Genetic Variants in Epigenetic Pathways and Risks of Multiple Cancers in the GAME-ON Consortium

Abstract

Background

Epigenetic disturbances are crucial in cancer initiation, potentially with pleiotropic effects, and may be influenced by the genetic background.

Methods

In a subsets (ASSET) meta-analytic approach, we investigated associations of genetic variants related to epigenetic mechanisms with risks of breast, lung, colorectal, ovarian and prostate carcinomas using 51,724 cases and 52,001 controls. False-discovery-rate corrected p-values (q-values < 0.05) were considered statistically significant.

Results

Among 162,887 imputed or genotyped variants in 555 candidate genes, SNPs in eight genes were associated with risk of more than one cancer type. For example, variants in BABAM1 were confirmed as a susceptibility locus for squamous cell lung, overall breast, ER-negative breast, overall prostate, overall and serous ovarian cancer; the most significant variant was rs4808076 (odds ratio (OR)=1.14, 95% confidence interval (CI)=1.10–1.19, q=6.87*10⁻⁵). DPF1 rs12611084 was inversely associated with ER-negative breast, endometrioid ovarian, overall and aggressive prostate cancer risk (OR=0.93, 95% CI=0.91–0.96, q=0.005). Variants in L3MBTL3 were associated with colorectal, overall breast, estrogen receptor (ER)-negative breast, clear cell ovarian, and overall and aggressive prostate cancer risk (e.g. rs9388766: OR=1.06, 95% CI=1.03–1.08, q= 0.02). Variants in TET2 were significantly associated with overall breast, overall prostate, overall ovarian and endometrioid ovarian cancer risk, rs62331150 showing bidirectional effects. Analyses of sub-pathways did not reveal gene subsets that contributed disproportionately to susceptibility.

Conclusion

Functional and correlative studies are now needed to elucidate the potential links between germline genotype, epigenetic function, and cancer etiology.

Impact

This approach provides novel insight into possible pleiotropic effects of genes involved in epigenetic processes.

INTRODUCTION

Genetic and epigenetic alterations are hallmarks of cancer initiation and progression and can influence each other to work cooperatively (1). Dysfunction of epigenetic processes, such as DNA methylation, chromatin remodeling and covalent histone modifications can be as important in carcinogenesis as the change of the genetic material itself (2). Since the first studies that described the global hypomethylation of cancer genomes and the hypermethylation of the promoter sequence of mainly tumor suppressor genes, several “pan-cancer” DNA methylation patterns (patterns across multiple cancer types) have been identified (reviewed in (3)). The CpG island methylator phenotype (CIMP) was first described in colorectal cancer (4) and later similar patterns were observed in several other tumor types. Highlighting the interplay between genetic and epigenetic changes, CIMP subtypes usually present with characteristic genetic alterations. CIMP-H colorectal cancers are frequently characterized by BRAF mutations, while CIMP-L tumors tend to harbor KRAS mutations (5). Non-CIMP colorectal cancer, the B-CIMP-negative breast cancer and the low methylated tumor group of serous ovarian cancers frequently acquire TP53 mutations (5–8).

Furthermore, somatic mutations in epigenetic regulatory genes that are either carcinogenic driver or passenger mutations are known to exist. Important mutations have been shown for example in DNMTs, IDH1, IDH2 and TETs (as important players of DNA methylation), in EZH2 and KDM1A (involved in histone modifications) and in ARID1A (participant of chromatin remodeling) (reviewed in (2)). In addition, inherited genetic variants related to epigenetic regulatory processes were described in association with multiple cancers (9, 10). Given the fundamentality of epigenetic processes, germline variants in genes related to epigenetic pathways presumably have pleiotropic effects on the initiation of different cancers.

As part of the U.S. National Cancer Institute’s Genetic Associations and Mechanisms in Oncology (GAME-ON) Network (http://epi.grants.cancer.gov/gameon/), we have previously shown the value of cross-cancer analyses in inflammation pathways (11).

