In estrogen receptor (ER)–negative breast cancer, high tumor glucocorticoid receptor (GR) expression has been associated with a relatively poor outcome. In contrast, using a meta-analysis of several genomic datasets, here we find that tumor GR mRNA expression is associated with improved ER+ relapse-free survival (RFS; independently of progesterone receptor expression). To understand the mechanism by which GR expression is associated with a better ER+ breast cancer outcome, the global effect of GR-mediated transcriptional activation in ER+ breast cancer cells was studied. Analysis of GR chromatin immunoprecipitation followed by high-throughput sequencing in ER+/GR+ MCF-7 cells revealed that upon coactivation of GR and ER, GR chromatin association became enriched at proximal promoter regions. Furthermore, following ER activation, increased GR chromatin association was observed at ER, FOXO, and AP1 response elements. In addition, ER associated with GR response elements, suggesting that ER and GR interact in a complex. Coactivation of GR and ER resulted in increased expression (relative to ER activation alone) of transcripts that encode proteins promoting cellular differentiation (e.g., KDM4B, VDR) and inhibiting the Wnt signaling pathway (IGFBP4). Finally, expression of these individual prodifferentiation genes was associated with significantly improved RFS in ER+ breast cancer patients. Together, these data suggest that the coexpression and subsequent activity of tumor cell GR and ER contribute to the less aggressive natural history of early-stage breast cancer by coordinating the altered expression of genes favoring differentiation.
Implications: The interaction between ER and GR activity highlights the importance of context-dependent nuclear receptor function in cancer. Mol Cancer Res; 14(8); 707–19. ©2016 AACR.
This article is featured in Highlights of This Issue, p. 673
In the past decade, it has become increasingly clear that breast cancer subtypes have very different biologic behaviors. For example, improved early-stage estrogen receptor–positive (ER+) breast cancer prognosis appears to correlate with high ER and progesterone receptor (PR) expression and activity (1), while mechanisms of triple-negative breast cancer (TNBC) prognosis have yet to be well characterized (2). Our laboratory used a retrospective meta-analysis of more than 1,000 early-stage ER+ breast cancer patients and found that high tumor glucocorticoid receptor (GR/NR3C1) mRNA expression was associated with an improved prognosis compared with low or absent GR/NR3C1 expression (3). This result suggested that in the presence of ER, high GR expression leads to a less malignant phenotype by altering the regulation of genes associated with tumor aggressiveness. This finding was surprising because the same retrospective meta-analysis of more than 300 early-stage ER-negative (ER−) tumors found that high versus low GR mRNA expression was associated with a relatively poor prognosis (3); the latter conclusion is supported by a recent immunohistochemical analysis of GR expression in ER− breast cancer (4). Interestingly, consistent with results in early-stage ER− breast cancer (5–7), GR activity appears to be associated with the expression of tumor cell survival and chemoresistance genes in a number of other cancers, including ovarian (8–10) and castration-resistant prostate cancer (11, 12).
Studies examining GR association with chromatin in the presence of activated ER have suggested that both ER and GR have chromatin remodeling activity (13–15). For example, both ER and GR can increase chromatin accessibility for each other in ER+/GR+ murine mammary cells (13). In a model system using ER response elements (ERE), known EREs from the prolactin receptor gene regulatory region demonstrated increased GR association following ER activation, suggesting ER and GR act cooperatively (14). Conversely, GR activation was shown to displace liganded ER from AP1-associated chromatin regions near the transcriptional start sites (TSS) of the well-established ER target genes CCND1 and TFF1 (15). On the basis of these observations and our own work, we hypothesized that ER could modulate the chromatin localization of activated GR, thereby affecting GR-dependent gene expression pathways that contribute to the better outcome of ER+/GR+ breast cancer patients.
To begin to test this hypothesis, we first examined whether ER and GR expression is independent of the known good prognostic association of high PR expression in early-stage ER+ breast cancers. We then employed GR and ER chromatin immunoprecipitation with and without their respective ligands, followed by high-throughput sequencing (ChIP-seq) and genome-wide transcript profiling in GR+/ER+ MCF-7 cells. We found that coactivation of GR and ER was associated with a relative increase in the expression of prodifferentiating genes compared with individual receptor activation alone. ER and GR coactivation also resulted in decreased expression of GR-regulated endothelial-to-mesenchymal transition (EMT)-related genes compared with GR activation alone. These findings form the framework for understanding how high GR expression, chromatin remodeling, and transcriptional activity result in improved patient outcome in early-stage ER+/GR+ breast cancer.
