The DNMT1 Target Recognition Domain Resides in the N Terminus*

DNA-cytosine-5-methyltransferase 1 (DNMT1) is the enzyme believed to be responsible for maintaining the epigenetic information encoded by DNA methylation patterns. The target recognition domain of DNMT1, the domain responsible for recognizing hemimethylated CGs, is unknown. However, based on homology with bacterial cytosine DNA methyltransferases it has been postulated that the entire catalytic domain, including the target recognition domain, is localized to 500 amino acids at the C terminus of the protein. The N-terminal domain has been postulated to have a regulatory role, and it has been suggested that the mammalian DNMT1 is a fusion of a prokaryotic methyltransferase and a mammalian DNA-binding protein. Using a combination of in vitro translation of different DNMT1 deletion mutant peptides and a solid-state hemimethylated substrate, we show that the target recognition domain of DNMT1 resides in the N terminus (amino acids 122–417) in proximity to the proliferating cell nuclear antigen binding site. Hemimethylated CGs were not recognized specifically by the postulated catalytic domain. We have previously shown that the hemimethylated substrates utilized here act as DNMT1 antagonists and inhibit DNA replication. Our results now indicate that the DNMT1-PCNA interaction can be disrupted by substrate binding to the DNMT1 N terminus. These results point toward new directions in our understanding of the structure-function of DNMT1.

Vertebrate genomes are modified by methylation of ϳ60 -80% of the cytosines residing at the CG dinucleotide sequence (1). The distribution of methylated cytosines is not random, resulting in gene-and tissue-specific patterns of methylation (2). A large body of evidence supports the hypothesis that both methylation patterns and activity of DNA methyltransferases (DNMTs) 1 play critical roles in development and in controlling genome functions such as differential gene expression, chromosome imprinting, and X-chromosome inactivation (3)(4)(5). It has also been suggested that DNMT1 is a downstream effector of many oncogenic pathways and a potential target for anticancer therapy (6 -11). We have previously demonstrated that inhibition of DNMT1 leads to an inhibition of DNA replication (12). Recently, it has been shown that DNMT1 is able to form a complex with Rb, E2F, and HDAC1 and repress E2F-responsive expression (13,14). Furthermore, it has been shown that DNMT1 can establish a transcriptional repressive complex with HDAC2 and DMAP1 at replication foci (15). These data suggest that DNMT1 has multiple functions in the cell. However, because the DNMT1 target recognition domain is unknown it is not possible to determine how these multiple functions and protein-protein interactions relate to its target specificity.
If DNA methylation patterns contain significant information, there must be a mechanism that ensures its proper inheritance in cell lineages. Razin and Riggs (16) have proposed that patterns of methylation are inherited, because DNMT1 is more proficient in methylating hemimethylated DNA than nonmethylated DNA. This hypothesis has been verified by a number of experiments (17,18). Another level of specificity is the ability of DNMT1 to recognize CG sequences almost exclusively (19,20). Thus, DNMT1 exhibits both substrate and sequence specificity.
The mammalian DNMT1 is a protein postulated to be composed, based on its similarity to other cytosine DNA methyltransferases, of at least three structural components (21)(22)(23)(24)(25)(26). These domains are as follows: a catalytic domain at the C terminus, an N-terminal domain that is responsible for localization of the protein to the nucleus and replication foci, and another poorly characterized central domain. It is unclear as yet which segment is responsible for determining its specificity for hemimethylated CG sequences. Previous reports have shown that when the N-terminal domain is cleaved by proteolysis, the enzyme loses its ability to discriminate between hemimethylated and unmethylated DNA (23). It has been suggested that the N-terminal domain performs a regulatory role by inhibiting the de novo methylation activity of the C-terminal domain (23). Recently, it has been demonstrated that a mouse prokaryotic methyltransferase hybrid, DNMT1-HhaI, containing the intact N terminus of DNMT1 and most of the coding sequence of HhaI, has a 2.5-fold preference for hemimethylated DNA, whereas HhaI by itself has preference for unmethylated DNA (27). This finding indicates that the N terminus is responsible for binding specificity to hemimethylated DNA. DNA binding activity has been shown to reside within the N-terminal domain, but this binding activity does not show sequence specificity, and it has been suggested that it determines the distance traversed between replication and methylation (28).
