Genome-Wide Identification of Direct Target Genes Implicates Estrogen-Related Receptor α as a Determinant of Breast Cancer Heterogeneity

Breast tumor development and progression are complex processes that involve the contribution of numerous signaling pathways, depicting the high degree of heterogeneity of the disease (1). Estrogen receptor α (ERα) represents an important prognostic and therapeutic factor in breast cancer (2). In addition, ERBB2 amplification occurs in 15% to 20% of breast tumors and strongly associates with poor prognosis (3). Other primary tumors express both ERα and ERBB2, whereas the basal-like subtype expresses neither signaling proteins. Therefore, the identification of factors contributing to breast cancer initiation or progression and possessing the potential to modulate both ERα and ERBB2 signaling pathways represents an interesting avenue for the development of more effective and all-encompassing therapies.

Estrogen-related receptor α (ERRα; NR3B1) is an orphan member of the superfamily of nuclear receptors (4). Recent molecular and genetic analyses have shown that ERRα primary function is to regulate bioenergetic pathways required for cell- and tissue-specific functions (5). In particular, ERRα acts in concert with the coregulators peroxisome proliferator-activated receptor-γ coactivator (PGC) 1α and β to control the expression of genes involved in tricarboxylic acid cycle, oxidative metabolism, and mitochondrial biogenesis in tissues with high-energy demands. ERRα also possesses many molecular attributes suggesting that it could play a role in the etiology of breast cancer. First, ERRα shares both structural and functional features with ERα (6). In particular, ERRα and ERα can bind to each other's cognate hormone response element in vitro and regulate reporter constructs harboring either type of binding sites in transfected cells. Second, the transcriptional activity of ERRα can be modulated by the epidermal growth factor/ERBB2 signaling pathway (7, 8). Third, the expression of ERRα in breast tumors correlates with negative prognosis and positively and inversely correlates with ERBB2 and ERα status, respectively (9, 10). Taken together, these findings suggest that ERRα could contribute to the biology of several subtypes of breast cancer.

Mapping of binding sites on a genome-wide scale using chromatin immunoprecipitation–based approaches has allowed the unbiased identification of large gene networks under the direct control of nuclear receptors (11). In particular, the ERα cistrome has provided insights into functional interactions between the receptor and other transcription factors in estrogen-dependent tumors (12–15). Likewise, analysis of global gene expression profiles of breast tumor samples, and more recently of the surrounding stroma, has led to the definition of gene expression signatures that can classify distinct subtypes of breast cancer and/or predict clinical outcome and response to therapies (16–18). However, these expression signatures reflect the overall transcriptional milieu of the cells, thereby limiting the development of therapies aimed at a precise molecular target. Here, we describe a study in which we interconnected genome-wide identification of direct target genes of ERRα and gene expression data sets. Our results identify ERRα as a determinant of breast cancer heterogeneity.

Cell culture, reagents, and antibodies. MCF-7 and SKBr3 cells were cultured as described previously (19). SKBr3 cells were treated with XCT-790 (Sigma) for 4 or 6 h before harvesting mRNA. ERRα chromatin immunoprecipitation assays were done using an anti-human ERRα polyclonal antibody raised in our laboratory as described previously (20). Other antibodies used were anti-RNA polymerase II (PolII; 8WG16) and anti-ERRα (Millipore) and anti–amplified in breast cancer 1 (AIB1), anti-PGC-1β, anti-ERα, and anti-SRC-1 (Santa Cruz Biotechnology). Small interfering RNAs (siRNA) for ERα, ERRα, and control (ON-Target Plus siRNA pool) were obtained from Dharmacon.

