Methylation-mediated Silencing of TMS1/ASC Is Accompanied by Histone Hypoacetylation and CpG Island-localized Changes in Chromatin Architecture*

Aberrant methylation of CpG-dense islands in the promoter regions of genes is an acquired epigenetic alteration associated with the silencing of tumor suppressor genes in human cancers. In a screen for endogenous targets of methylation-mediated gene silencing, we identified a novel CpG island-associated gene, TMS1, which is aberrantly methylated and silenced in response to the ectopic expression of DNA methyltransferase-1. TMS1 functions in the regulation of apoptosis and is frequently methylated and silenced in human breast cancers. In this study, we characterized the methylation pattern and chromatin architecture of the TMS1 locus in normal fibroblasts and determined the changes associated with its progressive methylation. In normal fibroblasts expressing TMS1, the CpG island is defined by an unmethylated domain that is separated from densely methylated flanking DNA by distinct 5′ and 3′ boundaries. Analysis of the nucleoprotein architecture of the locus in intact nuclei revealed three DNase I-hypersensitive sites that map within the CpG island. Strikingly, two of these sites coincided with the 5′- and 3′-methylation boundaries. Methylation of the TMS1 CpG island was accompanied by loss of hypersensitive site formation, hypoacetylation of histones H3 and H4, and gene silencing. This altered chromatin structure was confined to the CpG island and occurred without significant changes in methylation, histone acetylation, or hypersensitive site formation at a fourth DNase I-hypersensitive site 2 kb downstream of the TMS1 CpG island. The data indicate that there are sites of protein binding and/or structural transitions that define the boundaries of the unmethylated CpG island in normal cells and that aberrant methylation overcomes these boundaries to direct a local change in chromatin structure, resulting in gene silencing.

Aberrant methylation of CpG-dense islands in the promoter regions of genes is an acquired epigenetic alteration associated with the silencing of tumor suppressor genes in human cancers. In a screen for endogenous targets of methylation-mediated gene silencing, we identified a novel CpG island-associated gene, TMS1, which is aberrantly methylated and silenced in response to the ectopic expression of DNA methyltransferase-1. TMS1 functions in the regulation of apoptosis and is frequently methylated and silenced in human breast cancers. In this study, we characterized the methylation pattern and chromatin architecture of the TMS1 locus in normal fibroblasts and determined the changes associated with its progressive methylation. In normal fibroblasts expressing TMS1, the CpG island is defined by an unmethylated domain that is separated from densely methylated flanking DNA by distinct 5 and 3 boundaries. Analysis of the nucleoprotein architecture of the locus in intact nuclei revealed three DNase I-hypersensitive sites that map within the CpG island. Strikingly, two of these sites coincided with the 5-and 3methylation boundaries. Methylation of the TMS1 CpG island was accompanied by loss of hypersensitive site formation, hypoacetylation of histones H3 and H4, and gene silencing. This altered chromatin structure was confined to the CpG island and occurred without significant changes in methylation, histone acetylation, or hypersensitive site formation at a fourth DNase I-hypersensitive site 2 kb downstream of the TMS1 CpG island. The data indicate that there are sites of protein binding and/or structural transitions that define the boundaries of the unmethylated CpG island in normal cells and that aberrant methylation overcomes these boundaries to direct a local change in chromatin structure, resulting in gene silencing.
Cytosine methylation is a post-replicative modification of DNA that plays an important role in epigenetic inheritance. In vertebrates, methylation occurs primarily at cytosines within the dinucleotide CpG. Both the distribution of CpG sites and their methylation status are non-random in the human genome, generating a pattern of DNA methylation that varies with respect to density across the chromosome. CpG sites occur relatively infrequently throughout much of the human genome, except in discreet regions of CpG-dense DNA known as CpG "islands." These islands are ϳ200 -1000 bp in length and often coincide with the 5Ј-ends of genes. Accordingly, the identification of such regions in the human genome sequencing projects has been useful in gene prediction (1). Current estimates suggest that there are ϳ29,000 CpG islands in the human genome (2). Whereas non-CpG island DNA is generally methylated, most CpG islands remain methylation-free in both expressing and non-expressing tissues (3,4). Specialized physiologic cases in which CpG islands are methylated include genes subject to parental imprinting and genes on the inactive X chromosome (5,6). Here methylation functions as part of an epigenetic mechanism ensuring the stable propagation of allele-specific gene repression.
