Association Between Single-Nucleotide Polymorphisms in Hormone Metabolism and DNA Repair Genes and Epithelial Ovarian Cancer: Results from Two Australian Studies and an Additional Validation Set

Ovarian cancer is the leading cause of death from gynaecologic malignancy. The vast majority of malignant ovarian cancers are of epithelial origin and can be classified into four major subtypes: serous, mucinous, endometrioid, and clear cell (1). Mutations in the high-risk breast cancer susceptibility genes, BRCA1 and BRCA2, as well as the mismatch repair genes, MSH2 and MLH1, underlie most “hereditary” ovarian cancers (2). The known ovarian cancer susceptibility genes have been estimated to explain ∼40% of the excess familial risk of ovarian cancer (3). Thus, it is likely that other ovarian cancer susceptibility genes exist. Several genetic models may explain residual familial clustering, but other highly penetrant genes are likely to be rare because mutations in BRCA1/2 are responsible for most families containing three or more ovarian cancer cases. A more plausible alternative is that the remaining familial clustering is driven by variants at multiple loci, each conferring a more moderate risk of the disease. Such variants will also confer risks of nonfamilial ovarian cancer.

The ovarian surface epithelium or epithelial-lined inclusion cysts within the ovarian cortex have traditionally been considered to be the most likely site of origin of epithelial ovarian cancers. The most widely cited hypothesis for the etiology of these cancers is that proposed by Fathalla (4), which states that repeated ovulatory cycles increase the risk of ovarian cancer because of the resulting proliferation of the surface epithelium. If the incessant ovulation hypothesis of ovarian cancer is correct, one might expect that errors arising during DNA synthesis might confer elevated risks of ovarian cancer. Genetic variants in double-strand DNA repair genes, including those involved in homologous recombination, such as XRCC2, XRCC3, BRCA2, and RAD52, may also influence cancer risk.

However, the incessant ovulation model does not take into account all of the known epidemiologic risk factors for ovarian cancer, and these other factors suggest a role for hormones. In particular, ovarian cancer risk is decreased with increasing duration of contraceptive pill use and parity (5, 6). It has been specifically suggested that exposure of the ovarian surface epithelium to androgens may increase cancer risk and exposure to progesterones confer a protective effect (7, 8). Reproductive hormones control normal ovarian function by regulating processes, such as cell proliferation, differentiation, and apoptosis. At ovulation, ovarian surface epithelial cells are exposed to very high levels of estrogens contained within follicular fluid (8), and Syed et al. (9) have shown that an increased concentration of estrogen stimulates ovarian surface epithelial cell proliferation in vitro (9). Exposure to exogenous estrogens has been shown to be related to increased ovarian cancer risk in most studies (10-12).

In this study, we examined the role of specific, putative functional single-nucleotide polymorphisms (SNP) in the SRD5A2, CYP19A1, HSB17B1, and HSD17B4 genes, encoding components of the hormone synthetic pathways. The product of the SRD5A2 gene, 5-α-reductase, catalyzes the conversion of testosterone to the more biologically active dihydrotestosterone. A common nonsynonymous coding SNP (V89L; rs523349) seems to affect the rate of this conversion. Recombinant protein carrying the leucine residue has been shown to result in 30% less testosterone production (13-15). CYP19 encodes the enzyme cytochrome P450c19α aromatase, which catalyses the conversion of androgens to estrogens (16). An SNP in the 3′ untranslated region (rs10046) has been described previously, and the T allele has been linked to “high enzyme activity” and increased CYP19 mRNA levels (17). An association has also been shown between the C allele and decreased circulating estradiol levels (18). The final step of estradiol synthesis (conversion of estrone to the more biologically active estradiol) is catalyzed by type I 17β-hydroxysteroid dehydrogenase, which is encoded by the HSD17β1 gene (16). This enzyme also catalyses the conversion of the weak androgen, androstenedione, to testosterone (19). A nonsynonymous change in the protein sequence (S313G; rs605059) has been studied previously (20), although site-directed mutagenesis shows little effect on catalytic or immunologic activity of the protein (21). In this study, we evaluated the role of HSD17β1 S313G and another nonsynonymous change (A238V) on ovarian cancer risk. Further regulation of estradiol levels occurs by unidirectional oxidation of estrone by type 4 peroxisomal 17β-hydroxysteroid dehydrogenase, encoded by the HSD17β4 gene, and so, we also genotyped a previously unstudied SNP in the 3′ end of this gene (W511R; rs17145454). To our knowledge, this is the first investigation of these SNPs in SRD5A2, CYP19A1, HSB17B1, and HSD17B4 in ovarian cancer.

