The Importance of Aberrant DNA Methylation in Cancer

Cancer has long been considered primarily a genetic disease, caused by different mutations throughout the genome. In 1983., Feinberg and Vogelstein discovered that the level of DNA methylation significantly varies between primary human malignant tumors and their normal counterparts (Feinberg & Vogelstein, 1983). Before this publication, there had been a paper describing changes in DNA methylation in cancer cell cultures, including the influence of N-methyl-N-nitrosourea on the level of DNA methylation in Raji cells (Boehm & Drahovsky, 1981). Currently, we are presented with much experimental data showing multilevel changes in cancer cells. In this context, two major areas of epigenomic research DNA methylation and histone modifications appear most promising in understanding the multistep nature of carcinogenesis. Additionally, they seem to have the potential for being cancer biomarkers, useful in early detection, in predicting the biological behavior of tumors and for therapy monitoring, as recently reviewed (Rodriguez-Paredes & Esteller, 2011; Baylin & Jones, 2011). Finally, epigenetic changes are well-recognized targets for cancer therapy, alone, or in combination with various cytostatics (Ren et al, 2011). Epigenetic changes are also of the greatest importance in chemoprevention, as there is increasing data relating to possibly reversing epigenetic changes in the earliest phase of carcinogenesis, when genetic changes have yet to develop (reviewed, Huang et al., 2011). It is not easy to understand the particular rules applicable to epigenomic processes. If one specific epigenetic change, relating to a specific gene/its promoter, exists in a majority of tumors of a specific type, it does not necessarily mean that the same change exists in another type of tumor. This is consequential, and represents one reason for obvious differences in responses to epigenomic therapy. Recently, we wrote a review article on some aspects of epigenomic changes in which we used the term „epigenetic networking“. If we imagine each of our living cells as an orchestra performing the symphony of life, then each player (a gene) of the orchestra needs to play in concert with 30,000 other players. The communication that produces the network of our epi-genome is established at many levels: transcriptionally, post-transcriptionally, through protein translation, at the level of post-translational protein modifications, through their orchestrated interactions and, finally, their interaction with the DNA that can be modified in order to accept or reject the protein partner. This is the way of

The problem becomes even more prominent in comparative analyses, when tumor tissue needs to be compared to non-tumorous, adjacent tissue. Our group was not the only one that has shown, unexpectedly, the change in imprinting status of IGF2 in a tissue adjacent to laryngeal cancer (Grbeša et al., 2008). It seemed "normal" to the surgeon, and, back then in 2005, the simplest thought was that we mixed up tumorous and non-tumorous samples. Even at that time, we were quite careful with tumor samples, as years of experience taught us to take only a small portion of tissue for analyses, leaving at least one small piece of tissue in our tumor bank (Spaventi et al. 1994). After obtaining confusing results, this residual piece of tissue was given to a very experienced pathologist who needed to answer our question: "Is this the tumor"? Morphologically, it was not the tumor. Epigenetically, it showed loss of IGF2 imprinting. Based on that finding -it did not appear as "normal tissue". We still think that, especially in smoking-related cancer, this specific change may be the first sign of "abnormality".
In addition to obvious problems relating to exploring the DNA methylation status in native tumors, there are also very specific problems when using cell cultures. As shown by Asada, who used several different rat liver cell lines (including a primary cell line), methylation level increases significantly after 10 passages. Hence, the authors concluded that, at least in their experimental model, "a cautious approach is required when cell lines are utilized to study methylation-related carcinogenesis, metastatic or tumoricidal mechanisms" (Asada et al., 2006).
Based on this brief but, hopefully, informative data relating to objective limitations of the system based on factors beyond our control, we enter the field of cancer epigenomics.

Cancer and DNA methyltransferases
DNA methyltransferases (DNMTs) are the only enzymes which have been shown to mediate the transfer of a methyl group from S-adenosylmethionine (SAM) to the C-5 position of cytosine, mainly in CpG nucleotides, in mammalian genomes. Although detected, cytosine methylation is very rare in the outside of CpG sequences, at least in differentiated cells. For example, 99.98% of all methylation in mature fibroblasts occur at CpG dimers. This number is significantly reduced in both embryonic and induced stem cells (Lister et al., 2009).
In mammals, there are four DNMTs: DNMT1, DNMT2, DNMT3A, DNMT3B. While DNMT1 has the highest importance in maintaining post-replicative DNA methylation patterns, DNMT3A and -3B are considered critical players in establishing de novo methylation patterns. They also assume maintenance activity. DNMT3L, discovered in 2000 (Aapola et al., 2000), is a regulatory factor for de novo methylation. Its amino acid sequence is very similar to that of DNMT3a and DNMT3b but lacks the residues required for DNA methyltransferase activity in the C-terminal domain.
It was shown that the fidelity in replicating methylation patterns in human non-cancerous, dividing cells reaches 99.85-99.92% per site, per generation in CGs reach promoters and 99.56-99.83% in CGs outside the promoters (Ushijima et al., 2003). Human cancer gastric cell lines showed decreased fidelity in maintaining the methylation pattern which manifested through an increased level of de novo DNA methylation in promoter regions of five tested cancer-related genes and 4-to 8-fold higher expression of de novo DNMT3B. This increase was highest in two cell lines that showed the highest level of decreased fidelity (Ushijama et al., 2005). The question remains: was the increased level of DNMT3B alone sufficient to induce so prominent change at the promoters of these genes?

