Androgen Receptor Exon 1 CAG Repeat Length and Breast Cancer in Women Before Age Forty Years

JNCI Journal of the National Cancer Institute, Jun 1999

BACKGROUND: We conducted a population-based, case-control-family study to determine whether androgen receptor (AR) exon 1 polymorphic CAG repeat length (CAGn) was a risk factor for early-onset breast cancer in the Australian population. METHODS: Case subjects under 40 years of age at diagnosis of a first primary breast cancer and age-matched control subjects were interviewed to assess family history and other risk factors. AR CAGn length was determined for 368 case subjects and 284 control subjects. Distributions in the two groups were compared by linear and logistic regression, allowing adjustment for measured risk factors. All statistical tests were two-tailed. RESULTS: When analyzed as either a continuous or a dichotomous variable, there was no association between CAGn length and breast cancer risk, before or after adjustment for risk factors. Mean (95% confidence interval [CI]) CAGn lengths were 22.0 (21.8-22.2) for case subjects and 22.0 (21.7-22.3) for control subjects (P = .9). The frequency (95% CI) of alleles with 22 or more CAGn repeats was 0.531 (0.494-0.568) for case subjects and 0.507 (0.465-0.549) for control subjects (P = .4). After adjustment, the average effect on log OR (odds ratio) per allele was 0.16 (95% CI = −0.03 to 0.40; P = .2), and the effect of any allele was equivalent to an OR of 1.40 (95% CI = 0.94-2.09; P = .1). Stratification by family history also failed to reveal any association. Similar results were obtained when alleles were defined by other cutoff points. CONCLUSION: We found no evidence for an association between AR exon 1 CAGn length and breast cancer risk in women under the age of 40, despite having 80% power to detect modest effects.

A PDF file should load here. If you do not see its contents the file may be temporarily unavailable at the journal website or you do not have a PDF plug-in installed and enabled in your browser.

Alternatively, you can download the file locally and open with any standalone PDF reader:

https://jnci.oxfordjournals.org/content/91/11/961.full.pdf

Androgen Receptor Exon 1 CAG Repeat Length and Breast Cancer in Women Before Age Forty Years

Amanda B. Spurdle 0 1 2 Gillian S. Dite 0 1 2 Xiaoqing Chen 0 1 2 Carol J. Mayne 0 1 2 Melissa C. Southey 0 1 2 Leigh E. Batten 0 1 2 Hun Chy 0 1 2 Lynne Trute 0 1 2 Margaret R. E. McCredie 0 1 2 Graham G. Giles 0 1 2 Jane Armes 0 1 2 Deon J. Venter 0 1 2 John L. Hopper 0 1 2 Georgia Chenevix-Trench 0 1 2 0 Oxford University Press 1 Affiliations of authors: A. B. Spurdle, X. Chen, G. Chenevix-Trench (Cancer Unit), C. Mayne (Epide- miology Unit), Joint Experimental Oncology Pro- gramme, Queensland Institute of Medical Research, and The University of Queensland , Brisbane, Aus- tralia; G. S. Dite, J. L. Hopper , Centre for Genetic Epidemiology, The University of Melbourne , Carl- ton , Australia; M. C. Southey, L. E. Batten, H. Chy, L. Trute , Department of Pathology and Research, Peter MacCallum Cancer Institute, Melbourne; M. R. E. McCredie, Cancer Epidemiology Research Unit, NSW Cancer Council , Kings Cross , Australia , and Department of Preventive and Social Medicine, University of Otago , Dunedin , New Zealand; G. G. Giles , Cancer Epidemiology Centre, Anti-Cancer Council of Victoria, Australia; J. Armes, Victorian Breast Cancer Research Consortium and Depart- ment of Pathology, Peter MacCallum Cancer Insti- tute , Melbourne; D. J. Venter , Department of Pathol- ogy and Research, Peter MacCallum Cancer Institute, and Department of Pathology, The Univer- sity of Melbourne. Cancer Unit, Queensland Institute of Medical Re- search, P.O. Royal Brisbane Hospital , Queensland, 4029 , Australia ( 2 Journal of the National Cancer Institute , Vol. 91, No. 11, June 2, 1999 - Background: We conducted a population-based, casecontrol-family study to determine whether androgen receptor (AR) exon 1 polymorphic CAG repeat length (CAGn) was a risk factor for early-onset breast cancer in the Australian population. Methods: Case subjects under 40 years of age at diagnosis of a first primary breast cancer and age-matched control subjects were interviewed to assess