Oncogene (2001) 20, 3139 ± 3155 ã 2001 Nature Publishing Group All rights reserved 0950 ± 9232/01 $15.00 www.nature.com/onc

DNA methylation, methyltransferases, and cancer

Keith D Robertson*,1

1Epigenetic Gene Regulation and Cancer Section, National Cancer Institute, NIH, Bldg. 41, 41 Library Dr., Bethesda, Maryland, MD 20892, USA

The ®eld of epigenetics has recently moved to the repression, chromatin structure modulation, X chro- forefront of studies relating to diverse processes such as mosome inactivation, genomic imprinting, and the transcriptional regulation, chromatin structure, genome suppression of the detrimental e€ects of repetitive and integrity, and tumorigenesis. Recent work has revealed parasitic DNA sequences on genome integrity (Baylin how DNA methylation and chromatin structure are and Herman, 2000; Jones and Laird, 1999; Robertson linked at the molecular level and how methylation and Wol€e, 2000). While DNA methylation clearly anomalies play a direct causal role in tumorigenesis enhances the ability of cells to regulate and package and genetic disease. Much new information has also the genetic information it also adds an additional come to light regarding the cellular methylation burden. Genomic methylation patterns are frequently machinery, known as the DNA methyltransferases, in altered in tumor cells with global hypomethylation terms of their roles in mammalian development and the accompanying region-speci®c hypermethylation events. types of proteins they are known to interact with. This When hypermethylation events occur within the information has forced a new view for the role of DNA promoter of a tumor suppressor gene this can silence methyltransferases. Rather than enzymes that act in expression of the associated gene and provide the cell isolation to copy methylation patterns after replication, with a growth advantage in a manner akin to deletions the types of interactions discovered thus far indicate that or mutations. Recent work has revealed that DNA DNA methyltransferases may be components of larger methylation is an important player in many processes complexes actively involved in transcriptional control and including DNA repair, genome stability, and chroma- chromatin structure modulation. These new ®ndings will tin structure. This review will ®rst discuss what is likely enhance our understanding of the myriad roles of known about the enzymes that catalyze the methyl- DNA methylation in disease as well as point the way to transfer reaction, the DNA methyltransferases or novel therapies to prevent or repair these defects. DNMT's. I will then discuss the role of methylation Oncogene (2001) 20, 3139 ± 3155. in genetic disease and cancer in terms of the types of methylation defects, the functional consequences, and Keywords: DNA methylation; cancer; DNA methyl- models for how these defects arise. Lastly I will transferase; ICF syndrome summarize what is known about proteins which interact with one of the best studied DNA methyl- transferases and why these interactions may be important for understanding why methylation patterns Introduction go awry in tumors.

It is now clear that the genome contains information in two forms, genetic and epigenetic. The genetic The DNA methylation machinery information provides the blueprint for the manufac- ture of all the proteins necessary to create a living Global cytosine methylation patterns in mammals thing while the epigenetic information provides appear to be established by a complex interplay of at instructions on how, where, and when the genetic least three independently encoded DNA methyltrans- information should be used. Ensuring that genes are ferases (DNMT's): DNMT1, DNMT3A, and turned on at the proper time is as important as DNMT3B (Figure 1). A fourth DNA methyltransfer- ensuring that they are turned o€ when not needed. ase, DNMT2 (Figure 1), has been cloned and The major form of epigenetic information in mamma- characterized but catalytic activity of this enzyme has lian cells is DNA methylation, or the covalent addition yet to be demonstrated in vitro or in vivo (Okano et al., of a methyl group to the 5-position of cytosine 1998b; Yoder and Bestor, 1998). The reaction mechan- predominantly within the CpG dinucleotide. DNA ism of 5-methylcytosine DNA methyltransferases is methylation has profound e€ects on the mammalian rather unusual. The target cytosine is extruded from genome. Some of these e€ects include transcriptional the double helix into the active site cleft of the enzyme where it can be reacted upon by a conserved active site cysteine (Klimasauskas et al., 1994). DNMT1 was the *Correspondence: KD Robertson ®rst methyltransferase to be discovered (Bestor et al., DNA methylation, methyltransferases, and cancer KD Robertson 3140

Figure 1 Structure of the known DNA methyltransferases (DNMT's) and DNMT-like proteins. DNMT1, 3A, and 3B can be divided into two domains, regulatory and catalytic. Conserved motifs (roman numerals) involved in catalysis are indicated with black boxes. Other structural features such as the replication foci targeting domain and zinc binding region of DNMT1 or the cysteine-rich PHD (plant homeodomain (Aasland et al., 1995)) region of DNMT3A/3B are also indicated

1988) while the DNMT3 family was only recently prior to the 8 somite stage and a nearly 70% reduction discovered and characterized. in their genomic 5-methylcytosine content (Li et al., 1992). Recently the DNMT1 gene was inactivated by somatic cell knockout methods in a colon adenocarci- DNMT1 noma cell line. Remarkably, these cells retained DNMT1 has been shown to have a 10 ± 40-fold approximately 80% of their normal methylation levels preference for hemimethylated DNA (Pradhan et al., and lacked any profound growth aberrations indicating 1999; 1997) and is the most abundant methyltransfer- that the DNMT3 family of enzymes (next section) may ase in somatic cells (Robertson et al., 1999). While the be able to act as maintenance enzymes in certain enzyme was cloned 12 years ago (Bestor et al., 1988) its situations or that there are additional, as yet sequence remained incomplete for many years (Yoder undiscovered, methyltransferases that can compensate et al., 1996) which complicated studies of the DNMT1 for the loss of DNMT1 (Rhee et al., 2000). promoter (Rouleau et al., 1992) as well as interpreta- tion of the e€ect of its overexpression on cell growth The DNMT3 family (Tucker et al., 1996; Vertino et al., 1996). DNMT1 localizes to replication foci via several independent In 1998, Okano et al. (1998a) reported the cloning and domains (Leonhardt et al., 1992; Liu et al., 1998), and initial characterization of the DNMT3 family of interacts with the proliferating cell nuclear antigen methyltransferases (Figure 1). The mouse and human (PCNA) (Chuang et al., 1997). This set of features is enzymes are highly conserved (approximately 95% why DNMT1 is often referred to as the `maintenance' identical at the amino acid level) (Xie et al., 1999) methyltransferase since it is believed to be the primary and homologous genes have been identi®ed in zebra- enzyme responsible for copying methylation patterns ®sh, Arabidopsis thaliana, and maize (Cao et al., 2000; after DNA replication. The generation of DNMT1- Okano et al., 1998a). Mouse knockouts of the Dnmt3 knockout mice has revealed that DNMT1 is required family enzymes have revealed that they are required for for proper embryonic development, imprinting, and X- the wave of de novo methylation that occurs in the inactivation (Beard et al., 1995; Li et al., 1992; 1993). genome following embryo implantation as well as de Mice de®cient for DNMT1 show arrested development novo methylation of newly integrated retroviral

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3141 sequences in murine embryonic stem (ES) cells (Okano al., 1996). This result was the motivating factor for a et al., 1999). Dnmt3a knockout mice are born live but continued search for new DNMT's that eventually led die at about 4 weeks of age in contrast to the Dnmt3b to the discovery of the DNMT3 family (Okano et al., knockout mice which are not viable (Okano et al., 1998a). Additional support for this classi®cation 1999). Dnmt3b7/7 mutant embryos show numerous scheme comes from the previously mentioned study developmental defects and growth impairment after E utilizing Drosophila expressing Dnmt1 and/or Dnmt3a. 9.5, which is close to the time the Dnmt1 knockout In this study, Dnmt1 exhibited no de novo methylation mice begin to exhibit growth defects (Li et al., 1992; activity while Dnmt3a expression resulted in low level Okano et al., 1999). Data from the Dnmt3 knockout methylation (Lyko et al., 1999). It could be argued mice, in addition to limited in vitro data indicating that however, that Drosophila has lost a trans-acting factor the DNMT3 enzymes have an equal preference for necessary for the in vivo activity of Dnmt1 when this hemimethylated and unmethylated DNA substrates, lineage lost the ability to methylate its genome has led to them being classi®ed as `de novo' (although a recent report indicates that Drosophila methyltransferases (Okano et al., 1998a). A speci®c does have low-level methylation ± primarily at non- function for Dnmt3b, revealed by the mouse knockout CpG sequences (Lyko et al., 2000)). Further support experiments, is maintenance of DNA methylation of for the de novo/maintenance scheme comes from the minor satellite repeats adjacent to centromeres. knockout experiments which showed that homozygous Similar supportive data for this role comes from three deletion of Dnmt3a and Dnmt3b did not alter pre- papers in 1999 describing mutations in the catalytic existing methylation patterns in ES cells (Okano et al., domain of the human DNMT3B gene in patients with 1999), whereas homozygous deletion of Dnmt1 resulted ICF syndrome (immunode®ciency, centromeric in- in a *70% reduction in 5-methylcytosine content (Li stability, facial anomalies) (Hansen et al., 1999; Okano et al., 1992). et al., 1999; Xu et al., 1999). As will be discussed Evidence against the de novo versus maintenance further in the next section, individuals with this disease classi®cation comes from several experiments. In the exhibit profound losses of DNA methylation from ®rst, it was demonstrated that enforced overexpression satellite 2 and 3 sequences adjacent to the centromeres of DNMT1 in cancer cell lines leads to de novo of chromosomes 1, 9 and 16 resulting in massive methylation of endogenous CpG islands (Vertino et al., instability of these chromosomes (Ji et al., 1997). A 1996). In addition, Rhee et al. (2000) recently reported speci®c function for Dnmt3a was not detectable with that somatic cells lacking DNMT1 retain approxi- the knockout model but studies using the various mately 80% of their normal methylation levels and Dnmt-knockout ES cells in addition to transgenic that the expression levels of DNMT3A and 3B were Drosophila melanogaster expressing Dnmt3a revealed not profoundly altered. It will be of great interest to that this enzyme may be specialized to methylate non- determine how these cells are maintaining so much of CpG sequences like CpA, and CpT although the their DNA methylation in the absence of DNMT1 and function of non-CpG methylation in ES cells is may be good evidence for the existence of additional unknown (Ramsahoye et al., 2000). The sequence DNA methyltransferases that can compensate for the speci®city of Dnmt3b remains to be determined since loss of DNMT1. It is probable that all three DNMT's it was not analysed in the previously mentioned study. possess both de novo and maintenance functions in vivo The catalytic activity of both Dnmt3a and 3b appears and that speci®c methyltransferases will be responsible to be quite low. In vitro assays utilizing recombinant for the methylation of certain genomic regions via their proteins expressed in baculovirus indicated the activity interaction with other nuclear proteins or DNA- of the Dnmt3's was roughly 20 times lower than the binding factors, a notion strongly supported by the activity of Dnmt1 (Dnmt3a was slightly more active ®nding that mutations in DNMT3B in patients with than Dnmt3b) (Okano et al., 1998a). In vivo over- ICF syndrome lead to loss of methylation in very expression studies showed that both enzymes could discreet locations within the genome. methylate a stably maintained episome to di€ering extents (Dnmt3a4Dnmt3b) and there appeared to be Methyltransferase motifs in proteins that do not some sequence preference as well (Hsieh, 1999). It may methylate DNA be that the in vitro assays that have been developed and work so well for DNMT1 are not optimal for the An increasing number of proteins are being identi®ed DNMT3 family or that they require small-molecule or which contain all or a subset of the conserved DNA protein co-factors for ecient catalytic activity. methyltransferase motifs (Kumar et al., 1994) yet do not appear to be involved in methylating DNA (Figure 1). The ®rst example of this was DNMT2, a protein Methyltransferases ± maintenance versus de novo homologous to the yeast pmt1 gene. This protein has DNA methyltransferases are commonly classi®ed as de all of the conserved methyltransferase motifs but novo (DNMT3) or maintenance (DNMT1) and there is exhibits no methyltransferase activity in vitro and evidence both supporting and refuting this classi®ca- transgenic mice with a targeted mutation of the tion scheme. Evidence supporting this scheme origin- putative Dnmt2 catalytic site showed no defects in ally came from the observation that Dnmt1 knockout cellular methylation patterns (Okano et al., 1998b). ES cells retained de novo methylating activity (Lei et More recently, proteins with a subset of the most