To dissect the role that DNMT1 plays in different biological functions using structure-function analysis, one has to determine which domain of the DNMT1 is responsible for target recognition. Identifying the target recognition domain of the enzyme is also critical for developing direct inhibitors of this enzyme as potential anticancer agents. To map the target recognition domain, we utilized a solid-state hemimethylated DNMT1 substrate to test the binding affinities of in vitrotranslated DNMT1 deletion mutant peptides. This method enabled us to determine which segment of DNMT1 per se is responsible for target recognition. It has been previously demonstrated that these modified hairpin oligonucleotide substrates are able to form a stable complex with DNMT1 and inhibit its activity (29). Furthermore, inhibition of DNMT1 with these modified hairpin oligonucleotides results in both inhibition of DNA replication (12) and transcriptional up-regulation of the tumor supressor p21 (30). However, the mechanism by which inhibition of DNA replication occurs remains unclear. In this manuscript, we show that the target recognition domain of DNMT1 does not reside in the previously defined catalytic domain but rather in the N-terminal region of DNMT1, between amino acids 122 and 417. This result demonstrates the fundamental disparity between mammalian and bacterial DNA methyltransferases and the inadequacy of amino acid sequence-sequence similarities per se in predicting the functional role of protein domains. The identification of the target recognition domain in the N terminus points toward new directions in our understanding of the structure-function relationship of mammalian DNA methyltransferases.

EXPERIMENTAL PROCEDURES
Generating DNMT1 Deletion Mutant Constructs-DNMT1 cDNA deletion mutants were generated by RT-PCR (31) from 1 g of total RNA prepared from the human small cell lung carcinoma cell line H446 (ATCC number HTB-171) using the following set of primers: 5Јccccatcggtttccgcgcgaaaa 3Ј (sense) and 5Јgcatctgccattcccactct 3Ј (antisense) for codons 1 to 125; 5Јgcaaacagaaataaagaatc 3Ј (sense) and 5Јgtgatggtggtttgcctggt 3Ј (antisense) for codons 122 to 170; 5Јggcaagggaaaagggaaggg 3Ј (sense) and 5Јgtccttagcagcttcctcct 3Ј (antisense) for codons 1113 to 1616; 5Јttatccgaggagggctacct 3Ј (sense) and 5Јcccttccctttgtttccagggc 3Ј (antisense) for codons 122 to 1112; 5Јttatccgaggagggctacct 3Ј (sense) and 5Јcgccggcgcttaaaggcgtt 3Ј (antisense) for codons 122 to 652. Amplification conditions were as follows: 95°C for 0.5 min, 60°C for 0.5 min, and 68°C for 5 min for 30 cycles using Promega Taq polymerase. The PCR products were cloned in a pCR3.1 vector (Invitrogen), and the sequence of the cDNAs was verified by the dideoxy-chain termination method (32) using a T7 DNA sequencing kit (Amersham Pharmacia Biotech) and alignment to the published human Dnmt1 sequence (33,34). To generate construct M, we cleaved the pCR3.1 bearing codons 1113-1616 with EcoRI, and the fragment was blunted and ligated to a pCR3.1 vector bearing codons 122 to 1212, which was cleaved at the 3Ј EcoRV site. Construct FTR bears the RT-PCR product encoding codons 122-652 inserted in the EcoRI site of pCR3.1. Constructs 5Ј-FTR and 3Ј-FTR were generated by digestion of construct FTR with BstEII and NotI. The resulting fragments were blunted and ligated separately into pCR3.1. Construct N-FTR bears an RT-PCR product encoding codons 122 to 1112 inserted in the EcoRI site of pCR3.1. To generate construct N-CAT, the blunted fragment bearing codons 1113-1616 was inserted into the EcoRV site of a pCR3.1 bearing codons 122-652. To generate construct CAT, we inserted the 1113-1616 coding fragment into the EcoRV site of a pCR3.1 bearing codons 122 to 168. To generate construct DNMT1, we cleaved construct M with XbaI and ligated it into a pCR3.1 vector bearing codons 1-122, which was cleaved at an internal XbaI site. To generate construct PS (the cysteine at the catalytic dipeptide proline-cysteine is replaced with a serine), the catalytic domain fragment encoding codons 1113 to 1616 was subcloned in the EcoRI site of the pAlter (Promega) site-directed mutagenesis vector. Mutagenesis was performed according to the manufacturer's protocol using a primer containing a single mismatch in the codon encoding the cysteine in the catalytic dipeptide, 5ЈAAGCCCTGGGAGGGCGGCC 3Ј (the underlined G is mismatched). The mutation was verified by sequencing, and the mutated catalytic domain was cleaved with EcoRI, blunted, and inserted into the EcoRV site of a pCR3.1 vector bearing codons 122-1112. Construct DNMT1 corresponds to the DNA-cytosine-5-methyltransferase initiated at the upstream ATG site, and construct M corresponds to the one initiated at the downstream ATG site. The insertion of the catalytic domain into the EcoRV site of pCR3.1 leaves a linker between the GK repeat and the ATG sequence (SRILQIIRLG). To ascertain that this linker does not impair the activity of the catalytic domain, it was inserted in the same position in the full-length DNMT1 constructs M and DNMT1 with no observed effect on DNMT1 binding or enzymatic activities (data not shown).
