Epigenetic Inactivation of the HOXA Gene Cluster in Breast Cancer

Using an integrated approach of epigenomic scanning and gene expression profiling, we found aberrant methylation and epigenetic silencing of a small neighborhood of contiguous genes—the HOXA gene cluster in human breast cancer. The observed transcriptional repression was localized to ∼100 kb of the HOXA gene cluster and did not extend to genes located upstream or downstream of the cluster. Bisulfite sequencing, chromatin immunoprecipitation, and quantitative reverse transcription-PCR analysis confirmed that the loss of expression of the HOXA gene cluster in human breast cancer is closely linked to aberrant DNA methylation and loss of permissive histone modifications in the region. Pharmacologic manipulations showed the importance of these aberrant epigenetic changes in gene silencing and support the hypothesis that aberrant DNA methylation is dominant to histone hypoacetylation. Overall, these data suggest that inactivation of the HOXA gene cluster in breast cancer may represent a new type of genomic lesion—epigenetic microdeletion. We predict that epigenetic microdeletions are common in human cancer and that they functionally resemble genetic microdeletions but are defined by epigenetic inactivation and transcriptional silencing of a relatively small set of contiguous genes along a chromosome, and that this type of genomic lesion is metastable and reversible in a classic epigenetic fashion. (Cancer Res 2006; 66(22): 10664-70)

Epigenetic inactivation linked to aberrant methylation of CpG islands is a fundamental participant in human carcinogenesis and is often coupled to associated changes in histone state, chromatin structure, and gene silencing (1). Aberrant DNA methylation in cancer cells is involved in the pinpoint inactivation of individual genes and, recently, has also been shown to afflict large chromosomal domains (2, 3). These lesions are analogous to the genetic events of gene mutation and chromosomal deletion. The HOXA gene cluster is a family of homeotic genes that encode transcriptional regulators that play critical roles in the development and differentiation of many multicellular organisms (4), and HOXA1 to HOXA11 are expressed in normal breast epithelium (Supplementary Fig. S1). Results from the study presented herein show that this small neighborhood of contiguous genes, which inhabits ∼100 kb of DNA, undergoes selective epigenetic inactivation in human breast cancer. Gene expression analysis indicates that this epigenetic inactivation is limited to the HOXA cluster and does not extend to genes upstream or downstream of this region. Overall, these results suggest that inactivation of the HOXA gene cluster in breast cancer is an example of epigenetic microdeletion. This new type of lesion is defined by an epigenetic inactivation that is targeted to a contiguous set of genes along a chromosome and structurally and functionally resembles a genetic microdeletion in that it affects a relatively small genomic region that results in the loss of expression of the genes in the region. Unlike genetic microdeletions, in the case of epigenetic microdeletions, the inappropriate gene silencing is metastable and reversible in a classic epigenetic fashion.

Cell culture. All cell lines were maintained and treated with 5-aza-2′ deoxycytidine (5-aza-dCyd) as previously described (5). Cells were treated with trichostatin A alone for 24 hours or added to the 5-aza-dCyd treatments for the final 24 hours before RNA isolation.

Breast tumor specimens. Flash-frozen surgical specimens from normal or cancerous breast tissue were obtained from breast cancer patients. All patients signed surgical and clinical research consents for tissue collection in accordance with the University of Arizona Institutional Review Board and Health Insurance Portability and Accountability Act regulations.

Nucleic acid isolation. RNA and DNA were isolated as previously described (6).

Bisulfite sequencing. Genomic DNA was analyzed by bisulfite sequencing as described (7). The HOXA CpG islands were amplified from the bisulfite-modified DNA by two rounds of PCR using nested primers. Primer sequences are available in Supplementary Table S1.

Real-time reverse transcription-PCR. Quantitative reverse transcription-PCR (RT-PCR) analysis was done as previously described (6). Primer/probe sequences or ABI Assay ID numbers are available on request.

Chromatin immunoprecipitations. Chromatin immunoprecipitations and quantitative real-time PCR were done as previously described (5). Antibodies used for the analysis of histone modifications were obtained from Upstate Biotechnology (Temeculah, CA). Primer sequences are available on request.

