The genotyped and 1000 Genomes–imputed SNPs and indels were tested for association with categorical nicotine dependence (mild, moderate and severe) using ProbAbel software30 in each sample with linear regression models that included age, sex and sample-specific covariates (if applicable, see Supplementary Information). The four European American and Italian samples also included principal component eigenvectors to minimize bias owing to population stratification; for each sample, we selected the number of eigenvectors needed to account for >75% of the variability in nicotine dependence.
