High-risk susceptibility genes explain <40% of the excess risk of familial ovarian cancer. Therefore, other ovarian cancer susceptibility genes are likely to exist. We have used a single nucleotide polymorphism (SNP)–tagging approach to evaluate common variants in 13 genes involved in cell cycle control—CCND1, CCND2, CCND3, CCNE1, CDK2, CDK4, CDK6, CDKN1A, CDKN1B, CDKN2A, CDKN2B, CDKN2C, and CDKN2D—and risk of invasive epithelial ovarian cancer. We used a two-stage, multicenter, case-control study. In stage 1, 88 SNPs that tag common variation in these genes were genotyped in three studies from the United Kingdom, United States, and Denmark (∼1,500 cases and 2,500 controls). Genotype frequencies in cases and controls were compared using logistic regression. In stage 2, eight other studies from Australia, Poland, and the United States (∼2,000 cases and ∼3,200 controls) were genotyped for the five most significant SNPs from stage 1. No SNP was significant in the stage 2 data alone. Using the combined stages 1 and 2 data set, CDKN2A rs3731257 and CDKN1B rs2066827 were associated with disease risk (unadjusted P trend = 0.008 and 0.036, respectively), but these were not significant after adjusting for multiple testing. Carrying the minor allele of these SNPs was found to be associated with reduced risk [OR, 0.91 (0.85–0.98) for rs3731257; and OR, 0.93 (0.87–0.995) for rs2066827]. In conclusion, we have found evidence that a single tagged SNP in both the CDKN2A and CDKN1B genes may be associated with reduced ovarian cancer risk. This study highlights the need for multicenter collaborations for genetic association studies. [Cancer Res 2007;67(7):3027–35]

Several genes have been identified that confer high-penetrance susceptibility to epithelial ovarian cancer. The main protagonists are BRCA1 and BRCA2. These genes are responsible for about half of all families containing two or more ovarian cancer cases in first-degree relatives and most families containing several ovarian cancer cases (≥3) or in which ovarian and breast cancers occur together (1, 2).

However, the known susceptibility genes explain <40% of the excess familial ovarian cancer risk (3). If other highly penetrant ovarian cancer susceptibility genes exist, they are likely to be rare. One possible explanation for the residual risks is the existence of several common but only moderately or low-penetrance susceptibility alleles in the population (4).

There have been several studies that have attempted to identify common variants in genes that may be associated with increased risk of ovarian cancer. Candidate genes have usually been selected based on biological plausibility. These include genes in pathways controlling steroid hormone metabolism (e.g., progesterone receptor, androgen receptor, CYP17, prohibitin; refs. 514), DNA repair (e.g., BRCA1, BRCA2, RAD51, MSH2; refs. 1517) and in putative oncogenes and tumor suppressor genes (TP53, RB1, HRAS1, STK15; refs. 1823). Thus far, reported positive associations include an increased ovarian cancer risk for two PGR haplotypes (8); a protective effect for the PGR promoter +331A allele in endometrioid ovarian tumors (9); an increased risk of borderline ovarian cancer associated with the Pro72Arg polymorphism in the TP53 gene (20); and an increased risk associated with the polymorphism in the STK15 oncogene (23). However, none of these associations are conclusive due to the small size of the initial studies, and there are no reports indicating that they have been validated in additional populations.

There are no published reports describing the association between variants in cell-cycle control genes and ovarian cancer susceptibility. In general, cancer is associated with a breakdown in the mechanisms that regulate cell division. In mammalian cells, cell division is controlled by the activity of cyclin-dependent kinases (CDK) and their essential activating coenzymes, the cyclins and CDK inhibitors (reviewed in refs. 24, 25). CDK activity is regulated on several levels, including cyclin synthesis and degradation, phosphorylation, and dephosphorylation. The interaction between CDKs and cyclins is tightly controlled to ensure an ordered progression through the cell cycle from G1 to S to G2. The events occurring before DNA synthesis are probably the most thoroughly understood. The key event in cell cycle regulation is transversion of the restriction (R) point late in G1, which is crucial to the cell's destiny toward division, differentiation, senescence, or apoptosis. It is believed that once the R point has been overcome, cell cycle progression occurs almost automatically. In the event of cellular stress, there are several proteins that can inhibit the cell cycle in G1; for example, in response to DNA damage, P53 accumulates in the cell and induces P21/CDKN1A–mediated inhibition of cyclin D/CDK.

Somatic alterations of genes involved in the G1 phase of the cell cycle, including the cyclins, CDKs, and CDK inhibitors, are common events in neoplastic development for multiple tumor types; but different cell cycle proteins seem to be targets for different cancers. A reason for this could be that although the basic mechanisms of proliferative control are identical in different tissues, the maintenance of normal cell-cycle control is also partly regulated in a tissue-specific manner. For ovarian cancer, overexpression of cyclin D1 (CCND1) has been reported in most borderline and invasive ovarian cancers (26, 27); cyclin D2 was found to be overexpressed in about 40% of ovarian tumors compared with normal ovarian tissues (28); P16INK4a (CDKN2A) is relatively frequently lost due to deletion, loss of expression, and hypermethylation in ovarian cancer cell lines and primary tumors (29, 30); and in some primary ovarian cancers, homozygous deletions of P16INK4a involving the neighboring P15INK4b (CDKN2B) gene are associated with a poor prognosis for ovarian cancer cases (31). Together, these data suggest an important role for CDKs, cyclins, and CDK inhibitors in ovarian cancer development.

The aim of this study was to evaluate the association between common variants in 13 genes coding for cyclins, CDKs, and CDK inhibitors and susceptibility to epithelial ovarian cancer. We used a two-stage case-control study design in which single nucleotide polymorphisms (SNP) that tag all known common variants in these genes were first genotyped in three studies. The most significant hits were then genotyped in eight additional ovarian cancer case-control studies from the international Ovarian Cancer Association Consortium (OCAC).

Study subjects. In total, 11 different ovarian cancer case-control studies (summarized in Table 1A and 1B) contributed data to this study, three for stage 1 and eight for stage 2 (see also refs. 810, 16, 17, 21, 23, 32, 33 for further details). Most subjects were white Europeans in origin. Ten studies used population-based ascertainment for cases and controls, and one study was clinic based. All studies have received ethical committee approval, and all study subjects provided informed consent.

Table 1A.

Characteristics of the 11 ovarian cancer case control studies (listed alphabetically) included in the Ovarian Cancer Association Consortium pooled study