CpG island microarrays. Six thousand eight-hundred clones from a human CpG island library (8) were arrayed, DNA sequence validated, and prepared for microarray as described (9).

HOXA tiling array. PCR primers were designed using Primer3 for a region spanning the HOXA cluster (chr7:26900200-27019000) using the May 2004 build of the human genome. The average size of the 170 products was 496 bp with gaps between amplicons averaging 199 bp. PCR primers were also designed for a region spanning GAPDH (chr12:6512873-6519114) and ACTB (chr7:5338000-5344715) each with an average amplicon length of 492 bp to serve as controls. Primer sequences are available on request.

Methylcytosine immunoprecipitation. DNA was immunoprecipitated with a 5-methylcytosine antibody, and the immunoprecipitated and input DNA were fluorescently labeled for microarray hybridization as described (10). Microarray slides were hybridized, washed, and scanned as described (9).

McrBC methylation analysis of DNA. McrBC digestion and methylation analysis were done as previously described (9).

CpG Island and HOXA microarray hybridization, washing, and scanning. Microarray slides were hybridized, washed, and scanned as described (9).

Microarray data processing. Median signal intensity of each spot was extracted using GenePix software and used for analysis.

McrBC methylation data were normalized using Bioconductor with the LIMMA package (11). Spots flagged as bad or not found were not used. Robust spline normalization and Gquantile were used intra-array and interarray, respectively. Normalized log 2 ratios were analyzed using BRB ArrayTools⁶

to find CpG islands differentially methylated between tumor and normal breast specimens. Only probes with <50% of missing values were used for statistical testing. Parameters of testing were the following: significance level, 0.01; confidence level of false discovery rate assessment, 90%; and maximum allowed proportion of false-positive genes, 15%.

Data obtained from the methylcytosine immunoprecipitation and HOXA arrays were normalized in Excel. First, intensity-based normalization was done according to X_k = log₂(R_k / G_k) − A_k × f, where X_1k is the normalized ratio and R_k and G_k are red and green foreground median spot intensities of spot k, respectively. A_k is overall magnitude of the spot intensity [log₂(R_k) + log2(G_k)] / 2 and f is array-specific normalization factor. The value of normalization factor f was calculated iteratively to minimize the sum of SDs of replicated spots on the array. In the second normalization step, the median ratio of all housekeeping genes was subtracted from X_k. For visualization of DNA methylation profile, normalized ratios were ordered according to physical position in HOXA cluster and smoothed using average over three neighboring loci.

Affymetrix microarray expression data. Affymetrix HG-UI33A plus 2.0 GeneChips were used according to the protocols of the manufacturer. CEL files were analyzed using the GC-RMA algorithm (12) to produce normalized transcript-level signal. The median normal tissue expression was than used for per-gene normalization. To exclude low-quality data, all probes with raw signal below threshold (median of all intensities) in more than 80% of the arrays were removed.

Hierarchical clustering and data visualization. All hierarchical clustering was done in Cluster 3.0 software (University of Tokyo, Human Genome Center)⁷

and heat maps were created in Java TreeView 1.0.12 software.

Normal breast tissue and invasive breast cancers were analyzed for DNA methylation state using CpG island microarrays (refs. 9, 13; see Supplementary Table S2 for pathologic assessment of the samples and the assays done on each specimen). 271 CpG island microarray elements that displayed significant differences between normal and cancerous tissues were identified using BRB array tools with a false discovery rate of 15% (Supplementary Table S3) and then analyzed based on their genomic location, such as promoter or satellite sequences. Hierarchical clustering of the data showed that tumor tissue can be distinguished from healthy tissue based on genome-wide DNA methylation patterns (Fig. 1). Differences between normal and cancer tissues could be seen throughout the genome, including single-copy CpG island promoters as well as highly repetitive satellite and Alu sequences. In breast cancer in general, satellite sequences became relatively hypomethylated whereas single-copy CpG island elements and Alu sequences became aberrantly methylated, consistent with earlier work (13–15).

