Triple Analysis of the Cancer Epigenome: An Integrated Microarray System for Assessing Gene Expression, DNA Methylation, and Histone Acetylation¹

Microarray approaches used to study functional DNA-protein interactions (1, 2, 3) have revealed recently that many transcription regulators are linked to chromatin remodeling (3, 4), placing this type of epigenetic change at the center of gene regulation. Repressed chromatin and gene silencing are associated with changes in DNA methylation and histone acetylation (5), and whereas these epigenomic modifications are widely recognized as contributing factors in human tumorigenesis, their molecular basis is not understood yet. One model suggests that methylated DNAs at the 5′-end regulatory regions recruit MBD⁴ proteins, which are known to complex with HDACs and other transcriptional corepressors (6). Deacetylation of lysine groups on histones 3 and 4 occurs via HDACs, resulting in a tighter interaction between negatively charged DNA and positively charged lysine, and a closed, repressive chromatin configuration (5, 6). How repressive chromatin structures assemble onto DNA is not clear, but changes in methylation status of CpG islands in gene promoters presumably play a central role (5). We developed recently a microarray approach called differential methylation hybridization for screening CpG methylation and identifying loci susceptible to epigenetic modifications in various cancers (7, 8, 9). However, to fully elucidate the functional relationship between DNA methylation and histone acetylation in gene silencing, a genomic microarray system for detecting changes in gene expression, DNA methylation, and histone acetylation would be necessary.

We developed an integrated “triple” microarray system to decipher the hierarchies of epigenetic regulation of gene expression in cancer cells. The microarray panel used in this novel approach contains 1507 ECISTs, short genomic fragments (0.2–2-kb) located at the 5′-end regulatory regions of genes (10). We used the GC-rich components of ECISTs for screening methylated CpG sites, the exon-containing portions (i.e., the first exons) for measuring levels of the corresponding transcripts, and the promoter sequences within ECISTs for identifying chromatins immunoprecipitated with antibodies against acetylated histones. It is well known that DNA methylation and histone acetylation work in concert to regulate gene expression, and this new microarray system provides an effective means of segregating at specific loci expression changes that occurred as a consequence of reversing promoter hypermethylation status by epigenetic treatments.

A human epithelial ovarian cancer cell line CP70 (gift from Dr. Robert Brown, University of Glasgow, Glasgow, United Kingdom) was cultured in the presence of vehicle (PBS) or DAC (0.5 μm; medium changed every 24 h). After 4 days, cells were either harvested or treated with TSA (0.5 μm) for 12 h and then harvested. Some cells were also treated with TSA alone for 12 h before harvest. DNA and RNA were isolated using the QIAamp Tissue and RNeasy kits (Qiagen), respectively.

To identify ECISTs (including the first exons), RLCS (11) was used to prepare targets for screening of CpG island clones derived from a genomic library, CpG Island library (12). In the presence of T4 RNA ligase, an RNA adapter (0.5 nmol, 5′-ACC GGA GCG GCA CGG GAA AUA GAG CAA CAG GAA A) was ligated to the 5′-ends of decapped mRNAs derived from the Stratagene Human Universal Reference RNAs. After reverse transcription, full-length cDNAs were amplified by long RT-PCR (TaqPlus Long PCR system; Stratagene) with the flanking 5′- and 3′-adapters (5′-GCA CGG GAA ATA GAG CAA CAG and 5′-GGC CGA CTC ACT GCG CGT CTT CTG, respectively). A low number of PCR cycles (18–25) were used to preserve the linearity of amplification. Amplified products were labeled with Cy3 fluorescent dyes as described (10) and hybridized to the CGI microarray panel. Hybridization and posthybridization procedures were performed.⁵ Hybridized slides were scanned with the GenePix 4000A (Axon). The acquired images and data were transferred to Excel spreadsheets for additional analysis using GenePix Pro 3.0. CGI loci with signal intensities 2-fold greater than local background were scored as positive for containing expressed sequences.

Preparation of methylation amplicons was carried out essentially as described (7). Briefly, CP70 DNA (∼1 μg) was digested with MseI and then ligated to a PCR-linker. The ligated DNA was digested with methylation-sensitive endonucleases BstUI and HpaII, and amplified with a linker primer by PCR. DNA obtained from a normal ovary tissue was prepared similarly. Genomic fragments containing methylated sites were protected from enzymatic restrictions and could be amplified; however, fragments containing unmethylated sites were digested and, thus, not present in the amplified samples. CP70 amplicon was labeled with Cy5 (red), whereas the control amplicon was labeled Cy3 (green). Both samples were cohybridized onto an ECIST microarray slide and processed as described (7).

