Double-Strand Break Repair Gene Polymorphisms and Risk of Breast or Ovarian Cancer

Deficiencies in the DNA repair process have been hypothesized to increase cancer risk. Excess cancer incidence is a feature of inherited disease caused by defects in DNA damage recognition and repair, including mutations of the BRCA1, BRCA2, ATM, and Chek2 genes (1-5). Epidemiologic evidence suggests that double-strand DNA damage may contribute to the etiology of breast cancer in that ionizing radiation, which primarily causes double-strand DNA breaks, increases risk of breast cancer (6-8). Furthermore, in vitro studies have shown reduced double-strand break repair (DSBR) proficiency after radiation-induced damage in breast cancer patients both with and without a reported family history of breast cancer (9). It has been proposed that common variants in genes of the DSBR pathways that potentially result in altered protein expression or function may confer a modest increase in breast cancer susceptibility. Ionizing radiation has also been associated with increased incidence of ovarian cancer among survivors of atomic bomb explosions (10) but the association with therapeutic radiation is inconsistent (11-13).

To date, the largest population-based case-control study assessing the role of DSBR gene variants in breast cancer susceptibility investigated 15 variants in seven DSBR genes, in up to 2,200 cases and controls (14). This study reported a significant increased risk associated with the rare homozygote genotype for two amino acid substitution polymorphisms, the XRCC2 R188H G > A AA genotype [odds ratio (OR), 2.6; 95% confidence interval (95% CI), 1.0-6.7] and the XRCC3 T241M C > T TT genotype (OR, 1.3; 95% CI, 1.1-1.6). Although there was no statistically significant increase in risk associated with the RAD51 5′ untranslated region −135 G > C polymorphism, the point estimate was increased for the rare homozygote (OR, 2.5; 95% CI, 0.6-10.9).

Using a case-control design, we assessed whether these three polymorphisms (RAD51 5′ untranslated region −135 G > C, XRCC2 R188H G > A, XRCC3 T241M C > T) were associated with risk of breast or ovarian cancer in Australian women.

Approval for this study was obtained from the ethics committees of The University of Melbourne, the New South Wales Cancer Council, The Cancer Council Victoria, and the Queensland Institute of Medical Research. Written informed consent was obtained from all participants.

Details of the methods of enrollment, study conduct, collection of peripheral blood, and DNA extraction have been published previously (15-19) and a brief description is given below.

Breast cancer cases (n = 1,456) and controls (n = 793) were from a population-based case-control-family study, Australian Breast Cancer Family Study, conducted in 1992 to 2000 (15). Cases comprised women aged <60 years living in Sydney or Melbourne who were diagnosed with a first primary invasive breast cancer; controls were a randomly selected population-based sample of unaffected women recruited using the electoral rolls, frequency matched to the cases by age. Interviews were conducted for 1,579 (69%) of 2,304 eligible cases and 1,021 (67%) of 1,531 eligible controls, and DNA was available for 1,456 (92%) of participating cases and 793 (78%) of participating controls. Caucasian ethnicity was reported by 1,305 (90%) cases and 665 (84%) controls, and 975 (67%) cases and 601 (76%) controls were born in Australia. Deleterious germline mutations in BRCA1 or BRCA2 have been detected in 36 cases.

Ovarian cancer cases (n = 548) were women aged 18 to 95 years (median 58 years) with incident epithelial ovarian cancer diagnosed at Royal Brisbane Hospital, Queensland, Australia, between 1985 and 1996 (n = 186) or recruited in New South Wales, Victoria, and Queensland for a population-based case-control study of ovarian cancer conducted in 1992 to 1995 (n = 362; ref. 17). Participation rates for the hospital-based collection were not formally documented. The original participation rate for the case-control study was 89% and DNA samples were available for 44% of cases (blood samples were not collected from controls in this study). For practical reasons, women with DNA were more likely to come from Queensland and to live in metropolitan areas, but the groups did not differ with regard to established ovarian cancer risk factors. Histopathologic information regarding tumor behavior (low malignant potential or invasive), histology, stage, and grade was available for all women; information on ovarian cancer risk factors including age, ethnicity, country of birth, parity, oral contraceptive use, tubal ligation, hysterectomy, and age at menarche was available for the cases from the population-based study. Limited information including age, ethnicity, and country of birth was available for Royal Brisbane Hospital patients. Overall, Caucasian ethnicity was reported by 453 of 483 (94%) women with this information available and 347 of 489 (71%) were born in Australia.

