Epithelial Progeny of Estrogen-Exposed Breast Progenitor Cells Display a Cancer-like Methylome

Abstract

Estrogen imprinting is used to describe a phenomenon in which early developmental exposure to endocrine disruptors increases breast cancer risk later in adult life. We propose that long-lived, self-regenerating stem and progenitor cells are more susceptible to the exposure injury than terminally differentiated epithelial cells in the breast duct. Mammospheres, containing enriched breast progenitors, were used as an exposure system to simulate this imprinting phenomenon in vitro. Using MeDIP-chip, a methylation microarray screening method, we found that 0.5% (120 loci) of human CpG islands were hypermethylated in epithelial cells derived from estrogen-exposed progenitors compared with the non–estrogen-exposed control cells. This epigenetic event may lead to progressive silencing of tumor suppressor genes, including RUNX3, in these epithelial cells, which also occurred in primary breast tumors. Furthermore, normal tissue in close proximity to the tumor site also displayed RUNX3 hypermethylation, suggesting that this aberrant event occurs in early breast carcinogenesis. The high prevalence of estrogen-induced epigenetic changes in primary tumors and the surrounding histologically normal tissues provides the first empirical link between estrogen injury of breast stem/progenitor cells and carcinogenesis. This finding also offers a mechanistic explanation as to why a tumor suppressor gene, such as RUNX3, can be heritably silenced by epigenetic mechanisms in breast cancer.

Introduction

Prolonged exposure to estrogen and related endocrine disrup-tors during early development increases the risk of developing breast cancer (1, 2). These hormonally active agents have been shown to disrupt normal development of breast epithelia and support the growth of breast tumors (1, 2). Animal and epidemiologic studies suggest an imprinting phenomenon in which early exposure to estrogenic chemicals potentiates a carcinogenic process that is not observed until later in adult life (1, 3). Recent studies imply that estrogen imprinting can be mediated by an epigenetic mechanism (3, 4), a process that influences the phenotypic outcome of a gene without genetic changes of the underlying DNA sequence. DNA methylation, a key epigenetic modification, occurs frequently in CpG islands located near the promoters of silenced genes during disease development (5). Hypermethylation of promoter CpG islands is now considered to be a common mechanism for the heritable maintenance of gene silencing in human cancer (6).

Although developmental exposure to estrogens results in aberrant imprinting and subsequent transformation of breast epithelia (1), the cellular process underlying this phenomenon remains unknown. We hypothesize that breast stem and progenitor cells are primary targets of estrogenic exposure. These slowly dividing cells are capable of self-renewing and, in response to specific signaling, give rise to an amplifying population of progenitors that subsequently differentiate into different epithelial lineages (7). As the property of self-renewal allows for a long life span of stem cells, we speculate that these undifferentiated cells are highly susceptible to environmental injuries over time and transmit their “injury memory” to differentiated progeny through an epigenetic mechanism.

Using a methylation screening method, we report here that epithelial progeny of estrogen-exposed breast progenitors exhibits aberrant patterns of DNA methylation similar to those of known tumor suppressor genes in primary tumors. This cancer-like DNA methylation epigenome, or methylome (8), is associated with abnormal clonal expansion of epithelial progeny. Furthermore, the prevalence of estrogen-induced hypermethylation in tumors and adjacent normal tissues suggests that these epigenetic alterations occur early during breast carcinogenesis.

Materials and Methods

Tissue samples and cell lines

Tissues were collected in accordance with the protocols approved by the Institutional Review Boards of the University of South Florida and the Ohio State University. Breast tissues from patients that underwent mastectomy were collected. Specimens were marked for tumor and in some cases including its four surrounding zones (1, 2, 3, and 4 cm away from the grossly visible tumor boundary; see Fig. 5A). Samples from each zone were collected and frozen in liquid nitrogen. In contralateral mastectomy cases, normal tissues were taken from the four quadrants of the breast. Tissues were also obtained from healthy individuals undergoing breast reduction mammoplasty. Frozen tissues mounted on slides were marked and microdissected as described previously (9). Adipose tissues were trimmed away; the tumor and “normal” tissues were separated and stored in liquid nitrogen. Histologic examination of tissue sections and microdissection of samples were conducted in close collaboration with a pathologist to ensure consistency in the clinical diagnoses. Tissues adjacent to tumors were defined to be mostly unremarkable breast ducts or breast with cystic changes, mild ductal hyperplasia, or other nonproliferative changes. DNA was isolated from these samples as described previously (9). Breast cancer cell lines Hs578t, MCF-7, MDA-MB-134, MDA-MB-231, MDA-MB-435s, MDA-MB-453, T47D, and ZR75-1 were obtained from the American Type Culture Collection.

Figure 5.

