Obesity, hormone therapy, estrogen metabolism and risk of postmenopausal breast cancer

Authors


Abstract

Hormone therapy (HT) and body mass index (BMI) have been associated with postmenopausal breast cancer. Because estrogen metabolism may affect breast cancer risk and can be altered by weight and HT, it might play a role in the HT–BMI–breast cancer associations. We undertook a nested case-control study within the Observational Study of the Women's Health Initiative. Baseline levels of 2- and 16α-hydroxy estrone (2-OHE1 and 16α-OHE1) were measured in 200 women who developed breast cancer during follow-up and 200 healthy controls matched to cases by ethnicity, enrollment date, clinic site, type of HT and years since menopause. Wilcoxon nonparametric tests were used to compare estrogen metabolite levels between cases and controls. Conditional logistic regression was used to assess the relationship between BMI, estrogen metabolites and breast cancer risk. 16α-OHE1 levels were modestly but significantly higher in HT users among cases (median 356 pg/ml vs. 315 pg/ml) and controls (354 pg/ml vs. 298 pg/ml). 2-OHE1 levels were substantially and significantly higher in HT users among cases (369 pg/ml vs. 125 pg/ml) and controls (347 pg/ml vs. 134 pg/ml). For non-HT users only, greater BMI and higher 16α-OHE1 levels were individually and jointly associated with increased breast cancer risk (OR for women with high BMI and high 16α-OHE1 compared to those with low BMI and low 16α-OHE1 = 3.51, 95% CI = 1.34–9.16). No associations between BMI, estrogen metabolism and breast cancer risk were found for HT users. Estrogen metabolism differs according to both BMI and HT use, potentially explaining the interaction between BMI and HT in relation to breast cancer risk. © 2005 Wiley-Liss, Inc.

The Women's Health Initiative (WHI) clinical trial recently confirmed observational study findings1, 2 that progestin-containing hormone therapy (HT) formulations are associated with an increased risk of postmenopausal breast cancer.3 Elevated circulating estrogen levels may underlie this association, as substantial data, including markers of long-term estrogen exposure,4 implicate estrogen in the etiology of postmenopausal breast cancer. However, for a given dose of estrogen, estrogen–only HT formulations appear to impart less risk than combined estrogen and progestin therapies.1, 3, 5, 6, 7, 8 In particular, the WHI clinical trial reported no increase in risk associated with an estrogen-only formulation,8 but a 24% increase in risk among women assigned to a combination estrogen–progestin formulation compared to placebo.3 Interestingly, this increase is less than expected, given the magnitude of the increase in circulating estrogens resulting from oral estrogen preparations9, 10 and among studies that associate higher blood levels of estrogen with increased breast cancer risk.11, 12

Body size has also been implicated in postmenopausal breast cancer.13, 14 For example, risk increases by 3% for each kilogram per meter square increase in body mass index (BMI).15 Again, circulating elevated estrogens levels, produced by the aromatization of androgens in adipose tissue, are believed to underlie this association, although other mechanisms, including effects of growth factors and inflammation, have been proposed.16 Interestingly, the breast cancer–BMI association appears limited to women not using HT,13, 17, 18, 19, 20 and the mechanism mediating the BMI–HT interaction remains unknown.

The fact that both HT use and high BMI elevate circulating estrogen levels suggests that estrogens might mediate the observations described above. In the liver, estradiol, the estrogen most strongly associated with breast cancer,21 is first (reversibly) converted to estrone, which is irreversibly converted to either 2- or 16α-hydroxy estrone in target cells.22 Because the 2-OHE1 and 16α-OHE1 metabolic pathways compete for a limited substrate pool, an increase in metabolism of estrone via one pathway will reduce the amount of product in the competing pathway.22 Both metabolites have estrogenic properties, although to varying degrees. 16α-OHE1 retains potent hormonal activities by binding strongly to the estrogen receptor.22 Laboratory data23, 24, 25 and some but not all epidemiologic studies26, 27, 28, 29, 30, 31, 32, 33, 34 suggest a positive association between 16α-OHE1 levels and breast cancer risk. In contrast, 2-OHE1 binds to the estrogen receptor with reduced affinity35 and because of O-methylation is cleared rapidly from circulation.36 Data suggest that this metabolite does not increase breast cancer risk and, in fact, may impart some protection.26, 27, 28, 37, 38

While genetic factors may play a role in determining whether the 2-OHE1 or the 16α-OHE1 pathway predominates,39 body weight is also important. In thin women, the 2-OHE1 pathway predominates whereas in overweight women, the 16α-OHE1 pathway predominates.40 Evidence also suggests that exogenous agents, including both pharmaceuticals and dietary factors, may alter the way by which estrogens are metabolized.41, 42 Of relevance here, HT regimens containing a progestin may cause a greater shift to the 16α-OHE1 pathway than estrogen-only regimens.43 The observations that both weight and type of HT influences estrogen metabolism led us to investigate whether estrogen metabolism may explain, in part, the less-than-expected increase in breast cancer risk associated with HT, the greater risk imparted by progestin-containing HT formulations compared to estrogen-only formulations and the interaction between BMI and HT in relation to breast cancer risk.

