Telomere structure and maintenance gene variants and risk of five cancer types

Abstract

Telomeres cap chromosome ends, protecting them from degradation, double-strand breaks, and end-to-end fusions. Telomeres are maintained by telomerase, a reverse transcriptase encoded by TERT, and an RNA template encoded by TERC. Loci in the TERT and adjoining CLPTM1L region are associated with risk of multiple cancers. We therefore investigated associations between variants in 22 telomere structure and maintenance gene regions and colorectal, breast, prostate, ovarian, and lung cancer risk. We performed subset-based meta-analyses of 204,993 directly-measured and imputed SNPs among 61,851 cancer cases and 74,457 controls of European descent. Independent associations for SNP minor alleles were identified using sequential conditional analysis (with gene-level P-value cutoffs ≤3.08×10⁻⁵). Of the thirteen independent SNPs observed to be associated with cancer risk, novel findings were observed for seven loci. Across the TERT-CLPTML1 region, rs12655062 was associated positively with prostate cancer, and inversely with colorectal and ovarian cancers, and rs115960372 was associated positively with prostate cancer. Across the TERC region, rs75316749 was positively associated with colorectal, breast, ovarian, and lung cancers. Across the DCLRE1B region, rs974404 and rs12144215 were inversely associated with prostate and lung cancers, and colorectal, breast, and ovarian cancers, respectively. Near POT1, rs116895242 was inversely associated with colorectal, ovarian, and lung cancers, and RTEL1 rs34978822 was inversely associated with prostate and lung cancers. The complex association patterns in telomere-related genes across cancer types may provide insight into mechanisms through which telomere dysfunction in different tissues influences cancer risk.

Introduction

Telomeres are complex nucleoprotein structures that cap chromosome ends (1,2), protecting them from degradation, double strand breaks, and end-to-end fusions (1,2). Thus, telomeres play an essential role in preserving genomic stability. Telomeres are maintained by the enzyme telomerase, which is made up of a reverse transcriptase encoded by TERT, and an RNA template encoded by TERC (1,2), with several other associated proteins encoded by DKC1, NOP10, NHP2, NAF1, and GAR1 (1). The telomere structure itself is composed of simple tandem TTAGGG repeats bound by six proteins (encoded by TERF1, TERF2, TINF2, TERF21P, ACD, and POT1), termed shelterin. Other proteins that interact with shelterin are encoded by OBFC1, RTEL1, DCLRE1B, TNKS, PINX1, and TEP1 (1). Germline SNPs in TERC, TERT, RTEL1, NAF1 (3), and OBFC1 (3,4) have been associated with telomere length in genome-wide association studies (GWAS). Additional genes associated with telomere length include: BICD1 (5), ACYP2, ZNF208, MPHOSPH6 (3), and DCAF4 (6).

Susceptibility loci for multiple cancer types have been identified in the TERT and adjoining CLPTM1L gene region in GWAS. Both increased and decreased risk associations have been reported for some loci for different cancers (7–9), suggesting complex patterns of associations across cancer types which could be due to tissue specificity or interactions with risk factors. Because properly functioning telomeres are vital for genomic stability and chromosomal integrity, genetic variants in other telomere structure and maintenance genes may affect cancer risk. Therefore, we sought to examine whether pleiotropic associations for variants in telomere structure and maintenance genes are observed across cancer types within the Genetic Associations and Mechanisms in Oncology Network (GAME-ON) (10) and the Genetic and Epidemiology of Colorectal Cancer Consortium (GECCO) (11).

GAME-ON was established by the National Cancer Institute (NCI) to foster collaborative post-GWAS research across consortia of colorectal, breast, prostate, ovarian, and lung cancers (10). The extensive genomic data available through GAME-ON and GECCO, including over 61,000 cases and 74,000 controls, were utilized to identify and systematically characterize patterns of associations between independent variants in 22 telomere structure and maintenance gene regions and risk of colorectal, breast, prostate, ovarian, and lung cancers.

Materials and Methods

Study Population

Our analysis included 61,851 cancer cases and 74,457 controls of European descent from 45 GWAS (12) (Table 1). Details of each study have been described previously (10–19) (Supplementary Table 1); at minimum, cases were frequency-matched to controls on age and sex. Each study obtained informed consent from participants; study procedures including certifications required for data sharing in accordance with National Institutes of Health policies were approved by all Institutional Review Boards.

Table 1.

Characteristics of genome-wide association studies included in consortium-based meta-analyses of colorectal, breast, prostate, ovarian, and lung cancers

Cancer Type/Subtype-Consortium	Cases (N)	Controls (N)	GWAS (N)	Genotyping Platform	Covariates
Colorectal- GECCO	10,314	12,857	13	Illumina 300/240S, 300K, 550K, 610K, 730K; Affymetrix 100K, 500K	age, sex, PCA, center, batch effecta, smokingb
Colorectal- CORECT	5,100	4,831	6	Affeymetrix Axiom	age, sex, PCA
Breast- DRIVE	15,748	18,084	11	llumina 240K/317K/370K/550K/610K/610K+Cyto12/660K/670K/1.2M; Affymetrix 5.0/6.0	age, PCA
ER- negative	4,939	13,128	8		age, PCA
Prostate- ELLIPSE	14,160	12,724	6	Illumina 550K/610K/2.5M/iSELECT; Affymetrix GeneChip 5.0	age, study
Aggressive	4,450	12,724	6		age, study, PCA
Ovarian- FOCI	4,369	9,123	3	Illumina 317K/370K/550K/610K/670K/2.5M	site, PCA, age
Endometrioid	715	9,123	3		site, PCA, age
Serous	2,556	9,123	3		site, PCA, age
Lung- TRICL	12,160	16,838	6	Illumina 317K/370K/550K/610K	age, sex, PCA
Adenocarcinoma	3,718	15,871	6		age, sex, PCA
Squamous	3,422	16,015	6		age, sex, PCA
Total	61,851	74,457	45

Abbreviations: CORECT- ColoRectal Transdisciplinary Study; DRIVE- Discovery, Biology, and Risk of Inherited Variants in Breast Cancer; ELLIPSE- Elucidating Loci Involved in Prostate Cancer Susceptibility; ER-estrogen receptor; FOCI- Follow-up of Ovarian Cancer Genetic Association and Interaction Studies; GWAS- genome wide association studies; N- number; PCA- principal components analysis; TRICL- Transdisciplinary Research in Cancer of the Lung.