Mapping geographic zones of RUNX3 DNA methylation in histologically normal tissues adjacent to breast tumor. A, microdissected breast tissues from patients undergoing mastectomy were collected. Mastectomy specimens were marked for tumor and its surrounding zones N1 (1 cm), N2 (2 cm), N3 (3 cm), and N4 (4 cm) from the grossly visible tumor boundary. In bilateral and prophylactic mastectomy cases, normal breast tissues were taken from the four quadrants of the breast. UO, upper outer; Ul, upper inner; LO, lower outer; LI, lower inner. B, the graphs show methylation ratios of the primary breast tumor and adjacent normal tissues of indicated distance (cm) from tumor boundary of each patient. Data of contralateral (CL) prophylactic mastectomy normal epithelia of patient 10486 are also shown. Dotted lines, mean methylation ratio for breast reduction normal tissues as reference.

Mammosphere culture and estrogen treatment

Normal breast tissues were obtained from healthy individuals (age range, 15–60 y; average, 34) undergoing reduction mammoplasties mainly due to macromastia. The tissues were dissociated mechanically and enzymatically, and single cells were grown into mammospheres in ultralow attachment plates (Cloning) in serum-free mammary epithelial growth medium (Cambrex) containing 20 ng/mL epidermal growth factor (EGF), 20 ng/mL basic fibroblast growth factor, l × B27, and 4 µg/mL heparin as described (10). Breast stem/progenitor cells were dissociated from primary mammospheres and treated with 70 nmol/L 17β-estradiol (E2; Sigma) or DMSO in phenol red-free medium for 2 wk (medium changed twice weekly). This estrogen treatment scheme has previously been shown to trigger neoplastic development of normal breast epithelial cells (11). After a 2-wk exposure period, mammosphere-derived cells were cultured on a collagen substratum (BD Biosciences) for an additional 2 to 3 wk. The cells were differentiated into epithelia in estrogen-free conditions using phenol red-free DMEM/F12 medium supplemented with 5% charcoal-dextran-treated fetal bovine serum (FBS; Hyclone), 5 µg/mL insulin, 1 µg/mL hydrocortisone, 10 ng/mL cholera toxin, and 10 ng/mL EGF as described previously (10). Total RNA, isolated from mammospheres and the derived epithelial cells using Trizol (Sigma), was used for real-time quantitative reverse transcription-PCR (RT-PCR) as described previously (12). Specific primers for amplification are available on request. DNA was extracted using QIAamp DNA Mini kit (Qiagen).

Immunofluorescence staining

Two thousand cells dissociated from mammospheres were seeded onto collagen-coated disks (BD Biosciences), placed on 6-cm-diameter dishes, and cultured as described above. Cells were fixed with 4% paraformaldehyde for 10 min and permeabilized with 0.1% Triton X-100 for 10 min. After blocking with 10% normal goat serum (Vector Laboratories) for 30 min, the cells were incubated with antibodies against estrogen receptor α (ERα; 1:50; Santa Cruz Biotechnology), CK14 (1:50; Santa Cruz Biotechnology), or CK18 (1:40; Santa Cruz Biotechnology) at room temperature overnight. Corresponding secondary FITC-conjugated antibody was applied followed by 4′,6-diamidino-2-phenylindole staining (Molecular Probes). The disks were mounted with Vectashield medium (Vector Laboratories) and images were captured by Zeiss fluorescence microscopy (Zeiss).

Colony formation assay

Single-cell suspensions from mammospheres exposed with or without estrogen were plated on collagen-coated plates at a density of 5,000 viable cells/10-cm-diameter dish as described (10). After 21 d, the colonies were fixed with methanol/acetic acid (3:1) and stained with 0.5% crystal violet solution for 10 min. The size of the colonies was measured and counted. Two-tailed Student’s t test was used to assess the significance of the observed differences in colony formation assay.

MeDIP assay and CpG island microarray hybridization

MeDIP was performed on 2 µg sonicated genomic DNA (300–1,000 bp) using a modification of the method described previously (13). Methylated DNA was immunoprecipitated at 4°C overnight with 30 µL of polyclonal antibody against 5-methylcytosine (Abcam). After serial washings, DNA was purified by phenol-chloroform extraction. The immunoprecipitated DNA and input DNA (10 ng) were then ligated to linkers and amplified by linker-mediated PCR as described previously (13). MeDIP-enriched and input DNAs (2 µg) were labeled with Cy5 and Cy3 dyes, respectively (12), and cohybridized to a human CpG island microarray (Agilent Technologies). Oligonucleotide probes (~ 185,000, 45-mers to 60-mers) were selected to target 27,800 CpG islands (7–24 probes per CpG island; Agilent Technologies). The hybridized slides were scanned by a GenePix 4000B scanner (Axon) and the acquired images were analyzed with GenePix Pro 6.0 (Axon). Dye-swap experiments were conducted, and Lowess-normalized data of duplicated microarray experiments were used to identify significant methylated loci (P < 0.001) using a peak detection algorithm in the ChIP Analytics 1.3 software (Agilent). To confirm candidate methylated genes identified by MeDIP-chip, primers targeting the enriched region were designed. PCR was performed with immunoprecipitated or input DNA Specific primers for amplification are available on request.