Study nameLocationTotal number of cases*Total number of controls*Age range (cases)Age range (controls)ReferenceDNA sourceDays from diagnosis to blood drawNumber of SNPs genotyped
AOCS and ACS (ovarian cancer) Multiregional, Australia 802 (71) 933 (42) 23–80 19–81 Blood lymphocytes 0–1,663 (103) 
FROC (Family Registry for Ovarian Cancer) Study Stanford, CA 327 (39) 429 (59) 23–64 19–66 16, 17, 21, 23 Blood lymphocytes; mouthwash 55–1,477 (281) 88 
Hawaii Ovarian Cancer Study Hawaii 313 (241) 574 (426) 18–87 19–88 33 Blood lymphocytes; mouthwash 16–2,026 (375) 
HOPE (hormones and ovarian cancer prediction) study Pittsburgh 57 (1) 152 (8) 34–80 44–79 – Blood lymphocytes; mouthwash 0–280 (149) 
LAC-CCOC Los Angeles, CA 659 (211) 819 (198) 18–84 21–78 Blood lymphocytes 46–1,340 (252) 
MALOVA (Malignant Ovarian Cancer study) Copenhagen, Denmark 446 1221 (0) 32–80 35–79 16, 17, 21, 23 Blood lymphocytes All preoperative 88 
Mayo Clinic Rochester, ovarian cancer case-control study Minnesota 317 (13) 462 (35) 28–91 23–90 32 Blood lymphocytes 0–365 (31) 
NCOCS North Carolina 610 (91) 843 (156) 22–74 20–75 Blood lymphocytes 29–1,259 (187) 
POCS Warsaw and Lodz 264 625 (0) 24–74 24–74 – Blood lymphocytes; mouthwash 0–525 (82) 
SEARCH ovarian cancer Cambridge, United Kingdom 731 (42) 855 (4) 21–74 39–77 16, 17, 21, 23 Blood lymphocytes 140–4,850 (1,231) 88 
Study nameLocationTotal number of cases*Total number of controls*Age range (cases)Age range (controls)ReferenceDNA sourceDays from diagnosis to blood drawNumber of SNPs genotyped
AOCS and ACS (ovarian cancer) Multiregional, Australia 802 (71) 933 (42) 23–80 19–81 Blood lymphocytes 0–1,663 (103) 
FROC (Family Registry for Ovarian Cancer) Study Stanford, CA 327 (39) 429 (59) 23–64 19–66 16, 17, 21, 23 Blood lymphocytes; mouthwash 55–1,477 (281) 88 
Hawaii Ovarian Cancer Study Hawaii 313 (241) 574 (426) 18–87 19–88 33 Blood lymphocytes; mouthwash 16–2,026 (375) 
HOPE (hormones and ovarian cancer prediction) study Pittsburgh 57 (1) 152 (8) 34–80 44–79 – Blood lymphocytes; mouthwash 0–280 (149) 
LAC-CCOC Los Angeles, CA 659 (211) 819 (198) 18–84 21–78 Blood lymphocytes 46–1,340 (252) 
MALOVA (Malignant Ovarian Cancer study) Copenhagen, Denmark 446 1221 (0) 32–80 35–79 16, 17, 21, 23 Blood lymphocytes All preoperative 88 
Mayo Clinic Rochester, ovarian cancer case-control study Minnesota 317 (13) 462 (35) 28–91 23–90 32 Blood lymphocytes 0–365 (31) 
NCOCS North Carolina 610 (91) 843 (156) 22–74 20–75 Blood lymphocytes 29–1,259 (187) 
POCS Warsaw and Lodz 264 625 (0) 24–74 24–74 – Blood lymphocytes; mouthwash 0–525 (82) 
SEARCH ovarian cancer Cambridge, United Kingdom 731 (42) 855 (4) 21–74 39–77 16, 17, 21, 23 Blood lymphocytes 140–4,850 (1,231) 88 
*

Numbers of subjects that are non-white are given in parentheses.

The mean is given in parentheses.

Table 1B.

Additional characteristics of the 11 ovarian cancer case-control studies included in the Ovarian Cancer Association Consortium pooled study

StudyCase ascertainmentControl ascertainment*Participation rates
AOCS and ACS Australia Diagnosed from 2002 onward; recruited through surgical treatment centers throughout Australia and cancer registries of Queensland South Australia and West Australia (AOCS) and cancer registries of New South Wales and Victoria (ACS) Randomly selected from Commonwealth electoral roll. Frequency matched to cases for geographic region Cases: 84%; controls: 47% 
FROC, USA Consecutive cases diagnosed from 1997–2002 in Greater Bay Area Cancer Registry San Francisco Random-digit dial identification from study area Cases: 63%; controls: 85% 
Hawaii, USA Rapid case ascertainment through Hawaii Tumor Registry from 1993 onward Randomly selected from Hawaii Department of Health Annual Survey of the representative households Cases: 58.1%; controls: 62.8% 
HOPE, USA Variable source including physician offices cancer registries and pathology databases from counties of western Pennsylvania, eastern Ohio, and western New York Identified in same regions. All participants undergo home interviews Cases: 69%; controls: 81% 
LAC-CCOC, USA Rapid case ascertainment through Los Angeles Cancer Surveillance program from 1993 onward Neighborhood-recruited controls Cases: −73%; controls: −73% 
MALOVA, Denmark Incident cases (35–79 years) diagnosed 1994–1999 from municipalities of Copenhagen and Frederiksberg and surrounding counties Random sample of general female population (35–79 years) in study area selected using computerized Central Population Register Cases: 79%; controls: 67% 
Mayo, USA Cases attending Mayo Clinic diagnosed from 2000 onward identified in a six-state surrounding region Identified through Mayo Clinic; healthy women seeking general medical examination Cases: 84%; controls: 65% 
NCOCS, USA Cases from 1999 onward identified from 48 counties within the region by rapid-case ascertainment Controls identified from the same region as cases Cases: 70%; controls: 63% 
POCS, Poland Cases collected from cities of Warsaw and Lodz 2001–2003 by rapid ascertainment at participating hospitals Identified at random through the Polish electronic system; stratified by city and 5-year age categories Cases: 71%; controls: 67% 
SEARCH, UK Cases <70 years from East Anglian West Midlands and Trent regions of England; prevalent cases diagnosed 1991–1998; incident cases diagnosed 1998 onward Selected from the EPIC-Norfolk cohort of 25,000 individuals aged 45–74 based in the same geographic regions as the cases Cases: 67%; controls: 84% 
StudyCase ascertainmentControl ascertainment*Participation rates
AOCS and ACS Australia Diagnosed from 2002 onward; recruited through surgical treatment centers throughout Australia and cancer registries of Queensland South Australia and West Australia (AOCS) and cancer registries of New South Wales and Victoria (ACS) Randomly selected from Commonwealth electoral roll. Frequency matched to cases for geographic region Cases: 84%; controls: 47% 
FROC, USA Consecutive cases diagnosed from 1997–2002 in Greater Bay Area Cancer Registry San Francisco Random-digit dial identification from study area Cases: 63%; controls: 85% 
Hawaii, USA Rapid case ascertainment through Hawaii Tumor Registry from 1993 onward Randomly selected from Hawaii Department of Health Annual Survey of the representative households Cases: 58.1%; controls: 62.8% 
HOPE, USA Variable source including physician offices cancer registries and pathology databases from counties of western Pennsylvania, eastern Ohio, and western New York Identified in same regions. All participants undergo home interviews Cases: 69%; controls: 81% 
LAC-CCOC, USA Rapid case ascertainment through Los Angeles Cancer Surveillance program from 1993 onward Neighborhood-recruited controls Cases: −73%; controls: −73% 
MALOVA, Denmark Incident cases (35–79 years) diagnosed 1994–1999 from municipalities of Copenhagen and Frederiksberg and surrounding counties Random sample of general female population (35–79 years) in study area selected using computerized Central Population Register Cases: 79%; controls: 67% 
Mayo, USA Cases attending Mayo Clinic diagnosed from 2000 onward identified in a six-state surrounding region Identified through Mayo Clinic; healthy women seeking general medical examination Cases: 84%; controls: 65% 
NCOCS, USA Cases from 1999 onward identified from 48 counties within the region by rapid-case ascertainment Controls identified from the same region as cases Cases: 70%; controls: 63% 
POCS, Poland Cases collected from cities of Warsaw and Lodz 2001–2003 by rapid ascertainment at participating hospitals Identified at random through the Polish electronic system; stratified by city and 5-year age categories Cases: 71%; controls: 67% 
SEARCH, UK Cases <70 years from East Anglian West Midlands and Trent regions of England; prevalent cases diagnosed 1991–1998; incident cases diagnosed 1998 onward Selected from the EPIC-Norfolk cohort of 25,000 individuals aged 45–74 based in the same geographic regions as the cases Cases: 67%; controls: 84% 
*

All controls were frequency matched to cases for age and ethnicity.

Participation rates are the percentage of those taking part as proportion of those invited to take part.

Tag SNP selection. A set of tagging SNPs (tSNP) was identified for 13 genes comprising four cyclins (CCND1, CCND2, CCND3, CCNE1); three CDKs (CDK2, CDK4, CDK6); six CDK inhibitors (CDKN1A, CDKN1B, CDKN2A, CDKN2B, CDKN2C, CDKN2D). The principal hypothesis underlying the study was that one or more common SNPs in these genes were associated with an altered risk of ovarian cancer. We therefore identified a set of tagging SNPs that efficiently tagged all the known common variants [minor allele frequency (MAF) ≥0.05] and were likely to tag most of the unknown common variants.

