After excluding samples with call rates <95%, 2549 African-American subjects remained for analysis (1308 cases, 1241 controls). Compared to controls, cases were more likely to be male, to be older, to smoke, and to have smoked a greater number of pack-years and cigarettes per day (Table S2
). Cases and controls had similar levels of sub-Saharan African ancestry and European ancestry (Table S2
, Figure S1
). After excluding monomorphic SNPs and SNPs that did not meet call-rate or HWE thresholds in at least two of the three study sites, 336 SNPs remained for analysis.
Seven SNPs were significantly associated with lung cancer risk at a Bonferroni-corrected level (P < 1.49 × 10−4) (Table ). One of these is a novel association detected near CHRNA1 on Chromosome 2q31.1. The other six SNPs are located in the 15q25.1 region previously associated with lung cancer. None of the SNP associations showed statistically significant heterogeneity across study site, as measured by Cochran's Q or I2.
Significant SNP associations from the case-control analysis of lung cancer in African-Americans
Located <3kb upstream of CHRNA1, rs3755486 showed a strong association with lung cancer risk (OR = 1.40, 95% CI = 1.18-1.67, P = 1.0 × 10−4). CHRNA1, a neuromuscular nAChR subunit gene, has not been previously associated with lung cancer risk or with smoking behaviors. Adjustment for number of pack-years smoked did not attenuate the rs3755486 association (OR = 1.42, 95% CI = 1.19-1.70, P = 8.7 × 10−5). Although rs3755486 did not show a statistically significant association with lung cancer risk among never-smokers (OR = 1.33, 95% CI = 0.84-2.12, P = 0.22), this subgroup includes only 126 cases and 388 controls and statistical power is severely diminished.
Targeted imputation of an ~150 kb region on 2q31.1 containing the CHRNA1
gene permitted calculation of association statistics for an additional 305 SNPs, three of which had smaller p-values than the most significant genotyped SNP (rs4972458, rs2646151 and rs935864) (Figure and Table S3
). While the most significantly associated genotyped SNP surpassed the Bonferroni threshold, the additional associations identified through imputation further indicate that the association peak is located near the CHRNA1
gene promoter. The most significant of these imputed SNPs was rs4972458 (OR = 1.33, 95% CI = 1.16-1.52, P = 3.8 × 10−5
Association of 16 genotyped and 305 imputed SNPs on chromosome 2q31.1 with lung cancer in African-Americans
Because SNP imputation in admixed populations, such as African-Americans, tends to be less accurate than in other populations, imputation accuracy was assessed with TaqMan genotyping. Imputed genotype at the rs4972458 risk SNP was strongly correlated with TaqMan genotype in a subset of 183 cases and 221 controls from UCSF undergoing genotyping at this locus (κweighted = 0.78, 95% C.I. = 0.72- 0.83; r = 0.83; 55/404 discordant genotypes). Odds ratios and p-values calculated from imputed genotypes were comparable to those calculated from TaqMan genotypes (OR = 1.53, P = 0.011 and OR = 1.45, P = 0.024, respectively). Further, the imputed minor allele frequency was similar to the minor allele frequency derived from TaqMan genotypes among cases (0.30 vs. 0.29, respectively) and among controls (0.21 vs. 0.22, respectively). Although we observed acceptable genotype concordance at the most significantly associated imputed SNP, it remains to be determined if the imputation was similarly accurate at neighboring markers.
Of the six significantly associated SNPs in the 15q25.1 region, the strongest association was observed at rs17486278 in intron 1 of CHRNA5 (OR = 1.37, 95% CI = 1.21-1.55, P = 6.4 × 10−7). Of the significantly associated 15q25.1 SNPs listed in Table , only rs16969968 remained significantly associated with lung cancer when the model was conditioned on rs17486278 (OR = 1.36, P = 0.02). Likewise, only rs17486278 remained significantly associated when the model was conditioned on rs16969968 (OR = 1.22, P = 0.0050), indicating that multiple independent lung cancer risk loci may exist in the 15q25.1 region. Rs16969968 is a missense variant located in the fifth exon of CHRNA5 (OR = 1.60, 95% CI = 1.27-2.01, P = 5.9 × 10−5). This SNP is an A>G transition, changing an aspartic acid residue to an asparagine residue. No 15q25.1 SNPs in CHRNB4 were significantly associated with lung cancer, but a single SNP in the first intron of CHRNA3 did reach statistical significance (rs4243084, OR = 1.33, 95% CI = 1.15-1.53, P = 1.1 × 10−4). Adjustment for smoking modestly attenuated SNP associations on 15q25.1 (Figure ). Furthermore, when analysis was restricted to never-smokers, rs2036527 remained associated with lung cancer risk (OR = 1.58, 95% CI = 1.12-2.26, P = 9.9 × 10−3).
Associations of number of cigarettes smoked per day (CPD) with top ranked SNPs from the case-control analysis of lung cancer in African-Americans, stratified by lung cancer status
Association of SNPs on chromosome 15q25.1 with lung cancer, with and without adjustment for pack-years smoked
To explore whether the nAChR SNPs may influence cancer risk through modification of smoking behavior, associations between SNPs and the number of cigarettes smoked per day (CPD) were calculated separately among individuals with lung cancer and among controls. Five of the seven SNPs significantly associated with lung cancer in our study were also associated with CPD among the control subjects, with cancer risk alleles associated with heavier smoking. However, none of these SNPs were associated with CPD among lung cancer cases (Table ). In controls, the strongest association with CPD was observed at rs17486278. This is also the SNP most strongly associated with lung cancer risk in the case-control analysis. For each additional copy of the lung cancer risk allele, an increase of 2.54 CPD was observed in the controls (95% CI = 1.15-3.94, P = 3.6 × 10−4).
To further explore the functional role of SNPs significantly associated with lung cancer risk, the relationship between genotype and gene expression level was assessed in three cell types. Functional data were not available for the 15q25.1 risk SNPs, but were available for rs3755486 on 2q31.1. CHRNA1
mRNA expression, as measured by probe ILMN_1798700, was significantly associated with rs3755486 genotype in lymphoblastoid cell lines (ρ = −0.403, P = 3.0 × 10−4
). The lung cancer risk allele was associated with increased levels of CHRNA1
expression (Figure S2
In an independent sample of 1154 European-ancestry cases with non-small cell lung cancer and 1137 European-ancestry controls collected at M.D. Anderson [6
], rs3755486 was not associated with risk of lung cancer (OR = 1.01, 95% CI = 0.89-1.14, P = 0.896), nor was rs2244340. Associations were also null when the sample was restricted to cases with adenocarcinoma and to cases with squamous cell carcinoma.