Genome-wide association study of familial lung cancer

From the GWAS meta-analysis of familial lung cancer risk, novel and imputed rare variants on 4p15.31 and 9p21.3 were identified for squamous cell carcinoma. Several other variants were also confirmed as associated with familial lung cancer including many on chromosome 15q25.1 that reached genome-wide levels of significance (P < 5 × 10^–8).

Abstract

To identify genetic variation associated with lung cancer risk, we performed a genome-wide association analysis of 685 lung cancer cases that had a family history of two or more first or second degree relatives compared with 744 controls without lung cancer that were genotyped on an Illumina Human OmniExpressExome-8v1 array. To ensure robust results, we further evaluated these findings using data from six additional studies that were assembled through the Transdisciplinary Research on Cancer of the Lung Consortium comprising 1993 familial cases and 33 690 controls. We performed a meta-analysis after imputation of all variants using the 1000 Genomes Project Phase 1 (version 3 release date September 2013). Analyses were conducted for 9 327 222 SNPs integrating data from the two sources. A novel variant on chromosome 4p15.31 near the LCORL gene and an imputed rare variant intergenic between CDKN2A and IFNA8 on chromosome 9p21.3 were identified at a genome-wide level of significance for squamous cell carcinomas. Additionally, associations of CHRNA3 and CHRNA5 on chromosome 15q25.1 in sporadic lung cancer were confirmed at a genome-wide level of significance in familial lung cancer. Previously identified variants in or near CHRNA2, BRCA2, CYP2A6 for overall lung cancer, TERT, SECISPB2L and RTEL1 for adenocarcinoma and RAD52 and MHC for squamous carcinoma were significantly associated with lung cancer.

Introduction

Lung cancer is estimated to account for over 13% of all new cancer diagnoses and almost 27% of all cancer deaths (http://SEER.cancer.gov/statfacts/html.lungb.html). Despite diagnostic advances in recent years, lung cancer usually presents at advanced stages and mortality remains high. While most lung cancer is attributable to tobacco smoke exposure, many studies have identified genetic variation that influences lung cancer risk (1–12). Because cancers arise through an accumulation of deleterious somatic mutations, environmental or inherited risk factors due to inherited deleterious germline genetic variants are critical (12).

The Genetic Epidemiology of Lung Cancer Consortium (GELCC) has conducted the only family-based lung cancer linkage study to identify rare, highly penetrant lung cancer susceptibility genes, providing evidence of linkage on chromosomal regions 6q23-25, 18p11.23, 2p22.2, 14q13.1, 16p13 and 20q13.11 (4,5,10–13). Linkage analysis is designed to identify highly penetrant uncommon variants but does not identify more common variants. Genome-wide association studies (GWAS) have provided highly significant and reproducible results associating common genetic variations to lung cancer. For example, a locus on chromosome 15q25, spanning a neuronal nicotinic acetylcholine receptor gene cluster comprising CHRNA3, CHRNA5 and CHRNB4 subunits, is significantly associated with lung cancer risk (5,7,14). Regions on chromosomes 6p21 (BAG6, MSH5), 5p15 (TERT, CLPTM1L), 13q (BRCA2) and 22q (CHEK2) have also been identified from GWAS as being consistently associated with histology-specific lung cancer risk (5,6,11,15).

To further identify genetic variation associated with familial lung cancer risk, we conducted a GWAS meta-analysis focused on familial lung cancer cases. Familial lung cancer cases used in the analysis came from the GELCC family-based study and from the Transdisciplinary Research in Cancer of the Lung (16) and the International Lung Cancer Consortium (TRICL-ILCCO) (11). Lung cancer genome-wide association studies enriched for familial lung cancer cases represent a novel application of GWAS methodology to a population most likely to have a genetic component to their risk. Selecting cases with a family history improves the power to detect uncommon variants that are likely to have a larger inherited impact on lung cancer risk (12,17).

