The objective of this study was to evaluate the coexpression patterns of hormonal markers in breast cancer tissue and their relationship with pathologic characteristics and epidemiologic risk factors. We evaluated the expression of 17 markers by immunohistochemistry in 842 invasive breast carcinomas collected in a population-based case-control study conducted in Poland. Based on marker correlations, factor analysis identified four major coexpression patterns (factors): “nuclear receptor factor” [estrogen receptor (ER)-α, progesterone receptor, androgen receptor, cyclin D1, and aromatase], “estrogen metabolism/ER-β factor” (ER-β, peroxisome proliferator-activated receptor-γ, steroid sulfatase, estrogen sulfonotransferase, and cytochrome P450 1B1), “HER2 factor” (human epidermal growth factor receptor 2, E-cadherin, cyclooxygenase-2, aromatase, steroid sulfatase), and “proliferation factor” (cytokeratin 5, cytokeratin 5/6, epidermal growth factor receptor, P53). Three of these factors corresponded to molecular subtypes previously defined by expression profiling; however, the estrogen metabolism/ER-β factor seemed to be distinctive. High scores for this factor were associated with high tumor grade (P heterogeneity = 0.02), younger age at menarche (P heterogeneity = 0.04), lower current body mass index among premenopausal women (P heterogeneity = 0.01), and older age at menopause (P heterogeneity = 0.04). High scores for the proliferation factor were also associated with early menarche (P heterogeneity < 0.0001), and in contrast to the estrogen metabolism/ER-β factor, higher current body mass index among premenopausal women (P heterogeneity = 0.03). Our analysis of hormonal pathway markers independently confirmed several previously defined molecular subtypes identified by gene expression profiling and augmented these findings by suggesting the existence of additional relationships related to ER-β and enzymes involved in hormone metabolism. [Cancer Res 2007;67(21):10608–17]
- hormonal markers
- breast cancer
- factor analysis
- population-based study
- molecular subtype
Estrogen plays important roles in the pathogenesis and development of breast cancer. Increased cumulative estrogen exposure is hypothesized to mediate the effects of many breast cancer risk factors including early age at menarche, late age at menopause, nulliparity, postmenopausal obesity, and possibly other factors ( 1). Estrogen exposure may contribute to mammary carcinogenesis through several mechanisms including increased cell cycling, genotoxic effects, and induction of aneuploidy ( 2). If etiologic exposures act through different mechanisms, then the relationships between these exposures and breast cancer may vary by expression patterns of markers that reflect these underlying mechanistic differences.
Analysis of molecular markers in breast cancer tissues may reveal information about hormonal actions in the breast that are not shown by evaluating risk factors or serum hormone levels. Hormone levels were found to be higher in breast tumor tissue than blood ( 3), especially among postmenopausal women with low circulating hormone levels. Breast tissues express the enzyme aromatase, which is involved in estrogen synthesis, and steroid sulfatase (STS) and estrogen sulfonotransferase (EST), which, respectively, increase and decrease estrogen bioavailability ( 4). Expression of these enzymes may be higher in breast cancer than in normal breast tissues ( 5, 6). Administration of aromatase inhibitors has shown therapeutic value in treating estrogen receptor (ER)–positive breast cancer and has potential utility in chemoprevention ( 7). However, data linking the expression of enzymes involved in estrogen metabolism to tumor and patient characteristics are limited.
Once formed, estrogens may exert different actions by activating ER-α or ER-β. In contrast to the established importance of ER-α as a breast cancer marker, the prognostic and predictive relevance of ER-β remains unclear. In addition to its receptor-mediated actions, estradiol may serve as a substrate for cytochrome P450 1B1 (CYP1B1), leading to the formation of 4-hydroxyestradiol, which represents a putative genotoxin ( 8). Furthermore, hormone pathways are known to interact with other factors related to inflammation, DNA repair, cell cycle/apoptosis, and other mechanisms in complex ways ( Fig. 1 ).
