Common 5p15.33 and 6p21.33 variants influence lung cancer risk

Support for inherited genetic susceptibility to lung cancer has recently come from genome-wide association studies that have demonstrated that 15q25.1 variation influences lung cancer risk^1–3.

To identify risk variants for lung cancer, we carried out a GWA study. Using Illumina HumanHap550 BeadChips, we genotyped 561,466 SNPs in 1,978 cases (Supplementary Methods online). After application of quality control criteria, genotypes were available for 1,952 cases. We were able to satisfactorily genotype 552,947 SNPs (98.5%) with mean sample call rate 99.7%. For controls, we used publicly accessible HumanHap550 genotype data in 1,438 individuals from the 1958 Birth Cohort⁴ (Supplementary Methods). Genotypes were available for 541,327 SNPs (97.5% of 555,352 SNPs typed) and 524,714 SNPs were common to cases and controls. Applying quality control filters, we excluded 8,534 SNPs monomorphic in either cases or controls; 2,744 with call rates < 95%; 770 showing departure from Hardy-Weinberg equilibrium (HWE; P < 10⁻⁵ in cases or controls) and 747 with minor allele frequency (MAF) <1% in cases or controls; leaving 511,919 informative SNPs for analysis.

Comparison of observed and expected distributions showed little evidence for inflation of allele test statistics (inflation factor λ = 1.03; Supplementary Fig. 1 online), excluding the possibility of significant hidden population substructure, cryptic relatedness or differential genotype calling. The distribution of association P values was significantly skewed from the null distribution with 116 SNPs having P value ≤10⁻⁴, greater than the 51 expected (P ~ 10⁻⁴). In keeping with previous studies, 15q25.1 SNPs were most strongly associated; excluding these SNPs, 98 were associated at P ≤ 10⁻⁴.

To replicate associations, we are now genotyping the most strongly associated SNPs in an additional case-control series. In the interim, we have sought to identify new associations by conducting a meta-analysis pooling our UK-GWA study with data from two other studies: the IARC-GWA study of 1,989 cases and 2,625 controls², summary data from which is publicly available; and the Texas-GWA study of 1,154 non–small-cell lung cancer cases and 1,137 controls¹. In both studies genotyping was done using Illumina HumanHap300 Bead-Chips. Pooling was based on the 223,891 autosomal SNPs genotyped in all three GWA studies that had MAFs >1% and no departure from HWE (P < 10⁻⁵ in cases or controls).

We derived meta-analysis odds ratios (ORs) and confidence intervals (CIs), and associated P values (Supplementary Methods). As expected, the strongest associations were obtained for SNPs mapping to 15q25.1. After exclusion of the 36 SNPs mapping to this locus (76.4–76.8 Mb), there remained evidence of enrichment of associated variants: 77 had P values ≤10⁻⁴, compared with 22 expected (Supplementary Table 1 online). Genomic control values for the Texas-GWA and IARC–GWA studies were 0.99 and 1.03, respectively (Supplementary Fig. 1). For the combined dataset λ was 1.04 and 0.92 under fixed and random effects models, respectively (Supplementary Fig. 1), providing little evidence of confounding from population substructure as a source of bias in our meta-analysis.

Aside from 15q25, the strongest evidence for a lung cancer risk loci was found at 6p21.33 (Supplementary Fig. 2 online). Associations were significant for rs3117582 and rs3131379 after adjustment for multiple testing (OR = 1.30, 95% CI = 1.19–1.42; P = 5.71 × 10⁻⁹; OR = 1.26, 95% CI = 1.16–1.38; P = 1.91 × 10⁻⁷, respectively; Supplementary Table 1), assuming a Bonferroni correction. An additional 12 SNPs mapping to this region also had an association with risk (P ≤ 10⁻⁵).

We observed strong support for an association between rs3117582 and rs3131379 and risk in the UK-GWA and IARC-GWA studies, with P values in each of borderline genome-wide significance (P = 6.24 × 10⁻⁶ and 4.41 × 10⁻⁶, respectively). Support was limited in the Texas-GWA study, reflected in the random-effects model (P = 6.63 × 10⁻³, P_het = 0.02 and P = 0.013, P_het = 0.03 for rs3117582 and rs3131379, respectively). As all three GWA studies were based on subjects with European ancestry, ancestry-related differences are unlikely to underlie between-study heterogeneity, and the nonsignificant association in the Texas-GWA study may reflect study power, especially as these SNPs have low MAFs. To further validate the association between 6p and risk, we genotyped rs3117582 in an additional 2,484 cases and 3,036 controls (UK-Replication series; Supplementary Methods). Genotypes were obtained for 2,448 (98.6%) cases and 2,983 (98.3%) controls. As previously, the C allele was associated with a significantly increased risk (OR = 1.16, 95% CI = 1.04–1.29; P = 7.30 × 10⁻³). Pooling data from all series provided unequivocal evidence for a relationship between rs3117582 and lung cancer risk (P = 4.97 × 10⁻¹⁰, P_het = 0.02, I² = 71%; Fig. 1). ORs associated with AC and CC genotypes were 1.20 (95% CI = 1.11–1.29, P = 7.12 × 10⁻⁶; P_het = 0.06, I² = 60%) and 1.80 (95% CI = 1.41–2.30, P = 2.25 × 10⁻⁶; P_het = 0.29, I² = 19%).