We have examined SNPs in some of these genes in a previous case-control comparison (22-24) but have now extended this analysis to the Australian Ovarian Cancer Study (AOCS) to provide more statistical power, particularly to evaluate rare variants and different histologic subtypes of ovarian cancer. We present here our analysis of nine SNPs in a subset of genes involved in steroid hormone synthesis and DNA repair using two Australian case-control studies, comprising a total of 1,466 cases and 1,821 controls, as well as the attempted validation of the most convincing finding from this analysis in an independent set of 1,479 cases and 2,452 controls from United Kingdom, United States, and Denmark.

Clinical and epidemiologic data and DNA samples for genotyping were obtained from women who participated in two separate Australian studies of epithelial ovarian cancer and, for the SRD5A2 validation set, from a consortium of three ovarian cancer studies from United Kingdom, United States, and Denmark (25, 26). Approval for all studies was obtained from the relevant Human Research Ethics Committees, and all participants provided informed consent.

This study has been described in detail elsewhere (23, 24, 27-29) but, in brief, it included a total of 510 Caucasian women newly diagnosed before the age of 82 with epithelial ovarian cancer (including borderline tumors) who were recruited as part of a population-based case-control study conducted in New South Wales, Victoria, and Queensland between 1992 and 1995 (n = 341) or treated at the Royal Brisbane Hospital, Queensland, between 1985 and 1996 (n = 169). The comparison group came from two sources: 294 unrelated adult monozygotic twins (one from each pair) recruited into a multicenter study between 1992 and 1993 (30) and 686 control women recruited into a population-based case-control study of breast cancer between 1992 and 2000 (31). As the monozygotic twin control set was not a population-based sample, we compared the distributions of genotypes in the two groups. There were no significant differences, so the two comparison groups were pooled. Epidemiologic data relating to potential confounders, such as parity, oral contraceptive use, hormone replacement therapy, and smoking status were only available for the case and control women recruited through the population-based studies.

The second study was a national population-based case-control study of ovarian cancer, the AOCS. Between January 2002 and June 2006, eligible case women ages between 18 and 79 years and newly diagnosed with epithelial ovarian cancer (including borderline tumors) were recruited primarily in specialized gynecologic oncology units by research nurses. Additional case women missed at the major treatment centers were identified through cancer registries and, with the treating doctor's permission, invited to participate (recruitment through the New South Wales and Victorian Cancer Registries was conducted in parallel under a separately funded study: the Australian Cancer Study). Women who were unable to give informed consent due to language difficulties, mental incapacity, or illness were excluded, as were those whose diagnosis was not histopathologically confirmed. Of the 4,005 women identified as potentially eligible for the study, 806 (20%) were excluded due to death (n = 304), illness (n = 201), or inability to give informed consent (n = 301). Of those invited to participate, 2,714 (85%) agreed to take part (68% of those identified). Of these, a further 730 women (27%) were excluded after pathology review because they were not confirmed as having an eligible cancer. Two researchers independently abstracted information on site of tumor, histologic subtype, invasiveness, and grade from histology reports, and discrepancies were resolved by consensus. A formal review of a full set of diagnostic slides was undertaken for a sample of 87 women by one of a group of gynaecologic pathologists. There was agreement of 97% on tumor site, 98% on tumor behavior, and 99% on tumor subtype between the results of the formal review and the abstracted data.

Control women, frequency matched to the entire case series, were randomly selected from the Australian Electoral Roll (enrollment to vote is compulsory in Australia) after stratifying for age (in 5-year groups) and state of residence. Selected women were mailed study information and subsequently contacted by telephone by research nurses. At least five attempts were made to telephone each woman, and those not contacted were sent a second letter. Women were excluded if they reported a previous history of ovarian cancer or a previous bilateral oophorectomy, as were those women who were unable to give informed consent due to illness, mental incapacity, or language difficulties. Of the potential control women contacted and invited to participate, two percentage were excluded on the basis of illness or language difficulties and of the remaining women, 1,614 (47%) agreed to take part. Of these, 104 women were excluded from analyses because they reported a prior ovarian cancer (n = 7) or bilateral oophorectomy (n = 97).

