The human Dnmt2 has residual DNA-(cytosine-C5) methyltransferase activity.

The human Dnmt2 protein is one member of a protein family conserved from Schizosaccharomyces pombe and Drosophila melanogaster to Mus musculus and Homo sapiens. It contains all of the amino acid motifs characteristic for DNA-(Cytosine-C5) methyltransferases, and its structure is very similar to prokaryotic DNA methyltransferases. Nevertheless, so far all attempts to detect catalytic activity of this protein have failed. We show here by two independent assay systems that the purified Dnmt2 protein has weak DNA methyltransferase activity. Methylation was observed at CG sites in a loose ttnCGga(g/a) consensus sequence, suggesting that Dnmt2 has a more specialized role than other mammalian DNA methyltransferases.

In mammals, DNA methylation is the only known physiological modification of DNA. It primarily occurs at CG sites, which are methylated to 70 -80% and encode epigenetic information on the DNA. DNA methylation is implicated in the regulation of gene expression, control of development, X chromosome inactivation, parental imprinting, and the protection of the genome against parasitic genetic elements such as transposons, retrotransposons, and viruses (for review see Refs. 1, 2-4). Aberrant methylation is among the most important causes of the inactivation of tumor suppressor genes in cancer (5). DNA-(Cytosine-C5) methyltransferases (MTases) 1 are characterized by a set of highly conserved amino acid motifs (6) that easily allows the identification of putative enzymes based on genomic sequences. In mammals, four candidate enzymes have been identified (1,7). Dnmt1 is known to have a high preference for hemimethylated CG sites and has an important role in maintenance of methylation (8,9). Dnmt3a and Dnmt3b do not recognize preexisting patterns of methylation and are de novo MTases (10 -12). However, it is also clear that in vivo the functions of maintenance and de novo methylation overlap, because accurate maintenance of the methylation at certain sequences also requires the presence of Dnmt3a and 3b and de novo methylation also relies on support by Dnmt1 (13)(14)(15)(16). Dnmt1, Dnmt3A, and Dnmt3B are large proteins comprising an N-terminal part implicated in protein targeting and regulation and a C-terminal part that contains all 10 motifs characteristic for DNA-(Cytosine-C5) MTases. All three MTases are essential in mammals. Knock-out mice die during embryogenesis or shortly after birth (8,11). The fourth putative DNA MTase in the genome of mammals, Dnmt2, has been discovered in 1998 (17,18). Expression of Dnmt2 has been found in many human and mouse tissues albeit at low levels (10,17,18). Lacking a large N-terminal part, the human protein comprises 391 amino acid residues, which show clear homology to prokaryotic DNA-(Cytosine-C5) MTases. Members of the Dnmt2 family are found in many species including man and mice, which also contain other DNA MTases, but also in D. melanogaster and S. pombe where these enzymes are the only obvious DNA MTase candidate genes. The strict conservation of all of the motifs characteristic for prokaryotic Cytosine-C5 MTases in Dnmt2 including the catalytically important amino acid residues suggests a function as DNA MTase. This hypothesis was supported by the finding that D. melanogaster DNA contains methylated cytosines that most probably is attributed to the activity of Dnmt2 (19,20). The structure of Dnmt2 was solved in complex with AdoMet, the cofactor for DNA methylation (21). It shows strong similarities to other DNA MTases, supporting the assumption that the protein is a DNA MTase. Interaction with DNA could be detected in gel shift experiments. In addition, it has been reported that Dnmt2 covalently binds to DNA (21). This observation also connects the protein to DNA MTases, which form a covalent enzyme-DNA intermediate during catalysis (for review see Refs. 1,22,23). Nevertheless, despite considerable efforts no catalytic activity of Dnmt2 could be detected so far (10,17,18,21). In addition, Dnmt2deficient embryonic stem cells are viable and do not show any obvious difference in the DNA methylation pattern (10) and Dnmt2 knock-out animals are viable and fertile with minor defects. 2

EXPERIMENTAL PROCEDURES