Materials and methods

GELCC study subjects

The GELCC enrolled lung cancer cases with a family history of lung cancer from the following data collection sites: Karmanos Cancer Institute/Wayne State University, University of Cincinnati College of Medicine, Mayo Clinic, Medical College of Toledo, and Louisiana State University Health Sciences Center—New Orleans. Local Institutional Review Boards approved all protocols and each participant provided written informed consent. Familial lung cancer cases (ICD-9 162.0–162.9) were defined as individuals with a histologically confirmed diagnosis of lung cancer and at least one first or second degree relative with lung cancer. Family history information was collected by interview. The diagnosis of lung cancer in both the index case and relatives was verified with medical records, tumor registries and death certificates when possible. For the GWAS, one lung cancer case from each of 685 families of European descent with at least two first or second-degree relatives with lung cancer was selected. 744 controls of European descent were selected from persons related by marriage to lung cancer cases or from the general population in which the cases were identified, and did not have a lung cancer diagnosis and had no first or second-degree relatives with lung cancer (Supplementary Table 1, available at Carcinogenesis Online). Controls were frequency matched to cases on age (5-year groups), sex and smoking status as closely as possible. The detailed demographic information on GELCC study is presented in Supplemented Table 2, available at Carcinogenesis Online.

Participation included completing a questionnaire (either by phone, in the clinic, or by mail), providing a blood or buccal/saliva sample for DNA, and providing access to medical records for confirmation of diagnosis and histology type. The study questionnaire included basic demographics, health history including history of other lung diseases, smoking history, and family history of cancer. Clinical data about diagnostic procedures, stage of disease, and histology type were obtained from medical records.

GELCC study genotyping

DNA samples were obtained from several sources including blood (68.2%), mouthwash (0.3%), saliva (2.3%) and whole genome amplification from blood (29.2%). Germline DNA was genotyped at the Center for Inherited Disease Research (CIDR) using the Illumina Human OmniExpressExome-8v1 array, designed to human genotype build 37. The GenTrain Version 1.0 calling algorithm in Genome Studio version 2011.1 was used, as was zCall for post-processing rare variant calling. Subsequent quality control steps were conducted collaboratively between the Genetic Analysis Center at the University of Washington in Seattle and the statistical group at Dartmouth College. The array included 18 642 pairs and 51 trios of duplicated SNPs. Where SNPs were duplicated, the SNP with the lowest missing rate was retained for analysis. In the analysis of the array, 49 blind duplicate samples were analyzed of which 31 were study-specific duplicates and 18 were derived from HapMap referent panel. One HapMap control and one internal control were included on each 96-well plate that was analyzed. The median discordance rate for genotypes for these 49 blind duplicates was 1.7 × 10^–5. The genotyped HapMap samples included five trios, and of 936 945 unique markers, there were 1008 markers with one Mendelian error and 75 with two or more errors all of which were excluded from further analysis. A selected set of 907 markers from the whole genome amplified samples analyzed at the Mayo Clinic, one of the participating sites, provided different genotype frequencies from other participating sites. Results from the Mayo Clinic for those specific SNPs were removed from analyses. The median call rate for genotypes on this array was 99.9% and the error rate assessed from 31 pairs of study sample duplicates was 8.5 × 10^–6. All genotyped samples had a call rate of 97% or higher and were retained for further quality control checks. Relatively few SNPs were removed due to quality control steps, but on this array 14.47% of SNPs were monomorphic and were therefore removed from analyses (Supplementary Table 3, available at Carcinogenesis Online). Of the 951,117 SNPs assayed, 29% had minor allele frequencies of 1% or less. The detailed QC processes are presented in Supplementary Figure 1, available at Carcinogenesis Online.

