Transcriptional Gene Silencing Promotes DNA Hypermethylation through a Sequential Change in Chromatin Modifications in Cancer Cells

Methylation of DNA occurs at cytosine residues at CpG dinucleotides by the addition of a methyl group to the carbon-5 position of cytosine through the action of the DNA methyltransferase enzymes (1). Approximately 70% of all CpG dinucleotides in the mammalian genome are methylated and these sites commonly occur in repetitive DNA elements (2). In contrast, most unmethylated CpG sites are located within the 29,000 CpG islands found spanning the promoter and first exon regions of housekeeping genes and tumor suppressor genes. The precise mechanism by which these regions remain methylation free is unclear, but active transcription appears to be an essential component (3, 4). Methylation of CpG islands, however, is associated with gene silencing such as in X chromosome inactivation and genomic imprinting (5).

In cancer cells, DNA hypermethylation of tumor suppressor genes is thought to play a critical role in carcinogenesis (6). Methylation-induced suppression is thought to occur either by directly interfering with the binding of transcription factors or through the action of methylated DNA binding (MBD) proteins and chromatin modification. The specific MBD proteins, which include MeCP2, MBD1, MBD2, MBD3, and MBD4, all have a conserved methyl-CpG binding domain and specific methylated DNA binding properties (7). MeCP2 was the first described member of the MBD family and can bind to single methylated CpG dinucleotides regardless of sequence context (8). MeCP2 is reported to repress transcription by binding to methylated DNA and recruiting Sin3 through its transcriptional repression domain, which interacts with histone deacetylases (HDACs; Ref. 9). MBD2 can also bind to a single methylated CpG site in solution binding assays and localizes to major satellite DNA (10). MBD2 was reported to be the methyl binding component of the MeCP1 complex (11), another transcriptional repression complex, which interacts with the Sin3 (12) or NuRD complex (13), and is also associated with HDAC1 and HDAC2 and MBD3. In contrast to MeCP2, the MeCP1 complex requires a densely methylated DNA fragment (11 or more methyl CpG sites) for productive binding.

The structure of chromatin and the activity of the associated genes are clearly mediated by multiple modifications of its components. Histone proteins are subject to many different chemical modifications, including phosphorylation, acetylation, and methylation, and these modifications affect access of regulatory factors and complexes to chromatin and influence gene expression. Acetylation of lysine residues on histones H3 and H4 leads to formation of open chromatin structure (14), whereas deacetylation is associated with repressed chromatin. Methylation at lysine 4 of histone H3 (H3-K4) is associated with promoters of active genes, whereas methylation at H3-K9 is associated with promoters of inactive genes (15, 16).

To understand the process that triggers hypermethylation of CpG island-associated genes in cancer cells, it is important to understand the temporal relationship between CpG methylation, the binding of MBD proteins, and the acetylation and methylation of associated histones. In a previous study, we reported that a combination of both gene silencing and random seeds of CpG methylation is necessary to trigger hypermethylation of the unmethylated CpG island spanning the promoter and first exon of the glutathione S-transferase (GSTP1) gene in prostate cancer cells (3). The CpG island promoter region of the GSTP1 gene becomes methylated in the majority of prostate tumors. In normal tissues, the gene is expressed and unmethylated; however, in prostate cancer, the gene is inactivated, and both alleles are extensively methylated (17). In this study, we have followed the sequential changes in histone H3 acetylation and methylation and binding of MeCP2 and MBD2 during the process of DNA hypermethylation of the GSTP1 gene in prostate cancer cells. Our findings have important implications for understanding the relationship between chromatin modification and abnormal hypermethylation of CpG islands in cancer cells.

LNCaP prostate cancer cells that had been stably transfected with GSTP1 plasmid constructs were cultured in T-medium with 10% heat-inactivated FCS, 1% penicillin/streptomycin (Invitrogen), and 900 μg/ml Geneticin (G418; Life Technologies, Inc.), as described previously (3). The cells were grown at 37°C in 5% CO₂ atmosphere and split 1:3 every 4–5 days.

