Opposing Effects of Runx2 and Estradiol on Breast Cancer Cell Proliferation: In Vitro Identification of Reciprocally Regulated Gene Signature Related to Clinical Letrozole Responsiveness

The mitogenic effect of estrogens in breast epithelial cells is one of the best examples of hormonal carcinogenesis. Which of the many primary and secondary estrogen-responsive genes in breast epithelial cells drive hormonal carcinogenesis is difficult to discern. Although genes such as c-Myc, cyclin D, and cyclin E have been implicated in estrogen-driven breast cancer (1), there is limited understanding of why some are better correlated than others with disease progression and drug responsiveness. Recent high-throughput gene expression studies have led to the identification of gene groups, whose baseline mRNA levels and hormone responsiveness correlated well with clinical and histologic features of disease progression (2–6). Interestingly, first-generation prognostic gene signatures were enriched for proliferation- and cell-cycle–related genes and were particularly valid for estrogen receptor alpha (ERα)-positive tumors (7, 8). Better understanding of the biological significance of such gene signatures will potentially lead to novel pharmacologic approaches targeting key molecular nodes that regulate breast cancer progression.

The Runx family of mammalian transcription factors plays fundamental roles in the differentiation of osteoblasts and chondrocytes (Runx2; ref. 9), hematopoietic cells (Runx1; ref. 10, 11) and neurons (Runx3; ref. 12). Runx proteins are also increasingly implicated in cancer progression, both positively and negatively (13, 14). Contrasting their prometastatic role, which was mostly studied in advanced breast and prostate cancer (14–20), Runx proteins are well known for their tumor suppressor properties. Runx3 is a bona fide tumor suppressor gene, whose methylation contributes to gastric cancer (21–23), ablation of Runx1 activity leads to leukemia (24, 25), and Runx2 inhibits cell-cycle progression in osteoblasts and prostate cancer cells (20, 26–28).

Runx2 physically interacts with the DNA-binding domains of both ERα and the androgen receptor. These interactions usually result in reciprocal inhibition of the respective transcriptional activation activities (29–31), although different outcomes have been reported (32–35). The physical interaction between ERα and Runx2 (29), and our more recent observation that E2 independently regulates half of the Runx2-responsive genes in breast cancer cells (see herein), led to the speculation that the oncogenic potential of estradiol in breast epithelial cells might be related in part to antagonizing Runx2. In this study, we first define gene sets coregulated by estrogen and Runx2 signaling in breast cancer cells in vitro. Then, by correlating expression pattern of these genes in breast cancer tumor biopsies with clinical response to aromatase inhibition therapy, we show that genes upregulated by E2 and downregulated by Runx2 are more informative than those defined in vitro based on their response to E2 alone.

MCF7 breast cancer cells were obtained from the American Type Culture Collection. To establish MCF7 cell subline that conditionally express Runx2, we employed the recently described lentivirus-based pSLIK vector system, which allows tight doxycycline (dox)-inducible, RNA PolII-mediated transcription of a gene of interest (36). Lentiviral particles encoding dox-inducible Flag-Runx2 (MASN isoform, type-2) and the Hygromycin B selection marker were constructed and packaged as previously described (20). Transduced MCF7/Rx2^dox cells were selected in Dulbecco's modified Eagle's medium (DMEM) containing 10% FBS and 100 μg/mL Hygromycin B (GIBCO).

Proteins were separated on 10% SDS-PAGE and Western blotting was performed using mouse monoclonal anti-FLAG M2 (Sigma), mouse monoclonal anti-Runx2 (Invitrogen), and goat anti-GAPDH (glyceraldehyde-3-phosphate dehydrogenase, V-18; Santa Cruz Biotechn., Inc.) as primary antibodies. After incubation with either horseradish peroxidase–conjugated goat anti-mouse secondary antibodies (sc2031) or donkey anti-goat antibodies (sc2020) from Santa Cruz Biotech., Inc., proteins were visualized with enhanced Chemiluminescence Plus Western blotting detection kit (GE Healthcare UK limited).

Runx2 and ERα were visualized with the respective primary antibodies and secondary antibodies conjugated to either rhodamine or fluorescein, respectively. Cells were viewed using an LSM 510 Zeiss confocal microscope at 60× magnification.

MCF7/Rx2^dox cells were suspended in 0.4% agarose prepared in DMEM containing 10% charcoal-stripped serum (CSS) and plated over 0.5% base agar at a density of 5,000 cells per well in 6-well plates. Cells were incubated at 37°C with media change every 2 days and stained with 0.005% Crystal violet after 21 days of incubation. Colonies were counted using a dissecting microscope.

