Mammalian Genome

, Volume 22, Issue 3, pp 249–259

DNA methylation changes in murine breast adenocarcinomas allow the identification of candidate genes for human breast carcinogenesis

  • Deanna Acosta
  • Masako Suzuki
  • Diana Connolly
  • Reid F. Thompson
  • Melissa J. Fazzari
  • John M. Greally
  • Cristina Montagna
Article

DOI: 10.1007/s00335-011-9318-6

Cite this article as:
Acosta, D., Suzuki, M., Connolly, D. et al. Mamm Genome (2011) 22: 249. doi:10.1007/s00335-011-9318-6

Abstract

Epigenetic inactivation due to aberrant promoter methylation is a key process in breast tumorigenesis. Murine models for human breast cancer have been established for nearly every important human oncogene or tumor suppressor gene. Mouse-to-human comparative gene expression and cytogenetic profiling have been widely investigated for these models; however, little is known about the conservation of epigenetic alterations during tumorigenesis. To determine if this key process in human breast tumorigenesis is also mirrored in a murine breast cancer model, we mapped cytosine methylation changes in primary adenocarcinomas and paired lung metastases derived from the polyomavirus middle T antigen mouse model. Global changes in methylcytosine levels were observed in all tumors when compared to the normal mammary gland. Aberrant methylation and associated gene silencing was observed for Hoxa7, a gene that is differentially methylated in human breast tumors, and Gata2, a novel candidate gene. Analysis of HOXA7 and GATA2 expression in a bank of human primary tumors confirms that the expression of these genes is also reduced in human breast cancer. In addition, HOXA7 hypermethylation is observed in breast cancer tissues when compared to adjacent tumor-free tissue. Based on these studies, we present a model in which comparative epigenetic techniques can be used to identify novel candidate genes important for human breast tumorigenesis, in both primary and metastatic tumors.

Introduction

Changes in human DNA methylation patterns are a major event in cancer initiation and progression (Esteller 2008). The cancer genome is characterized mainly by promoter hypermethylation of tumor suppressor genes concurrently with an overall, genome-wide decrease in the level of 5-methyl cytosine (Jones and Baylin 2007). Promoter hypermethylation thus serves as one epigenetic mechanism used to deregulate tumor suppressor gene activity. In contrast, genome-wide hypomethylation affects the heterochromatic regions of DNA, resulting in chromosomal instability and an increase in mutation events (Chen et al. 1998; Tsai et al. 2008). Recently, aberrantly hypomethylated promoters have been observed in tumors when compared to normal tissue, suggesting that the opposite scenario is also possible (Figueroa et al. 2010). Therefore, the development of tools and comparative genetic approaches designed to narrow candidate genes and pinpoint cancer-specific epigenetic changes is essential to identify genes responsible for tumor transformation. This progress is facilitated by the construction of comparative maps and the sequencing of the mouse genome (Okazaki et al. 2002; Waterston et al. 2002). Conservation of DNA sequences between species is well studied (Waterston et al. 2002). In contrast, few studies have been performed to determine whether epigenetic patterns are also conserved, and if so, to what degree (Bernstein et al. 2005; Demircan et al. 2009; Vu et al. 2006). Therefore, it is important to establish whether mouse models for breast cancer are suitable for studying epigenetic changes involved in human breast tumorigenesis. This would allow investigators to perform in vivo studies that would otherwise be difficult to carry out using human samples.

Modeling human cancer in the mouse has become an indispensable tool to dissect the sequence of molecular events that occur during tumorigenesis (Hutchinson and Muller 2000). Comparative cytogenetic analysis has been performed on primary adenocarcinomas from a variety of murine breast cancer models (Bowen et al. 2005; Dorritie et al. 2004; McNeil et al. 2003; Montagna et al. 2002, 2003; Ried et al. 2004; Weaver et al. 1999, 2002). These studies established that (1) murine models for breast cancer, as for humans, are characterized by nonrandom chromosomal aneuploidies; (2) chromosomal changes recapitulate rearrangements observed in humans; and (3) chromosomal alterations observed in advanced stage IV adenocarcinomas are independent of the original oncogenic event driving tumorigenesis. Such findings underscore the importance of using murine models to study chromosomal translocations significant in human tumorigenesis and for the identification of novel oncogenes (Montagna et al. 2003).

In this study we query epigenetic differences between polyomavirus middle T antigen (PyMT) primary and paired lung metastatic tumors and normal mammary glands. Tumors in PyMT mice arise in the mammary gland with short latency and frequently result in pulmonary metastases (Guy et al. 1992). These tumors represent a model for studying epigenetic changes occurring in both primary tumors and metastatic cells.