Methylation-specific PCR and bisulfite sequencing

DNA was isolated from breast tissues and cell lines using QIAamp DNA Mini kit and 0.5 µg from each sample was treated with sodium bisulfite using EZ DNA Methylation kit (ZYMO Research). Methylation-specific PCR (MSP) primers targeting the MeDIP-enriched regions of MXI1, RPRM, TFAP2C, and WNT5A genes were designed using Methyl Primer Express Software v1.0 (Applied Biosystems). Primer sequences and conditions for amplification are available on request. The primers used for RUNX3 have been described previously (14). Amplified products were electrophoresed on 2% agarose gels and visualized with ethidium bromide. For sequencing, bisulfite-treated DNA was amplified using primers for the RUNX3 promoter region (forward primer, YGGGGTAAATGTTAGAAATTTGT; reverse primer, ACCCCAAAC-TACTTAAAATCCCT). PCR products were cloned into TOPO TA cloning kit (Invitrogen). Five to 10 randomly picked clones were sequenced using ABI 3730 DNA analyzer (Applied Biosystems) and analyzed using the BiQ Analyzer.

Quantitative methylation analysis using MassARRAY

Bisulfite-treated DNA (1 µg) was amplified with specific primers for the RUNX3 promoter region. MassARRAY platform was used to perform quantitative methylation analysis (Sequenom). This system uses matrix-assisted laser desorption ionization time-of-flight mass spectrometry in combination with RNA base cleavage (MassCLEAVE). Details of quantitative analysis of methylation status have been described elsewhere (15). Two-tailed Student’s t test was used to assess the significance of the observed differences in MassARRAY when the averaged methylation ratios of different breast tissue types were compared. Statistical significance was assigned at P < 0.05.

5-Aza-2′-deoxycytidine treatment and real-time quantitative RT-PCR

MCF-7 cells were treated with 5 µmol/L 5-aza-2′-deoxycytidine (5-aza-dC) for 5 d in MEM containing 10% FBS and 6 ng/µL insulin. During the final 72 h of 5-aza-dC treatment, cells were hormone deprived by culturing in phenol red-free MEM supplemented with 4% charcoal-dextran-treated FBS. During the final 48 h of hormone deprivation, cells were treated with 1 µmol/L 4-hydroxy-tamoxifen (4-OHT) or vehicle (DMSO) followed by treatment with vehicle (0-h time point) or 10 nmol/L E2 for the indicated time period. RNA (2 µg) was extracted and reverse transcribed with Superscript III reverse transcriptase (Invitrogen). RUNX3 mRNA levels were measured using SYBR Green (Applied Biosystems) on a 7500 Real-Time PCR System apparatus. TATA-binding protein (TBP) mRNA levels were also measured as a control. Primer sequences and conditions for amplification are available on request.

Gene expression analyses

Total RNA was extracted from micro-dissected breast tissues [23 invasive ductal carcinomas (IDC), 28 adjacent normal tissues, and 10 reduction mammoplasties], and the quality of RNA samples was assessed using Agilent 2100 Bioanalyzer. Microarray hybridizations were performed by the Core Facility at the Dana-Farber Cancer Institute. The RNA was amplified and labeled according to Affymetrix guidelines and applied to Affymetrix GeneChip Human Genome U133 Plus 2.0 arrays (Affymetrix). The quality of hybridization data was assessed using normalized unscaled SE and relative log expression statistics (16) and MAS5.0 report quality scores. Normalized unscaled SE and relative log expression statistics were computed using the affyPLM package in Bioconductor. Expression values were computed from the raw (.cel) Affymetrix files and normalized using robust multichip averaging RMA (Bioconductor release 2.4.0). Detailed sample histology and clinical information and raw (.cel) and RMA-normalized data for these 61 gene expression profiles are available in ArrayExpress (accession number: E-TABM-276). The gene expression levels of the two Affymetrix probe sets for RUNX3 (204197_s_at, 204198_s_at) were highly correlated (r = 0.87) and were not significantly different (P > 0.05, Student’s t test). Further analyses of RUNX3 expression were performed on the mean gene expression level of these two probe sets. One IDC sample (T8703A1) was excluded as outlier in all analyses. To evaluate RUNX3 expression in breast tissue, we performed a general linear model ANOVA between RUNX3 gene expression and tissue histology and patient ERα status. To test the association of RUNX3 and ESR1 gene expression in breast tissue samples, the Pearson’s product-moment correlation coefficient was calculated. Further details, normalized RUNX3 expression data, and the R code to perform these statistical analyses are available online.⁸ RUNX3 expression levels (RMA-normalized log2 values) derived from Affymetrix microarray experiments of a panel of 51 breast cell lines were also extracted from a recent report (17). Four noncancerous cell lines (HBL100, MCF10A, MCF12A, and SUM190PT) were excluded for analysis. Two-tailed Student’s t test was used to test if RUNX3 expression (mean of the two probe sets as described above) was significantly different in breast cancer cell lines according to ERα status. Statistical significance was assigned at P < 0.05. Statistically significant results were also obtained when all 51 cell lines were included in the analysis.

