Estrogenic Activity and Risk of Invasive Breast Cancer Among Postmenopausal Women in the Nurses' Health Study

Breast cancer is a leading cause of cancer incidence and mortality among women (1). Estrogen, an essential steroid hormone, promotes breast carcinogenesis and high endogenous levels are a well-confirmed risk factor for postmenopausal breast cancer (2). Prior epidemiologic studies have consistently linked the most common circulating estrogens, here termed classic estrogens ([i.e., estradiol (E2), estrone (E1), and E1 sulfate (E1S)], to risk, with approximately a 2-fold increase in breast cancer risk comparing the top with bottom quartile of estrogen levels (3–6).

The biologic effects of the classic estrogens are mediated through binding to the estrogen receptors (ERα and ERβ) leading to expression of multiple ER target genes (2). Endogenous estrogen metabolites as well as environmental exposures, such as phytoestrogens (e.g., soy products) and xenoestrogens (e.g., phthalates and some pesticides), can also bind to ER and alter its function (7–10). Given varying binding affinities, these compounds may activate (or inhibit) estrogen pathway signaling leading to, in the case of activation for example, increased cellular proliferation (11). The influence of both classic estrogens and these other compounds on estrogen pathway activation should provide a cumulative, biologically relevant predictor of breast cancer risk, as well as offer insights into the contribution of these other compounds to risk. Several estrogen bioassays have been developed to assess the cumulative effect of estrogenic components, both endogenous and exogenous estrogens, on total estrogen pathway activity (EA; refs. 12–14).

Prior epidemiologic studies of EA and breast cancer were conducted using either a yeast-based assay (12, 13) or a cervical cancer cell assay (14) and observed positive associations. However, when specifically evaluating risk of breast cancer these assays may not be as physiologically relevant as using a breast cancer cell line, leading to an underestimation of the association. For example, ER signaling in yeast cells is different than in humans due to variations in cell membrane permeability and ER coregulator recruitment (15, 16). A bioassay, using T47D human breast cancer cells transformed by transfecting estrogen-responsive elements, has been shown to quantify total estrogenic (and antiestrogenic) activity that activates the ER and induces gene transcription (17).

We conducted a nested case–control study to evaluate the relationship between plasma EA as assessed in breast cancer cells and risk of invasive postmenopausal breast cancer in the Nurses’ Health Study (NHS), a large ongoing prospective cohort. We further evaluated whether the association was independent of the classic estrogens which would suggest a potentially important association with other estrogenic compounds in driving breast cancer development.

The NHS is a U.S.-based cohort that began in 1976 with 121,700 female, registered nurses, ages 30 to 55 years. Participants are mailed biennial questionnaires to provide information on health and lifestyle factors (e.g., demographics, reproductive history, diet, disease diagnosis). Cohort details have been described previously (18), and all study protocols were approved by the Institutional Review Boards at Brigham and Women's Hospital and the Harvard T.H. Chan School of Public Health in Boston, MA, the University of Massachusetts Amherst in Amherst, MA, and those of participating registries as required. Heparin plasma blood samples were provided by 32,826 women (ages 43 to 69 years) from 1989 to 1990. Blood collection kits were sent to each woman, and after their blood was drawn, the samples were shipped with an ice pack to the laboratory for processing and storage in nitrogen freezers at −130°C or colder (19).

In our study, all women were postmenopausal and not currently using hormone therapy, with no prior cancer history, except nonmelanoma skin cancer, at time of blood draw. Women were defined as postmenopausal if they reported having no menses in the past 12 months, a hysterectomy and/or a bilateral oophorectomy, or were at least 54 years of age for current smokers and 56 years of age for nonsmokers; these ages represent when 90% of the cohort had reported natural menopause.