Chromatin immunoprecipitation assays, sequential chromatin immunoprecipitation, and ChIP-on-chip on extended promoter arrays. Chromatin immunoprecipitation was done as described previously (19). Quantification of chromatin immunoprecipitation enrichment by quantitative real-time PCR (Q-RT-PCR) were carried out using the LightCycler 480 instrument (Roche). For sequential chromatin immunoprecipitation, enriched chromatin from the initial chromatin immunoprecipitation was eluted twice at room temperature, diluted in chromatin immunoprecipitation dilution buffer, and re-immunoprecipitated with the second antibody. For ChIP-on-chip analyses on Agilent extended promoter arrays, chromatin was prepared from MCF-7 cells cultured in phenol red–free medium supplemented with hormone-deprived serum (for ERRα ChIP-on-chip) or treated with 10⁻⁷ mol/L estradiol (E₂) before harvesting (for ERα ChIP-on-chip) or from SKBr3 cells (for ERRα ChIP-on-chip). The chromatin immunoprecipitation primers are listed in Supplementary Table S1.

Computational motif discovery. De novo motif discovery was done with MDScan.⁶

siRNA. siRNAs pool (Dharmacon) for ERα and control were transfected in MCF-7 cells cultured in regular DMEM supplemented with complete serum using the HyperFect reagent (Qiagen). Similarly, siRNAs pool (Dharmacon) for ERRα and control were transfected in SKBr3 cells using the HyperFect reagent (Qiagen) according to the long-term protocol provided. Forty-eight hours before harvesting for Q-RT-PCR, the cells were cotransfected with 300 ng ERRspPGC-1 (generous gift from Dr. McDonnell, Duke University) using FuGene transfection reagent (Roche Diagnostics).

Expression analysis in MCF-7 and SKBr3 cells. mRNA from E₂-treated MCF-7 cells or from SKBr3 cells transfected with the ERRα siRNA or treated with XCT-790 (5 μmol/L) was reverse transcribed into cDNA using SuperScript (Invitrogen) and analyzed by Q-RT-PCR with SYBR Green–based RT-PCR (Roche). Primer pairs used for Q-RT-PCR are listed in Supplementary Table S2.

Gene Ontology and enrichment of biological processes. Identification of the biological processes enrichment was done using the FatiGo functional enrichment tool⁹

Relation of ERRα target genes to breast cancer expression subtypes. The ERRα gene list was ranked by interquartile range; genes above the 0.85 quantile were selected to build the heat map. Genes and samples were grouped by hierarchical clustering (ward linkage, 1-Pearson correlation). Please refer to the Supplementary Materials and Methods for the association between cluster membership and clinical characteristics.

Genome-wide ChIP-on-chip analyses differentiate ERRα from ERα transcriptional networks in human breast cancer cells. We used chromatin immunoprecipitation coupled to microarray analysis (ChIP-on-chip) to identify genes directly controlled by ERRα in breast cancer cells. To relate the binding events to a specific target gene, we performed two distinct ChIP-on-chip experiments using tiled genomic DNA arrays covering the extended promoter regions (from −5.5 to +2.5 kb from transcriptional start site) of ∼17,000 human genes. First, we examined ERRα binding sites in the ERα-positive MCF-7 human breast cancer cell line cultured in phenol red–free DMEM supplemented with hormone-depleted serum. Second, to avoid any interference with ERα, we identified ERRα binding sites in the SKBr3 human breast cancer cell line, which does not express ERα and displays amplification of the ERBB2 locus (7). Analysis of the ChIP-on-chip data sets identified 414 and 795 high-confidence ERRα binding sites mapping to the promoters of 447 and 799 genes in MCF-7 and in SKBr3 cells, respectively (Supplementary Fig. S1A and B; Supplementary Tables S3 and S4). Together, a total of 1026 ERRα-bound extended promoter regions were identified in the two breast cancer cell lines studied (Fig. 1A).

To determine whether ERRα and ERα share DNA binding sites in vivo, we performed an ERα chromatin immunoprecipitation experiment on the same platform using chromatin obtained from MCF-7 treated with E₂. The experiment identified a total of 1,184 high-confidence ERα-bound segments mapping to the promoters of 1,165 different genes (Fig. 1A; Supplementary Fig. S1C; Supplementary Table S5).

