Copy number variation underlies complex phenotypes in domestic dog breeds and other canids

Abstract

Extreme phenotypic diversity, a history of artificial selection, and socioeconomic value make domestic dog breeds a compelling subject for genomic research. Copy number variation (CNV) is known to account for a significant part of inter-individual genomic diversity in other systems. However, a comprehensive genome-wide study of structural variation as it relates to breed-specific phenotypes is lacking. We have generated whole genome CNV maps for more than 300 canids. Our data set extends the canine structural variation landscape to more than 100 dog breeds, including novel variants that cannot be assessed using microarray technologies. We have taken advantage of this data set to perform the first CNV-based genome-wide association study (GWAS) in canids. We identify 96 loci that display copy number differences across breeds, which are statistically associated with a previously compiled set of breed-specific morphometrics and disease susceptibilities. Among these, we highlight the discovery of a long-range interaction involving a CNV near MED13L and TBX3, which could influence breed standard height. Integration of the CNVs with chromatin interactions, long noncoding RNA expression, and single nucleotide variation highlights a subset of specific loci and genes with potential functional relevance and the prospect to explain trait variation between dog breeds.

Figure 1.

Global and specific differences in CN between breed dogs, village dogs, and wolves. (A) Copy number-based principal component analysis of breed dogs (orange), village dogs (green), and wolves (purple). (B,C) Depictions of the copy number values for two highly differentiated loci: the HBB chain gene cluster (ENSCAFT00000009978, ENSCAFT00000022860, ENSCAFT00000009984) and SLIT2. Discrete copy number values for all samples are depicted in rectangle plots. Purple: wolves, green: village dogs, orange: modern dogs. Top panel: V_ST values for the same genomic windows; bottom panel: copy number window values.

Figure 2.

Analysis of the associations between CNVs and two phenotypes. (A,D) Manhattan plots of the copy number GWAS for breed standard height, retinal atrophy susceptibility, respectively (Jones et al. 2008). Red line: Bonferroni correction (−log₁₀P-value = 6.417). Blue line: one order of magnitude below Bonferroni correction (−log₁₀P-value = 5.417). P-values were calculated using different tests (Supplemental Table S3 and Methods). (B,C,E) Close-up of the relevant regions for each trait, respectively: SMAD2 (Chr 7: 43,787,168–43,801,320) and MED13L (Chr 26: 12,739,546–12,754,676) loci for height, and DMBT1 (Chr 28: 32,220,591–32,260,415) for retinal atrophy. Each sample corresponds to a line along the y-axis and is ordered according to the trait in question. The x-axis shows the genomic position of each window. CN windows for each sample are colored according to their normalized distance to the median CN in the window; the darker the shade of blue, the more the CN of a sample differs from the window median. Gray CN windows correspond to uncertain genotypes.

Figure 3.

Instances of CNVs associated with phenotypes overlapping chromatin contacts. (A,B) Hi-C interaction heat maps for significant contacts on Chromosomes 26 and 9. The white squares on the x-axis mark the position the GWAS hits. The arrows mark the position and directionality of the most significant CTCF motifs of the area. The dashed circles within the Hi-C plot mark the significant interactions involving the relevant genes and the associated CNVs.

Figure 4.

Instances of loci where specific clades have a different CN. (A–C) Representation of high inter-breed V_ST regions for the NLRP8 and NLRP13 genes, CTNNA3, and SLC7A (ENSCAFT00000025818 and ENSCAFT00000025829), respectively. Normalized copy number is represented by color, where the modal CN is the lightest color, and deviations from the mode (either deletions or duplications) are colored darker. Each row represents a sample ordered by clade, as described in Parker et al. (2017).