Two days before treatment MCF7/Rx2^dox cells were switched to phenol red-free DMEM containing 5% CSS. Cells were treated with 0.5 μg/mL dox to induce Runx2 expression and/or with 10 nmol/L E2. After 48 hours of treatment, RNA was extracted using Aurum Total RNA Mini Kit (BioRad) and submitted to the Southern California Genotyping Consortium (SCGC) for microarray analysis using HumanRef-8 v3.0 Expression BeadChips (Illumina Inc.). Raw data processing was done using GenomeStudio (Illumina Inc.) and extended analyses were carried out using Partek Genomics Suite 6.6 (Partek, Inc.), or the R 2.11.1 package (http://cran.r-project.org), as indicated. Comparisons between gene expression under the various treatment conditions were done by one-way ANOVA and differentially expressed genes were defined based on 2 criteria: an FDR-adjusted P value less than 0.05 and a fold change of 1.3-fold or more. Hierarchical clustering was statistically assessed based on average silhouettes distance, average Pearson gamma, entropy, and within-between cluster ratio (see Supplementary Table S1). The web-based Database for Annotation, Visualization, and Integrated Discovery (DAVID) system (37) was used for functional annotation clustering of gene sets. The complete microarray dataset has been deposited in the Gene expression Omnibus (GEO) database with the accession number GSE30597.

Quantitative real-time PCR was carried out on an OPTICON2 (MJ Research) in triplicates using Maxima SYBR Green/Fluorescein Master Mix (Fermentas, Inc.). All transcript levels were normalized to that of GAPDH. Primer sequences used for PCR are listed in Supplementary Table S2.

Three independent Affymetrix HG-U133A microarray datasets profiling gene expression in untreated ER-positive and ER-negative breast cancer patients were downloaded from the NCBI GEO data repository. The first dataset (GSE2034) known as the Rotterdam cohort represented 180 lymph node–negative relapse-free patients and 106 lymph node–negative patients that developed a distant metastasis (3). The second dataset TRANSBIG (GSE7390) consisted of 186 lymph node–negative primary breast cancer patient samples (38) and the third cohort (GSE11121) involved 200 node-negative breast cancer patients treated at the Department of Obstetrics and Gynecology of the University of Mainz between 1988 and 1998 (39).

The open-label, multicenter phase II preoperative letrozole (POL) trial has been previously described (40). Eligible hormone receptor–positive patients were administered 2.5 mg oral letrozole daily for 16 to 24 weeks before surgery, with biopsies taken at baseline and after 4 weeks of treatment, with change in proliferation marker Ki67 being as one of the primary outcomes of response. Using RNA extracted from the biopsies, gene expression profiles were generated on Agilent 4 × 44K whole human genome platform on 80 patients at baseline, whereas 58 had paired 4-week on-treatment gene expression profiling. Probe-level gene expression data were collapsed to the gene level by median representative and standardized by median centering and robust scaling by interquartile range. All statistical analysis was carried out in R 2.11.1 (http://cran.r-project.org) and all tests were 2-sided at the 5% significance level unless otherwise noted. The detailed description of the POL data analyses and independent letrozole study GSE5462 downloaded from GEO (41) is provided in Supplementary Materials.

Luminal-like MCF-7 breast cancer cells were transduced with lentiviruses containing the dox-regulated Flag-tagged Runx2 cassette, described previously (20). Neither Western blotting nor immunofluorescence imaging detected endogenous Runx2 protein in untreated MCF7/Rx2^dox cells (Fig. 1A and B). Runx2 was expressed in response to dox (Fig. 1A and B) at levels that were comparable with the endogenous Runx2 in T47D breast cancer cells and much lower than those seen in osteoblasts (Fig. 1A). ERα was readily detected in both control and E2-treated cells and reorganized in subnuclear domains, together with Runx2, when the latter was expressed (Fig. 1B). The colocalization of ERα and Runx2 in MCF7/Rx2^dox cells is similar to that previously reported in COS7 cells and is attributable to physical interaction between the 2 proteins (29).