A microarray approach was used to analyze cytosine methylation at 71 genes, including loci known to be epigenetically silenced in human breast cancer and novel candidate genes and control loci. Such genes included the homeobox a and d domains and the Gata2 locus. Homeobox-containing transcription factors control vital networks during tissue development and differentiation, and their silencing through epigenetic mechanisms has been associated with breast morphological abnormalities and oncogenesis (Novak et al. 2006; Tommasi et al. 2009). Gata2 is of particular interest because the Gata family of proteins is critical in maintaining the nontransformed cell state in the mammary gland (Kouros-Mehr et al. 2008) and Gata3 is an essential regulator of mammary gland morphogenesis and differentiation (Asselin-Labat et al. 2007). Our array allows for the detection and analysis of known differentially methylated genes and novel candidates in the PyMT model.

We mapped epigenetic alterations in metastatic cells and compared them with changes observed in the primary tumors. We describe the identification of differentially methylated loci in the mouse that are known to be epigenetically dysregulated in human primary breast cancer. These methylation changes were found to correlate with altered gene expression, supporting their functional significance in the mouse. The most relevant finding of our study is that these candidate genes are also silenced in human breast cancer. Our results provide new insight into how murine models of breast cancer can be used to study epigenetic differences between primary tumors and their metastases, and most importantly to identify significant epigenetic changes that occur in the mouse and are maintained in human genome.

Materials and methods

Mouse and human primary

We dissected tumors from two PyMT mice (FVB/N genetic background, gift from Dr. Jeffrey Pollard, Einstein) at 10 weeks of age. Primary tumors were removed from the mammary glands (PT1 and PT2) and matching lung metastases (LM1 and LM2) were isolated from the same animals. The fourth inguinal mammary gland was dissected from an age-matched normal FVB/N mouse and used as normal control (MG1). Tumors were divided and processed for genomic DNA isolation, mRNA extraction, and primary cultures to obtain metaphases for spectral karyotyping (SKY) analysis. DNA and mRNA isolation was performed as previously described (Khulan et al. 2006). An additional 11 primary tumors were isolated from 12 to 20 week-old PyMT mice and 5 age-matched normal FVB/N mice for biological validation of candidate genes. Flash-frozen human breast tumors (n = 59) and tumor-free adjacent areas (n = 7) used for qRT-PCR were provided by Dr. Adrian Harris (University of Oxford). Additional normal breast tissue from reduction mammoplasty (n = 10) were obtained from the Cooperative Human Tissue Network. DNA from primary breast carcinomas and adjacent tumor-free tissue was collected by Dr. Greally’s laboratory.

Spectral karyotyping (SKY)

Cultures of primary tumors were maintained at passage numbers lower than 5. Cells were processed for chromosome preparations and SKY hybridization following standard protocols (Montagna et al. 2002, 2003). Images were acquired with an Olympus BX51 microscope equipped with a spectracube controlled by ASI full acquisition and analysis software (Applied Spectral Imaging, Inc., Carlsbad, CA). Chromosome aberrations were defined using the nomenclature rules from the International Committee on Standardized Genetic Nomenclature for Mice (Davisson 1994). Seven to ten metaphases were analyzed for each tumor. The average chromosomal aneuploidy level for each tumor was calculated as the total number of chromosomes gained or lost in each sample divided by the number of cells analyzed. The average number of chromosome breaks for each tumor was calculated as the total number of chromosome breaks observed in each sample divided by the number of cells analyzed.

HELP assay

Ligation-mediated PCR for fragment amplification and hybridization was performed as described (Khulan et al. 2006; Oda et al. 2009). The LM-PCR products were cohybridized to our custom-designed 12-plex microarray (Selzer et al. 2005). The array design is based on the mm5 build of the mouse genome and contains 1339 HpaII-amplifiable fragments in the range of 200-2000 bp (covering 6.2 Mb of the mouse genome), with more than 13,000 oligonucleotides on the array. There are 71 genes on the array, plus 8 additional splice isoform variants, for a total of 79 genes. The microarray is designed to interrogate diverse genomic contexts, including autosomes, sex chromosomes, constitutively active and tissue-specific genes, CG-depleted and CpG island-rich regions (e.g., Hox gene clusters), and an imprinted locus (H19/Igf2). A subset of the loci included in the array have been previously shown to undergo epigenetic changes in various cancer types, specifically in human breast cancer: Hoxa and Hoxd domains (Makiyama et al. 2005); Pou5f1 (Ben-Porath et al. 2008), and H19/Igf2 imprinted region (Antoniou et al. 2009); while other genomic regions represent novel candidates (Gata2) and control loci for comparison. The full list of the genomic regions included in the array is listed in Supplementary Table 1. Samples were hybridized in triplicate.