Microarray data

The MeDIP-chip microarray data are accessible from the Gene Expression Omnibus (GSE7844). The raw data for expression profiling of human breast tumors and adjacent tissues are available at ArrayExpress (E-TABM-276).

Supplementary data

The supplementary data include four supplementary figures and one supplementary table.

Results

Estrogen preexposure of breast progenitors alters hormone receptor signaling and promotes clonal proliferation of their epithelial progeny

An in vitro protocol has recently been developed to enrich and propagate breast stem and progenitor cells using tissue samples obtained from reduction mammoplasties (10). These cells formed spherical colonies, called mammospheres, in suspension culture. Although a mammosphere may contain progenitor cells that are capable of differentiating into different types of epithelial cells, a subpopulation (<0.1%) of cells with self-renewal ability is believed to be stem cells (10). We independently established this culture method in the laboratory and showed that the isolated mammospheres exhibited the same progenitor characteristics as previously described (see Supplementary Fig. S1).

To investigate whether the model system can be used to simulate an estrogen imprinting event, mammospheres derived from reduction mammoplasties were continuously exposed to E2 (70 nmol/L) or DMSO (control) for 2 weeks (Fig. 1A), a treatment scheme known to induce transformation of the human breast epithelial cell line MCF-10F (11). Estrogen treatment stimulated self-renewal of breast progenitors, indicated by a significant increase in the number of mammospheres (P < 0.01; Supplementary Fig. S1A). After the exposure, E2 was removed from culture dishes, and progenitor cells were placed in fresh hormone-free medium on two-dimensional collagen substratum for 2 to 3 weeks (Fig. 1A). Under this two-dimensional culture condition, breast progenitors differentiated into epithelial cells, which were confirmed by immunostaining with epithelial markers (Supplementary Fig. S1B).

Figure 1.

Preexposure of breast progenitor cells to estrogen increases proliferation of progeny epithelial cells. A, the breast progenitor cells were propagated as floating spherical colonies, called mammospheres, and treated with 70 nmol/L E2 or DMSO solvent control for 2 wk. To induce differentiation, cells were seeded on a collagen substratum in the absence of E2 for 2 to 3 wk. Phenotypic and epigenetic analyses were then performed on the progeny epithelial cells. B, immunofluorescence shows that some progenitor-derived epithelial cells express ERα. After E2 stimulation (10 nmol/L, 3 h), translocation of ERα protein from the cytoplasm and into the nucleus was seen, suggesting functional estrogen signaling. In contrast, epithelial cells preexposed to high-dose estrogen exhibited nuclear ERα localization before E2 stimulation, reminiscent of MCF-7 cells. The percentage, subcellular localization, and staining intensity of ERα-positive cells are shown in the bar chart. The numbers of cells listed in each category were independently scored by two researchers. C, left, cell proliferation was measured by colony formation assay. Representative results of two independent experiments are shown. Exposure of progenitor cells to estrogen significantly increased the number of large-sized colonies (P = 0.02; middle) and the overall size of colonies (P = 0.038; right) formed by epithelial progeny cells. Columns, mean of two independent experiments; bars, SD. Asterisks, statistical significance.

We then determined whether this estrogen exposure resulted in changes of ERα signaling in progenitor-derived epithelial cells. Immunofluorescence analysis showed that a low level of cytoplasmic ERα was present in 80% of control epithelial cells. Following E2 stimulation (10 nmol/L for 3 h), we observed ERα translocating into the nucleus in some of the differentiated cells (Fig. 1B). However, this nuclear localization of ERα before E2 stimulation was already evident in epithelial cells derived from estrogen-exposed progenitors. In addition, stronger nuclear staining of these cells was observed after E2 stimulation, a condition reminiscent to that of receptor-positive breast cancer cells (18). These affected progeny also formed significantly larger colonies and exhibited an increased rate of proliferation compared with the control cells (P = 0.02–0.038; Fig. 1C). The findings indicate that an imprinting process may occur in breast progenitors preexposed to estrogen, resulting in an alteration of nuclear receptor signaling and clonal expansion in their epithelial progeny.

Estrogen-induced DNA hypermethylation occurs in genes commonly associated with the maintenance of stem and progenitor functions

To determine whether epigenetic processes play a role in this estrogen imprinting, we compared profiles of DNA methylation between estrogen preexposed and control progeny using a methylation screening approach, called MeDIP-chip (13). Methylated DNA, immunoprecipitated with an antibody specific for 5-methylcytosine, and input DNA were labeled with different fluorescent dyes and cohybridized to a microarray panel containing nearly all the annotated human CpG islands (n = 27,800). MeDIP-chip analysis in two replicates showed that 0.5% (120 loci) of the CpG islands were hypermethylated in epithelial cells derived from estrogen-exposed progenitors compared with the nonexposed control cells (Supplementary Table S1; Fig. 2A). Sequence analysis showed that 111 of these loci are located near the transcription start site (TSS) of genes. Thirteen gene-related loci with diverse MeDIP enrichments were randomly selected for confirmation analyses. Primers flanking the enriched regions were designed for gene-specific PCR assays, and estrogen-induced hypermethylation was confirmed in 92% (12 of 13) of these loci (Fig. 2B). Expression analysis of four such loci (RUNX3, MXI1, RPRM, and WNT5A) further indicates that this hypermethylation event is associated with down-regulation of the corresponding genes (Fig. 2C). In addition, MSP was conducted to determine whether the estrogen-induced hypermethylation could be observed in other epithelial cells prepared from different mammoplasties (Fig. 2D). Hypermethylation of three loci (MXI1, TFAP2C, and WNT5A) could be confirmed in some, but not all, epithelial progeny samples, likely attributed to individual variability.

