A Transcriptome-Wide Association Study Among 97,898 Women to Identify Candidate Susceptibility Genes for Epithelial Ovarian Cancer Risk

Epithelial ovarian cancer (EOC) has a substantial heritable component with a heritability estimated to be 22% (1). Genome-wide association studies (GWAS) have identified approximately 35 loci associated with EOC risk (2–12). Most reported associations are specific to the most common histologic subtype, serous EOC (2–7, 9–12). Together, known GWAS-identified variants account for approximately 6.4% of EOC risk in the general population (12), indicating that additional susceptibility variants remain to be identified. In addition, genes that underlie the large majority of GWAS-identified risk loci remain unknown; most GWAS-identified variants are located in noncoding genomic regions that may be involved in regulation of gene expression. Recent mechanistic studies have demonstrated that GWAS-identified variants are more frequently located in active chromatin regions, and highly-enriched with expression quantitative trait loci (eQTL; refs. 13, 14). This evidence underscores the importance of transcriptional regulation in influencing human traits and disease susceptibility.

Prior studies on genetically regulated gene expression were largely limited to easily accessible sources, such as adipose tissue and peripheral blood cells (15). Although the sample size in eQTL studies of peripheral blood cells recently reached the thousands, a relatively small number of genes are expressed in blood cells compared with other tissue types (14). Conclusions from eQTL studies in tumor tissue (e.g., TCGA) should also be interpreted with caution due to the inherent complexity of transcriptional regulation caused by acquired somatic alterations (16). The Genotype-Tissue Expression (GTEx) project provides high-density genotype data and RNA sequencing (RNA-seq) transcriptome data from 53 tissues (14). We used these data to build models of genetically regulated expression for 17,121 genes. We investigated the association between these genetically predicted gene expressions and EOC risk using data from 97,898 women including 29,396 EOC cases. We identified 35 genes at 14 loci associated with EOC risk, and provide additional evidence of a potential role for dysregulated ovarian function and imbalanced ovarian hormone production in ovarian carcinogenesis.

The GTEx preliminary cleaned genome-wide genotype data and RNA-seq transcriptome data across 53 unique tissues (released on 2015-01-12) were downloaded from dbGaP (accession phs000424.GTEx.v6.p1). It included 183 GTEx donors genotyped on Illumina's Omni 5M and 267 GTEx donors genotyped on Omni 2.5M. Genomic and transcriptomic data were processed according to the GTEx protocol (http://www.gtexportal.org/home/documentationPage). The Omni 2.5M portion of hard-called genotypes from the Omni 2.5M or Omni 5M across all 450 donors were extracted and merged for analysis. We excluded variants with a genotyping call rate < 98%, with differential missingness between Omni 2.5M and Omni 5M arrays, with Hardy–Weinberg equilibrium P < 10⁻⁶ (for subjects of European ancestry), or with batch effects. Genotype data were imputed to the Haplotype Reference Consortium reference panel using minimac3 for imputation and SHAPEIT for prephasing (17). Variants with high imputation quality (R² ≥ 0.8), minor allele frequency (MAF) ≥ 0.05, and inclusion in the HapMap Phase 2 project were used to build predicted expression models.

We used gene level expression in Reads Per Kilobase of transcript per Million mapped reads (RPKM) from RNA-SeQC for gene expression data. For ovarian transcriptomic data, genes were required to have expression in ≥10 individuals with >0.1 RPKM and raw counts >6. For our analysis of cross-tissue derived transcriptomic data (below), genes were filtered on mean expression levels with >0.1 RPKM and RPKM >0 required in at least 3 individuals (18). We performed quantile normalization to transform the expression profile of each sample to the same scale, and performed inverse quantile normalization for each gene to map each set of expression values to a normal distribution. Residual expression was calculated by regressing transformed expression data against three top principal components (PC) derived from common genetic variants (MAF ≥0.05), top 15 or 35 probabilistic estimation of expression residuals (PEER) factors, respectively, for ovarian tissue and cross-tissue derived models (below; ref. 19), sex (for cross-tissue only) to correct for batch effects and other potential experimental confounders.

The ancestral analysis was conducted with 2,836 ancestry informative markers for 450 GTEx individuals and 1,092 individuals included in the 1000 Genome project (Phase 1; ref. 20). Of the individuals with both genotype and transcriptome data available, 369 were clustered together with EUR populations (CEU, FIN, GBR, IBS, and TSI) on the multidimensional scaling plot of the pairwise Identity-By-State distance and were included in the analysis, 68 of whom had transcriptome data available for ovarian tissue.

Mixed effect models were used to decompose gene expression levels into subject-specific and subject-by-tissue–specific components (18). GTEx data consisted of expression measurements from multiple tissues for each subject. The expression level of a gene at a given tissue for individual i was considered to be composed of a cross-tissue component represented as Y_i^CT and a tissue-specific component that was estimated as the difference between the expression levels and cross-tissue components given the lack of replicated measurement for a specific tissue/subject pair (18). Z′_i represents a vector of covariates that have effects of β on the expression levels of the gene, such as PEER factors, ancestry information derived from the principal component analysis, and sex. The expression of a gene for individual i in tissue t, Y_i,t, is modeled as

The mixed effect model parameters were estimated using the lme4 package in R. Posterior modes of the subject level random intercepts were used as estimates of the cross-tissue components (18). Cross-tissue model included gene expression from 6,124 GTEx tissue samples from 369 unique European individuals who had genome-wide genotype data available.

We built an expression prediction model for each gene using the elastic net method as implemented in the glmnet R package, with a ridge-lasso mixing parameter of α = 0.5 and a penalty parameter lambda chosen through 10-fold cross-validation (18, 21, 22). The elastic net method with α = 0.5 is a compromise between the ridge-regression penalty (α = 0) for solutions with many parameters (each of small effects) and the lasso penalty (α = 1) for solutions with fewer parameters (each of large effects; ref. 18). The genetically regulated expression for each gene was estimated by including SNPs within 1 Mb of the gene start or end, as defined by GENCODE V19 gene annotations. Expression prediction models were built for protein-coding genes, long noncoding RNAs (lncRNA), miRNAs, processed transcripts, immunoglobulin genes, and T-cell receptor genes, according to categories described in the GENCODE V19 gene annotation file. Pseudogenes were not included in this study because of potential concerns of inaccurate calling (23). Prediction r² values (the square of the correlation between predicted and observed expression) were generated to estimate the prediction performance for each gene in our prediction models.