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3142 conserved methyltransferase motifs have been identi®ed ingly chromosomes 1, 9, and 16 contain large blocks of and include double-stranded RNA (dsRNA) adenosine classical satellite long tandem repeat arrays (satellites 2 deaminase, which contains regions with homology to and 3) adjacent to the centromere (Tagarro et al., motifs IV, VI, and VIII (Hough and Bass, 1997), and 1994). These regions are normally heavily methylated DNMT3L (Figure 1) which contains regions with in somatic cells but ICF patients demonstrate a marked homology to motifs I, IV, and VI (Aapola et al., hypomethylation of these regions (Jeanpierre et al., 2000). DNMT3L is most homologous to the DNMT3 1993; Tagarro et al., 1994), indicating that DNA family and also includes the PHD domain in the N- methylation may be essential for proper centromere terminal region of DNMT3A and 3B. This protein is structure and stability. Repetitive elements elsewhere in almost certainly non-functional since the conserved the genome, including a subtelomeric repeat (Kondo et `PC' active site is `mutated' to `PL'. Since DNMT3L al., 2000), and single copy sequences on the inactive X retains the PHD domain but lacks a functional chromosome (Hansen et al., 2000; Miniou et al., 1994) catalytic site it may act to antagonize the activities of have also been shown to undergo hypomethylation in DNMT3A and 3B by competing for the binding of ICF cells. In contrast to the drastic hypomethylation of interacting proteins or occupying the preferred DNA satellite 2 and 3 sequences, overall genomic 5- binding sites of DNMT3A and 3B. In the case of ds methylcytosine levels in ICF versus normal lympho- RNA adenosine deaminase, an enzyme believed to be blastoid cell lines (LCLs) was unchanged (a result involved in RNA editing, the `PC' active site is present likely in¯uenced by cell culture), and primary ICF but the lack of many of the other motifs important for brain tissue showed only a 7% decrease in 5- catalysis of the methyltransferase reaction make it methylcytosine content (Tuck-Muller et al., 2000). unlikely to be a functional methyltransferase (Hough Three studies in 1999 revealed that ICF syndrome and Bass, 1997; Kumar et al., 1994). While it cannot was associated with mutations in the DNMT3B gene yet be ruled out that DNMT2 is a functional (Hansen et al., 1999; Okano et al., 1999; Xu et al., methyltransferase with a highly speci®c recognition 1999) located on 20q11.2 (Robertson et al., 1999; Xie sequence or a preference for cytosine in non-CpG et al., 1999), the previously mapped ICF susceptibility sequences, proteins not involved in DNA methylation locus (Wijmenga et al., 1998). A number of di€erent may have utilized these motifs for DNA binding, mutations have been identi®ed in DNMT3B and are mismatch recognition, or the catalysis of other summarized in Figure 2. Most of the mutations are nucleotide modifying or repair activities in which the heterozygous and, with one exception, a€ect the unique ability to extrude the base from the double catalytic domain of DNMT3B. The e€ect of one helix with minimal distortion of the surrounding DNA naturally occurring mutation (D809G) was examined enhanced the catalytic or DNA binding function of the in vivo using an episomal system and revealed a protein. marked reduction in catalytic activity (Xu et al., 1999). As was mentioned previously, mice with a homozygous knockout of the Dnmt3b gene recapitulate ICF syndrome the aberrant patterns of pericentromeric hypomethyla- tion seen in humans with ICF syndrome and may serve ICF syndrome is a very rare autosomal recessive as a model for ICF syndrome if a less severe knockout disorder with fewer than 40 cases reported over the can be generated (Okano et al., 1999). Interestingly, no last 20 years and only two of these in the United States patients with mutations that would result in homo- (Carpenter et al., 1988; Sawyer et al., 1995). A€ected zygous loss of the N-terminal region of DNMT3B were individuals demonstrate variable immunode®ciency identi®ed indicating that complete loss DNMT3B may consisting of an absence or severe reduction in at least be embryonic lethal to humans as it is in mice. This two immunoglobulin isotypes and often su€er from also implies that the ICF-speci®c mutations in severe respiratory tract infections (Franceschini et al., DNMTB may not be complete loss of function alleles. 1995; Smeets et al., 1994). Developmental defects The mutant forms of DNMT3B may retain low level include a variable degree of mental impairment, methyltransferase activity in vivo at its normal sites of delayed developmental milestones, and peculiar facial action or, as will be discussed later for DNMT1, features such as low-set ears, hypertelorism, ¯at nasal additional roles for this protein in gene expression or bridge, micrognathia, and macroglossia (Smeets et al., chromatin structure may exist (and remain functional 1994). Perhaps the most remarkable characteristic of in the mutants). It remains unclear how DNMT3B is ICF syndrome is the dramatic elongation of juxtacen- targeted to pericentromeric heterochromatin but is tromeric heterochromatin in lymphocytes from a€ected most likely a result of the interaction of DNMT3B individuals. Profound chromosomal abnormalities, with another DNA binding protein or protein complex including multibranched con®gurations (multiradials), which is targeted to pericentromeric heterochromatin. deletions or duplications of entire chromosome arms, Interestingly, a recent report showed that ICF cells are isochromosomes, and centromeric breakage involving extremely sensitive to ionizing radiation despite posses- almost exclusively chromosomes 1, 9 and 16 have been sing intact cell cycle checkpoints (Narayan et al., 2000). observed (Franceschini et al., 1995). Telomeric associa- DNMT3B may be required for the proper methylation, tions between non-acrocentric chromosomes were also and thus proper gene expression patterns and chroma- recently reported (Tuck-Muller et al., 2000). Interest- tin structure, of a select group of as yet unidenti®ed

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3143

Figure 2 Mutations in the human DNMT3B gene on chromosome 20q11.2 in ICF syndrome. Most individuals are compound heterozygotes for two independent mutations (denoted with an `h'). With a single exception, all mutations occur within the catalytic domain. Conserved methyltransferase motifs within the catalytic domain are indicated with roman numerals. Several mutations (mt.) result in altered splicing patterns (Alt.) and the insertion (Ins) or deletion (D) of coding sequence genes necessary for normal brain development. Cen- which are normally heavily methylated. This may result tromeric heterochromatin regions exist in distinct foci in increased transcription from transposable elements within the nucleus. These foci are enriched in both and increased genomic instability. It has been proposed heterochromatin binding proteins like HP-1, which are that the ancestral function of DNA methylation was in involved in transcriptional silencing, as well as fact to restrain the spread of parasitic elements as transcriptionally silent genes themselves (Brown et al., genomes became larger and more complex and the 1997). It has been proposed that loss of methylation dangers to genome integrity from unrestrained trans- from centromeric regions might prevent these foci from position events increased (Bestor and Tycko, 1996; forming or reduce the ability of these regions to recruit Yoder et al., 1997). This genome defense system was silencing proteins which could alter transcription and then later utilized as a method of gene regulation. genome structure in profound ways (Xu et al., 1999). Regions that are frequent targets of hypermethylation ICF syndrome now joins a growing list of genetic events are CpG islands. CpG islands are GpC and diseases including Rett, Fragile X, and ATR-X (alpha- CpG-rich regions of approximately 1 kilobase (kb) that thalassemia, mental retardation, X-linked) syndromes are usually associated with the promoter or 5'-end of that have in common defects in the cellular methyla- genes (Figure 3). It has been estimated that there are tion machinery (Robertson and Wol€e, 2000). Com- 45 000 CpG islands in the human genome and these are mon to all these diseases is a variable degree of mental associated with roughly half of all genes (Antequera impairment, indicating that DNA methylation-depen- and Bird, 1993). CpG island methylation is rare in dent gene control pathways or chromatin structure normal cells. It plays a role in X-chromosome modi®cations may be particularly important for brain inactivation in females and genomic imprinting, development. Further support for a unique role for increases with age and in vitro cell culture. Abnormal DNA methylation in the brain comes from reports methylation of CpG islands can eciently repress demonstrating that DNA methyltransferase activity is transcription of the associated gene in a manner akin high in neurons despite their terminally di€erentiated to mutations and deletions and act as one of the `hits' state (Goto et al., 1993) and that DNA methyltransfer- in the Knudsen two-hit hypothesis for tumor genera- ase activity may contribute to induced ischemic brain tion (Figure 3) (Baylin and Herman, 2000; Jones and damage in mice (Endres et al., 2000). Laird, 1999). There are now numerous examples of aberrant CpG island promoter hypermethylation of tumor suppressor genes, genes involved in cell-cell DNA hypermethylation and cancer adhesion, and genes involved with DNA repair and these are summarized in Table 1. Nearly ®fteen years ago it was recognized that DNA methylation patterns in tumor cells are altered relative A direct role for DNA methylation in tumorigenesis to those of normal cells (Feinberg et al., 1988; Goelz et al., 1985). Tumor cells exhibit global hypomethylation Does DNA methylation have a direct role in of the genome accompanied by region-speci®c hyper- carcinogenesis? There are now numerous lines of methylation events (Baylin and Herman, 2000; Jones evidence that indicate that the answer to this question and Laird, 1999). Most of the hypomethylation events is yes. For example, promoter-region hypermethylation appear to occur in repetitive and parasitic elements, of the retinoblastoma (pRb) (Stirzaker et al., 1997)

