The Androgen Receptor CAG Repeat Polymorphism and Risk of Breast Cancer in the Nurses’ Health Study¹ Free

Estrogen and progesterone regulate multiple normal physiological processes, including the growth and development of the mammary gland (1). Through binding to their respective receptors, these hormones stimulate breast cell proliferation, resulting in greater opportunity for the occurrence and clonal propagation of nucleotide sequence errors caused by carcinogens or errors in DNA replication. The established reproductive breast cancer risk factors, such as early age at menarche and late age at menopause, provide an indirect link between dose and duration of exposure to estrogens and progesterone and breast cancer risk (2). The most direct evidence supporting the role of endogenous steroid hormones in the etiology of breast cancer has come from prospective epidemiological studies in which plasma estrogen levels are higher in women subsequently diagnosed with breast cancer compared with those who remain disease free (3, 4, 5, 6).

Among postmenopausal women, androgen levels are also positively associated with breast cancer risk (4, 5, 6, 7, 8, 9). The role of endogenous androgens in breast cancer development, however, is less clear, and hypotheses have emerged surrounding their etiological potential (2, 10). Androgens may influence breast cancer risk indirectly through their conversion to estradiol; directly, by binding to the AR³ and either promoting or opposing breast cell growth, or through binding to ER-α and directly stimulating breast cell proliferation (11). Epidemiological studies support the indirect hypothesis of androgen action; the strength of the association between total testosterone levels and breast cancer risk weakens after controlling for estradiol levels (5, 6, 7). Positive associations between adrenal androgens DHEA, DHEAS and androstene-3β, 17β-diol and breast cancer risk have also been observed (4, 6, 8, 9), although whether this effect is mediated by the AR, through their conversion to testosterone, is unclear. Laboratory studies provide support for the indirect and direct effects of adrenal androgens in breast cancer development. Physiologically relevant androgen concentrations have been shown to activate ER-α and stimulate proliferation of breast cancer cell lines (11), and intratumoral aromatase activity has been shown to promote in situ conversion of androgens to estrogens and tumor growth in animal models (12).

On the basis of the hypothesized roles of endogenous androgens in breast cancer development, variation in genes involved in androgen metabolism or hormone signaling may influence susceptibility to breast cancer. Genetic modification of the AR has been studied extensively in association with prostate cancer (13, 14, 15, 16). An association has been reported between CAG repeat number within exon 1 of the AR and prostate cancer risk, with an increase in risk observed for shorter [CAG]_n repeat lengths (14, 15, 16). In vitro studies also provide evidence of biological function for the repeat polymorphism as longer alleles have decreased AR transactivational activity (17, 18). Genetic variation in AR repeat length has also been evaluated in relationship with breast cancer risk (19, 20, 21, 22, 23, 24). Among BRCA1 mutation carriers, Rebbeck et al. (19) observed that women with longer AR alleles were diagnosed with breast cancer at a significantly younger age, suggesting that androgens may modulate breast cancer penetrance among women predisposed to early-onset breast cancer. However, subsequent studies among women from high-risk breast cancer families (20, 21) and most (23, 24) but not all (22) studies of sporadic breast cases do not support the association between AR [CAG]_n repeat length and breast cancer risk.

We conducted a prospective, case-control study nested within the Nurses’ Health Study cohort to evaluate AR [CAG]_n repeat allele length in relation to breast cancer risk among predominantly postmenopausal Caucasian women. We also examined whether AR repeat alleles modified cancer risk associated with established breast cancer risk factors and whether repeat length is associated with circulating levels of endogenous steroid hormones.

The Nurses’ Health Study was initiated in 1976, when 121,700 United States registered nurses between the ages of 30 and 55 returned an initial questionnaire reporting medical histories and baseline health-related exposures. Updated information has been obtained by questionnaire every 2 years, including data on reproductive variables, oral contraceptive and postmenopausal hormone use, cigarette smoking, and since 1980, dietary intake. Incident breast cancers were identified by self-report and confirmed by medical record review. Between 1989 and 1990, blood samples were collected from 32,826 women. Approximately 97% of the blood samples were returned within 26 h of blood draw, immediately centrifuged, aliquoted into plasma, RBCs, and buffy coat fractions, and stored in liquid nitrogen freezers. Follow-up has been >90% in all subsequent questionnaire cycles for this subcohort.

