rate at 5% is p = 1.40 × 10−4. Since the LLFS cohort is ~40% smaller than WHI cohort (i.e., the association analysis in the replication cohort had Lower statistical power than that conducted on the discovery cohort), we considered p = 0.1 as indicative of successful replication, while bearing in mind that this “trend-level” requirement is a necessary compromise and true replication will require additional subject samples. To verify whether the association observed were due to the role of CYP2A6 in nicotine metabolism, we conducted an interaction test, analyzing the difference between regression coefficients in nicotine-exposed vs. nicotine-unexposed subjects. Because the LLFS cohort includes closely related subjects, we performed the association and interaction analyses using the R package GWAF28 to fit a generalized estimating equations (GEE) model to adjust for correlations among related individuals. Genotype-Tissue Expression (GTEx) Version 6 data (available at http://www.gtexportal.org/) were used to analyze the effect of alleles investigated on CYP2A6 expression across 36 human tissues29. A detailed description of the GTEx methods (i.e., preprocessing, expression quantification, and association analysis) used is available at http://www.gtexportal.org/static/doc/analysis/Portal_Analysis_Methods_v6_08182016.pdf. Briefly, the association between genetic variant and gene expression was conducted using a linear regression analysis considering the covariates (i.e., top-3