All participants were asked to complete a detailed health and life-style questionnaire, and 87% of eligible cases (n = 1717) and 85% of controls (n = 1287) provided a sample of blood. The questionnaire covered demographic and physical characteristics, family history, medical and surgical history, life-style habits (including smoking and alcohol consumption), and reproductive and contraceptive histories. Missing information and/or inconsistencies were clarified during a subsequent telephone interview. For this study, we genotyped 956 AOCS cases and 841 AOCS controls for whom DNA was available at the time.

Three case-control studies from United Kingdom (SEARCH), Denmark (MALOVA), and United States (GEOC), comprising 1,479 cases and 2,452 controls, were used for a validation set. These studies have been described in detail previously (25, 26). Analysis included only 1,400 cases and 2,393 controls of self-reported Caucasian ancestry for whom genotype information was available. Tumor histologic subtype breakdown for cases was as follows: serous 47%, mucinous 11%, endometrioid 16%, clear cell carcinoma 8%, other 19%.

DNA extraction methods for study 1 have been described earlier (27, 28). For study 2, DNA was extracted from peripheral blood using a salt extraction method modified from (ref. 32) or QiAMP blood kit (Qiagen, Inc.). All DNA samples were quantitated using a NanoDrop ND-1000 spectrophotometer, and then 5 ng of each gDNA was aliquoted into 384-well plates and dried down at room temperature. Case and control DNAs were randomly assigned to well positions. Each plate included 36 randomly chosen, repeat samples to confirm assay reproducibility and four negative, template-free controls.

SNPs were genotyped using MALDI-TOF mass spectrophotometric mass determination of allele-specific primer extension products using Sequenom's MassARRAY system. The design of oligonucleotides was carried out according to the guidelines of Sequenom, Inc., and was done using MassARRAY assay design software (Version 1.0). Primer sequences are available on request. Multiplex PCR amplification of amplicons containing SNPs of interest was done using Qiagen HotStart Taq Polymerase and Perkin-Elmer GeneAmp 2400 thermal cycler. Primer extension reactions were carried out according to manufacturer's instructions for either homogenous MassEXTEND or iPLEX chemistries. Assay data were analyzed using Sequenom TYPER software (Version 3.0). Seven of the nine SNPs were genotyped with both the homogenous MassEXTEND and iPLEX chemistries to obtain confidence with the new iPLEX chemistry and to confirm deviations from Hardy-Weinberg equilibrium with a different chemistry. The concordance was 99.88% to 100% for each SNP. Whenever the two chemistries provided different results (0.12% genotypes), we omitted the discrepant sample.

The V89L SNP was not amenable to TaqMan (Applied Biosystems) genotyping technology, so for the SRD5A2 validation study, we genotyped rs632148, which was reported to tag the V89L (rs523349) SNP with a pair-wise correlation coefficient (r²) of 0.945. To estimate r² in our own population, we genotyped 30 individuals (33 cases and 57 controls) with each of the three rs523349 genotypes for the rs632148 SNP.

The two Australian data sets were initially analyzed separately to provide a test and replication set and were then combined to increase the power, particularly to examine the different histologic subtypes. Analyses were restricted to Caucasian women (510 cases and 980 controls for study 1 and 956 and 841 for study 2) who made up >95% of each study population. The Hardy-Weinberg equilibrium assumption was first assessed in case and control groups for each SNP using a standard χ² test. Univariate tests of association for each SNP were carried out using the likelihood ratio test with 2 df. Odds ratios (OR) and 95% confidence intervals (95% CI) were calculated using unconditional logistic regression. All models were adjusted for age (at diagnosis for cases and at first contact for controls) as a continuous term, and when data from more than one study were combined, a term for study was also included. Other factors, such as parity (0, 1-2, ≥3), pregnancies (≥6 months of duration), and use of oral contraceptives (never, <5 years, ≥5 years of duration) were not included in final models as they changed point estimates by <10%. Stratification was used to investigate gene-environment interactions, and if this suggested interaction, the statistical significance of the relevant multiplicative term was assessed. Heterogeneity between the results from different studies was assessed using an interaction term for study by each gene. A nominal P value of <0.05 was considered statistically significant. Analysis of the validation set was as above.

