A potent cell-active allosteric inhibitor of murine DNA cytosine C5 methyltransferase.

The major DNA cytosine methyltransferase isoform in mouse erythroleukemia cells, Dnmt1, exhibits potent dead-end inhibition with a single-stranded nucleic acid by binding to an allosteric site on the enzyme. The previously reported substrate inhibition with double-stranded substrates also involves binding to an allosteric site. Thus, both forms of inhibition involve ternary enzyme-DNA-DNA complexes. The inhibition potency of the single-stranded nucleic acid is determined by the sequence, length, and most appreciably the presence of a single 5-methylcytosine residue. A single-stranded phosphorothioate derivative inhibits DNA methylation activity in nuclear extracts. Mouse erythroleukemia cells treated with the phosphorothioate inhibitor show a significant decrease in global genomic methylation levels. Inhibitor treatment of human colon cancer cells causes demethylation of the p16 tumor suppressor gene and subsequent p16 re-expression. Allosteric inhibitors of mammalian DNA cytosine methyltransferases, representing a new class of molecules with potential therapeutic applications, may be used to elucidate novel epigenetic mechanisms that control development.

The patterns of DNA cytosine methylation in mammals evolve throughout development (1). This essential process regulates imprinted genes, X chromosome inactivation, the inactivation of repetitive elements, and the expression of tissuespecific genes (2,3). Methylated DNA is recognized by several proteins and assembled into transcriptionally silent chromatin structures. Gene regulatory regions, including normally undermethylated CpG islands, become hypermethylated with age and in tumors, thereby repressing the expression of essential genes (2,4,5). Changes in DNA methylation and AdoMet-dependent DNA cytosine methyltransferase (DCMTase) 1 activity appear early in tumorigenesis. These and other observations have motivated and continue to motivate the development of DCMTase inhibitors as potential "epigenetically based" drugs (2,6,7). Thus, antisense oligonucleotides that interfere with DCMTase expression inhibit tumor formation (8), and the cancer drug 5-aza-deoxycytidine (5AC; decitabine) functions by inhibiting DCMTase (9,10). 5AC virtually abolished adenoma formation in mice genetically prone to colon tumors (11). 5AC is a clinically administered mechanism-based inhibitor of DC-MTase. Unfortunately, it may be too carcinogenic and mutagenic for most applications (10). Targeting DCMTase function with novel inhibitors has great therapeutic potential, and novel regulators could help elucidate the aspects of epigenetic control that occur throughout development (2,6,12).
The major eukaryotic DCMTase, Dnmt1, has been cloned and sequenced from at least five animal sources: mouse, human, chicken, frog, and sea urchin (13)(14)(15)(16)(17). Several variants have been identified that arise from separate gene loci and alternatively spliced mRNAs (1,3). Dnmt1 is the largest DC-MTase, with a molecular mass of 184 kDa. The smaller Cterminal domain shares sequence homology with all DCMTases and contains all or the majority of the residues required for catalysis. Mammalian methylation occurs predominately within the context of the CpG dinucleotide. The N-terminal domain contains a phosphorylation site (18) as well as zinc (19), nucleic acid (20 -23), and protein-binding elements (24). The large N-terminal domain is likely to play an active role in the complex interplay with several proteins including histone deacetylases and other Dnmts, leading to the active restructuring of chromatin (1). A previous report postulating such protein-protein interactions proposed that the substrate inhibition of Dnmt1 at high DNA concentrations results from the loss of activating Dnmt1-Dnmt1 interactions with adjacent enzyme molecules on the same DNA scaffold (25).