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3144

Figure 3 Model for how global loss and region-speci®c gain in methylation can occur within the same cell and potentially contribute to malignant transformation. A representative segment of DNA showing a transcriptionally active gene (light gray boxes are exons) with an unmethylated (white lollipops) CpG island promoter (bent arrow). CpG sites within the body of the gene or in repetitive elements (black boxes and `LINE') are generally hypermethylated (black lollipops). (a) In a normal cell, a DNMT- interacting factor (`X') may guide the DNMT to the regions that are to be methylated or allow the DNMT access by opening chromatin. Additionally, there may be a factor (`Y') which actively blocks access of the DNMT to the CpG island or recruits CpG islands to subnuclear regions from which DNMT's are restricted. (b) In a premalignant cell, a defect in the methylation system occurs which could include one or more of the following: mutation in X, loss of Y, or an inappropriate timing or expression level of X or Y during the cell cycle. This would grant the DNMT access to the CpG island, a region rich in potential methylation sites, and `titrate' the enzyme away from other regions which are normally methylated. (c) Continued cell division and/or additional defects in the methylation system exacerbate the methylation errors eventually resulting in silencing of the gene and potentially reactivation of transcription or transposition from parasitic elements. Repetitive regions that lose methylation may become decondensed and prone to mitotic recombination. `MBD' is methyl-CpG binding protein, `HDAC' is histone deacetylase, and `TF' is transcription factor

gene and the von Hippel Lindau (VHL) (Herman et methylation of the mismatch repair gene hMLH1. That al., 1994) gene has been detected in familial cases of transcriptional silencing of the hMLH1 gene is unilateral retinoblastoma and renal cancer, respec- primarily mediated by DNA hypermethylation was tively, and is the primary inactivating event (Baylin shown by treating cell lines containing hypermethylated and Herman, 2000; Jones and Laird, 1999). In hMLH1 alleles with the DNA methyltransferase addition, studies of sporadic cases of colorectal inhibitor 5-aza-2'-deoxycytidine (5-azaCdR). This re- carcinomas exhibiting microsatellite instability have sulted in re-expression of hMLH1 and partial restora- revealed a high frequency of promoter region hyper- tion of mismatch repair ability (Herman et al., 1998).

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3145 Table 1 CpG-island-associated genes involved in cell growth control or metastasis that can become hypermethylated and silenced in tumors Gene Function References pRb Regulator of G1/S phase transition (Sakai et al., 1991; Stirzaker et al., 1997) p16INK4a Cyclin-dependent kinase inhibitor (Gonzalez-Zulueta et al., 1995; Merlo et al., 1995) p15INK4b Cyclin-dependent kinase inhibitor (Herman et al., 1996) ARF Regulator of p53 levels (Esteller et al., 2000b; Robertson and Jones, 1998) hMLH1 DNA mismatch repair (Herman et al., 1998; Kane et al., 1997) APC Binds b-catenin, Regulation of actin cytoskeleton? (Hiltunen et al., 1997) VHL Stimulates angiogenesis (Herman et al., 1994) BRCA1 DNA repair (Dobrovic and Simpfendorfer, 1997; Rice and Futscher, 2000) LKB1 Serine/threonine protein kinase (Esteller et al., 2000a) E-cadherin Cell ± cell adhesion (Graff et al., 1995; Yoshiura et al., 1995) ER Transcriptional activation of estrogen-responsive genes (Issa et al., 1994) GSTPI Protects DNA from oxygen radical damage (Esteller et al., 1998; Lee et al., 1994) O6-MGMT Repair/removal of bulky adducts from guanine (Esteller et al., 2000c; Qian and Brent, 1997) TIMP3 Matrix metalloproteinase inhibitor (Bachman et al., 1999) DAPK1 Kinase required for induction of apoptosis by g interferon (Katzenellenbogen et al., 1999) p73 Apoptosis?, structurally similar to p53 (Corn et al., 1999; Kawano et al., 1999)

Furthermore, examples in which one copy of a tumor di€erent tumor types. Certain tumor types, like breast, suppressor gene is either mutated or lost and the wild- head and neck, and testicular, displayed a relatively type copy is transcriptionally silenced by hypermethy- low frequency of aberrant methylation while other lation have been found. For example, the HCT116 tumors such as colon, glioma, and acute myeloid colon adenocarcinoma cell line is defective in the Rb leukemia, displayed a signi®cantly higher frequency of pathway due to loss of expression of the cyclin- aberrant methylation events. It was estimated that an dependent kinase inhibitor p16INK4a. It has been shown average of 608 (up to as many as 4500) CpG islands that one allele of p16INK4a is mutated resulting in were aberrantly hypermethylated in tumors. Methyla- premature termination and loss of function while the tion abnormalities were detectable in both low and second copy is wild-type but silent due to methylation high-grade malignancies, again supporting the notion of the promoter region in HCT116 cells (Myohanen et that methylation changes can be an early event in al., 1998). Lastly, it has recently been recognized that tumor progression. Furthermore, CpG island methyla- aberrant hypermethylation events can occur early in tion was not random indicating that certain CpG tumorigenesis and disrupt pathways that may predis- islands may be more susceptible to de novo methylation pose cells to malignant transformation. For example, than others or that selective loss of expression from methylation of the p16INK4a promoter has been detected certain CpG island associated genes may be favored if in both experimentally induced tumor models as well loss of that gene provides the cell with a growth as the natural setting in lung cancer and mammary advantage (Costello et al., 2000). The non-random epithelial cells. Aberrant methylation was detectable in nature of CpG island hypermethylation between tumor pre-neoplastic lesions and the frequency of aberrations types may even provide a useful signature for increased with disease progression (Belinsky et al., classi®cation if a more high-throughput method for 1998; Nuovo et al., 1999; Wong et al., 1999). Thus, carrying out this type of analysis can be developed and ample evidence exists to support the notion that DNA could also provide important information on the hypermethylation events can act as a primary inacti- molecular defects (particularly epigenetic defects) which vating event contributing directly to tumorigenesis. contribute to the generation of a tumor cell.

CpG island hypermethylation in tumors is widespread How does DNA methylation silence transcription? The identi®cation of CpG island-associated genes that become methylated in tumors has relied primarily on a A connection between DNA methylation and tran- candidate gene approach; a potential tumor suppressor scriptional silencing in vertebrates has been recognized gene is identi®ed with a CpG island promoter then for over 20 years, yet evidence directly connecting the tumor cell lines are examined for methylation of this two has only recently been obtained (Jones et al., 1998; region. Table 1 lists only 17 such genes/CpG islands Nan et al., 1998). Hypermethylated promoters are out of an estimated 45 000 CpG islands in the human almost always transcriptionally silent, packaged into a genome. How widespread are hypermethylation events chromatin structure resistant to nucleases, and enriched in tumors? A study by Costello et al. (2000) begins to in hypoacetylated core histones (Eden et al., 1998; address this question by utilizing a restriction land- Jones and Laird, 1999). Local cytosine methylation of mark genomic scanning (RLGS) approach to analyse a particular sequence can directly interfere with the the methylation status of 1184 CpG islands in an binding of certain transcription factors (Tate and Bird, unbiased manner from 98 tumor samples. Results 1993) but this is unlikely to be a widespread indicated that aberrant methylation of CpG islands mechanism for transcriptional silencing since most was di€erent between individual tumors and between transcription factors do not have CpG dinucleotides

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3146 within their binding sites. Furthermore, this direct recent studies have shown that DNMT1 can interact interference model cannot easily account for the wide with both HDAC1 and 2 (Fuks et al., 2000; Robertson range of biological phenomena that rely on methyla- et al., 2000a; Rountree et al., 2000). The exact tion for the global silencing of large domains or even functional signi®cance of this interaction is not fully entire chromosomes, such as X inactivation in females. understood but two interesting scenarios can be An alternative mechanism involves proteins that bind imagined. In one, histone deacetylation may actually selectively to methylated DNA. The ®rst such protein, be necessary for ecient methylation of a given region MeCP2, identi®ed in 1992 (Lewis et al., 1992), can be and the DNMT1-HDAC interaction has no direct divided into two structural domains: a methyl-CpG e€ect on transcriptional repression. In the second, binding domain (MBD) which recognizes a symme- DNMT1, once targeted to the region to become trically methylated CpG dinucleotide through contacts methylated, may act as a master repressor by in the major groove of the double helix (Wake®eld et simultaneously methylating CpG sites in the promoter al., 1999), and a transcriptional repression domain and also e€ecting histone deacetylation and chromatin (TRD) which interacts with several other regulatory compaction by its associated HDAC's. Once DNMT1 proteins (Nan et al., 1997). Since that time it has been completes methylation of the region (and presumably found that MeCP2 is only one member of a family of departs), MBD's would be recruited and these in turn methyl-CpG binding proteins (MBD's) which now would recruit HDAC activity to maintain or even include MBD1-4 (Hendrich and Bird, 1998). Thus a potentiate the repressed state. second mechanism by which DNA methylation could inhibit transcription is by sterically blocking access of transcription factors by binding of the MBD's. DNA hypomethylation ± roles in cancer and genome A critical ®nding that directly linked not only DNA stability methylation and transcriptional silencing, but also DNA methylation and histone hypoacetylation, was The emphasis in studies of DNA methylation and that MeCP2 could recruit histone deacetylase (HDAC) cancer has focused primarily on hypermethylation (Jones et al., 1998; Nan et al., 1998). This allowed for a events as was discussed in the preceding paragraphs. rational mechanism for how DNA methylation could The global loss of genomic DNA methylation, which repress transcription and result in a chromatin occurs concomitantly with CpG island hypermethyla- structural change: recruitment of MBD's and their tion, may be of equal importance to the generation of associated HDAC's to methylated DNA would result a transformed cell. It is important to stress that in local deacetylation of core histone tails, which would tumors exhibit a global defect in the DNA methyla- result in tighter packaging of DNA and reduced access tion system with hypo- and hypermethylation events of transcription factors to their binding sites (Robert- occurring in the same cell (Figure 3). The reason for son and Wol€e, 2000). Recent experiments have linked the loss of methylation in tumors is not known, four of the methyl-CpG binding domain-containing although several models will be proposed in the next proteins, MeCP2, MBD1, MBD2, and MBD3 with section. aspects of the chromatin remodeling machinery in As was mentioned previously, one primary function addition to HDAC. In Xenopus eggs for example, of DNA methylation is suppression of transcription MBD3 is a component of the Mi-2 chromatin and expansion of parasitic elements like transposons remodeling complex which also includes Rpd3 (Xeno- (this include SINES and LINES as two examples) pus HDAC1/2) and RbAp46/48 (Wade et al., 1999). (Yoder et al., 1997). The vast majority of methylated Furthermore, MBD2, HDAC1, HDAC2, and RbAp46/ CpG's do in fact reside within repetitive elements 48 co-purify in HeLa cell nuclear extracts and are (Yoder et al., 1997) and methylation of the long components of the MeCP1 repressor complex (Ng et terminal repeat (LTR) promoters of many parasitic al., 1999). Evidence for a mechanistic link between elements inhibits their activity (Kochanek et al., 1995). DNA methylation and histone deacetylation has also Experimentally-induced demethylation of the genome been demonstrated by treating cells with a combination by homozygous knockout of the Dnmt1 gene also of the DNA methyltransferase inhibitor 5-azaCdR and supports this notion by demonstrating that transpo- the histone deacetylase inhibitor trichostatin A (TSA). sable elements become demethylated and re-express in Low doses of 5-azaCdR resulted in low-level re- Dnmt17/7 ES cells (Walsh et al., 1998). Similar expression and minimal demethylation of hypermethy- observations have been made in primary human and lated CpG-island-associated genes but a combination rodent tumors samples (Flori et al., 1999; Grassi et al., of 5-azaCdR and TSA resulted in robust activation of 1999). Transcription from many strong promoters these same genes (TSA alone had no e€ect) (Cameron could globally alter transcription patterns by altering et al., 1999). This revealed not only that DNA transcription factor levels or by negatively e€ecting methylation and histone deacetylation worked together speci®c growth-regulatory genes in which the reacti- to silence transcription but also that DNA methylation vated elements reside. Gene function could be dis- was dominant over histone acetylation status. An rupted not only by direct insertion of a new additional, newly discovered player in DNA methyla- transposable element into a coding exon but also by tion-mediated transcriptional repression may be transcriptional interference, transcriptional initiation DNMT1 itself. As will be discussed in detail later, within a coding region (introns), or generation of an

