Recombinant Human DNA (Cytosine-5) Methyltransferase

A method is described to express and purify human DNA (cytosine-5) methyltransferase (human DNMT1) using a protein splicing (intein) fusion partner in a baculovirus expression vector. The system produces ∼1 mg of intact recombinant enzyme >95% pure per 1.5 × 109 insect cells. The protein lacks any affinity tag and is identical to the native enzyme except for the two C-terminal amino acids, proline and glycine, that were substituted for lysine and aspartic acid for optimal cleavage from the intein affinity tag. Human DNMT1 was used for steady-state kinetic analysis with poly(dI-dC)·poly(dI-dC) and unmethylated and hemimethylated 36- and 75-mer oligonucleotides. The turnover number (k cat) was 131–237 h−1 on poly(dI-dC)·poly(dI-dC), 1.2–2.3 h−1 on unmethylated DNA, and 8.3–49 h−1 on hemimethylated DNA. The Michaelis constants for DNA (K m CG) andS-adenosyl-l-methionine (AdoMet) (K m AdoMet) ranged from 0.33–1.32 and 2.6–7.2 μm, respectively, whereas the ratio ofk cat/K m CGranged from 3.9 to 44 (237–336 for poly(dI-dC)·poly(dI-dC)) × 106 m −1 h−1. The preference of the enzyme for hemimethylated, over unmethylated, DNA was 7–21-fold. The values of k cat on hemimethylated DNAs showed a 2–3-fold difference, depending upon which strand was pre-methylated. Furthermore, human DNMT1 formed covalent complexes with substrates containing 5-fluoro-CNG, indicating that substrate specificity extended beyond the canonical CG dinucleotide. These results show that, in addition to maintenance methylation, human DNMT1 may also carry out de novo and non-CG methyltransferase activities in vivo.

Methylated cytosine is found in the genome of organisms ranging from prokaryotes to mammals (1). Methylation of DNA in eukaryotes is implicated in various biological and developmental processes, such as gene regulation (2), DNA replication (3), genomic imprinting (4), embryonic development (5), carcinogenesis (6), and genetic diseases (7). The bulk of the methylation takes place during DNA replication in the S-phase of the cell cycle (8). The maintenance methylation ensures the propagation of tissue-specific methylation patterns established during mammalian development. The methyl transfer reaction proceeds via nonspecific binding of the enzyme to DNA, recognition of the specific DNA target site, and recruitment of the methyl group donor S-adenosyl-L-methionine (AdoMet) 1 to the active site of the enzyme. DNA (cytosine-5) methyltransferases (m 5 C MTase) introduce a methyl group onto carbon 5 of the target cytosine through a covalent intermediate between the protein and the target cytosine (9). During this process, the cytosine is flipped 180°out of the DNA backbone into an active site pocket of the enzyme (10). After completion of the methyl transfer reaction, the products, methylated DNA and S-adenosyl-L-homocysteine (AdoHcy), are released. Previous studies on the mechanism of methylation were mainly limited to prokaryotic m 5 C MTases although some limited kinetic studies have also been reported with mouse and human DNMT1 (11,12).
Eukaryotic m 5 C MTases have been cloned from various organisms such as mouse (13), human (14), Xenopus (15), sea urchin (16), chicken (17), Arabidopsis (18), and pea (19). The human DNMT1 has been localized to the chromosomal site 19 p13.2-p13.3 by fluorescence in situ hybridization (14). The enzyme consists of a large N-terminal region and a smaller Cterminal region linked by a run of Gly-Lys dipeptide repeats (14). The large N-terminal domain is unique to the eukaryotic m 5 C MTases. The smaller C-terminal domain has strong homology with prokaryotic m 5 C MTases and contains the elements necessary for catalysis (20). The function of the N terminus is poorly understood, and there is little sequence homology between plant and animal enzymes. Sequences within this domain have been implicated in Zn 2 binding (21), interaction with proliferating cell nuclear antigen (22), and targeting to replication foci (8). The full-length mouse m 5 C MTase is about 1700 amino acids in length and has a molecular mass of 183.5 kDa (23).
Prokaryotic m 5 C MTases are usually part of restriction-modification systems and rarely discriminate between unmethylated or hemimethylated DNA substrates. In contrast, mammalian m 5 C MTases show a much higher reaction velocity on a hemimethylated DNA substrate (maintenance methylation), the product of semiconservative DNA replication, than on an unmethylated DNA substrate (de novo methylation). Methylation of single-stranded DNA by the mammalian m 5 C MTase is strongly stimulated by the presence of nearby 5-methylcytosines (24). This rate of methylation was comparable with that of hemimethylated DNA (24). It has been demonstrated that the human enzyme can also methylate non-B-DNA structures such as hairpins and mismatches (12). Recently, a family of mammalian m 5 C MTases has been identified and hypothesized to serve as de novo methyltransferases (25).