Eligible cases in this study consisted of women with pathologically confirmed incident breast cancer from the subcohort who gave a blood specimen. Cases with a diagnosis any time after blood collection up to June 1, 1996 with no previously diagnosed cancer except for nonmelanoma skin cancer were included. Controls were randomly selected participants who gave a blood sample and were free of diagnosed cancer (except nonmelanoma skin cancer) up to and including the interval in which the case was diagnosed. Controls were matched to cases on year of birth, menopausal status, postmenopausal hormone use, as well as time of day, month, and fasting status at blood draw. Women were defined as postmenopausal at the time of a bilateral oophorectomy or after having no menstrual cycle within the last 12 months prior to blood draw. Women who had a hysterectomy with one or both ovaries left intact were classified as premenopausal until the age at which 10% of the cohort had undergone natural menopause (46 years for smokers and 48 years for nonsmokers) and as postmenopausal at the age at which 90% of the cohort had undergone natural menopause (54 for smokers and 56 for nonsmokers); in the intervening years, these women were classified as being of uncertain menopausal status. For postmenopausal cases not using postmenopausal hormones within 3 months prior to blood draw, we matched a second control to increase statistical power in plasma steroid hormone analyses. The nested case-control study consists of 727 incident breast cancer cases and 969 matched controls. The study sample for the plasma hormone analysis is composed of 474 postmenopausal controls not using hormone replacement therapy within 3 months of blood draw. The protocol was approved by the Committee on Human Subjects at Brigham and Women’s Hospital.

Information regarding breast cancer risk factors was obtained from the 1976 baseline questionnaire, subsequent biennial questionnaires, and a questionnaire completed at the time of blood sampling. Menopausal status and use of postmenopausal hormones was assessed at blood draw and updated until date of diagnosis for cases and the equivalent date for matched controls. Histopathological characteristics such as tumor stage and size and estrogen and progesterone receptor status were ascertained from medical records and used in case subgroup analyses when available.

DNA was extracted from buffy coat fractions using the Qiagen QIAamp blood kit (Qiagen, Chatsworth CA). Oligonucleotide primers flanking the [CAG_n] repeat were constructed (5′-TCCAGAATCTGTTCCAGAGCGTGC-3′ and 5′-GCTGTGAAGGTTGCTGTTCCTCAT-3′). Genomic DNA was PCR amplified using fluorescently labeled primers. The length of these fragments varied by the number of CAG repeats. The fragments were run on a 5% denaturing polyacrylamide gel, and amplicon length was determined relative to a Gene Scan size standard (Applied Biosystems) by automated fluorescence detection (Genescan: Applied Biosystems, Foster City, CA) in the Dana-Farber Cancer Institute Molecular Biology Core Facility and the Harvard Center for Cancer Prevention Laboratory. Genotyping was performed by laboratory personnel blinded to case-control status, and blinded quality control samples were inserted to validate genotyping procedures; concordance for the blinded samples was 100%.

Steroid hormones estrone sulfate, estrone, estradiol, testosterone, androstenedione, DHEA, and DHEAS were assayed in four separate batches. Estrone sulfate from batches 1 and 2 were assayed in the laboratory of Dr. C. Longcope (University of Massachusetts Medical Center, Worcester, MA). All other analyses were performed by Nichols Institute (San Juan Capistrano, CA). The intra-batch coefficients of variation were ≤13.6%. With the exception of estrone sulfate, estrone, testosterone, and DHEA, there was little or no batch-to-batch variation. For the hormones with batch-to-batch variation, the means for identical quality control samples varied by 23–46% between batches. Hence, to account for this variation, we controlled for batch in all analyses. Methods for plasma hormone assays and information regarding laboratory precision and reproducibility have been published previously (6, 25).

The Wilcoxon signed-rank test was used to compare the distributions of the average repeat length {([CAG]_n of allele 1 + allele 2)/2} and the shorter and the longer of a woman’s two AR [CAG]_n alleles among cases and controls. A global test to evaluate differences in allele frequencies between cases and controls was performed using a Mantel Haenszel χ² test across matched case-control sets. This test was also used to evaluate case-control differences in the frequency of AR alleles using the cutpoint designated by Spurdle et al. (Ref. 23; [CAG]_n ≥22 versus <22), corresponding closely to the average [CAG]_n repeat length in our population.

