On the substrate specificity of DNA methyltransferases. adenine-N6 DNA methyltransferases also modify cytosine residues at position N4.

Methylation of DNA is important in many organisms and essential in mammals. Nucleobases can be methylated at the adenine-N6, cytosine-N4, or cytosine-C5 atoms by specific DNA methyltransferases. We show here that the M.EcoRV, M.EcoRI, and Escherichia coli dam methyltransferases as well as the N- and C-terminal domains of the M. FokI enzyme, which were formerly all classified as adenine-N6 DNA methyltransferases, also methylate cytosine residues at position N4. Kinetic analyses demonstrate that the rate of methylation of cytosine residues by M.EcoRV and the M.FokI enzymes is reduced by only 1-2 orders of magnitude in relation to methylation of adenines. This result shows that although these enzymes methylate DNA in a sequence specific manner, they have a low substrate specificity with respect to the target base. This unexpected finding has implications on the mechanism of adenine-N6 DNA methyltransferases. Sequence comparisons suggest that adenine-N6 and cytosine-N4 methyltransferases have changed their reaction specificity at least twice during evolution, a model that becomes much more likely given the partial functional overlap of both enzyme types. In contrast, methylation of adenine residues by the cytosine-N4 methyltransferase M.BamHI was not detectable. On the basis of our results, we suggest that adenine-N6 and cytosine-N4 methyltransferases should be grouped into one enzyme family.

In higher eukaryotes post-replicative, enzymatic methylation of DNA is involved in genomic imprinting (1,2), X-chromosome inactivation (3), gene regulation (4), and carcinogenesis (5)(6)(7). In eukaryotes methylation is the only known covalent modification of DNA. At least in mammals it is absolutely essential (8). In prokaryotes, methylation of DNA is involved in processes like post-replicative repair (9) and control of DNA replication (10). In addition, DNA methylation is required in restriction-modification systems found in almost all bacteria (11). Today, over 200 prokaryotic DNA methyltransferases (MTases) 1 have been identified and characterized; most of which are components of restriction-modification systems (12)(13)(14). These enzymes recognize short DNA sequences and spe-cifically methylate these sequences at a defined position yielding 6-methyladenine, 4-methylcytosine, or 5-methylcytosine. All MTases contain a weakly conserved F_G_G amino acid motif responsible for the interaction with the cofactor AdoMet. Amino acid sequence comparisons show that cytosine-C 5 MTase share nine additional conserved amino acid motifs (12,15,16). In contrast, adenine-N 6 MTases are much more heterogenic. Besides the F_G_G motif, they only contain one moderately conserved (D/N)PP(Y/F) motif, which forms part of the active center of these enzymes. Some additional weakly conserved motifs could only be identified in structure-guided alignments (17). Interestingly, cytosine-N 4 MTases also contain a set of motifs comparable with adenine-N 6 MTases. However, most of them contain an SPP(Y/F) motif instead of the (D/ N)PP(Y/F) motif present in adenine-N 6 Mtases 2 (17). The conserved motifs of adenine-N 6 and cytosine-N 4 MTases may appear in variable distance and order. According to the arrangement of the motifs these enzymes can be subdivided into three groups, ␣, ␤, and ␥ (15,17). Most cytosine-N 4 MTases belong to the ␤ group, however, some are members of the ␣ group.