New and previously identified single-copy CpG islands that become aberrantly methylated in breast cancer were identified by our CpG island arrays. An example of a CpG island previously reported to be methylated and its associated gene silenced in breast cancer was CAVEOLIN-1 (16). New aberrantly methylated CpG islands identified included the islands associated with HOXA1 and HOXA7. Interestingly, in addition to HOXA1 and HOXA7, the rest of the HOXA genes within the cluster are also associated with CpG islands at their 5′ ends, and aberrant methylation of the HOXA5 and HOXA9 promoter in breast cancer has been reported (17, 18). Together, these results led us to hypothesize that the entire contiguous region of the HOXA cluster is a target of aberrant methylation and epigenetic gene silencing in human breast cancer.

To address this hypothesis, we analyzed the gene expression data for the HOXA, HOXB, HOXC, and HOXD gene clusters obtained by Affymetrix transcription profiling from a random subset of 22 of these breast cancers and 4 normal breast tissues, as well as 4 breast cancer cell lines, the nontumorigenic breast epithelial cell line MCF10A, and a normal human mammary epithelial cell strain designated HMEC (Fig. 2A). Overall, there was a selective decrease in the expression of the entire HOXA cluster in the cancerous samples when compared with the noncancerous samples, whereas the paralogous HOX clusters HOXB, HOXC, and HOXD showed no consistent changes in gene expression. Furthermore, the transcriptional repression of the HOXA gene cluster was localized to this neighborhood of genes as analysis of expression of genes extending 2.5 Mb 5′ and 3′ of the HOXA cluster does not show decreases in expression, thereby showing that the repression is localized to the HOXA gene cluster (Fig. 2B). To confirm the Affymetrix gene expression data, we analyzed HOXA gene expression by quantitative real-time RT-PCR (Fig. 2C; Supplementary Table S4). Results from these analyses confirmed a significant decrease or loss of expression of the HOXA1 to HOXA10 genes in invasive breast cancer when compared with normal tissue. This statistically significant diminution in the HOXA cluster did not extend to HOXA11 and HOXA13. The repression of the HOXA cluster seen in vivo extended to in vitro cell line models as well (Fig. 2D). Additional analysis of this data revealed that in this gene cluster, there was a high correlation in expression of neighboring genes in the cluster, suggesting that epigenetic silencing of one HOXA gene may influence the expression of other HOXA genes in their proximity, and therefore the expression of particular gene may not be driven only by the epigenetic state of the promoter region of an individual gene but is also dependent on the epigenetic state of whole HOXA cluster (Supplementary Fig. S2).

To confirm and extend the results obtained from CpG island microarrays, we used bisulfite sequencing to analyze the methylation state of the 5′ CpG islands of the HOXA cluster in a set of the invasive breast cancer samples, normal breast tissue, and in vitro models. The bisulfite analysis of the methylation state of the HOXA1 and HOXA7 CpG island also served to validate the CpG island microarray results. The CpG island regions analyzed are shown schematically in Fig. 3A, and representative results of this analysis are shown in Fig. 3B. (Supplementary Fig. S3 provides the remaining portion of the bisulfite sequencing data set.) Despite contaminating normal breast and stromal tissues present in the breast cancer specimens, bisulfite sequencing confirmed aberrant DNA methylation of HOXA1 and HOXA7 initially revealed by the CpG island microarray results, and extended the results to CpG islands associated with the other HOXA genes, with the exception of HOXA13. Analysis of in vitro models recapitulated the observations made in clinical breast cancer specimens that loss of HOXA gene expression was similarly linked to extensive methylation of HOXA CpG islands. Bisulfite sequencing analysis of MDA-MB-231 and Bt549 revealed widespread methylation of the HOXA gene cluster (Fig. 3B and Supplementary Fig. S3). In contrast, a normal human mammary cell strain (HMEC), which expressed all HOXA genes, except HOXA13, revealed a strikingly different result, showing largely unmethylated CpG islands for the HOXA genes similar to those seen in the normal in vivo setting. Overall, the results from this high-resolution analysis show that aberrant DNA methylation of multiple HOXA CpG islands is a common event in invasive human breast cancer.

