Resistance to hormonal therapies is a major clinical problem in the treatment of estrogen receptor α–positive (ERα+) breast cancers. Epigenetic marks, namely DNA methylation of cytosine at specific CpG sites (5mCpG), are frequently associated with ERα+ status in human breast cancers. Therefore, ERα may regulate gene expression in part via DNA methylation. This hypothesis was evaluated using a panel of breast cancer cell line models of antiestrogen resistance. Microarray gene expression profiling was used to identify genes normally silenced in ERα+ cells but derepressed upon exposure to the demethylating agent decitabine, derepressed upon long-term loss of ERα expression, and resuppressed by gain of ERα activity/expression. ERα-dependent DNA methylation targets (n = 39) were enriched for ERα-binding sites, basal-up/luminal-down markers, cancer stem cell, epithelial–mesenchymal transition, and inflammatory and tumor suppressor genes. Kaplan–Meier survival curve and Cox proportional hazards regression analyses indicated that these targets predicted poor distant metastasis–free survival among a large cohort of breast cancer patients. The basal breast cancer subtype markers LCN2 and IFI27 showed the greatest inverse relationship with ERα expression/activity and contain ERα-binding sites. Thus, genes that are methylated in an ERα-dependent manner may serve as predictive biomarkers in breast cancer.

Implications: ERα directs DNA methylation–mediated silencing of specific genes that have biomarker potential in breast cancer subtypes. Mol Cancer Res; 15(2); 152–64. ©2016 AACR.

This article is featured in Highlights of This Issue, p. 115

Estrogen receptor α (ERα, ESR1) has proven to be the single most important target in breast cancer. Approximately 70% to 80% of breast cancers are ERα+, for which routine testing is used to predict response to antihormonal therapy (1). As demonstrated by genome-wide studies, ERα is a global regulator of gene transcription in breast cancer that orchestrates well-integrated hormonal responses that promote proliferation and survival and inhibit apoptosis (2–5).

As a result of regulating expression of thousands of genes, the presence of ERα drives the luminal classification of breast cancer. There are five intrinsic tumor subtypes, luminal A, luminal B, HER2-enriched, claudin-low, and basal-like, as well as a normal breast-like group. Patients with either luminal B, HER2-enriched, basal-like, or claudin-low tumors experience worse clinical outcome than patients with luminal A tumors (6–8).

ERα has been shown to negatively regulate gene expression, but not much is currently known on how it can achieve this. Epigenetic marks, namely DNA methylation of cytosine at specific CpG sites (5mCpG), are frequently associated with ERα+ status in human breast cancers. ERα may play a role in directing DNA methylation to target genes, as specific 5mCpG marks associate with ERα status in human breast cancer and predict risk of tumor recurrence (9–12).

Methylation of cytosine at CpG dinucleotide sites (5mCpG) by DNA methyltransferases (DNMT) in transcriptional regulatory regions mediates stable epigenetic gene silencing. In cancer cells, DNA methylation is highly correlated with repressive chromatin marks, such as trimethylated H3K27 (H3K27me3; ref. 13). H3K27 trimethylation is catalyzed by EZH2, the histone methyltransferase enzymatic subunit of the polycomb repressor complex 2 (PRC2; ref. 14). Together, EZH2 and PRC2 then recruit DNMTs (13, 15). Methylated CpG sites near transcriptional start sites (TSS) can silence gene expression by interacting with effectors, such as methyl-CpG–binding domain proteins, that impede binding of transcription factors, block transcriptional initiation, and recruit histone deacetylases (HDAC) to promote chromatin compaction (16).

The relationship between ERα and DNA methylation patterning in breast cancer has been reported. In a comprehensive bioinformatics study, methylation of CpG sites near ERα-binding regions tended to be lower in ERα+ tumors than ERα tumors. This indicated a passive role for ERα in preventing gene silencing. The methylation status of DNA sequences at ERα-binding sites is tightly coupled with ERα activity (12). Differentially methylated genes have also been identified in antihormone-resistant versus wild-type MCF-7 cells (17, 18), and in ERα RNAi-depleted versus nondepleted MCF-7 cells (19). Consistent with this notion, loss of ERα activity leads to silencing of estrogen-responsive genes, such as PgR (18, 19). Yet, ERα may also play an active role in promoting silencing. A functional link between ERα and DNA hypermethylation has been demonstrated at the CYP1A1 locus, whose gene product converts 17β-estradiol (E2) into a metabolite that inhibits proliferation; ERα silenced CYP1A1 by recruiting DNMT3B (20).

We sought to identify ERα targets for CpG methylation–mediated silencing by selecting the intersection of (i) genes upregulated (i.e., derepressed) by the demethylating agent decitabine (DAC); (ii) genes upregulated by loss of ERα expression in a series of antihormone-resistant T47D and MCF7 cell lines; and (iii) genes downregulated by E2 reexposure or increased ERα expression in antihormone-resistant T47D and MCF7 cells. Additional experiments verified the functional dependence on ERα for silencing and DNA methylation of the basal breast cancer subtype markers LCN2 and IFI27 in wild-type and antihormone-resistant T47D-based cell lines. Therefore, we show that ERα targets genes for DNA methylation–mediated silencing that may potentially be predictive biomarkers of breast cancer subtypes.

Cell lines

Sources and culture conditions of cell lines generated in this study are provided in Supplementary Materials and Methods. A schema representing the derivation of antihormone-resistant cells is shown in Supplementary Fig. S1. The fulvestrant (FUL)–resistant cell lines (T47D/FUL, MCF7/FUL) and the estrogen deprivation (ED)–resistant cell lines (T47D/ED1, T47D/ED2) were generated by continuous culture (8 weeks to >1 year) of wild-type T47D and MCF-7 cells in estrogenized media (RPMI1640 plus 10% whole FBS) supplemented with 100 nmol/L fulvestrant or in estrogen-free media (phenol red-free RPMI1640 plus 10% dextran-coated charcoal-stripped FBS), as appropriate. Antihormone-resistant cells were maintained as polyclonal populations. All cell lines were authenticated by gene expression microarrays, morphology, and by verifying ERα, PgR, HER2, LCN2, and IFI27 levels and cell line growth responses to estrogen, ED, and fulvestrant.

The lentiviral cell lines, T47D/ED1/VC, T47D/ED1/VC+E2, T47D/ED1/ERα, and T47D/ED1/ERα+E2 were generated by infecting ERα T47D/ED1 cells with an ERα-expressing lentivirus or an empty vector control (VC) lentivirus, as appropriate. Infected cells were maintained in estrogen-free or in 1 nmol/L E2–supplemented medium for 12 weeks. After initial recovery from infection and again 4 weeks later, infected cells were sorted for the lentiviral ZsGreen fluorescent marker using a Becton Dickinson FACS-VantageSE/DiVa cell sorter. To produce the lentiviral vectors, ERα's coding region was excised from pHEGO using EcoRI and inserted into the EcoRI site of the lentiviral vector pLVX-EF1α-IRES-ZsGreen1 (Clontech Laboratories).

RNA isolation

RNA was purified using Qiagen's RNeasy Plus Kits. RNA samples were required to exhibit an RNA integrity number of 9.8 to 10.0 on an Agilent 2100 Bioanalyzer.

qRT-PCR assays

qRT-PCR assays were carried out as described previously (21) but using AMV First-Strand cDNA Kit, predesigned TaqMan assays, TaqMan Universal PCR Master Mix, and a 7900HT Fast Real-Time PCR system (Thermo Fisher Scientific). Data were analyzed by comparison with a serial dilution series of cDNA. All qPCR data represent the mean and SDs of three independent biological replicates and two technical replicates per biological replicate.