Data processing and analysis

Raw data were preprocessed and normalized by applying the analytical pipeline for HELP assays (Thompson et al. 2009). After normalization we obtained 1385 amplifiable fragments, with HpaII/MspI ratios having signal-to-noise values suitable for analysis. From the normalized HpaII and MspI intensities, we generated log2 ratios of HpaII/MspI fragments (M). Positive and negative values represent hypo- and hypermethylated loci, ranging from a minimum of −3.77 to a maximum of +3.50. Analysis of the normalized M values by hierarchical clustering was performed using the statistical software R v2.8 (Ihaka 2008).

The values of the triplicates were averaged for each sample to determine probe-specific standard deviation, and the 95% confidence interval was calculated for each locus. To determine a cutoff to classify each locus as methylated or nonmethylated based on the log2 HpaII/MspI ratio, we used the observed distribution of epigenome-wide normalized methylation values measured from the normal mammary gland (MG). We plotted the density of the M values for MG and, based on the bimodal distribution of the data, we set the cutoff at the lowest point in the trough of the curve that divides the two populations (Supplementary Fig. 1, red). Therefore, for further descriptive analysis based on categorized loci and for the purpose of selecting biologically relevant genes, any M value above 1.01 was considered hypomethylated, any value between 0 and 1.01 was considered as having intermediate methylation, and values below 0 were considered hypermethylated (Supplementary Fig. 1, vertical black line).

For prioritization of genomic loci and for the identification of biologically significant DNA methylation changes, we first calculated the change in methylation value (each tumor minus MG) to generate a delta value for each locus (∆M) similar to previous epigenome-wide studies (Irizarry et al. 2009). Due to our small sample size, we chose to focus largely on the observed fold change as the primary locus-specific summary. Positive ∆M values represent loss of methylation in tumors, while negative values represent increased methylation. Next, we plotted a histogram of the ∆M values for all tumor samples (Supplementary Fig. 2). We used a 1.5-fold cutoff representing the top 10% of differentially methylated loci (Supplementary Fig. 2, black lines) to identify candidate loci. For genomic compartment analyses, the promoter region was defined as ±2 kb from the annotated transcription start site of annotated RefSeq genes (mm5 build of the UCSC genome browser), the gene body as the remaining region to the transcription termination site, and all other loci as intergenic.

Quantitative DNA methylation analysis by MassARRAY EpiTyping

To validate the data obtained from the HELP assay, we performed MALDI-TOF mass spectrometry using EpiTyper by MassARRAY (Sequenom, San Diego, CA) on bisulfite-converted DNA (Ehrich et al. 2005). For bisulfite conversion we used the Zymo kit (Zymo Research Corp., Orange, CA). MassARRAY primers were designed using MethPrimer (http://www.urogene.org/methprimer/) to cover flanking HpaII sites for Hoxa7 and Gata2.

Gene expression analysis by qRT-PCR

Five micrograms of total RNA was used to synthesize cDNA using Superscript II (Invitrogen, Carlsbad, CA). The expression values of our candidate genes were determined by quantitative RT-PCR (qRT-PCR) in mouse and human tissue samples using Fast SYBR® Green and the StepOnePlus Real-Time PCR from Applied Biosystems (Foster City, CA). We ran three technical replicates for each sample and calculated the average of the CT values for each gene. Data were normalized against GAPDH. The ΔCT and ΔΔCT methods were used to determine relative gene expression levels (Schmittgen and Livak 2008). All primer sequences are available as supplementary data (Supplementary Table 2).

Results

Molecular cytogenetic profiling of breast adenocarcinomas and lung metastasis

We selected mice in which tumorigenesis is induced by the overexpression of polyomavirus middle T antigen because in this model, tumorigenesis recapitulates morphological alterations observed in human breast cancer (Guy et al. 1992; Lin et al. 2003). We have previously performed comparative cytogenetic profiling of this model (Montagna et al. 2003; Ried et al. 2004) and demonstrated that when taking human and mouse synteny into account, chromosomal DNA copy number changes (gains and losses) occurring in the PyMT animals carry a cytogenetic profile similar to that observed in human breast cancer.