There was no evidence of deviation from the Hardy-Weinberg equilibrium in controls from study 1 for eight of the nine SNPs investigated, although there was borderline significant deviation from Hardy-Weinberg equilibrium for the HSD17β1 Gly313Ser SNP (rs605059) gene (P = 0.05). In study 2, there was also little evidence of deviation from the Hardy-Weinberg equilibrium but there was a statistically significant deviation from Hardy-Weinberg equilibrium for BRCA2 (rs144848; P = 0.04) with the results from both homogenous MassEXTEND and iPLEX chemistry (33). Cases and controls from study 2 (mean age, 58 and 57 years, respectively) and cases from study 1 (mean age, 57 years) were on average of very similar in age, whereas the controls from study 1 (mean age, 44 years) were significantly younger (P < 0.0001); all results were thus adjusted for age. The distribution of histologic subtypes of tumors was similar for both studies (Table 1).

The risk of epithelial ovarian cancer associated with the various SNPs was determined separately for study 1 and study 2 (Table 2). For study 1, the only statistically significant association was seen for SRD5A2 (rs523349). The rare allele was associated with an increased risk in a codominant manner (P_trend = 0.02) with the GC genotype showing a 30% increase in risk compared with the GG genotype (1.30; 95% CI, 1.01-1.68) and the CC genotype having a slightly higher risk (OR, 1.44; 95% CI, 0.93-2.24), albeit not significant at the 5% level. Having a C allele for HSD17β4 (rs17145454) was associated with a nonsignificant decreased risk of ovarian cancer (OR, 0.79; 95% CI, 0.56-1.12 for CT versus TT; OR, 0.31; 95% CI, 0.07-1.53 for CC versus TT; P_trend = 0.07), but no associations of note were seen between the other investigated SNPs and risk of ovarian cancer.

A statistically significant association between SRD5A2 genotype and ovarian cancer risk was also seen in study 2 (P_trend = 0.0001). As for study 1, there was a small increase in risk associated with the GC genotype compared with the GG genotype (OR, 1.14; 95% CI, 0.93-1.40) and a larger statistically significant OR associated with the CC genotype (OR, 1.83; 95% CI 1.37-2.44; P = 0.00004). There was also a significant trend of increasing risk per C allele (30% increase in risk per C allele, P = 0.0001). In study 2, the trend of decreasing risk with one or more C alleles was not as marked for HSD17β4 as that seen in study 1, but the CC homozygotes again had a nonsignificant decreased risk of ovarian cancer (OR, 0.59; 95% CI, 0.17-2.11; P = 0.4). In study 2, the AC genotype of BRCA2 (rs144848) was associated with a 36% increase in risk (OR, 1.36; 95% CI, 1.12-1.66) compared with the AA genotype, but those with CC genotypes did not seem to be at increased risk. It is of note that there was significant deviation from Hardy-Weinberg equilibrium for BRCA2 in this group of control women with a deficiency of heterozygotes, but, as described above, this could not be attributed to genotyping errors.

When the data from study 1 and study 2 were combined, there was no evidence of heterogeneity between the studies for any of the genotypes (P_interaction > 0.05). In the combined data set, the association for SRD5A2 rs523349 became highly significant (P_trend = 0.00002). Both heterozygotes (OR, 1.16; 95% CI, 1.00-1.36; P = 0.06) and CC homozygotes (OR, 1.70; 95% CI, 1.35-2.16, P = 0.000009) had increased risk, and there was a significant trend of increasing risk per extra C allele (26% per C allele, P = 0.00002). In the combined analysis, no statistically significant associations were found for HSD17β4, but a small significant risk persisted for those heterozygous for the BRCA2 SNP (OR, 1.27; 95% CI, 1.09-1.49; P < 0.005). In addition, a small, but statistically significant, increase in risk was associated with the TC (OR, 1.20; 95% CI, 1.01-1.43; P = 0.04) but not the CC (OR, 1.15; 95% CI, 0.93-1.41) genotype of CYP19A1 (rs10046).

