Alterations in histones, chromatin-related proteins, and DNA methylation contribute to transcriptional silencing in cancer, but the sequence of these molecular events is not well understood. Here we demonstrate that on disruption of estrogen receptor (ER) α signaling by small interfering RNA, polycomb repressors and histone deacetylases are recruited to initiate stable repression of the progesterone receptor (PR) gene, a known ERα target, in breast cancer cells. The event is accompanied by acquired DNA methylation of the PR promoter, leaving a stable mark that can be inherited by cancer cell progeny. Reestablishing ERα signaling alone was not sufficient to reactivate the PR gene; reactivation of the PR gene also requires DNA demethylation. Methylation microarray analysis further showed that progressive DNA methylation occurs in multiple ERα targets in breast cancer genomes. The results imply, for the first time, the significance of epigenetic regulation on ERα target genes, providing new direction for research in this classical signaling pathway.
The steroid hormone estrogen is important for normal breast development, but it is also important for growth and progression of breast cancer. The molecular actions of estrogen are mediated by estrogen receptors (ERs), ERα and ERβ. On ligand binding, ERα functions as a transcription factor by either binding to DNA targets or tethering to other transcription factors, such as AP-1 and SP-1 (1) . These molecular interactions have been shown to positively or negatively modulate the activity of ERα downstream genes important to breast epithelial development.
It is known that estrogen signaling regulates the growth of some breast tumors, and antiestrogen therapies can effectively block this growth signaling, resulting in tumor suppression (2) . However, most tumors eventually develop resistance to this endocrine therapy, and antiestrogens are mostly ineffective in patients with advanced disease (2) . Mechanisms underlying this hormonal resistance are complex, involving intricate interactions between ERα and kinase networks (1 , 2) . In addition, epigenetic silencing of ERα is known to contribute to the antiestrogen resistance (1 , 2) . An emerging theme not yet investigated in this field is the subsequent influence on the expression of ERα downstream target genes.
Epigenetics can be defined as the study of heritable changes that modulate chromatin organization without altering the corresponding DNA sequence. DNA methylation, the addition of a methyl group to the fifth carbon position of a cytosine residue, occurs in CpG dinucleotides (3) and is a key epigenetic feature of the human genome. These dinucleotides are usually aggregated in stretches of 1- to 2-kb GC-rich DNA, called CpG islands, located in the promoter and first exon of ∼60% of human genes (3 , 4) . Promoter methylation is known to participate in reorganizing chromatin structure and also plays a role in transcriptional inactivation (3 , 5) . Studies have suggested that the CpG island in an active promoter is usually unmethylated, with the surrounding chromatin displaying an “open” configuration, allowing for the access of transcription factors and other coactivators to initiate gene expression (6, 7, 8) . Furthermore, transcription factor occupancy may make the promoter inaccessible to repressors or other chromatin-remodeling proteins. In contrast, the CpG island in an inactive promoter may become methylated, with the associated chromatin exhibiting a “closed” configuration. As a result, the methylated area is no longer accessible to transcription factors, disabling the functional activity of the promoter (7 , 9 , 10) .
Recent studies have shown that establishing transcriptional silencing of a gene involves a close interplay between DNA methylation and histone modifications (7 , 11) . This process may be achieved by recruiting histone-modifying enzymes, such as histone deacetylases, which mediate posttranslational modification at the NH2 terminus ends of histones (7 , 11) . As a result, chromatin modifications form distinct patterns, known as the “histone code,” that may dictate gene expression (12, 13, 14) .
Two models have been offered to describe the molecular sequence leading to the establishment of epigenetic gene silencing. One model suggests that histone modifications are the primary initiating event in transient repression (15 , 16) . DNA methylation subsequently accumulates in the targeted CpG island, creating a heterochromatin environment to establish a heritable, long-term state of transcriptional silencing. However, a second model is that DNA methylation can actually specify unique histone codes for maintaining the silenced state of a gene (17, 18, 19, 20) . In this case, DNA methylation may precede histone modifications. Clearly, this epigenetic process is complex, and multiple systems may be implemented for genes participating in different signaling pathways.
In this study, we investigated whether the removal of ERα signaling triggers changes in DNA methylation and chromatin structure of ERα target promoters. By using RNA interference (RNAi) to transiently disable ERα in breast cancer cells, we show, for the first time, that polycomb repressors and histone deacetylases assemble on the promoters of interrogated ERα target genes to participate in long-term transcriptional silencing. These events are later accompanied by a progressive accumulation of DNA methylation in the promoter regions of the now silent targets, leaving a heritable “mark” that may be stably transmitted to cell progeny.