family history and other risk factors. AR CAGn length was determined for 368 case subjects and 284 control subjects. Distributions in the two groups were compared by linear and logistic regression, allowing adjustment for measured risk factors. All statistical tests were two-tailed. Results: When analyzed as either a continuous or a dichotomous variable, there was no association between CAGn length and breast cancer risk, before or after adjustment for risk factors. Mean (95% confidence interval [CI]) CAGn lengths were 22.0 (21.822.2) for case subjects and 22.0 (21.722.3) for control subjects (P = .9). The frequency (95% CI) of alleles with 22 or more CAGn repeats was 0.531 (0.4940.568) for case subjects and 0.507 (0.465 0.549) for control subjects (P = .4). After adjustment, the average effect on log OR (odds ratio) per allele was 0.16 (95% CI = 0.03 to 0.40; P = .2), and the effect of any allele was equivalent to an OR of 1.40 (95% CI = 0.942.09; P = .1). Stratification by family history also failed to reveal any association. Similar results were obtained when alleles were defined by other cutoff points. Conclusion: We found no evidence for an association between AR exon 1 CAGn length and breast cancer risk in women under the age of 40, despite having On a population basis, female breast cancer is a familial disease, in that it is observed to cluster in families more often than expected by chance alone. This familial aggregation could be due to genetic and/or environmental factors shared by relatives. The established lifestyle risk factors identified to date by epidemiologic studies have at best a modest effect on risk and themselves display a modest degree of familial aggregation. On face value, mathematical models suggest that lifestyle factors explain less than 15% of the familial aggregation of breast cancer, although this percentage may be greater once measurement error and misclassification of these easily measured surrogates for an underlying hormonal etiology are taken into account (1). On the other hand, although the recently discovered genes BRCA1 and BRCA2 are associated with a substantial, dominantly inherited increased risk of disease of at least 10- to 20-fold in female mutation carriers (Hopper JL, Southey MC, Dite GS, Jolley DJ, Giles GG, McCredie MR, et al.: manuscript submitted for publication), mathematical models suggest that, because of the rarity of such carriers (somewhere in the order of one in 100 to one in 1000), these genes will explain much less than 50% of familial aggregation of breast cancer, perhaps at the most 20% (2). Therefore, it is possible that there are low-risk genetic factors (an order of magnitude more common than the highrisk mutations in BRCA1 or BRCA2) that explain a substantial proportion of familial aggregation of breast cancer (2). Candidates for such low-risk genes would include genes likely to be involved in cancer predisposition that contain common but subtle variants. Such genes would include those mediating a range of functions, such as DNA repair, steroid hormone metabolism, signal transduction, and cell cycle control. Variants of particular interest would include genetic polymorphisms that affect gene expression or function through modified transcription of DNA, through altered stability, through processing or translation of messenger RNA, or by amino acid substitution in the expressed protein. Exposures to endogenous and exogenous steroid hormones are known to influence breast cancer risk, and hormonal signals are manifested via hormone receptors. The androgen receptor (AR) gene is involved in various pathways, including differentiation, development, and regulation of cell growth, and a role of the AR gene in cancer predisposition is indicated by reported associations between prostate cancer risk and the length of the polymorphic exon 1 CAG repeat (CAGn) within the AR transactivation domain (37). For prostate cancer, an increased risk for smaller repeat lengths (such as CAGn <20 or <22) has been observed (37). One of the larger studies of 269 high-grade prostate cancer case subjects and 588 control subjects found a relative risk of 2.1 (95% confidence interval [CI] 4 1.14.0) for CAGn <19 (6), while a smaller study of 281 prostate cancer case subjects