Figure 2.

Preexposure of progenitor-derived epithelial cells to estrogen causes DNA hypermethylation and transcriptional silencing of tumor suppressor genes. A, differentially methylated genes in estrogen-preexposed progenitor-derived epithelial cells were identified by MeDIP-chip using a microarray panel containing 27,800 CpG islands. The maximum MeDIP enrichment along 120 hypermethylated loci in control and preexposed cells are shown in the heat map (see Supplementary Table S1 for a complete list of the hypermethylated loci). The heat map represents the averaged values of two independent experiments. B, to confirm candidate methylated genes determined by microarray, PCR primers targeting the MeDIP-enriched region were designed and the immunoprecipitated control/preexposed DNA samples were amplified along with input DNA. PCRs of no template (H₂O) control are also shown. Twelve of 13 tested genes were validated by PCR analyses. C, DNA methylation landscaping maps by plotting the enrichment ratio of each probe of the hypermethylated genes. The preexposed cells (red line) displayed an enriched methylated region of multiple probes, within the CpG island, located in proximity to the TSS (arrow); no significant enrichment was observed in control cells (blue line). Genes showing DNA hypermethylation in preexposed cells were associated with transcriptional suppression [shown by real-time quantitative RT-PCR (qRT-PCR); upper right, inset]. Columns, mean of three independent experiments; bars, SD. D, methylation status of tumor suppressor gene promoter regions was confirmed by MSP in control/preexposed DNA samples derived from three individuals. MSP PCR primers targeting MeDIP-enriched regions were designed. Preexposed cells of at least one set of samples showed DNA methylation in these tumor suppressor genes. M, PCR product with primers specific for methylated genes; U, unmethylated genes.

Ontological analysis of the 111 loci revealed significant methylation enrichment of genes related to nucleic acid and cellular metabolism, transcriptional control, and Wnt signaling (Supplementary Fig. S2A). Interestingly, 23% (n = 25) of these loci are known targets of polycomb proteins that play a role in maintaining pluripotency of stem and progenitor cells (Supplementary Fig. S2B; ref. 19). Promoter hypermethylation of polycomb target genes is frequently observed during tumorigenesis (20). Furthermore, eight of the positive loci (EGR2, FANCF, MXI1, PTPRG, RPRM, RUNX3, TFAP2C, and WNT5A) identified by MeDIP-chip are known tumor suppressor genes, the down-regulation of which was observed in various types of cancer (Supplementary Fig. S2C; refs. 14, 21–24). Taken together, our results indicate that key developmental regulators can be epigenetically targeted by estrogen, leading to deregulation of self-renewal of stem and progenitor cells and also tending to promote tumorigenesis.

Estrogen-induced DNA hypermethylation can frequently be seen in primary tumors

To determine whether estrogen-induced methylation changes observed in progenitor-derived epithelial cells also occur in breast tumors, we examined five loci (MXI1, RPRM, RUNX3, TFAP2C, and WNT5A) in a cohort of breast samples. As indicated earlier, these five loci have previously been reported to be tumor suppressors in different cancer types. MSP analysis was performed on 31 primary tumors, 8 cancer cell lines, and 10 control samples. Hypermethylation of these genes was observed in 23% to 74% of these tumors, but the methylation frequencies (25–100%) were slightly higher in cancer cell lines (Fig. 3). The results are consistent with the previously reported methylation frequencies (37% and 52%) for two genes, RPRM and RUNX3, in breast tumors (14, 24). In contrast, low methylation frequencies were detected in the reduction controls. The prevalence of these estrogen-induced epigenetic changes in breast tumors therefore provides evidence for a mechanistic basis underlying how promoter CpG islands acquire DNA methylation during cancer development (20).

Figure 3.

Estrogen-induced hypermethylation of tumor suppressor loci is also present in breast tumors. A, MSP analyses for tumor suppressor gene promoter methylation in breast reduction normal tissues (top) and primary tumors (bottom). Representative gel pictures show data for 10 of a total of 31 tumor tissues examined and 10 breast reduction normal tissues. B, summary of methylation data for all tissue samples and cancer cell lines. Aberrant methylation of these tumor suppressor genes occurred in a high percentage of primary tumors (23–74%) and cancer cell lines (25–100%) compared with reduction normal controls (0–20%).

