Background: We previously reported the risks of ovarian carcinoma for common polymorphisms in one-carbon transfer genes. We sought to replicate associations for DPYD rs1801265, DNMT3A rs13420827, MTHFD1 rs1950902, MTHFS rs17284990, and TYMS rs495139 with risk of ovarian carcinoma overall and to use the large sample of assembled cases to investigate associations by histologic type.
Methods: Associations were evaluated in the Ovarian Cancer Association Consortium, including 16 studies of 5,593 epithelial ovarian carcinoma cases and 9,962 controls of white non-Hispanic origin. Odds ratios (OR) and 95% confidence intervals (CI) were adjusted for age and study site.
Results: The five polymorphisms were not associated with ovarian carcinoma overall (Ptrend > 0.13); however, associations for the minor allele at TYMS rs495139 were observed for carcinomas of mucinous type (OR, 1.19; 95% CI, 1.03-1.39; P = 0.02), clear cell type (OR, 0.86; 95% CI, 0.75-0.99; P = 0.04), and endometrioid type (OR, 0.90; 95% CI, 0.81-0.99; P = 0.04; Pheterogeneity = 0.001). Restriction to low-grade mucinous carcinomas further strengthened the association for the mucinous type (OR, 1.32; 95% CI, 1.07-1.62; P = 0.01). TYMS rs495139 was not associated with serous type (OR, 1.06; 95% CI, 1.00-1.13; P = 0.05).
Conclusions: TYMS rs495139 may be associated with a differential risk of ovarian carcinoma types, indicating the importance of accurate histopathologic classification.
Impact: Biomarkers that distinguish ovarian carcinoma types are few, and TYMS rs495139 may provide a novel clue to type etiology. Cancer Epidemiol Biomarkers Prev; 19(7); 1822–30. ©2010 AACR.
One-carbon (1-C) transfer reactions are essential for DNA synthesis and replication, particularly for rapidly dividing cells, as well as for the biosynthesis of S-adenosyl methionine, an essential supplier of methyl groups for the methylation of many compounds including DNA (1). Perturbation of gene expression and gene product function in the 1-C transfer pathway can have pleiotropic consequences, leading to tumor initiation and progression (2). Incessant demand for DNA synthesis and preservation of DNA integrity through methylation on a genetically susceptible background may possibly increase the risk of ovarian carcinomas.
We previously reported that genetic variation in the 1-C transfer pathway was associated with ovarian carcinoma risk among cases and controls from the upper Midwest and North Carolina (3). Ten common nonsynonymous and tagging single nucleotide polymorphisms (SNP) in eight genes were statistically significant at P ≤ 0.05 in either an ordinal (per minor allele) model or codominant model comparing heterozygotes and homozygotes for the minor allele separately to homozygotes with the common allele. In the current report, our first aim was to replicate the findings of five SNPs from our U.S. study (3) with risk of ovarian carcinomas using data from the international Ovarian Cancer Association Consortium (OCAC; ref. 4). The five SNPs selected for follow-up genotyping in OCAC were chosen from preliminary analyses before publication of the final report (3) with consideration to available funds to assay ∼19,500 samples from among several promising SNPs that were nominated for genotyping by other OCAC members. We weighted our decision for which five SNPs to genotype using the criteria of statistical significance from the preliminary analyses and the known biology of the enzymes' pivotal roles at critical junctions in 1-C transfer. At that time, our preliminary analyses did not identify what would become our most promising SNP (SHMT1 rs9909104) with ovarian carcinoma risk (3), and it explains its absence in this report.
Molecular and genetic-based analyses of ovarian carcinomas show that ovarian cancer is several diseases with different patterns of genetic mutations (5), biological markers (6), survival outcomes (7), and cells of origin (8). Recent advances in histopathologic typing, based on distinct molecular alterations, have led to more accurate classification of ovarian carcinoma types (5, 6), resulting in a lower prevalence for mucinous and endometrioid ovarian carcinomas than previously thought (9). Associations in these rare types might have been diluted in previous analyses due to nondifferential misclassification. Using the large sample size of the OCAC, our second aim was to evaluate associations at the five 1-C SNPs with histologic types of ovarian carcinomas. For this aim, we considered information on grade and histology (10, 11) to “reassign” the histologic types to correct for potential misclassification.