In human cancers, CpG island methylation occurs aberrantly and is associated with the inappropriate silencing of tumor suppressor genes and other genes that function in the suppression of the malignant phenotype. Methylation-mediated silencing contributes to malignant progression at several levels and has been implicated in the inactivation of genes involved in tumor suppression (e.g. Rb, VHL, and CDKN2A), the regulation of DNA repair and genome integrity (e.g. MLH1, MGMT, BRCA1), and the suppression of metastases (e.g. CDH1 and TIMP3) (reviewed in Refs. [7][8][9]. Recent evidence suggests that aberrant methylation also contributes to human carcinogenesis by conferring resistance to cell death signals through the silencing of genes that promote apoptosis (10 -13). In each of these cases, aberrant methylation of promoter region CpG islands is associated with loss of gene expression in tumor-derived cell lines and primary tumors relative to normal somatic cells. That methylation plays a primary role in the silencing of these genes is supported by the fact that treatment with a demethylating agent such as 5-aza-2Ј-deoxycytidine is sufficient to restore expression and, in many cases, gene function.
At present, it is not known how or why particular CpG islands succumb to aberrant methylation in cancer cells, and the precise mechanism in which an unmethylated, actively transcribed gene progresses to a methylated and inactive state is unclear. One established consequence of CpG island methylation is transcriptional repression. Methylation of cytosine can affect gene expression directly, by interfering with the binding of transcription factors like E2F and AP-2 (14,15), or indirectly, through the recruitment of proteins that bind preferentially to methyl CpG through a methyl CpG binding domain (MBD) 1 (reviewed in Refs. 16,17). The MBD proteins are found within complexes containing histone deacetylases (HDAC) and chromatin remodeling factors (18 -21). The accidental methylation of a CpG island during carcinogenesis could serve to re-direct the MBDs and chromatin remodeling machinery to newly methylated DNA, resulting in aberrant gene silencing. The recent observation that the DNA methyltransferases (DNMTs), DNMT1, DNMT3a, and DNMT3b, possess intrinsic transcriptional repressor activity and interact directly with HDACs (22)(23)(24)(25)(26) also raises the intriguing possibility that methylation-dependent gene silencing may be directed by the DNA methyltransferases themselves through the recruitment HDACs to their methylation targets.
To study the factors involved in aberrant methylation targeting and its consequences in somatic cells, we have utilized a human fibroblast model system in which de novo methylation is induced by the ectopic expression of DNA methyltransferase-1 (DNMT1). In previous studies, we showed that the overexpression of DNMT1 results in the progressive de novo methylation of endogenous CpG island sequences (27,28). Using this model, we recently identified a novel CpG islandassociated gene, TMS1 (Target of Methylation-induced Silencing), that is aberrantly methylated and silenced in response to DNMT1-driven methylation (13). Further studies indicated that TMS1 is silenced in association with aberrant methylation in human breast cancer cell lines and primary tumors, implicating TMS1 as a putative tumor suppressor in breast cancer (13). TMS1 encodes a caspase recruitment domain protein and functions as a positive mediator of apoptosis (29). TMS1 was independently identified (and named ASC) by Masumoto et al. (30) who showed that decreased expression of TMS1 results in reduced sensitivity to anticancer agents. Methylation-mediated silencing of TMS1 thus may contribute to breast carcinogenesis by allowing cells to bypass apoptosis and may confer resistance to cancer chemotherapeutic agents or other genotoxic stress.
In this study, we used bisulfite genomic sequencing, DNase I-hypersensitive site mapping and chromatin immunoprecipitation to characterize the methylation pattern and the chromatin architecture of the TMS1 locus in normal human fibroblasts, and to determine the changes associated with its progressive methylation in immortalized and DNMT1-overexpressing derivatives. We find that in normal fibroblasts, the TMS1 CpG island is comprised of an unmethylated domain with distinct 5Ј-and 3Ј-methylation boundaries. Three DNase I-hypersensitive (HS) sites mapped within the CpG island, with two coinciding almost precisely with the positions of the methylation boundaries. De novo methylation of the CpG island in DNMT1-overexpressing cells was accompanied by a loss of CpG island-specific HS formation, localized hypoacetylation of histones H3 and H4, and gene silencing. We propose that there are protein binding sites that demarcate the boundaries of TMS1 CpG island in normal cells and that aberrant methylation overcomes these boundaries to direct a localized change in chromatin architecture, resulting in gene silencing.

EXPERIMENTAL PROCEDURES
Cell Culture and Drug Treatments-Normal diploid human fibroblasts (IMR90) and SV40 immortalized IMR90 cells (referred to as "90SV") were obtained from the NIA (National Institutes on Aging) Cell Repository (AG02804C) and were maintained in EMEM with 2 mM glutamine and 10% fetal calf serum. The generation of a 90SV derivative cell line stably overexpressing the human DNMT1 (HMT.1E1) has been reported (27). HMT.1E1 cells were maintained in EMEM plus 2 mM glutamine, 10% fetal calf serum, and 400 g/ml G418. HMT.1E1 cells were seeded in 75-cm 2 flasks overnight and treated the next day with 100 ng/ml trichostatin A (TSA) (Sigma) or 500 nM 5-aza-2Ј-deoxycytidine. For the combined treatments, cells were treated with 500 nM 5-aza-2Ј-deoxycytidine for 72 h and 100 ng/ml TSA was added for the last 24 h.