Materials and Methods
Cell culture and reagents
MCF-7 cells were obtained from ATCC and were revalidated within 6 months of use using short tandem repeat profiling (DDC Medical). MCF-7 cells were grown at 37°C in 5% CO2 in DMEM (Lonza) supplemented with 10% FBS (Gemini Bio Products) and 1% penicillin/streptomycin (Lonza). Estradiol (E2, Sigma-Aldrich) and dexamethasone (Sigma-Aldrich) were dissolved in vehicle [ethanol (EtOH)] in 1 mmol/L stock solutions and further diluted in EtOH for the various cell culture experiments. Protease and phosphatase inhibitor tablets (Roche Diagnostics) were dissolved in respective lysis buffers per the manufacturer's protocol.
GR and ER ChIP assay and analysis
MCF-7 cells were incubated in phenol red-free DMEM containing 2.5% charcoal-stripped serum for 4 days total (with a media change after 48 hours) and treated with either EtOH (60 minutes), 100 nmol/L dexamethasone (60 minutes), or 100 nmol/L E2 (75 minutes) or pretreated with E2 for fifteen minutes, followed by cotreatment with dexamethasone (i.e., 75 minutes E2 and 60 minutes dexamethasone). These hormone concentrations were used previously in GR ChIP-seq experiments (13). DNA and associated proteins were crosslinked with 1% formaldehyde, and lysates were sonicated and immunoprecipitated as described previously (3). ChIP experiments were conducted using the ChIP Assay Kit and the manufacturer's protocol (EMD Millipore). ChIP-grade rabbit polyclonal anti-GR (sc-1003x, Santa Cruz Biotechnology, recognizing the N-terminal region of GR) and anti-ER HC-20 (sc-543x, Santa Cruz Biotechnology, recognizing the C-terminal region of ERα) were used for immunoprecipitation. We also used normal rabbit IgG (sc-2027; Santa Cruz Biotechnology) as a negative control. Eluted ChIP DNA was purified using the PCR Purification Kit (Qiagen). GR ChIP samples, ER ChIP samples, and corresponding input samples were analyzed in triplicate.
GR ChIP-seq and ER ChIP-seq was performed on the Illumina HiSeq platform, generating raw reads for analysis (see Supplementary Materials and Methods). The sequence alignment and identification of peaks is described briefly. Sequence quality was assessed and aligned to the human genome (version hg19), and peaks were detected using two algorithms: rgt-ODIN v0.3.2 (16) and MACS2 v126.96.36.19940616 (17). The final lists of peaks contained concordant peaks between the two algorithms, with an FDR of 0.05 (MACS2). Next, ChIP-seq peaks were annotated on the basis of genome features using Homer v4.5, and binding region sequences were determined using CentriMo v4.10.0. Androgen receptor (AR) and GR share the same response element (RE) sequence (18); thus, a CentriMo-identified “AR RE” is likely a functional glucocorticoid receptor response element (GRE) in the context of dexamethasone treatment and anti-GR ChIP-seq. Likewise, EREs in this study include Homer-defined REs that include ER full and half-sites (such as REs for ESR1, ESR2, ESRRA, ESRRB, NR2F1, and NR2F2). GR enrichment at specific genomic features was calculated, and log2-enrichment scores for Homer-defined regions were obtained (see Supplementary Materials and Methods for additional description). GR and ER ChIP-seq peaks for specific genes (KDM4B, IGFBP4, and VDR) were visualized using the Integrative Genomics Viewer (the Broad Institute, Cambridge, MA).
Global gene expression profiling and data analysis
MCF-7 cells were incubated in phenol red-free DMEM containing 2.5% charcoal-stripped serum for 4 days total (with a media change after 48 hours). Next, the cells were treated with either vehicle, dexamethasone (100 nmol/L), E2 (10 nmol/L), or dexamethasone/E2 for 4 hours (to identify early gene expression resulting from GR and/or ER chromatin association) in 2.5% charcoal-stripped serum (following incubation in stripped serum as described above). Cells were lysed in an RNA lysis buffer (Ambion, Invitrogen), and RNA was isolated and submitted to The University of Chicago Genomics Facility for reverse transcription and Affymetrix microarray as described previously (7), using the Human Genome U133 Plus 2.0 array platform. Global transcript profiling was performed in triplicate for each sample, and resulting data (available on GEO:GSE79761) were normalized using the RMA method from Affymetrix Power Tools v1.10.2. The dexamethasone, dexamethasone/E2, and E2 treatments were compared with the vehicle to estimate relative expression values. Probes were filtered to include those with at least two of three replicates exhibiting absolute relative expression values of ±1.5-fold change (3). Ingenuity Pathway Analysis (IPA, Qiagen) was performed with averaged expression values over two biologic replicates to generate lists of canonical pathways for the dexamethasone/E2 and E2 treatment conditions (compared with vehicle).
