DSD. RefSeq gene annotation is shown below the plot (CanFam3.1).

https://doi.org/10.1371/journal.pone.0186331.g006

Download:

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

Table 6. Canine embryonic gonad pairs collected for PRO-seq.

https://doi.org/10.1371/journal.pone.0186331.t006

An epigenomic model for gonadal mosaicism.

An epigenomic model for gonadal mosaicism in the canine XX DSD model could account for the wide range of XX DSD phenotypes observed in related individuals. Phenotypically, gonads observed in the canine XX DSD model are phenocopies of those resulting from sex chromosome mosaicism, which are composed of XX and XY clones developing as a mosaic of ovary and testis. The proportion of ovary or testis depends upon the proportion of XX and XY clones. Similarly, we propose a model of gonadal mosaicism in which the canine XX DSD genetic defect disrupts epigenomic mechanisms that control cell commitment to the testis or ovary pathway. Epigenomic mosaicism has been described in mice having variable expression at the agouti locus. This causes mosaic coat colors in individuals and highly variable expression between littermates according to parent of origin [72]. A similar process is proposed for XX DSD in the canine model. The canine XX DSD associated insertion is located within a putative CpG promoter. These sites have the potential to affect gene regulation over large areas of DNA and can be sensitive to genomic background and methylation patterns that differ by parent of origin. For example, a recent study showed that multiple methylation sites imprinted by parent of origin influenced gene expression over a 1.9MB region on human chromosome 15 [73]. Their population analyses of differentially methylated regions predict that loss of imprinting at such sites could cause congenital defects that have incomplete penetrance or variable expression. This may be applicable to XX DSD, as differences in methylation pattern and chromatin marks were found in a RevSex duplication shared by an XX DSD patient, the XY father, and unaffected paternal grandmother [74]. The epigenomic mosaic model we propose predicts that epigenomic differences in canine XX DSD lead to varying degrees of RSPO1 downregulation between gonadal clones. In some clones, RSPO1 downregulation is insufficient to prevent cell commitment to the ovary pathway. In others, it is so severe that cells commit to the testis pathway. Thus a mosaic pattern of ovarian and testicular clones would develop within the same gonad. This mechanism would also account for the wide range of gonadal phenotypes observed in affected individuals from the same family, as well as differences in penetrance observed with changes in canine genomic background. Further studies are needed, since this would be a novel mechanism for inherited canine XX DSD, and possibly for human families segregating this XX DSD subtype.

The canine XX DSD model mirrors the rare human XX DSD subtype.

The wide range of XX DSD gonadal phenotypes observed in the canine model is strikingly similar to those observed in human families in which both nonsyndromic OT-XX DSD and T-XX DSD occur [20–27]. The canine XX DSD model closely resembles the nonsyndromic human disorder in expressing both phenotypes within one family and the chromosomal location of the associated insertion. Although the cause in humans is unknown, this type of XX DSD has been mapped to a large region of chr17 (69100001–76800000; OMIM 278850). The XX DSD associated guanine insertion we have identified lies within the canine ortholog for this region.

While rodent models have been useful, they do not have the same range of phenotypic variation observed in these human families. Furthermore, exploration of epigenetic mechanisms in most models has been uncommon. The XX DSD gonadal phenotypes observed in the canine model resemble human patients with RSPO1 variants, where the RSPO1 null is associated with testes, and RSPO1 deficiency is associated with ovotestes. However, Rspo1 knockout mice develop only ovotestes [5]. In Rspo1/Sox9 double knockout mice, ovotestes develop in XX mice and other genes in the SOX family are thought to induce testis cells [8]. However, in that model, XY mice develop hypoplastic testes, whereas XY siblings of human XX DSD patients are normal, as are XY males in the canine model. Therefore, none of these rodent models mirrors this subtype of human XX DSD.