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3147 antisense transcript if the element is integrated in the Potential causes of aberrant methylation in tumor cells antisense orientation (Robertson and Wol€e, 2000). While the loss of methylation in tumor cells may The exact nature of the defect in the cellular predispose them to genomic instability via the tran- methylation machinery in tumor cells remains un- scriptional activation and movement of parasitic known however it is important that models take into elements, these elements also pose a signi®cant threat account the observations that methylation is both lost to the genome as mediators of homologous recombina- and gained from di€erent regions of the genome in tion. DNA methylation may act to stabilize the transformed cells (Figure 3). One possible defect that genomes of organisms containing large amounts of has been studied extensively is related to the expression repetitive DNA by `masking' or inhibiting homologous level of the DNMT's. Numerous studies have examined recombination between repeats (Colot and Rossignol, expression of the DNA methyltransferases in tumor 1999). Several examples of recombination events tissue, almost exclusively at the RNA level, and most between repeats have been identi®ed in the human have reported variable degrees of overexpression, population and these have had deleterious conse- particularly for DNMT1 (Issa et al., 1993; Lee et al., quences (Puget et al., 1997; Rouyer et al., 1987; Small 1996; Robertson et al., 1999). Induced overexpression et al., 1997). Work with the fungus Ascobolus immersus of DNMT1 in tissue culture cells has been shown to has demonstrated that induced methylation of a gradually induce CpG island hypermethylation and meiotic recombination hotspot reduced the frequency result in cellular transformation (Vertino et al., 1996; of crossing-over within this region by several hundred- Wu et al., 1993). The exact degree of overexpression of fold (Maloisel and Rossignol, 1998). The most direct methyltransferase in tumors and the frequency remains evidence that DNA methylation suppresses homolo- somewhat controversial but low-level overexpression gous recombination has come from the work in (2 ± 4-fold) is likely to be relatively common. There are, Ascobolus, but several studies support a similar role however, many examples of tumors which do not in mammalian cells. For example, V(D)J recombina- overexpress (Eads et al., 1999; Robertson et al., 1999) tion is reduced more than 100-fold when the and even if this was a universal feature of tumor cells it recombination substrate is methylated (Hsieh and would be dicult to imagine how this could be Lieber, 1992). In addition, Dnmt1 knockout ES cells responsible for both hypermethylation of CpG islands exhibit a 10-fold increased mutation rate involving and global genome hypomethylation. gene rearrangements (Chen et al., 1998), and indivi- Another recently proposed mechanism, which could duals with ICF syndrome or cultured cells treated with disrupt the regulation of DNA methylation patterns, 5-azaCdR show increased numbers of chromosomal relates to inappropriate DNMT expression during the translocations (Ji et al., 1997; Miniou et al., 1994). cell cycle. Several studies have revealed that DNMT1, Several studies have now clearly demonstrated that 3A, and 3B are expressed di€erentially during the cell repetitive elements become demethylated in tumors and cycle (Robertson et al., 2000b; Szyf et al., 1991; that the degree of hypomethylation correlates with Tatematsu et al., 2000) and that DNMT RNA levels disease progression (Narayan et al., 1998; Qu et al., and overall DNA methyltransferase activity is higher in 1999). Many of the rearrangements observed in growth-arrested normal cells than in growth-arrested primary tumors involve the centromere of chromo- tumor cells (Robertson et al., 2000b). Aberrant somes 1 and 16 (Heim and Mitelman, 1995). Pericen- expression of one of the DNMT's during G1 for tromeric rearrangements involving chromosome 1 are example, could give rise to methylation errors, or de over-represented in cancers relative to the total novo methylation events at normally unmethylated percentage of pericentromeric heterochromatin se- CpG sites, which would be copied by DNMT1 after quences in the genome (Mertens et al., 1997). cell division (Jones, 1996). Inappropriate timing of Aberrations include whole-arm deletions and unba- expression of a critical level of DNMT activity during lanced translocations between 1q and 16p with the multiple rounds of cell division could then lead to a fusion of these arms occurring within centromeric progressive increase in aberrant methylation over time. heterochromatin. It remains unknown how DNA This type of defect could also be responsible for the methylation may suppress homologous recombination global hypomethylation observed in tumor cells if, for but potential mechanisms include masking of the example, certain DNMT's are localized to speci®c recombination initiation site, maintenance of a highly regions of the nucleus or are responsible for propagat- condensed chromatin structure through the recruitment ing methylation patterns of DNA segments which of DNA organizing proteins, destabilization of the replicate during speci®c times during S phase and the recombination intermediate, or interference with the required amount of enzyme is not present (Robertson assembly of the recombination machinery. Thus loss of et al., 2000b). DNA methylation from repetitive elements may occur The ®nal, and perhaps most speculative model of early in tumor progression and may predispose cells to the molecular nature of the methylation defect in genome rearrangements via mitotic recombination tumors is that DNA methyltransferase nuclear which could inactivate critical growth regulatory genes localization, sequence targeting, or regulated enzy- or, via whole chromosome arm gain and/or loss, result matic activity is disrupted in tumor cells due to in an inappropriate level of expression of a tumor aberrant protein-protein interactions with DNMT- suppressor gene or growth-promoting gene. associated proteins or protein complexes in which

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3148 DNA methyltransferases reside. In this model, dia- which require only the CpG dinucleotide or in some grammed schematically in Figure 3, DNA methyl- cases a CpA and CpT dinucleotide as their recognition transferase is targeted to the genomic regions which sequence (Bestor and Ingram, 1983; Ramsahoye et al., are to be methylated and/or restricted access to 2000). In this section I will review the proteins known regions which are to be kept methylation-free via its to interact with DNMT1 and their potential func- interaction with other proteins. In a premalignant or tional signi®cance. tumor cell, the normal distribution of DNA methyl- transferase is disrupted because of aberrations in the p23 DNMT- interaction partners, via mutation, loss, inappropriate expression level or timing of expression A listing of the known DNMT1-interacting proteins is of the interaction partner, or defective post-transla- shown in Table 2 and a summary of the interaction tional modi®cation. This could then allow access of sites of these proteins on DNMT1 is shown in Figure DNA methyltransferase to regions that would not 4. Most of these interacting proteins have been normally be methylated, like CpG islands. At the identi®ed in yeast two-hybrid screens or by biochemical same time, this could `titrate' DNA methyltransferase fractionation methods. Two of the earliest reported away from regions which are supposed to be DNMT1-intertacting factors, which still remain rather methylated, like the `bulk' genomic DNA or repetitive enigmatic, are p23 (Zhang and Verdine, 1996) and sequences. Indirect support for this model comes from annexin V (Ohsawa et al., 1996). p23 was originally several recent reports, to be discussed in detail in the identi®ed as a phosphoprotein associated with the next section, that DNMT1 can interact with a number cytoplasmic form of the progesterone receptor (PR) of di€erent proteins that could clearly a€ect its and subsequently shown to interact with other steroid nuclear localization (annexin V, p23, PCNA) as well hormone receptors and hsp90 as a co-chaperone as access to its DNA target sites in chromatin (Buchner, 1999; Johnson et al., 1994). The exact (HDAC1, HDAC2, and pRb). It is likely that the function of this well-conserved protein, however, identi®cation of DNMT-interacting factors will be a remains unclear. p23 resides in the nucleus as well as major focus in the methylation ®eld in the coming the cytoplasm, it is the only protein known to interact years and it will be critically important to examine the with the catalytic domain of DNMT1 (although it was integrity of these interactions in transformed cells. reported that this interaction did not a€ect catalytic activity in vitro), and it has also been shown to associate with telomerase (Holt et al., 1999; Zhang and Proteins known to interact with DNMT1 and their Verdine, 1996). Unfortunately searches for interacting potential functions factors have been biased toward the N-terminal regulatory domain of DNMT1 and important interac- As the heading of this section suggests, the discussion tions that could directly a€ect binding site speci®city on DNMT-interacting proteins will focus on proteins and catalytic activity may have been missed. The known to interact with DNMT1. This is simply signi®cance of the interaction of p23 with DNMT1 because DNMT1 has been far more extensively remains unknown but may aid in the proper folding of studied. Given the critical role of the DNMT3 family the catalytic domain or could tether DNMT1 to the in the de novo establishment of DNA methylation cytoskeleton or nuclear matrix due to the ability of patterns during embryogenesis (Okano et al., 1999) it p23, via hsp90, to interact with actin ®laments (Miyata is likely that interacting factors which regulate their and Yahara, 1991). catalytic activity, substrate speci®city, and sequence targeting will be critical and the identi®cation of such factors will no doubt be a major area of study in the next few years. As was previously mentioned, cellular Table 2 DNMT1-associated proteins and their proposed functions methylation patterns are not random. Certain genomic Protein Reference Proposed functions regions, like pericentromeric heterochromatin, im- printed regions (one allele), genes on the inactive X p23 (Zhang and Verdine, 1996) Folding Linkage to nuclear matrix chromosome in females, and bulk genomic DNA Annexin V (Oshawa et al., 1996) Linkage to nuclear matrix which includes many of the parasitic DNA elements, PCNA (Chuang et al., 1997) Targeting to replication foci are hypermethylated while other regions, like CpG Targeting to repair sites islands often associated with transcriptionally active HDAC1 (Fuks et al., 2000) Transcriptional repression genes, are hypomethylated (Robertson and Wol€e, Chromatin remodeling HDAC2 (Rountree et al., 2000) Transcriptional repression 2000; Yoder et al., 1997). In tumor cells the normal DMAP1 Chromatin remodeling methylation patterns become reversed due to an as yet Maturation of chromatin unknown defect or defects in the methylation after DNA replication machinery (Baylin and Herman, 2000; Jones and pRb (Robertson et al., 2000a) Transcriptional repression HDAC1 Chromatin remodeling Laird, 1999). This apparent speci®city or targeting of Methylation targeting methylation to particular regions of the genome MBD3 (Tatematsu et al., 2000) Targeting to hemi- contrasts with in vitro data indicating little sequence methylated DNA speci®city of the known DNA methyltransferases, Targeting to replication foci