^aAdjusted for batch effect only in the Association STudy Evaluating RISK for Sporadic Colorectal Cancer (ASTERISK) study

^bAdjusted for smoking only in the Physician’s Health Study (PHS)

Consortium-based Imputation and Meta-analysis

Genotyping was performed using Illumina and Affymetrix GWAS platforms. Each consortium imputed unmeasured single nucleotide polymorphisms (SNPs) for their GWAS data from the 1000 Genomes (Phase 1) March 2012 Build 37 reference panel using MACH, IMPUTE, or Minimac (10–19) Supplementary Table 2. Within each consortium, per-allele odds ratios (ORs) and 95% confidence intervals (CIs) for each SNP and cancer risk were calculated using unconditional logistic regression. Study-specific results were combined using fixed-effects meta-analysis.

Gene Selection

We examined 204,993 SNPs within one mega-base upstream and downstream of the transcription start and end sites of the following genes, selected either because of their relevance to telomere structure and maintenance, or telomere length: ACD, ACYP2, BICD1, DCLRE1B, GAR1, MPHOSPH6, NAF1, NHP2, NOP10, OBFC1, PIK3C3, PINX1-TNKS, POT1, RTEL1, TEP1, TERC, TERF1, TERF2, TERF2IP, TERT-CLPTM1L, TINF2, and ZNF208. The chromosomal location and number of SNPs evaluated in each gene is in Supplementary Table 3 (20).

Cross-Cancer Association Analysis

ASSociation analysis based on SubSET (ASSET) meta-analysis allows for identification of associations that may be in the same, or opposite, direction for some cancer types versus others (21). We performed one-sided and two-sided ASSET analyses using summary data for each of the five cancer types, and repeated analyses additionally including the following cancer subtypes: estrogen receptor (ER) negative breast; aggressive prostate (defined as Gleason score ≥8, disease stage ‘distant’, prostate-specific antigen level >100 ng/ml, or death from prostate cancer (17)); endometrioid and serous ovarian; and adenocarcinoma and squamous lung. Other tumor subtypes were not independently evaluated due to low frequencies. ASSET takes into account matrices of overlapping cases and controls across datasets including overlap between cancer types and subtypes (Supplementary Table 4), and adjusts for correlations across studies. ASSET groups cancer types by the direction of their associations and identifies the strongest associations, so multiple testing penalties may be incurred, widening the CIs of summary results (21). A Manhattan plot of P-values from our two-sided unconditional ASSET analysis was produced in R Studio [http://www.rstudio.com]. Forest plots of two-sided unconditional ASSET meta-analysis results for individual SNPs were generated by cancer type and subtype. Because ASSET takes into account overlap between cancer types and subtypes, associations appearing statistically significant for a given cancer type (or subtype) may be included in the “null” category if the association is actually driven by that cancer’s subtype(s) included in the ‘positive” or “inverse” category. Statistically significant positive or inverse associations are only interpretable within ASSET if the overall one-sided (positive or negative) test is statistically significant.

Gene-level association tests to evaluate all SNPs within a gene and cancer risk after taking linkage disequilibrium (LD) into account were performed using VEGAS2 (22) on the overall two-sided unconditional ASSET meta-analysis P-values for all SNPs +/−50kb of each gene.

Identifying SNPs in Linkage Disequilibrium

Because GAME-ON and GECCO data included summary statistics for each SNP, not individual-level data, we could not calculate LD directly. Instead, we determined LD using individual-level data from European ancestry subjects in the Cancer Genetic Markers of Susceptibility (CGEMS) Project (23) and the Environmental and Genetics in Lung Cancer Etiology (EAGLE) study (24). To be comparable to the summary data used for our analyses, we imputed SNPs with IMPUTE2 (25) from 1000 Genomes (Phase 3) October 2014 Build 37 in CGEMS and EAGLE (26). Of 204,718 SNPs in the summary data, 7,015 SNPs could not be imputed in CGEMS and EAGLE because they were not present in 1000 Genomes (Phase 1) data used to impute the GAME-ON and GECCO data (12,13). Additionally, 8,977 SNPs failed quality-control measures (information score <0.3) and 96 were multi-allelic, leaving 188,630 markers in CGEMS and EAGLE for analysis. We identified sets of SNPs with r²>0.70 in CGEMS and EAGLE using Haploview (27). Given the complicated LD patterns in TERT-CLPTM1L and TERC, we generated LD plots of all significant SNPs from ASSET analyses, to the extent possible, with r²<0.70 (27).

Determining Gene-level P-value Thresholds

We used the Genetic type 1 Error Calculator (GEC) to calculate the effective number of independent tests (M_e) and statistical significance P-value threshold for each gene (28). This method, developed to address the issue of multiple testing with SNPs in LD, utilizes eigenvalues derived from matrices of association test P-values between SNPs to calculate M_e. For each gene, the P-value threshold required to keep type I error at 5% equals alpha divided by M_e. Before applying the GEC, for simplicity we removed redundant SNPs (r²>0.98) from CGEMS and EAGLE using gPLINK version 1.07 [http://pngu.mgh.harvard.edu/purcell/plink/] ensuring that directly measured SNPs in our dataset were not eliminated, leaving 98,783 markers. M_e and P-value thresholds for each gene are in Supplementary Table 3.

Conditional Analysis

To identify independent associations, we performed sequential conditional analysis using Yang et al.’s method for summary-level data (29). For each gene, SNPs were ranked by P-value, and in each step, a single SNP was added to the ASSET analysis, conditioning on SNPs that were most significantly associated in previous steps. This process was repeated until the two-sided P-value for the most significant SNP for a step remained below the M_e P-value. To avoid collinearity, in each step, the program assesses r² between the next SNP to add and the SNPs that are already included in the model, and skips SNPs that are correlated (in this case, r²>0.80). To evaluate if this resulted in over-fitting, we performed a sensitivity analysis by conducting sequential conditional analysis of TERT-CLPTM1L using pruned variants with r²≤0.70. No evidence of over-fitting was observed (data not shown).

For SNPs with two-sided P-values that reached multiple comparison-adjusted gene-level significance, we assessed whether both the positive and inverse results contributed to the association (versus the association being driven primarily by one-sided results) by evaluating whether the two-sided P-value was smaller than the most significant one-sided P-value. We used an arbitrary P-value cutoff of 0.01 for the contributing one-sided associations, and considered P-values between 0.01–0.05 as suggestive.

Functional annotations for SNPs with observed associations that have not been previously reported were obtained from HaploReg Version 4.1 on June 14^th, 2016 (30). HaploReg is a data repository which integrates information on sequence conservation, regulatory protein binding, epigenomic evidence, expression quantitative trait loci, and regulatory motifs from several sources including the ENCODE project, the GRASP database, GTEx, SiPhy, and multiple other studies (30).