The selection of tSNPs is most reliable where the gene has been resequenced in a sample of individuals sufficiently large to identify all common variants. The National Institute of Environmental Health Sciences (NIEHS) Environmental Genome Project (EGP) is currently resequencing candidate genes for cancer across a panel of 90/95 individuals representative of U.S. ethnicities. The original panel (P1-PDR90) of 90 individuals consists of 24 European Americans, 24 African Americans, 12 Mexican Americans, 6 Native Americans, and 24 Asian Americans, but the ethnic group identifiers are not available. Because it is known that there is greater genetic and haplotype diversity in individuals of African origin, we have identified and excluded 28 of the samples with the greatest African ancestry in this population by comparing the genotypes of the resequenced sample with genotypes for the same SNPs from the National Heart, Lung, and Blood Institute Variation Discovery Resource Project African American panel (http://pga.gs.washington.edu/finished_genes.html). Data from the remaining 62 individuals were used to identify tSNPs. Ideally, samples from Native American, Hispanic American, and Asian American individuals should also be removed; but there is less genetic diversity between these groups, and they cannot be excluded with any certainty.

When resequencing data were not available, we used data from the International HapMap Project (release 34/5 21-06-2005: last public release used in this study) to select tSNPs. The HapMap Project has genotyped a large number of SNPs in several populations, including 30 parent-offspring trios collected in 1980 from U.S. residents with northern and western European ancestry by the Centre d'Etude du Polymorphisme Humain (CEPH).

The best measure of the extent to which one SNP tags another SNP is the pairwise correlation coefficient (rp2) because the loss in power incurred by using a marker SNP in place of a true causal SNP is directly related to this measure. We aimed to define a set of tSNPs such that all known common SNPs had an estimated rp2 > 0.8 with at least one tSNP. However, some SNPs are poorly correlated with other single SNPs but may be efficiently tagged by a haplotype defined by multiple SNPs, so-called “aggressive tagging,” thus reducing the number of tSNPs needed. As an alternative, therefore, we aimed for the correlation between each SNP and a haplotype of tSNPs (rs2) to be at >0.8. The program Tagger21

uses a strategy that combines the simplicity of pairwise methods with the potential efficiency of multimarker approaches (3). It begins by selecting a minimal set of markers such that all alleles to be captured are correlated at an rp2 ≥ 0.8 with a marker in that set. Certain markers can be forced into the tag list or explicitly prohibited from being chosen as tags. After this, it tries to “peel back” the tag list by replacing certain tags with multimarker tests. Tagger avoids overfitting by only constructing multimarker tests from SNPs, which are in strong linkage disequilibrium with each other, as measured by a pairwise LD score. If an assay for one of the chosen tSNPs could not be designed, the SNP tagging process was repeated with that SNP excluded from the possible set of tSNPs. Table 2 shows details of the tSNP selection for each of the genes analyzed in this study.

Table 2.

Details of tSNP selection for cell cycle control genes

GeneSize (kb)Data% of gene resequenced by EGPNo. common variantsNo. tSNPs
Additional SNPs genotyped*Mean r2Number with r2 > 0.8r2 for poorly tagged SNPsLD blocks
Successfully genotypedAdditional assays failed
CCND1 16.0 EGP 100 12 0.82 10 0.08–0.24 
CCND2 34.0 HapMap 93 24 14 0.89 21 0.18–0.74 
CCND3 9.0 EGP 88 13 0.92 12 0.09 
CCNE1 13.5 EGP 84 26 11 0.57 0.10–0.73 
CDK2 7.6 EGP 98 0.54 0.19–0.30 
CDK4 6.7 EGP 96 0.86 0.29 
CDK6 220.8 HapMap 18 41 13 0.96 41 N/A 
CDKN1A 10.9 EGP 85 27 10 0.85 20 0.07–0.69 
CDKN1B 5.7 HapMap 100 0.90 0.23 
CDKN2A 28.8 EGP 81 12 0.97 12 N/A 
CDKN2B 13.3 EGP 94 20 0.96 20 N/A 
CDKN2C 8.9 EGP/HapMap 87 1.00 N/A 
CDKN2D 4.1 EGP 63 1.00 N/A 
GeneSize (kb)Data% of gene resequenced by EGPNo. common variantsNo. tSNPs
Additional SNPs genotyped*Mean r2Number with r2 > 0.8r2 for poorly tagged SNPsLD blocks
Successfully genotypedAdditional assays failed
CCND1 16.0 EGP 100 12 0.82 10 0.08–0.24 
CCND2 34.0 HapMap 93 24 14 0.89 21 0.18–0.74 
CCND3 9.0 EGP 88 13 0.92 12 0.09 
CCNE1 13.5 EGP 84 26 11 0.57 0.10–0.73 
CDK2 7.6 EGP 98 0.54 0.19–0.30 
CDK4 6.7 EGP 96 0.86 0.29 
CDK6 220.8 HapMap 18 41 13 0.96 41 N/A 
CDKN1A 10.9 EGP 85 27 10 0.85 20 0.07–0.69 
CDKN1B 5.7 HapMap 100 0.90 0.23 
CDKN2A 28.8 EGP 81 12 0.97 12 N/A 
CDKN2B 13.3 EGP 94 20 0.96 20 N/A 
CDKN2C 8.9 EGP/HapMap 87 1.00 N/A 
CDKN2D 4.1 EGP 63 1.00 N/A 

Abbreviation: N/A, not applicable.

*

Additional SNPs that were genotyped that were not common in the data set used for tagging. These SNPs do not contribute to the estimates of tagging efficiency.

One common SNP in EGP (environmental genome project; rs3212867) was monomorphic in the CEU HapMap population.

Assay for rs3176359 was too rare in our populations (MAF = 0.002).

Genotyping. Genotyping was carried out for 9 of the 11 studies using TaqMan (Applied Biosystems, Foster City, CA) according to the manufacturer's instructions. Primers and probes were Assays-by-Design (Applied Biosystems). Following PCR amplification, end reactions were read on the ABI Prism 7900 using Sequence Detection Software (Applied Biosystems). Two studies [Australian Ovarian Cancer Study (AOCS) and Australian Cancer Study (ACS); Table 1A and B] used iPlex technology (Sequenom) to perform genotyping. Failed genotypes were not repeated. Plates with genotype calling <95% were discarded. If there was genotyping discordance between duplicates within an experiment, the data were discarded and either repeated or the assay replaced with a different assay with similar tagging properties.

Each group carried out genotyping on their own samples. Therefore, we tested the quality of genotyping between laboratories for the five tSNPs analyzed by all groups using the HAPMAPPT01 panel of CEPH-Utah trios-standard plate supplied by Coriell.22

This panel includes 90 different DNA samples, five duplicate samples, and a negative template control in a 96-well plate format. We compared genotype call rates and concordance between studies. Call rates on these plates ranged from 96% to 99%. Overall concordance on these plates was >99%.

Statistical methods. Deviations of genotype frequencies in the controls from those expected under Hardy-Weinberg equilibrium (HWE) were assessed by χ2 tests [1 degree of freedom (d.f.)]. We treated the design as a single, two-stage multicenter study rather than hypothesis-generating and replication studies because joint analysis has been shown to be more efficient than replication analysis in a staged design (34). The primary tests of association were the univariate analyses between each tSNP and ovarian cancer. Genotype frequencies in cases and controls were compared for each study separately using χ2 tests for heterogeneity (2 d.f.). The data were then pooled, and a genetic model free test was carried out by comparing genotype frequencies in cases and controls using unconditional logistic regression with terms for genotype and study and an appropriate likelihood ratio test (Phet, 2 d.f.). We tested for heterogeneity between studies by comparing logistic regression models with and without a genotype-study interaction term using likelihood ratio tests. Genotypic specific risks with the common homozygote as the baseline comparator were estimated as odds ratios with associated 95% confidence limits by unconditional logistic regression. We also tested for rare allele dose effect (assuming a multiplicative genetic model) using χ2 tests for each study separately and unconditional logistic regression for the pooled data (Phet, 1 d.f.). Primary analyses were restricted to study subjects of white European origin. Secondary analyses included all study subjects with additional adjustment for ethnic group.