MATERIALS AND METHODS
Cell Lines and Clinical Samples.
The breast cancer cell line MCF-7 and its derived subline, C4-12, were routinely maintained in our laboratories. For the demethylating treatment, cells were plated at a density of 2 × 106 cells per 10-cm dish and pretreated with 2 or 5 μmol/L 5-aza-2′-deoxycytidine (5-AzadC; Sigma, St. Louis, MO) for 5 days before treatment with 17β-estradiol (E2; 10 nmol/L, 24 hours). Thirty-two invasive ductal carcinomas were obtained from patients undergoing breast surgery at the Ellis Fischel Cancer Center (Columbia, MO), in compliance with the institutional review board. Seven tumor-free breast parenchymas were used as controls. The ER status of tumor tissue was determined by immunohistochemical staining (21) .
Transfection of Estrogen Receptor α Small Interfering RNAs.
MCF-7 cells (60% confluent in a 3.5-cm–diameter culture dish) were starved in serum-free medium (minimal essential medium only) for 72 hours, followed by the addition of 10 nmol/L E2 (E2758; Sigma) for 24 hours. The cells were then transfected with small interfering RNAs (siRNAs) for 4 to 5 hours with DMRIE-C reagent (Invitrogen, Carlsbad, CA). Double-stranded siRNA was generated using the Silencer siRNA Construction Kit (Ambion, Austin, TX). The siRNA oligonucleotides designed according to the ERα mRNA sequence (GenBank accession numbers AF_258449, 258450, and 258451) are as follows: (a) target sequence 1 (5′-AACCTCGGGCTGTGCTCTTTT), sense strand siRNA primer 5′-CCTCGGGCTGTGCTCTTTTTTCCTGTCTC and antisense strand siRNA primer AAAAGAGCACAGCCCGAGGTTCCTGTCTC; and (b) target sequence 17 (5′-AAACAGGAGGAAGAGCTGCCA), sense strand siRNA primer 5′-ACAGGAGGAAGAGCTGCCATTCCTGTCTC and antisense strand siRNA primer 5′-TGGCAGCTCTTCCTCCTGTTTCCTGTCTC.
Media were changed after transfection. The cells were then harvested for total RNA (RNeasy Kit; Qiagen, Valencia, CA) and genomic DNA (QIAamp; Qiagen) isolation at various time periods after siRNA treatment.
Transfection of Estrogen Receptor α Expression Vector.
C4-12 cells were transfected with pcDNA-ERα (C4-12/ER) or empty vector (C4-12/vec) using LipofectAMINE Plus Reagent (Life Technologies, Inc., Carlsbad, CA) and then exposed to an antibiotic (G418; 0.5 mg/mL) for 3 weeks. Expression of ERα in G418-resistant colonies was detected by immunoblotting with an anti-ER antibody (Chemicon, Temecula, CA).
Real-Time Reverse Transcription-Polymerase Chain Reaction.
Total RNA (2 μg) was treated with DNase I to remove potential DNA contamination and then reverse transcribed using the SuperScript II reverse transcriptase (Invitrogen). Real-time polymerase chain reactions (PCRs) were then performed using puReTaq Ready-To-Go PCR beads (Amersham Biosciences, Piscataway, NJ) and monitored by SYBR Green I (BioWhittaker, Walkersville, MD) using a Smart Cycler Real-Time PCR instrument (Cepheid, Sunnyvale, CA) for 42 cycles. PCR products of the expected size were also visualized on agarose gels stained with ethidium bromide. Alternatively, the reverse transcription-PCR (RT-PCR) reaction was conducted using iQ SYBR Green Supermix (Bio-Rad, Hercules, CA) in an iCycler system (Bio-Rad) for PR transcripts (22) . The relative mRNA level of a given locus was calculated by Relative Quantitation of Gene Expression (Applied Biosystems, Foster City, CA) with glyceraldehyde-3-phosphate dehydrogenase (GAPDH) or β-actin mRNA as an internal control. The primers used for RT-PCR reactions are listed in Supplementary Table S1.
Immunofluorescence and Western Blot Analysis.