Results

We examined 204,993 SNPs in 22 telomere structure and maintenance gene regions and colorectal, breast, prostate, ovarian, and lung cancer risk in 61,851 cancer cases and 74,457 controls (Table 1). ASSET unconditional two-sided analysis combined P-values for each SNP are shown in the Manhattan plot (Supplementary Figure 1). VEGAS2 gene-based association tests evaluating all SNPs in each gene in aggregate and cancer risk were statistically significant for DCLRE1B (P=1.1×10⁻⁵), TERT-CLPTM1L (P=1.0×10⁻⁶), and RTEL1 (P=9.4×10⁻⁴). Using the per-gene P-value threshold for the effective number of independent tests, we observed significant associations with cancer risk for 89 DCLRE1B, 153 TERC, 1 GAR1, 95 TERT-CLPTM1L, 2 POT1, 1 TERF2, and 7 RTEL1 SNPs (Supplementary Table 5). After removing SNPs in LD at r²>0.70 with the lead SNP, 3 DCLRE1B, 19 TERC, 1 GAR1, 23 TERT-CLPTM1L, 2 POT1, 1 TERF2, and 2 RTEL1 SNPs remained (Table 2). Correlations between these SNPs (r² and D′) are in Supplementary Table 6. Supplementary Table 7 includes r² correlations between these SNPs and all other significantly associated SNPs by gene. Even after pruning, 3 SNP pairs in TERC remained correlated with r²>0.70 (rs75982374 and rs76925190; rs80304993 and rs969217; rs59758024 and rs9865021), as did 2 SNP pairs in TERT (rs35953391 and rs37004; rs3816659 and rs37005). LD between these highly correlated SNPs and the variants retained is in Supplementary Figure 2.

Table 2.

Unconditional ASSET two-sided meta-analysis results across five cancer types^a

Gene (Chr.)		Ref: MA	MAF	Combined P-value	Positively Associated		Inversely Associated		Cancer types
SNP	Position				OR (95% CI)	P-value	OR (95% CI)	P-value	Positively Associated	Negatively Associated
DCLRE1B (Chr. 1)
rs974404	114382025	T:G	0.449	9.19E-06e	1.04 (1.01–1.07)	2.47E-02f	0.94 (0.91–0.97)	2.43E-05e	Breast, Ovarian	Prostate, Lung
rs7523862b	114443419	G:A	0.379	1.09E-05e			0.94 (0.91–0.97)	1.17E-05e		Prostate, Lung
rs12144215	114187155	G:T	0.131	1.50E-05e			0.90 (0.87–0.94)	2.11E-06e		Colorectal, Prostate
TERC (Chr. 3)
rs80304993	170097606	G:A	0.230	6.54E-15e			0.82 (0.78–0.86)	1.51E-15e		Prostate
rs71277158	169999216	T:G	0.162	8.88E-15e			0.80 (0.76–0.84)	3.64E-16e		Prostate
rs76925190	170066339	A:C	0.173	1.25E-14e			0.80 (0.76–0.84)	6.27E-16e		Prostate
rs75982374	170063227	A:G	0.140	4.65E-13e			0.79 (0.74–0.84)	2.62E-14e		Prostate
rs55953261	170121598	G:A	0.488	4.58E-09e			0.89 (0.85–0.92)	3.57E-09e		Prostate
rs77085460	170127536	A:G	0.063	7.30E-09e			0.76 (0.70–0.83)	3.81E-10e		Prostate
rs59758024c	170119352	A:T	0.447	9.03E-09e	1.12 (1.08–1.17)	7.08E-09e			Prostate
rs75316749	168761423	A:G	0.041	1.38E-06e	1.14 (1.08–1.20)	1.38E-06e			Colorectal, Breast, Ovarian, Lung
rs75313056	170017609	G:A	0.082	1.51E-08e			0.80 (0.74–0.86)	9.54E-10e		Prostate
rs12487040	170103592	T:C	0.372	1.70E-08e	1.05 (1.01–1.09)	2.46E-02f	0.89 (0.85–0.93)	3.13E-08e	Breast, Ovarian	Prostate
rs10804842	170135700	T:C	0.234	1.73E-08e			0.86 (0.82–0.90)	1.94E-09e		Prostate
rs969217	170159134	C:T	0.391	2.73E-07e	1.06 (1.02–1.11)	2.62E-03	0.90 (0.87–0.94)	5.47E-06e	Breast	Prostate
rs77964281	169916180	T:C	0.117	3.49E-07e			0.85 (0.80–0.90)	2.65E-08e		Prostate
rs62293480	170106672	G:T	0.388	1.35E-06e			0.89 (0.84–0.93)	5.97E-07e		Prostate
rs10936633	170158128	G:A	0.493	1.49E-06e			0.90 (0.87–0.94)	7.74E-07e		Prostate
rs9865021	170146881	C:T	0.487	2.48E-06e	1.10 (1.06–1.14)	2.11E-06e			Prostate
rs74677551	168861788	T:G	0.032	3.20E-06e	1.15 (1.08–1.22)	3.20E-06e			Colorectal, Breast, Prostate, Ovarian, Lung
rs9809168	168803900	T:C	0.033	1.20E-05e	1.15 (1.08–1.22)	1.20E-05e			Colorectal, Breast, Ovarian, Lung
rs2901621	170057704	G:C	0.098	1.77E-05e			0.93 (0.91–0.96)	3.52E-06e		Colorectal, Prostate
GAR1 (Chr. 4)
rs17042238d	111745854	A:G	0.003	6.33E-06e			0.04 (0.01–0.16)	6.33E-06e		Prostate
TERT-CLPTM1L (Chr. 5)
rs37004b	1356684	C:T	0.239	2.27E-11e			0.84 (0.81–0.88)	1.29E-12e		Lung
rs37005b	1356450	C:T	0.460	1.98E-10e			0.87 (0.84–0.91)	9.85E-12e		Lung
rs3816659b	1317820	G:A	0.441	2.44E-10e			0.88 (0.85–0.91)	9.97E-12e		Lung
rs2736100b	1286516	C:A	0.500	3.38E-10e	1.05 (1.01–1.09)	1.72E-02f	0.90 (0.86–0.93)	7.54E-10e	Colorectal, Prostate	Lung
rs7725218b	1282414	G:A	0.359	3.02E-09e	1.12 (1.07–1.17)	3.14E-07e	0.90 (0.85–0.96)	4.04E-04	Lung	Prostate
rs35953391b	1312329	C:T	0.201	6.44E-09e			0.87 (0.83–0.91)	1.43E-09e		Lung
rs2735940b	1296486	G:A	0.499	7.44E-09e	1.09 (1.05–1.14)	1.91E-05	0.93 (0.90–0.96)	1.70E-05	Lung	Colorectal, Prostate
rs2736099b	1287340	G:A	0.344	8.62E-09e	1.12 (1.08–1.17)	1.75E-07e	0.95 (0.92–0.98)	2.18E-03	Lung	Colorectal, Breast, Prostate
rs35029535	1284976	C:T	0.352	5.54E-08e	1.09 (1.03–1.15)	2.09E-03	0.92 (0.89–0.95)	1.28E-06e	Prostate	Breast, Ovarian, Lung
rs7713218	1283312	G:A	0.497	5.78E-08e	1.10 (1.06–1.14)	6.60E-07e	0.95 (0.91–0.98)	4.23E-03	Ovarian, Lung	Colorectal, Prostate
rs10866498	1285162	C:T	0.472	6.36E-08e	1.05 (1.01–1.09)	6.76E-03	0.91 (0.88–0.94)	4.57E-07e	Colorectal, Prostate	Ovarian, Lung
rs2735948b	1299213	G:A	0.418	7.70E-08e			0.88 (0.85–0.92)	5.56E-09e		Lung
rs36019446	1339890	A:G	0.484	1.57E-07e			0.88 (0.85–0.92)	1.05E-08e		Lung
rs2736098b	1294086	C:T	0.234	2.48E-07e	1.08 (1.04–1.12)	1.81E-04	0.93 (0.90–0.97)	7.16E-05	Prostate, Lung	Colorectal, Breast, Ovarian
rs7717443	1283486	C:T	0.483	5.37E-07e	1.10 (1.06–1.14)	2.24E-06e	0.95 (0.91–0.989)	1.31E-02f	Ovarian, Lung	Colorectal, Prostate
rs115960372	1518494	C:T	0.104	6.94E-07e	1.19 (1.1–1.27)	2.97E-06e	0.90 (0.83–0.98)	1.29E-02f	Prostate	Lung
rs2735944b	1304432	C:T	0.132	1.27E-06e			0.85 (0.80–0.90)	1.38E-07e		Lung
rs2853677b	1287194	A:G	0.400	1.33E-06e	1.11 (1.06–1.16)	1.54E-06e	0.97 (0.93–1.000)	4.99E-02f	Lung	Colorectal, Breast, Prostate
rs12655062	1890877	G:A	0.354	1.65E-06e	1.12 (1.06–1.18)	3.53E-05	0.95 (0.92–0.98)	2.72E-03	Prostate	Colorectal, Ovarian
rs2736109b	1296759	C:T	0.392	2.99E-06e	1.11 (1.06–1.16)	5.88E-06e	0.96 (0.93–0.996)	3.08E-02f	Lung	Colorectal, Breast, Ovarian
rs33961405b	1277577	A:G	0.491	1.20E-05e	1.11 (1.06–1.16)	4.55E-06e			Lung
rs55901723	1342154	T:C	0.232	2.14E-05			0.88 (0.84–0.93)	2.94E-06e		Lung
rs6861230	304003	T:C	0.042	2.70E-05	1.25 (1.13–1.37)	5.83E-06e			Breast, Ovarian
POT1 (Chr. 7)
rs116895242	123946403	T:A	0.041	5.21E-05			0.83 (0.77–0.90)	6.99E-06e		Colorectal, Ovarian, Lung
rs74986217	123465182	A:C	0.041	2.54E-04e	1.31 (1.16–1.48)	2.17E-05e			Ovarian
TERF2 (Chr. 16)
rs117496043c	69590365	C:T	0.003	4.28E-05	1.66 (1.33–2.06)	6.14E-06e			Prostate
RTEL1 (Chr. 20)
rs34978822c	62291599	C:G	0.015	2.14E-05			0.71 (0.62–0.82)	3.17E-06e		Prostate, Lung
rs114220381c	61477960	T:A	0.048	1.21E-04	1.31 (1.16–1.48)	1.13E-05e			Prostate