With genome-wide genomic data and RNAseq-based tissue transcriptome data, we built an ovarian tissue cis genetically regulated expression model for 8,580 genes that had predicted performance of r² > 0.01 and a cross-tissue cis genetically regulated expression model for 14,085 genes that had predicted performance of r² > 0.01.

Associations between predicted gene expression levels and EOC risk were evaluated using MetaXcan (22). Briefly, the formula:

was used to estimate the Z-score of the association between predicted gene expression and ovarian cancer risk. Here |${w_{lg}}$| is the weight of SNP |$l$| for predicting the expression of gene |$g$|⁠, |${\hat{\beta }_l}$|⁠, and |${\rm{se}}({\hat{\beta }_l})$| are the association regression coefficient and its standard error for SNP |$l$| in GWAS, and |${\hat{\sigma }_l}$| and|$\ {\hat{\sigma }_g}$| are the estimated variances of SNP |$l$| and the predicted expression of gene |$g$|⁠, respectively. The input variables for the MetaXcan analyses include the weights for gene expression predicting SNPs, GWAS summary statistics results, and correlations between predictor SNPs. We integrated prediction models of gene expression levels with summary statistics from GWAS of EOC risk for 97,898 European women with 29,396 EOC cases from the Ovarian Cancer Association Consortium (OCAC) and Consortium of Investigators of Modifiers of BRCA1/2 (CIMBA; ref. 12) based on the variance and covariance matrix of genetic variants derived from 1000 Genome phase 3 EUR population (N = 503). The performance of MetaXcan has been shown to be similar to PrediXcan that uses individual-level genetic data for the identification of genes with expression that is associated with disease risk (21, 22).

Details of individual contributing studies were previously reported (12). Briefly, the OCAC summary statistics were based on analysis of 40,941 controls and 25,509 population-based EOC cases (22,406 invasive cases and 3,103 borderline cases). OCAC cases included 1,954 serous borderline ovarian cancers, 1,149 mucinous borderline ovarian cancers, 1,417 mucinous invasive ovarian cancer, 1,012 low-grade serous ovarian cancers, 13,037 high-grade serous ovarian cancers, 2,810 endometrioid ovarian cancers, 1,366 clear-cell ovarian cancer and 2,764 other EOC cases. The CIMBA summary statistics were based on the analysis of 19,036 BRCA1 and 12,412 BRCA2 mutation carriers, of whom 2,933 and 954, respectively, were diagnosed with EOC. Details of the genotyping procedure and QC have been described elsewhere (12). In brief, samples were excluded if they had a genotyping call rate < 95%, excessively low or high heterozygosity, if they were not female or had ambiguous sex, or were duplicates (cryptic or intended) (12). SNPs were excluded for a call rate <95%, deviating from Hardy–Weinberg equilibrium (P < 10⁻⁷ in controls or unrelated samples in CIMBA and P <10⁻¹² in cases) and concordance <98% among 5,280 duplicate pairs (12). All participants provided written informed consent and each contributing study was approved by the appropriate local institutional ethical review board. The studies were conducted in accordance with Declaration of Helsinki.

We used a Bonferroni-corrected P value threshold of 2.21 × 10⁻⁶ (adjusting for 22,665 gene–tissue pairs) to determine a statistically significant association in our analysis. This threshold was conservative as 5,544 genes appeared in both ovarian and cross-tissue models. We did the primary analysis for high-grade serous EOC, as this had the largest sample size. In our secondary analyses, we also evaluated other histotypes or the combined histotypes, even though power to discover novel gene associations was relatively low for some (i.e., clear-cell, endometrioid, or low-grade serous). To determine whether associations identified between genetically predicted gene expression and EOC risk were influenced by variants previously identified by GWAS, we conducted conditional analyses adjusting for index SNPs. Briefly, we performed conditional analyses developed by Yang and colleagues (24) (GCTA-COJO) to calculate association betas and standard errors of SNPs with ovarian cancer risk after adjusting for index SNPs of interest. This was followed by reperforming MetaXcan analyses using updated summary statistics.

We constructed genetically regulated expression models based on genome-wide genotype data and RNA-seq transcriptome data from the GTEx project (Supplementary Fig. S1; ref. 14). Ovarian transcriptome data were available for 68 European individuals, and 8,580 genes achieved a prediction performance of r² ≥ 0.01 in the ovarian model (Supplementary Table S1). Because a large portion of cis expression regulation is shared across multiple tissues (14, 18), we also used transcriptome data for 6,124 tissue samples from 369 European individuals to build cross-tissue models for 14,085 genes with a prediction performance of r² ≥ 0.01 (Supplementary Table S1).

We evaluated associations between predicted gene expression levels and EOC risk using MetaXcan (22) with summary statistics for individual GWAS SNPs from 97,898 European women including 29,396 EOC cases from OCAC and CIMBA (Supplementary Fig. S1; ref. 12). Our primary analysis focused on high-grade serous EOC; secondary analyses included other EOC histotypes (Supplementary Fig. S1).