An additional value is that our datasets include large numbers of cancer subtypes that were not studied in The Cancer Genome Atlas (TCGA). The present study was focused and approved by the GAME-ON consortium for the overall analyses of pleiotropy, where we aimed to identify cross-cancer associations of epigenetically related polymorphisms that advance our understanding of the role of epigenetics in cancer development. Given the central role of epigenetic processes in carcinogenesis, germline variants in genes related to epigenetic pathways show pleiotropic effects on the initiation of different cancers. Consequently, we investigated whether common polymorphisms in epigenetic genes are associated with risk of multiple cancer types (breast, colorectal, lung, ovarian and prostate cancer) and their subtypes.

METHODS

Study population

Within the GAME-ON Network, 32 studies from North America and Europe participated in this investigation (12–21). Studies included frequency matched cases and controls on at least age, and all subjects were of European descent based on ancestry analyses. The study characteristics are summarized in Table 1. In total, 51,724 cancer patients (breast, colorectal, lung, ovarian and prostate with respective subtypes) and 52,001 controls were included in the analysis.

Table 1.

Overview of cancer types and studies participating in the original meta-analyses

Cancer site Studies	Genotyping platforms	Subtype	Covariates	No. of Studies	N cases	N controls
Colorectal MECC, CFRa, KY, ACS, Australia, NF	Affymetrix Axiom	All	Age, sex, PCs	6	5100	4831
Breast ABCFS, HEBCS, UK2, SASBAC, MARIE, CPSII, EPIC, MEC, NHS2, PBCS, PLCO	Illumina arrays (317K-1.2M)	All	Age, PCs (vary by studies)	11	15569	18204
		ESR1 (ER)-negative	Age, PCs (vary by studies)	8	4760	13248
Lung UK, MDACC, IARC, NCI, SLRI, HGF	Illumina arrays (317K-610K)	All	Age, sex, PCs	6	12527	17285
		Adenocarcinoma	Age, sex, PCs	6	3804	16289
		Squamous cell	Age, sex, PCs	6	3546	16434
Ovary UKGWAS, USGWAS, U19GWAS	Illumina arrays (317K-2.5M)	All	Study, PCs	3	4368	9123
		Endometrioid	Study, PCs	3	2553	9123
		Serous	Study, PCs	3	715	9123
		Clear cell	Study, PCs	3	355	9123
Prostate BPC3, CRUK1, CRUK2, CAPS	Illumina and Affymetrix arrays	All	Age, study	6	14160	12712
		Aggressive subtype	Age, study, PCs	6	4446	12724

^aColon-CFR: 1660 cases, 1393 controls

ER = estrogen receptor; PCs = principal components representing residual European ancestry.

Gene and variant selection, pathway assignment

Genes (n=634) involved in epigenetic processes were identified using GO and GeneCards databases by searching for the following keywords: DNA methylation, DNA demethylation, histone acetylation, deacetylation, methylation, demethylation, and other histone modification, chromatin remodeling, chromatin modification and histones. The recent literature was also reviewed. After excluding genes on sex chromosomes and those not covered in all cancer sites, 555 genes were included in the analysis, which were categorized into one or more of epigenetic sub-pathways (Supplementary Table S1).

We analyzed all single nucleotide polymorphisms (SNPs) residing within 50 kb of the largest transcript for each gene (Databases see in Supplementary Table S2). Overall, 162,887 polymorphisms were included in the final analysis. In the combined dataset, the major alleles (according dbSNP) were used as reference alleles.

Statistical analysis

Cancer sites were further divided into subtypes and for each cancer type and subtype, a fixed effect meta-analysis was conducted to combine results from individual studies (Table 1). This method used log-additive models adjusted for age, European principal components, and sex (where appropriate).

The beta values and standard errors for each cancer or cancer subtype were then combined using the association analysis based on a subsets (ASSET) meta-analytic approach, which allows for disease heterogeneity and potential opposite directions of the same genetic variant on different cancer types (22). It searches for the most parsimonious grouping based on the test statistics using any of the five cancers or cancer subtypes simultaneously as the outcome variables. Overlapping subjects amongst cancer subtypes (e.g. overlapping cases and controls between overall lung cancer and its subtypes) and across cancer types (e.g. UK ovary and UK breast GWAS both used controls from Welcome Trust Case Control Consortium, WTCCC) were accounted for in the covariance matrix when estimating the standard errors (11). The resulting p-values were adjusted using false-discovery rate (FDR) correction. Results with FDR q<0.05 were considered statistically significant (Supplementary Table S3). All association analyses were performed in R (3.2.5).