DNA Methyltransferase 1 (DNMT1)
Homozygous knockout of DNMT1 is lethal to the embryo in mammals. On the other hand, studies on DNMT1-overexpression in embryonic stem cells also resulted in lethality of the embryo, suggesting that accurate expression of DNMT1 is a key factor in maintaining embryonic development (Biniszkiewicz et al., 2002).
For maintaining the methylation pattern during cell division, the cellular machinery uses DNMT1. After replication, 5-mC is present only on one parental DNA strand and the methylation of cytosines on the newly synthesized strand takes place on the cytosine that lies diagonally opposite to 5-mC in the parent DNA strand. Keeping the methylation pattern as inheritable modification that needed to be preserved during cell division was originally published in 1975 by three independent researchers/research groups (Holliday & Pugh 1975;Riggs, 1975). Since then, our knowledge has been significantly advancing, especially as a result of fast developing molecular techniques. However, it does not matter how rapidly our research progresses, the importance of the discovery published in 1975 remains astonishing, even from the most sophisticated molecular perspective.
During the S-phase of the cell cycle, DNMT1 was found to be localized to DNA replication foci through its interaction with proliferating cell nuclear antigen (PCNA). The precise cellcycle-dependent localization of DNMT1 depends on the protein UHRF1, also known as ICBP and NP95. This protein shows strong preferential binding to hemimethylated CG sites through its methyl DNA binding domain, and tethers DNMT1 to replication fork (Bostick et al., 2007;Sharif at al., 2007). The DNMT1 also interacts with histone deacetylases resulting in repressing gene expression or forming heterochromatin structure.

DNMT1 and post-translational modifications
Little is known about post-translational modification of DNMT1, that may, possibly, change its functioning, especially in cancer. There are several in vitro studies pointing out the www.intechopen.com The Importance of Aberrant DNA Methylation in Cancer 335 protein kinases involved in DNMT1 phosphorylation. From the perspective of cancer research, the AKT and PKC certainly are very promising candidates that may help us to better understand the functioning of DNMT1. Both kinases were shown to phosphorylate recombinant DNMT1 at Ser127. AKT additionally phosphorylates Ser143. This modification decreases the ability of DNMT1 to interact with PCNA and UHRF1. As a consequence, DNMT1 shows increased cell-cycle-dependent stability (Esteve et al., 2011).

DNMT2
DNMT2 is expressed in most human and mouse adult tissues (Goll & Bestor, 2005) and its role seems to be related to methylation of cytosine 38 (C38) of RNA Asp (Goll et al., 2006). There are no strong evidence on DNA methylation activity of DNMT2. The Dnmt2 defficient mouse embryonic stem cells do not show measurable alteration of genomic DNA methylation pattern (Okano et al., 1998). Additionally, in contrast to exclusive nuclear localization of Dnmt1 and Dnmt3, Dnmt2 is primarily localized in the cytoplasm of transfected mouse 3T3 fibroblasts (Goll et al., 2006).
The level of DNMT2 expression in human cancer cell line is quite variable: high in K562 (leukemia) and MCF-7 (breast cancer) and very low, almost undetectable in A549 (lung cancer ) and HepG2 cells (liver cancer) (Schaefer et al., 2009). It has been shown that the treatment with 5-azacytidine inhibits C38 methylation at RNA Asp . These findings open the possibility that DNMT2 may contribute to neoplastic process through a novel pathways, related to RNA methylation. Clearly, much research should be performed in this area in order to understand all possible roles of DNMT2.

DNMT3 family
In mature cells which divide, DNMT1 is predominant DNA methyltransferase. However, there are two other DNMTs, DNMT3a and DNMT3b, which cannot differentiate between unmethylated and hemimethylated CpG sites. Their role is primarily de novo DNA methylation. Accordingly, they are highly expressed in early embryonic cells when programmed waves of de novo methylation occur. Their level is considerably lower after differentiation and in adult somatic tissues, but it significantly increases in cancer cells. Both enzymes contain large N-terminal parts which interact with other proteins. The C-terminal domain represents the catalytic center (Gowher & Jeltsch, 2002). In 1999, mice with targeted disruption of the Dnmt3a and Dnmt3b genes was an excellent model for exploring the activity of these two enzymes. Experiments showed lack of de novo methylation in embryonic stem cells and early embryos but without any effect on the maintenance of imprinted methylation patterns (Okano et al., 1999).

DNMT3B
The significance of this enzyme in cancer has been well recognized. The most recent research publications present its role in silencing tumor suppressor genes, through methylation of their promoters. In a study of hepatocellular carcinoma, DNMT3B overexpression was correlated to the level of promoter methylation and expression of MTSS1 (Metastasis Suppressor 1). There was negative correlation with DNMT3B expression and MTSS1 expression, but not with its promoter methylation. The DNMT3B was found to be directly bound to the 5'-flanking MSS1 region that was sparsely methylated and methylation inhibitors failed to recover the MSS1 expression. Based on these findings, the conclusion was that DNMT3B may repress MTSS1 through a DNA methylationindependent mechanism (Fan et al., 2011). This should not be surprising, keeping in mind that different protein complexes include the DNMT3B. The thought is that DNMT3B, through a partnership with a transcription repressive complex, inhibits gene expression without necessarily exhibiting its genuine de novo methyltransferase function.

DNMTs in cancer
As shown in a comprehensive review in 2011, incorporating most available data relating to the level of DNMTs in cancer, these enzymes are increased in all tested cancer clinical specimens and cancer cell lines (Daniel et al., 2011). The methods used for the quantification were primarily Real-Time PCR and immunochistochemical methods, or both. When we perform these experiments, we must question: Does the amount of mRNA reflect the amount of the protein? Can we reach any conclusion without measuring protein activity? On all these, the answer is, or should be, a resounding "no". However, it has happened all too many times that we do not see clearly and that we reach our conclusion prematurely. If premature -then it is, unfortunately often -wrong. The consequence of all too many examples of this kind of unfortunate mistake is an enormous waste of time, as it takes years to get back on the right track. Many recently retracted papers, including those published in journals with the highest impact factors, are extremely consequential. Many researchers who are initially on the right path, change their hypotheses after reading what they had been led to believe, mistakenly, to be a break-through article. This mistaken action took them straight into the disaster zone of irreproducible results. It takes years for an article to be retracted. Meanwhile, many scientific careers are damaged in an effort to reproduce a result that cannot be reproduced.