The constructs used in this study have been described in detail in Song et al. (3). Briefly, The Shuttle vector contains the entire GSTP1 CpG island extending from 1.2 kb upstream of the transcription start site to intron 4 cloned into pBluescript SK⁺. A modification of the Nar1 site resulted in an extra CpG dinucleotide at CpG –23, which was used as an informative marker to distinguish the endogenous GSTPI gene from transfected GSTP1 construct. The Sp1 deletion vector was constructed by deletion of the minimal promoter region [Nar1(1150)/StuI(1216) fragment] from the GSTP1 shuttle vector. The seeded shuttle vector and the seeded Sp1 deletion vector were prepared by methylation of the shuttle and Sp1 deletion vectors with HpaII methylase (New England Biolabs) according to the manufacturer’s protocol. Complete methylation of the HpaII sites was confirmed by resistance to restriction by HpaII enzyme and by bisulfite sequencing.

DNA was extracted from the LNCaP cells using the Puregene extraction kit (Gentra Systems). The bisulfite reaction was carried out on 2 μg of restricted DNA for 16 h at 55°C under conditions as described previously (18). After neutralization, the bisulfite-treated DNA was ethanol precipitated, dried, resuspended in 50 μl of H₂O, and stored at −20°C. Approximately 2 μl of DNA were used for each of the nested PCR amplifications (3). The GSTP1 primers used for the amplification were GST9 and GST10 for first round and GST11 and GST12 for second round, as described by Millar et al. (17). At least three independent PCR reactions were performed to ensure a representative methylation profile. PCR fragments were directly purified using the Wizard PCR DNA purification system and cloned into the pGEM-T-Easy Vector (Promega) using the Rapid Ligation Buffer System (Promega). Approximately 12 individual clones were sequenced from the pooled PCR reactions using the Dye Terminator cycle sequencing kit with AmpliTaq DNA polymerase FS (Applied Biosystems) and the automated 373A NA Sequencer (Applied Biosystems). The methylation status for each CpG site was determined, and the average was calculated from the sequencing data of at least 12 individual clones.

RNA was extracted using Trizol reagent (Invitrogen) according to the manufacturer’s protocol. cDNA was reverse transcribed from 2 μg of total RNA using SuperScript III RNase H⁻ Reverse Transcriptase (Invitrogen-Life Technologies, Inc.), according to the manufacturer’s instructions. The reaction was primed with 200 ng of random hexamers (Roche). The reverse transcription reaction was diluted 1:20 with sterile H₂O before addition to the reverse transcription-PCR. Expression from the GSTP1-transfected plasmid was quantitated using a fluorogenic real-time detection method using the ABI Prism 7000 Sequence Detection System. Five μl of the reverse transcription reaction were used in the quantitative real-time PCR reaction using 2× SYBR Green 1 Master Mix (P/N 4309155) with 50 ng of each primer. The forward primer (5′-TCAAAGCCTCCTGCCTATACG-3′) was designed to cover the exon 3/exon 4 junction of GSTP1, and the reverse primer (5′-GCGAGCTCTAGCATTTAGGTGA-3′) was designed to the bovine growth hormone 3′-polyadenylation signal, which had been cloned into the GSTP1 shuttle vector (3). To control for the amount and integrity of the RNA, the Human 18S rRNA kit (P/N 4308329; Applied Biosystems), containing the rRNA forward and reverse primers and rRNA VIC probe, was used. Five μl of the reverse transcription were used in a 20-μl reaction in TaqMan Universal PCR Master Mix (P/N 4304437) with 1 μl of the 20× Human 18S rRNA mix. The reactions were performed in triplicate, and the SD was calculated using the Comparative method (ABI PRISM 7700 Sequence Detection system User Bulletin no. 2, P/N 4303859). The cycle number corresponding to where the measured fluorescence crosses a threshold is directly proportional to the amount of starting material. The mean expression levels are represented as the ratio between GSTP1 and 18S rRNA expression.