The diverse role played by DNA methylation in mammals has motivated several groups to express mammalian m 5 C MTases in Escherichia coli (26), baculovirus (23,27), and mammalian (COS) cells (28). In the E. coli and COS cell expression systems, the amount of protein produced was low, whereas protein expression was high using the baculovirus system. Insect cells are rich in proteases, and long purification protocols often result in partial proteolytic degradation of proteins. To overcome the problem of degradation, we have constructed and expressed the human DNMT1 as a fusion protein with a Saccharomyces cerevisiae VMA1 intein gene (29). The C terminus of the human DNMT1 was fused to the N terminus of the intein. A small chitin-binding domain from Bacillus circulans has been added to the C terminus of the intein for affinity purification. Thus, the three-part fused human DNMT1 protein can be immobilized using chitin beads as an affinity matrix. 2 Addition of dithiothreitol (DTT) at low temperature initiates cleavage on the column between the intein and the recombinant human DNMT1. 2 Whereas the intein and chitin-binding domain remain attached to the column, the purified recombinant enzyme is eluted. This purified recombinant human DNMT1 is identical with the native enzyme except for the two C-terminal amino acids. In this report, we have studied its steady-state kinetic parameters on double-stranded, unmethylated, and hemimethylated DNA substrates, which are representative of human genes. We also assayed methylation of non-CG sequences by human DNMT1.

EXPERIMENTAL PROCEDURES
Human DNA (Cytosine-5) Methyltransferase Transfer Vector-Human DNMT1 expression constructs were derived from pBKSHMT5.0 (b) (gift of Prof. S. Baylin, The Johns Hopkins University). This plasmid had the full-length human DNMT1 cDNA based on the previously published sequence (14). A polymerase chain reaction of the 3Ј end of the cDNA was used to incorporate an SmaI restriction endonuclease site in place of the stop codon of the cDNA and to provide the optimal amino acids for protein cleavage (sense primer, 5Ј-GGAATTCCATAT-GCATGACAGGAAGAACGGCCGCAGC-3Ј and antisense primer, 5Ј-T-GACCCGGGAGCAGCTTCCTCCTCCTTTATTTTAGC-3Ј). The restriction sites are underlined. This change causes the C-terminal amino acids lysine and aspartic acid to become proline and glycine. Substitution of these two amino acids at the C terminus of the mammalian DNMT1 had no effect on the methyl transfer reaction as compared with the native sequence enzyme as judged by methylase assay (data not shown). This polymerase chain reaction product of about 500 bp was digested with NdeI and SmaI and ligated to a baculovirus vector pVIC1 2 to give p⌬VICHMT-4. This construct contains the 3Ј end of the human DNMT1 cDNA in frame with the intein chitin-binding domain of the transfer vector pVIC1. 2 p⌬VICHMT-4 was digested with NruI and EagI and a 4.1-kbp EcoRV-EagI fragment from pBKSHMT5.0(b) was ligated to give the functional human DNMT1 transfer vector pVICHMT (Fig. 1). The ligated junctions in the final pVICHMT construct were verified by DNA sequencing.
Insect Cell Culture, Viral Transfection, and Recombinant Protein Expression-A pupal ovarian cell line (SF9) from the worm Spodoptera frugiperda was used for co-transfection and expression of the human DNMT1 as described previously (23) with the following modifications. SF9 cells were maintained as a suspension culture in TNM-FH media (JRH Biosciences) supplemented with fetal calf serum 10% (v/v) and an antibiotic/antimycotic solution at a final concentration of 5 units of penicillin, 50 g of streptomycin, and 0.125 g of amphotericin B ml Ϫ1 at 27°C on a Bellco steering platform at 70 rpm. Co-transfection of a monolayer of SF9 insect cells was carried out using BaculoGold DNA, a modified, linearized Autographa californica nuclear polyhedrosis virus (AcNPV) DNA (PharMingen) and the transfer vector pVICHMT. Four days after transfection, the supernatant was screened for recombinant baculovirus using the agarose overlay technique (31). Individual plaques were purified and amplified 2 to 3 times in order to reach a viral titer above 2 ϫ 10 8 ml Ϫ1 . Recombinant clones from individual plaque isolates were used for test expression by infecting 1.5 ϫ 10 6 SF9 cells in a 60-mm Petri dish. Forty-eight hours post-infection the cell extracts were checked for methyltransferase activity and fusion protein expression by Western blot analysis using anti-intein antibody (New England Biolabs). For routine protein expression, SF9 cells were grown in spinner culture flasks. SF9 cells at a density of 1.4 ϫ 10 6 ml Ϫ1 were infected at a multiplicity of infection between 5 and 7. The cells were kept at 27°C at 60 rpm and harvested 48 h post-infection after a final wash with phosphate-buffered saline.