Cloning, Site-directed Mutagenesis, and Protein Purification-Dnmt2 cDNA was obtained from the Deutsches Ressourcenzentrum fü r Genomforschung GmbH (Berlin, Germany) (clone number IMAGp998C184250Q2) amplified by PCR and cloned into pET28aϩ (Novagen). A Dnmt2 variant in which the highly conserved active site residues Cys-79 and Glu-119 were exchanged by Ala was prepared by site-directed mutagenesis following standard procedures (24). The sequence of all of the expression constructs was verified by DNA sequencing. Dnmt2 and the Dnmt2 mutant were expressed in BL21 Rosetta(DE3) pLysS cells (Novagen) and purified to homogeneity by nickel-nitrilotriacetic acid affinity chromatography following a procedure described for Dnmt3a with the exception that no protease inhibi-* This work has been supported in part by the Bundesministerium fü r Forschung and Bildung BioFuture program, the Deutsche Forschungsgemeinschaft (Grant JE252/1), and the Fonds der Chemischen Industrie. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. tors were added (12). A MALDI-TOF mass spectrometry analysis revealed masses of the M ϩ H ϩ ions of 46631.0 and 46537.1 Da for Dnmt2 (Fig. 1B) and the Dnmt2 mutant (data not shown), respectively. These values are within experimental accuracy identical to the corresponding theoretical masses (46629.7 and 46538.6 Da) calculated after removal of the initial Met residue. A mock Dnmt2 preparation was carried out using BL21 Rosetta(DE3) pLysS cells containing the pET28a ϩ vector without the Dnmt2 insert.
Mass Spectrometry-Molecular masses were determined by MALDI-TOF mass spectrometry using a Vision 2000 mass spectrometer (Finnegan MAT, Bremen, Germany). The protein samples were desalted using Centricon-30 concentrators (Amicon, Beverly, MA). 1 l of protein solution was mixed with 1 l of matrix solution (9 mg of 2,5-dihydroxybenzoic acid and 1 mg of 2-hydroxy-5-methoxybenzoic acid/ml, 0.1% (v/v) TFA, 30% (v/v) acetonitrile) and allowed to air-dry. Ions were generated by irradiation with a pulsed nitrogen laser (emission wavelength: 337 nm; laser power density: ϳ10 6 watts cm Ϫ2 ), and positive ions were accelerated and detected in the reflector mode. Spectra were calibrated using bovine serum albumin (Sigma) as external standard.
DNA and Oligodeoxynucleotides-Purified oligodeoxynucleotides were purchased from MWG-Biotech (Ebersberg, Germany). DNA methylation experiments were carried out with a 558-mer PCR fragment derived from pAT153 that contains 34 CG sites. The fragment was produced by PCR and purified over PCR spin columns (Qiagen), and its concentration was determined from A 260 nm .
DNA Methylation Reactions-For the DNA methylation reactions, 1-2 g of DNA was incubated with 10 M Dnmt2 in 40 l of methylation buffer (20 mM HEPES, pH 7.5, 1 mM EDTA) containing 1 mM AdoMet (Sigma) for 16 h at ambient temperature. After the methylation reactions, DNA was purified by phenol/chloroform extraction or by passing over PCR spin prep columns (Qiagen). As a negative control, the DNA was incubated at the same condition with the same dilution of a Dnmt2 mock preparation. As positive control, the DNA was modified using 16 units of M.SssI (New England Biolabs) for 1 h at 37°C.
HPLC Separation of Nucleosides-HPLC separation of the nucleosides was carried out basically as described previously (19,25,26). The DNA (1-2 g) was degraded to nucleosides using 2 g of P1 nuclease (Sigma), 1.5 g of Serratia marcescens nuclease (ϳ1500 units), and 2 units of shrimp alkaline phosphatase (Amersham Biosciences). The S. marcescens nuclease was kindly provided by G. Meiss (Institut fü r Biochemie, Giessen, Germany). It is also commercially available as Benzonase from Merck. Subsequently, 0.1-1 g of the DNA was subjected to reverse-phase HPLC column Apex 1-octadecylsilyl(C-18), 5-m-particle size (Jones Chromatography, Llanbradach, Wales), equilibrated with buffer A (100 mM triethylammonium acetate, pH 6.5). HPLC runs were performed at a flow rate of 0.5 ml/min at ambient temperature using a biphasic linear gradient consisting of buffers A and B (100 mM triethylammonium acetate, pH 6.5, 30% acetonitrile): t ϭ 0 min, 0% buffer B; t ϭ 20 min, 15% buffer B; t ϭ 40 min, 100% buffer B; and t ϭ 50 min, 100% buffer B. As reference, a synthetic oligonucleotide containing 5-methylcytosine was used. Either peaks corresponding to cytidine, methylcytidine, thymine, adenosine, or guanosine were collected or fractions were collected for 30-s intervals during the entire run. The fractions were dried in vacuum, dissolved in water and reevaporated twice, and finally dissolved in 15 l of water.