MTases of all classes show structural similarity in their catalytic domains comprising the overall fold and topology, the cofactor binding site and location of the catalytic center (18 -22). The structure of the cytosine-C 5 MTase M.HhaI⅐DNA complex has shown that these enzymes flip their target base out of the DNA helix during catalysis (23), which is also very likely to occur in adenine-N 6 and cytosine-N 4 MTases (20, 24 -32). All DNA MTases studied so far prefer binding to substrates that contain a mismatch base pair or even an abasic site at the target position (27,30,(32)(33)(34). This result can be rationalized because base flipping requires disruption of the Watson-Crick hydrogen bonds of the target base, which is facilitated if the target base forms weaker hydrogen bonds to its partner base. However, these results also imply that the target base may not be recognized very accurately by MTases, at least in the ground state of the enzyme⅐DNA complex. Given this finding and the structural similarity of adenine-N 6 and cytosine-N 4 MTases as well as the similarity of the active site sequence motifs of both enzyme families, the question arises whether adenine-N 6 MTase might also be able to methylate cytosine residues and vice versa. TC Methylation reactions with M.EcoRV (0.9 M) were carried out in 50 mM HEPES, pH 7.5, 50 mM NaCl, 1 mM EDTA. Methylation reactions with M.FokI(1-367) (3.7 M) and M.FokI(335-647) (9.6 M) were carried out in 50 mM Tris/HCl, pH 7.5, 5 mM dithioerythritol, 2 mM EDTA, 10 ng/l bovine serum albumin. Methylation reactions with M.EcoRI (4 units/ l), M.BamHI (4 units/l), and dam MTase (8 units/l) were carried out in the buffers recommended by the supplier. After the methylation reactions 10 l of the reaction mixture was incubated with 5 mM MgCl 2 , 3 2 g (3.6 ϫ 10 Ϫ3 units) P1 nuclease (Sigma), and 1.4 g (1400 units) Serratia marcescens nuclease for 16 h at 37°C in a total volume of 30 l. Then, 2 l (2 units) of shrimp alkaline phosphatase (Amersham Pharmacia Biotech), 4 l 10ϫ SAP buffer (200 mM Tris/HCl, pH 8.0, 100 mM MgCl 2 ), and 4 l of water were added, and the reaction mixture was incubated for 30 min at 37°C. The samples were heated to 75°C for 2 min and mixed with 120 l of HPLC buffer A (100 mM triethylammonium acetate, pH 6.5). 80 l of the mixture were applied onto a reversed-phase HPLC column (Apex 1-octadecylsilyl(C-18), 5 M particle size, 25 ϫ 0.45 cm purchased from Jones Chromatography, Llanbradach, Wales) equilibrated with buffer A. HPLC runs were performed at a flow rate of 1 ml/min at an 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 40 min, 20% buffer B; t ϭ 55 min, 100% buffer B. The column eluate was monitored at 260 nm. Fractions of 0.5 ml were collected and radioactivity analyzed by scintillation counting. The identity of the nucleosides was confirmed by injection of standard dC, dT, dA, and dG (Sigma) that were treated with nucleases under the same conditions as the oligodeoxynucleotides.
Kinetic Analyses Using the Biotin/Avidin Microtiter Plate Assay-Methylation reactions with M.EcoRV were carried out in 50 mM HEPES, pH 7.5, 50 mM NaCl, 1 mM EDTA, 50 ng/l bovine serum albumin in the presence of 0.76 M labeled [methyl-3 H]AdoMet at an ambient temperature. Oligonucleotide substrates used in these experiments carry a biotin group on one end. Enzyme and oligonucleotide concentrations were varied between 0.1 and 1 M. Separation of methylated oligonucleotides and free AdoMet was performed using avidin-coated microtiter plates. To coat microtiter plates with avidin, 100 l of avidin (10 g/ml, Sigma) in 100 mM NaHCO 3 (pH 9.6) were pipetted into each well and incubated overnight at 4°C. The wells were washed five times with 200 l of PBST (140 mM NaCl, 2.7 mM KCl, 4.3 mM Na 2 HPO 4 , 1.4 mM K 2 HPO 4 , 0.05% v/v Tween 50, pH 7.2). 10 l of 10 mM unlabeled AdoMet (Sigma) in 10 mM H 2 SO 4 were pipetted into each well, and an aliquot of the methylation reaction mixture containing 2 pmol oligonucleotide was added. PBST was added to a total volume of 50 l, and the mixture was incubated for 30 min to allow binding of the oligonucleotides to the microtiter plate. The wells were washed five times with 200 l of PBST supplemented with 500 mM NaCl to remove the unreacted AdoMet and enzyme⅐AdoMet complexes. To quantify the radioactivity incorporated into DNA, the DNA was degraded using 0.7 g (700 units) S. marcescens nuclease in 100 l of 50 mM Tris/HCl, pH 8.0, 5 mM MgCl 2 for 30 min at an ambient temperature, and the released radioactivity was analyzed by liquid scintillation counting. As shown in Fig. 4 (only experiments with M.EcoRV were carried out with the biotin/avidin assay), the assay is linear with respect to time and the signal/noise ratio is much better than in the curves determined with the thin layer chromatography assay. Further controls showed that the assay is very sensitive and accurate. 4 Methylation Reactions with the M.FokI Enzymes-Methylation reactions with the M.FokI enzymes were carried out in 50 mM Tris/HCl, pH 7.5, 5 mM dithioerythritol, 2 mM EDTA, 1 mg/ml bovine serum albumin at an ambient temperature as described (35) using a thin layer chromatography assay (31). Oligonucleotide concentrations were 0.2 M; enzyme concentrations were between 0.5 and 9 M. [methyl- 14 C]AdoMet was used at a concentration of 50 M.

