The role of androgens and polymorphisms in the androgen receptor in the epidemiology of breast cancer.

Testosterone binds to the androgen receptor in target tissue to mediate its effects. Variations in testosterone levels and androgen receptor activity may play a role in the etiology of breast cancer. Here, we review the epidemiologic evidence linking endogenous testosterone to breast cancer risk. Paradoxically, results from observational studies that have examined polymorphisms in the androgen receptor suggest that the low-activity androgen receptor increases breast cancer risk. We review the quality of this evidence and conclude with a discussion of how the androgen receptor and testosterone results coincide.


Introduction
The role of androgens in breast cancer etiology has been a subject of both curiosity and confusion. It is still unclear by which mechanisms testosterone exerts its activity in the female breast, and whether the effects are predominantly proliferative or anti-proliferative on breast cells at physiologic levels. In the present review we evaluate the results from epidemiologic studies on the role of circulating testosterone and a functional polymorphism in the androgen receptor (AR) in breast cancer. We also highlight some of the epidemiologic challenges in addressing these questions.

Sources of endogenous testosterone
There are two main sources of androgens in women. Testosterone is produced directly by the ovary and by conversion of the adrenal androgens dehydroepiandrosterone and dehydroepiandrosterone-sulfate into androstenedione, and then further to testosterone in peripheral tissue [1]. In premenopausal women, approximately 25% of circulating testosterone is secreted directly from the adrenal gland and 25% from the ovary, whereas the remaining 50% is produced by peripheral conversion of androstenedione [2]. Testosterone levels vary over the menstrual cycle with peak levels mid-cycle, and diurnally with highest levels in the early morning [3].
Testosterone and androstenedione are produced by the interstitial cells of the ovarian stroma and may continue to respond to gonadotopins and produce testosterone after the menopause [4]. In normal postmenopausal women the ovarian vein has been observed to have higher concentrations of testosterone than is found in peripheral blood; bilateral oophorectomy results in reductions in testosterone levels by as much as 50% [5].
Several smaller cross-sectional studies have found lower testosterone levels in postmenopausal than premenopausal women [6][7][8] or lower levels in perimenopausal than premenopausal women [2]. Large longitudinal studies that have followed women through the menopausal transition have observed either no significant change in testosterone [2,9] or a 15% decrease in both testosterone and androstenedione at menopause [10]. In one study of women aged 50-89 years testosterone levels were lowest at the time of the menopause, whereas women older than 70 years or more than 20 years postmenopause had levels approximating those of premenopausal women [11].

The role of androgens and polymorphisms in the androgen receptor in the epidemiology of breast cancer
In summary, there is increasing evidence that the ovary continues to produce androstenedione and testosterone in healthy postmenopausal women. Levels may either remain the same or decrease slightly at menopause. However, women with bilateral oophorectomy may be androgen deficient.

Testosterone and breast cancer risk
Prospective studies that examined the association between testosterone and breast cancer Data from eight prospective cohort studies have been reported on the association between endogenous testosterone levels and breast cancer risk using testosterone measured from blood samples gathered at baseline from postmenopausal women [12][13][14][15][16][17][18][19]. Six of these studies were nested case-control studies [12,[14][15][16][17][18]; one was a case-cohort study [19] and one a full cohort study [13]. Only one of these studies reported results for premenopausal women [12].
Six of the eight studies reported a statistically significant increase in postmenopausal breast cancer risk with increasing levels of endogenous testosterone [14][15][16][17][18][19]. A recently conducted pooled analysis [20] of these eight prospective studies estimated that the relative risk for breast cancer in women whose levels of testosterone were in the top quintile as compared with women in the bottom quintile was 2.22 (95% confidence interval 1.59-3.10). A statistically significant dose-response relationship was also observed (P for trend < 0.001) [20]. Two of the studies also reported statistically significantly increasing breast cancer risk with increasing levels of free testosterone [15,19], a measure of bioavailable testosterone.
The study of premenopausal women [12] found no statistically significant differences between cases and noncases in mean levels in either premenopausal or postmenopausal women, but the sample size was small (premenopausal women: 17 cases, 67 controls; postmenopausal women: 22 cases, 88 controls).