XX DSD patients with CNVs near SOX9 are also relevant for comparison, and such syndromic and nonsyndromic patients were recently reviewed [75]. To summarize, the CNV identified are presumed to alter testis-specific regulatory regions of SOX9 to cause XX DSD. Recent studies refined this region from 42 to 24kb [76,77]. In one of these patients, epigenetic differences were proposed to explain phenotypic expression in the XX DSD patient and absence of expression in the paternal grandmother having the same duplication [74]. Specifically, differences in chromatin signature and methylation patterns identified in adult fibroblast cultures suggested that chromatin modifications may regulate sex determining gene expression. At present, the molecular cause of XX DSD remains unknown in many cases, particularly those patients in which CNVs in the RevSex region or SOX9 duplications have not been identified [78,79]. Also, none of the XX DSD patients from families in which both nonsyndromic testicular and ovotesticular XX DSD have been reported [20–27] has been mapped to the RevSex region. The variant identified in this study that is associated with canine XX DSD is relevant to such patients. Examination of the orthologous region in 17q24, surrounding and containing BTBD17, is warranted. In addition to further patient studies, canine studies should be useful to identify differences in gene expression and epigenetic marks in embryonic gonads instead of, or in addition to, adult fibroblasts. Such studies could define the steps in molecular pathogenesis, including the role of epigenetic factors in phenotypic penetrance, and the cause of ovarian pathway suppression in developing gonads.

Diagnostic criteria for canine XX DSD.

In previous studies of XX DSD in the model pedigree, all affected dogs had a normal female karyotype (78,XX). Diagnosis in this study was determined by two criteria: 1) presence of seminiferous tubules in either gonad and 2) absence of SRY in genomic DNA (below). The proportion of seminiferous tubules in each gonad was estimated from serial histologic sections as: entirely testis (t), ovotestis with greater than one half composed of seminiferous tubules (ovt >0.5t), or ovotestis with less than one half composed of seminiferous tubules (ovt <0.5t, S2 Fig). Neonatal or prepubertal gonads were obtained from most affected dogs, and adult gonads were obtained from breeding stock. For purebred pet dogs, affected dogs were identified by attending veterinarians and diagnosed by the PI (M-W) using the same criteria as in the model pedigree. Where gonadal serial sections from XX DSD pet dogs were unavailable to the PI, gonads were ranked as testis (t), ovotestis (ovt), or ovary (ov) based upon a histopathology report from a veterinary pathologist.

Each animal was genotyped at the SRY locus by PCR. Genomic DNA was extracted by the phenol chloroform method [80] or a commercial kit (DNeasy blood and tissue kit, Qiagen). Samples were assayed in triplicate for SRY and in duplicate with positive control primers in parallel. Primers were designed from canine SRY (GenBank AF107021, [34]): CfSRY 378F: CGGAGGAAACGGTAGAGACA; CfSRY 675R: GGCTGCAGGTAGCAATTTGT. These flank the HMG box, avoiding hybridization to conserved sequence. Positive control primers were HMGN2F: GCCATGTCAGAAACAGTTGG and HMGN2R: AAAGGCAGATGCTAACTGAGG. These primer sets produce 298 bp and 425 bp amplicons, respectively. Reactions were run in a standard 4 step thermocycling protocol, reducing annealing temperatures stepwise from 63C to 51C for a total of 37 cycles. The CfSRY product was not generated from templates of affected dogs or control females (SRY-negative), but was generated from male controls (SRY-positive). An HMGN2 product was generated from all templates.

GWAS genotyping.

Genotyping of 27 dogs (Table 1) was performed using the Affymetrix version 2 canine SNP array (www.broadinstitute.org/mammals/dog/caninearrayfaq.html) and processed with the GeneChip Mapping 250K Sty Assay protocol (Affymetrix Inc., Santa Clara, CA). Genotypes were generated from batched sets of the Affymetrix probe result files (CEL) and were analyzed with canine genotyping software [81]. To identify suggestive variants for further analysis, we analyzed the genome-wide set of genotypes using the GMMAT software [82] to account for confounding effects of genetic relatedness and breed structure among samples, where the random effects matrix was constructed in GEMMA [83] and five Principal Components (PCs) calculated from the genotype matrix filtered for MAF <0.01 and thinned for every 100^th SNP were included as fixed effects.