To confirm differential gene expression, the Custom Profiler RT2 PCR Array (Qiagen, 4-hour compound treatment) and/or individual quantitative RT-PCR (qRT-PCR) of mRNA were performed on three separately prepared biologic replicates. Transcript levels were normalized to the housekeeping gene RPLP0. Genomic contamination control primers for GRIE3 were also used to assure RNA purity. Relative differences (at least 15% change in either direction) in fold change gene expression (normalized initially to vehicle treatment) between E2 versus dexamethasone/E2 treatment groups were calculated using a dataset of averaged significantly up- and downregulated microarray and custom array genes. Additional qRT-PCR was performed for individual genes of interest as follows: KDM4B, 8-hour compound treatment; IGFPB4, 8 hours; VDR, 2 hours; these hormone treatment times were selected after a time course optimization for maximal transcript levels between 0 to 8 hours. PCR primer sets for individual qRT-PCR are listed in the Supplementary Materials and Methods. Fold change was calculated by normalizing C t values to RPLP0 and then vehicle controls; error was calculated as SD (propagated error; ref. 19), and P values were generated by a paired t test.
Kaplan–Meier plots of gene expression and relapse-free survival
Affymetrix ESR1 probe 205225_at expression was used to classify patients as ESR1 + and ESR1 − based on the cut-off value previously determined using the ROC analysis that best predicted ER+ status by IHC (3). PR (PGR)-high and -low expression groups were defined as being above or below the median expression of the Affymetrix PGR probe 208305_at among all 1,378 patients (3). Among the 1,024 (74.3%) ESR1 + patients, 664 (64.8%) were classified as PGR-high. For all other genes (NR3C1, KDM4B, VDR, and IGFBP4), high and low expression groups were defined as the top and bottom 25% of expression levels, respectively. For KDM4B, VDR, and IGFBP4, we used the corresponding Affymetrix probeID, in which we observed differential gene expression in the initial genome-wide microarray experiments.
Relapse-free survival (RFS) was estimated using the method of Kaplan–Meier, and groups were compared using the log-rank test. Hazard ratios (HRs) were estimated using Cox proportional hazards regression model. We determined whether tumor NR3C1 expression was associated with differential RFS among all ESR1 + patients, followed by the high- and low-PGR gene expression subgroups of ESR1 + patients. All NR3C1 probes hybridize to GR-α, and probe 216321_at maps to both GR-α and GR-β (3). Pearson correlation coefficient between expression levels of the PGR probe (208305_at) and each of the four NR3C1 probes was calculated among ER+ patients. Similarly, we determined whether tumor KDM4B, VDR, and IGFBP4 expression was associated with differential RFS for all ESR1 + patients.
Knockdown of prodifferentiating genes
Individual transient siRNA knockdown of KDM4B, IGFBP4, and VDR were performed in MCF-7 cells. Briefly, the MCF-7 cells were cultured in 5% charcoal-stripped serum for 4 days, with a media change after 48 hours. On day 4, reverse transfection was performed using RNAiMax (Invitrogen) and a pool of four siRNAs for each gene as well as a nontargeting control sequence siRNA pool (ON-TARGETplus human siRNA SMARTpool; Dharmacon; siRNA sequences in the Supplementary Information). The RNAiMax protocol was performed using 150 pmol siRNA per well of a 12-well plate. Live cell number was counted over the course of 72 hours by using either phase-contrast images (20) from IncuCyte Live Cell Imager (Essen) or trypan blue exclusion assay. Knockdown efficiency of each gene compared with control siRNA was measured by qRT-PCR (primers are in the Supplementary Information).
Tumor GR expression is associated with improved RFS in a meta-analysis of early-stage ER+ breast cancer patients
Previous studies suggest that high tumor GR expression is significantly associated with a shortened RFS in ER− breast cancer patients and, conversely, with an improved RFS in ER+ breast cancer (3). These findings prompted us to further examine the association between GR and RFS in ER+ breast cancer. Thus, we performed a retrospective meta-analysis of 1,024 ER+ breast cancer patients (3) to determine whether GR mRNA (NRC31) expression associated with RFS. Indeed, we found that high tumor NRC31 expression was significantly associated with improved outcome in four of four Affymetrix NR3C1 probes, with HRs for recurrence ranging from 0.35 to 0.65 (Fig. 1A).
It is well established that ER+ breast cancer patients with high tumor PR expression have improved RFS compared with patients with low tumor PR expression (1). PR has been recently shown to alter ER chromatin association as a possible mechanism for an improved outcome in ER+/PR+ breast cancer patients (21). To determine whether the significant association of high tumor GR expression with improved RFS in ER+ patients is dependent on PR expression, we examined our meta-analysis dataset and stratified ER+ breast cancers based on high versus low PR transcript (PGR) expression. We found that high tumor GR expression was associated with improved RFS in patients with both ER+/PR-high tumors (Fig. 1B, HRs ranging 0.35–0.55) and in ER+/PR-low tumors (Fig. 1C, HRs ranging 0.39–0.84). In addition, the statistical interaction between PR positivity (based on the single available PGR probe) and each GR (NR3C1) probe was nonsignificant in Cox regression models, suggesting that tumor GR and PR expression independently affects patient outcome. To further support this observation, we found that the correlation between tumor GR mRNA expression (all four NR3C1 probes) and PR expression (one PGR probe) was low in the ER+ patient subset (Pearson correlation coefficients ranging from −0.02 to 0.2). Taken together, these data suggest that tumor GR expression is associated with improved RFS in ER+ breast cancer patients, independently of tumor PR expression.
