Background: Estrogen exposure is a major risk factor for breast cancer. Increased estrogen responsiveness of breast epithelium may enhance this effect. We examined the relationship between breast cancer diagnosis and 1) the presence and absence of estrogen receptor expression in benign breast epithelium, 2) the level of expression and 3) its variation during the menstrual cycle, and 4) other established risk factors. e.g., age, age at menarche, parity, and family history. Methods: We measured estrogen receptor expression (as % of positive cells) by immunohistochemistry in normal breast epithelium from 376 women undergoing diagnostic or therapeutic breast surgery. Data on established risk factors were collected prior to surgery and those on menstrual cycle dates at the time of surgery. Logistic regression was used to assess risks (odds ratios [ORs]). Results: The crude OR for breast cancer in women with estrogen receptorpositive breast epithelium versus those without was 3.16 (95% confidence interval [CI] = 1.89–5.28), with an OR of 2.49 (95% CI = 1.25–4.96) for premenopausal and an OR of 3.32 (95% CI = 1.43–7.68) for postmenopausal women. The ORs remained high and statistically significant after controlling for age and other breast cancer risk factors. The level of estrogen receptor expression was higher in patients with breast cancer than in control subjects and it was related to breast cancer risk in postmenopausal women (P trend <.005). Expression declined as expected in premenopausal control subjects as the menstrual cycle progressed but rose in breast cancer patients (P trend <.015). Conclusions: The overexpression of estrogen receptors in normal breast epithelium may augment estrogen sensitivity and hence the risk of breast cancer. [J Natl Cancer Inst 1997;89:37–42]
Estrogen exposure is a major contributor to the risk of developing breast cancer, but the biologic mechanisms involved are poorly understood. One of the links is probably through the induction of proliferation of breast epithelium because every mitotic event provides an opportunity for genetic mishaps (1-3). Other possible mechanisms include a genotoxic role for estrogen metabolites (4) and high levels of both total and free estradiol in the sera of breast cancer case subjects compared with those of control subjects (5-7). Enhanced estrogen responsiveness of the target organ (breast epithelium) may be partly responsible for breast cancer susceptibility but has not been systematically explored. This forms the basis of this study.
Estrogen response requires transport of estrogen into the cell, binding of estrogen to estrogen receptors, binding to DNA, and transcription of estrogen-responsive genes (8), one of which is the gene for progesterone receptor (9,10). Given the obligate role of estrogen receptor in estrogen response and the fact that steroid receptor content appears to limit cellular response to steroids (11,12), we hypothesized that the histologically normal breast epithelium of women with breast cancer (case subjects) may demonstrate an increased estrogen receptor content when compared with women with benign breast disease (control subjects). We have previously reported data on 120 women, which show a significantly greater prevalence of estrogen receptor-positive epithelium in case subjects over control subjects (85% versus 55%) (13). We next hypothesized that estrogen receptor alpha content of breast epithelium may be determined by factors, such as age at menarche, parity, and menopause, or it may be influenced by use of exogenous estrogens. We now report data on a total of 376 women with an examination of the relationship of estrogen receptor positivity to other risk factors that are endocrine related, e.g., differences in thresholds for estrogen receptor positivity, and of estrogen receptor variation during the menstrual cycle.
Our study subjects were recruited from the Breast Care Center at University Hospital, Syracuse, NY; the study was approved by the Institutional Review Board. All patients completed a self-administered questionnaire regarding breast cancer risk factors at their first clinic visit. For most women, surgery was performed within 1 month of questionnaire completion. If the interval between questionnaire completion and study participation exceeded 1 year, the questionnaire information was updated by direct questioning. Potential study subjects were asked to participate by allowing use of a sample of their normal breast tissue (i.e., tissue of normal gross appearance) from the surgical specimen. All participants signed a document of informed consent. Case subjects were women with newly or previously diagnosed invasive or in situ breast cancer who required further breast surgery, and control subjects were women without a prior history of breast cancer who required diagnostic breast biopsy, but proved not to have breast cancer. Seven cases were previously diagnosed with breast cancer; the interval from cancer diagnosis to the biopsy that resulted in study participation ranged from 7 months to 3 years. The indication for biopsy in six of these women was the appearance of a new breast lump or mammographic density in the contralateral breast. One woman underwent a prophylactic contralateral mastectomy 3 years after the diagnosis of her original duct carcinoma in situ. Indications for diagnostic biopsy included palpable lumps, mammographic abnormalities, and nipple discharge. Benign lesions encountered in these biopsy specimens ranged from fibroadenoma to various types of nonproliferative fibrocystic disease and proliferation without atypia, including sclerosing adenosis. Women with atypical hyperplasia on biopsy were not specifically excluded from being control subjects, but in fact only one control subject was found to display atypical proliferation in her biopsy specimen.