We next employed the MCF7/Rx2^dox cell culture model to investigate the individual and combined effects of E2 and Runx2 on anchorage-independent growth in vitro. Cells were suspended in soft agar and colonies were enumerated after 21 days of incubation in the presence of dox (to induce Runx2), E2, or both. As shown in Fig. 2, E2 increased by 3.3-fold and Runx2 decreased by 4-fold the number of colonies formed; when present together, Runx2 completely antagonized the ability of E2 to enhance growth of MCF7/Rx2^dox cells in soft agar.

To investigate the interaction between E2- and Runx2-regulated gene expression in breast cancer, we carried out genome-wide mRNA analysis of MCF7/Rx2^dox cells after treatment with vehicle (control; C), dox (D), estradiol (E), or both together (DE; see Materials and Methods and Fig. 1B). Thus, the Runx2 response in the absence of E2 is defined for each gene as the ratio between the D and the C values, and Runx2 response in the presence of E2 is defined as the ratio between the DE and the E values. Similarly, E2-response in the absence and presence of Runx2 is defined as the E versus C and the DE versus D ratios. The complete list of differentially regulated genes in each array comparison is provided in Supplementary Table S3.

A global view of the influence of E2 on Runx2-driven changes in gene expression revealed diverse outcomes. Most of the Runx2-stimulated genes were enhanced to a lesser extent or even inhibited by Runx2 in the presence of E2 (Fig. 3A and Supplementary Table S4). This is graphically shown in Fig. 3A by the larger number of data points below versus above the diagonal in the right half of the scatter plot. The fraction of genes where E2 attenuated Runx2-mediated stimulation progressively increased from 0.7 for genes stimulated by less than 1.5-fold to 0.9 for genes stimulated by 3-fold or more (Fig. 3A, bar graph and Supplementary Table S4). That E2 attenuated the Runx2-mediated stimulation of gene expression is equally demonstrable when counting only the statistically significant changes (Supplementary Table S4). Similarly, E2 attenuated the inhibitory effects of Runx2 on genes expression (Fig. 3A, left half, and Supplementary Table S4). Exceptions to the attenuation phenomenon, however, were numerous and are represented in Fig. 3A by data points above the diagonal for dox-stimulated genes and below the diagonal for dox-repressed genes. Similar analysis in the reciprocal orientation revealed that Runx2 modulated E2-mediated regulation of gene expression in a locus-specific manner, with attenuation again being the predominant mode of interaction (Fig. 3B). These results are consistent with our original observation of E2-mediated inhibition of Runx2 activity (29) but suggest 2 amendments. First, although attenuation of Runx2 activity by E2 was observed for the vast majority of genes particularly responding to Runx2, many exceptions were observed for genes that Runx2 stimulated or inhibited only mildly (Fig. 3A, bar graphs). Second, in the reciprocal orientation, Runx2 generally attenuated the responses to E2, in particular for E2-inhibited genes (Fig. 3B). In both orientations, the outcomes of the interactions were gene dependent.

To investigate the relationship between Runx2 and E2 signaling in human breast cancer tumors, we initially assessed the correlation between the respective responsive genes, as defined in our MCF7/Rx2^dox culture model, using publicly available gene expression data from breast cancer biopsies. Because of the significant overlap between Runx2- and E2-responsive genes in the MCF7/Rx2^dox model, we disregarded the “common” genes (Fig. 4A) to avoid bias and analyzed the correlation between genes that responded only to dox (D-only) or only to E2 (E-only; Fig. 4A). We defined a D-only^↑ gene set as 52 D-only genes that Runx2 stimulated by 2-fold or more and a similarly sized E-only^↑ gene set as 53 E-only genes that E2 stimulated by 2.5-fold or more in the MCF7/Rx2^dox model (Fig. 4A and Supplementary Table S5). For each breast cancer patient in the GSE11121 (39), GSE2034 (3), and GSE7390 public datasets (38), we determined the D-only^↑ and E-only^↑ metagenes as the average of standardized expressions of matching genes in the GEO datasets (Supplementary Table S5). We observed a significant negative correlation between the D-only^↑ and E-only^↑ metagenes in each of these 3 independent cohorts (Fig. 4B). The negative correlation likely reflects the E2-mediated attenuation of the Runx2 response (Fig. 3A) and/or the Runx2-mediated attenuation of the E2 response (Fig. 3B) revealed using the MCF7/Rx2^dox culture model.