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3149

Figure 4 A summary of proteins currently known to interact with DNMT1 and the location of the interacting region. The MBD3- DNMT1 interaction has not been mapped (indicated by arrow and `?'). Potential functions of these interactions are summarized in Table 2

with both the leading and lagging strands as a trimer Annexin V (Kelman, 1997). PCNA is also essential for both Annexin V is a calcium-dependent phospholipid mismatch (Umar et al., 1996) and nucleotide excision binding protein which has been found in both the repair (Nichols and Sancar, 1992). PCNA is the target nucleus and cytoplasm and its localization can vary of the cell cycle regulator p21WAF1/CIP1 (Waga et al., with growth state of the cell and with calcium 1994). In response to DNA damage p21 is upregulated concentration (Barwise and Walker, 1996; Raynal et (El-Deiry et al., 1993), binds to PCNA, and inhibits al., 1996). The annexin V in the nucleus is associated DNA synthesis (Waga et al., 1994). The DNMT1- with the nuclear matrix and was found to interact with PCNA interaction may allow for the newly synthesized the N-terminus of DNMT1 (Figure 4) (Altieri et al., daughter strands to be rapidly remethylated before 1996; Ohsawa et al., 1996). Interestingly, a second being packaged into chromatin. This may be essential annexin V binding protein identi®ed by the same group for maintenance of proper methylation patterns since was ATRX. The ATRX gene encodes a putative ATP- histone HI, one of the principal chromatin organizing dependent chromatin remodeling factor and helicase of proteins, has been shown to inhibit DNA methylation the SNF2 family that is mutated in patients with ATR- (Carotti et al., 1996). X syndrome (Picketts et al., 1996). One interesting Loss of p21, either by mutation of the p21 gene itself feature of this disease is that patients have subtle or loss of p53, which activates p21 expression in defects in cellular methylation patterns which include response to DNA damage, is a common event in both hypo- and hypermethylation events at select cancer (Vogelstein et al., 2000). It has been proposed repetitive elements (Gibbons et al., 2000). ATRX, that loss of p21, which also associates with PCNA at DNMT1, and annexin V are highly expressed in the sites of DNA repair, might allow access of DNMT1 to brain (Cardoso et al., 1998; Goto et al., 1993; Ohsawa the damaged regions resulting in inappropriate de novo et al., 1996) and both DNMT1 and ATRX have been methylation at CpG sites that are not normally found to be associated with the nuclear matrix (Berube methylated since DNMT1 and p21 interact with PCNA et al., 2000; Burdon et al., 1985). The functional in a mutually exclusive manner (Chuang et al., 1997). consequence of this interaction remains unclear but While this scenario may be true, another possibility may be to anchor DNMT1 to the nuclear matrix. should be considered in light of more recent results. DNA damage results in disruption of normal chroma- tin structure and may also result in loss of DNA PCNA methylation if the repaired region contains CpG sites The 1997 report by Chuang et al. (1997) that DNMT1 (Gaillard et al., 1996; Kastan et al., 1982). It is associates with the replication-associated protein essential for genome integrity that not only the DNA PCNA (proliferating cell nuclear antigen) received sequence be properly repaired, but also the original much attention. It is believed that this interaction DNA methylation pattern and chromatin structure of may target DNMT1 to replication foci although the damaged region be restored to ensure proper gene several independent domains within the N-terminus expression patterns. Interestingly, it has been found of DNMT1 appear to be capable of mediating this that the chromatin assembly factor CAF-1 (speci®cally recruitment (Leonhardt et al., 1992; Liu et al., 1998). the p150 subunit of CAF-1), which assembles nucleo- PCNA, the polymerase processivity factor or sliding somes onto replicating DNA (Kaufman et al., 1995), clamp, is required for DNA replication and associates interacts directly with PCNA and the presence of

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3150 PCNA molecules left on the newly replicated DNA acetylated regions are replicated in early S phase and mark these regions for chromatin assembly (Shibahara transcriptionally silent, hypoacetylated, heterochro- and Stillman, 1999). A similar process also appears to matic regions are replicated late in S phase, the be operational at sites of DNA damage (Gaillard et al., recruitment of HDAC2 selectively to late replication 1996). Thus the presence of DNMT1 at sites of DNA foci may be instrumental in deacetylating the newly damage, rather than being a liability for the cell, may deposited histones (Rountree et al., 2000). Thus the be essential for the restoration of the original selective interaction of DNMT1 with HDAC2 during S methylation patterns of the region as well as the phase may be critical for the maintenance of these two original chromatin structure of the damaged region. epigenetically distinct compartments of the genome. A PCNA, by virtue of its interaction with DNMT1 and similar situation may also be operational with HDAC1 CAF-1, would act as a cellular memory device for the although this has not yet been investigated. epigenetically encoded information of the genome (Shibahara and Stillman, 1999). The ability of DNMT1 DMAP1 to interact with HDAC, which will be discussed in the next section, may also be critical for the proper Further evidence that DNMT1 is more than just a maintenance and ®delity of the epigenetic component maintenance methyltransferase comes from studies of the genome. indicating that it can repress transcription indepen- dently of its ability to catalyze the methyl-transfer reaction. Three reports revealed that the N-terminus of HDAC's DNMT1 could interact with HDAC1 and 2 and Three recent papers have generated great interest by repress transcription in a partially-HDAC-dependent demonstrating that DNMT1 can interact with histone manner when tethered to a promoter (Fuks et al., 2000; deacetylases 1 and 2 (Table 2, Figure 4) (Fuks et al., Robertson et al., 2000a; Rountree et al., 2000). The 2000; Robertson et al., 2000a; Rountree et al., 2000). HDAC-insensitive component of DNMT1-mediated An association between transcriptionally inactive transcriptional repression may be due to its interaction regions, histone hypoacetylation, and DNA hyper- with DMAP1, which has also been shown to repress methylation has been recognized for some time (Eden transcription when fused to a heterologous DNA et al., 1998). The ®nding that methyl-CpG binding binding domain but this repression was insensitive to proteins (MBD's) associate with histone deacetylases inhibitors of histone deacetylase (Rountree et al., provided a mechanistic link for these observations 2000). DMAP1, isolated in a two-hybrid screen, is a (Jones et al., 1998; Nan et al., 1998). The situation may small protein with a potential coiled-coil domain as now be more complex given that DNMT1 itself can well as a SANT-like domain within the central portion interact with HDAC1 and 2 as well a novel co- of the molecule. It interacts with the most N-terminal repressor, termed DMAP1, which appears to repress portion of DNMT1 (Figure 4). The function of SANT transcription in an HDAC-independent manner (Roun- domains remain unclear but they may be DNA binding tree et al., 2000). DNMT1 can directly repress motifs or sites of protein-protein interaction (Stewart transcription when fused to a heterologous DNA and Gibson, 1996). Interestingly, DMAP1, which maps binding domain (Fuks et al., 2000; Rountree et al., to human chromosome 1p32.2 ± 34.2, is close to a 2000) or when targeted to promoters containing E2F- previously described methylation modi®er locus binding sites by virtue of a newly discovered interac- (MEMO-1) which a€ects the methylation status at tion between DNMT1 and the retinoblastoma gene certain class I HLA-loci (Cheng et al., 1996). Little is product pRb (Robertson et al., 2000a). The traditional known about the exact function of DMAP1, although view of DNMT1 as a maintenance methyltransferase it has been shown to interact with the tumor which simply copies methylation patterns after replica- susceptibility gene 101 (tsg101). tsg101 was isolated in tion must now be reconsidered. The interaction of a mutagenesis screen of NIH3T3 cells as a protein, DNMT1 with PCNA and HDAC2 may be critical for which when mutated, resulted in transformation of the restoration of chromatin structure after DNA NIH3T3 cells (Li and Cohen, 1996). Interestingly, replication given that newly synthesized histones arrive tsg101 has also been shown to have co-repressor at replicated DNA in hyperacetylated form but must activity and interacts with steroid hormone receptors become deacetylated for the restoration of proper (Hittelman et al., 1999). chromatin structure (Annuziato and Seale, 1983). Rountree et al. (2000) showed that DNMT1 recruited pRb to replication foci interacts with HDAC2 in a temporally distinct manner. Early S-phase replication Another protein recently reported to interact with the foci contained DNMT1 and the co-repressor DMAP1 N-terminal region of DNMT1 is the Rb tumor and it was proposed that the function of this complex suppressor gene (Table 2, Figure 4). It was demon- was to temporarily repress transcription during passage strated by biochemical fractionation that DNMT1 is of the replication fork as well as copy the parental associated with Rb and the Rb-associated DNA- methylation patterns. Late S-phase replication foci binding protein and transcriptional activator E2F-1. were found to contain DNMT1, DMAP1, and This same report showed that HDAC1 is also HDAC2. Since most transcriptionally active, hyper- associated with DNMT1 (Robertson et al., 2000a).