RESULTS
Adenine-N 6 and cytosine-N 4 MTases have similar structures and catalytic centers. Moreover, biochemical evidence suggests that MTases do not very accurately recognize their target base (27,30,(32)(33)(34). Hence, the question arose whether adenine-N 6 MTases might be able to methylate cytosine residues at the N 4 position, which is the subject of this study. Most of our experiments were carried out with the M.EcoRV and M.FokI adenine-N 6 MTases. M.EcoRV recognizes GATATC and methylates the first adenine residue within this sequence (36). M.FokI is a tandem enzyme consisting of two fused MTases that are catalytically active independently of each other (35,37,38). Both of the M.FokI enzymes recognize the asymmetric DNA sequence GGATG/CATCC (39,40), the N-terminal domain modifies the upper GGATG-strand, the C-terminal domain has a strong preference for the lower CATCC-strand (35,37,38). In this study we have used purified preparations of both individual domains, viz. M.FokI(1-367) comprising amino acid 1-367 and M.FokI(335-647) comprising amino acids 335-647 (35).
First we tested whether M.EcoRV might flip a cytosine residue located at the target position out of the DNA helix, a prerequisite for enzymatic activity. To this end, the binding constants of M.EcoRV to the canonical hemimethylated substrate RV_ME containing a GATATC/G m ATATC site and to RV_C/A were determined. RV_C/A contains a GCTATC/ G m ATATC site in which the unmethylated target adenine of RV_ME is replaced by cytosine and the "target" cytosine is located in a C:T base mismatch. The hemimethylated substrate RV_ME is bound with an affinity of K a ϭ 5 ϫ 10 6 M Ϫ1 , similar to binding constants determined for an unmethylated 20-mer (31) and 30/33-mer (41). In contrast the RV_C/A substrate is bound with K a ϭ 6 ϫ 10 7 M Ϫ1 . This preference of M.EcoRV for binding to the mismatch substrate is very similar to results obtained with other mismatch substrates (30,32). It suggests that M.EcoRV flips cytosine residues if they are located in a GCTATC/G m ATATC sequence.
To find out if cytosine residues can be modified by ade- 3 Depending on the EDTA concentration, MgCl 2 was added to a final concentration of 5 mM free Mg 2ϩ . 4  AdoMet results in strong methylation of adenine residues but no methylation occurs at cytosine residues (Fig. 2, dashed lines). If fully methylated reference substrates are used, again no methylated cytosine is detectable (data not shown). In contrast, if the C/A substrates are incubated with the corresponding MTase no 6-methyladenine is detectable, 5 but in all reactions with C/A substrates, a significant amount of 4methylcytosine is formed (Fig. 2, continuous lines). This result is highly reproducible and the amount of 4-methylcytosine is far beyond any background fluctuations. We conclude that all three adenine MTases studied are able to methylate cytosine residues.