Because cross-sectional analyses can only test correlation, we directly tested how loss of estrogen signaling influences Runx2-regulated genes by comparing gene expression in breast cancer biopsies obtained from patients in the POL clinical trial (40) at baseline versus 4 weeks into treatment with the aromatase inhibitor letrozole. We first confirmed that, similar to the negative correlation observed in the GEO datasets, there was a negative correlation between the D-only^↑ and E-only^↑ metagenes in the POL cohort at baseline (Fig. 4C). We also confirmed the expected decrease in expression of the E-only^↑ metagene in response to letrozole treatment (Fig. 4D). Most importantly, 4 weeks of letrozole therapy increased expression of the D-only^↑ metagene in most of the POL breast cancer patients (Fig. 4D), with P = 0.01 for the entire cohort (paired sample t test). The increase in expression of the D-only^↑ metagene after letrozole treatment is attributable to loss of the inhibitory effect of estrogens on Runx2-stimulated genes.

Given that E2 and Runx2 stimulated and inhibited, respectively, colony formation by MCF7/Rx2^dox cells in soft agar, even in the absence of the other stimulus (Fig. 2), we turned our attention to genes that were independently regulated by each of these opposing poles. We subjected the “common” genes, those regulated by E2 and Runx2 even in the absence of the other stimulus (Fig. 4A) to unsupervised clustering based on their expression levels in the absence and presence of dox (to induce Runx2) and/or E2. Of the 4 major branches (Fig. 5A), genes in clusters I and III were similarly regulated by Runx2 and E2, either down (cluster I) or up (cluster III). More interesting were clusters II and IV, comprised of genes inversely regulated by the 2 signals; up by Runx2 (dox) and down by E2 (cluster II) or vice versa (cluster IV). Cotreatment with both dox and E2 generally resulted in additive effects (Fig. 5A). The genes comprising each cluster, the matching genes in the POL and GEO datasets, as well as microarray-based standardized expression values for each of these genes in the MCF7/Rx2^dox model, are listed in Supplementary Table S5. Quantitative reverse transcriptase PCR (qRT-PCR) analysis of selected genes in 3 independent experiments confirmed the expected expression pattern of genes in cluster I (NPNT and MATN3), cluster II (S100A6 and RHOB), cluster III (CD44 and HEY2), and cluster IV (PLK4, CCNB2, CCNB1, and CDC2; Fig. 5B). These and additional qRT-PCR assays of a total of 20 genes confirmed reproducibility of the microarray data (90% validation rate) and also suggested similar gene regulation in T47D/shRx2^dox cells (42), in which endogenous Runx2 was silenced by dox treatment (Supplementary Table S7). Functional annotation analysis of clusters I–IV revealed significant enrichment of each of them for particular biological functions, but the only strong significant association was that between cluster IV and cell-cycle–related genes, with a P value of 1.9 × 10⁻⁶ (Fig. 5A). Thus, cluster IV may be enriched in genes responsible for the opposing effects of E2 and Runx2 signaling on breast cancer cell growth (Fig. 2).

We speculated that genes in cluster IV, which were stimulated by E2 and inhibited by Runx2 in the MCF7/Rx2^dox model, represent a critical growth regulatory node, and that the sensitivity of these genes to letrozole would therefore correlate with clinical outcome better than genes selected based on their stimulation by E2 alone. We approached this conjuncture by comparing cluster IV to other metagenes with regard to their letrozole response and the correlation with clinical outcome in the POL study.

We initially redefined metagenes I–IV, consisting of the respective genes in clusters I–IV from the Illumina BeadChip analysis of the MCF7/Rx2^dox cells (Fig. 5A), which were also represented in the Agilent Microarrays employed in the POL study (Supplementary Table S6). Figure 6A describes arrow plots connecting expression levels at baseline and after 4 weeks of letrozole treatment for each metagene in each tumor. As expected, expression of the clusters III and cluster IV metagenes, representing E2-stimulated genes in vitro, was decreased in response to letrozole in most patients, whereas expression of the clusters I and cluster II metagenes, representing E2-inhibited genes in vitro (Fig. 5A) was generally stimulated in response to letrozole (Fig. 6A). Remarkably, however, the letrozole response was much stronger for the clusters II and IV metagenes (Fig. 6A). The more significant response of cluster II and IV genes (P values of 5 × 10⁻¹⁰ and 3 × 10⁻¹⁰, respectively) as compared with cluster I and III metagenes (P values of 0.1 and 0.0002, respectively) is attributable to the unleashed activity of Runx2 upon loss of the counteracting activity of E2.