Materials and Methods
Sixteen studies of ovarian cancer contributed data to this analysis and are described in Table 1 (see also refs. 4, 12, 13). Thirteen studies used population-based ascertainment for cases and controls; one study was clinic-based; and one was a case-control study nested within a cohort. One population-based study, the North Carolina Ovarian Cancer Study (NCO), was evaluated in two batches: NCO samples 0001 to 1040 (henceforth called NCO1) were included in our original report along with the Mayo Clinic Ovarian Cancer Case Control Study (MAY) samples from which initial observations for the five SNPs of interest were made (3). NCO samples 1041 to 1771 (henceforth called NCO2) were genotyped in the current replication investigation. Thus, 14 studies (including NCO2) served as replication studies, and two studies (NCO1 and MAY) were included from our original report.
Each study received ethics committee approval, and all study subjects provided written informed consent. Key clinical and questionnaire data on study subjects including case-control status, ethnicity/race, tumor behavior, histology, age at diagnosis (or comparable reference date for controls), and history of prior cancers were merged into a common data set. The data were checked for consistency and completeness, and discrepancies were followed-up with individual study investigators.
We excluded subjects with missing information on age and tumor behavior, subjects with nonepithelial ovarian tumors, and those with a prior history of ovarian cancer.
Genotyping and quality control
The five SNPs that were assayed were as follows: dihydropyrimidine dehydrogenase (DPYD) rs1801265, DNA (cytosine-5-)-methyltransferase 3 α (DNMT3A) rs13420827, methylenetetrahydrofolate dehydrogenase (NADP+ dependent) 1 (MTHFD1) rs1950902, 5,10-methenyltetrahydrofolate synthetase (5-formyltetrahydrofolate cyclo-ligase) (MTHFS) rs17284990, and thymidylate synthetase (TYMS) rs495139. Genotyping of the 14 replication studies was done on 384-well plates using a semicentralized approach with centrally supplied probes at 11 different centers: the Malignant Ovarian Cancer Study (MAL) and the United Kingdom Ovarian Cancer Population Study (UKO) were genotyped in the same laboratory, as were the Diseases of the Ovary and their Evaluation (DOV), the Hormones and Ovarian Cancer Prediction Study (HOP) and the Los Angeles County Case-Control Studies of Ovarian Cancer (USC) studies, and the Studies of Epidemiology and Risk Factors in Cancer Heredity-Ovarian Cancer (SEA) and the Genetic Epidemiology of Ovarian Cancer Study (STA). All samples except Australian Ovarian Cancer Study and Australian Cancer Study-Ovarian Cancer (AUS) were genotyped using the 5′ nuclease Taqman allelic discrimination assay (Taqman, Applied Biosystems). The Australian Cancer Study-Ovarian Cancer used the Sequenom iPLEX gold genotyping technology (Sequenom, Inc.). Each assay was carried out using 10 ng DNA in a 2.5 or 5-μL reaction volume as previously described (4).
The following criteria were used as measures of acceptable genotyping for each SNP and each study: (a) >3% sample duplicates included, (b) concordance for duplicate samples of ≥98%, (c) overall SNP call rate by study of ≥95%, (d) call rate for each 384-well plate of >90%, (e) a difference in call rate between cases and controls of <5%, and (f) <25% overall failed plates. Studies failing one of these criteria were excluded for particular SNPs: for example, Nurses' Health Study (NHS) was excluded for DPYD; MAL, the New England–based Case-Control Study (NEC), NHS, and UKO were excluded for MTHFS; and MAL was excluded for TYMS. Therefore, the number of studies/samples successfully genotyped varied for each polymorphism. Further, among white non-Hispanic control subjects, genotypes were compared with those expected under Hardy-Weinberg equilibrium in each study separately, and no deviations from Hardy-Weinberg equilibrium among controls were observed at P < 0.05. In addition, consistency across laboratories was confirmed by genotyping a common set of 95 DNAs (90 CEPH trios and five duplicate samples; HAPMAPPT01 provided by Coriell) with the requirement of >98% concordance in genotype calls. Five SNPs across four studies had one mismatch genotype for the HAPMAPPT01 (>99.9% call rate overall; range across studies, 99.9-100%). Finally, to evaluate genotype consistency across discovery and replication sets, NCO1 samples (n = 1,040) were regenotyped using the replication assay. Genotype call rate concordance between Illumina (discovery set) and Taqman (replication set) assays was very high (DNMT3A = 99.5%, DPYD = 99.3%, MTHFS = 99.8%, MTHFD1 = 99.2%, and TYMS = 99.0%).
