RAD51 colocalizes with both BRCA1 and BRCA2, and genetic variants in RAD51 would be candidate BRCA1/2 modifiers. We searched for RAD51 polymorphisms by sequencing 20 individuals. We compared the polymorphism allele frequencies between female BRCA1/2 mutation carriers with and without breast or ovarian cancer and between population-based ovarian cancer cases with BRCA1/2 mutations to cases and controls without mutations. We discovered two single nucleotide polymorphisms (SNPs) at positions 135 g→c and 172 g→t of the 5′ untranslated region. In an initial group of BRCA1/2 mutation carriers, 14 (21%) of 67 breast cancer cases carried a “c” allele at RAD51:135 g→c, whereas 8 (7%) of 119 women without breast cancer carried this allele. In a second set of 466 mutation carriers from three centers, the association of RAD51:135 g→c with breast cancer risk was not confirmed. Analyses restricted to the 216 BRCA2 mutation carriers, however, showed a statistically significant association of the 135 “c” allele with the risk of breast cancer (adjusted odds ratio, 3.2; 95% confidence limit, 1.4–40). BRCA1/2 mutation carriers with ovarian cancer were only about one half as likely to carry the RAD51:135 g→c SNP. Analysis of the RAD51:135 g→c SNP in 738 subjects from an Israeli ovarian cancer case-control study was consistent with a lower risk of ovarian cancer among BRCA1/2 mutation carriers with the “c” allele. We have identified a RAD51 5′ untranslated region SNP that may be associated with an increased risk of breast cancer and a lower risk of ovarian cancer among BRCA2 mutation carriers. The biochemical basis of this risk modifier is currently unknown.
Genetic factors are important in breast and ovarian cancer, but less than 10% are attributable to the inheritance of mutations in a single gene, such as BRCA1 and BRCA2 (1) . Although relatively uncommon, female carriers of mutations in the BRCA1 and BRCA2 genes are estimated to have breast cancer risks of 37% to over 85% by age 70 and ovarian cancer risks approximately one half as high (2, 3, 4, 5) . Both BRCA1 and BRCA2 appear to confer similarly high risks of female breast cancer, whereas ovarian cancer risk is higher for BRCA1 mutation carriers and male breast cancer risk is largely confined to BRCA2 carriers (1 , 6) . The heterogeneity of risk in breast and ovarian cancer between different families segregating mutations in the same gene and the estimated average risk of breast cancer of about 50% both suggest the existence of important environmental or genetic modifiers of risk. Some potential genetic modifiers have been identified, including rare polymorphisms in H-ras for ovarian cancer (7) and variation in the androgen receptor gene for breast cancer risk (8) .
The precise mechanisms by which mutations in BRCA1/2 predispose to breast and ovarian cancer are unknown. They are large proteins that share little sequence similarity with each other or with other known sequences. Both appear to be involved in double-strand DNA break repair, particularly because of their association with RAD51 (9) . RAD51 is a homologue of the Escherichia coli recA protein and is involved in recombination and the repair of double-strand DNA breaks (10) . RAD51 shares 60% protein identity with the yeast scRad51, which has been studied extensively (10 , 11) . The relative consistency of the findings of BRCA1 and BRCA2 associating with RAD51, their possible complementary roles as DNA damage repair proteins, and the availability of a model organism system in which to investigate findings led us to consider RAD51 as a candidate modifier locus.
Materials and Methods
Polymorphisms in RAD51 were investigated by complete sequence analysis of the gene in 10 individuals who had developed both breast cancer and at least one other primary cancer obtained from a national cohort of radiological technologists participating in a prospective epidemiological study (12) , and the 5′UTR7 was sequenced in an additional 10 individuals (subjects 01 and 02 from each of the Center d’Etude du Polymorphisme Humain families 1331, 1332, 1347, 1362, and 1413; Coriell Institute, Camden, NJ). Sequencing these subjects revealed a g→c polymorphism at position 135 and a g→t at position 172 of the published cDNA (GenBank accession no. D14134.1)
Two groups were studied to test for allelic association, including: (a) a series of female BRCA1/2 mutation carriers; and (b) Jewish subjects from an Israeli ovarian cancer case-control study. All of the subjects with individually identifiable samples gave informed consent for participation. Study group (a) included all of the known female BRCA1 and/or BRCA2 heterozygous mutation carriers from each of four study centers. Disease associations for the two SNPs were first assessed in two groups of subjects from the NCI. One group was from the Washington Ashkenazi Study, a community-based sample of 86 females who carried the185delAG or 5382 insC mutations in BRCA1 and/or the 6174delT mutation in BRCA2 (3) . The second NCI group consisted of 100 female members of 27 breast/ovarian cancer families with identified BRCA1 or BRCA2 mutations from the NCI Genetic Epidemiology Branch family registry (13) .
