An alternate, indirect mechanism in which genomic destabilization occurs is through DNA methylation-induced global changes in gene expression that alter transcriptional regulation of proteins responsible for repeat stabilization (Table 1), an effect that has been observed in cancer. In general, the genome of tumor cells is hypomethylated, particularly in regions of repetitive DNA. Widespread hypomethylation contributes to a hyper-mutation state in cancer, which may generate oncogenic mutations or loss of heterozygosity of tumor suppressor genes. Additionally, tumor cells also demonstrate sites of local hypermethylation, frequently at CpG islands that would, under normal circumstances, be hypomethylated (Robertson and Wolffe 2000). Site-specific CpG island hypermethylation affects genomic stability when it targets promoters of MMR genes. Along these lines, MSI is a common pathway to tumor development, and many MSI tumors contain mutations in MMR genes or epigenetic repression of MMR genes (Laghi et al. 2008).

An underlying methylation defect has been described in a subset of colorectal cancers in which promoter CpG islands are hypermethylated at high frequency, a condition termed the CpG island methylator phenotype (CIMP; Issa 2004; Toyota et al. 1999). One class of CIMP tumors displays MSI and concurrent inactivation of MLH1 by promoter hypermethylation (Issa 2004). The first evidence of CIMP came from the finding that colorectal tumors lacking MLH1 expression displayed MLH1 promoter hypermethylation but no MLH1 mutations (Kane et al. 1997), and subsequent reports have characterized MSI tumors as having a strong association with promoter hypermethylation compared to non-MSI tumors (Ahuja et al. 1997; Veigl et al. 1998). Moreover, experiments introducing a transgenic reporter gene into cancer cell lines revealed that expression of the reporter was associated with genetic instability: upon introduction of the transgene, cell lines deficient in MMR showed strong repression of the reporter gene compared to MMR proficient lines (Lengauer et al. 1997). The contrasting behavior of the transgene was linked to cell line-specific methylation differences that were negated by 5-azadC treatment, rescuing transgene expression (Lengauer et al. 1997). These reports are consistent with the idea that CIMP is associated with increased DNMT1 expression or enhanced DNA methyltransferase activity (Teodoridis et al. 2008). Additional evidence from experiments in mice has further highlighted the importance of DNA methyltransferase activity in CIMP tumors: mice deficient for Mlh1 expression due to Mlh1 promoter hypermethylation were less susceptible to intestinal tumor formation when Dnmt1 function was also reduced (Trinh et al. 2002). Interestingly, although Mlh1/Dnmt1-deficient mice were less susceptible to intestinal tumorigenesis, they were more susceptible to lymphoid tumor development, a result linked to hypomethylation-induced genomic instability (Trinh et al. 2002). In Dnmt1-deficient mice, a similar phenotype was observed including genomic hypomethylation, increased lymphomagenesis, and marked genomic instability, particularly trisomy of chromosome 15 (Gaudet et al. 2003). Thus, DNA methylation is a double-edged sword in maintaining proper growth controls in that too much methylation enhances tumor-associated repression of repair genes, but too little methylation enhances genomic instability in repetitive regions that require a heterochromatic state for their stability. Obtaining a better understanding of the mechanisms of maintaining the proper balance and distribution of methyl marks in normal tissues will be critical for understanding how disruption of this balance contributes to cancer.

Trinucleotide repeat expansion is a hallmark of more than 20 human diseases including fragile X syndrome, Huntington's disease, and myotonic dystrophy (Mirkin 2007). These diseases are characterized by an “anticipation” phenomenon in which microsatellite expansion and disease severity worsen with each generation (Mirkin 2007). DNA methylation defects may directly contribute to the etiology of some repeat expansion diseases, because the developmental stages in which MSI occurs coincide with two key periods during which epigenetic reprogramming occurs (Gorbunova et al. 2004; Pearson 2003; Reik et al. 2001; Yoon et al. 2003). Moreover, DNA methylation was linked to stabilization of the CGG repeats whose expansion causes fragile X syndrome (Wohrle et al. 1998).