Functional annotation

The overall approach of the functional annotation is summarized in Supplementary Fig. S1. For each gene with more than five significant SNPs (FDR q<0.05 in the ASSET meta-analysis), we selected tagSNPs to represent these regions in subsequent analysis. Specifically, a linkage disequilibrium (LD) map was prepared using the Haploview 4.2 software and tagSNPs were identified with the tagger algorithm of Haploview using 1000Genomes data (release 20130502). Variants with more than two alleles based on 1000Genomes were excluded from LD mapping. As a result we were able to investigate SNPs that were not covered in the original meta-analysis but potentially have functional effect on the genes in the region of interest.

To assess if any of the epigenetic sub-pathways shown in Supplementary Table S1 were enriched with genes containing significant associations with cancer types or subtypes, pathway analyses were conducted using the ALIGATOR algorithm of the SNPath R package.

The possible functional annotation of the tagSNPs and the region-representative SNPs (functional follow-up (FFU) SNPs) were then assigned using the FunciSNP R/Bioconductor package (23). Using the package, we identified all the corresponding SNPs of our tagSNPs using 50 kb searching window and r²>=0.8 as a linkage threshold. In the next step, FunciSNP package checks if the corresponding SNPs or the tagSNPs show overlap with DNA segments with predicted functional importance. To annotate these biofeatures, we used the combined genome segmentation assessed by the ENCODE Project Consortium. These results represent ChIP-seq data for eight chromatin marks (H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K27ac, H3K27me3, H3K36me3, H4K20me1), RNA Polymerase II and the CTCF transcription factor, as well as DNase-seq and FAIRE-seq data. This data is processed with ChromHMM and Segway software which segments the genome into seven disjoint segments based on their predicted functional role (24). Since the goal of the study was to identify those polymorphisms that change the function of the epigenetic related genes, we interpreted polymorphisms that overlap with a predicted transcribed region only, if they were in the gene of interest. We used the data available on Huvec, H1hesc and Gm12878 cell lines. Unfortunately, comprehensive information for the genome segmentation track was not available for all cell lines of the respective cancer types. We thus decided to use data from normal cell lines. Additionally, an ENCODE Uniform transcription factor binding site (TFBS) track was used, that encompasses data for 161 transcription factors from 91 cell types. Supplementary Table S4 summarizes the functional annotation of all SNPs based on FunciSNP (SNPs that were annotated as not functional are not listed). Furthermore, the functionality of the ASSET-identified SNPs as well as their corresponding SNPs were annotated using RegulomeDB, version 1.1.

All software packages and databases that were used are listed in Supplementary Table S2.

RESULTS

The results of the original (individual study based) meta-analyses and the ASSET-based risk associations are summarized in Figure 1. Ovarian cancer was associated with the largest number of variants (98), followed by prostate (70), lung (50), breast (46) and colorectal (10) cancers. Interestingly, all of the endometrioid ovarian cancer specific SNPs also showed an association with overall prostate cancer risk. These polymorphisms were mainly located in the RUVBL1 gene regions. Variants in the flanking region of MORF4L1 on 15q25 were mainly associated with lung and ovarian cancer. Due to the proximity of MORF4L1 to CHRNA5, CHRNA3 and CHRNB4 genes and their well-known association with lung cancer, we excluded this region from further analysis (25–27). The number of remaining SNPs that were associated with lung cancer risk was 35 and with ovarian cancer was 83. Furthermore, variants in PHC3 (3q26) were solely associated with risk of overall prostate cancer and will not further be discussed.

Figure 1.

Manhattan plot showing the original meta-analyses (A) and the results of the ASSET-based meta-analysis (C) on the selected SNPs available for all studies. Variants with –log₁₀ (p values) higher than 20 are not shown. Regions showing significant pleiotropic association in the ASSET analysis are marked in green.