Ovarian cancer controls (n = 335) were an additional group of women included in the ovarian cancer analyses, because blood samples were not collected from controls from the ovarian cancer case-control study and the breast cancer control group was somewhat younger than the ovarian cancer cases. This control group comprised (a) 300 unrelated adult female monozygotic twins (one per pair) from a multicenter study, the Semistructured Assessment for the Genetics of Alcoholism (19), which reported ∼70% response rate for study interview in 1992 to 1993 and (b) 35 controls ages 60 to 69 years recruited through the Australian Breast Cancer Family Study in 1996 to 2000 but excluded from the breast cancer analyses because of the <60-year age restriction on that study. Twin controls were recruited from the eastern states of Australia, and were selected to best match date-of-birth distribution of ovarian cancer cases. Limited information including age, ethnicity, country of birth, parity, and age at menarche was available to assess confounding. Age ranged from 30 to 90 years (median 54 years) for controls overall. Caucasian ancestry was reported by 291 of 311 (94%) and birth within Australia by 251 of 256 women (98%) with this information available.

Genotyping for the RAD51, XRCC2, and XRCC3 polymorphisms was carried out using the ABI Prism 7700 Sequence Detection System (Applied Biosystems, Foster City, CA) methodology as previously described (16). Details of primers and probes are available on request. Genotyping success rates were ≥99% for each genotype for all subgroups, with the exception of XRCC2, with a success rate of 95% for ovarian cases. Reproducibility was absolute for the 186 to 213 individuals replated and rescreened.

The Hardy-Weinberg equilibrium assumption was assessed using standard maximum likelihood methods. Genotype frequencies were compared across groups using the χ² test or the Mantel-Haenzel test for linear trend (for assessment of XRCC3 genotype frequency differences and trends with ovarian cancer stage and grade). Fisher's exact tests were used for 2 × 2 tables when the expected frequency in any cell was <5. OR and 95% CI were calculated using unconditional logistical regression, with and without adjustment for measured risk factors. OR for ovarian cancer risk were calculated by comparison to the combined group of ovarian cancer and breast cancer controls. All statistical tests and P values were two-tailed. SPSS (11.5) and SAS (8.02) software were used for statistical analyses.

Tables 1 and 2 show the genotype distribution for different subject groups. Overall, there was no evidence of deviation from Hardy-Weinberg equilibrium for any of the polymorphisms in any group (P ≥ 0.1). Whereas too few ovarian cancer controls reported non-Caucasian ethnicity or birth outside Australia to evaluate differences in genotype distribution by ancestry, there was evidence of significant variation in genotype frequency by country of birth (Australia versus overseas) and/or self-reported ethnicity (Caucasian versus other) for the other subject groups. Also, statistically significant deviations from Hardy-Weinberg equilibrium were observed among breast cancer cases for RAD51 and XRCC3 (P ≤ 0.05) in the relatively small subgroups defined generically as “non-Caucasian.” Among Caucasian subgroups, there was deviation from Hardy-Weinberg equilibrium among breast cancer controls only (P = 0.005) and this deviation was not observed among the subset of individuals reporting Caucasian ancestry and Australian country of birth (P > 0.2). Given apparent differences in genotype frequencies by reported ancestry and/or country of birth and nonhomogeneity of the relatively small samples of individuals randomly grouped as reporting birth outside Australia or non-Caucasian ethnicity, a term for country of birth was included in all adjusted models and analyses were carried out including all women and also only women reporting Caucasian ethnicity.

There were no differences in genotype frequencies between breast cancer cases and controls for RAD51 (P = 0.5), XRCC2 (P = 0.8), or XRCC3 (P = 0.6). There were no significant associations between genotypes and breast cancer risk (Table 1). Estimates were little changed (<3%) by the exclusion of BRCA1 or BRCA2 mutation carriers, and there were no significant differences in genotype distribution between cases and controls for subgroups defined by age at onset (<40 or ≥40 years), or reported family history (all P > 0.3). Analysis of the effects of combined chest X-ray/radiation exposure (ever/never) or smoking exposure (ever/never or pack-years) provided no consistent evidence that these environmental exposures modified the effect of the RAD51, XRCC2, or XRCC3 genotypes (data not shown). In addition, the combined effect of the three genetic variants was assessed, and based on the individual estimates RAD51 GC/CC, XRCC2 GA/AA, and XRCC3 TT were considered to be the likely at-risk genotypes. In the Caucasian group, a single control (0.2%) and three cases (0.3%) carried all three risk genotypes; moreover, there was no difference between cases and controls in the distribution of the different combinations of the three risk genotypes (P = 0.8) or in the total number of high-risk genotypes (0, 1, or 2-3) carried (P_trend = 0.5).