The study population was accrued between December 1990 and December 1995. Fresh surgical specimens were reviewed by a pathologist, who released a grossly normal sample of breast tissue for study if hefelt that making an accurate diagnosis of the patient'ss condition would be unaffected by permitting the use of this sample. The potential study population comprised 1620 women undergoing breast surgery at University Hospital during this interval. Breast tissue was released for this study by the examining pathologist on 531 (32.8%) of these women, who formed the actual study population. The decision not to release tissue for research was based solely on specimen size and the need to evaluate surgical margins or exclude a diagnosis of carcinoma in situ. The mean age of the actual study
If the surgical procedure was a mastectomy, four samples were obtained, one from each quadrant. The samples were embedded in tissue-freezing medium (O.C.T., Miles Chemical Co., Elkhart, IN) and snapfrozen in liquid nitrogen. Cryostat sections were first evaluated by hematoxylin-eosin staining. The sections were processed further only if adequate normal epithelium was present. Adequacy was defined as a minimum of 10 ducts or lobular acini. For mastectomy specimens, the sample demonstrating the best normal epithelium on hematoxylin-eosin staining was chosen for immunohistochemical demonstration of receptor expression. Epithelial samples included in the study showed either morphologically normal epithelium or minimal nonproliferative benign change. The tissue was adequate for evaluation of receptor status in 398 women; 22 subjects were excluded for incomplete information on risk factors, leaving a final study population of 376 women. Of these, 219 were premenopausal (70 case subjects and 149 control subjects) and 157 were postmenopausal (104 case subjects and 53 control subjects). Exclusions for inadequate epithelium were mainly in postmenopausal control subjects whose samples consisted of fatty tissue with very scant epithelium.
Five-micron cryostat sections were fixed in 10% formaldehyde, methanol, and acetone, according to the manufacturer'ss instructions for the estrogen receptor-immunocytochemical assay and progesterone receptor-immunocytochemical assay kits (Abbott, Chicago, IL); primary antibodies were applied (antiestrogen receptor antibody H222 and antiprogesterone receptor antibody KD68, respectively) for 30 minutes and treated with bridging antibody (goat anti-rat). PAP (peroxidase) complex was added for 30 minutes followed by a phosphatebuffered saline wash and chromagen (diaminobenzidine).The dilution of reagents used was in accordance with the manufacturer'ss directions. A hematoxylin counterstain was used. Proliferative epithelium was excluded from analysis; only normal epithelium and tissue with mild degrees of adenosis (mild architectural distortion without hyperplasia) were analyzed. Sections were scored positive or negative for receptor expression based on counts of an average of 2000 cells at ×40 magnification. On the average, every third epithelium-containing field was counted. Labeling index was calculated as the percent of immunostained epithelial nuclei. The method of scoring benign tissue was developed in our laboratory and has been internally validated as having an interobserver variation of less than 20%. The threshold for positivity for both estrogen receptor and progesterone receptor was prospectively set at an labeling index of 1%. This threshold was chosen because it is equivalent to a level of 10 fmcytosol protein in ligand-binding assays (14). Negative control slides were processed for every patient in a similar manner.