Statistical analyses of global gene expression have led to the proposal of a novel breast cancer classification that identifies tumor types with distinctive biology and prognosis ( 9, 10); however, the functional pathways associated with the molecular subtypes are largely unknown. Understanding the relationships between hormonal pathways and breast cancers with particular pathologic features or risk factor associations may have value in identifying targeted prevention or treatment approaches. Accordingly, we did a focused analysis based on immunohistochemical expression of 17 candidate markers related to hormonal pathways for 842 invasive breast cancers collected in a population-based case-control study. Our goals were to assess marker coexpression patterns and relate these patterns to tumor characteristics, including previously defined molecular subtypes, and epidemiologic risk factors.
Materials and Methods
Study population. The study population has been previously described in detail ( 11). In brief, eligible subjects included women between the ages of 20 and 74 years who resided in Warsaw or Łódź, Poland, from 2000 to 2003. Breast cancer cases were defined as pathologically confirmed incident in situ or invasive breast carcinoma. Cases were identified through a rapid identification system organized at five participating hospitals (∼90% of cases) and via cancer registries to ensure complete case ascertainment. Control subjects were randomly selected using a population-based database, frequency-matched to cases on city and age in 5-year categories. A total of 2,386 cases (79% of eligible) and 2,502 controls (69% of eligible) agreed to participate in the study and provided informed consent as required by the National Cancer Institute and local Institutional Review Boards in Poland. We collected tumor tissue blocks from ∼70% of the participating cases. This report is based on data from all controls and a subset of 842 participating cases with invasive carcinomas whose blocks were available for preparation of tissue microarrays (TMA). Subject and tumor characteristics for the 842 cases were similar to the remaining invasive cases that were not included in the analyses.
Information on breast cancer risk factors was elicited through a personal interview. Women were considered premenopausal if still menstruating at the time of interview, postmenopausal if periods had stopped, and unclear menopausal status if menopausal hormone replacement therapy (HRT) was initiated before cessation of natural periods. Body mass index (BMI) was calculated using measured weight (kilograms) divided by standing height (meters) squared. For the ∼5% of subjects with missing measurements, BMI was calculated from self-reported information. Women who reported having had a benign breast biopsy 1 year prior to cancer diagnosis (cases) or interview (controls) were considered to have had a history of benign breast disease.
Pathology. Histopathologic features including histology, grade, tumor size, and axillary lymph node metastases were assessed using surgical pathology reports and independent evaluation by the study pathologist (M.E. Sherman). Routinely prepared formalin-fixed paraffin-embedded blocks of 842 invasive breast cancers were used to construct TMA blocks with 2-fold representation as 0.6-mm diameter cores (Beecher Instruments). Tissue sections of 5 μm thickness were placed on glass slides using a tape-transfer system (Instrumedics, Inc.) with UV cross-linking, dipped in paraffin, and stored at room temperature under nitrogen to prevent oxidation-related loss of immunoreactivity prior to staining.
We did immunohistochemical staining of TMA blocks for 17 markers including estrogen metabolism enzymes [aromatase, STS, EST, CYP1B1, cyclooxygenase-2 (COX-2)], nuclear receptors [ER-α, ER-β, progesterone receptor (PR), peroxisome proliferator-activated receptor-γ (PPAR-γ), and androgen receptor (AR)], growth factors (human epidermal growth factor receptor [EGFR] and HER2), tumor suppressor genes (p53, E-cadherin), ER-regulated cell cycle molecule cyclin D1, and basal markers [cytokeratins 5/6 (CK5/6) and CK5]. Antibodies and experimental procedures are described in Table 1 ( 12– 15). Staining was done with antigen retrieval prior to antibody incubation according to established protocols. Slides were digitized via whole slide scanning using an Aperio T2 instrument (Aperio Technologies) with a 20× objective, and the digital slides were reviewed by the study pathologist (M.E. Sherman) as well as by routine microscopic examination. Tumors without interpretable cores (3%) containing adequate viable tissue were omitted from the analysis. For tumors with two satisfactory cores (∼80%), results were averaged; for the remainder, results were based on a single interpretable core.