Early evidence indicated that Dnmt1 binds a second DNA molecule at an unidentified allosteric site (20,21) and that N-terminal domain-derived peptides bind nucleic acids (26). These interactions recently have been characterized in detail (22,(27)(28)(29). The N-terminal domain comprising the first 501 amino acids represses the methylation reaction of the human Dnmt1; this repression is relieved upon binding to methylated DNA (23). A similar stimulation was reported for murine Dnmt1 upon the addition of fully methylated double-stranded DNA, again leading to the proposal that the enzyme forms a ternary enzyme-DNA-DNA complex (30). Another form of allosteric Dnmt1 activation, in this circumstance involving single-stranded substrates with the potential of forming transient duplexes, has been described for oligonucleotides containing 5-methylcytosine adjacent to the CpG site that undergoes methylation (31,32). Many of these nucleic acid-mediated effects appear to depend on the nature of the DNA substrate upon which the enzyme is initially positioned. For example, although poly(dI⅐dC) and hemimethylated (CGG/CCG) 12 , a 36-bp fragment containing the triplet repeat sequence in which expansion in the FMR1 gene causes fragile-X syndrome, show complex kinetics consistent with DNA binding to an allosteric site, no such behavior is observed with unmethylated (CGG/ CCG) 12 (22,23). The biological role most frequently invoked for allosteric regulation of Dnmt1 via DNA binding is in the context of "methylation spreading," a process leading to de novo DNA methylation of previously ummethylated regions (30, 34 -36). This bidirectional spreading, which occurs in newly integrated viral DNA (36), in gene regulatory sequences during X chromosome inactivation (37), and during cellular immortalization (38), in transposed DNA sequences (39), may be a threat to the success of gene therapy (40).
Our interests were to address the mechanism of the well known substrate inhibition that underlies much of the complex kinetics reported for Dnmt1 and to further characterize the interactions of the enzyme with several single-stranded DNA molecules that were previously shown to be poor substrates for the enzyme (41). The enzyme binds these poor substrates with affinities comparable with substrates that are readily methylated (21), and we hypothesized that they might act as dead-end inhibitors. A previous report of nucleic acids that bind Dnmt1 but are poor substrates showed that various single-and doublestranded nucleic acids can act as inhibitors (42). Here we report the mechanism of action of one such single-stranded inhibitor and demonstrate its ability to inhibit methylation in vitro.

EXPERIMENTAL PROCEDURES
Materials-S-Adenosyl-L-[methyl-3 H]methionine (75 Ci/mmol, 1 mCi/ ml ϭ 37 GBq) was purchased from Amersham Biosciences. Unlabeled AdoMet (Sigma) was further purified as described (43). Routinely, 125 M AdoMet stocks were prepared at a specific activity of 5.8 ϫ 10 3 cpm/pmol. Two lots of poly(dI⅐dC-dI⅐dC) were purchased from Amersham Biosciences, with an average length of 6250 and 5000 base pairs. DE81 filters were purchased from Whatman. LipofectAMINE was purchased from Invitrogen. Other standard chemicals and reagents were purchased from Sigma or Fisher Scientific. Oligonucleotides were synthesized by Research Genetics (Huntsville, AL) and HPLC-purified on a Dynamax PureDNA column (Rainin Instrument Co.) according to the manufacturer's specifications. Oligonucleotides were stored in 10 mM Tris, pH 8.0, 1 mM EDTA. The following sequences were prepared (GC-box p MET , LC2, and the antisense sequence Concentrations were determined using calculated coefficients, and the annealing of GC-box a/b was as described previously (41). Dnmt1 was isolated from mouse erythroleukemia cells (MEL cells) as described (41).
Methyltransferase Assays-Filter binding assays monitored the incorporation of tritium-labeled methyl groups into DNA. Reaction and dilution buffers and filter processing were described previously (20). Incubations were at 37°C for 60 min. Each point represents the average of at least two reactions.
Substrate Inhibition-Initial velocity data were collected with a 6250-base pair poly(dI⅐dC) and a 30-base pair GC-box a/b substrate. The poly(dI⅐dC) concentrations were ϳ2.0 Inhibition with Single-stranded Oligonucleotides-Initial velocity data were collected with poly(dI⅐dC), AdoMet, GC-box b , and GC-box bMET at the concentrations indicated in Fig. 2 legend. Dnmt1 was either 3 or 4 nM (see Fig. 2 legend), AdoMet was 10 M when added as the nonchanging substrate, and poly(dI⅐dC) was 50 pM when added as the nonchanging substrate. IC 50 determinations with GC-box bMET , GC-box-pMET , GC-box cMET , GC-box dMET , GC-box eMET , and CRE used 4 nM Dnmt1 and 50 pM poly(dI⅐dC). Oligonucleotide concentrations ranged between 5-fold lower and 5-fold higher than the IC 50 .
Nuclear Extract Assays-MEL nuclear extracts were prepared as described previously (41). Freshly made nuclear extract (2 l) was combined with 12.5 M tritiated AdoMet, 1 nM poly(dI⅐dC), and varying concentrations of oligonucleotides in a volume of 20 l at 37°C. The reaction was stopped after 60 min, and label incorporation into DNA was determined as described (41).