The global pattern of GR association with chromatin is altered by ER activation
Given that high GR expression was associated with a significantly improved RFS in ER+ patients independently of PR expression (Fig. 1), we sought to uncover how GR and ER could coordinately activate pathways that contribute to a better patient outcome. To study the role of GR association with chromatin, we treated GR+/ER+ MCF-7 cells with the GR agonist dexamethasone (100 nmol/L, 60 minutes), ER agonist estradiol (E2, 100 nmol/L, 75 minutes), or the combination of both (Dex/E2). Bioinformatic analysis of GR ChIP-seq revealed GR-binding regions (GBRs), indicated by ChIP-seq peaks. To reduce the occurrence of false positives, GR-associated chromatin sequences were identified using concordant peaks from two algorithms, ODIN (16) and MACS (17). In the absence of ligand, only 726 genome-wide GR peaks were observed (Supplementary Fig. S1A). While the total number of dexamethasone-treated GR peaks remained approximately the same regardless of ER activation (Fig. 2A, dexamethasone vs. dexamethasone/E2), about 60% of dexamethasone/E2 peaks were unique to the dexamethasone/E2 condition (N = 5,963 genome-wide), demonstrating an altered GR chromatin landscape upon dual receptor activation. Furthermore, upon GR and ER coactivation, the change in GR location reflected significantly increased relative enrichment of GR chromatin association at promoter and DNA sequences that encode the 5′ and 3′ untranslated regions (5′UTR and 3′UTR; Fig. 2B). This dramatic change in GR chromatin topography suggests that activated ER can remodel chromatin, allowing increased accessibility of GR to proximal gene-regulatory regions.
To determine which of the genome-wide GBRs might be contributing to changes in transcriptional activity, we examined GBRs within 100 kb in either direction of known (annotated) TSSs (version hg19). We identified 7,500 GBRs (peaks) upon GR activation alone and 7,292 GBRs upon GR and ER coactivation (Fig. 2A, middle). Similar to our genome-wide peak analysis, about 60% of the GBRs within 100 kb of an annotated TSS were unique to the dexamethasone/E2 condition. To connect chromatin association with potential transcriptional activity, we identified the TSS-associated genes closest to the GR peaks in any treatment condition (Fig. 2A, right). We found a total of 5,289 GR-associated genes following GR activation alone and 5,128 genes following dual receptor activation, while approximately 40% of the annotated genes (N = 2,097) were unique to the dexamethasone/E2 treatment groups. Together, these data suggest that ER has an important role in determining the topography of GR chromatin association.
Dual ER and GR activation results in GR enrichment at GR- and ER-related REs
Because we found that coactivation of GR and ER resulted in altered GR association with DNA (GBRs), we next sought to characterize GBRs by identifying their associated REs. Using a CentriMo-based motif analysis, we found that in the absence of ER activation, GR associated most significantly with GREs (E value = 9.3e−1147, dexamethasone vs. vehicle; Fig. 2C). To study how ER activation changes GR enrichment, we performed a differential motif analysis (CentriMo) comparing dexamethasone/E2 with dexamethasone-only conditions. When GR and ER were coactivated, GR demonstrated significantly enriched association with GREs compared with GR activation alone (E value = 5.2e−1258, dexamethasone/E2 vs. dexamethasone; Fig. 2C). Also, upon dual receptor activation, GR associated more with EREs (E value = 4.3e−202; Fig. 2C), consistent with GR ChIP-seq data in murine mammary cells (13). Furthermore, ER activation led to an increase in GR association at REs for the cooperating transcription factor AP1 (E value = 1.5e−7) and the ER-associated pioneering factor FOXO (E value = 2.9e−49; refs. 22, 23). Because GR is not believed to recognize or associate directly with EREs (24), these results suggest that ER remodels chromatin and allows increased GR accessibility to EREs (indirect), as well as to GREs (direct).
GBRs frequently overlap with ER-binding regions
The relative increase in GR association at known ER, AP1, and FOXO REs prompted us to investigate whether, conversely, ER associates with GBRs during coactivation. Thus, ER ChIP-sequencing was performed in MCF-7 cells treated with dexamethasone/E2 and compared with vehicle treatment (peak counts summarized in Supplementary Fig. S1A and S1B). Indeed, genome-wide CentriMo motif analysis of ER ChIP-seq showed that upon dexamethasone/E2 treatment, ER was significantly enriched not only at EREs (E value = 8.4e−1530), AP1 REs (E value = 3.2e−62), and FOXO REs (E value = 8.1e−92), but also at GREs (E value = 6.5e−379; Supplementary Fig. S1C). These observations further suggest that GR and ER can interact in a complex and associate with each other's known primary chromatin-binding regions.