and 246 control subjects found a relative risk of 2.2 (95% CI 4 1.14.7) for CAGn <22 in a subgroup of relatively thin individuals (Quetelet index, computed as weight in kilograms divided by height in meters squared, <24.4) (5). Furthermore, case subjects with earlier onset disease have been reported to have shorter repeat lengths (8). Biological significance of the CAG repeat length variation is suggested by in vitro studies (9), which demonstrated that smaller repeat lengths exhibit greater transactivation capabilities. Furthermore, greatly expanded CAG repeat lengths (>39 repeats) are associated with spinal and bulbar muscular atrophy (SBMA) in vivo (10). The biological importance of repeat length variation is emphasized at even the extreme lengths within the SBMA range, with an increase in repeat length correlating with age at onset of this disorder and also with the likelihood of clinical manifestation of gynecomastia (11). In addition, the importance of repeat length variation within the normal range is suggested not only by the prostate cancer studies detailed above, but also by a more recent report of an association between CAG repeat length >28 and increased risk of impaired spermatogenesis (12). Several lines of evidence suggest a role for the AR gene in breast cancer risk. Inactivating mutations in the hormonebinding domain in male breast cancer patients have been documented (13,14). In addition, seven different human breast cancer samples and three different breast cancer cell lines have been shown to express high levels of an AR splice variant that lacks exon 3, the region encoding the second zinc finger of the DNA-binding transactivation domain (15). Furthermore, a report in abstract form of a study of 190 BRCA1 mutation carriers (16) purported to show that the AR exon 1 CAGn may act as a modifier of BRCA1-associated risk of breast cancer, with inheritance of at least one allele of CAGn 29 apparently associated with earlier age at disease onset. We undertook a population-based study to establish whether there was evidence for an association between earlyonset breast cancer and the length of the AR exon 1 CAG repeat. SUBJECTS AND METHODS A population-based, casecontrol-family study of early-onset breast cancer was carried out in Melbourne and Sydney, Australia, during the period 1992 to 1995 (17). The study protocol was approved by the relevant ethics committees, and written informed consent was obtained from each subject (17). Women under the age of 40 years at diagnosis of a first primary breast cancer were identified though the Victorian and New South Wales cancer registries. Control subjects (women without breast cancer) were selected from the electoral roll (registration for voting is compulsory in Australia) by use of stratified random sampling and frequency matched for age. Case subjects, control subjects, and relatives of both case and control subjects were administered the same questionnaire on risk factors. For each case subject and control subject, a detailed family history was systematically recorded for all first- and second-degree relatives and subsequently checked with their living relatives at the time of their interview. Verification of all family cancers reported by case subjects, control subjects, 962 REPORTS or relatives was sought through cancer registries, pathology reports, hospital records, treating clinicians, and death certificates. Blood samples were collected from all case subjects and control subjects at the time of interview. Of 644 eligible case subjects, 467 (72.5%) were interviewed. Attrition was due to death (1.7%), refusal (surgeon, 8.4%; patient, 11.8%), no response (surgeon, 0.6%; patient, 1.4%), or having changed residence (3.6%). Of the 633 eligible control subjects, refusals (25.8%) and no responses (9.8%) resulted in 408 control subjects being interviewed (64.4%). Blood samples were available from 393 case subjects (84.2% of participating and 61.0% of eligible case subjects) and from 295 control subjects (72.3% of participating and 46.6% of eligible control subjects). AR gene analysis was performed for 368 case subjects (78.8% of participating and 57.1% of eligible case subjects) and for 284 control subjects (69.6% of participating and 44.9% of