Further associated variant discovery in the GWAS region by genomic tiling array for SNP genotyping. DNA samples from 5 dogs were screened with a custom canine genomic tiling array (385K Nimblegen Sequence Capture System, Roche Diagnostic Corp., Indianapolis, IN) composed of 60 bp repeat masked probes overlapping by 3–5 bp in a 2MB region (2604935–3349954 and 5922063–7019389; CanFam3, Fig 2). Coverage for target bases in this design was 97%. Five libraries were prepared from DNA eluted from the array (DNA library preparation kit, Illumina, Inc., San Diego, CA) with approximately 100 fold amplification, judged by quantitative PCR. Libraries were directly sequenced, producing 86 nt reads (GAII, Illumina, BRC Genomics Facility, Cornell University). Read quality was checked (FASTX Toolkit), identifying 83 nt quality reads of single end sequence. Reads were aligned to the reference genome (CanFam2) using default parameters (Novoalign V2.05.04). After conversion to BAM format (Samtools), SNPs and Indels in aligned reads were identified using the Genome Analysis Toolkit (GATK) and genome browsers (IGV, USCS). A total of 8,300 SNPs and 2,182 Indels were selected for further evaluation. These were filtered, eliminating noninformative markers and those in repeats, and retaining those in conserved regions (PhastCons, genome.ucsc.edu ). From these, 5 SNP haplotype blocks of homozygous alternate SNP alleles (Regions 1–5) were identified that were the same in ACS1, ACS2 and XX DSD in the model pedigree, but different from those in controls and the reference genome. This provided 732 SNPs/Indels for further homozygosity mapping in SNP haplotype Regions 1–5. Of those, 108 loci were selected for genotyping based on the P-value in GWAS, high conservation (PhastCons, genome.ucsc.edu ), or low AF (<0.15) in village dog populations [84]. Sanger sequencing of PCR products was used to genotype 28 affected dogs in the XX DSD pedigree, unrelated affected dogs (8 ACS, 3 ECS, and 29 from 19 other breeds), and 7 female unaffected controls (6 BGL, A168).

Sanger sequencing of PCR products. For SNP genotyping, primers flanking the nucleotide of interest were designed from CanFam3.1 (genome.uscs.edu) using Primer3 software (frodo.wi.mit.edu) and in silico PCR (genome.uscs.edu). PCR was conducted using the 4 step thermocycling protocol, as for SRY PCR (above). PCR products were gel purified (QiaQuick gel extraction kit, Qiagen, Valencia, CA) and Sanger sequenced (ABI 3730xl capillary electrophoresis, BRC Genomics Facility, Cornell University). Sequence texts were evaluated with Clustal software (ebi.ac.uk/Tools/msa/clustalo/) and alleles were verified in electropherograms.

CNV detection and SNP genotyping: Region 3, canine RevSex ortholog, the region near CFA9:6048201, and the SOX9 coding region.

MLPA for CNV detection in Region 3 haplotype and the canine RevSex ortholog. MLPA probes were designed according to manufacturer’s directions (Version 12, www.mrc-holland.com) using the Oligo and Peptide design Tool (sigmaaldrich.com). For primer design, conserved sequence between human RevSex and the canine ortholog was chosen by using PhastCons, 100 vertebrates basewise conservation by PhyloP, Multiz Alignments of 100 vertebrates, and BLAT functions (genome.ucsc.edu ). Final probe mixes included control probes on other autosomes and 10 test probes in Region 3, or 7 probes in the canine RevSex ortholog (listed in S2 Table). DNA was extracted from blood or tissue samples using a commercial kit (DNeasy blood and tissue kit, Qiagen) and assayed according to manufacturer’s directions (one-tube protocol, June 2011 version, MDP-v002, MRC-Holland, The Netherlands). Reaction products were submitted for fragment analysis (Big Dye Terminator capillary electrophoresis, ABI 3730, Pathology Department, University of Melbourne Medical School, Melbourne, Australia). Dogs tested in Region 3 and the canine RevSex ortholog included both founder sires, proven carrier females, and XX DSD from the model pedigree, as well as XX DSD pet dogs unrelated to the model pedigree and female and male controls. Affected dogs tested in both regions included T-XX DSD and OT-XX DSD phenotypes (listed in S3 Table).