In total, we identified 35 genes with genetically predicted expression that were associated with EOC risk at a Bonferroni-corrected threshold of P < 2.21 × 10⁻⁶ (Fig. 1; Supplementary Figs. S2 and S3; Tables 1 and 2; Supplementary Table S2). One gene at 11q14.2 (FZD4), was more than 1 Mb away from any GWAS-identified EOC susceptibility variant (Fig. 1), suggesting a potential novel risk locus for this disease. High predicted FZD4 expression was associated with increased risk of high-grade serous EOC (Z = 5.08, P = 3.83 × 10⁻⁷; Fig. 1). The remaining 34 genes were located within 1 Mb of previously identified EOC susceptibility variants (Tables 1 and 2; Supplementary Tables S2 and S3), including 11 genes (at 8 loci) that were previously implicated in EOC risk using functional annotation, bioinformatic prediction, in vitro cellular models or known gene biology. Our study provides additional evidence to support these previous findings (Table 2; Supplementary Table S3). However, 23 genes (at 6 known risk loci) had not been reported to be associated with EOC risk in prior studies (Table 1; Supplementary Table S3). For 31 of these 34 genes, the associations were no longer statistically significant at P < 1.47 × 10⁻³ (multiple comparisons correction of 0.05/34) after adjustment for the nearest SNP identified by EOC GWAS (Supplementary Table S4), indicating that the previously identified GWAS SNPs for EOC at these 31 regions might regulate the expression of these associated gene to affect EOC risk. Associations for three genes (Z = 6.84 vs. 3.27 for DNALI1, Z = 5.16 vs. 3.81 for HOXD3 and Z = −8.60 vs. −4.18 for CCDC171; Tables 1 and 2; Supplementary Table S4) remained statistically significant at P < 1.47 × 10⁻³ after adjusting for the nearest EOC risk SNP, although the strength of the association was attenuated. Four loci (2q31.1, 9p22.3, 17q21.31, and 17q21.32) had multiple nearby genes associated with EOC risk (Tables 1 and 2). This may be partially due to coregulated gene expression in these chromosomal regions (Supplementary Table S5 and Supplementary Material).

Consistent with the etiologic heterogeneity of EOC (25), GWAS-identified risk variants differed across histologic subtypes (12). Therefore, we investigated associations between genes with P < 2.21 × 10⁻⁶ across all major histotypes of EOC (Supplementary Table S6). The majority of identified genes were associated with high-grade serous EOC risk, likely due to the large number of cases in our primary analysis. A few additional histotype specific associations were identified from secondary analyses. HOXD3 at 2q31.1 was associated with borderline mucinous EOC risk (Table 2; Supplementary Table S6: Z = 5.16, P = 2.42 × 10⁻⁷). RP11-403A21.1 at 18q11.2 was associated with low-grade or borderline serous EOC risk (Table 1; Supplementary Table S6: Z = −5.53, P = 3.13 × 10⁻⁸). ZNF546 at 19q13.2 was associated with mucinous EOC risk (Table 1; Supplementary Table S6: Z = 7.14, P = 9.07 × 10⁻¹³ for invasive/borderline mucinous EOC combined; Z = 5.99 and P = 2.14 × 10⁻⁹ for borderline mucinous EOC only). HOXD1 at 2q31.1 was associated with both invasive serous (Supplementary Table S6: Z = 4.92, P = 8.55 × 10⁻⁷) and borderline mucinous (Supplementary Table S6: Z = 5.24, P = 1.59 × 10⁻⁷) EOC risk.

Evidence from previous eQTL analyses of identified EOC susceptibility risk variants supports several currently identified gene associations (Table 2; Supplementary Table S3). Reduced OBFC1 expression was associated with risk allele of GWAS identified EOC SNP at 10q24.33 (12), and we found that higher predicted OBFC1 expression was associated with lower EOC risk. Similarly, reduced RCCD1 expression was associated with risk allele of GWAS identified EOC SNP (11), and we found that higher predicted RCCD1 expression was associated with reduced EOC risk at 15q26.1. In addition, multiple lines of evidence support our finding between higher predicted ABHD8 and increased EOC risk at 19p13.11. Increased ABHD8 expression was associated with risk allele of GWAS identified EOC SNP (26). Copy number variant analysis indicated that 46% of high-grade serous EOC had amplification at 19p13.11 that contains ABHD8 (3).

In this large transcriptome-wide association study (TWAS) among 97,898 women of European ancestry, we identified 35 genes with genetically predicted expression levels associated with EOC risk. One of these genes (FZD4) is located more than 1 Mb away from any previously identified GWAS EOC variant (25 Mb away from the nearest reported EOC risk variant; ref. 11), suggesting it is a potential novel risk locus. All other 34 genes identified were located within 1 Mb of known GWAS loci, including 23 genes at 6 loci that had not previously been associated with EOC risk. After adjustment for nearby known EOC GWAS-identified variants, the associations for 3 of the 34 genes retained.

FZD4 is a member of the frizzled gene family that encodes seven-transmembrane domain proteins (Fz) as the receptors for the secreted Wnts signaling ligands. Several Wnts and Fzs (including Fzd4 and Wnt4), as well as downstream targets of the canonical WNT signaling pathway, are expressed at different stages of ovarian follicular development, ovulation, and luteinization, suggesting specific functions for these signaling molecules in the mature ovary (27). Recent studies using transgenic mouse models demonstrated that Wnt4, Fzd4, and Ctnnb1 are required for normal folliculogenesis, luteogenesis and steroidogenesis, and that dysregulated WNT signaling leads to granulosa cell tumor development (27, 28). FZD4-null female mice are infertile and exhibit reduced progesterone production, reduced luteinization-associated gene expression, impaired corpora lutea formation and function, and impaired vascular development (28). Interestingly, WNT4 (1p36.12) encodes a potential Fzd4-binding ligand, which was also recently identified as a potentially causal gene underlying EOC risk by GWAS (Supplementary Table S3; ref. 7). Aberrant activation of WNT signaling in adult tissues has been implicated in the pathogenesis of several types of cancer, including colorectal cancer (29). The positive association between FZD4 expression and invasive serous EOC risk suggests that dysregulated corpus luteum function and/or progesterone production may contribute to EOC pathogenesis.