Women reported breast cancer diagnoses on the biennial questionnaires; permission to obtain medical records was requested from those reporting a new diagnosis. Medically trained cohort investigators confirmed diagnoses through medical record review. Over 99% of reported breast cancer diagnoses were confirmed after medical record review. In the NHS, tumor microarrays were constructed from paraffin blocks of breast cancer tumor tissue retrieved from the participant's hospital, and immunostained for ER and progesterone receptor (PR) using a Dako Autostainer (Dako Corporation). A breast pathologist reviewed the stains and determined positivity. Where tissue was not available, ER/PR status, histologic grade, and invasiveness were extracted from pathology reports (20).

Cases in this study were diagnosed with invasive breast cancer from 1994 to 2004. Because plasma samples are particularly limited for cases diagnosed early in follow-up, we chose a period that avoided use of the earliest diagnosed cases and extended 10 years to provide sufficient statistical power for our study. Cases were matched to controls by date and time of blood draw, fasting status, and year of birth using incidence density sampling (21). Two controls per case were included for all hormone assays except for data from 2002 and 2004, when only one control per case was assayed for the measurement of E1. The total number of women with measured classic estrogens in our study were as follows: 1,000 (E1), 1,049 (E1S), and 1,060 (E2).

Total EA was assessed using a receptor-mediated Chemical-Activated LUciferase gene eXpression bioassay performed at the laboratory of Dr. Martyn Smith (UC Berkeley, Berkeley, CA). Details of the plasma assay methods were described previously (17). Briefly, EA in human plasma was measured using transfected human breast cancer cells, T47D-Kbluc (ATCC), with a luciferase reporter. Each sample was run in triplicate. Reporter activity was measured by the fluorescent light emitted per well and results were averaged across three wells per sample; higher relative light units (RLU) indicated greater EA. RLUs were converted to picomoles using a 17β-estradiol concentration response standard curve (Tocris Bioscience), which was then converted to picograms per milliliter (pg/mL) for comparability with the measurement of the classic estrogens. For our project, plates with an average interwell coefficient of variation (CV) ≥24% (n = 14) or with a technical failure (n = 1) were rerun once. In total, we had 76 plates and the average within plate CV was 9.5% after these plates were rerun once.

For all plasma analyses, matched case–control sample pairs were included together within each batch and lab personnel were blinded to case–control status. Replicate quality control samples (10% of samples) from three postmenopausal plasma pools were included among case and control samples and were indistinguishable from these samples to calculate CVs.

Plasma E2, E1, and E1S assays were performed by the Nichols Institute at Quest Diagnostics and the Mayo Clinic. E2 and E1 were analyzed using radioimmunoassay (RIA) following organic extraction and celite column chromatography (1994–2000 follow-up cycles), and at the Mayo Clinic using LC/MS-MS (2002–2004 follow-up cycles). E1S (1994–2004 follow-up cycles) was extracted via an organic solvent and assayed by RIA after E1 extraction and enzymatically cleaving the E1S to release estrone by column chromatography (22). The average (range) of within batch coefficients of variation reported by each laboratory are as follows: Quest Diagnostics: E2 8.9% (3.9%–16.4%); E1 9.4% (8.7%–12.0%); and E1S 10.3% (5.5%–12.9%). The Mayo Clinic assays had an average (range) CV of 7.6% (4.0%–9.5%) for E2 and E1 (2000–2004 follow-up cycles).

Classic estrogens were previously average batch recalibrated to account for batch and participant characteristics that may have varied by batch, [i.e., age and body mass index (BMI) at blood collection], as described previously (4, 23).

Information on breast cancer risk factors was measured via NHS biennial questionnaires and the questionnaire administered at the time of blood collection. Self-reported weight at age 18, height, and age at menarche were obtained from the baseline questionnaire administered in 1976. Family history of breast cancer was queried from 1976 to 1988. History of benign breast disease confirmed by biopsy, age at natural menopause, parity/age at first birth, and alcohol intake were obtained from the biennial questionnaire administered in 1990. Date and time of blood draw, fasting status, age at blood draw, and current weight were recorded at the time of blood collection. Missing covariate information was substituted with the sample median value for continuous variables and the most frequent value for categorical variables.