Agilent gene expression microarrays

Genome-wide RNA profiling was carried out by the Genomics Facility at Fox Chase Cancer Center (Philadelphia, PA) using Agilent's Human Gene Expression 4 × 44 K v2 oligonucleotide microarrays. RNA labeling (one-color cyanine 3-CTP), hybridization to the arrays, and quality assessment of hybridizations were performed according to the manufacturer's instructions.

Immunoblot analyses

Immunoblots were done as described previously (21) but using RIPA buffer and 40 μg protein per lane. Antibodies used are listed in Supplementary Materials and Methods. Blots were visualized using the Odyssey Infrared Imaging System (Li-Cor Biosciences).

DNA methylation analysis by pyrosequencing

Genomic DNA was isolated using the DNeasy Blood and Tissue Kit (Qiagen) and treated with bisulfite [EpiTect Bisulfite Conversion Kit (Qiagen)] to change unmethylated cytosine nucleotides to thymines. Pyrosequencing reactions were carried out at EpigenDx as a service using their predesigned assays. In pyrograms, the ratios of methylated cytosines to thymines (which represent unmethylated cytosines) are internally normalized values. All pyrosequencing data represent the mean and SD of 4 replicates.

Human breast cancer cohorts

Breast cancer data from The Cancer Genome Atlas (TCGA) project were downloaded via the International Cancer Genome Consortium (ICGC) data portal (https://dcc.icgc.org/releases/release_18/Projects/BRCA-US). Methylation data were retrieved for 1,013 patients, 967 of whom also had ERα status available. CpG differential methylation by ERα status was assessed as described in Supplementary Materials and Methods.

The gene expression microarray cohort comprising 2,116 breast cancers was previously constructed in ref. 22 from publicly available datasets and is described in Supplementary Materials and Methods. Breast cancer–intrinsic subtypes were previously classified as described in ref. 22.

Metagenes

To analyze the composite expression level of gene sets in a tumor, gene sets were represented as metagenes and metagene scores, or single number summary values were determined across the expression array breast cancer cohort. These scores represent a linear combination of expression values of each gene in the gene set in individual tumors. Metagene scores were generated by determining the first “principle component” or “eigenvector” of each gene set in each tumor using singular value decomposition (SVD). The eigenvector produced by SVD was rescaled to a rank-based score between 0 and 1, with 0 relating to the lowest composite expression value for a gene set, and 1 relating to the highest. Thus, metagene scores capture the majority of variation in gene expression that is common to the majority of genes in a gene set across a population of samples.

To construct the ERα DNA methylation metagene, Entrez gene identifiers were used to match the Agilent probes from the expression microarrays used in this study to the Affymetrix probe sets used in the combined breast cancer cohort. This resulted in matching 34 of 39 ERα DNA methylation genes (Supplementary Table S1). The ERα status–associated metagene consists of the 100 most differentially expressed genes between ER+ and ERα tumors in the 2,116 breast cancer cohort as determined using the “limma” package (23) for R software (www.r-project.org). The specific genes comprising each metagene are provided in Supplementary Excel File S1.

Accession numbers

Microarray data are deposited in the NCBI GEO repository with accession number GSE85536.

Statistical analyses

Expression array data were log2 transformed for all comparisons. Differentially expressed genes were identified by serial pairwise comparisons using SAM (24) at an FDR <5% and a 2-fold cutoff, except a 1.5-fold cutoff was used when comparing T47D/ED2/E2 versus T47D/ED2 cells because the ERα levels in these cells were <5% that of wild-type T47D cells (Fig. 1A). Gene enrichment in Supplementary Excel Files S4, S5, and S10 was assessed by one-way Fisher exact tests using the R software application. Associations between ERα DNA methylation metagene scores and distant metastasis–free patient survival (DMFS) were evaluated by Kaplan–Meier analysis and log-rank tests, and by univariate and multivariable Cox proportional hazards regression models as described previously (22). The additional covariates used were age at diagnosis, intrinsic subtype, ERα status, tumor size, and tumor grade. Statistical tests used in Figs. 5 and 6 are specified in the figure legends and were carried out using Prism v4.03 (GraphPad Software). Where specified, one-way tests were employed because gene expression and CpG methylation was assumed a priori to be inversely related.

Figure 1.

Candidate ERα targets for DNA methylation and ERα mRNA levels in the cell lines used to identify the targets. A, ERα mRNA levels in matched wild-type (wt) fulvestrant (FUL)-resistant, ED-resistant, and ED-resistant reexposed to E2 (ED/E2) cell line models at the indicated weeks (wk) of derivation. The selection process schema is shown in Supplementary Fig. S1. ERα mRNA levels normalized to TBP mRNA were measured by qRT-PCR. B, The 39 candidate ERα DNA methylation targets. Cell lines were transcriptionally profiled using Agilent Human Gene Expression 4 × 44 K v2 microarrays. Shown is the intersection of DAC-regulated genes and genes whose expression consistently showed an inverse relationship to ERα expression/activity across all wild-type and antihormone-resistant cell lines. Genes are ranked by their average fold increase in expression in T47D/FUL, T47D/ED1, and T47D/ED2 versus wild-type T47D cells. Note, profiles of T47D/ED2/E2 week 38 and not week 24 cells were compared against T47D/ED2 cells for significantly differentially expressed genes. Basal-up/luminal-down genes were established according to references in Supplementary Excel File S10.

Figure 1.

Candidate ERα targets for DNA methylation and ERα mRNA levels in the cell lines used to identify the targets. A, ERα mRNA levels in matched wild-type (wt) fulvestrant (FUL)-resistant, ED-resistant, and ED-resistant reexposed to E2 (ED/E2) cell line models at the indicated weeks (wk) of derivation. The selection process schema is shown in Supplementary Fig. S1. ERα mRNA levels normalized to TBP mRNA were measured by qRT-PCR. B, The 39 candidate ERα DNA methylation targets. Cell lines were transcriptionally profiled using Agilent Human Gene Expression 4 × 44 K v2 microarrays. Shown is the intersection of DAC-regulated genes and genes whose expression consistently showed an inverse relationship to ERα expression/activity across all wild-type and antihormone-resistant cell lines. Genes are ranked by their average fold increase in expression in T47D/FUL, T47D/ED1, and T47D/ED2 versus wild-type T47D cells. Note, profiles of T47D/ED2/E2 week 38 and not week 24 cells were compared against T47D/ED2 cells for significantly differentially expressed genes. Basal-up/luminal-down genes were established according to references in Supplementary Excel File S10.

Close modal

Identification of genes inversely correlated with ERα expression/activity

To identify ERα targets for DNA methylation–mediated silencing, we sought to find the intersection of genes that fulfilled three conditions: (i) those genes derepressed by loss of ERα expression; (ii) those genes resilenced by increased ERα activity or expression; and (iii) those genes derepressed by loss of DNA methylation.

To begin, breast cancer cell line models were developed that exhibited loss of ERα to enable subsequent identification of genes, which inversely correlated with ERα expression/activity. We elected not to use RNAi-based methods, as acute depletion of ERα in estrogen-dependent cells leads to widespread cell death (unpublished observation). Therefore, starting with wild-type ERα+ T47D and MCF-7 luminal breast cancer cells, a panel of ERα-low/negative T47D and MCF7 breast cancer cells was derived by long-term selection of cells in 100 nmol/L fulvestrant or in estrogen-free media for 8 weeks to greater than 1 year (schema in Supplementary Fig. S1). Thus fulvestrant-resistant (T47D/FUL, MCF7/FUL) and ED-resistant (T47D/ED1, T47D/ED2) cell lines were derived. ERα mRNA levels were measured by qRT-PCR (Fig. 1A). T47D/ED1 cells lost 99.9%, T47D/FUL and T47D/ED2 cells lost ≥95%, and MCF7/FUL cells (at week 8 of derivation) lost 90% of ERα mRNA compared with respective wild-type parental cells. Immunoblotting also demonstrated similar ERα protein losses (Fig. 4).