ChIP assays were carried out according to the manufacturer (Upstate Biotechnology). Briefly, ∼1 × 10⁶ LNCaP cells, in a 10-cm dish, were fixed by adding formaldehyde at a final concentration of 1% and incubating for 10 min at 37°C. The cells were washed twice with ice-cold PBS containing protease inhibitors (1 mm phenylmethylsulfonyl fluoride, 1 μg/ml aprotinin, and 1 μg/ml pepstatin A), harvested, and treated with SDS lysis buffer for 10 min on ice. The resulting lysates were sonicated to shear the DNA to fragment lengths of 200–500 bp. The complexes were immunoprecipitated with antibodies specific for acetylated-histone H3 (no. 06-599), MeCP2 (no. 07-013), dimethyl-histone H3(lys9) (no. 07-212), and MBD2 (no. 07-198) from Upstate Biotechnology. Ten μl of antibody were used for each immunoprecipitation according to the manufacturer. No antibody controls were also included for each ChIP assay, and no precipitation was observed. The antibody/protein complexes were collected by salmon sperm DNA/protein A agarose slurry and washed several times following the manufacturer’s instructions. The immune complexes were eluted with 1% SDS and 0.1 m NaHCO₃, and the cross-links were reversed by incubation at 65°C for 4 h in the presence of 200 mm NaCl. The samples were treated with proteinase K for 1 h, and the DNA was purified by phenol/chloroform extraction, ethanol precipitation, and resuspended in 30 μl of H₂O.

The amount of GSTP1 DNA that was immunoprecipitated with each antibody was measured by Real-Time PCR using the ABI Prism 7900HT Sequence Detection System. Amplification primers: forward primer, GST 82U (1034–1048) 5′-GCTGCGCGGCGACTC-3′ or GST85U 5′-TTCCCCGGCCAGCTG-3′ and reverse primer, GST 83U (1091–1003) 5′-GGCGGCCGCTGCA-3′. A TaqMan probe [labeled with FAM (1056–1074) 5′-6FAM-TCCAGGGCGCGCCCCTCT-TAMRA-3′] was designed so that it only detected transfected GSTP1 sequences and did not detect endogenous GSTP1. A TaqMan probe [labeled with VIC (1054–1073) 5′-VIC-ACTCCAGGGCGCCCCTCTGC-TAMRA-3′) was designed to detect endogenous GSTP1 sequences. PCR reactions were set up according to the SDS compendium (version 2.1) for the 7900HT Applied Biosystems Sequence Detector. Twenty-μl reactions were set up using the TaqMan Universal PCR Master Mix (2×) in a 396-well plate. Four μl of either immunoprecipitated DNA, no-antibody control, or input chromatin were used in each PCR, and the PCRs were set up in triplicate. Universal thermal cycling conditions were used: 50°C for 2 min, then 95°C for 10 min, followed by 95°C for 15 s and 60°C for 1 min repeated for 40 cycles. At the completion of the run, the threshold bar was set in the exponential phase of the logarithmic graph of the amplification plot and the C_T values recorded. The cycle number that the measured fluorescence crosses a threshold is directly proportional to the amount of starting material. Outlying samples were excluded from the analysis. SD was calculated using the Comparative method (ABI PRISM 7700 Sequence Detection System User Bulletin no. 2, P/N 4303859). For each sample, an average C_T value was obtained for immunoprecipitated material and for the input chromatin. The difference in C_T values (δ C_T) reflects the difference in the amount of material that was immunoprecipitated relative to the amount of input chromatin, and this was then expressed relative to the C_T value for LNCaP cells (ABI PRISM 7700 Sequence Detection system User Bulletin no. 2, P/N 4303859).