eligible control subjects). Selection of case subjects and control subjects for AR gene analysis was not made on the basis of measured risk information but rather on the basis of DNA availability. For case subjects and control subjects, there was no difference between those included and those not included in the AR gene analysis for factors shown to be associated with breast cancer in the full sample of case subjects and control subjects (17). A greater proportion of Victorian participants than of New South Wales participants was included in the AR gene analysis for resource reasons unrelated to AR genotype. For the case subjects, 151 (69%) of 219 participating Victorian subjects were included in the AR gene analysis versus 217 (88%) of 248 participating New South Wales subjects (P<.001). For control subjects, 154 (86%) of 180 participating Victorian subjects were included in the AR gene analysis versus 130 (57%) of 228 participating New South Wales subjects (P<.001). With regard to family history of breast cancer, 49 (13.3%) of the 368 case subjects in the AR gene analysis had an affected first-degree relative compared with 58 (12.4%) of all interviewed case subjects. For control subjects, these numbers (percentages) were 17 (6.0%) and 21 (5.2%), respectively. AR gene analysis was performed for 49 (84.5%) case subjects and for 17 (81.0%) control subjects with an affected first-degree relative and for 319 (78.0%) case subjects and for 267 (69.0%) control subjects without an affected first-degree relative. Molecular Analysis Collection of peripheral blood and DNA extraction were described previously (18). The AR exon 1 CAG trinucleotide repeat was amplified by polymerase chain reaction (PCR) with the use of primer sequences detailed by La Spada et al. (10), with inclusion of a 58-6-carboxy-4,7,28,78-tetrachlorofluorescein (58-TET)-labeled forward primer to generate a fluorescent product. The 10-mL reaction mixture contained 30 ng of DNA, primers (10 pmol each), deoxynucleotide triphosphates (200 nM), 1 Taq polymerase buffer (The Perkin-Elmer Corp., Foster City, CA), 1 U of Taq polymerase, 1.5 mM MgCl2, and 7% deionized formamide. Amplification conditions were as follows: 2 minutes at 94 C and 34 cycles at 94 C for 20 seconds, 62 C for 20 seconds, and 72 C for 20 seconds, followed by a 10minute extension at 72 C. Amplified samples were diluted 1 : 12 in formamide loading buffer, denatured for 2 minutes at 95 C, and separated by size on a 6% denaturing polyacrylamide gel. The ABI Prism 373 Genescan and Genotyper systems (The Perkin-Elmer Corp.) were used for detection and sizing of fluorescent products. Separation of the ABI TAMRA-350 size standard in each lane allowed for Genescan automated sizing of 58-TET-labeled PCR products. In addition, control samples of known size were separated at different positions across each gel. Consistent sizing of samples was indicated by the independent generation of matching size results for a random subset (18.2%) of samples separated on more than one gel. PCR-amplified samples from both case subjects and control subjects were loaded randomly on gels to further avoid any sizing bias. Immunohistochemical Studies Immunohistochemical studies were performed on sections obtained from paraffin blocks of tissues fixed in 10% neutral buffered formalin. Sections (3 mm) were cut from paraffin blocks, placed onto silane-coated slides, and dried at 60 C for 30 minutes. The sections were dewaxed in Histolene (Fronine, Riverstone, New South Wales, Australia) and rehydrated through graded alcohols. Antigen was retrieved by heating the sections for 2 minutes at pressure in a pressure cooker in 10 mM sodium citrate (pH 6.0). The Autostrainer (Dako Corp., Carpinteria, CA) was used for subsequent steps. All washes used 50 mM TrisHCl (pH 7.6), containing 0.05% Tween. The sections were treated with 3% hydrogen peroxide for 10 minutes to inactivate endogenous peroxidase activity. They were then incubated with monoclonal estrogen receptor (ER)- or progesterone receptor (PR)-specific antibodies (Dako Corp.) at a 1 : 50 dilution for 30 minutes at room temperature. Sections were incubated with biotinylated secondary antibody, followed by peroxidase-conjugated streptavidin by use of the Universal