To determine the proximity of GBRs and ER-binding regions (EBRs), we analyzed the location of GR (GR ChIP-seq) and ER (ER ChIP-seq) peaks following dexamethasone/E2. Of the total genome-wide GR (N = 9,740; Fig. 2) and ER (N = 14,353, Supplementary Fig. S1B) peaks, 4,390 GR and 6,355 ER peaks were found within ±100 kb of the same TSS (see Table 1, A). On the basis of gene annotation by the nearest TSS, we identified 2,674 genes in which GR and ER associated within ±100 kb of the TSS (Table 1, A). In a genome-wide analysis, we also identified 2,834 regions where chromatin immunoprecipitated GR and ER peaks overlapped (Table 1, B, partial peak overlap). Of these GR- and ER-overlapped peaks, 2,205 were within 100 kb of a TSS, revealing 1,734 TSS-annotated genes (Table 1, B). To determine those GR and ER peaks that overlapped exactly, we identified all instances where the summits of a GR and an ER peak fell within six or fewer nucleotides of each other (Table 1, C). Throughout the entire genome, 348 overlapping GR/ER summits were identified, and the majority of these summits (N = 266) were found within 100 kb of a nearest TSS. These 266 overlapping summits were associated with 259 genes (based on proximal TSSs; Table 1, C). Finally, we used motif analysis to determine the highest ranked transcription factor REs represented among proximal GBRs and EBRs. GREs and EREs were the top two significant sequences found when GR and ER bound in proximity, partially overlapped, or showed perfect overlap (Table 1, A–C). Overall, these analyses demonstrate that upon GR and ER coactivation approximately 30% (2,205 from Table 1) of the genome-wide GBRs (7,292 from Fig. 2A) demonstrate some overlap with EBRs. In these surprisingly frequent cases, GR and ER may concurrently complex with DNA, or alternatively, bind to the same chromatin region in a temporally independent fashion.
Coactivation of GR and ER results in differential gene expression compared with individual receptor activation
Because the coactivation of GR and ER extensively altered GBR locations, we sought to determine accompanying changes in early target gene expression. Using independent Affymetrix gene array experiments (N = 3), we measured steady-state gene expression changes after treatment with vehicle, dexamethasone, E2, or dexamethasone/E2 (4 hours). Known GR and ER target genes (e.g., SGK1, SNAI2, PER1 in dexamethasone treatment) and ER target genes (e.g., PGR, TFF1, GREB1, and IGFBP4 in E2 treatment) were identified as significantly induced ≥1.5-fold compared with control (vehicle) at 4 hours. Figure 3A shows a Venn diagram of all differentially expressed genes identified (±1.5-fold compared with vehicle) under each hormone condition. At 4 hours, we found that upon ER (E2) activation, a total of 1,420 genes were differentially expressed (Fig. 3A). Upon GR (dexamethasone) activation, we identified only 497 differentially expressed genes, whereas upon ER and GR (E2 and dexamethasone) coactivation, 1,093 genes were differentially expressed compared with vehicle. Of the 1,420 E2-regulated genes, 51% overlapped with the dual dexamethasone/E2 treatment, and of the 497 dexamethasone-regulated genes, 39% overlapped with genes regulated by the dexamethasone/E2 treatment. Upon dexamethasone/E2 treatment, 297 unique genes were differentially expressed.
As the overall goal of this study was to understand how GR expression and activation contribute to a more indolent ER+/GR-high breast cancer phenotype compared with ER+/GR-low breast cancers, we next examined genome-wide gene expression with GR and ER coactivation (dexamethasone/E2) compared with ER activation alone (E2). To determine differences in gene expression patterns and associated pathways, we performed IPA using the differentially expressed genes in the dexamethasone/E2 (vs. vehicle) and E2 (vs. vehicle) treatment groups. The most significant gene expression pathways for the E2 alone–regulated genes were related to cancer, cell differentiation, and cell proliferation (Fig. 3B). Interestingly, the most significant pathway for the dexamethasone/E2–regulated genes was related to cell differentiation, followed by pathways involved in cancer, inflammation, cell proliferation, and cell-cycle checkpoint regulation (25). We next intersected GR ChIP-seq (Fig. 2) and ER ChIP-seq (Supplementary Fig. S1) TSS-annotated genes to determine which dexamethasone/E2 pathway genes had direct GR- and/or ER-associated chromatin binding (Fig. 3B, right three columns). Most of the dexamethasone/E2–regulated pathways contained genes in which at least half had either a GBR or EBR or both within ±100 kb of their gene TSS. Furthermore, many dexamethasone/E2 pathway genes had both GR and ER localized within ±100 kb of the TSS (Fig. 3B, right column). Together, these data suggest that ER and GR coactivation directly and coordinately induce gene expression pathways affecting differentiation and proliferation.