In the current study, SKY was conducted to determine the similarity between the PyMT tumors and human cancer based on cytogenetic profiling. This analysis revealed that both structural and numerical alterations exist in these samples (Fig. 1a, b). A summary of cytogenetic alterations is available through the SKY/CGH database (http://www.ncbi.nlm.nih.gov/sky). In Primary Tumor 1 (PT1), the most recurrent abnormality is a T1,11 translocation resulting in the gain of distal 11E2 (orthologous to human 17q21–25), as frequently observed in murine models for human breast cancer (Fig. 1a, b); (Montagna et al. 2003). This marker chromosome is lost in the matching lung metastasis (LM1) where a high degree of whole-chromosome aneuploidy is present without recurrent structural rearrangements (ploidy = 13-80) (Fig. 1c). Primary Tumor 2 (PT2) carries a loss of chromosomes 2, 12, and X, plus a recurrent T14,2 and Rob15, resulting in gain of chromosome 15. No recurrent translocations could be detected in the two lung metastases. However, a recurrent gain of chromosome 6 (the chromosome to which Gata2 and Hoxa7 map) was observed in the lung metastasis.
Fig. 1

Cytogenetic profiling of mouse adenocarcinomas reveals similarities to human carcinomas. a SKY karyotype of a representative tetraploid metaphase derived from PT1. Pseudocolors depicting chromosomal abnormalities are shown. The full karyotype of this cell is (>4n) 77XX T(1,11), +1, & (4,7), +5, −8, −10, −13, −18, Dp(19), T(X,11), −x. b Representative hyperdiploid karyotype from PT1: (<2n) 41XX, T(X,11), +1. c Number of chromosomes indicating the ploidy are plotted for each cell analyzed in the four samples. d Summary of the number of chromosomal breaks present in each cell analyzed for the four tumor samples. On the bottom of each graph we calculated, the average number of chromosome gains and losses for each cell and the average number of chromosomal breaks for each cell respectively

Our analysis revealed that the average chromosomal aneuploidy is similar for the primary tumors (4 and 4.7 gains or losses) and is more variable in the metastases (2.7 and 9.2 gains or losses). The average number of chromosome breaks was similar in PT1, PT2, and LM2 (range = 1.7-2.4) and lower in LM1 (0.4), as summarized in Fig. 1d. From the cytogenetic analysis, we observed an increased level of chromosomal instability, measured as whole-chromosome aneuploidy, in the metastatic tumors. We concluded that these tumors show rearrangements characteristic of the PyMT model and that they carry a cytogenetic profile similar to what is observed in human breast tumors (Ried et al. 2004).

Mouse breast adenocarcinomas acquire distinctive DNA methylation profiles

We investigated whether cytosine methylation changes in these tumors are orthologous to changes observed in human breast cancer. The HELP assay was performed on selected regions of the mouse genome (Supplementary Table 1). Some of these regions, including the Hoxa and Hoxd domains, have been shown to be differentially methylated in human breast cancers (Novak et al. 2006, 2008).

The HpaII fragments were assigned as hypo- or hypermethylated using density plots of the M values for each sample as described in Materials and methods (Supplementary Fig. 1). We observed that the normal mammary gland (MG, red line) has a greater portion of loci above the 1.01 threshold, and thus has a higher number of hypomethylated loci relative to the tumor samples (green and blue lines). Consistent with this observation, there are a larger number of hypermethylated fragments in each of the tumors than in the normal gland (Table 1). To identify the most significant differentially methylated fragments, we selected the top 10% of HpaII fragments (~150) that were differentially methylated in each tumor sample when compared to MG based on a ∆M cutoff value of ±1.5 (Supplementary Fig. 2).
Table 1

Number of hypo- and hypermethylated fragments for each sample

 

MG

PT1

LM1

PT2

LM2

Hypomethylated

302

211

177

145

131

Hypermethylated

1083

1174

1208

1240

1254

MG mammary gland; PT1 primary tumor 1; LM1 lung metastasis 1; PT2 primary tumor 2; LM2 lung metastasis

The Venn diagram in Fig. 2a summarizes the distribution of methylation changes of these HpaII fragments between the four tumor samples. Though not the focus of our study, we observed that there were 5 hypermethylated loci specific to the primary tumors and 11 loci (9 hyper- and 2 hypomethylated) specific to the metastatic samples. Combining primary and metastatic tumors, there were 59 fragments (24 hyper- and 35 hypomethylated) characterizing the common methylation changes between all of the tumors. The relative methylation (M values) of these fragments is represented in the heat map in Fig. 2b. The technical replicates for each sample are plotted in the heat map to show the consistency of the data.
Fig. 2