To determine global changes in gene expression, which correlated with loss of ERα expression, transcriptional profiling was performed using Agilent 4 × 44 K v2 oligonucleotide microarrays. ERα-low/negative cell lines, T47D/FUL, ED1, ED2, and MCF7/FUL (week 8), were compared against their respective wild-type parental T47D or MCF-7 cell line (four separate pairwise comparisons). Supplementary Excel Files S2 and S3 list the 324 and 153 significantly up- and downregulated genes, respectively. These differentially regulated genes were examined for enrichment of functional gene groups consistent with acquired antihormone resistance using one-way Fisher exact tests (tables and P values in Supplementary Excel Files S4 and S5; Supplementary Fig. S2). As expected, genes with ERα-binding sites were overrepresented. ERE-regulated genes were taken from previously published datasets (see Supplementary Fig. S3 and its legend for a list of genes and references). Importantly, basal markers were very significantly enriched among the upregulated genes, whereas luminal markers were very significantly enriched among the downregulated genes. Also observed was enrichment of cancer stem cell (CSC), epithelial–mesenchymal transition (EMT), and tumor suppressor genes (TSG; see the legend of Supplementary Fig. S3 for references used). Examples of key genes in these functional groups are shown in Supplementary Fig. S3. This indicates that the antihormone-resistant ERα-low/negative T47D and MCF-7 cells transitioned to a differentiation state similar to the basal-like and claudin-low breast cancer subtype. Such a change in differentiation has previously been observed in T47D tumors in vivo following antiestrogen treatment or estrogen withdrawal and termed “luminobasal” (25).

To further refine the list of ERα inversely correlated genes, T47D/ED2 cells were reexposed to E2 for 38 weeks, resulting in T47D/ED2/E2 cells. Interestingly, ERα RNA (Fig. 1) and protein levels (Fig. 4) never rebounded, indicating permanent ERα silencing as observed elsewhere (26). In fact, ERα RNA levels actually decreased ∼50% more; this likely reflected a known E2–ERα negative feedback regulatory loop indicative of ERα transcriptional activity (27). MCF/FUL cells were also further selected. These cells at week 8 of derivation showed 90% loss of ERα, but after 13 weeks of additional exposure to fulvestrant (total 21 weeks), ERα levels rebounded to wild-type cell levels (Figs. 1A and 4E). Transcriptional profiling showed increased expression of well-known E2-stimulated genes in T47D/ED2/E2 versus T47D/ED2 cells (e.g., PGR, CA12, ERBB4) and in MCF7/FUL week 21 versus week 8 cells (e.g., CXCL12, GREB1, ERBB4), as well as decreased expression of E2-repressed genes (e.g., OASL, C3; both cell lines). Furthermore, the expression pattern of many (but not all) basal and luminal, CSC, EMT, and TSGs reversed upon E2 reexposure in T47D/ED2/E2 or increased ERα expression in MCF7/FUL (week 21) cells compared with respective parental cells (Supplementary Fig. S3).

Taking into account all cell line transcriptional profiles, 161 genes were identified that consistently inversely related with ERα expression/activity, whereas only 9 genes were directly related (Supplementary Excel Files S6 and S7, respectively).

Candidate ERα DNA methylation targets

As ERα inversely related genes whose expression was regulated by DNA methylation were sought, genes upregulated by the DNA demethylating agent DAC were identified. Wild-type T47D cells were treated with 1 μmol/L DAC or control (CON)-treated for 96 hours and then transcriptionally profiled. This resulted in the identification of 1,049 genes (Supplementary Excel File S8).

Subsequently, the intersection of ERα inversely related genes and DAC-induced genes was determined. This intersection represented the set of genes that fulfilled the following criteria: (i) genes induced by DAC versus CON-treated wild-type T47D cells; (ii) genes upregulated in each of the ERα-low/negative cell lines, that is, T47D/FUL, T47D/ED1, and T47D/ED2 cells, versus wild-type T47D cells; (iii) genes downregulated by E2 in T47D/ED2/E2 versus T47D/ED2 cells; (iv) genes upregulated in ERα-low MCF7/FUL week 8 versus wild-type MCF-7 cells; and (v) genes downregulated in ERα+ MCF-7/FUL week 21 versus ERα-low MCF-7/FUL week 8 cells. These selection criteria pinpointed 39 high-value candidates for ERα-mediated silencing via DNA methylation (Fig. 1B; Supplementary Excel File S9).

Initially, these 39 candidate genes were evaluated for methylation in human breast cancer (Supplementary Table S1). Using TCGA-processed breast cancer methylation data, a set of 1,996 CpGs associated with these genes was identified. These CpG sites were assessed for differential methylation between ERα+ and ERα breast cancers using one-sided Wilcoxon rank sum tests adjusted for FDR. Using a permutation analysis to determine whether similar results could be achieved using 1,000 sets of 39 random genes, it was concluded that the candidate ERα DNA methylation targets tended to display higher methylation levels in ERα+ compared with ERα tumors than would be expected for an identically sized set of randomly selected genes (permutation-based P = 0.011).

Next, the 39 candidate ERα DNA methylation targets were analyzed for enrichment of the same gene groups as the ERα inversely related genes (tables and P values in Supplementary Excel File S10; Supplementary Fig. S2). Similar to the earlier results, the candidate methylation targets were enriched for genes with ERα-binding sites, basal markers, CSC-upregulated genes, EMT-upregulated genes, and TSGs. The candidate methylation targets were also enriched for EMT-downregulated genes, but there were almost twice as many EMT-upregulated genes than downregulated (13 vs. 7, respectively).

Expression analysis of the candidate ERα DNA methylation gene set in breast cancer

The candidate ERα DNA methylation target gene set was analyzed relative to other important tumor-related gene sets and clinical variables in a cohort of 2,116 breast cancers. Gene sets were represented as a composite entity termed an “expression metagene” and a single value summary of the gene set's expression level in an individual tumor as a “metagene score.” To enable evaluation of the distribution of gene set expression levels, metagene scores were used to divide the breast cancers in the cohort into “tertiles” (lowest 33%, middle 33%, highest 33%).

ERα DNA methylation metagene scores were plotted versus ERα status, an ERα status–associated metagene, breast cancer intrinsic subtypes, luminobasal signature metagenes, EMT metagenes and CD44+/CD24−/low CSC metagenes (Fig. 2). The ERα status–associated metagene encapsulated the 100 most differentially expressed probe sets between ERα-positive and -negative tumors in the 2,116 breast cancer cohort. Congruent with the original selection criteria, the ERα DNA methylation metagene showed a clear negative association with ERα status and 100 other ERα status–associated genes, with ERα+ tumors tending to have lower scores (and thus indicating lower levels of gene expression; Fig. 2A). With regard to intrinsic subtype, luminal A and B subtypes displayed the lowest ERα DNA methylation metagene scores, whereas the basal-like subtype exhibited the highest scores (Fig. 2B). This was consistent with enrichment of basal-up/luminal-down genes as previously noted. Furthermore, the ERα DNA methylation metagene clearly directly related to the luminobasal signature metagenes (Fig. 2C), suggesting the ERα DNA methylation targets program this type of change in differentiation. Again as expected from the enrichment analysis, ERα DNA methylation metagene scores were associated with CD44+/CD24−/low metagenes (Fig. 2D) and selectively with the EMT-upregulated metagene (Fig. 2E). This helps explain why the ERα DNA methylation metagene scores also associated with the claudin-low breast cancer subtype (Fig. 2B).