Abbreviations: Chr.- chromosome; CI- confidence interval; MA- Minor Allele; OR- odds ratio; Ref- reference; SNP- single nucleotide polymorphism.

^aResults are presented for SNPs after pruning at r²<0.70.

^bSNPs that are directly measured and not imputed.

^cASSET meta-analytical results for these SNPs are based on 4 cancer types rather than all 5 studies.

^dASSET meta-analytical results for these SNPs are based on 2 cancer types rather than all 5 studies.

^eGene level P-value thresholds based on the number of effective tests are: DCLER1B P-value<2.65×10−5; TERC P-value<2.45×10−5; GAR1 P-value<2.44×10−5; TERT-CLPTM1 P-value<1.32×10−5; POT1 P-value<2.94×10−5; TERF2 P-value<3.08×10−5; RTEL1 P-value<1.86×10−5.

^fPositive or negative associations with P-values between 0.01 and 0.05 are considered to be suggestive.

For GAR1 and TERF2, only single SNPs reached gene-level significance, and the associations were entirely driven by prostate cancer. However, data were available only for prostate and ovarian cancers for the SNP in GAR1, and for colorectal, prostate, ovarian, and lung cancers for the SNP in TERF2. These two SNPs are “very rare” variants with minor allele frequencies (MAF) of 0.3%, making them difficult to impute. For POT1 and RTEL1, only one SNP in each gene was significantly associated in sequential conditional analyses. RTEL1 rs34978822 was associated with prostate and lung cancers (and was not investigated in breast cancer). rs34978822, and two SNPs in LD with it, are associated with chromatin structure changes in a large number of cell lines reported in HaploReg (30). POT1 rs116895242 was associated with colorectal, ovarian, and lung cancers (Table 2); this SNP creates six motif changes that may affect transcription factor binding (30).

Table 3 presents results from unconditional and sequential conditional analysis of DCLRE1B, TERC, and TERT-CLPTM1L gene regions, including all SNPs with ASSET two-sided results that reached gene-level P-value cutoffs. Sequential conditional analysis identified 11 independent SNPs associated with risk of multiple cancers. For all conditional results, two-sided P-values are smaller than one-sided P-values (data not shown).

Table 3.