Percentage (0–100%) of tumor cells stained and staining intensity (0 = negative, 1 = weak, 2 = intermediate, and 3 = strong) were recorded for each marker. Aromatase was scored according to published methods (ref. 12; Table 1). Stains for ER-β, PPAR-γ, EST, STS, and CYP1B1 were diffuse (∼100%), therefore, only the intensity of staining for these markers was recorded. The product of percentage and intensity ranging from 0 to 300 was created to represent the overall score for each core (100% was used as the percentage for markers with diffuse staining when creating the overall score). We limited our analysis to 698 of the 842 tumors that had interpretable staining results for all 17 markers evaluated.
Statistical analysis. Correlations between markers were assessed by Pearson's correlation using the overall score (percentage × intensity) for each marker as an input variable. Multivariate analysis was complicated by the high correlation between markers. Therefore, we used factor analysis (PROC FACTOR, with the VARIMAX option, SAS 8.0) to construct independent factors that explained much of the correlation between these markers ( 16). Kaiser's rule of keeping those factors corresponding to eigenvalues of the correlation matrix values >1 was used to decide how many factors to use in the subsequent analysis ( 17). The factor loading, which is the standardized scoring coefficient obtained from the SAS FACTOR procedure, was used to determine the contribution of a marker to a particular factor. A higher loading value indicates a greater contribution of the marker to that factor, whereas a negative factor loading means that the marker is inversely related to that factor. Factors were rotated by an orthogonal transformation in order to obtain a simpler structure to make interpretation easier. Markers with rotated factor values >0.36 for a particular factor were considered significant contributors for that factor. For each subject, a factor score for each of the four factors defined in this data set was calculated by summing up the overall expression level (percentage × intensity) for each marker weighted by the factor loadings. A high factor score means the tumor expressed high levels of markers that were positively related to that factor and low levels of markers that were negatively related to that factor.
To compare the four factors identified in the current analysis to previously defined molecular subtypes, we first classified the 698 tumors into the five molecular subtypes based on immunohistochemical profiles as described previously ( 18, 19). We then used χ2 tests to compare the observed to expected frequency of tumors with low (Q1 and Q2) and high (Q3 and Q4) factor scores for each of the four factors in each of the five molecular subtypes. We used factor scores (in quartiles) for each of the four factors defined in this data set to compare marker expression patterns to tumor characteristics and to breast cancer risk factors using multivariable logistic regression. In case-only analyses, ordinal logistic regression was used to estimate heterogeneity P values to test for significant differences in distributions of tumor characteristics and risk factors by scores for each factor. Variables in the multivariable model for the analysis of tumor features included age, study site, histology, grade, size, and axillary lymph node metastases. For the analysis of risk factors, models included age, study site, education level, age at menarche, menopausal status, age at menopause, number of full-term births, age at first full-term birth, current/recent oral HRT use (combined estrogen and progesterone) among postmenopausal women, history of benign breast disease, prior mammographic screening, family history of breast cancer among first-degree relatives, and BMI. We also included an interaction term between BMI and menopausal status in the regression models to account for the effect modification by menopausal status. In these analyses, menopausal age for women who had undergone hysterectomy without oophorectomy prior to menopause was defined as age at surgery. Results for analyses restricted to women who had undergone natural menopause yielded similar results (data not shown). Alcohol consumption (ever/never and duration), oral contraceptive use (ever/never, not common in the Polish population), and smoking overall were not significantly associated with breast cancer risk and thus were omitted from the final model. Odds ratios (OR) and 95% confidence intervals (95% CI) of breast cancer risk were obtained from logistic regression models in which cases with the highest quartile of factor scores were compared with the population-based controls (N = 2,502) for the breast cancer risk factors. All statistical analyses were done using SAS (version 8.0, SAS Institute, Inc.) software.