Figure 2.

Evaluation of the ERα DNA methylation target metagene in a 2,116 sample breast cancer cohort. A–E, ERα status and ERα status–associated metagene scores (A), intrinsic subtype (B), luminobasal metagene scores (C), CD44+/CD24−/low metagene scores (D), and EMT metagene scores (E) were plotted against ERα DNA methylation target metagene scores in the 2,116 sample breast cancer cohort. Metagene scores were used to divide the breast cancer samples into low, medium, and high tertiles. F, Kaplan–Meier survival curves of ERα DNA methylation metagene scores versus DMFS. Patients were grouped on the basis of metagene scores split at the 10th, 50th, and 90th percentile. Significance was assessed by log-rank tests. The number of patients in each group is shown in the legends; numbers of patients are less than the total cohort size of 2,116 due to missing survival data.

Figure 2.

Evaluation of the ERα DNA methylation target metagene in a 2,116 sample breast cancer cohort. A–E, ERα status and ERα status–associated metagene scores (A), intrinsic subtype (B), luminobasal metagene scores (C), CD44+/CD24−/low metagene scores (D), and EMT metagene scores (E) were plotted against ERα DNA methylation target metagene scores in the 2,116 sample breast cancer cohort. Metagene scores were used to divide the breast cancer samples into low, medium, and high tertiles. F, Kaplan–Meier survival curves of ERα DNA methylation metagene scores versus DMFS. Patients were grouped on the basis of metagene scores split at the 10th, 50th, and 90th percentile. Significance was assessed by log-rank tests. The number of patients in each group is shown in the legends; numbers of patients are less than the total cohort size of 2,116 due to missing survival data.

Close modal

The ERα DNA methylation metagene was next evaluated for predicting DMFS in the breast cancer cohort by Kaplan–Meier survival curves (Fig. 2F) and Cox proportional hazards regression models (Supplementary Table S4). In each analysis, patients were separated into two groups according to metagene scores split at the 10th, 50th, or 90th percentiles, and then the proportion of patients exhibiting DMFS in each group was plotted against time. Patients with metagene scores in the bottom 10th or top 90th percentile experienced significantly decreased DMFS. Likewise, univariate Cox proportional hazards regression analysis demonstrated that metagene scores split at the 10th and 90th percentiles associated with DMFS (P = 0.00003 and 0.035, respectively). ERα DNA methylation metagene scores split at the 10th percentile remained significantly associated with DMFS in a multivariable Cox proportional hazards regression model (P = 0.026), but not when split at the 90th percentile. These results suggested that some genes in the ERα DNA methylation metagene when expressed at low levels promoted poor DMFS, whereas others did so when expressed at high levels.

To determine which of the genes of the candidate ERα DNA methylation metagene when expressed at low or high levels may promote poor DMFS, Kaplan–Meier survival curves and univariate Cox proportional hazards regression analysis was conducted for each gene. On the basis of these analyses, the candidate ERα methylation targets were separated into low- and high-expression metagenes (defined in Supplementary Excel File S11). The low- and high-expression metagenes poorly correlated with each other, indicating they indeed likely represented different biological processes (Supplementary Table S5). Patients were then divided according to their tumor's low- and high-expression metagene scores split at the 50th percentile and evaluated for DMFS as before (Fig. 3A). The Kaplan–Meier plots showed clear separations of survival curves in which patients in the low expression metagene's bottom 50% group and patients in the high expression metagene's top 50% group displayed poor DMFS.

Figure 3.

Evaluation of the low- and high-expression ERα DNA methylation target metagenes in the combined breast cancer cohort. A, Kaplan–Meier survival curves of the low- and high-expression ERα DNA methylation metagenes. Genes were parsed into the low versus high metagene according to how each gene's expression level tended to associate with poor DMFS. In the survival curves, patients were grouped on the basis of metagene scores split at the 50th percentile. The log-rank test P value and the number of patients in each group are shown in the legends. B and C, Low-expression (B) and high-expression (C) ERα DNA methylation metagene scores plotted against tumor grade as well as proliferation, tumor suppressor, focal adhesion, and Th1 immune response metagene tertiles in the breast cancer cohort.

Figure 3.

Evaluation of the low- and high-expression ERα DNA methylation target metagenes in the combined breast cancer cohort. A, Kaplan–Meier survival curves of the low- and high-expression ERα DNA methylation metagenes. Genes were parsed into the low versus high metagene according to how each gene's expression level tended to associate with poor DMFS. In the survival curves, patients were grouped on the basis of metagene scores split at the 50th percentile. The log-rank test P value and the number of patients in each group are shown in the legends. B and C, Low-expression (B) and high-expression (C) ERα DNA methylation metagene scores plotted against tumor grade as well as proliferation, tumor suppressor, focal adhesion, and Th1 immune response metagene tertiles in the breast cancer cohort.

Close modal

Low- (Fig. 3B) and high-expression ERα DNA methylation metagenes (Fig. 3C) were next assessed for associations with various tumor-related metagenes in the breast cancer cohort. These results indicated that the low-expression metagene associated with tumor suppressor and focal adhesion gene expression in breast cancers. Accordingly, low levels of these types of genes would be predicted to promote metastasis. Conversely, the high-expression metagene associated with high-grade tumors, as well as proliferation and proinflammatory Th1 immune response gene expression in breast cancers; this would also promote poor DMFS. Supplementary Excel File S11 contains references that help provide a rationale for the segregation of genes into either the low- or high-expression metagenes.

Inverse relationship between LCN2 and IFI27 expression and ERα

IFI27 and LCN2 were the top two genes inversely related to ERα expression/activity in the T47D-based cell lines (Fig. 1B). Both LCN2 (5) and IFI27 (2) contain ERα-binding sites. Also, both are basal markers (7, 28, 29) and promote EMT (30, 31). Hence, LCN2 and IFI27 were selected for validation of ERα-dependent changes in expression and 5mCpG levels.

LCN2 mRNA and protein levels dramatically increased in a time-dependent manner after precipitous drops in ERα levels across all antihormone-resistant models (Fig. 4A–C, and E). Furthermore, LCN2 expression decreased in a time-dependent manner after extended E2 reexposure in T47D/ED2/E2 cells (Fig. 4D) and once ERα expression rebounded in MCF/FUL cells (Fig. 4E). LCN2 has previously been reported to upregulate the key EMT transcription factor slug (SNAI2; ref. 30); therefore, slug expression was examined. Across all antihormone-resistant models, changes in slug expression followed similar changes in LCN2, although in MCF7/FUL cells, slug induction was delayed until after LCN2 was upregulated from 8 to 12 weeks and silenced again due to ERα reexpression (Fig. 4B–E). Together, these results are consistent with LCN2 regulating slug expression and demonstrate that LCN2 may have promoted EMT via slug.

Figure 4.

LCN2 and IFI27 expression inversely relates to ERα expression/activity. A–E, Effect of long-term ED, fulvestrant (FUL), and E2 reexposure on ERα, LCN2, and slug expression. In T47D-based cell lines, both ED and fulvestrant led to silencing of ERα mRNA (A) and protein levels (B and C), followed by dramatic induction of LCN2 mRNA (A) and protein (B and C). In T47D/ED2/E2 versus T47D/ED2 cells, long-term E2 reexposure repressed LCN2 (D). In MCF7-based cell lines, fulvestrant exposure up to 12 weeks repressed ERα and induced LCN2 expression, but by 16 weeks of fulvestrant, ERα rebounded and LCN2 was again silenced (E). Slug protein levels increased or decreased after similar changes were seen in LCN2 expression (B–E). F, Effect of long-term ED and fulvestrant on IFI27 expression. IFI27 RNA was silenced in ERα+ wild-type T47D cells, highly upregulated in ERα-low/negative T47D/FUL and T47D/ED cells, and again silenced by reexposure to E2 in T47D/ED/E2 cells. RNA levels were measured by qRT-PCR and proteins levels by immunoblotting.