In addition to the univariate analyses, we carried out specific haplotype tests for those combinations of alleles that tagged specific SNPs. We also carried out a general comparison of common haplotype frequencies in each gene haplotype block using the data from all the tSNPs in that block. Haplotype blocks were defined such that the common haplotypes (>5% frequency) accounted for at least 80% of the haplotype diversity. We considered haplotypes with >2% frequency in at least one study to be common. Rare haplotypes were pooled. For both specific haplotype marker tests and the general comparison of haplotype frequencies by haplotype block, haplotype frequencies and subject-specific expected haplotype indicators were calculated separately for each study using the program TagSNPs. This implements an expectation substitution approach to account for haplotype uncertainty given unphased genotype data (35). Subjects missing >50% genotype data in each block were excluded from haplotype analysis. We used unconditional logistic regression to test the null hypothesis of no association between specific tSNP haplotype frequency and ovarian cancer stratified by study by comparing a model with terms for subject-specific haplotype indicator and study with a model including study term only. The global null hypothesis of no association between haplotype frequency (by haplotype block) and ovarian cancer was tested by comparing a model with multiplicative effects for each common haplotype (treating the most common haplotype as the reference) to the intercept-only model. Haplotype-specific odds ratios were also estimated with their associated confidence intervals.

Identifying Tagging SNPs in Cell Cycle Control Genes

Using a combination of NIEHS resequencing data and data from the international HapMap project, we identified a total of 199 common SNPs (MAF ≥0.05) in the 13 genes. One-hundred and sixty-seven SNPs were tagged with r2 > 0.8, of which 10 were tagged by 5 different multimarker haplotypes. The remaining SNPs were tagged with r2 between 0.07 and 0.74. These could not be tagged more efficiently because assays could not be designed for them. Eighty-one tSNPs were selected for this series. We also selected a further seven SNPs that were not found in the EGP or HapMap data sets used for tSNP selection. These SNPs do not contribute to the estimates of tagging efficiency. Thus, we identified 88 SNPs from 13 genes for genotyping. Eleven genes were tagged with an average rp2 > 0.8. For two genes (CDK2 and CCNE1), the best tagging we were able to achieve was rp2 0.54 and 0.57, respectively. These data are summarized in Table 2.

Genotyping in Ovarian Cancer Cases and Controls

Genotyping was done in two stages. In stage 1, the 88 tSNPs described above were genotyped in three case-control populations from the United Kingdom (SEARCH), Denmark [Malignant Ovarian Cancer (MALOVA)], and the United States (Stanford). Combined, these studies comprise ∼1,500 invasive ovarian cancer cases and 2,500 controls. In stage 2, we genotyped five SNPs from four genes that showed the strongest evidence of association in stage 1, in ∼3,000 additional cases and 4,400 controls from another eight case-control studies.

Stage 1. Nineteen out of 264 genotype distributions in controls deviated from HWE (P < 0.05). In the majority of cases, deviations from HWE occurred in only one of the studies. The discrimination of genotypes for these assays was good. For rs3176319 (CDKN1A), the deviations from HWE were substantial in all three studies (P < 10−7), and so this SNP was excluded from further analyses. rs3176359 in CDKN1A was also excluded from further analyses because it was found to be very rare in all three populations (MAF = 0.002).

We identified 28 SNPs with a P ≤ 0.2, of which 9 SNPs had a P ≤ 0.05 in tests for association. These data are summarized in Table 3. The complete data for all SNPs are given in Supplementary Tables S1 and S2. There was no evidence for association of genotype with age in controls, and as expected, age-adjusted risks were similar to unadjusted risks (data not shown). The haplotype frequencies and associated risks for the five multimarker haplotype tags are presented in Supplementary Table S3. There was no evidence for association at a significance level P ≤ 0.05 for SNPs in 6 of the 13 genes (CDK2, CDK4, CCND2, CDKN1A, CDKN2C, CDKN2D).

Table 3.

Genotype-specific risks for tSNPs with P < 0.2: combined analysis of 88 SNPs using SEARCH-MALOVA-Stanford studies

Gene (no. tSNPs)*SNP IDdbSNPMAFHetOR (95% CI)HomOR (95% CI)§P2dfP trend
CCND1 (7) ccnd107 rs602652 0.46 1.14 (0.98–1.33) 1.24 (1.03–1.50) 0.057 0.018 
 ccnd103 rs603965 0.44 1.06 (0.91–1.24) 1.29 (1.07–1.55) 0.021 0.010 
 ccnd108 rs3212879 0.49 0.85 (0.73–0.99) 0.82 (0.68–0.98) 0.057 0.028 
 ccnd110 rs3212891 0.46 0.86 (0.74–1.00) 0.83 (0.69–0.997) 0.076 0.034 
 ccnd101 rs7178 0.07 1.24 (1.03–1.49) 1.22 (0.52–2.88) 0.065 0.021 
CCND2 (14) ccnd202 rs3217805 0.40 1.07 (0.93–1.23) 0.86 (0.70–1.05) 0.078 0.33 
 ccnd215 rs3217916 0.28 1.00 (0.88–1.15) 0.76 (0.59–0.99) 0.11 0.17 
 ccnd216 rs3217925 0.26 0.98 (0.85–1.12) 0.74 (0.55–0.99) 0.13 0.13 
 ccnd207 rs3217936 0.33 1.01 (0.88–1.16) 0.77 (0.61–0.98) 0.069 0.15 
CCND3 (7) ccnd303 rs1051130 0.48 1.13 (0.97–1.32) 0.88 (0.73–1.06) 0.0090 0.23 
 ccnd306 rs9529 0.29 1.00 (0.87–1.15) 0.79 (0.62–1.02) 0.14 0.18 
 ccnd307 rs3218110 0.24 1.11 (0.97–1.28) 1.07 (0.80–1.42) 0.32 0.19 
CCNE1 (4) ccne102 rs997669 0.38 1.10 (0.95–1.27) 1.14 (0.94–1.40) 0.31 0.13 
 ccne112 rs3218036 0.31 1.07 (0.93–1.23) 1.26 (1.00–1.57) 0.12 0.052 
CDK2 (2) cdk202 rs2069408 0.34 0.90 (0.78–1.03) 0.90 (0.72–1.12) 0.27 0.16 
 cdk204 rs1045435 0.09 1.10 (0.93–1.31) 1.97 (0.89–4.34) 0.14 0.093 
CDK6 (13) cdk612 rs2285332 0.25 0.92 (0.80–1.06) 0.86 (0.65–1.14) 0.38 0.16 
 cdk611 rs3731348 0.06 1.07 (0.86–1.34) 0.28 (0.07–1.21) 0.19 0.93 
 cdk613 rs2237570 0.12 0.89 (0.76–1.05) 0.70 (0.37–1.32) 0.23 0.094 
 cdk610 rs2282991 0.10 0.84 (0.71–1.00) 0.85 (0.43–1.67) 0.14 0.054 
 cdk605 rs8 0.21 1.18 (1.02–1.35) 1.41 (1.01–1.96) 0.017 0.0042 
CDKN1A (11) cdkn1a06 rs2395655 0.29 1.12 (0.97–1.29) 1.16 (0.96–1.41) 0.19 0.082 
 cdkn1a01 rs1801270 0.08 0.91 (0.76–1.11) 0.52 (0.18–1.50) 0.32 0.20 
CDKN1B (7) cdkn1b05 rs3759217 0.12 1.13 (0.97–1.33) 1.03 (0.61–1.74) 0.30 0.17 
 cdkn1b02 rs2066827 0.26 0.88 (0.77–1.01) 0.68 (0.51–0.90) 0.0098 0.0035 
CDKN2A/2B (17) cdkn2a04 rs2811712 0.10 0.98 (0.83–1.16) 1.91 (1.07–3.42) 0.086 0.42 
 cdkn2a10 rs11515 0.14 1.0 (0.86–1.16) 1.55 (0.97–2.47) 0.19 0.38 
 cdkn2a13 rs3731257 0.26 0.90 (0.78–1.03) 0.80 (0.59–1.07) 0.14 0.046 
Gene (no. tSNPs)*SNP IDdbSNPMAFHetOR (95% CI)HomOR (95% CI)§P2dfP trend
CCND1 (7) ccnd107 rs602652 0.46 1.14 (0.98–1.33) 1.24 (1.03–1.50) 0.057 0.018 
 ccnd103 rs603965 0.44 1.06 (0.91–1.24) 1.29 (1.07–1.55) 0.021 0.010 
 ccnd108 rs3212879 0.49 0.85 (0.73–0.99) 0.82 (0.68–0.98) 0.057 0.028 
 ccnd110 rs3212891 0.46 0.86 (0.74–1.00) 0.83 (0.69–0.997) 0.076 0.034 
 ccnd101 rs7178 0.07 1.24 (1.03–1.49) 1.22 (0.52–2.88) 0.065 0.021 
CCND2 (14) ccnd202 rs3217805 0.40 1.07 (0.93–1.23) 0.86 (0.70–1.05) 0.078 0.33 
 ccnd215 rs3217916 0.28 1.00 (0.88–1.15) 0.76 (0.59–0.99) 0.11 0.17 
 ccnd216 rs3217925 0.26 0.98 (0.85–1.12) 0.74 (0.55–0.99) 0.13 0.13 
 ccnd207 rs3217936 0.33 1.01 (0.88–1.16) 0.77 (0.61–0.98) 0.069 0.15 
CCND3 (7) ccnd303 rs1051130 0.48 1.13 (0.97–1.32) 0.88 (0.73–1.06) 0.0090 0.23 
 ccnd306 rs9529 0.29 1.00 (0.87–1.15) 0.79 (0.62–1.02) 0.14 0.18 
 ccnd307 rs3218110 0.24 1.11 (0.97–1.28) 1.07 (0.80–1.42) 0.32 0.19 
CCNE1 (4) ccne102 rs997669 0.38 1.10 (0.95–1.27) 1.14 (0.94–1.40) 0.31 0.13 
 ccne112 rs3218036 0.31 1.07 (0.93–1.23) 1.26 (1.00–1.57) 0.12 0.052 
CDK2 (2) cdk202 rs2069408 0.34 0.90 (0.78–1.03) 0.90 (0.72–1.12) 0.27 0.16 
 cdk204 rs1045435 0.09 1.10 (0.93–1.31) 1.97 (0.89–4.34) 0.14 0.093 
CDK6 (13) cdk612 rs2285332 0.25 0.92 (0.80–1.06) 0.86 (0.65–1.14) 0.38 0.16 
 cdk611 rs3731348 0.06 1.07 (0.86–1.34) 0.28 (0.07–1.21) 0.19 0.93 
 cdk613 rs2237570 0.12 0.89 (0.76–1.05) 0.70 (0.37–1.32) 0.23 0.094 
 cdk610 rs2282991 0.10 0.84 (0.71–1.00) 0.85 (0.43–1.67) 0.14 0.054 
 cdk605 rs8 0.21 1.18 (1.02–1.35) 1.41 (1.01–1.96) 0.017 0.0042 
CDKN1A (11) cdkn1a06 rs2395655 0.29 1.12 (0.97–1.29) 1.16 (0.96–1.41) 0.19 0.082 
 cdkn1a01 rs1801270 0.08 0.91 (0.76–1.11) 0.52 (0.18–1.50) 0.32 0.20 
CDKN1B (7) cdkn1b05 rs3759217 0.12 1.13 (0.97–1.33) 1.03 (0.61–1.74) 0.30 0.17 
 cdkn1b02 rs2066827 0.26 0.88 (0.77–1.01) 0.68 (0.51–0.90) 0.0098 0.0035 
CDKN2A/2B (17) cdkn2a04 rs2811712 0.10 0.98 (0.83–1.16) 1.91 (1.07–3.42) 0.086 0.42 
 cdkn2a10 rs11515 0.14 1.0 (0.86–1.16) 1.55 (0.97–2.47) 0.19 0.38 
 cdkn2a13 rs3731257 0.26 0.90 (0.78–1.03) 0.80 (0.59–1.07) 0.14 0.046 
*