Expression and coexpression patterns of markers. The distribution of the breast cancer risk factors for tumors included in the TMA analysis (n = 842) was similar to that for all other invasive tumors in the Polish study (n = 1,302; Supplementary Table S1). As would be expected, tissues for research purpose were not available for some small tumors, which consequently were underrepresented in the TMA blocks (7% of cases with tissue samples in the TMA blocks had tumors ≤1 cm compared with 19% of cases not included in the TMA blocks; P < 0.0001). Other tumor characteristics (histology, grade, nodal status, and hormone receptor status) were not significantly different after adjusting for tumor size (Supplementary Table S1). The current analysis was based on data obtained from 698 of the 842 tumors that had interpretable staining results for all 17 markers evaluated. Among the tumors we analyzed, 65% were ductal, 26% were poorly differentiated, 50% were >2 cm, and 43% had axillary node metastases. Based on a threshold of >10% of cells stained as a positive result, 68% of tumors would have been classified as ER-α–positive and 50% as PR-positive.
Marker expression is shown in Table 1 and coexpression patterns annotated by significance of associations are represented in Table 2 . Three markers related to estrogen synthesis and bioavailability (aromatase, STS and EST) were significantly correlated (P < 0.0001). Correlations with other markers for ER-α and ER-β showed different patterns. Both ER-α and ER-β expression were significantly correlated with aromatase, whereas only ER-α was coexpressed with cyclin-D1, PR and AR, and negatively correlated with HER2, CK5, EGFR, P53, and EST; only ER-β was coexpressed with STS, EST, PPAR-γ, and CYP1B1 (all comparisons, P < 0.0001). As expected, basal cytokeratins, P53 and EGFR were correlated. COX-2 was positively correlated with most markers studied, whereas the results for PR were most notable for negative correlations with expression of other markers (HER2, COX-2, CK5, EGFR, and P53; P < 0.0001). Analysis of the correlations between marker pairs for cases stratified by age, menopausal status, ER-α, or PR were generally similar.
Identification of four marker patterns by factor analysis. We identified four major expression patterns (factors) by factor analysis ( Fig. 2 ), which constructs independent factors based on correlations between these markers. The markers that contributed significantly to each factor were as follows: (a) “nuclear receptor” factor: ER-α, cyclin D1, AR, PR, and aromatase; (b) “estrogen metabolism/ER-β factor”: ER-β, PPAR-γ, CYP1B1, EST, and STS; (c) “HER2 factor”: HER2, COX-2, E-cadherin, aromatase, and STS; and (d) “proliferation factor”: CK5, CK5/6, EGFR, and P53. Aromatase and STS were the only markers that contributed significantly to two factors: aromatase was a significant contributor to the nuclear receptor and HER2 factors, whereas STS was significantly related to the estrogen metabolism/ER-β and HER2 factors. Analyses restricted to postmenopausal women identified similar groupings of significant markers, except that within the nuclear receptor factor, aromatase was not a significant contributor and HER2 was a significant negative contributor. Similarly, factor analysis stratified by tumor size resulted in similar classifications. Given that the markers included in the estrogen metabolism/ER-β factor (ER-β, PPAR-γ, EST, STS, and CYP1B1) stained diffusely (or were negative), our analysis for these markers was limited to intensity measurements. Accordingly, we repeated the factor analysis using four levels of marker intensity data (negative, weak, intermediate, and strong) for all 17 markers and obtained similar results (data not shown).
Expression patterns and molecular subtypes. We compared these four factors to the molecular subtypes defined by a five-marker immunohistochemical profile described previously, as adopted from Carey et al. ( 18, 19). Overall, three factors resembled four of the molecular subtypes (nuclear receptor factor similar to luminal A, HER2 factor to luminal B and HER2-expressing subtypes, and proliferation factor to basal-like subtype; all comparisons P < 0.0001; Fig. 3 ). All distributions were as expected, except for the high frequency of tumors observed with high-proliferation factor scores (Q3 and Q4) among luminal A tumors. The estrogen metabolism/ER-β factor was not associated with any particular subtype, although high factor scores for this factor were observed more frequently among ER-α negative subtypes (HER2 expressing, P = 0.04; basal-like, P = 0.003; and normal-like, P = 0.05). In general, estrogen metabolism enzymes seemed distinct from previously defined molecular subtypes identified via global gene expression profiling.