Imputation in GELCC study

We performed genotype imputation in the familial lung cancer study, using the software package IMPUTE2 v2.3.0 (18) and 1000 Genomes Project Phase 1 as a reference panel, which contains whole genome sequence data from 1092 individuals including approximately 39 million variants, insertions/deletions, and structural variants. Imputation was done to 1000 Genomes Project Phase 1 version 3 integrated haplotypes produced with SHAPEIT2 for chromosome 1–22; for chromosome X, the previous version was used. The study samples were imputed in a two-stage procedure using SHAPEIT2 (version 2.r644, c2011-2012) to derive phased genotypes from the array SNPs and IMPUTE2 (version 2.3.0) to perform imputation of the phased data. We used the default parameters to derive phased genotypes with SHAPEIT2, increasing the number of conditioning haplotypes to be used during the phasing process to 200 (“--states 200 or –S 200”). The imputation was performed using 5MB segments over the length of each chromosome from the first to last position appearing in the reference panel. The flag “-use_prephased_g” was provided to indicate that prephased haplotypes were being used and the buffer region was increased to 500kb (“-buffer 500”). Imputation target variants were limited to those with at least two copies of the minor allele observed in any of the four reference panel super-populations, yielding 27 728, 264 total variants in the imputed dataset.

TRICL-ILCCO Data

The Transdisciplinary Research in Cancer of the Lung (16) and the International Lung Cancer Consortium (TRICL-ILCCO) is an international consortium that has assembled and analyzed GWAS data from lung cancer cases and controls. Among the studies that have collected data for TRICL-ILCCO, a subset had information about the presence of lung cancer in a first degree relative. From TRICL-ILCCO, 1,308 cases with a family history of lung cancer in a first degree relative from six studies were included (Supplementary Table 2, available at Carcinogenesis Online). An additional sample of 32 946 controls was available for analysis from the TRICL data. In all (GELCC plus TRICL-ILCCO), 1993 familial cases and 33 690 controls were included (Supplementary Table 1, available at Carcinogenesis Online). The same referent panel and procedures were used for the United States, Canadian and French sites for TRICL-ILCCO as were used for GELCC. The deCODE analysis included a referent set derived from Icelandic individuals who have been sequenced; it therefore provides more accurate imputation for rare variants than imputation analysis used by other sites.

Imputation in TRICL studies

Whole genome imputations for the MD Anderson Cancer Center (MDACC), deCODE, Samuel Lunenfeld Research Institute (SLRI) Study, Environmental and Genetics in Lung Cancer Etiology (EAGLE), and Prostate, Lung, Colon, and Ovary (PLCO) screening studies in TRICL-ILCCO consortium were conducted using IMPUTE2 v2.5 based on the 1000 Genomes Project Phase 1. For the International Agency for Research on Cancer (IARC) study imputation was performed using minimac (version 2012.10.3) based on 1000 Genome Project Phase 1 For MDACC, SLRI, EAGLE and PLCO studies, we used the default parameters to derive phased genotypes with SHAPEIT, increasing: the number of burn-in iterations used by the algorithm to reach a good starting point to 10 (“--burn 10”), the number of pruning iterations to find a parsimonious graph for each individual to 10 (“--prune 10”), and the number of iterations to compute transition probabilities in haplotype graphs to 50 (“--main 50”). We conducted imputation with IMPUTE2 using 5MB non-overlapping intervals. The flag “-use_prephased_g” was applied to indicate the prephased haplotypes were being used. The missing genotypes were replaced with imputed genotypes using the option “-pgs_miss” and the buffer region was increased to 500KB (“-buffer 500”). Supplementary Table 4, available at Carcinogenesis Online, provides the details for the number of SNPs that passed quality control on each study.

Statistical analysis

Using genotyped data from 1429 subjects of European descent in the GELCC study, we applied the EIGENSTRAT method to carry out principal components analysis. To assure robustness of our analytical approach, we performed association analyses adjusting for the top three eigenvectors based on the principal components analysis. Logistic regression was applied to evaluate the odds ratio per allele of each SNP, adjusting for age, sex and eigenvectors. Analyses were conducted using PLINK.