Two-dimensional TLC Analysis-Labeling of the deoxynucleosides in the HPLC fractions and two-dimensional TLC analysis were carried out essentially as previously described (19,26). 2 l of the HPLC samples were incubated with 30 nM deoxynucleoside kinase (kindly provided by H. Gowher, Institut fü r Biochemie, Giessen, Germany) in 15-l buffer (50 mM Tris-HCl, pH 8.5, 5 mM MgCl 2 ) containing 0.25 l of [␥-32 P]ATP (370 MBq/ml, PerkinElmer Life Sciences) for 2 h at 37°C. 2 l of the radioactive sample were applied to cellulose TLC plates (Cellulose CEL 400-10, 20 ϫ 20 cm, Macherey-Nagel, Dü ren, Germany) at ambient temperature. The first dimension was run in isobutyric acid/water/ ammonia (66:33:1, v/v/v). The plate was dried in air and developed in the second dimension with isopropyl alcohol/concentrated HCl/water (70:15:15, v/v/v). The plate was dried, and the radioactivity was analyzed using an instant imager (Canberra Packard).
Bisulfite Sequencing Analysis-Bisulfite sequencing was carried out as described previously (27,28). Purified methylated or unmethylated DNA (20 ng) was incubated NaOH (0.3 M) for 15 min at 37°C, sodium metabisulfite and hydroquinone were added to final concentrations of 2.4 M and 0.5 mM, and the mixture was incubated for 16 h at 55°C. The DNA was purified over PCR spin columns and incubated with NaOH (0.3 M) for 15 min at 37°C. The solution was neutralized by the addition of ammonium acetate, pH 7, and precipitated with ethanol. The lower strand of the converted DNA was amplified by PCR using TaqDNA polymerase and cloned using TOPO TA cloning kit for sequencing (Invitrogen). Finally, 50 clones of the DNA incubated with Dnmt2 and 50 clones of the control DNA were sequenced.

RESULTS
We have cloned human Dnmt2 into a bacterial expression vector and expressed and purified the protein to homogeneity (Fig. 1A). A MALDI-TOF mass spectrometry analysis confirmed the mass of the protein within the experimental error (experimental mass of M ϩ H ϩ : 46631.0 Da; theoretical mass: 46629.7 Da) (Fig. 1B). No significant peaks were detected at higher masses. This result shows that the purified protein does not carry covalently bound nucleotides or DNA. The protein was incubated with different radioactively labeled DNA fragments generated by PCR in the absence of cofactor as well as in the presence of AdoMet or S-adenosyl-L-homocysteine and analyzed on denaturing SDS-polyacrylamide gels. However, no covalent complex with DNA could be detected (data not shown). This result suggests that the covalent complex formation reported previously (21) might occur only with special DNA sequences.
HPLC/TLC Activity Assay-The catalytic activity of Dnmt2 was investigated using -DNA or a PCR fragment derived from pAT153 plasmid and radioactive [methyl-3 H]AdoMet. In addition, restriction protection studies were performed using several restriction endonucleases sensitive to methylation of cytosine residues. However, in agreement with previous studies (10,17,18,21), no catalytic activity could be detected. Because we have used purified Dnmt2 enzyme at high concentrations in these assays, this result indicates that the specific activity of Dnmt2 is severalfold lower than that of Dnmt1 (9) or Dnmt3a or Dnmt3b (10,12).