MCF-7 cells (2 × 105) treated with or without ERα siRNAs were permeablized with 0.5% Nonidet P-40/PBS and blocked with a 1:100 dilution of horse serum before incubation with primary anti-ERα antibody (1:1,000; mouse monoclonal antibody D-12; Santa Cruz Biotechnology, Santa Cruz, CA). Sample slides were washed with PBS and incubated in the dark with secondary antibody (1:500) conjugated with Texas Red (fluorescent antimouse IgG kit; Vector Laboratories, Burlingame, CA) for 1 hour. The slides were then mounted with Vectashield mounting medium with 4′,6-diamidino-2-phenylindole (Vector Laboratories) and observed under a fluorescence microscope (Zeiss Axioskop 40; Zeiss, Thornwood, NY). Images were captured by the AxioCam HRC camera and analyzed by AxioVision 5.05 software.
Small interfering RNA-treated cells and control cells were lysed in the presence of proteinase inhibitors. One hundred micrograms of protein were subjected to 7% SDS-PAGE and transferred to immunoblot membranes. The membranes were then incubated with mouse anti-ERα (MAB463; Chemicon) and labeled secondary antibody. GAPDH was used a loading control.
Chromatin Immunoprecipitation-Polymerase Chain Reaction.
Cultured cells (2 × 106) were cross-linked with 1% formaldehyde and then washed with PBS in the presence of protease inhibitors. The cells were resuspended in lysis buffer, homogenized using a tissue grind pestle to release nuclei, and then pelleted by centrifugation. SDS-lysis buffer from a chromatin immunoprecipitation (ChIP) assay kit (Upstate Biotechnology, Lake Placid, NY) was used to resuspend the nuclei. The lysate was sonicated to shear chromatin DNAs and then centrifuged to remove cell debris. The supernatants were transferred to new tubes and incubated overnight with an antibody against ERα, YY-1, or EZH2 (Santa Cruz Biotechnology); HDAC1, MBD2, or MeCP2 (Upstate Biotechnology); and DMNT1, DNMT3a, or DNMT3b (Imgenex, San Diego, CA). Agarose slurry was then added to the mixture, and the chromatin-bound agarose was centrifuged. The supernatant was collected and used for total input (it serves as a positive control) in the ChIP-PCR assay. After elution, proteins were digested from the bound DNA with proteinase K. Phenol/chloroform-purified DNA was then precipitated and used in ChIP-PCR assays for a progesterone receptor (PR) promoter region. The primer sequences were 5′-GGCTTTGGGCGGGGCCTCCCTA (sense strand) and 5′-TCTGCTGGCTCCGTACTGCGG (antisense strand). After amplification, 32P-incorporated PCR products were separated on 8% polyacrylamide gels and subjected to autoradiography using a Storm PhosphorImager (Amersham Biosciences).
Methylation-Specific Polymerase Chain Reaction.
Genomic DNA (1 μg) from each sample was bisulfite-converted using the EZ DNA Methylation Kit (Zymo Research Corp., Orange, CA), according to the manufacturer’s protocol. The converted DNA was eluted with 40 μL of elution buffer and then diluted 50 times for methylation-specific PCR (MSP). The primer sets designed for amplifying the methylated or unmethylated allele of the PR locus are listed in Supplementary Table S2. All PCR reactions were performed in PTC-100 thermocyclers (MJ Research, Watertown, MA) using AmpliTaq Gold DNA polymerase (Applied Biosystems). 32P-incorporated amplified products were separated on 8% polyacrylamide gels and subjected to autoradiography using a Storm PhosphorImager (Amersham Biosciences).
Combined Bisulfite Restriction Analysis.
Combined bisulfite restriction analysis (COBRA) was carried out essentially as described previously (23) . Bisulfite-modified DNA (∼10 ng) was used as a template for PCR with specific primers flanking the interrogated sites (TaqI or BstUI) of an ERα downstream target. Primer sequences used for amplification are listed in Supplementary Table S3. After amplification, radiolabeled PCR products were digested with TaqI or BstUI, which restrict unconverted DNA containing methylated sites. The undigested control and digested DNA samples were run in parallel on polyacrylamide gels and subjected to autoradiography. The percentage of methylation was determined as the intensity of methylated fragments relative to the combined intensity of unmethylated and methylated fragments.
Chromatin Immunoprecipitation on Chip.