Primary analyses were restricted to controls and ovarian carcinomas (invasive cases) among white non-Hispanic subjects in the 14 replication studies. SNP associations for ovarian carcinoma risk were assessed using unconditional logistic regression to estimate odds ratios (OR) and 95% confidence intervals (CI). Association testing assumed an ordinal (log additive) genotypic relationship with simple tests for trend, as well as separate comparisons of women with one copy and two copies of the minor allele to women with no copies (reference) using a two degrees-of-freedom test. Risk models were adjusted for age category (<40, 40-49, 50-59, 60-69, ≥70 y) in study-specific analyses, and adjusted for age category and study in pooled analyses. Before pooling data across studies, a test for heterogeneity of ORs was evaluated for significance using the likelihood ratio test comparing models with and without a product term for the ordinal coding of the genotype and the categorical variable for study (14). Pooled ORs and 95% CIs are presented with and without including the two studies (MAY and NCO1) in which the initial findings were generated.
Among the 16 studies, we simultaneously modeled the risk of each of four histologic types of epithelial ovarian carcinomas (serous, mucinous, endometrioid, and clear cell) under an ordinal genetic model using polytomous logistic regression. Risk models were adjusted for age category and study, and statistical heterogeneity of the SNP-ovarian carcinoma histology associations was tested (14). We also incorporated information from contemporary pathologic reviews to refine risk associations in the analyses of histologic type. Specifically, others have shown that a significant proportion of grade 3 mucinous ovarian carcinomas are, in fact, metastatic from the gastrointestinal tract (15), and up to 28% of endometrioid ovarian carcinomas are reclassified as high-grade serous ovarian carcinomas after contemporary histopathologic review (10). We, therefore, reassigned histologic type according to the expected distributions of histology combined with grade observed from rereviews of >1,000 ovarian carcinomas from large population-based series (10, 11). The reclassifications used were as follows: serous carcinomas (serous histology or ≥G3 + endometrioid histology) and metastatic mucinous carcinomas (≥G3 + mucinous histology). No reassignment was done for clear cell carcinomas.
For the first aim, two-sided P values of <0.05 were considered to be statistically significant for replication risk estimates with ovarian carcinomas overall. For the second aim, two-sided P values of <0.01 were identified as statistically significant associations after Bonferroni correction of the type 1 error rate for the five SNP by histologic cell type heterogeneity tests. Analyses were implemented using SAS (version 9.1, 2009, SAS Institute).
The characteristics of the 16 studies are shown in Table 1. The final data set including MAY and NCO1 studies comprised 6,583 ovarian carcinoma cases and 11,215 controls. Of these, 87% were of white non-Hispanic origin (5,593 cases, 9,962 controls). Other subjects were white Hispanic (169 cases and 189 controls), black non-Hispanic (134 cases and 199 controls), various Asian ethnicities (241 cases and 327 controls), and other races or ethnicities (198 cases and 371 controls); 248 cases and 167 controls had missing or unknown race or ethnicity information. An additional 1,634 cases had borderline/low malignant potential (LMP) tumors, of which 1,307 were of white non-Hispanic origin. Genotype counts, minor allele frequency (MAF), and Hardy-Weinberg equilibrium statistics for each SNP among white non-Hispanic women with ovarian carcinoma (invasive cases only) in each study population are shown in Supplementary Table S1.
None of the five SNPs evaluated in the 14 replication studies was associated with ovarian carcinoma overall (Table 2; Fig. 1). No statistical evidence for heterogeneity of ORs was observed across replication studies (Fig. 1). Study-specific ORs and 95% CIs are in Supplementary Table S2.