Unconditional and conditional ASSET two-sided meta-analysis results across five cancer types

Unconditional Results						Cancer types		Conditional Results
Gene (Chr.) SNP	Combined P-value	Positively Associated		Inversely Associated				Combined P-value	Positively Associated		Inversely Associated
		OR (95% CI)	P-value	OR (95% CI)	P-value	Positively Associated	Negatively Associated		OR (95% CI)	P-value	OR (95% CI)	P-value
DCLRE1B (Chr. 1)
rs974404a	9.19e-06b	1.04 (1.01–1.07)	2.47E-02c	0.94 (0.91–0.97)	2.43E-05b	Breast, Ovarian	Prostate, Lung
rs12144215	1.50E-05b			0.90 (0.87–0.94)	2.11E-06b	Lungd	Colorectal, Breastd, Prostate	2.07E-05b	1.05 (1.004–1.11)	3.31E-02c	0.93 (0.90–0.96)	4.33E-05
TERC (Chr. 3)
rs80304993a	6.54E-15b			0.82 (0.78–0.86)	1.51E-15b		Prostate
rs62293480	1.35E-06b			0.89 (0.84–0.93)	5.97E-07b		Prostate	2.16E-13b			0.84 (0.81–0.88)	1.44E-14b
rs75316749	1.38E-06b	1.14 (1.08–1.20)	1.38E-06b			Colorectal, Breast, Ovary, Lung		1.46E-06b	1.14 (1.08–1.20)	1.46E-06b
TERT-CLPTM1L (Chr. 5)
rs37004a,e	2.27E-11b			0.84 (0.81–0.88)	1.29E-12b		Lung
rs7717443	5.37E-07b	1.10 (1.06–1.14)	2.24E-06b	0.95 (0.91–0.99)	1.31E-02c	Ovary, Lung	Colorectal, Prostate	1.26E-07b	1.11 (1.06–1.15)	5.31E-07b	0.95 (0.91–0.990)	1.19E-02c
rs10866498	6.36E-08b	1.05 (1.01–1.09)	6.76E-03	0.91 (0.88–0.94)	4.57E-07b	Colorectal, Prostate	Ovary, Lung	9.27E-18b	1.07 (1.03–1.10)	4.01E-05	0.88 (0.85–0.91)	5.25E-15b
rs12655062	1.65E-06b	1.12 (1.06–1.18)	3.53E-05	0.95 (0.92–0.98)	2.72E-03	Prostate	Colorectal, Ovarian	1.13E-06b	1.12 (1.06–1.18)	2.67E-05	0.95 (0.92–0.98)	2.42E-03
rs115960372	6.94E-07b	1.19 (1.10–1.27)	2.97E-06b	0.90 (0.83–0.98)	1.29E-02c	Prostate	Lung	3.12E-06b	1.17 (1.09–1.26)	1.20E-05b	0.91 (0.84–0.98)	1.57E-02c
rs2736098e	2.48E-07b	1.08 (1.04–1.12)	1.81E-04	0.93 (0.90–0.97)	7.16E-05	Prostate, Lung	Colorectal, Breast, Ovarian	5.36E-06b	1.08 (1.02–1.14)	1.23E-02c	0.93 (0.91–0.96)	2.74E-05
POT1 (Chr. 7)
rs116895242	5.21E-05			0.83 (0.77–0.90)	6.99E-06b		Colorectal, Ovarian, Lung
RTEL1 (Chr. 20)
rs34978822f	2.14E-05			0.71 (0.62–0.82)	3.17E-06b		Prostate, Lung

Abbreviations: Chr.- chromosome; CI- confidence interval; OR- odds ratio; SNP- single nucleotide polymorphism.

^aThe most significant SNP is always conditioned on in the sequential conditional analysis (and therefore there are no conditional results for it)

^bGene level P-value thresholds based on the number of effective tests are: DCLER1B P<2.65×10−5; TERC P<2.45×10−5; TERT-CLPTM1 P<1.32×10−5; POT1 P-value<2.94×10−5; RTEL1 P-value<1.86×10−5.

^cPositive or inverse associations with P-values between 0.01 and 0.05 are considered to be suggestive.

^dAssociations with phenotypes were statistically significant in conditional analyses only

^eSNPs that are directly measured and not imputed.

^fASSET meta-analytical results for these SNPs are based on 4 cancer types rather than all 5 studies.

In TERC, three independent loci were identified (Table 3). We observed highly significant inverse associations with prostate cancer risk for the A allele of rs80304993 (P=1.51×10⁻¹⁵), and the T allele of rs62293480, particularly after conditioning on rs80304993 (P_conditional=1.44×10⁻¹⁴). Forest plots by cancer type and subtype (Figures 1A and 1B) show that these inverse associations were driven solely by overall prostate cancer (OR=0.82, 95% CI=0.78–0.86). In our sequential analysis, the next SNP (ranked by unconditional P-value) to add to the model conditioning on both rs80304993 and rs62293480 was rs4420873, but it was excluded due to collinearity (r²>0.80). Therefore, the next SNP, rs75316749, was evaluated in the model conditioning on rs80304993 and rs62293480, and had a combined conditional P-value of 1.46×10⁻⁶. Unlike the two other SNPs in TERC, rs75316749 was not associated with prostate cancer. The G allele was positively associated in conditional and unconditional analyses with colorectal, breast, ovarian, and lung cancers (Table 3), driven specifically by ER-negative breast cancer, and lung adenocarcinoma and squamous cancers, but not lung cancer overall (P=2.9×10⁻⁴; OR=1.17 95%CI=1.07–1.27; Figure 1C). While rs75316749 has been reported to result in a motif change and enhancer histone changes in breast and fat cell lines, SNPs in very high LD including rs115002293 and rs75963875 are associated with enhancer histone changes in a wide variety of cell lines, including breast and lung fibroblast cells (30).

Figure 1. Unconditional ASSET forest plots by cancer type and subtype for TERC SNPs rs80304993, rs62293480, and rs75316749.

(A) Forest plot associations for the A allele for rs80304993. (B) Forest plot associations for the T allele for rs62293480. (C) Forest plot associations for the G allele for rs75316749.