SNP genotyping in the canine RevSex ortholog by Sanger sequencing. The canine RevSex ortholog (CFA9:17567133–17643805; CanFam3) was identified using the minimum human RevSex region shared by patients DSD2 and DSD3 [13] and LiftOver function (genome.ucsc.edu). Primer design and PCR were conducted as for SRY PCR and Sanger sequencing (above). Genomic DNA was assayed from 5 dogs (ACS2, C3481, C3023, 2 BGL controls, Table 1) and XX DSD pet dogs from breeds unrelated to the model pedigree.

ddPCR to identify CNVs near CFA9:6048201 and within the SOX9 coding region. Droplet Digital PCR (Bio-Rad Laboratories, Inc., Cambridge, MA) was used to test 5 XX DSD, ACS1 and ACS2 from the model pedigree, and a female control (A168). DNA was BstEII digested prior to PCR (New England BioLabs, Inc., Ipswich, MA). A control and 5 test sets of ddPCR primers and probes (listed in S7 Table) were designed according to manufacturer’s recommendations (Bulletin 6407 Rev A, Bio-Rad Laboratories, Inc.) using public websites (idt.com, sigmaaldrich.com, appliedbiosystems.com). Sequence immediately flanking the 6048201 insertion did not meet probe design standards due to high GC content. The SOX9 probe binds to a portion of the HMG box in exon 2. Primers and double labeled probes (5’FAM, 3’BHQ1) were manufactured by Eurofinns (Louisville, KY) and Integrated DNA Technologies (Coralville, Iowa), respectively. The ddPCR was conducted according to manufacturer’s recommendations (Bio-Rad Laboratories, Inc., Hercules, CA), using the QX100 Droplet Digital PCR system (BRC Genomics Facility, Cornell University). Briefly, reactions (20ul) containing digested DNA (60 ng), 0.9uM of each primer, 0.25uM of each probe, 1X BioRad Supermix and sterile ultrapure water were assayed in duplicate as absolute quantification experiments (ABS, QuantaSoft software, QX100, Bio-Rad Laboratories).

WGS in the canine model to identify additional variants associated with XX DSD.

DNA was extracted from blood or tissue samples using a commercial kit (DNeasy blood and tissue kit, Qiagen, Valencia, CA) and evaluated by Bioanalyzer (Agilent Technologies, Santa Clara, CA). A barcoded DNA library for each dog was prepared (TruSeq kit, Illumina), then 2 libraries/lane were sequenced (2x150 bp sequencing, HiSeq Rapid Run, Illumina, BRC Genomics Facility, Cornell University). Using vcftools, we identified all variant SNP/Indel alleles matching the predicted segregation pattern in the 4 dogs.

Fine mapping by custom array including the WGS discovered variants to identify an XX DSD associated variant in purebred pet dogs. Affected Pets and Control Pets 1 and 2 (Table 2) were genotyped with a custom designed SNP array of 75 test variants selected from the 244 loci in the WGS region and 9 control SNPs. The array manufacturer (Agena Bioscience, San Diego, CA) designed allele-specific probes at each locus, using sequence provided by investigators. DNA (500ng) from 189 dogs was sheared and size selected, then hybridized to arrays under optimal conditions. For quality control, 3 samples from the Affected Pets group were assayed in duplicate. After removal of nonhybridized DNA, fluorescent signals were identified by array reader and aligned to the allele-specific probe location on the array, generating genotype calls at each locus (Agena Bioscience, SanDiego, CA). Genotype associations were analyzed with PLINK (pngu.mgh.harvard.edu/~purcell/plink/). Genotypes at probes excluded for low call rate were obtained by Sanger sequencing of PCR products, as above.