A key question is how do repeat expansions cause disease? One mechanism is through disruption of gene function: microsatellite expansions residing in exon or intron sequences disrupt the normal function of the gene product. Repeat expansions also produce local epigenetic changes. These expansions disrupt local chromatin structure and alter the expression of adjacent genes, an effect that has been extensively studied in myotonic dystrophy patients (Gatchel and Zoghbi 2005; Mirkin 2007). Myotonic dystrophy is associated with expansion of a CAG repeat within the 3′UTR of the DMPK gene. Repeat length is inversely correlated with expression of the flanking genes DMPK and SIX5/DMAHP, both of which have been implicated in the pathology of myotonic dystrophy (Klesert et al. 2000, 1997; Thornton et al. 1997). Analysis of chromatin from skeletal muscle of individuals with myotonic dystrophy showed that the region surrounding the CAG expansion was protected from DNase digestion (Otten and Tapscott 1995), presumably due to a change in chromatin structure resulting in chromatin condensation (Otten and Tapscott 1995). The normal CAG locus was also shown to contain a region of H3K9 methylated heterochromatin embedded in a euchromatic H3K4 methylated region (Cho et al. 2005). The expanded repeat allele, in contrast, was associated with spreading of heterochromatin (and H3K9 methylation) into gene containing regions. Heterochromatin boundaries are normally limited by binding of the zinc-finger insulator protein CTCF; CTCF binding, however, is lost at the CAG boundaries in myotonic dystrophy (Cho et al. 2005; Filippova et al. 2001). Methylation of CTCF DNA binding sites prevents CTCF binding, which may, in turn, permit the spread of heterochromatin into surrounding euchromatic regions (Filippova et al. 2001).

DNA methylation is well known to protect against unlicensed homologous recombination (HR), which is the basis for many mutational events. Following generation of a DNA double-strand break (DSB), HR is employed to form a synapse between homologous chromosomes, a circumstance that can generate gene conversions, insertions or deletions, or more extensive chromosomal loss. Transcription enhances HR frequency by exposing single-stranded DNA and allowing it to invade its homologous sequence, while DNA methylation-mediated transcriptional silencing inhibits HR (Dominguez-Bendala and McWhir 2004; Ikeda and Matsumoto 1979; Maloisel and Rossignol 1998). Several reports support the idea that DNMT1 represses HR. Dnmt1 deficiency in mice resulted in elevated loss of heterozygosity due to mitotic recombination (Eden et al. 2003), Dnmt1 deficiency in mouse ES cells enhanced gene targeting by HR (Dominguez-Bendala and McWhir 2004), and embryonic carcinoma cells carrying integrated transgenic reporter constructs displayed increased recombination among transgenic sequences in response to depletion of DNA methylation by 5-azadC (McBurney et al. 2001). In a complementary study, hypermethylation of CpG islands within extrachromosomal fragments reduced the occurrence of deletions within and dimerizations between fragments, presumably due to reduced HR (Rizwana and Hahn 1999). Finally, DNMT1 and DNMT3 activity at repetitive pericentromeric and centromeric loci effectively limits mitotic recombination frequency and stabilizes centromeric repeats against length changes (Jaco et al. 2008).

In addition to the role of DNA methylation in prevention of HR, DNMT1 unexpectedly performs a role in DSB repair that appears to be independent of its maintenance methylation activity. DNMT1 has been shown to either assist in DSB detection or operate in the repair process itself. Induction of DSBs by irradiation or by 5-azadC shows an accumulation of DNMT1, its binding partners PCNA and MBD4, and the repair-associated factor MLH1 at foci of DSBs, and DNMT1 deficiency reduces the recruitment of DNA repair proteins to these foci (Mortusewicz et al. 2005; Palii et al. 2008; Ruzov et al. 2009). The well-established PCNA-DNMT1 interaction has recently been shown to be nonessential for DNMT1's maintenance methylation function (Schermelleh et al. 2007; Spada et al. 2007), and DNMT1 associates with chromatin independently of replication (Easwaran et al. 2004), suggesting a role for the PCNA-DNMT1 interaction separate from maintenance methylation. PCNA acts in multiple DNA repair processes including MMR and nucleotide excision repair, and thus DNMT1 may facilitate this function (Mortusewicz et al. 2005; Nichols and Sancar 1992; Umar et al. 1996). In summary, DNMT1 functions in the DNA damage response, perhaps through a methylation-independent role, highlighting an important and novel function for DNMT1 that deserves further examination.

