Single-Nucleotide Polymorphisms in Nucleotide Excision Repair Genes, Cigarette Smoking, and the Risk of Head and Neck Cancer

Head and neck cancer (HNC) includes tumors, principally squamous cell carcinomas, of the oral cavity, pharynx, and larynx (1). In the United States, an estimated 52,610 incident HNC cases and 11,500 associated deaths occurred in 2012 (2). Cigarette smoking is a major risk factor for HNC incidence with case–control studies consistently reporting elevated ORs for ever smoking, as well as dose–response gradients with duration and frequency (3). Among nonalcohol drinking HNC cases, 25% of cases are attributed to cigarette smoking (4).

Cigarette smoke contains numerous carcinogens, such as benzo-a-pyrene, that are known to cause DNA damage, including adducts (3, 5–7). Nucleotide excision repair (NER) enzymes are principally responsible for removing bulky DNA adducts, and are therefore hypothesized to be independent risk factors for HNC, as well as important modifiers of the association between smoking and HNC (5–7). Several previous studies have considered associations between variants in NER genes and HNC risk, but studies vary with regard to which specific single-nucleotide polymorphisms (SNP) were investigated and often present inconsistent evidence for analysis of the same SNP (8–49). In general, most previous studies have evaluated only a few SNPs on a single NER gene among a few hundred HNC cases (8–49). Few studies have examined the association of NER SNPs and HNC among African-Americans (16), a group shown to have a stronger association for smoking and HNC (50). Studies of cigarette–SNP joint effects have also been limited by sparse numbers of NER variants and small sample sizes and present varying results, though some studies indicate stronger associations among smokers with polymorphisms in NER genes (8, 10–12, 14, 16, 17, 23, 25, 27–29, 31, 32, 34, 36–40, 44).

To comprehensively assess associations between cigarette smoking, NER genes, and HNC risk, we used data from the Carolina Head and Neck Cancer Epidemiology (CHANCE) Study to estimate main and joint effects of cigarette smoking and 84 SNPs across 15 NER genes on HNC risk among a racially diverse population including Whites (922 cases and 1,074 controls) and African Americans (305 cases and 251 controls).

The CHANCE Study is a population-based case–control study of 1,389 cases and 1,396 controls from 46 of 100 counties in North Carolina (50–52). Eligible participants were 20 to 80 years of age (50–52). Cases were identified from the North Carolina Central Cancer Registry between January 1, 2002 and February 28, 2006 using rapid case ascertainment (50–52). Tumors were classified according to ICD-O-3 codes: squamous cell carcinomas of the oral cavity (C02.0-C02.3, C03.0-C03.1, C03.9-C04.1, C04.8-C05.0, C06.0-C06.2, and C06.8-C06.9), oropharynx (C01.9, C02.4, C05.1-C05.2, C09.0-C09.1, C09.8-C10.4, and C10.8–C10.9), hypopharynx (C12.9-C13.2 and C13.8-C13.9), larynx (C32.0-C32.3 and C32.8-C32.9), and oral cavity/pharynx not otherwise specified (C02.8-C02.9, C05.8–05.9, C14.0, C14.2, and C14.8) were included in the study, whereas tumors of the salivary glands, nasopharynx, nasal cavity, and nasal sinuses were excluded (50–53). Controls were randomly sampled from the North Carolina Department of Motor Vehicle records and frequency matched to cases within strata of age, race, and sex (50–52).

For this analysis, we excluded cases and controls who did not provide blood or buccal cell samples, whose samples were insufficient for genotyping, or whose samples did not otherwise meet quality control criteria [115 (8.3%) cases and 53 (3.8%) controls] (52). We further excluded individuals who self-reported race other than White or African American because of sparse data [26 (1.9%) cases and 18 (1.3%) controls] and cases with lip cancers because of etiologic differences [21 (1.5%) cases; ref. 52]. Our final sample included 1,227 HNC cases and 1,325 controls.

Self-reported demographic and behavioral information was ascertained through nurse-administered questionnaires (50–52). Information on cigarette smoking included ever/never, current/former, frequency (cigarettes/day), and duration (years). Information on environmental tobacco smoke (ETS) included ever/never and duration (years) of exposure in the home and at work (50).