However, there are several possible explanations for this negative result, because the sensitivity of tritium transfer assays is limited and many DNA MTases carry tightly bound cofactor after purification. Bound cofactor interferes with the tritium transfer assay, in particular, if high amounts of the purified enzyme are used, which is required if low activities are to be detected. In addition, the conclusions that can be drawn from restriction protection experiments are restricted because methylation can only be detected if the recognition sequence of the restriction enzyme overlaps with that of the MTase. Therefore, we have also investigated the activity of Dnmt2 using an HPLC/TLC assay that is very sensitive and general (19,26). DNA was methylated in vitro and hydrolyzed to nucleosides, and the nucleosides were separated by HPLC. The nucleosides then were labeled with deoxynucleoside kinase (dNK) using [␥-32 P]ATP, and the nature of each nucleotide was confirmed by two-dimensional TLC.
As shown with methylated reference DNA, 5-methylcytidine elutes from the HPLC column after 12.9 min (Fig. 2). The elution profile of the nucleosides from the HPLC column was monitored by UV spectroscopy at 260 nm. No 5-methylcytidine peak was detectable with DNA after incubation with Dnmt2 and AdoMet, indicating that the level of cytosine methylation was below 1%. Therefore, the material eluting from the column around the elution time of the 5-methylcytidine was collected (11.5Ј-14Ј), and the nucleosides were labeled using dNK and subjected to two-dimensional TLC (Fig. 3). The position of the m C on the two-dimensional TLC plate is shown in Fig. 3A using M.SssI-modified DNA. However, also in the DNA incubated with Dnmt2, a clear m C spot was detected (Fig. 3B). The additional spots present in the TLCs shown in Fig. 3 correspond to C and T, which appear in the tails of the much larger HPLC peaks of the major components of the DNA. A guanine spot is not detected, because the dNK has a preference for pyrimidines (29). No m C was detected in DNA not incubated with a DNA MTase (data not shown).
To rule out that the methylation might be due to a contamination of the Dnmt2 preparation, Dnmt2 was expressed in Escherichia coli cells that do not contain a DNA-(Cytosine-C5) MTase. In addition, a control purification was carried out using the same cells containing the same vector without Dnmt2 insert. Cells were grown and harvested exactly as the expression cells, and the cell extract was applied to affinity chromatography under the same conditions as the Dnmt2 crude cell extract. The target DNA was incubated with the same volumes of the mock purification as used in the Dnmt2 methylation experiments and subjected to the same assay procedure as the DNA incubated with Dnmt2. However, no m C was detectable (Fig.  3C), demonstrating that the methylation activity is attributed to the Dnmt2 protein.
To confirm the nature of the 5-methylcytosine in an additional experiment, we collected fractions every 30 s to follow the elution profile of the compound from the HPLC column. As shown in Fig. 4, the m C peak was observed between 12 and 13 min exactly as expected from the absorbance profile of the HPLC run with the reference DNA. To estimate the amount of 5-methylcytosine in the original DNA, we have added defined amounts of cytosine to the HPLC samples and compared the peak intensities of cytidine and 5-methylcytidine. On the basis of these titrations, the amount of 5-methylcytosine in the original DNA could be estimated, yielding a ratio of 5-methylcy-tosine/cytosine of roughly 1:150 (0.7%) as compared with 1:16 -1:12 (6 -8%) as observed in human DNA from natural sources.
As additional control, a Dnmt2 variant was prepared in which the highly conserved active site residues Cys-79 and Glu-119 were exchanged by Ala. The variant was purified, and its identity was confirmed by MALDI-TOF mass spectrometry (experimental mass of M ϩ H ϩ : 46537.1 Da; theoretical mass: 46538.6 Da). After incubation of the DNA with the active site variant, a much lower amount of 5-methylcytosine was detected (Fig. 5). A quantitative analysis revealed that the amount of 5-methylcytosine obtained after incubation with the Dnmt2 variant was Ͼ10-fold lower than the amount obtained with Dnmt2. This result demonstrates that the active site residues of Dnmt2 are important for catalysis, confirming the conclusion that Dnmt2 is an active DNA MTase. Because in the Dnmt2 variant the two most important catalytic residues are removed, one might have expected a larger reduction in activity. However, we recently obtained similar results with the Dnmt3a MTase where also residual activity was detectable even after the removal of these residues (30). This result indicates that in addition to the established catalytic mechanism DNA MTases also have other means to accelerate the methylation reaction such as correct positioning of the flipped base and the cofactor and binding to the transition state of methyl group transfer.