DSB repair must operate in the context of complex higher-order chromatin structure and orchestrate restoration of chromatin and epigenetic marks at the site of DNA damage. Chromatin remodeling, like repair of the DNA itself, is critical for maintaining genomic integrity. Decondensation of chromatin in response to DNA damage may be required to loosen or open the chromatin structure and allow access for repair machinery (Ziv et al. 2006). Consistent with this notion, in Saccharomyces cerevisiae, HATs, including Gcn5 and Esa1, are actively recruited to sites of DSBs presumably to assist in promoting open chromatin (Groth et al. 2007; Tamburini and Tyler 2005; van Attikum and Gasser 2005). This activity is only transient, and HDACs rapidly remove acetylation marks likely to prevent aberrant transcriptional activity in the newly repaired region (Tamburini and Tyler 2005). Chromatin remodeling following DNA damage is also mediated by the DNA damage checkpoint kinase CHK1. CHK1 is normally associated with chromatin in undamaged cells and specifically phosphorylates histone H3 at threonine 11 (H3T11) promoting acetylation of H3K9 and continued maintenance of cell cycle regulatory genes in a transcriptionally active state. Upon DNA damage, CHK1 dissociates from chromatin, and levels of phosphorylated H3T11 and acetylated H3K9 decline (Shimada et al. 2008; Smits et al. 2006). This deacetylation may function to repress transcription at the site of damage, providing protection for the decondensed chromatin against further vulnerability to mutations during the repair process.

Another critically important histone modification associated with DNA damage is phosphorylation of H2A, particularly at serine 129 and serine 1 (van Attikum and Gasser 2005). Recruitment of phosphorylated H2A.X (γH2A.X), a variant of histone H2A, to sites of DSBs is a critical step in the repair process (Celeste et al. 2003, 2002). γH2A.X-containing nucleosomes assemble along the regions bordering DSBs, as observed by immunofluorescence microscopy, and aid in the recruitment and/or maintenance of DNA break repair factors and histone modifiers at the site of damage (Rogakou et al. 2000, 1998; van Attikum and Gasser 2005). Notably, recruitment of γH2A.X containing histones to sites of DSBs is also DNMT1 dependent (Palii et al. 2008). In addition, in both fertilized zygotes and cloned embryos, the occurrence of γH2A.X-populated DSBs increased in a temporal pattern that coincided remarkably well with the period of active, DNA replication-independent demethylation during zygotic pronuclear development (Wossidlo et al. 2010). Although the mechanism underlying this connection is unclear, the association between these two events and the requirement for DNMT1 in the recruitment of γH2A.X is consistent with a functional involvement of DNA methylation in DNA repair. It will be of great interest to determine the role of DNMT1 at sites of DNA damage and whether it is methylation dependent, as this may provide important information on how DNA methylation inhibitors exert their antitumor effects.

A functional centromere is vital for guarding against chromosomal instability during cell division by maintaining proper DNA segregation, averting chromosomal rearrangements, and preventing changes in ploidy. Cytologically, the centromere appears as a chromosomal constriction, but it actually consists of a molecularly unique and complex configuration of chromatin. Centromeric chromatin forms a densely compact structure that provides attachment sites for kinetochore proteins and spindle microtubules. The flanking pericentromeric chromatin recruits a high concentration of cohesin molecules providing the site for sister chromatid attachment (Allshire and Karpen 2008).