Methods

This nested case-control study was carried out within the WHI, a large, long-term, multi-center, multi-component study of major causes of morbidity and mortality in postmenopausal women. The design, recruitment, eligibility criteria and data collection for the WHI have been described elsewhere.44, 45, 46 Briefly, over 161,000 postmenopausal women were recruited from among 40 clinical centers across the United States between 1993 and 1998 to participate in at least 1 of the 3 WHI clinical trial components or an Observational Study (OS). Women were eligible for the WHI, in general, if they were postmenopausal, between the ages of 50 and 79 years at baseline, had no medical condition that was predictive of less than 3 years survival and likely to reside in the study area for at least 3 years. Women who were not eligible for or willing to be randomized to the clinical trial were invited to participate in the OS, a cohort study focused on investigating risk factors and biological markers of chronic disease, including cancer, heart disease and osteoporosis-related fractures. A total of 93,676 women enrolled in the OS to be followed for an average of 7 years by annual mailed self report questionnaires (medical history and exposure updates) and an additional clinical center visit for physical measurements 3 years after enrollment follow-up.47

Study participants

A total of 200 centrally-adjudicated incident, invasive breast cancer cases and 200 individually case-matched controls were randomly selected for the present study from among all WHI OS participants. Only those breast cancer cases with completed estrogen and progesterone receptor assay data were included. Potential cases were excluded if they met any of the following exclusion criteria: any self-reported history of cancer at baseline, breast cancer diagnosis less than 5 years past menopause, any locally adjudicated incident cancers other than breast cancer during follow-up; use of specific medications that might affect estrogen levels (i.e., antimycobacterial, imidazole-related anti-fungals, androgen-anabolic steroids, adrenal steroid inhibitors, hormone receptor modulators, histamine H2 receptor antagonists, cyclosporine analogs and herbal estrogens), self-report of unknown race or ethnicity or inadequate baseline serum available for analysis. In addition to these criteria, potential controls were excluded if they had any locally adjudicated incident breast cancer. Potential cases and controls were also restricted if their HT use did not fit into 1 of 3 categories: non-HT users (no self-reported use of hormones by interview or medications inventory at baseline or at the Year 3 visit), current estrogen (ERT) users (baseline self-report of daily conjugated equine estrogens [CEE] for at least 1 year and no non-CEE HT use in the past year and no progesterone use within 4 years of baseline or at the Year 3 visit); or current estrogen plus progestin (PERT) users (baseline self-report of daily CEE plus continuous or cyclic progesterone use for at least 1 year, no non-CEE HT use in the past year). HT users were excluded if they were currently using any estrogen or progesterone creams, shots or implants. Non-HT users were excluded if they ever had estrogen or progesterone use of any kind, including shots, pills, patches, creams and implants.

Based on the above criteria, out of the original 93,676 WHI–OS participants, a total of 497 potential breast cancer cases and 28,735 controls were identified, as of August 31, 2002. Controls were matched to cases by enrollment date, number of years from age at menopause to study entry, ethnicity, randomization clinic and type of HT used (none, estrogen-only, or estrogen–progestin combination). Ethnicity, randomization clinic and type of HRT were matched exactly, and the remaining continuous matching variables were selected based on criteria to minimize an overall distance measure.48 Matching was done in a time forward manner to ensure that each control had at least as much control time as its matched case. For example, a case diagnosed with breast cancer 2 years after enrollment would be matched with a control with at least 2 years of follow-up. SAS code is available to implement this matching scheme.48 Out of the 497 cases meeting study inclusion criteria, 468 cases were successfully matched with controls (29 unmatchable cases). A sample of 100 CEE current user pairs and 100 HT non-user pairs were selected for the study population via simple random sampling,49 giving a total of 200 case-control pairs.

Demographic and risk factor data

Data collected at baseline included self-reported sociodemographic information; medical, reproductive and family history; hormone use, current medications and supplements, health behaviors; dietary intake and psychosocial and behavioral characteristics. Physical measurements were also obtained at baseline and included height, weight, waist and hip circumference, blood pressure and resting pulse.

Fasting blood specimens were collected from each participant at baseline and processed locally according to a standardized, timed protocol into separate aliquots containing serum, plasma and buffy coat. These aliquots were frozen and then shipped frozen to a central repository, where they were stored long-term at −70°C.

Breast cancer ascertainment and adjudication

Potential breast cancer cases were identified during follow-up based on annual self-report medical history update questionnaires. Participants who reported a new diagnosis of cancer, whether as an inpatient or outpatient, were asked for additional provider information and release of information authorization to obtain medical records. A standardized document set for breast cancer adjudication, including hospitalization, diagnostic, pathology and operative or other procedure reports, was assembled and then reviewed by a trained, local physician adjudicator at the clinical center. All breast cancer adjudication case packets were sent to the Clinical Coordinating Center at Fred Hutchinson Cancer Research Center for central adjudication by trained cancer coders, according to Surveillance, Epidemiology and End Results (SEER) coding guidelines and under the supervision of a cancer epidemiologist and physician, with consultation as needed by a cancer pathologist with the local SEER registry.50