Expression patterns and tumor characteristics. Figure 4 shows the distribution of tumor characteristics by quartiles of each factor score. High factor scores (the fourth quartile) for the nuclear receptor factor were associated with well or moderately differentiated (P < 0.0001) and small tumors (P = 0.01), whereas high estrogen metabolism/ER-β factor scores were related to poorly differentiated carcinomas (P = 0.02). High HER2 factor scores were related to ductal histology (P < 0.0001) and both high HER2 and high proliferation factor scores were associated with poorly differentiated tumors (P < 0.0001, both comparisons). Axillary node status was not associated with any of the four factors, with or without adjustment for tumor size and grade.
Expression patterns and breast cancer risk factors. The distribution of breast cancer risk factors did not differ significantly by factor scores for the nuclear receptor factor and the HER2 factor (Supplementary Table S2). However, associations between exposures and factor scores for the estrogen metabolism/ER-β factor and the proliferation factor showed heterogeneity ( Tables 3 and 4 ).
In case-only analyses, we found that progressively higher estrogen metabolism/ER-β factor scores were associated with younger age at menarche (P heterogeneity = 0.04), lower current BMI among premenopausal women (P heterogeneity = 0.01), and older age at menopause (P heterogeneity = 0.04). Higher proliferation factor scores were also associated with younger age at menarche (P heterogeneity < 0.0001) and, in contrast to high estrogen metabolism/ER-β factor scores, higher BMI among premenopausal women (P heterogeneity = 0.03). Although based on small numbers (HRT use is not frequent in the Polish population), high proliferation factor scores were marginally associated with current or recent combined HRT use (P heterogeneity = 0.05). In stratified analyses by tumor size, number of full-term births and age at first full-term birth showed heterogeneity by estrogen metabolism/ER-β factor scores. Specifically, high estrogen metabolism/ER-β factor scores were significantly associated with being nulliparous (P heterogeneity = 0.008 for having one child versus no child and P heterogeneity = 0.002 for having multiple children versus no child) and late age at first full-term birth (P heterogeneity = 0.003) in small tumors (tumor size ≤2 cm). Risk associations with other exposures and other factors did not vary significantly by tumor size.
When comparing tumors with the highest quartile of factor scores for the estrogen metabolism/ER-β and proliferation factors to controls, we observed that the protective effect of older age at menarche was stronger in tumors with higher estrogen metabolism/ER-β factor scores (OR, 0.87; 95% CI, 0.77–0.98, per 2 years increase) and higher proliferation factor scores (OR, 0.81; 95% CI, 0.72–0.90), whereas the reduced risk associated with current high BMI among premenopausal women was only seen in tumors with higher estrogen metabolism/ER-β factor scores (OR, 0.64; 95% CI, 0.47–0.88, per 5 units of increase) but not in those with high proliferation factor scores (OR, 1.02; 95% CI, 0.78–1.33). High estrogen metabolism/ER-β factor scores were associated with increased breast cancer risk for older age at menopause (OR, 1.29; 95% CI, 1.03–1.62, per 5 years increase) and high proliferation factor scores were associated with increased risk for HRT use based on small numbers (OR, 2.50; 95% CI, 1.32–4.75).
Our analysis of the immunohistochemical expression of 17 selected hormonal markers in ∼700 invasive breast cancers showed four independent marker coexpression patterns (factors). In contrast to gene expression profiling strategies, our goal was to refine molecular subtypes based on immunohistochemical expression data for a biologically important pathway, hormone metabolism. We used factor analysis, which finds groupings or “factors” that reflect distinctive patterns of marker coexpression based on marker correlations.