Variant analysis in purebred pet dogs to identify an XX DSD associated variant. DNA samples from 3 purebred pet dog groups unrelated to the XX DSD pedigree were selected for analysis (Table 2). Affected Pets were diagnosed by the PI as above (see Animals). For Control Pets 1 and 2, DNA was purchased from the Cornell Veterinary BioBank, an archive of DNA from pet dogs examined at the Cornell University Hospital for Animals, College of Veterinary Medicine, Ithaca, NY [85]. Samples were paired to medical information from a standard medical screen, including physical examination of the external genitalia by board certified veterinary specialists. DNA for Control Pets was extracted using commercial reagents (Gentra PureGene Blood Kit, Qiagen, Valencia, CA). DNA for Affected Pets was extracted by the phenol chloroform method [80] or commercial reagents (DNeasy blood and tissue kit, Qiagen) and some were whole genome amplified (REPLI-g Midi kit, Qiagen Inc., Valencia, CA). Breeds by group are listed in S6 Table.

Genotyping at or near CFA9:6048201. Dogs in the model pedigree and experimental breedings (Fig 3) were genotyped by sequencing PCR products that included the 6048201 locus (S5 Fig), as in Sanger sequencing (above). Genotypes for pet dogs were confirmed by this method. To identify sequence variations near CFA9:6048201, PCR primers were designed to amplify products (230–992 bp) of contiguous sequence within 6046484–6048494 (CanFam3). A Taq enzyme for GC rich targets was used, according to manufacturer’s recommendations (KOD Xtreme Hot Start, Novagen, Merck KGaA, Darmstadt, Germany). Templates included DNA from XX DSD, ACS1 and ACS2 from the model pedigree, and female controls (beagles and A168).

Gonad collection for RNA-seq and PRO-seq analyses.

Embryos at risk of XX DSD and XY littermate controls were produced by ACS/BGL females bred to ACS2 or ACS/BGL XY males in the study pedigree. Because the causative mutation is unknown, and there are no ACS lines certified free of XX DSD, age-matched controls were collected from BGL parents of the same genetic background as those bred into the study pedigree. The day of embryo collection was timed from d0, the day of the preovulatory serum luteinizing hormone (LH) surge and concomitant rise in serum progesterone (P4) concentration in the dam [86]. By using serial preovulatory P4 concentrations alone to identify d0, unassisted parturition occurs at d65 (+/- 2). This method is 90% accurate in predicting gestation length in all breeds, regardless of litter size and body weight of the dam [87,88] and has been used to time ovulation and gestation for in vitro fertilization [89], embryo transfer [90], and previous gonadal expression studies [49,91]. Embryos were collected by hysterotomy under general anesthesia in aseptic conditions and maintained in RNAse free phosphate buffered saline (PBS, 4C) throughout microdissection. Crown rump length was measured and photographs were taken during microdissection (DP11 digital camera, Olympus America, Melville, NY). Samples for DNA, RNA, or chromatin, cataloged by embryo identification number, were flash frozen and stored in liquid nitrogen. Each embryo was developmentally staged according to photographed external morphology and genotyped by PCR at SRY and CFA9:6048201 loci, as above (Animals).

RNA-seq library preparation and analysis of gene expression in embryonic gonads.

For library preparation, total RNA from paired gonads of each canine embryo was extracted using commercial reagents (RNeasy Plus Mini Kit, Qiagen, Hilden, Germany). However, d34 gonads were too small to provide sufficient RNA, therefore gonads from these were pooled into two d34 control groups according to SRY assay result (+/-), and RNA was extracted from each pool. For all others (d35-44), barcoded cDNA libraries were prepared from total RNA of each gonad pair (TruSeq RNA Sample Prep Kit v2, Illumina, San Diego, CA), then pooled for sequencing according to group (XX DSD at risk or control). Libraries were sequenced with 75-100nt single end reads (NextSeq500 or HiSeq 2000/2500, Illumina, BRC Genomics Facility, Cornell University). Sequence output was sorted by barcode to generate one transcriptome per gonad pair. An overall average of 40M reads was generated per library. Prior to analysis, reads were trimmed with cutadapt v1.8 to remove low quality and adaptor sequences from the 3’ ends, and filtered to remove trimmed reads <20nt (parameters: -m 20 -q 20 -a AGATCGGAAGAGCACACGTCTGAACTCCAGTC—match-read-wildcards). Reads were mapped to the CanFam3 reference genome with tophat v2.0, using the Ensembl annotated transcriptome. Overall, an average of 30M reads per library was mapped to the reference genome.