Measurement of serum levels of 2-OHE1 and 16α-OHE1

Serum levels of 2-OHE1 and 16α-OHE1 were measured by Immuna Care Corporation (Blue Bell, PA), using the ESTRAMET™ Serum 2 and 16 enzyme immunoassay kits (ELISAs). Details of these assays have been previously reported.51 Briefly, the serum assays were developed directly from reagents and buffers used to measure conjugated metabolites in urine.52 The assays for urinary estrogen metabolites have been validated against gas-chromatography-mass spectroscopy methods.52 Serum assays were validated against the urine assays by establishing recovery with dilution of known amount of urinary metabolites added to serum samples and were shown to have a sensitivity (i.e., lower limit of detection) of less than 20 and 10 pg/ml, respectively, for 2-OHE1 and 16α-OHE1. For the assays used in this study, the coefficient of variations (CVs) for within-assay duplicates for positive control sera were less than 5% for both 2-OHE1 and 16α-OHE1. The between-assay CVs for 3 positive laboratory control sera run on each plate were 15.5, 16.0 and 14.4% for 2-OHE1 and 5.3, 5.9 and 4.0% for the 16α-OHE1.

All samples were sent to the lab in a single batch, with case and control specimens blinded and randomly mixed together. All samples were run in duplicate on the same plate, with the resulting analyte values averaged to assign the final metabolite level to a sample. For each sample, the ratio of the 2-OHE1 to 16α-OHE1 metabolites was calculated by dividing the value of the 2-OHE1 metabolite by the value of the 16α-OHE1 metabolite. 16α-OHE1 and 2-OHE1 levels were not detectable in 11 women (1 control and 10 cases; 7 non-HT users and 4 HT users).

To evaluate the reproducibility of the assays, 19 test specimens were split, re-labeled with a new ID and randomly mixed with the other samples. CVs of duplicate measurements of these blinded, split test samples were 25.7 and 10.0% for the 2-OHE1 and 16α-OHE1, respectively. All specimens were labeled with ID numbers, so that only lab personnel were masked to the identity and case/control status of the subjects.

Statistical analyses

All analyses were conducted stratified by HT use. Differences in baseline characteristics between case and control subjects were assessed using Wilcoxon rank sum tests for continuous variables and χ2 tests for categorical variables. For BMI and estrogen metabolite measurements, tertile or quartile cut points were determined based on distributions in control subjects. Wilcoxon tests were used to compare differences in metabolite levels between cases and controls. Because of the matched design, conditional logistic regression was used to assess the association between metabolite levels, BMI and risk of breast cancer while controlling for potential confounders. Potential confounders were identified by considering factors known to affect breast cancer risk (e.g., age, parity, age at menarche) or estrogen metabolism (e.g., smoking, physical activity). Included in the final model were age and those variables that altered univariate estimates by >10%.53 For non-HT users, covariates included in the final models were age (continuous), education (high school or less vs. more than high school), marital status (married vs. not married), smoking history never vs. former/current), and alcohol use at study entry (>1 drink per day vs. <1 drink per day/no use). For HT users, covariates included were age (continuous), smoking history (never vs. former/current), and surgical menopause (yes vs. no). The same approach was used to assess metabolite–BMI interactions. Moreover, to examine the possible effects of high BMI and metabolite levels on breast cancer risk, according to HT use, we categorized women according to lower BMI (tertiles 1 and 2) and lower metabolite levels (quartiles 1 and 2) and assessed the joint effects of BMI and metabolite levels, using conditional logistic regression as described earlier. Measures of effect are reported as adjusted odds ratios (ORs) with 95% confidence intervals (95% CIs).

Results

Breast cancer cases were predominantly early stage, and positive for both estrogen and progesterone receptors (Table I). Among cases, HT users were younger and had slightly more regional disease and more progesterone receptor positive tumors than non-HT users. Sociodemographic, reproductive and health behavior characteristics of the cases and controls stratified by HT status are shown in Table II. In general, cases and controls were similar with respect to these characteristics, except that cases not on HT were less educated and heavier than controls not on HT. Regardless of case/control status, women on HT tended to be younger and more highly educated than women not on HT. They also tended to be leaner and have less central adiposity, as measured by waist-to-hip ratio (WHR). Among HT users, 60% used a combined estrogen–progestin formulation, and most women had been on HT for more than 10 years. There were no differences between cases and controls for factors that served as WHI Clinical Trial exclusion criteria, such as history of myocardial infarction or coronary revascularization (CHD event), venous thromboemoblism, diabetes or hypertension or use of cholesterol lowering drugs. There were also no differences between HT users and nonusers for these factors, except that cases not using HT reported a greater history of hypertension (44% vs. 23%) as well as treatment for the condition (32% vs. 14%) than did cases using HT.

Table I. Distribution of Clinical Factors Among Breast Cancer Cases According to Hormone Therapy Use
Clinical factorHT usersNon-HT users (n = 100)All cases (n = 200)
ERT users (n = 40)PERT users (n = 60)All users (n = 100)
  • 1

    Values in parentheses indicate percentage.

  • 2

    Values in parentheses indicate (mean ± SD).