To gain a more complete picture of the DNA methylation state of the HOXA cluster, we constructed a HOXA tiling microarray that covered ∼125 kb on chromosome 7. The DNA methylation state of 10 breast tumors, three normal specimens, MDA-MB-231, Bt549, and HMEC was analyzed using the methylcytosine antibody immunoprecipitation approach (10). Results from these experiments are shown in Fig. 3C. Overall, extensive DNA methylation of the HOXA cluster that extended beyond the regions analyzed by bisulfite sequencing could be consistently detected in the tumor specimens and breast cancer cell lines. In contrast, normal specimens and HMEC were largely unmethylated in CpG islands as well as inter–CpG island regions, although regions associated with HOXA9 and HOXA10 did reveal some methylation in normal tissue. Overall, these results show that extended regions of the HOXA cluster, just beyond the CpG islands, become aberrantly methylated in breast cancer.

This aberrant methylation was tightly linked to the histone modification state as shown in Fig. 3D. Chromatin immunoprecipitations of various histone modifications were done on HMEC, MDA-MB-231, and Bt549 and regions associated with the various HOXA genes were analyzed by quantitative real-time PCR. Repressive histone modifications such as hypoacetylation of histones H3 and H4 were associated with inappropriate HOXA gene silencing, whereas the permissive histone modifications of histone H3 and H4 hyperacetylation and H3 K4 dimethylation were associated with HOXA gene expression. Taken together, the results presented in Figs. 1–3 show that inappropriate silencing of the HOXA gene cluster in breast cancer is widely seen in vivo and in vitro and is closely linked to the presence of aberrant methylation and repressive histone modifications of CpG islands in the HOXA cluster, and these repressive epigenetic marks are limited to a small contiguous stretch of chromosome 7 that is on the order of 100 kb in length.

To investigate the functional importance of aberrant DNA methylation and histone modification state in the transcriptional repression of the HOXA cluster, we treated breast cancer cells in vitro with the DNA methyltransferase inhibitor 5-aza-dCyd (19). All silenced HOXA genes in MDA-MB-231 and Bt549 were reactivated by this treatment, supporting an important role for aberrant DNA methylation in the inappropriate repression of HOXA gene expression in breast cancer cells (Fig. 4). In contrast, trichostatin A alone was not sufficient to reactivate silenced HOXA genes in these cell lines, supporting the concept of a hierarchy to the epigenetic layers of gene control where aberrant DNA methylation is dominant to the repressive epigenetic layer of histone hypoacetylation (20). Further support of an epigenetic hierarchy is provided by experiments showing that treatment of the breast cancer cells with 5-aza-dCyd before inhibition of histone deacetylation complexes with trichostatin A resulted in enhanced reexpression of the HOXA cluster. Interestingly, even HOXA13 gene activation occurred following treatment with 5-aza-dCyd plus trichostatin A. This activation of HOXA13 may occur as a result of coordinate regulation of the HOXA genes, which is supported by the correlations shown in Supplementary Fig. S2 suggesting that expression of one HOXA gene in the cluster may influence the expression of other HOXA genes in their proximity. Overall, these results show the importance of DNA methylation in the epigenetic silencing of the HOXA gene cluster in breast cancer.

In conclusion, the silencing of the HOXA gene cluster in breast cancer is associated with the acquisition of the repressive epigenetic mark of DNA hypermethylation and the loss of permissive histone modifications. This epigenetic gene silencing spans several genes over ∼100 kb of contiguous DNA and suggests the existence of a new genomic lesion in human cancer, epigenetic microdeletion. We predict that a number of these events occur in the cancer genome.

Grant support: Grants R01CA65662 and R33CA091351 (B.W. Futscher), Center Grants P30ES06694 and P30CA023074, and Training grants ES007091 and CA09213 (M.M. Oshiro, T. Jensen, and R.J. Wozniak).

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.