Mouse Erythroleukemia Cell Studies-MEL cells, prepared as described (41) to a density of 10 6 /ml, were treated with inhibitors (7.7 M GC-box pMET , 1.5 M 5AC, or 18 M antisense oligonucleotide) along with LipofectAMINE as described by the manufacturer. The mock treatment used TE (Tris, pH 8.0 (10 mM), and EDTA, 1 mM) in place of any inhibitor. The inhibitors or the mock were added only at the initiation of the experiment, and genomic DNA was isolated after 72 and 110 h of cell culture. 5-Methylcytosine content was determined as described (44). Briefly, MspI endonuclease was used to digest the genomic DNA, and the samples were then treated sequentially with calf intestinal alkaline phosphatase, T4 polynucleotide kinase (and [␥-32 P]ATP), and P1 nuclease. The 32 P end-labeled cytosines, either 5-m CMP or CMP within the CpG of the MspI ends, were separated on cellulose thin layer chromatography plates. A PhophorImager (Amersham Biosciences) was used to visualize and quantify the percent 5-methylcytosine. Each reported result is the average of at least two independent determinations.
Human Colon Cancer Cell Studies-HT29 cells were plated at 50% confluence in six-well plates and treated with 5AC at 1 M or Lipo-fectAMINE-transfected with either 10 M GC-box bMET or LC2 control oligonucleotides for 48 -72 h. The LC2 control is identical to GC-box bMET with the four central bases converted to adenines. DNA was isolated using the Qiagen blood and cell culture DNA kit (Qiagen) according to the manufacturer's instructions. Methyl-specific PCR (MSP) was performed as described previously (45) using primers described therein. Peripheral blood lymphocytes were used as a positive control for unmethylated p16, and peripheral blood lymphocyte DNA, methylated in vitro using SssI methylase (New England Biolabs) according to the manufacturer's instructions, was used as a positive control for methylated p16. RNA was isolated using Trizol (Invitrogen). cDNA was prepared from 1 g of RNA using a Superscript II reverse transcription system with random hexamers as primers (Invitrogen). PCR was performed using primers for p16 designed to cross a splice junction in the gene (GenBank TM accession number L27211), 5Ј-ATC ATC AGT CAC CGA AGG TC-3Ј (sense) and 5Ј-CCA CAT GAA TGT GCG CTT AG-3Ј (antisense), on 1 l of cDNA product. The reaction was initiated with a 3-min incubation at 94°C followed by 35 amplification cycles (94°C for 30 s, 58°C for1 min, 72°C for 1 min) and a final 10 min extension step of 10 min yielding a 355-bp product following electrophoresis on a 1.5% agarose gel. The gel was stained with ethidium bromide and photographed. Expression of ␤-actin was used as a standard for RNA integrity and equal gel loading. Primers for ␤-actin were 5Ј-GGA GTC CTG TGG CAT CCA CG-3Ј (sense) and 5Ј-CTA GAA GCA TTT GCG GTG GA-3Ј (antisense). Amplification was the same as p16 with the exception of using an annealing temperature of 60°C and 27 amplification cycles.

RESULTS AND DISCUSSION
The original observation by Linn and co-workers (25) that hemimethylated duplex plasmid DNA and single-stranded DNA are inhibitory at high concentrations was explained with an interesting model in which Dnmt1 associates with itself upon the DNA scaffold at low ratios of DNA to enzyme, thereby activating the enzyme. Monomeric enzyme-DNA complexes are stabilized with excess substrate, thus leading to a loss of the activated form. An alternative explanation for this behavior at high DNA concentrations is the formation of an inhibitory ternary complex involving the enzyme and two or more DNA molecules. Although recent reports provide data consistent with this alternative explanation (23,30), we sought to provide a simple probe of the underlying mechanism. Fig. 1 shows our results with a large multi-site 6250-bp poly(dI⅐dC) substrate and a short 30-bp GC-box a/b duplex containing a single CpG site. The GC-box a/b duplex is an excellent substrate for Dnmt1 and contains an Sp1 transcription factor recognition element (41). We and others have demonstrated that poly(dI⅐dC) shows inhibition at high DNA concentrations (20,23,46), which is revealed by the data in Fig. 1. The relative velocities versus S/K m plots conveniently normalize the data between the two diverse DNA substrates. The short 30-bp duplex nearly matches the footprint of the Dnmt1-DNA complex (20), and we reasoned that this substrate would be precluded from stabilizing the types of complexes proposed by Linn and coworkers (25). As shown in Fig. 1, the GC-box a/b duplex shows the same form of substrate inhibition as the much larger poly(dI⅐dC). The poorer affinity observed with the 30-bp duplex limited our ability to take the S/K m ratio as high as with the poly(dI⅐dC); nevertheless, both sets of data are nicely fit to a standard equation for substrate inhibition, invoking formation of an inhibitory ternary enzyme-DNA-DNA complex (47).