Stage1 (n)
 Localized32 (80.0)49 (81.7)81 (81.0)84 (84.9)165 (82.9)
 Regional8 (20.0)11 (18.3)19 (19.0)14 (14.1)33 (16.6)
 Distant0 (0.0)0 (0.0)0 (0.0)1 (1.0)1 (0.5)
Estrogen receptor status1 (n)
 Positive34 (85.0)57 (95.0)91 (91.0)90 (90.0)181 (90.5)
 Negative5 (12.5)3 (5.0)8 (8.0)10 (10.0)18 (9.0)
 Borderline1 (2.5)0 (0.0)1 (1.0)0 (0.0)1 (0.5)
Progesterone receptor status1 (n)
 Positive29 (72.5)49 (81.7)78 (78.0)74 (74.0)152 (76.0)
 Negative10 (25.0)11 (18.3)21 (21.0)24 (24.0)45 (22.5)
 Borderline1 (2.5)0 (0.0)1 (1.0)2 (2.0)3 (1.5)
Age at diagnosis2 (years)65.0 (6.6)64.2 (6.5)64.5 (6.5)69.9 (6.5)67.2 (7.0)
Time from study entry to diagnosis2 (months)34.8 (18.9)32.0 (19.8)33.1 (19.4)33.9 (19.4)33.5 (19.4)
Table II. Baseline Characteristics of Breast Cancer Cases and Controls According to Hormone Therapy Use1
Baseline characteristicHT usersHT nonusers
No. of cases2No. of controls3p-valueNo. of cases2No. of controls3p-value
  • 1

    Comparisons between cases and controls assessed by χ2-square tests or Fisher's exact tests when cell sizes were <5.

  • 2

    n = 100.

  • 3

    n = 100.

  • 4

    Values in parentheses indicate percentage.

  • 5

    No statistical test performed, since variable was used to match cases with controls.

  • 6

    For subjects with menarche at age 9 or less, values were truncated at 9; for subjects with menarche at age 17 or older, values were truncated at 17.