Figure 4.

LCN2 and IFI27 expression inversely relates to ERα expression/activity. A–E, Effect of long-term ED, fulvestrant (FUL), and E2 reexposure on ERα, LCN2, and slug expression. In T47D-based cell lines, both ED and fulvestrant led to silencing of ERα mRNA (A) and protein levels (B and C), followed by dramatic induction of LCN2 mRNA (A) and protein (B and C). In T47D/ED2/E2 versus T47D/ED2 cells, long-term E2 reexposure repressed LCN2 (D). In MCF7-based cell lines, fulvestrant exposure up to 12 weeks repressed ERα and induced LCN2 expression, but by 16 weeks of fulvestrant, ERα rebounded and LCN2 was again silenced (E). Slug protein levels increased or decreased after similar changes were seen in LCN2 expression (B–E). F, Effect of long-term ED and fulvestrant on IFI27 expression. IFI27 RNA was silenced in ERα+ wild-type T47D cells, highly upregulated in ERα-low/negative T47D/FUL and T47D/ED cells, and again silenced by reexposure to E2 in T47D/ED/E2 cells. RNA levels were measured by qRT-PCR and proteins levels by immunoblotting.

Close modal

Like LCN2, IFI27 mRNA expression was strikingly upregulated 270- to 1,900-fold across the ERα-low/negative T47D-based antihormone-resistant compared with wild-type cell lines and was dramatically repressed again upon E2 reexposure in T47D/ED2/E2 versus T47D/ED2 cells (Fig. 4F).

Direct relationship between methylation of LCN2 and IFI27 and ERα

Levels of selected 5mCpG sites near the TSSs of LCN2 and IFI27 were quantitated across the T47D-based models by pyrosequencing of bisulfite-treated gDNA. This analysis found LCN2 and IFI27 CpG sites to be significantly hypomethylated in ERα-low/negative T47D/FUL, T47D/ED1, and T47D/ED2 cells versus wild-type ERα+ T47D cells (Fig. 5A), and significantly hypermethylated upon E2 reexposure in T47D/ED2/E2 cells compared with parental T47D/ED2 cells (Fig. 5B). Therefore, LCN2 and IFI27 CpG methylation levels directly associated with ERα expression/activity.

Figure 5.

LCN2 and IFI27 CpG methylation levels are directly related to ERα expression/activity. A, Decreased LCN2 and IFI27 CpG methylation in ERα-low/negative cell lines compared with wild-type T47D cells. B, Increased LCN2 and IFI27 CpG methylation in E2 reexposed T47D/ED2/E2 compared with T47D/ED2 cells. C, ERα, PgR, LCN2, and IFI27 mRNA expression in lentiviral VC and ERα-infected cells. ERα and the ERα-responsive gene PgR were substantially upregulated, whereas LCN2 and IFI27 were downregulated in cells expressing lentiviral ERα and maintained in E2. RNA levels normalized to TBP were measured by qRT-PCR. D, Increased CpG methylation levels of LCN2 and IFI27 in lentiviral ERα compared with VC cells. A and C, Significance was assessed by repeated measures one-way ANOVA, followed by Dunnett multiple comparison tests for subgroup analysis. B, Significance was assessed by one-tailed paired t tests. Genomic DNA was bisulfite treated and methylation was quantitated by pyrosequencing. TSS, transcriptional start site.

Figure 5.

LCN2 and IFI27 CpG methylation levels are directly related to ERα expression/activity. A, Decreased LCN2 and IFI27 CpG methylation in ERα-low/negative cell lines compared with wild-type T47D cells. B, Increased LCN2 and IFI27 CpG methylation in E2 reexposed T47D/ED2/E2 compared with T47D/ED2 cells. C, ERα, PgR, LCN2, and IFI27 mRNA expression in lentiviral VC and ERα-infected cells. ERα and the ERα-responsive gene PgR were substantially upregulated, whereas LCN2 and IFI27 were downregulated in cells expressing lentiviral ERα and maintained in E2. RNA levels normalized to TBP were measured by qRT-PCR. D, Increased CpG methylation levels of LCN2 and IFI27 in lentiviral ERα compared with VC cells. A and C, Significance was assessed by repeated measures one-way ANOVA, followed by Dunnett multiple comparison tests for subgroup analysis. B, Significance was assessed by one-tailed paired t tests. Genomic DNA was bisulfite treated and methylation was quantitated by pyrosequencing. TSS, transcriptional start site.

Close modal

Next, a causal relationship between ERα expression and CpG methylation of LCN2 and IFI27 was tested. ERα T47D/ED1 cells were infected with an ERα-expressing lentivirus or an empty VC lentivirus generating T47D/ED1/ERα and T47D/ED1/VC cells, respectively. These infected cells were maintained with and without E2 for 12 weeks and subjected to two rounds of cell sorting for the lentiviral ZsGreen fluorescent marker. Characterization of these lentiviral cells lines demonstrated functional ERα signaling as ERα and PgR mRNAs were expressed at high levels in ERα-infected cells. Also, LCN2 and IFI27 mRNA levels were downregulated in an ERα-dependent manner (plus E2 for IFI27; Fig. 5C). Next, LCN2 and IFI27 CpG methylation levels were quantitated by pyrosequencing and found to be significantly increased in lentiviral ERα plus E2 (LCN2) or just ERα (IFI27) compared with VC cells. (Fig. 5D). Therefore, increased CpG methylation of LCN2 and IFI27 was dependent on ERα plus E2 stimulation. In the case of IFI27, repression of its expression did not occur until its CpG methylation levels were maximally increased by the presence of E2, indicating that perhaps a methylation threshold was needed to cause its repression.

LCN2 and IFI27 expression and CpG methylation in breast cancer cell lines

LCN2 and IFI27 were examined in a panel of 11 breast cancer cell lines. Initial characterization showed 4 cell lines were ERα+ and 4 were HER2+, where only BT-474 cells were positive for both prognosticators (Fig. 6A). Then, LCN2 protein (Fig. 6B) and IFI27 mRNA levels (Fig. 6C) were measured and found to be significantly lower in ERα-positive compared with ERα cells. Next, methylation levels of CpG sites near the TSSs of LCN2 and IFI27 were quantitated by pyrosequencing of bisulfite-treated gDNA. Correlations between expression and 5mCpG levels were determined by Spearman ρ (Fig. 6D). Methylation of all 5 of LCN2's tested CpG sites and 4 of IFI27′s 9 tested sites (CpG sites +404, +438, +508, and +550) significantly correlated with each respective gene's expression levels. Finally, an association between CpG methylation levels and ERα status was evaluated. Although all 5 of LCN2′s CpGs were evaluated, only those 4 CpGs of IFI27 that significantly correlated with expression were considered. This analysis showed for both LCN2 and IFI27, that 5mCpG levels were significantly higher in ERα+ than ERα cells (Fig. 6E).

Figure 6.