A locus 17q21.31 was previously identified by GWAS as associated with EOC risk (10, 30). This region contains a 900-kb inversion in Europeans that has extensive linkage disequilibrium likely due to restriction from crossovers in individuals who are heterozygous with respect to inversion (31). The H2 haplotype is less frequent (20% in Europeans) and is associated with higher number of children born to women (31). Interestingly, minor alleles of genetic variants in this region were almost universally associated with reduced breast cancer risk but increased EOC risk at genome-wide significance levels (Supplementary Table S7; Supplementary Material; refs. 10, 30). Permuth-Wey and colleagues (10) investigated several of these genes, including KIF18B, C1QL1, DCAKD, NMT1, PLCD3, ACBD4, HEXIM1, HEXIM2, FMNL1, C17orf46, MAP3K14, ARHGAP27, PLEKHM1, CRHR1, IMP5, and MAPT; extensive functional analysis suggested that ARHGAP27 and PLEKHM1 may be EOC susceptibility genes (10). One of the other candidate genes at this region, CRHR1, is involved in regulating ovarian function; it is expressed in ovarian thecal cells, granulosa cells and luteal cells (32), and upregulated in EOC (10). High CRHR1 expression was almost universally associated with minor alleles of multiple genetic variants in this chromosome 17 region (Supplementary Table S8; Supplementary Material; ref. 33). Enhanced CRHR1 activation in the ovary leads to reduced production of testosterone (32) and estrogen (32, 34–36), but increased progesterone accumulation and production (32). This may explain the lower breast cancer risk associated with variants in this region from lower estrogen exposure and higher progesterone exposure associated with multiparity (31, 37). Similarly, this also suggests that imbalanced estrogen and/or progesterone production contributes to EOC pathogenesis.

Two of the candidate genes at the 17q21.32 locus, HOXB2 and HOXB3, belong to the homeobox gene family, which is important for normal vertebrate limb and organ development. This gene family was also recently shown to be enriched for genes underlying serous EOC risk by GWAS (38). Inconsistent tumorigenic effects of HOXB2 and HOXB3 were reported across several types of cancers (breast, pancreatic, lung, cervical cancer, and acute myeloid leukemia; refs. 39–43). This may be due to context-dependent effects from specific tumor microenvironments (39, 43). With regard to ovarian cancer, increased HOXB2 and HOXB3 expression were associated with reduced EOC risk; potential molecular mechanisms underlying HOXB suppressive effect on EOC warrant further investigation.

Several additional findings from this study are noteworthy. The precise function of DNALI1 at 1p34.3 is not known. It is a potential candidate gene for primary cilia syndrome or Kartagener syndrome, in which the action of cilia lining the respiratory tract and fallopian tube is compromised (44). A marked reduction in fertility was observed in female patients with Kartagener syndrome due to dysfunction of the oviductal cilia (45). The predicted expression of CCDC171 at 9p22.3 was associated with reduced EOC risk. CCDC171 was shown to interact with KRAS by a stringent screening for Ras synthetic lethal genes (46). Several lncRNAs were associated with EOC risk, including RP11-403A21.1 at 18q11.2 (Table 1). Little is known about their particular function in either tumor initiation or tumor development, but lncRNAs have been increasingly implicated in many classic cancer biology pathways (47). In addition to HOXD3 and HOXD1 at 2q31.1 (Table 2; Supplementary Table S3; refs. 4, 8), ZNF546 at 19q13.2 was identified as a novel candidate gene for mucinous EOC. Enrichment for expression in gonadal tissues (14) supports a potential role in EOC pathogenesis. Because of the complexity of mucinous EOC, and undetermined cell/tissue of origin, identification of associated genetic variants and/or genes is particularly important (8, 25).

The tissue samples used in building gene expression models in GTEx (V6) came most from the people who recently died of traumatic injury (for these young donors) or cardio-cerebrovascular diseases (for the old donors). There were no overlaps between the tissues used in building gene expression models and the samples used in EOC GWAS in OCAC or CIMBA. Our ability to detect genes significantly associated with EOC risk is affected by tissue specificity and the sample size of the dataset used to build genetic prediction models for gene expression. Four genes were identified from both ovarian and cross-tissue models; 8 genes were only identified on the basis of ovarian models, and 23 genes were only identified from cross-tissue models (Supplementary Table S2). The ovarian tissue transcriptome that we used to model gene expression was potentially derived from multiple ovarian cell types, including surface epithelial cells, oocytes, granulosa cells, Theca cells, luteal cells, and other interstitial cells. Because of the importance of tissue or cell-specific regulators (i.e., transcription factors or epigenomic features) in governing development and function, the ovarian-specific model should best capture transcriptional regulatory mechanisms of the ovary. However, in light of abundant shared cis regulation of expression across multiple tissues (14, 18), we also pooled constitutive variant-dependent regulatory information across tissues and built cross-tissue gene expression models. We would expect this model to yield greater power as the number of tissues in which a variant is functional increases. By coupling both tissue-specific and cross-tissue models, we aimed to robustly capture genetically regulated genes expression using a large sample size. Because of insufficient samples in the GTEx project, we did not build Fallopian tube–specific models.

In summary, we identified one novel locus (FZD4) and 34 genes at 13 known EOC risk loci associated with EOC risk, and these findings may help improve our mechanistic understanding of EOC pathogenesis. In line with tentative observations of increased borderline EOC risk from ovarian hormone dysregulation for women who received fertility drug treatment with in vitro fertilization (48–50), the known biology of FZD4 and CRHR1 in the ovary implicates the potential of long-term dysregulated ovarian function or imbalanced ovarian hormone production as a possible mechanism underlying EOC pathogenesis.

No potential conflicts of interest were disclosed.