EA and hormone concentrations were natural log (ln) transformed to improve normality. One woman with an EA value below the limit of detection, we assigned a value of 0.5 pmol/L, which is half the limit of detection for this bioassay. We performed the generalized extreme Studentized deviate test (24) to identify statistical outliers (n = 7 for EA assay) which were removed from our analyses.

Our primary analysis examined the results of the EA assay without modification. To then assess the contribution of EA to risk, independent of the classic estrogens, we created two estrogen-adjusted residual variables [termed here E2-adjusted and 3-estrogen- (i.e., E1, E2, E1S) adjusted] by regressing EA on ln-transformed, continuous E2 only or E1, E2, and E1S simultaneously, taking the residual EA value, and then adding the EA mean to the residual value for each woman (25). We also calculated an E2 subtraction variable by subtracting the plasma E2 value from the EA value, and adding the minimum estradiol value, rounded to the nearest whole number, as in prior research (26). Quartiles for each EA variable were created on the basis of the distribution among the controls. We calculated Pearson correlation coefficients with 95% confidence intervals (CI) between classic estrogens and EA concentrations and BMI at blood collection.

Participant characteristics between cases and controls were assessed using t tests for continuous variables with reported means and SDs, and the Mantel–Haenszel tests for categorical variables with reported total number and frequency per category. For the primary analysis, conditional logistic regression was used to calculate ORs of incident breast cancer and 95% CI for each quartile of EA compared with the lowest quartile (referent). Tests for trend across quartiles were conducted by modeling the median of each EA quartile category as a continuous variable. Cubic splines were used to test for nonlinearity between EA concentrations and breast cancer risk. To assess differences by hormone receptor subtype, heterogeneity methods outlined by Wang and colleagues (27) for conditional logistic regression models were used, and the Wald test for heterogeneity was reported. We evaluated whether associations varied across categories of family history of breast cancer (yes/no) and BMI (>25 vs. ≥25 kg/m²) using unconditional logistic regression to maximize statistical power and determined statistical significance via likelihood ratio tests comparing models with and without multiplicative interaction terms. In multivariable models, we adjusted for BMI at age 18 (continuous), weight change from age 18 to blood draw (continuous), family history of breast cancer (yes/no), history of benign breast disease (yes/no), age at menarche (continuous), parity/age at first birth (nulliparous, 1–2 children/<25 years, 1–2 children/≥25 years, ≥3 children/<25 years, ≥3 children/≥25 years), age at natural menopause (continuous), and alcohol intake (continuous) selected a priori; matching factors were included in unconditional multivariable models. Controlling for BMI/weight change would account for potential confounding but could also result in overcontrol if these variables are on the same biologic pathway, thus we assessed models with and without these covariates, and results were essentially unchanged.

We observed substantial between batch variation in the bioassay (CV = 27.3%). Therefore, for unconditional analyses, we created a categorical variable that grouped batches together based on the median bioassay value observed in each batch. These groupings resulted in a four-category batch variable to account for batch effects. Results from unconditional logistic regression models including this batch effect variable were similar to results of conditional logistic regression models.

Among all women who provided a blood sample in 1989–1990, 300 women were invited to participate in the Hormone Stability Substudy by providing two additional blood samples over the following 2 to 3 years, as described previously (28). The intraclass correlation coefficients (ICC) from this substudy ranged from 0.68 (95% CI, 0.59–0.80) to 0.75 (95% CI, 0.67–0.84) for the classic estrogens. From this substudy, we selected up to 30 women with measured classical estrogens to assess EA reproducibility over time.

All statistical analyses were performed using SAS version 9.4 (SAS Institute Inc.) for UNIX.

A total of 374 cases and 735 controls with an EA value and at least one successful hormone (E2, E1, E1S) measurement were included in our study. After excluding outliers, the final sample size consisted of 371 cases and 731 controls.

