Dog-human gene network colocalization links synaptic signalling and ubiquitination to compulsive disorders

Obsessive-compulsive disorder (OCD) is highly polygenic, and the largest genome-wide association study (GWAS) to date — 53,660 cases and 2,044,417 controls — identified only 30 loci, a small fraction of those expected to cause disease and too few to resolve the underlying biological pathways. Some dogs naturally develop a compulsive disorder (dogCD) with parallels to human OCD, and the distinctive population structure of pet dogs reduces sample sizes needed for discovery of associated loci. Here, we generated high-density genotypes (SNPsₘₐₓ = 9,885,182, MAF ≥ 1%) for 2,584 dogs with measures across 14 items and 3 factors describing the dogCD phenotypes using low-pass sequencing (mean depth = 0.83, s.d. = 0.35, N = 2,211) or SNP arrays (Axiom Canine Array A+B with 1,011,992 SNPs, N = 373) followed by imputation with the Dog10K reference panel. We tested for association using mixed-model approaches that account for population structure: MLMA-LOCO for the three latent factors and POLMM for 14 ordinal survey items and combined results to define a consensus set of non-overlapping associated loci. We then use cross-species network colocalization to combine the GWAS in 2,584 dogs, recruited through the community science project Darwin's Ark, with the OCD GWAS, identifying a 165-gene network not discernible from the current human data alone. Combining dog and human GWAS implicates an inherited risk architecture largely outside current OCD pharmacotherapy and establishes pet dogs as a tractable model for psychiatric gene network discovery, particularly for conditions underpowered in human studies alone. Data in storage includes Darwin's Ark genotype files: - affy_round3 = Axiom Canine Array A+B with 1,011,992 SNPs, N = 411 mapped to CanFam3.1 - DA_MERGED_GENCOVE_AXIOM_QC5 = 3328 dogs & 12518078 SNPs mapped to CanFam4. Imputed with the Dog10K reference panel. Dataset is merged from Axiom Canine array (lifted and imputed, N=411) and Gencove low-pass sequencing (mapped to Cf4 and imputed, N=2917). This is the dataset before separating them into 17 dogCD phenotypes. - 17 sets of sumstats (ITEM*sumstats & CCDF*_sumstats) from the GWAS of each items (N=14) and factor (N=3). - ALLFAM_sex_age_phenos.txt includes all phenotypes and covariates for dogs included in each separate plink gwas files (OBS for CCDF1 and CCF2: dogs with factor scores with only one item loading to that factor are excluded).