Conception and design: Y. Lu, J.M. Schildkraut, H. Anton-Culver, B.K. Arun, P.A. Fasching, R.L. Nussbaum, P.H.M. Peeters, V.W. Setiawan, M.C. Southey, S.A. Gayther, A. Berchuck, W. Zheng, J. Long

Development of methodology: Y. Lu, X. Guo, H.K. Im, C.F. Singer, W. Zheng

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Lu, L. Wu, J.M. Schildkraut, B.M. Reid, K.B. Moysich, I.L. Andrulis, H. Anton-Culver, B.K. Arun, E.V. Bandera, R.B. Barkardottir, J. Benitez, L. Bjorge, J. Brenton, R. Butzow, T. Caldes, M.A. Caligo, I.G. Campbell, J. Chang-Claude, K.B.M. Claes, F.J. Couch, D.W. Cramer, M.B. Daly, A. DeFazio, O. Diez, S.M. Domchek, T. Dork, D.F. Easton, D.M. Eccles, P.A. Fasching, R.T. Fortner, G. Fountzilas, E. Friedman, P.A. Ganz, J. Garber, G.G. Giles, A.K. Godwin, D.E. Goldgar, M.T. Goodman, M.H. Greene, J. Gronwald, U. Hamann, F. Heitz, M.A.T. Hildebrandt, C.K. Høgdall, A. Hollestelle, P.J. Hulick, E.N. Imyanitov, C. Isaacs, A. Jakubowska, B.Y. Karlan, L.E. Kelemen, L.A. Kiemeney, S.K. Kjaer, A. Kwong, N.D. Le, F. Lesueur, D.A. Levine, A. Mattiello, T. May, I.A. McNeish, F. Modugno, M. Montagna, S.L. Neuhausen, H. Nevanlinna, L. Nikitina-Zake, R.L. Nussbaum, K. Offit, E. Olah, O.I. Olopade, S.H. Olson, H. Olsson, A. Osorio, S.K. Park, P.H.M. Peeters, T. Pejovic, P. Peterlongo, C.M. Phelan, M.A. Pujana, S.J. Ramus, G. Rennert, H. Risch, G.C. Rodriguez, C. Rodríguez-Antona, M.A. Rookus, M.A. Rossing, I.K. Rzepecka, D.P. Sandler, R.K. Schmutzler, V.W. Setiawan, P. Sharma, W. Sieh, J. Simard, C.F. Singer, H. Song, M.C. Southey, A.B. Spurdle, R. Sutphen, A.J. Swerdlow, M.R. Teixeira, S.H. Teo, M. Thomassen, M. Tischkowitz, A. Trichopoulou, N. Tung, S.S. Tworoger, E.J. van Rensburg, A. Vega, D.V. Edwards, P.M. Webb, J.N. Weitzel, N. Wentzensen, E. White, A. Wolk, A.H. Wu, D. Yannoukakos, K.K. Zorn, S.A. Gayther, A.C. Antoniou, E.L. Goode, G. Chenevix-Trench, T.A. Sellers, P.D.P. Pharoah, W. Zheng

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Lu, A. Beeghly-Fadiel, L. Wu, X. Guo, B. Li, Y.A. Chen, J.K. Teer, K.B. Moysich, D.R. Barnes, D.R. Barnes, S.M. Domchek, R.T. Fortner, J. Gronwald, N.D. Le, K. Offit, H. Risch, G.C. Rodriguez, I. Romieu, W. Sieh, C.F. Singer, M.C. Southey, A.E. Toland, N. Tung, A. Vanderstichele, N. Wentzensen, A.C. Antoniou, A. Berchuck, E.L. Goode, W. Zheng

Writing, review, and/or revision of the manuscript: Y. Lu, A. Beeghly-Fadiel, L. Wu, B. Li, H.K. Im, J.B. Permuth, B.M. Reid, J.K. Teer, K.B. Moysich, I.L. Andrulis, H. Anton-Culver, B.K. Arun, E.V. Bandera, R.B. Barkardottir, J. Benitez, L. Bjorge, J. Brenton, I.G. Campbell, J. Chang-Claude, K.B.M. Claes, F.J. Couch, D.W. Cramer, M.B. Daly, A. DeFazio, O. Diez, S.M. Domchek, T. Dork, D.F. Easton, D.M. Eccles, P.A. Fasching, R.T. Fortner, G. Fountzilas, E. Friedman, P.A. Ganz, J. Garber, G.G. Giles, A.K. Godwin, M.T. Goodman, M.H. Greene, J. Gronwald, U. Hamann, F. Heitz, C.K. Høgdall, A. Hollestelle, P.J. Hulick, E.N. Imyanitov, C. Isaacs, A. Jakubowska, P. James, B.Y. Karlan, L.E. Kelemen, L.A. Kiemeney, S.K. Kjaer, A. Kwong, N.D. Le, A. Mattiello, T. May, M.A. Merritt, F. Modugno, S.L. Neuhausen, R.L. Nussbaum, E. Olah, O.I. Olopade, H. Olsson, A. Osorio, S.K. Park, P.H.M. Peeters, S.J. Ramus, G. Rennert, H. Risch, G.C. Rodriguez, I. Romieu, M.A. Rossing, D.P. Sandler, R.K. Schmutzler, V.W. Setiawan, P. Sharma, W. Sieh, J. Simard, C.F. Singer, H. Song, M.C. Southey, A.B. Spurdle, R. Sutphen, A.J. Swerdlow, S.H. Teo, M. Tischkowitz, A.E. Toland, A. Trichopoulou, N. Tung, S.S. Tworoger, E.J. van Rensburg, A. Vega, P.M. Webb, J.N. Weitzel, N. Wentzensen, E. White, A. Wolk, A.H. Wu, D. Yannoukakos, S.A. Gayther, A.C. Antoniou, A. Berchuck, E.L. Goode, G. Chenevix-Trench, T.A. Sellers, P.D.P. Pharoah, W. Zheng, J. Long

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Lu, L. Wu, K.B. Moysich, J. Dennis, T. Dork, M.A.T. Hildebrandt, B.Y. Karlan, L.E. Kelemen, L.A. Kiemeney, G. Leslie, F. Lesueur, L. McGuffog, F. Modugno, K. Offit, O.I. Olopade, M.T. Parsons, S.J. Ramus, M.A. Rookus, P. Sharma, M.C. Southey, J.N. Weitzel, E. White, A.C. Antoniou, A. Berchuck, W. Zheng

Study supervision: P.J. Hulick, D.G. Huntsman, L.A. Kiemeney, C.F. Singer, M.C. Southey, R. Sutphen, S.A. Gayther, W. Zheng, J. Long