With every cell generation, each chromosome's centromere is habitually re-generated at the same location. Providing a single site for kinetochore attachment is necessary to ensure proper chromosomal segregation and stability, and numerous lines of evidence have established that centromere location/identity is determined epigenetically (Allshire and Karpen 2008; Amor and Choo 2002; Henikoff et al. 2001; Karpen and Allshire 1997; Sullivan et al. 2001). For example, cross species comparison shows that centromeric DNA sequences vary widely suggesting the lack of a single centromere recognition sequence. Likewise, human α-satellite repeat sequence within centromeric DNA is insufficient to generate a functional centromere, since chromosomes that contain two regions of centromeric α-satellite DNA, still possess only one functional kinetochore attachment region. Additionally, in the absence of a functional centromere, a neocentromere, which is a functional ectopic centromere, may form at alternate non-centromeric genetic loci that are devoid of α-satellite repeats (Amor and Choo 2002; Choo 2000). Having epigenetic centromeric determinants as opposed to sequence-specific determinants is advantageous because it provides a reliable mechanism for transmitting genetic material in the event of mutational damage to the centromere locus. Epigenetic determination also allows for adaptive karyotypic evolution and may provide an important mechanistic means for speciation (Henikoff et al. 2001). Thus, the epigenetically determined centromere provides a versatile system for consistent and faithful transmission of chromosomal material and for accommodation of karyotype alterations.

Human centromeres are characterized by several thousand kilobases of a 171-base pair AT-rich repeat referred to as α-satellite DNA (Amor and Choo 2002). Satellite DNA sequences typically do not possess similarity among chromosomes or across species, but their DNA-binding proteins are frequently conserved. One such important conserved centromeric chromatin protein is CENP-A, a centromere-specific core histone that substitutes for histone H3 in the nucleosome core (Palmer et al. 1987). Centromeric chromatin is composed of alternating blocks of H3 and CENP-A-containing nucleosomes (Blower et al. 2002). The structural significance of this CENP-A distribution is not known, but it may impart directionality onto the centromere structure that is necessary for kinetochore attachment and for bi-directional, anaphase migration of sister chromatids (Allshire and Karpen 2008). Importantly, CENP-A is necessary and sufficient for recruitment of kinetochore proteins for mitosis (Palmer et al. 1991; Palmer et al. 1987). CENP-A deficiency disrupts kinetochore assembly and progression of mitosis, whereas CENP-A overexpression induces ectopic spreading of centromeric chromatin to non-centromeric locations and subsequent redistribution of kinetochore proteins (Allshire and Karpen 2008; Sullivan et al. 2001). Interestingly, CENP-A is recruited to DSBs, and cell survival following generation of DSBs is proportional to CENP-A activity (Zeitlin et al. 2009). Thus, CENP-A is multifaceted in its involvement in genomic maintenance and stability. H2A.Z, another histone variant replacing H2A in non-CENP-A-containing centromeric and pericentromeric nucleosomes, is also required for centromere structure and sister chromatid adhesion prior to chromosomal segregation (Greaves et al. 2007). The interplay between CENP-A and H2A.Z histone variants along with their unique arrangements of histone modifications (see below) appear to be key determinants in the organization and function of the centromere.