Coupled in Vitro Transcription and Translation-The peptides encoded by the constructs described above were transcribed and translated by coupled transcription-translation using the Promega TNT reticulocyte lysate kit (according to manufacturer's protocol) using 2 g of each construct and 40 Ci of [ 35 -S]methionine (l,000 Ci/mmol; Amersham Pharmacia Biotech) per 50 l of reaction volume.
Solid-state DNMT1 Substrate and Other Oligonucleotides-All phosphorothioate hairpin oligonucleotides used for the binding study (see Fig. 2 for sequences) were synthesized at Hybridon Inc. using standard phosphoramadite chemistry as previously described (10). Strepavidincoated magnetic beads (Dynabeads M-280 streptavidin) were purchased from Dynal. The 5Ј biotinylated phosphorothioate hemimethylated hairpin oligonucleotides (B1 and B2; see Fig. 2) (3 M) were incubated with 5 mg of beads in a buffer containing l0 mM Tris-HCl, EDTA (pH 7.4), and 1 M NaCl for 10 min at room temperature. The beads were then washed and resuspended in 500 1 of the same buffer. The final concentration of the oligonucleotide bound to the beads was 150 pmol per mg. To assay binding to the hairpin-bound beads, 3 l of in vitrotranslated polypeptides were preincubated in a 30-l reaction mixture including nonspecific or specific competitor oligonucleotides (l00 M), l0 mM Tris-HCl, 1 mM EDTA (pH 7.4), protease inhibitors (phenylmethylsulfonyl fluoride, aprotinin, and sodium vanadate, 1 mg/ml), and 40 mM NaCl. For the experiments in which CAT and FTR peptides were concurrently present (see Fig. 5), 3 l of each of the in vitro-translated peptides was utilized where the plus sign (ϩ) is indicated, and 6 l and 9 l were used where indicated by a triangle. After 30 min of pre incubation with competitor oligonucleotide, the mixture was applied to 500 g of hairpin-bound beads (mix 1/1 of oligonucleotides B1-and B2-coated beads) for 5 min. The beads and supernatant were separated by a magnet, and the beads were washed twice with 50 1 of l0 mM Tris-HCl, EDTA (pH 7.4) buffer followed by one wash in a 1 M NaC1containing buffer. The beads were resuspended in 30 l of l0 mM Tris-HCl, EDTA (pH 7.4), 0.1% SDS solution and boiled for 5 min. The supernatant (fraction S, corresponding to unbound proteins) and boiled fraction (fraction B, corresponding to bound proteins) were loaded onto an SDS-PAGE gel, dried, and exposed to autoradiography. The autoradiograms were scanned, the amount of proteins in each fraction was quantified, and the percentage of protein stably bound to the beads was calculated as boiled/(boiled ϩ supernatant) ϫ 100.
Cell Culture and Transient Transfections-HEK 293 cells, a human adenovirus type 5 transformed human embryonal kidney cell line (35) (ATCC number CRL 1573), were grown in Dulbecco's modified Eagle's medium (high glucose) supplemented with 10% fetal calf serum and 2 mM glutamine. For transient transfection experiments, HEK 293 cells were plated 18 h prior to transfection at a density of 7 ϫ 10 5 cells/ 100-mm tissue culture dish. The cells were transfected with 5 g of Xpress-FTR plasmid using the calcium phosphate protocol (36). The medium was replaced 24 h after transfection, and the cells were harvested 48 h after transfection.