Because cluster IV was highly enriched in cell-cycle–related genes (Fig. 5A), we asked whether the letrozole response of the corresponding metagene in individual patients correlated with the percentage of Ki67-positive cells in tumor sections from blocks obtained during surgery 16 to 24 weeks after commencement of letrozole treatment. Posttreatment Ki67 proliferative index is a well-documented surrogate marker for clinical outcome (43, 44). First, patients were dichotomized into a “High” and a “Low” group (N = 29 each) based on the median change in the expression of the cluster IV metagene during the initial 4 weeks of letrozole treatment. The same patients were also dichotomized in a similar fashion based on the response of clusters I, II, and III. We then compared the “High” and “Low” groups per each metagene in terms of the Ki67 index measured at the time of surgery months later. Remarkably, the letrozole response of the cluster IV metagene significantly correlated with the Ki67 index at surgery (Wilcoxon rank sum test, P = 0.0043), whereas no significant correlation was found with the letrozole response of clusters I–III (Fig. 6B). Furthermore, the change in the expression of the cluster IV metagene associated with the Ki67 index at surgery better than any of 1,000 random metagenes of similar size sampled from the E-only stimulated genes in MCF7/Rx2^dox cells (Fig. 4A, Fig. 6C). Even the most upregulated E-only^↑ metagene returned a less significant P value of 0.01. Thus, among all E2-responsive metagenes tested, the letrozole response of cluster IV during the initial 4 weeks of treatment best correlated with the Ki67 index in tumor biopsies taken after 16 to 24 weeks of treatment.

We further tested the correlation between the initial letrozole response in the expression of cluster IV genes and tumor phenotype in an independent study (41, 45). Changes in tumor gene expression before versus after 2 weeks of letrozole were first determined from the corresponding GSE5462 trial dataset (41) in a similar fashion to that described above for the POL study. Because this study did not include Ki67 staining, the patients with high versus low cluster IV response were compared in terms of the efficacy of letrozole in reducing tumor volume (45). As shown in Fig. 6D, the change in the cluster IV metagene expression over the initial 2 weeks of treatment significantly discriminated between clinically responsive and resistant patients assessed after 3 months of letrozole treatment (P = 0.0004). The changes in the expression of cluster I, cluster II, cluster III, the E-only^↑, and the D-only^↑ metagenes were either not or more than 7-fold less discriminative of tumor size reduction as compared with the cluster IV metagene (Supplementary Table S8).

The nuclear hormone receptor ERα and Runx family transcription factors play important roles in breast cancer, but little is known about their functional interaction. This work highlights Runx2 as an antagonist of the mitogenic property of E2 signaling in breast cancer tumors. Runx2 diminished E2-mediated colony formation by MCF7/Rx2^dox cells in soft agar (Fig. 2). Runx2 and ERα colocalized in distinct subnuclear domains of MCF7/Rx2^dox cells (Fig. 1) and genome-wide expression profiling of cells treated with dox (to induce Runx2) and/or E2 showed that Runx2 generally attenuated effects of E2 on gene expression (Fig. 3). Consistent with these in vitro observations, letrozole treatment of breast cancer patients not only decreased expression of genes upregulated by E2 in our in vitro system but also mimicked the response of genes to Runx2 (Fig. 4D). Finally, we identified a group of 84 genes (cluster IV), many with known roles in cell-cycle control, which were stimulated by E2 and inhibited by Runx2 in the MCF7/Rx2^dox culture model and whose short-term inhibition by letrozole in breast cancer tumors strongly correlated with clinical parameters assessed months later—posttherapy Ki67 proliferative index in our POL clinical study and change in tumor volume in an independent clinical study (Fig. 6). The opposing effects of Runx2 and E2 on expression of cluster IV genes and the targeting of this genomic node could therefore provide the basis for novel therapeutic approaches against hormone receptor–positive breast cancer.