DAKO LSAB2 kit (Dako Corp.) at 10 minutes for each step. Staining was visualized by use of 3-amino-9-ethyl-carbazole; the sections were washed in water and counterstained with hematoxylin. Crystal Mount (Biomedia, Foster City, CA) was applied to the sections and dried on a 60 C hot plate. For histology, the sections were then mounted under coverslips with DPX Mountant (Fluka Chemical Corp., Ronkonkoma, NY). Staining of invasive carcinoma cells with the use of ER- and PR-specific antibodies was scored for intensity and proportion of positive cells by a modification of the method described by Allred et al. (19). The proportion score represented the estimated fraction of positive staining cells (0 4 10%; 1 4 11%25%; 2 4 26%50%; 3 4 51%75%; 4 4 76%90%; 5 4 91%). The intensity score represented the estimated average staining intensity of positive cells (0 4 none; 1 4 weak; 2 4 moderate; 3 4 strong). The results were then analyzed to give a semiquantitative estimate of the expression levels of antigen in the tissue. Intensities of 0 or 1 were designated as negative to weak expression. For intensity scores of 2 and 3, a combined score was derived by adding the intensity and proportion scores. Combined scores of 2 and 3 were designated as negative to weak expression; combined scores from 4 to 6 were designated as moderate expression, and scores of 7 or 8 were designated as strong expression. Only nuclear staining was scored. Statistical Methods The distributions of the average, the smaller, and the larger CAGn alleles were compared between case subjects and control subjects by Students t test and by analysis of covariance, allowing adjustment for measured risk factors. Analyses of allele frequencies were carried out as described by Southey et al. (18), with the use of logistic regression. For each analysis, two alleles were defined according to the CAG repeat length. For example, the major analysis involved defining one allele as CAGn <22 and the other as CAGn 22. This cutoff point was chosen because it divides the distribution of CAG repeat lengths in controls approximately in half and is near to the mode of CAGn 4 21. It also happens to be one of the cutoff points reported to show an association with prostate cancer (5). Further analyses used other cutoff points, such as CAGn <20 as described by Ingles et al. (4) and CAGn 29 as described by Rebbeck et al. (16), although there were few observations in those extreme categories. Under HardyWeinberg equilibrium, the maximum likelihood estimator of the frequency of a particular allele is f 4 (2n11 + n01)/2n, where n 4 n11 + n01 + n00 and nij is the observed number of subjects with the ij genotype (i 4 0; j 4 1), where 1 represents presence of the allele or a group of alleles and 0 represents the absence and has asymptotic standard error (SE) [(f[1 f])/2n]1/2. The Hardy Weinberg equilibrium assumption was assessed by comparing the observed numbers of individuals with different genotypes with those expected under HardyWeinberg equilibrium for the estimated allele frequency and comparing the Pearson goodnessof-fit statistic with a x2 distribution with 1 df. Given no evidence of departure from Hardy Weinberg equilibrium, the allele frequency was analyzed and modeled as a function of risk factors by use of linear logistic regression, by assuming that the number of 1 alleles (as described above) was a binomial variable with n 4 2. The influence of the AR genotype on risk of breast cancer was assessed, as in standard casecontrol analyses, by use of unconditional multiple linear logistic regression, with and without adjustment for measured risk factors. Genotype was modeled six ways: by number of alleles (two parameters), by a linear effect per number of alleles (one parameter), by an effect of any allele, and by linear effects of the average allele (i.e., average of smaller and larger alleles), the smaller allele, and the larger allele sizes (one parameter each). All analyses were performed with the use of STATA statistical software (20). All statistical tests and P values were two-tailed. Following convention, statistical significance was taken as a nominal P value of less than .05. RESULTS The distributions of the AR exon 1 CAG repeat length genotypes for case subjects