Bisulfite Sequencing Activity Assay-To investigate the catalytic activity of Dnmt2 with an independent assay, DNA was incubated with Dnmt2 in vitro and then subjected to bisulfite treatment (31,32). This procedure converts all of the cytosine residues to uracil, giving rise to thymine after amplification by PCR. Only methylated cytosines are refractory to the deamination. Bisulfite-treated DNA was cloned, and 44 clones were sequenced (ϳ350 nucleotides/clone) (Supplemental Fig. 1). As a control, 41 clones of DNA incubated with the Dnmt2 mock preparation were sequenced. In the Dnmt2-treated DNA, 24 cytosine residues were detected in comparison to 11 in the control DNA (Table 1). However, because of polymerase errors in the PCR of the bisulfite-treated DNA, not all of these cytosines correspond to methylated cytosines in the original DNA. Eight of the cytosines detected in the control DNA and nine in the Dnmt2-treated DNA were observed at positions corresponding to a thymine in the original DNA. Thus, these cytosines must have been introduced by polymerase errors, which cannot be avoided, because TaqDNA polymerase must be used to amplify uracil-containing DNA (28). In addition, three cytosines in the control DNA occur at positions corresponding to cytosine in the original DNA. Because the control was not methylated, they can be the result of incomplete deamination during bisulfite treatment or polymerase errors after successful conversion of the original cytosine to uracil. However, the distribution of cytosines observed at the position of T or C in the original DNA is highly biased, whereas in the control DNA only three cytosines were observed at positions corresponding to a cytosine in the original DNA. In the Dnmt2treated DNA, 15 cytosines were found at such sites. This distribution of cytosines found at the place of cytosine versus cytosine found at the place of thymine (15:9 for Dnmt2-treated DNA but 3:8 for the control DNA) cannot be due to statistical fluctuations (p ϭ 1.1 ϫ 10 Ϫ4 ) and is a strong indication that Dnmt2 is an active DNA MTase. Since we observed 15 5-methylcytosine residues in the 44 clones sequenced, the ratio of 5-methylcytosine/cytosine can be estimated to be 1:250 on the basis of the bisulfite sequencing assay, close to what has been concluded from the results of the HPLC/TLC assay.
We have analyzed the sequences flanking the 15 positions where methylation has been observed in the bisulfite analysis. Because in the control DNA, three cytosines were observed that resulted from polymerase errors or incomplete conversion of the DNA, up to three or four cytosines of the Dnmt2-treated DNA also could be the result of this effect. Therefore, one cannot expect a fully defined recognition sequence. The most significant deviation of the distribution of the occurrence of A, T, G, and C from a random distribution was seen at position ϩ1 with respect to the cytosine where a G is highly favored (p ϭ 0.019). This result shows that Dnmt2, similar to all of the other mammalian DNA methyltransferases identified so far, prefers methylation of CG sites. At all of the other positions, smaller flanking sequence preferences are observed. If one considers an occurrence of 6 in 15 as significant (2-fold overrepresentation), the following preferences can be derived. At positions Ϫ3 and Ϫ2, a preference for T is observed. At position Ϫ1, no significant preference for any base occurs. At position ϩ2, G is over-represented, and at position ϩ3, A is most common base. At position ϩ4, G and A are enriched. Therefore, a consensus sequence of ttnCGga(g/a) can be derived for Dnmt2 from our data (Table II). It is interesting to note that Dnmt2 might recognize its CG target site in a palindromic TTCCGGAA sequence context. However, the small number of methylated sequences analyzed so far precludes a final statement on the specificity of Dnmt2, and the consensus sequence might be modified and sharpened as a result of further studies. DISCUSSION Genome projects have provided a wealth of sequencing data for many species including man and mice. For example, in human DNA, ϳ30,000 -40,000 genes have been found (33,34). However, in many cases, the functions of the genes identified are unknown. For example, approximately one-third of the predicted