Analysis of differentially methylated fragments within each tumor. a Venn diagram of the top 10% of the ΔM fragments. Hypermethylated fragments are highlighted in bold. In red are ΔM fragments common to all tumors, primary tumors and lung metastases. b Heat map of all samples (columns) and differentially methylated fragments common to all tumors (rows). Samples are displayed in triplicate to demonstrate technical reproducibility. Hypermethylated fragments are shown in blue, hypomethylated fragments are shown in yellow. c Distribution of ΔM fragments in the context of genomic compartments. The number of hypo- or hypermethylated fragments relative to the normal mammary gland were plotted for each tumor based on its mapping to promoter, gene body, or intergenic regions of the genome

We next analyzed the genomic compartments where the methylation changes were occurring (Fig. 2c). The differentially methylated HpaII fragments were assigned to genomic compartments as described in Materials and methods. When compared to the normal mammary gland, the tumors were hypermethylated both in the promoter and the gene body with a greater degree of difference at the promoter. The majority of differentially methylated fragments appeared in intergenic regions, and the tumors were relatively more hypomethylated, with the exception of LM1. These observations are similar to what is observed in the human cancer genome, which is frequently characterized by hypermethylation at the promoter of specific genes concurrent with an overall decrease in the level of 5-methyl cytosine (Ordway et al. 2007).

We identified a subset of genomic loci differentially methylated in the tumors relative to normal samples. The analysis conducted on the PyMT murine model revealed a system suitable for detection of methylation changes not only between normal and tumor samples, but also with matching solid metastatic tumors (Fig. 2a, Supplementary Fig. 3). Collectively, the data suggest that the epigenome of murine breast adenocarcinomas in vivo follow cytosine methylation changes similar to what is observed in human breast tumorigenesis.

Identification of differentially methylated genes in mouse breast adenocarcinomas

Upon mapping the differentially methylated fragments to promoters, we found that 23 genes were differentially methylated in the tumors relative to the normal. Of those that underwent methylation changes, 1 was hypomethylated in the tumors and 22 were hypermethylated when compared to MG (Table 2).
Table 2

Genes with differentially methylated promoters between normal mammary gland and tumor samples

Hypomethylated in tumors

Hypermethylated in tumors

Psors1c2

Cchcr1

 

Evx1

 

Evx2

 

Gata2

 

Hoxa1

 

Hoxa10

 

Hoxa11

 

Hoxa2

 

Hoxa3

 

Hoxa6

 

Hoxa7

 

Hoxa9

 

Hoxd12

 

Hoxd13

 

Lsp1

 

Mrpl23

 

Pou5f1

 

Psors1c2

 

Skap2

 

Sphk1

 

Srebf2

 

Tnni2

In bold genes selected for biological validation

We selected Hoxa7 (homeobox gene A7) and Gata2 for quantitative validation studies, based on (1) the degree of alteration of HELP values during tumorigenesis, (2) the mapping of these changes at the promoter region, and (3) prior implication of the gene in human carcinogenesis. The M values for the three technical replicates for these genes, the standard deviation, and 95% confidence intervals (CI) are shown in Supplementary Table 4. In addition, their chromosomal position and the genomic regions identified by the HELP assay are listed in Table 3 and Supplementary Table 3.
Table 3

Validated differentially methylated loci, their chromosomal locations, and mapping of primer pairs used for validation

Gene

Location of DMRa

Transcription start site

Strand

Primer 1

Primer 2

Primer 3

Primer 4

Primer 5

Primer 6

Hoxa7

chr6:52167054–52167817

chr6: 52164283

−1739

−2872

−1065

   

Gata2

chr6:88529921–88530722

chr6: 88533505

+

−2849

−2540

−2254

−2041

−3951

−4707

aDifferentially methylated region

bImprinting control region

The differentially methylated region (DMR) in the Hoxa7 promoter identified by the HELP assay is graphically represented in Fig. 3a. The HELP data indicate that this region becomes hypermethylated in both primary and metastatic breast tumors with respect to the normal mammary gland. Similar results were obtained for Gata2 (Supplementary Fig. 4). Bisulfite MassARRAY, used to validate the Hoxa7 promoter region, confirmed a consistent increase in methylation in all of the tumors tested (Fig. 3b). Likewise, bisulfite MassARRAY analysis of the Gata2 promoter showed a consistent increase in methylation in each of the tumor samples tested (Fig. 3d).
Fig. 3

