Sex steroid hormones and risk of breast cancer: a two-sample Mendelian randomization study

Breast cancer (BC) is the most common cancer in women worldwide and is the leading cause of cancer mortality in females [1]. Early menarche and a later age at menopause have been shown to be associated with an increased risk of breast cancer [2]. Furthermore, a study conducted on postmenopausal women showed that a higher number of lifetime cumulative menstrual cycles increased BC risk [3]. Taken together, susceptibility to BC appears to be associated with ovarian hormones related to the menstrual cycle, although the biological basis for this is still not understood [4].

The association of oral contraceptive use and hormone replacement therapy (HRT) with BC risk provides further evidence for the role of ovarian hormones in BC. A systematic review that included 44 BC studies showed that oral contraceptive use increased the risk of BC [5]. A large-scale meta-analysis combining case–control data from 58 studies found that HRT use was associated with an increased risk of BC within 4 years of current use, with the increasing risk associated with a longer duration of current use [6].

Analyses looking specifically at blood levels of nine sex steroid hormones and BC risk concur with evidence surrounding factors associated with the menstrual cycle, oral contraceptive and HRT use with BC risk. A pooled analysis of nine prospective studies on 663 BC cases and 1765 controls found that increasing concentrations of oestrone, androstenedione, dehydroepiandrosterone (DHEA), dehydroepiandrosterone sulphate (DHEAS) and testosterone were associated with increased risk of BC in postmenopausal women [7]. Whilst most of these associations were thought to be due to the conversion of androgens (DHEA, DHEAS, testosterone and androstenedione [8, 9]) to estradiol, these associations remained even after adjustment for circulating estradiol levels [7, 10]. Androgen receptors have been shown to increase proliferation when expressed in triple-negative breast cancer (TNBC), further providing evidence for the role of androgens in BC risk independent of estradiol [11]. Positive associations with premenopausal BC were also found for estradiol, androstenedione, DHEAS and testosterone in a pooled analysis of seven prospective studies including 767 women with BC and 1699 controls [12]; however, a much larger study conducted in UK Biobank among 30,565 premenopausal women and 133,294 postmenopausal women found that testosterone and sex hormone-binding globulin (SHBG) increased and decreased BC risk in postmenopausal women, respectively, but did not influence premenopausal BC risk [13].

Common metabolic pathways may underlie the relationship between sex hormones and BC risk. They are all produced from cholesterol and are synthesized in the gonads, adrenal cortex and placenta [14]. Cholesterol is first transported into the mitochondrion and converted to pregnenolone—the precursor for all sex hormones (Fig. 1) [15‐17]. Whilst approximately half of the testosterone originates from the adrenal glands and the ovaries, the remainder is derived from the conversion of proandrogens (DHEA, DHEAS and androstenedione) in the periphery [18]. In postmenopausal women, the primary source of estradiol is from the conversion of androgens [19].

Much of the current evidence surrounding sex hormones and BC risk comes from observational epidemiological studies. However, these studies are prone to confounding, selection bias and other biases [20, 21]. The most reliable method for evaluating the effects of exogenous sex hormones on BC risk is through conducting randomized controlled trials (RCTs), but these are time-consuming and costly [22], especially in the case of primary prevention trials of cancer. For this reason, other approaches to causal inference such as Mendelian randomization (MR) can be used to provide evidence for or against the role of sex hormones. MR uses genetic variants that predict changes in exposures (e.g. hormone levels) and assesses their effect on outcome (e.g. BC) [21, 23, 24]. MR is analogous to an RCT as genetic variants are randomly allocated at conception, similar to random allocation of intervention at the start of a trial [25, 26], and fixed thereafter. This reduces the impact of confounding encountered in observational epidemiology [25].

Previous MR studies have been carried out looking at the effect of testosterone (total and bioavailable) and sex hormone-binding globulin (SHBG) on overall, ER+ and ER–BC risk [27, 28]. In this two-sample MR study, we expanded the analysis to include seven other sex steroid hormones as well as investigating the effect of the hormones on subtype-specific BC risk (luminal A-like BC, luminal B-like BC, luminal B/HER2-negative-like BC, HER2-enriched-like BC, TNBC and BRCA1 mutated TNBC).

In this study, we aimed to assess whether nine sex steroid hormones and SHBG affect BC risk using an MR framework. Overall, we found that an increase in TT, BT, estradiol and DHEAS was associated with overall BC and/or ER + BC.

The association of TT and BT with overall BC as well as ER + BC risk has been reported before using MR methods [27]. However, our study also investigated subtype-specific associations using MR in ER+ tumours luminal A-like, luminal B-like and luminal B/HER2-negative-like BC. Associations with ER+ tumours involving testosterone may be explained by two possible mechanisms—the first involves its conversion to estradiol [19] which binds to the ER and induces transcription of growth-positive genes and reduces expression of negative regulators of cell growth, therefore increasing breast cancer cell proliferation [58]. The second possible explanation for the association may be due to ER expression acting as a proxy for androgen receptor (AR) expression of which AR expression is positively correlated with ER expression in tumours [59, 60]. This is further supported by the finding that only 20–30% of ER–BCs express AR [61]. The literature suggests that resistance to ER therapies may be due to tumour adaptation towards androgen dependence and AR signalling instead, and it has been suggested that patients with ER+/AR+ tumours would most likely benefit from combination therapies targeting both receptors [62]. In order to try and untangle the mechanism through which testosterone acts in breast cancer, genetic association studies on tumour subtypes stratified based on AR expression and ER expression are required.

