Common, germline genetic variations in the novel tumor suppressor BAP1 and risk of developing different types of cancer

Abstract

BRCA1 associated protein-1 (BAP1) is a novel tumor suppressor that has recently been shown to be somatically mutated in several cancers. The BAP1 gene also carries rare germline mutations in families with a high incidence of several types of cancers, such as mesothelioma, uveal melanoma, lung adenocarcinoma, melanocytic neoplasms, and renal cell carcinoma. To test the hypothesis that common, germline genetic variants in BAP1 may also contribute to the risk of developing different types of cancer, we genotyped germline single nucleotide polymorphisms (SNPs) for BAP1 in a large population of patients with cancer, including 2,340 with colorectal cancer, 1,436 with bladder cancer, 3,313 with lung cancer, 1,325 with renal cell carcinoma, and 1,162 with esophageal cancer. We identified significant association of rs11708581 (P = 0.0034) and rs390802 (P = 0.015) with risk of renal cell carcinoma and rs12163565 (P = 0.038) with risk of lung cancer. Expression quantitative trait loci analysis in renal cell carcinoma using publicly available data from TCGA showed that the proxy SNPs for rs11708581 and rs390802 were negatively associated with the expression level of BAP1. Our study indicate that common germline genetic variants of BAP1 play a role in mediating the risk of developing renal cell carcinoma and lung cancer.

INTRODUCTION

BRCA1-associated protein 1 (BAP1) is a nuclear ubiquitin carboxy-terminal hydrolase or deubiquitinating enzyme (DUB) that play a role in regulating several cellular functions, such as cell cycle, differentiation, DNA damage response, and cell proliferation (reviewed by Carbone et.al. [1]). BAP1 was originally discovered as a binding partner for BRCA1 [2] and shown to inhibit cancer cell growth. Later studies showed that BAP1 was not responsible for deubiquitylation of BRAC1/BARD1 [3] and demonstrated that BAP1’s effect on cell growth was independent of BRCA1 [4]. BAP1’s role as a tumor suppressor has been supported by an in vivo tumorigenicity study, in whch BAP1 activity and nuclear localization restricted the growth of lung cancer cell lines [4]. That study also showed that BAP1 slows cell growth through altered G1-S checkpoint regulation. These functions in the regulation of the cell cycle, cell growth, and proliferation are mediated through interactions with HCF-1 (host cell factor-1) [5] and transcription factor YY1 (Ying Yang 1) [6], which result in transcriptional control of genes involved in these processes.

Recently, several studies have identified somatic and germline BAP1 mutations in a wide-range of cancers, including mesothelioma [7–9], uveal melanoma [10–12], lung adenocarcinoma [11], renal cell carcinoma [13, 14], meningioma [11], and melanocytic neoplasms [15]. These studies have shown that rare genetic alterations in BAP1 can be drivers of the tumorigenetic process in these tumors.

Lung cancer, colorectal cancer, bladder cancer, esophageal cancer, and renal cell carcinoma comprise an estimated 31.7% of the total cancers diagnosed this year in the United States and up to 43.8% of predicted cancer deaths [16]. The pathogenesis of these cancers is multi-factorial with a combination of both exposure and lifestyle factors (i.e. age, sex, co-morbidities, obesity, and occupation) as well as a genetic component. Large scale genome-wide association studies (GWASs) have been conducted for each cancer site and have identified a number of novel susceptibility loci [17–30]. However, the effects of known susceptibility loci do not account for the entire genetic component of these cancers, according to estimates of heritability [31]. Therefore, there is a need to identify new genetic factors that contribute to cancer risk as a step towards enhanced risk prediction.

