Oncogene (2002) 21, 5361 – 5379 ª 2002 Nature Publishing Group All rights reserved 0950 – 9232/02 $25.00 www.nature.com/onc

DNA methylation and chromatin – unraveling the tangled web

Keith D Robertson*,1

1Epigenetic Gene Regulation and Cancer Section, National Cancer Institute, National Institutes of Health, Bethesda, Maryland, MD 20892, USA

Methylation of cytosines within the CpG dinucleotide by most prevalent epigenetic modification of DNA in DNA methyltransferases is involved in regulating tran- mammalian genomes. There are currently three known, scription and chromatin structure, controlling the spread catalytically active DNMTs, DNMT1, 3a, and 3b and of parasitic elements, maintaining genome stability in the each one appears to play a distinct and critical role in face of vast amounts of repetitive DNA, and X the cell (Bestor, 2000). CpG methylation profoundly chromosome inactivation. Cellular DNA methylation is influences many processes including transcriptional highly compartmentalized over the mammalian genome regulation, genomic stability, chromatin structure and this compartmentalization is essential for embryonic modulation, X chromosome inactivation, and the development. When the complicated mechanisms that silencing of parasitic DNA elements (Baylin et al., control which DNA sequences become methylated go 2001; Jones and Laird, 1999; Robertson, 2001). These awry, a number of inherited genetic diseases and cancer diverse processes nevertheless appear to share a may result. Much new information has recently come to common characteristic, that is, they all exert a light regarding how cellular DNA methylation patterns stabilizing effect which promotes genomic integrity may be established during development and maintained in and ensures proper temporal and spatial gene expres- somatic cells. Emerging evidence indicates that various sion during development. chromatin states such as histone modifications (acetyla- Genomic DNA methylation patterns are not tion and methylation) and nucleosome positioning randomly distributed. Rather, discrete regions, includ- (modulated by ATP-dependent chromatin remodeling ing most repetitive and parasitic DNA, are machines) determine DNA methylation patterning. Addi- hypermethylated, while other regions, such as CpG- tionally, various regulatory factors interacting with the rich regions often associated with the regulatory DNA methyltransferases may direct them to specific regions of genes (CpG islands), are hypomethylated DNA sequences, regulate their enzymatic activity, and (Yoder et al., 1997). Furthermore, DNA methylation allow their use as transcriptional repressors. Continued patterns change dramatically during embryonic devel- studies of the connections between DNA methylation and opment. Genome wide demethylation after fertilization chromatin structure and the DNA methyltransferase- is followed by waves of de novo methylation upon associated proteins, will likely reveal that many, if not all, embryo implantation. Not all sequences in the genome, epigenetic modifications of the genome are directly however, are demethylated upon fertilization and not connected. Such studies should also yield new insights all sequences become de novo methylated after into treating diseases involving aberrant DNA methyla- implantation. These exceptions further emphasize the tion. regional specificity of genomic DNA methylation Oncogene (2002) 21, 5361 – 5379. doi:10.1038/sj.onc. patterning (Reik et al., 2001). Evidence of the great 1205609 importance of these methylation patterns can be gleaned by examining the effects of disrupting them Keywords: DNA methyltransferase; chromatin; DNA in vivo. Engineered disruption of factors governing methylation; histone deacetylase; histone methylase; DNA methylation patterns in mice has revealed their transcriptional repression vital role in embryonic development (Li et al., 1992; Okano et al., 1999). Naturally occurring mutations in genes involved in controlling DNA methylation patterns, including one of the DNA methyltransferases, Introduction result in ICF, Rett, ATRX and fragile X syndromes (Robertson and Wolffe, 2000). Disruption of normal DNA methylation is a complex process whereby one of DNA methylation patterns is one of the most common three DNA methyltransferases (DNMTs) catalyzes the features of transformed cells and a number of studies addition of a methyl group from the universal methyl have revealed that methylation changes are early events donor S-adenosyl-L-methionine, to the 5-carbon posi- in the tumorigenesis process and contribute directly to tion of cytosine. This modification, occurring transformation. In tumor cells the normal regulation of predominantly within the CpG dinucleotide, is the the DNA methylation machinery is severely disrupted, such that the regional specificity of methylation patterns begins to be reversed, resulting in de novo *Correspondence: KD Robertson; E-mail: [email protected] methylation of CpG islands and hypomethylation of DNA methylation and cancer KD Robertson 5362 repetitive DNA (Baylin et al., 2001; Jones and Laird, 1992), which was mapped to chromosome 19p13.2. 1999; Robertson, 2001). Generally, DNMTs are believed to be composed of The role of DNA methylation in cancer has been two parts, a diversified amino terminal region and a reviewed extensively and will not be discussed at length relatively conserved carboxy terminal region. DNMT1 here (Baylin et al., 2001; Jones and Laird, 1999; has the largest amino terminal region of all the Robertson, 2001). Rather, I will focus on emerging mammalian DNMTs, which has roles in regulating data that may help to answer one of the most pressing the activity of the carboxy terminal catalytic domain, and intriguing questions in the DNA methylation field, nuclear localization, zinc binding, and in mediating namely how cellular DNA methylation patterns are protein – protein interactions (Figures 1 and 4, Table 1; established during mammalian development and then Bestor, 2000; Robertson, 2001). The carboxy terminal properly maintained in somatic cells. Clues have been region comprises the catalytic domain that is a contributed by numerous studies in the last few years, common feature of all cytosine DNA methyltrans- which indicate that DNA methylation and chromatin ferases. structure, or the ‘tightness’ of packaging of the DNA The biochemical and enzymatic properties of in nucleosomes and the higher order structures they DNMT1 have been studied in considerable detail. form, are physically and functionally linked (Bird, DNMT1 has a significant preference for hemimethy- 2002). For example, all known catalytically active lated double-stranded DNA relative to unmethylated DNA methyltransferases interact with histone deacety- double-stranded DNA. This unique property is why lases and the use of inhibitors of each of these DNMT1 is commonly referred to as the maintenance processes has revealed that they work together to methyltransferase (Bacolla et al., 1999; Flynn et al., repress transcription (Cameron et al., 1999). Studies in 1996; Glickman et al., 1997; Pradhan et al., 1997, 1999; plants, mice, and humans, using naturally occurring or Yokochi and Robertson, 2002). Analysis of the engineered mutations in chromatin remodeling subnuclear localization of DNMT1 supports this machines, have indicated that chromatin structure itself assignment. During the G1 and G2 phases of the cell may dictate cellular DNA methylation patterns (Bird, cycle, DNMT1 shows a diffuse nucleoplasmic distribu- 2001). I will first discuss what is currently known about tion pattern, but associates with sites of DNA synthesis the components of the cellular DNA methylation (replication foci) throughout S phase (Leonhardt et al., machinery, namely the DNA methyltransferases. I will 1992; Liu et al., 1998). These results led researchers to then summarize current knowledge of the interacting a classic and still attractive hypothesis. DNMT1, as the protein partners of the DNMTs and why the study of primary maintenance methyltransferase, is required to such factors may yield clues to the means by which maintain the epigenetic information encoded by methylation patterns are established. Lastly I will genome-wide DNA methylation patterns due to the discuss the connections between DNA methylation semiconservative nature of DNA replication. This and chromatin remodeling and describe a model of process results in two hemimethylated daughter how chromatin states may dictate genome-wide DNA chromosomes that must be fully methylated in order methylation patterning. for DNA methylation patterns to be properly main- tained. In contrast to this simple model, the expression patterns of DNMT1 in certain cell types has been The mammalian DNA methylation machinery – found to be somewhat paradoxical. If DNMT1 properties of the DNA methyltransferases (DNMTs) expression were strictly linked to DNA replication, then expression of DNMT1 would correlate with In this section, I will provide an overview of the proliferative state of the cell. However, Northern blot enzymes responsible for establishing and maintaining analysis showed that DNMT1 is highly expressed not cellular DNA methylation patterns in mammalian cells, only in the placenta and lung, but also in low namely the (cytosine-5) DNA methyltransferases proliferative tissues such as the heart and brain (Figure 1). Particular emphasis will be placed on the (Robertson et al., 1999; Yen et al., 1992). This domain structure of these proteins, their tissue-specific unexpected result may suggest an additional function patterns of expression, subnuclear localization, alter- for DNMT1 in addition to maintenance of methylation natively spliced isoforms, and their catalytic activity. at replication foci. Emerging lines of evidence suggest that tissue-specific alternative splicing produces several forms of DNMT1, which result in enzymes designed to DNMT1 carry out distinct roles in DNA methylation metabo- The (cytosine-5) DNA methyltransferase 1, now lism. properly referred to as DNMT1 (Figure 1), was the first enzyme to be isolated as a mammalian DNA DNMT1b methyltransferase, and more importantly, the only one that was identified via biochemical fractionation An alternative splice variant of human DNMT1 in methodology (Bestor, 2000). All other genes described somatic tissues was reported by two groups and was below were identified by database search. Following named DNMT1b (Bonfils et al., 2000; Hsu et al., the isolation of murine DNMT1 (Bestor et al., 1988), 1999). This isoform of DNMT1 contains an extra 48 its human homolog was identified in 1992 (Yen et al., base pairs between exons 4 and 5, generating a protein

Oncogene DNA methylation and cancer KD Robertson 5363

Figure 1 Schematic structure of all known mammalian DNA methyltransferases and DNMT-like proteins. Specific motifs involved in localization, catalysis (PC active site), or of unknown function (HRX-like, ATRX-like, and PWWP) are indicated with boxes with an additional 16 amino acids derived from this DNMT1o is localized in the cytoplasm, and region (Figure 2; Hsu et al., 1999). Dnmt1b mRNA is transiently relocates to the nucleus during the ubiquitously expressed in all cell lines tested and the eight-cell stage. This observation led to the sugges- protein possessed enzymatic properties comparable to tion that the cytoplasmic-nuclear translocation of DNMT1 in vitro (Bonfils et al., 2000). One group DNMT1o at a particular stage of embryogenesis is described the mRNA expression level of DNMT1b as essential for establishment of normal methylation 40 – 70% of the level of DNMT1 (Hsu et al., 1999), patterns in imprinted regions (Cardoso and Leon- while another group reported that DNMT1b protein hardt, 1999; Doherty et al., 2002). was present at only 2 – 5% of the level of DNMT1 protein (Bonfils et al., 2000). Thus, the relative DNMT1p abundance of DNMT1b and DNMT1 in somatic tissues remains unclear. The murine DNMT1 gene A larger form of DNMT1 mRNA was detected also undergoes alternative splicing and potentially exclusively in pachytene spermatocytes (Mertineit et generates an isoform (murine DNMT1b) similar to al., 1998), and thus, the transcript was named human DNMT1b, but would differ by two amino acids DNMT1p (Figure 2; Doherty et al., 2002). It was (Lin et al., 2000). A biological role for human and originally concluded that DNMT1p was not translated, murine DNMT1b in somatic tissues and the functional possibly due to the inhibitory effect of multiple short differences between them in vivo remains unknown. upstream open reading frames (Mertineit et al., 1998). However, the patterns of transcription and translation of this isoform remain controversial. In 2000, it was DNMT1o reported that an alternative DNMT1 transcript in Alternative splicing of sex-specific 5’ exons produces skeletal muscle, specifically differentiated myotubes, at least two DNMT1 variants at the mRNA level, was identical to DNMT1p (Aguirre-Arteta et al., which have been termed DNMT1o and DNMT1p 2000). It was also shown that this muscle-specific (Figure 2). A shorter isoform of DNMT1 isoform was translatable in both transfected cells and (DNMT1o), which is specific to oocytes, yields a in differentiated muscle cells. As with DNMT1o, large amount of enzymatically active protein which DNMT1p may play an important role during oogen- accumulates in oocytes during the growth phase esis, gametogenesis, or myogenesis. (Carlson et al., 1992; Mertineit et al., 1998). In fact, DNMT1o is the sole isoform of DNMT1 not only DNMT2 in the oocyte but also in the preimplantation embryo. It is noteworthy that DNMT1o exhibits In 1998, attempts to identify novel putative de novo an unusual trafficking movement. The majority of DNA methyltransferases in mammalian cells resulted

Oncogene DNA methylation and cancer KD Robertson 5364 Table 1 DNA methyltransferase-associated proteins involved in transcriptional repression and chromatin modification DNA methyltransferase Interacting protein Function of interacting protein How do they work together?

