Complex disease and phenotype mapping in the domestic dog

Abstract

The domestic dog is becoming an increasingly valuable model species in medical genetics, showing particular promise to advance our understanding of cancer and orthopaedic disease. Here we undertake the largest canine genome-wide association study to date, with a panel of over 4,200 dogs genotyped at 180,000 markers, to accelerate mapping efforts. For complex diseases, we identify loci significantly associated with hip dysplasia, elbow dysplasia, idiopathic epilepsy, lymphoma, mast cell tumour and granulomatous colitis; for morphological traits, we report three novel quantitative trait loci that influence body size and one that influences fur length and shedding. Using simulation studies, we show that modestly larger sample sizes and denser marker sets will be sufficient to identify most moderate- to large-effect complex disease loci. This proposed design will enable efficient mapping of canine complex diseases, most of which have human homologues, using far fewer samples than required in human studies.

Figure 1. Significant across-breed disease GWAS results.

Manhattan and quantile–quantile plots, showing the statistical significance of each marker (−log₁₀ scale) as a function of genomic position for (a) hip dysplasia (CHD, as measured by Norberg Angle, n=921), (b) elbow dysplasia (ED, 113 cases and 633 controls). Colours of circles indicate the amount of LD with top-associated marker, ranging from black (r²=0–0.2) to red (r²=0.8–1). Red lines on the Manhattan plots are the significance thresholds, at P=4 × 10⁻⁷. Inflation factors (λ values) are shown on the quantile–quantile plots.

Figure 2. Significant within-breed disease GWAS results.

Manhattan and quantile–quantile plots, showing the statistical significance of each marker (−log₁₀ scale) as a function of genomic position for (a) granulomatous colitis in Boxers and Bulldogs (46 cases, 91 controls), (b) idiopathic epilepsy in Irish Wolfhounds (34 cases, 168 controls), (c) lymphoma in Golden Retrievers (34 cases, 48 controls), (d) MCT in Labrador Retrievers (152 cases, 106 controls). Colours of circles indicate the amount of LD with top-associated marker, ranging from black (r²=0–0.2) to red (r²=0.8–1). Red lines on the Manhattan plots are the significance thresholds, calculated by a Bonferroni correction of unlinked markers. Inflation factors (λ values) are shown on the quantile–quantile plots.

Figure 3. Body size association results.

Manhattan and quantile–quantile plots of (a) breed-average male weight^0.38 (n=1,873) and (b) breed-average male height (n=1,873), showing the 17 significant loci, four of which are novel (shown in bold). Red lines on the Manhattan plots are the significance thresholds, at P=5 × 10⁻⁶ (FDR of <0.5% and <0.75% for weight and height, respectively). Inflation factors (λ values) are shown on the quantile–quantile plots. (c) Proportion of variance explained (R²) by the 17 size loci in a linear model for weight (blue bars) and height (green bars), with sex and inbreeding corrections. Shown are the results for individual breed dogs (with breed included in the model shown in grey), individual village dogs and among breeds. (d) Proportion of variance explained (pve) by SNPs on each chromosome for individual weight (n=2,072) by the length of the chromosome. Points are plotted as chromosome numbers.

Figure 4. Epistasis plots for fur phenotypes.

(a) Breed average shedding (on a scale from 0=minimal to 1=heavy), showing the interaction between RSPO2 and MC5R alleles, (b) breed fur length (on a scale from 1=short to 5=long), showing the interaction between MC5R and RSPO2 alleles, FGF5 and RSPO2 alleles, and MC5R and FGF5 alleles. a=ancestral allele, d=derived allele. Breed images are used with permission from the American Kennel Club (AKC).

Figure 5. Simulation GWAS results.

(a) Between-breed and within-breed GWAS designs using a dense (1 SNP every 2 kb) array (red) and the current (1 SNP every 13 kb) array (black) with different numbers of cases and controls. Shown is the power to detect causal loci with effect size of 0.75σ. (b) Power to detect loci of different effect sizes using a between-breed GWAS design and a dense array (red) and current array (black) with 500 cases/controls and 1,000 cases/controls. (c) Proportion of significant loci that are false positives using a dense (red) and the current (black) array with between-breed and within-breed GWAS designs and different numbers of cases and controls.