Differential methylation at the promoters of HoxA7 and Gata2 correlate with gene silencing. a Custom wiggle tracks representing the HELP data obtained for the Hoxa7 gene locus mapped to the mm5 build of the UCSC genome browser. Bars above 0 indicate hypomethylated fragments, and bars below 0 indicate hypermethylated regions. Each track represents data from a different sample (red, MG; dark blue and dark green, PT1 and LM1; light blue and light green, PT2 and LM2). The position of the gene, the direction of transcription (right to left), and the location of CpG islands (green bars) are depicted in the track. The HpaII restriction sites are mapped accordingly, and the location of the PCR fragments used for MassARRAY analysis are mapped in purple bars. (DMR, differentially methylated region; the star pinpoints the transcription start site). b MassARRAY validation of Hoxa7 methylation data. Each row represents a different sample and each box in the row represents the analyzable CpG dinucleotides present within each PCR fragment. Gaps within a fragment represent CpGs that were not analyzable. Hypomethylated CpG dinucleotides are yellow (0% methylation), while blue boxes represent hypermethylated CpG dincleotides (100% methylation). The asterisks mark CpGs associated with HpaII sites for HELP validation. c Expression of Hoxa7 in each tumor sample normalized to the expression in the normal mammary gland (MG) sample as determined by qRT-PCR. d The methylation status of individual CpG dinucleotides from each fragment analyzed by MassARRAY is graphed below the RefSeq annotation of Gata2 (gene in light blue, MassARRAY fragments in purple). MassARRAY data depicted as described in (b). e qRT-PCR data depicting the mRNA levels of Gata2 in the tumors normalized to the normal mammary gland

Hypermethylation at promoter regions of candidate genes correlates with gene silencing

Hypermethylation of the promoter is a mechanism by which tumor suppressors and other genes are silenced (Jones and Baylin 2007). Since the promoters of Hoxa7 and Gata2 undergo increased methylation, as detected by both HELP and MassARRAY, we expect transcriptional silencing at these loci in the tumors. We tested this hypothesis by qRT-PCR to measure the level of mRNA in the normal mammary gland and in the four tumor samples for the two genes tested by bisulfite MassARRAY. All of the tumor samples showed decreased gene expression associated with promoter methylation for both Hoxa7 (Fig. 3c) and Gata2 (Fig. 3e). From these experiments, we concluded that hypermethylation at the promoters of Hoxa7 and Gata2 correlates with gene silencing in mouse breast cancer.

Differential methylation and expression of candidate genes in an expanded set of samples

To validate our findings that Hoxa7 and Gata2 are differentially expressed in mouse breast adenocarcinomas and that methylation studies conducted in the PyMT mouse model have the potential to identify biologically significant changes for human breast carcinogenesis, we proceeded to analyze a larger bank of mouse and human primary tumors. First, we analyzed Hoxa7 and Gata2 expression in an additional set of 11 mouse tumors and 5 age-matched normal controls. Consistent with the original sample set, we found that Hoxa7 and Gata2 expression is significantly reduced in the tumor samples (p < 0.0001 and p = 0.0166, respectively) (Fig. 4a, c). We next tested the expression of our candidate genes in a bank of cDNA generated from 59 human tumor breast tissue samples and 17 normal breast tissue samples. This panel of cDNA included tumors and matching adjacent tumor-free tissue (solid circles), as well as normal breast tissue from reduction mammoplasties (open circles). Data for both Hoxa7 and Gata2 revealed that these candidate genes were also downregulated in the human breast tumor samples relative to the normal breast tissue (Fig. 4b, d, p = 0.0292 and p < 0.0001, respectively). Of additional biological relevance, when human breast tumors and matching tumor-free adjacent tissues were analyzed together for changes in gene expression levels, we observed that all pairs showed large differences in mRNA levels, with four of the six pairs showing a decrease in mRNA levels in tumors compared to the normal tissue (Supplementary Fig. 5). Taken together, both Hoxa7 and Gata2 expression are reduced in mouse and human primary tumor samples when compared to normal mammary tissue. To confirm that differential DNA methylation occurs at the HOXA7 promoter, we analyzed a small bank of human primary tumors and matching adjacent tumor-free tissue. Hypermethylation was observed in the tumors compared to the adjacent tissue in six of the eight pairs analyzed (Fig. 4e). Due to the limited amount of DNA material available for analysis, GATA2 promoter methylation could not be tested. Our finding validates previous studies, where HOXA7 methylation and expression is altered in tumors, and proposes GATA2 as a novel candidate gene in both mouse and human breast tumorigenesis.
Fig. 4