Dnmt1 HDAC1/2 Histone deacetylase Modification of chromatin by histone deacetylation, targeting methylation? pRb Tumor suppressor Sequester DNMT1 in non-dividing cell, Cell-cycle regulation target or modulate DNMT activity at replication foci? DMAP1 Co-repressor Recruiting other repressors, transcriptional repression PML-RAR Oncogenic transcription factor DNA-binding and interaction with other transcriptional co-regulators, targeting methylation MBD2/3 Methyl-CpG binding proteins Transcriptional repression in methylated regions, possible targeting of DNMT1 to hemimethylated DNA at replication foci?

Dnmt3a HDAC1 Histone deacetylase Modification of chromatin by histone deacetylation, targeting methylation? RP58 Transcription factor Sequence-specific DNA binding, targeting repression, maybe methylation as well? PML-RAR Oncogenic transcription factor DNA-binding and interaction with other transcriptional co-regulators, targeting methylation

Dnmt3b HDAC1 Histone deacetylase Modification of chromatin by histone deacetylation, targeting methylation? SUMO-1/Ubc9 Sumo ligase Modification of protein by sumoylation, altered localization or enzymatic activity?

Figure 2 Alternatively spliced isoforms of DNMT1. The mRNA splicing structure of the relevant region of DNMT1 (the 5’ end of the gene only) is indicated on the left and the resulting proteins are shown at the right. The presence or absence of important motifs and protein – protein interaction sites is also shown for each of the DNMT1 variants

in the identification of the DNMT2 gene (Figure 1), Yoder and Bestor, 1998). The characteristic motifs which shares homology with the pmt1 gene of fission found in all other active DNA methyltransferases, yeast (Okano et al., 1998b; Yoder and Bestor, 1998). including the conserved proline-cysteine dipeptide at Although the human DNMT2 gene was mapped to the active site, are well conserved in both the human chromosome 10p12-10p14 (Yoder and Bestor, 1998), and murine DNMT2 genes. Interestingly, DNMT2 may the human genome project revealed that its actual be the most highly conserved of all DNMTs across location is 10p15.1. DNMT2 mRNA is ubiquitously species since, unlike DNMT1 and DNMT3, DNMT2- expressed at very low levels (Okano et al., 1998b; like proteins have been found in yeast, plants, and the

Oncogene DNA methylation and cancer KD Robertson 5365 fruit fly Drosophila (Hung et al., 1999; Lyko et al., suggesting that DNMT3a expression is not regulated 2000; The Arabidopsis Genome Initiative, 2000; in a cell cycle-dependent manner like DNMT1 Wilkinson et al., 1995). At this time, however, no (Robertson et al., 2000b). In contrast to DNMT1, methyl-transfer activity of the DNMT2 gene product DNMT3a was found at discrete nuclear foci through- has been reported using in vitro assays (Dong et al., out the cell cycle that were not associated with DNA 2001; Okano et al., 1998b; Yoder and Bestor, 1998), synthesis. During late S phase, however, when although the DNMT2 protein does form stable DNA- heterochromatic regions of the genome are known protein complexes in vitro (Dong et al., 2001). to be replicated, some of the DNMT3a-enriched foci Elucidation of the exact function of this protein in appeared to overlap with replication foci (Bachman vivo will be of great importance. DNMT2 could, for et al., 2001). example, be the catalytic subunit of a DNA methylase complex that is inactive when expressed without one or DNMT3b and its multiple isoforms more of its associated factors. Alternatively it may possess a more complex DNA recognition sequence The human DNMT3b gene was mapped to chromo- beyond the CpG dinucleotide. some 20q11.2 (Robertson et al., 1999; Xie et al., 1999) and is 85% identical to murine DNMT3b (Figure 1). The catalytic domain, located at the carboxy terminus, DNMT3a is well conserved between DNMT3a and DNMT3b Low level, stable de novo methylation activity could be (more than 80% identity), whereas their amino observed in embryonic stem (ES) cells lacking terminal regions are poorly conserved (less than DNMT1, supporting the notion that enzymes other 30%). This further underscores the notion that the than DNMT1 contribute to de novo methylation in vivo variety in function of each of the DNMTs is most (Lei et al., 1996). DNMT3a and DNMT3b (Figure 1), likely due to their diverse N-terminal regions, while the two genes sharing significant homology to DNMT1, highly conserved C-terminal regions, which are were identified by EST database searches in 1998 using common among all mammalian DNMTs, have similar the conserved methyltransferase motifs as bait (Okano functions. DNMT3b, like DNMT3a, was shown to be et al., 1998a). The human DNMT3a gene was mapped an active DNA methylstransferase in vivo and in vitro to chromosome 2p23 (Robertson et al., 1999) and (Aoki et al., 2001; Hsieh, 1999; Okano et al., 1998a; shows 96% amino acid identity to its murine counter- Qiu et al., 2002). part (Xie et al., 1999). This is significantly higher than Compared to DNMT3a, the expression levels of the identity between human and murine DNMT1 DNMT3b are very low in most tissues. The testis, (78%), and suggests a critical function for this enzyme however, expressed high levels of DNMT3b, suggesting preserved throughout evolution. The carboxy terminal a crucial function for DNMT3b in spermatogenesis portion of DNMT3a includes the catalytic motifs (Okano et al., 1998a; Robertson et al., 1999; Xie et al., highly conserved in all cytosine DNA methyltrans- 1999). DNMT3b shows a diffuse distribution pattern ferases. DNMT3a has been shown to be enzymatically throughout the nucleus in NIH3T3 cells, while active both in vitro and in vivo by several groups, targeting to pericentromeric heterochromatin was although these reports differ somewhat in the exact observed in undifferentiated ES cells (Bachman et al., substrate preference of DNMT3a (Aoki et al., 2001; 2001). As will be discussed later in this review in the Gowher and Jeltsch, 2001; Hsieh, 1999; Lyko et al., context of ICF syndrome, DNMT3b appears to be 1999; Okano et al., 1998a; Yokochi and Robertson, specialized for the establishment and/or maintenance of 2002). DNA methylation of the minor satellite repeats Proteins of the DNMT3 family commonly have a (satellites 2 and 3 on human chromosomes 1, 9, and cysteine-rich domain in the amino terminal region, 16) and its co-localization with pericentromeric hetero- which is referred to as the PHD (plant homeo- chromatin in consistent with this function. domain) region or ATRX-like domain (Figure 1) In contrast to DNMT3a, there are several isoforms because of its homology with the PHD region of the (five for human and eight for mouse) of DNMT3b that ATRX gene. ATRX is a member of the SNF2/SWI2 result from alternative splicing. Three major isoforms, family of ATP-dependent chromatin remodeling namely DNMT3b1, 3b2, and 3b3, have been identified enzymes. This similarity suggests that DNMT3 (Figure 3) (Okano et al., 1998a), and were shown to be proteins may be associated with structural changes expressed in a tissue-specific manner (Robertson et al., in chromatin via protein – protein interactions at the 1999). DNMT3b1 is the longest form and is usually amino terminal region. DNMT3a transcripts were regarded as the typical gene product of DNMT3b. All ubiquitously expressed in adult tissues, most tumor other splice variants encode smaller proteins. cell lines, early embryos, and embryonic stem (ES) DNMT3b2 also demonstrated methyltransferase activ- cells (Okano et al., 1998a; Robertson et al., 1999; Xie ity in vitro, however DNMT3b3, an even shorter et al., 1999). Northern blot analysis showed that isoform lacking 63 amino acids within the central DNMT3a activity is particularly high at embryonic region of the catalytic domain, did not (Aoki et al., day 10.5 (Okano et al., 1998a). Unlike DNMT3b, the 2001; Okano et al., 1998a; Qiu et al., 2002). Several level of DNMT3a mRNA was less affected when cells other splice variants, DNMT3b4 and DNMT3b5 were arrested in the G0/G1 phase of the cell cycle, (Figure 3) were identified and are expressed predomi-

Oncogene DNA methylation and cancer KD Robertson 5366

Figure 3 Alternatively spliced isoforms of the human DNMT3b protein. Important motifs are denoted with boxes. Motifs I, IV, VI, IX, and X correspond to the highly conserved methyltransferase motifs involved in catalysis. The shading in the amino-terminal regions of DNMT3b4 and DNMT3b5 indicate that the exact splicing structure in region-1 has not been determined. The hatch marks at the extreme C-terminus of DNMT3b5 denote a change in reading frame after splicing (Robertson et al., 1999)

nantly in the testis (Robertson et al., 1999). Interest- Insights into functions of the DNA methylation ingly, these forms of DNMT3b lack several of the machinery derived from knockout models conserved catalytic motifs and possess additional novel sequences resulting from frameshifts after alternative The recent history of DNA methylation research has splicing and are therefore unlikely to be catalytically been punctuated by several elegant studies using active. These splice variants may possess the critical murine knockout models to elucidate the role of functions encoded by the N-terminal region of DNA methylation in general, and of the functions of DNMT3b, including protein – protein interaction sites, the individual DNMTs in development, transcriptional but lack catalytic activity. They may therefore act to regulation, chromatin structure maintenance and inhibit some function of DNMT3b1-2, like de novo genomic imprinting. Comparison of the transgenic methylation of certain sequences, by interacting with knockout mice with the corresponding knockout ES the same set of targeting proteins. cells provides additional data on the role of each of the DNMTs in a pluripotent cell versus a more differ- entiated cell type. This section of the review will DNMT3L summarize the effects of inactivating each of the The human DNMT3L (DNMT3-like, Figure 1) gene DNMTs at the individual cell and the whole organism located on human chromosome 21q22.3 was originally levels. isolated by database analysis of the genome sequence in 2000 (Aapola et al., 2000). The identification of the DNMT1 murine DNMT3L gene (Aapola et al., 2001), which is 61% identical to its human counterpart, soon followed. Murine ES cells lines homozygous for a DNMT1 Similarity between DNMT3L and DNMT3a and knockout were obtained by targeted disruption (Li et DNMT3b is restricted mainly to the cysteine-rich al., 1992). The mutant ES cells were viable and showed region encompassing the PHD/ATRX-like region in normal morphology. Only residual DNA methyltrans- the amino terminus of DNMT3L. DNMT3L lacks the ferase activity was observed in lysates derived from critical catalytic motifs commonly seen in all other these homozygous mutant cells. DNA from homo- DNA methyltransferases, including the ‘FGG’ zygous mutant cells was found to have a 5- sequence in conserved motif I, the ‘PC’ catalytic active methylcytosine content of roughly 30% of the DNA site in motif IV, and the ‘ENV’ sequence in motif VI from wild type cells, in which about 60% of all CCGG believed to be involved in cofactor binding. Thus, the sites were methylated (that is, a 70% decrease in 5- gene product is almost certainly catalytically inactive, methylcytosine content in the CCGG sequence) (Li et although this has yet to be tested directly. DNMT3L is al., 1992). Interestingly, targeted disruption of the highly expressed in testis and mouse embryos (Aapola DNMT1 gene in a human colorectal carcinoma cell line et al., 2001), and is likely involved in gametogenesis for yielded somewhat different results (Rhee et al., 2000). the process of establishing genomic imprints (Bourc’his DNA from the DNMT1 deficient HCT116 cell line had et al., 2001). a 5-methylcytosine content of about 80% of the DNA