MCF-7 cells (2 × 107) were used to conduct ChIP with an antibody specific for ERα following the protocol described (see Chromatin Immunoprecipitation-Polymerase Chain Reaction). After chromatin coimmunoprecipitation, DNA was labeled with Cy5 fluorescence dye and hybridized to a genomic microarray panel containing ∼9,000 CpG islands (24) . Microarray hybridization and posthybridization washes have been described previously (25) . The washed slides were scanned by a Gene Pix 4000A scanner (Axon, Union City, CA), and the acquired microarray images were analyzed with GenePix Pro 4.0 software. This ChIP-on-chip experiment was conducted twice.
Positive CpG island clones were sequenced, and the derived sequences were used to identify putative transcription start sites by Blastn 5 or Blat. 6 Both Genomatrix 7 and TFSEARCH 8 programs were then used to localize the consensus sequences of the estrogen response elements (EREs) and other related transcription factor binding sites (AP-1, SP-1, cAMP-responsive element binding protein, and CEBP).
Differential Methylation Hybridization.
Differential methylation hybridization (DMH) was performed essentially as described previously (25 , 26) . Briefly, 2 μg of genomic DNA were digested by the 4-base frequent cutter MseI, which restricts bulk DNA into small fragments but retains GC-rich CpG island fragments (24) . H-24/H-12 PCR linkers (5′-AGGCAACTGTGCTATCCGAGGGAT-3′ and 5′-TAATCCCTCGGA-3′) were then ligated to the digested DNA fragments. The DNA samples were further digested with two methylation-sensitive endonucleases, HpaII and BstUI, and amplified by PCR reaction using H-24 as a primer. After amplification, test DNA from siRNA-treated cell lines or clinical samples was labeled with Cy5 (red) dye, whereas control DNA from the mock-transfected cell lines or normal female blood samples was coupled with Cy3 (green) dye. Equal amounts of test and control DNAs were cohybridized to a microarray slide containing 70 ERα promoter targets (average, 500 bp) identified from the ChIP-on-chip results. Posthybridization washing and slide scanning are described above. Normalized Cy5/Cy3 ratios of these loci were calculated by GenePix Pro 4.0.
Shrunken Centroids Analysis.
DMH microarray data were analyzed by the procedure described online. 9 This program incorporates graphic methods for automatic threshold choice and centroid classification.
Differences of methylation or mRNA levels in experimental studies were analyzed by a paired t test. Methylation differences between two tumor groups were determined with a Pearson’s χ2 test. P < 0.05 was considered statistically significant.
RNA Interference Transiently Knocks Down Estrogen Receptor α Expression in Breast Cancer Cells.
Although several in vitro systems and mouse models are available for analysis of estrogen signaling, to our knowledge, the recently described RNAi (27) has not been actively used in this area of research. We therefore used this technology to specifically repress ERα gene expression via targeted RNA degradation (28 , 29) . Six different ERα siRNAs, two of which have sequences homologous to a splice variant, were synthesized (Fig. 1A) ⇓ . These siRNAs (40 nmol/L) were individually transfected into MCF-7, an ERα-positive human breast cancer cell line. MCF-7 cells were cultured in the presence of E2. Quantitative RT-PCR analysis showed that, 24 hours after transfection, two siRNAs, siRNAs 1 and 17, were capable of repressing ERα transcripts (Fig. 1B) ⇓ . Specifically for siRNA 1, we observed a >93% decrease of ERα mRNA. Immunofluorescence (Fig. 1C) ⇓ and Western blot (Fig. 1D) ⇓ analyses confirmed that this RNAi also dramatically reduced ERα protein synthesis. This inhibitory effect appeared to be transient, and the expression of ERα protein reappeared in cultured cells 4 weeks after RNAi withdrawal (Fig. 1D) ⇓ .
Epigenetic Silencing of the PR Gene Is Triggered by Estrogen Signal Disruption.
We hypothesized that disruption of ERα signaling by siRNA may lead to the silencing of some positively regulated ERα targets governed by epigenetic mechanisms. To this end, a known ERα downstream target, the PR gene, was investigated in detail. In Fig. 2A and B ⇓ , quantitative RT-PCR analysis showed that by 36 hours after treatment of MCF-7 cells with siRNA 1, the level of PR transcripts (PR-A and PR-B) was reduced by >95% (paired t test, P < 0.0001). Next, ChIP-PCR was performed to determine the status of chromatin remodeling at the 5′-end of the PR gene. The protein-DNA complexes were immunoprecipitated with antibodies to ERα or to specific modified histones (acetyl-H3, acetyl-H3-K9, and methyl-H3-K4) known to specify active transcription (7 , 30) . As shown in Fig. 2C ⇓ , the presence of these active chromatin components was diminished over a period of 36 hours, coinciding with decreased ERα binding to the PR promoter region.