Although these previous studies have provided the functional characterization of BAP1 and identified rare mutations in several cancers, the role of common germline genetic variants in BAP1 on the risk of developing different types of cancer remains largely unknown. A recent large scale meta-analysis showed that common genetic variants in APC and MLH1 genes, two genes known to have high-penetrance mutations contributing to familial colorectal cancer, had strong cumulative epidemiological evidence for a significant association with colorectal cancer risk [32]. In this study, we used a candidate gene based approach, which has been shown to play an important role in identifying additional genetic variants associated with a disease of interest even after the advent of GWAs [33]. We hypothesized that common, germline genetic variants in BAP1 may also contribute to the risk of developing different types of cancer. To ensure a comprehensive coverage of common genetic variants in the BAP1 gene, we used the tagging single nucleotide polymorphism (SNP) approach with an r² of 0.80 and a minor allele frequency of >5% for SNP selection and further supplemented with two additional SNPs based on the basis of putative function. We performed this analysis in a large, multi-cancer site, case-control study comprised of 9,576 cancer cases (bladder cancer, lung cancer, renal cell carcinoma, esophageal cancer, and colorectal cancer) and a pool of 4,945 controls.

RESULTS

Host characteristics

This multi-cancer site analysis included a total of 9,576 cancer cases and 4,945 healthy controls (Table 1 and Supplementary Table 1). All patients included in the analysis were white. The number of cancer cases by type was as follows: bladder cancer, 1,436; colorectal cancer, 2,340; esophageal cancer, 1,162; renal cell carcinoma, 1,325; and lung cancer, 3,313.

Table 1. Study populations.

Variables	Cases, N(%)	Controls, N(%)	P-value
Bladder Cancer
Total	1436	1436
Age, mean(SD)	65.40(11.09)	65.40(11.07)	1.00
Sex			1.00
Male	1147(80)	1147(80)
Female	289(20)	289(20)
Smoking status			<0.001
Never	340(28)	491(34)
Ever	877(72)	941(66)
Colorectal Cancer
Total	2340	2340
Age, mean(SD)	56.48(11.89)	56.52(11.83)	0.91
Sex			1.00
Male	1350(58)	1350(58)
Female	990(42)	990(42)
Smoking status			<0.001
Never	1270(54)	899(39)
Ever	1070(46)	1432(61)
Esophageal Cancer
Total	1162	1162
Age, mean(SD)	62.67(10.76)	62.68(10.74)	0.99
Sex			1.00
Male	1027(88)	1027(88)
Female	135(12)	135(12)
Smoking status			<0.001
Never	270(27)	398(34)
Ever	718(73)	760(66)
Renal Cell Carcinoma
Total	1325	1325
Age, mean(SD)	59.54(10.80)	59.54(10.78)	1.00
Sex			1.00
Male	882(66.57)	882(67)
Female	443(33.43)	443(33)
Smoking status			<0.001
Never	580(46)	488(37)
Ever	687(54)	834(63)
Lung Cancer
Total	3313	3313
Age, mean(SD)	62.40(10.81)	62.32(10.78)	0.78
Sex			1.00
Male	1824(55)	1824(55)
Female	1489(45)	1489(45)
Smoking status			1.00
Never	589(18)	589(18)
Ever	2724(82)	2724(82)

*SD–standard deviation. Number for each variable may not add up to total due to missing data.

BAP1 genetic variants and risk of developing different types of cancer

Overall, three BAP1 genetic variants were significant, two with the risk of renal cell carcinoma and one with the risk of lung cancer (Table 2). The most significant association was observed for rs11708581. A 26% decrease in the risk of developing renal cell carcinoma (95% confidence interval [CI]= 0.61-0.91, P=0.0034) was observed for those carrying either the heterozygous or homozygous variant genotype for rs11708581. Significant reduced risk of renal cell carcinoma was observed for those carrying variant alleles of rs390802 (odds ratio [OR]=0.80, 95% CI= 0.67-0.96, P=0.015). Individuals carrying variant alleles of rs12163565 had a significantly increased risk of developing lung cancer (OR=1.11, 95% CI= 1.01-1.24, P=0.038) as well as an increased risk of developing bladder cancer (OR=1.17, 95% CI=0.99-1.39, P=0.070), although that increased risk was not statistically significant. However, none of the variants were significantly associated with the risk of esophageal cancer and colorectal cancer and the similar results were observed for esophageal cancer with the additional adjustment for alcohol consumption (data not shown). The association for rs11708581 with renal cell carcinoma remained significant even after adjustment for multiple comparison using the Bonferroni correction of 0.01 (=0.05/5 tests) while none of the other associations remained significant after Bonferroni correction.