Oncogene DNA methylation and cancer KD Robertson 5367 from wild type cells, in which approximately 4% of all and may play a particularly important role in brain cytosines were methylated (that is, a 20% decrease in development and function as has been suggested total genomic 5-methylcytosine content). A direct previously (Fan et al., 2001; Robertson, 2001; Tucker, comparison of the results between these two experi- 2001). mental systems is somewhat difficult due to the different methods used for measuring the amount of DNMT1o DNA methylation in the original wild type cells. Regardless of whether this discrepancy arises from Since DNMT1o demonstrated a unique nuclear- different experimental methods or from the radically cytoplasmic trafficking movement during oogenesis different cell types used in each study, these results and preimplantation development, a functional role suggest that methyltransferases other than DNMT1 for DNMTo in genomic imprinting in the oocyte was contribute significantly to the homeostasis of DNA proposed (Carlson et al., 1992; Mertineit et al., 1998). methylation patterns. Transgenic mice heterozygous for Although males homozygous for a deletion of the the DNMT1 knockout were indistinguishable from maternal-specific exon were normal and fertile, homo- wild type littermates. Homozygous mutation of zygous females were infertile. That is, most DNMT1 was embryonic lethal and mutant embryos heterozygous fetuses of homozygous females died failed to develop beyond mid-gestation, strongly during the last third of gestation, indicating that suggesting that DNA methylation is essential for deletion of DNMT1o causes a pure-maternal pheno- mammalian development. DNMT1 mutants demon- type during oogenesis (Howell et al., 2001). DNA strated significant hypomethylation of several genes, methylation at certain imprinted loci, but not the which affected a variety of epigenetic events including whole genome, was lost in the heterozygous embryos. genomic imprinting (Li et al., 1993), X chromosome Therefore it appears that DNMT1o is required to inactivation (Beard et al., 1995), and the suppression of maintain methylation patterns at specific imprinted loci transcription from parasitic elements (Walsh et al., during the fourth embryonic S phase. 1998). More recently, conditional knockouts of DNMT1 DNMT2 have been generated to examine its role in the development and maintenance of specific tissues, Murine ES cells with a homozygous disruption of the particularly the central nervous system. In one study, DNMT2 gene appeared to be normal, suggesting that the cre/lox system was used to delete the DNMT1 locus DNMT2 is not essential for cell proliferation, growth, from primary fibroblasts in culture (Jackson-Grusby et and differentiation. No significant change in the al., 2001). The resulting DNMT1-null cells displayed methylation status of both genomic DNA and newly extensive genomic hypomethylation and uniform cell integrated provirus DNA was observed in the death by apoptosis within 6 days of DNMT1-deletion. DNMT27/7 ES cells. These results indicated that The connection between reduced DNA methylation DNMT2 is required for neither maintenance nor de and apoptosis was further underscored by showing that novo methylation in vivo (Okano et al., 1998b). interfering with p53 function, a key mediator of apoptosis, prolonged the life span of the DNMT1-null DNMT3 cells and reduced the level of apoptosis. Microarray expression studies revealed that 4 – 10% of all genes ES cell lines homozygous for a DNMT3a knockout were upregulated and 1 – 2% were down-regulated (DNMT3a7/7) retained their undifferentiated when DNMT1 was deleted and that these genes were morphology and showed normal de novo methylation involved in a multitude of cellular functions (Jackson- activity on newly integrated proviral DNA following Grusby et al., 2001). Conditional deletion of DNMT1 retroviral infection (Okano et al., 1999). DNMT3a in neural precursor cells early in development resulted heterozygous mice were normal and fertile. Although in genome-wide demethylation and embryonic lethality, DNMT3a7/7 homozygous mice appeared to be although brain structure appeared normal. Mice normal at birth, undergrowth of such mutant mice by delivered by caesarian section exhibited aberrant 18 days was obvious and all of the animals died by 4 breathing and their lungs failed to inflate, indicating weeks of age. potential defects in respiratory rhythmogenesis or A homozygous knockout of the DNMT3b gene in neurotransmission. Deletion of DNMT1 at this early ES cells yielded a phenotype similar to that of developmental stage in a small fraction of neural DNMT3a disruption. DNMT3b7/7 ES cell lines precursor cells resulted in a marked selection against and embryos exhibited comparable degrees of demethy- the demethylated cells such that the mice developed lation to those of DNMT3a7/7 homozygous ES cells normally and were devoid of demethylated, DNMT1- and embryos and a comparable ability to de novo deficient cells (Fan et al., 2001). Deletion of DNMT1 methylate retroviral sequences (Okano et al., 1999). from post-mitotic neurons later in development was However, an analysis of early embryos revealed clear compatible with both normal development and normal evidence that DNMT3a and DNMT3b have indepen- DNA methylation patterns. Thus, these studies dent functions. Unlike DNMT3a7/7 mice, no viable strongly suggest that DNA methylation is critical for DNMT3b7/7 mice were recovered at birth. Interest- normal mammalian development during embryogenesis ingly, the minor satellite repeats in the pericentromeric

Oncogene DNA methylation and cancer KD Robertson 5368 region were significantly hypomethylated in Robertson, 2001). This section of the review will focus DNMT3b7/7 but not in DNMT3a7/7 cells, specifically on chromatin-related proteins known to indicating that the minor satellite repeats are specific associate with the DNMTs and discuss the functional targets of DNMT3b. As will be discussed later, consequences of these interactions. naturally occurring mutations in the human DNMT3b gene give rise to ICF syndrome, which is also The retinoblastoma protein, Rb characterized by hypomethylation of pericentromeric repeats (Hansen et al., 1999; Okano et al., 1999; Xu et The retinoblastoma protein, Rb, is a protein with al., 1999). intimate links to transcriptional regulation in the DNMT3a7/7, DNMT3b7/7 double mutant ES context of chromatin. Robertson et al. (2000a) initially cells were also generated and were viable. However, de identified an interaction between Rb, E2F1, and novo methylation of newly integrated retroviral DNMT1 via biochemical fractionation, and this result sequences was no longer observed, suggesting that the was recently confirmed by others (Figure 4; Pradhan gene products of DNMT3a and DNMT3b have and Kim, 2002). The amino terminal region of overlapping functions in ES cells with regard to de DNMT1 could interact with the A/B pocket region novo methylation of parasitic elements. Embryonic of Rb (Robertson et al., 2000a), and also appears to be defects in the double mutant mice were more severe capable of interacting with the B/C pocket (Pradhan than those of DNMT3a7/7 single mutant mice. and Kim, 2002). Rb and DNMT1 cooperate to repress Global methylation levels in ES cell lines and E9.5 transcription from E2F-responsive promoters in vivo day embryos were dramatically reduced in although this repression did not depend on the DNMT3a7/7, DNMT3b7/7 knockout mutants catalytic activity of DNMT1 (Robertson et al., compared to the single mutants (Okano et al., 1999), 2000a). A recent report demonstrated that Rb binding supporting the notion that DNMT3a and DNMT3b to DNMT1 inhibited the ability of DNMT1 to bind to have at least partially overlapping functions in the DNA in vitro and exerted a strong negative effect on establishment of cellular DNA methylation patterns DNMT1 catalytic activity. Overexpression of Rb in during development. cells resulted in a significant reduction in total genomic 5-methylcytosine levels (Pradhan and Kim, 2002). Rb, like DNMT1, also co-localizes with DNA replication DNMT3L foci, specifically early S phase perinucleolar replication Targeted disruption of DNMT3L gene was recently foci (Kennedy et al., 2000). The role of Rb at these reported (Bourc’his et al., 2001). Homozygous animals sites remains unclear. of both sexes were viable but sterile. Bisulfite genomic Hypophosphorylated Rb interacts with E2F family sequencing revealed that a defect in DNMT3L resulted members and inhibits their transactivation function. As in the disruption of maternal methylation imprints in cells prepare to divide, Rb is phosphorylated and homozygous oocytes, while genome-wide methylation dissociates from E2F, allowing it to recruit co- patterns appeared to be normal, indicating that the activators and stimulate transcription of genes involved DNMT3L protein contributes to establishment of in cell cycle progression (Dyson, 1998). Interestingly, genomic imprints during oogenesis. As was discussed Rb has been shown to repress transcription at E2F- in the previous section, DNMT3L is almost certainly responsive promoters by recruitment of both HDACs not a functional DNA methyltransferase therefore the and histone methyltransferases (HMTs) and subse- observation that its disruption results in any loss of quent binding of the methylated lysine binding proteins DNA methylation indicates that DNMT3L may target of the heterochromatin protein (HP) 1 family (Brehm other functional DNA methyltransferases, such as et al., 1998; Luo et al., 1998; Nielsen et al., 2001). As DNMT3a and DNMT3b, to specific loci in the genome will be discussed in the next section, DNMT1 can also via direct or indirect protein – protein interaction. interact directly with HDAC1 and HDAC2 (Robertson et al., 2000a; Rountree et al., 2000) and methylated cytosine itself serves as a recognition site for another Chromatin-associated proteins that interact with class of repressors of the methyl-CpG binding protein DNMTs (MBD) family. The MBDs have also been shown to recruit HDACs to methylated DNA and repress In vitro studies of the DNMTs have shown that they transcription (Bird and Wolffe, 1999). exhibit little sequence preference beyond the CpG What are the functional consequences of the dinucleotide, while the knockout studies emphasize the interaction between Rb and DNMT1? We have non-random nature of DNA methylation and the previously proposed a model wherein DNMT1 binding distinctive roles for the individual DNMTs. A prime to Rb at E2F-containing promoters may be a mediator of these distinctive roles is most likely the mechanism to sequester DNMT1 from the genome complement of proteins that interact with the DNMTs and prevent promiscuous DNA methylation events at specific stages of development and differentiation, (Robertson, 2001). This is supported by recent findings and within the environment of chromatin. A number of that the catalytic activity of DNMT1 is inhibited by proteins have now been identified which interact with binding to Rb (Pradhan and Kim, 2002). This study one or more of the DNMTs (Figure 4, Table 1; also speculated that DNMT1 bound to Rb would not

