A Rigorous and Comprehensive Validation: Common Genetic Variations and Lung Cancer

Recently, eight genome-wide association single nucleotide polymorphism (SNP) studies (GWAS) have reported association between chromosome loci and lung cancer susceptibility (1-8). Of the five top candidate SNPs, some reside in known genes (rs402710, rs2736100, and rs16969968) and some are de novo (rs8034191 and rs4324798), have been validated (9, 10), or are under validation. rs402710 and rs2736100 are located in chromosome 5p15.33, containing two known genes: the human telomerase reverse transcriptase (TERT) gene and the cleft lip and palate transmembrane 1–like (CLPTM1L, alias CRR9) gene. rs2736100 is located in intron 1 of the TERT gene, and rs402710 is in a region of high linkage disequilibrium (LD) that includes the promoter regions of TERT and the entire coding region of the CLPTM1L gene (intron 16). The overall estimated allelic odds ratios (ORs) for rs402710 and rs2736100 are 1.18 and 1.14, respectively (4). rs8034191 and rs16969968 are located in chromosome 15q25, containing six known genes, three of which encode nicotinic acetylcholine receptor subunits (CHRNA5, cholinergic receptor nicotinic α 5; CHRNA3, cholinergic receptor nicotinic α 3; and CHRNB4, cholinergic receptor nicotinic β 4). The remaining three are IREB2 (iron-responsive element-binding protein 2), PSMA4 (implicated in DNA repair), and LOC123688 (unknown function). In rs16969968, a nonsynonymous variant in CHRNA5 and rs8034191, an unknown locus showed association with lung cancer susceptibility [allelic ORs at 1.30 (1.23-1.38) and 1.32 (1.21-1.45), respectively; ref. 2]. rs4324798 is located in chromosome 6p21.33 within an extended region of high LD near the MHC containing >20 genes (2); genotyping of rs4324798 in five validation studies provided evidence of association with lung cancer risk, at an OR of 1.28 (1.16-1.40; ref. 2).

However, these GWAS were primarily conducted in cigarette smokers except for one study, which did not support the observed association in never smokers (6). We report results of validating the five SNPs in relation to lung cancer susceptibility when they were separately evaluated in smokers and never smokers using unrelated and full-sibling controls. Importantly, history of chronic obstructive pulmonary disease (COPD) was carefully considered as a confounding factor because the well-established shared etiology with lung cancer from tobacco smoking and genetic susceptibility to both diseases (11-13). Studies since 1980 have shown COPD to be an independent risk factor for lung cancer (14). Moreover, we also evaluated their prognostic value for lung cancer survival.

Lung cancer patients were identified and enrolled at Mayo Clinic between 1997 and 2006. The research protocol and consent form were approved by the Mayo Clinic Institutional Review Board; detailed study design and procedure were reported previously (11). Unrelated controls were selected from community residents who were identified by having had a general medical examination and a leftover blood sample from routine clinical tests (11). All full siblings, who were free of cancer and who donated a blood sample, were recruited as controls through lung cancer cases (11).

Demographic and other risk information was obtained from all subjects via a combination of a structured interview, self-administered questionnaire, and medical records (11, 15). Never smokers were defined as having smoked fewer than 100 cigarettes during their lifetimes, and second-hand smoking history was collected as previously reported (16). Cigar or pipe smokers were excluded. Ever smoking included current and/or previous use. History of COPD was determined based on explicit diagnosis recorded in the medical history. Family history of lung cancer in first-degree relatives (parents, siblings, and children) included age at diagnosis and vital status.

Five candidate SNPs were selected from eight recent GWAS: (1-8) rs402710 (G->A), rs2736100 (C->A), rs4324798 (G->A), rs16969968 (G->A), and rs8034191 (T->C). The LD structure of each SNP constructed by Haploview (17) illustrates the known genes or candidate locus regions (Supplementary Fig. S1A-C). Genotyping, performed in the Mayo Clinic Genomic Shared Resource, used TaqMan (Applied Biosystems) according to the manufacturer's instructions. Primers and probes were Assay-by Design (Applied Biosystems). Quality control procedures of genotyping tests are provided in Supplementary Material and Table S1.

Our strategy to rigorously and comprehensively evaluate the role of top SNPs was accomplished by testing the specified hypothesis in the targeted subgroup while best controlling for the strong confounding effect of cigarette smoking history in risk assessment (case-control study) and treatment in survival outcome (patient follow-up study). Other potential confounders included age, sex, COPD history, lung cancer stage and histology, and progression or recurrence. Cases and controls are described in Supplementary Table S2 where eight data sets were defined and respective hypotheses specified. The three main categories are total cases (1,735) and controls (2,242), cigarette smokers (1,406 cases; 1,053 controls), and never smokers (329 cases; 757 controls); under each main category, two control groups were used, unrelated community residents and unaffected full-siblings of cases (11). The sixth group had a limited sample size and was not analyzed further. Survival analysis included 1,742 consecutive patients who were diagnosed between 1997 and 2006. The following contrasting groups were analyzed separately: 1,418 cigarette smokers and 324 never smokers. Among smokers, non–small cell lung carcinoma and SCLC were further separated according to surgical resection: 849 surgically resected NSCLC, 334 NSCLC without surgery, and 235 SCLC (all without surgery). Adjustments for covariates are specified in the footnotes of the result tables.