To further explore the transcriptional consequences of GR and ER coactivation versus ER activation alone, we next determined which genes were increased or decreased by at least 15% in the dexamethasone/E2 treatment compared with the E2 treatment alone. This analysis revealed that dexamethasone/E2 treatment resulted in the relatively differential expression of 308 genes (142 upregulated and 166 downregulated) compared with E2 alone. This comparison, when limited to genes with GBRs and/or EBRs within 100 kb of a TSS, revealed several putative direct GR (N = 225; 123 up and 102 down) and ER (N = 191; 110 up and 81 down) target genes (Fig. 3C). Moreover, the majority of these GR and ER target genes contained both GR and ER-binding regions within ±100 kb of the gene TSS (Fig. 3C), suggesting that ER and GR coordinately contribute their differential regulation. Finally, among the GR- and/or ER-bound genes with differential expression in the dexamethasone/E2 treatment group (compared with E2 alone) were several genes that encode “master regulator” transcription factors, their upstream kinases, and chromatin remodeling proteins (see Supplementary Table S1). Therefore, in addition to affecting gene expression involved in promoting cellular differentiation, the activation of GR and ER alters the expression of genes encoding master transcriptional regulators, suggesting that GR may act as a “master regulator of master regulators.”
Coactivation of GR and ER results in increased expression of genes associated with cellular differentiation
To further understand why ER+/GR-high tumors exhibit a better clinical outcome compared with ER+/GR-low tumors, we next analyzed our microarray and qRT-PCR data to determine which differentiation-related genes differed in expression. We sought to limit our study to genes where GR associated within ±100 kb of their TSS (GR ChIP-seq in dexamethasone/E2, see Supplementary Data). We found relatively increased gene expression in dexamethasone/E2 versus E2 of several genes associated with cellular differentiation such as VDR, chromatin remodelers such as KDM4B, as well as negative regulators of pro-oncogenic Wnt signaling, such as IGFBP4 and CCDC88C (DAPLE; Supplementary Table S1).
In addition to relatively increased expression of genes expected to favor differentiation and less aggressive tumor growth, conversely, hallmark EMT genes SNAI2 and SOX2 (26, 27) were markedly decreased upon coactivation of ER and GR compared with GR activation alone (Supplementary Data). This is consistent with our previous observation that in TNBC, GR drives pathways promoting cell survival and EMT (3). Furthermore, we also observed relatively repressed expression of other genes encoding proteins related to cellular dedifferentiation and EMT, such as zinc finger transcription factors (EGR3, KLF9, TRERF1), chromatin remodelers (SUV39H2), cytoskeletal organization (ARHGEF26, RHOU, RHOBTB1, ARHGAP36, TBC1D8), and cell junctions and cell polarity (RET, DOCK4, CXCL12, LAMA3; see Supplementary Table S1 and Supplementary Data for additional genes).
To further understand how GR chromatin association and individual gene expression may be altered by ER and GR coactivation, we chose three exemplary genes from the gene expression pathways (Fig. 3B) to study: Wnt signaling inhibitor insulin-like growth factor–binding protein 4 IGFBP4 (28), a chromatin remodeling enzyme associated with GR and ER function, lysine (K)-specific demethylase 4B KDM4B (29–31), and the vitamin D receptor VDR (32, 33). We first compared GR and ER chromatin binding within regulatory regions of these genes following single and dual hormone treatments (Fig. 4A). For IGFBP4, a known ER target gene, we observed ER chromatin association in the E2 and dexamethasone/E2 treatments, as well as some unliganded ER association in the vehicle and dexamethasone treatments (Fig. 4A, left). However, GR recruitment was only detected in the dexamethasone/E2 treatment, suggesting that ER remodels chromatin for GR binding (Fig. 4A and B, left). Indeed, following an analysis of RE motifs within the EBR and GBR peak of IGFBP4, we found FOXO REs and EREs among the most represented motifs in the dexamethasone/E2 treatment (Fig. 4A and Supplementary Fig. S3). An examination of regulatory regions for KDM4B revealed that the coactivation of GR and ER (dexamethasone/E2) resulted in novel GBRs as well as EBRs, implicating both GR and ER as chromatin remodeling factors (Fig. 4A and B, middle). Motif analysis of the dexamethasone/E2 EBR peaks revealed a high frequency of FOXO, AP2, and ER REs (Fig. 4A, middle). GREs were also identified within the EBR peaks, but less frequently (Supplementary Fig. S3). Finally, for VDR, we observed that activated GR associated with chromatin in both dexamethasone and dexamethasone/E2 treatment groups, whereas ER was only associated in the dexamethasone/E2 treatment. Motif analysis of the EBRs and GBRs resulting from the dexamethasone/E2 treatment for VDR showed the presence of GREs and EREs, as well as several FOXO and AP2 REs (Fig. 4A, right, and Supplementary Fig. S3). These data suggest that GR remodels chromatin, creating a new EBR in the regulatory region of VDR (Fig. 4A and B, right).