Bisulfite Sequencing and Methylation-specific PCR-Genomic DNA was treated with sodium bisulfite as previously described (31). For methylation-specific PCR, ϳ50 ng of bisulfite-modified DNA was amplified by PCR under the following reaction conditions: 67 mM Tris-HCl (pH 8.8), 16.6 mM NH 4 SO 4 , 6.7 M EDTA, 10 mM ␤-mercaptoethanol, 6.7 mM MgCl 2 , and 1 M of each primer in a 25-l reaction. A hot start was performed (5 min, 95°C) followed by the addition of 0.5 unit of Taq polymerase (Invitrogen) and 35 cycles of PCR (95°C, 30 s; 58°C, 30 s; 72°C, 30 s) with a final extension of 4 min at 72°C. Primers were designed from the interpolated sequence after bisulfite conversion assuming DNA was either methylated or unmethylated at three CpG sites. Primers used for methylated reactions were 5Ј-TTG TAG CGG GGT GAG CGG C-3Ј and 5Ј-AAC GTC CAT AAA CAA CAA CGC G-3Ј. Primers used for the unmethylated reactions were 5Ј-GGT TGT AGT GGG GTG AGT GGT-3Ј and 5Ј-CAA AAC ATC CAT AAA CAA CAC A-3Ј. Reaction products were separated by electrophoresis on a 6% polyacrylamide/Tris borate-EDTA gel, stained with ethidium bromide, and photographed.
For genomic sequencing, 50 ng of bisulfite-modified DNA was amplified by PCR using the same reaction conditions described above except that the annealing temperature was 55°C and the final extension was increased to 10 min. Primers were designed to avoid potential methylation sites (e.g. CpG) such that both methylated and unmethylated DNA would be amplified equally. The resulting amplification pools were cloned into the pCR2 vector using the TOPO TA cloning kit (Invitrogen). Eight to twelve individual subclones per PCR reaction were isolated and sequenced. Primer pairs used for bisulfite genomic sequencing were: DNase I Hypersensitivity-Cells (1 ϫ 10 7 ) were resuspended in nuclei isolation buffer (RBS) (10 mM Tris, pH 7.4, 5 mM MgCl 2 , 0.5 mM dithiothreitol, 0.3 mM sucrose, 0.4 mM phenylmethylsulfonyl fluoride, and 0.1% Nonidet P-40) and swollen on ice for 10 min. This was followed by Dounce homogenization, and nuclei were collected by centrifugation at 1500 ϫ g. Nuclei were resuspended in RBS without Nonidet P-40 and immediately added to prepared aliquots of DNase I (Invitrogen) and 0.1 mM CaCl 2 for 10 min at room temperature. As controls, nuclei were incubated in the presence of CaCl 2 without DNase I (endogenous nuclease control) or in the absence of DNase I (nuclease-free reference). Reactions were stopped by the addition of an equal volume of 20 mM EDTA, 1% SDS, 10 g/ml RNase A and incubated at 37°C for 30 min. DNA was then isolated by Proteinase K digestion, phenol:chloroform extraction, and ethanol precipitation. Ten micrograms of DNA was digested with HindIII and electrophoresed on a 25-cm, 1.0% agarose gel. The gel was transferred to a nylon membrane and hybridized to a random-prime-labeled 423-bp HindIII-XbaI fragment anchored to the 3Ј HindIII site of the TMS1 locus. Blots were washed to a final stringency of 0.1ϫ SSC, 0.1% SDS at 65°C and exposed to x-ray film (Biomax-MS, Kodak).
