Use of a predictive model derived from in vivo endophenotype measurements to demonstrate associations with a complex locus, CYP2A6

Abstract

This study demonstrates a novel approach to test associations between highly heterogeneous genetic loci and complex phenotypes. Previous investigations of the relationship between Cytochrome P450 2A6 (CYP2A6) genotype and smoking phenotypes made comparisons by dividing subjects into broad categories based on assumptions that simplify the range of function of different CYP2A6 alleles, their numerous possible diplotype combinations and non-additive allele effects. A predictive model that translates CYP2A6 diplotype into a single continuous variable was previously derived from an in vivo metabolism experiment in 189 European Americans. Here, we apply this model to assess associations between genotype, inferred nicotine metabolism and smoking behaviors in larger samples without direct nicotine metabolism measurements. CYP2A6 genotype is not associated with nicotine dependence, as defined by the Fagerström Test of Nicotine Dependence, demonstrating that cigarettes smoked per day (CPD) and nicotine dependence have distinct genetic correlates. The predicted metric is significantly associated with CPD among African Americans and European American dependent smokers. Individual slow metabolizing genotypes are associated with lower CPD, but the predicted metric is the best predictor of CPD. Furthermore, optimizing the predictive model by including additional CYP2A6 alleles improves the fit of the model in an independent data set and provides a novel method of predicting the functional impact of alleles without direct metabolism measurements. Lastly, comprehensive genotyping and in vivo metabolism data are used to demonstrate that genome-wide significant associations between CPD and single nucleotide polymorphisms are the result of synthetic associations.

Figure 1.

Distribution of modeled nicotine metabolism in European Americans.

Figure 2.

Distribution of modeled nicotine metabolism in African Americans.

Figure 3.

CYP2A6*1H and reference alleles, modeling the association with the estimated metric in dependent European Americans. The fit of the association between CPD and the predicted metabolism metric determined by the model, scanning allele parameters used to calculate the prediction between zero activity (0.0) and maximum activity (1.0). Essentially, the map demonstrates a variety of possible models, with differing relationships between their allele parameters, and the relative ability of each model to predict a metabolism metric that is significantly correlated with CPD. Points along the diagonal from lower left to upper right correspond to models where the two scanned parameters are equal. Reference alleles are all alleles excluding CYP2A6*1A, *2, *4, *9, *12 and the scanned allele *1H. The color key refers to the –log₁₀ of the P-value for the association between the metabolism metric and CPD in dependent European Americans. Coordinates color coded as >1.3 (orange) represent significant associations (P < 0.05).

Figure 4.

CYP2A6*1H and null alleles, modeling the association with the estimated metric in dependent European Americans. The fit of the association between CPD and the predicted metabolism metric determined by the model, scanning allele parameters used to calculate the prediction between zero activity (0.0) and maximum activity (1.0). Essentially, the map demonstrates a variety of possible models, with differing relationships between their allele parameters, and the relative ability of each model to predict a metabolism metric that is significantly correlated with CPD. Points along the diagonal from lower left to upper right correspond to models where the two scanned parameters are equal. Assumed null alleles are CYP2A6*2, *4 and *12. The color key refers to the –log₁₀ of the P-value for the association between the metabolism metric and CPD in dependent European Americans. Coordinates color coded as >1.3 (orange) represent significant associations (P < 0.05).

Figure 5.

CYP2A6*17 and reference alleles, modeling the association with the estimated metric in dependent African Americans. The fit of the association between CPD and the predicted metabolism metric determined by the model, scanning allele parameters used to calculate the prediction between zero activity (0.0) and maximum activity (1.0). Essentially, the map demonstrates a variety of possible models, with differing relationships between their allele parameters, and the relative ability of each model to predict a metabolism metric that is significantly correlated with CPD. Points along the diagonal from lower left to upper right correspond to models where the two scanned parameters are equal. Reference alleles are all alleles excluding CYP2A6*1A, *2, *4, *9, *12 and the scanned allele *17. The color key refers to the –log₁₀ of the P-value for the association between the metabolism metric and CPD in dependent African Americans. Coordinates color coded as >1.3 (dark yellow) represent significant associations (P < 0.05).

Figure 6.

CYP2A6*17 and null alleles, modeling the association with the estimated metric in dependent African Americans. The fit of the association between CPD and the predicted metabolism metric determined by the model, scanning allele parameters used to calculate the prediction between zero activity (0.0) and maximum activity (1.0). Essentially, the map demonstrates a variety of possible models, with differing relationships between their allele parameters, and the relative ability of each model to predict a metabolism metric that is significantly correlated with CPD. Points along the diagonal from lower left to upper right correspond to models where the two scanned parameters are equal. Assumed null alleles are CYP2A6*2, *4 and *12. The color key refers to the –log₁₀ of the P-value for the association between the metabolism metric and CPD in dependent African Americans. Coordinates color coded as >1.3 (dark yellow) represent significant associations (P < 0.05).

Figure 7.

LD (R² and D′) plot of functional variants in CYP2A6 and rs3733829 in EGLN2 among European Americans. *4 refers to the CYP2A6 deletion allele. Numbers represent the R² value expressed as a percentile. Red squares represent pairs with logarithm of odds (LOD) scores for LD ≥ 2 and D′ = 1.0, pink represent LOD ≥ 2 and D′ < 1.0, blue represent D′ = 1 but LOD < 2, white squares represent LOD < 2 and D′ < 1.0. Plot generated using HaploView v4.2.