Cancer and global DNA hypomethylation
The reasons for global DNA hypomethylation combined with hypermethylation at many 5' gene or promoter regions in cancer is not understood. In prior years, research related to this phenomenon was performed on several models: prostate carcinomas, Wilms's tumors and gastric cancer (Ehrlich et al., 2002;Kaneda et al., 2004;Santourlidis et al., 1999). In order to clarify this phenomenon, Ehrlich and co-workers analyzed the relationship of cancer-linked hypermethylation and hipomethylation at 55 gene loci (mostly CpG islands overlapping the 5' promoter regions), three classes of repetitive elements and global hypomethylation profile in epithelial ovarian malignant tumors (19 ovarian carcinomas, 20 LMP (low malignant potential) tumors and 21 cystadenomas) (Ehrlich et al., 2006). They proved that promoter 5' gene hypermethylation and both satellite or global DNA hypomethylation occur independently. This was shown in a multivariate regression analysis where, in a final model, hypermethylation variables and hypomethylation variables independently predicted the degree of malignancy in ovarian tumors as follows: LTB4R (P0.005), MTHFR1 (P=0.006), CDH1 (P=0.005) and Sat (P=0.005). After making an adjustment for multiple comparisons, the LTB4R and MTHFR1 showed an association of DNA methylation with DNMT1 mRNA levels (P0.01), in carcinomas. However, this association was not seen when combining them with LMP tumors.

www.intechopen.com
The Importance of Aberrant DNA Methylation in Cancer

337
These examples lead to the following thought: if the total amount of the enzyme is increased, and the system is (globally) hypomethylated, then something has to impact its function. The focus should be on protein interactions because, as shown very recently, the interaction between DNMT1, PCNA and UHFR1 may be disrupted in human and mice astrocytes and glial progenitor cells. This specific change was shown to be an oncogenic event . The same paper shows that gliomagenesis relates to the decrease of 5-mC, the expression level of Dnmt1 remains stabile, but the catalytic activity of the enzyme decreases. This knowledge was applied to analyses measuring maintaining DNMT1 activity in 45 glioma patients who were divided into two groups: those with low (N=23), and those with high level of methyltransferase activity (N=22). Very significant differences in survival time was found between these two groups (p=0.0019), indicating that the level of DNMT1 activity, rather than the absolute amount, could be used as a survival prognostic factor. However, this conclusion must be taken with caution because of the limited number of patients. The results also clearly show that DNMT1/PCNA/UHRF1 interactions inversely correlate with the level of DNMT1 phosphorylation, reflecting, as proved in the cited paper, that the DNMT1 phosphorylation represents the hallmark of DNMT1/PCNA/UHRF1 interaction. The loss of this interaction represents a milestone for chromosomal instability induced by hypomethylated DNA repeat elements and also mediates overexpression of several very potent oncogenes such H-ras and survivin. However, it has to be noted, once again, that the number of patients was rather small and more research, based on a larger number of patients, must be performed in order to convert very strong indications into conclusions relating to DNMT1 phosphorylation as a prognostic cancer marker.

Loss of Imprinting (LOI) and cancer
Genomic imprinting is an epigenetic phenomenon that ensures monoallelic gene expression in a parent-of-origin dependent manner. Accordingly, imprinted genes are expressed only from a paternal or maternal allele. If we consider the biallelic expression as a full activity of a certain gene, then the imprinted gene gives "half" of the information which makes it very vulnerable to pathogenetic processes. If the gene is biallelically expressed, then any kind of damage affecting one allele still leaves 50% of overall function. As is the case with tumor suppressor genes, this may be sufficient for normal functioning. If the same happens with the active copy of the imprinted gene, the other allele, silenced through established imprinting marks, cannot add to the functioning. Hence, there is a haplo-insufficiency related to imprinted genes that makes them more "vulnerable".

Regulation of genomic imprinting
Estimation of the total number of imprinted genes in the human genome varies according to the methodology used. There are ~100 imprinted genes in the mammalian genome and ~70 imprinted genes have been experimentally verified and catalogued (Morison et al., 2001). These genes are not randomly scattered throughout the genome. They are clustered in the domains containing regulatory DNA elements -imprinting control regions, ICRs. These cisregulatory elements are methylated only on one allele and that is the reason for calling them differentially methylated regions/domains, DMRs/DMDs. DNMT1 has the most important role in DNA methylation maintenance at ICRs. In addition to DNA methylation, other epigenetic modifications (post-translational histone tail modifications, binding of polycomb proteins, non-coding RNAs) play an important role in regulating ICRs.