Pie charts (B) show the number of variants that were significant in the ASSET analysis. Numbers in brackets depict the number of independent risk loci. Each diagram represents a gene region and the numbers of SNPs associated with a specific cancer type (in the same colors as indicated in the Manhattan plot (A)) are shown. SNPs associated with multiple cancer types are counted in each of the respective cancer sections. Overlap is not visualized.

When combining genes into epigenetic sub-pathways (see above), we observed no significant risk association with more than one cancer type or subtype (p values>0.05) indicating that all pathways were similarly important for cancer risk.

Overall 99 SNPs in 8 genes (excluding MORF4L1: 84 SNPs in 7 genes) showed significant associations (FDR q<0.05) with risk of more than one cancer type (Supplementary Fig. S2. A and B). Genes with associated SNPs were: RUVBL1 (3q21), TET2 (4q24), L3MBTL3 (6q23), HDAC9 (7p21), BRCA2 (13q12), MORF4L1 (15q25), BABAM1 (19p13) and DPF1 (19q13) (Table 2, Supplementary Fig. S3). Previous GWAS-identified cancer risk associations in these and other genes located in these regions are listed in Supplementary Table S5.

Table 2.

Summary of gene regions significantly associated with more than one cancer.

Candidate genes	Select. region	Nr. of SNPs/region	Nr. of associated SNPsa	Associated cancers	Strongest association	Previously in GWAS associated cancersb
RUVBL1	3: 127733628-127922757	346	27 (1)	Endometrioid ovarian cancer, overall prostate cancer, colorectal cancer	Endometrioid ovarian cancer, Overall prostate cancer: rs144609957 - OR: 1.13; 95% CI: 1.08–1.19; p: 3.44*10−7	Prostate cancer
TET2	4: 106017842-106250960	573	9 (3)	Overall prostate cancer, overall ovarian cancer, endometrioid ovarian cancer, overall breast cancer, clear cell ovarian cancer, colorectal cancer	Overall prostate cancer, endometrioid ovarian cancer: rs6839705 - OR: 1.11; 95% CI: 1.07–1.16; p: 3.32*10−7	Prostate cancer, breast cancer
L3MBTL3	6: 130289728-130512594	590	11 (2)	Colorectal cancer, overall breast cancer, ER-neg breast cancer, clear cell ovarian cancer, overall prostate cancer, aggressive prostate cancer	Colorectal cancer, overall breast cancer, ESR1 (ER)-negative breast cancer, clear cell ovarian cancer, overall prostate cancer, aggressive prostate cancer: rs9388766 - OR 1.06; 95% CI: 1.03–1.08; p: 1.07*10−6	None
HDAC9	7: 18076572-18758466	1307	1 (1)	Lung adenocarcinoma, squamous cell lung cancer, colorectal cancer, clear cell ovarian cancer	Lung adenocarcinoma, squamous cell lung cancer, colorectal cancer, clear cell ovarian cancer: rs190505819 - OR: 1.88; 95% CI: 1.44–2.45; p: 2.95*10−6	None
BRCA2	13: 32839617-33023809	196	1 (1)	Overall lung cancer, squamous cell lung cancer, colorectal cancer	Overall lung cancer, squamous cell lung cancer, colorectal cancer: rs56404467 - OR: 1.30; 95% CI: 1.15–1.48; p: 3.26*10−5	Breast cancer, lung cancer
MORF4L1	15: 79115123-79240081	327	15 (2)	Overall lung cancer, lung adenocarcinoma, squamous cell lung cancer, clear cell ovarian cancer	Overall lung cancer, lung adenocarcinoma, clear cell ovarian cancer: rs7179953 - OR: 0.93; 95% CI: 0.91–0.96; p: 5.34*10−7	Lung cancer
BABAM1	19: 17328232-17443811	404	33 (5)	Squamous cell lung cancer, ER-neg breast cancer, serous ovarian cancer, overall breast cancer, overall ovarian cancer, overall prostate cancer,	Squamous cell lung cancer, ESR1 (ER)-negative breast cancer, serous ovarian cancer: rs4808076 - OR: 1.14; 95% CI: 1.10–1.19; p: 1.77*10−10	Breast cancer, ovarian cancer
DPF1	19: 38651649-38770317	284	2 (1)	Overall prostate cancer, ER-neg breast cancer, endometrioid ovarian cancer, aggressive prostate cancer	Overall prostate cancer, ER-neg breast cancer, endometrioid ovarian cancer, aggressive prostate cancer: rs12611084 - OR 0.93; 95% CI: 0.91–0.96; p: 8.40*10−8	Prostate cancer