Other (variant classification): A.B. Spurdle

The Ovarian Cancer Association Consortium is supported by a grant from the Ovarian Cancer Research Fund thanks to donations by the family and friends of Kathryn Sladek Smith (PPD/RPCI.07). The scientific development and funding for this project were in part supported by the US National Cancer Institute GAME-ON Post-GWAS Initiative (U19-CA148112). This study made use of data generated by the Wellcome Trust Case Control consortium that was funded by the Wellcome Trust under award 076113. The results published here are in part based upon data generated by The Cancer Genome Atlas Pilot Project established by the National Cancer Institute and National Human Genome Research Institute (dbGap accession number phs000178.v8.p7). The OCAC OncoArray genotyping project was funded through grants from the NIH [U19-CA148112 (to T.A. Sellers), R01-CA149429 (to C.M. Phelan), and R01-CA058598 (to M.T. Goodman); Canadian Institutes of Health Research (MOP-86727 (to L.E. Kelemen)] and the Ovarian Cancer Research Fund (to A. Berchuck). The COGS project was funded through a European Commission's Seventh Framework Programme grant (agreement number 223175 - HEALTH-F2-2009-223175). Funding for individual studies: AAS: NIH (RO1-CA142081); AOV: The Canadian Institutes for Health Research (MOP-86727); AUS: The Australian Ovarian Cancer Study Group was supported by the U.S. Army Medical Research and Materiel Command (DAMD17-01-1-0729), National Health & Medical Research Council of Australia (199600, 400413 and 400281), Cancer Councils of New South Wales, Victoria, Queensland, South Australia and Tasmania and Cancer Foundation of Western Australia (Multi-State Applications 191, 211, and 182). The Australian Ovarian Cancer Study gratefully acknowledges additional support from Ovarian Cancer Australia and the Peter MacCallum Foundation; BAV: ELAN Funds of the University of Erlangen-Nuremberg; BEL: National Kankerplan; BGS: Breast Cancer Now, Institute of Cancer Research; BVU: Vanderbilt CTSA grant from the National Institutes of Health (NIH)/National Center for Advancing Translational Sciences (NCATS) (ULTR000445); CAM: National Institutes of Health Research Cambridge Biomedical Research Centre and Cancer Research UK Cambridge Cancer Centre; CHA: Innovative Research Team in University (PCSIRT) in China (IRT1076); CNI: Instituto de Salud Carlos III (PI 12/01319); Ministerio de Economía y Competitividad (SAF2012); COE: Department of Defense (W81XWH-11-2-0131); CON: National Institutes of Health (R01-CA063678, R01-CA074850; R01-CA080742); DKE: Ovarian Cancer Research Fund and National Institutes of Health 1R01CA211574 (J.M. Schildkraut); DOV: National Institutes of Health R01-CA112523 and R01-CA87538; EMC: Dutch Cancer Society (EMC 2014-6699); EPC: The coordination of EPIC is financially supported by the European Commission (DG-SANCO) and the International Agency for Research on Cancer. The national cohorts are supported by Danish Cancer Society (Denmark); Ligue Contre le Cancer, Institut Gustave Roussy, Mutuelle Générale de l'Education Nationale, Institut National de la Santé et de la Recherche Médicale (INSERM) (France); German Cancer Aid, German Cancer Research Center (DKFZ), Federal Ministry of Education and Research (BMBF) (Germany); the Hellenic Health Foundation (Greece); Associazione Italiana per la Ricerca sul Cancro-AIRC-Italy and National Research Council (Italy); Dutch Ministry of Public Health, Welfare and Sports (VWS), Netherlands Cancer Registry (NKR), LK Research Funds, Dutch Prevention Funds, Dutch ZON (Zorg Onderzoek Nederland), World Cancer Research Fund (WCRF), Statistics Netherlands (The Netherlands); ERC-2009-AdG 232997 and Nordforsk, Nordic Centre of Excellence programme on Food, Nutrition and Health (Norway); Health Research Fund (FIS), PI13/00061 to Granada, PI13/01162 to EPIC-Murcia, Regional Governments of Andalucía, Asturias, Basque Country, Murcia and Navarra, ISCIII RETIC (RD06/0020) (Spain); Swedish Cancer Society, Swedish Research Council and County Councils of Skåne and Västerbotten (Sweden); Cancer Research UK (14136 to EPIC-Norfolk; C570/A16491 and C8221/A19170 to EPIC-Oxford), Medical Research Council (1000143 to EPIC-Norfolk, MR/M012190/1 to EPIC-Oxford) (United Kingdom); GER: German Federal Ministry of Education and Research, Programme of Clinical Biomedical Research (01 GB 9401) and the German Cancer Research Center (DKFZ); GRC: This research has been co-financed by the European Union (European Social Fund - ESF) and Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program of the General Secretariat for Research & Technology: SYN11_10_19 NBCA. Investing in knowledge society through the European Social Fund; GRR: Roswell Park Cancer Institute Alliance Foundation, P30 CA016056; HAW: U.S. National Institutes of Health (R01-CA58598, N01-CN-55424 and N01-PC-67001); HJO: Intramural funding; Rudolf-Bartling Foundation; HMO: Intramural funding; Rudolf-Bartling Foundation; HOC: Helsinki University Research Fund; HOP: Department of Defense (DAMD17-02-1-0669) and NCI (K07-CA080668, R01-CA95023, P50-CA159981 MO1-RR000056 R01-CA126841); HUO: Intramural funding; Rudolf-Bartling Foundation; JGO: JSPS KAKENHI grant; JPN: Grant-in-Aid for the Third Term Comprehensive 10-Year Strategy for Cancer Control from the Ministry of Health, Labour and Welfare; KRA: This study (Ko-EVE) was supported by a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), and the National R&D Program for Cancer Control, Ministry of Health & Welfare, Republic of Korea (HI16C1127; 0920010); LAX: American Cancer Society Early Detection Professorship (SIOP-06-258-01-COUN) and the National Center for Advancing Translational Sciences (NCATS), Grant UL1TR000124; LUN: ERC-2011-AdG 294576-risk factors cancer, Swedish