Immunoprecipitations and Western Blot Analysis-Nuclear extracts were isolated as described previously (37). For the immunoprecipitation assay, 400 g of nuclear extracts were incubated with 10 l of the agarose-conjugated PCNA antibody (PC10; Santa Cruz Biotechnology) at 4°C overnight. For the competition experiments the reactions were preincubated with 100 nM final concentration of competitor oligonucleotides (Sc and H1) for 1 h before addition of the PCNA antibody. After the overnight incubation, the beads were spun and washed three times with phosphate-buffered saline. The immune complexes were resolved by a 7.5% SDS polyacrylamide gel electrophoresis. After transferring to a polyvinylidene difluoride membrane and blocking the nonspecific binding with 5% milk, Xpress-FTR protein was detected using mouse anti-Xpress antibody (Invitrogen) at a 1:5000 dilution, followed by peroxidase-conjugated anti-mouse secondary antibody (Jackson ImmunoResearch) at a 1:20000 dilution, and enhanced chemiluminescence detection kit (Amersham Pharmacia Biotech).
Nuclear Extract Binding Assay-Nuclear extracts were isolated as previously described (37), and binding was performed utilizing 500 g of hairpin-bound beads (mix 1/1 of oligonucleotides B1-and B2-coated beads) for 5 min. The beads and supernatant were separated by a magnet, and the beads were washed twice with 50 1 of l0 mM Tris-HCl, EDTA (pH 7.4) buffer followed by one wash in a 1 M NaC1-containing buffer. The beads were resuspended in 30 l of l0 mM Tris-HCl, EDTA (pH 7.4), 0.1% SDS solution and boiled for 5 min. The supernatant (fraction S, corresponding to unbound proteins) and boiled fraction (fraction B, corresponding to bound proteins) were loaded onto an SDS-PAGE gel, and Western blot analysis was performed as above.

Coupled in Vitro Transcription/Translation of Recombinant
Human DNMT1 Deletion Mutants-To map the DNMT1 target recognition domain, DNMT1 deletion mutant peptides were created using a coupled in vitro transcription/translation rabbit reticulocyte system (Fig. 1, A and B). The constructs were designed as follows: DNMT1 bears the entire coding sequence of the enzyme starting at the upstream ATG initiation site (34), construct M bears the coding sequence of the DNMT1 starting at the downstream ATG initiation site (33), construct PS bears a single base mutation converting the previously proposed catalytic site proline-cysteine dipeptide to proline-serine (24), construct N-FTR bears the N-terminal fork-targeting region and the central domain region contained within amino acids 122-1112, construct N-CAT bears the N-terminal and catalytic domains with a deletion of the central domain from amino acid 652 to 1113, construct CAT encodes the entire catalytic domain as proposed by homology to bacterial cytosine methyltransferases plus the N-terminal portion responsible for nuclear localization (21,23,24), construct FTR bears the N-terminal region between amino acids 122 and 652, construct 5Ј-FTR is a deletion of FTR containing the region between amino acids 122 and 417, and construct 3Ј-FTR contains the region between amino acids 417 and 731 (Fig. 1A). We utilized a luciferase construct as a control (Fig. 1A). Following coupled in vitro transcription/translation in the presence of [ 35 -S]methionine, the peptides were size-fractionated by SDS-PAGE and visualized by autoradiography (Fig. 1B). All the peptides migrated according to their expected sizes.
Oligonucleotide Design-Hemimethylated and nonmethylated CG-containing phosphorothioate hairpin oligonucleotides were designed as probes and competitors for our binding assays (Fig. 2). Because oligonucleotides B1 and B2 are biotinylated at the 5Ј arm of the hairpin, they can be conjugated to streptavidin-coated magnetic beads and used as solid-state probes for peptide binding experiments. The remaining oligonucleotides (Sc, H1, and N1) were utilized as competitors. The 5Ј arm of the hairpins containing methylated cytosines behaves as the methylation-guiding parental strand and the 3Ј arm behaves as the methyl-acceptor nascent strand of replicating DNA. The hairpin substrates we utilized contained an inosine (I) instead of the methyl acceptor, cytosine. These substrates have been previously shown to form a high affinity complex with DNMTl that could be dissociated only by boiling (29). Results from these experiments have also indicated that the inosine (I) group substitution, as well as synthesis of the backbone with a phosphorothioate modification, results in increased hairpin binding affinity to DNMT1 (29). Moreover, the CG sites present in the hairpins are separated by 4 bases to avoid tandem CGs, which have been demonstrated to be poor substrates for DNA methyltransferases (29).