Recombinant DNA (Cytosine-5) Methyltransferase Purification-For protein purification, infected cells (5.6 ϫ 10 8 ) were resuspended in 15 ml of buffer M (50 mM Tris-HCl, pH 7.4, 1 mM Na 2 EDTA, protease inhibitor mixture containing 4-(2-aminoethyl)benzenesulfonyl fluoride, pepstatin A, E64, bestatin, leupeptin and aprotinin (Sigma), 0.2% (v/v) per ml of cell extract, 7 g/ml phenylmethylsulfonyl fluoride, and 500 mM NaCl). The cell suspension was sonicated on ice for 30 s using a model W-225R (Heatsystem-Ultrasonics) sonicator in pulsed mode at 50% duty cycle. The extract was incubated on ice for 30 min with occasional shaking and was centrifuged at 11,000 ϫ g for 30 min. If the supernatant remained cloudy, then one more centrifugation step was included. Purification from larger numbers (Ͼ1 ϫ 10 9 ) of cells required an initial 40 -70% ammonium sulfate fractionation. This supernatant was loaded on a chitin bead column (New England Biolabs) equilibrated with buffer M at 0.4 ml/min. The nonspecifically bound proteins were removed by washing with 30-column volumes of buffer M. On-column cleavage of the target protein was initiated by passing 2-column volumes of buffer M supplemented with 50 mM DTT. The column was closed, and the target protein was cleaved from the intein chitin binding domain by overnight incubation at 4°C. Human DNMT1 was eluted with buffer M and dialyzed against buffer M supplemented with 0.2% (v/v) protease inhibitor mixture (Sigma), 7 g/ml phenylmethylsulfonyl fluoride, 100 mM NaCl, 50% glycerol (v/v), and 1 mM DTT. The purified protein was stored at Ϫ20°C. The purity of the protein was checked by SDS-PAGE (4 -20% Tris/glycine/SDS gradient gel). Purified protein was quantitated using the Bradford assay with bovine serum albumin as standard.
DNA (Cytosine-5) Methyltransferase Assay and Data Analysis-m 5 C MTase assays were carried out at 37°C, for 30 min in duplicate with a total volume of 25 l of reaction mix. A typical reaction contained S-adenosyl-L-[methyl-3 H]methionine (AdoMet) (specific activity 15 Ci/ mmol, Amersham Pharmacia Biotech), substrate DNA, and enzyme in assay buffer (50 mM Tris-HCl, pH 7.8, 1 mM Na 2 EDTA, pH 8.0, 1 mM DTT, 7 g/ml phenylmethylsulfonyl fluoride, 5% glycerol, and 100 g/ml bovine serum albumin). For kinetic analysis, the CG or CI concentrations were calculated from either substrate oligonucleotides or poly(dI-dC)⅐poly(dI-dC). One nM double-stranded oligonucleotide with 1 CG site is 2 nM CG. For poly(dI-dC)⅐poly(dI-dC), an average length of 7000 bp was used to determine the concentration of CI. For doublereciprocal plots, the concentrations of CG or CI and AdoMet were varied. The MTase reactions were stopped by transferring the reaction tubes to an ethanol/dry ice bath and were processed by spotting the reaction mix on DE81 paper circles (Whatman). These circles were washed sequentially with 4 ϫ 1 ml of cold 0.2 M ammonium bicarbonate, 4 ϫ 1 ml of Milli Q water, and 4 ϫ 1 ml of ethanol using a manifold (Millipore) connected to a vacuum pump. The filters were dried; 3 ml of Opti-fluor (Packard) was added to each, and tritium incorporation was measured. To calculate counting efficiency, tritiated plasmid DNA (pUC DNA methylated with tritiated AdoMet using M.SssI) was either spotted on DE81 paper, processed, and counted or directly counted with Opti-fluor. The efficiency of [ 3 H]DNA measurement was ϳ55%, and all calculations were corrected accordingly. Data obtained were plotted by regression analysis using the GraphPad PRISM program (GraphPad Software Inc.) or SigmaPlot (SPSS Inc., Chicago, IL). The equations used for fitting the data points and to obtain the kinetic constants, V max , K m CG , K m AdoMet , are reported elsewhere (32). These figures represent estimated values.
Southern Blot Analysis-Genomic DNA was isolated from SF9 cells 48 h post-infection with recombinant virus using the Easy DNA Kit (Invitrogen). The purified DNA was digested either by AatII, BsiEI, or SwaI, and the digested DNA fragments were separated on a 1% agarose gel in TBE buffer gel. The DNA in the gel was blotted on to Hybond-Nϩ and probed with a random-primed human DNMT1 clone. The blot was washed as described (33) and autoradiographed.
Western Blot Analysis-Cell extracts were mixed with SDS loading dye with 2 M Tris-(2-carboxyethyl) phosphine in place of DTT, boiled at 95°C for 5 min, and loaded on a 4 -20% Tris/glycine/SDS gradient gel. Tris-(2-carboxyethyl) phosphine is a strong reducing agent and does not take part in the cleavage process (34). The proteins were blotted on an Immobilin-P membrane and probed with an anti-intein antibody 2 M. E. Scott, S. Pradhan, and M. Q. Xu, submitted for publication.