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3151 Rb is known to bind to E2F1 via the A/B pocket `titrating' the enzyme away from the relatively CpG- region and repress transcription from E2F-responsive poor bulk genomic DNA resulting in gradual hypo- promoters by masking the transcriptional activation methylation of these regions. Such a model is clearly domain of E2F1 as well as by recruitment of HDAC speculative but testable. It is also intriguing that (Brehm and Kouzarides, 1999). When cells are DNMT1 interacts with the A/B pocket region of Rb stimulated to divide, Rb becomes hyperphosphorylated since this is the most frequently mutated region of Rb and no longer binds to E2F1, allowing E2F-responsive in tumors as well as the region targeted by transform- genes (genes that are involved in cell cycle progression) ing viruses like SV40 and HPV (Moran, 1993). to be activated and cells progress into S phase (Dyson, 1998). Interestingly, DNMT1 binding to Rb also MBD3 required an intact A/B pocket (Robertson et al., 2000a). One important functional consequence of these Another recent and rather intriguing report has interactions was that DNMT1 was capable of enhan- indicated that DNMT1 may interact, directly or cing Rb-mediated transcriptional repression at E2F-1 indirectly, with the methyl-CpG binding protein responsive promoters (Robertson et al., 2000a). The MBD3 (Tatematsu et al., 2000). Tatematsu et al. Rb-DNMT1 interaction has several important implica- (2000) found that MBD2 and MBD3 could form tions. The ®rst is that the presence of associated E2F-1 homo- and hetero-dimers and that the heterodimer could allow for a mechanism of targeting methylation possessed the novel property of binding to hemi- to genomic regions containing E2F binding sites, methylated DNA (Tatematsu et al., 2000). The exact although this has yet to be demonstrated. Interestingly, binding speci®city of MBD3 has been the subject of a recent report indicated that the binding of E2F itself some controversy, however none of the MBD's have may be regulated by methylation (Campanero et al., been shown to have anity for hemi-methylated DNA 2000). Second, the DNMT1-Rb interaction may be a (Bird and Wol€e, 1999). Furthermore, it was demon- way of enhancing Rb-mediated repression of cell cycle strated that MBD2 and MBD3 were localized to progression. This could be through a combination of replication foci in late S-phase and were co-localized methylation of promoters containing E2F binding sites with DNMT1 at the same time. Interestingly, co- which could then recruit methyl-CpG binding proteins immunoprecipitation experiments with epitope-tagged and their associated HDAC activities or by the direct MBD3 revealed that DNMT1 and MBD3 were present recruitment of HDAC1 and 2 to these promoters by in a complex, although the exact nature of the DNMT1. Such a mechanism could be utilized in cells interaction (direct or indirect) and the interacting undergoing terminal di€erentiation to ensure that genes domains on each protein were not de®ned in this involved in cell cycle progression are locked in the `o€' study (Figure 4) (Tatematsu et al., 2000). The ability of position. One ®nal important implication of the Rb- the MBD2-MBD3 heterodimer to bind to hemi- DNMT1 interaction is related to the methylation methylated DNA may indicate a previously unrecog- defects observed in cancer. Rb itself, or components nized role for these proteins in the deacetylation of of the Rb pathway, are mutated in nearly all tumor newly synthesized histones after DNA replication in a cells (Hanahan and Weinberg, 2000). Loss of func- manner akin to that previously described for DNMT1. tional Rb, either by direct loss of the Rb protein or It will be interesting to determine if DNMT1 and disruptions in the Rb pathway that a€ect its MBD3 are components of a larger complex and if the phosphorylation status, may result in improper nuclear formation of this complex is regulated in a cell-cycle localization of DNMT1 which could in turn result in dependent manner. the methylation defects observed in tumor cells as shown schematically in Figure 5. In this scenario, DNMT1 bound to Rb at transcriptionally repressed, Concluding Remarks E2F-responsive promoters, may be a mechanism of keeping the de novo methyltransferase activity of Although the list of interacting proteins may seem long DNMT1 in check. It is critical that de novo it is likely that we have only just scratched the surface methylation is tightly regulated in non-dividing cells of a large array of interacting factors that will regulate to prevent unwanted or aberrant de novo methylation not only DNA target-site speci®city of methylation but events as these would be copied with each subsequent also interactions that will make use of DNA methyl- cell division. When a cell is scheduled to divide, Rb transferases for purposes other than catalysis of the becomes phosphorylated, no longer binds to E2F1, and methyltransferase reaction, such as transcriptional could then release DNMT1 to go to replication foci repression and chromatin remodeling. For example, and perform its functions in maintenance methylation mutations in ATP-dependent chromatin remodeling and histone modi®cation (although the e€ect of Rb factors like ATRX in mammals and DDM1 in phosphorylation status on binding to DNMT1 has yet arabidopsis have revealed subtle to profound defects, to be investigated) (Robertson et al., 2000a). Loss of respectively, in DNA methylation patterns (Gibbons et functional Rb may grant DNMT1 free access to the al., 2000; Jeddeloh et al., 1999). The closest mamma- genome which could allow for aberrant de novo lian homolog to DDM1, termed Lsh (lymphoid speci®c methylation of CpG islands, regions rich in potential helicase), has recently been disrupted in mice (Geiman DNMT1 binding sites, while at the same time and Muegge, 2000) and it will be of great interest to

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3152

Figure 5 Model for one potential functional consequence of the DNMT1-Rb interaction. (A) In a normal resting cell Rb recruited DNMT1 may locally methylate the promoter potentiating the repressed state by recruitment of MBD's and their associated HDAC's. Enhanced repression may also occur via direct recruitment of HDAC by DNMT1 without methylation. DNMT1 activity would be restrained in G0/G1 when de novo methylation events would be highly undesirable. (b) When a cell commits to division, Rb is phosphorylated (`PO4'), no longer binds to E2F1, and the associated DNMT1 is free to go to replication foci (the e€ect of Rb phosphorylation on DNMT1 binding is not known). (c) In a tumor cell, mutations in Rb or the Rb pathway that result in constitutive Rb hyperphosphorylation are extremely common. This would allow DNMT1 access to the genome at all phases of the cell cycle and result in a gradual accumulation of methylation errors. `GPG' is growth promoting gene, `AF' is an activating factor, and `PolII' is RNA polymerase II. White and black lollipops represent unmethylated and methylated CpG's, respectively

learn the e€ects of this mutation on global methylation DNMT1 is an enzyme that only copies pre-existing patterns. methylation patterns will likely be need to be thrown The discovery and characterization of such interac- out. Its interactions with Rb, DMAP1, and potentially tions may ®nally reveal the exact relationship between (and indirectly) tsg101, may indicate a more global role DNA methylation patterns and chromatin structure, for this protein in transcriptional repression. For i.e. do methylation patterns determine chromatin example, tsg101 has been shown to interact with the structure or vice versa. Further interesting questions glucocorticoid receptor and repress its transcriptional that are likely to be answered by characterizing activation function (Hittelman et al., 1999). Could DNMT-interacting factors include the necessity of recruitment of DNMT1 and its associated co-repressor chromatin remodeling enzymes for access of the functions be involved in this repression in vivo? methylation machinery to its target sequences and the DNMT1 could in fact be viewed as an ideal role of the methylation machinery in in¯uencing local transcriptional repressor since it can utilize at least chromatin structure and gene expression through its two of the most potent repression mechanisms known associated HDAC's distinct from its ability to catalyze in mammalian cells. Although little is known of the the methyl-transfer reaction. Given the number of DNMT3 family with regard to interacting factors it is repressor activities associated with DNMT1 that have reasonable to assume that they may also be similarly been reported in only the last year, the old notion that multi-functional molecules of transcriptional regula-

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3153 tion. Given the exciting discoveries related to DNMT- study in the methylation ®eld. It is also likely that the interacting factors in the past year it is likely that the characterization of these interactions will shed light on roles of the methylation machinery in transcriptional the nature of the methylation defect in tumor cells and regulation, chromatin structure, DNA repair, and may even lead to novel therapies to reverse aberrant genome stability will become the subject of intense methylation patterns and restore growth control.