The DNMT1 Target Recognition Domain Resides in the Nterminal Region-Previous studies have shown that DNMT1 forms a low affinity complex with both hemimethylated and nonmethylated DNA (19). It has been proposed that the initial binding of DNMT1 with DNA does not discriminate between hemimethylated and nonmethylated DNA (19) and that discrimination between the two substrates occurs during catalysis (19). In our binding assays (see Figs. [3][4][5], the ability of a polypeptide to form a stable complex with the magnetic bead- To test the binding specificity of the M peptide for the hemimethylated solid-state substrate, we incubated [ 35 -S]methionine-labeled in vitro-translated M peptide with the substrate in the presence of increasing concentrations of either a hemimethylated (H1) or a nonspecific (Sc) oligonucleotide. As observed (Fig. 3B), in the absence of any competitors (N), M is able to form a stable complex with the hemimethylated hairpin. This is indicated by the presence of M exclusively in fraction B (Fig. 3B). Complex formation is not challenged by an excess of 100 M nonspecific oligonucleotide (Sc), because M remains exclusively in fraction B. Conversely, complex formation is partly abolished by a challenge with 10 M of cold hemimethylated competitor (H1) and is completely abolished by a challenge with 100 M cold competitor, as indicated by the disappearance of M from B and its presence in S.
To determine which domain of DNMT1 interacts specifically with the hemimethylated substrate, we incubated the different [ 35 -S]methionine-labeled recombinant polypeptides with the solid-state hemimethylated substrate in the presence of 100 M of either hemimethylated (H1), nonmethylated (N1) CG bearing hairpin oligonucleotides or a nonspecific competitor (Sc) (Fig. 2). As observed in the autoradiograms (Fig. 4A) and in the graphical representations (Fig. 4B), all the polypeptides form a  (N), 100 M of a nonspecific phosphorothioate oligonucleotide (Sc), hemimethylated hairpin (H1), or nonmethylated CG-bearing hairpin (N1). Following 30 min of preincubation at room temperature, avidin-coated magnetic beads bound to biotinylated hemimethylated hairpins (Bl and B2) were added to the mix as described under "Experimental Procedures." The hairpin-bound (B) and unbound (S) fractions were separated on SDS-PAGE and exposed to autoradiography. B, bound (B) and unbound (S) fractions were quantified by densitometry, and the results were plotted as percentage bound.
stable complex with the solid-state substrate in the absence of any competitors. This is indicated by their presence in fraction B, suggesting that both C-terminal and N-terminal domains can recognize and form a stable complex with the substrate. However, the binding of the catalytic domain peptide (CAT) to the hemimethylated substrate is completely abolished by nonspecific (Sc), as well as specific, competitors (H1 and N1). This is evident from the presence of the peptide exclusively in fraction S and its complete absence from fraction B. The binding of peptides bearing either the N-terminal domain (N-FTR, N-CAT, FTR, and 5Ј-FTR) or the complete protein (M and DNMT1) is not competed by the nonspecific competitor (Sc) (Ͼ80% remain bound) but is abolished (Ͻ40% remain bound) by the hemimethylated competitor (H1). The nonmethylated CG-bearing hairpin (N1) is a partial competitor as indicated by the distribution of the polypeptide in both the bound and unbound fractions. In all of these cases the hemimethylated hairpin is 2-4-fold more effective in competing out binding to the solid-state substrate than the nonmethylated homolog (Fig.  4B). PS also shows a preference for the hemimethylated substrate, but interestingly it is competed less efficiently by the H1/N1 competitor pair, indicating the formation of a tighter complex. The fact that the competition profiles of DNMT1 and N-CAT are similar indicates that the central region of the enzyme (amino acids 652-1113) is not required for target specificity. This is confirmed by the similarity of competition profile of these two peptides with the FTR peptide competition profile. Moreover, results indicating that binding of F-CAT to the hemimethylated substrate is competed as efficiently by the nonspecific competitor (Sc) as it is by the hemimethylated competitor (H1) further confirm that the central region is not essential for target recognition. The 5Ј-FTR peptide still retains hemimethylated binding specificity, which is indicated by the fact that H1 competes for binding more efficiently than N1 and Sc. Interestingly, the binding specificity of 3Ј-FTR peptide for the hemimethylated substrate is weaker than the more inclusive FTR peptides, because all competitors have a similar effect on binding. Taken together, these results support the hypothesis that the target recognition domain of DNMT1 resides in the N-terminal segment (amino acids 122-417) of the protein, overlapping with the fork-targeting region.