Cufflinks v2.2 (cuffnorm/cuffdiff) [92] was used to generate gene expression values in FPKM units for all annotated genes, and to perform statistical analysis of differential gene expression between samples from XX DSD gonads at risk and XX age-matched controls and at gestational age groups d37-39 and d42-44. Because FPKM units are normalized to the length of the transcript, these units are proportional to gene expression levels and can be compared across genes and between samples. The cuffdiff software uses q-value cutoff of 0.05 to distinguish genes with significantly different expression between groups. A more stringent list of differentially expressed genes was used for each age group by requiring a minimum 2-fold change between XX DSD gonads at risk and XX age-matched control groups [abs(log2FC)≥1] and at least one group to have an average of FPKM ≥4.

The Jmp Pro 11 software (SAS, Cary, NC) was used for principal components analysis (PCA) and hierarchical clustering. PantherDB (pantherdb.org) was used for statistical over representation tests for GO terms [93]. PCA comparing differential expression in gonads of all groups and ages confirmed expression clustered according to SRY assay result (+/-), age, and XX DSD at risk or control status. Because gene expression in XX DSD gonads at risk was more similar to that of XX age-matched controls than XY littermates, we used PCA to compare expression in XX DSD at risk and XX age-matched controls at the d37-39 and d42-44 age groups. Although expression in most samples clustered appropriately by age and status, four samples in the d37-39 group clustered with XX age-matched controls of the older group. Because developmental criteria were the same for d37 controls (A1097, A1098) and XX DSD C3512 and C3515, the latter were analyzed with the d37-39 group. Developmental criteria indicated XX DSD C3694 and C3696 and XY littermate C3691 were >d39 but <d42 in comparison to controls, and these were analyzed with the d42-44 group.

PRO-seq library preparation and analysis of gene expression in embryonic gonads.

Canine embryonic gonad pairs were assigned to test or control groups according to SRY assay result (+/-) and genotype at the insertion locus (Table 6). Gonad pairs were homogenized by mechanical disruption (Pestle Motor Mixer) in NUN buffer (20 mM HEPES, pH 7.5, 7.5 mM MgCl2, 0.2 mM EDTA, 0.3 M NaCl, 1M Urea, 1% NP-40, 1 mM DTT, 20 units/ml SuperaseIn RNase Inhibitor, 1X Protease Inhibitor Cocktail). Chromatin was pelleted by centrifugation (30 mins, 4C) and re-suspended in storage buffer (50mM Tris-HCl, pH 8, 25% glycerol, 5mM MgAc2, 0.1mM EDTA, 5mM DTT, 40 U/ml SUPERase In RNase Inhibitor) by sonication (10 mins, Duty cycle: 30 sec on/30 sec off) in a Bioruptor (Diagenode Inc., Denville, NJ).

PRO-seq library preparation was completed as previously described [47,94] with three minor modifications. 1) The run on reaction was completed on isolated chromatin. 2) Only two biotinylated (U and C) and two non-biotinylated (G and A) nucleotides were used, providing high resolution for the RNA polymerase active site location at an economical cost per library. 3) Adapters were modified to contain a 6 nucleotide molecule-specific barcode ID at the 5’ end of each sequenced RNA. Samples were sequenced using a single lane (Illumina NextSeq500, BRC Genomics Facility, Cornell University).

Data were processed and samples aligned using our publicly available alignment pipeline (github.com/Danko-Lab/utils/tree/master/proseq, commit: 425de21). Briefly, after removing reads that failed Illumina quality filters, duplicate reads were removed and the 6 bp molecule-specific ID was trimmed from the 5’ end of each read using prinseq lite [95]. Adapters were trimmed using cutadapt with a 10% error rate [96]. Finally, reads were mapped with BWA [97] to a reference genome comprised of CanFam3.1 and a single copy of the Pol I ribosomal RNA (GenBank U13369.1). Samples were visualized using custom track hubs on the WashU epigenome browser [98]. The dREG software program was used to identify the location of promoters and active enhancers in combined PRO-seq data [99], combining male and female samples to improve statistical power for identifying transcriptional regulatory elements.