We also analyzed tSNPs in the genes CDK4 (two SNPs); CDKN2C (two SNPs); CDKN2D (two SNPs), but none of these reached the required level of statistical significance. Data from the combined analysis of all SNPs are given in Supplementary Tables S1 and S2. Confidence intervals that do not reach or cross 1.00 and P values <0.05 are in bold type.

Minor allele frequency.

Heterozygous odds ratios (95% confidence intervals).

§

Homozygous odds ratios (95% confidence intervals).

Evidence of association at the P = 0.05 level was found for the following: five of seven SNPs in CCND1 (strongest association P trend, 0.01); one of seven SNPs in CCND3 (P2df = 0.0 09); 1 of 13 SNPs in CDK6 (Ptrend = 0.004); one of eight SNPs in CDKN1B (P2df = 0.01); and 1 of 17 SNPs in CDKN2A/2B (P2df = 0.046). Further details are provided in Table 3.

The global test of association with common haplotypes was not significant for any of the 13 genes (data not shown). One haplotype that showed a significant frequency difference between cases and controls was observed in CDK6 block 3 (haplotype frequency, 0.2 in controls). This haplotype, comprising the rare allele of rs8 and common alleles of rs479049, rs445, and rs992519, showed increased risk for ovarian cancer compared with the most common haplotype in this block (OR, 1.19 [1.06–1.35]; P = 0.004).

Stage 2. Five SNPs were genotyped in the stage 2 studies. These were rs7178 and rs603965 (both in CCND1), rs8 (CDK6), rs2066827 (CDKN1B), and rs3731257 (CDKN2A/2B). The evidence for association for these SNPs from the stage 1 analyses (adjusted for age and ethnicity) ranged from P trend 0.003 to 0.046. There was some evidence for between-study heterogeneity for rs8 (P = 0.036). None of the SNPs were significantly associated with ovarian cancer when considering the stage 2 studies as a replication set (Table 4A); rs3731257 was the most significant (Ptrend = 0.13).

Table 4.

Age-adjusted ovarian cancer risks associated with five tSNPs in cell cycle control genes for white cases and white controls only

GenedbSNPMAFNo. casesNo. controlsHetOR (95% CI)HomOR (95% CI)P2dfPtrend
A. Combined analysis of all stage 2 studies         
CCND1 rs7178 0.08 2,194 3,294 1.01 (0.86–1.18) 1.01 (0.50–2.06) 0.99 0.91 
CCND1 rs603965 0.45 1,874 2,833 1.00 (0.87–1.15) 1.01 (0.86–1.20) 0.98 0.90 
CDK6 rs8 0.21 2,188 3,298 1.02 (0.91–1.15) 0.91 (0.69–1.19) 0.69 0.83 
CDKN1B rs2066827 0.23 2,202 3,297 1.05 (0.93–1.18) 0.88 (0.69–1.12) 0.36 0.82 
CDKN2A/2B rs3731257 0.27 2,190 3,290 0.92 (0.82–1.04) 0.89 (0.72–1.10) 0.29 0.13 
         
B. Combined analysis for all studies, white subjects only         
CCND1 rs7178 0.08 3,607 5,725 1.09 (0.96–1.22) 1.12 (0.64–1.97) 0.37  0.12 
CCND1 rs603965 0.45 3,285 5,236 1.02 (0.92–1.13) 1.13 (0.99–1.28) 0.15 0.084 
CDK6 rs8 0.21 3,597 5,720 1.09 (1.00–1.19) 1.06 (0.86–1.31) 0.17 0.082 
CDKN1B rs2066827 0.25 3,618 5,719 0.98 (0.89–1.07) 0.79 (0.65–0.95) 0.044 0.036 
CDKN2A/2B rs3731257 0.27 3,601 5,705 0.89 (0.81–0.97) 0.87 (0.73–1.03) 0.021 0.0080 
GenedbSNPMAFNo. casesNo. controlsHetOR (95% CI)HomOR (95% CI)P2dfPtrend
A. Combined analysis of all stage 2 studies         
CCND1 rs7178 0.08 2,194 3,294 1.01 (0.86–1.18) 1.01 (0.50–2.06) 0.99 0.91 
CCND1 rs603965 0.45 1,874 2,833 1.00 (0.87–1.15) 1.01 (0.86–1.20) 0.98 0.90 
CDK6 rs8 0.21 2,188 3,298 1.02 (0.91–1.15) 0.91 (0.69–1.19) 0.69 0.83 
CDKN1B rs2066827 0.23 2,202 3,297 1.05 (0.93–1.18) 0.88 (0.69–1.12) 0.36 0.82 
CDKN2A/2B rs3731257 0.27 2,190 3,290 0.92 (0.82–1.04) 0.89 (0.72–1.10) 0.29 0.13 
         