The primary relation of interest was that of estrogen receptor positivity with the occurrence of breast cancer. Twenty-two subjects were excluded at the outset for incomplete risk factor information because their risk factor questionnaires were missing. Women with missing information on one or two parameters were included in the calculation of crude and age-adjusted ORs but were excluded from the multivariate analysis reported in Table 1. Since prior studies have indicated differences in breast cancer risk factors in premenopausal and postmenopausal women, the results were stratified by menopausal status. Menopausal status was determined from the self-administered questionnaire according to patient response. Women who indicated that they were postmenopausal were considered to be so. Women who had undergone hysterectomy without oophorectomy were considered postmenopausal. Only three of these women were less than 50 years old. Analysis of the data involved a three-stage process. A univariable analysis was conducted first to determine the distributions of the study variables. This was followed by a bivariable analysis for investigation of possible confounders, outliers, and collinearity. Covariates in the analyses were defined as age at breast tissue sampling, age at menarche, parity, use of synthetic hormones (hormone-replacement therapy [HRT] or birth control pills), body mass index (kg2 ), five categories of alcohol use (never, less than once a month, one to eight times per month, three to six times per week, and more than six times per week), previous diagnosis of cancer at any site, history of breast cancer in mother andsisters, prior exposure to radiation therapy, ethnicity, history of breast-feeding, prior hysterectomy andoophorectomy, smoking history, marital status, and educational level. In the adjusted analyses, only present oral contraceptive and HRT use was considered. Thus, a postmenopausal woman who was currently using HRT, but had used oral contraceptives 20 years previously, was considered an HRT user. Some information was missing on selected covariates, since not every woman who had a benign tissue sample completed all items on the questionnaire. These were coded as missing and were treated as such in the final analysis.
An additional variable was generated to indicate the nulliparous interval, i.e., the number of years between menarche and first-term pregnancy in parous women and menarche and menopause in nulliparous women. This variable reflects the period of time that the breast epithelium was exposed to ovarian hormones, without the effect of a term pregnancy. We developed the nulliparous interval parameter as a biologically sound way of increasing the power of the interaction analyses. The nulliparous interval was included in the analysis as a continuous variable.
Univariable and bivariable analyses were performed using SYSTAT (SYSTAT Inc., Evanston, IL). Initial comparisons of the distribution of estrogen and progesterone receptor-labeling indices were performed using the Wilcoxon rank sum test.
Multiple logistic regression analysis was used to investigate possible effect modification and to adjust for confounding (15). Receptor positivity was set at 1% or more of epithelial cells demonstrating nuclear stain. When investigating the effect of increasing positivity with risk, the categories were set arbitrarily at less than 1%, 1.00–4.99; and 5% or greater; for tests of trend, these categories were coded 0, 1, and 2. Unconditional logistic regression was performed using EGRET (Statistics and Epidemiology Research Corporation, Seattle, WA). Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for receptor positivity were calculated. The likelihood ratio test was used to assess the significance of variables in the model.
To analyze the effect of menstrual cycle on estrogen receptor expression, data were collected on premenopausal women regarding the day in the menstrual cycle when the tissue biopsy was performed. Day 1 in the cycle was the first day of menstruation. The menstrual cycle was a priori divided into six phases based on known differences in plasma concentrations of estradiol and progesterone throughout the cycle. These were days 1–5, 6–9, 10v14, 15–18, 19–24, and 25–28. The length of cycle for each woman was normalized to a 28-day cycle, maintaining a constant 14-day luteal phase. Median differences of estrogen receptor positivity were investigated for case subjects and control subjects, and ORs across the six phases were calculated.
Case subjects and control subjects had similar demographic characteristics, except that case subjects were significantly older than control subjects (mean age, 54 versus 41 years). The proportion of case subjects and control subjects with a positive family history (i.e., an affected firstdegree relative) was similar (22.4% and 22.3%); 21% of the case subjects experienced menarche prior to the age of 12 years versus 22% of the control subjects; parity before the age of 30 years occurred in 17% of the case subjects and in 11% of the control subjects; and 24% of the case subjects experienced menopause after the age of 50 years as opposed to 15% of the control subjects. None of these differences were statistically significant.