Can the observed association between testosterone levels and breast cancer risk be due to bias? Effects of measurement biases
One limitation of the studies reviewed is that serum testosterone may not be the ideal measure of testosterone. Total testosterone includes both free testosterone and bound testosterone. Furthermore, serum levels do not take into account the peripheral conversion of precursor androgens into testosterone in the breast tissue itself. The effect of this measurement error is most likely to be nondifferential, therefore biasing results toward the null.
Measurement biases due to use of a single hormone measurement or degradation of hormones in stored specimens over time would most likely be nondifferential, resulting in attenuated estimates of disease risk. All of the existing prospective studies (Table 1) analyzed testosterone measured from only one blood draw, which might not be representative of the cumulative exposure to testosterone; however, given the prospective design of the studies, any inaccuracy in measurement would probably bias risk estimates toward the null.
The consideration of time of day that blood was drawn and fasting status can help to avoid the biases due to using a single hormone measurement. To avoid this bias, three of the studies either matched with respect to time of blood draw [14,18] or restricted individuals to having their blood drawn in the morning [13]. Because the effect of these sporadic variations would be to bias the results toward the null, this may help to explain one of the null associations observed [12].
The studies that observed an association between testosterone and breast cancer risk attempted to reduce measurement bias due to degradation by matching cases to controls on the date of blood draw [14][15][16][17][18] and storage conditions such as sample location/shelf in the freezer [15]. This was not done in the studies reporting no association between testosterone and breast cancer risk [12,13] or in one of the positive studies [19].
Laboratory assay variation would also most likely be nondifferential because cases and controls were analyzed concurrently in these studies. The intra-assay and interassay coefficients of variation in these studies were rather good, ranging from 4% to 14%. However, the coefficient of variation was not reported in one of the null studies [12].
Thus, although there may be some attenuation in effect estimates in all of the studies, it is not clear whether measurement bias due to degradation can explain the discrepancies between the two null studies and the positive studies.

Temporal Bias
If breast cancer development increases testosterone levels, then studies that included individuals diagnosed shortly after baseline hormone measurement may have artificially elevated estimates of the risk for breast cancer associated with testosterone levels. Two of the positive studies and both of the null studies excluded those who were diagnosed 6-24 months after baseline [12][13][14]17]. However, the study with the most conservative cut-point of 24 months [14] reported a significant positive association between testosterone levels and breast cancer risk. Thus, although it is possible that temporal bias played a role in the four positive studies with no exclusions, this latter study suggests that temporal bias cannot explain the association between testosterone and breast cancer risk.

Effects of confounding
Lack of control for body mass index (BMI) or age at menopause could result in a positive bias away from the null. All of the studies that reported a significant association between testosterone levels and breast cancer risk included either BMI or height and weight as covariates in the statistical model. All but one of the null studies [13] considered either the amount of time menopausal [12,14,16] or age at menopause [15,[17][18][19] as a covariate to control for the effects of menopause on testosterone levels. Thus, it is unlikely that confounding by these variables can explain the associations observed between testosterone levels and breast cancer risk.

Comparability/generalizability
All of the prospective studies that examined the testosterone-breast cancer association were conducted using cohorts from Caucasian populations. Only two studies [15,16] were conducted outside the USA, one in Italy [15] and one on the island of Guernsey [16]. It is therefore unlikely that differences in the populations studied can explain the discrepant results between studies. There are, as far as we know, no prospective data from nonwhite populations.

Studies of testosterone measured after diagnosis that examined the testosterone-breast cancer association
Several case-control studies published during the past 20 years have evaluated the association between testosterone levels and breast cancer risk. Comparisons with levels in control individuals have shown that breast cancer patients who were postmenopausal [21][22][23][24][25] and those who were premenopausal [26][27][28] had significantly elevated testosterone levels. Table 2 presents the odds ratios of breast cancer associated with categories of serum testosterone, whereas Table 3 presents results of studies that compared mean levels of testosterone. The measures of testosterone in cases were almost twice those of controls. Although these retrospective studies support the prospective study results showing an association between increased testosterone and increased breast cancer risk, these results may not be as readily interpretable as those from the prospective studies, given the possibility that the presence of cancer, or the treatment for it, might have increased testosterone levels.

Summary
Both retrospective and prospective studies have reported statistically significant associations between increased levels of testosterone and increased breast cancer risk. These associations are unlikely to be due to measurement biases, the influence of disease, or lack of adjustment for the confounding effects of BMI or age at menopause. Age adjusted. 4 Analysis also conducted excluding past estrogen users. 5 Adjusted for BMI and age only. 6 Adjustment did not change results from the crude analysis. BMI, body mass index; CI, confidence interval; NS, not significant; NR, not reported; RR, relative risk; SHBG, sex hormone binding globulin; T, testostrone.

