Comprehensive multidimensional phenotypes provide unique opportunities to distill phenotypic associations and to further validate genetic findings. Our study is one of the first to examine different nicotine dependence phenotypes as a means of clarifying the association between identified genetic variants and the clinical/behavioral features of smoking. We focused on four genetic variants that have passed the threshold of genome-wide significance in large-scale meta-analyses using CPD as the primary phenotype.
We first examined rs16969968, a variant that changes an amino acid in the α5 nicotinic receptor protein. Our results are consistent with the previous finding that this gene cluster is associated with a broad range of nicotine dependence phenotypes (Baker et al., 2009
). In addition, consistent with prior findings and our hypotheses, we found that across different dependence instruments, rs16969968 was associated with “primary” dependence: for example, CPD from the FTND, “Craving” from DSM-V
criteria, “Drive” from the NDSS, and “Loss of control” from the WISDM. That is, rs16969968 is associated with the dependence features that are assessed in the PDM: smoking that is heavy, out of control, and manifests in strong craving. Consistent with this, and as predicted, rs16969968 shares strong relations with the NDSS total score (2.8 × 10−7
) and the PDM composite (3.4 × 10−7
). Craving is a key element of this core dimension, consistent with its correlations (r
= .69) with CPD, strong genetic associations with CHRNA5
, and its theoretical basis as one of the fundamental motivational processes for nicotine dependence (Piper, Bolt, et al., 2008
The key role of smoking heaviness is revealed by the finding that none of the comprehensive nicotine dependence phenotypes including DSM-IV
, NDSS, and WISDM strengthened the genetic association with rs16969968 beyond the CPD measure. In fact, the level of statistical significance was relatively weak for DSM-IV
symptom count compared with the other measures. This may represent a bias in our sample, given that it was selected based on the FTND criteria. However, the strong genetic relations between rs16969968 and the other nicotine dependence phenotypes such as NDSS and WISDM argue against this view. The weaker genetic association with DSM-IV
symptom count is consistent with twin data of heritability for nicotine dependence. The heritability of nicotine dependence is lower when defined by DSM-IV
(56%) than when defined by the Heaviness of Smoking Index (71%), an abbreviated version of the FTND that includes only two items, CPD and TTF of the day (Heatherton et al., 1989
) or CPD alone (70%; Lessov et al., 2004
Perhaps the heritability of the different definitions of nicotine dependence is related to the origins of these measures of smoking. The FTND, NDSS, and WISDM were developed specifically for the assessment of nicotine dependence, whereas DSM-IV nicotine dependence criteria were developed as part of a general measure for all substance dependence. The application of general dependence criteria in DSM-IV is parsimonious but may not capture important characteristics, including genetic features, that are specific to nicotine dependence.
As compared with the chromosome 15 finding, the other variants studied were not as strongly associated with nicotine dependence. The pattern of phenotypic association with rs6474412 was similar to that with rs16969968. Significant associations were seen with FTND, NDSS, and WISDM scores, especially those associated with heavy smoking and craving (e.g., WISDM “Tolerance” and NDSS “Drive”). Associations with DSM-IV-defined nicotine dependence or symptom count were not significant. Therefore, rs6474412, like rs16969968, appeared to be significantly associated with compulsive heavy smoking and craving. It is unclear why the FTND item “Can’t refrain from smoking” showed the strongest association with rs6474412, but this could be related to its significant correlation with CPD (r = .59). The strongest association with rs6474412 was with the WISDM “Tolerance” scale, which comprises items, such as “I consider myself a heavy smoker,” again affirming that rs16969968 and rs6474412 are associated with overlapping domains of nicotine dependence characterized by heavy smoking.
Before the biological mechanisms are clarified for CHRNA5 and CHRNB3, the possibility of having distinct but related phenotypes with these two variants cannot be ruled out. The third variant, rs3733829 in the EGLN2 gene about 40 kb from the 3′ end of CYP2A6, showed a modest association with nicotine dependence. When the genetic association was modest, we were unable to differentiate the degree of association based on the different phenotypic definitions.