In the TERT-CLPTM1L region, six independent loci were identified (Table 3). The T allele of the SNP with the lowest P-value, rs37004, was inversely associated with lung cancer overall and specifically lung adenocarcinoma (P=2.2×10⁻¹¹; OR=0.83 95%CI=0.79–0.88; Figure 2A). After conditioning on rs37004, rs7717443 had the lowest combined P-value (P_conditional=1.26×10⁻⁷). The T allele was associated with increased ovarian and lung cancer risks, and suggestive decreased colorectal and prostate cancer risks in conditional and unconditional analyses. The unconditional ASSET forest plot by cancer type and subtype for rs7717443 (Figure 2B) illustrates that the positive associations apply to serous and endometrioid ovarian cancer subtypes (but not overall ovarian cancer) and lung adenocarcinoma only (P=2.0×10⁻⁸; OR=1.20 95%CI=1.13–1.28), and inverse associations are for overall colorectal and prostate cancers, and aggressive prostate cancer (P=3.3×10⁻²; OR=0.94 95%CI=0.89–1.00). Next, after conditioning on both rs37004 and rs7717443, the combined P-value for rs10866498 was highly significant (9.27×10⁻¹⁸). In conditional and unconditional analyses, the T allele of rs10866498 was associated positively with colorectal and prostate cancers, and inversely with ovarian and lung cancers (Table 3). Associations with colorectal cancer, and with both overall and aggressive prostate cancers (P=0.01; OR=1.06 95%CI=1.01–1.10) were positive, and inverse associations were observed with overall lung cancer and lung adenocarcinoma, and serous ovarian cancer (P=4.6×10⁻⁷; OR=0.88 95%CI=0.83–0.92; Figure 2C). After conditioning on the top 3 TERT-CLPTM1L SNPs, rs12655062 was the next most significant (P_conditional=1.13×10⁻⁶). In both unconditional and conditional analyses, the rs12655062 A allele was associated positively with prostate and inversely with colorectal and ovarian cancers (Table 3). Positive associations were driven by both overall prostate cancer and aggressive prostate cancer (P=1.7×10⁻⁴; OR=1.14 95%CI=1.06–1.21), and inverse associations by overall colorectal cancer and endometrioid and serous ovarian cancers (P=4.1×10⁻²; OR=0.95 95%CI=0.90–1.00; Figure 2D). The rs12655062 A allele is associated with reduced expression of IRX4 and CTD02194D22.3 in prostate tissue (31), alters six motifs, and results in enhancer histone changes in breast and gastrointestinal cell lines (30). The next most significant SNP in sequential analysis after conditioning on the top four TERT-CLPTM1L SNPs was rs115960372 (P_conditional=3.12×10⁻⁶). The T allele was associated positively with prostate cancer and suggestively inversely associated with lung cancer in conditional and unconditional analyses (Table 3). This SNP results in changes to chromatin structure in several cell lines (including fetal, adult, and carcinoma lung cell lines) and two motif changes (30). The unconditional ASSET forest plot by cancer type and subtype revealed that positive associations were driven by overall prostate cancer (OR=1.19 95%CI=1.09–1.29); however, inverse associations were not significant in cancer subtype analyses (P=7.5×10⁻²; Figure 2E). The last significant SNP identified from sequential conditional analysis after conditioning on the above five TERT-CLPTM1L variants was rs2736098 (P_conditional=5.36×10⁻⁶). The T allele was suggestively positively associated with prostate and lung cancers, and inversely associated with colorectal, breast, and ovarian cancers, in both conditional and unconditional analyses (Table 3). Positive associations were driven not only by overall prostate and lung cancers, but also lung adenocarcinoma (P=1.9×10⁻³; OR=1.09 95%CI=1.03–1.16; Figure 2F).

Figure 2. Unconditional ASSET forest plots by cancer type and subtype for TERT-CLPTM1L SNPs rs37004, rs7717443, rs10866498, rs12655062, rs115960372, and rs2736098.

(A) Forest plot associations for the T allele for rs37004. (B) Forest plot associations for the T allele for rs7717443. (C) Forest plot associations for the T allele for rs10866498. (D) Forest plot associations for the A allele for rs12655062. (E) Forest plot associations for the T allele for rs115960372. (F) Forest plot associations for the T allele for rs2736098.

In DCLRE1B, rs974404 had the lowest combined P-value (P=9.19×10⁻⁶). The G allele was inversely associated with prostate and lung cancers, and suggestively positively associated with breast and ovarian cancers. The inverse associations were driven by overall prostate cancer and lung adenocarcinoma (P=1.3×10⁻³; OR=0.93 95%CI=0.88–0.97), but not by overall lung cancer, or squamous cell lung cancer; positive associations were no longer significant in analyses by cancer subtype (P=0.23) (Figure 3A). Considerable evidence supports that rs974404 and correlated SNPs alter gene function. rs974404 results in 27 altered motifs (30), and twelve SNPs in LD with rs974404 are associated with increased expression of DCLRE1B in whole blood (30).

Figure 3. Unconditional ASSET forest plots by cancer type and subtype for DCLRE1B SNPs rs974404 and rs12144215.

(A) Forest plot associations for the G allele for rs974404. (B) Forest plot associations for the T allele for rs12144215.

After conditioning on rs974404, the most significant SNP in DCLRE1B was rs12144215 (P_{unconditional}=1.50×10⁻⁵; P_conditional=2.07×10⁻⁵). In unconditional analyses, the rs12144215 T allele was inversely associated with colorectal and prostate cancers, and after conditioning on rs974404, a suggestive positive association with lung cancer and a significant inverse association with breast cancer were additionally observed (Table 3). The unconditional inverse association was driven by overall colorectal and prostate cancers and the ovarian cancer endometrioid subtype (P=1.1×10⁻⁵; OR=0.90 95%CI=0.86–0.94; Figure 3B); the positive association with lung cancer observed in conditional analyses was no longer observed in unconditional analyses by cancer subtype (P=1.00). Several SNPs in LD with rs12144215 change chromatin structure in multiple cell lines, including mammary epithelial and lung (30).

Discussion

Our conditional subset-based meta-analysis of GWAS data from five different cancer types identified 13 independent SNPs in DCLRE1B, TERC, TERT-CLPTM1L, RTEL and POT1 gene regions that are associated with risk of multiple cancers. Across the DCLRE1B region, we identified two novel loci: rs974404, which is associated with increased DCLRE1B expression (30) and was associated with prostate and lung cancer risk, and rs12144215, which may be associated with chromatin structure alterations and was associated with colorectal, breast, and ovarian cancers risk. While the observed associations between two SNPs near the TERC gene, rs80304993 and rs62293480, and prostate cancer risk have been reported in GWAS previously (32), we show that the association between rs62293480 and prostate cancer is much more significant after conditioning on rs80304993 (P_{unconditional}=1.35x⁻⁰⁶, P_conditional=2.16x⁻¹³). We also report a novel finding in the TERC region; after conditioning on both rs80304993 and rs62293480, rs75316749 was associated with colorectal, breast, ovarian, and lung cancer risk. There is some evidence that this SNP and/or others in LD with it result in enhancer histone changes (30). Across the TERT-CLPTML1 regions, we detected six susceptibility loci where strong associations with lung and/or prostate cancer risk were generally observed. We report similar associations previously observed in GWAS for four TERT-CLPTM1L SNPs and lung and prostate cancer (7,9), but observe novel findings for two SNPs, rs12655062 and rs115960372. The rs12655062 variant is associated with reduced expression of the gene IRX4 in prostate tissue, and rs115960372 may alter chromatin structure in multiple tissue types (30). Our study demonstrated that for rs10866498, after controlling for the top two hits in TERT-CLPTM1L, the p-value for the inverse association with lung and ovarian cancer was even more significant (P_{unconditional}=6.36×10⁻⁸, P_conditional=9.27×10⁻¹⁸). We also observed associations between rs116895242 in the POT1 region and colorectal, ovarian and lung cancer risk, and between rs34978822 in RTEL1 and prostate and lung cancer. There is limited evidence to support that these SNPs alter gene function (30).

