The study sample was composed of participants from three smoking cessation clinical trials: (a) open-label trial of nicotine patch versus nicotine nasal spray (n
= 285; Lerman et al., 2004
), (b) placebo-controlled trial of bupropion (n
= 317; Collins et al., 2004
), and (c) placebo-controlled trial of extended duration nicotine patch versus standard duration (n
= 428; Schnoll et al., 2010
). Analysis of pretreatment data was carried out on participants of European ancestry who had genotypes for the SNPs rs578776 and rs1051730 and complete baseline data for NMR and CPD. Within this subsample, 13 individuals had participated in more than one of the studies and were excluded, leaving a final sample of 1,030. Genotypes on the third SNP, rs16969968, were not collected in the bupropion trial, and 36 participants in the other two studies had missing data for the SNP, leaving 677 participants available for analyses involving rs16969968.
In the overall sample, 47.6% of participants were female, 40.9% were college graduates, and the mean age was 44.98 (SD = 11.01). At baseline, the mean CPD was 22.63 (SD = 9.32), mean nicotine dependence (FTND) score was 5.38 (SD = 2.20), and mean NMR was 0.416 (SD = 0.217).
Genotyping for SNPs rs16969968, rs578776, and rs1051730 was performed using the TaqMan SNP Genotyping Assays in a 384-well microplate format along with SNP-specific control samples (Applied Biosystems). The genotype distribution for the three SNPs was as follows: For rs16969968, 97 participants (14.3%) were of the A/A genotype, 325 (48.0%) were A/G, and 255 (37.7%) were G/G; for rs578776, 49 (4.8%) were A/A, 367 (35.6%) were A/G, and 614 (59.6%) were G/G; and for rs1051730, 152 (14.8%) were A/A, 491 (47.7%) were A/G, and 387 (37.6%) were G/G.
Pairwise LD (|D′| and r2
) was determined using Haploview (Barrett, Fry, Maller, & Daly, 2005
) from the subset of 677 participants who had data for all three SNPs. For rs16969968 and rs578776, |D′| = 1.0 and r2
= .182; for rs578776 and rs1051730, |D′| = 0.99 and r2
= .182; and for rs16969968 and rs1051730, |D′| = 0.98 and r2
= .957 (Supplementary Figure S1
), which are consistent with previous studies (Bierut et al., 2008
; N. L. Saccone et al., 2009
). Genotype frequencies were in Hardy–Weinberg equilibrium.
NMR data were determined by liquid chromatography—tandem mass spectrometry (Dempsey et al., 2004
) using blood samples collected at the pretreatment visit in each of the three studies. Participants were not given specific instructions regarding smoking behavior prior to collection; however, all participants reported smoking at least 10 CPD. The range of NMR values in this dataset was 0.01–2.08, consistent with previous studies (Johnstone et al., 2006
; Swan et al., 2009
). Previous studies at our center observed significant differences in smoking cessation between smokers in the lowest quartile of NMR and smokers in all other quartiles (Lerman et al., 2010
; Schnoll et al., 2009
). Based on those studies, we defined slow metabolizers as those in the lowest quartile (NMR < 0.27) and normal metabolizers as those in the top three quartiles (NMR > 0.27).
Chi-square tests and one-way ANOVAs were used to check for differences by genotype group on sex, education, and age. These variables did not differ by genotype on any of the three SNPs (p
values > .05). Three linear regression models of CPD were then estimated. In each model, the predictors were sex, age, education (college graduate vs. other), NMR, and a set of two indicator variables denoting genotype on one of the three SNPs of interest. For each SNP, the reference category was the genotype with the lowest projected risk (i.e., associated with lower CPD) based on previous association studies (S. F. Saccone et al., 2007
; Stevens et al., 2008
). For rs16969968 and rs1051730, the indicator variables denoted genotypes of A/G and A/A, with G/G the reference category; for rs578776, the indicator variables denoted genotypes of A/G and G/G, with A/A the reference category. In each model, a set of two indicator variables representing the potential interaction of NMR by genotype was also included. All predictors were entered as a block after which the interaction terms were allowed to drop out if nonsignificant (p
> .05). To obtain effect sizes for NMR within genotype, regressions of CPD on sex, age, education, and NMR were performed separately within each of the three genotype groups for each of the three SNPs.
Due to the high LD across SNPs, we used latent-class haplotypes to model CPD using the haplo.stats package for the statistical program R (Sinnwell & Schaid, 2009
). Models included haplotype class and NMR as predictors of interest and sex, age, and education as covariates to control for error. Haplotype classes were restricted to those with ≥5% frequency, lumping the remainder into “other.” We used the likelihood ratio to test the interaction between haplotype and NMR.