IGF2/H19 imprinting
IGF2, coding for IGF2 mitogenic peptide and H19, a protein non-coding gene, at the human chromosome 11p15.5, are reciprocally imprinted, in most tissues studied to date. This is controlled by the IGF2/H19 ICR which lies upstream of H19 and is methylated only on the paternal allele (Tremblay et al., 1997). Accordingly, H19 promoter is also methylated on the paternal allele and H19 is silent (Zhang et al, 1993).
The insulator model (Figure 1) describes, roughly, how IGF2/H19 ICR regulates monoallelic expression of IGF2 and H19. The insulators are DNA sequences which block contact between promoters and nearby enhancers/silencers. The IGF2/H19 ICR is positioned between IGF2 and H19, ~100 kb downstream of the IGF2. The downstream enhancers are shared by IGF2 and H19 (Leighton et al., 1995). On the maternal allele, the CCCTC binding factor (CTCF) binds to unmethylated IGF2/H19 ICR and insulates Igf2 promoters from the enhancers (Bell & Felsenfeld, 2000;Hark et al., 2000). The human IGF2/H19 ICR region has seven CTCF binding sites, but only the methylation of the sixth one acts as a key regulatory domain (Takai et al., 2001) through abolishing the CTCF binding to the paternal IGF2/H19 ICR, leading to IGF2 expression (Bell & Felsenfeld, 2000;Hark et al., 2000). In humans, the CTCF binding to both IGF2/H19 ICR and the IGF2 promoters P2-P4, and insulation of the IGF2 promoters from enhancers on the maternal allele, involves long-range intrachromosomal interactions (Vu et al., 2010). In humans, there are still some missing links relating to IGF2/H19 imprinting mechanisms. One of them includes two additional differentially methylated regions ( Figure 1): DMR0, surrounding IGF2 promoter P0 and methylated on the paternal allele (Murrell et al., 2008) and DMR2, in IGF2 exon 9 (Murrell et al., 2008). DMR0 does not bind CTCF (Vu et al., 2010). The existence of two additional imprinted genes in this region IGF2AS (Okutsu et al., 2000), and PIHit (paternally expressed Igf2/H19 intergenic transcript) (Court et al., 2011) adds even more complexity to the whole chromosomal locus; together with the CTCF paralogue, BORIS/CTCFL (Brother of Regulator of Imprinted Sites/CTCF-like) (Loukinov et al., 2002).
Both proteins, CTCF (Uniprot: P49711) and BORIS (Uniprot: Q8NI51), share 74% homology in their 11-Zn-finger, DNA binding domains. Their N' and C' termini have less than a 10% sequence homology. This implies that they can bind to the same DNA sequences (for example, IGF2/H19 ICR) but each of them interacts with different protein partners and has different function. For example, BORIS is involved in the establishment of IGF2/H19 imprinting marks in the male germline (Jelinic et al., 2006). CTCF is involved in the interpretation of these imprinted marks in somatic cells (Hore et al., 2008).

Loss of imprinting in cancer
Loss of imprinting (LOI) in cancer is manifested as either activation of normally epigenetically silenced allele resulting in biallelic expression, or as silencing of normally active allele ( Theoretically, in the case of imprinted tumor suppressor genes, loss of the expression from only one functional allele could contributes to tumorigenesis by mimicking "the second hit", according to Knudson's Two Hits Hypothesis (Knudson, 1971). The same effect on the cell dividing potential and growth has the biallelic expression of the imprinted, growth promoting gene; mice with Igf2 LOI in intestines have less differentiated intestines and develop twice as many adenomas, in comparison with control (Sakatani et al., 2005). This is, of course, simplified presentation which may not be realistic at all due to existence of many different regulating processes and signaling molecules included in the process of carcinogenesis.

Loss of IGF2 and H19 imprinting in cancer
IGF2 and H19 LOI has been demonstrated in many different types of cancer (reviewed in Jelinic & Shaw, 2007). We were the first to analyze the IGF2/H19 imprinting status in laryngeal squamous cell carcinoma samples (LSCCs) (Grbesa et al., 2008) where we detected IGF2 LOI in 33% of LSCCs and 28% of adjacent non-tumorous laryngeal tissues. This finding was already discussed. At that time, the IGF2 LOI in normal tissues has been detected only in colon mucosa of patients with colorectal cancer (Cui et al., 2003), and it was also known that IGF2 LOI exists in peripheral blood lymphocytes, in 10% of normal population (Sakatani et al., 2001). It is difficult to imagine the IGF2 LOI as a cancer biomarker as, presently, we detect LOI at the level of easily degraded mRNA (that needs to be entirely free of DNA). Additional problem presents restriction of analysis to the polymorphic sites (SNP), that are not necessarily informative for certain markers, in certain populations (Kaaks et al., 2009). We have also detected H19 LOI in 23% of LSCCs, in line with el-Naggar's results (el-Naggar et al., 1999). Increased H19 expression in LSCCs was reconfirmed recently (Mirisola et al., 2011).
In a non-tumorous cell, based on the insulator model, one expects existence of MOI (maintenance of imprinting) through monoallelic IGF2 and monoallelic H19 expression. If there is a LOI, the insulator model proposes biallelic expression of IGF2 or H19, and no expression of the partner (for example, IGF2 biallelically expressed, H19 silenced). But, this is not the case. We have detected, in a small subset of samples (2/10) informative for both IGF2 and H19 imprinting analyses, biallelic IGF2 expression (IGF2 LOI) joined with H19 MOI. In remaining eight samples, the imprinting was maintained. This was also observed in a broad group of head and neck cancers (among them, 14 LSCCs) (el-Naggar et al., 1999). Again, we are here facing the challenge related to the presence of "contaminating", nontumorous cell with an open possibility that they contributed to the "mixed" result. The solution could be the usage of methods which enable analysis of IGF2/H19 imprinting in the individual cells (for example RNA fluorescent in situ hybridization).
We have also analyzed the methylation of the 6 th CTCF-binding site (CBS6) within the ICR by methylation restriction PCR, MR-PCR (Ulaner et al., 2003a). In the samples with IGF2 and H19 MOI, the CBS6 was hemimethylated, while its methylation appeared aberrant in the tissue samples with IGF2 or H19 LOI (Grbesa et al., 2008). The analysis of CTCF binding to the CBS6 by chromatin immunoprecipitation (ChIP) was not performed due to the well known problem in this part of molecular oncology: the limited amount of tissue. However, based on ours and other groups results, it seems that IGF2 and H19 LOI cannot be explained solely on the basis of the level of CBS6 methylation (Cui et al., 2002) and its occupancy by the CTCF (Ulaner et al., 2003b). In cancer cell lines with IGF2 LOI, the whole 3-D structure of www.intechopen.com The Importance of Aberrant DNA Methylation in Cancer 341 the IGF2/H19 locus is dramatically changed, in comparison with normal cells and cancer cells with IGF2 MOI (Vu et al., 2010). As the prototype of the method that should be used for this kind of analysis (The Chromosome Conformation Capture Original Copy Assay (3C-OC) coupled to QPCR and 3D software analysis) does not represent the standard technique applicable for clinical tissue specimen analyses, it will take some time before implementing this type of research on native clinical material.
In our LSSC samples, BORIS was expressed with both maintained IGF2 and H19 imprinting but also in the samples with IGF2/H19 LOI (Grbesa I, unpublished results).
Recently, 23 BORIS transcript variants that may potentially produce 17 BORIS polypeptides were discovered (Pugacheva et al., 2010). In human tissues, polypeptides that correspond to calculated molecular weight of some of the BORIS isoforms, have been detected with polyclonal anti-BORIS antibody  but the role of different BORIS isoforms in establishment and maintenance of IGF2/H19 imprinting remains to be elucidated.