We speculated that this initial transcriptional inactivation might trigger further recruitment of repressor molecules to the PR promoter CpG island to subsequently establish a long-term silencing state. ChIP-PCR assays were conducted with a panel of antibodies raised for the polycomb repressors YY-1 and EZH2, histone deacetylase HDAC1, methyl-CpG–binding proteins MBD2 and MeCP2, and DNA methyltransferases DMNT1, DNMT3a, and DNMT3b. At 36 hours after siRNA treatment, YY-1 and EZH2 were bound to the promoter region (Fig. 2D) ⇓ . These polycomb proteins have previously been shown to target the regulatory regions of homeobox genes, the resulting repression of which can be tissue specific and important for early embryonic development (31 , 32) . Here we demonstrate for the first time that these proteins have an additional role in repressing an ERα target gene. Furthermore, the PR promoter was seen, at 36 hours, to recruit HDAC1 (Fig. 2D) ⇓ , a protein known to deacetylate histone protein tails, creating a repressive heterochromatin environment in the targeted promoter area (7 , 9 , 10) . However, at this early time point (36 hours), ChIP-PCR analysis did not detect the presence of the DNA methyltransferase DNMT1 in the PR promoter CpG island area (Fig. 2D) ⇓ , nor did we observe the presence of MBD2 or MeCP2, which are known to bind methylated CpG sites. Only a faint band corresponding to DNMT3b was detected in the PR promoter area by 36 hours after ERα siRNA treatment of MCF-7 cells. Except for DNMT3a, the recruitment of these repressive proteins to the PR promoter CpG island was evident by 168 hours after siRNA treatment.
To further determine whether the recruitment of these epigenetic components could trigger de novo DNA methylation, MSP assays (33) were conducted to survey two 5′-end regions of PR at different time periods after siRNA treatment. As shown in Fig. 3A ⇓ , PR methylation was detected in amplified bisulfite-treated DNA only at 36 hours after treatment. This observation was independently confirmed by conducting semiquantitative COBRA (23) . In the assay, ∼10% of MCF-7 cells showed methylation in one (Fig. 2A ⇓ , TaqI-2) of the two PR TaqI sites analyzed 36 hours after siRNA treatment (Fig. 3B) ⇓ . Both of these sites became methylated at a later time point (168 hours) of treatment (Fig. 3B) ⇓ . This study implies that acquired DNA methylation is a late event and that the density of DNA methylation may gradually accumulate at the 5′-end of PR after disrupting ERα signaling by siRNA.
Next, we determined whether this acquired promoter methylation could be observed in ERα-negative breast tumors. COBRA was therefore conducted in 32 primary tumors (16 ERα-negative and 16 ERα-positive tumors) and 7 normal controls (see representative examples in Fig. 3C ⇓ ). Consistent with the in vitro findings, PR promoter hypermethylation occurred more frequently in ERα-negative tumors (45%) than in ERα-positive tumors (10%) (χ2 test, P < 0.05).
Reexpression of PR Requires Both Estrogen Signal Restoration and DNA Demethylation.
The in vitro experimental results described above are based on transient siRNA treatment. To determine whether this signal disruption has a lasting impact on PR expression, we took advantage of an ERα-negative cell subline, C4-12, derived from ERα-positive MCF-7 cells by long-term hormonal depletion (34) . A recent study has indicated that PR gene expression is absent in this cell line (35) . We therefore examined whether stably reexpressing ERα could restore PR gene activity in several established C4-12 subclones (C4-12/vec, C4-12/ER#1, C4-12/ER#50, and C4-12/ER#86; see examples in Fig. 4A ⇓ , inset).