Oncogene DNA methylation and cancer KD Robertson 5369

Figure 4 Summary of known DNMT1, DNMT3a, and DNMT3b interacting proteins. The interacting region on each DNMT pro- tein is indicated with a horizontal bar. An arrow not pointing to a specific region indicates that the interaction domain has not been mapped

be capable of binding to PCNA, another DNMT1- HDAC-independent components. The HDAC-depen- associated protein thought to recruit DNMT1 to dency can be demonstrated using the HDAC inhibitor replication foci. Therefore this model may require trichostatin A (TSA), which relieves a substantial revision to take into account these new observations. amount of the DNMT1-mediated repression (Fuks et The exact nature of the complexes formed between al., 2000; Robertson et al., 2000a). DNMT1 was shown DNMT1, Rb, and PCNA in silent chromatin versus to bind HDAC1 via a transcriptional repression region replication foci will require extensive additional study other than the HRX-homology domain (Figure 4; Fuks however we propose a model (Figure 5) for how the et al., 2000). A direct interaction between the amino nature of these interactions may change during the cell terminal half of DNMT1 and HDAC2 has also been cycle and how they may effect the catalytic activity of demonstrated (Figure 4; Rountree et al., 2000). At DNMT1. It is likely that temporal changes in the present, there is no evidence to believe that there are complement of DNMT1-associated proteins, particu- major functional differences between HDAC1 and larly Rb and HDAC2, which associate with early and HDAC2 since they are highly similar proteins (85% late replication foci, respectively, will be critical identical at the amino acid level in humans). Thus, the modulators of DNMT1 activity at this site. interaction of DNMT1 with HDAC1 or HDAC2 likely has the same functional consequence. The TSA- insensitive component of DNMT1 repression may be HDAC1 and HDAC2 mediated by its interaction with DNMT1-associated Several studies have now shown that DNMT1, protein (DMAP) 1 (Rountree et al., 2000). DMAP1 DNMT3a, and DNMT3b associate with HDAC1 and binds to the extreme amino terminus of DNMT1, a HDAC2 in vitro and in vivo (Figure 4 and Table 1; region missing in certain germ-cell specific DNMT1 Bachman et al., 2001; Fuks et al., 2000, 2001; splice variants (Figures 2 and 4). DMAP1 interacts Robertson et al., 2000a). Histone deacetylases can with another potent transcriptional repressor TSG101, remove acetyl groups from the amino terminal core although the existence of a DNMT1, DMAP1, and histone tails, which are critical modulators of chroma- TSG101 ternary complex has not been reported. tin structure, leading to the assembly of tightly packed DMAP1 co-localizes with DNMT1 at replication foci chromatin and rendering the sequence inaccessible to throughout S phase (Figure 5) and may affect catalytic the transcription machinery (Jenuwein and Allis, 2001). activity or targeting of DNMT1. DNMT1-mediated transcriptional repression has been The DNMT3s also interact with HDAC1 through shown to be comprised of both HDAC-dependent and their ATRX-homology/PHD regions (Bachman et al.,

Oncogene DNA methylation and cancer KD Robertson 5370

Figure 5 Model for how DNMT1 and Rb may function to regulate cellular DNA methylation patterns. In resting cells (top), DNMT1 binding to Rb may inhibit the catalytic activity of DNMT1 and prevent unscheduled methylation of the genome. DNMT1 may also potentiate Rb-mediated repression of E2F-responsive genes. During early S phase (middle), DNMT1, PCNA, Rb, and DMAP1 co-localize with replication foci. The catalytic activity of DNMT1 at this time, when transcriptionally active hypomethy- lated sequences are replicated, may be low due to the presence of Rb (denoted by thin arrow), to prevent aberrant methylation. Alternatively, DNMT1 may be completely inactive in early S phase and maintenance methylation is accomplished by another DNMT. In late S phase (bottom), Rb is no longer present at replication foci and HDAC2 begins to co-localize with DNMT1. The departure of Rb may stimulate DNMT1 catalytic activity at a time when transcriptionally inactive hypermethylated regions are replicated and a highly active enzyme would be necessary (denoted by heavy arrow). An ATP-dependent chromatin remodeling enzyme or complex (‘ATPase’) appears to be crucial for proper DNA methylation throughout the replication process. Filled lolli- pops represent methylated CpG sites and open lollipops represent unmethylated CpG sites. The assembly of potential proteins dur- ing replication is shown for only one strand. The remainder of the macromolecular DNA replication machinery is not shown

2001; Fuks et al., 2001), which are not present in to the temporal association of DNMT1 and HDAC2 DNMT1 (Figure 4 and Table 1). DNMT3a and at replication foci (Figure 5). Heterochromatic, DNMT3b also exhibited a TSA-insensitive transcrip- hypermethylated sequences enriched in hypoacetylated tional repression component, however the protein – histones are replicated in late S phase. The recruitment protein interactions mediating this repression are of HDAC2 to replication foci by DNMT1 may help to unknown (Bachman et al., 2001). Therefore, it is likely coordinate remethylation of the newly synthesized that the repression capability of DNMT1 and DNA and deacetylation of newly deposited histones DNMT3s are the result of distinct protein – protein in late S phase (Rountree et al., 2000). It is possible interactions, although histone deacetylase activity that DNMT recruitment to, and methylation of, a appears to be involved in both. specific genomic region then brings in HDACs to What is the functional consequence of the HDAC- deacetylate core histones in the newly methylated DNMT interactions? This extremely important ques- region and allow for chromatin compaction and tion remains unanswered. The potential importance is transcriptional silencing. This implies that DNA further underscored by the now universal nature of the methylation is the primary event in transcriptional association. All known catalytically active DNMTs silencing and chromatin modifications follow. Evidence have been shown to interact with HDACs (Table 1). gathered from experimental systems in which gene One scenario that has been proposed previously relates silencing occurs in a regulated manner, such as X

Oncogene DNA methylation and cancer KD Robertson 5371 chromosome inactivation in females and host genome island. Co-immunoprecipitation studies revealed an defense-mediated silencing and de novo methylation of interaction between PML – RAR and DNMT1 and retroviral DNA, suggests that other mechanisms may DNMT3a, and immunofluorescence studies showed co- also exist. In X chromosome inactivation, transcrip- localization of PML – RAR and DNMT1 and Dnmt3a tional silencing and histone modifications (both when over expressed. It appeared that DNMT3a could deacetylation and methylation) occur well before de interact with regions on both the PML and RAR novo DNA methylation (Csankovszki et al., 2001; portions of the fusion protein. Treatments of cells with Heard et al., 2001). Furthermore, transcriptional the DNA methylation inhibitor 5-aza-2’-deoxycytidine silencing of newly introduced retroviral sequences (5-aza-dC) or the HDAC inhibitor TSA, revealed that occurs before de novo methylation and can occur in the PML – RAR-mediated transcriptional repression the complete absence of the de novo DNA methyl- and differentiation blockage was due to the temporally transferases (Pannell et al., 2000), which have been distinct recruitment of both histone deacetylases (at shown to be responsible for methylation of this class of early time points) and DNA methyltransferases (at sequence (Okano et al., 1999). Therefore an alternative later time points) (Di Croce et al., 2002). The scenario is one in which HDACs are recruited to a functional significance of the interaction between region destined to undergo long-term transcriptional DNMT1 and DNMT3a and PML in normal cells is silencing via interaction with other proteins. Once the unclear, although it is possible that the DNMTs may region is deacetylated and possibly with the help of carry out some critical function in the PODs. other chromatin remodeling factors, the DNMTs gain Interestingly, another PML-associated POD protein access to the DNA or are attracted to a particular and transcriptional repressor, Daxx, has been shown to chromatin structural feature, the region is methylated, associate with DNMT1 and it remains possible that and stable, heritable, long-term gene silencing and Daxx bridges the interaction between PML – RAR and chromatin compaction is achieved (Figures 5 and 6). DNMT1 (Li et al., 2000; Michaelson et al., 1999; Alternatively, both models may operate and depend on Zhong et al., 2000b). Perhaps most significantly are the the transcriptional and replicative state of the cell. implications of this work in cancer since there have been no prior studies showing how regional hyper- methylation events, common to so many tumor cells, PML – RAR might occur. As has been mentioned previously, there is a great need to better understand how cellular DNA methylation RP58 patterns are targeted to, or restricted from, certain regions of the genome. A fascinating recent study, DNMT3a and DNMT3b were recently shown to directly relevant to the targeting issue, showed that interact directly with a protein called RP58 via the DNMT1 and DNMT3a can interact with the leukemia- ATRX-like domain (Figure 4 and Table 1; Fuks et al., promoting PML – RAR fusion protein (Figure 4 and 2001). This is also the region of DNMT3a and Table 1; Di Croce et al., 2002). Acute promyelocytic DNMT3b that interacts with HDAC1 (and most likely leukemia (APL) is caused by a reciprocal translocation HDAC2). RP58 is a sequence-specific zinc finger DNA of the retinoic acid receptor a (RARa) gene on binding protein and transcriptional repressor asso- chromosome 17 to one of several other chromosomes. ciated with heterochromatin (Aoki et al., 1998; Meng To generate the PML – RAR fusion, the most common et al., 2000). It contains a POZ domain and several in APL, the RARa gene is fused to the promyelocytic Kru¨ ppel-type zinc fingers commonly seen in other leukemia gene (PML) on chromosome 15. PML is a transcriptional repressors (Ryan et al., 1999). The critical component of discrete nuclear structures repression activity of RP58 was enhanced by co- referred to as PML-oncogenic domains (PODs, ND10 expression of DNMT3a and the cooperative effect did and nuclear bodies). PODs, which are composed of a not require the catalytic activity of DNMT3a (Fuks et number of proteins, including Sp100, Sp140, SUMO-1 al., 2001). This suggests that DNMT3a acts as a and Daxx, may be involved in transcriptional activa- structural component, rather than an active enzyme, in tion, DNA replication, and apoptosis (Li et al., 2000; this repression pathway. Nuclear localization studies Muller et al., 2001; Zhong et al., 2000a,b). The PML – using a DNMT3a fragment lacking the catalytic RAR fusion protein, which retains the DNA and domain also support this notion. The isolated amino ligand binding domains of RARa, disrupts the PODs, terminal regulatory domain of DNMT3a co-localized but they can be restored by treatment of cells with with heterochromatin-associated proteins like HP1a as retinoic acid (RA). RA treatment and POD reorgani- well as methyl-CpG binding proteins like MeCP2 zation correlates with differentiation of the APL cells, (Bachman et al., 2001). Thus, the co-localization of indicating that PODs may have a critical role in DNMT3a with these types of proteins suggests that it differentiation of promyelocytes (Di Croce et al., 2002; may be an important component of densely methy- Zhong et al., 2000a). lated, pericentromeric heterochromatin. Although the Di Croce et al. (2002) showed that PML – RAR ability of DNMT3a and RP58 to cooperate in could repress a model RA target gene, RARb2,and transcriptional repression was not dependent on the that the repression coincided with de novo methylation catalytic activity of DNMT3a, it still remains possible of the 5’ end of the endogenous RARb2 promoter CpG that one function of RP58 could be to target DNMT3a