The relationship between AR repeat length and phenotypic variation in waist:hip ratio, a physiological characteristic for which androgens may play a role, was evaluated by age-adjusted Spearman correlation. ORs and CIs were calculated using conditional and unconditional logistic regression. In conditional analyses, 9 controls were excluded due to incomplete matching. In the majority of analyses, we used the cutpoint [CAG]_n ≥22 repeats, which is the average repeat length among our controls. Indicator variables for all three genotypes (0 alleles ≥22, 1 allele ≥22, and 2 alleles ≥22) were created using 0 alleles ≥22 as the reference category in multivariate models. Genotype was evaluated as a dichotomous variable with one or two alleles ≥22 combined and as the linear effect/CAG repeat of the shorter, longer, and average repeat lengths. We also evaluated cutoff points throughout the [CAG]_n repeat (19–29) and average repeat length (19–27) distributions. In addition to the matching variables, we adjusted for the breast cancer risk factors BMI (kg/m²) at age 18 (continuous), weight gain since age 18 (<5 kg, 5–19.9, ≥20), age of menarche (<12 years, 12, 13, >13), parity/age at first birth (nulliparous, 1–2 children/age at first birth ≤24 years, 1–2 children/age at first birth >24, 3+ children/age at first birth ≤24, 3+ children/age at first birth >24), first-degree family history of breast cancer (yes/no), history of benign breast disease (yes/no), and duration of postmenopausal hormone use (never; past user <5 years, ≥5; current user, <5 years and ≥5). We also adjusted for age at menopause (continuous in years) in analyses limited to postmenopausal women. Fisher’s Exact Test was used in case-case analyses to evaluate associations between case characteristics [histological subtype, stage of disease, and ER (ER+/−) and progesterone receptor (+/−) status] and AR genotype. Unconditional logistic regression models, including terms for the matching variables and other potential confounders, were used to assess the association between AR repeat alleles and breast cancer risk when stratifying by established breast cancer risk factors. Interactions were evaluated by including interaction terms between genotype and risk factor variables in unconditional multivariate logistic regression models. The likelihood ratio test was used to assess the statistical significance of these interactions.

Linear regression models were used to evaluate the association between AR repeat length and geometric mean plasma hormone levels among controls, controlling for BMI at blood draw, laboratory batch, and the matching variables. Genotype was evaluated as the linear effect (percent change in hormone level) per average CAG repeat length increment. The natural logarithm of the plasma hormone values was used in the analyses to reduce the skewness of the regression residuals. Hormones were measured in four different batches, and all hormones were not assayed for all women due to insufficient plasma. Subjects with hormone levels below detectable limits (n = 6) were assigned the lowest detectable value (estrone, n = 3; testosterone, n = 1; DHEAS, n = 2). Within each batch, hormone values greater than three times the interquartile range above the 75th percentile were treated as outliers and excluded (estrone sulfate, n = 11; estrone, n = 3; estradiol, n = 8; testosterone, n = 2; androstenedione, n = 1; DHEA, n = 3; DHEAS, n = 1). We used the Statistical Analysis System (SAS) for all analyses (26).

There were 304 premenopausal and 1236 postmenopausal women with mean ages of 48.5 (SD, 3.1) and 61.0 (SD, 5.0), respectively, at blood draw. Compared with controls, cases had the same mean BMI at blood draw (25.5 versus 25.6 kg/m²) and similar ages at menarche (12.5 versus 12.6 years), first birth (25.0 versus 24.9 years) and menopause (48.0 versus 48.2 years). The proportion of women with a first-degree family history of breast cancer was significantly higher among the cases (21% versus 15%; P = 0.001). The distribution of AR alleles for cases and controls was similar (Fig. 1) as was the average repeat lengths (cases, 22.0; 95% CI, 21.8–22.1; controls, 21.8; 95% CI, 21.7–22.0; P = 0.15). The mean repeat lengths of the shorter and longer alleles were nearly identical between cases and controls (shorter allele: cases, 20.4; 95% CI, 20.2–20.5; controls, 20.3; 95% CI, 20.1–20.4; P = 0.35; longer allele: cases, 23.6; 95% CI, 23.4–23.8; controls, 23.4; 95% CI, 23.3–23.6; P = 0.18). We observed no significant difference in allele frequencies between cases and controls dichotomized by [CAG]_n ≥22 (χ² = 0.007; df = 1; P = 0.93), and the expected genotype distributions based on the allele frequencies among controls were similar to the observed distributions.

