Statistical analysis was performed using PLINK [29] and R software. For the binary phenotypes, SI-1 and SC, the association tests were performed using logistic regression analysis, with age, sex, and geographic area of recruitment as covariates under the additive and dominant models. The cumulative logit model was fit to the ordinal phenotypes SI-2 and SQ [64]. LD analysis was performed using Haploview software [30] with the option of determining haplotype blocks according to the definitions proposed by Gabriel et al. [31]. Haplotypes were reconstructed using SNPs within the genomic region containing the CHRNA5/A3/B4 cluster with sliding window sizes of 3 to 5. Haplotype-based association analysis was carried out with the haplo.stats R statistics package [33] under additive and dominant models, adjusting for age, sex, and geographic area. As we consider this study a replication of reported significant association of variants of this gene cluster with smoking behavior in an independent sample [14], [15], [16], [17], [24], we did not correct for multiple comparisons for the two genetic models and three smoking-related phenotypes, primarily because they are related, when we interpreted