Meta-Analysis

Meta-analysis of the GELCC and TRICL-ILCCO studies was conducted using METAL software (http://csg.sph.umich.edu/abecasis/metal/index.html) that combined test statistics and standard errors across studies (Supplementary Table 5, available at Carcinogenesis Online). In addition, we determined whether there were differences between case groups according to the most common histologic subtypes of lung cancer: squamous (n = 367) and adenocarcinoma (n = 775). For the overall meta-analysis, GELCC and 6 TRICL-ILCCO studies were used (Supplementary Table 6A, available at Carcinogenesis Online). Histology specific meta-analysis was performed on GELCC and 5 TRICL-ILCCO studies (excluding the IARC study) (Supplementary Tables 6B and C, available at Carcinogenesis Online). For some histology specific analyses, some populations contributed small numbers of cases, a situation where rare variants can create unstable estimates of the effect size. Therefore, we excluded markers with imputation quality info score less than 0.5 and removed results when Odd Ratios (OR) were greater than 100 or less than 0.01 for each study to reduce false positive findings. Common SNPs that were present in at least five studies including GELCC plus at least four studies out of five TRICL-ILCCO studies were used for histology specific analyses. We also did not consider SNPs when the Mantel–Haenszel heterogeneity P-value test was ≤ 0.05. The lambda values showed minimal inflation except for squamous carcinoma, 1.015 and 1.004, respectively, before and after adjustment to reflect a standardized sample size of 1000 cases and 1000 controls (since lambda increases with sample size). Lambda values for histology-specific analyses were 1.040 and 1.026 (adenocarcinoma) and 1.079 and 1.108 (squamous cell) before and after correction based on 1000 cases and 1000 controls, respectively. Q-Q plots are presented in Supplementary Figure 2, available at Carcinogenesis Online. We used GCTA to evaluate joint and conditional models with chromosome 15q21.1 region markers (19,20).

Lung expression QTL analysis

Whole-genome gene expression profiling in 409 adjacent normal lung samples obtained from patients undergoing resection for lung cancer therapy was performed on a custom Affymetrix array (GPL10379). Microarray pre-processing and quality controls were described previously (21,22). Genotyping was carried out on the Illumina Human 1M-Duo BeadChip array. Expression traits were adjusted for age, sex and smoking status. Cis-eQTL were evaluated for transcripts (probe sets) located 1 Megabase up- and downstream of the sentinel lung cancer associated SNPs. Association tests were carried out using PLINK version 1.9. For the chromosome 9 region transcripts that were evaluated included IFNA gene cluster and the CDKN2A/CDKN2B region.

Results

In the meta-analysis of the GELCC and 6 TRICL-ILCCO studies, 44 and 41 variants (Supplementary Tables 7A and B, available at Carcinogenesis Online) were identified as associated with overall lung cancer and squamous cell carcinoma lung cancer (367 cases) in family history positive lung cancer cases, respectively, at a genome-wide level of significance (P < 5 × 10^–8) (Supplementary Figure 3, available at Carcinogenesis Online). No variants were identified for adenocarcinoma (775 cases) at a genome-wide significant level. In the analysis of overall lung cancers, the top hits (P < 5 × 10^-8) fall primarily within the CHRNA3/CHRNA5/CHRNB4 nicotinic acetylcholine receptor gene region on chromosome 15 (Supplementary Figure 4A, available at Carcinogenesis Online) (23). The most significant SNP on chromosome15, rs8040868, is in the region of CHRNA3. As presented in Table 1, the estimate of risk associated with this relatively common SNP (MAF = 0.42) in familial lung cancer is OR = 1.33 (P = 4.38 × 10⁻¹³). Even though this is a synonymous change, statistically significant changes in gene expression for CHRNA3, as well as CHRNA5 and PSMA4, are associated with this polymorphism, based on expression quantitative trait analysis (eQTL) from 409 noncancerous lung tissues removed for lung cancer resections, providing additional support for the role of this polymorphism or another variant in linkage disequilibrium with it in influencing lung cancer risk (Table 2; Supplementary Figure 5, available at Carcinogenesis Online).