We then tested whether the FTR peptide still binds specifically to hemimethylated DNA when both CAT and FTR peptides are present in the same reaction, as is the case with the natural product. We tested their affinity to hemimethylated DNA under two conditions, in the absence or in the presence of competitor DNA. The presence of competitor DNA mimics the in vivo scenario, where DNMT1 has to discriminate between its specific target and the bulk of non-CG DNA. The results (Fig.  5A) demonstrate that in the absence of any competitors both CAT and FTR peptides can bind the target, in agreement with results from Fig. 4, showing that both peptides have high affinity to DNA. However, in the presence of a nonspecific competitor (Fig. 5B), as is the case when DNMT1 interacts with the genome in vivo, only FTR is bound to hemimethylated DNA. Having both CAT and FTR peptides concurrently also verifies that the differences observed between distinct peptides in the competition experiments depicted in Fig. 4 are not the result of impurities in the in vitro translation reaction of one peptide versus another. For clarity only the bound fractions were loaded.
PCNA/FTR Complex Can Be Disrupted by a Hairpin Oligonucleotide-Previous reports have shown that modified hairpin oligonucleotides can work as bona fide antagonists of DNMT1 and inhibit DNA replication (12,29). Because the target recognition of DNMT1 may overlap with its PCNA binding site, we tested the hypothesis that the DNMT1/PCNA complex could be disrupted by FTR binding to the modified hairpin oligonucleotides. We transfected HEK 293 cells with an Xpress-tagged FTR expression vector followed by PCNA immunoprecipitations in the presence of a nonspecific oligonucleotide (Sc), a hemimethylated hairpin (H1), or no competitor (N). The results (Fig. 6A, top panel) indicate that the PCNA/DNMT1 complex can be disrupted by H1 but not by Sc. The presence of IgG and PCNA in all the immunoprecipitations was also tested (Fig. 6A,  middle and bottom panels). We also immunoblotted the supernatants of these immunoprecipitations to confirm FTR expression (Fig. 6A, bottom panel). Because the previous experiments had been performed using in vitro-translated peptides, we tested the ability of the FTR peptide to bind the hairpin solidstate substrate in HEK 293 nuclear extracts. The results (Fig.  6B) demonstrate that the FTR peptide does indeed bind the solid substrate under these conditions, as indicated by the FTR presence in the bound fraction (B). These results suggest that the disruption of the PCNA/DNMT1 complex might be a possible mechanism by which these modified hairpin oligonucleotides inhibit DNA replication. DISCUSSION Recognition of hemimethylated CGs, the DNMT1 target sequence, is a critical step in the replication of the DNA methylation pattern and the epigenetic information that it encodes. Although the first mammalian dnmt has been cloned and its cDNA sequenced more than a decade ago (21,22), the domain responsible for its target recognition has not yet been clearly defined. Based on sequence homology between mammalian and bacterial cytosine methyltransferases the entire catalytic domain, including the target recognition domain, has been proposed to reside in the C terminus of the protein (24). The additional N-terminal and central domains of the protein were proposed to perform regulatory roles (23,28). This hypothesis is based on the assumption that the structure of the mammalian DNMT1 follows the rules laid out in bacteria and that DNMT1 is an evolutionary hybrid of a primordial DNMT plus at least two additional regulatory protein modules (23). Our results are in agreement with recent target specificity studies utilizing a mouse prokaryotic hybrid DNMT, DNMT1-HhaI, containing the intact N terminus of DNMT1 and most of the coding sequence of HhaI, demonstrating that it has a 2.5-fold preference for hemimethylated DNA, whereas HhaI by itself has preference for unmethylated DNA (27). Moreover, structural analysis of de novo DNMTs, DNMT3a and DNMT3b (4,38), which lack the DNMT1 N-terminal region and recognize nonmethylated CGs, further supports the hypothesis that the hemimethylated target recognition domain resides within the DNMT1 N terminus.