Cancer Society, Swedish Research Council, Beta Kamprad Foundation; MAC: NIH (R01-CA122443, P30-CA15083, P50-CA136393); Mayo Foundation; Minnesota Ovarian Cancer Alliance; Fred C. and Katherine B. Andersen Foundation; Fraternal Order of Eagles; MAL: Funding for this study was provided by research grant R01- CA61107 from the National Cancer Institute, Bethesda, MD, research grant 94 222 52 from the Danish Cancer Society, Copenhagen, Denmark; and the Mermaid I project; MAS: Malaysian Ministry of Higher Education (UM.C/HlR/MOHE/06) and Cancer Research Initiatives Foundation; MAY: National Institutes of Health (R01-CA122443, P30-CA15083, P50-CA136393); Mayo Foundation; Minnesota Ovarian Cancer Alliance; Fred C. and Katherine B. Andersen Foundation; MCC: Cancer Council Victoria, National Health and Medical Research Council of Australia (NHMRC) grants number 209057, 251533, 396414, and 504715; MDA: DOD Ovarian Cancer Research Program (W81XWH-07-0449); MEC: NIH (CA54281, CA164973, CA63464); MOF: Moffitt Cancer Center, Merck Pharmaceuticals, the state of Florida, Hillsborough County, and the city of Tampa; NCO: National Institutes of Health (R01-CA76016) and the Department of Defense (DAMD17-02-1-0666); NEC: National Institutes of Health R01-CA54419 and P50-CA105009 and Department of Defense W81XWH-10-1-02802; NHS: UM1 CA186107, P01 CA87969, R01 CA49449, R01-CA67262, UM1 CA176726; NJO: National Cancer Institute (NIH-K07 CA095666, R01-CA83918, NIH-K22-CA138563, P30-CA072720, and P30-CA008748) and the Cancer Institute of New Jersey; NOR: Helse Vest, The Norwegian Cancer Society, The Research Council of Norway; NTH: Radboud University Medical Centre; OPL: National Health and Medical Research Council (NHMRC) of Australia (APP1025142) and Brisbane Women's Club; ORE: OHSU Foundation; OVA: This work was supported by Canadian Institutes of Health Research grant (MOP-86727) and by NIH/NCI 1 R01CA160669-01A1; PLC: Intramural Research Program of the National Cancer Institute; POC: Pomeranian Medical University; POL: Intramural Research Program of the National Cancer Institute; PVD: Canadian Cancer Society and Cancer Research Society GRePEC Program; RBH: National Health and Medical Research Council of Australia; RMH: Cancer Research UK, Royal Marsden Hospital; RPC: National Institute of Health (P50 CA159981, R01CA126841); SEA: Cancer Research UK (C490/A10119 C490/A10124); UK National Institute for Health Research Biomedical Research Centres at the University of Cambridge; SIS: NIH, National Institute of Environmental Health Sciences, Z01 ES044005 and Z01-ES049033; SMC: The Swedish Research Council; SON: National Health Research and Development Program, Health Canada, grant 6613-1415-53; SRO: Cancer Research UK (C536/A13086, C536/A6689) and Imperial Experimental Cancer Research Centre (C1312/A15589); STA: NIH grants U01 CA71966 and U01 CA69417; SWE: Swedish Cancer foundation, WeCanCureCancer and årKampMotCancer foundation; SWH: NIH (NCI) grant R37-CA070867; TBO: National Institutes of Health (R01-CA106414-A2), American Cancer Society (CRTG-00-196-01-CCE), Department of Defense (DAMD17-98-1-8659), Celma Mastery Ovarian Cancer Foundation; TOR: NIH grants R01 CA063678 and R01 CA063682; UCI: NIH R01-CA058860 and the Lon V Smith Foundation grant LVS-39420; UHN: Princess Margaret Cancer Centre Foundation-Bridge for the Cure; UKO: The UKOPS study was funded by The Eve Appeal (The Oak Foundation) and supported by the National Institute for Health Research University College London Hospitals Biomedical Research Centre; UKR: Cancer Research UK (C490/A6187), UK National Institute for Health Research Biomedical Research Centres at the University of Cambridge; USC: P01CA17054, P30CA14089, R01CA61132, N01PC67010, R03CA113148, R03CA115195, N01CN025403, and California Cancer Research Program (00-01389V-20170, 2II0200); VAN: BC Cancer Foundation, VGH & UBC Hospital Foundation; VTL: NIH K05-CA154337; WMH: National Health and Medical Research Council of Australia, Enabling Grants ID 310670 & ID 628903. Cancer Institute NSW Grants 12/RIG/1-17 & 15/RIG/1-16; WOC: National Science Centren (N N301 5645 40). The Maria Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Warsaw, Poland. We are grateful to the family and friends of Kathryn Sladek Smith for their generous support of the Ovarian Cancer Association Consortium through their donations to the Ovarian Cancer Research Fund. The OncoArray and COGS genotyping projects would not have been possible without the contributions of the following: Per Hall (COGS); Kyriaki Michailidou, Manjeet K. Bolla, Qin Wang (BCAC), Marjorie J. Riggan (OCAC), Rosalind A. Eeles, Ali Amin Al Olama, Zsofia Kote-Jarai, Sara Benlloch (PRACTICAL), Jonathan P. Tyrer, Siddhartha Kar, Alison M. Dunning, Andrew Lee, and Ed Dicks, Craig Luccarini and the staff of the Centre for Genetic Epidemiology Laboratory, Anna Gonzalez-Neira and the staff of the CNIO genotyping unit, Daniel C. Tessier, Francois Bacot, Daniel Vincent, Sylvie LaBoissière and Frederic Robidoux and the staff of the McGill University and Génome Québec Innovation Centre, Stig E. Bojesen, Sune F. Nielsen, Borge G. Nordestgaard and the staff of the Copenhagen DNA laboratory, and Julie M. Cunningham, Sharon A. Windebank, Christopher A. Hilker, Jeffrey Meyer and the staff of Mayo Clinic Genotyping Core Facility. We pay special tribute to the contribution of Professor Brian Henderson to the GAME-ON consortium and to Olga M. Sinilnikova for her contribution to CIMBA and for her part in the initiation and coordination of GEMO until she sadly passed away on the 30th June 2014. We thank the study participants, doctors, nurses, clinical and scientific collaborators, health care providers and health information sources who have contributed to the many studies contributing to this manuscript.