Our results provide compelling support for a mechanism involving the binding of a second double-stranded DNA molecule to an allosteric site on Dnmt1. The observed inhibition with the short duplex DNA (Fig. 1) is unlikely to come from the preferential stabilization of monomeric enzyme forms, because these short substrates were designed to be incapable of stabilizing such complexes under any enzyme concentrations. In this circumstance limited to unmethylated duplexes (poly(dI⅐dC) or GC-box a/b ), the binding of a second DNA molecule is inhibitory. We recently confirmed this result with unmethylated DNA and another assay, showing that the ternary enzyme-DNA-DNA complexes dramatically decrease the processive action of the enzyme on multi-site substrates by in-creasing the off-rate from the substrate DNA. 2 Such an increased off-rate could have the same effect with the single-site substrate (GC-box a/b ) if the off-rate and methylation rates are comparable. In contrast to these inhibitory allosteric effects, we and others (Ref. 30 and footnote 2) have shown that methylated DNA activates Dnmt1. This occurs in trans, with 5-methylcytosine attached to DNA other than that being methylated, and in cis, when introduced on the same molecule at positions adjacent to the target methylation sequence (CpG) (31,48)). 3 It remains unclear whether the activation and inhibition effects are mediated by nucleic acid binding to the same allosteric site.
Dnmt1 shows good activity with single-stranded plasmidderived substrates (25) and single-stranded oligonucleotides (Refs. 31 and 34, and footnote 3); we previously determined this with single-stranded versions of CRE, a cis-regulatory transcriptional element (41). Some single-stranded oligonucleotides are proposed to form transient duplexes that are then acted upon by the enzyme (31); the CRE and GC-box-derived single strands used in our study show no such behavior (41). In contrast to the CRE-derived strands, single-stranded substrates mimicking the GC-box element, showed no detectable activity (41) despite the ability of the enzyme to bind such sequences with good affinity (21). We therefore speculated that the single-stranded GC-box oligonucleotides could act as deadend inhibitors of the enzyme. We submitted GC-box b singlestranded DNA to a detailed inhibition analysis; the double reciprocal plots are shown in Fig. 2A. The pattern of lines intersects far to the left of the y axis and is best fit by a standard equation for noncompetitive inhibition. The inhibition constants were determined to be K is ϭ 3.6 Ϯ 1.5 M and K ii ϭ 6.8 Ϯ 1.2 M (49). K ii describes the inhibition constant for the enzyme-substrate complex, and K is the inhibition constant for the free enzyme. The K ii /K is ratio is a measure of which binding event is preferred (49), and in this case, its value of 1.9 suggests that this inhibitor slightly favors addition to the free enzyme over the enzyme-DNA complex. Thus, the pattern shows that the free enzyme can bind to single-stranded GC-box b , as can the binary enzyme-poly(dI⅐dC) complex. In both cases the enzyme is inhibited. The latter ternary complex requires that GC-box b binds to an allosteric site.
Dead-end inhibition with a CpG methylated homolog of GCbox b , GC-box bMET , also exhibited noncompetitive inhibition (Fig. 2B). Remarkably, a single 5-m C substitution reduces the inhibition constant associated with the allosteric site 340-fold, as K ii ϭ 20 Ϯ 3 nM. The pattern of lines is visibly less convergent than in Fig. 2A, and K ii /K is ϭ 0.57. The unmethylated and methylated inhibitors have distinct partitioning preferences as suggested by the changes in binding affinity and K ii /K is . These results strongly suggest that GC-box bMET and poly(dI⅐dC) can and do bind preferentially to distinct sites on Dnmt1, with the single methyl moiety driving that preference in the favor of the allosteric site.