rs3117582 (31,728,499 bp) localizes to intron 1 of BAT3 and rs3131379 (31,829,012 bp) localizes to intron 10 of MSH5 at 6p21.33 (Fig. 2). Genotypes are highly correlated (r² = 0.99), hence on the basis of flanking recombination hot spots they define a single locus at 31,676,001–32,303,001 bp. The association could be mediated through LD with a number of transcripts; however, BAT3 and MSH5 both represent strong candidates for lung cancer susceptibility. BAT3 is implicated in apoptosis and the protein complexes with E1A-binding protein p300, required for acetylation of p53 in response to DNA damage⁵. MSH5 is involved in DNA mismatch repair (MMR) and meiotic recombination, and deficiency of MMR has been documented to have a role in lung cancer^6–8.

There is some evidence for a risk locus at 6p22.1 (rs9295740; 27,797,481 bp; OR = 1.20, 95% CI = 1.12–1.29; P = 3.63 × 10⁻⁷; random effects P = 3.43 × 10⁻⁷; Supplementary Table 1 and Fig. 2). LD across 6p21.33–6p22.1 is extensive, and although recombination rates across the region are compatible with an independent susceptibility locus, the moderate LD between rs9295740 and both rs3117582 and rs3131379 (r² values of 0.38 and 0.39, respectively) suggests that the associations may be mediated through correlation with the same causal variant.

The most consistent evidence for a new disease locus outside 6p was attained at 5p15.33 (rs401681; OR = 0.88, 95% CI = 0.83–0.93; P = 4.40 × 10⁻⁶; P_het = 0.94, I² = 0%; Supplementary Table 1). rs401681 localizes to intron 13 of CLPTM1L⁹ within a 60-kb region of LD (1,353,580–1,412,838 bp; Fig. 2 and Supplementary Fig. 3 online) frequently amplified in early-stage NSCLC¹⁰. Genotyping rs401681 in the UK-Replication series provided further validation of the association. We obtained genotypes for 2,396 (99.7%) cases and 3,001 (98.8%) controls. The A allele was associated with a significantly decreased risk (OR = 0.92, 95% CI = 0.88–0.97; P = 4.95 × 10⁻⁴). Pooling data from all series provided unequivocal evidence for a relationship between rs401681 and risk (OR = 0.87, 95% CI = 0.84–0.92; P = 7.90 × 10⁻⁹; P_het = 0.99, I² = 0%; Fig. 1). ORs associated with GA and AA genotypes were 0.86 (95% CI = 0.80–0.92; P = 2.12 × 10⁻⁵, P_het = 0.53, I² = 0%) and 0.77 (95% CI = 0.70–0.84; P = 3.54 × 10⁻⁸, P_het = 0.99, I² = 0%), respectively.

We examined for clinicopathological relationships with rs3117582 and rs401681 in the UK and Texas datasets. The only significant association was between rs3117582 and family history of lung cancer (P = 0.03), nonsignificant after adjustment for multiple comparisons(Supplementary Table 2 online). These data suggest that, despite differences in the biology of NSCLC and SCLC, the causal variants affect the risk of all forms of lung cancer, compatible with epidemiological data showing that familial lung cancer risks are not subtype dependent and that intrafamilial histological concordance is poor¹¹.

The power of our analysis to identify the 15q25.1 and 6p loci at P = 2.0×10⁻⁷ was high (>80%). In contrast, power to detect alleles with smaller effects and MAFs (for example, those of rs401681) was low. By implication, variants with similar profiles may constitute a larger class of susceptibility loci, whether because of smaller effects or submaximal LD with tagging SNPs.

The 15q25.1 and 6p variants are unlikely to account for >1% of the familial risk, hence a large number of low-risk variants remain to be identified. Further efforts to expand the scale of GWA meta-analyses, in terms of both sample size and SNP coverage, and to increase the number of SNPs taken forward to large-scale replication should identify additional risk variants.