Each study (including the case-cohort studies [20]) undertook analyses of up to four smoking traits using RAREMETALWORKER [21] or RVTESTS [22] (Supp. Table 2), which generated single variant score statistics and their covariance matrices within sliding windows of 1Mb. CPD and pack-years were analysed using linear models or linear mixed models. Smoking initiation and smoking cessation were analysed using logistic models or linear mixed models. All studies adjusted each trait for age, sex, at least three genetic principal components and any study-specific covariates (Supp. Table 2). Chromosome X variants were analysed using the above-described approach, but coding males as 0/2. This coding scheme ensures that on average females and males have equal dosages and so is optimal for genes that are inactivated (due to X chromosome inactivation) and is valid for genes that do not undergo X chromosome activation. Males and females were analysed together adjusting for sex as a covariate.