To replicate the findings with RAD51:135 g→c, study group (a) was expanded to include subjects from three additional centers. Subjects from the Fox Chase Cancer Center were recruited from high-risk and breast cancer clinics at the Fox Chase Cancer Center and affiliated hospitals in the Delaware Valley, as well as through self- and physician-referral. Subjects from Australia were either members of multiple-case families ascertained in family cancer clinics by the Kathleen Cuningham Foundation Consortium for Research into Familial Breast Cancer (14) or the South Australian Clinical Genetics Service or from the population-based case-control-family studies of the Australian Breast Cancer Family Study/Cooperative Family Registry for Breast Cancer Studies (15) and the Australian Jewish Breast Cancer Study (16) . All of the Jewish BRCA1/2 mutation carriers ascertained from Memorial Sloan Kettering Cancer Center were also studied.
Study group (b) included a subset of subjects from a case-control study of all of the cases of invasive epithelial ovarian cancer diagnosed among Jews in Israel between 1994–1999 and age- and ethnicity-matched population controls tested for the three founder BRCA1/2 mutations (17 , 18) . For study group (b), RAD51:135 g→c SNP typing was completed for 213 (86%) of 249 cases who were BRCA1/2 mutation positive and for random subsets of 304 of the 598 mutation-negative cases and 221 of the 777 mutation-negative controls from this study.
Most BRCA1/2 mutations were identified in the laboratories of the individual study centers using a number of techniques, whereas some were identified by complete sequencing performed at Myriad Genetics Laboratories (Salt Lake City, UT). We were relatively conservative in our definition of mutations, because only two of the 73 included in this study were BRCA1 missense mutations (C61G, observed to cosegregate with disease in multiple families worldwide, and V1713A, an extremely rare variant in a family with a logarithm of the odds score of 1.3 for linkage to BRCA1). The remaining mutations were nonsense, frameshift, splice junction, or large genomic rearrangements. Among group (a) subjects, the two most common mutations were BRCA1:185delAG (n = 216) and BRCA2:6174delT (n = 124).
To identify RAD51 polymorphisms, because the genomic structure of RAD51 was unpublished, we extracted total RNA from lymphoblastoid cell lines, performed reverse transcription-PCR, and sequenced the cDNA in overlapping fragments (conditions and primer sequences available on request). The two 5′UTR SNPs identified could be detected in one single-strand conformational polymorphism/heteroduplex analysis assay using genomic DNA as template and the primers 21F (5′-ccgagccctaaggagagtgcggcgcttc-3′) and 232R (5′-tactcggtccgcagcgctcctctctccag-3′). Thirty-five samples were sequenced and analyzed with single-strand conformational polymorphism/heteroduplex analysis to verify the correlation between the observed banding patterns and the sequences at the two SNPs. Over 97% of the samples could be scored for both loci by blinded review by two individuals. The RAD51:135 g→c SNP was assayed in some subjects using a standard PCR and BstNI digestion specific for the wild-type “g” allele, using the primers 64F (5′-gggaactgcaactcatctgggtt) and 170R (5′-tgg cac gcg ccc gac) at 200 nm concentration to amplify a 107-bp product. The population allele frequencies for the polymorphisms were estimated by genotyping 40 unrelated Ashkenazi spouses and 62 non-Jewish spouses from the NCI registry families and 79 males from the Washington Ashkenazi Study. The RAD51:135c allele frequency was 0.05 among Jews and 0.07 among non-Jews, and the RAD51:172t allele frequencies were 0.43 and 0.41, respectively. The genotype frequencies did not deviate significantly from Hardy-Weinberg equilibrium.