Chromatin Immunoprecipitation-Chromatin immunoprecipitation was performed using antibodies specific for the acetylated isoforms of histone H4 and H3 in accordance with the protocol provided by the supplier (Upstate Biotechnology Inc., Lake Placid, NY). DNA fragments immunoprecipitated with anti-acetylated H3 or anti-acetylated H4 were analyzed by multiplex PCR using primers spanning HS1, HS2, HS3, or HS4 of the TMS1 locus in combination with primers specific for the ␤-actin promoter region. Multiplex PCR was carried out under the following reaction conditions: 67 mM Tris-HCl (pH 8. Reverse Transcriptase-PCR-Total RNA was isolated from log-phase cells by the Chomczynski and Sacchi method (32). Six micrograms of total RNA was pretreated with amplification grade DNase I (Invitrogen) and reverse-transcribed using random hexamer primers and Moloney murine leukemia virus-reverse transcriptase (Invitrogen). Onethirtieth of the reverse transcriptase reaction (200 ng of starting RNA) was used in a PCR reaction. The PCR conditions were: 67 mM Tris-HCl (pH 8.8), 16.6 mM NH 4 SO 4 , 6.7 M EDTA, 10 mM ␤-mercaptoethanol, 4.7 mM MgCl 2 , 10% Me 2 SO, and 400 nM of each primer in a 25-l reaction. A hot start was performed (5 min, 95°C), followed by 35 cycles of 95°C, 30 s; 55°C, 30 s; 72°C, 30 s. TMS1 primers were 5Ј-TGG GCC TGC AGG AGA TG-3Ј and 5Ј-ATT TGG TGG GAT TGC CAG-3. ␤-Actin primers were 5Ј-CCT TCC TGG GCA TGG AGT CCT G-3Ј and 5Ј-GGA GCA ATG ATC TTG ATC TTC-3Ј. For multicycle experiments, samples were withdrawn from a 100-l PCR reaction after 20, 30, and 40 cycles for TMS1 or after 15, 25, 35 cycles for ␤-actin. Samples were run on a 2% agarose gel, transferred to a nylon membrane, and hybridized to random prime-labeled TMS1 or ␤-actin cDNA probe. Blots were washed to a final stringency of 0.1ϫ SSC, 0.1% SDS at 65°C and exposed to x-ray film (Biomax-MS, Kodak).

RESULTS
We identified TMS1 in a PCR-based subtractive cDNA screen to isolate transcripts that were down-regulated in human fibroblasts stably overexpressing DNMT1 (13). We have shown previously that TMS1 is expressed in normal human diploid fibroblasts (IMR90) and their SV40-immortalized counterparts (90SV cells) but is silent in the 90SV derivative cell line expressing 50-fold increased levels of DNMT1 (HMT.1E1 cells) (13). These cell lines represent a useful model for studying the genesis and consequences of aberrant CpG island methylation in that they are derived from a common parent (IMR90) yet differ in their expression and methylation status at the TMS1 locus.
To examine the role of DNA methylation in TMS1 silencing, we first characterized the TMS1 locus with regards to structure and methylation status in normal diploid fibroblasts (IMR90). The coding sequence of TMS1 covers ϳ1.5 kb on chromosome 16p11.2. The promoter region of TMS1 lacks a defined TATA box but contains a 600-bp CpG island predicted on the basis of CϩG content (69%) and CpG frequency (CpG observed/expected ϭ 0.82), which extends from ϳ200 bp upstream to 400 bp downstream of a single transcription start site (position 1177, Fig. 1A and data not shown). Genomic sequencing of bisulfite-modified DNA was used to construct a map of the methylation status of 110 CpGs across 2.5 kb of the TMS1 locus. Five overlapping regions covering the TMS1 locus were amplified from bisulfite-treated DNA. PCR products were then subcloned, and individual clones were sequenced. This allows for the assessment of the methylation status of independent alleles derived from the cell population. Fig. 1B shows the percentage of alleles methylated at each CpG site as determined from the analysis of eight to sixteen individual alleles for each primer set. The TMS1 locus was characterized by an unmethylated domain that extended from ϳ100 bp upstream to ϳ1 kb downstream of the transcription start site and encompassed the entire predicted CpG island plus ϳ500 bp of additional downstream sequence (Fig. 1B). CpGs within this region were methylated in no more than one allele per eight to sixteen analyzed with the exception of one CpG site in which three of eight alleles analyzed were methylated. In contrast, sequences upstream of nucleotide 1100 and downstream of 2100 were densely methylated, with distinct 5Ј and 3Ј boundaries separating the unmethylated CpG island region from heavily methylated flanking DNA (Fig. 1B). Primary mammary epithelial cells are also completely devoid of methylation in the region from 1100 to 2000 (data not shown). Thus, the pattern shown here appears representative of normal tissues.
To determine the relationship between methylation of TMS1 and gene silencing, we next analyzed the methylation pattern of 500 bp surrounding the transcription start site (primer set B, see Fig. 1) in the IMR90 fibroblasts, their SV40 immortalized derivatives, the 90SV cells, and the DNMT1 overexpressing derivative HMT.1E1 cells. As discussed above, TMS1 was almost completely unmethylated on individual alleles down-

FIG. 1. Pattern of methylation at the TMS1 locus in normal fibroblasts.