Comparison of the data sets from both receptors revealed that a total of 212 ERRα target genes are also targets of ERα (18% of all ERα targets; Fig. 1A). Extensive chromatin immunoprecipitation validation showed that ERRα and ERα display very strict binding site specificities in intact chromatin (Fig. 1B; Supplementary Fig. S1B-D). Examination of the bound segments revealed four different classes of promoter regions targeted by the two receptors (Fig. 1C). For example, the promoter of ABCG2 contains a segment that is bound only by ERα, whereas the promoter of TAPBL is recognized specifically by ERRα. The binding of ERα and ERRα to the promoters of the shared target genes occurs either through distinct segments in the extended promoter (e.g., FRAT2) or through a common segment that can be recognized by both receptors (e.g., C1orf151). Of the 212 common promoters, 131 include a segment bound by both receptors (Supplementary Fig. S1D; Supplementary Tables S6 and S7). The degree of overlap between ERRα and ERα targets detected with the SKBr3 cells (20%) is similar to the overlap observed with the MCF-7 cells (24%), suggesting that the absence of ERα in SKBr3 cells does not lead to the extension of the repertoire of ERRα target promoters.

We next tested whether ERRα and ERα recruitment to extended promoter regions relates to the control of gene expression by the receptors. We treated MCF-7 cells with E₂ or used siRNA to silence ERRα activity in SKBr3 cells and monitored the expression of target genes identified by ChIP-on-chip. As shown in Fig. 1D, the levels of expression of genes with unique ERRα or ERα binding sites respond specifically to the cognate regulatory molecules. In contrast, the expression of genes with promoters recognized by both receptors is modulated by both effectors (Fig. 1D).

Motif search identifies a novel mechanism of target gene specificity by nuclear receptors. The results of the ChIP-on-chip analysis were further validated using a motif-finding algorithm (21). A search for de novo motifs with length between 6 and 18 bp predominantly identified the known consensus ERRE (TNAAGGTCA) and ERE (AGGTCANNNTGACCT) for the ERRα- and ERα-specific segments, respectively (Fig. 2A). Significant enrichment of the cognate ERE (P = 3.8e-152) and ERRE (P = 6.1e-69) was also observed in the segments bound specifically by ERα and ERRα, respectively, and in the common promoters bearing distinct binding segments for each receptor (Fig. 2B). Remarkably, examination of the common segments bound by both receptors using de novo motif finding revealed the presence of a novel 18-bp extended motif consisting in a consensus ERRE embedded within a consensus ERE (TCAAGGTCANNNTGACCT; Fig. 2C). The 18-bp ERRE/ERE motifs are present in ∼40% of the common bound segments, which represents 6.5% of the total of ERRα-bound segments (Fig. 2D). In addition, ∼20% of the segments bound by both ERRα and ERα contain distinct but closely spaced consensus ERRE and ERE.

Molecular mechanisms of gene regulation via interplay between ERRα and ERα in human breast cancer cells. We examined the functional relationship between ERRα and ERα at the common ERRE/ERE sites. To test whether ERRα and ERα can co-occupy these sites when both are present in the same cells, we performed serial chromatin immunoprecipitation experiments in MCF-7 cells treated with E₂. As shown in Fig. 3A, whereas re-chromatin immunoprecipitation for ERα generated further enrichment at the common binding sites located in the FAM100A and ENO1 promoters, re-chromatin immunoprecipitation with the ERRα antibody failed to produce an additive enrichment at these sites. As controls, we observed enrichment with the ERRα antibody at a common segment bearing distinct sites within the CSAD locus and no further enrichment at a segment specifically recognized by ERα in the GSK3B promoter (Fig. 3A). Because both receptors are not simultaneously recruited to common ERRE/ERE sites, ERRα and ERα should compete for occupancy at these sites in cells where they are both expressed. Indeed, depletion of ERα protein using siRNA pools in MCF-7 cells cultured in regular DMEM supplemented with complete serum led to a decrease in ERα recruitment concurrent with an increase in ERRα recruitment to common segments on the promoters of FAM100A, ENO1 and C1orf151 loci bearing the ERRE/ERE motifs (Fig. 3B). In contrast, recruitment of ERRα to common segments bearing independent but closely located ERRE and ERE in the CSAD or ESRRA promoters or to specific ERRα segments in TAPBPL and SLC25A23 loci was not affected by manipulating ERα levels (Fig. 3B). In addition, treatment of the MCF-7 cells with E₂ led to a reduction of ERRα recruitment to common segments containing a ERRE/ERE motif while leaving the ERRα-specific and common promoters with distinct sites unaffected (Fig. 3C).