Poly(ADP-ribosyl)ation in regulation of DNA methylation
Since the seminal work by Feinberg and Vogelstein pointing to the global hypomethylation in tumor cells (Feinberg & Vogelstein, 1983), many reports followed documenting aberrant aquisition of methylation marks at discrete loci in the genome, most notably those comprising genes involved in cell cycle control. Such methylation pattern is opposite to the bimodal methylation pattern, characterized by global DNA methylation and hypomethylation of CpG islands, that is physiologically aquired at the time of embryo implantation and faithfully maintained throughout life (Brandeis et al., 1993). The search for cis-and trans-acting factors that orchestrate such bimodal methylation pattern has since been the focus of scientific interest.
Twenty years ago, linker histone H1 has been identified to have inhibitory effect on DNA methylation (Caiafa et al., 1991). Shortly thereafter, the difference between H1 histone isoforms, at that time termed as "tightly-bound" and "loosely-bound," in regulation of DNA methylation were observed. In contrast to "typical" loosly-bound histone H1, tightly-bound histone H1 has been shown to facilitate methylation of linker DNA (Santoro et al., 1993). The histone H1, which is able to bind CpG-rich DNA sequences and inhibit double-stranded DNA methylation, has later been identified as variant H1e (Santoro et al., 1995) that promotes chromatin condensation or, upon poly(ADP-ribosyl)ation (pARylation), chromatin decondensation (D'Erme et al., 1996). Appart from the importance of this histone variant and its pARylation in chromatin decondensation, which allows recruitment of transcription factors, the same group has demonstrated the mandatory role of pARylation in the maintenance of hypomethylated state of CpG islands in mouse fibroblasts (Zardo & Caifa, 1998).
Inhibition of pARylation by competitive PARP inhibitor, 3-aminobenzamide (3-AB), enhances DNA metylation (Zardo & Caifa, 1998). pARylation is process catalyzed by poly(ADP-ribose) polymerases (PARP), which use NAD as a substrate to build up polymers of ADP-ribose on acceptor proteins, including PARP-1 (D' Amours et al., 1999). It is the founding member of this enzyme family, accounting for more than 90% of cellular p A R y l a t i n g c a p a c i t y .  , 1982). The negative charge of ADP-ribose polymers makes them resemble nucleic acids and compete with them for binding different protein partners. ADPribose polymers, either covalently linked to acceptor proteins or protein-free, are also able to non-covalently bind proteins (Malanga & Althaus, 2005). The binding of ADP-ribose polymers to acceptor proteins is dependent on the presence of amino-acids consensus, poly(ADP-ribose)-binding motifs, that allows for non-covalent binding (Pleschke et al., 2000). Those consensuses are present in a wide variety of proteins with very divergent functions, ranging from structural proteins such as histones (Althaus et al., 1995), proteins involved in DNA repair to enzymes involved in regulation of DNA topology (Malanga & Althaus, 2005). Binding of negatively charged poly(ADP-ribose) polymers functionaly and structuraly modifies acceptor proteins (Panzeter et al., 1992).
The DNMT1 has two amino-acid consensus motifs for binding poly(ADP-ribose) polymers in its N-terminal domain. It was demonstrated that pARylated PARP-1 and DNMT1 form complex in vivo, and that either PARP-1-associated or free poly(ADP-ribose) polymers are able to inhibit catalytic activity of DNMT1 (Reale et al., 2005). The majority of PARP-1 molecules in normal cell is unmodified (D'Amours et al., 1999) and the mechanism directing the minority of, under physiological conditions, automodified PARP-1 molecules to CpG islands remains an open question. In an effort to elucidate the possible involvement of pARylation in the regulation of Dnmt1 gene promoter, Guastafierro et al. (2008) have examined transcription factors known to be subject to covalent poly(ADP-ribosyl)ation (Hassa et al., 2006). Highly conserved multifunctional transcription factor, CTCF (Ohlsson et al., 2001), attracted their attention based on its role in protection of DNA from methylation and functional dependence on pARylation. The key role of pARylation of CTCF in its insulator/ enhancer blocking function has been reported by Yu et al. in 2004. The role of CTCF in regulation of IGF2/H19 imprinting has been abolished by treatment with PARP-1 inhibitor 3-AB. To establish whether the lack of CTCF pARylation is indeed responsible for the loss of IGF2 imprinting, the association of poly(ADP-ribose) polymers with H19 ICR was examined by ChIP on wild-type and mutant-type ICR containing CTCF-binding sites. The pARylation mark has been present only if the wild-type allele has been inherited maternally.
In the lack of specific antibodies recognizing pARylated CTCF, however, it cannot be ruled out that CTCF binding to its target sites is necessary for activation of PARP-1 or other members of poly(ADP-ribose) polymerase family that would pARylate proteins other than CTCF in that region. Indeed, it has been demonstrated that, in addition to its previously recognized characteristic of being acceptor of poly(ADP-ribose) polymers, CTCF is interacting www.intechopen.com The Importance of Aberrant DNA Methylation in Cancer 343 with PARP-1 and is able to activate PARP-1 in the absence of DNA nicks, whereby both proteins become pARylated and negatively affect DNA methylation machinery (Guastafierro et al., 2008). The epigenetic regulation of tumor suppressor p16 INK4a provided some insight into CTCF and PARP-1 DNA-binding and the influence of their pARylation relative to the expression of this gene and several other CTCF-regulated genes (Witcher & Emerson, 2009). Transcription of p16 INK4a depends on CTCF binding to a chromatin boundary ~ 2kb upstream of its transcription start site. In the absence of CTCF binding, p16 is silenced in various types of tumor cells. When associated with the boundary element, in p16-expressing cells, CTCF is pARylated. However, no direct association of pARylated CTCF and PARP-1 could be detected in those p16-expressing cells. In p16-silenced cells, CTCF was not pARylated and bound to the methylated boundary element, but CTCF-PARP-1 complex could be readily detected by co-immunoprecipitation. Moreover, the authors found that pARylated CTCF dissociates from PARP-1, whereas pARylated CTCF remains associated with PARP-1 and loses its function, at this boundary element. Therefore, it is conceivable that deregulation of pARylation may impart the aberrant association of CTCF and PARP-1 and change the association of CTCF with its DNA binding sites. Relevant to the possible influence of the CTCFs binding to DNA and its protective role against DNA methylation is the recent report on the ability of CTCF to form an unusual DNA structure (MacPherson & Sadowski, 2010). Considering that, in addition to the classical view that PARP-1 is activated by DNA nicks, various non-linear DNA structures are able to activate this enzyme (Lonskaya et al., 2005), the property of CTCF to loop DNA may be yet another facet in connecting processes of DNA methylation and poly(ADP-ribosyl)ation.