A, genomic map of the TMS1 locus. The position of 110 CpG dinucleotides (vertical lines) analyzed in this study are indicated above the map. Open boxes, TMS1 exons; numbers indicate nucleotide position relative to an upstream HindIII site (H). Connected arrows represent the locations of the five sets of overlapping primers (A-E) used in bisulfite sequencing. B, methylation pattern of the TMS1 locus in normal fibroblasts. DNA from IMR90 cells was modified with sodium bisulfite, which deaminates cytosines to uracil but leaves methyl cytosines unaltered. Bisulfite-modified DNA was amplified by PCR using primer pairs A-E. The resulting PCR products were subcloned, and at least eight individual subclones were sequenced for each amplification reaction. Percent methylation was determined from the number of alleles containing a methylated CpG at each position relative to the total number of alleles analyzed. stream of nucleotide 1100 in IMR90 cells ( Fig. 2A). Of the few sites that showed methylation, they appeared to be random in that no single CpG site was methylated in more than one allele analyzed ( Fig. 2A). Relative to IMR90, 90SV cells showed a partially methylated pattern. There did not appear to be any methylation hot spots in these cells, although there was somewhat more methylation in general in the 5Ј-end than the 3Ј-end of the region analyzed. Rather, the analysis of individual subclones was suggestive of two populations of alleles: those that were predominantly methylated and those that were predominantly unmethylated ( Fig. 2A). This pattern is consistent with a mixed cell population in which some cells are methylated for TMS1 and others are not, or with allelic heterogeneity within each cell. The partially methylated pattern of the 90SV cells correlated with a decreased expression of TMS1 relative to IMR90 cells (Fig. 2B). In stark contrast, HMT.1E1 cells showed no detectable TMS1 expression and were methylated at nearly every CpG site on all alleles analyzed (Fig. 2, A and B).
Next we used DNase I to probe the nucleoprotein architecture of the TMS1 locus in intact nuclei. Four DNase I-hypersensitive sites, designated HS1-HS4, were identified in the IMR90 fibroblasts (Fig. 3). Interestingly, mapping of the DNase I-hypersensitive sites relative to the DNA methylation map from the same cells (Fig. 1) indicated that HS1, HS2, and HS3 lay within the unmethylated domain, with HS1 and HS3 corresponding almost precisely to the location of the 5Ј-and 3Јmethylation boundaries (compare Fig. 3A with Fig. 1B). The fourth site (HS4) mapped to about 1 kb downstream of the end of the TMS1 coding region (Fig. 3A). Similar experiments using a probe anchored at the 5Ј HindIII site confirmed the position of the DNase I-hypersensitive sites (data not shown). To determine the relationship between CpG island methylation, gene expression, and chromatin structure, DNase I hypersensitivity assays were also performed on the 90SV and HMT.1E1 cells. All four DNase I-hypersensitive sites found in the IMR90 cells (HS1-HS4) were also observed in the 90SV cells, which express TMS1 and contain a mixed population of unmethylated and methylated alleles (Fig. 3B). In contrast, complete methylation of the TMS1 CpG island in the HMT.1E1 cells resulted in loss of the CpG island-associated hypersensitive sites HS1, HS2, and HS3 but had no effect on the formation of HS4 (Fig. 3B). HS4 is thus "constitutive" in that its formation is independent of TMS1 expression and methylation status at the CpG island, whereas the CpG island-associated hypersensitive sites (HS1, HS2, and HS3) form only when the CpG island is at least partially unmethylated and TMS1 is expressed. Analysis of the methylation status of the 500 bp surrounding HS4 (primer set E, see Fig. 1) in the 90SV and HMT.1E1 cells indicated a similar pattern to that observed in IMR90 cells (data not shown). These data show that, in normal cells, the chromatin architecture adopted by the unmethylated CpG island is bound by DNase I HS and that methylation-mediated silencing of TMS1 is accompanied by localized changes in the conformation of this domain.
Recent evidence suggests that methylation-associated gene repression involves the recruitment of histone deacetylases (HDACs) and other chromatin-modifying factors (22)(23)(24)(25)(26). To determine whether methylation-mediated silencing of TMS1 is accompanied by changes in histone acetylation, we analyzed the relative levels of acetylated histone H3 and H4 at the regions corresponding to HS1, HS2, HS3, and HS4 in IMR90, 90SV, and HMT.1E1 cells. Formaldehyde-cross-linked chromatin was immunoprecipitated with antibodies against the acetylated isoforms of histone H3 or histone H4. Immunoprecipitated DNA was analyzed by multiplex PCR using a primer pair specific to each of the hypersensitive site regions (HS1-HS4) in conjunction with primers to the ␤-actin promoter under conditions of linear amplification of input DNA. ␤-Actin was used as a positive control, because it is expressed at similar levels in all three cell lines (Fig. 2B). The ratio of intensities of the TMS1 product and ␤-actin product was determined and normalized to that of the IMR90 cells to allow for comparison between the three cell lines.