Hoxa7 and Gata2 expression is reduced in an additional set of mouse and human tumors relative to normal breast tissue. aHOXA7 expression in normal and PyMT tumor mouse mammary tissue (p < 0.0001). b Expression of Hoxa7 in a panel of normal and tumor human breast tissue (p = 0.0292). cGata2 expression in normal and PyMT tumor mouse mammary tissue (p = 0.0166). d Expression of GATA2 in the panel of normal and tumor human breast tissue (p < 0.0001). For human samples, open circles represent normal tissue from reduction mammoplasty, while closed circles represent from adjacent tumor-free tissue. e The methylation status of individual CpG dinucleotides within the promoter region of HOXA7 was analyzed for eight human breast tumors (T) and matching adjacent tumor-free tissue (N). Six of eight pairs show hypermethylation in the tumor for the region analyzed. Data are represented as boxplots showing the average level of methylation across the CpGs analyzed within this locus (the central bar marks the median, the lower and upper limits of the box mark the first and third quartiles, and the whiskers extend the 1.5 interquartile range from the box. The black dots represent outlier CpGs)

Discussion

Gene expression profiles and cytogenetic analysis of mouse models for breast cancer have been widely used to identify genes important for tumor initiation and progression (Andrechek et al. 2003; Bowen et al. 2005; Dorritie et al. 2004; Herschkowitz et al. 2007; McNeil et al. 2003; Montagna et al. 2002, 2003; Ried et al. 2004; Wang et al. 2007; Weaver et al. 2002; Ye et al. 2004). Although the importance of DNA methylation changes in human breast tumorigenesis has been established (Ordway et al. 2007; Widschwendter and Jones 2002), cytosine methylation alterations in mouse models of human cancer have not yet been demonstrated by genome-wide studies as a mechanism responsible for altered gene expression in mouse tumorigenesis. This mechanism is suggested by the observation that reduction of DNA methylation by DNA methyltransferase 1 knockdown results in mice that develop aggressive thymic tumors (Gaudet et al. 2003), and from an in vitro analysis of mouse cell lines (Demircan et al. 2009). Our study indicates that alterations of methylcytosine levels occur in vivo in PyMT-induced breast tumors and that these alterations mimic changes observed in human breast cancer. In addition, we demonstrated that the use of the PyMT mouse model to carry out epigenetic studies allowed for the discovery of new epigenetically dysregulated candidate genes that are important for human breast tumorigenesis.

We used the PyMT mouse model because it has been studied extensively (Guy et al. 1992), and gene expression analysis showed that these tumors have a “luminal” cell phenotype (Herschkowitz et al. 2007) similar to the most prevalent types of human breast tumors. By SKY analysis, we established that the isolated tumors had cytogenetic profiles that mimic human chromosomal rearrangements. We determined the level of aneuploidy and the frequency of chromosome rearrangements and observed that LM1, the sample with the lowest rate of chromosomal breaks, was also the tumor that maintained hypermethylation in the intergenic regions. The epigenome of somatic differentiated cells was heavily methylated in repetitive elements to preserve them in an inactive state. During tumorigenesis, hypomethylation of the genome largely affects intergenic and intronic regions and results in chromosomal instability and increased mutation events (Wilson et al. 2007). Hypomethylation at intergenic regions could thus represent a mechanism that drives genomic instability and formation of chromosomal rearrangements. The presence of hypomethylation in cancer cells at the premalignant stage and its effect on chromosomal instability has not been fully investigated. Due to the small portion of the genome present on our array and to the small sample size, we could not fine-map methylation changes to recurrent breakpoints or to specific types of repetitive elements. However, by performing integrative epigenetic and cytogenetic analyses, we learned about the contribution of epigenetic dysregulation to chromosome instability. Mouse models for breast cancer could therefore represent an excellent model for integrative cytogenetic and epigenetic studies to gain insights into the mechanisms that lead to chromosomal translocations in vivo.