(New England Biolabs) followed by chemiluminescent detection of the human DNMT1.
5-Fluoro-2Ј-deoxycytidine Assay-Duplex oligonucleotides containing 5-fluoro-2Ј-deoxycytidine (FdC) were 32 P-end-labeled using polynucleotide kinase and [␥-32 P]ATP. Human DNMT1 (10 nM) was used with 5 nM fluorocytosine-containing oligonucleotide duplex in the presence of 50 M cold AdoMet in 1ϫ buffer M at 37°C for 30 min. The reaction was stopped by adding SDS gel loading dye (one-third of the reaction volume) and boiling the sample at 100°C for 5 min. The boiled protein/ DNA mixtures were loaded and resolved on a 4 -20% Tris/glycine/SDS gradient gel. The gel was dried and autoradiographed.
Oligonucleotides-Poly(dI-dC)⅐poly(dI-dC), average length 7000 bp, was obtained from Amersham Pharmacia Biotech. The following oligonucleotides were synthesized at New England Biolabs. M is 5-methylcytosine; and F is 5-fluoro-2Ј-deoxycytidine. Duplexes were prepared by annealing equimolar amounts of appropriate single strands. Complementary oligonucleotides were incubated in annealing buffer (50 mM Tris-HCl, pH 7.4, 1 mM Na 2 EDTA, and 100 mM NaCl) at 95°C for 5 min followed by 65°C for 15 min and 37°C for 15 min. For the oligonucleotides corresponding to the CGG⅐CCG repeat of the FMR-1 locus, the annealing step was followed by cleavage with S1 nuclease. This additional step was included to eliminate any singlestranded nucleotide that resulted from imperfect annealing between the two complementary 36-mers. PAGE analyses of the annealed duplexes showed that ϳ90% of the products had the expected length of 36 bp, whereas ϳ10% were one repeat shorter (11 repeats, 33 bp in length). The full-length duplexes were purified on a 15% polyacrylamide gel, dissolved in TE, and stored at Ϫ20°C.

RESULTS
Expression and Purification of the Biologically Active Human DNA (Cytosine-5) Methyltransferase-Co-transfection of the human DNMT1 transfer vector, pVICHMT, with linear AcNPV DNA (BaculoGold DNA, PharMingen) resulted in homologous recombination of the polyhedrin promoter and the DNMT1 cDNA carried by the construct (Fig. 1, A and B). Only the recombinant viruses are viable. The transfection supernatant was harvested and used for purification of several single plaques. Recombinant viral DNA from each of the plaques was isolated and checked for correct recombination of the human DNMT1-intein-chitin binding domain construct using restriction enzyme digestion and Southern blotting. The probe used was the full-length human DNMT1 cDNA. As expected, an SwaI digest gave an ϳ10-kbp band, whereas AatII and BsiEI gave two bands each (Fig. 1C). Several clones had double homologous integration with the human DNMT1 coding sequence lying downstream of the polyhedrin promoter, and one, VICHMT8, was used for all further expression and purification studies. Two different insect cell lines, SF9 and High Five (Trichoplusia ni), were used for the evaluation of protein levels. Different multiplicities of infection were used, and the extracts were evaluated at 48 h post-infection for DNMT1 expression using poly(dI-dC)⅐poly(dI-dC) as substrate. High Five cells have been reported to express between 5-and 10-fold higher protein levels than SF9 cells (35), but for human DNMT1 the protein expression did not improve (data not shown). All further experiments used the SF9 cell line for expression. To follow expression and find optimal conditions for production, extracts of the human DNMT1 clones were assayed by Western blots using anti-intein antibody. The optimal expression of the enzyme was 48 h post-infection (Fig. 1D), and the human DNMT1-specific band starts decreasing after 54 h post-infection infection (data not shown).
Mouse DNMT1 is highly unstable and susceptible to proteolysis (13). The protease-susceptible domains in mouse and human DNMT1 are similar (14). To minimize degradation, all steps of the purification were carried out at 4°C using buffers containing a mixture of protease inhibitors supplemented with additional E64, a strong inhibitor of the cysteine protease, which is abundant in insect cells. During sonication, the presence of 500 mM NaCl ensures that most of the enzyme, which is normally tightly bound to DNA, remains soluble (36). Most of the enzyme was bound to the chitin-bead affinity matrix. Overnight incubation at 4°C with 50 mM DTT induced cleavage on the column and released most (Ͼ90%) of the full-length recombinant human DNMT1 (Fig. 2). This method yielded protein that was at least 95% pure as determined by SDS-PAGE on a 4 -20% Tris/glycine gradient gel and Coomassie staining (Fig. 2). The dialyzed protein was stored at Ϫ20°C for 4 weeks with little loss of activity. However, significant loss of activity appeared upon prolonged storage (Ͼ6 weeks) at either Ϫ20 or Ϫ80°C.