Relative to the IMR90 cells, both 90SV and HMT.1E1 cells showed decreased levels of acetylated histones H3 and H4 at all four regions of the TMS1 locus analyzed (Fig. 4). At the CpG island-localized DNase I-hypersensitive sites (HS1, HS2, and HS3), where the density of CpG is highest, the relative degree of histone hypoacetylation in the three cell lines correlated well with the degree of methylation. As the level of methylation in the CpG island increased from Ͻ10% in the IMR90 cells to ϳ50% in the 90SV to nearly 100% in HMT.1E1, the relative levels of histone H4 acetylation decreased by ϳ50 -60% in the 90SV cells and Ͼ80% in the HMT.1E1 cells. The relative levels FIG. 2. TMS1 silencing is associated with complete methylation of the CpG island. A, bisulfite sequencing analysis of the methylation status of 53 CpG sites in the TMS1 CpG island. DNA from IMR90, 90SV, and HMT.1E1 cells was modified with sodium bisulfite and amplified by PCR using primer set B indicated in Fig. 1. The amplification product was subcloned and sequenced. Each row represents the methylation pattern of an individual allele; E, unmethylated CpG; q, methylated CpG. CpG positions are indicated and are relative to the 5Ј HindIII site as in Fig. 1. B, TMS1 expression. Expression of TMS1 was determined by semi-quantitative reverse transcription-PCR analysis. RNA from IMR90, 90SV, and HMT.1E1 cells was reverse-transcribed and amplified with primers specific to TMS1 (top) or ␤-actin (bottom) for the indicated number of cycles. TMS1 silencing in HMT.1E1 cells is correlated with the near-complete methylation of all 53 CpG sites in the CpG island. of acetylated histone H3 at HS1, HS2, and HS3 were also decreased in the 90SV and HMT.1E1 cells, although to a somewhat lesser extent than acetylated histone H4 (Fig. 4B). In contrast, there was little correlation between methylation and histone acetylation outside of the CpG island, in the region of HS4. Although the levels of acetylated histones associated with HS4 in the 90SV and HMT.1E1 cells was decreased relative to IMR90 cells, there was little difference between the TMS1expressing 90SV cells and the TMS1 silent HMT.1E1 cells (Fig.  4). Together with the finding that the formation of HS4 is independent of the CpG island methylation and gene expression, these data suggest that the effect of aberrant methylation at the TMS1 CpG island on chromatin structure is confined to the immediate vicinity of the CpG island.
To further examine the relative contributions of methylation and histone deacetylation in TMS1 silencing, we determined the effect of treatment with the demethylating agent 5Ј-aza-2Јdeoxycytidine (5-azadC) or the histone deacetylase inhibitor TSA on the expression and methylation of TMS1 in the HMT.1E1 cells. Treatment of HMT.1E1 cells with 5-azadC resulted in a time-dependent re-activation of TMS1 (Fig. 5A). This re-activation was accompanied by partial demethylation of the TMS1 CpG island (Fig. 5B). In contrast, treatment of HMT.1E1 cells with TSA had no effect on TMS1 expression or methylation of the CpG island and did not significantly alter the degree of re-expression when combined with 5-azadC. The data indicate that the inhibition of histone deacetylation alone is insufficient to re-activate TMS1. As has been found for other CpG island-associated genes silenced in human cancers (33,34), these data suggest that methylation plays a primary role in TMS1 silencing. DISCUSSION TMS1/ASC is an apoptotic signaling protein whose expression is frequently lost in human breast cancers via aberrant methylation. TMS1 is associated with a 5Ј-CpG island that is unmethylated in normal fibroblasts and mammary epithelial cells but becomes aberrantly methylated in response to over-expression of DNMT1 and in human breast cancer cell lines and primary tumors. In this study, we showed that the CpG island of the TMS1 gene is unmethylated in normal somatic cells and is separated from densely methylated flanking DNA by distinct boundaries at both the 5Ј-and 3Ј-ends. The methylation boundaries coincided with DNase I HS sites that form only when the CpG island is unmethylated. The de novo methylation of the TMS1 CpG island was accompanied by local changes in CpG island conformation and activity, including the hypoacetylation of histones H3 and H4, remodeling of CpG island-specific DNase I HS sites, and gene silencing. An understanding of the consequences of aberrant methylation of TMS1 on chromatin structure and gene expression is important not only because of its role in breast cancer but because TMS1 may be representative of other CpG island-associated tumor suppressor genes that are silenced in cancer.
Recently, 5Ј-methylation boundaries have been mapped for the CpG islands of the GSTP1 and the BRCA1 genes (35,36). This demarcation was absent in cancer cells in which the CpG island is aberrantly methylated and silenced. In the case of GSTP1, the methylation boundary overlapped a small AϩTrich region with a repeated sequence of ATAAA (35). Although the 5Ј-methylation boundary of TMS1 contains one copy of an ATAAA sequence, neither the 5Ј-nor 3Ј-methylation boundary is particularly AϩT-rich. We note here that one thing the TMS1 boundaries have in common with those identified at the GSTP1 and BRCA1 genes is small gaps of 50 -100 bp that lack CpGs altogether (note the gaps in CpG density in Fig. 1A). The 5Ј and 3Ј boundaries of the TMS1 CpG island contain a stretch of 60 and 134 base pairs, respectively, that, although CϩG-rich, are devoid of CpGs. These regions of non-CpG DNA abutting the CpG island DNA may act as a buffer zone ensuring that methylation does not spread into the CpG island.