and control subjects are displayed in Table 1. The distributions of the average, the smaller, and the larger of the two CAGn alleles were each approximately normal, and the cumulative distribution curves were virtually identical for case subjects and control subjects (Fig. 1). For example, for the average of repeat lengths, the mean (95% CI) of the average was 22.0 (21.822.2) in case subjects, which was no different from 22.0 (21.7 22.3) in control subjects (P 4 .9). After adjustment for the previously identified risk factors for breast cancer, the difference was 0.01 and remained no different from zero (P 4 .5). Similarly, the means of the smaller and the larger of the CAGn alleles were almost identical. The SE of the difference in means was about 0.17, so there was 80% power at the .05 level of significance to detect differences in means in excess of 0.43 repeat length, or 0.2 standard deviation. The frequencies of alleles defined by the cutoff CAGn 22 are shown in Table 2 for case subjects and control subjects. There was no evidence of deviation from HardyWeinberg equilibrium in case subjects (x2 4 0.36; df 4 1; P 4 .5), in control subjects (x2 4 0.22; df 4 1; P 4 .6), in total (x2 4 0.02; df 4 1; P 4 .9), or in any group or subgroup defined by casecontrol and family history status. There was no difference in the frequency of alleles between case subjects and control subjects overall (P 4 .4), in those with a reported family history of breast Fig. 1. Androgen receptor exon 1 CAGn allele distribution in breast cancer case subjects and control s u b j e c t s . T h e t e r m s small allele refer to the smaller of two alleles in a given genotype and large allele to the larger of two alleles in a given genotype, while average represents the average of the two alleles of an individual genotype. Family historyyes Family historyno cancer (P 4 .4), or in those without a family history of breast cancer (P 4 .5). When women with and without a reported family history were compared, there was no difference overall (P 4 .7), in case subjects (P 4 .8), or in control subjects (P 4 .7). For different definitions of reported and verified family history (any firstdegree relative affected; any first- or second-degree relative affected), the allele frequency for women with a family history of breast cancer was no different from that in women without a family history overall (P 4 .2.7), in case subjects (P 4 .4.9), and in control subjects (P 4 .09.7). For example, the allele frequency (95% CI) was 0.51 (0.460.56) in women 964 REPORTS who reported at least one affected first- or second-degree relative and 0.52 (0.49 0.56) otherwise (P 4 .7). The frequency of alleles did not differ according to age, country of birth, state of residence, highest level of education (as a surrogate for socioeconomic status), marital status, or any of the other risk factors for breast cancer measured by the questionnaire. After adjustment for these factors, there was no difference in allele frequency between case subjects and control subjects (OR 4 1.41; 95% CI 4 0.95 2.09; P 4 .09). There was also no difference between women with or without a family history of breast cancer, whether in case subjects and control subjects combined (OR 4 0.99; 95% CI 4 0.661.50; P 4 1.0), in case subjects only (OR 4 1.00; 95% CI 4 0.581.74; P 4 1.0), or in control subjects only (OR 4 0.98; 95% CI 4 0.521.88; P 4 1.0). Table 3 shows that, irrespective of how the CAGn 22 allele status was modeled, there was no association with breast cancer, either before or after adjustment for the risk factors identified in the full dataset (17). After adjustment, the average effect on log OR per allele 22 CAGn was 0.16 (95% CI 4 0.03 to 0.40; P 4 .2), and the effect of any allele 22 CAGn was equivalent to an OR of 1.40 (95% CI 4 0.942.09; P 4 .1). The SEs on the log OR scale were about 0.12, so that effects equivalent to an OR of 1.35 or more would have been detectable at the No. of alleles with CAGn 22 Linear Average No. of repeats Effect per repeat Larger No. of repeats Effect per repeat Smaller No. of repeats Effect per repeat Crude OR (95% confidence interval) Adjusted OR (95% confidence interval) Referent 1.40 (0.922.15); x2 4 2.56 1.40 (0.872.26); P 4 1.0 1.10 (0.881.37); P 