LCN2 and IFI27 expression inversely associates, whereas CpG methylation directly associates with ERα status in breast cancer cell lines. A, Characterization of HER2 and ERα protein expression. B and C, LCN2 protein (B) and IFI27 RNA (C) expression levels. For both LCN2 and IFI27, expression levels were scaled relative to their median value (ZR751 cells for LCN2, and T47D cells for IFI27). LCN2 and IFI27 expression inversely associated with ERα+ status. For both genes, expression values were log2 transformed because their variances were significantly different between ERα+ and ERα cell lines (both P < 0.0001, F test). Significance was assessed using two-tailed unpaired t tests. HER2, ERα, and LCN2 protein levels were measured by immunoblotting and IFI27 mRNA levels by qRT-PCR. D, Correlation between CpG methylation and expression in breast cancer cell lines. Correlations were determined using Spearman ρ coefficient and a one-tailed P value. E,LCN2 and IFI27 CpG methylation levels positively associated with ERα status. Only those CpG sites that showed a significant inverse correlation between methylation and gene expression by Spearman ρ were evaluated for an association with ERα status. Significance was assessed considering all tested CpG sites together using two-tailed paired t tests in which CpG methylation levels were paired by site location. Individual CpG sites are presented to show pairings. The line in the ERα-positive and -negative subgroups represents the mean methylation value. Methylation levels were quantitated by pyrosequencing of bisulfite-treated gDNA.

Figure 6.

LCN2 and IFI27 expression inversely associates, whereas CpG methylation directly associates with ERα status in breast cancer cell lines. A, Characterization of HER2 and ERα protein expression. B and C, LCN2 protein (B) and IFI27 RNA (C) expression levels. For both LCN2 and IFI27, expression levels were scaled relative to their median value (ZR751 cells for LCN2, and T47D cells for IFI27). LCN2 and IFI27 expression inversely associated with ERα+ status. For both genes, expression values were log2 transformed because their variances were significantly different between ERα+ and ERα cell lines (both P < 0.0001, F test). Significance was assessed using two-tailed unpaired t tests. HER2, ERα, and LCN2 protein levels were measured by immunoblotting and IFI27 mRNA levels by qRT-PCR. D, Correlation between CpG methylation and expression in breast cancer cell lines. Correlations were determined using Spearman ρ coefficient and a one-tailed P value. E,LCN2 and IFI27 CpG methylation levels positively associated with ERα status. Only those CpG sites that showed a significant inverse correlation between methylation and gene expression by Spearman ρ were evaluated for an association with ERα status. Significance was assessed considering all tested CpG sites together using two-tailed paired t tests in which CpG methylation levels were paired by site location. Individual CpG sites are presented to show pairings. The line in the ERα-positive and -negative subgroups represents the mean methylation value. Methylation levels were quantitated by pyrosequencing of bisulfite-treated gDNA.

Close modal

LCN2 and IFI27 expression as predictors of DMFS in human breast cancer

LCN2 and IFI27 RNA expression levels were examined in the breast cancer cohort with respect to DMFS by Cox proportional hazards regression analysis (Supplementary Table S6). In univariate models, LCN and IFI27 both significantly associated with DMFS (P = 0.040 and 0.0023, respectively), but this did not hold in multivariable models.

We hypothesized that ERα may regulate gene expression in part via DNA methylation as methylation of specific CpG sites associates with ERα+ status in human breast cancer. This hypothesis was tested by identifying genes normally silenced in ERα+ breast cancer cell lines but which were derepressed upon exposure to the demethylating agent DAC, derepressed upon long-term loss of ERα expression, and resuppressed by gain of ERα activity/expression. On the basis of these criteria, 39 candidate ERα DNA methylation targets were found. These 39 targets were used to construct an ERα DNA methylation metagene that inversely associated with ERα status in human breast cancers and directly associated with expression signatures of basal-like and claudin-low breast cancer subtypes (25). Congruent with these associations, the candidate ERα DNA methylation targets were enriched for basal markers, CSC, and EMT genes.

LCN2 and IFI27 were the top two ERα inversely related genes identified and were selected for validation. Both LCN2 (5) and IFI27 contain ERα-binding sites (2), are basal markers (7, 28, 29), and involved in EMT (30, 31). First, LCN2 and IFI27 were originally silenced in wild-type T47D and MCF-7 cells, but their expression dramatically increased upon loss of ERα, while their 5mCpG levels significantly decreased in all antihormone-resistant T47D cell lines. Second, LCN2 and IFI27 were resilenced upon E2 reexposure in T47D/ED2 cells while their 5mCpG levels increased. Third, lentiviral ERα plus E2 in T47D/ED1 cells also repressed LCN2 and IFI27 expression while increasing their 5mCpG levels. Fourth, LCN2′s and IFI27′s 5mCpG levels positively associated with ERα status but inversely correlated with expression in a panel of 11 breast cancer cell lines. Together, these results provide correlative and functional evidence that ERα directed DNA methylation–mediated silencing of LCN2 and IFI27.

As ERα plays such a pivotal role in a more favorable outcome in breast cancer, genes targeted by ERα for DNA methylation–mediated silencing likely play important roles in disease progression. In addition to the CSC and EMT genes, Kaplan–Meier survival curve analyses indicated that the candidate ERα DNA methylation targets consisted of two classes of genes that predicted poor DMFS, one when expressed at low levels and a second when expressed at high levels. The low expression class associated with tumor suppressor and focal adhesion gene expression in breast cancer. Conversely, the high-expression class associated with proliferation and inflammatory response gene expression in breast cancer. In addition, the two validated targets for methylation, LCN2 and IFI27, predicted DMFS in univariate Cox proportional hazards models. Thus, genes methylated and silenced in an ERα-dependent manner may be good targets for therapeutic intervention in ERα breast cancer where they are expressed.

How might ERα direct DNA methylation to specific genes? We propose it may begin with transcriptional repression. E2 actually represses transcription of more genes than it stimulates (3, 4). Studies on E2-dependent transcriptional repression have demonstrated that ERα recruits coregulators (corepressors; refs. 32–36) and coactivators that act as corepressors (36, 37). The coregulators serve as scaffolds to interact with HDACs and a host of additional cofactors (32, 33, 38, 39), such as EZH2 (34, 40), that together remove activating histone marks, add repressive marks, and restructure chromatin structure (38, 39, 41–43). We hypothesize that not only does ERα direct epigenetic silencing via histone modification, but also via cytosine methylation at CpG sites.

Genome-wide kinetics of DAC-induced DNA demethylation and subsequent remethylation after drug withdrawal in breast cancer cells showed that CpGs differ in both their susceptibility to demethylation and propensity for remethylation after drug removal (44).

This is also plausible as EZH2 recruits DNMTs directly and indirectly through PRC2 (13, 15). Other protein–protein interactions exist as well that could support formation of a multicomponent complex containing ERα and DNMTs, such as those between ERα and EZH2 (45, 46), between HDACs and DNMTs (47–49), and between corepressors and DNMTs (40, 49). Evidence for such a complex exists at least at the CYP1A1 promoter, where it was demonstrated that ERα and DNMT3B interacted (20). Thus, it is possible that ERα could silence targeted genes via DNA methylation by directly and indirectly recruiting corepressors, HDACs, EZH2 in PRC2, and DNMTs (model shown in Fig. 7).

Figure 7.

Model of ERα-mediated silencing via DNA methylation. ERα initially represses transcription by recruiting corepressors (CoR) and HDACs that in turn recruit EZH2 to modify histones with repressive H3K27me3 marks. EZH2, a component of PRC2, then tethers DNMTs to catalyze C methylation in CpG dinucleotides to maintain long-term silencing. Gene silencing of basal markers, CSC, and EMT genes may in part program ERα breast cancers as the luminal subtype. See Discussion for additional details.

Figure 7.

Model of ERα-mediated silencing via DNA methylation. ERα initially represses transcription by recruiting corepressors (CoR) and HDACs that in turn recruit EZH2 to modify histones with repressive H3K27me3 marks. EZH2, a component of PRC2, then tethers DNMTs to catalyze C methylation in CpG dinucleotides to maintain long-term silencing. Gene silencing of basal markers, CSC, and EMT genes may in part program ERα breast cancers as the luminal subtype. See Discussion for additional details.