We previously reported that de novo methylation of the CpG island, spanning the promoter and first exon of a transfected copy of the GSTP1 gene, can be triggered in LNCaP prostate cancer cells, if the gene is inactive (deletion of Sp1 sites) and has a low level of methylation (3). In this study, to address what the relationship is between the process that triggers DNA hypermethylation and modification of the associated chromatin, we have used a subset of the GSTP1 constructs for transfection into LNCaP cells, summarized in Fig. 1,A. These include two pairs of vectors, one pair that express GSTP1 (shuttle) and one pair where the Sp1 sites had been deleted (Sp1 deletion). We used quantitative real-time reverse transcription-PCR to confirm that the GSTP1 shuttle vector was transcriptionally active and the Sp1 deletion vector was transcriptionally inactive (Fig. 1,B). For each pair of constructs, one partner was transfected in the unmethylated state, and one was transfected with seeds of methylation after HpaII methylation. After transfection and selection by G418 resistance, colonies were selected and pooled. DNA was isolated after a number of serial passages and analyzed for methylation of the transfected GSTP1 CpG island by bisulfite genomic sequencing. Fig. 2 summarizes the methylation status of the endogenous and transfected GSTP1 gene constructs. In contrast to the endogenous GSTP1 CpG island that is densely methylated in LNCaP cells (Fig. 2,A), the two expressing constructs (shuttle and seeded shuttle) remained unmethylated or became substantially demethylated (Fig. 2, B and C). This is in accordance with our previous findings where we also found that active transcription from the GSTP1 shuttle vector promoted demethylation of the HpaII seeds of methylation (3). Hypermethylation of the CpG island was only triggered in the construct that was transcriptionally inactive (by removal of the Sp1 sites) and seeded with a low level of methylation at HpaII sites. In this construct, de novo methylation of the CpG sites flanking the HpaII sites occurred, and in addition, the density of methylation was shown to increase across the island with additional doublings (Fig. 2, E and F).

To address how the seeds of methylation aid in triggering the subsequent hypermethylation of the island, we analyzed the chromatin modification across the GSTP1 CpG island in the various constructs after transfection. We used a ChIP assay using antibodies against acetylated and methylated histone H3 and measured the amount of GSTP1-specific DNA that was released from the immunoprecipitates by quantitative real-time PCR. To differentiate the chromatin state of the transfected GSTP1 sequences from the endogenous GSTP1 gene, we designed Taqman probes specifically to either the transfected GSTP1 sequence or the endogenous GSTP1 sequence (Fig. 3,A). All of the transfected GSTP1 constructs have an extra CG at base 1061. A FAM-labeled Taqman probe was designed to detect this extra CG, and a VIC-labeled Taqman probe was designed that was specific to the wild-type GSTP1 gene for the same region using the same reverse primer and a different forward primer (GST85U). The FAM probe was totally specific for transfected GSTP1 DNA, and it did not detect the endogenous gene sequence (Fig. 3,B). The VIC probe detected the endogenous gene in preference to the transfected sequence by 5 cycles (Fig. 3 B); however, because of the reduced specificity, we only used this probe to measure endogenous GSTP1 chromatin modification in untransfected LNCaP cells. The specificity of the FAM probe ensured we would only detect transfected GSTP1 sequence in the chromatin immunoprecipitation assays.

To determine whether seeding methylation triggered subsequent DNA hypermethylation by altering the acetylation state of the associated histones, we studied the H3 histone acetylation of the GSTP1 promoter region. Using ChIP and a polyclonal antibody generated to acetylated H3, we performed ChIP analysis on LNCaP cells and LNCaP cells transfected with different GSTP1 constructs (Fig. 1). After immunoprecipitation, the DNA was released, and the GSTP1 promoter region, associated with the transfected DNA, was analyzed by real-time PCR and distinguished from the endogenous gene by using the FAM-labeled Taqman probe. Fig. 4 A summarizes the extent of GSTP1 promoter region associated with acetylated H3 histones in the various GSTP1 constructs relative to the endogenous GSTP1 promoter from LNCaP cells. In the LNCaP cells, the endogenous GSTP1 promoter is deacetylated, i.e., it does not bind acetylated histone H3 antibodies, and this correlates with the promoter being extensively methylated and the gene silent. In contrast, the transfected shuttle or HpaII-seeded shuttle vector, which are both hypomethylated and actively transcribed, are associated with acetylated histone H3. Interestingly, the chromatin associated with the Sp1 deletion construct is also acetylated at histone H3, although in this construct, GSTP1 expression is inactive because of the promoter deletion. Deacetylation of histone H3 was only triggered in the chromatin of the HpaII-seeded GSTP1 construct that had become hypermethylated with increasing doublings (25d and 36d). Therefore, inactive transcription alone was insufficient to change the acetylation state of the histones: extensive DNA methylation triggered by seeds of methylation was required before histone deacetylation occurred.

