GWAS analyses were conducted in each ancestry group to test the imputed SNP and indel genotype dosages for association with drug abuse using logistic regression models in ProbABEL(Aulchenko et al., 2007). Drug abuse was coded as a dichotomous variable: ‘1’ corresponding to UHS cases who reported abusing drugs (regardless of type of drug category or frequency of abuse) and ‘0’ corresponding to population controls. The regression models were adjusted for sex and selected principal component eigenvectors to remove potential bias due to population stratification. For each ancestry group, 10 eigenvectors were generated using EIGENSTRAT(Price et al., 2006) analyses with a pruned set of genotyped SNPs in linkage equilibrium (r2<0.5). We selected the number of eigenvectors needed to cumulatively explain >90% of the phenotypic variance, resulting in the inclusion of three eigenvectors as covariates for each ancestry group (Table S8). The ancestry-specific GWAS results were combined via fixed-effects, sample-size weighted meta-analysis, as done in other multiancestry GWAS meta-analyses,(Guo et al., 2013) using the METAL program.(Willer et al., 2010) Discovery analyses could not account for smoking or alcohol abuse history because these data were not collected in the UHS and not available in some of control data sets.