Centromeric and pericentromeric chromatin contain unique and largely non-overlapping configurations of histone marks that are critical for centromere function (Fig. 1). Interestingly, despite the highly condensed nature of centromeric chromatin, its array of histone modifications is quite distinct from constitutive heterochromatin. Centromeric heterochromatin contains a characteristic combination of both euchromatic and heterochromatic marks (Lam et al. 2006; Sullivan and Karpen 2004). For example, condensed transcriptionally inert heterochromatin is defined by H3K9 di- and tri-methylation, but immunofluorescence studies showed that centromeric chromatin, despite being highly condensed, does not contain these marks (Sullivan and Karpen 2004). Surprisingly, CENP-A containing centromeric chromatin contains H3K4me2, a mark associated with euchromatin and transcriptional permissiveness (Sullivan and Karpen 2004). However, other marks which confer an “open” euchromatic and transcriptionally active signature, such as H3K4me3 and histone acetylation, are absent from centromeres (Sullivan and Karpen 2004). Pericentromeric heterochromatin, on the other hand, more closely resembles canonical heterochromatin. It is hypoacetylated and enriched for both H3K9me2 and H3K9me3; however, H3K9me3 domains are located distally from the centromere, a configuration that may function to establish centromeric boundaries (Sullivan and Karpen 2004). Consistent with a functional role for H3K9 trimethylation in pericentromeric heterochromatin, the H3K9 methylases, SUV39H1 and SUV39H2, are critical for mitotic progression. SUV39H1 and SUV39H2 accumulate at centromeric domains during mitosis, and their loss, or the loss of H3K9 methylation, in pericentromeric regions is detrimental to chromosome segregation (McManus et al. 2006; Melcher et al. 2000; Peters et al. 2003, 2001; Rice et al. 2003). Also within pericentromeric domains, two additional histone methylases, SUV4-20H1 and SUV4-20H2, were shown to recognize H4K20 substrates for methylation, a modification necessary for chromosomal condensation and stability during mitosis (Oda et al. 2009; Schotta et al. 2004). SUV39H1/2 and SUV4-20H1/2 enzymes are also associated with recruitment and stabilization of heterochromatic protein 1 (HP1) to pericentromeric chromatin to mediate gene silencing and chromosomal condensation (Bannister et al. 2001; Lachner et al. 2001; Maison et al. 2002; Schotta et al. 2004; Souza et al. 2009). Suv39H1/2 activity is also important for CENP-A assembly into centromeric chromatin, since the fission yeast homolog of SUV39, Clr4, along with HP1 and other heterochromatin associated factors, is necessary for loading CENP-A nucleosomes into centromeric regions (Folco et al. 2008). These studies emphasize how critical the distinctive patterns of centromeric and pericentromeric histone marks and their associated methylases are for the proper recruitment and stabilization of centromeric proteins. Global changes in histone methylation are also critical for whole chromosomal condensation and centromere function during mitosis. Inhibition of overall histone methyltransferase activity just prior to prophase causes defects in chromosome alignment and segregation due to expansions and loosening of centromeric chromatin, unstable microtubule associations with kinetochores, and improper kinetochore protein localization (Heit et al. 2009). Fig. 1

Chromosomal definition by histone modifications. Many molecules, including histone-modifying enzymes, histone variants, and DNA methyltransferases, some of which are depicted here on the left, establish the epigenetic marks and chromatin structure required for chromosomal organization and function. On the right, distinct combinations of histone modifications define chromatin regions including the centromere and pericentromere that organize and stabilize the chromosome as a whole and maintain genetic stability throughout repeated cell divisions

Centromeric and pericentromeric DNA is heavily methylated at CpG dinucleotides, a status that persists throughout the cell cycle and is important for centromere function (Monier et al. 2007; Wong et al. 2006). The requirement for this centromere-specific methylation is quite apparent in diseases characterized by chromosomal segregation defects, such as some cancers and ICF syndrome. These diseases, which demonstrate large-scale gain or loss of chromosomes, are often associated with loss of DNA methylation at centromeric and/or pericentromeric loci (Gisselsson et al. 2005; Hansen et al. 1999; Lengauer et al. 1997; Wong et al. 2001). For example, a subset of hepatocellular carcinomas has been defined by particular chromosome 1 rearrangements stemming from a pericentromeric breakpoint, which was highly correlated with hypomethylation of pericentromeric satellite-2 CpG dinucleotides (Wong et al. 2001). The requirement for DNA methylation is also apparent in patients with ICF syndrome. ICF syndrome patients exhibit hypomethylation of heterochromatin, particularly of pericentromeric satellite-2 repeats within chromosomes 1, 9, and 16, resulting in decondensation of pericentromeric chromatin (Gisselsson et al. 2005; Hansen et al. 1999; Miniou et al. 1994, 1997). DNMT3B's PWWP domain is essential for methylation of these repeats, and significantly, the majority of ICF syndrome patients have mutations in DNMT3B that at least partially disrupt its methylation activity (Chen et al. 2004; Hansen et al. 1999; Xu et al. 1999).