Receptor Positivity in Case Subjects and Control Subjects
Estrogen receptor and progesterone receptor positivity were examined initially as continuous variables and both were found to be positively skewed. The median estrogen receptor labeling index was 7.7 in the case subjects and 3.2 in the control subjects (P = .001). The median progesterone receptor labeling index was 12.4 in the case subjects and 14.6 in the control subjects (P =.829). Estrogen receptor positivity (threshold labeling index, 1%), examined as a categoric variable by menopausal status, was significantly different between case subjects and control subjects. In premenopausal women, there was a 17.6% difference in the prevalence of estrogen receptor positivity in benign breast epithelium, with 81.4% (57 of 70) of the case subjects being estrogen receptor positive versus 63.8% (95 of 149) of the control subjects (chi-squared test, P <.008). Among postmenopausal women, the difference was 18.7% (88.5% [92 of 104]) in the case subjects and 69.8% (37 of 53) in the control subjects (P <.004).
The crude ORs for breast cancer were elevated in women showing at least 1% breast epithelial cells positive for estrogen receptor: among all estrogen receptorpositive women, it was 3.16 (95% CI = 1.89–5.28). When stratified by menopausal status, the crude OR for premenopausal women was 2.49 (95% CI = 1.25–4.96) and for postmenopausal women was 3.32 (95% CI = 1.43–7.68). There was no significant difference in the relation between estrogen receptor positivity and breast cancer for premenopausal and postmenopausal women (P = .480 for interaction). When adjusted for age and other known breast cancer risk factors (age at menarche and parity, family history, history of cancer at other sites, body mass index, alcohol use, and hormone use), the OR of carrying a diagnosis of breast cancer in estrogen receptor-positive women was 2.63 (95% CI = 1.47–4.70). For premenopausal women, the OR was 2.04 (95% CI = 0.97–4.3) and for postmenopausal women there was a stronger association between estrogen receptor positivity of breast epithelium and breast cancer, with an OR of 3.8 (95% CI = 1.5–9.8).
Effect of Increasing Estrogen Receptor Content
We next examined the possibility of an increasing effect on breast cancer risk with increasing estrogen receptor positivity. Crude and adjusted ORs for three different categories of estrogen receptor positivity are shown in Table 1. The test for trend in estrogen receptor positivity for all women was statistically significant (P <.001), although the crude ORs for women with estrogen receptor labeling index 1.00–4.99 and estrogen receptor labeling index greater than or equal to 5.00 were similar (3.1 and 3.2, respectively). With the inclusion of additional covariates, the OR estimates for increasing levels of estrogen receptor positivity remained elevated but nonsignificant in premenopausal women. However, a statistically significant increase in the estimated relative risk of developing breast cancer in postmenopausal women was found with the inclusion of covariates. The adjustment for other study covariates (history of lactation, recent birth, history of radiation, smoking, marital status, education, and ethnicity) did not appreciably alter the final ORs.
Effect of Menstrual Cycle Variation
Since estrogen receptor expression in breast epithelium has been described by several authors to vary with the menstrual cycle in premenopausal women (16,17), we examined the effect of timing of sampling relative to the menstrual cycle in premenopausal women. This analysis is shown in , where the median estrogen receptor positivity of case subjects is compared with the median value for control subjects. Each phase contained approximately one-third case subjects and two-thirds control subjects, although the absolute numbers of women in each phase varied from 14 to 76 (). The difference in median estrogen receptor labeling index differed significantly in the 25–28day interval (P = .032); the difference did not reach statistical significance in the other intervals shown, probably because of small numbers. The data suggest that case subjects and control subjects demonstrate opposite trends in the estrogen receptor positivity as the menstrual cycle progresses. In women without breast cancer, an expected decreasing trend in estrogen receptor expression in the latter part of the cycle was observed, as reported by others (16-19). In the case of women with breast cancer, estrogen receptor expression tended to increase with the progression of the menstrual cycle. ORs for breast cancer in relation to estrogen receptor positivity across the six phases of the menstrual cycle were unstable, but generally revealed an increasing trend from OR = 1.25 in days 1–5 to OR = 4.38 in days 25–28 (chi-squared trend = 5.90; P = .015).