Estrogen-induced hypermethylation of RUNX3 can frequently be seen in histologically normal tissue adjacent to the tumor site

We choose RUNX3, a developmental transcription factor regulated by polycomb repressors (19), for detailed methylation analysis in histologically normal tissues adjacent to primary tumors. Although inactivation of RUNX3 is known to play a role in tumorigenesis (14, 25), the mechanism of how this Cancer Res 2008; 68: (6). March 15, 2008 silencing event occurs in early breast tumorigenesis has not been elucidated.

MassARRAY, an assay simultaneously quantifying multiple methylated sites based on their molecular mass (15), was used to determine the methylation status of a 2-kb RUNX3 promoter region in nine sets of microdissected tumors and adjacent normal tissues (n = 40; Fig. 4A). These normal tissues were obtained from four successive zones, each 1 to 4 cm away from the tumor site of the ipsilateral breast (Fig. 5A, left). In one case (patient 10486), normal tissues from four quadrants of the contralateral breast were used as controls (Fig. 5A, right). In addition, eight unrelated tissues obtained from reduction mammoplasties were included in the analysis.

Figure 4.

DNA hypermethylation of RUNX3 in breast tumors and adjacent histologically normal tissues. A, genomic map of the RUNX3 promoter CpG island. Arrow, RUNX3 TSS in the genome sequence; inverted arrow, potential estrogen-responsive element (ERE; predicted by MATCH using the TRANSFAC database). Red bar, MeDIP-enriched region. Locations of the RUNX3 promoter fragments interrogated by MassARRAY (blue bars) and bisulfite sequencing (gray bar). B, quantitative methylation results from MassARRAY analysis of RUNX3 promoter region in primary breast tissues. Six amplicons covering a total of 101 CpG sites, located approximately 0.3 to 2.3 kb upstream of the RUNX3 TSS, were designed to analyze 9 sets of breast tumor (T)/adjacent normal (N) tissues and 8 breast reduction normal tissues (BRN; 48 samples total). CpG site measurements in the analyzed region of the RUNX3 promoter for all samples are depicted in the heat map. Blue color intensity, methylation ratios (see scale bar); gray, missing data values. Each row represents methylation data from one tissue sample. Tissues from each cancer patient are separated by black lines. Each column (spaced by white lines) represents methylation data from each amplicon. C, summation of the overall methylation levels of each tissue type in a genomic context. Red dotted lines, methylation ratios of tumor tissues; green dotted lines, methylation ratios of adjacent normal tissues; blue dotted lines, methylation ratios of breast reduction normal tissues; gray lines, CpG content in the amplification region calculated for a moving 100-bp window. D, bisulfite sequencing analysis of RUNX3 promoter region in one set of breast tumor/adjacent normal tissues and three breast reduction normal tissues. Black dots, methylated CpG sites; white dots, unmethylated sites. The underlined region is near the beginning of the methylation plateau observed by MassARRAY. The percentages of CpG methylation of the overall and underlined regions are indicated for each tissue. Breast tumor and adjacent histologically normal tissues showed significantly higher methylation levels compared with breast reduction normal tissues, in agreement with the MassARRAY analysis. Additional bisulfite sequencing results are shown in Supplementary Fig. S3.

We observed hypermethylation (e.g., regions 3, 4, and 5) of RUNX3 in at least five (4993, 9993, 9348, 10486, and 10501) of the nine primary tumor sites (Fig. 4B). The overall methylation levels in their corresponding normal tissues (1–4 cm away from tumor sites) also seemed to be higher than those of the eight reduction controls. The boundaries of this hypermethylation were clearly observed at ~ 0.5 and 1.7 kb upstream of the TSS (Fig. 4C). In this regard, methylation levels through this plateau were higher in tumors compared with adjacent normal tissues (P = 0.00038) or reduction controls (P = 0.0027). Notably, methylation levels were higher (P = 1.88e–11) in adjacent normal tissues compared with reduction controls. Bisulfite sequencing and MassARRAY analyses of an overlapping fragment (region 1) yielded concordant results (Fig. 4D). At the distal end of the sequencing region, where the beginning of the RUNX3 methylation plateau was observed by MassARRAY, dense methylation was detected in tumor 4993 and its adjacent normal tissues but not in three reduction controls. In addition, in the first exon region of RUNX3, extensive methylation was observed in three tumor pairs and adjacent normal tissues, whereas only a few CpG sites were methylated in reduction normal tissue (Supplementary Fig. S3).

We then performed an in-depth individual gradient analysis using the area with the greatest methylation difference (Fig. 5). Five patient samples with elevated methylation levels in the primary tumor site showed a gradual decrease in DNA methylation with increasing distance up to 4 cm (Fig. 5B, top). However, methylation levels in tumors from the other four patients were almost identical to those seen in reduction normal tissues (Fig. 5B, bottom). The four quadrant control tissues from the contralateral breast of patient 10486 also exhibited similar basal methylation level. Taken together, these results show that RUNX3 methylation may precede morphologic transformation of normal breast epithelia in a subset of patients. This observation supports the concept that estrogen-induced hypermethylation in progenitor cells is an early carcinogenic event and may create a large field of cancerization in the human breast.