Table 1.

Meta-analysis association test results for top SNPs, by histology type

Strata	SNP	Chr	Position	Nearest Gene	Function	Allelea	EAF	OR (95% CI)	P-value	Mean of QC Info Score
Overall	rs8040868	15q25.1	78911181	CHRNA3	Synonymous (V-V)	T_C	0.42	1.33 (1.23, 1.43)	4.38 × 10−13	0.99
SQC	rs71603396	4p15.31	18012035	LCORL	Intron	G_A	0.14	1.92 (1.55, 2.37)	1.73 × 10−9	0.88
	rs12686364	9p21.3	21670087	CDKN2A	Intergenic	A_G	0.004	8.49 (4.13, 17.46)	6.21 × 10−9	0.96

OR, odds (log additive) ratio; 95% CI, 95% confidence interval; SQC, squamous cell carcinoma.

^aDenotes reference_effect(coded) allele; Genome positions relative to GRCh37; EAF, effective allele frequency.

Table 2.

Gene expression QTL analysis in the lung for the three top SNPs

SNP	Chr	Major allele	Minor allele	MAF	# probes tested	ProbeSet	Gene	BETA	SE	P-valuea	% variance
rs8040868	15	T	C	0.474	45	100154936_IF147302_TGI_at	AF147302	−0.235	0.023	7.84 × 10−22	20.3
						100154936_IREB2_TGI_at	IREB2	−0.244	0.027	4.23 × 10−18	16.9
						100156434_TGI_at	CHRNA5	−0.121	0.019	3.43 × 10−10	9.3
						100148404_TGI_at	PSMA4	0.059	0.014	1.9 × 10−5	4.4
						100123649_TGI_at	CHRNA3	−0.151	0.039	1.5 × 10−4	3.5
						100312267_TGI_at	AJ584709	−0.109	0.032	0.001	2.7
						100143446_TGI_at	IREB2	−0.066	0.023	0.005	1.9
						100145993_TGI_at	CHRNA3	0.037	0.015	0.012	1.5
						100154605_TGI_at	CRABP1	0.078	0.035	0.025	1.3
						100142282_TGI_at	TBC1D2B	−0.027	0.013	0.031	1.2
						100135897_TGI_at	Not known	0.071	0.033	0.032	1.1
						100126596_TGI_at	DNAJA4	0.057	0.028	0.044	1.0
rs71603396	4	G	A	0.113	21	100160526_TGI_at	MED28	−0.063	0.026	0.01498	1.5
						100142682_TGI_at	LF383968	−0.128	0.054	0.01769	1.4
						100158710_TGI_at	BX648830	0.082	0.039	0.03589	1.1
rs12686364	9	A	G	0.0049	38	100144400_TGI_at	IFNA8	−0.390	0.191	0.04117	1.0

^aOnly genes with statistically significant changes in gene expression associated with SNP genotypes are presented.

In histology-specific analyses, two of the most significant associations were with familial lung squamous cell carcinoma. The first, rs71603396 on chromosome 4p15.31 is in the region of the ligand dependent nuclear receptor corepressor (LCORL) gene (MAF = 0.14, OR = 1.92, P = 1.73 × 10⁻⁹) (Supplementary Figure 4B, available at Carcinogenesis Online). This SNP rs71603396 in the LCORL gene was associated with gene expression in eQTL analysis (Table 2). The second significant SNP, rs12686364, is located in an intergenic region (Supplementary Figure 4C, available at Carcinogenesis Online, on chromosome 9p21.3 (MAF = 0.004, OR = 8.49, P = 6.21 × 10⁻⁹) that is upstream of IFNA8 and downstream of CDKN2A). eQTL analysis demonstrated an association between this SNP and expression of interferon-α8 (IFNA8) (Table 2). Results by study center are presented in Supplementary Table 8, available at Carcinogenesis Online.