We used the combined dataset to investigate whether risk associated with the various SNPs might be modified by exposures known to influence ovarian cancer risk. Stratifying analyses by use of oral contraceptives (ever versus never), use of hormone replacement therapy (ever versus never), smoking status (ever versus never), and age (<50 years compared with ≥50 years) did not provide any evidence that the genotype effects varied according to these characteristics. A borderline significant interaction (P = 0.045) associated with parity was found for the RAD52 SNP (rs4987208). Women who were nulliparous and were GT heterozygotes for this SNP had a significant 75% decrease in risk of ovarian cancer compared with TT homozygotes (OR, 0.25; 95% CI, 0.06-0.98), whereas OR for parous women with the GT genotype was 1.06 (95% CI, 0.64-1.75). No other significant interactions were seen for parity.

Variations in risk between histologic subtypes of epithelial ovarian cancer were investigated using the combined data from study 1 and study 2 (Table 3). The positive association between the C allele for SRD5A2 and risk persisted for borderline and invasive tumors of the serous and mucinous subtypes (but did not reach statistical significance for borderline serous tumors) and seemed to be somewhat stronger for invasive mucinous cancers (OR, 2.81; 95% CI, 1.23-6.38 for CC versus GG genotype). We also considered site of tumor as the case groups included women with primary peritoneal and fallopian tube cancers. Although these cancers are clinically and histologically similar to their ovarian counterparts, our previous analyses suggest that there are etiologic differences between ovarian/fallopian tube and primary peritoneal tumors.¹¹

Two associations were statistically significant for invasive clear cell cancers. The CT genotype of HSD17β4 was associated with a 65% decrease in risk of clear cell cancer (OR, 0.35; 95% CI, 0.13-0.96; P = 0.04). There were no women with clear cell tumors who were homozygous CC for HSD17β4. Risk of clear cell cancers was also associated with the HSD17β1 Gly³¹³Ser (rs605059) SNP. Having the GG genotype was associated with a more than 2-fold increase in risk of clear cell cancer compared with having the AA genotype (OR, 2.33; 95% CI, 1.14-4.79; P = 0.02), and there was a significant trend of risk increase associated with each additional G allele (52% per G allele; P = 0.02).

The only nominally statistically significant finding for the DNA repair SNPs related to serous borderline tumors. The AG genotype for XRCC2 (rs3218536) was associated with a 51% decreased risk compared with the GG genotype (OR, 0.49; 95% CI, 0.25-0.95; P = 0.03), although AA homozygotes had a nonsignificant 2-fold increase in risk (OR, 2.12; 95% CI, 0.59-7.58).

Given our similar findings for SRD5A2 (rs523349) in both the Australian case-control studies, we sought to validate this result in a large consortium of three case-control studies from United Kingdom, United States, and Denmark (25, 26). The rs523349 was not amenable to genotyping by TaqMan, so instead we genotyped a 3′ untranslated region tagging SNP in SRD5A2 (rs632148), which tags rs523349 with an r² of 0.945 based on the HapMap genotype data for the CEU samples (30 trios of European origin). To determine the r² in our own population, we genotyped 90 individuals from the AOCS with both the rs523349 and rs632148 SNPs and also showed that, in this sample, the r² between the two SNPs was 0.9.

There was marginal evidence of deviation from the Hardy-Weinberg equilibrium in cases for the validation set overall (P = 0.06), which seemed to be driven by a deficiency of heterozygotes in the SEARCH sample. Although there was no evidence for Hardy-Weinberg disequilibrium in controls overall, there was a similar deficiency of heterozygotes in controls in the MALOVA sample (P = 0.03).

We found no association between this SRD5A2 SNP and ovarian cancer risk in any of these validation case-control sets individually nor when the three studies were pooled (P_trend = 0.9; Table 4). There was no evidence for heterogeneity between the three sample sets, and the combined OR for the validation set was not significantly different from 1 for the heterozygote genotype (OR, 0.93; 95% CI 0.80-1.07) or the homozygote genotype (OR, 1.12; 95% CI 0.89-1.40). Analysis of subgroups defined by histologic subtype revealed no evidence for an increased risk within any particular subtype (data not shown).