Table 2. Effects of BAP1 genetic variation on the risk of developing different types of cancer.

SNP	Cases, N (cc/cv/vv)	Control, N (cc/cv/vv)	Model	*OR(95% CI)	p-value
Bladder Cancer
rs11708581	1066\310\16	1064\325\23	Dom	0.90(0.74-1.10)	0.30
rs123598	1260\135\1	1272\133\2	Dom	1.04(0.79-1.36)	0.79
rs12163565	893\461\50	945\422\53	Dom	1.17(0.99-1.39)	0.070
rs390802	922\427\40	941\422\46	Dom	0.99(0.83-1.18)	0.89
rs13094687	617\632\135	613\631\163	Dom	1.06(0.89-1.26)	0.51
Esophageal Cancer
rs11708581	880\247\20	848\261\20	Dom	0.94(0.76-1.16)	0.54
rs123598	1033\114\2	1046\95\2	Dom	1.07(0.79-1.45)	0.67
rs12163565	755\340\33	759\345\39	Dom	1.00(0.83-1.21)	1.00
rs390802	772\333\37	758\344\33	Dom	0.93(0.77-1.13)	0.49
rs13094687	502\488\135	516\484\142	Dom	1.00(0.88-1.13)	0.96
Renal Cell Carcinoma
rs11708581	1028\251\9	982\295\18	Dom	0.74(0.61-0.91)	0.0034
rs123598	1173\127\1	1169\123\3	Dom	0.99(0.71-1.38)	0.95
rs12163565	825\420\56	861\391\55	Dom	1.05(0.86-1.29)	0.62
rs390802	918\340\34	861\400\35	Dom	0.80(0.67-0.96)	0.015
rs13094687	571\584\139	613\547\136	Dom	1.09(0.93-1.28)	0.30
Lung Cancer
rs11708581	2521\685\41	2467\721\42	Dom	0.93(0.82-1.04)	0.21
rs123598	2937\320\3	2900\330\6	Dom	0.93(0.79-1.10)	0.39
rs12163565	2060\1073\135	2122\1001\129	Dom	1.11(1.01-1.24)	0.038
rs390802	2207\970\76	2163\970\94	Dom	0.96(0.86-1.06)	0.41
rs13094687	1476\1435\342	1468\1404\357	Dom	1.00(0.90-1.11)	0.99
Colorectal Cancer
rs11708581	1767\503\30	1744\514\29	Dom	1.00(0.86-1.17)	0.80
rs123598	2083\224\1	2056\234\2	Dom	0.89(0.72-1.09)	0.93
rs12163565	1506\729\89	1501\700\95	Dom	1.07(0.94-1.22)	0.30
rs390802	1553\690\61	1526\697\64	Dom	0.97(0.85-1.11)	0.73
rs13094687	1062\1024\230	1054\992\241	Dom	1.00(0.88-1.13)	0.96

* adjusted for age, sex, and smoking status.

Dom – dominant,

c – common allele, v – variant allele.

Stratified analysis for BAP1 genetic variants and risk of of developing different types of cancer

To determine whether the effects of these significant variants were modified by epidemiological factors, we performed stratified analyses by age, smoking status, and sex (Table 3). The protective effects of both rs11708581 and rs390802 against renal cell carcinoma remained significant among male subjects, ever smokers and subjects younger than 65 yrs. The effect of rs12163565 with the risk of lung cancer remained significant among male and ever smokers. However, none of the interactions between these three SNPs and the stratification variables was significant.