At blood collection, women with breast cancer were more likely to have a family history of breast cancer, greater weight change from age 18 to blood draw, a slightly later age at menopause, and a higher concentration (pg/mL) of EA, E2, E1, and E1S compared with their matched controls (Table 1). Cases also were generally more likely to have a history of benign breast disease.

Among controls, levels of E1, E1S, and E2 were moderately correlated with EA [r = 0.30 (0.23–0.37), r = 0.34 (0.27–0.41), and r = 0.41 (0.35–0.47), respectively] (Table 2). E2-adjusted, 3E-adjusted, and E2 subtraction EA were highly correlated with EA [0.91 (0.90–0.92), 0.89 (0.87–0.91), and 0.76 (0.73–0.79), respectively], though were minimally correlated, as expected, with the classic estrogens. Notably, EA was modestly correlated with BMI at blood collection [r = 0.23 (0.16–0.29)], whereas there was no correlation observed once the variation due to E2, E1, and E1S was removed.

In analyses not adjusting for classic estrogens, women in the highest quartile had a 96% higher risk of invasive breast cancer (OR_Q4vsQ1, 1.96; 95% CI, 1.24–3.11; P_trend = 0.003) compared with women in the lowest quartile. This association was modestly attenuated after accounting for covariates (OR_Q4vsQ1, 1.86; 95% CI, 1.16–2.97; P_trend = 0.01). Additional attenuation was observed after accounting for the variation due to E2 in EA among women with measured E2-adjusted EA (OR_Q4vsQ1, 1.27; 95% CI, 0.75–2.17; P_trend = 0.51) and E2 subtracted EA (OR_Q4vsQ1, 1.08; 95% CI, 0.65–1.78; P_trend = 0.42). The association was null after accounting for the cumulative contribution of the classic estrogens to EA (3E-adjusted EA OR_Q4vsQ1, 1.01; 95% CI, 0.56–1.84; P_trend = 0.93; Table 3). When modeled individually, and comparing the top versus bottom quartile categories, the association of EA with breast cancer risk was 1.86 (1.16–2.97) and for E2 and breast cancer risk was 2.19 (1.45–3.31). When mutually adjusted for, these EA and E2 associations were both somewhat attenuated with OR_Q4vsQ1 = 1.42 (0.79–2.55) and OR_Q4vsQ1 = 1.89 (1.17–3.07), respectively.

In analyses considering hormone receptor status of the tumor, we observed a 2.5-fold increase in association between EA and invasive breast cancer risk among ER⁺/PR⁺ subtype (OR_Q4vsQ1, 2.50; 95% CI, 1.34–4.68), in analyses not adjusting for classic estrogens. Similar to our primary analyses, the association was attenuated for E2-adjusted, 3E-adjusted, and E2 subtraction EA (OR_Q4vsQ1, 1.49; 95% CI, 0.76–2.93; OR_Q4vsQ1, 1.17; 95% CI, 0.55–2.50; OR_Q4vsQ1, 1.14; 95% CI, 0.60–2.15, respectively; Table 4). In the ER⁺/PR⁻ and ER⁻/PR⁻ subtypes, the associations were null with wide CIs. Tests for heterogeneity between hormone receptor subtypes for each assessment of EA were not statistically significant (P_{heterogeneity} > 0.34). When examining associations of EA and invasive breast cancer risk stratified by BMI at blood collection and family history of breast cancer, no statistically significant heterogeneity was observed (P_interaction > 0.15).

When examining the reproducibility over 2–3 years of EA among a subgroup of our participants, the ICC was 0.87 (95% CI = 0.76–0.94). For the estrogen-adjusted EA variables, ICCs were 0.80 (95% CI = 0.62–0.90) for E2-adjusted EA and 0.57 (95% CI = 0.31–0.79), among 26 women, for 3E-adjusted EA.