B. Combined analysis for all studies, white subjects only         
CCND1 rs7178 0.08 3,607 5,725 1.09 (0.96–1.22) 1.12 (0.64–1.97) 0.37  0.12 
CCND1 rs603965 0.45 3,285 5,236 1.02 (0.92–1.13) 1.13 (0.99–1.28) 0.15 0.084 
CDK6 rs8 0.21 3,597 5,720 1.09 (1.00–1.19) 1.06 (0.86–1.31) 0.17 0.082 
CDKN1B rs2066827 0.25 3,618 5,719 0.98 (0.89–1.07) 0.79 (0.65–0.95) 0.044 0.036 
CDKN2A/2B rs3731257 0.27 3,601 5,705 0.89 (0.81–0.97) 0.87 (0.73–1.03) 0.021 0.0080 

Combined analysis. As joint analysis with appropriate adjustment for multiple testing has been shown to be more powerful than replication analysis (34), we also carried out a combined analysis of all the data. Figure 1 shows the genotype-specific risks (estimated as odds ratios) for each study and for the combined data for the analyses restricted to white subjects. Only two of the five SNPs were significant at the 0.05 level (Table 4B). These were rs2066827 (CDKN1B) and rs3731257 (CDKN2A-2B; Ptrend = 0.036 and 0.008, respectively). The rare alleles for both SNPs were associated with reduced ovarian cancer risks [rs2066827 HetOR, 0.98 (0.89–1.07); HomOR, 0.79 (0.65–0.95); and rs3731257 HetOR, 0.89 (0.81–0.97); HomOR, 0.87 (0.73–1.03)]. There was little change in the association of rs2066827 when the data for all ethnic groups were included (Ptrend = 0.006); but for rs3731257, the association was almost completely attenuated (P = 0.16). Standard methods for correcting for multiple testing, such as the Bonferroni correction, are too conservative when hypotheses are correlated, as is the case here. We have therefore used a permutation procedure modified to account for the staged design (36). A P value smaller than the most significant result was observed in 368 out of 1,000 permutations, which is equivalent to an adjusted Ptrend of 0.37.

Figure 1.

Genotype-specific risks by study. Heterozygous and homozygous genotype-specific odds ratios are illustrated for White subjects only for all five SNPs analyzed in all studies. Lines, 95% confidence intervals; size of the box, proportionate size of the study (number of white subjects genotyped) compared with other studies. There were no homozygous odds ratios for some studies because the minor allele was too rare in the population. Key to studies in Table 1A and B: ACS, ACS Australia; AOC, AOCS Australia; HAW, Hawaii Ovarian Cancer Study Hawaii; HOP, HOPE study, Pittsburgh, PA; MAY, Mayo Clinic, Minnesota; MAL, MALOVA study, Denmark; NCO, North Carolina Ovarian Cancer Study (NCOCS), North Carolina; POL, Polish Ovarian Cancer Study (POCS), Poland; SEA, SEARCH (Studies of Epidemiology and Risk Factors in Cancer Heredity) study, United Kingdom; USC, Los Angeles County Case-Control Studies of Ovarian Cancer (LAC-COCC), southern California.

Figure 1.

Genotype-specific risks by study. Heterozygous and homozygous genotype-specific odds ratios are illustrated for White subjects only for all five SNPs analyzed in all studies. Lines, 95% confidence intervals; size of the box, proportionate size of the study (number of white subjects genotyped) compared with other studies. There were no homozygous odds ratios for some studies because the minor allele was too rare in the population. Key to studies in Table 1A and B: ACS, ACS Australia; AOC, AOCS Australia; HAW, Hawaii Ovarian Cancer Study Hawaii; HOP, HOPE study, Pittsburgh, PA; MAY, Mayo Clinic, Minnesota; MAL, MALOVA study, Denmark; NCO, North Carolina Ovarian Cancer Study (NCOCS), North Carolina; POL, Polish Ovarian Cancer Study (POCS), Poland; SEA, SEARCH (Studies of Epidemiology and Risk Factors in Cancer Heredity) study, United Kingdom; USC, Los Angeles County Case-Control Studies of Ovarian Cancer (LAC-COCC), southern California.

Close modal

Statistical power to identify subgroup effects from these studies is limited; but there were sufficient data to carry out an analysis with cases restricted to serous and nonserous histopathologic subtypes. In total, there were ∼2,000 cases of serous histology in the 11 studies combined and 1,600 nonserous cases. The analysis based on this histologic subgroup stratification suggested marginal associations for rs3731257 with both serous (Ptrend = 0.030) and nonserous cancers (Ptrend = 0.042) and for rs7178 (Ptrend = 0.034), rs8 (P2df = 0.046), and rs2066827 (Ptrend = 0.023) with nonserous cancers only (Table 5).

Table 5.

The effects of histologic subtype on ovarian cancer risks associated with five tSNPs in cell cycle control genes

Histologic typeGenedbSNPMAFNumber of casesNumber of controlsHetOR (95% CI)HomOR (95% CI)P2dfPtrend
Serous CCND1 rs7178 0.08 2,014 5,725 0.99 (0.87–1.12) 1.11 (0.95–1.29) 0.06 0.74 
 CCND1 rs603965 0.45 1,807 5,236 0.99 (0.87–1.12) 1.11 (0.95–1.29) 0.26 0.27 
 CDK6 rs8 0.21 2,009 5,720 1.04 (0.93–1.16) 1.14 (0.88–1.47) 0.53 0.28 
 CDKN1B rs2066827 0.25 2,022 5,719 1.02 (0.92–1.14) 0.82 (0.65–1.03) 0.18 0.30 
 CDKN2A/2B rs3731257 0.27 2,012 5,705 0.90 (0.81–1.01) 0.86 (0.69–1.06) 0.11 0.03 
Nonserous CCND1 rs7178 0.08 1,593 5,725 1.09 (0.93–1.28) 2.16 (1.18–3.95) 0.034 0.034 
 CCND1 rs603965 0.45 1,478 5,236 1.07 (0.93–1.22) 1.15 (0.97–1.36) 0.26 0.12 
 CDK6 rs8 0.21 1,588 5,720 1.16 (1.03–1.31) 1.00 (0.75–1.34) 0.046 0.063 
 CDKN1B rs2066827 0.25 1,596 5,719 0.93 (0.82–1.05) 0.75 (0.58–0.97) 0.058 0.023 
 CDKN2A/2B rs3731257 0.27 1,589 5,705 0.87 (0.78–0.99) 0.88 (0.69–1.11) 0.073 0.042 
Histologic typeGenedbSNPMAFNumber of casesNumber of controlsHetOR (95% CI)HomOR (95% CI)P2dfPtrend
Serous CCND1 rs7178 0.08 2,014 5,725 0.99 (0.87–1.12) 1.11 (0.95–1.29) 0.06 0.74 
 CCND1 rs603965 0.45 1,807 5,236 0.99 (0.87–1.12) 1.11 (0.95–1.29) 0.26 0.27 
 CDK6 rs8 0.21 2,009 5,720 1.04 (0.93–1.16) 1.14 (0.88–1.47) 0.53 0.28 
 CDKN1B rs2066827 0.25 2,022 5,719 1.02 (0.92–1.14) 0.82 (0.65–1.03) 0.18 0.30 
 CDKN2A/2B rs3731257 0.27 2,012 5,705 0.90 (0.81–1.01) 0.86 (0.69–1.06) 0.11 0.03 
Nonserous CCND1 rs7178 0.08 1,593 5,725 1.09 (0.93–1.28) 2.16 (1.18–3.95) 0.034 0.034 
 CCND1 rs603965 0.45 1,478 5,236 1.07 (0.93–1.22) 1.15 (0.97–1.36) 0.26 0.12 
 CDK6 rs8 0.21 1,588 5,720 1.16 (1.03–1.31) 1.00 (0.75–1.34) 0.046 0.063 
 CDKN1B rs2066827 0.25 1,596 5,719 0.93 (0.82–1.05) 0.75 (0.58–0.97) 0.058 0.023 
 CDKN2A/2B rs3731257 0.27 1,589 5,705 0.87 (0.78–0.99) 0.88 (0.69–1.11) 0.073 0.042 

NOTE: White invasive cases (serous/non serous) are compared with white controls.

We have assessed the effect of tSNPs in 13 cell cycle control genes on the risks of invasive epithelial ovarian cancer. We chose to analyze genes involved in the control of cell division (cyclins, CDK, and CDK inhibitors) because several of these genes have previously been implicated in the somatic development of multiple tumor types. To our knowledge, this is the first report describing the association between these genes and ovarian cancer susceptibility.