Oncogene DNA methylation and cancer KD Robertson 5372

Figure 6 Model for how HDACs, ATP-dependent chromatin remodeling enzymes, and DNA methyltransferases may cooperate to set up region-specific DNA methylation patterns. The histones within a transcribed or transcriptionally competent region destined for silencing (top) may first be deacetylated (and potentially methylated) and this likely initiates transcriptional silencing. It remains to be determined exactly how the HDAC would first be targeted to this region. The chromatin remodeler can now recognize the chromatin and mobilize nucleosomes in an ATP-dependent manner to allow the DNMT access to its target DNA sites or create a particular chromatin signature or ‘epitope’ that is recognized by the DNMT or DNMT-containing complex. Ample evidence exists for a DNMT-HDAC containing complex but it remains to be determined if the ATPase is directly or indirectly associated with the DNMT complex. Once the region is methylated, the methylated cytosines will recruit methyl-CpG binding proteins (MBD) and their associated co-repressors to further reinforce transcriptional silencing and chromatin compaction

to specific DNA sequences that are destined to become which contains a DNA modification module (DNMT), de novo methylated. Post-translational modifications, a methylcytosine reocognition subunit (methyl-CpG other protein cofactors expressed in a tissue-specific binding protein), and a histone modifying subunit manner (RP58 is highly expressed in the brain for (HDAC). This complex could have roles in directing example (Meng et al., 2000)), missing from the cell DNMT1 to hemimethylated sequences following DNA culture system used to first characterize this interaction, replication, silencing of genes during S phase, or could influence the function of the RP58 – DNMT3a deacetylation of newly deposited histones in a manner complex. akin to the previously described DNMT-DMAP1- HDAC2 interactions. MBD2 and MBD3 A physical interaction between DNMT1 and methyl- ICF syndrome – a disease caused by aberrant DNA CpG binding proteins was also reported (Figure 4 and methylation and chromatin structure Table 1). DNMT1 was co-immunoprecipitated with MBD2 and MBD3 as one potential complex (Tate- The identification of numerous interactions between matsu et al., 2000). MBD2 and MBD3 have recently DNA methyltransferases and chromatin associated been reported to be components of the large macro- proteins like histone deacetylases, Rb, and RP58 molecular MeCP1 repressor complex, which is capable provides a clear link between DNA methyltransferases of preferentially binding, remodeling, and deacetylating and transcriptional regulation and chromatin structure methylated DNA-containing nucleosomes in vitro modulation. Is there additional in vivo evidence, (Feng and Zhang, 2001). MBD2 and MBD3 co- particularly in human cells, that these processes are localized with DNMT1 at replication foci in late S connected? While knockout studies in mouse models phase. Furthermore, the MBD2 – MBD3 complex have been very revealing, the severity of the phenotype, demonstrated binding affinity for both hemimethylated namely embryonic lethality for DNMT1 and and fully methylated DNA and repressed transcription DNMT3b, limits the information that can be gained in a TSA-sensitive manner (Tatematsu et al., 2000). to embryonic development. Are there models or Thus, these results suggest the possibility of an ultimate diseases involving less severe mutations in the DNA ‘all-in-one’ type transcriptional repression complex, methylation machinery that could provide clues? The

Oncogene DNA methylation and cancer KD Robertson 5373 recent finding of a human genetic disease caused by DNMT3b lead to highly selective losses of methylation mutations in a DNA methyltransferase gene, the only from the genome (Kondo et al., 2000; Tuck-Muller et such disease known, has provided a wealth of al., 2000). Several other regions have been shown to information about the more subtle roles a particular become hypomethylated in ICF patients including, a- DNMT may play in determining genomic DNA satellite and Alu repeats (Miniou et al., 1997; methylation patterns. Schuffenhauer et al., 1995), the non-satellite repeats ICF syndrome (immunodeficiency, centromere D4Z4 and NBL2 (Kondo et al., 2000), the H19 gene instability, facial anomalies) is a very rare recessive (Schuffenhauer et al., 1995), and several genes (G6PD, disorder caused by mutations in the DNMT3b gene SYBL1, AR and PGK1) on the inactive X chromosome (Hansen et al., 1999; Okano et al., 1999; Xu et al., of female ICF patients (Hansen et al., 2000). Biallelic 1999). Most ICF patients are compound heterozygotes expression of several genes and advanced replication for their DNMT3b mutations and, with one exception, timing of the normally late S phase replicating inactive all of these mutations occur within the carboxy X chromosome have also been noted (Hansen et al., terminal catalytic domain of DNMT3b and likely fully 2000). The marked lack of autosomal genes in the list or partially impair catalytic activity (Robertson, 2001; of affected DNA sequences in ICF syndrome suggests a Xu et al., 1999). Phenotypically, ICF syndrome is specialized role for DNMT3b in gene-poor hetero- characterized by a profound immunodeficiency with an chromatin. The transcribed genes (as opposed to absence or severe reduction in at least two immuno- repeats of some kind) most consistently affected in globulin isotypes, variable impairment in cellular ICF syndrome all appear to reside on the inactive X immunity, unusual facial features, neurologic and chromosome, whose methylation is likely regulated in a intestinal dysfunction, and delayed developmental manner distinct from autosomal sequences. milestones (Franceschini et al., 1995; Smeets et al., 1994). DNA methylation and aberrant chromatin structure in At the cytogenetic level, primary and cultured cells ICF syndrome from ICF patients exhibit marked elongation of juxtacentromeric heterochromatin. This elongation In ICF syndrome it appears that loss of methylation occurs most consistently on chromosomes 1 and 16, from pericentromeric heterochromatin and the inactive and to a lesser extent chromosome 9. Abnormalities X chromosome results in aberrant chromatin structure. which have been observed include multiradial chromo- Pericentromeric heterochromatin is massively decon- somes involving multiple arms (3 – 12) of one or more densed and promoter regions of genes on the inactive of the decondensed chromosomes, whole-arm chromo- X chromosome showing reactivation demonstrate some deletions or duplications, translocations, increased susceptibility to nucleases, indicative of a centromeric breakage, and in rare cases telomeric more open chromatin configuration (Hansen et al., associations (Franceschini et al., 1995; Smeets et al., 2000). ICF syndrome also demonstrates that hypo- 1994; Tuck-Muller et al., 2000). These observations methylation results in aberrant transcription, which strongly suggest that defective forms of DNMT3b lead may also be directly related to chromatin structure to chromosome instability and large-scale changes in changes since a number of genes whose expression is chromatin structure. altered in ICF syndrome do not display DNA methylation changes (Ehrlich et al., 2001). This suggests that DNA methylation is critical for the Molecular aspects of ICF syndrome long-term maintenance of repressive condensed chro- One of the most consistent features of ICF syndrome is matin. Since the regions of the genome which juxtacentromeric repeat sequence hypomethylation on DNMT3b is responsible for methylating are never chromosomes 1, 9, and 16 (Jeanpierre et al., 1993). properly methylated in ICF cells, the subsequent Interestingly, these chromosomes contain the largest recruitment of other proteins involved in maintaining blocks of classical satellite long tandem repeat arrays or reinforcing chromatin compaction, such as the (satellite 2 for chromosomes 1 and 16, and satellite 3 MBDs and their associated HDACs, or the MeCP1 for chromosome 9) adjacent to their centromeres repressor complex, never occurs (Bird, 2002; Feng and (Jeanpierre et al., 1993; Tagarro et al., 1994). These Zhang, 2001). Therefore ICF syndrome can be regions are normally hypermethylated in somatic cells regarded not only as a disease of aberrant DNA and such methylation is likely essential for proper methylation patterns, but also of aberrant chromatin centromere structure and stability (Jeanpierre et al., structure. The chromatin structural changes may be 1993; Tuck-Muller et al., 2000). A recent study, using directly or indirectly related to the ability of DNMT3b bisulfite genomic sequencing, revealed that satellite 2 to act as a transcriptional repressor via both HDAC- repeat methylation was reduced from roughly 70% in dependent and independent mechanisms. Alternatively, normal lymphoblasts, to 20% in ICF cells (Hassan et reduced methylation may result in global imbalances in al., 2001). Although there is a drastic loss of transcription factor binding by allowing transcription methylation from satellite sequences in ICF patients, factors access to sites that would normally be blocked the overall reduction in cellular 5-methylcytosine levels by closed chromatin configuration, or by excesses of is relatively small (about 7% in primary ICF brain transcriptional repressors, such as the methyl-CpG tissue) further underscoring the idea that mutations in binding protein-containing complexes, that may gain

Oncogene DNA methylation and cancer KD Robertson 5374 promiscuous access to important transcriptional packed nucleosomes characteristic of transcriptionally control regions. The exact nature of defects in DNA inactive heterochromatin (Vignali et al., 2000; Wolffe methylation and chromatin structure in ICF cells will and Pruss, 1996). Inherited mutations in SNF2-like require considerable study, however, this disease genes give rise to a number of human diseases, further emphasizes the connection between DNA including Werner, Bloom and Cockayne syndromes, methylation, DNA methyltransferases, and chromatin and Xeroderma pigmentosum (Ellis, 1997). In the next structure. section we will review the connections between DNA methylation and chromatin remodeling enzymes of the SNF2 family and finally propose a model for how DNA methylation and chromatin remodeling DNA methylation, histone modification, and chroma- tin remodeling may act in concert to establish and We have previously discussed connections between maintain genomic DNA methylation patterns. DNA methyltransferases and chromatin in the context of histone tail modifying proteins, namely histone DDM1 deacetylases. HDACs are associated with all active DNA methyltransferases and likely play an important The phenotypic consequences of mutations in the role in determining which regions of the genome are Arabidopsis DDM1 (decrease in DNA methylation 1) to be methylated. A significant body of data has gene provided the first evidence that chromatin accumulated showing that hypermethylated regions remodeling may be essential for proper DNA methyla- are transcriptionally inactive, enriched in hypoacety- tion patterning. DDM1 is not a DNA lated histones, and the chromatin is tightly packed, methyltransferase, but rather a member of the SNF2 reducing the access of transcription factors to these family (Jeddeloh et al., 1999). DDM1 does not readily regions (Eden et al., 1998). Work over the last few fit into one of the three SNF2 subfamilies listed in years has provided tantalizing evidence for yet another Table 2 due to its lack of motifs commonly associated connection between DNA methylation and chromatin with members of each family, but may be distantly – the possibility that DNA methylation is directly related to the ISWI-subfamily. Mutations in DDM1 connected to, or targeted by, chromatin remodeling result in loss of about 70% of the total genomic 5- machines. In this last section I will discuss the methylcytosine content, primarily at repetitive elements evidence for this linkage and propose models for like satellites and ribosomal DNA (Jeddeloh et al., how these processes may be coordinated in mamma- 1999; Martienssen and Henikoff, 1999). With increas- lian cells. ing generations of DDM1 inbreeding, DNA methylation at single copy loci, including imprinted regions, is also lost, suggesting that DDM1 may be ATP-dependent chromatin remodeling enzymes involved in maintenance methylation following DNA Helicases are a large group of proteins, involved in a replication (Jeddeloh et al., 1999; Vielle-Calzada et al., host of RNA and DNA metabolic processes, which 1999). DDM1 mutant plants exhibited defects in possess a set of seven conserved motifs involved in flowering time, floral and leaf morphology, and fertility ATP binding and hydrolysis. The helicase group can be (Jeddeloh et al., 1999). Similar effects were observed in broken down into several families. SNF2 family plants expressing antisense to the Arabidopsis DNMT1- members are involved in chromatin remodeling (ISWI), like gene MET1, except that the defects in the MET1 transcription (SNF2), DNA repair (ERCC6), and plants tended to occur after fewer generations recombination (RAD54) (Table 2). SNF2 (sucrose (Finnegan et al., 1996). Demethylation and activation non-fermenter) was first identified in yeast as a gene of transposition from transposable elements has also essential for transcription of genes involved in sucrose been observed in DDM1 mutant plants (Miura et al., fermentation and mating type switching (Eisen et al., 2001). 1995). Since then, a large number of SNF2-like genes have been discovered, most of which have homologs in ATRX organisms ranging from yeast to humans. The SNF2 family can be further divided into three subfamilies, the The ATRX gene is mutated in a human genetic disease SNF2-like subfamily, the ISWI-like subfamily, and the called X-linked, a-thalassemia mental retardation CHD subfamily, based on the presence of other (ATR-X) syndrome (Gibbons et al., 1997). Features conserved domains (Table 2; Havas et al., 2001; of this disease include developmental abnormalities, Varga-Weisz, 2001). None of the SNF2 family severe mental retardation, facial dysmorphism, and a- members appear to act as helicases, instead they utilize thalassemia. Many of the mutations in the ATRX gene the energy derived from ATP hydrolysis to disrupt occur within the PHD region, a region highly histone-DNA interactions and allow for the physical homologous to the PHD regions of DNMT3a and movement, or sliding, of nucleosomes on the DNA. In DNMT3b (Gibbons et al., 1997). Structurally, ATRX this way SNF2 proteins can reorganize the chromatin belongs to the CHD subfamily (Table 2) of SNF2-like structure of a region, to one more permissive for proteins, although it has yet to be purified from cells transcription factor binding and transcriptional activa- and shown to possess chromatin remodeling activity tion, or to one with more regularly spaced, tightly (Havas et al., 2001). ATRX localizes to pericentromeric