Androgen receptor, the AR-CAG repeat, and breast cancer risk
The main receptor for testosterone is the AR. A functional polymorphism in the AR gene has been examined in female breast cancer, and the literature is reviewed to shed light on the possible mechanisms by which testosterone may affect breast cancer risk.

Androgen receptor protein and breast cancer
The AR is expressed in the majority of breast cancers [29][30][31][32][33][34][35]. Several studies have been conducted to examine the effects of androgens on the growth of AR-positive breast cancer cell lines. These studies have reported both inhibitory [36,37] and stimulatory [38,39] effects. These divergent effects have been observed to be specific to the cell line under study [40].
To our knowledge, the only in vivo study of the effect of testosterone on breast cell proliferation was conducted in rats and showed that treatment with testosterone results in both tumor regression and a reduction in estrogen receptor expression [41]. However, it is unclear whether the testosterone levels used represent physiologic doses. No in vivo or epidemiologic studies have examined the association between serum or tissue testosterone levels and breast cell proliferation in tumors with varying degrees of AR expression.
In summary, the effects of androgens on breast cancer cell growth are still unclear. In contrast to the epidemiologic observation of a consistent association between serum testosterone levels and increasing breast cancer risk, in vivo studies reported an antiproliferative effect and in vitro studies reported both proliferative and antiproliferative effects.

The androgen receptor gene and a polymorphic CAG repeat
The AR is encoded by a single 90 kilobase gene on the X chromosome (Xq11-q12), which encodes a 11-kilobase mRNA transcript composed of eight exons [42][43][44][45][46]. Epidemiologic evidence for a role of the AR gene in breast cancer was first suggested by studies of male breast cancer patients. A mutation in AR in the DNA-binding domain resulting in an inability to bind androgens was first reported in a pair of brothers with breast cancer [47]. In a study of 13 male breast cancer patients, one was observed to carry a similar mutation [48]. In another small study of 11 male breast cancer patients [49], this mutation was not observed. These results suggested that the mutation may play a role in the development of breast cancer in some males.
Within the first exon of AR lies a polymorphic CAG repeat that encodes a polyglutamine tract of variable length. The normal size range of these repeats is between 6 and 39 repeats [50,51]. Between 40 and 66 repeats have been observed in patients with a rare, neurodegenerative disorder called spinal and bulbar muscular atrophy [52], which is characterized by androgen insensitivity with gynecomastia, testicular atrophy, oligospermia, azoospermia, and elevated serum gonadotropins.

AR-CAG repeat length and breast cancer
The association between the length of the AR-CAG repeat polymorphism and breast cancer risk has been examined in several case-control studies (Table 4) [63][64][65][66][67][68]. The long AR-CAG repeat, which is representative of the less active AR, was associated with a statistically significant increase in breast cancer risk in a population of women from Quebec [66] and in a population of BRCA1 mutation carriers [63]. Four additional studies [64,65,67,69] reported slightly increased risk for breast cancer associated with the long allele, but none of these findings were statistically significant. A study nested within the Nurses' Health Study cohort [68] found no increased breast cancer risk associated with the long AR allele overall, but an increased risk was observed when analyses were limited to those individuals with a firstdegree family history of breast cancer (odds ratio 1.70, 95% confidence interval 1.2-2.4). Another trinucleotide repeat in the AR, a GGC repeat, has been observed to be associated with prostate cancer risk [59,60,62,70,71].
One of three studies that examined the GGC repeat length and breast cancer [65,67,69] found a significant association in women diagnosed before age 45 years [69], but no evidence of an interaction between the CAG and GGC repeat with breast cancer risk was observed.
The three studies that reported a significant association between long AR-CAG repeat and breast cancer risk [63,66,68] included both premenopausal and postmenopausal women. One study stratified with respect to menopausal status and found that the significant association with the long AR-CAG repeat was observed only in postmenopausal women (odds ratio 3.22, 95% confidence interval 1.54-6.75) and not in premenopausal women (odds ratio 1.03, 95% confidence interval 0. 43-2.48). If this effect modification is true then it may explain, at least in part, the nonsignificant results in the studies restricted to women aged under 40 years [64].