Total RNA (100 μg) was prepared from control (vehicle treated) CP70 cells, or cells cultured with TSA and/or DAC. The RLCS method was used to generate full-length cDNAs. For quality control, the Rapid Amplification of cDNA Ends method was used to determine the integrity of 5′-ends of a few cDNA sequences (10). Cy5-labeled cDNAs from treated cells and Cy-3-labeled cDNAs from untreated cells were cohybridized to the ECIST panel, and microarray images obtained were processed accordingly.

The protocol used to identify immunoprecipitated E2F1 targets (2) was adapted for this study. To obtain a network of DNA-protein biopolymers, treated or untreated CP70 cells (2 × 10⁷ cells/assay) were cross-linked using 1% formaldehyde. Cell nuclei were collected by microcentrifugation, and cross-linked chromatin fibers were isolated and fragmented to ∼600-bp by sonication. Immunoprecipitation was carried out with 5 μg of antiacetylated histone H3 or H4 rabbit polyclonal antibody (Upstate) or no-antibody (negative control). DNA was additionally released by digesting the immunocomplex with proteinase K. Purified chromatin DNA (a total of ∼1 μg) was recovered from 10–15 preparations for fluorescent labeling. Microarray hybridization, posthybridization washing, and slide scanning have been described previously by us (2).

The Cy3 and Cy5 fluorescence intensities of hybridized ECIST spots were obtained for each experiment. Because Cy5 and Cy3 labeling efficiencies varied among samples, the Cy5:Cy3 ratio of each spot was normalized according to the global ratio in each microarray image. As described in our previous studies (7, 9, 10), the derived normalization factor was additionally verified based on 14 internal controls of which the adjusted ratios were expected to be 1. Microarray experiments were repeated twice. A self-hybridization study using two equal portions of a test DNA sample was conducted for quality control. These self-hybridizing spots usually had adjusted Cy5:Cy3 ratios approaching 1.

Plasmid DNA was prepared from ECISTs and sequenced using the DyeDeoxy Terminator reaction (Applied Biosystems) and the ABI PRISM 377 sequencer. The sequencing results were compared with GenBank for known sequence identities.

Sodium bisulfite modification of genomic DNA, which converts unmethylated but not methylated cytosine to uracil, was performed using the CpG Genome modification kit (Intergen). COBRA was performed as described (13). Briefly, ∼200 ng of treated DNA were used as the template for PCR with specific bisulfite primers (Table 1) for a given locus. ³²P-labeled PCR products were digested with BstUI, separated on 8% polyacrylamide gels, and subjected to autoradiography using a PhosphorImager (Amersham-Pharmacia).

cDNA and chromatin DNA were prepared as described earlier. Diplex PCR (for both test and control targets) was performed using the AmpliTaq Gold polymerase (Perkin-Elmer). For RT-PCR, primer pairs were used to amplify a region (average 200-bp) from the 3′-end of a test gene, whereas for ChIP-PCR, primers were designed to amplify a fragment in the promoter or first exon region (average 200-bp) of the test gene (see Table 1 for primer information). After 20–25 cycles of amplification, radiolabeled PCR products were run on 5–8% polyacrylamide gels. A PhosphorImager was used to analyze the dried gels, and densitometric analysis of the observed bands was performed using ImageQuant (Molecular Dynamics). The relative levels of gene expression or histone acetylation were normalized with the level of the control run in the same lane.

Using RLCS, we screened a library of ∼9000 CGIs (12) and recovered 1,507 ECIST-positive loci. To confirm whether these ECISTs were located at the 5′-ends of genes, nucleotide sequencing was performed on 250 of these loci. Sequencing data showed that: (a) 79% (198) contained sequences located in the promoter and first exon of known genes; (b) 16% (40) matched genomic sequences and may contain as yet uncharacterized expressed sequences; and (c) 5% (12) contained non-exon 1 expressed sequences. These results suggest that the ECIST loci identified here can be effectively used to assess epigenetic alterations in cancer cells.

To assess gene expression, DNA methylation, and histone acetylation in parallel, CP70 cells were treated with a demethylating agent, DAC, and/or an inhibitor of HDACs, TSA, and then subjected to triple microarray procedures (Fig. 1,A). Representative individual gene loci are marked by arrows in Fig. 1,B. At a hypermethylated locus in untreated CP70 cells (Fig. 1,B, top panels), DAC plus TSA treatments increased expression (normalized Cy5:Cy3 = 5.5) and histone hyperacetylation (3.4-fold relative to the control) of this gene. The combined treatment of DAC plus TSA also increased expression and histone hyperacetylation of a locus that was not hypermethylated in untreated CP70 cells (Fig. 1 B, bottom panels).

