Once the relevant data were permuted, statistical tests were performed for each permuted dataset. Ordinary least squares regression was used to test whether the interaction between CSA and each SNP within each gene was associated with cannabis dependence symptoms. Analyses of main effects were conducted in the same manner and were controlled for in interaction testing. Case status (opioid dependent vs. non-dependent), sex, age quintile, and three ancestrally informative principal components were entered as covariates for all analyses. In accordance with recommendations for GxE analyses, additional covariates were entered representing the interactions between the SNP and each covariate of no interest as well as CSA score and each covariate of no interest (Keller, 2014). In order to restrict the gene-level test statistic to only potentially informative SNPs and thus minimize the influence of gene size and within-gene linkage disequilibrium patterns on significance, the maximum number of SNPs within each gene passing the nominal uncorrected significance threshold of p < .05 in a single permutation was used as the number of SNPs (NmaxSNPs) across which the SNP-level ΔR2 values were averaged
