Molecular Enzymology of the Catalytic Domains of the Dnmt3a and Dnmt3b DNA Methyltransferases* 210

The C-terminal domains of the mammalian DNA methyltransferases Dnmt1, Dnmt3a, and Dnmt3b harbor all the conserved motifs characteristic for cytosine-C5 methyltransferases. Whereas the isolated catalytic domain of Dnmt1 is inactive, we show here that the C-terminal domains of Dnmt3a and Dnmt3b are catalytically active. Neither Dnmt3a nor Dnmt3b shows a significant preference for the satellite 2 sequence, although Dnmt3b is required for methylation of these regions in vivo. However, the catalytic domain of Dnmt3a methylates DNA in a distributive reaction, whereas Dnmt3b is processive, which accelerates methylation of macromolecular DNAin vitro. This property could make Dnmt3b a preferred enzyme for methylation at satellite 2 repeats, since they are highly CG-rich. We have also analyzed the catalytic activities of six different mutations found in ICF (immunodeficiency, centromeric instability, and facial abnormalities) patients in the catalytic domain of Dnmt3b. Five of them display catalytic activities reduced by 10–50-fold; one mutant was inactive in our assay (residual activity <1%). These results confirm that a reduced catalytic activity of Dnm3b causes ICF. However, the mutations in general do not completely abrogate catalytic activity. This finding may explain why ICF patients are viable, whereas nmt3b knock-out mice die during embryogenesis.


Humaira Gowher and Albert Jeltsch ‡
From the Institut fü r Biochemie, FB 8, Justus-Liebig-Universitä t, Heinrich-Buff-Ring 58, 35392 Giessen, Germany The C-terminal domains of the mammalian DNA methyltransferases Dnmt1, Dnmt3a, and Dnmt3b harbor all the conserved motifs characteristic for cytosine-C5 methyltransferases. Whereas the isolated catalytic domain of Dnmt1 is inactive, we show here that the Cterminal domains of Dnmt3a and Dnmt3b are catalytically active. Neither Dnmt3a nor Dnmt3b shows a significant preference for the satellite 2 sequence, although Dnmt3b is required for methylation of these regions in vivo. However, the catalytic domain of Dnmt3a methylates DNA in a distributive reaction, whereas Dnmt3b is processive, which accelerates methylation of macromolecular DNA in vitro. This property could make Dnmt3b a preferred enzyme for methylation at satellite 2 repeats, since they are highly CG-rich. We have also analyzed the catalytic activities of six different mutations found in ICF (immunodeficiency, centromeric instability, and facial abnormalities) patients in the catalytic domain of Dnmt3b. Five of them display catalytic activities reduced by 10 -50-fold; one mutant was inactive in our assay (residual activity <1%). These results confirm that a reduced catalytic activity of Dnm3b causes ICF. However, the mutations in general do not completely abrogate catalytic activity. This finding may explain why ICF patients are viable, whereas nmt3b knock-out mice die during embryogenesis.
In vertebrate DNA cytosine residues are modified by cytosine-C5 methylation mainly at CG sequences (reviewed in Refs. [1][2][3]. Approximately, 70 -80% of all CG sequences are modified in a cell type specific pattern. Methylation is involved in epigenetic processes like gene regulation during the embryonic development and cell differentiation, genomic imprinting, and X-inactivation. By silencing the expression of repetitive sequences, DNA methylation protects the genome against selfish genetic elements and helps to maintain genomic integrity. Hypermethylation and hypomethylation of DNA contributes to cancerogenesis and tumor progression (reviewed in Refs. 4 -6). DNA methylation is introduced by DNA methyltransferases (MTases), 1 which use S-adenosylmethionine (AdoMet) as donor for an activated methyl group (reviewed in Ref. 7). Depending on the developmental state of the cell, the DNA methylation pattern either has to be created de novo, or the existing pattern of methylation has to be maintained. To accomplish this purpose, vertebrates contain a maintenance methyltransferase, Dnmt1, and de novo methyltransferases, Dnmt3a and Dnmt3b, although there is now evidence accumulating that the functions of these proteins overlap (8 -10). 2 Dnmt1 is responsible for propagation of methylation pattern through cell generations by methylating the hemimethylated sites created after every round of replication. The Dnmt3 MTases have been assigned the role of de novo methylation, since they do not show a preference for hemimethylated DNA (8,12). De novo methylation of DNA by Dnmt3a and Dnmt3b was also demonstrated in vivo after expression in human cell lines (13) and by expression of Dnmt3a in transgenic Drosophila melanogaster (14). Dnmt3a and Dnmt3b are highly expressed in embryonic tissues, whereas only low expression is observed in differentiated cells (8,15), suggesting that these proteins are involved in the re-methylation of the genome in early embryogenesis that occurs after a massive demethylation immediately after fertilization (reviewed in Ref. 16). There are several alternative splice variants of Dnmt3b, two of which are active, one is not (8,17).
The Dnmt1 and both Dnmt3 proteins consist of an N-terminal part, which has regulatory and targeting functions and a C-terminal catalytic domain, which contains 10 characteristic amino acid motifs that are conserved among all cytosine-C5 MTases (reviewed in Ref. 18). The Dnmt3a and Dnmt3b proteins from mouse comprise 908 and 859 amino acid residues, respectively, and share 36% amino acid sequence identity with each other (Ͼ80% in their C-terminal domains). The N-terminal part of Dnmt1 is an important regulator of enzyme activity (19 -21) that has been shown to interact with many other proteins like PCNA, transcription factors, and histone deacetylases (22)(23)(24)(25)(26) and to be involved in targeting Dnmt1 to replication foci (27,28). The C-terminal domain of Dnmt1 was cloned several times, always showing that it is catalytically inactive and does not act as an independent methyltransferase despite the presence of all the amino acid motifs characteristic for DNA MTases (21,29,30). The N-terminal parts of Dnmt3a and Dnmt3b target the enzymes to heterochromatin regions in murine ES cells (31). They contain a Cys-rich region that is similar to the ATRX zinc finger, which interacts with the histone deacetylase HDAC1 (26) and a PWWP domain, which interacts with the DNA (32). Dnmt3a also interacts with RP58, a DNA-binding transcriptional repressor protein found at transcriptionally silent heterochromatin (26). The isolated catalytic domains of Dnmt3a and Dnmt3b have not yet been investigated.
Transgenic mice lacking Dnmt3a and Dnmt3b, singly and in combination, are hypomethylated and die at embryonic stages (Dnmt3a Ϫ/Ϫ /Dnmt3b Ϫ/Ϫ , Dnmt3b Ϫ/Ϫ ) or shortly after the birth (Dnmt3a Ϫ/Ϫ ), indicating the critical role of these enzymes during the development (33,34). Dnmt3b Ϫ/Ϫ mice show a massive demethylation at minor satellite sequences. Mutations in the Dnmt3b gene have been shown to be associated with the ICF syndrome (immunodeficiency, centromeric instabilities, facial abnormalities), a rare, genetic disease (34 -36) that is associated with hypomethylation of satellites 2 and 3 in pericentromic heterochromatin of chromosomes 1, 9, and 16 and in heterochromatic regions of the Y and inactive X chromosome (37). The hypomethylation of chromosomes at pericentromeric regions leads to the formation of complex multiradiate chromosomes (38). Hypomethylation of the DNA causes disregulation of gene expression which interferes with normal development of the immune system (39). The observation that missense mutations in the DNMT3B genes of ICF patients all occur at or near to the catalytic domain of Dnmt3b suggests that reduced catalytic activity of the enzyme is responsible for the disease (34 -36, 40). When combined, the results obtained with Dnmt3b Ϫ/Ϫ mice and ICF patients strongly suggest that Dnmt3b is responsible for methylation of minor satellite sequences like satellite 2 in vivo and that Dnmt3a cannot replace Dnmt3b in this function.
In this work we investigate whether the isolated C-terminal domains of Dnmt3a and Dnmt3b are catalytically active DNA MTases. We characterize the enzymes and show that the enzymatic properties of Dnmt3a and Dnmt3b might contribute to the functional specificity of Dnmt3b to methylate DNA at satellite 2 repeats in vivo and explain why Dnmt3a cannot substitute Dnmt3b in this function. Finally, we determined the catalytic activities of several Dnmt3b variants, which carry amino acid exchanges observed in ICF patients.