Acknowledgements for individual studies: AOV: We thank Jennifer Koziak, Mie Konno, Michelle Darago, Faye Chambers and the Tom Baker Cancer Centre Translational Laboratories; AUS: The AOCS also acknowledges the cooperation of the participating institutions in Australia and acknowledges the contribution of the study nurses, research assistants and all clinical and scientific collaborators to the study. The complete AOCS Study Group can be found at www.aocstudy.org. We would like to thank all of the women who participated in these research programs; BEL: We would like to thank Gilian Peuteman, Thomas Van Brussel, Annick Van den Broeck and Joke De Roover for technical assistance; BGS: The BGS is funded by Breast Cancer Now and the Institute of Cancer Research (ICR). ICR acknowledges NHS funding to the NIHR Biomedical Research Centre. We thank the Study staff, study participants, doctors, nurses, health care providers and health information sources who have contributed to the study; BVU: The dataset(s) used for the analyses described were obtained from Vanderbilt University Medical Center's BioVU, which is supported by institutional funding, the 1S10RR025141-01 instrumentation award, and by the Vanderbilt CTSA grant UL1TR000445 from NCATS/NIH; CAM: This work was supported by Cancer Research UK; the University of Cambridge; National Institute for Health Research Cambridge Biomedical Research Centre; CHA: Innovative Research Team in University (PCSIRT) in China (IRT1076); CHN: To thank all members of Department of Obstetrics and Gynaecology, Hebei Medical University, Fourth Hospital and Department of Molecular Biology, Hebei Medical University, Fourth Hospital; COE: Gynecologic Cancer Center of Excellence (W81XWH-11-2-0131); CON: The cooperation of the 32 Connecticut hospitals, including Stamford Hospital, in allowing patient access, is gratefully acknowledged. This study was approved by the State of Connecticut Department of Public Health Human Investigation Committee. Certain data used in this study were obtained from the Connecticut Tumor Registry in the Connecticut Department of Public Health. The authors assume full responsibility for analyses and interpretation of these data; DKE: OCRF; EPC: To thank all members and investigators of the Rotterdam Ovarian Cancer Study. Dutch Cancer Society (EMC 2014-6699); GER: The German Ovarian Cancer Study (GER) thank Ursula Eilber for competent technical assistance; HOC: The study was supported by the Helsinki University Research Fund; JGO: JSPS KAKENHI grant; KRA: This study (Ko-EVE) was supported by a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), and the National R&D Program for Cancer Control, Ministry of Health & Welfare, Republic of Korea (HI16C1127; 0920010); LUN: ERC -2011-AdG, Swedish Cancer Society, Swedish Research Council; MAS: We would like to thank Famida Zulkifli and Ms Moey for assistance in patient recruitment, data collection and sample preparation. The Malaysian Ovarian Cancer Genetic Study is funded by research grants from the Malaysian Ministry of Higher Education (UM.C/HIR/MOHE/06) and charitable funding from Cancer Research Initiatives Foundation; MCC: MCCS cohort recruitment was funded by VicHealth and Cancer Council Victoria. The MCCS was further supported by Australian NHMRC grants 209057, 251553 and 504711 and by infrastructure provided by Cancer Council Victoria. Cases and their vital status were ascertained through the Victorian Cancer Registry (VCR) and the Australian Institute of Health and Welfare (AIHW), including the National Death Index and the Australian Cancer Database; MOF: the Total Cancer Care Protocol and the Collaborative Data Services and Tissue Core Facilities at the H. Lee Moffitt Cancer Center & Research Institute, an NCI designated Comprehensive Cancer Center (P30-CA076292), Merck Pharmaceuticals and the state of Florida; NHS: The NHS/NHSII studies thank the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, and WY; NJO: L. Paddock, M. King, U. Chandran, A. Samoila, and Y. Bensman; OPL: Members of the OPAL Study Group (http://opalstudy.qimrberghofer.edu.au/); RPC: National Institute of Health (P50 CA159981, R01CA126841); SEA: SEARCH team, Craig Luccarini, Caroline Baynes, Don Conroy; SIS: The Sister Study (SISTER) is supported by the Intramural Research Program of the NIH, National Institute of Environmental Health Sciences (Z01-ES044005 and Z01-ES049033); SON: National Health Research and Development Program, Health Canada, grant 6613-1415-53; SRO: To thank all members of Scottish Gynaecological Clinical Trails group and SCOTROC1 investigators; SWE: Swedish Cancer Foundation, WeCanCureCancer and årKampMotCancer Foundation; SWH: The SWHS is supported primarily by NIH grant R37-CA070867. We thank the participants and the research staff of the Shanghai Women's Health Study for making this study possible; UCI: The UCI Ovarian Cancer Study is supported by the National Institutes of Health, National Cancer Institute grants CA58860, and the Lon V Smith Foundation grant LVS-39420; UHN: Princess Margaret Cancer Centre Foundation-Bridge for the Cure; UKO: We particularly thank I. Jacobs, M.Widschwendter, E. Wozniak, A. Ryan, J. Ford and N. Balogun for their contribution to the study; UKR: Carole Pye; VAN: BC Cancer Foundation, VGH & UBC Hospital Foundation; WMH: We thank the Gynaecological Oncology Biobank at Westmead, a member of the Australasian Biospecimen Network-Oncology group, which is funded by the National Health and Medical Research Council Enabling Grants ID 310670 & ID 628903 and the Cancer Institute NSW Grants 12/RIG/1-17 & 15/RIG/1-16.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.