References

Aapola U, Shibuya K, Scott HS, Ollila J, Vihinen M, Heino Colot V and Rossignol J-L. (1999). BioEssays, 21, 402 ± 411. M, Shintani A, Kawasaki K, Minoshima S, Krohn K, Corn PG, Kuerbitz SJ, vanNoesel MM, Esteller M, Antonarakis SE, Shimizu N, Kudoh J and Peterson P. Compitello N, Baylin SB and Herman JG. (1999). Cancer (2000). Genomics, 65, 293 ± 298. Res., 59, 3352 ± 3356. Aasland R, Gibson TJ and Stewart AF. (1995). TIBS, 20, Costello JF, Fruhwald MC, Smiraglia DJ, Rush LJ, 56 ± 59. Robertson GP, Gao X, Wright FA, Fermisco JD, Altieri F, Maras B and Turano C. (1996). Biochem. Biophys. Peltomaki P, Lang JC, Schuller DE, Yu L, Bloom®eld Res. Commun., 225, 448 ± 454. CD, Caligiuri MA, Yates A, Nishikawa R, Huang H-JS, Annuziato AT and Seale RL. (1983). J. Biol. Chem., 258, Petrelli NJ, Zhang X, O'Dorisio MS, Held WA, Cavenee 12675 ± 12684. WK and Plass C.(2000). Nature Genet., 25, 132 ± 138. Antequera F and Bird A. (1993). Proc. Natl. Acad. Sci. USA, Dobrovic A and Simpfendorfer D. (1997). Cancer Res., 57, 90, 11995 ± 11999. 3347 ± 3350. BachmanKE,HermanJG,CornPG,MerloA,CostelloJF, Dyson N. (1998). Genes. Dev., 12, 2245 ± 2262. Cavanee WK, Baylin SB and Gra€ JR. (1999). Cancer Eads CA, Danenberg KD, Kawakani K, Saltz LB, Dannen- Res., 59, 798 ± 802. berg PV and Liard PW. (1999). Cancer Res., 59, 302 ± 306. Barwise JL and Walker JH. (1996). FEBS Lett., 394, 213 ± Eden S, Hashimshony T, Keshet I, Cedar H and Thorne AW. 216. (1998). Nature, 394, 842. Baylin SB and Herman JG. (2000). TIG, 16, 168 ± 174. El-Deiry WS, Tokino T, Velculescu VE, Levy DB, Parsons Beard C, Li E and Jaenisch R. (1995). Genes. Dev., 9, 2325 ± R, Trant JM, Lin D, Mercer WE, Kinzler KW and 2334. Vogelstein B. (1993). Cell, 75, 817 ± 825. Belinsky SA, Nikula KJ, Palmisano WA, Michels R, Endres M, Meisel A, Biniszkiewicz D, Namura S, Prass K, Saccomanno G, Gabrielson E, Baylin SB and Herman RuscherK,LipskiA,JaenischR,MoskowitzMAand JG. (1998). Proc. Natl. Acad. Sci. USA, 95, 11891 ± 11896. Dirnagl U. (2000). J. Neurosci., 20, 3175 ± 3181. Berube NG, Smeenk CA and Picketts DJ. (2000). Hum. Mol. Esteller M, Avizienyte E, Corn PG, Lothe RA, Baylin SB, Genet., 9, 539 ± 547. Aaltonen LA and Herman JG. (2000a). Oncogene, 19, Bestor T, Laudano A, Mattaliano R and Ingram V. (1988). J. 164 ± 168. Mol. Biol, 203, 971 ± 983. Esteller M, Corn PG, Urena JM, Gabrielson E, Baylin SB Bestor TH and Ingram VM. (1983). Proc. Natl. Acad. Sci. and Herman JG. (1998). Cancer Res., 58, 4515 ± 4518. USA, 80, 5559 ± 5563. Esteller M, Tortola S, Capella G, Peinado MA, Baylin SB Bestor TH and Tycko B. (1996). Nature Genet., 12, 363 ± 367. and Herman JG. (2000b). Cancer Res., 60, 129 ± 133. Bird AP and Wol€e AP. (1999). Cell, 99, 451 ± 454. Esteller M, Toyota M, Sanchez-Cespedes M, Capella G, Brehm A and Kouzarides T. (1999). TIBS, 24, 142 ± 145. Peinado MA, Watkins DN, Issa JP, Sidransky D, Baylin Brown KE, Guest SS, Smale ST, Hahm K, Merkenschlager SB and Herman JG. (2000c). Cancer Res., 60, 2368 ± 2371. M and Fisher AG. (1997). Cell, 91, 845 ± 854. Feinberg AP, Gehrke CW, Kuo KC and Ehrlich M. (1988). Buchner J. (1999). TIBS, 24, 136 ± 141. Cancer Res., 48, 1159 ± 1161. Burdon RH, Qureshi M and Adams RLP. (1985). Biochim. Flori AR, Lower R, Schmitz-Drager BJ and Schulz WA. Biophys. Acta, 825, 70 ± 79. (1999). Brit. J. Canc., 80, 1312 ± 1321. Cameron EE, Bachman KE, Myohanen S, Herman JG and FranceschiniP,MartinoS,CiocchiniM,CiutiE,Vardeu Baylin SB. (1999). Nature Genet., 21, 103 ± 107. MP, Guala A, Signorile F, Camerano P, Franceschini D Campanero MR, Armstrong MI and Flemington EK. and Tovo PA. (1995). Eur. J. Pediatr., 154, 840 ± 846. (2000). Proc. Natl. Acad. Sci. USA. Fuks F, Bergers WA, Brehm A, Hughes-Davies L and Cao X, Springer NM, Muszynski MG, Phillips RL, Kaeppler Kouzarides T. (2000). Nature Genet., 24, 88 ± 91. S and Jacobsen SE. (2000). Proc. Natl. Acad. Sci. USA, 97, Gaillard P-HL, Martini EM-D, Kaufman PD, Stillman B, 4979 ± 4984. Moustacchi E and Almouzni G. (1996). Cell, 86, 557 ± 896. Cardoso C, Timsit S, Villard L, Khrestchatisky M, Fontes M Geiman TM and Muegge K. (2000). Proc. Natl. Acad. Sci. and Colleaux L. (1998). Hum. Mol. Genet., 7, 679 ± 684. USA, 97, 4772 ± 4777. Carotti D, Funiciello S, Lavia P, Caiafa P and Strom R. Gibbons RJ, McDowell TL, Raman S, O'Rourke DM, (1996). Biochemistry, 35, 11660 ± 11667. Garrick D, Ayyub H and Higgs DR. (2000). Nature Carpenter NJ, Fillpovich A, Blaese M, Carey TL and Berkel Genet., 24, 368 ± 371. I. (1988). J. Pediat., 112, 757 ± 760. Goelz SE, Vogelstein B, Hamilton SR and Feinberg AP. Chen RZ, Pettersson U, Beard C, Jackson-Grusby L and (1985). Science, 228, 187 ± 190. Jaenisch R. (1998). Nature, 395, 89 ± 93. Gonzalez-Zulueta M, Bender CM, Yang AS, Nguyen T-D, Cheng NC, Chan AJK, Beitsma MM, Speleman F, Beart RW, VanTornout JM and Jones PA. (1995). Cancer Westerveld A and Versteeg R. (1996). Hum. Mol. Genet., Res., 55, 4531 ± 4535. 5, 309 ± 317. Goto K, Numatas M, Momuram J-I, Ono Y, Bestor T and Chuang LS-H, Ian H-I, Koh T-W, Ng H-H, Xu G and Li Kondo H. (1993). Di€erentiation, 56, 39 ± 44. BFL. (1997). Science, 277, 1996 ± 2000.

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3154 Gra€ JR, Herman JG, Lapidus RG, Chopra H, Xu R, Kawano S, Miller CW, Gombart AF, Bartram CR, Matsuo Jarrard DF, Isaacs WB, Pitha PM, Davidson NE and Y, Asou H, Sakashita A, Said J, Tatsumi E and Koe‚er Baylin SB. (1995). Cancer Res., 55, 5195 ± 5199. HP. (1999). Blood, 94, 1113 ± 1120. Grassi M, Girault JM, Wang WP, Thiery JP and Jouanneau Kelman Z. (1997). Oncogene, 14, 629 ± 640. J. (1999). Gene, 233, 59 ± 66. Klimasauskas S, Kumar S, Roberts RJ and Cheng X. (1994). Hanahan D and Weinberg RA. (2000). Cell, 100, 57 ± 70. Cell, 76, 357 ± 369. Hansen RS, Stoger R, Wijmenga C, Stanek AM, Can®eld Kochanek S, Renz D and Doer¯er W. (1995). FEBS Lett., TK,LuoP,MatarazzoMR,D-EspositoM,FeilR, 360, 115 ± 120. Gimelli G, Weemaes CMR, Laird CD and Gartler SM. Kondo T, Bobek MP, Kuick R, Lamb B, Zhu X, Narayan A, (2000). Hum. Mol. Genet., 9, 2575 ± 2587. Bourc'his D, Viegas-Pequignot E, Ehrlich M and Hanash Hansen RS, Wijmenga C, Luo P, Stanek AM, Can®eld TK, SM. (2000). Hum. Mol. Genet., 9, 597 ± 604. Weemaes CMR and Gartler SM. (1999). Proc. Natl. Acad. Kumar S, Cheng X, Klimasauskas S, Mi S, Posfai J, Roberts Sci. USA, 96, 14412 ± 14417. RJ and Wilson GG. (1994). Nucleic Acids Res., 22, 1±10. Heim S and Mitelman F. (1995). Wiley-Liss: New York, Lee PJ, Washer LL, Law DJ, Boland CR, Horon IL and pp. 369 ± 388. Feinberg AP. (1996). Proc. Natl. Acad. Sci. USA, 93, Hendrich B and Bird A. (1998). Mol. Cell. Biol., 10, 6538 ± 10366 ± 10370. 6547. Lee W-H, Morton RA, Epstein JI, Brooks JD, Campbell PA, Herman JG, Jen J, Merlo A and Baylin SB. (1996). Cancer Bova GS, Hsieh W-s, Isaacs WB and Nelson WG. (1994). Res., 54, 722 ± 727. Proc. Natl. Acad. Sci. USA, 91, 11733 ± 11737. HermanJG,LatifF,WengY,LermanMI,ZbarB,LiuS, LeiH,OhSP,OkanoM,JuttermannR,GossKA,Jaenisch Samid D, Duan D-SR, Gnarra JR, Linehan WM and R and Li E. (1996). Development, 122, 3195 ± 3205. Baylin SB. (1994). Proc.Natl.Acad.Sci.USA,91, 9700 ± Leonhardt H, Page AW, Weier H and Bestor TH. (1992). 9704. Cell, 71, 865 ± 873. Herman JG, Umar A, Polyak K, Gra€ JR, Ahuja N, Issa J- Lewis JD, Meehan RR, Henzel WJ, Maurer-Fogy I, PJ, Markowitz S, Willson JKV, Hamilton SR, Kinzler Jeppesen P, Klein F and Bird A. (1992). Cell, 69, 905 ± 914. KW,KaneMF,KolodnerRD,VogelsteinB,KunkelTA Li E, Beard C and Jaenisch R. (1993). Nature, 366, 362 ± 365. and Baylin SB. (1998). Proc. Natl. Acad. Sci. USA, 98, Li E, Bestor TH and Jaenisch R. (1992). Cell, 69, 915 ± 926. 6870 ± 6875. Li L and Cohen SN. (1996). Cell, 85, 319 ± 329. Hiltunen MO, Alhonen L, Koistinaho J, Myohanen S, Liu Y, Oakeley EJ, Sun L and Jost J-P. (1998). Nucleic Acids Paakkonen M, Marin S, Kosma VM and Janne J. Res., 26, 1038 ± 1045. (1997). Int. J. Cancer, 70, 644 ± 648. Lyko F, Ramsahoye BH and Jaenisch R. (2000). Nature, 408, Hittelman AB, Burakov D, Iniguez-Lluhi JA, Freeman LP 538 ± 540. and Garabedian MJ. (1999). EMBO J., 18, 5380 ± 5388. Lyko F, Ramsahoye BH, Kashevsky H, Tudor M, Mas- Holt SE, Aisner DL, Baur J, Tesmer VM, Dy M, Ouellette trangelo M-A, Orr-Weaver TL and Jaenisch R. (1999). M, Trager JB, Morin GB, Toft DO, Shay JW, Wright WE Nature Genet., 23, 363 ± 366. and White MA. (1999). Genes. Dev., 13, 817 ± 826. Maloisel L and Rossignol J-L. (1998). Genes. Dev., 12, Hough RF and Bass BL. (1997). RNA, 3, 356 ± 370. 1381 ± 1389. Hsieh C-L. (1999). Mol. Cell. Biol., 19, 8211 ± 8218. MerloA,HermanJG,MaoL,LeeDJ,GabrielsonE,Burger Hsieh C-L and Lieber MR. (1992). EMBO J., 11, 3115 ± 325. PC, Baylin SB and Sidransky D. (1995). Nature Med., 1, Issa J-P, Ottaviano YL, Celano P, Hamilton SR, Davidson 686 ± 692. NE and Baylin SB. (1994). Nature Genet., 7, 536 ± 540. Mertens F, Johansson B, Hoglund M and Mitelman F. IssaJ-P,VertinoPM,WuJ,SazawalS,CelanoP,Nelkin (1997). Cancer Res., 57, 2765 ± 2780. BD, Hamilton SR and Baylin SB. (1993). J. Natl. Cancer Miniou P, Jeanpierre M, Blanquet V, Sibella V, Bonneau D, Inst., 85, 1235 ± 1240. Herbelin C, Fischer A, Niveleau A and Viegas-Pequignot Jeanpierre M, Turleau C, Aurias A, Prieur M, Ledeist F, E. (1994). Hum. Mol. Genet., 3, 2093 ± 2102. Fischer A and Viegas-Pequignot E. (1993). Hum. Mol. Miyata Y and Yahara I. (1991). J. Biol. Chem., 266, 8779 ± Genet., 2, 731 ± 735. 8783. Jeddeloh JA, Stokes TL and Richards EJ. (1999). Nature Moran E. (1993). Curr. Opin. Genet. Dev., 3, 63 ± 70. Genet., 22, 94 ± 97. Myohanen SK, Baylin SB and Herman JG. (1998). Cancer Ji W, Hernandez R, Zhang XY, Qu GZ, Frady A, Varela M Res., 58, 591 ± 593. and Ehrlich M. (1997). Mutat. Res., 379, 33 ± 41. Nan X, Campoy FJ and Bird A. (1997). Cell, 88, 471 ± 481. Johnson JL, Beito TG, Krco CJ and Toft DO. (1994). Mol. NanX,NgH-H,JohnsonCA,LahertyCD,TurnerBM, Cell. Biol., 14, 1956 ± 1963. Eisenman RN and Bird A. (1998). Nature, 393, 386 ± 389. Jones PA. (1996). Cancer Res., 56, 2463 ± 2467. NarayanA,JiW,ZhangX-Y,MarrogiA,Gra€JR,Baylin Jones PA and Laird PW. (1999). Nature Genet., 21, 163 ± 166. SB and Ehrlich M. (1998). Int. J. Cancer, 77, 833 ± 838. Jones PL, Veenstra GJC, Wade PA, Vermaak D, Kass SU, Narayan A, Tuck-Muller C, Weissbecker K, Smeets D and Landsberger N, Strouboulis J and Wol€e AP. (1998). Ehrlich M. (2000). Mutat. Res., 456, 1 ± 15. Nature Genet., 19, 187 ± 191. Ng H-H, Zhang Y, Hendrich B, Johnson CA, Turner BM, Kane MF, Loda M, Gaida GM, Lipman J, Mishra R, Erdjument-Bromage H, Tempst P, Reinberg D and Bird Goldman H, Jessup JM and Kolodner R. (1997). Cancer A. (1999). Nature Genet., 23, 58 ± 61. Res., 57, 808 ± 811. Nichols AF and Sancar A. (1992). Nucleic Acids Res., 20, Kastan MB, Gowans BJ and Lieberman MW. (1982). Cell, 2441 ± 2446. 30, 509 ± 615. Nuovo GJ, Plaia TW, Belinsky SA, Baylin SB and Herman Katzenellenbogen RA, Baylin SB and Herman JG. (1999). JG. (1999). Proc. Natl. Acad. Sci. USA, 96, 12754 ± 12759. Blood, 93, 4347 ± 4353. Ohsawa K, Imai Y, Ito D and Kohsaka S. (1996). Kaufman PD, Kobayashi R, Kessler N and Stillman B. J. Neurochem., 67, 89 ± 97. (1995). Cell, 81, 1105 ± 1114.