Table 3 presents results for previously identified SNPs that influence lung cancer risk in European descent individuals and that had previously reached genome-wide levels of significance. Of the previously identified SNPs, only the rs55781567 variant on chromosome 15q25.1 reached a genome-wide level of significance. The nicotinic acetylcholine region of chromosome 15q25.1 has consistently been the most strongly associated region for lung cancer risk among unselected cases of European descent lung cancer. We further examined the two SNPs on 15q25.1; rs8040868 from familial lung cancer and rs55781567 from unselected lung cancer, which are 53,195 bp apart. The linkage disequilibrium between the two SNPs is D’ = 0.95 (R² = 0.73), indicated that these SNPs are in high linkage disequilibrium. We, therefore, performed the conditional analysis to evaluate whether rs8040868 (P =4.38 × 10⁻¹³) has association signal conditioning on rs55781567 (P =1.69 × 10⁻¹¹) (22). The conditional test of rs8040868 after adjusting for rs55781567 yielded P = 3.4 × 10⁻³. This indicates that rs8040868 still has association signal for familial lung cancer risk. While rs8040868 provided an independent signal, many SNPs in this region are strongly correlated and the signal from rs804868 could reflect effects from another causal variant in the region such as rs16969968, a causal SNP in CHRNA5. Conditional analysis of rs804868 allowing for effects from rs16969968 also showed a residual effect with P = 0.002. On chromosome 9p21.3, the effects between the previously identified SNP rs885518 and the newly identified SNP rs12686364 was very low (R² = 0.007, D’ = 0.452), with the very low R² value reflecting very different allele frequencies of the two variants. After conditioning on the previously identified SNP rs885518, which predisposes to adenocarcinoma, the association of rs12686364 remained highly significant (P = 5.93 × 10⁻⁹), supporting completely independent effects of these two SNPs on lung cancer risk.

Table 3.

Comparison to familial lung cancer study using the recently published susceptible loci in OncoArray lung cancer

Strata	Chromosome	SNP	Position	Gene	Allelea	EAF	OR (95%CI)	P-value
Lung	1p31.1	rs71658797	77967507	FUBP1	T_A	0.099	1.12 (0.99–1.27)	7.54 × 10−2
Lung	6q27	rs6920364	167376466	RNASET2	G_C	0.449	1.08 (1.00–1.17)	5.25 × 10−2
Lung	8p21.1	rs11780471	27344719	EPHX2, CHRNA2	G_A	0.063	0.80 (0.67–0.95)	1.00 × 10−2
Lung	13q13.1	rs11571833	32972626	BRCA2	A_T	0.010	1.66 (1.14–2.41)	7.78 × 10−3
Lung	15q21.1	rs66759488	47577451	SEMA6D	G_A	0.362	1.05 (0.97–1.14)	1.91 × 10−1
Lung	15q25.1	rs55781567	78857986	CHRNA5	C_G	0.366	1.31 (1.21–1.41)	1.69 × 10−11
Lung	19q13.2	rs56113850	41353107	CYP2A6	C_T	0.429	0.91 (0.83–0.99)	2.86 × 10−2
Adeno	3q28	rs13080835	189357199	TP63	G_T	0.489	0.96 (0.86–1.08)	5.28 × 10−1
Adeno	5p15.33	rs7705526	1285974	TERT	C_A	0.343	1.28 (1.13–1.46)	1.45 × 10−4
Adeno	8p12	rs4236709	32410110	NRG1	A_G	0.225	1.10 (0.96–1.26)	1.56 × 10−1
Adeno	9p21.3	rs885518	21830157	MTAP/CDNK2A	A_G	0.110	1.37 (1.16–1.62)	2.20 × 10−4
Adeno	10q24.3	rs11591710	1.06E+08	OBFC1	A_C	0.143	1.14 (0.97–1.33)	1.21 × 10−1
Adeno	11q23.3	rs1056562	1.18E+08	MPZL3/AMICA1	C_T	0.471	1.08 (0.96–1.21)	1.89 × 10−1
Adeno	15q21.1	rs77468143	49376624	SECISBP2L	T_G	0.249	0.87 (0.76–0.99)	4.00 × 10−2
Adeno	20q13.33	rs41309931	62326579	RTEL1	G_T	0.119	1.35 (1.14–1.60)	6.14 × 10−4
SQC	6p21.33	rs116822326	31434111	MHC	A_G	0.161	1.26 (1.03–1.54)	2.36 × 10−2
SQC	12p13.33	rs7953330	998819	RAD52	G_C	0.313	0.79 (0.66–0.94)	7.05 × 10−3
SQC	22q12.1	rs17879961	29121087	CHEK2	A_G	0.004	2.50 (0.76–8.15)	1.29 × 10−1