We next investigated the mechanisms used by the two receptors to confer transcriptional regulation to common target genes. We monitored by chromatin immunoprecipitation the recruitment of two coactivators, AIB1/SRC-3 and PGC-1β, known to be preferentially recruited by ERα and ERRα, respectively (22, 23), at shared or specific sites (Fig. 4). In MCF-7 cells, E₂ treatment preferentially induced the recruitment of ERα, AIB1, and PolII at common ERRα/ERα segments containing an ERRE/ERE motif located in the promoter regions of ENO1 and C1orf151 (Fig. 4A; Supplementary Fig. S2). In contrast, preferential recruitment of ERRα, PGC-1β, and PolII was observed on the same sites in SKBr3 cells. Differential recruitment of distinct cofactors was also observed on sites specific for each receptor (Fig. 4B and C).

Biological processes regulated by ERRα and ERα in breast cancer cells. We evaluated the biological processes associated with genes with promoter regions that are recognized specifically by ERRα, ERα, or both receptors using Gene Ontology. Target genes specific for ERα or ERRα were enriched for processes that were significantly different between the two receptor subtypes, indicating that ERRα mediates functions that are distinct from ERα signaling in breast cancer cells (Fig. 5A). The specific ERα target genes were mostly enriched in tissue development and cell proliferation, whereas the specific ERRα target genes showed enrichment for various metabolic-related functions (Fig. 5B). We observed specific ERRα recruitment to the promoters of several enzymes involved in the tricarboxylic acid cycle, to many of the subunits of the mitochondrial respiratory chain (e.g., NDUFS2, ATP5B, and UCP2), and to enzymes involved in glycolysis (e.g., ENO1 and ALDOA). ERRα ChIP-on-chip targets also includes genes involved in tumor growth and proliferation such as cell cycle (e.g., CCNE1 and NEK2), metastasis (e.g., CXCR4), apoptosis (e.g., MDM4 and BAK1), and transcription (e.g., ETS2 and FOXE3; Supplementary Fig. S3A and B).

A similar analysis revealed that many of the promoters bound by both receptors are associated with important genes in breast tumor biology (e.g., GATA3, STK11, ABCC5, KRT13, and SPRY1), indicating the high degree of biological relevance of the relatively small number of genes coregulated by both receptors to the disease (Supplementary Fig. S3C; Supplementary Table S8). Noteworthy, a considerable amount of the genes with promoters that are bound by both receptors are linked to ERBB2 signaling (e.g., ERBB2, CLND4, G3BP1, GRB7, JUN, and RPL19).

Expression profiling of ERRα target genes recapitulates breast cancer subtypes and predicts clinical outcome in breast cancer. The interplay between ERRα, ERα, and ERBB2 signaling in breast cancer cells described herein and the previous findings that ERRα expression correlates with poor prognosis suggest that ERRα signaling could be relevant across different tumor subtypes. Therefore, we examined the expression profile of ERRα direct target genes in microarray expression data from two distinct cohorts of human breast tumor samples (24, 25). Unsupervised hierarchical clustering of the tumors according to the expression of the ERRα targets revealed four main clusters that significantly and specifically (NKI rand index = 0.44, randomization P = 0.008, Chin rand index = 0.63, and randomization P < 0.001) recapitulated previously established breast tumor subtypes (ref. 26; Fig. 6A; Supplementary Fig. S4), suggesting that ERRα displays subtype-specific signaling.