Nr=Number; SNP=Single nucleotide polymorphism; GWAS=Genome-wide association study; OR=Odds ratio; CI=Confidence interval; ER=Estrogen receptor

^aIndependent associations

^bassociated with breast, colorectal, lung, ovarian or prostate cancer

The most pleiotropic genes were TET2, BABAM1, DPF1and especially L3MBTL3 (Figure 1, Table 2). Eleven variants in L3MBTL3 were associated with cancer risk, all with pleiotropic effects. The highest OR (odds ratio) in this region was 1.06 (rs9388766, 95% CI (confidence interval) =1.03–1.08, FDR q= 0.02), which was associated with risk of colorectal, overall breast, ESR1 (ER)-negative breast, clear cell ovarian, overall and aggressive prostate cancer. L3MBTL3 is a member of the putative Polycomb group (PcG) proteins. Two SNPs, rs9375694 and rs6569648, were previously identified as eQTLs (expression quantitative trait locus) for L3MBTL3 (RegulomeDB score: 1d and 1f, respectively) (28). The variant allele of rs6899976 may also be functionally important, since it overlaps with CTCF enriched regions in all cell lines as well as a transcription factor binding site. However, this variant has a RegulomeDB score of only 4.

TET2 (tet methylcytosine dioxygenase 2) at 4q24 encodes a protein catalyzing the conversion of methylcytosine to 5-hydroxymethylcytosine. Nine variants at this locus were significantly associated with risk of at least two cancer types, of which one variant (rs6825684) was associated with decreased risk of four cancers or subtypes: colorectal, overall prostate, overall and endometrioid ovarian cancer (OR=0.89, 95% CI=0.85–0.93, FDR q=0.02) and one polymorphism (rs62331150) showed a bidirectional effect. The variant allele of rs62331150 increased the risk of overall breast and serous ovarian cancer (OR:1.09, 95% CI: 1.02–1.15, p-value=0.009) and decreased the risk of clear cell ovarian and prostate cancer (OR=0.91, 95% CI= 0.87–0.96, p-value=0.0004) with a combined q-value of 0.04 (Figure 2). Most of the variants were positioned within TET2. The non-synonymous rs34402524 was predicted to be deleterious (SIFT) and possibly damaging (PolyPhen). Among the ASSET identified and FFU SNPs, further functional annotation singled out polymorphisms with a possible functional role. Rs62331150, (RegulomeDB score =2b), overlaps with a transcription start site, a transcription factor binding site and an enhancer region.

Figure 2.

(A) Linkage disequilibrium (LD) plot encompassing the significant SNPs in the TET2 region. Selected SNPs representing each LD block with respective forest plots are shown for (B) rs62331150 representing the single-variant block A; (C) rs17508261representing block B1 and B2; and (D) rs2007403 representing block C.

33 variants, all pleiotropic, showed an association with cancer susceptibility in the region containing BABAM1, a known ovarian and breast cancer locus. The strongest association was observed for rs4808076, which conferred 14% increased risk of ESR1 (ER)-negative breast, serous ovarian and squamous cell lung cancer (OR=1.14, 95% CI=1.10–1.19, FDR q=6.87*10⁻⁵). Five variants decreased the risk of six cancer types and subtypes; overall prostate, overall breast, ESR1 (ER)-negative breast, squamous cell lung, overall and serous ovarian cancer risk (strongest signal for rs8100241: OR=0.95, 95% CI=0.93–0.97, FDR q=1.78*10⁻³). Besides BABAM1, the captured region (19p13) additionally contains ANKLE1, ABHD8 and USHBP (Supplementary Table S5). BABAM1 was selected for its involvement in chromatin modifications, namely ubiquitination as part of the BRCA1 A complex. The ASSET identified SNPs in this region were in LD with several variants that may play an important role in regulatory processes. The most important ones are shown in Table 3. Apart from the variants in regulatory regions, five SNPs were in coding sequences. Important features of these variants, as well as their SIFT (29) and PolyPhen (30) scores are shown in Table 4.