EAF, effective allele frequency; OR, odds (log additive) ratio; 95% CI, 95% confidence interval; Adeno, adenocarcinoma; SQC, squamous cell carcinoma.

^aDenotes reference_effect(coded) allele. Variants in bold are associated with familial lung cancer at nominal significance level of 0.05. Genome positions relative to GRCh37.

Supplementary Tables 6A–C, available at Carcinogenesis Online, present results stratified by analysis in GELCC alone, TRICL alone or combined. Results for the GELCC study only identified several additional variants that were not replicated on analysis of the data from the TRICL studies. Variants in BCHE, PDGFRA and CFTR genes were highly significant (P < 5 × 10⁻⁷) for overall lung cancer, but all protective of developing lung cancer and did not replicate in the TRICL analysis. Similarly, highly significant associations (P < 5 × 10⁻⁷) were noted for adenocarcinoma including the GSG1L and ZC3H12D genes and for squamous associations were noted for RP11-335M9.1, EPDR1, CNTN4 and a gene desert on chromosome 14, but none of these associations were supported when combined with the TRICL data.

Discussion

Prior studies of familial lung cancer have identified rare variants that substantially increase lung cancer risk (24,25). Common genetic variants have been associated with sporadic lung cancer risk in GWAS studies (26). The study presented is the largest study to use a GWAS approach to more broadly identify genetic risk in familial lung cancer. Selecting cases with a family history of disease increases the power to detect rare variants that increase cancer risk (27). This study of familial lung cancer confirmed genetic variation on CHRNA3 and CHRNA5 on chromosome 15q25.1 previously reported in sporadic lung cancer. For all lung cancers, the top hits (P < 5 × 10⁻⁸) fall primarily within the CHRNA3/CHRNA5/CHRNB4 nicotinic acetylcholine receptor regions on chromosome 15. The most significant SNP on chromosome15, rs8040868, has shown to be associated with lung cancer risk regardless of family history (16) (Supplementary Figure 4A, available at Carcinogenesis Online). In European descent individuals, rs8040868 is very strongly associated (R² = 0.78) with the causal variant rs16969968 that changes signal transduction of CHRNA5 (28,29) The variant rs55781567 that was previously identified in unselected lung cancer cases (11) as most strongly associated with lung cancer SNP was also highly significant in our study (MAF = 0.37; OR = 1.31, P = 1.69 × 10⁻¹¹). This variant was previously shown to tag an insertion/deletion polymorphism in the promoter of CHRNA5 that affects its expression (30,31). The estimate of risk we report in familial lung cancer is similar to risk estimates for SNPs in the chromosome 15q25.1 region reported for sporadic lung cancer (5,7).