To find out if this behavior is a general property of adenine-N 6 MTases, we have carried out additional experiments with M.EcoRI (recognition sequence GAATTC, modified base is underlined) and E. coli dam MTase (recognition sequence GATC). In these experiments, unmethylated substrates con- 5 The small peak eluting after 45 min in the reaction with M.EcoRV represents a degradation product of AdoMet that is observed even in runs without oligonucleotide and enzyme under these conditions. taining a cytosine at the target position in one DNA strand are used, viz. GCATTC/GAATTC for M.EcoRI(RI_C/A) and GCTC/ GATC for the dam MTase (dam_C/A). Both substrates are modified at the target adenines in the lower DNA strand with good yields (data not shown). Whereas in both cases no 4-methylcytosine is observed with the control substrates (RI_L20 and dam_L19), a small but significant 4-methylcytosine peak appears with the C/A substrates (Fig. 3, A and B). This result demonstrates that M.EcoRI and the dam MTase also modify cytosine residues. However, smaller amounts of 4-methylcytosine are detected than with M.EcoRV and both M.FokI enzymes, presumably because M.EcoRV and both M.FokI enzymes were available at high concentrations and used in the assays in micromolar concentrations, whereas M.EcoRI and the dam MTase were commercial enzyme preparations containing much lower concentrations of enzyme. We conclude from these results that most, if not all, adenine-N 6 MTases also modify cytosine residues at position N 4 , a very unexpected finding.
These experiments demonstrate that cytosine residues can be methylated by adenine-N 6 MTases if they are located in a C:T mismatch base pair at the target position of the enzyme. However, we have shown that these mismatch substrates are better bound at least by M.EcoRV, and similar results were also obtained with other MTases (27, 33, 34). Hence, the ques-tion arises whether the methylation of cytosine residues by adenine-N 6 MTases depends on the fact that the cytosine is located in a base mismatch in the C/A substrate. To answer this question, an additional substrate has been used for M.EcoRV that contains a GCTATC/GATAGC site (RV_CG). Thus, RV_CG contains a GCTATC "star" site of M.EcoRV that differs in one base pair from GATATC and not only in one nucleotide as the C/A substrates. In this substrate the target cytosine is base paired to a guanine in the other DNA strand to form a G:C base pair that is even stronger than A:T base pair in the canonical recognition sequence. If this substrate is incubated with M.EcoRV 4-methylcytosine is formed although less than with the GCTATC/G m ATATC substrate (Fig. 3C). This result shows that the methylation of cytosine residues can occur in regular DNA and is not dependent on the facilitated flipping of this base when it is located in a base mismatch.
It was tempting to investigate whether the cytosine-N 4 MTase M.BamHI (recognition sequence: GGATCC) could modify adenine residues. To this end a substrate was used in which the target cytosine of M.BamHI is replaced by an adenine in one DNA strand, i.e. GGATAC/GGATCC (Bam_A/C). This substrate is methylated by M.BamHI at the target cytosine in the lower DNA strand (data not shown), but no formation of 6-methyladenine could be detected (Fig. 3D). However, given the low amounts of 4-methylcytosine observed in the reactions with commercial enzymes, the quantitative meaning of this result remains unclear so far. Moreover, it is uncertain whether this result represents a common property of all cytosine-N 4 MTases.
Taken together, all data presented so far demonstrate that adenine-N 6 MTases can also modify cytosine residues at position N 4 , a result that has not been described before. Hence, we have tried to verify this conclusion by independent experimental evidence. To this end, we have measured the rate of methylation of the C/A substrates by M.EcoRV and the two M.FokI enzymes. These experiments were carried out with both M.FokI enzymes using the thin layer chromatography assay used in our previous work (31,35,42). This, however, was not possible with M.EcoRV, because in the case of M.EcoRV a high background of radioactivity was observed possibly due to AdoMet bound to the enzyme which obscured DNA methylation at low rates. Therefore, we have developed a new methylation assay. In this assay, oligonucleotide substrates that are biotinylated in one DNA strand are methylated using labeled [methyl-3 H]AdoMet. After the methylation reaction, the substrates are bound to an avidin-coated microtiter plate. After extensive washing, the DNA is degraded by nuclease treatment. Thereby, the radioactivity incorporated into the DNA is released into solution and can be analyzed quantitatively by liquid scintillation counting. This assay proved to be very sensitive, reproducible, and accurate.