The adjusted OR for the effect per average [CAG]_n repeat increment was 1.04 (95% CI, 0.99–1.10). We observed similar results for the shorter (OR, 1.04; 95% CI, 0.99–1.08) and longer (OR, 1.02; 95% CI, 0.98–1.07) repeat alleles. Compared with women with both alleles <22, those with one or two alleles ≥22 did not have a substantially increased risk of breast cancer (Table 1). No material change in relative risk estimates was evident when cutpoints above 22 were selected (women with 1 or 2 alleles ≥23, 25, 27, and 29 compared with those with two short alleles, both alleles <22; Table 1). Among premenopausal and postmenopausal women, no significant association was observed between having one or more [CAG]_n repeat alleles ≥22 and breast cancer risk (OR, 1.38; 95% CI, 0.68–2.82 and OR, 1.01; 95% CI, 0.78–1.31, respectively) or when alternate allele cutpoints were evaluated (Table 1). Risk was not significantly greater for women homozygous for long alleles (versus both alleles <22; both alleles ≥25: OR, 1.42; 95% CI, 0.81–2.50). Similarly, an association with breast cancer risk was not evident among women with two short alleles compared with those with both alleles ≥22 (both alleles ≤20: OR, 0.92; 95% CI, 0.62–1.36; ≤19: OR, 1.06; 95% CI, 0.61–1.86). We also did not observe a consistent elevation in breast cancer risk for cutpoints along the average AR allele length distribution (Table 1). Using the same cutpoints, no associations were observed for shorter or longer repeat alleles (data not shown).

In analyses by breast cancer histological subtype and clinical parameters, carriers of one or more [CAG]_n alleles ≥22 were not significantly overrepresented among invasive cancers (OR, 1.10; 95% CI, 0.86–1.40). We observed no difference in the frequency of longer alleles, shorter alleles, or the number of women with an average repeat length ≥22 by estrogen or progesterone receptor status, number of involved lymph nodes, tumor size, or degree of differentiation (all Ps ≥ 0.3). We also did not observe a significant correlation between AR repeat length and waist:hip ratio (r = 0.05; P = 0.17), a physiological factor associated with higher circulating androgen levels.

We next evaluated the relationships between AR alleles and the risk of breast cancer according to established breast cancer risk factors. We observed no significant interactions between longer AR alleles ([CAG]_n ≥22) and BMI among postmenopausal women (P = 0.92, test for interaction), history of benign breast disease (P = 0.60, test for interaction), age at menarche (P = 0.71, test for interaction), age at first birth (P = 0.96, test for interaction), parity (P = 0.39, test for interaction), oral contraceptive use (P = 0.19, test for interaction), or hormone replacement therapy (P = 0.10, test for interaction).

Because the data of Rebbeck et al. (19) suggest that BRCA1-associated breast cancer risk may be modified by AR [CAG]_n repeat length, we examined the potential interaction between AR genotype and first-degree family history of breast cancer. The test for interaction between first-degree family history of breast cancer and AR genotype (cutpoint, 22 repeats) was statistically significant (P = 0.04; Table 2). In stratified analyses, among women without a first-degree family history of breast cancer, the OR for carriers of 1 or 2 alleles ≥22 (versus noncarriers) was 0.92 (95% CI, 0.71–1.17). A nonsignificant positive association was observed among women with a first-degree family history of breast cancer (1 or 2 alleles ≥22: OR, 1.64; 95% CI, 0.93–2.89). When evaluating longer repeat cutpoints (23, 25, and 27), positive associations were limited to the family history-positive strata. We also examined the cross-classification of AR genotype and first-degree family history of breast cancer (Table 2). Compared with women with both alleles <22 and no family history, we observed a significant positive association limited to women with both a first-degree family history of breast cancer and longer alleles (Table 2). Point estimates did not change much when adjusting for all established risk factors in multivariate models.

We also evaluated the relationship of AR repeat length with circulating steroid hormone levels among postmenopausal controls not currently using postmenopausal hormones (Table 3). The AR gene is located on the X chromosome, and because one X chromosome is randomly inactivated in women, we analyzed the average of both AR repeat alleles to evaluate the systemic effects of AR genotype on endogenous steroid hormone levels. With increasing repeat number, we observed a nonsignificant inverse linear association with plasma testosterone levels and a nonsignificant-positive association with DHEAS levels. Relationships between the hormones androstenedione, DHEA, estradiol, estrone, and estrone sulfate and average repeat length were all nonsignificant.