proteins in the S. cerevisiae genome does not show clear homology to any protein of known function. For another one-third of all genes, hypotheses for the functions of proteins can be derived on the basis of amino acid sequence homology (35,36). One example of a human protein belonging to the second group is Dnmt2, which shows clear cut amino acid sequence similarity to DNA-(Cytosine-C5) MTases and also has a structure closely resembling the common structure of enzymes from this class. Nevertheless, catalytic activity of Dnmt2 could not been detected by several groups. Therefore, the function of Dnmt2 as DNA MTases has been questioned. Using a combined HPLC/TLC assay that is very sensitive and specific, we show here that Dnmt2 is an active DNA MTase. Its low in vitro activity explains the failure to detect activity in previous studies and may in part explain the weak phenotype observed after deletion of Dnmt2 in mice. After sequencing many clones of bisulfite-treated DNA modified by Dnmt2 and statistical analysis of the results, we could define a loose consensus sequence for Dnmt2 (ttnCGga(g/a)) showing that Dnmt2 also modifies DNA at CG residues similar to all of the other mammalian DNA MTases identified so far. Therefore, the other MTases might compensate for the function of Dnmt2, which also could explain the weak knock-out phenotype. The consensus sequence rules out that the enzyme recognizes CCWGG sites as suggested by the fact that this sequence specificity has been reported for a mutant of the S. pombe Dnmt2 homolog pmt1 (37) and by the observation that CCWGG methylation has been detected in mammals (cf. Ref. 38 and references therein). However, CCWGG is the recognition sequence for the E. coli dcm MTase, which might have caused the CCWGG methylation activity observed in the recombinant protein preparation of the pmt1 variant. The CCWGG methylation identified in human DNA could be attributed to other enzymes, like Dnmt3a or Dnmt3b, which also methylate DNA at non-CG sites (12,39,40).
The consensus sequence derived from the in vitro methyla- FIG. 5. Analysis of the catalytic activity of a Dnmt2 active site variant. 0.2 g of DNA was incubated with Dnmt2 degraded to nucleosides and subjected to HPLC analysis. Fractions were collected between 12Ј and 13Ј, labeled with dNK, and analyzed by two-dimensional TLC. The activity of the active site variant was Ͼ10-fold reduced as compared with Dnmt2.

TABLE I
Results of the bisulfite sequencing methylation analysis n Seq , number of clones sequenced; n C (at C), number of cytosine residues found at positions where a C was present in the original DNA; n C (at T), number of cytosine residues found at positions where a T was present in the original DNA.

Dnmt2
Control n Seq 44 41 n C (at C) 15 3 n C (at T) 9 8 tion reactions (ttnCGga(g/a)) overlaps with the general target site of ets-transcription factors (GGA(G/A)), which have important roles in regulation of gene expression in different signaling pathways and interact with many other proteins involved in transcriptional regulation (for review see Ref. 41). Since it has already been shown that CG methylation inhibits DNA binding by different ets-transcription factors (42,43), the consensus sequence suggests that Dnmt2 could be involved in the regulation of DNA binding and transactivation by an ets-transcription factor. Therefore, Dnmt2 possibly has a more specialized role in vivo than the other known mammalian DNA MTases.
Finally, it should be noted that Dnmt2 is the only mammalian DNA MTase that does not have a large N-terminal domain, which usually is involved in the targeting and regulation of the MTases. It is interesting to note that Dnmt1, which is highly active in vitro, is not active after removal of the N-terminal part (9). Thus, the catalytic domain of Dnmt1 depends on activation by the N-terminal domain. It is possible that an activator protein for Dnmt2 is present in certain tissues where it could play the role of the N-terminal part in Dnmt1 and stimulate the activity of Dnmt2. However, it should be mentioned that the catalytic domains of Dnmt3a and Dnmt3b are active in the absence of the N-terminal parts of the proteins (44), demonstrating that not all of the mammalian MTases require an N-terminal part for activity.