Table 3. Effect of BAP1 genetic variants by sex, smoking status, and age.

SNP	Group	Case N(cc/cv/vv)	Control N (cc/cv/vv)	Model	*OR(95% CI)	P-value
Sex
Renal Cell Carcinoma
rs11708581	Male	689\165\5	652\197\16	Dom	0.75(0.60-0.95)	0.017
	Female	339\86\4	330\98\2	Dom	0.88(0.63-1.23)	0.45
rs390802	Male	625\217\22	575\263\26	Dom	0.77(0.63-0.95)	0.014
	Female	293\123\12	286\137\9	Dom	0.93(0.69-1.25)	0.65
Lung
rs12163565	Male	1128\594\73	1203\531\59	Dom	1.19(1.04-1.37)	0.013
	Female	932\479\62	919\470\70	Dom	0.99(0.85-1.15)	0.90
Smoking Status
Renal Cell Carcinoma
rs11708581	Never	447\109\6	372\103\6	Dom	0.87(0.65-1.17)	0.36
	Ever	538\128\3	607\192\12	Dom	0.73(0.57-0.94)	0.014
rs390802	Never	390\159\14	329\140\11	Dom	0.96(0.74-1.25)	0.78
	Ever	485\168\18	529\260\24	Dom	0.73(0.58-0.91)	0.0051
Lung
rs12163565	Never	377\187\18	375\177\29	Dom	1.00(0.78-1.27)	0.98
	Ever	1683\886\117	1747\824\100	Dom	1.12(1.00-1.26)	0.047
Age
Renal Cell Carcinoma
rs11708581	Subjects<65	679\157\5	639\196\15	Dom	0.72(0.57-0.91)	0.0065
	Subjects≥65	349\94\4	343\99\3	Dom	0.92(0.66-1.27)	0.60
rs390802	Subjects<65	601\226\18	552\265\32	Dom	0.77(0.62-0.95)	0.014
	Subjects≥65	317\114\16	309\135\3	Dom	0.90(0.67-1.21)	0.47
Lung
rs12163565	Subjects<65	1103\584\72	1138\554\72	Dom	1.07(0.93-1.23)	0.32
	Subjects≥65	956\489\63	984\446\57	Dom	1.13(0.97-1.31)	0.12

*adjusted for age, and smoking status in sex stratification; adjusted for age and sex in smoking status stratification.

Dom - dominant.

c – common allele, v – variant allele.

eQTL analysis

Because rs11708581 and rs390802 were not genotyped by The Cancer Genome Atlas (TCGA), we used 4 proxy SNPs for rs11708581 and 3 proxy SNPs for rs390802 (Table 4) to assess the association of these SNPs with expression level of BAP1. We found that BAP1 expression was negatively associated with two proxy SNPs for rs11708581 (rs17052053: r²=0.9, P value=0.006 and rs11713914: r²=1.0, P value=0.05) and one proxy SNP for rs390802 (rs11714402: r²=0.98, P value=0.023). The association of rs17052053 with BAP1 expression remained significant after Bonferroni correction of 0.007 (=0.05/7 tests).

Table 4. eQTL analysis on the effect of Proxy SNPs on BAP1 expression level for renal cell carcinoma using data downloaded from TCGA.