Table 3.

Variants in the region 19p13 with a putative functional effect using ENCODE combined genome segmentation assessed in Huvec, H1hesc and Gm12878 cell lines

SNP ID	SNP position	RegulomeDB score	TFBSa	Promoter flanking	Enhancer	Weak enhancer	Transcription start site (TSS)	CTCF rich
rs10406920	17389648	3a	+
rs113299211	17400765	5				Gm12878
rs11540855	17403361	2a	+				Huvec, H1hesc, Gm12878
rs11667661	17390579	2b
rs11669059	17400453	2b	+				Huvec
rs12982178	17371568	3a				H1hesc		Huvec
rs2363956	17394124	5						Huvec
rs34084277	17387176	1f
rs35686037	17359535	2b	+		H1hesc
rs4808616	17403033	2b			H1hesc		Gm12878
rs55924783	17404072	5			H1hesc, Gm12878
rs56069439	17393925	4	+		H1hesc			H1hesc
rs66753001	17394839	5						H1hesc
rs73509996	17393449	4	+		Huvec		Huvec, H1hesc
rs8100241	17392894	4	+		Huvec, H1hesc		Gm12878
rs8108174	17393530	2b	+				Huvec, H1hesc, Gm12878
rs8170	17389704	4	+

^aIndicated with + when overlapping with the ENCODE Uniform transcription factor binding site (TFBS) track

Table 4.

Pleiotropic polymorphisms in the BABAM1 region that are located in exons and their predicted effect on the proteins

SSNP	Position	Gene	Nucleotide change	Amino acid change	EUR MAF	SIFT	PolyPhen-2
rs8170	19:17389636	BABAM1	AAG ⇒ AAA	K [Lys] ⇒ K [Lys]	A:0.16	n.a.	n.a.
rs10425939	19:17389155	ANKLE1	GGC ⇒ GGT	G [Gly] ⇒ G [Gly]	T:0.16	n.a.	n.a.
rs10425939	19:17389155	ANKLE1	GCT ⇒ GTT	A [Ala] ⇒ V [Val]	T:0.16	n.a.	n.a.
rs8100241	19:17392893	ANKLE1	GCG ⇒ ACG	A [Ala] ⇒ T [Thr]	A:0.58	deleterious(0)	probably_damaging (0.998)
rs8108174	19:17393529	ANKLE1	CTG ⇒ CAG	L [Leu] ⇒ Q [Gln]	A:0.58	deleterious(0)	probably_damaging (1)
rs2363956	19:17394123	ANKLE1	TTG ⇒ TGG	L [Leu] ⇒ W [Trp]	G:0.57	deleterious(0.03)	probably_damaging (0.999)

SNP= Single nucleotide polymorphism; EUR MAF= Minor allele frequency in European population

DPF1 is part of the neuron-specific chromatin remodeling complex (nBAF complex). One variant (rs12611084) was significantly associated with endometrioid ovarian, ESR1 (ER)-negative breast, overall and aggressive prostate cancer risk (OR=0.93, 95% CI=0.91–0.96, FDR q=0.005) and one variant (rs8100395) additionally with lung adenocarcinoma (OR=0.93, 95% CI=0.90–0.96, FDR q=7.2*10⁻³). Both variants were located upstream of DPF1, some were overlapping with other genes in this region, PPP1R14A and SPINT2, and were captured by one tagSNP in the FunciSNP analysis. Seven FFU SNPs showed a possible functional role, among them rs7250689, which was previously reported to be an eQTL for PPP1R14A (28). Based on RegulomeDB, rs8100395 and rs12611084 (both significant in the ASSET analysis) likely affect binding, and additionally overlap with enhancer regions as well as transcription factor binding sites and, in the case of rs8100395, overlaps with a CTCF enriched region.