4 .4 1.18 (0.941.49); P 4 .2 1.00 (0.931.08); P 4 .9 1.02 (0.941.11); P 4 .6 1.00 (0.941.07); P 4 .9 1.01 (0.941.08); P 4 .8 1.00 (0.951.07) P 4 .9 1.02 (0.961.09); P 4 .5 .05 level of significance with more than 80% power. For the average number of repeats, the crude linear effect per repeat was 0.00 (95% CI 4 0.07 to 0.08) on the log odds scale and 0.02 (95% CI 4 .06 to 0.10) after adjustment (P 4 .6). A similar result occurred for analysis of the larger and of the smaller of the repeat lengths, demonstrating virtually no effect when unadjusted (Fig. 1) and a small and not significant effect after adjustment for risk factors (P 4 .8 and .5, respectively). Given the SEs of 0.04 and the standard deviation of the CAGn distribution, the study had 80% power at the .05 level of significance to detect effects in excess of 10% per repeat length. For women with a reported family history of breast cancer (first- or seconddegree relative), the crude OR for presence of the CAGn 22 allele was 1.26 (95% CI 4 0.632.49; P 4 .5); after adjustment for risk factors as in Table 3, it was 1.36 (95% CI 4 0.632.95; P 4 .4). For women without a reported family history, the crude OR was 1.24 (95% CI 4 0.801.91; P 4 .3); after adjustment, it was 1.35 (95% CI 4 0.842.15; P 4 .2). Therefore, there was no evidence that having 22 or more CAG repeats has an effect on risk of breast cancer in women with a family history of breast cancer or in women without a family history. Case subjects were stratified by morphology to assess whether CAG repeat length exhibited an association with cancers of a particular morphologic subtype. The majority of tumors in case subjects (83.5%) were classified as ductal cancers, and analysis of this subgroup gave OR estimates of 1.21 (95% CI 4 0.831.77; P 4 .3) and 1.32 (95% CI 4 0.881.98; P 4 .2) for unadjusted and adjusted analyses, respectively, for a linear effect of number of alleles with CAGn 22. ER and PR status was determined for tumors from a proportion of case subjects. ER and PR status is a known prognostic indicator, with less differentiated and presumably higher grade tumors lacking receptors, and is a common basis for stratification of breast cancers. Case subjects were stratified on the basis of ER and PR status to evaluate the possibility that CAG repeat length may be a risk factor for breast cancers of a particular ER or PR expression level. There was no evidence of an association between the presence of the allele CAGn 22 and moderate or strong ER expression (91 case subjects), with unadjusted and adjusted OR estimates of 1.09 (95% CI 4 0.323.78; P 4 .9) and 1.38 (95% CI 4 0.316.21; P 4 .7), respectively. Similarly, there was no evidence of an association between the presence of the allele CAGn 22 and moderate or strong PR expression (92 case subjects), with unadjusted and adjusted OR estimates of 0.81 (95% CI 4 0.252.61; P 4 .7) and 0.91 (95% CI 4 0.223.72; P 4 .9), respectively. Most of the samples included in this study have also been typed for the ER codon 325 amino acid substitution polymorphism (18). Although our study of 388 case subjects and 294 control subjects found no evidence for an association of the ER codon 325 polymorphism with breast cancer in women under the age of 40 years, we analyzed the data from the 648 individuals typed for both the ER and AR polymorphisms. There was no evidence for a genegene interaction between the AR CAG repeat polymorphism (as defined by the cut point 22 CAGn) and the ER codon 325 polymorphism in the etiology of breast cancer before age 40, either with (OR for genegene interaction 4 0.89; 95% CI 4 0.402.01; P 4 .8) or without (OR 4 0.93; 95% CI 4 0.441.98; P 4 0.9) adjustment for risk factors previously identified (18). Analyses using the extreme cutoff points of CAGn <20 and CAGn 29 were also undertaken to allow comparison with published data for prostate cancer risk (5) and breast cancer risk (16). The allele frequency (95% CI) for CAGn <20 was 0.168 (0.1420.198) in case subjects, no different from 0.176 (0.1460.210) in control subjects (P 4 .7), whereas the allele frequency (95% CI) for CAGn 29 was 0.026 (0.0160.040) in case subjects and 0.020 (0.0100.034) in control subjects (P 4 .5). Similar results were also obtained for other cut