Epigenomic therapy
The inhibition of DNMTs has been used in epigenetic cancer therapy, based on the idea that seems to be quite simple: what is hypermethylated, needs to be normomethylated. So, if we consider the act of removing the methylation marks from hypermethylated promoters of tumor suppressor genes whose protein products are involved in regulation of cell cycle, apoptosis and DNA repair as a therapeutic act, we may be well on a right way.
There are two kinds of DNA methylation inhibitors: nucleoside (Fig. 2) and non-nucleoside analogues. The consequences of nucleoside analogue incorporation into DNA (in this situation, in lieu of cytosines) is DNMT binding and blocking, causing depletion of overlay active enzyme molecules with DNA methyltransferase activity. Two DNA methylation inhibitors, cytidine analogs, were approved by the US Food and Administration (FDA), for the treatment of Myelodysplastic Syndrome and certain forms of leukemias: 5-azacytidine (azacytidine, Vidaza TM ), which was approved in May, 2004 and 5-aza-2'-deoxycytidine (5-azaCdR, decitabine, Dacogen TM ), which is a deoxyribose analog of 5-azacytidine, approved in May, 2006 ( Figure 2).
The antineoplastic effects caused by these two drugs are related to targeted DNA demethylation (and consequential restoration of gene activity necessary for differentiation) and a direct cytotoxic effect on abnormal, rapidly dividing hematopoietic cells in the bone marrow. Non-proliferating cells are relatively insensitive to these two drugs, but their inherently toxic effects do produce certain side-effects.

5-azacytidine and decitabine
5-azacytidine was described as a DNMT inhibitor more than 30 years ago (Jones & Taylor, 1980). The first approved clinical targets for 5-azacytidine were Myelodysplastic Syndrome (MDS) and acute myeloid leukema. The drug can be applied subcutaneously or through IV infusion.
This nucleotide analogue becomes incorporated into DNA in place of cytosine after being modified by ribonucleotide reductase and subsequent phosphorylation. When DNMT1 "recognizes" it as a unmethylated substrate in a newly synthesized DNA strand and "approaches" it, it becomes trapped through covalent binding to the incorporated analogue. The reduced level of DNA methylation that follows represents the consequence of passive demethylation in consequent cell cycles due to the lack of functional enzyme. 5-azacytidine can also be phosphorylated by uridine-cytidine kinase and, as such, incorporated into RNA. When applied subcutaneously, it may cause: nausea, anemia, thrombocytopenia, vomiting, pyrexia, leukopenia, diarrhea, injection site erythema, constipation, neutropenia and ecchymosis. Most common adverse reactions through IV application, according to the FDA, included: petechiae, rigors, weakness and hypokalemia.
MDS is also the primary therapeutic indication for Dacogen TM , as in the case with Vidaza TM . The most prominent side-effects associated with Dacogen TM treatment are neutropenia and thrombocytopenia.
The third analogue, zebularine, has several advantages over the two previously mentioned compounds. It is more stable, highly selective for cancer cells and, hence, far less toxic (Cheng et al., 2004). However, this potential drug seemed to fail after being very successful in a small pilot study, because a very high dosage of the drug were needed to obtain the desired antitumorous effect (Goffin & Eisenhauer, 2002). In all three cases, the facts which include non-selective DNA targeting resulting in side-effects, were the basis for approaching the problem in a different way. That is, developing compounds which target DNMTs directly, without prior DNA incorporation requirement.