Proxy SNP	r2	SNP of interest	Coefficient	Standard error	P-value
rs1769	0.94	rs11708581	-0.062	0.042	0.142
rs6793317	0.9	rs11708581	-0.079	0.040	0.051
rs17052053	0.9	rs11708581	-0.113	0.041	0.006
rs11713914	1.0	rs11708581	-0.084	0.042	0.050
rs6763882	0.93	rs390802	-0.070	0.038	0.067
rs11714402	0.98	rs390802	-0.086	0.038	0.023
rs123602	0.98	rs390802	-0.050	0.041	0.222

DISCUSSION

BAP1 has generated much attention owing to the consistent findings of rare germline or somatic mutations in a wide-range of cancer sites [7–15]. However, the role of common germline genetic variants in BAP1 on the risk of developing different types of cancer is largely unknown. This novel tumor suppressor is intriguing because of not only its known association with BRCA1 but also its established functions in the regulation of cell cycle, cell growth, and cell proliferation, among other key cellular functions. A recent report further established the potential role of BAP1 in modulating the response to metabolic stress [34]. Prior mutational analyses of the BAP1 gene used a family-based design and focused on the high risk individuals to identify rare variants with MAF <0.01. The sample sizes in these published mutational analyses of the BAP1 gene were well below 100. In addition, various techniques such as comparative genomic hybridization array, targeted next-generation sequencing (NGS), whole exome NGS, sanger sequencing, and immunohistochemical analysis have been used to detect BAP1 mutations [35, 36]. However, due to the limitations of these techniques, detection of BAP1 mutation is still challenging and an ongoing research field. In this study, we performed an extensive investigation in different types of cancer that included nearly 10,000 cancer cases and 5,000 healthy controls to determine the association of the common germline genetic variants in BAP1 with the risk of developing different types of cancer. TaqMan genotyping assay was used in this study which is quite mature and has been used in many small scale SNP genotyping studies and also in the validation of the top candidate SNPs from genome-wide association studies. The goal of this study is to establish the role of common variations as susceptibility loci for cancer, providing additional knowledge of the impact of this gene.

BAP1 is located at chromosome 3p21.31-p21.2 in close vicinity to two other known genes, DNAH1 (dynein, axonemal, heavy chain 1) and PHF7 (PHD finger protein 7). This genomic region of interest, which was defined at 10 kilobases (kb) upstream and downstream of the BAP1 gene for a total of 28,985 base pairs (bp), contains a total of 135 common germline genetic variants that were genotyped as part of the HapMap project (Phase 3 data). Among white population, four tagging SNPs were identified, representing common genetic variation across this entire genomic region of interest. We also genotyped two other variants in this region identified in dbSNP owing to the potential functional roles of these variants. rs123598 is located in the 3’-UTR of BAP1 and rs56238158 was identified as a non-synonymous variant in the whole genome sequencing of J. Craig Venter [37]. The rs56238158 variant was not polymorphic in our population of nearly 15,000 individuals, suggesting that it is a private variant or one occurring with a very low frequency in the population. The overall lack of genetic variation in this region is intriguing, particularly when considering the potential role of BAP1 in the tumorigenic process for a number of cancers.

In the risk analysis by cancer site, the most significant association was for rs11708581 in renal cell carcinoma, resulting in a 26% reduction in risk that remained significant after Bonferroni correction for multiple comparisons. We also found that rs390802 was associated with a decreased risk of renal cell carcinoma. These two SNPs are located within 10 kb downstream of BAP1. Because of the close proximity of other genes to BAP1, rs11708581 is a synonymous variant of DNAH1 and rs390802 is an intronic variant of DNAH1. Using HaploReg [38] and RegulomeDB [39] from the ENCyclopedia Of DNA Elements (ENCODE) project data [40], we found that two regions covering these SNPs and their correlated variants (r²>0.8) had a length of 250 kb for rs11708581, including 80 SNPs and 224 kb for rs390802, including 56 SNPs (Supplementary Tables 2 and 3). Interestingly, rs123602 (an intronic variant of BAP1) and rs498946 (61 bps 3’ of BAP1) are in high linkage disequilibrium (LD) with rs390802 (r²=0.98 and 0.95, respectively) and moderate LD with rs11708581 (r²=0.56 and 0.56, respectively). Because genotyping data for rs11708581 and rs390802 were not available in TCGA for renal cell carcinoma, we used the proxy SNPs in the eQTL analysis and found that two proxy SNPs for rs11708581 and one proxy SNPs for rs390802 exhibited an significantly negative association with the expression level of BAP1. Genotyping data for rs123602 were available but did not show any association with BAP1 expression. Both rs11708581 and rs390802 are in the DNase I hypersensitivity region which is a signature of open chromatin that is typically associated with transcriptionally active regions of the genome [41]. Therefore, it is possible that rs11708581 and rs390802 affect the risk of developing renal cell carcinoma through BAP1.