Steady-state Kinetic Properties of Recombinant Human DNA (Cytosine-5) Methyltransferase on Poly(dI-dC)⅐Poly(dI-dC) Substrate-Historically, poly(dI-dC)⅐poly(dI-dC) has been used to define the biochemical properties of mammalian m 5 C MTases (36,37). Each molecule of poly(dI-dC)⅐poly(dI-dC) provides a large number of potential (CI) dinucleotide sites for methylation. It has also been claimed that this polymer can undergo methylation at rates comparable to a hemimethylated DNA substrate (38). Thus, we used this substrate to define the biochemical properties of the recombinant human DNMT1. The time course of methylation for poly(dI-dC)⅐poly(dI-dC) was examined at constant DNA and enzyme concentrations. The reaction was found to be linear for the first 30 min (Fig. 3A), during which time each enzyme molecule undergoes several rounds of catalysis. This is evident from the fact that 1 nmol of human DNMT1 incorporated several nanomoles of methyl groups onto the substrate DNA (Fig. 3, A and B). Similar reactions were carried out with fixed concentrations of DNA and AdoMet but variable amounts of enzyme. The rate of methylation was linear at enzyme concentrations between 1 and 4 nM (Fig. 3B). The resulting curves (Fig. 3, A and B) were used to determine the optimal enzyme and substrate concentrations under which linear methylation reaction was obtained. Based on the above observations, a series of double-reciprocal plots was made at six different AdoMet concentrations or six differ-ent CI (DNA) concentrations, with poly(dI-dC)⅐poly(dI-dC) having an average length of 7000 bp. With this procedure the reciprocal of the amount of [ 3 H]CH 3 transferred to the DNA in 1 min by 1 nM DNMT1 (1/v) is plotted as a function of 1/substrate concentration (either 1/CI or 1/AdoMet) in the presence of fixed concentrations of the co-substrate. For bireactant enzymes, which use two substrates and give two products such as DNMT1, these double-reciprocal plots usually yield linear regressions, and their slopes and intercepts are then replotted to derive all the kinetic constants. However, for poly(dI-dC)⅐poly(dI-dC), both double-reciprocal plots, i.e. those with 1/CI as the variable substrate (Fig. 4A) as well as those with 1/AdoMet as the variable substrate (Fig. 4B), were not linear. In particular, the increases in 1/v at low 1/CI (i.e. at high DNA concentration) indicated that the DNA substrate inhibited the reaction, especially in the presence of low amounts of AdoMet (Fig. 4A). We conclude that poly(dI-dC)⅐poly(dI-dC) acts both as a substrate and as an inhibitor of the methyl transfer reaction with human DNMT1. The regressions from the plots of 1/v versus 1/AdoMet (Fig. 4B) were concave, suggesting some positive cooperativity due to enzyme activation at high AdoMet concentrations. Overall, these data indicate that the methyl transfer reaction with poly(dI-dC)⅐poly(dI-dC) by human DNMT1 does not follow simple kinetics.
Since the derivation of the Michaelis constants and V max requires an estimate of the slopes and intercepts from the lines in Fig. 4, the two data points at low 1/CI that deviated from linearity at the four lowest AdoMet concentrations (Fig. 4A) were excluded from the regressions (32, 39). Fig. 5, A and B, shows that, with the exception of 1 M AdoMet, such slopes and intercepts varied linearly with respect to 1/AdoMet concentration, which enabled all the kinetic constants to be evaluated (32). The intercept values from Fig. 4B were also replotted ( Fig.  5C) and indicated that V max(app) varied linearly with the DNA at infinite AdoMet concentrations. Furthermore, the 1/V max obtained from these data (y axis intercept of Fig. 5C) agreed closely with that estimated from Fig. 5B (y axis intercept), giving confidence in the method used. Due to their curvature, the slopes of Fig. 4B were not used to verify the kinetic constants obtained from Fig. 4A (40). The equation is based on the assumption that the reaction is ordered, with AdoMet binding before the DNA to the catalytic center of DNMT1. However, this assumption needs to be verified. The K AdoMet is the dissociation constant for AdoMet, for which the equation gives a value of ϳ1.0 M. K m AdoMet , K m CG , k cat , and K i CI are listed in Table I. The value for k cat for the recombinant human enzyme is 184 (131-237) h Ϫ1 , at least 30-fold higher than the values found from the MEL cells or recombinant mouse enzymes (11,23). The K m value of CI was 0.5 (0.3-0.7) M, whereas that for the AdoMet was 7.2 (5.3-9.1) M. The K i for DNA is estimated between 0.2 and 1.2 M CI. These constants can be used as the basis for a comparison of different enzyme preparations.