Following Bonferroni adjustment, only TYMS rs495139 showed statistical heterogeneity of ORs across histologic types (Pheterogeneity = 0.001; Table 3) despite no significant association with ovarian carcinoma overall (Table 2). Each copy of the TYMS rs495139 minor allele was associated with an increased risk of mucinous ovarian carcinomas (OR, 1.19; 95% CI, 1.03-1.39; P = 0.02), and a decreased risk of endometrioid (OR, 0.90; 95% CI, 0.81-0.99; P = 0.04) and clear cell (OR, 0.86; 95% CI, 0.75-0.99; P = 0.04) ovarian carcinomas. The association with serous ovarian carcinomas (OR, 1.06; 95% CI, 1.00-1.13; P = 0.05) was not statistically significant. None of the other SNPs were associated differentially with histologic types (Ptumor heterogeneity > 0.30; data not shown).
To correct for potential misclassification of histologic type, a combination of grade and type was used to reassign type. Grade information was available on 2,903 of the 4,407 cases in the TYMS analyses (Table 3, second row). As shown in Table 3 (last row), the association among mucinous ovarian carcinomas strengthened (OR, 1.32; 95% CI, 1.07-1.62; P = 0.01; 183 cases) when 21 high-grade, and presumably metastatic, mucinous carcinomas were excluded (OR for these 21 cases, 0.88; 95% CI, 0.47-1.64; P = 0.69). The null association among serous carcinomas persisted with the inclusion of 197 high-grade endometrioid carcinomas, which were presumed to be misdiagnosed high-grade serous carcinomas (OR, 1.02; 95% CI, 0.95-1.09; P = 0.62; 1,972 cases). The decreased risk among endometrioid ovarian carcinomas attenuated and was no longer statistically significant; however, 306 endometrioid cases were lost from the analyses due to missing information on grade. TYMS rs495139 was not associated with LMP tumors of the different histologic types: serous LMP (OR, 1.01; 95% CI, 0.90-1.13; P = 0.92; 680 cases), mucinous LMP (OR, 0.99; 95% CI, 0.87-1.13; P = 0.88; 504 cases), endometrioid LMP (OR, 1.38; 95% CI, 0.74-2.54; P = 0.31; 20 cases), and clear cell LMP (OR, 1.13; 95% CI, 0.39-3.27; P = 0.83; 7 cases).
This investigation profited from the collaborative efforts of investigators in the international OCAC, which combined their samples into one of the largest series of ovarian cancer cases and controls to investigate associations of genetic susceptibility. Our findings do not support significant associations for DPYD rs1801265, DNMT3A rs13420827, MTHFD1 rs1950902, MTHFS rs17284990, and TYMS rs495139 with risk of ovarian carcinoma overall. When we considered the different histologic types separately, the risks at TYMS rs495139 were statistically heterogeneous, showing an increase in risk of mucinous ovarian carcinomas. This association was strengthened when we excluded putative metastatic carcinomas, suggesting the importance of homogeneous histopathological classification and the utility of our approach to correct for potential misclassification and refine risk associations in studies of ovarian carcinoma types. Decreased risks of clear cell and possibly endometrioid types were also observed. Our findings are reinforced by the application of rigorous quality control standards applied to our genotyping protocol and the centralized repository of key clinical variables from each study that underwent logic checks before merging into a common data set to ensure data integrity. To limit any effect of population stratification, we restricted our analysis to women of white non-Hispanic ancestry.
Histologic types of ovarian carcinoma are distinct diseases that vary according to the characteristics of the precursor lesions and the genetic events during oncogenesis (5), and distinct biomarker expression profiles (6). The progression of mucinous ovarian tumors is hypothesized to develop from benign cysts or LMP tumors rather than arising de novo (16). In support of the progression model of mucinous ovarian tumors is the coexistence of benign, LMP, and malignant areas within the same mucinous tumor (17) and the sharing of identical KRAS mutations (18). It is conceivable that genetic variants may have different functional relevance for different ovarian histologic types. In the present report, TYMS rs495139 was not associated with LMP tumors, but the positive association with mucinous ovarian carcinomas suggests that it may serve to drive the progression of LMP mucinous tumors to mucinous carcinomas. Although speculative, TYMS rs495139 may play a protective role in the early development of clear cell and endometrioid carcinomas, which share a similar precursor lesion and molecular features (19). Precisely how the SNP would differentially affect histologic types requires further investigation. The null association at TYMS rs495139 with the overall ovarian cancer phenotype observed in the current analysis is not surprising given that associations with the overall phenotype are often driven by the greater proportion of the serous ovarian carcinomas.