We identified three factors that resembled cognate molecular subtypes defined by global analysis of gene expression ( 10): the nuclear receptor factor was related to luminal A, the HER2 factor to luminal B and HER2, and the proliferation factor to basal tumors. Distributions of factor scores for these three factors in most of the molecular subtypes were as expected. For example, ER-α–negative subtypes (HER2, basal, and normal-like) were associated with low nuclear factor scores (P < 0.0001); similarly, most luminal A tumors had low HER2 factor scores (P < 0.0001). Unexpectedly, we did not observe a negative association between high proliferation factor scores and luminal A tumors (P = 0.07). These results suggest that luminal A molecular subtype, defined as ER/PR-positive and HER2-negative, may itself represent a heterogeneous class. In contrast to these three factors, scores for the estrogen metabolism/ER-β factor (identified in the current analysis) showed a similar distribution across most of the previously defined molecular subtypes ( 10), suggesting that it represents a subtype distinct from those previously described. Similarly, the incompletely characterized “normal-like” molecular subtype included tumors with high scores for all factors, except the nuclear receptor factor, suggesting that this group may be heterogeneous with respect to hormonal pathways.
The coexpression of ER-β along with estrogen metabolism enzymes is consistent with a recent report showing that expression profiles for ER-α and ER-β in breast tumors differ markedly ( 20). The importance of high STS and CYP1B1 expression in the estrogen metabolism/ER-β factor prompts the hypothesis that increased availability of estrogen and its conversion to genotoxic metabolites may contribute to the development of this tumor subset. High PPAR-γ receptor levels also contributed to this factor, consistent with data indicating that PPAR-γ expression is correlated with ER-β levels and might modulate estrogen actions ( 21).
Our findings of coexpression of aromatase and ER-α in the nuclear factor are consistent with previous observations that high intramammary estradiol levels might favor the development of ER-α–positive breast cancers ( 22). Aromatase levels were also significant contributors to the HER2 factor. Levels of aromatase transcripts have been associated with both normal-like and HER2-expressing tumors, irrespective of ER-α expression ( 23). A recent clinical trial in which patients with ER+/HER2+ tumors were treated with an aromatase inhibitor and trastuzumab found that a subgroup of patients responded, suggesting that these tumors might be biologically heterogeneous; however, expression of estrogen metabolism enzymes were not measured ( 24).
Our analysis showed that aromatase, HER2 and COX-2 were correlated, consistent with previous findings ( 25, 26). Experimental animal models show that amplification of HER2 up-regulates COX-2, which in turn, activates the transcription of aromatase ( 27). However, elevated COX-2 and aromatase levels were also found in the nuclear receptor factor which is not associated with increased HER2 levels, so other mechanisms for up-regulating COX-2 may exist. Consistent with mechanistic studies, we found that COX-2 and EGFR were coexpressed (P < 0.0001). The observed high coexpression of COX-2, hormone synthesizing enzymes (aromatase, STS), and growth factors (HER2, EGFR) supports the recent proposal that COX-2 may be a potential molecular target for chemoprevention and treatment strategies.
Associations between the nuclear, HER2 and proliferation factors and pathologic features were as expected given their similarities with the luminal A, HER2, and basal-like molecular subtypes, respectively. The HER2 factor, which includes high E-cadherin expression, was associated with ductal histology as expected given that low E-cadherin levels are associated with lobular differentiation ( 28). Published results on the prognostic significance of ER-β levels are mixed, although several studies based on the assessment of protein levels suggested that these cancers are aggressive ( 29– 31).