Epigenetic suppression of RUNX3 expression is mediated by estrogen signaling in hormone receptor-positive breast cancer

Although underexpression of RUNX3 frequently occurs in breast cancer compared with normal epithelium (14), the expression of RUNX3 in histologic normal tissue adjacent to the primary tumor and the relationship between RUNX3 expression and ERα status have not been investigated. We thus determined mRNA levels of RUNX3 in a panel of 23 microdissected primary IDC tissues, 28 adjacent normal tissues, and 10 reduction tissues using expression data derived from Affymetrix microarray experiments. RUNX3 expression in IDC was significantly down-regulated (P < 0.0005) compared with both reduction normal tissues and adjacent normal tissues (Fig. 6A), in agreement with a recent report (14). When the tumors were segregated by ERα status, RUNX3 expression in ERα-positive IDC was significantly lower (P = 0.028) compared with ERα-negative IDC (Fig. 6B). Although no difference in overall expression of RUNX3 was observed in adjacent normal tissues versus reduction normal tissues (Fig. 6A), RUNX3 expression was significantly lower (P = 0.048) in normal breast tissues adjacent to ERα-positive breast tumors versus ERα-negative counterparts (Fig. 6B). Overall, a strong negative correlation (r = −0.52; P = 2.51e–5) of RUNX3 and ERα (ESR1) gene expression was observed in this set of breast samples (Supplementary Fig. S4). In addition, examination of RUNX3 expression from an Affymetrix data set of 47 breast cancer cell lines (17) conclusively indicated that RUNX3 expression was significantly lower (P = 0.0011) in ERα-positive compared with ERα-negative cell lines (Fig. 6C). Collectively, the MassARRAY and gene expression analyses show that RUNX3 expression is inversely correlated with hypermethylation of RUNX3 and ERα status.

Figure 6.

Inverse correlation between RUNX3 expression and ERα status. A, RUNX3 mRNA expression was significantly lower in IDC tissues compared with breast reduction normal tissues. No difference in RUNX3 expression was observed in adjacent normal tissues with unremarkable breast ducts and fibrocystic/nonproliferative changes versus reduction normal tissues. B, ERα-positive IDC had significantly reduced RUNX3 mRNA expression compared with ERα-negative tumors. Notably, normal breast tissues adjacent to ERα-positive breast tumors also had significantly lower expression than ERα-negative counterparts. C, evaluation of RUNX3 mRNA expression in a panel of breast cancer cell lines. ERα-positive cell lines had significantly lower RUNX3 expression than ERα-negative cell lines, in agreement with the data from primary breast tissues. In A to C, expression levels are represented in log2 values. In the box plots, the upper and lower boundaries of the box indicate 75th and 25th percentiles, respectively. Line within the box, median; bars above and below the box, 90th and 10th percentiles, respectively. The numbers of tissues/cell lines studied are also indicated. Asterisks, statistical significance. D, real-time quantitative RT-PCR analysis of RUNX3 in MCF-7 cells after a 5-d treatment with 5 µmol/L 5-aza-dC. Black line, 5-aza-dC-treated MCF-7 cells were hormone deprived for 72 h and treated with vehicle or 10 nmol/L E2 for the indicated times (h); gray line, 5-aza-dC-treated and hormone-deprived cells were treated with 4-OHT (1 µmol/L for 48 h) and then treated with vehicle or 10 nmol/L E2 for the indicated times (h). Results are presented relative to no 4-OHT treatment control (black line, 0 time point). Points, mean of triplicates; bars, SD.

To determine if estrogen signaling plays a role in RUNX3 silencing, we performed a gene reactivation study using quantitative RT-PCR in ERα-positive MCF-7 breast cancer cells displaying RUNX3 hypermethylation (Fig. 3B). RUNX3 expression was first reactivated by 5-aza-dC treatment (data not shown); subsequent treatment with E2 resulted in transient down-regulation of RUNX3 mRNA levels (Fig. 6D). Pretreatment with 4-OHT, an ERα antagonist, partially inhibited the E2-induced down-regulation, providing the direct evidence that estrogen suppression of RUNX3 is, in part, mediated by ERα.

Discussion

Developmental exposure to estrogen and related endocrine disrupters is known to increase breast cancer risk in women (1, 2). As indicated earlier, epidemiologic studies have suggested an imprinting mechanism in which early exposure of environmental estrogens potentiates a carcinogenic process observed later in life (1, 3). In the past, it is difficult to study how this imprinting mechanism is operated at the molecular level. Presently, we show that mammospheres, which contain enriched breast stem and progenitor cells (10), can be a model system for this type of study. Our rationale is that long-lived, slow-growing breast progenitors are more susceptible to environmental injuries than terminally differentiated epithelial cells, which have a high turnover rate in the breast duct. With this prior notion, we designed the experiment by exposing mammospheres to estrogen (i.e., E2) and later completely removed E2 from the culture medium and allowed breast progenitors to undergo epithelial differentiation in two-dimensional collagen stratum.