Treatment of these subclones (e.g., C4-12/ER#86 in Fig. 4A ⇓ ) with E2, however, failed to induce PR mRNA expression, demonstrating that reintroduction of ERα alone was insufficient to reactivate expression of a silent PR gene. To determine whether loss of PR expression was due to DNA methylation, C4-12/vec (i.e., cells stably transfected with empty vector) and C4-12/ER#86 cells were pretreated with 5-AzadC, a DNA demethylating agent, before E2 treatment. As shown in Fig. 4A ⇓ , sequential treatment with 5-AzadC followed by E2 resulted in reexpression of PR mRNA in C4-12/ER#86 cells, but not in C4-12/vec cells, demonstrating that both ERα expression and DNA demethylation are required to restore PR expression. To further confirm that reactivation of the PR gene was due to DNA demethylation, the methylation status of the PR promoter CpG island region was examined by MSP (Fig. 4B) ⇓ . In contrast to MCF-7 cells in which the PR promoter CpG island was unmethylated (Fig. 3A) ⇓ , methylation was observed in both C4-12/vec and C4-12/ER cells (Fig. 4B ⇓ , Lanes 1, 2, and 5–10). However, after treatment with 5-AzadC, PR promoter methylation was partially reversed in C4-12/vec cells (Fig. 4B ⇓ , Lanes 3 and 4) and completely removed in C4-12/ER#86 cells (Fig. 4B ⇓ , Lanes 11 and 12). Together, these results demonstrate that the silencing of PR is maintained, in part, by DNA methylation in the ERα-negative C4-12 cells and that reactivation of the PR promoter requires both the presence of ERα and DNA demethylation.
DNA Methylation of Multiple Estrogen Receptor α Downstream Targets Is Triggered by Disrupting Receptor Signaling.
To determine whether this epigenetically mediated silencing is a generalized event, we used ChIP-on-chip, a novel microarray-based method developed in our laboratory (36 , 37) , for a genome-wide screening of ERα downstream targets. In this case, we probed a panel of ∼9,000 arrayed CpG island fragments with anti-ERα–coimmuoprecipitated chromatins. Putative target sequences were used to search for the presence of ERα binding motifs, EREs, and other related binding sites (e.g., AP-1, SP-1, cAMP-responsive element binding protein, and CEBP) by using the Genomatrix 7 and TFSEARCH 8 programs. These computational algorithms identified a total of 70 unique ERα promoter targets, which were used to construct a subpanel genomic microarray (see a partial list of the genes in Supplementary Table S4). The previously described DMH method (25 , 26) was then used to determine the DNA methylation status of these ERα targets in siRNA-treated versus mock-treated MCF-7 cells. Amplicons representing genomic pools of methylated DNAs were prepared from these treated cells using our established protocols (25 , 26) . Cy5 (red dye)- and Cy3 (green dye)-labeled DNAs were prepared from siRNA- and mock-treated cells, respectively, and cohybridized to microscope slides containing the arrayed 70 unique ERα targets. ERα target loci methylated in siRNA-treated cells, but not in mock-treated cells, were expected to show greater Cy5/Cy3 hybridization signals. This is because methylated CpG sites are protected from methylation-sensitive restriction (i.e., HpaII and BstUI) and could thus be amplified by a linker-PCR approach during amplicon preparation. In contrast, unmethylated CpG sites were restricted by the methylation-sensitive enzymes, could not be amplified by PCR, and were thus devoid of hybridization signals.
To analyze our microarray data, we adapted the “shrunken centroids method” (38) to define the threshold setting for class prediction of methylated ERα target loci. This approach can be used to uniquely define the threshold level that statistically discriminates ERα loci commonly methylated in siRNA-treated cells from the same loci in mock-treated cells. After initial evaluation of the microarray data, we chose the threshold value 2.0 that generates less error (≤0.3) for cross-validation (data not shown). When the cross-validation variances from individual samples were plotted (Fig. 5A) ⇓ , many ERα target loci could be used to discriminate between siRNA-treated cells and mock-treated counterparts (manifested as having many loci with no misclassification error) at the 168 hour time point. However, this threshold level was not sufficiently stringent to discriminate between the mock- and siRNA-treated cell samples at 24 or 36 hours (manifested as having very few loci with low misclassification error). In Fig. 5B ⇓ , the actual methylation status of individual loci, in comparison with the predicted centroids, is plotted to present an overall change of DNA methylation at different time periods of siRNA treatment. Relative to the overall predicted centroids, a positive value of a locus indicates more methylation during the treatment, whereas a negative value indicates less methylation. This shrunken centroids map revealed that de novo DNA methylation can be detected in a subset of ERα targets 168 hours after siRNA treatment, but not in cells treated for only 24 or 36 hours after treatment.