DCLRE1B plays an important role in protecting telomeres by interacting with the shelterin complex to suppress DNA damage-sensing machinery during and after replication (20,33). The SNPs that we observed to be associated with risk of prostate and lung cancers (rs974404 in PTPN22), and colorectal, breast, and ovarian cancers (rs12144215 in MAGI3), have been previously associated in GWAS with rheumatoid arthritis and Grave’s disease, respectively (34,35). To date, only one SNP in the DCLRE1B gene, rs11552449, has been shown to be associated with breast cancer risk in a meta-analysis of nine GWAS and 41 studies in the Breast Cancer Association Consortium (P-value=1.8×10⁻⁸) (16). However, this SNP did not reach gene-level statistical significance in our analyses.

TERC is essential for telomerase expression because it encodes the RNA component of telomerase required for elongation of telomeric repeats (1,20). Variants in the 3q26 TERC region have been associated with risk of several different cancers in GWAS, including melanoma, glioma, bladder, colorectal, nasopharyngeal, chronic lymphatic leukemia, and multiple myeloma (36–42). In a GWAS of >25,000 prostate cancer cases and controls, Kote-Jarai et al. reported that rs10936632 was associated with a 10% decrease in prostate cancer risk (P-value 1.0×10⁻¹³) (32). In our unconditional ASSET analysis we also observed that rs10936632, which is in high LD (r²=0.97) with rs55953261, was significantly associated with reduced prostate cancer risk. It should be noted that 27% of prostate cancer cases and 26% of controls in Kote-Jarai et al. (32) were also included in our investigation.

Our additional TERC findings for rs80304993 and rs62293480 and prostate cancer risk have been observed previously in a multi-ethnic meta-analysis of GWAS (43). These SNPS are located in the SKIL gene, which regulates cell growth and differentiation (20). Our study findings for SNP rs75316749 and colorectal, breast, ovarian, and lung cancer risk are novel. SNP rs75316749 lies approximately 40kb 3′ of the MECOM gene which encodes a protein involved in hematopoiesis, apoptosis, development, and cell differentiation and proliferation (20).

The TERT gene, at 5p15.33, encodes the catalytic subunit of telomerase (1,20,33) and thus plays a vital role in maintaining telomerase expression and facilitating elongation of telomeric repeats. The 5′-end of TERT adjoins CLPTM1L, which is overexpressed in lung and pancreatic cancers (9,44,45). There is extensive LD between the two genes, and susceptibility loci in this combined gene region have been associated with multiple cancer types (7–9,46–48). The most commonly associated risk variants in the TERT-CLPTM1L regions are rs2736100 and rs401681, respectively. In GWAS, rs2736100 has been associated with lung, glioma, and testicular cancer risk (9,45) while rs401681 has been linked to lung, bladder, pancreas, prostate, and skin cancer risk (9,45). Our unconditional ASSET analyses corroborate the associations observed between these variants and lung cancer risk.

Mocellin et al. performed a systematic review of TERT-CLPTM1L polymorphisms and cancer risk in 85 studies including 27 GWAS of predominantly individuals of European ancestry (87%) (9). Of the 67 SNPs and 24 tumor types examined, statistically significant associations were reported for 22 SNPs, 19 of which were linked to lung cancer. In our investigation, unconditional ASSET analysis confirmed associations with lung cancer for 13 of these at our gene level cutoff (P-value <1.32×10⁻⁵) and for four more at P-value<0.05. Of particular interest from Mocellin et al.’s study was the highly significant association reported between rs2736098 and lung (4 studies, P-value=2.2×10⁻¹³) and bladder (3 studies, P-value=8.6×10⁻¹⁰) cancer risk and the association between rs451360 and lung cancer risk (2 studies, P-value=4×10⁻³) (9). Our findings are in agreement with these observations. In our analysis, rs37004, 2kb 5′ of CLPTM1L and in high LD (r²=0.89) with rs451360, was the SNP in the TERT-CLPTM1L region with the lowest P-value, due entirely to its association with lung cancer risk; rs2736098, located within TERT, was associated with lung cancer risk as well as prostate, colorectal, breast, and ovarian cancer. However, we did not observe the lung cancer association reported by Mocellin et al. for rs1801075 and we could not evaluate the association with rs4246742 because no data on lung cancer were available for this SNP.

Similar results were reported for these variants by Wang et al., who utilized the same ASSET meta-analytic approach that we used to examine common susceptibility alleles in TERT-CLPTM1L across six cancer types (lung, prostate, pancreatic, testicular, glioma, and bladder), in 34,248 cases and 45,036 controls (7). A large proportion of prostate (60.3%) and lung (46.9%) cancer cases from that study were also included in our investigation. Using sequential conditional ASSET analyses, Wang et al. identified five independent risk loci in individuals of European ancestry. These loci are included in the LD plot of our significant unconditional ASSET two-sided SNPs retained following LD pruning at r²>0.70 (Supplementary Figure 2). In one region, rs13170453 was associated positively with pancreatic and testicular cancer (P-value=4.38×10⁻¹³) and inversely with lung cancer risk (P-value=9.5×10⁻⁸). Our conditional ASSET findings for rs37004, which is in high LD (r²=0.88) with rs13170453, confirm the lung cancer association observed by Wang et al. for this SNP. In a second region, Wang et al. observed that rs2736098 was associated positively with lung, prostate, and bladder cancer (P-value=2.58×10⁻⁸) and inversely with pancreatic and testicular cancer (P-value=4.89×10⁻⁶). In our investigation, rs2736098, located within TERT, was similarly positively associated in conditional analyses with lung and prostate cancer, but inversely with colorectal, breast, and ovarian cancer. In a third region, Wang et al. reported rs4449583 as being associated positively with glioma, and inversely with testicular, prostate, and pancreatic cancers. In our unconditional ASSET analysis, this SNP was associated positively with ovarian and lung cancer, and inversely with prostate cancer. Cancer associations for the remaining two TERT-CLPTM1L regions including rs10069690 and rs13172201 in Wang’s study were not replicated in our investigation. Associations for these regions were also not confirmed in supplementary analyses conducted by Wang et al. across additional cancer types (esophageal, gastric, breast, endometrial, prostate, osteosarcoma, ovarian, renal, and additional prostate cancers) including 11,385 cases and 18,322 controls (7). We examined a larger region around TERT-CLPTM1L than did Wang et al. (chr5: 1,250,000–1,450,000), thus they did not assess rs12655062 and rs115960372 which lie outside of that region. Our significant conditional ASSET associations between TERT SNPs rs7717443 and rs10866498 and colorectal, prostate, ovarian, and lung cancer risk, which are not highly correlated with variants observed in Wang et al. have not been previously reported. In summary, our study confirms the Wang et al. findings for three of the five significant TERT-CLPTM1L SNPs that they reported among European subjects (conditionally for rs2736098 and rs37004 (r²=0.88 with rs13170453), and unconditionally for rs4449583); however, our study did not corroborate their findings for rs10069690 or rs13172201. Additionally, in our study, of the six conditionally significant TERT-CLPTM1L risk loci detected among European subjects, Wang et al. did not report findings for SNPs rs7717443 and rs10866498 nor did they examine associations for SNPs rs12655062 and rs115960372 which lied outside of the regions that they evaluated. Nonetheless, a Japanese fine-mapping study of 1,583 prostate cancer cases and 2,480 controls reported a highly significant association with rs115960372, (OR=1.31, P-value=7.76×10⁻¹⁰) which is in the LPCAT1 gene (49). The association between rs12655062, which is in the CTD-2194D22.4 gene, and colorectal, breast, and prostate cancer risk has not been previously reported.