Overall, there were no significant differences between the distributions of RAD51, XRCC2, or XRCC3 genotypes between ovarian cancer cases and the combined group of breast cancer and ovarian cancer controls (P = 0.3, 0.9, and 0.3, respectively), and no significant association with risk (Table 2). Further adjustment for oral contraceptive use and parity among the subgroup of women with this information reduced the OR by 5% to 10% (results not shown). There was no statistically significant variation in genotype frequency across the different histologic subtypes of ovarian cancer or by tumor stage or grade (data not shown). There was also no variation in XRCC3 or RAD51 between invasive and borderline tumors; however, the XRCC2 GA/AA genotype was significantly more common among women with invasive cancer (16 versus 4%, P = 0.002). As a result, there was no association between the XRCC2 GA/AA genotype and risk of invasive cancer (OR, 1.09; 95% CI, 0.81-1.48) but a statistically significant inverse association was seen for borderline tumors (OR, 0.26; 95% CI, 0.09-0.71). A similar pattern was seen for both serous (13% versus 6%) and particularly mucinous (26% versus 3%) tumors.

Overall, two ovarian cancer cases and two controls (one breast cancer control and one ovarian cancer control) carried all three risk genotypes; all four women were Caucasian. There was no difference between cases and controls in the overall frequency of the different combinations of the three risk genotypes (P = 0.8) or in the total number of high-risk genotypes (0, 1, or 2-3) they carried (P_trend = 0.8).

Our data provide no evidence to support an increase in risk of either breast cancer or ovarian cancer associated with the Rad51 −135 G > C, XRCC2 R188H G > A or XRCC3 T241M C > T polymorphisms. Risk estimates for breast or ovarian cancer were either close to unity, below unity, or driven by the heterozygote genotype. Whereas our analysis of risk of breast cancer might be considered translatable to only women ages <60, there was no evidence for differences in allele or genotype frequency for different age groups or between our control group and those from other studies, suggesting it is also applicable to older women.

The nonsignificant increased risk for breast cancer associated with the RAD51 C variant was driven by the heterozygote genotype, with ORs below unity for the rare homozygous variant. Despite the same allele frequency in controls, this pattern contrasts with the 0.8-fold and 2.5-fold breast cancer risks reported by Kuschel et al. (14) for the GC and CC genotypes, respectively. The RAD51 C variant has not been investigated in any other large case-control studies although it has been reported to modulate breast cancer risk among BRCA1/2 mutation carriers (20-22). The relative risk estimates in our study and that by Kuschel et al. (14) were imprecise, especially for the rare CC genotype, so we cannot exclude the possibility of an increased breast cancer risk associated with the homozygote variant. However, assuming a rare homozygote frequency of 0.3% and a risk estimate of 2.5 based on the combined analysis of published studies to date, even if the association is causal, this genotype would account for <0.3% of all breast cancers.

Several other association studies have reported attempts to replicate the XRCC2 findings of Kuschel et al. (14). A British case-control study (521 cases/895 controls) supported an association with increased breast cancer risk (GA: OR, 1.5; 95% CI, 1.0-2.2; AA: OR, 2.4; 95% CI, 0.5-12.2; ref. 23) and provided suggestive evidence that the risk might be greatest among young women with a family history. Our Australian study, which oversampled women aged <40 years and had detailed family history information on cases and controls, found no evidence of this. A nested case-control study from the Harvard Nurses' Health Study (1,004 cases/1,385 controls; ref. 24) like us found no evidence for an association with the XRCC2 polymorphism (GA: OR, 1.1; 95% CI, 0.8-1.4; AA: OR, 1.3; 95% CI, 0.4-4.2). The frequency of the rare allele differed little between control groups of the four studies (0.06-0.08), and a combined analysis of the crude results from all studies (4,494 cases, 4,104 controls) suggests that there is no increased risk associated with the GA genotype (OR, 1.02; 95% CI, 0.90-1.15) but that an association with the rare AA genotype cannot be excluded (OR, 1.65; 95% CI, 0.95-2.86). However, even assuming a causal relationship, the population-attributable risk for this rare variant would be minor, accounting for <0.4% of all breast cancers.