Overall, 27 polymorphisms in RUVBL1 were associated with risk of prostate and endometrioid ovarian cancer, while one SNP was additionally associated with colorectal cancer risk. The strongest association was observed for rs144609957 with increased risk of prostate and endometrioid ovarian cancer (OR=1.13, 95% CI=1.08–1.19, FDR q=0.01). None of the SNPs had reached genome-wide significance in the original meta-analysis. RUVBL1 plays a role in chromatin organization. All associated SNPs belonged to the same LD block and were captured by one tagging SNP. Further, FunciSNP analysis revealed seven variants that overlapped with multiple biofeatures (transcription factor binding site, weak enhancer region and promoter flanking region). These variants also had low RegulomeDB scores, the lowest being 2b for rs9879865 and rs9879866, variants that likely affect binding.

DISCUSSION

We performed the first large-scale association study of variants in epigenetic-related genes and cancer risk utilizing the extensive genomic data on 51,724 cancer patients and 52,001 controls and identified eight epigenetic-related genes with pleiotropic effects on cancer risk.

Epigenetic disturbances are common drivers of carcinogenesis, yet, effects of germline variants and their potential pleiotropic mechanisms are not well understood. Thus, we investigated the risk association of SNPs related to epigenetic processes with multiple cancers. Using a subset-based meta-analysis, we were able to account for different subsets of cancer types and subtypes even with contrasting risk associations.

The L3MBTL3 gene on 6q26 is a member of the putative Polycomb group (PcG). It contains a methyl-lysine reader Malignant Brain Tumor (MBT) domain that is responsible for the recognition of the mono- and di-methylated lysines of H3 and H4 histone tails. MBT domain proteins are associated with gene expression repression and their dysregulation has been shown to contribute to different diseases (31). In our analysis, two variants (rs9375694 and rs6569648), which were previously identified as eQTLs, were significantly associated with risk of prostate and breast cancer (and their subtypes), and to a lesser extent with risk of clear cell ovarian and colorectal cancer (28). Interestingly, previous GWAS identified an association of rs6569648 and rs6899976, both hits in our analysis, with height (32) and height is associated with risk of several cancers including breast, ovarian, prostate and colorectal cancer (33). Our findings suggest the link between height and cancer risk may be vis-a-vis altered epigenetic processes, but this requires further investigations.

Several SNPs located in and around TET2 showed significant associations with risk of overall prostate, overall ovarian, endometrioid ovarian, overall breast and colorectal cancer. Previous studies reported significant associations of variants at the TET2 locus with risk of cancer including ovarian and breast cancer (9, 21, 34). A large number of functional variants were identified in this region forming multiple pleiotropic linkage blocks that support the role of TET2 and its germline variants in the development of multiple cancer types. Furthermore, an association between rs62331150 and TET2 gene expression in breast normal and tumor tissue was recently shown (9). The bidirectional association of the rs62331150 variant allele implies that the effect of TET2 genetic variation may be of a different nature for distinct cancers, increasing the risk of breast cancer, but decreasing the risk of prostate cancer. Similar associations were observed for a group of highly linked polymorphisms, namely rs2007403, rs2047409, rs6533183, rs6839705, rs11097882 and rs13147502 confirming previous studies (21, 35); however, with only one statistically significant risk direction.

Several functional variants were found at 19p13 with significant associations observed for risk of ESR1 (ER)-negative breast cancer, serous ovarian cancer and squamous cell lung cancer, but also with overall ovarian, breast and prostate cancer. BABAM1 is involved in chromatin modifications (ubiquitination), as part of the BRCA1 complex and regulates the retention of BRCA1 at double strand DNA breaks to maintain stability of this complex at the sites of DNA damage (36). Previous GWAS associated this region with breast (37) and ovarian cancer (10), with some of the SNPs showing triple-negative breast cancer specificity (38). However, to our knowledge we are the first to describe an association with squamous cell lung cancer or overall prostate cancer risk. Demonstrating a limitation of our selective candidate gene approach, new evidence suggests that nearby 19p13 genes ANKLE1 and/or ABHD8, rather than BABAM1, may be the functional drivers in breast and ovarian cancer [Lawrenson et al, Nature Communications in press]. The complexity of this region requires detailed functional follow up to disentangle the combined effect of individual variants and to understand their role in carcinogenesis.