The rates of methylation of the C/A substrates (RV_C/A, Fok_up_C/A, and Fok_low_C/A) by M.EcoRV and both M.FokI enzymes were compared with the rates of methylation of hemimethylated substrates (RV_ME, Fok_up_ME, and Fok_low_ME) and, more importantly, with that of fully methylated substrates in which both target adenine residues are replaced by 6-methyladenine (RV_MM and Fok_MM). For each MTase, the C/A substrate and the fully methylated reference substrate differ from each other at only one nucleotide position, whereas in the C/A substrate, a cytosine is present at the target position of the MTase in one DNA strand, a 6-methyladenine is present at this position in the fully methylated substrate. If, therefore, the C/A substrate is modified faster than the fully methylated substrate, this methylation almost certainly takes place at the cytosine residue not present in the fully methyl-  Fig. 4 and Table I. 6 Similar results were obtained with M.EcoRV at all substrate concentrations tested (0.1-1.0 M, data not shown). In agreement with the results obtained in the HPLC analyses, the kinetic data demonstrate that M.EcoRV, M.FokI(1-367), and M.FokI(335-647) modify the C/A substrates, albeit at rates reduced by about 1-2 orders of magnitude with respect to the hemimethylated substrates. However, in all cases the C/A substrates were modified about 20-fold faster than the fully methylated reference substrates. This result confirms our conclusion that the adenine-N 6 MTases M.EcoRV, M. FokI(1-367), and M.FokI(335-647) are able to methylate cytosine residues.

DISCUSSION
Enzymes enhance the rates of chemical reactions by several orders of magnitude in comparison with the uncatalyzed situation. This is achieved by a selective stabilization of the transition state of the chemical reaction which, in general, depends on a very precise stereochemical arrangement of the functional groups of the substrate and enzyme. Therefore, enzyme-substrate interfaces are usually highly optimized with respect to both the structure of the substrate and the chemical reaction to be catalyzed. Thus, enzymes possess a high degree of substrate specificity. In many cases, even subtle alterations of the substrate diminish catalytic activity dramatically. DNA MTases, for example, methylate only nucleotides located within a precisely defined target sequence on the DNA.
In some cases, enzymes catalyze reactions that would not occur spontaneously. One example of this type is the methylation of adenine residues at the N 6 position. In solution the N 1 -atom of adenines is more reactive than N 6 , and methylation would take place at N 1 . Adenine-N 6 MTases, however, selectively enhance the reactivity of the N 6 group leading to a highly regiospecific direct methylation of adenine residues at the N 6 position (43). It remains unclear how methylation is selectively directed to the N 6 atom of adenines. One model to explain this regioselectivity would be that the enzyme positioned the flipped adenine in a precise stereochemical configuration with respect to the methyl group donor, such that only the N 6 position is accessible for the methyl group. Surprisingly, despite their catalytic efficiency and their DNA sequence specificity, we show here that adenine-N 6 DNA MTases do not have a pronounced specificity for their target base, because these enzymes also catalyze methylation of cytosine residues. This result suggests mechanistic similarities between the methylation of adenine and cytosine residues. The catalytic mechanism of adenine-N 6 and cytosine-N 4 MTases is not precisely known, but models have been proposed. It has been suggested that a deprotonation of the target amino group could be catalyzed by the conserved Asn/Asp or Ser in the (D/N/ S)PP(Y/F) motif (20). Alternatively, it has been suggested that the transition state of the methylation reaction is stabilized by cationinteractions between the flipped base and aromatic amino acid residues (44). In a third model, the deprotonation of the exocyclic amino group of the adenine is catalyzed by a transient protonation of the adenine ring accompanied by a tautomeric change (45). This protonation could occur at N 1 , which is the most basic nitrogen in the adenine ring. In the next step of the reaction, N 1 could be deprotonated, and by a tautomeric shift, the N 6 position would become more nucleophilic and could directly attack the AdoMet. Forming and breaking of a hydrogen bond to N 1 could have a similar effect. As a similar protonation (or hydrogen bond contact) could occur in cytosines at N 3 , this model is in accordance with our results, whereas important catalytic roles of the N 3 or N 7 atoms of the adenine ring are unlikely in light of our results, because these positions have no counterparts in cytosines. Note that these mechanistic models are not excluding each other.