A two-stage design was used for this study. In the first stage, 88 tSNPs were genotyped in three ovarian cancer case-control studies. The best five SNPs that were marginally associated with epithelial ovarian cancer were then genotyped in a further eight case-control studies from the international Ovarian Cancer Association Consortium, but none was significant in the second stage. However, the power to detect an effect in the replication studies was just 70% to detect an allele with frequency and effect size similar to CDKN1B rs2066827 (the most significant association from stage 1) at a type I error rate of 0.01. Because a combined analysis with adjustment for multiple testing has been shown to have greater power than a replication analysis, we also carried out a joint analysis of the stages 1 and 2 data. Two SNPs, CDKN1B-rs2066827 and CDKN2A/2B-rs3731257, showed weak evidence for association, but neither was highly statistically significant (naïve Ptrend = 0.036 and 0.008, respectively), but these were not significant after adjustment for multiple testing using a permutation procedure that allows for the correlation between the SNPs. As an alternative to correcting for multiple testing, some authors have suggested that simple but stringent criteria should be applied to statistical tests for genetic association; e.g., P < 10−4 for candidate gene studies or even P < 10−7 for genomewide significance (37). Using these criteria, it can be estimated that 26,448 cases and an equal number of controls would be needed to confirm the effect for rs2066827 to detect a codominant allele with a minor allele frequency of 0.25 that confers a risk of 1.08 with 90% power at a type I error rate of 10−4.

Because the selection of SNPs in this study was based on a tagging approach rather than on putative function, we are unable to comment in great detail on the possible functional significance of these findings. The rs2066827 variant in CDKN1B is potentially functional. It encodes a nonsynonymous amino acid change Val109Gly in CDKN1B and is situated in the interaction surface of CDKN1B and its negative regulator p38jab1, in the region spanning amino acid residues 97 to 151. This SNP could therefore alter the interaction between CDKN1B and p38jab1. The Val109Gly variant has previously been reported in association with advanced prostate cancer risk (38).

The rs3731257 SNP in CDKN2A/2B is of no obvious functional significance, but it could be in linkage disequilibrium with another functional SNP within the gene. CDKN2A is a G1 CDK inhibitor that binds to CDK4 and CDK6 and prevents their association with D-type cyclins, thereby facilitating CDK4/6–cyclin D–mediated phosphorylation and inactivation of retinoblastoma protein and entry into S-phase. CDKN2B inhibits MDM2-mediated degradation of P53; thus, its loss would lead to a reduction in levels of the P53 protein (39, 40). The known function of both proteins suggests a plausible role in tumor development.

An alternative explanation for a false positive association is hidden population stratification. This occurs when allele frequencies differ between population subgroups, and cases and controls are drawn differentially from those subgroups. However, this study comprised 11 different case-control populations from England, Denmark, Poland, Australia, and several states throughout the United States. It is unlikely that population stratification is important because any population stratification will be study specific. Furthermore, the combined study was of sufficient size to allow the analyses to be restricted to white-only cases and controls, which represented the vast majority of all subjects.

Disease heterogeneity could also lead to false-positive reporting or mask the presence of true associations. In this study, we stratified cases according to histologic subtype and did the analysis for the combined data after dividing cases into serous and nonserous subgroups. Inevitably, this leads to a loss in statistical power, and although some marginal associations were found, these data should be treated with caution. Cases were collected from multiples of different centers, and there are likely to be variations between studies in the completeness of these data and in the reporting of different histopathologic subtypes.

There was no evidence of association with disease risk for polymorphisms in CCND2, CCNE1, CDK2, CDK4, CDKN1A, CDKN2C, and CDKN2D at the 0.05 level. The SNPs tested in this study were selected to tag all the common variants in each gene and not because of their putative functional effects. However, we were unable to design assays for 31 of the selected tSNPs, and 35 of the 199 common SNPs were tagged with rp2 < 0.8. Furthermore, although tSNPs based on HapMap and EGP data are likely to tag most of the common SNPs, there is a possibility that other unknown common variants were poorly tagged.

There have been many candidate SNP/gene association studies for ovarian cancer published over the past few years. Few of the published studies report results that are statistically significant; but most of these studies have had insufficient statistical power to detect moderate risks even for common genetic variants. Furthermore, very few studies have used comprehensive tagging approaches to capture all the common variation in a gene. Where associations have been found, the existence of a susceptibility allele remains unproven because results await confirmation, or there are conflicting results in follow-up studies. The current study illustrates the value of large consortia to follow-up putative positive associations for common polymorphisms in complex diseases to clarify their risks. In the future, a consortium approach to genetic epidemiology studies might also enable the analysis of rarer genetic variants and gene-gene/gene-environment interactions where individual studies have inadequate power.

Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).

S.A. Gayther and H. Song contributed equally to this work.

Grant support: Funded by a grant from WellBeing (H. Song); B.A.J. Ponder is a Gibb Fellow; P.D.P. Pharoah is a Senior Clinical Research Fellow; S.A. Gayther is a Higher Education Funding Council for England–funded senior lecturer; the Mermaid component of the Eve Appeal (S.J. Ramus); D.F. Easton is a Principal Research Fellow of Cancer Research United Kingdom. E.L. Goode is a Fraternal Order of Eagles Fellow. G. Chenevix-Trench and D. Whiteman are supported by Fellowships from the National Health and Medical Research Council; P. Webb is supported by a Fellowship from the Queensland Cancer Fund.

Stage 1 genotyping was funded by a grant from Cancer Research United Kingdom. Stage 2 genotyping and the Ovarian Cancer Association Consortium are funded by the Ovarian Cancer Research Fund, thanks to generous donations from the family and friends of Kathryn Sladek Smith.

Additional support was provided by the Roswell Park Alliance and the National Cancer Institute (CA71766 and core grant CA16056; Stanford study); the Mayo Foundation (Mayo study); Intramural Research Funds of the National Cancer Institute, Department of Health and Human Services, United States (Polish Ovarian Cancer Study); U.S. Army Medical Research and Materiel Command under DAMD17-01-1-0729, the Cancer Council Tasmania and Cancer Foundation of Western Australia (AOCS study); The National Health and Medical Research Council of Australia (199600; ACS study); National Cancer Institute grants K07-CA80668 and R01CA095023; and Department of Defence grant DAMD17-02-1-0669 [the Hormones and Ovarian Cancer Prediction (HOPE) project]; the California Cancer Research Program grants 00-01389V-20170 and 2110200; U.S. Public Health Service grants CA14089, CA17054, CA61132, CA63464, N01-PC-67010, and R03-CA113148, and the California Department of Health Services subcontract 050-E8709 as part of its statewide cancer reporting program (University of Southern California).

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.

We thank Joan MacIntosh, Hannah Munday, Barbara Perkins, Clare Jordan, Kristy Driver, Mitul Shah, the local general practices and nurses and the East Anglian Cancer Registry for recruitment of the United Kingdom cases: the EPIC-Norfolk investigators for recruitment of the United Kingdom controls; Craig Luccarini and Don Conroy for expert technical assistance. Dr. Stephen Chancok of the Core Genotyping Facility, Division of Cancer Epidemiology and Genetics of the National Cancer Institute (United States), Dr. Neonila Szeszenia-Dabrowska of the Nofer Institute of Occupational Medicine (Lodz, Poland), and Dr. Witold Zatonski of the M. Sklodowska-Curie Institute of Oncology and Cancer Center (Warsaw, Poland) for their contribution to the Polish Breast Cancer Study; Debby Bass, Shari Hutchison, Carlynn Jackson, Jessica Kopsic, and Mary Hartley for the HOPE project; Karin Goodman and Renee Weatherly for subject recruitment; and Robert Vierkant, Zachary Fredericksen, and Ludi Fan for data management in the Mayo Clinic study.

Finally, we express our profound thanks to all the study participants who contributed to this research.

