The sum of DSM-IV criterion counts for all EA OD-exposed individuals from the Yale-Penn 1-3 samples were used as an ordinal (quantitative) trait in association analysis (Supplementary figure 1). GEMMA (10) was performed for each cohort separately, with an adjustment for sex, age, the first three PCs of ancestry, and the degree of relatedness among subjects. In our previous study, we also adjusted for the number of other substance dependence criteria, but did not do so here. This was intended to capture more OD information, at the cost of possible confounding by association driven in part by substance use comorbidities. Summary statistics from GEMMA were reformatted for PLINK1.9 for meta-analysis, and all variants were evaluated by matching their chromosomal positions and two alleles among the four GWAS datasets. Variants with conflicting allele information that could not be resolved by being flipped were excluded from the meta-analysis that used PLINK1.9. The inverse variance method implemented in PLINK1.9 was used to generate fixed-effects meta-analysis P-values (meta-P). In the above process, ∼8 M distinct variants present in at least in one cohort were reported in the meta-analysis; among these, ∼4.7 M variants were present among all four cohorts.