Our data do not support the previously reported association with rs1329650 on chromosome 10. Most nicotine dependence phenotypes showed no evidence of association (p
> .50). Though DSM-IV
nicotine dependence demonstrated a weak association, the effect was in the opposite direction from what was previously reported; thus, our results must be interpreted as nonreplication (TAG, 2010
). Based on the allele frequency and our sample size, our study has sufficient power to detect an association with OR
of 1.14 or higher (Gauderman & Morrison, 2006
There are several limitations to our study. Our sample selection for cases and controls was based on extreme FTND scores, so a comparison of FTND results with the other nicotine dependence phenotypes cannot be directly made. We purposely examined the strength of genetic associations within FTND measures as one set of analyses. In a separate analysis, we examined the level of genetic association among other nicotine dependence phenotypes (DSM-IV
, NDSS, and WISDM). In order to adjust for the ascertainment bias that was built into our study design, we performed additional analyses correcting for the bias (Lin & Zeng, 2009
) and obtained similar results.
Second, caution is needed in interpreting these intertwined clinical constructs. There is moderate to high correlation among the examined phenotypes and subphenotypes. As a consequence, the pattern of significant findings can be somewhat misleading. For instance, the DSM “Craving” item has a significant stepwise association with rs16969968, but the WISDM “Craving” item did not (instead, the WISDM “Loss of Control” subscale was significant in the stepwise tests). In fact, both craving measures are similarly highly associated with rs16969968 as indicated by their CIs (), suggesting no real inconsistency. Also, because all the dependence measures are meaningfully correlated with one another (given that they measure a common construct), there is limited power to demonstrate significant differences among them in their relations with genetic variants. However, even given that, this study shows that the key rs16969968 and rs6474412 genetic variants were reliably associated only with certain types of dependence measures: those reflecting heavy out-of-control smoking accompanied by strong craving. For none of the genetic variants were the associations of the WISDM SDM scales as strong as those for the PDM scales.
Another limitation is that only selected genetic variants were tested. These variants were identified in several large Genome-Wide Association Studies (GWAS) of smoking, which have used the simple measure of CPD as the primary phenotype (Liu et al., 2010
; TAG, 2010
; Thorgeirsson et al., 2010
). These variants were selected on the basis of their prior associations with CPD, so our finding of CPD as one of the strongest association findings may be the result of a bias based on the original phenotype that identified these variants. The more comprehensive measures of nicotine dependence are not widely available in genetic samples, and no GWAS has been performed on the WISDM or NDSS. Until GWAS using these other phenotypic definitions is performed, we will not be able to determine if there are additional novel variants associated with these other phenotypes to be identified.
Finally, our goal was to examine phenotype–genotype relations by comparing the genetic associations across phenotypes and characterizing the subphenotypes’ best capturing known genetic associations. This work informs our understanding of phenotypes and the mission to improve diagnostic validity. For example, a proposed test of the validity of mental disorder diagnosis for DSM
includes identified genetic risk factors (Carpenter et al., 2009
). If biological or genetic factors are an important test of the validity of mental disorder diagnosis, this evidence suggests the importance of measuring smoking heaviness and “Craving” to reflect part of the underlying genetic risk. Though the variance explained by these genetic polymorphisms is small (2.4% by four variants for FTND), the current evidence suggests that measures of smoking heaviness, in particular, the number of cigarettes smoked per day, is an important measure of nicotine dependence that captures part of the genetic variance related to nicotine dependence. Adding “Craving” as a symptom criterion in DSM
is supported in our work. Although CPD is a simple measure from a psychiatric or psychological perspective, it is an important measure used in medicine because it assesses the toxic exposure from smoking, which is a determinant of diseases, such as lung cancer and chronic obstructive pulmonary disease. Therefore, we suggest that CPD be considered in the future assessment of nicotine use disorders, including DSM-V