To test for association between the RAD51:135 g→c SNP and cancer risk among the series of 652 female BRCA1/2 mutation carriers (study group a) for breast cancer analyses, the 345 subjects who had been diagnosed with breast cancer were classified as cases and the remaining 307 subjects as controls. Because analyses of ovarian cancer risk were undertaken after those for breast cancer, we excluded all of the subjects who had developed breast cancer, leaving 44 subjects with ovarian cancer and 263 control subjects with neither breast nor ovarian cancer. A subject’s age was taken as the age at diagnosis for cases and the current age, age at bilateral preventive surgery (mastectomy or oophorectomy), or age at death for controls. The ORs comparing SNP carrier frequency between those with and without cancer were obtained using logistic regression models. All of the analyses were adjusted for age, ethnicity (Jewish, non-Jewish), gene (BRCA1 or BRCA2), and study center, although the estimates changed very little with adjustment. Because observations from related subjects may not be independent, adjusted CLs for all of the ORs and Ps were calculated using a bootstrap resampling technique. Two thousand resamplings of the data, grouped on family, were drawn with replacement (19) . The 2.5% and 97.5% empirical values of the regression coefficients were used as the 95% CLs, and the significance level was calculated as twice the proportion of replicates that resulted in a regression coefficient less than zero.
For subjects from the ovarian cancer case-control study (study group b), we compared the prevalence of the RAD51:135 g→c SNP between BRCA1/2 mutation-positive cases versus all of the others with a Mantel-Haenszel OR, adjusted for ethnicity (Ashkenazi/non-Ashkenazi). All of the statistical analyses were performed using S-Plus 2000 Professional (Mathsoft, Inc.).
By complete sequencing of RAD51 cDNA in 10 individuals, no differences were observed in the amino acid-coding portions of RAD51 (compared with GenBank accession no. D14134.1), but we observed a polymorphism in the 5′UTR at nucleotide position 172 g→t. To analyze this polymorphism in our sample of female BRCA1/2 mutation carriers, in which only genomic DNA was available for most, we developed a genomic PCR-based assay that upon sequencing in 10 unrelated individuals from the Center d’Etude du Polymorphisme Humain collection revealed another, less common 5′UTR variant nucleotide at position 135 g→c (GenBank accession no. AF233739.1).
To test for association between the SNPs and cancer risk, we compared the frequencies between female BRCA1/2 mutation carriers with and without breast and/or ovarian cancer with the characteristics shown in Table 1⇓ . Approximately one half of the subjects had breast cancer and approximately one third carried a BRCA2 mutation. The average number of subjects/family ranged from one in the Washington Ashkenazi Study group to 3.7 for the Genetic Epidemiology Branch families. In the hypothesis-generating data from NCI, 14 (21%) of the 67 female BRCA1/2 mutation carriers who had developed breast cancer carried at least one RAD51:135c allele, whereas 8 (7%) of 119 subjects without breast cancer carried a “c” allele (adjusted OR, 3.1; 95% CL, 1.1–14; P = 0.04). There was no difference in the frequency of the three genotypes at RAD51:172 g→t between breast cancer cases (14% t/t; 49% g/t) and controls (18% t/t; 38% g/t).
We sought to replicate the association by determining the RAD51:135 g→c genotypes of an additional 466 BRCA1/2 mutation carriers from three study locations. The frequency of the RAD51:135c was only slightly higher among breast cancer cases (12%) versus subjects without breast cancer (10%; adjusted OR, 1.3; 95% CL, 0.7–2.8; P = 0.3). Combining all of the subjects, 13% of breast cancer cases carried a RAD51:135 “c” allele versus 9% of controls, with an adjusted OR of 1.6 (95% CL, 1.0–3.4; P = 0.06). If we exclude subjects with ovarian cancer in this comparison, the frequency of RAD51:135c was 43 of 317 (14%) among breast cancer cases versus 25 of 263 (10%) among women without cancer. The RAD51:172 g→t SNP was tested in all of the subjects except those from Australia and was again not associated with cancer risk, with observed allele frequencies of 0.41 versus 0.38 among BRCA1-positive breast cancer cases versus those without breast cancer and 0.38 versus 0.37 among BRCA2-positive subjects.
