Across the replication samples, the 1000 Genomes–imputed additive genotype dosages for selected SNPs and indels were tested for association with FTND-defined nicotine dependence (mild, moderate and severe) using linear regression models unless otherwise stated. Adjustments were made for age, sex and eigenvectors (first three for Yale-Penn and the number needed to account for >75% of the phenotypic variability for UW-TTURC, GAIN and nonGAIN). Sample-specific results were combined, first across the replication samples and then across all GWAS and replication samples, using inverse variance-weighted meta-analysis. Odds ratio estimates, computed as eβ for moderate vs mild dependence and e2β for severe vs mild dependence, were compared across all the samples using the Forest Plot Viewer.40