Close modal

Taken together, our data indicate that ERα can silence genes via DNA methylation, such as LCN2 and IFI27. Moreover, ERα may direct DNA methylation–mediated silencing of a subpopulation of basal markers, CSC, and EMT genes that may potentially enforce luminal differentiation of breast cancer cells.

No potential conflicts of interest were disclosed.

Conception and design: E.A. Ariazi, J. Boyd

Development of methodology: E.A. Ariazi, J.C. Taylor

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.A. Ariazi, J.C. Taylor, E. Nicolas, J. Boyd

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.A. Ariazi, M.A. Black, M.J. Slifker

Writing, review, and/or revision of the manuscript: E.A. Ariazi, M.A. Black, M.J. Slifker, D.J. Azzam, J. Boyd

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.A. Ariazi, J.C. Taylor, M.J. Slifker

Study supervision: E.A. Ariazi, J. Boyd

The authors thank Dennis DeSimone and Trung Nguyen, clinical fellows in the laboratory for technical support. The authors also thank the Expression Microarray facility, the Genotyping and Real-Time PCR facility, and the Flow Cytometry facility at Fox Chase Cancer Center for technical support.

This work was supported by the Commonwealth Universal Research Enhancement (CURE) Program Award from the Pennsylvania Department of Health (to J. Boyd) and NIHP30 CA006927 (Fox Chase Cancer Center Core Grant).

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.