Age at study entry (years)  0.76  0.74
 55 or younger18 (18.0)422 (22.0) 5 (5.0)5 (5.0) 
 56–6550 (50.0)49 (49.0) 28 (28.0)33 (33.0) 
 66 or older32 (32.0)29 (29.0) 67 (67.0)62 (62.0) 
Race/ethnicity  N/A5  N/A5
 Caucasian92 (92.0)92 (92.0) 90 (90.0)90 (90.0) 
 African–American1 (1.0)1 (1.0) 10 (10.0)10 (10.0) 
 Hispanic5 (5.0)5 (5.0) 0 (0.0)0 (0.0) 
 Asian/Pacific Islander2 (2.0)2 (2.0) 0 (0.0)0 (0.0) 
Education  0.37  0.04
 Less than high school2 (2.0)0 (0.0) 2 (2.0)6 (6.1) 
 High school/G.E.D.14 (14.0)14 (14.2) 28 (28.3)15 (15.1) 
 More than high school84 (84.0)85 (85.9) 69 (69.7)78 (78.8) 
Marital status  0.66  0.78
 Married/partnered74 (74.0)66 (66.0) 50 (50.4)52 (52.5) 
 Separated/divorced13 (13.0)16 (16.0) 15 (15.2)19 (19.2) 
 Widowed10 (10.0)14 (14.0) 27 (27.3)22 (22.2) 
 Never married3 (3.0)4 (4.0) 7 (7.1)6 (6.1) 
Body mass index  0.96  0.05
 Lower tertile (≤23.77 kg/m2)36 (36.0)35 (35.0) 19 (19.2)31 (31.0) 
 Middle tertile (≤27.37 kg/m2)34 (34.0)36 (36.0) 27 (27.3)32 (32.0) 
 Upper tertile (>27.37 kg/m2)30 (30.0)29 (29.0) 53 (53.5)37 (37.0) 
Waist-to-hip ratio  0.25  0.81
 Lower tertile (≤0.772)49 (49.0)38 (38.0) 23 (23.0)27 (27.0) 
 Middle tertile (≤0.826)31 (31.0)41 (41.0) 28 (28.0)27 (27.0) 
 Upper tertile (>0.826)20 (20.0)21 (21.0) 49 (49.0)46 (46.0) 
Age at menarche6  0.19  0.59
 12 or younger50 (50.0)45 (45.0) 54 (54.0)48 (48.0) 
 13–1443 (43.0)40 (40.0) 34 (34.0)41 (41.0) 
 15 or older7 (7.0)15 (15.0) 12 (12.0)11 (11.0) 
Parity  0.22  0.64
 Nulliparous12 (12.0)12 (12.0) 16 (16.0)15 (15.0) 
 1–231 (31.0)44 (44.0) 27 (27.0)30 (30.0) 
 3–446 (46.0)33 (33.0) 40 (40.0)44 (44.0) 
 5 or more11 (11.0)11 (11.0) 17 (17.0)11 (11.0) 
Age at 1st pregnancy lasting ≥6 months  0.49  0.68
 Never pregnant/full term15 (15.0)19 (19.2) 25 (25.0)25 (25.2) 
 Less than 2013 (13.0)8 (8.1) 6 (6.0)10 (10.1) 
 20–2967 (67.0)64 (64.6) 58 (58.0)56 (56.6) 
 30 or older5 (5.0)8 (8.1) 11 (11.0)8 (8.1) 
Age at menopause (years)  0.31  0.67
 Less than 4518 (18.0)29 (29.0) 9 (9.0)10 (10.0) 
 45–4932 (32.0)26 (26.0) 25 (25.0)29 (29.0) 
 50–5437 (37.0)32 (32.0) 48 (48.0)49 (49.0) 
 55 or older13 (13.0)13 (13.0) 18 (18.0)12 (12.0) 
Type of menopause  0.14  0.52
 Natural55 (55.0)62 (62.0) 76 (76.0)81 (81.0) 
 Surgical7 (7.0)12 (12.0) 3 (3.0)4 (4.0) 
 Undetermined38 (38.0)26 (26.0) 21 (21.0)15 (15.0) 
Hysterectomy  0.89  0.60
 No59 (59.0)61 (61.0) 78 (78.0)82 (18.0) 
 Yes41(41.0)39 (39.0) 22 (22.0)18 (18.0) 
Type of current HT use  N/A5  N/A
 Estrogen alone40 (40.0)40 (40.0) 0 (0.0%)0 (0.0%) 
 Estrogen + Progesterone60 (60.0)60 (60.0) 0 (0.0%)0 (0.0%) 
Years HT use (ever users)  0.63   
 Less than 515 (15.0)20 (20.0)  
 5 to less than 1026 (26.0)26 (26.0)  
 10 or more59 (59.0)54 (54.0)  
Alcohol intake  0.69  0.28
 Never/past drinker23 (23.0)26 (26.0) 30 (30.0)22 (22.2) 
 <1 drink per week29 (29.0)32 (32.0) 28 (28.0)37 (37.4) 
 ≥1 drink per week48 (48.0)42 (42.0) 42 (42.0)40 (40.4) 
Smoking status  0.07  0.92
 Never42 (42.9)55 (56.1) 52 (52.0)50 (50.5) 
 Past49 (50.0)41 (41.8) 41 (41.0)43 (43.4) 
 Current7 (7.1)2 (2.0) 7 (7.0)6 (6.1) 
CHD Event  1.00  1.00
 No98 (98.0)98 (99.0) 94 (96.1)94 (96.1) 
 Yes2 (2.0)1 (1.0) 3 (3.9)3 (3.9) 
Stroke  1.00  1.00
 No99 (99.0)98 (98.0) 97 (97.0)97 (97.0) 
 Yes1 (1.0)2 (2.0) 3 (3.0)3 (3.0) 
Venous Thromboembolism  0.37  0.44
 No96 (96.0)99 (99.0) 90 (90.0)94 (94.0) 
 Yes4 (4.0)1 (1.0) 10 (10.0)6 (6.0) 
Hypertension  0.32  0.11
 No76 (76.8)68 (68.0) 55 (56.1)69 (69.0) 
 Yes, untreated9 (9.1)10 (10.0) 11 (11.2)11 (11.0) 
 Yes, treated14 (14.1)22 (22.0) 32 (32.7)20 (20.0) 
Diabetes  0.28  0.37
 No98 (98.0)94 (94.0) 96 (96.0)92 (92.0) 
 Yes2 (2.0)6 (6.0) 4 (4.0)8 (8.0) 
Cholesterol Lower Drug Use  1.00  0.68
 No86 (86.8)85 (87.8) 82 (83.7)79 (81.4) 
 Yes12 (13.2)13 (12.2) 16 (16.3)18 (18.6) 

HT use was associated with a clinically modest, yet statistically significant increase in serum 16α-OHE1 levels in both breast cancer cases (median, 356 pg/ml vs. 315 pg/ml; p < 0.0001) and controls (median, 354 pg/ml vs. 298 pg/ml; p < 0.0001). Among HT users, 16α-OHE1 levels appeared slightly greater among users of a progestin-containing formulation than an estrogen formulation, although these differences were not statistically significant (Fig. 1).

Figure 1.

Distribution of 16-hydroxy metabolite levels by HT use and BMI among breast cancer cases (top half) and control subjects (bottom half). The rectangles depict the interquartile range and the lower and upper ends of the vertical lines represent the 10th and 90th percentiles. The number in parentheses represent median metabolite level (pg/mL). HT: hormone therapy. ERT: Estrogen only therapy. PERT: Estrogen + Progesterone therapy. For Cases, the p-value for the BMI × HT use interaction was 0.27. For Controls, the p-value for the BMI × HT use interaction was 0.37.