When the data from all five studies were combined, there was no evidence of heterogeneity between the studies (P = 0.15), and the pooled OR suggested a statistically significant 40% increase in risk of ovarian cancer among rare homozygotes (P_trend = 0.002).

We used two large Australian case-control populations to test for associations between nine SNPs in genes involved in steroid hormone synthesis and DNA repair and ovarian cancer risk. All analyses were restricted to Caucasians to reduce the potential problems of population stratification. For many of the SNPs we genotyped, we and others had previously reported no or weak associations with ovarian cancer risk (22-24, 29). The main rationale for this study was therefore to have additional power to look for associations with different histologic subtypes of ovarian cancer. In study 1, the only significant finding was seen for SRD5A2 (rs523349), and this increased risk associated with the GC and CC genotypes was also noted in study 2 with a 30% increased risk per C allele (P = 0.0001). This association was seen for both invasive and borderline tumors of the serous and mucinous subtypes of ovarian cancer, but there was no association for endometrioid and clear cell tumors. The exclusion of women who died before recruitment or were very sick means that very aggressive disease may be underrepresented in the case group. However, this is unlikely to have introduced any bias unless the etiology of aggressive cancers differs markedly from that of less aggressive cancers. The borderline significant findings with HSD17B4 (rs17145454) in study 1 and with BRCA2 (rs144848) and CYP19A1 (rs10046) in study 2 were not replicated in the other study. In addition, the modest associations seen for clear cell cancers in the combined analysis are likely to be due to chance as a result of the relatively large number of tests done. No associations were seen for any of the SNPs we genotyped in DNA repair genes in either study 1 or study 2.

SRD5A2 catalyses the conversion of testosterone to dihydrotestosterone, a more potent androgen than testosterone. Functional studies of SRD5A2 variants using recombinant enzyme constructs have shown differential enzyme activities (15). In particular, the valine variant of V89L has an increased turnover rate compared with the leucine variant. Little is known of the role of testosterone in the ovary, but high levels of androgens have been linked to polycystic ovary syndrome (reviewed in ref. 34), and there is some evidence that women with polycystic ovary syndrome are at increased risk of developing ovarian cancer (35). The SRD5A2 V89L SNP has been examined in several prostate cancer studies with inconsistent results (36-43). However, a comprehensive metaanalysis of SRD5A2 V89L frequencies in prostate cancer association studies clearly excluded any increased risk conferred by this allele (44).

We sought to validate the consistent association that we found in studies 1 and 2 for SRD5A2 V89L (rs523349) and ovarian cancer risk in the SEARCH/MALOVA/Stanford consortium (25, 26) using a tagging SNP (rs632148) that correlated with rs523349 (r² = 0.9). The combined sample size of the three validation studies had >95% power to detect an OR of 1.7 for the homozygote genotype; thus, it should have been large enough to see an association even allowing for the less than perfect correlation between the SNPs. Despite this, we did not find any evidence of an association between SRD5A2 rs632148 and ovarian cancer risk in any of these three case-control studies or in combined analyses, although a 40% increase in risk for the rare homozygote could not be excluded (combined OR, 1.1; 95% CI, 0.9-1.4). Despite the strong association seen in the initial studies and the lack of association in the three validation studies, overall, there was no significant heterogeneity between the results of the five studies. Given this lack of heterogeneity, the best estimate of the association is obtained from the combined analysis of all five studies which suggested a significant 40% increase in risk of ovarian cancer among women with the rare homozygote genotype of SRD5A2 V89L or rs632148. Given the key role for SRD5A2 in androgen metabolism, further analysis of SNPs in the SRD5A2 gene is warranted in even larger studies, as would be a comprehensive analysis of SNPs in CYP19A1, HSB17B1, HSD17B4, XRCC2, XRCC3, BRCA2, and RAD52.

Management Group: D. Bowtell (Peter MacCallum Cancer Center, PMCC), G. Chenevix-Trench, A. Green, P. Webb (Queensland Institute of Medical Research, QIMR), A. deFazio (Westmead Hospital), D. Gertig (University of Melbourne).