The combined data set had sufficient numbers of subjects to perform analyses restricted to subsets of the data (Table 2)⇓ . There was a statistically significant increased risk of breast cancer associated with the RAD51:135c allele among the 216 BRCA2 mutation carriers (OR, 3.2; 95% CL, 1.4–40; P = 0.01). The median age at breast cancer diagnosis among BRCA2 mutation carriers and at least one RAD51:135c allele was 37 years versus 41 years for those without the SNP (P = 0.02; two-sided nonparametric median test; Ref. 20 ). The magnitude of the risk was strongest for the two study groups in which all of the subjects were Jewish (Washington Ashkenazi Study and Memorial Sloan Kettering Cancer Center). In analyses restricted to the 400 Jewish subjects, an increased risk was present for BRCA1 and BRCA2 mutation carriers combined, with 13% of breast cancer cases carrying the RAD51:135c versus 7% without breast cancer (OR, 2.4; 95% CL, 1.2–5.8; P = 0.02). The associations among the Jewish subjects were relatively consistent across study centers and were stronger for BRCA2 mutation carriers, but small sample sizes precluded reliable statistical testing of subgroups.
When considering the entire group of female BRCA1/2 mutation carriers, there were sufficient numbers of subjects with ovarian cancer to compare the RAD51:135 g→c SNP allele frequencies. But because these analyses were undertaken after the increased SNP frequency was observed among breast cancer cases, we excluded women who had been diagnosed with breast cancer. Ovarian cancer cases were only about one half as likely to carry at least one RAD51:135c allele (2 of 44; 5%) compared with controls (25 of 263; 10%; OR, 0.4; 95% CL, 0.1–1.5; P = 0.3). If breast cancer cases were not excluded, 5 of 72 (7%) ovarian cancer cases carried at least one RAD51:135c allele compared with 68 of 580 (12%) controls.
The suggestion of a lower risk of ovarian cancer led us to determine the RAD51:135 g→c SNP genotype in a completely different study setting. Among a large group of invasive epithelial ovarian cancer cases and controls from Israel, the RAD51:135c allele was present in 6.5% of 62 BRCA2 mutation-positive cases, 10% of 151 BRCA1 mutation-positive cases, 13% of 304 BRCA1/2-negative cases, and 14% of 221 controls (Table 3)⇓ . The OR associated with the RAD51:135c allele, adjusted for ethnicity, was 0.66 (95% CL, 0.4–1.15) for BRCA1/2-positive cases versus all of the other subjects.
We did not observe any sequence variation in the amino acid-coding portion of the RAD51 gene among 10 individuals completely analyzed, which is consistent with another report (21) . Although an amino acid variant has been identified in two Japanese patients with breast cancer (22) , it appears that common amino acid variants in RAD51 are not cancer susceptibility alleles. We did observe two SNPs at nucleotide positions 135 g→c and 172 g→t in the 5′UTR of the gene. The 135 g→c SNP was present in approximately 14% of non-Jewish control individuals in the United States and in approximately 10% of Jewish controls from the United States. On the basis of the frequency of the 135 g→c allele among population-based controls from the ovarian cancer case-control data from Israel presented here, it appears that the SNP is more common in non-Ashkenazi Jews in Israel, because approximately 19% of controls were heterozygous carriers. All of the subjects who were homozygous 135c/c were also homozygous 172 g/g, and among heterozygous 135 g/c individuals, where phase could be determined, the 135c occurred only on the 172 g chromosome, suggesting that the SNPs are in linkage disequilibrium.
We evaluated the two RAD51 SNPs as potential modifying alleles, initially in a series of 186 female BRCA1/2 heterozygous mutation carriers from NCI. Because of the relatively small sample size, we combined BRCA1 and BRCA2 mutation carriers and observed a statistically significant association between the risk of breast cancer and carrying the rarer “c” allele at nucleotide 135 of the RAD51 gene. Female carriers of either a BRCA1 or a BRCA2 mutation who also carried this SNP had about 3-fold increased odds of developing breast cancer compared with BRCA1/2 mutation carriers without the SNP. We attempted to replicate this finding by determining the RAD51:135 g→c SNP in an additional set of 446 female BRCA1/2 mutation carriers from three study centers. Analyzing BRCA1/2 mutation carriers together, the association was not confirmed, with the RAD51:135c allele nearly equally frequent among women with and without breast cancer.