TYMS encodes thymidylate synthetase (TS), which catalyzes the transformation of dUMP to dTMP and is the only de novo source of thymidylate used for DNA (pyrimidine) biosynthesis (20). TS is present in proliferating cell types and is considered an important target in cancer chemotherapy (21). The TYMS rs495139 polymorphism is located in the 3′ untranslated region and is also situated in the 3′ region of the enolase superfamily member 1 (ENOSF1) gene. ENOSF1 encodes two proteins that downregulate TS expression (22, 23): one of these, rTS α, is an antisense transcript that binds, cleaves, and inactivates human TS RNA (24). Because the levels of rTS α vary between cell lines (22), the relative abundance between cell types could control the rate of TS expression and cell type proliferation. The antisense transcript function of ENOSF1 mRNA is suggestive of the role that microRNAs play in translational repression or transcript degradation. The 3′ untranslated region of TYMS and ENOSF1 both have predicted sites for binding several microRNA families (25-29), and altered expression of several microRNAs has been reported in ovarian carcinoma (both serous and unspecified type) compared with normal tissue (29-31). Further, predicted microRNA target sites (32) at TYMS contain at least two polymorphisms, rs699517 (MAF = 0.27) and rs16948421 (MAF = not available), among Caucasians in the HapMap database. The TYMS rs495139 variant investigated in the present report is in modest linkage disequilibrium (LD) with the microRNA target site SNP rs699517 (r2 = 0.24, HapMap Consortium release 27). We speculate that if TYMS rs495139 or a variant in LD with TYMS rs495139 influences the binding of one or more microRNAs or affects ENOSF1 antisense transcript binding and subsequently TS cleavage in different cell types, including ovarian histologic types, then this may be one potential mechanism by which genetic variation alters TS transcript regulation and, therefore, cellular proliferation. Interestingly, another SNP, rs3819102, located in the 3′ flanking region of TYMS and in an intron of the ENOSF1 gene was associated recently with endometrial cancer among Chinese women (33); however, among Caucasians, this SNP is rare (MAF = 0.025) and is not in LD with TYMS rs495139 (r2 = 0.02, HapMap Consortium release 27).
We were unable to confirm the initial observed associations between DPYD rs1801265, DNMT3A rs13420827, MTHFD1 rs1950902, MTHFS rs17284990, and TYMS rs495139 with risk of ovarian carcinoma overall in the OCAC. It remains possible that other variants in these genes with which these five SNPs were in strong LD in the samples from the original report may have shown an association with ovarian carcinoma had they been genotyped in OCAC. In selecting the best correlated variant for a single locus for follow-up genotyping, we relied, in part, on the criteria of statistical significance. However, it has been shown that variants with the strongest support in the original discovery stage can be replaced by other variants in high LD that have much stronger effects in replication studies (34). Although the samples in our initial report had reasonably good SNP coverage (r2 ≥ 0.80) of variation present in the HapMap Caucasian samples, this may have been offset by lower power to detect weak effects in our previous study (3), which may have led to SNPs not being selected for validation that, by themselves or in a combination with other variants, may influence the risk of ovarian carcinoma overall or by specific cell type. Thus, the absence of associations with the five SNPs examined in this report does not preclude other variants in these genes to be important determinants of ovarian carcinoma. Using an even larger number of samples within the OCAC, we intend to pursue further genotyping in several of the genes from our initial report (3) in a more comprehensive manner using SNPs from the 1000 Genomes project (35).