The ORs for progesterone receptor positivity were not significantly different from unity; for all women, the OR was 0.996 (95% CI = 0.98–1.01). Progesterone receptor-positive premenopausal women exhibited an OR of 1.0 (95% CI = 0.99–1.02), and postmenopausal women exhibited an OR of 0.999 (95% CI = 0.98–1.02).
Interaction With Other Risk Factors
There were no significant first-order interactions between estrogen receptor positivity and other covariates, with case- control status as the outcome. In particular, there were no interactions with early menarche, late parity, or late menopause. There were 39 case subjects and 97 control subjects with a history of any oral contraceptive use in the past, and estrogen receptor positivity was equally distributed between them. Postmenopausal hormone use was relatively infrequent: 23 case subjects and 30 control subjects described using HRT at any time, and only 14 case subjects and 20 control subjects were current users. Estrogen receptor positivity was significantly more frequent in nonusers if they were case subjects (88.6% versus 55.2%); but among the users of HRT, proportions of women showing estrogen receptor-positive epithelium was roughly equal, regardless of case-control status (85.7% versus 85.0%), as shown in Table 2. However, there was no significant interaction between current HRT use and estrogen receptor positivity in terms of breast cancer risk (P = .106).
Nulliparous interval was modeled as a continuous variable and was found to be a significant predictor of case status, with risk increasing by 6% for each additional year (OR = 1.06; 95% CI = 1.03–1.11). There was no significant interaction between nulliparous interval and estrogen receptor status of the benign epithelium. The fact that variations in nulliparous interval could result from oral contraceptive use in some women was not a point of analysis here.
Table 3 shows the characteristics of women with estrogen receptor-positive and -negative tissue. The mean age of estrogen receptor-positive women was higher than those who were estrogen receptor negative. On the whole, there was a trend toward postmenopausal women being estrogen receptor positive more frequently than premenopausal women (P <.07), with a significant positive correlation between age and estrogen receptor positivity in the entire population (r = .2; P <.001). The effect of age on estrogen receptor positivity was no longer significant when women were stratified by menopausal status. Premenopausal women who consumed alcohol regularly were more frequently estrogen receptor positive (8.6%) than women who denied alcohol consumption (4.5%). Among postmenopausal women, estrogen receptor positivity was more likely if they were nulliparous, had had a prior cancer other than that of the breast (uterus, ovaries, colon, and lymphoma), had a mother or sister with breast cancer, were using HRT, or had a greater time interval from menarche to either first full-term pregnancy or menopause. Body mass index was not significantly different, both between case subjects and control subjects and between estrogen receptor-positive and -negative women.
We have compared estrogen receptor alpha and progesterone receptor expression in normal, nonhyperplastic breast epithelium from breast cancer case subjects with that in control subjects with only benign disease. This model has the advantage of comparing preneoplastic epithelium from a high-risk group with a defined probability of developing new breast cancers (0.75% per year) (20) to similar epithelium from women whose future risk of breast cancer is close to that of the general population. We find that the odds of a woman with estrogen receptor-positive breast epithelium having a cancer of the breast are significantly elevated, with the crude OR being 3.16 (95% CI = 1.89–5.28) and the adjusted OR after controlling for known breast cancer risk factors (including age) being 2.63 (95% CI = 1.47–4.70). The effect of estrogen receptor positivity is stronger for postmenopausal than for premenopausal women, but it should be noted that estrogen receptor expression in premenopausal women varies with the menstrual cycle. Since our samples were not collected in a specific phase of the menstrual cycle, the random variation of surgical timing within the menstrual cycle may be diluting the effect of estrogen receptor positivity thus biasing our results toward the null in the premenopausal subset.
Progesterone receptor positivity was equally prevalent in the breast epithelium of case subjects and control subjects, with no significant differences in the proportion of positive cells, with crude ORs in both premenopausal and postmenopausal women being very close to one. This is in agreement with our previous finding that progesterone receptor positivity is uniformly prevalent in most women and is a constant feature of normal breast epithelium (13) [reviewed in (21) ].