Our analysis revealed differences in strength of relatedness of epidemiologic factors to the estrogen metabolism/ER-β and the proliferation factors, but not for the nuclear receptor and HER2 factors. Higher estrogen metabolism/ER-β factor scores were associated with younger age at menarche and older age at menopause, suggesting that prolonged menstrual span and increased cumulative estrogen exposure may be etiologically important for these tumors. This factor was also strongly associated with low current BMI among premenopausal women, a characteristic that has been related to higher estrogen levels and elevated breast cancer risk ( 32, 33). The associations of this factor with two other hormone-related risk factors (nulliparity and old age at first full-term birth) in small tumors further support a hormonal etiology for these tumors. Molecular changes in small tumors might show stronger associations with causal etiologic factors because alterations related to genetic instability or tumor progression (i.e., changes unrelated to tumor development per se) may be fewer. Younger age at menarche was also strongly related to tumors with higher proliferation factor scores. However, in contrast to the estrogen metabolism/ER-β factor, increased proliferation factor scores were associated with higher BMI among premenopausal women. These results are consistent with our previous findings for basal tumors ( 19). Basal tumors have been reported to occur more commonly among premenopausal African-American women compared with postmenopausal African-American and non–African-American women ( 18), but further examination of this association accounting for BMI are needed. The contrasting associations for current BMI among premenopausal women and the estrogen metabolism/ER-β and the proliferation factors, strongly suggest etiologic differences between these cancers.
The distinctiveness of the estrogen metabolism enzymes/ER-β factor may have clinical relevance, given that a substantial percentage of women with ER-α expressing tumors do not respond clinically to antiestrogenic agents. Specifically, clinical tests that comprehensively assess markers related to hormone synthesis, availability, or action might improve the selection of specific antiestrogenic agents for cancer treatment. In addition, developing risk factor profiles that can identify healthy women at increased risk for breast cancers that develop through particular hormonal pathways might help identify optimal chemoprevention. Future analyses are needed to examine treatment responses and outcomes for breast cancers characterized by their estrogen metabolism/ER-β factor scores.
The strengths of our study include the population-based sample, the rich collection of patient and tumor characteristics, the TMA design to achieve standardized marker measurement, and the comprehensive selection of hormonal markers. Our statistical approach, factor analysis, allowed us to construct linear combinations of markers, “factors”, which fully accounted for the correlations between markers and their levels of expression. Markers that belong to more than one factor (such as aromatase and STS) could be identified. In subsequent analyses, tumors expressing different levels of signature markers could be analyzed, which may provide further biological insights than that would not be revealed by cluster analysis. Nonetheless, our study had limited power to assess relationships with rare exposures. In addition, subjective assessment of immunostains might be imprecise, especially when based on TMA cores containing limited tumor tissue. However, ER and PR expression levels in this study population were also assessed by two other methods: (a) clinical reports mainly determined by immunostaining whole tissue sections and (b) a novel quantitative immunofluorescent method, automated quantitative analysis on TMA cores ( 34). High correlations were observed among all three methods, suggesting that subjective evaluation of TMA cores may be adequate. Our ability to reproduce several previously established associations for marker coexpression also reassured us of the quality of our measurements. Furthermore, we also confirmed the robustness of our factor analysis by showing that we could identify similar factors in analyses based on intensity only, using marker exclusion strategies, and analyses stratified by age, tumor size, and menopausal status.
In summary, this population-based analysis of breast cancer stratified by 17 hormonally related markers recapitulated three of the molecular subtypes previously identified by analysis of gene expression profiling and suggested the existence of a new subtype related to ER-β, STS, EST, and CYP1B1. Future studies of breast tissues to understand the pathogenesis of hormone-dependent and independent breast cancer may expand our understanding of the etiology and pathogenesis of this disease and facilitate the development of prevention and treatment targeted to pathways operative in carcinogenesis.
Grant support: National Cancer Institute and U.S. Army Medical and Research materials Command under grant DAMD17-03-1-0229 (T.R. Sutter).
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.
We thank Anita Soni and Elena Adrianza (Westat, Rockville, MD) for their work on study management; Pei Chao (IMS, Silver Spring, MD) for her work on data and sample management; Lori Charette (TMA Core Facility, Yale University, New Haven, CT) for her technical support on TMA construction; and the physicians, pathologists, and nurses from participating centers in Poland as well as interviewers and study participants for their efforts during field-work.
- Received June 11, 2007.
- Revision received August 12, 2007.
- Accepted September 5, 2007.
- ©2007 American Association for Cancer Research.