De Novo and Maintenance Methylation Catalyzed by the Recombinant Human DNA (Cytosine-5) Methyltransferase-Find-
ing a 30-fold difference in the k cat value of poly(dI-dC)⅐poly(dI-dC) by the recombinant human and mouse DNMT1 prompted us to address the question of the de novo and maintenance methylation properties of the enzyme. Two sets of oligonucleotides were evaluated kinetically. One set of oligonucleotides corresponds to the imprinted locus SNRPN (small nuclear riboprotein-associated peptide N) exon-1, and the other set was from a short stretch of the FMR-1 locus (fragile X mental retardation syndrome). Both DNA sequences mimic human genes and are methylated during development or disease progression. Three sets of duplex DNAs were made with either both strands unmethylated or one strand methylated (hemimethylated). Two hemimethylated duplexes contained either the coding (upper) or the complementary strand (lower) in methylated form. The FMR-1 locus contained either 24 (un-  Table I. FIG. 5. Intercept and slope replots of the methylation rate in poly(dI-dC)⅐poly(dI-dC). Replots of the slopes and y axis intercepts are from Fig. 4. A and B represent the slopes versus 1/AdoMet and 1/V max(app) versus 1/AdoMet from the fixed AdoMet data in Fig. 4A. C represents 1/V max(app) versus 1/CI from the data in Fig. 4B. Error bars are indicated. The kinetic constants V max , K m AdoMet , and K m CI were calculated from these plots. methylated duplex) or 12 (hemimethylated duplex) cytosine residues available for methylation. Similarly, the SNRPN exon-1 locus contained either 16 or 8 target cytosines (assuming that the CG dinucleotide is the only target site for methylation).
A series of double-reciprocal plots were made with six different AdoMet or six different DNA concentrations following the same principles described for poly(dI-dC)⅐poly(dI-dC). For the unmethylated duplex, the CG concentration ranged between 0.5 and 5.0 M, and for the hemimethylated substrate it was between 0.1 and 1.0 M. The AdoMet concentration was varied from 4.0 to 15.1 M for unmethylated DNA and 2.0 to 12.1 M for hemimethylated DNA. At each given DNA and AdoMet concentration, less than 5% of the cytosines were converted to 5 methylcytosine, and the AdoMet concentration was in vast excess over that of the end product, AdoHcy. Double-reciprocal plots were obtained using the SNRPN exon-1 locus either as an unmethylated or hemimethylated substrate. With increasing concentrations of the co-substrate, the slope of the plots decreased, as expected. Representative double-reciprocal plots of SNRPN exon-1 at fixed AdoMet and at fixed DNA are shown in Fig. 6, A and B. Replots of 1/V max(app) versus 1/AdoMet (from Fig. 6B), slopes of 1/v versus 1/AdoMet (from Fig. 6B), 1/V max -(app) versus 1/CG (from Fig. 6A), and slopes of 1/v versus 1/CG (from Fig. 6A) were all linear (data not shown). k cat for the unmethylated imprinted SNRPN exon-1 locus was about 2.3 h Ϫ1 . However, the k cat value for hemimethylated SNRPN exon-1 was 23 and 49 h Ϫ1 for the lower and upper methylated strand duplexes, respectively (Table II). The values obtained for K m CG were similar for both hemimethylated duplexes. Similar experiments were carried out with the FMR-1 locus substrate. In this case, the double-reciprocal plots and replots were linear for (CGG⅐CCG) 12 and (m 5 CGG⅐CCG) 12 but were curved for (CGG⅐Cm 5 CG) 12 . Together with the data from poly(dI-dC)⅐poly(dI-dC), these curved velocity patterns reveal that the sequence composition and methylation status of the DNA template may dramatically alter the kinetics of the reaction. The kinetic constants measured for the FMR-1 locus are shown in Table III. Unmethylated trinucleotide repeats had a very low k cat (1.2 h Ϫ1 ). The k cat of hemimethylated triplet repeat DNA increased about 10 -20-fold, depending on the methylated strand (Table IV), as observed for the SNRPN exon-1 locus. Furthermore, since k cat measures the rate of dissociation of the products (methylated DNA and AdoHcy) from the enzyme, the increases in k cat for the hemimethylated DNA substrates are consistent with human DNMT1 having a greater affinity for hemimethylated DNA (41) than for fully methylated DNA. Finally, the finding that the k cat values varied 2-3-fold between the upper and the lower hemimethylated strands suggests that DNA product release is additionally influenced by the sequence composition flanking the CG substrate site.