To determine whether aberrant cytosine methylation is a key process for breast cancer transformation in the PyMT mouse model, we established a detailed map of DNA methylation changes that occur in mouse primary tumors by applying the high-resolution DNA methylation HELP assay. Our analysis revealed that in the mouse, as in humans, tumorigenesis results in promoter-specific hypermethylation. Though our array had limited coverage, we observed that the majority of DNA methylation changes in the mouse tumors occurred at human breast cancer-specific targets (Hoxa and Hoxd clusters). Acquisition of CpG methylation in homeobox genes during breast tumorigenesis is a frequent event in humans (Novak et al. 2006; Tommasi et al. 2009). The HELP assay we conducted did not have the statistical power to reveal biologically significant cytosine methylation changes, but it served as a tool to identify candidate genes. Expression analysis of a larger sample set supported the hypothesis that the two candidate genes identified in the mouse (Hoxa7 and Gata2) are downregulated in both mouse and human primary breast tumors. The observation that expression differences were more significant in the PyMT model than in human samples could be explained by the heterogeneity of human breast cancer. Our tissue bank includes human samples from a variety of breast cancer subtypes (including luminal, basal breast cancers), while the PyMT samples are all of luminal origin, with little genetic variability between tumors isolated from different animals (Herschkowitz et al. 2007).

The novel candidate gene Gata2 is a member of the Gata family of transcription factors, which are key regulators of gene expression in hematopoietic progenitors and nonhematopoietic embryonic stem cells. This is the first study that shows that Gata2 undergoes promoter hypermethylation and gene silencing in both mouse and human breast tumor samples. This observation is of particular interest because Gata3 is an essential regulator of mammary gland morphogenesis and differentiation (Asselin-Labat et al. 2007). Furthermore, the Gata family of proteins has been shown to have a critical role in maintaining the normal (nontransformed) cell state in the mammary gland (Kouros-Mehr et al. 2008). Our findings place Gata2 as a candidate player in breast tumorigenesis.

There are clear biological differences between mice and humans, and thus potential limitations to using the mouse as an experimental model. However, murine models have been essential to elucidating key processes in human breast carcinogenesis. The major advantage, as demonstrated by our studies, is the ability to compare epigenetic and cytogenetic variations in primary and metastatic tumors. This finding opens a new field of investigation where epigenetic changes, specific to metastasis, could be easily interrogated, and where integrative cytogenetic and epigenetic analysis could shed light on mechanisms that lead to chromosomal translocations during tumorigenesis. By performing genome-wide high-resolution studies using murine models for breast cancer, we can obtain key information on changes occurring at the premalignant-to-malignant transition and compare these alterations to metastatic-specific changes in a genetically controlled system.

Acknowledgments

We thank members of the Greally’s lab for constructive discussion and Zhixia Yang for her assistance with sample preparation. We thank the Shared Resources at Albert Einstein College of Medicine: GIF (Genome Imaging Facility) for help with the SKY; Dr. Shahina Maqbool and the Center for Epigenomics for assistance with the epigenomic studies; Brent Calder and the Computational Genomic Core for help with the data processing. We are grateful to Dr. Thomas Ried for providing the mouse SKY kits, to Dr. Jeffrey Pollard for providing the PyMT mice, and to Dr. Maria Figueroa for help with the R scripts.

Supplementary material

335_2011_9318_MOESM1_ESM.doc (744 kb)
Supplementary material 1 (DOC 744 kb)
335_2011_9318_MOESM2_ESM.tif (8.3 mb)
Supplementary material 2 (TIFF 8482 kb)
335_2011_9318_MOESM3_ESM.tif (8.3 mb)
Supplementary material 3 (TIFF 8509 kb)
335_2011_9318_MOESM4_ESM.tif (4.5 mb)
Supplementary material 4 (TIFF 4597 kb)
335_2011_9318_MOESM5_ESM.tif (6.4 mb)
Supplementary material 5 (TIFF 6578 kb)
335_2011_9318_MOESM6_ESM.tif (4.9 mb)
Supplementary material 6 (TIFF 4984 kb)

Copyright information

© Springer Science+Business Media, LLC 2011

Authors and Affiliations

  • Deanna Acosta
    • 1
  • Masako Suzuki
    • 1
  • Diana Connolly
    • 1
  • Reid F. Thompson
    • 1
  • Melissa J. Fazzari
    • 1
    • 2
  • John M. Greally
    • 1
  • Cristina Montagna
    • 1
    • 3
    • 4
  1. 1.Department of GeneticsAlbert Einstein College of Medicine of Yeshiva UniversityBronxUSA
  2. 2.Department of Epidemiology and Population HealthAlbert Einstein College of Medicine of Yeshiva UniversityBronxUSA
  3. 3.Department of PathologyAlbert Einstein College of Medicine of Yeshiva UniversityBronxUSA
  4. 4.Albert Einstein College of MedicineBronxUSA

Personalised recommendations