Cluster IV genes are extreme examples of antagonism between E2 and Runx2 signaling because they are independently upregulated by E2 and downregulated by Runx2 (Fig. 5). Additional mechanisms of antagonism between estrogen and Runx2 signaling do not require independent regulation by the 2 pathways. For example, both the microarray data (Supplementary Table S3) and the immunofluorescence results (Fig. 1) suggest that Runx2 inhibits ERα expression. Runx2 may also destabilize ERα similar to Runx3 (23). Yet another mechanism could be the physical interaction between ERα and Runx2, which may result not only in decreased association of Runx2 with its target genes (29) but also decreased association of ERα with its targets. In support of this idea, reanalysis of data from 2 recent studies (46, 47) indicated that genomic ERα target genes are more likely to be antagonized by Runx2 than nongenomic targets (Supplementary Table S9). Although the relative contribution of each of these mechanisms is under investigation, the global reciprocal inhibitory effects between estrogen and Runx2 signaling observed in vitro (Fig. 3) and the inverse correlation between E2- and Runx2-stimulated genes observed in breast cancer tumors (Fig. 4B and C) provide a novel insight into breast cancer carcinogenesis. In particular, being independently stimulated by E2 and inhibited by Runx2 (Fig. 5), as well as being enriched for cell-cycle regulatory genes (Fig. 5A), cluster IV seems to represent a critical regulatory node where the mitogenic property of E2 meets the antimitogenic property of Runx2. It is tempting to speculate that the direct opposition between E2 and Runx2 signaling at this node (Supplementary Fig. S1) renders cluster IV a sensor for the status of interaction between E2 and Runx2 signaling. Tumors with the highest tension between the 2 poles will respond to loss of E2 signaling with the sharpest decline in the expression of cluster IV genes. In turn, unopposed Runx activity after letrozole treatment of these tumors will most effectively suppress their growth (Fig. 6). Unlike the response of cluster IV to letrozole treatment, no significant correlation was observed between clinical outcome and the baseline expression levels of cluster IV, Runx2, Runx2-resposive genes, or E2-responsive genes (data not shown).

Cluster IV is enriched in genes associated with the mitotic polo-like kinase pathway, such as PLK4, CDC2, CCNB2, KIF11, PRC1, and PTTG1 (Supplementary Table S6 and Supplementary Fig. S1), which regulates multiple mitotic events, including centrosome maturation, bipolar spindle formation, DNA damage adaptation, mitotic entry, activation of anaphase promoting complex, and cytokinesis, processes that must be well balanced to ensure normal mitotic progression and cell division (48). Polo-like kinase 4 (PLK4) in particular plays a crucial role in centriole duplication, cytokinesis, and maintenance of chromosomal stability, and deregulation of PLK4 has been implicated in carcinogenesis (48). Previous reports showed that estrogens stimulate and letrozole inhibits genes regulating mitotic spindle formation and cell proliferation (41, 49). Therefore, inhibition of these genes by Runx2 and, possibly, other members of the Runx family (14, 18, 23) may constitute an important mechanism of tumor suppression.

In summary, by manipulating both estrogen and Runx2 signaling in vitro, we showed diverse interactions between the pathways. Most commonly, E2 attenuated responses to Runx2 and vice versa, although many exceptions to this general rule were observed, the significance of which remains to be investigated. Another type of interaction occurs when E2 and Runx2 each regulate the same genes in the absence of the other stimulus. Of particular interest is the E2-stimulated and Runx2-inhibited PLK pathway-enriched cluster IV, whose 2- to 4-week response to initial letrozole treatment in 2 independent clinical studies correlated well with the Ki67 index or tumor volume reduction measured after 3 to 6 months of treatment. This correlation contrasts with the weaker or lack of association between other E2-regulated gene signatures determined here in vitro (e.g., E-only^↑, cluster I–III) or by Miller and colleagues based on clinical short-term response to letrozole alone (45). Future studies are warranted to investigate other roles of Runx proteins in breast cancer hormonal carcinogenesis, including the roles of Runx1 and Runx3 (14, 18, 23), as well as the interactions of Runx proteins with signals elicited by selective estrogen receptor modulators. Finally, while harnessing the tumor suppressor functions of Runx proteins, future translational research will have to take into account that these proteins can also display oncogenic properties (14, 18–20), much like estrogens can also display tumor suppressor properties (8, 50).

No potential conflicts of interest were disclosed.

The authors thank Dr. Alice E. Kohn-Gabet and Ms. Yunfan Shi for their expert technical assistance, Dr. Joseph DeYoung at the Southern California Genotyping Consortium (SCGC) for the microarray hybridization, and Dr. Yibu Chen and Ms. Meng Li at the USC Keck School of Medicine Bioinformatics Service Group for help with data analysis.

The work received financial support from NIH grants RO1 DK071122 and R01 CA109147 to B. Frenkel and R01 CA095614 to M.J. Ellis. B. Frenkel holds the J. Harold and Edna L. LaBriola Chair in Genetic Orthopedic Research at USC, M.J. Ellis holds the Anheuser Busch Chair in Medical Oncology, and D. Tripathy holds the Priscilla and Art Ulene Chair in Women's Cancer.

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.