The effects of DNA methylation and DNMT activity in pericentromeric and centromeric heterochromatin formation appear to be mediated through histone-modifying enzymes, heterochromatin factors, and centromeric proteins. Depletion of DNMT1 prevents accumulation of H3S10P foci, perhaps by affecting the subnuclear localization or the binding affinity of Aurora kinases to the centromere (Monier et al. 2007). DNMT3B also facilitates centromeric heterochromatin formation and chromosomal condensation through an association with condensin complexes (Geiman et al. 2004). DNMT3A and DNMT3B are recruited by H3K9me3-containing nucleosomes and SUV39H1 to pericentromeric domains, where they also interact with the heterochromatin factors HP1β and HP1α (Fuks et al. 2003; Geiman et al. 2004; Lehnertz et al. 2003).

DNA methylation and centromere function are further linked by the recent finding that DNMT3B interacts with the constitutive centromere protein CENP-C. CENP-C is an essential kinetochore protein that binds centromeric α-satellite DNA and is required for kinetochore assembly during mitosis (Kalitsis et al. 1998; Kwon et al. 2007; Politi et al. 2002). DNMT3B and CENP-C interact and become co-enriched at centromeric foci, reaching peak enrichment during metaphase (Gopalakrishnan et al. 2009). Importantly, this co-enrichment requires the mutual activity of both factors, since siRNA knockdown of either DNMT3B or CENP-C reduces the other's binding to centromeric and pericentromeric regions. This enrichment also facilitates DNMT3B's enzymatic activity, since CENP-C knockdown resulted in significant reduction in DNA methylation at both centromeric α-satellite DNA and pericentromeric satellite-2 sequence. Both CENP-C and DNMT3B are required for HP1α recruitment to centromeric and pericentromeric regions, and their loss affects chromosomal segregation and mitotic progression. Based on these findings, it is possible that loss of centromeric CENP-C and HP1α recruitment in DNMT3B-deficient ICF patients contributes to the marked centromeric instability that is a hallmark of this disorder (Gopalakrishnan et al. 2009).

Importantly, DNA methylation within the centromere is not entirely ubiquitous and unrestrained, but it instead consists of specific non-methylated sites within extensively methylated regions. Non-methylated regions are particularly important for the kinetochore protein CENP-B. CENP-B, which interacts with both CENP-A and CENP-C, binds a specific 17-base pair sequence within α-satellite DNA and provides an important mechanism for centromere identity (Ando et al. 2002; Masumoto et al. 1989). Loss of CENP-B does not affect kinetochore assembly or function (Choo 2001; Hudson et al. 1998; Kapoor et al. 1998; Perez-Castro et al. 1998), but it instead functions to effectively limit multi-centromere formation and promote heterochromatin formation via SUV39H1 recruitment and DNA methylation (Okada et al. 2007). The centromeric pattern of methylation consisting of unmethylated CENP-B binding sites intermingled with extensively methylated regions is important for these functions since aberrant DNA methylation or 5-azadC treatment disrupts binding of CENP-B to these loci and alters its distribution (Miniou et al. 1997; Mitchell et al. 1996; Okada et al. 2007; Tanaka et al. 2005).

Interestingly, centromeric and pericentromeric DNA methylation is distinctive in that its function is not simply to promote widespread transcriptional silencing. Despite the dense DNA methylation in these regions, these genetic loci show signs of transcriptional activity (Lehnertz et al. 2003). Analysis of DNA methylation of neocentromeres showed that most CpGs are hypermethylated, except for small domains of hypomethylation corresponding to transcriptionally permissive sites (Wong et al. 2006). Furthermore, centromeres on human artificial chromosomes showed loss of function when transcription within the centromere was altered (Nakano et al. 2008). Either strong transcriptional repression or broad transcriptional activation caused loss of CENP-A, -B, and -C binding, loss of kinetochore assembly, and chromosomal missegregation (Nakano et al. 2008). Thus, studies using artificial chromosomes and neocentromeres suggest that a delicate balance between transcriptional activity and repression is required to maintain centromere functionality.