New methods in exploring activity of methylation inhibitors
In 2008, Illumina Golden Gate arrays were used for direct characterization of the effects of azacytidine application in three different leukemia cell lines (HEL, HL-60, K562) and ten patients who fulfilled the WHO criteria for MDS. In the cell lines, the effect of the drug on www.intechopen.com The Importance of Aberrant DNA Methylation in Cancer 345 DNMT1 protein level differed, the lowest being in HL-60, while HEL cells appeared relatively resistant to DNMT1 depletion. Accordingly, HL-60 was considerably demethylated, while the HEL cell line did not exhibit significant change in global methylation level. After performing an array-based methylation profiling (1,505 CpGs representing 807 cancer-associated genes), the results were very interesting. In untreated cell lines, the number of methylated CpGs exceeded 80%. After treatment, more than 80% of spots became demethylated, but only in HL-60 and K562 cell lines. There was no consistent demethylation trend in HEL cells. Of importance, flow cytometry analysis showed similar overall cell cycle profiles in all three cell lines. The results obtained on patients' samples (6/10, as six patients completed at least one treatment course consisting of four cycles) differed significantly. In three patients, the methylation levels remained the same, while in three other persons the level of methylation decreased significantly, through a cyclic demethylation, following the cyclic administration of the drug Stresmann & Lyko 2008).
In 2011., a genome-scale Infinium analysis (27,578 CG nucleotides; more than 14,475 associated genes) was performed on two human colon cancer cell lines (HCT116 and double knockout (DKO) HCT116, lacking DNMT1 and DNMT3B) and a HL-60 leukemia cell line treated with both azacytidine and decitabine. The bimodal peaks of methylation distribution was found in both treated and untreated cells, representing spots with low and high levels of methylation. These experiments not only showed more potent demethylation activity of decitabine when compared with azacytidine, but also preferential demethylation at specific loci and demethylation resistance of certain number of CGs, in HCT116. The results from these in vitro study shed new light on problems encountered in clinical work with these drugs: not only was the degree of demethylation of the whole genomic DNA higher than gene-specific demethylation (this is something that we do not want to happen, as these drugs were implemented in the clinic in order to demethylate the hypermethylated promoters of tumor suppressor genes), but also the spatial distribution of demethylated CpGs mimicked the distribution found in DKO HCT116. However, when the cluster of cancer-related genes associated CGs was analyzed separately, it turned out that 906 out of 2,125 were hypermethylated and both drugs were very efficient in removing methylation marks. It is hard to distinguish which gene (and joining CGs) represents the clean "cancer-related" gene. Many genes that were considered to be "inflammatory genes" or "metabolic genes", turned out to be "cancer genes", as well. One should be careful regarding this kind of clustering because we are currently far away from a complete understanding of how certain signaling proteins/pathways interact, regardless of to which cluster they were primarily asigned.
When performing computational modeling for the presence of transcription factors binding sites in 851 CpGs representing 644 genes, demethylation -sensitive and -resistant CGs showed different types of enrichment. For example, binding sites of Forkhead box (Fox) transcription factors were enriched in demethylation sensitive genes, while basic Helix-Loop-Helix transcription factor binding sites turned out to be enriched in demethylation resistant genes (Hagemann et al., 2011).

Non-nucleoside compounds
There are several more, potentially promising, non-nucleoside, candidates. Some of them are well known drugs/healing compounds, such as curcumin.
-Procaine (a well known local anesthetic) and its derivative, procainamide (a well known drug for treating cardiac arrhythmia) have shown demethylating activity in cancer cell lines of different origin. They were shown to be a specific inhibitors of DNMT1 (Lee et al, 2005). For example, in prostate cancer cells, procainamide restores GSTP1 expression through demethylation of GSTP1 promoter (Lin X et al., 2001). In lung cancer, these drugs induce demethylation of WIF-1 (Wnt Inhibitory Factor) promoter, a negative regulator of the Wnt-signaling pathway (Gao et al., 2009). -Hydralazine: The methylation inhibitory role was shown to be specifically related to the inhibition of DNMT (Angeles et al., 2005). Its combination with magnesium valproate, seems to be promising in treating different types of malignant disease, including MDS (Candelaria et al., 2011). -The inhibitory effect of (-)-epigallocatechin-3-gallate (EGCG), the healing compound from green tea, was shown in 2003 (Fang et al., 2003). It was then shown for the first time, that inhibition of DNA methylation can be inhibited by a commonly consumed dietary constituent. At the same time, these results suggested the potential use of EGCG for the prevention of cancer-related gene silencing. The authors measured the DNMT1 catalytic activity and performed molecular modeling of the interaction between EGCG and DNMT1. Finally, they proved reversal of hypermethylation through the reactivation of expression of several genes (RAR, MGMT, p16INK4a, and hMLH1). -Genistein, the soy bean isoflavone, was also shown to influence DNMTs. Based on a literature search, there seems to be only one study exploring its efficacy as a DNA methyltransferase inhibitor (Li et al., 2009).

Curcumin
Curcumin (diferulolymethane) is the yellow pigment found in the cooking spice turmeric (Curcuma longa linn). Curcumin is a strong inhibitor of the NF-B signaling pathway (Gupta et al., 2011). Having in mind the central role of NF-B in many different signaling pathways, it is not surprising that this compound shows anti-inflammatory, anti-oxidant, antimicrobial and, finally, anticancer activity. Curcumin is currently being investigated for its chemopreventative efficacy in a variety of solid tumors. So far, most of the controlled clinical trials of curcumin are in phase I (Hatcher et al., 2008), suggesting that oral curcumin is more likely to be effective as a therapeutic agent in cancers of the gastrointestinal tract than in other tissues (Sharma et al., 2005). The results of one non-randomized, open-label, phase II clinical trial conducted in the U.S. were published recently, reporting on the first 25 patients with advanced pancreatic cancer. The patients did not receive any concomitant chemotherapy or radiotherapy. There was partial response in one patient and disease stabilization in other patient, for approximately 2.5 years (Dhillon et al., 2008). Another clinical phase I/II trial included 21 gemcitabine-resistant pancreatic cancer patients who received, like in Dhillon's study, 8 grams of curcumin daily, together with gemcitabine-based chemotherapy in this instance. This combination was shown to be "safe and feasible in patients with pancreatic cancer and warrants further investigation into its efficacy" .
There are many efforts to improve curcumin's bioavailability. The most recent results reported on nanoparticle curcumin (Theracurmin), show that this form of curcumin can safely increase plasma curcumin levels in a dose-dependent manner at least up to 210 mg, without saturating the absorption system .

www.intechopen.com
The Importance of Aberrant DNA Methylation in Cancer

347
We have shown that curcumin selectively inhibits the H19 transcription in several different tumor cell lines, but not in non-tumorous cells. We do not think that protein-non-coding H19 mRNA itself necessarily exerts any kind of vital oncogenic or tumor-suppressive function, but we do think that its mRNA presence indicates vivid, globally deregulated cellular transcription (Novak Kujundzic et al., 2008).
It has been confirmed that curcumin interacts directly with 33 proteins, one of them being DNMT1 . It is considered that this binding causes direct inhibition of the enzyme and represents the molecular basis for the DNA hypomethylating activity of curcumin ).