DNA MTases can be subdivided according to the position that is modified into cytosine-C 5 MTases, which methylate aromatic ring carbon atoms, and N-MTases comprising adenine-N 6 and cytosine-N 4 MTases, which methylate exocyclic amino groups. These groups of enzymes differ in the reaction mechanism and the sequences of conserved amino acid motifs involved in catalysis. Assigning adenine-N 6 and cytosine-N 4 MTases into a common group is reasonable, because cytosine-N 4 and adenine-N 6 MTases display some sequence similarities at the F_G_G as well as SPP(Y/F) and (D/N)PP(Y/F) motifs. In fact, these enzymes cannot be readily distinguished on the basis of their conserved amino acid motifs, because the cytosine-N 4 MTase M.BamHI contains a DPPY motif, which is  Table I and "Experimental Procedures"). characteristic for adenine-N 6 MTases. Moreover the structures of all N-MTases are similar, and we have shown here that the catalytic activities of adenine-N 6 and cytosine-N 4 MTases are overlapping.
The overlap in the reaction specificities of adenine-N 6 and cytosine-N 4 MTases has implications on the molecular evolution of this class of N-MTases, because it opens an evolutionary pathway to change the reaction specificity of adenine-N 6 MTases to cytosine-N 4 and vice versa. So far, three groups of N-MTases are known (␣, ␤, and ␥) (17). These groups differ in the positions of the integration of the DNA recognition domain into the catalytic domain and by a circular permutation of the amino acid sequence in the catalytic domain (46), which lead to different arrangements and distances of the conserved amino acid sequence motifs (Fig. 5A). So far, cytosine-N 4 MTases have been found in the ␣ and ␤ groups. Two scenarios could account for the this situation: N-MTases could have diverged into adenine-N 6 and cytosine-N 4 MTases first, and, then differentiated into the three groups. However, in light of the large number of possible ways for circular permutations and possible sites for integration of a DNA recognition domain into the catalytic domain, it is unlikely that adenine-N 6 and cytosine-N 4 independently diverged into ␣ and ␤ groups having the same topology. Alternatively, N-MTases could have branched into the three groups first, and, afterward, members of the ␣ and ␤ group could have changed their substrate specificity from adenine-N 6 to cytosine-N 4 (Fig. 5B). This evolutionary pathway implies that the substrate specificities of N-MTases must have changed at least twice. Our data support the latter model, because they demonstrate that the catalytic framework of N-MTases is, in principle, able to support methylation of adenine and cytosine residues.
In contrast to the results obtained with adenine-N 6 MTases, the cytosine-N 4 MTase M.BamHI appears not to modify adenine residues at a detectable rate. Similar experiments with other cytosine-N 4 MTases will reveal if this result represents a general property of cytosine-N 4 MTases. The result obtained with M.BamHI can be interpreted in terms of molecular recognition, because the large active site of adenine-N 6 MTases can also accommodate cytosine, whereas the converse is less likely. This situation may be analogous to that of aminoacyl-tRNA synthetases, because a proofreading mechanism is required to prevent misloading of tRNAs with amino acid residues, which are smaller than the canonical residue.
Conclusions-We have shown here that five adenine-N 6 DNA methyltransferases also methylate cytosine residues at position N 4 . The rates of cytosine methylation by M.EcoRV and both domains of M.FokI are reduced by only 1-2 orders of magnitude, demonstrating that these enzymes have a low spec-ificity with respect to the target base. Our data confirm that adenine-N 6 and cytosine-N 4 belong to one enzyme family of N-MTases and suggest that these enzymes have repeatedly changed their specificity during evolution.