It is currently unknown what, if anything, constitutes a methylation boundary in normal cells. We found that the unmethylated domain of the TMS1 CpG island is flanked by DNase I-hypersensitive sites that form at the boundaries be- FIG. 4. Histone acetylation at the TMS1 locus. The indicated cell lines were treated with formaldehyde to cross-link protein to DNA, and acetylated histones were immunoprecipitated using antibodies specific to the acetylated isoforms of histone H3 (A) or histone H4 (B). Immunoprecipitated DNA was analyzed by multiplex PCR using primers specific to the regions surrounding HS1, HS2, HS3, or HS4 of the TMS1 locus coupled with a common primer set specific to the ␤-actin promoter. Immunoprecipitation reactions lacking antibody (No Ab) or of 1/40 of chromatin input (Input) were used as negative and positive controls for each of the four primer pairs (only the reaction with HS3 primers is shown). Amplification products were separated by electrophoresis on a 2% agarose gel, stained with ethidium bromide, and photographed. A representative experiment is shown. C and D, band intensities were quantified from the digital images using Scion Imager software and the ratio of intensities of the TMS1 product to the ␤-actin product were calculated for each reaction. Ratios obtained for the IMR90 cells at each site were set to one to allow for comparison between the cell lines. The data represent the mean Ϯ S.D. of four independent immunoprecipitation experiments using anti-acetylated Histone H3 (C) or anti-acetylated Histone H4 (D). tween the unmethylated CpG island and the surrounding methylated DNA. This finding is significant in that it shows that the CpG island boundary is more than just a transition in methylation pattern; it suggests the existence of a physical boundary occurring at the level of chromatin. These sites were lost when the CpG island was methylated. One model that stems from these findings is that the maintenance of a chromatin boundary at HS1 and HS3, which could be mediated by the binding of a specific protein or complex, plays an important role in preventing the methylation of the CpG island in normal cells. Loss of this boundary function could contribute to aberrant methylation. The idea that there are cis-acting elements that protect CpG islands or other regions from methylation or that specific proteins might block de novo methylation is not without precedent. The binding of transcription factors and other DNA binding proteins blocks de novo methylation by bacterial and mammalian DNA methyltransferases in vivo (37)(38)(39). A region of the chicken ␤-globin locus defined by a constitutive DNase I HS site was shown to block the progressive de novo methylation and silencing of an integrated transgene (40). Subsequent studies have shown that this site is bound by the methylation-sensitive transcription factor CTCF (41). CTCF binding to the imprint control region of the H19/ IGF2 locus is important in the maintenance of normal imprinted expression of H19 and IGF2 (42,43). Interestingly, acquired methylation of these sites in human colon cancers and Wilms tumors leads to loss of imprinting and bi-allelic expression of IGF2 (44,45).
Previous studies have suggested that a cis element in the 5Ј-end of the mouse or hamster APRT gene containing Sp1-like sites is important in protecting the CpG island from de novo methylation (46 -49). Mutation or deletion of the Sp1 binding sites leads to de novo methylation of the CpG island when propagated in transgenic mice or transfected into embryonic cells. Whether Sp1 binding mediates the protective effect is not entirely clear, however, because neither the Sp1 consensus sites alone nor those from a heterologous locus protected the APRT gene from de novo methylation in transgenic embryos (50). Moreover, targeted disruption of Sp1 does not lead to widespread methylation of CpG islands (51). Nevertheless, it is noteworthy that all three CpG island-associated HS sites (not HS4) are in the vicinity of a canonical Sp1 binding site.
If the boundary were occupied by a methylation-sensitive DNA binding protein, the competition between de novo methylation of the boundary element and its occupancy by the DNA binding protein during DNA replication might be the first step toward the methylation of the CpG island. In the model system utilized here, that balance might be tipped in favor of aberrant methylation by overexpression of DNMT1. We have shown previously that the hypermethylation of affected CpG island loci was observed only in clones expressing Ͼ9-fold increased levels of DNMT1 (27), consistent with a competition model or the existence of a saturable control mechanism. Alternatively, loss of a trans factor that mediates the boundary function in the HMT.1E1 cells or other cell types might allow for aberrant methylation. Indeed, we find that in samples exhibiting incomplete methylation of TMS1, such as the 90SV cells studied here or breast cancer cell lines and primary tumors, 2 there are no "hot spots" of methylation, rather individual alleles are either predominantly methylated or predominantly unmethylated, suggesting that loss of a boundary function may be a limiting factor.