points, such as 19, 21, and 27. Finally, of the 11 case subjects we have identified to date who carry a germline protein-truncating mutation in BRCA1 (Hopper JL, Southey MC, Dite GS, Jolley DJ, Giles GG, McCredie MR, et al.: manuscript submitted for publication), none carried a CAGn 29. The mean (95% CI) of the average repeat lengths was 21.5 (20.222.8) as compared with 22.0 (21.822.2) for case subjects and 22.0 (21.722.3) for control subjects, respectively. DISCUSSION This Australian population-based study showed no evidence for an association between AR exon 1 CAG repeat length and breast cancer risk in women under the age of 40 years, even when restricted to women with or without a family history of breast cancer. If CAGn was a risk factor for breast cancer, one would anticipate a different distribution of CAGn in women with a family history of breast cancer, whether they be case subjects or control subjects, but we found no evidence of this. Furthermore, we found no evidence of larger CAGn or more CAGn 29 in BRCA1 mutation carriers, although we had little power to address this hypothesis. The hypotheses arising from the studies of prostate cancer (38) and breast cancer (1316) discussed in the introduction were thoroughly tested by the analyses carried out. Cutoff points based on the median size in control subjects (CAGn <22), allele sizes reported to confer risk for prostate cancer (CAGn <20), and modifying risk allele sizes purported for BRCA1 mutation carriers (CAGn 29), and other cut points, all failed to reveal an influence on risk of breast cancer before the age of 40. This study was sufficiently large to have good statistical power to detect modest effect sizes, such as a difference in means of 0.2 standard deviation or an effect of 1.35 or more for CAGn 22. Furthermore, the inability to distinguish between the active and inactive X allele of female case subjects and control subjects was obviated by testing the risk differences between individuals with no, one, or two alleles within a risk category group. Neither that analysis nor the analysis of average allele (on the presumption of random X inactivation of the AR gene in target tissues) showed an association. We intend to genotype other candidate genes involved in steroid hormone metabolic pathways and to test for genegene interactions. We will also test for gene environment interactions by comparing the distributions of environmental and lifestyle factors measured by questionnaire across case subjects with different genotypes. Because such analyses involve a multitude of statistical tests, nominally significant findings must be treated with caution. Replication is essential in establishing credible results. As in this report, we shall publish the data in its raw form, so as to allow pooling with other similar population-based studies, and we encourage others to do likewise. Supported by the National Health and Medical Research Council of Australia, the Victorian Health Promotion Foundation, the NSW Cancer Council, and the Peter MacCallum Cancer Institute. We thank Alana Goldman and Joanne Voisey of the Queensland Institute of Medical Research for technical assistance with this project. Confirmation of the results was possible through the provision of size standards by Steve Edwards of the Institute of Cancer Research, Sutton, Surrey, Najah Nassif of Sydney University, and Wayne Tilley and Grant Buchanan of Flinders University. We are grateful to the physicians, surgeons, and oncologists in Victoria and New South Wales who endorsed this project, to the interviewing staff, and to the many women and their relatives who participated in this research. Manuscript received November 12, 1998; revised March 5, 1999; accepted April 5, 1999.


This is a preview of a remote PDF: https://jnci.oxfordjournals.org/content/91/11/961.full.pdf

Amanda B. Spurdle, Gillian S. Dite, Xiaoqing Chen, Carol J. Mayne, Melissa C. Southey, Leigh E. Batten, Hun Chy, Lynne Trute, Margaret R. E. McCredie, Graham G. Giles, Jane Armes, Deon J. Venter, John L. Hopper, Georgia Chenevix-Trench. Androgen Receptor Exon 1 CAG Repeat Length and Breast Cancer in Women Before Age Forty Years, JNCI Journal of the National Cancer Institute, 1999, 961-966, DOI: 10.1093/jnci/91.11.961