The entire group of 652 BRCA1/2 mutation carriers was large enough to conduct analyses of subgroups, and we analyzed BRCA1 and BRCA2 mutation carriers separately. In the hypothesis-generating data set from NCI, the association was stronger among BRCA2 mutation carriers. In the replication sets, although there was clearly no association among the BRCA1 mutation carriers, more BRCA2 mutation carriers with breast cancer (12%) carried the 135c SNP compared to those without breast cancer (8%). In the entire group of 216 BRCA2 mutation carriers studied, carrying the RAD51:135c allele was associated with a significant risk of breast cancer (OR, 3.2; 95% CL, 1.4–40). Because only nonsense, frameshift, or splicing BRCA2 mutations were studied, 26% of the studied mutations would be expected to have lost all of the eight BRC repeats, the regions of BRCA2 involved in the RAD51 interaction (23) , whereas an additional 69%, including the common 6174delT mutation, would be predicted to have lost at least some of the terminal BRC repeats.
We observed the opposite effect with regard to ovarian cancer risk for the RAD51:135 g→c SNP. None of the 11 BRCA2 mutation-positive subjects with ovarian cancer (and no breast cancer) carried the RAD51:135c allele, compared with 6% of 84 subjects with neither breast nor ovarian cancer. The RAD51:135c allele was also less common in these two groups of BRCA1 mutation carriers, present in 6% of 33 ovarian cancer cases versus 13% of controls. This apparent lower risk of ovarian cancer was explored further in an entirely different study population. Among 731 subjects from the National Israeli Study of Ovarian Cancer characterized as to BRCA1/2 founder mutations and RAD51:135 g→c SNP genotypes, BRCA1/2 mutation-positive ovarian cancer cases were 34% less likely to carry the RAD51:135c allele compared with BRCA1/2 mutation-negative cases and controls, with BRCA2 mutation-positive cases the least likely to carry the SNP. Although not formally statistically significant, the size of the effect was similar to that predicted from the hypothesis-generating observation in the series of female BRCA1/2 mutation carriers.
The populations of BRCA1/2 mutation carriers we studied are not ideal, but given the low frequency of mutations among the general population and the technical difficulties of detecting mutations, optimal epidemiological designs to identify factors that may modify mutation carriers’ cancer risk require inordinately large sample sizes. The female BRCA1/2 mutation carriers consisted predominantly of members of families initially identified in high-risk clinical settings. Almost all of the unaffected mutation carriers were relatives of cancer cases in whom the mutation was initially identified. All of the Washington Ashkenazi Study participants volunteered for a single cross-sectional study. Breast cancer “cases,” therefore, were survivors who volunteered after a variable length of time since their diagnosis and were not a random or epidemiological sampling of incident cases. If survival is different between individuals with different RAD51 SNP genotypes, this could lead to differential participation and bias the results. Survival bias should be somewhat less of a problem in the largely family-based study groups, which are followed prospectively and include both living and deceased individuals.
Is the RAD51:135 g→c SNP association with cancer risk real? The statistical evidence from this study is not overwhelming, with the initial observation, made by analyzing a combined group BRCA1 and BRCA2 mutation carriers, not confirmed in a replication set of over 400 BRCA1/2 mutation carriers. But the more consistent association across study sites for BRCA2 mutation carriers, the supportive findings in an entirely different epidemiological study group from the National Israeli Ovarian Cancer Study, and the replication of the association among BRCA2 mutation carriers by an independent group, in which 17% of 46 mutation carriers with breast or ovarian cancer from Israel carried the RAD51:135c allele compared with 5% of 41 healthy carriers (24) , suggests that it is.