Although the mammosphere model system cannot exactly imitate a real-life scenario, the experimental design is useful to determine whether an imprinting phenomenon could be recreated in vitro. The E2 dose (70 nmol/L) used in the present study is higher than typically seen in women under normal physiologic conditions or having prior history of estrogen exposure. In a previous study by Russo et al. (11), they found that, at this dose of treatment, estrogen caused cellular transformation of a breast epithelial cell line, MCF-10F. When transformed cells with aggressive behavior were further filtered through an invasion chamber, these clonal cells could grow into tumors in a xenograft model. Unlike MCF-10F, our primary epithelial cells derived from estrogen-treated progenitors did not produce tumors in female athymic mice after 6 months of transplantation (data not shown). Additional molecular alterations need to be acquired in these primary epithelial cells to increase their tumorigenic potential. Future studies can also be performed to test tumorigenicity of different doses of estrogen and other endocrine disruptors using this mammosphere system.

There are limitations in this mammosphere model system. Genetic variations and other confounding factors (e.g., menopausal status and age) may affect estrogenic susceptibility of individuals. Therefore, we could not anticipate concordant results when different breast progenitor samples are exposed to the same dose of estrogen (see Fig. 2D). The experimental findings derived from this study may be biased toward patients undergoing breast reduction procedures. Ethically, we could not obtain breast tissue samples from healthy individuals with a known exposure history of endocrine disruptors. Nevertheless, the model system proposed here represents a new paradigm, simulating a condition of early breast carcinogenesis that is associated with exposure to environmental estrogen. We have shown that estrogen exposure stimulates self-renewal of breast progenitors (Supplementary Fig. S1A) and increases proliferation of the derived epithelial progeny (Fig. 1C), both of which are associated with alterations in the methylome. Because such epigenetic alterations were not observed in the differentiated epithelial cells in vitro (Fig. 2) and in vivo (Fig. 3), we strongly contend that the methylation changes were not a consequence of cell differentiation. Alternatively, we propose that estrogen induces epigenetic alterations in breast progenitors leading to mammary carcinogenesis.

Our proof-of-principle study shows that critical genes controlling tumor suppression, such as RUNX3, can be down-regulated on estrogen exposure to breast progenitors. Recent reports have shown that breast progenitor cells express ERα (26, 27). Thus, the down-regulation may be mediated, in part, by the canonical pathway of estrogen signaling (see Fig. 6D). This finding is also supported by the presence of nuclear ERα in the affected epithelial progeny; this nuclear internalization is commonly seen in hormone receptor-positive breast cancer with functional estrogen signaling (18). As reported in our previous study (12), estrogen can induce H3K9 dimethylation, a repressive chromatin modification (28), and transiently suppresses a set of ERα-responsive genes in breast cancer cells. The presence of dimethyl H3K9 mark may promote a repressive mode that renders target genes more vulnerable to long-term silencing during cancer development (29). Based on existing reports and our present study, we propose that progressive accumulation of DNA methylation plays a final role in both maintaining permanent silencing of tumor suppressor genes and transmitting this heritable information from breast progenitors to their epithelial progeny. Although our MeDIP data only provide methylation information of the affected epithelial progeny, a sensitive immunoprecipitation technique (30) can be implemented in the future to directly interrogate the methylome of mammospheres containing only 2,000 to 10,000 cells. Comparing the methylomes between the hormone-treated mammospheres and their derived epithelial progeny will lead to better understanding of the estrogen epigenetic imprinting phenomenon.

Although the present study reports estrogen-induced hyper-methylation, we also observed preferential MeDIP enrichment of ~ 0.5% of CpG island loci analyzed in the control epithelial progeny relative to those of the affected progeny (data not shown). This could be a hypomethylation event, which may require a functional demethylase yet to be characterized in the affected progeny. Hypomethylation of non-CpG islands is known to occur in repeat sequences of the cancer genome (6). Because our microarray panel contains only single-copy CpG islands, the issue of whether estrogen induces hypomethylation of repeat sequences during epithelial differentiation remains to be explored. One more plausible explanation is that promoter methylation of these CpG islands occurs during normal maturation of particular epithelial branches in the breast duct. It is possible that, as a result of estrogen exposure, these promoter CpG islands escape methylation-mediated silencing, and subsequent activation of the corresponding genes may promote clonal proliferation of the affected epithelial progeny. Future research direction will require isolation of primary epithelial cells from different lineages by flow cytometry using defined surface markers and separately analyze their methylomes by MeDIP-chip. These challenging studies will determine whether the estrogen treatment can induce a true hypomethylation event or prevent an epigenetically mediated maturation process in the affected progeny.

In conclusion, the mammosphere model described in this study may provide a new paradigm for assessing cancer risk due to exposure to endocrine disruptors. The in vitro system may serve as a viable alternative to animal models and other toxicologic assays for testing new estrogenic compounds. Furthermore, the altered methylome cataloged in this study can be used to identify biomarkers for early breast cancer detection and biosensors for environmental estrogens.