In this prospective study of premenopausal plasma sex steroids and SHBG and breast cancer risk, we found positive associations between androgens and risk of both invasive and ER+/PR+ breast cancer, as well as luteal estradiol and risk of ER+/PR+ tumors. Several of the associations varied by menopausal status at breast cancer diagnosis. Total and free estradiol levels in the follicular phase were positively, but not significantly, associated with premenopausal invasive and ER+/PR+ disease but significantly inversely associated with postmenopausal disease. Testosterone levels also appeared more strongly associated with postmenopausal breast cancer, although these differences were not statistically different. We observed no significant associations when evaluating all cancers combined.
The positive associations between endogenous estrogens and androgens and postmenopausal breast cancer risk are well established in prior epidemiologic studies [4
], and biologic mechanisms are well described for estrogens [1
]. Estrogens are associated with increased proliferation and decreased apoptosis, and may promote proliferation of cells with genetic mutations [1
]. The biologic mechanism between androgens and breast cancer is less established, with androgens demonstrating both growth inhibitory [2
] and proliferative [3
] effects in breast cancer cell lines. In animal models, androgens have been shown to inhibit proliferation [35
] and the androgen receptor antagonist flutamide has been shown to increase epithelial cell proliferation [35
]. Breast tissue can convert androgens to estrogens via aromatase [38
] and, therefore, androgens, through their conversion to estrogen, also may be indirectly associated with breast cancer risk.
There is an inherent complexity in measuring estrogens in premenopausal women given the variation in estrogen levels across the menstrual cycle, and prior studies have accounted for this in different ways. The eight prior prospective studies [5
] either did not account for menstrual cycle day [11
], matched cases and controls on cycle day [5
], matched cases and controls on cycle day and used a spline regression model to adjust for cycle differences [9
], or, as in our prior analysis, restricted sample collection to specific windows in the menstrual cycle and matched on cycle day [13
]. Six of the eight prior studies [5
] had fewer than 100 cases, with four of the six [5
] finding non-significant or suggestive positive associations. Kaaks et al.
], in the largest prior study (case n
= 285) in the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort, found no association with invasive breast cancer risk for either estrone (4th
quartile OR: 1.2, 95%: 0.7 to 1.9, Ptrend
= 0.46) or estradiol (OR: 1.0, 95%: 0.7 to 1.5, Ptrend
= 0.89). A prior analysis in a subset of the population included in the current analysis [13
], in women predominantly premenopausal at diagnosis (n
= 197 cases; n
= 192 premenopausal at diagnosis), found a significant positive association between follicular estradiol with breast cancer overall (4th
quartile OR: 2.1, 95% CI: 1.1 to 4.1, Ptrend
= 0.08), that appeared somewhat stronger for invasive and ER+/PR+ disease. Results in the present study, with the addition of 317 timed cases, are attenuated relative to our prior findings, though we continue to see the suggestion of a weak positive association for follicular estradiol and invasive breast cancer diagnosed in premenopausal women. In our prior analysis, we saw no clear association between luteal estradiol and breast cancer; results in the current analysis are suggestive of an association between luteal estradiol and invasive and ER+/PR+ tumors.
The association between circulating premenopausal androgens and breast cancer risk has been investigated in seven prospective studies [5
], in addition to the NHSII [13
]. In the EPIC cohort (n
= 370 cases), significant positive associations with breast cancer risk were observed for both testosterone (4th
quartile OR: 1.7, 95% CI: 1.2 to 2.6, Ptrend
= 0.01) and DHEAS (OR = 1.5, 95% CI: 1.0 to 2.1, Ptrend
= 0.10) [12
]. Recent results from the New York University Women's Health Study (NYUWHS) (n
= 356 cases) [15
], are in agreement, where an approximately two-fold increase in risk was observed in the highest quintile of both testosterone (OR: 2.2, 95% CI: 1.3 to 3.5, Ptrend
= 0.03) and free testosterone (OR: 1.9, 95% CI: 1.2 to 2.9, Ptrend
= 0.01), though a weaker, non-significant positive association was observed with DHEAS (OR: 1.3, 95% CI: 0.8 to 2.1, Ptrend
= 0.20). In our initial report in the NHSII, positive associations were seen for luteal testosterone (OR: 1.6, 95% CI: 0.9 to 2.8, Ptrend
= 0.10), free testosterone (OR: 1.4, 95% CI: 0.8 to 2.5, Ptrend
= 0.14) [13
], and DHEAS (OR: 1.3, 95% CI: 0.9 to 2.1, Ptrend
= 0.08) [14
]. In the current much larger study, our findings are somewhat weaker, although a number of associations remained statistically significant. Thus, overall, data from prospective studies are quite consistent in showing that premenopausal androgen levels predict later breast cancer risk.