We next examined the prognostic value of ERRα targets in six cohorts of expression data from human breast tumor samples by univariate Cox regression analysis and identified 86 ERRα target genes significantly associated with clinical outcome in at least two independent data sets (Fig. 6B). Significant enrichment of this univariate predictor is observed in five of the data sets (Ivshna, P < 2.2e-16; McGill, P < 2.2e-16; Wang, P < 2.2e-16; Bild, P < 8.9e-13; and Chin, P < 2.2e-16). This prognosis-associated ERRα direct target gene signature is subsequently called ERRDTG. The genes from the ERRDTG signature mediate a variety of biological processes including regulation of cell cycle, transcription, and metabolism (Supplementary Table S9). Twenty-three (26.6%) promoters also contained segments bound by ERα in MCF-7 cells, a similar ratio than the one observed with the complete lists of targets, indicating that the ERRα direct target genes associated with outcome are not specifically enriched for common ERRα/ERα targets. However, 19 of the 23 (82%) shared promoters contain segments with overlapping binding sites for ERRα and ERα (Supplementary Table S10), suggesting that competitive and differential recruitment of the receptors could be of clinical importance in breast cancer progression. We next show that five genes of the ERRDTG signature (NEK2, C14orf156, KIAA0406, UBE2C, and CDC20) display independent prognostic value over ERα and ERBB2 status by multivariate Cox regression analysis, indicating that ERRα signaling can specify clinical outcome independently of ERα and of ERBB2 status in breast tumors (Fig. 6B, genes in red). Finally, treatment of SKBr3 cells with the compound XCT-790 that preferentially antagonizes ERRα (27) significantly affected the expression of several of the genes within the ERRDTG signature, suggesting that the expression of genes included in the signature could be amenable to pharmacologic manipulation (Fig. 6C).

In this study, the identification of ERRα target genes in human breast cancer cells established an unequivocal distinction between ERRα and ERα transcriptional activities at a genomic, functional, and mechanistic levels in breast cancer cells. We showed that ERRα controls a vast network of metabolic genes in breast cancer cells, specific components of ERBB2 signaling pathways, and a small subset of ERRα/ERα coregulated genes that are highly significant in breast cancer biology. The biological relevance of the ERRα target genes was further probed by intersecting the targets of ERRα and gene expression data from breast tumors. The resulting expression profiling-based clustering of the tumors recapitulates previously established breast cancer subtypes, suggesting the involvement of ERRα transcriptional activity in breast cancer heterogeneity. Finally, identification of the ERRDTG signature associated with clinical outcome suggests that ERRα signaling may contribute to important pathways leading to breast cancer progression including pathways linked to both ERα and ERBB2 signaling.

A prominent feature of this study is the observation that ERRα and ERα, two closely related members of the nuclear receptor superfamily, display a very strict DNA binding site specificity in vivo. Indeed, a de novo motif search of the segments specifically bound by ERRα and ERα identified the consensus ERRE and ERE as the prevailing motifs for each receptor, respectively. More striking, however, was the finding that the predominant motif present in segments bound by both ERRα and ERα is a novel chimeric hormone response element of 18 bp in length composed of an ERRE embedded within an ERE. This uncovers a mechanism by which two nuclear receptors that display strict binding site specificity can regulate the same gene via a hybrid hormone response element. This finding was somewhat unexpected as previous studies have shown that the ERRα and ERα could bind to each other's response element in vitro and regulate synthetic reporter constructs harboring either type of binding site in transfected cells (28–30). This work shows that nuclear receptors possess a much superior ability to discriminate among binding sites in intact chromatin than in nonnatural settings.

The binding site specificity displayed by ERRα and ERα is also reflected in the finding that the two receptors co-occupy a much smaller subset of promoters than originally anticipated. Our ChIP-on-chip approach showed that only 21% of ERRα target promoters are also recognized by ERα through either distinct or common binding sites, representing ∼14.5% and 6.5% of all ERRα targets, respectively. This indicates that direct competitive binding by the two receptors at extended promoter regions is a minor component in their regulatory cross-talk. Whole-genome studies have shown the predominant recruitment of ERα to regions distal to the transcriptional start site of target genes, therefore suggesting that some common distal targets could have been missed by limiting our study to extended promoter regions. Future whole-genome studies will reveal the exact distribution of ERRα binding sites in relation to the transcriptional start site of target genes. However, when considering previously reported ERα binding sites (12, 15) located up to 50 kb away from the transcriptional start site of our ERRα targets, the scope of ERRα/ERα common promoters was not significantly changed (21-23%), showing that ERRα/ERα transcriptional cross-talk is relatively limited in human breast cancer cells. This conclusion is further supported by a recent study reporting a limited overlap between genes regulated by the ERRα coregulator PGC-1α and E₂ in breast cancer cells (31).