As with other CpG islands that become aberrantly methylated in disease states, such as the FMR1 locus in fragile X syndrome patients or the BRCA1 locus in human breast cancer cell lines (36,52), methylation of the TMS1 CpG island was associated with hypoacetylation of histones H3 and H4. We found that, within the CpG island, there was a good correlation between the levels of methylation, histone deacetylation, and expression. Given the methylation pattern in the 90SV cells, the intermediate levels of expression and histone acetylation observed most likely reflect a mixed population of cells/alleles, with some alleles being unmethylated, associated with hyperacetylated histones and expressing TMS1, and others methylated, associated with hypoacetylated histones and silent. These data are consistent with a role for HDACs in TMS1 silencing. The aberrant methylation of TMS1 may drive gene silencing through the binding of MBDs. The MBDs are found in complexes containing HDACs and chromatin remodeling factors and could serve to recruit these complexes to methylated DNA (18 -21). This model is supported by recent evidence showing an in vivo association of MBD2 with the aberrantly methylated promoter of the p16/INK4A gene in human cancer cells (34). An alternative mechanism would be one in which the aberrant methylation and deacetylation of histones occurs in a coupled process orchestrated by the DNA methyltransferases. DNMT1, DNMT3a, and DNMT3b have been shown to interact directly with HDACs and to repress transcription independently of their methyltransferase activities (22)(23)(24)(25)(26). This may occur through direct interaction with sequence-specific transcription factors (22,24). The DNMTs may thus direct HDAC activity to their methylation targets. This latter possibility may be particularly relevant in the case of TMS1, because it was identified as a gene subject to DNMT1-driven gene silencing. At this point, we cannot distinguish between MBD-dependent and -independent silencing mechanisms. However, once established, methylation appears to be the predominant mediator of TMS1 silencing, because reactivation of TMS1 required demethylation of the CpG island induced by 5-aza-2Ј-deoxycytidine and was not affected by the histone deacetylase inhibitor TSA. This is consistent with other CpG island-associated genes silenced 2 J. Levine and P. Vertino, manuscript in preparation. HMT.1E1 cells were treated for the indicated times with 500 nM 5-aza-2Ј-deoxycytidine (AZA), 100 ng/ml TSA, or a combination of both. A, analysis of TMS1 expression by reverse transcription-PCR. Following treatment, total RNA was isolated, reverse-transcribed, and amplified by PCR using primers specific to the TMS1 transcript (top) or the ␤-actin transcript (bottom). Control reactions in which the reverse transcriptase was omitted (ϪRT) were amplified under the same conditions. B, methylation of the TMS1 locus analyzed by methylation-specific PCR. DNA isolated from parallel HMT.1E1 cultures treated as described in panel A was modified with sodium bisulfite and amplified using primers in the TMS1 CpG island that overlap three potential methylation sites each. Parallel reactions were performed using primers specific to the unmethylated (U) or methylated (M) DNA. in cancer where methylation plays the dominant role in the maintenance of the repressed state (33).
The association between hypermethylation of TMS1, the deacetylation of histones, and the remodeling of DNase I-hypersensitive sites was specific to the vicinity of the CpG island. At HS4, which maps to only 2 kb downstream of the 3Ј-end of the TMS1 CpG island, the hypersensitive site formed independently of the changes in methylation and chromatin conformation occurring at the CpG island. The differences in histone H3 acetylation between the cell lines were also less dramatic at HS4, and there was no difference in methylation at this site. Therefore, silencing of TMS1 does not appear to be due to widespread changes in methylation or the structure of the locus but to the local effects of methylation on the CpG island region. Recently, Schubeler et al. (53) showed that targeted insertion of a methylated transgene resulted in specific changes in transgene conformation, including loss of a promoter DNase I HS site, hypoacetylation of histones H3 and H4, and transcriptional repression (53). These changes occurred without affecting the methylation or chromatin structure of surrounding DNA. Although the promoter used in that study was not a CpG island, both studies are consistent with the idea that methylation can direct changes in nucleosome modification and chromatin structure within a defined domain and that these changes are sufficient to drive gene silencing.
We have shown that, in normal somatic cells, the TMS1 CpG island is embedded in an unmethylated domain that adopts a chromatin structure characterized by hyperacetylated histones and whose boundaries are defined by DNase I HS sites. Aberrant methylation of TMS1 results in loss of function at these boundaries, hypoacetylation of histones, and re-organization of CpG island chromatin. Importantly, it can be concluded from these findings that the CpG island functions as an independent domain at the level of chromatin and that the remodeling of this domain represents a critical step in the silencing associated with aberrant methylation. A similar scenario may be operative at other CpG island-associated tumor suppressor genes silenced by methylation in cancer.