Blood (∼90%) or buccal cell (∼10%) samples were collected from cases and controls at the time of interview for DNA extraction (52). An Illumina GoldenGate assay with Sentrix Array Matrix and 96-well standard microtiter platform was used to genotype 1,536 SNPs, including 129 SNPs in 15 NER genes (52, 54). Seventy-one tag SNPs in NER genes were selected on the basis of a case–control study of HNC at MD Anderson Cancer Center (Houston, TX), which queried National Institute of Environmental Health Sciences–Environmental Genome Project (NIEHS-EGP) and HapMap databases using selection criteria of r² ≥ 0.80, a minor allele frequency (MAF) ≥ 0.05, 1 to 2 kb flanking regions, and the Utah Residents with Northern and Western European Ancestry (CEU) population (Supplementary Table S1; Q. Wei; personal communication; refs. 55, 56). Another 58 SNPs in NER genes were selected on the basis of several criteria including association in other cancer studies and/or potential function (Supplementary Table S1). We excluded 14 SNPs for which genotyping resulted in poor signal intensity or genotype clustering (52), as well as SNPs with a MAF less than 0.05 (31 SNPs among Whites and 36 SNPs among African Americans; Supplementary Table S1). Most excluded SNPs had been selected on the basis of previous literature and/or function (Supplementary Table S1). Among the remaining SNPs, genotype frequencies for 7 SNPs in Whites and 7 SNPs in African Americans were inconsistent with Hardy–Weinberg equilibrium (HWE; P < 0.05; Supplementary Table S1); however, because genotyping scatter plots showed reasonable genotype clustering, these SNPs were included in analyses but interpreted with caution (57). Our final analysis included 84 SNPs in 14 NER genes among Whites and 79 SNPs in 15 NER genes among African Americans.

Unconditional logistic regression models were used to estimate ORs with 95% intervals (I) for the main effects of cigarette smoking and ETS on HNC risk. Adjusted cigarette smoking and ETS models included matching factors (age, sex, and race), education, and lifetime consumption of alcohol (categorical milliliters of ethanol). ETS ORs were additionally adjusted for duration of cigarette smoking (continuous, years), as well as stratified by ever/never cigarette smoking. Information on human papillomavirus (HPV) infection is not currently available in CHANCE, and was therefore not considered in analyses. Cigarette smoking and ETS models were considered in the overall study population and stratified by race (White and African American).

For SNPs, race-stratified hierarchical unconditional logistic regression was used to estimate ORs and 95% CIs for the main effects of SNPs on HNC risk (as well as tumor site–specific risk) by including a SNP-gene matrix to account for clustering of SNP data by gene (58, 59). Because the conventional logistic regression approach of modeling one SNP at a time with P values corrected for multiple comparisons using the Bonferroni method is overly conservative as it assumes tests are independent, which is not the case with potentially correlated exposures, we selected a hierarchical approach (58, 59). Results from the conventional approach are provided in Supplementary Tables.

We used a two-stage hierarchical model:

where p_i represents the probability of case status in the sample, X_ij contains indicators of SNPs, and W_i represents important covariates or potential confounders (58, 59).

where β_j represents the coefficients for the effects of the SNPs, Z_j represents the matrix linking SNPs with their associated genes, and δ_j represents independent errors, which are normally distributed with a mean of zero and a variance of τ² (58, 59). To avoid over-parameterization by modeling one large SNP-gene matrix (i.e., including all 84 SNPs across 15 genes) in a single model, 15 models, one for each gene, were used to shrink estimates for SNPs on the same gene toward a common gene effect (i.e., the Z matrix was a single column representing a single gene, with rows of 1′s for each SNP). Because SNPs on the same gene were included in the same model, we excluded some SNPs due to extreme colinearity (estimated correlation ρ > 0.98; 11 SNPs in Whites and 5 SNPs in African Americans). A semi-Bayes approach was used to set τ² to 0.05, as this corresponded with a plausible range of expected ORs for the association between SNPs and HNC based on previous literature (i.e., 0.6 to 1.6; ref. 58). Sensitivity analyses with τ² = 0.01, τ² = 0.10, and τ² = 1.0 evaluated robustness of this choice.