Oncogene DNA methylation and cancer KD Robertson 5375 Table 2 Listing of SNF2-like proteins, and their proposed functions, in each of the three SNF2 subfamilies SNF2 subfamily Protein name Proposed function(s) Reference

SNF2-like Swi2/Snf2 (S. cerevisiae) DNA-stimulated ATPase, disrupt nucleosome Whitehouse et al., 1999 spacing and stimulate factor binding Brahma (Drosophila) Nucleosome spacing, transcription co-activator Kal et al., 2000 hBRG1 (Human) DNA-dependent ATPase, remodeling Wang et al., 1996 nucleosomal arrays, octamer transfer hBRM (Human) Activities similar to BRG-1 Havas et al., 2001

ISWI-like ISWI (Drosophila) Nucleosome-dependent ATPase, generate Deuring et al., 2000; regularly-spaced nucleosomal array Ito et al., 1999 hSNF2H Component of RSF, hACF, hCHRAC complexes, LeRoy et al., 2000; DNA/histone/nucleosome stimulated ATPase, Poot et al., 2000 nucleosome spacing and remodeling hSNF2L Less well characterized, likely similar to hSNF2H

CHD CHD1 (S. cerevisiae) Nucleosome disruption activity Tran et al., 2000 CHD4 (Drosophila) Mi-2 complex, nucleosome-stimulated ATPase, Brehm et al., 2000 interacts with HDAC CHD4 (Xenopus) Similar to dCHD4, in a complex with MBD3, Wade et al., 1999 DNA methylation-mediated gene silencing? ATRX Unknown, DNA methylation remodeling? Gibbons et al., 2000

The list is by no means comprehensive and more complete descriptions can be found in Havas et al., 2001; Varga-Weisz, 2001; Vignali et al., 2000

heterochromatin and may act as a transcriptional Lsh7/7 mice demonstrated defects in cell prolifera- regulator within a chromatin environment (Berube et tion and high levels of apoptosis (Geiman and al., 2000). Interestingly, it was found that ATR-X Muegge, 2000). Given the homology between Lsh patients demonstrate DNA methylation defects in and DDM1 (50% identity in the helicase region), select regions of the genome. The ribosomal DNA genomic DNA methylation patterns were examined in repeats (where ATRX has also been shown to localize) the Lsh7/7 mice. Remarkably, the knockout mice were significantly hypomethylated. Conversely, a Y exhibited profound methylation defects (Dennis et al., chromosome-specific repeat (DYZ2) was found to be 2001). An examination of repetitive elements, including hypermethylated in ATR-X patients (Gibbons et al., the major and minor satellite repeats, intracisternal A- 2000). Therefore, unlike the effects of DDM1 muta- particle retroviral sequences, and Sine B1 repeats, all of tion, ATRX mutations result in both aberrant losses which are normally heavily methylated in cells, and gains in DNA methylation in the genome. This revealed significant hypomethylation. Several single result implies that aberrant chromatin structures, which copy loci were also examined, including the b-globin, may be the result of an improperly functioning or Pgk-2 and Pgk-1 genes, and the imprinting control improperly targeted chromatin remodeling protein, region upstream of the H19 gene, and all revealed may be able to target DNA methylation to regions losses of methylation relative to normal mice. Total that would not normally be methylated. This also genomic 5-methylcytosine levels were reduced by 50 – supports the notion that chromatin structure changes 60% and all tissues appeared to be similarly affected may precede, and potentially dictate, patterns of DNA (Dennis et al., 2001). Thus it appeared that most, but methylation. not all DNA methylation was affected by the Lsh mutation and unlike ATRX mutation, resulted only in losses of DNA methylation. No aberrant hypermethy- Lsh lation was reported. The Lsh work provides the most Lsh (lymphoid-specific helicase, Hells, PASG), the compelling evidence that chromatin structure and murine homolog of DDM1, was initially identified as chromatin remodeling are critical determinants of a protein highly expressed in lymphoid tissue and cellular DNA methylation patterns in mammalian thought to be involved in recombination (Jarvis et al., cells. 1996). Since then, its expression has been found to be less restricted, however Lsh expression appears to be tightly correlated with cell proliferation (Geiman and DNA methylation and chromatin structure – how does it Muegge, 2000; Raabe et al., 2001). Lsh knockout mice all fit together? were generated and developed relatively normally. Lsh7/7 mice were born live but died shortly there- The consistent theme arising from studies on DDM1, after, possibly due to renal failure. The Lsh knockout ATRX, and Lsh, is that these chromatin remodeling mice show a reduced birth weight, renal lesions, and enzymes, most likely acting within large macromole- reduced numbers of lymphoid cells. T cells from cular remodeling complexes, may remodel chromatin

Oncogene DNA methylation and cancer KD Robertson 5376 to allow the DNA methyltransferases access to their Neurospora, absolutely require DNA methylation for target sites, or set up a particular chromatin config- proper development. uration which is recognized by DNA How might DNA methylation and chromatin methyltransferases and/or their associated proteins. remodeling be coupled mechanistically? How are the Thus one could imagine that, (1) the chromatin HDACs also involved in this process? Unfortunately, remodeling enzyme is directly associated with a DNA the in vitro biochemical properties of the DDM1, methylase complex which can both remodel chromatin ATRX and Lsh proteins have not been investigated. and methylate DNA or (2), the chromatin remodeling The biochemical properties of other SNF2-like chro- complex may remodel a particular region destined to matin remodeling enzymes, including ones in the ISWI undergo DNA methylation, depart, then the DNA subfamily, have been investigated in considerable detail methylation machinery would follow. The work with and we may be able to gain insights into potential DDM1, ATRX, and Lsh firmly establishes an indirect mechanisms of Lsh or DDM1 using the properties of connection between DNA methylation and chromatin these related proteins as a model. In the case of remodeling but has not provided any evidence for a recombinant Drosophila ISWI, or the ISWI-containing direct connection, such as a direct interaction, between complex NURF, it has been demonstrated that the a chromatin remodeling enzyme and a DNA methyl- histone H4 tail, which protrudes from the nucleosomal transferase (or DNA methyltransferase-associated core particle, is essential for ISWI chromatin remodel- protein). Given the universal association of HDACs ing activity (Clapier et al., 2001; Corona et al., 2002; with DNMTs it does not seem unreasonable to Hamiche et al., 2001). The H3 and H4 tails are subject speculate that a similar direct association between to numerous post-translation modifications including one or more of the DNMTs and SNF2 family acetylation, methylation and phosphorylation (Jenu- members may soon come to light. wein and Allis, 2001). In particular, it appears that the A model which has been proposed previously is that base of the H4 tail, combined with nucleosome-bound proteins like DDM1 and Lsh may facilitate access of DNA, is the ‘epitope’ recognized by ISWI in the DNMT to newly replicated DNA following DNA chromatin. Interestingly, it was recently shown that replication (Martienssen and Henikoff, 1999). DDM1 acetylation of lysines (K12 and K16) near the base of and Lsh may therefore remodel newly deposited the H4 tail inhibited the chromatin remodeling ability histones in such a way as to allow DNMT1 to carry of ISWI (Clapier et al., 2002). Indirect genetic evidence out its maintenance methylation function. The hemi- presented in the previous section indicates that methylated DNA would be converted to fully chromatin remodeling is essential for proper DNA methylated DNA, and the associated HDAC2 may methylation patterns. Thus we propose a model then deacetylate newly deposited histones and allow for (Figure 6) where these three activities, histone maintenance of a heterochromatic state once the DNA deacetylase, ATPase, and DNA methyltransferase, rely replication machinery has passed (Rountree et al., on each other. In this model, histone deacetylation 2000). The strict correlation between Lsh expression would occur first. This may be the point at which and cell proliferation lends support for this idea transcriptional shutdown of the region would occur. (Raabe et al., 2001). This model is highly feasible for Access of the HDACs themselves to chromatin may methylation in the context of DNA replication require other chromatin remodeling activities (Tong et (maintenance methylation), however evidence from al., 1998). Following histone deacetylation (and other cellular processes, particularly related to de novo possibly methylation), the chromatin remodeling methylation and methylation remodeling events occur- complex binds and alters the chromatin structure or ring during embryogenesis, indicates that there may be nucleosome positioning in such a way as to directly additional mechanisms. As was mentioned previously, facilitate access of the DNMT to the nucleosomal silencing of newly introduced retroviral sequences, and DNA or create a particular ‘epitope’ or signature of the X chromosome copy destined for inactivation, recognized by a DNMT or DNMT-associated occurs well before de novo DNA methylation of those complex. The region is then methylated which locks sequences. A similar situation may also be occurring the chromatin in a silent mode, which would then be during the rapid genome-wide methylation remodeling further enhanced by the recruitment of methyl-CpG events (both demethylation and de novo methylation) binding proteins and their associated repressive activ- that occur during embryogenesis (Reik et al., 2001). ities (Bird, 2002). There is ample evidence for DNMTs Therefore chromatin remodeling and histone modifica- and HDACs as components of one complex. The ATP- tion would seem to set the stage for DNA methylation dependent chromatin remodeling enzyme may also be in some cases. Additional support for this notion part of the complex or act separately. Although comes from recent studies in Neurospora and Arabi- speculation at this time, this model is testable using dopsis, which showed that loss of histone methylation chromatin reconstituted in vitro and recombinant (on histone H3, lysine 9) resulted in a complete or remodeling proteins and will likely be the subject of partial loss of genomic DNA methylation, respectively future studies. Previously proposed models, where (Tamaru and Selker, 2001; Jackson et al., 2002). DNA methylation occurs first and is followed by Homologous HMTs exist in mammalian cells (Rea et histone deacetylation and gene silencing (Baylin et al., al., 2000), but it remains unclear whether a similar 2001), may also operate and are by no means excluded control system exists in mammals, which, unlike by the model proposed here.