Oncogene DNA methylation, methyltransferases, and cancer KD Robertson 3155 Okano M, Bell DW, Haber DA and Li W. (1999). Cell, 99, Stewart AF and Gibson T. (1996). TIBS, 21, 87 ± 88. 247 ± 257. Stirzaker C, Millar DS, Paul CL, Warnecke PM, Harrison J, Okano M, Xie S and Li E. (1998a). Nature Genet., 19, 219 ± Vincent PC, Frommer M and Clark SJ. (1997). Cancer 220. Res., 57, 2229 ± 2237. Okano M, Xie S and Li E. (1998b). Nucleic Acids Res., 26, Szyf M, Bozovic V and Tanigawa G. (1991). J. Biol. Chem., 2536 ± 2540. 266, 10027 ± 10030. PickettsDJ,HiggsDR,BachooS,BlakeDJ,QuarrellOWJ Tagarro I, Fernandez-Peralta AM and Gonzales-Aguilera and Gibbons RJ. (1996). Hum. Mol. Genet., 5, 1899 ± 1907. JJ. (1994). Hum. Genet., 93, 383 ± 388. Pradhan S, Bacolla A, Wells RD and Roberts RJ. (1999). Tate PH and Bird AP. (1993). Curr. Opin. Genet. Dev., 3, J. Biol. Chem., 274, 33002 ± 33010. 226 ± 231. PradhanS,TalbotD,ShaM,BennerJ,HornstraL,LiE, Tatematsu K-i, Yamazaki T and Ishikawa F. (2000). Genes Yaenisch R and Roberts RJ. (1997). Nucleic Acids Res., to Cells, 5, 677 ± 688. 25, 4666 ± 4673. Tucker KL, Talbot D, Lee MA, Leonhardt H and Jaenisch Puget N, Torchard D, Serova-Sinilnikova OM, Lynch HT, R. (1996). Proc. Natl. Acad. Sci. USA, 93, 12920 ± 12925. Feunteun J, Lenoir GM and Mazoyer S. (1997). Cancer Tuck-Muller CM, Narayan A, Tsien F, Smeets DFCM, Res., 57, 828 ± 831. Sawyer J, Fiala ES, Sohn OS and Ehrlich M. (2000). Qian XC and Brent TP. (1997). Cancer Res., 57, 3672 ± 3677. Cytogenet. Cell Genet., 89, 121 ± 128. Qu G-z, Dubeau L, Narayan A, Yu MC and Ehrlich M. Umar A, Buermeyer AB, Simon JA, Thomas DC, Clark AB, (1999). Mutat. Res., 423, 91 ± 101. Liskay RM and Kunkel TA. (1996). Cell, 87, 65 ± 73. Ramsahoye BH, Biniszkiewicz D, Lyko F, Clark V, Bird AP Vertino PM, Yen R-WC, Gao J and Baylin SB. (1996). Mol. and Jaenisch R. (2000). Proc. Natl. Acad. Sci. USA, 97, Cell. Biol., 16, 4555 ± 4565. 5237 ± 5242. Vogelstein B, Lane D and Levine AJ. (2000). Nature, 408, Raynal P, Kuijpers G, Rojas E and Pollard HB. (1996). 307 ± 310. FEBS Lett., 392, 263 ± 268. WadePA,GegonneA,JonesPL,BallestarE,AubryFand Rhee I, Jair K-W, Yen R-WC, Lengauer C, Herman JG, Wol€e AP. (1999). Nature Genet., 23, 62 ± 66. Kinzler KW, Vogelstein B, Baylin SB and Schuebel KE. Waga S, Hannon GJ, Beach D and Stillman B. (1994). (2000). Nature, 404, 1003 ± 1007. Nature, 369, 574 ± 578. Rice JC and Futscher BW. (2000). Nucleic Acids Res., 28, Wake®eld RID, Smith BO, Nan X, Free A, Soteriou A, 3233 ± 3239. Uhrin D, Bird AP and Barlow PN. (1999). J. Mol. Biol, Robertson KD, Ait-Si-Ali S, Yokochi T, Wade PA, Jones PL 291, 1055 ± 1065. and Wol€e AP. (2000a). Nature Genet., 25, 338 ± 342. Walsh CP, Chaillet JR and Bestor TH. (1998). Nature Genet., Robertson KD and Jones PA. (1998). Mol. Cell. Biol., 18, 20, 116 ± 117. 6457 ± 6473. Wijmenga C, van den Heuvel LPWJ, Strengman E, Luyten Robertson KD, Keyomarsi K, Gonzales FA, Velicescu M JAFM, van der Burgt IJAM, de Groot R, Smeets DFCM, and Jones PA. (2000b). Nucleic Acids Res., 28, 2108 ± Draaisma JMT, van Dongen JJ, De Abreu RA, Pearson 2113. PL, Sandkuijl LA and Weemaes CMR. (1998). Am. J. Robertson KD, Uzvolgyi E, Liang G, Talmadge C, Sumegi J, Hum. Genet., 63, 803 ± 809. Gonzales FA and Jones PA. (1999). Nucleic Acids Res., 27, Wong DJ, Foster SA, Galloway DA and Reid BJ. (1999). 2291 ± 2298. Mol. Cell. Biol., 19, 5642 ± 5651. Robertson KD and Wol€e AP. (2000). Nature Rev. Genet., 1, Wu J, Issa JP, Herman J, Bassett DE, Nelkin BD and Baylin 11 ± 19. SB. (1993). Proc. Natl. Acad. Sci. USA, 90, 8891 ± 8895. Rouleau J, Tanigawa G and Szyf M. (1992). J. Biol. Chem., XieS,WangZ,OkanoM,NogamiM,LiY,HeW-W, 267, 7368 ± 7377. Okumura K and Li E. (1999). Gene, 236, 87 ± 95. Rountree MR, Bachman KE and Baylin SB. (2000). Nature Xu G-L, Bestor TH, Bourc'his D, Hsieh C-L, Tommerup N, Genet., 25, 269 ± 277. Bugge M, Hulten M, Qu, Russo JJ and Viegas-Pequignot Rouyer F, Simmler MC, Page D and Weissenbach J. (1987). E. (1999). Nature, 402, 187 ± 191. Cell, 51, 417 ± 425. Yoder JA and Bestor TH. (1998). Hum. Mol. Genet., 7, 279 ± Sakai T, Toguchida J, Ohtani N, Yandell DW, Papaport JM 284. and Dryja TP. (1991). Am.J.Hum.Genet.,48, 880 ± 888. Yoder JA, Walsh CP and Bestor TH. (1997). TIG, 13, 335 ± Sawyer JR, Swanson CM, Wheeler G and Cunni€ C. (1995). 340. Am.J.Med.Genet.,56, 203 ± 209. Yoder JA, Yen R-WC, Vertino PM, Bestor TH and Baylin Shibahara K-i and Stillman B. (1999). Cell, 96, 575 ± 585. SB. (1996). J. Biol. Chem., 271, 31092 ± 31097. Small K, Iber J and Warren ST. (1997). Nature Genet., 16, Yoshiura K, Kanai Y, Shimoyama Y, Sugimura T and 96 ± 99. Hirohashi S. (1995). Proc. Natl. Acad. Sci. USA, 92, Smeets DFCM, Moog U, Weemaes CMR, Vaes-Peeters G, 7416 ± 7419. Merkx GFM, Niehof JP and Hamers G. (1994). Hum. Zhang X and Verdine GL. (1996). FEBS Lett., 392, 179 ± Genet., 94, 240 ± 246. 183.

Oncogene