SNPs were defined using a dominant genetic model given the large portion of SNPs with few cases and controls homozygous for the variant allele (∼7% among Whites and ∼33% among African Americans). The referent allele for both Whites and African Americans was assigned to be the major allele based on controls from the overall study population (which was concurrent with the race-specific major allele for 98% of SNPs in Whites and 92% of SNPs in African Americans). Because genetic exposures were based on germline DNA, which would not reflect the influences of smoking, drinking, or HPV infection, SNP models were only adjusted for matching factors (sex and age) and ancestry (continuous proportion African ancestry), as informed by our directed acyclic graph (DAG) analysis (60). On the basis of previous studies of cancer among Whites and African Americans in North Carolina, 145 ancestral informative markers (AIMS) were selected on the basis of differences in allele frequencies between European and African HapMap populations and used to estimate the proportion of African ancestry in each participant based on Fisher's information criterion (FIC; refs. 52, 61–63).

ORs and 95% CIs for the joint effects of cigarette smoking and SNPs in NER genes were estimated using conventional and hierarchical logistic regression. Joint effects were modeled using disjoint indicator variables for (i) individuals who smoked and had the referent genotype, (ii) individuals who did not smoke and had the variant genotype, and (iii) individuals who smoked and had the variant genotype (58). As described by Hung and colleagues, hierarchical models included a 3 × 2 gene–environment matrix to account for clustering of disjoint indicator variables by SNP and cigarette effects (i.e., the Z matrix had 2 columns, one representing SNP effects and one representing smoking effects, and 3 rows, each representing the disjoint indicator variables, with 1′s and 0′s entered according to concordance of rows and columns; ref. 58). A τ² of 0.35 was used to correspond to expected ORs between 0.3 and 3.0 for each indicator variable (58). Sensitivity analyses with τ² = 0.05 evaluated robustness of this choice. Joint effects models were stratified by self-reported race. Only joint effect estimates among Whites are presented because small cell counts among African Americans prohibited reliable estimation for most SNP–cigarette effects. Joint effects models were adjusted for matching factors (sex and age), education, alcohol drinking, and ancestry as both behavioral and genetic exposures were being modeled. Interactions between SNPs and cigarette smoking were assessed on the additive scale using the relative excess risk due to interaction (RERI), with 95% CIs calculated using the Hosmer and Lemeshow method (64). All statistical analyses were conducted using SAS 9.3 (Cary, NC) (65).

The study population included 922 cases and 1,074 controls who self-reported race as White and 305 cases and 251 controls who self-reported African American (Table 1). The majority of cases (76.4%) and controls (69.7%) were male. Approximately, one-third of cases (33.6%) and controls (30.2%) were between the ages of 55 and 65 years. Controls were more highly educated than cases with 60.7% of controls attending college compared with 38.6% of cases.

The adjusted OR for ever compared with never cigarette smoking was elevated in the overall (2.28; 95% CI, 1.81–2.88; Table 2) and race-stratified study populations (1.97; 95% CI, 1.54–2.53 among Whites and 7.75; 95% CI, 3.57–16.83 among African Americans). Furthermore, the risk of HNC increased with increasing frequency and duration of cigarette smoking (P_trend < 0.0001). In contrast, we did not observe strong associations between ETS and HNC (Supplementary Table S2). Adjusted ORs for ever compared with never ETS exposure were not elevated when stratified by race (0.87; 95% CI, 0.63–1.19 among Whites and 0.91; 95% CI, 0.45–1.82 among African Americans) or by active cigarette smoking (0.84; 95% CI, 0.54–1.33 among never cigarette smokers and 0.92; 95% CI, 0.62–1.37 among ever cigarette smokers). Duration of ETS exposure at work or home was also not associated with HNC risk (Supplementary Table S2).