1.
Ariazi
EA
,
Jordan
VC
. 
Estrogen receptors as therapeutic targets in breast cancer
.
In:
Ottow
E
,
Weinmann
H
,
editors
.
Nuclear receptors as drug targets. Methods and principles in medicinal chemistry
.
Mörlenbach, Germany
:
Wiley-VCH
; 
2008
.
p.
127
99
.
2.
Fullwood
MJ
,
Liu
MH
,
Pan
YF
,
Liu
J
,
Xu
H
,
Mohamed
YB
, et al
An oestrogen-receptor-alpha-bound human chromatin interactome
.
Nature
2009
;
462
:
58
64
.
3.
Lin
CY
,
Vega
VB
,
Thomsen
JS
,
Zhang
T
,
Kong
SL
,
Xie
M
, et al
Whole-genome cartography of estrogen receptor alpha binding sites
.
PLoS Genet
2007
;
3
:
e87
.
4.
Frasor
J
,
Danes
JM
,
Komm
B
,
Chang
KC
,
Lyttle
CR
,
Katzenellenbogen
BS
. 
Profiling of estrogen up- and down-regulated gene expression in human breast cancer cells: insights into gene networks and pathways underlying estrogenic control of proliferation and cell phenotype
.
Endocrinology
2003
;
144
:
4562
74
.
5.
Seth
P
,
Porter
D
,
Lahti-Domenici
J
,
Geng
Y
,
Richardson
A
,
Polyak
K
. 
Cellular and molecular targets of estrogen in normal human breast tissue
.
Cancer Res
2002
;
62
:
4540
4
.
6.
Perou
CM
,
Borresen-Dale
AL
. 
Systems biology and genomics of breast cancer
.
Cold Spring Harb Perspect Biol
2011
;
3
:
pii:a003293
.
7.
Prat
A
,
Parker
JS
,
Karginova
O
,
Fan
C
,
Livasy
C
,
Herschkowitz
JI
, et al
Phenotypic and molecular characterization of the claudin-low intrinsic subtype of breast cancer
.
Breast Cancer Res
2010
;
12
:
R68
.
8.
Prat
A
,
Perou
CM
. 
Deconstructing the molecular portraits of breast cancer
.
Mol Oncol
2011
;
5
:
5
23
.
9.
Ronneberg
JA
,
Fleischer
T
,
Solvang
HK
,
Nordgard
SH
,
Edvardsen
H
,
Potapenko
I
, et al
Methylation profiling with a panel of cancer related genes: association with estrogen receptor, TP53 mutation status and expression subtypes in sporadic breast cancer
.
Mol Oncol
2011
;
5
:
61
76
.
10.
Fackler
MJ
,
Umbricht
CB
,
Williams
D
,
Argani
P
,
Cruz
LA
,
Merino
VF
, et al
Genome-wide methylation analysis identifies genes specific to breast cancer hormone receptor status and risk of recurrence
.
Cancer Res
2011
;
71
:
6195
207
.
11.
Hill
VK
,
Ricketts
C
,
Bieche
I
,
Vacher
S
,
Gentle
D
,
Lewis
C
, et al
Genome-wide DNA methylation profiling of CpG islands in breast cancer identifies novel genes associated with tumorigenicity
.
Cancer Res
2011
;
71
:
2988
99
.
12.
Ung
M
,
Ma
X
,
Johnson
KC
,
Christensen
BC
,
Cheng
C
. 
Effect of estrogen receptor alpha binding on functional DNA methylation in breast cancer
.
Epigenetics
2014
;
9
:
523
32
.
13.
Schlesinger
Y
,
Straussman
R
,
Keshet
I
,
Farkash
S
,
Hecht
M
,
Zimmerman
J
, et al
Polycomb-mediated methylation on Lys27 of histone H3 pre-marks genes for de novo methylation in cancer
.
Nat Genet
2007
;
39
:
232
6
.
14.
Kim
KH
,
Roberts
CW
. 
Targeting EZH2 in cancer
.
Nat Med
2016
;
22
:
128
34
.
15.
Vire
E
,
Brenner
C
,
Deplus
R
,
Blanchon
L
,
Fraga
M
,
Didelot
C
, et al
The Polycomb group protein EZH2 directly controls DNA methylation
.
Nature
2006
;
439
:
871
4
.
16.
Kulis
M
,
Esteller
M
. 
DNA methylation and cancer
.
Adv Genet
2010
;
70
:
27
56
.
17.
Fan
M
,
Yan
PS
,
Hartman-Frey
C
,
Chen
L
,
Paik
H
,
Oyer
SL
, et al
Diverse gene expression and DNA methylation profiles correlate with differential adaptation of breast cancer cells to the antiestrogens tamoxifen and fulvestrant
.
Cancer Res
2006
;
66
:
11954
66
.
18.
Stone
A
,
Valdes-Mora
F
,
Gee
JM
,
Farrow
L
,
McClelland
RA
,
Fiegl
H
, et al
Tamoxifen-induced epigenetic silencing of oestrogen-regulated genes in anti-hormone resistant breast cancer
.
PLoS One
2012
;
7
:
e40466
.
19.
Leu
YW
,
Yan
PS
,
Fan
M
,
Jin
VX
,
Liu
JC
,
Curran
EM
, et al
Loss of estrogen receptor signaling triggers epigenetic silencing of downstream targets in breast cancer
.
Cancer Res
2004
;
64
:
8184
92
.
20.
Marques
M
,
Laflamme
L
,
Gaudreau
L
. 
Estrogen receptor alpha can selectively repress dioxin receptor-mediated gene expression by targeting DNA methylation
.
Nucleic Acids Res
2013
;
41
:
8094
106
.
21.
Ariazi
EA
,
Kraus
RJ
,
Farrell
ML
,
Jordan
VC
,
Mertz
JE
. 
Estrogen-related receptor alpha1 transcriptional activities are regulated in part via the ErbB2/HER2 signaling pathway
.
Mol Cancer Res
2007
;
5
:
71
85
.
22.
Nagalla
S
,
Chou
JW
,
Willingham
MC
,
Ruiz
J
,
Vaughn
JP
,
Dubey
P
, et al
Interactions between immunity, proliferation and molecular subtype in breast cancer prognosis
.
Genome Biol
2013
;
14
:
R34
.
23.
Ritchie
ME
,
Phipson
B
,
Wu
D
,
Hu
Y
,
Law
CW
,
Shi
W
, et al
limma powers differential expression analyses for RNA-sequencing and microarray studies
.
Nucleic Acids Res
2015
;
43
:
e47
.
24.
Tusher
VG
,
Tibshirani
R
,
Chu
G
. 
Significance analysis of microarrays applied to the ionizing radiation response
.
Proc Natl Acad Sci U S A
2001
;
98
:
5116
21
.
25.
Haughian
JM
,
Pinto
MP
,
Harrell
JC
,
Bliesner
BS
,
Joensuu
KM
,
Dye
WW
, et al
Maintenance of hormone responsiveness in luminal breast cancers by suppression of Notch
.
Proc Natl Acad Sci U S A
2012
;
109
:
2742
7
.
26.
Pink
JJ
,
Bilimoria
MM
,
Assikis
J
,
Jordan
VC
. 
Irreversible loss of the oestrogen receptor in T47D breast cancer cells following prolonged oestrogen deprivation
.
Br J Cancer
1996
;
74
:
1227
36
.
27.
Ellison-Zelski
SJ
,
Solodin
NM
,
Alarid
ET
. 
Repression of ESR1 through actions of estrogen receptor alpha and Sin3A at the proximal promoter
.
Mol Cell Biol
2009
;
29
:
4949
58
.
28.
Neve
RM
,
Chin
K
,
Fridlyand
J
,
Yeh
J
,
Baehner
FL
,
Fevr
T
, et al
A collection of breast cancer cell lines for the study of functionally distinct cancer subtypes
.
Cancer Cell
2006
;
10
:
515
27
.
29.
Charafe-Jauffret
E
,
Ginestier
C
,
Monville
F
,
Finetti
P
,
Adelaide
J
,
Cervera
N
, et al
Gene expression profiling of breast cell lines identifies potential new basal markers
.
Oncogene
2006
;
25
:
2273
84
.
30.
Yang
J
,
Bielenberg
DR
,
Rodig
SJ
,
Doiron
R
,
Clifton
MC
,
Kung
AL
, et al
Lipocalin 2 promotes breast cancer progression
.
Proc Natl Acad Sci U S A
2009
;
106
:
3913
8
.
31.
Li
S
,
Xie
Y
,
Zhang
W
,
Gao
J
,
Wang
M
,
Zheng
G
, et al
Interferon alpha-inducible protein 27 promotes epithelial-mesenchymal transition and induces ovarian tumorigenicity and stemness
.
J Surg Res
2015
;
193
:
255
64
.
32.
Augereau
P
,
Badia
E
,
Balaguer
P
,
Carascossa
S
,
Castet
A
,
Jalaguier
S
, et al
Negative regulation of hormone signaling by RIP140
.
J Steroid Biochem Mol Biol
2006
;
102
:
51
9
.
33.
Gurevich
I
,
Flores
AM
,
Aneskievich
BJ
. 
Corepressors of agonist-bound nuclear receptors
.
Toxicol Appl Pharmacol
2007
;
223
:
288
98
.
34.
Hwang
C
,
Giri
VN
,
Wilkinson
JC
,
Wright
CW
,
Wilkinson
AS
,
Cooney
KA
, et al
EZH2 regulates the transcription of estrogen-responsive genes through association with REA, an estrogen receptor corepressor
.
Breast Cancer Res Treat
2008
;
107
:
235
42
.
35.
Ye
Y
,
Xiao
Y
,
Wang
W
,
Yearsley
K
,
Gao
JX
,
Barsky
SH
. 
ERalpha suppresses slug expression directly by transcriptional repression
.
Biochem J
2008
;
416
:
179
87
.
36.
Stossi
F
,
Madak-Erdogan
Z
,
Katzenellenbogen
BS
. 
Estrogen receptor alpha represses transcription of early target genes via p300 and CtBP1
.
Mol Cell Biol
2009
;
29
:
1749
59
.
37.
Cvoro
A
,
Tzagarakis-Foster
C
,
Tatomer
D
,
Paruthiyil
S
,
Fox
MS
,
Leitman
DC
. 
Distinct roles of unliganded and liganded estrogen receptors in transcriptional repression
.
Mol Cell
2006
;
21
:
555
64
.
38.
Green
KA
,
Carroll
JS
. 
Oestrogen-receptor-mediated transcription and the influence of co-factors and chromatin state
.
Nat Rev Cancer
2007
;
7
:
713
22
.
39.
Manavathi
B
,
Dey
O
,
Gajulapalli
VN
,
Bhatia
RS
,
Bugide
S
,
Kumar
R
. 
Derailed estrogen signaling and breast cancer: an authentic couple
.
Endocr Rev
2013
;
34
:
1
32
.
40.
Kiskinis
E
,
Hallberg
M
,
Christian
M
,
Olofsson
M
,
Dilworth
SM
,
White
R
, et al
RIP140 directs histone and DNA methylation to silence Ucp1 expression in white adipocytes
.
EMBO J
2007
;
26
:
4831
40
.
41.
Mann
M
,
Cortez
V
,
Vadlamudi
RK
. 
Epigenetics of estrogen receptor signaling: role in hormonal cancer progression and therapy
.
Cancers
2011
;
3
:
1691
707
.
42.
Magnani
L
,
Lupien
M
. 
Chromatin and epigenetic determinants of estrogen receptor alpha (ESR1) signaling
.
Mol Cell Endocrinol
2014
;
382
:
633
41
.
43.
Osmanbeyoglu
HU
,
Lu
KN
,
Oesterreich
S
,
Day
RS
,
Benos
PV
,
Coronnello
C
, et al
Estrogen represses gene expression through reconfiguring chromatin structures
.
Nucleic Acids Res
2013
;
41
:
8061
71
.
44.
Bell
JS
,
Kagey
JD
,
Barwick
BG
,
Dwivedi
B
,
McCabe
MT
,
Kowalski
J
, et al
Factors affecting the persistence of drug-induced reprogramming of the cancer methylome
.
Epigenetics
2016
;
11
:
273
87
.
45.
Shi
B
,
Liang
J
,
Yang
X
,
Wang
Y
,
Zhao
Y
,
Wu
H
, et al
Integration of estrogen and Wnt signaling circuits by the polycomb group protein EZH2 in breast cancer cells
.
Mol Cell Biol
2007
;
27
:
5105
19
.
46.
Lee
ST
,
Li
Z
,
Wu
Z
,
Aau
M
,
Guan
P
,
Karuturi
RK
, et al
Context-specific regulation of NF-kappaB target gene expression by EZH2 in breast cancers
.
Mol Cell
2011
;
43
:
798
810
.
47.
Fuks
F
,
Burgers
WA
,
Brehm
A
,
Hughes-Davies
L
,
Kouzarides
T
. 
DNA methyltransferase Dnmt1 associates with histone deacetylase activity
.
Nat Genet
2000
;
24
:
88
91
.
48.
Fuks
F
,
Burgers
WA
,
Godin
N
,
Kasai
M
,
Kouzarides
T
. 
Dnmt3a binds deacetylases and is recruited by a sequence-specific repressor to silence transcription
.
EMBO J
2001
;
20
:
2536
44
.
49.
Cai
Y
,
Geutjes
EJ
,
de Lint
K
,
Roepman
P
,
Bruurs
L
,
Yu
LR
, et al
The NuRD complex cooperates with DNMTs to maintain silencing of key colorectal tumor suppressor genes
.
Oncogene
2014
;
33
:
2157
68
.

Supplementary data