Genome-wide interaction study of smoking and bladder cancer risk

Abstract

Bladder cancer is a complex disease with known environmental and genetic risk factors. We performed a genome-wide interaction study (GWAS) of smoking and bladder cancer risk based on primary scan data from 3002 cases and 4411 controls from the National Cancer Institute Bladder Cancer GWAS. Alternative methods were used to evaluate both additive and multiplicative interactions between individual single nucleotide polymorphisms (SNPs) and smoking exposure. SNPs with interaction P values < 5 × 10(-) (5) were evaluated further in an independent dataset of 2422 bladder cancer cases and 5751 controls. We identified 10 SNPs that showed association in a consistent manner with the initial dataset and in the combined dataset, providing evidence of interaction with tobacco use. Further, two of these novel SNPs showed strong evidence of association with bladder cancer in tobacco use subgroups that approached genome-wide significance. Specifically, rs1711973 (FOXF2) on 6p25.3 was a susceptibility SNP for never smokers [combined odds ratio (OR) = 1.34, 95% confidence interval (CI) = 1.20-1.50, P value = 5.18 × 10(-) (7)]; and rs12216499 (RSPH3-TAGAP-EZR) on 6q25.3 was a susceptibility SNP for ever smokers (combined OR = 0.75, 95% CI = 0.67-0.84, P value = 6.35 × 10(-) (7)). In our analysis of smoking and bladder cancer, the tests for multiplicative interaction seemed to more commonly identify susceptibility loci with associations in never smokers, whereas the additive interaction analysis identified more loci with associations among smokers-including the known smoking and NAT2 acetylation interaction. Our findings provide additional evidence of gene-environment interactions for tobacco and bladder cancer.

Fig. 1.

Quantile–quantile plots for interaction P values from multiplicative models without and with independence assumption. Quantile–quantile plots for multiplicative interaction P values of smoking–SNP genome scan. P values were computed using two different methods to test for multiplicative interactions. The first method (A) used a likelihood ratio test performed by comparing two logistic regression models, one with and one without an interaction term for a SNP and smoking, did not assume independence between a SNP and smoking, and assumed an additive genetic model for each SNP. The logistic regression models were adjusted for study, age (5-year categories), gender and an interaction term for smoking and an indicator variable for the PLCO study to account for stratified sampling of controls by smoking status. The second method (B) assumed that SNP and smoking exposure are independent, using a retrospective likelihood, which exploits the gene–environment independence assumption in a general logistic regression framework.

Fig. 2.

Quantile–quantile plots for interaction P values from additive models without and with independence assumptions. Quantile–quantile plots for additive interaction P values of smoking–SNP genome scan. P values were computed using two different methods to test for additive interactions. The first method (A) does not assume gene–environment independence and was calculated using a likelihood ratio test using logistic regression models comparing saturated and additive models (27); under the null hypothesis of the additive model, the OR for the combined effect of a given SNP and smoking status is constrained so that the risk difference associated with one exposure (e.g. smoking) is constant across levels of other exposure (e.g. SNP), or the reverse, and models were adjusted for study, age (5-year categories), gender and an interaction term for smoking and an indicator variable for the PLCO study to account for stratified sampling of controls by smoking status. All tests for additive interactions were performed using categorical variables (each SNP was coded as a dichotomous variable indicating the presence of any risk allele) to avoid complex numerical issues related to non-standard model fitting procedures when using continuous variables, such as log-additive effect of SNP alleles. For testing additive interactions using a gene–environment independence assumption, we used a method proposed by Han et al. (27), which is based on the retrospective likelihood by Chatterjee et al. (25).