To delineate the DNMT1 target recognition domain, we performed experiments utilizing a solid-state hemimethylated substrate and in vitro-translated deletion mutant peptides. Competition assays with hemimethylated and nonmethylated substrates revealed that the N-terminal domain specifically recognizes a hemimethylated CG substrate. The previously described catalytic domain, which has been proposed to contain all the elements required for catalytic activity including target recognition (24), binds DNA but does not exhibit specificity to hemimethylated CG duplex DNA as determined by competition assays. Moreover, when the catalytic domain and the N-terminal domain are concurrently present in the reaction with nonspecific DNA, the substrate binds to the N-terminal domain suggesting that it bears the target recognition domain. Interestingly, the PS mutant is competed less efficiently by the H1/N1 competitor pair, indicating the formation of a tighter complex. This result is supported by a previous study indicating that EcoRII DNMT mutants with substitution of the conserved cysteine for serine bind tightly to DNA (39). These mutants resemble the wild-type enzyme in that their binding to substrate is not eliminated by the presence of nonspecific DNA in the reaction (39).
Deciphering whether the catalytic components of DNMT1 reside entirely in the C-terminal domain or whether they reside elsewhere is critical for both structure-function and crystal structure analysis of the protein, as well as for drug design. Identifying the location of the different components of the catalytic domains of DNMTs is also critical for interpretation of knockout experiments by homologous recombination, which target putative catalytic domains (5). If these alleles, which are inactivated at the C terminus, still maintain the sequencespecific target recognition domain some of the previously described biological effects of this knockout could be attributed to the DNA binding effects of the truncated protein rather than inhibition of methylation.
The catalytic domain should, by definition, include the substrate recognition domain, otherwise there is no catalysis. Our results shed doubt on the previous designation of the DNMT1 C terminus as the exclusive catalytic domain. The data presented in this paper suggest that this conclusion might be revisited. In contrast to many bacterial DNMTs (40), the target recognition domain of the mammalian DNMT1 resides at the N terminus, at a significant distance from the conserved catalytic and AdoMet binding domains. Our data is consistent with the hypothesis that the N-terminal is not just a regulatory domain, as has been previously suggested (23,41), but it is the substrate recognition module of the enzyme. Therefore, the catalytic component of the vertebrate enzyme is composed of two modular domains, a target recognition domain in the N terminus, in addition to the previously described AdoMet binding and methyl-transfer domains in the C terminus of the enzyme. The fact that the target recognition domain and methyl-transfer domain are located at such a distance on the linear amino acid sequence must not necessarily translate to a similar distance in the tertiary structure of the enzyme. The results presented here might explain the failure of previous attempts to demonstrate DNA methylation activity in the previously postulated catalytic domain of the enzyme (41). Despite the striking structural homology of the C terminus of the vertebrate to bacterial DNMTs, its expression in bacterial or mammalian cells does not result in DNA methylation activity (26). These results illustrate the risk in predicting biochemical functions exclusively from sequence homology.
We had previously demonstrated that the hairpin oligonucleotides utilized here can form a stable complex with DNMT1 (29), inhibit its activity, and inhibit DNA replication in parallel with a transcriptional up-regulation of the tumor supressor p21 (12,30). With the knowledge that the PCNA interaction domain of DNMT1 and its target recognition domain are overlapping, we were able to show that the PCNA/DNMT1 complex could be disrupted by the hemimethylated hairpins. These results might provide an alternative mechanism by which DNMT1 inhibition can supress oncogenic transformation (42). Because DNMT1 is a multifunctional protein, identifying the domains important for transformation is critical for DNMT1 anti-oncogenesis drug design. Moreover, the identification of the target recognition domain allows a more complete structure-function analysis of DNMT1, and it will help the development of novel direct inhibitors of this enzyme (7).