In contrast to the modest increase in serum 16α-OHE1 levels associated with HT use, substantial increases in serum 2-OHE1 levels were observed with the use of HT (Fig. 2). Specifically, median serum 2-OHE1 levels among HT cases (369 pg/ml) and controls (347 pg/ml) were nearly 3-fold higher than those observed in non-HT users (125 and 134 pg/ml, respectively, p < 0.0001 for both comparisons). These substantial differences persisted (p < 0.0001) after separate adjustment for BMI and WHR, both of which were univariately associated (inversely) with serum 2-OHE1 levels (data not shown). In addition, the ratio of 2:16α-OHE1 among controls using any type of HT was 0.99 compared to 0.46 for those not on HT (p < 0.0001, Fig. 3), further supporting the substantial increase in 2-OHE1 levels among HT users vs. nonusers. Consistent with results for serum 16α-OHE1 levels, serum 2-OHE1 levels appeared somewhat greater for progestin-containing formulations than estrogen-only formulations (Fig. 2), and the 2:16α-OHE1 ratio also appeared slightly higher among PERT users than ERT users (Fig. 3), but the differences were not significant.

Figure 2.

Distribution of 2-Hydroxy metabolite levels by hormone therapy use and body mass index (BMI) among breast cancer cases (top half) and control subjects (bottom half). The rectangles depict the interquartile range, the lower and upper ends of the vertical lines represent the 10th and 90th percentiles. The numbers in parentheses represent median metabolite level. HT: hormone therapy. ERT: Estrogen only therapy. PERT: Estrogen + Progesterone therapy. For Cases, the p-value for the BMI × HT use interaction was 0.42. For Controls, the p-value for the BMI × HT use interaction was 0.81.

Figure 3.

Distribution of the ratio of 2-Hydroxy:16-Hydroxy metabolite levels by hormone therapy use among breast cancer cases (left side) and control subjects (right side). The rectangles depict the interquartile range, the lower and upper ends of the vertical lines represent the 10th and 90th percentiles. The numbers in parentheses represent median metabolite level. HT: hormone therapy. ERT: Estrogen only therapy. PERT: Estrogen + Progesterone therapy.

There was no relationship between either of the estrogen metabolites of interest and breast cancer risk among women on HT (Table III). Among non-HT users, 16α-OHE1 but not the 2-OHE1 was associated with an increased risk of breast cancer (adjusted OR for 16α-OHE1: 1.21, 1.64 and 2.47 for quartiles 2, 3 and 4 compared to quartile 1, respectively; p for trend, 0.06). Consistent with other studies, higher BMI was associated with increased risk of breast cancer in women not on HT (adjusted OR = 3.27, 95% CI = 1.40–8.40, highest tertile compared to lowest), but not in women using HT (adjusted OR = 1.47, 95% CI = 0.67–3.22). Despite the observation that 2-OHE1 levels were not associated with risk in women not on HT, non-HT users with low 2-OHE1 levels (quartiles 1 and 2) and high BMI (tertile 3) exhibited an increase in breast cancer risk (adjusted OR = 3.67, 95% CI = 1.26–10.65) compared to those with high 2-OHE1 and lower BMI. Not surprisingly, women not on HT but with high BMI (tertile 3) and high levels of 16α-OHE1 (quartiles 3 and 4) were at a significant increase in risk compared to those with lower BMI and low 16α-OHE1 levels (adjusted OR = 3.51, 95% CI = 1.34–9.16). No metabolite–BMI interactions were evident among HT users.

Table III. Matched (Conditional) Odds Ratios for Breast Cancer in Relation to Anthropometric Measures and Serum Metabolites According to HT Use
Anthropometric measure/serum metaboliteHT usersHT nonusers
No. of casesNo. of ctrlsCrude ORAdj. OR195% CINo. of casesNo. of ctrlsCrude ORAdj. OR295% CI
  • 1

    Adjusted for age (continuous), smoking history (never vs. former/current) and surgical menopause (yes vs. no).

  • 2

    Adjusted for age (continuous), education (high school or less vs. more than high school), marital status (married vs. not married), smoking history never vs. former/current) and alcohol use at study entry (≥1 drink per day vs. <1 drink per day/no use). Missing values are as follows: for HT users — 4 subjects missing metabolite data and 4 subjects missing smoking history data; for HT nonusers — 7 subjects missing metabolite data and 10 subjects missing alcohol use, smoking history, education or marital status data (some subjects were missing more than 1 data item).

2 Hydroxy  (n = 192)(n = 184)   (n = 186)(n = 178) 
 Quartile 1262411292211
 Quartile 219240.750.450.18–1.1224250.740.840.37–1.89
 Quartile 327241.10.890.36–2.2015250.460.470.18–1.17
 Quartile 424240.970.920.38–2.2225210.890.800.32–1.98
 p for trend   0.94    0.44 
16 Hydroxy  (n = 192)(n = 184)   (n = 186)(n = 178) 
 Quartile 1212311152211
 Quartile 226261.070.980.39–2.4620251.21.210.47–3.11
 Quartile 329231.391.760.70–4.4528231.911.640.65–4.16
 Quartile 420240.911.020.40–2.5730232.112.470.90–6.80
 p for trend   0.68    0.06 
Body mass index (kg/m2)  (n = 200)(n = 192)   (n = 198)(n = 188) 
 Tertile 1363511193111
 Tertile 234360.900.950.43–2.1027321.451.330.55–3.31
 Tertile 330291.011.470.67–3.2253362.663.271.40–8.40
 p for trend   0.36    0.006 
BMI and 2 hydroxy  (n = 192)(n = 184)   (n = 184)(n = 176) 
 BMI tertile 1–2; 2 hydroxy quartiles 3–4403411162811
 BMI tertile 1–2; 2 hydroxy quartiles 1–226330.570.490.18–1.2228291.751.700.64–4.54
 BMI tertile 3; 2 hydroxy quartiles 3–411140.660.790.28–2.1624172.432.891.10–7.60
 BMI tertile 3; 2 hydroxy quartiles 1–219151.151.530.55–4.6024182.633.671.26–10.65
BMI and 16 hydroxy  (n = 192)(n = 184)   (n = 184)(n = 176) 
 BMI tertile 1–2; 16 hydroxy quartiles 1–2363211213211
 BMI tertile 1–2; 16 hydroxy quartiles 3–430350.730.910.40–2.0223251.381.730.68–4.42
 BMI tertile 3; 16 hydroxy quartiles 1–211170.520.740.25–2.2214151.362.820.80–9.98
 BMI tertile 3; 16 hydroxy quartiles 3–419121.502.220.87–6.2034202.853.511.34–9.16

