by meta-analysis. SNPs were excluded from the meta-analyses if they were not present (either directly genotyped or well-imputed) in all studies. There were 7,445,053 SNPs for the AA sample, 5,675,219 SNPs for the EA sample, and 4,795,232 SNPs for the EA+AA sample. Regional associations were evaluated with plots generated using LocusZoom (http://locuszoom.sph.umich.edu/locuszoom).46 For laboratory participants, BQSU and MNWS effects were analyzed with linear mixed models using JMP Pro (v10.0.0) software (SAS Institute Inc.). The following independent variables were included in the models as fixed effects: rs31746 genotype (with number of risk allele copies coded additively as 0, 1 and 2), sex, population (EA or AA), time point (pre- or post-nicotine) and the interaction of time point with rs31746 genotype. A random effect was included to account for correlations between the repeated measures from each subject. We performed additional analyses to control for the potential confounding effects of differences in baseline smoking behavior, evaluating the fold-change in BQSU with models that included the following independent variables: sex, race, baseline nicotine plasma concentration, baseline cotinine plasma concentration, and baseline nicotine metabolite ratio.