In this updated analysis, with six more years of follow-up, we observed an attenuation of our previously published results, particularly for follicular estradiol. Cases included in the previous analysis were similar to the new cases with regards to case characteristics (for example, invasive vs. in situ
, hormone receptor status, grade, stage), and risk factor status (for example, family history of breast cancer). Data from our prior reproducibility study [20
] suggest androgens and SHBG are quite stable over at least three years (ICC range: 0.73 (testosterone) to 0.89 (SHBG)) and do not change substantially at menopause while ICCs for the estrogens are lower (ICC range: 0.33 (estrone) to 0.45 (estradiol)), and levels change markedly around menopause. It is possible, particularly for the estrogens, that steroid hormones measured closer to diagnosis are a stronger predictor of disease, or are most relevant to risk as promoters of early stage tumorigenesis.
There are little prior data to suggest a differential effect of premenopausal sex steroids in relation to pre- vs. postmenopausal disease. In the current analysis, follicular estradiol was associated with a suggestively increased risk of premenopausal breast cancer and significantly inverse risk of postmenopausal disease. Given this novel finding, and the considerably different hormonal milieu in pre- vs. postmenopausal women, further investigation is warranted. Although speculative, because estradiol levels are substantially lower after menopause, women with higher premenopausal follicular estradiol may experience the greatest change in estradiol levels from pre- to post menopause, which may in turn confer a lower risk, at least over the short term. There is some indirect evidence to support a role for changes in endogenous estrogen levels in breast cancer etiology. For example, discontinuation of exogenous postmenopausal hormones (PMH) is associated with reduced proliferation in ER+ tumors [39
] and postmenopausal weight loss, which is associated with a decrease in estrogen levels, is associated with lower breast cancer risk [40
Our androgen/breast cancer associations appeared stronger in women postmenopausal at diagnosis, although the differences by menopausal status were not statistically significant. This has been evaluated in few other cohorts. In the NYUWHS, the association between testosterone and breast cancer risk was suggestively stronger in women postmenopausal at diagnosis (for doubling of testosterone: premenopausal, OR: 1.4, 95% CI: 0.8 to 2.3, Ptrend
= 0.23; postmenopausal, OR: 1.7, 95% CI: 1.1 to 2.6, Ptrend
= 0.02), although this difference also was not statistically significant (P
for interaction > 0.15) [15
]. Similarly, in the Columbia, Missouri cohort, results for total testosterone and bioavailable testosterone were stronger among women ages 55 and older at diagnosis (for example, 4th
quartile, overall results for testosterone, OR: 3.3, 95% CI: 1.5 to 7.5, Ptrend
= 0.006; among women ≥ 55 at diagnosis, OR: 4.5, 95% CI: 1.6 to 13.0, Ptrend
= 0.009) [5
]. However, in contrast to these findings, in the EPIC study associations were somewhat stronger for both testosterone and DHEAS in women ages ≤ 49 years at diagnosis [12
]. Clearly, additional studies are necessary to address this potential heterogeneity. The influence of androgens on breast tissue in a high estrogen environment may differ from those in a low estrogen environment though this hypothesis has not been adequately explored. Androgen levels do not change substantially at menopause [42
], and prior prospective studies consistently link postmenopausal androgens to breast cancer risk [4
]. Therefore, it is possible that premenopausal androgen levels are a marker for postmenopausal levels, which are then determinants of postmenopausal breast cancer risk.
Among prior prospective assessments of premenopausal progesterone levels and breast cancer risk, three [6
] of four small studies (case n < 100) found non-significant inverse associations, and the larger EPIC cohort reported a significant inverse association [12
]. In our prior analyses in the NHSII, we observed no association between progesterone and breast cancer risk [13
], and this was confirmed in the current expanded analysis. The ICC over three years for progesterone is 0.29, evidence that it is not well measured with one sample. Prior prospective studies [5
] did not find a statistically significant association between premenopausal SHBG and breast cancer risk, in agreement with our findings.
Our study has both strengths and limitations. This is the largest prospective study to date, although we had limited power in analyses stratified by menopausal status at diagnosis (particularly postmenopausal cases). Our samples were carefully timed in the menstrual cycle. We used highly specific and sensitive assays, and laboratory CVs were excellent. Additionally, we are among the first to present data for premenopausal hormones and breast cancer by tumor hormone receptor status. Although we are limited to samples collected during one menstrual cycle, a prior reproducibility study provides evidence of reasonable stability across a three-year period (follicular estradiol: 0.38 to DHEAS: 0.86), except for luteal progesterone (ICC = 0.29) [20
]. We also used these reproducibility data to correct for measurement error and showed that several of the associations may be substantially stronger than observed.