The initial velocity patterns obtained by varying AdoMet and GC-box bMET concentrations were best fit to a standard equation for competitive inhibition (Fig. 2C). The intersection of the fit lines on the 1/velocity axis, K is ϭ 25 Ϯ 10 nM, suggests that GC-box bMET and AdoMet bind competitively to the same poly(dI⅐dC)-bound form of the enzyme but not necessarily to the same site on the enzyme. The inhibition constants determined for GC-box bMET at the proposed allosteric site, K is in Fig. 2C and K ii in Fig. 2B, are in good agreement at about 30 nM. In the absence of poly(dI⅐dC), GC-box bMET binds free enzyme with a dissociation constant of 880 nM (21). Therefore, poly(dI⅐dC) stabilizes GC-box bMET binding 29-fold. The binding of these duplex and single-stranded nucleic acids to a site on Dnmt1 distinct from the substrate-binding site is intriguing. The evidence for the existence of such an allosteric site is compelling: (i) the substrate inhibition described in Fig. 1 for both long and short substrates is best explained by the binding of a second DNA molecule; (ii) noncompetitive, dead-end inhibition by single-stranded oligonucleotides (Fig. 2,  A and B) was dramatically enhanced when the inhibitor contained 5-methylcytosine; (iii) no inhibition with either GC-box b or GC-box bMET was observed with two bacterial DNA cytosine methyltransferases, M.HhaI and M.SssI (data not shown); (iv) previous product inhibition studies support a Dnmt1-DNA-DNA ternary complex (20,22,23,25); (v) DNA concentrations 10 times higher than K m produce a second, less mobile band by gel mobility shift assays (20); (vi) short peptides from the N terminus of Dnmt1 have primary sequence similarity to zinc fingers and bind double-stranded DNA, although singlestranded DNA was not tested (24).
Allosteric enzyme regulators provide a novel approach to alter enzyme function and thus to introduce new strategies for drug design (50). Because GC-box bMET is a potent allosteric inhibitor, we sought to define those features that are essential for its activity. Derivatives of GC-box bMET were tested to determine how length, base, and backbone composition affect Dnmt1 inhibition (Table I). The IC 50 , defined as that concentration of inhibitor that causes 50% inhibition of the methylation reaction with poly(dI⅐dC), increased 2-fold in going from 30 to 50 nucleotides, being 15 and 30 nM, respectively. Significantly weaker binding was observed with 22 and 14 nucleotide lengths. Thus the enzyme appears to interact with at least 30 nucleotides through this allosteric site. Changing the backbone to a phosphorothioate, GC-box pMET , decreased the IC 50 from 15 nM for the comparable 30-nucleotide GC-box bMET to 5 nM. This 3-fold better binding could derive from various effects, including altered electrostatics to subtle changes in the allowed conformations of the oligonucleotides. Sequence specificity was demonstrated with CRE aMET , a relatively adenine/thyminerich element compared with the GC-box inhibitors. CRE aMET

TABLE I In vitro inhibition analysis of oligonucleotides
The concentration of inhibitor required to achieve 50% inhibition of a control reaction was determined for each and is shown as the IC 50 . K ii values were estimated from the mechanistic studies in Fig. 2 (ND, not determined). All data were collected in triplicate, and estimated errors are less than 10%. was at least 60-fold less inhibitory than our most potent inhibitor, GC-box p MET (Table I), and showed competitive inhibition against poly(dI⅐dC) (data not shown). We suggest that CRE aMET and the substrate DNA (poly(dI⅐dC)) bind the same form of the enzyme, a binding that most likely occurs at the active site. This is supported by our observations that the nonmethylated version of CRE aMET (CRE a) is an excellent substrate for Dnmt1 (41) and that CRE aMET shows competitive inhibition.
Our data provide the first evidence that an allosteric site on Dnmt1 shows DNA sequence preferences, which almost certainly opens up novel regulatory pathways. The data provided here and by others show that one or more allosteric site(s), most likely residing within the N-terminal domain, can bind double-or single-stranded nucleic acids. This binding is highly dependent on methylation status and DNA sequence, as well as on length and backbone composition. Although much needs to be learned about the site within the protein, its interactions with nucleic acids, and its biological role, there can be no question that the function of Dnmt1 is regulated by an allosteric mechanism.