To validate the findings of the shrunken centroid analysis, unsupervised cluster analysis was performed on the microarray data, using the top 21 methylated loci selected by machine training (“heat map” shown in Fig. 5C ⇓ ). The result reaffirms the shrunken centroid data in that replicates of each treatment type are clustered together and that the level of methylation increased with the extent of siRNA treatment. A paired t test revealed that the methylation status of these 21 loci was significantly different (P < 0.05) between the mock-treated (ERα-positive) and siRNA-treated (ERα-negative) cells.
This microarray observation was independently validated by conducting expression and DNA methylation analyses on three newly identified ERα downstream targets, TRIP10, Kr-Znf1, and DCC. In general, the decreased levels of these mRNAs preceded the emergence of DNA methylation at their respective promoter CpG islands (Fig. 6A and B) ⇓ . This epigenetically mediated silencing also indirectly influenced the expression of MTA3, a gene known to be regulated via a downstream ERα target and to participate in Mi-2/NuRD nucleosome remodeling (Fig. 6B ⇓ ; ref. 39 ).
DNA Methylation of ERα Downstream Targets is Preferentially Observed in ERα-Negative Tumors.
We next determined whether this in vitro finding could be seen in vivo. DMH was therefore conducted using the aforementioned 32 primary breast tumors and 7 normal controls. The derived microarray data were then analyzed by the shrunken centroid method. Although the methylation results of these 70 ERα target loci did not clearly segregate tumor samples into subclasses, we observed a general trend that methylated loci appear more frequently in ERα-negative tumors than in ERα-positive tumors (P < 0.05). Fig. 5D ⇓ presents a heat map of the 12 most methylated loci in the studied breast tumors. As shown, we observed higher overall methylation in the ERα-negative tumors (6 of 16 tumors had >40% methylation in the loci analyzed) than in the ERα-positive tumors (only 1 of 16 tumors achieved the same level of methylation). Also, the total number of loci showing DNA methylation was greater in ERα-negative tumors, when compared with ERα-positive tumors. Only four loci showed a low level of methylation in normal breast samples. Methylation analysis by MSP was further conducted for TRIP10 in these breast samples (Fig. 6C) ⇓ . Consistent with the microarray finding, TRIP10 promoter hypermethylation was detected in 50% (8 of 16) of ERα-negative tumors but in none of the 16 ERα-positive tumors analyzed (χ2 test, P < 0.005).
Understanding the sequence of how complex epigenetic events are established can provide important insights into the molecular mechanisms underlying gene silencing in cancer. However, the “chicken and egg” issue of which comes first, DNA methylation, histone modification, or others, is an ongoing debate in the epigenetic research community. Many early studies of this issue come from nonmammalian systems. Mutations in a histone methyltransferase specific for H3-K9 resulted in loss of DNA methylation in Neurospora crassa (15 , 16) , suggesting that histone methylation can initiate DNA methylation. In Arabidopsis, it has been shown that CpNpG methylation depends on a histone H3 methyltransferase (40) , also indicating that histone methylation can direct DNA methylation. New evidence suggests that the reverse scenario can occur in heterochromatin (41) . In this case, a self-reinforcing system is implemented, allowing for feedback from DNA methylation to histone methylation for the long-term maintenance of a heterochromatin state in a gene (41) . However, this epigenetic paradigm remains to be explored in mammalian systems. Earlier studies have shown that in vitro methylated transgenes can be targets for methyl-CpG–binding proteins, which in turn recruit repressor complexes containing histone deacetylases (17 , 18) . Fahrner et al. (19) suggested that DNA methylation of hMLH1 can specify unique histone codes for the maintenance of a silenced state. They detected methyl histone 3-lysine 9 in the DNA methylated, transcriptionally silenced promoter CpG island of hMLH1 in a cancer cell line. Treatment with the DNA demethylating agent 5-AzadC alone, but with not the histone deacetylase inhibitor trichostatin A, resulted in reversal of this repressive histone modification. Taken together, these reports, as well as other studies, imply that in contrast to other organisms, histone modifications may be secondary to DNA methylation in initiating gene silencing in mammalian cells (17 , 18 , 20 , 42) .
A study by Bachman et al. (43) , however, presents a different view with respect to the silencing of the p16 gene in an experimental system using somatic knockout cells. These authors suggest that chromatin modifications are not totally dependent on prior DNA methylation to initiate gene silencing. In support of this observation, Mutskov and Felsenfeld (44) have recently demonstrated that histone modifications are the primary event associated with the silencing of a transgene, ILR2. In this case, a gradual increase in DNA methylation density in and around the ILR2 promoter was observed after transfection. In contrast to previous observations, these two recent studies therefore suggest that DNA methylation sets up an epigenetic “mark” for the maintenance of long-term silencing, rather than initiating it. Clearly, this epigenetic process is complex and multifaceted, and it is possible that the sequence of epigenetic events for establishing and maintaining the silenced state of a gene can be locus or pathway specific.