In this large prospective nested case–control study, we observed a 2-fold higher risk of invasive breast cancer among those in the top versus bottom quartile of ER activated estrogen pathway activity as measured by total plasma EA concentrations. However, this association was substantially attenuated after accounting for plasma levels of E2 and no association was observed after accounting for the three classic estrogens together.

A number of estrogen bioassays have been developed previously (12–14). The assays are intended to measure the cumulative impact on ER activity by both classic estrogens and a range of endogenous (e.g., estrogen metabolites) and exogenous compounds (e.g., phytoestrogens and certain pesticides) not typically measured in epidemiologic studies. Although ER binding affinities for these other compounds are generally lower than for E2, the compounds have similar molecular structure to E2, and thus can compete with classic estrogens for ER binding and activation (or inhibition; ref. 29). Furthermore, because E2 levels are relatively low among postmenopausal women, these other compounds could play a particularly important role by influencing the ER (2, 29).

ERα and ERβ are nuclear receptors that regulate the expression of ER target genes involved in various biological processes, for example, cell proliferation, and are responsible for binding as dimers to estrogen response elements for transcription (2, 30). Both ERs have high affinity for E2, and although similar homology is observed in their ligand-binding domains (53%), the ERs respond differently to various ligands (31). For example, ERβ has a greater affinity for phytoestrogens, and ERα has both antagonist and agonist effects in the presence of selective ER modulator drugs (e.g., tamoxifen; refs. 32, 33). Notably, ERα and ERβ have opposing effects in the promoter region of various proliferation genes, and therefore, understanding the ratio and fluctuation of both ERs in target tissue may be an important component in assessing risk (2, 31).

To date, three prior epidemiologic studies have utilized EA reporter bioassays to evaluate the relationship between EA and breast cancer in postmenopausal women (12–14). All three studies separately evaluated ERα and ERβ activity whereas our study reported cumulative ERα and ERβ activity. Each observed significant positive associations of similar magnitude to our findings, in at least a subset of the data. Two studies utilized the same yeast reporter bioassay (12, 13) and the third used a cervical cancer HELA cell assay (14). In a retrospective assessment (12) with 182 cases, women in the highest (vs. lowest) category were at a significant 2.5-fold higher risk for both ERα and ERβ activity, assessed separately or together, in the same model. In the prospective nested case–control study using a yeast-based assay (ref. 13; n = 200 ER⁺ cases), no significant association was observed overall, although significant positive associations of a similar magnitude to associations seen in our study, for both ERα and ERβ activation, were observed in cases diagnosed >2 years since blood collection (95 cases). In the prospective nested case–control study using the HELA cell assay (ref. 14; 169 cases), positive associations for ERα were suggestively stronger than for ERβ [comparable comparison, OR_Q4vsQ1 = 2.80 (95% CI, 1.58–4.97) for ERα and 1.74 (95% CI, 0.89–3.41) for ERβ]. Overall, these studies provide further support that activation of ERs is an important etiologic pathway in breast tumorigenesis.

Prior studies also evaluated whether associations of EA and breast cancer were attenuated after accounting for E1 and/or E2, although differences in the estrogen assays used limit comparisons. In two studies (12, 13), direct assays that often have lower sensitivity and specificity in detecting estradiol (or estrone) specifically were used, which would lead to misclassification in postmenopausal women (34, 35). Perhaps because of this, E2 concentrations were not associated with breast cancer risk in those studies, limiting the ability to evaluate how E2 adjustment influenced associations for the EA assay results. Furthermore, the yeast-based assay used was not correlated with serum E2 (r ≤ 0.08), possibly due to the use of a non-human cell type for the EA assay or measurement error in the EA or E2 assay. In contrast, we observed a correlation of 0.41 (95% CI = 0.35–0.47) between plasma E2 and the EA bioassay. In Fourkala and colleagues (13), E1 was significantly positively associated with breast cancer. However, in the smaller subgroup with 95 cases where EA associations were observed, E1 and E2 were only weakly related to risk and they did not substantially alter the EA associations. In Lim and colleagues (14), where more sensitive and specific LC/MS assays were used, E1 and E2 were nonsignificantly positively associated with risk. E1, E2, and ERα bioassay values were included in the same model, and only the ERα bioassay remained significantly associated with risk. Our use of sensitive and specific estrogen assays, additional consideration of E1S, and use of a regression method that may better remove variation due to the classic estrogens, may account, at least in part, for the greater attenuation we observed in the EA/breast cancer association. Given the limited data to date, further assessment of this issue is warranted.