Discussion

Data from the present study confirm previous findings that breast cancer risk in postmenopausal women is associated with relatively high BMI.13, 14 We further confirmed that these associations are limited to women not using HT.13 Several possible explanations for these two observations have been posited, including mechanisms involving estrogens, androgens, insulin and growth factors.54, 55 For example, in postmenopausal women not on HT, peripheral conversion of androgens to estrogens by aromatase in adipose tissue is the main source of circulating estrogens.56 Hence, heavier postmenopausal women have higher circulating estrogen levels than thinner women,57 thereby, potentially increasing their risk of breast cancer.21 In contrast, increases in estrogen levels by HT are independent of a woman's weight,58 and exogenous estrogens increase circulating levels to such a degree that endogenous production by adipose tissue may become inconsequential.

Data from the present study suggest another potential mechanism mediating the obesity–breast cancer association and the observed interaction with HT: the pathway by which estrogen is metabolized differs according to both BMI and HT use. In general, we found that among women not on HT, breast cancer risk increased with increasing levels of the 16α-OHE1 metabolite. Levels of the 2-OHE1 metabolite did not appear to alter breast cancer risk. We further found that among women not using HT, 16α-OHE1 levels were higher in heavier women. These relationships were not evident among women on HT. Together, these observations may provide a potential mechanistic explanation for the increase in breast cancer risk observed among heavier postmenopausal women not using HT but not among heavier women on HT.13, 14

Our results may also explain the modest increase in breast cancer risk associated with HT use in general and the difference in risk associated with the type of HT (opposed vs. unopposed). If higher estrogen levels underlie breast cancer risk, then one would anticipate that the high levels of estrogens induced by HT use would dramatically increase breast cancer risk. To the contrary, most studies have observed only a modest increase in risk with long-term use and this increased risk is greater for progestin-containing formulations.1, 3, 5, 6, 7, 8 Our data suggest that a possible explanation for those observations is that, independent of the type of HT (opposed or unopposed), the estrogens in HT are metabolized mostly to the benign 2-OHE1 metabolite rather than to the risk-bearing 16α-OHE1 metabolite. Unfortunately, our study was not designed to assess the differences in estrogen metabolism between HT formulations. However, because the WHI findings noting a difference in breast cancer risk between ERT and PERT formulations were reported after our study was begun, we used our limited data set to determine if estrogen metabolism might explain the ERT/PERT differences. As shown in Figures 1 and 2, our data suggest that opposed formulations may produce a greater amount of 16α-OHE1, despite the fact that they contain the same amount of exogenous estrogen as unopposed formulations. This difference in metabolite production may explain, in part, the reason that progestin-containing HT formulations are associated with a greater risk of breast cancer than estrogen-alone. This hypothesis, however, must be examined in a larger data set, because our sample size for examining the ERT/PERT differences was small (a total of 100 HT users) and underpowered to find any significant difference of relevance.

Most but not all case-control studies investigating the role of estrogen metabolism in postmenopausal breast cancer risk support the hypothesis that greater 2-OHE1 levels protect against breast cancer while greater 16α-OHE1 may increase risk.26, 27, 28, 29, 30, 31 Differences in these findings may be an artifact of study design, including small sample sizes (usually less than 50 cases), noncomparable controls and the use of spot urine samples to measure metabolite levels, which may be sensitive to changes in excretory patterns and function. Moreover, the presence of the tumor or an endocrine response to treatment, both of which may be present in women recruited after diagnosis, may alter metabolite levels.59 In addition, changes in diet, exercise or body composition, as a result of treatment,60, 61 all of which can alter estrogen metabolite levels,40, 41 may further affect metabolite measurements in a case-control setting. These methodologic limitations may explain the contradictory findings in case-control studies published to date.