In summary, we did not find significant associations for DPYD rs1801265, DNMT3A rs13420827, MTHFD1 rs1950902, MTHFS rs17284990, and TYMS rs495139 with ovarian carcinoma overall, but our findings suggest differential risk with ovarian histologic types for a common TYMS polymorphism. To reduce the misclassification of ovarian histologic types in the absence of contemporary pathological rereview, epidemiologic research may consider combining information on grade and histology using the reclassification based on expected distributions of ovarian types proposed in this report.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
The Australian Ovarian Cancer Study Management Group (AOCS; D. Bowtell, G. Chenevix-Trench, A. deFazio, D. Gertig, A. Green, and P.M. Webb) thanks all the clinical and scientific collaborators (see http://www.aocstudy.org/) for their contribution. The Australian Cancer Study Management Group comprises A. Green, P. Parsons, N. Hayward, P.M. Webb, and D. Whiteman. The German Ovarian Cancer Study (GER or GOCS) thank Ursula Eilber and Tanja Koehler for competent technical assistance for the German Ovarian Cancer study. The United Kingdom Ovarian Cancer Population Study (UKOPS) study thank all members of the research team, including research nurses, research scientists, data entry personnel, and consultant gynaecological oncologists for their help in establishing the UKOPS case-control collection.
Grant Support: Ovarian Cancer Research Fund, provided by the family and friends of Kathryn Sladek Smith. L.E. Kelemen is supported by career awards from the Alberta Cancer Research Institute (project numbers 23905 and 24258) and the Canadian Institutes of Health Research (grant number MSH-87734). The Australian Ovarian Cancer Study (AOCS) and the Australian Cancer Study (ACS) was supported by the U.S. Army Medical Research and Materiel Command (DAMD17-01-1-0729), the Cancer Council Tasmania and Cancer Foundation of Western Australia (AOCS study), and the National Health and Medical Research Council of Australia (199600; ACS study). G. Chenevix-Trench and P.M. Webb are supported by the NHMRC of Australia. The Diseases of the Ovary and their Evaluation study (DOV) was supported by the U.S. NIH (CA87538, CA112523). The Hawaii Ovarian Cancer Study (HAW) group was supported by the U.S. NIH (R01 CA 58598, N01-CN-55424, N01-PC-67001). The German Ovarian Cancer Study (GER or GOCS) was supported by the German Federal Ministry of Education and Research of Germany, Programme of Clinical Biomedical Research (01 GB 9401), the genotyping in part by the state of Baden-Württemberg through Medical Faculty of the University of Ulm (P.685), and data management by the German Cancer Research Center. The MALOVA study (MAL) was supported by grants from Mermaid 1, The Danish Cancer Society, and U.S. NIH (RO1 CA 61107). The Mayo Clinic Ovarian Cancer Case-Control Study (MAY) was supported by U.S. NIH (R01 CA 122443, R01 CA 88868) and the Mayo Foundation. The NHS was supported by U.S. NIH (P01 CA87969, R01 CA49449). M.A. Gates is supported by U.S. NIH (T32 CA009001). The Stanford study (STA) was supported by U.S. NIH grants U01 CA71966 (A.S. Whittemore and V. McGuire), R01 CA16056 (R.A. DiCioccio), and U01 CA69417 for recruitment of controls by the Northern California Cancer Center. The Los Angeles County Case-Control Studies of Ovarian Cancer (USC) was supported by the California Cancer Research Program (00-01389V-20170, 2110200), U.S. NIH (CA14089, CA17054, CA61132, CA63464, N01-PC-67010, and R03-CA113148), and California Department of Health Services (subcontract 050-E8709) as part of its statewide cancer reporting program. The United Kingdom Ovarian Cancer Population Study (UKOPS) study was supported by the OAK Foundation. A portion of this work was done at UCLH/UCL within the “women's health theme” of the National Institute for Health Research University College London Hospitals/University College London Comprehensive Biomedical Research Centre supported by the Department of Health.
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.
Note: Supplementary data for this article are available at Cancer Epidemiology, Biomarkers & Prevention Online (http://cebp.aacrjournals.org/).
- Received December 29, 2009.
- Revision received March 29, 2010.
- Accepted April 5, 2010.