Because associations with cancer risk may vary by histology, some studies have assessed SNPs across TERT-CLPTMIL in relation to cancer subtypes. Of particular interest were the lung cancer ASSET meta-analytic results reported by Wang et al. (50). Based on data from five GWAS, highly significant associations were reported between TERT-CLPTMIL rs7717443 (OR=1.24, P-value=4.90×10⁻¹⁰), rs10866498 (OR=0.81, P-value=3.28×10⁻¹¹) and rs37004 (OR=0.78, P-value=2.52×10⁻¹²) and lung adenocarcinoma risk; an association between rs37004 (OR=0.82, P-value=7.94×10⁻⁸) and squamous lung cancer risk was also observed. In on our unconditional ASSET forest plot analysis, we observed similar associations between these variants and lung adenocarcinoma, but not with squamous lung cancer.

The associations reported here for variants in the GAR1, TERF2, POT1 and RTEL1 gene regions and colorectal, lung, breast, ovarian, and/or prostate cancers have not been reported in GWAS of these cancers. We advise caution in interpreting results for GAR1 and TERF2 variants with low MAF (0.3%). Although these SNPs passed imputation accuracy cutoffs in some consortium-specific meta-analyses, SNPs with such low MAFs are known to be difficult to impute accurately. The RTEL1 gene at 20p13.3 encodes a DNA helicase involved in stabilization, protection and elongation of telomeres (9,20). This gene interacts with shelterin complex proteins and variants in this gene have been associated in previous GWAS with high-grade glioma risk (37). POT1 at 7q31.33 and TERF2 at 16q22.1 are protein-coding genes that are components of the shelterin complex (20,33). Variants in POT1 have been previously associated with risk of chronic lymphocytic leukemia in GWAS (41).

To our knowledge, this is the largest meta-analysis of GWAS data on telomere structure and maintenance genes and cancer risk. With over 61,000 cancer cases and nearly 75,000 controls, our study is highly powered to detect significant associations for variants with common allele frequencies. Our study is unique in that we evaluated risk of multiple cancer types as well as risk of specific histologic or molecular subtypes of cancer and subtypes related to aggressiveness. Our subset-based meta-analysis also permitted us to examine the magnitude and direction of genetic associations allowing for heterogeneity of associations across cancer sites. Compared to traditional methods, ASSET helps minimize false-positives through multiple testing penalties and improves detection power (21). We were able to determine independent associations between SNPs and cancer types by conditioning on the effects of SNPs with lower P-values. Because there is considerable evidence linking TERT-CLPTM1L variants to risk of many different cancer types, and several other telomere structure and maintenance genes have been implicated in GWAS of various cancer types, we used gene-level P-value thresholds to define statistical significance. Although we were able to interrogate a very large number of SNPs in telomere structure and maintenance genes, we did not assess SNPs across all known telomere structure and maintenance genes, and most of the SNPs (97.5%) we examined were imputed. Our study was not well-designed to examine imputed rare variants since these SNPs may be poorly represented or poorly tagged on genotyping arrays. While we were able to use the available aggregate data to evaluate whether variation in all SNPs combined in each gene was associated with risk, we could not evaluate haplotypes.

In summary, our results indicate that patterns of association in telomere structure and maintenance genes observed across cancer types and subtypes are complex. These findings may provide insight into the mechanisms through which telomere dysfunction in different tissues influences cancer risk. In our investigation, seven of the thirteen conditional associations identified were novel. While we observed suggestive pleiotropic associations within the DCLRE1B, TERC, TERT-CLPTM1L, POT1 and RTEL1 gene regions, fine-mapping studies with the ability to assess haplotypes are needed to evaluate the relationship between alleles at different loci in order to help identify potential variants that may have gone undetected. Replication and mechanistic studies are also needed to help provide insight regarding the function and variability of risk across cancers for these telomere structure and maintenance SNPs.

Supplementary Material

Supp Fig S1

Supplementary Figure 1. Manhattan plot of associations between 204,993 SNPs in telomere structure and maintenance genes and five cancer types.

The dotted line indicates the genome-wide significance level (P-value=5×10⁻⁸), and the individual solid lines indicate the gene-level P-value cutoffs taking into account the number of effective tests. Genes are evaluated in the order listed in the key.

Supp Fig S2

Supplementary Figure 2. TERT-CLPTM1L and TERC LD plots for statistically significant ASSET variants that were retained following LD pruning at r²>0.70.

Associations between alleles across variants are based on CGEMS and EAGLE datasets. SNPs circled in black are variants identified as statistically significant in ASSET conditional analyses. ^aSNPs identified as statistically significant in ASSET conditional analyses in the Wang et al. study among individuals of European ancestry [7]. A) LD plot for TERT-CLPTM1L variants. B) LD plot for TERC variant.

Novelty & Impact Statements.

Utilizing the novel ASSociation analysis based on SubSET (ASSET) meta-analytic approach, we examined associations between >200,000 variants in 22 telomere structure and maintenance gene regions and colorectal, breast, prostate, ovarian, and lung cancer risk. We observed pleiotropic associations across cancer types in the DCLRE1B, TERC, TERT-CLPTM1L, POT1, and RTEL1 gene regions. Additional studies clarifying the mechanisms through which these complex association patterns in telomere-related genes influence cancer risk are needed.

Abbreviations used

ASSET

ASSociation analysis based on SubSET

CGEMS

Cancer Genetic Markers of Susceptibility Project

CI

confidence interval

EAGLE

Environmental and Genetics in Lung Cancer Etiology study

ER

estrogen receptor

GAME-ON

Genetic Associations and Mechanisms in Oncology Network

GEC

Genetic type 1 Error Calculator

GECCO

Genetic and Epidemiology of Colorectal Cancer Consortium

GWAS

genome-wide association studies

LD

linkage disequilibrium

MAF

minor allele frequency

M_e

effective number of independent tests

OR

odds ratio

SNPs

single nucleotide polymorphisms