Among Whites, most ORs were close to the null value for associations between SNPs and HNC (Table 3). The SNP rs4150403 on the excision repair cross-complementing 3 (ERCC3) gene, also known as xeroderma pigmentosum B (XPB), however, was statistically significantly associated with elevated HNC risk (AA+AG vs. GG; OR, 1.28; 95% CI, 1.01–1.61). In addition, another SNP on ERCC3 (XPB), rs4150496, suggested a possible reduced HNC risk among Whites (AA+AG vs. GG; OR, 0.80; 95% CI, 0.62–1.02). When we considered associations between these SNPs and each tumor site separately, associations between rs4150403 and oral cavity cancer resulted in the largest magnitude OR (1.32; 95% CI, 1.01–1.71; Supplementary Table S3). For rs4150496, associations with oral cavity and oropharyngeal cancers resulted in the smallest magnitude ORs (OR, 0.79; 95% CI, 0.60–1.04 and OR, 0.77; 95% CI, 0.56–1.06, respectively).

Among African Americans, one SNP on ERCC6 (also known as Cockayne Syndrome B, CSB), rs4253132, was significantly associated with reduced HNC risk (CC+CT vs. TT; OR, 0.62; 95% CI, 0.45–0.86; Table 4). Because of low cell counts, we were unable to assess the association between this SNP and all tumor sites among African Americans. We did find, however, that rs4253132 was significantly associated with reduced risk of laryngeal cancer (OR, 0.65; 95% CI, 0.44–0.97; Supplementary Table S4).

No other significant SNP–HNC associations were detected, including none of the extensively studied associations between SNPs in ERCC2 (also known as XPD), ERCC1, or ligase 1 (LIG1) and HNC risk. In particular, we did not find an association between rs13181 in ERCC2 (XPD) and HNC among Whites (GG+TG vs. TT; OR, 1.05; 95% CI, 0.76–1.45; Table 3) nor among African Americans (OR, 1.01; 95% CI, 0.75–1.37; Table 4). In sensitivity analyses, results from Tables 3 and 4 were robust to further adjustment for cigarette smoking and alcohol drinking and variation of τ² (i.e., results were similar when adjusting for cigarette smoking and alcohol drinking or when τ² = 0.01, 0.10, and 1.0 rather than 0.05, though the OR for rs4150403 among Whites was nonsignificantly elevated when adjusting for cigarette smoking and alcohol drinking or when τ² = 0.01; data not shown). Compared with the hierarchical model, ORs (95% CIs) for the conventional model were similar though less stable, with a few additional SNP–HNC associations implicated at 0.05 α level but none at a Bonferroni-corrected significance level of 0.0006 (Supplementary Tables S5 and S6).

Using the conventional method (Table 5), interactions between ever cigarette smoking and 3 SNPs suggested possible departures from the null on the additive scale at an uncorrected 0.05 α level among Whites. Specifically, the interaction between cigarette smoking and rs4253132 on ERCC6 (CSB, RERI = 0.70; 95% CI, 0.14–1.26) and rs2291120 on DDB2 (XPE, RERI = 0.68; 95% CI, 0.11–1.26) seemed to be more than additive, whereas the interaction between cigarette smoking and rs744154 on ERCC4 (XPF, RERI = −1.02; 95% CI, −2.02 to −0.02) seemed to be less than additive. However, RERI estimates were generally imprecise and none were significant at a Bonferroni-corrected significance level (Table 5). Furthermore, genotype frequencies for rs4253132 on ERCC6 among Whites seemed inconsistent with HWE at a 0.05 α level, although the genotype-clustering plot seemed reasonable, and should therefore be cautiously interpreted. ORs (95% CIs) for joint effects from the hierarchical model (Table 6) were similar to estimates from the conventional method. RERI point estimates were also similar between the two methods, but we were unable to estimate 95% CIs for RERI estimates using hierarchical regression. Joint effects of SNPs and former/current cigarette smoking as well as SNPs and ETS among Whites are provided in Supplementary Tables S7 and S8, respectively, and highlight a few other potential gene–environment interactions. Among African Americans, no significant ever cigarette–SNP interactions were noted; however, estimates were unreliable due to relatively low cell counts and are therefore not presented.