A prospective design, as was employed in the current study, can overcome methodologic issues due to treatment effects and, potentially, tumor presence. Three other prospective studies have examined the estrogen metabolism–breast cancer association. In the Guernsey III study, a higher 2:16 ratio was associated with a lower risk of breast cancer in postmenopausal women.32 However, in both the ORDET33 and SOF34 breast cancer studies, no association between estrogen metabolite levels at baseline and subsequent breast cancer development was found. While the ORDET study was limited in size (only 71 cases) and did not control for HT use, the SOF study was relatively large (272 cases) and limited analyses to non-HT users only. However, SOF is unique in that the participants were all older white women over 70 years of age who were, on average, more than 20 years postmenopause. Moreover, previous analyses in these women have not found associations between breast cancer development and well-established breast cancer risk factors.62 Hence, it is possible that the unique characteristics of the women participating in SOF may explain the differences in their findings and the ones reported here.

Although our findings are intriguing, we are careful to note that the relationship between metabolite levels and subsequent breast cancer development are prospective, while the data comparing HT use and non-use are cross-sectional. Therefore, we cannot draw any conclusions about the underlying mechanisms mediating the obesity–HT–breast cancer association; rather our data are suggestive of a possible explanation and point to the need to evaluate the role of estrogen metabolism in breast cancer risk, within a controlled clinical trial of hormone therapy. Several other limitations of this study warrant discussion. First, our sample size was relatively small, therefore, limiting our power to detect modest associations and precluding our ability to conduct analyses by HT formulation (ERT vs. PERT, Table III). We further limited HT users to oral formulations containing a CEE. Hence, the findings do not apply to HT containing other types of estrogens or those employing other modes of administration (e.g., patches, shots, implants or creams). Furthermore, because more than 90% of the subjects were white, these findings cannot be generalized to other racial and ethnic groups. As well, although this study was nested within a prospective cohort, average follow-up between baseline and breast cancer diagnosis in cases was less than 3 years; hence, it is possible that observations were influenced by disease status. In addition, serum measurements of estrogen metabolites may not necessarily reflect tissue-specific exposures. We further note that the coefficient of variation was 25.7 for the 2-OHE1 metabolite, quite likely reflecting the low levels of the metabolite in women not on HT. Although this CV was somewhat high, the large difference in 2-OHE1 metabolite levels between HT users and nonusers suggests that the variability was not likely to affect our results. As in any study, we cannot exclude the possibility that some of our findings are due to chance, especially given the large number of comparisons.

There are several issues that must be mentioned regarding the WHI OS cohort that might have influenced the study findings and limited its generalizability. Women participating in the WHI OS comprised those who were excluded from the clinical trial or who did not wish to be randomized. While the OS participants are less different from the US population than participants of other recent studies of postmenopausal women,63, 64 in general, OS women are better educated than same-aged women in the US.47 However, although OS participants may have been expected to have a higher prevalence of health conditions or behaviors used as clinical trial exclusion criteria, such as diabetes, hypertension or cardiovascular disease, in practice, they appear to be healthier than the general population of women aged 50–70.47 For example, the prevalence of coronary disease and stroke at baseline was only about half of that expected using NHANES-III estimates, and current smoking was about 30% of what was expected.65 These differences quite likely reflect a “healthy volunteer” effect. In the subset of OS participants included in the current analyses, the distribution of these factors was similar to that of the overall WHI OS cohort,47 suggesting that, like the overall cohort, the women in this study were healthier than the general population from which they were recruited. We did find a difference in reported history of hypertension between cases not using HT and cases using HT. Although hypertension does not appear to be associated with breast cancer risk,66 questions remain about the association between antihypertensive medications and breast cancer risk.67, 68, 69 If these medications alter estrogen metabolism, it is possible that the use of such medications could have influenced our reported results. However, we found no case-control differences in reported history of hypertension as well as in history of cardiovascular disease or diabetes, and so it is unlikely that these factors substantially influenced our findings. Because of our small sample size and the relative rarity of these conditions in our sample, we could not stratify our analyses according to these factors to further explore their impact on our findings and on its generalizability. Hence, our findings may not apply to women with cardiovascular disease, diabetes or hypertension. Finally, because most of the cases taking HT had used the therapy for more than 10 years, it is possible that these women were less sensitive to the effects of HT. However, the fact that the distribution, by years, of HT use did not differ between cases and controls suggests that if this were the true, it quite likely did not substantially impact our findings.

In conclusion, the results of this study support the observation that metabolism of estrogen via the 16α-hydroxylation pathway is associated with increased breast cancer risk; however, this finding is limited to women not using HT. Given that BMI and HT both influence how estrogen is metabolized, our findings may provide some insight into potential mechanisms mediating the associations among BMI, HT and breast cancer risk.

Disclosures

T. Klug is the president of ImmunaCare Corporation, the laboratory that performed the assays. All specimens were sent to Dr. Klug's lab in a masked fashion so that he and all lab personnel were unaware of the case control status of the subject. All data generated by the lab was sent to the WHI Coordinating Center where it was merged with the risk factor and outcome data before being analyzed by Dr. Kip. Dr. Klug was not involved in data analysis or interpretation. Beyond performing the laboratory assays and helping to draft the laboratory methods section of this manuscript, Dr. Klug had no other role in this project.

Acknowledgements

We thank Chandra Marriott and Dr. Jeffrey L. Eppinger for their help with this manuscript.

Ancillary