The total number of ECISTs up-regulated ≥4-fold by epigenetic treatments was determined. Treatment with DAC or TSA alone resulted in up-regulation of 29 (1.9% of 1507 loci) or 17 (1.1%) loci, respectively; however, a greater number of genes (150 or 10.4%; P < 0.001 versus either treatment alone) were up-regulated after the combined treatment (Fig. 2, A–C). The epigenetic treatments also resulted in down-regulation of a few ECIST loci (≤0.25-fold), but this response was not the focus of our investigation. Histone hyperacetylation was measured in the combined treatment and scored when a locus showed a normalized Cy5:Cy3 ratio 2-fold greater in the treated cells than that of untreated cells (2). Using this cutoff, hyperacetylated loci were detected in 3.6% (55; red circles in Fig. 2 C) of the 1507 ECISTs examined.

To identify hypermethylated ECISTs, a normalized Cy5:Cy3 ratio ≥1.5 relative to the control was used. This cutoff ratio was used by us to reliably identify hypermethylated CpG islands in various cancers (7, 9, 14). The genes up-regulated by the combined treatment of DAC plus TSA were additionally divided into two groups (Fig. 2,D): hypermethylated (group 1, yellow spots; see Table 2) and no detectable methylation (group 2, blue spots; see Table 3). As shown in Fig. 2 C, up-regulation of group 1 loci is more closely associated with histone hyperacetylation than that of group 2 loci (64%; 22 of 34 loci versus 28%; 33 of 116 loci).

Within group 1 genes, increased expression of only a few loci (n = 11) was observed after treatment with DAC alone; however, the combined treatment of DAC and TSA resulted in up-regulation of 34 loci (Fig. 2,D). No significant change in expression of group 1 genes was seen in CP70 cells treated with TSA alone. To confirm the microarray findings, three gene loci from Group 1 (HSPA.2, CYP27B1, and EIF1A) were additionally analyzed. Hypermethylation of the HSPA.2 CpG island in CP70 cells was confirmed using COBRA (Fig. 3, row 1, left panel), and no expression of HSPA.2 was detected in untreated CP70 cells using RT-PCR (Fig. 3, row 1, middle panel). However, HSPA.2 expression was increased by DAC treatment, remained unchanged after treatment with TSA alone, and was markedly increased by the combined treatment of DAC and TSA (Fig. 3, row 1, middle panel). Furthermore, after treatment of CP70 cells with DAC plus TSA, histones H3 and H4 in the promoter region of HSPA.2 were determined to be hyperacetylated using ChIP-PCR (Fig. 3, row 1, right panel). These results support previous reports (3, 4) that the concerted action of DNA demethylation and histone hyperacetylation resulted in synergistic re-expression of methylation-silenced genes.

In untreated CP70 cells, partial methylation of the CYP27B1 CpG island was observed, and expression of CYP27B1 was low; however, treatment of CP70 cells with DAC plus TSA resulted in histone hyperacetylation and increased expression of CYP27B1 (Fig. 3, row 2). Contrariwise, despite the strong hyperacetylation observed at the EIF1A locus, expression of EIF1A remained largely unaffected by the epigenetic treatments. The EIF1A locus we identified, located on human chromosome 1 (15), was determined to be hypermethylated in CP70 cells by using COBRA. It has been reported that multiple copies of EIF1A exist at different chromosomal regions, e.g., chromosomes X and Y (16), and it seems reasonable to suggest that one or more of these loci remain unmethylated, and, thus, contribute to the basal expression of EIF1A detected by RT-PCR (Fig. 3, row 3).

A total of 116 loci were up-regulated (≥4-fold) by the epigenetic treatments (blue spots; see also Fig. 2,D), but expression of these loci appeared to be unrelated to DNA methylation. From this group, 8 loci were additionally analyzed using COBRA, RT-PCR, and ChIP-PCR (Fig. 4, A and B). The loci were unmethylated in CP70 cells, and expression of these loci was low or absent in untreated CP70 cells. Increased expression of some of these loci was observed after treatment with DAC or TSA alone. The combined treatment induced expression of all 8 of the loci, but histone hyperacetylation was seen in only the promoter regions of MDS1, SC13C2, and UNG2 (Fig. 4 A). On the basis of the response of these 8 loci to the epigenetic treatments, we additionally subdivided the methylation-independent loci into two groups: group 2a, methylation-independent, histone acetylation-enhanced genes (n = 33) and group 2b, methylation- and histone acetylation-independent genes (n = 83).