Two of the most significant associations were with familial lung squamous cell carcinoma. The first SNP, rs71603396 on chromosome 4p15.31, is in the region of the LCORL gene. The association between familial lung cancer risk and this SNP in the LCORL gene was also supported by the eQTL analysis. LCORL regulates transcription from RNA polymerase II and mutations in LCORL have frequently been noted in all histologies of lung cancer. In TCGA data, deletions of LCORL were observed in 0.5% of squamous lung cancers (32) and LCORL is often deleted in pancreatic cancers (5.5% of cases) (33). The second significant SNP associated with familial squamous cell lung cancer, rs12686364, is located in an intergenic region on chromosome 9p21.3, upstream of IFNA8 and downstream of CDKN2A. eQTL analysis again provided support for this association. Published results show that IFNA8 was deleted in 30% of glioblastomas (34), 24.8% of pancreatic cancers (33), and deleted or mutated in 12.9% of squamous lung cancers (32). Deletions of IFNA8 are associated with improved survival in glioma patients (35,36).

Our results were also compared to the recently published variants associated with risk of sporadic lung cancer from a very large study (Table 3) (11). There is some heterogeneity in findings for familial lung cancer compared with those previously reported for the largest study to date of unselected lung cancer cases. In particular, variants in TP63 have been reproducibly associated with lung adenocarcinoma risk, but showed no association in this study. The rare variant rs17879961 of CHEK2 in sporadic lung cancer is highly protective from lung cancer development but in these familial cancers, the association analysis yielded strongly positive association (odds ratio of 2.5). Given the small sample size, this association was not significant and larger studies are needed to evaluate this finding further. While this is the largest study to date of a GWAS for familial lung cancer, the sample size is still quite limited, which may largely explain why many known associations for sporadic lung cancer were not associated with significant findings in this study. Another limitation in this study is the difference in family history between the Genetic Epidemiology of Lung Cancer Consortium, which recruited families with two or more first or second degree relatives and the restriction to families with one or more first degree relatives for the other studies. However, in conducting this study we sought to identify reliable findings that met genome wide significance and to be able to evaluate the findings in multiple independent studies.

In summary, a novel variant rs71603396 on chromosome 4p15.31 and an imputed rare variant rs12686364 on 9p21.3 were identified at a genome-wide level of significance for familial squamous cell carcinomas. Also, the associations of CHRNA3 and CHRNA5 on chromosome 15q25.1 were confirmed in this familial lung cancer data. The identification of high risk variants in individuals selected through a positive family history provides insights into the etiology of lung cancer and may improve our ability to target selected individuals for lung cancer screening.

Supplementary material

Supplementary materials can be found at Carcinogenesis online.

Funding

National Institute of Health (U01CA76293, U19CA148127, P30CA22453, HHSN26820100007C); The Intramural Research Program of the National Human Genome Research Institute; National Institute of Health (U19CA148127, CA148127S1) (to the Transdisciplinary Research for Cancer in Lung (TRICL)); Cancer Care Ontario Research Chair of Population Studies (to RH); Canadian Cancer Society Research Institute (no. 020214) (to the SLRI GWAS); Ontario Institute for Cancer Research (to RH); National Institute of Health (R01CA060691, R01CA87895) (to Detroit GELCC study). National Institute of Health -National Cancer Institute (CA77118, CA80127, CA84353) (to Mayo Clinic Lung Study); National Institute of Health (P30ES006096) (to Cincinnati study); Genotyping services for the GELCC study were provided by the Center for Inherited Disease Research (CIDR). CIDR is fully funded through a federal contract from the National Institutes of Health to The Johns Hopkins University, contract numbers HHSN268200782096C and HHSN268201100011.

Abbreviations

eQTL

expression quantitative trait loci

EAGLE

Environment And Genetics in Lung cancer Etiology

GELCC

The Genetic Epidemiology of Lung Cancer Consortium

GWAS

Genome-wide association studies

ILCCO

International Lung Cancer Consortium

IARC

International Agency for Research on Cancer

MDACC

MD Anderson Cancer Center

PLCO

Prostate, Lung, Colon, and Ovary screening studies

SNP

single nucleotide polymorphism

TRICL

Transdisciplinary Research in Cancer of the Lung