Our haplotype tests have greater power, but diplotype tests were CHIR99021 included for the information they provide regarding inheritance models. Quantitative phenotypes (WISDM-68 scales, withdrawal, and days to relapse) were tested as both continuous and dichotomous variables to attain a comprehensive appraisal of their genotype relationships. Categorical scoring of these phenotypes was included because research suggested nonlinear relationships between the individual WISDM-68 PDM scales and dependence criteria (Piper et al., 2004) and inspection of the relationships between dependence scale deciles and the frequency of HA relative to HC in early-onset smokers in the present sample indicated nonlinearity for some scales. Categorical coding of WISDM-68 scales and withdrawal was based on median splits of the entire sample across age-at-onset conditions.

For the dichotomous 90-day relapse variable (0=no relapse, 1=relapse), the 46 participants who were lost to follow-up within the first 90 days after quitting but before they reported a relapse were assumed to have relapsed the day following their last follow-up contact. Logistic regression was used to test associations between genotypes and dichotomous phenotypes. In logistic regression haplotype analyses, HA was chosen as the reference condition and the HA versus HC contrast was of primary interest; in diplotype analyses, homozygous diplotype AA was chosen as the reference condition. Given our a priori prediction that CHRNA5-A3-B4 effects on nicotine dependence severity would be more robust in early-onset smokers (Weiss et al.

, 2008), all haplotype�Cnicotine dependence relationships were tested for age-at-onset interactions using logistic regression analyses, since interaction tests can be biased by nonlinearity. If the interaction was significant, associations between the haplotype or diplotype and the phenotype were tested by age-at-onset condition; otherwise, they were tested across age-at-onset conditions. Continuous WISDM-68 scale scores and withdrawal were analyzed using the general linear model (GLM). Significant haplotype or diplotype effects were followed by Tukey pairwise comparisons of the least-squares haplotype or diplotype means. Although all pairwise genotype comparisons were tested, we made an a priori prediction, based on our previous results with the FTND (Weiss et al.

, 2008), that HA would be associated with greater nicotine dependence than HC. For the nondichotomous days-to-relapse measure, a nonparametric Cox model survival analysis was tested using Kaplan�CMeier estimation of relapse probability and the Mantel�CHaenszel method for calculating the log-rank test. If the overall haplotype or diplotype test was significant, pairwise comparisons were made. In survival analyses, the days-to-relapse variable was censored for subjects lost to follow-up within 90 days who had not reported a relapse by the date of their Cilengitide last follow-up assessment.