Biochemical evidence as to how this SNP might affect cancer risk would greatly bolster the association but is lacking. Although there are some known regulatory elements associated with 5′UTRs (25 , 26) , it is not immediately clear how the SNPs identified in this study might affect gene function. One possibility is that it affects mRNA splicing, but we did not observe consistent, alternate reverse transcription-PCR products that correlated with the SNP. Other mechanisms, such as regulation of transcription (e.g., through methylation), translation (e.g., through internal ribosomal entry segments), or mRNA stability, remain to be studied. This polymorphism could also be in linkage disequilibrium with another sequence change in a regulatory region of the RAD51 gene or with another nearby gene that affects the incidence of breast cancer in BRCA1/2 mutation carriers. Studies correlating the presence of the polymorphism with RAD51 expression and function will be needed to support the statistical association we observed.
The RAD51:135 g→c SNP was associated with an increased risk of breast cancer and a lower risk of ovarian cancer among the series of female BRCA1/2 mutation carriers studied. This divergence in findings may be attributable to chance or may reflect different biological effects in the two tissues. The differences may also be attributable to the fact that to a large extent breast and ovarian cancer are competing end points. If RAD51 is primarily active in one tissue, e.g., if the SNP is associated with an increased risk of breast cancer, fewer polymorphism carriers would then survive to be diagnosed with ovarian cancer, and one might expect to see opposite associations for the two end points. It is unknown whether the RAD51:135 g→c SNP is also associated with an increased breast cancer risk among the majority of women who do not carry BRCA1/2 mutations.
We thank Washington Ashkenazi Study investigators Patricia Hartge, Sholom Wacholder, and Lawrence Brody; Kathleen Cuningham Foundation Consortium for Research into Familial Breast Cancer investigators Joe Sambrook, Heather Thorne, Jenny Leary, Judy Kirk, Graham Mann, Ted Edkins, Eric Haan, Geoff Lindeman, Kathy Tucker, and Michael Gattas; Australian Breast Cancer Family Study/Cooperative Family Registry for Breast Cancer Studies investigators John Hopper, Deon Venter, Graham Giles, Margaret McCredie, Melissa Southey, Gillian Dite, and Sarah Steinborner; Australian Jewish Breast Cancer Study investigators Carmel Apicella, Leslie Andrews, and Michael Friedlander; South Australian Clinical Genetics Service investigators Frank Firgaira and Barney Rudzski; and National Israeli Study of Ovarian Cancer investigators Shmuel Anderman, Shaul U. Anteby, Jack Atad, Dan Beck, Uzi Beller, Gilad Ben-Baruch, Walter Gotlieb, Ram Dgani, Amiram Fishman, Alexander Kopmer, Tal Levi, Liat Lerner-Geva, Zohar Liviatan, Joseph Menczer, Moshe Oettinger, Benjamin Piura, David Schneider, Shifra Zohar, Eitan Friedman, Haim Yafe, Yeuda Ben-David, Amiram Baraam, Marko Altaras, and Moshe Ben-Ami for contributing data, Xiaoqing Chen (Queensland Institute of Medical Research, Queensland, Australia) for technical assistance, and Robert Clifford (Laboratory of Population Genetics, NCI) for assistance with sequence analysis.
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.
↵1 Supported by the Koodish Foundation, the Queensland Cancer Fund, the Evan M. Frankel Foundation, and the Gloria and Sidney Danziger Foundation.
↵2 Kathleen Cuningham Foundation Consortium for Research into Familial Breast Cancer.
↵3 Australian Breast Cancer Family Study/Cooperative Family Registry for Breast Cancer Studies.
↵4 Australian Jewish Breast Cancer Study.
↵5 National Israeli Study of Ovarian Cancer.
↵6 To whom requests for reprints should addressed, at Laboratory of Population Genetics, Building 41/D702, 41 Library Drive, Bethesda, MD 20892-5060. Phone: (301) 435-8953; Fax: (301) 435-8963; E-mail:
↵7 The abbreviations used are: UTR, untranslated region; SNP, single-nucleotide polymorphism; NCI, National Cancer Institute; OR, odds ratio; CL, confidence limit.
- Received December 31, 2000.
- Revision received May 11, 2001.
- Accepted June 11, 2001.