We next confirmed key gene expression changes initially detected by whole-genome transcript profiling using qRT-PCR. The dual activation of GR and ER resulted in increased expression of IGFBP4, KDM4B, and VDR compared with ER activation alone (Fig. 4C). Together with the chromatin binding data shown in Fig. 4A, we conclude that ER, GR, or both receptors can remodel chromatin in association with enhanced gene expression.
To determine the phenotypic effect of these three genes in primary breast cancer, we examined the RFS of early-stage ER+ breast cancer patients in the meta-dataset analyzed in Fig. 1 (3). A Kaplan–Meier plot of the top quartile versus bottom quartile of gene expression was generated. High tumor gene expression for IGFBP4 (P = 6.5e−11; refs. 34, 35), KDM4B (P = 6e−6), or VDR (P = 0.0028; ref. 33) was significantly associated with improved long-term RFS (Fig. 4D).
To further explore how expression of IGFBP4, KDM4B, and VDR affects ER+ breast cancer phenotype in a cell-based assay, we measured cell number over time following individual gene knockdown. Increased proliferation is associated with poorly differentiated, aggressive ER+ breast cancer, and decreased proliferative indices are observed in well-differentiated, more indolent ER+ breast cancer (36, 37). We therefore predicted that depletion of IGFBP4, KDM4B, or VDR gene expression would result in increased cell proliferation. Indeed, in comparison with MCF-7 cells expressing a control sequence, we observed a significant increase in cell proliferation (Fig. 4E) following siRNA depletion (Supplementary Fig. S4), suggesting these proteins have prodifferentation functions. These studies suggest that GR and ER coordinately alter gene expression in ER+ breast cancer, thereby contributing to a favorable tumor phenotype and patient outcome.
The efforts of many laboratories have contributed to our current understanding of the GR cistrome and transcriptome in mammalian epithelial cells (38–42). For example, GR [and other nuclear receptors (NRs)] interact with DNA in multifaceted complexes composed of coregulators (43, 44), transcription factors (22, 45), and more recently appreciated, other NRs (11–15, 21, 46). Thus far, the interplay between GR and other NRs is most evident in hormone-dependent cancers, such as prostate and breast cancer. Data from this study and others suggest that GR and ER influence each other's chromatin accessibility, as well as subsequent hormone-dependent transcriptional activity (3, 13–15).
Because GR is known to activate TNBC cell survival gene expression pathways (5, 6, 10–12) and high tumor GR expression is associated with poor ER− patient outcome (3, 4), we were struck by the observation that high GR expression was conversely associated with a significantly improved outcome in ER+ breast cancer. Through a retrospective meta-analysis, we found an association between high tumor GR expression and longer RFS in patients with ER+ tumors. Moreover, we found that the association of GR expression with improved RFS was independent of PR expression. Overall, it appears that ER+/PR− breast cancer patients with low GR–expressing tumors have markedly shortened RFS compared with those with high GR expression (Fig. 1), suggesting that GR status could be used to guide adjuvant treatment choice.
Our studies also support the notion that the cistrome resulting from GR and ER coactivation likely contributes to a more indolent phenotype in ER+ breast cancer. Interestingly, we found that the higher individual expression of putative cell-differentiating genes (KDM4B, VDR, IGFBP4) was associated with improved RFS in ER+ breast cancer patients. The chromatin remodeling protein KDM4B has been shown previously to be required for E2-mediated proliferation (29–31). However, our results suggest that increased KDM4B expression may play a different role in the context of GR and ER coactivation. Several genes related to EMT, such as SNAI2 and SOX2, were downregulated upon coactivation of ER and GR compared with GR activation alone. These data suggest that altered patterns of GR and ER chromatin binding are associated with changes in the expression of differentiation-related genes; future experiments using CRISPR to delete these binding regions could more definitively demonstrate their requirement. Finally, in addition to the increase in gene expression associated with cellular differentiation, GR activation may also inhibit ER-regulated proliferative genes (47). We are currently studying how GR activation alters global ER chromatin association, ER target gene expression, and ER-mediated cell proliferation.
Other NRs, such as PR and AR also appear to play a role in ER+ breast cancer by altering ER-mediated gene expression. The well-characterized MCF-7 cell line has been used extensively to study ER activity (48); however, these cells also express PR and some AR (49), both of which can crosstalk with GR and ER (11–12, 21, 39, 50–52). The consensus GRE is similar to AR and PR REs (18), suggesting a potential interplay between GR, AR, and PR at the chromatin level. It is possible that other NRs (such as PR and AR) may form complexes with GR and ER to influence each other's activity in subsets of breast cancer. We are currently examining whether the individual members of the NR3C nuclear receptor family share common chromatin rearrangement characteristics in ER+ breast cancer models.