The biological processes associated to ERα and ERRα direct target genes are mostly distinct with ERRα targets regulating metabolic pathways, particularly oxidative metabolism and the tricarboxylic acid cycle. This result unambiguously establishes the involvement of ERRα in the direct regulation of breast cancer cell metabolism. This correlates with recent studies that reported the ERRα-dependent up-regulation of some genes implicated in oxidative metabolism and tricarboxylic acid cycle in breast cancer cell lines (31, 32). We also observed recruitment of ERRα to the promoters of genes involved in mitochondrial dysfunction (e.g., SOD2 and UCP2) and glycolysis (e.g., ALDOA and ENO1). Therefore, it would be of particular interest to investigate whether and how changes in ERRα activity contribute to the metabolic adaptation of cancer cells to the tumor microenvironment. We also show that ERRα binds to the promoters of a large number of genes involved in many processes linked to tumor growth and proliferation, which uncovers previously unrecognized and less studied functions for ERRα. Noteworthy, several ERRα target genes in the prognosis-associated ERRDTG signature have been linked to adverse clinical outcome in previous studies on breast tumors, including CCNE1 (33), S100P (34), S100A14 (35), STRA13 (36), UBE2C (37), HSPC111 (38), and RPL19 (39).

Although our findings imply that the ERRα transcriptional network is mostly independent of ERα, the small fraction of common ERRα/ERα targets comprises several genes of major importance in breast tumor biology. For instance, the prognosis-associated ERRDTG signature includes common ERRα/ERα targets consistently present in the ERBB2 amplicon as well as GATA3, a transcription factor with expression that has been linked to good prognosis (18, 40) and strongly associates with ERα expression in breast cancer (41). In addition, the list of ERRα/ERα common promoters includes genes encoding nuclear receptors (ESRRA, NR0B2, NR1D1, RARA, RORC, SF1, and THRA), suggesting extensive cross-regulation of transcriptional networks by members of the nuclear receptor superfamily in breast cancer cells.

The observation that the differential expression of ERRα targets significantly segregates the tumors in clusters that are in close agreement with previously established subtypes (26) suggests that ERRα displays subtype-specific signaling and might contribute to breast cancer heterogeneity. These observations are in agreement with recent studies showing the importance of ERRα in the growth of both ERα-negative and ERα-positive breast cancer cells (31, 42). Notably, the identification of the ERBB2-positive cluster is driven by the expression of ERRα targets located within the ERBB2 amplicon, some of which were significantly associated with outcome in the ERRDTG signature (e.g., PERLD1, GRB7, and RPL19). Interestingly, we also identified ERBB2 itself as a gene controlled by ERRα at a site shared with ERα. Previous studies indicated that estrogen signaling can down-regulate ERBB2 expression in ERα-positive breast cancer cells (43, 44), and our study indicates that ERRα may also play a role in dictating ERBB2 expression in breast tumors.

We have shown that several ERRα direct target genes are known to play key roles in breast cancer biology. Furthermore, the approach described in this study combining identification of direct target genes and gene expression data sets could be used to identify linkage between lists of direct target genes of other nuclear receptors, gene expression signatures, and biological responses in both normal physiology and disease state of any tissue of interest. Finally, our work implicating ERRα as a determinant of breast cancer heterogeneity shows that the value of ERRα as a therapeutic target resides in its potential to influence several major biologically relevant pathways active in all subtypes of breast cancer.

No potential conflicts of interest were disclosed.

Grant support: Canadian Institutes for Health Research, Terry Fox Foundation Program Project Grant from the National Cancer Institute of Canada, Fonds de la Recherche en Santé du Québec studentship (G. Deblois), and National Sciences and Engineering Research Council of Canada studentship (M-C. Perry).

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.