The threshold for estrogen receptor and progesterone receptor positivity was prospectively chosen as 1%, since this level has been reported to correspond to the commonly used threshold of 10 fm mg cytosol protein in breast cancer samples (14). Other thresholds for estrogen receptor positivity between greater than zero and 2% did not alter the ORs appreciably; a threshold of 5% resulted in lower ORs (data not shown). However, analysis using three categories of estrogen receptor expression (labeling index <1, 1- portions of estrogen receptor-positive cells result in higher ORs for postmenopausal women (P trend .005) but not for premenopausal women (Table 1). This difference in trends may be related to the variability of estrogen levels in premenopausal women, and needs to be examined further.
Estrogen receptor positivity of breast epithelium does not appear to be modulated by the endocrine risk factors, such as early menarche, late first-term pregnancy and late menopause, and does not explain the high risk associated with them. We found no significant interactions between estrogen receptor positivity and these risk factors in our study population, which contained 76 women with early menarche, 31 with late menopause, 32 with late first-term pregnancy, and 74 with nulliparity. Since the breast epithelium is particularly susceptible to transforming events between menarche and the terminal differentiation associated with term pregnancy, we collapsed the endocrinerelated risk categories into one, defined above as the nulliparous interval (see“ Methods” section). The odds of having breast cancer increased by 6% for each additional year that the breast epithelium was exposed to ovarian cycles without a term pregnancy, but the duration of the nulliparous interval did not influence estrogen receptor positivity in either case subjects or control subjects, and we found no interaction between nulliparous interval, estrogen receptor positivity, and the occurrence of breast cancer. These results were unaffected by censoring premenopausal nulliparous women who still have not completed their reproductive life span.
The use of HRT by postmenopausal women resulted in a dramatic increase in the proportion of estrogen receptorpositive control subjects, so that there was no difference between estrogen receptorpositivity rates between case subjects and control subjects in this subset (Table 2). This finding needs to be pursued further in a larger study, with pre-and post-HRT assessments of estrogen receptor expression. If confirmed, it would suggest that HRT promotes estrogen receptor positivity in postmenopausal women and may contribute to breast cancer risk through this mechanism.
Estrogen receptor expression in benign breast epithelium has been described to decline late in the menstrual cycle. Our data are consistent with this finding in the control subjects, but in case subjects with breast cancer the reverse trend was seen, with median estrogen receptor labeling index increasing late in the cycle. Additionally, we observed a statistically significant trend for increasing odds of breast cancer in women whose breast epithelium was estrogen receptor positive late in the cycle (P <.015). These data suggest that dysregulated estrogen receptor expression, as reflected by estrogen receptor positivity in the luteal phase, carries a particularly strong association with the presence of breast cancer, and lead us to speculate that loss of the normal regulatory mechanisms that control expression of estrogen receptor in normal breast epithelium may confer an increased risk for the development of breast cancer. These findings also imply that future studies examining the relationship of estrogen receptor expression in benign breast epithelium and breast cancer risk must control for menstrual cycle dates in premenopausal women, and that estrogen receptor expression in the luteal phase deserves special attention.
In conclusion, further investigation of estrogen receptor expression in breast epithelium may identify points in the estrogen response pathway that may be interrupted to avoid the cancer-promoting effects of estrogen on breast epithelium. In that context, strategies that decrease estrogen receptor expression may prove protective against the carcinogenic effects of estrogen and other compounds collectively called xenoestrogens, which have affinity for estrogen receptors.
Supported by American Cancer Society grant 9333 and by the Department of Surgery, SUNY Health Science Center at Syracuse.
We acknowledge the assistance of Jane Dantoni and Patricia Brady in patient recruitment and risk factor data collection and Christine Jones for technical assistance in the processing of slides for immunohistochemistry.
Manuscript received May 28, 1997; revised August 20, 1997; accepted October 9, 1997.