Some of the hemimethylated templates, such as SNRPN exon-1 and FMR-1 substrates (methylated lower strand , Tables  II and III), showed a lower K m AdoMet than their unmethylated counterpart. According to the reaction scheme used to derive the velocity equations for the methyl transfer reaction by DNMT1 (32), K m AdoMet is defined as the ratio between k cat and the forward rate constant for the binding of AdoMet to the catalytic center. For all the DNA templates tested, k cat was greater for the hemimethylated substrates than for the unmethylated ones, which increases K m AdoMet in the former molecules. However, since K m AdoMet values were lower for hemimethylated duplexes, we believe that such low K m AdoMet values are due to substantially higher rates of AdoMet binding. In the accompanying paper (32), we provide evidence that hemimethylated DNA activates the reaction rates by binding to an allosteric site in DNMT1. We propose that the consequence of such allosteric binding is to increase the accessibility of AdoMet to the catalytic center.
The second-order rate constant defined by the ratio of k cat / K m CG is also a good measure of the catalytic efficiency of an enzyme. For human DNMT1 the catalytic efficiency for hemimethylated DNA was 3-10-fold higher than the corresponding unmethylated DNA (Table IV). Thus we conclude that human  DNMT1 is catalytically more efficient on hemimethylated than on unmethylated DNA. Non-CG Methylation in Vitro by the Recombinant Human DNA (Cytosine-5) Methyltransferase-The above data clearly demonstrate that the full-length human DNMT1 prefers hemimethylated DNA as a substrate. However, there is methylation in mammals at non-CG sequences (42,43). To test whether human DNMT1 can methylate non-CG sequences, a series of FdC-containing oligonucleotide duplexes were made with non-CG sites such as FXG, where X was either C, A, or T and F was 5-fluoro-2Ј-deoxycytidine. The complementary strand of the FdC strand was methylated to create a hemimethylated duplex (see "Experimental Procedures"). A stable complex is formed between FdC-containing oligonucleotide and DNMT1 only after transfer of the methyl group from AdoMet to the target cytosine. The complex results from the inability of the fluorine atom on C-5 of cytosine to serve as a leaving group and permit release of the enzyme (9). Following SDS-polyacrylamide gel electrophoresis of the reaction mixture, the complex can be visualized by autoradiography. This technique has been used successfully to cross-link Dcm DNA m 5 C MTase from E. coli (44). From the autoradiography it is clear that human DNMT1 forms a complex with non-CG sites (Fig. 7). The efficiency of complex formation was in the order FG Ͼ FCG Ͼ FWG, where the 5Ј F is the methyl acceptor in each case. Hence we conclude from these results that human DNMT1 also methylates non-CG sequences. DISCUSSION Previously, several purification attempts were described for mammalian m 5 C MTases using human placental tissue (37), murine erythroleukemia cells (13), and murine ascites cells (45). However, because of the small amount of the protein and the many purification steps required to approach homogeneity, it has been difficult to purify these proteins. The problem is compounded by the presence of a protease-susceptible domain in the first 300 amino acids of the N-terminal domain and the large molecular mass of the m 5 C MTases, about 183.5 kDa (13,23). This gene is selectively expressed in a few tissue types such as placenta, lung, brain, and heart (14). Recently, recombinant expression systems using COS cells (28) and baculovirus (23) have been reported for the murine DNMT1. In COS cells the expression was very low, but the baculovirus system led to a much higher level of protein expression. However, even in the baculovirus system, purification involved several chromatographic steps leading to a reduced yield of full-length protein.
The addition of a hexa-histidine tag to the N terminus of the DNMT1 and nickel column affinity purification was attempted (23). However, in this purification method several other proteins were present after the affinity elution, and further chromatographic purification was required. Thus we switched to an intein-based purification system (29). 2 The full-length DNMT1 gene was placed in a vector such that it was fused in frame to an intein and a chitin-binding domain. The fusion precursor is immobilized on a chitin matrix where the intein can be activated to catalyze its own cleavage and release the full-length human DNMT1 protein. Although the intein-chitin domain remains bound to the affinity beads, the target protein is eluted. We have been successful in obtaining 1 mg of protein/1.5 ϫ 10 9 cells (a liter of infected SF9 cells). This purification protocol yields intact protein at over 95% purity. Often a secondary band close to 55 kDa was observed which was later established to be the viral (AcNPV)-encoded chitinase (data not shown). To facilitate cleavage from the purification tag, the two C-terminal amino acids were changed from lysine and aspartic acid to proline and glycine. This had no detectable effect on activity.