Our study has both strengths and limitations. It is the first to use a breast cancer cell line–based bioassay, which may be more biologically relevant than previously used bioassays. Our bioassay did not, however, distinguish ERα from ERβ activation and the T47D-Kbluc cell line expresses somewhat higher levels of ERβ than ERα (17). Our exposure was assessed using a single blood sample, and we noted laboratory variation in our cell-based assay, which would attenuate the observed associations. However, we saw very good reproducibility of the bioassay over a 2-year period both overall and after removing the variation due to E2. Although after accounting for all three classic estrogens, the ICC appeared somewhat attenuated. We also assessed the contribution of classic estrogens to EA, using state-of-the-art and precise measurements of E2, E1, and E1S. Our sample size was sufficient to evaluate the primary association between EA and total breast cancer risk; however, it was small for assessing potential interactions by BMI and family history of breast cancer, and when considering less common tumor hormone receptor subtypes. Our population was largely non-Latina white, and given the distribution of EA levels varies across racial/ethnic groups (26), additional studies are warranted.

Our findings suggest a broader assessment of plasma compounds contributing to ER activation, as measured specifically using the T47D-Kbluc cell line bioassay, does not provide additional information on breast cancer risk beyond measuring classic estrogens alone. Further large prospective assessments that can distinguish ERα and ERβ pathway activation using a breast cancer relevant assay, and that can also account for classic estrogen levels, would provide additional insight.

S.S. Sanchez reports no other relationships/conditions/circumstances that present a potential conflict of interest. A.H. Eliassen reports grants from NIH during the conduct of the study. S.S. Tworoger reports grants from NIH/NCI during the conduct of the study; grants from NIH/NCI, State of Florida, DOD, BMS; personal fees from AACR, Ovarian Cancer Research Alliance, Ponce Health Sciences University, German Cancer Research Center, and Harvard T.H. Chan School of Public Health outside the submitted work; and is member of external advisory committee of California Teachers Study (City of Hope; unpaid) and The Tomorrow Project (Alberta Cancer Center; unpaid). S.E. Hankinson reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

E.X. Holder: Formal analysis, writing–original draft, analyzed and interpreted the data. S.C. Houghton: Writing–review and editing, contributed to the analysis and interpretation of data. S.S. Sanchez: Writing–review and editing. A.H. Eliassen: Writing–review and editing. J. Qian: Writing–review and editing, provided statistical expertise. E.R. Bertone-Johnson: Writing–review and editing. Z. Liu: Writing–review and editing. S.S. Tworoger: Writing–review and editing. M.T. Smith: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing, contributed to the interpretation of the data. S.E. Hankinson: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing, contributed to the interpretation of the data.

We would like to thank the participants and staff of the NHS for their valuable contributions and the Breast Cancer Research Foundation as well as the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY. The authors assume full responsibility for analyses and interpretation of these data. We thank Dana-Farber/Harvard Cancer Center in Boston, MA, for the use of the Specialized Histopathology Core, which provided histology and IHC service. Dana-Farber/Harvard Cancer Center is supported in part by an NCI Cancer Center Support Grant #NIH 5 P30 CA06516. Research reported in this publication was supported by the NCI of the NIH under award numbers UM1CA186107, P01CA87969 awarded to A.H. Eliassen, and R01CA49449 and R01CA207369 awarded to S.E. Hankinson. E.X. Holder was funded in part by R01CA207369. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.