To determine whether there was a difference in the H3 histone methylation state of the various GSTP1 constructs, we performed ChIP analysis using a polyclonal antibody generated to dimethylated H3-K9 on LNCaP cells and LNCaP cells transfected with different GSTP1 constructs. Fig. 4,B summarizes the extent of H3-K9 methylation associated with the various GSTP1 constructs relative to the endogenous GSTP1 promoter from LNCaP cells. In the LNCaP cells, the endogenous GSTP1 promoter histones are methylated, i.e., the promoter region binds methylated histone H3-K9 antibodies, and this correlates with the promoter being extensively methylated and the gene silent. Conversely, the shuttle and HpaII-seeded shuttle, which are unmethylated and actively transcribed, are not associated with H3-K9 histone methylation. The Sp1 deletion, which is unmethylated but is transcriptionally silent, is also not associated with methylated H3-K9 histones. Interestingly, the Sp1 deletion HpaII-seeded construct was also not significantly associated with methylated H3-K9 histones, although the promoter region is hypermethylated. Increasing the number of doublings (25d to 36d) increased the level of DNA methylation (Fig. 3,F) but did not increase the level of histone methylation (Fig. 4 B). Therefore, we conclude that DNA methylation of the GSTP1 promoter precedes histone methylation.

To determine whether hypermethylation and deacetylation of the GSTP1 was associated with binding of MeCP2 proteins, we performed ChIP analysis using a polyclonal antibody generated to MeCP2 on LNCaP cells and LNCaP cells transfected with different GSTP1 constructs. Fig. 4,C summarizes the extent of MeCP2 binding associated with the GSTP1 constructs relative to the endogenous GSTP1 promoter from LNCaP cells. MeCP2 was found to bind to the endogenous GSTP1 promoter from LNCaP cells, which is extensively methylated and not expressed. In contrast, there was no significant level of MeCP2 binding to the actively transcribing unmethylated shuttle and HpaII-seeded shuttle, nor to the unmethylated but silent SpI deletion construct. In addition, there was little MeCP2 binding across the GSTP1 promoter from the HpaII-seeded Sp1 deletion construct. However, there was a slight increase in MeCP2 binding after additional doublings (25d to 36d; Fig. 4,C), and this was associated with an increase in hypermethylation density across the GSTP1 promoter region (Fig. 2, E and F). However, the level of MeCP2 binding relative to binding on the endogenous promoter is minimal and therefore does not appear to be associated with triggering hypermethylation of the GSTP1 promoter.

To determine whether MDB2 is associated with initiating hypermethylation, we performed ChIP analysis using a polyclonal antibody generated to MBD2 on LNCaP cells and LNCaP cells transfected with different GSTP1 constructs. Fig. 4 D summarizes the extent of MBD2 binding associated with the various GSTP1 constructs relative to the endogenous GSTP1 promoter from LNCaP cells. Similar to the MeCP2 results, MBD2 was found to bind to the endogenous GSTPI promoter from LNCaP cells that is extensively methylated and not expressed. Also similar to the MeCP2 ChIP results, there was no significant binding of MBD2 in the actively transcribing unmethylated shuttle or the HpaII-seeded shuttle nor in the unmethylated but silent SpI deletion construct. However, in contrast to MeCP2 lack of binding, MBD2 protein was found to be associated with the silent HpaII-seeded vector at a similar level to the endogenous GSTP1 gene. Therefore, although MBD2 and MeCP2 are both found to bind to the endogenous GSTP1 gene, MBD2 binding appears to bind initially to this region and is associated with the spread of de novo methylation of the GSTP1 CpG island.