Novel in vitro modulators of DCMTase activity could possibly alter genomic methylation patterns, alter gene expression and control the unrestricted proliferation of cancerous cells (3, 6, 51). The epigenetic inactivation of genes is as important a driving force in tumorigenesis as the inactivation of genes by mutation (52). Epigenetic transcriptional repression has been demonstrated in diverse tumor types including tumor suppressor genes, DNA repair genes, cell-cycle genes, and genes involved in invasion and metastasis (5,12,52). The re-expression of such genes in tumor cells leads to cell growth suppression and altered sensitivity to existing cancer therapies. However, despite encouraging results with available approaches to alter cellular DNA methylation, there is clearly a need for potent Dnmt1 inhibitors with improved drug characteristics (3,6,51). Based on the nanomolar potency of GC-box pMET , its novel mechanism of inhibition, and its phosphorothioate backbone, we sought to determine whether this compound could alter methylation in nuclear extracts as well as in cells. We tested the effectiveness of GC-box pMET in nuclear extracts from MEL cells (41). Potent inhibition of DNA methylation was observed with an IC 50 of 45 nM (Fig. 3A). A direct comparison with our results obtained with purified Dnmt1 (Table I) is complicated by several factors, such as Dnmt1 concentration and allosteric site accessibility. If significant levels of other DCMTase family members exist in MEL nuclear extracts (1, 3), then the inhibitor appears to regulate them also, because methylation activity was potently decreased to background levels.
To test the efficacy of the inhibitor in intact cells, we treated MEL cells with a mock inhibitor solution, 1.5 M 5AC, 7.7 M GC-box pMET , or 18 M of an antisense to Dnmt1 mRNA. The antisense oligonucleotide was previously shown to alter cellular DNA methylation levels and is in Phase II clinical trials for the treatment of cancer (8). Genomic DNA was isolated after 72 and 110 h of cell culture, and the percentage of 5-m C within CpG sites was used to assess genomic methylation levels (44). The 5-m C content decreased in a time-dependent manner in all treatment cases (Fig. 3B). A decrease in 5-m C was evident after 110 h using 5AC (Fig. 3B). Likewise, GC-box pMET was effective in reducing the 5-m C content, whereas the Dnmt1 antisense oligonucleotide was the least effective. Although it was not as effective as 5AC under these conditions, GC-box pMET did not appear to damage the cells; 5AC did damage the cells, resulting in significant degradation of cellular DNA. These results on genomic methylation levels were complemented by our characterization of the impact of the inhibitor on the methylation of a specific locus, in which inappropriate methylation was previously shown to cause gene expression changes leading to tumorigenesis. The HT29 human colon cell line was used to examine the ability of LipofectAMINE-transfected GC-box bMET to cause gene-specific demethylation of the tumor suppressor gene, p16 (12,53). The aberrant methylation of the CpG island associated with the p16 gene occurs in diverse human cancers with frequencies up to 48% (lymphoma), 39% (pancreatic), and 37% (colon) (53). Reversal of this methylation with 5AC treatment results in re-expression of p16. MSP was used to determine the methylation status of individual CpG sites (54). Treatment of the HT29 cells with 5AC results in demethylation (Fig. 4A, lane 4) and expression of p16 (Fig. 4B, lane 3) and was used as a positive control. A similar experiment performed with LipofectAMINE-transfected GC-box bMET showed specific demethylation of p16 (Fig. 4A, lane 3) and re-expression of the gene (Fig. 4B, lane 4), whereas a control oligonucleotide (LC2, see "Experimental Procedures") had no effect on p16-specific demethylation in HT29 (Fig 4, A, lane 2 and B, lane 5). The LC2 control results argue persuasively for a specific effect deriving from the inhibitor on the cellular processes leading to the p16 CpG island methylation.
In conclusion, we and others have previously identified low affinity interactions between Dnmt1 and various nucleic acids (22,23,25,30,41,48,55), effects that predominately involve the active site of the enzyme. Although the underlying mechanisms remain obscure in most of these cases, it is clear that they are distinct from the inhibition described in this report. Cell-active, tight binding, allosteric Dnmt1 inhibitors may form the basis of a new class of cancer drugs. Such therapies are likely to involve treatment with other epigenetically based drugs, such as histone deacetylase inhibitors, and to complement chemotherapies that are vulnerable to methylation-dependent resistance mechanisms (33). Although oncogenic transformation has been reversed with 5AC, allosteric inhibitors can perhaps separate the beneficial effects of gene reactivation from the known cytotoxic effects of 5AC.