Oncogene DNA methylation and cancer KD Robertson 5377 Concluding remarks (4) do each of the three catalytically active DNMTs recognize the same chromatin ‘epitope’ for methylation The past decade has seen amazing advances in our or do the de novo and maintenance methyltransferases understanding of the ways in which DNA methylation utilize fundamentally different signals, and finally; (5) contributes to transcriptional regulation and tumor- have we indeed accounted for all of the DNA igenesis. Findings from the last few years in particular methylating activities in mammalian cells? Such have revealed the first glimpses of how the myriad important questions will likely be answered in the next epigenetic control mechanisms that exist in mammalian few years as intensive research identifies and char- cells, including DNA methylation, histone acetylation, acterizes DNMT-associated proteins, and in vitro DNA histone methylation, and ATP-dependent nucleosome methylation and chromatin remodeling systems are positioning, may be directly connected. Histone established. Such findings will likely be highly relevant acetylation and methylation patterns may recruit to diseases involving aberrant DNA methylation certain chromatin remodeling activities, which in turn patterns, including cancer, as well as ICF, Rett, and may create docking sites for DNA methyltransferase ATRX syndromes, and may provide the understanding complexes. The CpG methylated region may recruit to devise completely novel strategies for reversing the further chromatin modulatory activities, such as the defects. Thus, our blurred image of the ‘tangled web’ methyl-CpG binding protein complexes and their of DNA methylation, histone modifications, and associated repressive activities, and finally lock a given chromatin remodeling is gradually resolving into that region of the genome in a silent state. Emerging of a precisely patterned, logically woven fabric. evidence strongly suggests that histone modifications set the stage for DNA methylation. Issues that still need to be resolved include the following: (1) which ATP-dependent chromatin remodeling machines are Acknowledgments KD Robertson is a Cancer Scholar supported by the involved in establishing and maintaining DNA methy- National Cancer Institute (NIH grant CA84535-01). I lation patterns; (2) is the association between the thank Andrea Kahler Robertson for critical reading of ATPases and the DNMTs direct or indirect; (3) does the manuscript. This article is dedicated to my former the nature of the interaction change between undiffer- mentor and friend, Alan Wolffe, who died tragically last entiated, differentiated, and transformed cellular states; year.

References

Aapola U, Lyle R, Krohn K, Antonarakis SE and Peterson Bourc’his D, Xu G-L, Lin C-S, Bollman B and Bestor TH. P. (2001). Cytogenet. Cell Genet., 92, 122 – 126. (2001). Science, 294, 2536 – 2539. Aapola U, Shibuya K, Scott HS, Ollila J, Vihinen M, Heino Brehm A, Langst G, Kehle J, Clapier CR, Imhof A, M, Shintani A, Kawasaki K, Minoshima S, Krohn K, Eberharter A, Muller J and Becker PB. (2000). EMBO Antonarakis SE, Shimizu N, Kudoh J and Peterson P. J., 19, 4332 – 4341. (2000). Genomics, 65, 293 – 298. Brehm AB, Miska EA, McCance DJ, Reid JL, Bannister AJ Aguirre-Arteta AM, Grunewald I, Cardoso MC and and Kouzarides T. (1998). Nature, 391, 597 – 601. Leonhardt H. (2000). Cell Growth Differ., 11, 551 – 559. Cameron EE, Bachman KE, Myohanen S, Herman JG and Aoki A, Suetake I, Miyagawa J, Fujio T, Chijiwa T, Sasaki H Baylin SB. (1999). Nature Genet., 21, 103 – 107. and Tajima S. (2001). Nucleic Acids Res., 29, 3506 – 3512. Cardoso MC and Leonhardt H. (1999). J. Cell Biol., 147, Aoki K, Meng G, Suzuki K, Takashi T, Kameoka Y, 25 – 32. Nakahara K, Ishida R and Kasai M. (1998). J. Biol. Carlson LL, Page AW and Bestor TH. (1992). Genes. Dev., 6, Chem., 273, 26698 – 26704. 2536 – 2541. Bachman KE, Rountree MR and Baylin SB. (2001). J. Biol. Clapier CR, Langst G, Corona DFV, Becker PB and Chem., 276, 32282 – 32287. Nightingale KP. (2001). Mol. Cell. Biol., 21, 875 – 883. Bacolla A, Pradhan S, Roberts RJ and Wells RD. (1999). J. Clapier CR, Nightingale KP and Becker PB. (2002). Nucleic Biol. Chem., 274, 33011 – 33019. Acids Res., 30, 649 – 655. Baylin SB, Esteller M, Rountree MR, Bachman KE, Corona DFV, Clapier CR, Becker PB and Tamkun JW. Schuebel K and Herman JG. (2001). Hum. Mol. Genet., (2002). EMBO Rep., 3, 242 – 247. 10, 687 – 692. Csankovszki G, Nagy A and Jaenisch R. (2001). J. Cell Biol., Beard C, Li E and Jaenisch R. (1995). Genes. Dev., 9, 2325 – 153, 773 – 783. 2334. Dennis K, Fan T, Geiman T, Yan Q and Muegge K. (2001). Berube NG, Smeenk CA and Picketts DJ. (2000). Hum. Mol. Genes. Dev., 15, 2940 – 2944. Genet., 9, 539 – 547. Deuring R, Fanti L, Armstrong JA, Sarte M, Papoulas O, Bestor T, Laudano A, Mattaliano R and Ingram V. (1988). J. Prestel M, Daubresse G, Verardo M, Moseley SL, Berloco Mol. Biol., 203, 971 – 983. M, Tsukiyama T, Wu C, Pimpinelli S and Tamkun JW. Bestor TH. (2000). Hum. Mol. Genet., 9, 2395 – 2402. (2000). Cell, 5, 355 – 365. Bird A. (2001). Science, 294, 2113 – 2115. Di Croce L, Raker AA, Corsaro M, Fazi F, Fanelli M, Bird A. (2002). Genes. Dev., 16, 6 – 21. Faretta M, Fuks F, Coco FL, Kouzarides T, Nervi C, Bird AP and Wolffe AP. (1999). Cell, 99, 451 – 454. Minucci S and Pelicci PG. (2002). Science, 295, 1079 – Bonfils C, Beaulieu N, Chan E, Cotton-Montpetit J and 1082. MacLeod AR. (2000). J. Biol. Chem., 275, 10754 – 10760.

Oncogene DNA methylation and cancer KD Robertson 5378 Doherty AS, Bartolomei MS and Schultz RM. (2002). Dev. Hung M-S, Karthikeyan N, Huang B, Koo H-C, Kiger J and Biol., 242, 255 – 266. Shen C-KJ. (1999). Proc. Natl. Acad. Sci. USA, 96, Dong A, Yoder JA, Zhang X, Zhou L, Bestor TH and Cheng 11940 – 11945. X. (2001). Nucleic Acids Res., 29, 439 – 448. ItoT,LevensteinME,FyodorovDV,KutachAK, Dyson N. (1998). Genes. Dev., 12, 2245 – 2262. Kobayashi R and Kadonaga JT. (1999). Genes. Dev., 13, Eden S, Hashimshony T, Keshet I, Cedar H and Thorne AW. 1529 – 1539. (1998). Nature, 394, 842. Jackson JP, Lindroth AM, Cao X and Jacobsen SE. (2002). Ehrlich M, Buchanan KL, Tsein F, Jiang G, Sun B, Uicker Nature, advance online publication, 17 March.. W, Weemaes CMR, Smeets D, Sperling K, Belohradsky Jackson-Grusby L, Beard C, Possemato R, Tudor M, BH,TommerupN,MisekDE,RouillardJ-M,KuickR FambroughD,CsankovzkiG,DausmanJ,LeeP,Wilson and Hanash SM. (2001). Hum. Mol. Genet., 10, 2917 – C, Lander E and Jaenisch R. (2001). Nature Genet., 27, 2931. 31 – 39. Eisen JA, Sweder KS and Hanawalt PC. (1995). Nucleic Jarvis CD, Geiman T, Vila-Storm MP, Osipovich O, Akella Acids Res., 23, 2715 – 2723. U, Candeias S, Nathan I, Durum SK and Muegge K. Ellis NA. (1997). Curr.Opin.Genet.Dev.,7, 354 – 363. (1996). Gene, 169, 203 – 207. Fan G, Beard C, Chen RZ, Csankovszki G, Sun Y, Siniaia Jeanpierre M, Turleau C, Aurias A, Prieur M, Ledeist F, M, Biniszkiewicz D, Bates B, Lee PP, Kuhn R, Trumpp A, Fischer A and Viegas-Pequignot E. (1993). Hum. Mol. Poon C-S, Wilson CB and Jaenisch R. (2001). J. Neurosci., Genet., 2, 731 – 735. 21, 788 – 797. Jeddeloh JA, Stokes TL and Richards EJ. (1999). Nature Feng Q and Zhang Y. (2001). Genes. Dev., 15, 827 – 832. Genet., 22, 94 – 97. Finnegan EJ, Peacock WJ and Dennis ES. (1996). Proc. Natl. Jenuwein T and Allis CD. (2001). Science, 293, 1074 – 1080. Acad. Sci. USA, 93, 8449 – 8454. Jones PA and Laird PW. (1999). Nature Genet., 21, 163 – 166. Flynn J, Glickman JF and Reich NO. (1996). Biochemistry, Kal AJ, Mahmoudi T, Zak NB and Verrijzer CP. (2000). 35, 7308 – 7315. Genes Dev., 14, 1058 – 1071. Franceschini P, Martino S, Ciocchini M, Ciuti E, Vardeu Kennedy BK, Barbie DA, Classon M, Dyson N and Harlow MP, Guala A, Signorile F, Camerano P, Franceschini D E. (2000). Genes Dev., 14, 2855 – 2868. and Tovo PA. (1995). Eur. J. Pediatr., 154, 840 – 846. Kondo T, Bobek MP, Kuick R, Lamb B, Zhu X, Narayan A, Fuks F, Bergers WA, Brehm A, Hughes-Davies L and Bourc’his D, Viegas-Pequignot E, Ehrlich M and Hanash Kouzarides T. (2000). Nature Genet., 24, 88 – 91. SM. (2000). Hum. Mol. Genet., 9, 597 – 604. Fuks F, Burgers WA, Godin N, Kasai M and Kouzarides T. LeiH,OhSP,OkanoM,JuttermannR,GossKA,Jaenisch (2001). EMBO J., 20, 2536 – 2544. R and Li E. (1996). Development, 122, 3195 – 3205. Geiman TM and Muegge K. (2000). Proc. Natl. Acad. Sci. Leonhardt H, Page AW, Weier H and Bestor TH. (1992). USA, 97, 4772 – 4777. Cell, 71, 865 – 873. Gibbons RJ, Bachoo S, Picketts DJ, Aftimos S, Asenbauer LeRoy G, Loyola A, Lane WS and Reinberg D. (2000). J. B,BergoffenJ,BerrySA,DahlN,FryerA,KepplerK, Biol. Chem., 275, 14787 – 14790. Kurosawa K, Levin ML, Masuno M, Neri G, Pierpont Li E, Beard C and Jaenisch R. (1993). Nature, 366, 362 – 365. ME, Slaney SF and Higgs DR. (1997). Nature Genet,, 17, Li E, Bestor TH and Jaenisch R. (1992). Cell, 69, 915 – 926. 146 – 148. Li H, Leo C, Zhu J, Wu X, O’Neil J, Park E-J and Chen JD. GibbonsRJ,McDowellTL,RamanS,O’RourkeDM, (2000). Mol. Cell. Biol., 20, 1784 – 1796. Garrick D, Ayyub H and Higgs DR. (2000). Nature Genet., Lin M-J, Lee T-L, Hsu D-W and Shen C-KJ. (2000). FEBS 24, 368 – 371. Lett., 469, 101 – 104. Glickman JF, Pavlovich JG and Reich NO. (1997). J. Biol. Liu Y, Oakeley EJ, Sun L and Jost J-P. (1998). Nucleic Acids Chem., 272, 17851 – 17857. Res., 26, 1038 – 1045. Gowher H and Jeltsch A. (2001). J. Mol. Biol., 309, 1201 – Luo RX, Postigo AA and Dean DC. (1998). Cell, 92, 463 – 1208. 473. Hamiche A, Kang J-G, Dennis C, Xiao H and Wu C. (2001). Lyko F, Ramsahoye BH, Kashevsky H, Tudor M, Proc. Natl. Acad. Sci. USA, 98, 14316 – 14321. Mastrangelo M-A, Orr-Weaver TL and Jaenisch R. Hansen RS, Stoger R, Wijmenga C, Stanek AM, Canfield (1999). Nature Genet., 23, 363 – 366. TK, Luo P, Matarazzo MR, D’Esposito M, Feil R, Gimelli Lyko F, Whittaker AJ, Orr-Weaver TL and Jaenisch R. G, Weemaes CMR, Laird CD and Gartler SM. (2000). (2000). Mech. Dev., 95, 215 – 217. Hum. Mol. Genet., 9, 2575 – 2587. Martienssen R and Henikoff S. (1999). Nature Genet., 22, 6– Hansen RS, Wijmenga C, Luo P, Stanek AM, Canfield TK, 7. Weemaes CMR and Gartler SM. (1999). Proc. Natl. Acad. Meng G, Inazawa J, Ishida R, Tokura K, Nakahara K, Aoki Sci. USA, 96, 14412 – 14417. K and Kasai M. (2000). Gene, 242, 59 – 64. Hassan KMA, Norwood T, Gimelli G, Gartler SM and MertineitC,YoderJA,TaketoT,LairdDW,TraslerJMand Hansen RS. (2001). Hum. Mol. Genet., 109, 452 – 462. Bestor TH. (1998). Development, 125, 889 – 897. Havas K, Whitehouse I and Owen-Hughes T. (2001). Cell. Michaelson JS, Bader D, Kuo F, Kozak C and Leder P. Mol. Life Sci., 58, 673 – 682. (1999). Genes. Dev., 13, 1918 – 1923. Heard E, Rougeulle C, Arnaud D, Avner P, Allis CD and Miniou P, Jeanpierre M, Bourc’his D, Barbosa ACC, Spector DL. (2001). Cell, 107, 727 – 738. Blanquet V and Viegas-Pequignot E. (1997). Hum. Genet., Howell CY, Bestor TH, Ding F, Latham KE, Mertineit C, 99, 738 – 745. Trasler JM, Trasler JM and Chaillet JR. (2001). Cell, 104, Miura A, Yonebayashi S, Watanabe K, Toyama T, Shimada 829 – 838. H and Kakutani T. (2001). Nature, 411, 212 – 214. Hsieh C-L. (1999). Mol. Cell. Biol., 19, 8211 – 8218. Muller S, Hoege C, Pyrowolakis G and Jentsch S. (2001). Hsu D-W, Lin M-J, Lee T-L, Wen S-C, Chen X and Shen C- Nature Rev. Mol. Cell Biol., 2, 202 – 210. KJ. (1999). Proc.Natl.Acad.Sci.USA,96, 9751 – 9756.