The present study suggests that gene inactivation and histone modifications occur before DNA methylation at some ERα target loci. Depicted in Fig. 7 ⇓ is a hypothetical gene containing an ERE site within the promoter area, the active transcription of which is directly dependent on estrogen signaling. On the removal of this signaling, down-regulation of this gene occurs immediately. Transcriptional repressors (e.g., polycomb proteins) and histone deacetylases are then assembled to its promoter to initiate long-term transcriptional repression. Subsequent recruitment of DNA methyltransferases to the repressor complex methylates CpG sites in the adjacent area. This process may be gradual, with methylation density increasing over time in the targeted area (see the heat map in Fig. 5C ⇓ ). The buildup of DNA methylation could set up a heritable mark that may eventually replace some of the original repressors to establish a heterochromatin state of long-term silencing. In this case, reactivation of ERα target genes could no longer be achieved by reestablishing estrogen signaling alone (see the example of PR in Fig. 4A ⇓ ); it also requires DNA demethylation. In addition to the PR gene, we suggest that establishment of epigenetic memory may occur in other critical ERα downstream loci in some breast cancer cells.
The occurrence of DNA methylation in a pathway-specific manner also has a new implication. Altered DNA methylation was originally thought to be a generalized phenomenon arising from a stochastic process in earlier studies (45 , 46) . This random methylation in tumor suppressor genes at their promoter CpG islands, thus silencing their transcripts, would provide tumor cells with a growth advantage. The specific epigenetic patterns observed in particular cancer types would therefore be derived from clonal selection of the proliferating cells. Some studies (26 , 47 , 48) , however, have indicated that this epigenetic event is not random and that remodeling of the local chromatin structure of a gene may influence its susceptibility to specific DNA methylation. The present study provides some answers to this conundrum. Here we show that dysregulation of normal signaling in cancer cells may result in stable silencing of downstream targets, maintained by epigenetic machinery. This implies that the altered epigenetic condition is pathway specific, rather than a stochastic process in the ERα signaling pathway.
In conclusion, the present study implicates, for the first time, epigenetic influence (i.e., chromatin remodeling and DNA methylation) on transcription of ERα downstream target genes and thus provides a new direction for research in this classical signaling pathway. Unlike irreversible genetic damage, epigenetic alterations are potentially reversible, providing an opportunity for therapeutic intervention in breast cancer. Histone deacetylase inhibitors, alone or together with DNA demethylating agents, may represent novel treatment approaches that could be combined with currently available chemotherapies. Our experimental evidence therefore provides a rationale for such treatment strategies designed to alter aberrant epigenetic processes in hormone-insensitive but receptor-positive breast tumors.
The authors wish to thank Drs. Curt Balch and Phil Abbosh (Bloomington, IN) and Diane Peckham (Columbia, MO) for constructive review of the manuscript.
Grant support: National Cancer Institute grants R01 CA-69065 (T. Huang) and R01 CA-85289 (K. Nephew); United States Army Medical Research Acquisition Activity, Award Numbers DAMD 17-02-1-0418 and DAMD 17-02-1-0419 (K. Nephew); American Cancer Society Research and Alaska Run for Woman Grant TBE-104125 (K. Nephew); and funds from The Ohio State University Comprehensive Cancer Center-Arthur G. James Cancer Hospital and Richard J. Solove Research Institute (P. Yan and T. Huang).
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.
Note: Supplementary data for this article can be found at Cancer Research Online (http://cancerres.aacrjournals.org). T. Huang is a consultant to Epigenomics, Inc., Berlin, Germany.
Requests for reprints: Tim H-M. Huang, Human Cancer Genetics Program, Department of Molecular Virology, Immunology, and Medical Genetics, Comprehensive Cancer Center, The Ohio State University, 420 West 12th Avenue, Columbus, OH 43210. Phone: 614-688-8277; Fax: 614-292-5995; E-mail:
- Received June 9, 2004.
- Revision received August 27, 2004.
- Accepted September 23, 2004.
- ©2004 American Association for Cancer Research.