Issues with the studies of AR-CAG repeat length and breast cancer risk
The gene for the AR lies on the X chromosome, and therefore women carry two alleles whereas men carry only a S, short allele; L, long allele. 2 Analyses using 29 and 30 repeats as the cut-point produced progressively higher significant risk estimates and progressively earlier age of onset, no trend test. 3 Published odds ratio models the SS genotype as the high-risk allele. BMI, body mass index; CI, confidence interval; RR, relative risk.
single allele. In general, normal women are a mosaic, with one allele randomly expressed in each cell. A recent study [72] reported that 13% of young (27-45 years old) breast cancer cases exhibited preferential activation of one of the AR alleles as measured by genotyping of peripheral blood DNA, but there was no preference toward the allele with the longer or shorter CAG repeat. Analyses of the AR in women that only consider the length of the CAG repeat on one allele assume that this is the active allele in the breast tissue. Analyses that use the average of the CAG repeat lengths or the sum of the repeats consider the contributions of both alleles; however, if only one allele is preferentially expressed then this would result in misclassification. There is a high rate of heterozygosity in the AR-CAG repeat length, and therefore this is likely to be a major misclassification problem, which should bias the results toward the null. Genotyping methods can be optimized to detect better whether there is a preferentially active AR allele by either genotyping tumor tissue or serum DNA using methylation sensitive enzymes [72].

Summary
The studies conducted thus far suggest that the long AR-CAG repeat (less active AR) may be associated with increased breast cancer risk in women who are postmenopausal, have a first-degree family history of breast cancer or who have a known BRCA1 mutation. The location of the AR gene on the X chromosome means that results from epidemiologic studies will be biased toward the null as long as we do not know which allele is expressed.

Discussion
If the long AR-CAG repeat (less active AR) is associated with increased breast cancer risk in postmenopausal women, then how do these results coincide with results showing that increased testosterone levels increase postmenopausal breast cancer risk?
One hypothesis to explain this apparent paradox is that the less active AR may be involved in a physiologic feedback associated with increased circulating testosterone. However, the only data available discount this hypothesis. Two studies have examined the association between the AR-CAG repeat length and circulating testosterone levels in normal women [68,73]. AR-CAG repeat length was inversely associated with testosterone levels. In other words, the less active AR was associated with lower circulating testosterone levels, and the results were statistically significant both in a study of premenopausal women [73] and in a study of postmenopausal women [68].
If the AR is not involved in a feedback mechanism to influence testosterone levels in postmenopausal women, then it is possible that the effect of testosterone on the breast epithelium does not act through binding to the AR. Testosterone may exert its effect on breast tissue through con-version of testosterone to estrone, which is then aromatized into estradiol in adipose tissue, and the increased estradiol levels may result in increased breast cell proliferation and breast cancer risk.
Testosterone may also exert an indirect effect on breast cancer proliferation by sequestering sex hormone binding globulin, leaving more estradiol in the non-protein-bound state and able to act on breast tissue [25,74]. Approximately 66% of total testosterone is bound to sex hormone binding globulin, 31% is bound to albumin, and 2% is bound to cortisol binding protein [75]. Two of the studies suggesting an association between testosterone and breast cancer [16,17] reported that this association disappeared when adjusting for estradiol levels. However, in the pooled analysis [20] the significant association between testosterone and breast cancer risk remained after adjustment for estradiol [20].
Finally, it is possible that further studies will show that AR-CAG repeat length is not linked to breast cancer risk.

Conclusion
Prospectively conducted epidemiologic studies have found that increased levels of serum testosterone are associated with an increase in postmenopausal breast cancer risk. However, a number of questions remain. Several lines of evidence suggest a role of AR in breast cancer risk, and sparse epidemiologic data suggest that a long AR-CAG repeat yielding a less active AR may be associated with increased risk. There still remain a number of questions on how testosterone increases breast cancer risk. Although in vitro studies report both proliferative and antiproliferative effects of testosterone on the growth of various breast cancer cell lines, we still need to further understand under which in vivo circumstances does testosterone exert these effects. Finally, we do not know whether androgens affect breast cancer risk in premenopausal women. Further analyses of the role of AR-CAG repeat length and breast cancer using genotyping methods that assess which allele is the active AR allele are clearly needed. Additional data are also needed to help elucidate the apparent paradox between the AR-CAG repeat length, testosterone levels, and breast cancer risk.

Competing interests
None declared.