We also found that rs12163565 was associated with an increased risk of developing lung cancer. Similar to rs11708581, rs12163565 is a missense variant of DNAH1 and is located within 10kb of BAP. The region covering its correlated variants (r²>0.8) had a length of 201 kb, including five SNPs and none of the SNPs in the BAP1 gene had high LD with rs12163565 (Supplementary Figure 1). Therefore, its role on the functional consequence of BAP1 remains unknown.

Overall, the results of this analysis of BAP1 common germline genetic variations in a large population of nearly 10,000 patients with cancer from five different cancer sites support the idea that common germline genetic variants in BAP1 playing a role in mediating the risk of developing renal cell carcinoma and lung cancer.

MATERIALS AND METHODS

Study population

Cancer patients with newly diagnosed, histologically confirmed bladder cancer, esophageal cancer, renal cell carcinoma, lung cancer, or colorectal cancer were recruited from the University of Texas MD Anderson Cancer Center. Additional patients with colorectal cancer were recruited through the TexGen Consortium which focused on cancer patients seen at institutions throughout the Texas Medical Center. There were no recruitment restrictions on age, sex, ethnicity or cancer stage. Control subjects were defined as individuals with no prior history of cancer, excluding non-melanoma skin cancer, and were recruited in parallel with cases through two mechanisms: 1) random digit dialing and 2) from Kelsey-Seybold Clinics, the largest multi-site and multispecialty physician group in Houston, Texas. Details regarding the case and control recruitment for the on-going studies were described elsewhere [42–46]. Controls were frequency matched to cases by sex, age (±5), and ethnicity for all cancer types except for lung cancer, which was additionally matched by smoking status. All study participants provided written informed consent and the study was approved by the Institutional Review Boards of MD Anderson Cancer Center, Baylor College of Medicine, and Kelsey-Seybold Clinic. The analysis was restricted to self-reported white patients to minimize confounding by ethnicity.

Data collection

A standardized questionnaire was used to collect epidemiologic information such as age, sex, ethnicity, medical history, and smoking history. A never smoker was defined as an individual who had smoked fewer than 100 cigarettes in his or her life, a former smoker was an individual who had smoked more than 100 cigarettes, but had quit smoking more than 1 year prior to diagnosis for cases and interview for controls. And a current smoker was an individual who was currently smoking or quit less than 1 year prior to diagnosis for cases or interview for controls, respectively. For the purposes of the current study, former and current smokers were grouped together as ever smokers. A blood specimen was collected from all study participants for genetic analyses. DNA was isolated from each blood sample using the QIAamp DNA Blood Maxi Kit (QIAGEN, Valencia, CA) following standard procedures and stored at -80 °C until use.

SNP selection and genotyping

International HapMap Project data for the CEU population was used to select tagging SNPs within 10 kb of BAP1 with an r² of 0.80 and a minor allele frequency of >5%. Four variants were identified as tagging SNPs and an additional two variants were selected based on the basis of putative function: rs123598 and rs56238158 (Table 5). TaqMan genotyping assays for each SNP were purchased from ABI (Carlsbad, CA) and genotyping was completed following standard protocol. All genotyping was performed blinded to the status of each patient (i.e., case or control). Water negative controls and internal controls were included in each plate, and 5% of the samples were randomly selected and run in duplicate, with 100% concordance. The call rates were over 95% for all of the selected polymorphisms except for rs56238158, which was not detectable as a polymorphism in any of our study populations and was excluded from further analysis. The remaining five SNPs were tested for agreement with the Hardy–Weinberg using a goodness of-fit χ² test in the control genotyping data.