5-Hydroxymethylcytosine
As discussed in previous sub-chapters, methylated cytosine entered the spotlight of the international scientific community, primarily due to our understanding of how genes are, or need to be, regulated. We are just entering the era of full appreciation of the importance of one more cytidine modification, discovered in bacteriophage, in 1952. (Wyatt & Cohen, 1952). It was "rediscovered" in 1972. in rat tissue, but was neglected because the results did not seem to be reproducible (Penn et al., 1972). However, thanks to knowledge gained during these almost 60 years, combined with advances in technology, we are now learning about a sixth nucleotide in our genome (Münzel et al., 2011). Only two years ago, two papers in Science showed that mammalian DNA contains 5'-hydroxymethylcytosine (hmC; 5 hmC; 5-HOME dC, Figure 3) (Kriaucionis & Heintz, 2009;Tahiliani et al., 2009). Unmodified cytosine (a), 5-methylcytosine (b), 5-hydroxymethylcytosine (c).
To date, based on a few in vitro experiments, it has been thought that hmC presents the major oxidative product of mC (Bienvenu et al. 1996;Wagner & Cadet, 2010).
Ideal for detecting the methylated cytosines, bisulfite fails on hmC. As a result, there is a problem with positioning the sixth nucleotide in the DNA molecule. There is hope from nanopore sequencing, because the first published results show the difference between mC and hmC in single-, and double-stranded DNA (Wanunu et al., 2011).
So far, most of the work on 5-hmC and TET group of proteins (TET1, TET2 and TET3) was performed on embryonic stem cells and there are only a few papers dealing with "the sixth nucleotide" in cancer. The TET proteins which can modify 5-methylcytosine in humans were initially discovered through a computational search showing these proteins as mammalian homologs of the trypanosome proteins JBP1 and JBP2, the enzymes proposed to oxidize the 5-methyl group of thymine (Tahiliani et al., 2009). Predictably, hmC levels decrease upon RNA interference-mediated depletion of TET1.
The role of hmC in cancer and its role in participating or in creating epigenomic networks, is almost entirely unknown. There are a few, recently published papers, showing its influence on the affinity (a strong decrease) with which the MBD proteins bind to DNA "occupied" by hmC. The 5-hmC immunoassay (developed by Liu at coworkers) was applied to various healthy human tissues and four cancer cell lines. The differences in hmC content among the tested cell lines were minor and insignificant. The tissue analyses showed that brain tissue has the highest content of hmC. Among all tissues tested, the lowest level was detected in lung, breast and placenta. When compared with mC distribution in various tissues (which published data showed a range of 1-2.5), the differences in hmC tissue-specific content were very strong. In cancer tissues, when compared to adjacent, non-tumorous tissue, the level of hmC was significantly lower; in one case 7.7-fold, in the other 28-fold (Li & Liu, 2011). So, the question was asked: is it possible that this increase occurs because the global level of methylated cytosines decreases in cancer cells?
Part of the answer was given in a paper published in September, 2011 (Haffner et al., 2011). The authors analyzed 78 carcinomas and 28 normal tissue samples (prostate, breast and colon). They have shown, by using immunohistochemical staining they developed, a significant decrease of hmC in tumorous tissues. There was also a significant difference in hmC tissue distribution: in normal tissue, the signals were strongest in the terminally differentiated luminal cells and far less strong in basal cells. In cancer tissues, the differences were very clear at the border between the tumor and non-tum o r o u s t i s s u e . H o w e v e r , a l t h o u g h v e r y prominent, these changes did not allow for any association with clinicopathological features, including the tumor grade (level of differentiation). Although there is a significant similarity between Haffner's and Li's result, the methods they used were quite different, which -to be sure, does not diminish the quality and importance of their results. In any event, the hmC story will need to be explored on many clinical samples before we allow ourselves to conclude anything about their prognostic significance in cancer patients.
Williams and colleagues published a very extensive study on TET protein family member function, showing the necessity of TET1 time-specific expression during development. In their experimental model, TET1 localized to both gene bodies and transcription start sites (TSS), being especially enriched at genes with high CpG content, while mC localized in regions with low CpG content. The results indicate "that TET1, by converting mC to hmC serves an important function in the regulation of DNA methylation fidelity" (Willimas et al., 2011). What, one should ask, is the consequence of the hmC presence in the gene promoters/bodies? It seems that, in a pure in vitro system, based on CMV promoter and HeLa cells extracts, the presence of hmCs in the gene promoter inhibits transcription, while their presence in the gene body does not directly inhibit transcription (Robertson et al., 2011).

Conclusion
In this chapter, we have covered aspects of deregulated DNA methylation in cancer, including a review of older data and introducing the most recent findings. By using this approach, we have tried to show maybe the most intriguing and certainly the most emerging aspects of molecular biology of a cancer cell, at the time of preparing this chapter. Certainly, the new exciting discoveries in the field of cancer epigenomics that we are presenting here are only part of emerging sets of data. The new papers with exciting findings are coming to scientific community almost on a daily base does and, for that reason, we did not allow to ourselves offering any hard conclusion, at this time period. We are aware that there are many more issues and mechanisms for discussion, such as the interactions of DNA and proteins, methylation related and unrelated, that we did not discuss. For that reason, we all look forward to future books and articles providing insight on these and like topics.

Acknowledgement
We are grateful to Dr. Višnja Stepanić and Mr. Zlatko Pandžić for their technical help in preparing this chapter. Our special thanks go to Mr. Aaron Etra who helped us with a careful revision of the text. Our research is supported by grant #098-0982464-2511 (PI:KGT), from the Ministry of Science, Education and Sport of the Republic of Croatia.