To additionally define epigenetic modifications and order of epigenomic events at CpG islands on a global scale, we have developed a microarray system that combines gene expression, DNA methylation, and DNA-protein interaction analyses. To our knowledge, this represents the first report of a genomic approach that is capable of dissecting the complex hierarchy of transcriptional controls orchestrated by the epigenomic machinery. This integrated microarray system allows for both the identification of individual genes and a systematic analysis of the relationship among the epigenetic machinery, promoter targets, and downstream responses regulated by the epigenome.

It has been demonstrated that pharmacological reversal of promoter hypermethylation status results in global and specific changes in gene expression (3, 5); in addition, inhibiting DNA methylation has both primary (direct) and secondary (indirect) effects on gene expression (3, 17, 18). Using the triple analysis approach, we identified both primary and secondary responses, and additionally categorized those responses into three groups of genes based on their methylation status: group 1, methylation-dependent, and groups 2a and b, methylation-independent. For group 1 genes, transcriptional silencing is dominated by methylation (Fig. 3). Reactivation of genes silenced by CpG methylation would presumably involve a series of steps, including removal of MBD proteins from demethylated DNA and/or transcriptional repressors that are recruited by MBD proteins (5). Epigenetic complexes have been shown to possess chromatin-remodeling activity and produce structures refractory to transcriptional activation (5). Disrupting these complexes would presumably diminish their activity and result in a more open, transcriptionally active chromatin configuration. A physical association between methylated DNAs and deacetylated histones has been shown recently (19), and our observation that synergistic reactivation of methylation-silenced genes (group 1) could only be achieved by the combined treatment is suggestive of a functional interaction between the epigenetic modifications. Whether this functional relationship is because of a direct or indirect interaction between the molecular targets remains to be elucidated.

The triple array analysis revealed an effect of the drug treatments on methylation-independent gene expression. Group 2a represents a class of distinct genes with unmethylated promoters of which the increased expression is produced by TSA alone or the combined treatment, but not by DAC alone. It is unclear how DAC and TSA act mechanistically on unmethylated promoters, but it was shown recently that DNA methyltransferase 1 (3), in the absence of DNA methylation, can directly suppress transcription through actions with HDACs (3, 19). Our observation that enhanced histone hyperacetylation of MDS1 required both DAC and TSA supports a role for a methylation-independent effect of DNA methyltransferase 1 in ovarian cancer cells; furthermore, these observations indicate that HDAC activity may play a role in the epigenetic-associated control of group 2a gene expression. The majority of genes we identified in this triple array analysis belonged to group 2b, which showed enhanced expression independent of both DNA demethylation and histone acetylation. Up-regulation of these loci by DAC plus TSA treatments is most likely because of an event downstream of modulations in the epigenetic cascade. There are several possible, but not mutually exclusive, mechanisms that may account for this secondary effect, including increased post-transcriptional processing (RNA stability), reactivation of an upstream transcription factor, or regulation by target genes in an induced signal transduction pathway (20).

Induction of some of the genes in group 2 is likely to be associated with cellular responses to drug toxicity or stress, as shown recently by several groups using microarrays to examine gene expression profiles in DAC-treated human cancer cell lines (3, 17, 21). Furthermore, many of the stress-response genes induced by DAC show similar early and transient expression characteristics (21). For example, early induction of the apoptosis promoting factor BIK was observed after DAC treatment of a human lung cancer cell line, and BIK expression returned to control levels by 72 h after treatment with DAC (21). Interestingly, the BIK gene, which does not contain a CpG island, is also induced in a methylation-independent manner by TSA (3). In contrast, DAC treatment gradually induces expression of methylation-dependent genes and their downstream targets (17, 21), and expression of these genes has been shown to be prolonged or increased as demethylation progresses (17, 21). In this regard, our triple microarray system is well suited for distinguishing early stress-response genes from late genes induced by epigenetic treatments over time, and our future studies will investigate this and the effect of other modulators on epigenetic pathways.

The current study offers proof of principle for a triple microarray system capable of interrogating the complex hierarchy of epigenetic changes often seen in human cancer. This integrated approach is useful for identifying novel therapeutic targets and more fully understanding the mechanisms underlying epigenetic gene silencing. The continued development of the triple microarray will be useful for assessing the specificity of emerging epigenetic therapies based on reactivating the expression of methylation-silenced genes in cancer and other diseases.

We thank Diane Peckham for assistance in the preparation of this manuscript.