Project Managers: N. Traficante (PMCC), S. Moore (QIMR), J. Hung (Westmead Hospital).

Data Managers: S. Fereday (PMCC), K. Harrap, T. Sadkowsky (QIMR).

Research Nurses: NSW — A. Mellon, R. Robertson (John Hunter Hospital), T. Vanden Bergh (Royal Hospital for Women), J. Maidens (Royal North Shore Hospital), K. Nattress (Royal Prince Alfred Hospital), Y.E. Chiew, A. Stenlake, H. Sullivan (Westmead Hospital); QLD — B. Alexander, P. Ashover, S. Brown, T. Corrish, L. Green, L. Jackman, K. Martin, B. Ranieri (QIMR); SA — J. White (QIMR); TAS — V. Jayde (Royal Hobart Hospital); VIC — L. Bowes (PMCC), P. Mamers (Monash Medical Center); WA — T. Schmidt, H. Shirley, S. Viduka, Hoa Tran, Sanela Bilic, Lydia Glavinas (Western Australia Research Tissue Network).

Clinical Collaborators: NSW — A. Proietto, S. Braye, G. Otton (John Hunter Hospital), T. Bonaventura, J. Stewart (Newcastle Mater Misericordiae), M. Friedlander (Prince of Wales Hospital), D. Bell, S. Baron-Hay, A. Ferrier, G. Gard, D. Nevell, B. Young (until mid 2003; Royal North Shore Hospital), C. Camaris, R. Crouch, L. Edwards, N. Hacker, D. Marsden, G. Robertson (Royal Hospital for Women), P. Beale, J. Beith, J. Carter, C. Dalrymple, A. Hamilton, R. Houghton, P. Russell (Royal Prince Alfred Hospital), A. Brand, R. Jaworski, P. Harnett, G. Wain (Westmead Hospital); QLD — A. Crandon, M. Cummings, K. Horwood, A. Obermair, D. Wyld (Royal Brisbane and Women's Hospital, RBWH), J. Nicklin (RBWH and Wesley Hospital), L. Perrin (RBWH and Mater Misericordiae Hospitals), B. Ward (Mater Misericordiae Hospitals); SA — M. Davy, C. Hall, T. Dodd, T. Healy, K. Pittman (Royal Adelaide Hospital, Burnside Memorial Hospital), D. Henderson, S. Hyde (Flinders Medical Center), J. Miller, J. Pierdes (Queen Elizabeth Hospital); TAS — P. Blomfield, D. Challis, R. McIntosh, A. Parker (Royal Hobart Hospital); VIC — B. Brown, R. Rome (Freemasons Hospital), D. Allen, P. Grant, S. Hyde, R. Laurie, M. Robbie (Mercy Hospital for Women), D. Healy, T. Jobling, T. Maniolitas, J. McNealage, P. Rogers, B. Susil, A. Veitch, J. Constable, S. Ping Tong, I. Robinson, I. Simpson (Monash Medical Center), K. Phillips, D. Rischin, P. Waring, M. Loughrey, N. O'Callaghan, Bill Murray (PMCC), V. Billson, S. Galloway, J. Pyman, M. Quinn (Royal Women's Hospital); WA — I. Hammond, A. McCartney, Y. Leung (King Edward Memorial Hospital, St. John of God).

Scientific Collaborators: I. Haviv (PMCC), D. Purdie, D. Whiteman (QIMR), N. Zeps (WARTN).

We thank Adele Green and David Purdie for DNA and data from the Survey of Women's Health; Nick Martin for the DNA and data from SSAGA (study 1); New South Wales, Queensland, South Australian, Victorian, and Western Australian Cancer Registries, as well as all the collaborating institutions represented within the AOCS Study Group and the women who participated in the study, for their cooperation; Hannah Munday, Barbara Perkins, Clare Jordan, Mitul Shah, the local general practices and nurses, and Eastern Cancer Intelligence Unit for recruitment of U.K. cases; Lydia Quaye for genotyping the validation set; EPIC-Norfolk investigators for recruitment of U.K. controls; and all the study participants who contributed to this research and all the members of the AOCS Study Group who are listed on http://www.aocstudy.org.