The unexpectedly high incidence of partial (Table 1, B) or exact (Table 1, C) overlapping GBRs and EBRs suggests that GR and ER interact at functional regulatory regions (e.g., enhancers) and that GR and ER complex together when interacting with chromatin. Our identification of RE motifs for cooperative transcription factors (such as FOXO and GATA REs) within the shared GR/ER–binding regions of IGFBP4, KDM4B, and VDR (Fig. 4A) is consistent with finding these motifs at ER-dependent promoter and enhancer regions (22, 53). Additional studies to map known enhancers with these shared GR and ER–binding regions should be pursued. Although we did not perform experiments distinguishing simultaneous versus sequential GR and ER association with chromatin, we (Supplementary Fig. S2) and others (15) have observed GR and ER protein complex formation using immunoprecipitation in whole-cell MCF-7 lysates with and without ligand. We found that available ER coimmunoprecipitates with GR most efficiently in the presence of dexamethasone or dexamethasone/E2 (Supplementary Fig. S2), despite low-level steady-state GR (dexamethasone) and ER (dexamethasone/E2) due to ligand-induced receptor degradation (54, 55). Studies are ongoing to determine whether ligand activation recruits cooperative transcription factors, such as FOXO, to GR/ER complexes, thereby facilitating chromatin interaction.
Our analysis of GR genome-wide chromatin association suggests that liganded ER acts to remodel chromatin, resulting in altered GR accessibility to DNA. A role for ER in chromatin remodeling has been described previously (13, 56–58), but this function may be underestimated in ER+ breast cancer biology. Our observation of enriched GR binding at promoter regions upon ER activation suggests that ER alters chromatin configuration significantly. ER appears to remodel chromatin to expose de novo GBRs, and this in turn might allow GR to further modulate chromatin accessibility for other transcription factors (see model in Fig. 5).
Because ER has been successfully targeted for treatment in breast cancer, our studies suggest that the addition of GR modulation might be evaluated for therapeutic benefit. Indeed, a clinical trial of tamoxifen, chemotherapy, and the GR agonist prednisone was performed; however, patients were not stratified by GR or ER tumor expression, and thus, the results are difficult to interpret (59). Moreover, adjuvant use of daily prednisone was not tolerable. Our results suggest that treatment with an alternative GR modulator (with less toxicity than prednisone or dexamethasone) might promote the activation of prodifferentiation and antiproliferative gene expression pathways in ER+ breast cancer.
Disclosure of Potential Conflicts of Interest
Drs. Pan, Kocherginsky, and Conzen have a patent issued "methods and compositions related to glucocorticoid receptor (GR) antagonists and breast cancer." This patent covers ER− breast cancer and therefore is not directly relevant to this work in ER+ breast cancer. However, it may be considered broadly relevant to the work. No potential conflicts of interest were disclosed by the other authors.
Conception and design: D.C. West, D. Pan, C.F. Pierce, S.D. Conzen
Development of methodology: D.C. West, D. Pan, E.Y. Tonsing-Carter, C.F. Pierce, S.C. Styke, M. Kocherginsky, S.D. Conzen
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.C. West, D. Pan, E.Y. Tonsing-Carter, C.F. Pierce, S.C. Styke, K.R. Bowie, S.D. Conzen
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.C. West, D. Pan, K.M. Hernandez, C.F. Pierce, K.R. Bowie, T.I. Garcia, M. Kocherginsky, S.D. Conzen
Writing, review, and/or revision of the manuscript: D.C. West, E.Y. Tonsing-Carter, C.F. Pierce, S.C. Styke, K.R. Bowie, M. Kocherginsky, S.D. Conzen
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.C. West, E.Y. Tonsing-Carter, C.F. Pierce, S.C. Styke, K.R. Bowie, S.D. Conzen
Study supervision: D.C. West, C.F. Pierce, S.D. Conzen
The study was supported by NIH R01 CA089208, The University of Chicago Comprehensive Cancer Center NIH P30 CA014599, Susan G. Komen for the Cure IIR12223772, and the Prostate Cancer Foundation Movember Challenge Award.
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.
The authors thank Drs. Pieter Faber and Jaejung Kim of the University of Chicago Genomics Core Facility. The authors also thank Dr. Doug Turnbull of the University of Oregon Genomics Core Facility.
Note: Supplementary data for this article are available at Molecular Cancer Research Online (http://mcr.aacrjournals.org/).
- Received October 26, 2015.
- Revision received April 14, 2016.
- Accepted April 19, 2016.
- ©2016 American Association for Cancer Research.