1
Gayther SA, Russell P, Harrington P, et al. The contribution of germline BRCA1 and BRCA2 mutations to familial ovarian cancer: No evidence for other ovarian cancer-susceptibility genes.
Am J Hum Genet
1999
;
65
:
1021
–9.
2
Ford D, Easton DF, Stratton M, et al. Genetic heterogeneity and penetrance analysis of the BRCA1 and BRCA2 genes in breast cancer families.
Am J Hum Genet
1998
;
62
:
676
–89.
3
Antoniou AC, Easton DF. Risk prediction models for familial breast cancer.
Future Oncol
2006
;
2
:
257
–74.
4
Pharoah PD, Antoniou A, Bobrow M, et al. Polygenic susceptibility to breast cancer and implications for prevention.
Nat Genet
2002
;
31
:
33
–6.
5
Rowe SM, Coughlan SJ, McKenna NJ, et al. Ovarian carcinoma-associated TaqI restriction fragment length polymorphism in intron G of the progesterone receptor gene is due to an Alu sequence insertion.
Cancer Res
1995
;
55
:
2743
–5.
6
Spurdle AB, Webb PM, Purdie DM, et al. No significant association between progesterone receptor exon 4 Val660Leu G/T polymorphism and risk of ovarian cancer.
Carcinogenesis
2001
;
22
:
717
–21.
7
Lancaster JM, Berchuck A, Carney M, Wiseman RW, Taylor JA. Progesterone receptor gene and risk of breast and ovarian cancer.
Br J Cancer
1997
;
78
:
277
.
8
Pearce CL, Hirschhorn JN, Wu AH, et al. Clarifying the PROGINS allele association in ovarian and breast cancer risk: a haplotype-based analysis.
J Natl Cancer Inst
2005
;
97
:
51
–9.
9
Berchuck A, Schildkraut JM, Wenham RM, et al. Progesterone receptor promoter +331A polymorphism is associated with increased risk of endometrioid and clear cell ovarian cancers.
Cancer Epidemiol Biomarkers Prev
2004
;
13
:
2141
–7.
10
Spurdle AB, Webb PM, Chen X, et al. Androgen receptor exon 1 CAG repeat length and risk of ovarian cancer.
Int J Cancer
2000
;
87
:
637
–43.
11
Levine DA, Boyd J. The androgen receptor and genetic susceptibility to ovarian cancer: results from a case series.
Cancer Res
2001
;
61
:
908
–11.
12
Garner EI, Stokes EE, Berkowitz RS, Mok SC, Cramer DW. Polymorphisms of the estrogen-metabolizing genes CYP17 and catechol-O-methyltransferase and risk of epithelial ovarian cancer.
Cancer Res
2002
;
62
:
3058
–62.
13
Spurdle AB, Chen X, Abbazadegan M, et al. CYP17 promotor polymorphism and ovarian cancer risk.
Int J Cancer
2000
;
86
:
436
–9.
14
Spurdle AB, Purdie D, Chen X, Chenevix-Trench G. The prohibitin 3′ untranslated region polymorphism is not associated with risk of ovarian cancer.
Cancer Res
2003
;
90
:
145
–9.
15
Auranen A, Spurdle AB, Chen X, et al. BRCA2 Arg372His polymorphism and epithelial ovarian cancer risk.
Int J Cancer
2003
;
103
:
427
–30.
16
Auranen A, Song H, Waterfall C, et al. Polymorphisms in DNA repair genes and epithelial ovarian cancer risk.
Int J Cancer
2005
;
117
:
611
–8.
17
Song H, Ramus SJ, Quaye L, et al. Common variants in mismatch repair genes and risk of invasive ovarian cancer.
Carcinogenesis
2006
;
27
:
2235
–42.
18
Runnebaum IB, Tong XW, Konig R, et al. p53-based blood test for p53PIN3 and risk for sporadic ovarian cancer.
Lancet
1995
;
345
:
994
.
19
Lancaster JM, Brownlee HA, Wiseman RW, Taylor J. p53 polymorphism in ovarian and bladder cancer.
Lancet
1995
;
346
:
182
.
20
Buller RE, Sood A, Fullenkamp C, et al. The influence of the p53 codon 72 polymorphism on ovarian carcinogenesis and prognosis.
Cancer Gene Ther
1997
;
4
:
239
–45.
21
Song H, Ramus SJ, Shadforth D, et al. Common variants in RB1 gene and risk of invasive ovarian cancer.
Cancer Res
2006
;
66
:
10220
–6.
22
Weitzel JN, Ding S, Larson GP, et al. The HRAS1 minisatellite locus and risk of ovarian cancer.
Cancer Res
2000
;
60
:
259
–61.
23
Dicioccio RA, Song H, Waterfall C, et al. STK15 polymorphisms and association with risk of invasive ovarian cancer.
Cancer Epidemiol Biomarkers Prev
2004
;
13
:
1589
–94.
24
Funk JO. Cancer cell cycle control.
Anticancer Res
1999
;
19
:
4772
–80.
25
Lundberg AS, Weinberg RA. Control of the cell cycle and apoptosis.
Eur J Cancer
1999
;
35
:
1886
–94.
26
Barbieri F, Cagnoli M, Ragni N, et al. Increased cyclin D1 expression is associated with features of malignancy and disease recurrence in ovarian tumors.
Clin Cancer Res
1999
;
5
:
1837
–42.
27
Shigemasa K, Tanimoto H, Parham GP, et al. Cyclin D1 overexpression and p53 mutation status in epithelial ovarian cancer.
J Soc Gynecol Investig
1999
;
6
:
102
–8.
28
Chen CH, Shen J, Lee WJ, Chow SN. Overexpression of cyclin D1 and c-Myc gene products in human primary epithelial ovarian cancer.
Int J Gynecol Cancer
2005
;
15
:
878
–83.
29
Milde-Langosch K, Ocon E, Becker G, Loning T. p16/MTS1 inactivation in ovarian carcinomas: high frequency of reduced protein expression associated with hyper-methylation or mutation in endometrioid and mucinous tumors.
Int J Cancer
1998
;
79
:
61
–5.
30
Kusume T, Tsuda H, Kawabata M, et al. p16-cyclin D1/CDK4-pRb pathway and clinical outcome in epithelial ovarian cancer.
Clin Cancer Res
1999
;
5
:
4152
–7.
31
Kudoh K, Ichikawa Y, Yoshida S, et al. Inactivation of p16/CDKN2 and p15/MTS2 is associated with prognosis and response to chemotherapy in ovarian cancer.
Int J Cancer
2002
;
99
:
579
–82.
32
Sellers TA, Schildkraut JM, Pankratz VS, et al. Estrogen bioactivation genetic polymorphisms and ovarian cancer.
Cancer Epidemiol Biomarkers Prev
2005
;
14
:
2536
–43.
33
Goodman MT, McDuffie K, Kolonel LN, et al. Multiethnic case-control study of ovarian cancer and polymorphisms in genes involved in catechol estrogen formation and metabolism.
Cancer Epidemiol Biomarkers Prev
2001
;
10
:
209
–16.
34
Skol AD, Scott LJ, Abecasis GR, Boehnke M. Joint analysis is more efficient than replication-based analysis for two-stage genome-wide association studies.
Nat Genet
2006
;
38
:
209
–13.
35
Stram DO, Haiman CA, Hirschhorn JN, et al. Choosing haplotype-tagging SNPS based on unphased genotype data using a preliminary sample of unrelated subjects with an example from the Multiethnic Cohort Study.
Hum Hered
2003
;
55
:
27
–36.
36
Dudbridge F. A note on permutation tests in multistage association scans.
Am J Hum Genet
2006
;
78
:
1094
–5; author reply 1096.
37
Thomas DC, Haile RW, Duggan D. Recent developments in genome-wide association scans: a workshop summary and review.
Am J Hum Genet
2005
;
77
:
337
–45.
38
Kibel AS, Suarez BK, Belani J, et al. CDKN1A and CDKN1B polymorphisms and risk of advanced prostate carcinoma.
Cancer Res
2003
;
63
:
2033
–6.
39
Tomoda K, Kubota Y, Arata Y, et al. The cytoplasmic shuttling and subsequent degradation of p27Kip1 mediated by Jab1/CSN5 and the COP9 signalosome complex.
J Biol Chem
2002
;
277
:
2302
–10.
40
Bardeesy N, Aguirre AJ, Chu GC, et al. Both p16(Ink4a) and the p19(Arf)-p53 pathway constrain progression of pancreatic adenocarcinoma in the mouse.
Proc Natl Acad Sci U S A
2006
;
103
:
5947
–52.

Supplementary data