Table 5. Genotyped genetic variants for BAP1.

SNP	Chromosome (hg36)	Position	Nucleotide change	MAF*
rs11708581	chr3: 52428988	3’ Flanking	C>A	0.05
rs12163565	chr3: 52430526	3’ Flanking	G>A	0.17
rs390802	chr3: 52431671	3’ Flanking	G>A	0.13
rs123598	chr3: 52435860	3’ UTR	G>A	0.03
rs56238158	chr3: 52439302	Glu>Lys	C>T	0.13
rs13094687	chr3: 52450043	5’ Flanking	A>G	0.29

*MAF – minor allele frequency.

Statistical analysis

We compared the difference in the distribution of categorical variables (sex and smoking status) using the Pearson’s χ² test and the distribution of continuous variable (age) using the Student’s t-test. The association of each SNP with the risk of developing different types of cancer was estimated as ORs with 95% CIs using unconditional multivariate logistic regression, adjusting for age, sex, and smoking status for all five cancer types. The effect of rs11708581, rs123598, rs12163565, rs390802, and rs13094687 on the risk of developing different types of cancer was assessed using the dominant model. Stratified analyses were also performed to identify the effects of genetic variants in specific subgroups. The likelihood ratio test was used to assess the interaction of the genetic variants and the stratification variables by comparing the model with and the model without the interaction term. Gene track, H3K27AC mark, DNAse clusters, and transcription factor ChIP-seq plots in Supplementary Figure 1 were generated by UCSC genome browsers (http://genome.ucsc.edu) [47]. LD plot in Supplementary Figure 1 was generated by Haploview software [48] using European subjects from the 1000 Genome reference panel (phase I, April 2012). Statistical analyses were completed using the STATA software package (version 10, STATA, College Station, TX) and p-value<0.05 was considered statistically significant.

Expression quantitative trait loci (eQTL) analysis

We performed eQTL analysis for SNPs rs11708581 and rs390802 to elucidate potential mechanism underlying the association of these SNPs with renal cell carcinoma. Data on renal clear cell carcinoma were downloaded from The Cancer Genome Atlas (TCGA, dbGaP Study Accession: phs000178.v9.p8, data portal: https://tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp). Expression data (RNA-seq) and methylation (HumanMethylation450 chip) were measured in tumor samples, somatic copy number was inferred from the tumor samples and paired normal controls (Affymetrix SNP 6.0 platform), and genotyping data were measured from blood derived DNA samples (Affymetrix SNP 6.0 platform). For the BAP1 gene, expression data were available for 533 cases, methylation data for the 19 CpG site were available for 319 cases, copy number data were available for 528 cases, and genotyping data were available for 556 cases. Expression values for BAP1 were log₂-transformed. Because rs11708581 and rs390802 were not directly genotyped, we found five proxy SNPs for rs11708581 and three proxy SNPs for rs390802 (r²>0.8) using LD information from the 1000 Genome project. In the dbGaP data, SNP rs6809248 did not show high LD with the other proxy SNPs for rs11708581 (r² range from 0.0 to 0.26) and was removed from further analysis. Following Li et al 2013 [49], We first evaluated the effect of somatic copy number and methylation level on log₂-transformed expression level using linear regression. Univariate analysis identified the significant associations of somatic copy number and methylation levels of five CpG sites with BAP1 expression level. However, in the multivariate model including the somatic copy number and methylation levels of five CpG sites, only somatic copy number remained significant and none of the methylation levels were signficant. Therefore, we assessed the association of eight SNPs in the dominant model with the expression level of BAP1 using a linear regression model while adjusting for the effect of somatic copy number.