The mechanism responsible for initiating hypermethylation of CpG islands promoter regions in cancer cells is poorly understood. CpG islands in normal cells are hypomethylated, and we previously hypothesized that when the gene is actively transcribed, hypomethylation is maintained by a constant flux between a low level of de novo methylation and demethylation (3). However, if the gene is inactivated in the cell, either because of a lack of transcription factors or transiently, de novo methylation is favored and appears to spread from the seeds of methylation. Gene inactivity is also associated with changes in chromatin modification. In this study, we wished to investigate the temporal relationship between hypermethylation and chromatin modification. This is difficult to assess in genes that are already hypermethylated because the chromatin modification has already occurred in the cancer cell. Therefore, we decided to explore the change in histone modification of an introduced GSTP1 gene in LNCaP prostate cancer cells to establish what effect transcriptional gene silencing and seeds of methylation have on the chromatin structure during the process of hypermethylation of the GSTP1 CpG island.

The CpG island spanning the promoter of the endogenous GSTP1 gene is extensively methylated, and the gene is inactive in the majority of prostate cancer cells, including the prostate cancer cell line LNCaP (19). Using ChIP analysis, we found that the H3-K9 histones associated with this hypermethylated region were deacetylated and methylated. In addition, there was significant binding of the methyl binding proteins MeCP2 and MBD2. Previous studies on the endogenous GSTP1 in breast cancer cells (20) and in hepatocellular carcinoma cells (21) have also showed substantial binding of MBD2 to the methylated promoter, but in contrast to our findings, they did not observe MeCP2 binding. This difference in our results could simply reflect the different antibodies to MeCP2 used in our studies or could reflect a difference in protein levels in prostate cancer cells. However, the chromatin modification and binding of methyl binding proteins all correlate with the hypermethylated state of the gene and inactive transcription. In contrast to the endogenous gene, we found that the H3-K9 histones associated with the transfected GSTP1 gene were acetylated and unmethylated, and there was no binding of the MeCP2 or MBD2 correlating with the transfected GSTP1 CpG island remaining unmethylated and the gene active.

To determine whether inactivation of the GSTP1 gene alone could trigger a change in chromatin structure, we examined the modification of histones spanning the GSTP1 CpG island of a construct where the Sp1 sites had been deleted and thus resulted in gene silencing. Interestingly, in this construct, although the promoter was inactivated, the associated H3 histones remained acetylated and unmethylated at the lys9 residue. Therefore, silencing alone is insufficient to alter the chromatin modification of the GSTP1 CpG island. Deacetylation of histone H3-K9 only occurred when the CpG island was seeded with methylation in combination with gene inactivation. Therefore, a low level of methylation across the CpG island is necessary to induce deacetylation of the histones. Seeding methylation only occurs when the gene is inactive because when the gene is active demethylation of the island is constantly promoted. Our results indicate that a low level of CpG methylation can induce histone deacetylation but does not affect histone H3-K9 methylation. Interestingly, the endogenous GSTP1 gene from LNCaP cells, which was extensively methylated across the CpG island, was associated with methylated H3-K9 histones, whereas the transfected DNA that was less methylated (average 50% in comparison to 99%) was still associated with H3 histones unmethylated at the lysine 9 residue. Therefore, the degree of DNA methylation across the CpG island appears to dictate the methylation state of the associated histones.

To determine whether the seeds of methylation stimulated the methylated DNA binding proteins to associate with the GSTP1 CpG island, we assayed for binding of MeCP2 and MBD2 to the transfected constructs. Interestingly, we found that seeding methylation initiated MBD2 binding in preference to MeCP2 binding. We propose that the seeds of methylation initiate MBD2 binding and that MBD2, as part of the MeCP1 complex, recruits HDACs (11, 22), resulting in subsequent histone deacetylation of the GSTP1 CpG island. MBD2 has also been shown to bind DNMT1 (23), and DNMT1 and DNMT3a bind a variety of HDACs (24, 25, 26); therefore, a combination of MBD2 binding and HDAC and DNA methyltransferase recruitment would contribute to the additional spread and maintenance of DNA methylation across the CpG island. We found that MeCP2 binding only occurred with multiple passages after transfection and was associated with an increase in the average DNA methylation level of the GSTP1 CpG island, whereas MBD2 binding was independent of the degree of methylation. Interestingly, MeCP2 binding has been shown to facilitate H3-K9 methylation (27), suggesting that it is the later binding of MeCP2 that initiates subsequent histone H3-K9 methylation as seen in the GSTP1 endogenous gene.