Our study also found no evidence of an increased cancer risk associated with the XRCC3 T241M C > T polymorphism, although the upper confidence limit of 1.4 for our estimate of breast cancer risk for the TT genotype was similar to the 1.3-fold risk reported by Kuschel et al. (14). Three of five subsequent reports support our findings, reporting OR (95% CI) for the CT and TT genotypes, respectively, of 0.9 (0.7-1.1) and 1.0 (0.8-1.3) for the Nurses Health study nested case-control study (24), 0.9 (0.5-1.4) and 1.1 (0.5-2.3) for a relatively small U.S. hospital-based case-control study (25), and 1.0 (0.8-1.4) and 0.9 (0.6-1.4) for a Danish nested case-control study (26). Two other studies have, however, reported an increased risk for TT homozygotes with OR (95% CI) of 0.9 (0.6-1.3) and 1.4 (0.8-2.4) for a hospital-based case-control study (27) and 0.9 (0.7-1.3) and 1.4 (0.9-2.1) for a Canadian population-based study (28). The frequency of the rare allele in control groups ranged from 0.36 to 0.41, and a combined analysis of the crude results from all seven studies (5,584 cases, 5,012 controls) gave OR (95% CI) of 1.00 (0.92-1.09) and 1.14 (1.01-1.28) for the CT and TT genotypes, respectively. This suggests that the XRCC3 homozygote variant genotype may be associated with a very modest increased risk of breast cancer, which, if true, would account for ∼1.9% of all breast cancers.

The results from the ovarian cancer case-control comparisons largely mirrored those seen for breast cancer. We observed at most a nonsignificant 1.2-fold increase in ovarian cancer risk associated with the RAD51 C variant under a dominant model of inheritance. However, as for the breast cancer case-control comparison, this was driven by the heterozygote genotype. All other genotypes examined had an overall risk close to or below unity. Although there was some variation in risk for XRCC2 when we stratified by invasiveness, these comparisons were based on small numbers and any variation is likely to be due to chance. We conclude that these variants play no substantial role in determining risk of ovarian cancer. To our knowledge, the single other published study assessing ovarian cancer risk in relation to these variants only considered the RAD51 polymorphism among BRCA1/2 mutation carriers and found a reduction in risk of ovarian cancer associated with the C allele (20).

It is relevant and of interest to consider possible functional consequences of the variants under investigation. The only insight we have pertaining to the XRCC2 A allele 188H variant is that is has been shown by cellular complementation assays to slightly increase sensitivity to damage (23). Numerous studies have assessed function of the XRCC3 T241M variant. These reports include no detrimental affect on homozygous recombination (29, 30), no association with glycophorin-A somatic mutation frequency (31), no increased levels of chromosomal aberrations and single-strand breaks (32), a suggestion of slightly increased resistance to Camptothecin (29), and a capability of rescuing the endoreduplication phenotype in XRCC3 −/− cells (33). It is likely that these rather insubstantial differences in function have at most minimal effects on cancer risk.

Our study sample sizes were sufficiently large to detect dominantly inherited risks in the order of 1.4- to 1.5-fold for the rare RAD51 and XRCC2 variants, and 1.3- to 1.4-fold for the XRCC3 variant. Whereas we had 80% power to detect a 1.4- to 1.5-fold risk associated with the homozygote XRCC3 variant genotype, there was little statistical power to detect modest increases in cancer risk for the RAD51 and XRCC2 homozygote variant genotypes. We found no evidence that any of these DSBR variants were associated with a significant substantially increased risk of breast or ovarian cancer even after considering the effects of ethnicity on variant frequency and the effects of certain exposures where data were available. Similarly, combined analysis with other published studies indicated that there was at best evidence for a 1.14-fold increased risk of breast cancer, of borderline significance, associated with the XRCC3 T241M TT genotype. We conclude that the DSBR variants investigated in this study are unlikely to play a major role in the etiology of breast or ovarian cancer.

We thank the physicians, surgeons, and oncologists who endorsed this project; the interviewing staff; and the many women and their relatives who participated in this research; Margaret McCredie for her role in the establishment of Australian Breast Cancer Family Study; Adele Green for ovarian cancer case DNA and risk factor information; Nicholas Martin for ovarian cancer control DNA and risk factor information; Gillian Dite for database management; and Sarah Steinborner and Deon Venter for preparation of breast cancer case and control DNA.