DPF1 is part of the mSWI/SNF (also called BAF) chromatin remodeling complex with a central role in carcinogenesis (39). Mutations in DPF1 were seen in solid tumors (7). Furthermore, significant overexpression of DPF1 was observed in breast and squamous cell lung cancers (40). Our results also support a pleiotropic effect of DFP1 during carcinogenesis through potentially functional polymorphisms in this gene. However, as in each region of interest, we cannot exclude the potential relevance of the other genes in this region (PPP1R14A, SPINT2).

Polymorphisms in 3q21 were previously only observed in association with prostate cancer risk (41); however, our analysis has detected additional associations with endometrioid ovarian and colorectal cancer risk. RUVBL1 is a member of the INO80 family protein remodeling complex. It interacts with MYC and CTNNB1 (β-catenin), participates in many signal transduction pathways and is overexpressed in many cancer types (42). We have identified several polymorphisms with seemingly strong functional impacts. Interestingly, a proportion of endometrioid ovarian and colorectal cancers arise from common etiologies associated with hereditary non-polyposis cancer (HNPCC) or Lynch syndrome (43) and also show de novo promoter methylation silencing of DNA mismatch repair genes (44) and altered β-catenin signaling (45). RUVBL1 may represent novel susceptibility genes that further unify endometrioid ovarian and colorectal cancer development.

The major strength of this study is the large sample size of more than 100,000 subjects across five cancer types and their subtypes, some of which were not studied in TCGA. In addition, by searching the most parsimonious grouping based on the test statistics using any of the five cancers or cancer subtypes simultaneously as the outcome variables, the ASSET-subset-based meta-analysis (1) increased the power to detect associations, which may not have been detected in the individual analyses of the five cancer types, (2) allowed estimation of associations with opposing effects, and (3) provided new insights into pleiotropy that were not observed in the original analyses (22). Further, the overlapping subjects (cases and controls) are accounted for during the analysis (11). Finally, our focused approached reduced the genome-wide multiple testing burden and allowed for examination of functionally grouped subsets of epigenetic-related genes (i.e., sub-pathways). We were thus able to confirm established and identify new risk genes, including TET2 and L3MBTL3.

Although the odds ratios that are discovered as pleiotropic across cancer types may be considered modest, there is potential clinical significance. First, the ORs for individual cancers may be higher than the summary OR. Second, the combination of several SNPs with low ORs may become relevant through creation of a risk score, and third, the association of SNPs with disease may be modified and, in some instances, strengthened by environmental factors.

While our approach provides interesting insights into the pleiotropic effects of selected regions, it is limited with respect to the assignment of the identified predisposing variants to genes by chromosomal position rather than the actual cancer-initiating processes. Of note, several of the identified pleiotropic associations cannot clearly be linked to the selected epigenetic genes, as some of the regions additionally contain genes that were previously described for their effect on carcinogenesis.

Further investigations are required to elucidate the functional link between the identified pleiotropic variants and their impact on epigenetic processes such as the potential effect of TET2 polymorphisms on DNA methylation. Indeed, our pathway-based selection of epigenetic-related genes overlooked the subtleties of complex gene networks, and most genes are involved in multiple biological processes. Finally, this dataset did not allow for the investigation of interactions with other genetic or environmental factors, which are undoubtedly of great importance.

In summary, using a unique, large dataset, we identified novel pleiotropic variants in epigenetic-related genes that are associated with susceptibility to multiple cancer types and subtypes. This study provides the basis for future studies investigating the impact of these variants, their causal relationship to epigenetic processes, and the mechanisms leading to carcinogenic pleiotropy.