In our system, we have initiated gene silencing by genetic deletion of Sp1 binding sites and studied the subsequent effect on DNA hypermethylation and histone remodeling. In contrast, other studies have examined the relationship between aberrant DNA hypermethylation and key histone code components by reversing DNA methylation using 5-aza-deoxycytidine treatment and histone deacetylation using HDACs inhibitors to follow the change in histone modification after gene activation and concluded that DNA methylation acts as the dominant switch for directing the modification of the histone proteins and chromatin state (28, 29). Our results support this observation in that DNA methylation precedes any change in the histone state of the GSTP1 gene. In contrast, Bachman et al. (30) have reported that histone modification can occur independently of DNA methylation. In this study, the authors showed the wild-type p16^INKAa gene could be activated with demethylation in HCT116 cells that were lacking DNMT1 and 3B expression. With continued cell passaging, the p16^INKAa gene became inactivated, and the associated histones were methylated, but the DNA remained unmethylated and DNA methylation only occurred subsequently after additional passage. However, a low level of seeding DNA methylation may not have been detected because methylation-specific PCR and pooled genomic sequencing were used to study methylation levels. Histone methylation has also been shown to direct DNA methylation and gene silencing in Neurospora and Arabidopsis (31, 32): a reverse scenario to our system where we have shown that gene silencing can equally direct DNA methylation and subsequent histone methylation. It is possible that two alternate mechanisms can occur to remodel the chromatin and the mechanism used is governed by the transcriptional state of the gene.

The combination of our results allows us to suggest a model to explain the sequential process leading to DNA hypermethylation of the GSTP1 CpG island in prostate cancer cells (Fig. 5). In our model, if GSTP1 is inactivated in a prostate cell as the initial event, then the low level of random de novo methylation is no longer counterbalanced by demethylation, resulting in an accumulation of methylation that spreads slowly across the island from these methylation seeds. The seeds of methylation promote binding of MBD2, which in turn promotes binding of histone deacetylases and DNA methyltransferases, resulting in histone deacetylation and additional de novo methylation. MeCP2 binds after extensive DNA methylation of the GSTP1 CpG island and recruits histone methyltransferase resulting in H3-K9 methylation consolidation of an inactive chromatin structure. In this model, gene inactivation of tumor suppressor genes occurs before chromatin remodeling. We propose that gene silencing in cancer cells is stochastic and can occur randomly at a low frequency in any normal cell, and it is the silencing of the gene that promotes DNA hypermethylation and subsequent chromatin remodeling over successive cell generations.

Previously, we proposed that the genes susceptible to methylation in different cancer subtypes reflect the set of genes that are more likely to undergo random inactivation in the normal cell equivalent or those genes that once silent give the cell a selective growth advantage (3, 4, 33). We now suggest that if the genes are inadvertently silenced, their continued inactivation is ensured by a combination of DNA hypermethylation and chromatin remodeling. We find that deacetylation of the GSTP1 CpG island is independent of transcriptional gene silencing but is dependent on a low level of seeding methylation that accumulates with lack of gene transcription. Deacetylation of the promoter is associated with binding of MBD2 and the localized spread of DNA methylation from seeds within the CpG island. Histone methylation occurs subsequently and is associated with MeCP2 binding and extensive DNA methylation. These findings have important implications in understanding the temporal relationship between transcriptional gene silencing and the triggers associated with DNA hypermethylation and chromatin remodeling in cancer cells.

We thank Dr. Renee Poropat from Sydney University Prince Alfred Macromolecular Analysis Centre for help in discussion and design of the real-time probes used in this analysis and Rebecca Hinshelwood for help in figure preparation.