We tested for association between each SNP and lung cancer status using unconditional logistic regression for cases and unrelated controls, and conditional logistic regression for cases and sibling controls (18). We also tested the association of each SNP with survival time, defined as the time from lung cancer diagnosis to death or last follow-up, using Cox proportional hazards regression analysis (19). Significant covariates were selected and identified through forward and backward variable selection procedures. The level of P = 0.05 was chosen as our threshold for statistical significance. Multiple comparison correction was not applied because our goal was to validate each SNP independently. All analyses were performed using SAS software (SAS Institute, Inc. SAS/STAT User's Guide, v9.)

Seven predefined case-control sets are provided in Supplementary Table S2, and basic descriptions of age, sex, cigarette pack-year smoking, and prior medical history of COPD are in Supplementary Table S3. As an initial step to replicate published results in a comparable design, each of the five SNPs were assessed in our total cases and unrelated controls (Table 1, Set 1). Four of the five SNPs were significantly associated with lung cancer risk in univariate models; however, after accounting for previously adjusted risk factors, only rs402710 remained significant; and the significant association holds after further adjusting for COPD, suggesting individuals with the minor allele have a 17% reduced risk. Although all subjects were self-reported Caucasians, an alternative design using full sibling controls was applied to avoid subpopulation stratification; the results do not support an association of any tested SNPs and lung cancer risk (Table 1, Set 2).

Next, the five candidate SNPs and lung cancer risk in smokers and never smokers were assessed separately. Only two SNPs on chromosome 5 showed some significant results (Table 2): rs402710 was significant in all smokers, but was not significant when tested in heavy smokers only; whereas, rs2736100 was only significant in never smokers. When compared with siblings, none of the five SNPs remained significant. Specific to rs402710, the estimated OR was 0.78 (P = 0.002) when cases were compared with unrelated controls among all smokers, attenuated to 0.85 when restricted to heavy smokers (P = 0.117), and diminished to 0.91 (P = 0.211) when compared with siblings. A similar pattern was observed with rs2736100.

The prognostic role of each SNP was also tested for lung cancer overall survival, as shown in Table 3; a more detailed description of patients' characteristics, which are included in the multivariable Cox models, is provided in Supplementary Table S4. rs432478 is the only SNP that showed a significant effect in SCLC patients, with a minor allele associated with longer survival.

We rigorously evaluated the five top candidate SNPs that have been revealed to alter lung cancer susceptibility from multiple GWAS and a few validation studies. Our initial validation results, using a similar design as in the published studies, confirmed results for two SNPs on chromosome 5 (rs402710 and rs2736100). However, more rigorous evaluation by controlling for the effect of pre-existing COPD attenuated the association; then, using sibling controls diminished the association. Specifically noted is that for rs2736100, the estimated effect in never smokers was significant (OR, 1.23) but attenuated to 1.14 for related controls, no longer significant. These findings could be due to the small sample size; however, three alternative explanations are postulated: First, any genetic effects were dampened in the presence of heavy exposure to environmental carcinogens. Second, findings from previous studies were confounded by variable degrees of tobacco smoke exposure, likely a residual effect even after adjusting for cigarette smoking history (20). Indeed, in one of the previous validation studies, when never smokers were analyzed independently, rather than in the midst of smokers, no association with the top SNPs remained significant (6). As nicotine dependence phenotype confounds carcinogen exposure, the authors of one GWAS (1, 6) had interpreted their finding of chromosome 15q24/25.1 with no consensus as to the relative impact of the variants on the propensity to smoke versus a direct carcinogenic effect. Third, the association between these SNPs and lung cancer may, in part, be confounded by COPD (21); future studies should carefully adjust for COPD and evaluate the dual effects of the at-risk SNPs in both COPD and lung cancer etiology.

Finally, we have revealed one SNP, rs4324798, as an independent prognostic factor for overall survival in SCLC patients, calling for further validation by other studies. The prognostic value of this SNP and its context gene and region in treatment response and toxicities needs to be evaluated.

In conclusion, three unique strengths of our study are the dual-control design, multiple independent subsets, and the consideration of medical history of COPD. Although none of the five candidate SNPs were found to be significant, we did obtain comparable results when we chose a more liberal design that mimicked previously published studies, specifically, without adjusting for COPD and only using unrelated controls. Results are subject to limited sample size, calling for effective multicenter collaborations.

No potential conflicts of interest were disclosed.

We thank Susan Ernst, M.A., for her technical assistance with the manuscript.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.