Poly(dI-dC)⅐poly(dI-dC) as a substrate has been shown to have either comparable or greater methyl group acceptance ability than other DNAs (38). In this work, we demonstrate a complex behavior of the human DNMT1 with the poly(dI-dC)⅐poly(dI-dC) substrate. At higher concentrations of CI, substrate inactivation was apparent. It was suggested that mouse DNMT1 can form reversible, multimeric complexes at a higher protein concentration (46), and this may be the cause of the inactivation. Kinetic observations by Hitt et al. (47) suggest an irreversible binding of murine DNMT1 to DNA and aggregation at high molar excess of the enzyme. However, our kinetic measurements were carried out at 2 nM human DNMT1. In this concentration range, protein aggregation was not detected in a  gel shift assay (data not shown). Thus aggregation seems unlikely. Several other DNA substrates that we tested using the above conditions also do not support the aggregation hypothesis. However, an alternative explanation of enzyme inactivation at high poly(dI-dC)⅐poly(dI-dC) concentrations via the formation of a ternary complex consisting of DNA-enzyme-DNA may explain our observations. This hypothesis is supported by the presence of a DNA-binding peptide motif, B1, at the N terminus of the human DNMT1 protein (48). Motif B1, as well as a deleted portion of it ϪDB1, interacts and binds strongly to DNA. DB1 can bind oligonucleotide duplexes as small as 30 bp through the lysine residues (48). Thus, it is possible that poly(dI-dC)⅐poly(dI-dC) may have formed catalytically inactive complexes with human DNMT1 at high CI concentrations. It should be noted that proteolysis of the N terminus of murine DNMT1 leads to its catalytic activation (21), suggesting that this domain may be involved in repression of methyl transfer.
One potentially significant observation with recombinant human DNMT1 is the high turnover number compared with previous reports for poly(dI-dC)⅐poly(dI-dC) ( Table I). This difference could be due to the inherent nature of the human DNMT1 or could reflect a greater percentage of active protein because of our purification protocol. This could also be caused by the substrate, which has a limited search space for methylation. Our results are consistent with the k cat values of other AdoMet-dependent methyltransferases such as the bacterial m 5 C MTases M.HhaI, 78 h Ϫ1 on poly(dG-dC)⅐poly(dG-dC) (9), and tRNA (m5U54) methyltransferase, 108 h Ϫ1 (49).
The mammalian enzymes have been classified as maintenance m 5 C MTases. The mouse enzyme is found to be associated with replication foci where it likely methylates the newly synthesized DNA during the S-phase of the cell cycle (8). We analyzed two synthetic DNAs representing the whole exon-1 sequence of SNRPN and 12 CGG trinucleotide repeats from the FMR-1 gene. These are representative of human DNA sequences that undergo methylation during development (imprinting) or are implicated in disease (fragile X syndrome). Both sequences have multiple CGs, thus enabling the study of both de novo and maintenance methylation processes in well defined, biologically relevant sequences.
The k cat values for normal DNA substrates shown in Tables II and III are smaller than the turnover number for poly(dI-dC)⅐poly(dI-dC) shown in Table I but are comparable to other m 5 C MTases (11,49). However, the strand bias for methylation is evident for human DNMT1, once the k cat values for complementary hemimethylated duplexes are compared (Table IV). This could be the result of an unusual topological structure of the DNA, or the sequence context of 5mC may act as a signal for enhanced methylation in adjacent DNA regions as observed previously (24). Kinetic studies with more substrates will be necessary to assess whether this is due to a specific sequence effect. The human enzyme has a typical turnover number between 7 and 49 h Ϫ1 for hemimethylated DNA as compared with 1.2-2.3 h Ϫ1 for unmethylated DNA. Our data show that the turnover number for human m 5 C MTase and unmethylated DNA is at least 110-fold greater than previously reported (12). The preference for methylation between hemimethylated and unmethylated DNA is about 15-fold, as measured under optimal kinetic conditions (Table IV), whereas a 134-fold difference was observed by Kho et al. (12) for the human enzyme using oligonucleotide duplexes. A major reason for this discrepancy may be the quality of the enzyme since a much smaller human m 5 C MTase was used for kinetic analysis by Kho et al. (12) based on the FdC data presented. Our data show clearly (Table  IV) that the enzyme prefers hemimethylated DNA as a substrate and is capable of both de novo and maintenance methy-lation. However, within the nucleus, DNMT1 binds to the replication fork (8) along with many other factors, such as p21 and proliferating cell nuclear antigen (22). Thus, the enzyme is in a very different environment during DNA replication, and the rate of methylation in vivo may differ from that observed in vitro.
Apart from methylating the canonical CGs, the human DNMT1 can also methylate unusual structures such as slipped duplexes, snapbacks, and cruciforms (50). Clark et al. (42) have shown that mammalian cells were capable of maintaining methylation of cytosine at CNG sites within the transfected plasmids. There are also reports of non-CG methylation in the hypermethylated CG promoter region of the human L1 retrotransposon (43). Methylation at such sites and its maintenance were postulated to be mediated by a different enzyme(s) (25). We tested the ability of the recombinant human DNMT1 to carry out non-CG methylation by substituting the target cytosine with a fluorocytosine and allowing the reaction to proceed in the presence of AdoMet (44). Oligonucleotide duplexes containing FCG or FWG sequences formed covalent complexes with the human enzyme, indicative of a successful methyl transfer event. Thus, human DNMT1 can carry out de novo and maintenance methylation at the CG sites and can also maintain methylation of some non-CG sites.