Our study demonstrated a relationship between estradiol and BC risk in an MR framework. An SD increase in estradiol increased the risk of overall BC and ER + BC as well as the ER + BC subtype luminal B/HER2-negative-like. We used summary statistics from Pott et al. to identify suitable genetic instruments for estradiol [37]. Despite the more sensitive methods for measuring E2, the average levels of the hormones detected for LIFE-Adult and LIFE-Heart were 18.4 pmol/L and 11.1 pmol/L, respectively[37], compared to ≥ 200 pmol/L in premenopausal women [37]. This may be because the average age of women in these studies was 59.4 and 64.8 years, respectively, which indicates that a large percentage of these cohorts may have been postmenopausal. Postmenopausal women no longer produce estradiol from the ovaries, and so production of this hormone is through the conversion of androgens to estradiol which occurs at the tissue of interest [19]. Therefore, estradiol production in postmenopausal women is localized and may have resulted in lower detection of estradiol in the blood.

These low levels of estradiol may explain why we only had two instruments to proxy for the hormone and why they only explained 0.64% of the variance, with a low F statistic indicating that these were weak instruments (8.40). However, one SNP (rs2414098) near CYP19A1 shows evidence for a key biological role in affecting hormone levels, with suggestive evidence that it is associated with an increase in CYP19A1 expression in breast tissue and would theoretically result in increased conversion of testosterone to estradiol. This plausible biological role of the SNP supports the results, despite the F statistic suggesting weak instruments. We also find no strong evidence of genetic correlation between TT and BT, indicating that the estradiol effect on BC risk may be independent of testosterone.

Similarly, an SD increase in DHEAS was associated with an increased risk of ER + BC, and subtype analysis showed positive associations with the three ER + BC subtypes: luminal A-like, luminal B-like and luminal B/HER2-negative-like BC. These results support positive associations found in observational studies between DHEAS levels and BC risk in postmenopausal women [7, 63].

While observational studies have shown generally consistent results with regards to sex hormones and postmenopausal BC risk, few studies have looked at the association with premenopausal BC. ER + BC is generally found in older and postmenopausal women, and ER−BC is generally found in younger premenopausal women [64]. Analysis of seven prospective studies found that doubling concentrations of estradiol, androstenedione, DHEAS and testosterone all increased the risk of premenopausal BC [12]. Unlike postmenopausal women, premenopausal women produce estradiol in the ovaries which then circulates in the blood [15]. The difficulty in trying to understand the relationship between estradiol and BC in premenopausal women is due to the much smaller sample sizes of cases in prospective cohorts as well as difficulty in accounting for the phase of the menstrual cycle which impacts measures of estradiol [15]. For this reason, the association between estradiol and premenopausal BC is still unclear [15]. It is important to note that the sex steroid hormone GWASs used in this study have mostly been conducted on older-aged women of which a large percentage are postmenopausal. Since ER–BC tends to occur more commonly in premenopausal women, instruments robustly predicting hormonal levels in premenopausal women need to be identified and used instead.

Whilst our study showed associations between testosterone, estradiol, DHEAS and cortisol with various BC subtypes’ risk in an MR framework, it is not without limitations. Firstly, the sample sizes of the GWAS from which some of our exposure instruments were derived were relatively small, and therefore the instruments used were weak, especially in the case of androstenedione, aldosterone, progesterone and 17-OHP. In the case of a two-sample MR setting, using weak instruments will bias the causal estimate towards the null [65] and may explain some of the null associations observed. The lack of genome-wide significant SNPs for androstenedione and aldosterone may have been due to the small sample sizes of the GWASs for these hormone measurements (N = 712 and N = 686, respectively). In addition, participants in the LIFE-Heart study were selected based on suspected or confirmed coronary artery disease, indicating possible selection bias. Furthermore, we derived genetic instruments for DHEAS and cortisol from mixed populations, due to much larger sample sizes than female-specific GWASs. This means that larger GWASs specifically in females need to be performed to identify stronger genetic instruments for these hormones before definitive conclusions on null associations can be made.

Further limitations include that our study also does not differentiate between exogenous sources of these hormones or endogenous, which is important for public health interventions such as advising for or against oral contraceptives and HRT use. We also acknowledge that the effect sizes observed in this study are small—the highest effect association was found between BT and luminal A BC risk (OR 1.29, 95% CI 1.15–1.43), indicating that increased hormone levels may slightly increase the risk of BC and that perhaps higher levels obtained through exogenous sources increase this risk further. Furthermore, MR investigates the lifetime effect of an exposure [23], whereas these drugs are often taken at specific time points or for certain durations. Therefore, it is difficult to identify the duration for which these drugs or exogenous sources of hormones could be affecting the risk of disease. Moreover, genetic instruments used in MR studies typically proxy a small amount of variation in the exposure [66]. For some exposures, larger variations may be required to detect an effect on an outcome which MR would otherwise show as null.

Overall, our results suggest that increasing levels of testosterone, BT, estradiol and DHEAS may increase the risk of overall BC and/or ER+ (postmenopausal) BC risk, consistent with results from observational studies. For the remaining hormones, we found some suggestive associations but also acknowledge the possibility of weak instrument bias and the need for better genetic instruments. Our study provides new insights into the role of sex steroid hormones in BC risk using MR and may inform the eventual development of interventions aimed at BC prevention.