Oncogene DNA methylation and cancer KD Robertson 5379 Nielsen SJ, Schneider R, Bauer U-M, Bannister AJ, Tagarro I, Fernandez-Peralta AM and Gonzales-Aguilera Morrison A, O’Carroll D, Firestein R, Cleary M, Jenuwein JJ. (1994). Hum. Genet., 93, 383 – 388. T, Herrera RE and Kouzarides T. (2001). Nature, 412, Tamaru H and Selker EU. (2001). Nature, 414, 277 – 283. 561 – 565. Tatematsu K-i, Yamazaki T and Ishikawa F. (2000). Genes Okano M, Bell DW, Haber DA and Li E. (1999). Cell, 99, Cells, 5, 677 – 688. 247 – 257. The Arabidopsis Genome Initiative. (2000). Nature, 408, Okano M, Xie S and Li E. (1998a). Nature Genet., 19, 219 – 796 – 815. 220. Tong JK, Hassig CA, Schnitzler GR, Kingston RE and Okano M, Xie S and Li E. (1998b). Nucleic Acids Res., 26, Schreiber SL. (1998). Nature, 395, 917 – 921. 2536 – 2540. Tran HG, Steger DJ, Iyer VR and Johnson AD. (2000). Pannell D, Osborne CS, Yao S, Sukonnik T, Pasceri P, EMBO J., 19, 2323 – 2331. Karaiskakis A, Okano M, Li E, Lipshitz HD and Ellis J. Tucker KL. (2001). Neuron, 30, 649 – 652. (2000). EMBO J., 19, 5884 – 5894. Tuck-Muller CM, Narayan A, Tsien F, Smeets DFCM, Poot RA, Dellaire G, Hulsmann BB, Grimaldi MA, Corona Sawyer J, Fiala ES, Sohn OS and Ehrlich M. (2000). DFV, Becker PB, Bickmore WA and Varga-Weisz PD. Cytogenet. Cell Genet., 89, 121 – 128. (2000). EMBO J., 19, 3377 – 3387. Varga-Weisz P. (2001). Oncogene, 20, 3076 – 3085. Pradhan S, Bacolla A, Wells RD and Roberts RJ. (1999). J. Vielle-Calzada J-P, Thomas J, Spillane C, Coluccio A, Biol. Chem., 274, 33002 – 33010. Hoeppner MA and Grossniklaus U. (1999). Genes Dev., Pradhan S and Kim G-D. (2002). EMBO J., 21, 779 – 788. 13, 2971 – 2982. PradhanS,TalbotD,ShaM,BennerJ,HornstraL,LiE, Vignali M, Hassan AH, Neely KE and Workman JL. (2000). Jaenisch R and Roberts RJ. (1997). Nucleic Acids Res., 25, Mol. Cell. Biol., 20, 1899 – 1910. 4666 – 4673. Wade PA, Gegonne A, Jones PL, Ballestar E, Aubry F and Qiu C, Sawada K, Zhang X and Cheng X. (2002). Nature Wolffe AP. (1999). Nature Genet., 23, 62 – 66. Struct. Biol., 9, 217 – 224. Walsh CP, Chaillet JR and Bestor TH. (1998). Nature Genet., Raabe EH, Abdurrahman L, Behbehani G and Areci RJ. 20, 116 – 117. (2001). Dev. Dyn., 221, 92 – 105. Wang W, Cote J, Xue Y, Zhou S, Khavari PA, Biggar SR, ReaS,EisenhaberF,O’CarollD,StrahlBD,SunZ-W, Muchardt C, Kalpana GV, Goff SP, Yaniv M, Workman Schmid M, Opravil S, Mechtler K, Ponting CP, Allis CD JL and Crabtree GR. (1996). EMBO J., 15, 5370 – 5382. and Jenuwein T. (2000). Nature, 406, 593 – 599. Whitehouse I, Flaus A, Cairns BR, White MF, Workman JL Reik W, Dean W and Walter J. (2001). Science, 293, 1089 – and Owen-Hughes T. (1999). Nature, 400, 784 – 787. 1093. Wilkinson CRM, Bartlett R, Nurse P and Bird AP. (1995). Rhee I, Jair K-W, Yen R-WC, Lengauer C, Herman JG, Nucleic Acids Res., 23, 1995. Kinzler KW, Vogelstein B, Baylin SB and Schuebel KE. Wolffe AP and Pruss D. (1996). Cell, 84, 817 – 819. (2000). Nature, 404, 1003 – 1007. XieS,WangZ,OkanoM,NogamiM,LiY,HeW-W, Robertson KD. (2001). Oncogene, 20, 3139 – 3155. Okumura K and Li E. (1999). Gene, 236, 87 – 95. Robertson KD, Ait-Si-Ali S, Yokochi T, Wade PA, Jones PL Xu G-L, Bestor TH, Bourc’his D, Hsieh C-L, Tommerup N, and Wolffe AP. (2000a). Nature Genet., 25, 338 – 342. Bugge M, Hulten M, Qu Russo JJ and Viegas-Pequignot E. Robertson KD, Keyomarsi K, Gonzales FA, Velicescu M (1999). Nature, 402, 187 – 191. and Jones PA. (2000b). Nucleic Acids Res., 28, 2108 – 2113. Yen R-WC, Vertino PM, Nelkin BD, Yu JJ, El-Deiry W, Robertson KD, Uzvolgyi E, Liang G, Talmadge C, Sumegi J, Cumaraswamy A, Lennon GG, Trask BJ, Celano P and Gonzales FA and Jones PA. (1999). Nucleic Acids Res., 27, Baylin SB. (1992). Nucleic Acids Res., 20, 2287 – 2291. 2291 – 2298. Yoder JA and Bestor TH. (1998). Hum. Mol. Genet., 7, 279 – Robertson KD and Wolffe AP. (2000). Nature Rev. Genet., 1, 284. 11 – 19. Yoder JA, Walsh CP and Bestor TH. (1997). Trends Genet., Rountree MR, Bachman KE and Baylin SB. (2000). Nature 13, 335 – 340. Genet., 25, 269 – 277. Yokochi T and Robertson KD. (2002). J. Biol. Chem., 277, Ryan RF, Schultz DC, Ayyanathan K, Singh PB, Friedman 11735 – 11745. JR, Fredericks WJ and Rasucher FR. (1999). Mol. Cell. Zhong S, Salomoni P and Pandolfi PP. (2000a). Nature Cell Biol., 19, 4366 – 4378. Biol., 2, E85 – E90. Schuffenhauer S, Bartsch O, Stumm M, Buchholz T, ZhongS,SalomoniP,RonchettiS,GuoA,RuggeroDand Petropoulou T, Kraft S, Belohradsky B, Hinkel GK, Pandolfi PP. (2000b). J. Exp. Med., 191, 631 – 639. Meitinger T and Wegner R-D. (1995). Hum. Genet., 96, 562 – 571. Smeets DFCM, Moog U, Weemaes CMR, Vaes-Peeters G, Merkx GFM, Niehof JP and Hamers G. (1994). Hum. Genet., 94, 240 – 246.

Oncogene