If androgens act directly on breast epithelium via the AR to inhibit tumor growth, then women with decreased AR transactivation (carriers of longer repeat alleles) may have an elevated risk of breast cancer. Conversely, through a physiological feedback mechanism, women with less functional receptors may have elevated androgen concentrations and be at increased risk of breast cancer as a result of increased estrogen levels and an increase in the fraction of bioavailable estradiol unbound to sex hormone-binding globulin due to a greater affinity of sex hormone-binding globulin for testosterone (27). Our data do not directly support either of these hypotheses, as we observed no association between [CAG]_n repeat length in exon 1 of the AR gene and risk of breast cancer, nor did we observe AR allele length to be significantly associated with endogenous steroid hormone levels.

The association between AR [CAG]_n alleles and breast cancer risk has been evaluated in three case-control studies (22, 23, 24). In a study conducted among French-Canadians (cases, n = 255; controls, n = 461), Giguére et al. (22) observed a significant inverse association between AR repeat allele length and breast cancer risk (2 alleles ≤20: OR, 0.50; 95% CI, 0.27–0.82). A null association was reported between AR genotype and breast cancer risk among 368 cases and 284 age-matched controls <40 from an Australian population (1 allele ≥22: OR, 1.40; 95% CI, 0.92–2.15; 2 alleles ≥22: OR, 1.40; 95% CI, 0.87–2.26; Ref. 23). A British study consisting of 508 cases and 426 controls >50 years of age (24) also provided no evidence of an association between AR repeat length and breast cancer risk (versus alleles <22; 1 allele >23: OR, 0.82; 95% CI, 0.62–1.09; 2 alleles >23; OR, 1.31; 95% CI, 0.87–1.97). Unlike the results of Giguére et al. (22), our data from a prospective study among mainly Caucasian women predominantly >50 years of age support these findings. Small sample size and the analysis of multiple potential AR CAG cutpoints could potentially explain the inconsistent findings between studies. In a recent case series of 133 Italian women with primary breast cancer, shorter AR [CAG]_n repeat alleles were observed to be overrepresented among more aggressive forms of breast cancer (28). Our data also do not support these findings, as we observed no evidence of a relationship among cases between AR repeat length and breast cancer stage or grade.

In a study among 304 BRCA1 mutation carriers, Rebbeck et al. (19) reported that women with at least one long AR allele (≥28) were diagnosed with breast cancer at a significantly younger age. Recent studies among BRCA1/2 mutation carriers and women from high-risk breast/ovary cancer families have not confirmed these findings (20, 21). We did not observe a strong association between AR genotype and breast cancer risk; however, phenotypic and etiological differences between sporadic and BRCA1-associated breast cancers may be attributable to physiological variation in the effects of hormonal factors, such as AR genotype, on breast cancer development. We examined the hypothesis that having a family history of breast cancer may act as a modifier of breast cancer risk associated with AR genotype. Longer AR repeat alleles were overrepresented among cases with a first-degree family history of breast cancer; women in this dual category had a significantly elevated risk of breast cancer (OR, 1.70; 95% CI, 1.20–2.40). Although few, if any, of the women in this study are BRCA1 mutation carriers, our data are consistent with the observations of Rebbeck et al. (19) suggesting that AR genotype may influence family history-associated breast cancer risk.

In our study, AR repeat length was not a strong predictor of plasma steroid hormone levels. In a prospective study among males 40–70 years of age (n = 882; Ref. 29), the age-related decrease in serum testosterone was significantly attenuated among men with longer AR alleles (% change in free testosterone: +2.8% per three CAG repeats; 95% CI, 1.0–4.7). More recently, in a small cross-sectional study among premenopausal women (n = 239), Westberg et al. (30) observed a significant association between short AR repeat alleles and higher follicular phase testosterone levels. Our data do not support an association in postmenopausal women. In addition, the modest inverse association between testosterone levels and increasing [CAG]_n repeats that we observed, as well as the findings of Westberg et al. (30), do not support the hypothesis in women that longer, less functional AR alleles have increased plasma androgen levels. In addition, we did not observe even modest associations between AR repeat length and other hormones.

In summary, we did not observe a direct association between AR repeat length and breast cancer risk or evidence of a relationship between AR genotype and endogenous testosterone levels. However, our data are consistent with the hypothesis that longer AR repeat alleles may be involved in modifying family history-associated breast cancer risk.

We thank Jeanne Sparrow, Lisa Li, and Carsta Cieluch for technical and laboratory assistance. We are also indebted to the participants in the Nurses’ Health Study for their dedication and commitment.