EXPERIMENTAL PROCEDURES
Oligodeoxynucleotides-Purified oligonucleotides were purchased from MWG (Ebersberg, Germany). Duplex oligonucleotides were prepared by annealing equimolar amounts of complementary strands. 30_GpC (Bt-GAAGCTGGGACTTCCGGGAGGAGAGTGCAA/TTGCA-CTCTCCTCCCGGAAGTCCCAGCTTC) contains one central CG site. To test for methylation of satellite sequences, we used SAT (Bt-CGAA-TGGTATCGAATGGAATCATCGAATA/TATTCGATGATTCCATTCG-ATACCATTCG), which has the consensus sequence of human satellite 2 repeats. The 33_2CpG oligonucleotide (Bt-TGGGACTTCCGGGAGC-TTCCGGGAGGAGAGTG/CACTCTCCTCCCGGAAGCTCCCGGAAG-TCCCA), a 33-mer that also has two CG sites was used as a control for CG methylation in a non-satellite sequence context.
Cloning, Expression, and Purification of Dnmt3a and Dnmt3b Catalytic Domains-Murine Dnmt3a and Dnmt3b cDNA clones (Entrez accession numbers: AAC40177 and AAF74515) were kindly provided by Dr. En Li (Charlestown, MA). Catalytic domains (CD) of Dnmt3a and Dnmt3b were cloned into pET28a as N-terminal His 6 fusion proteins. CD-3b variants carrying the mutations identified in ICF patients (A609T, G669S, L670T, V726G, D823G, and V824M in mouse Dnmt3b) were prepared using a PCR megaprimer mutagenesis method (41,42). Protein expression was carried out in Escherichia coli BL21 DE3 pLysS cells by addition of 1 mM isopropyl-1-thio-␤-D-galactopyranoside at 0.35 A 600 nm . The overexpressed proteins were purified over Ni-NTA agarose as described for the full-length Dnmt3a (12). In general, protein preparations were Ͼ90% pure. Protein concentration was quantified from A 280 nm using calculated extinction coefficients for each protein. Western blots were carried out using an ␣-His 6 tag antibody (Amersham Biosciences) according to the instructions of the supplier.
Methylation Assay Using Radioactively Labeled [methyl-3 H]AdoMet-DNA methylation was measured by the incorporation of tritiated methyl groups from labeled [methyl-3 H]AdoMet (3048 GBq/ mmol, PerkinElmer Life Sciences) into biotinylated oligonucleotides as described (43). The methylation reactions were usually carried out at concentrations of 1 and 0.76 M of DNA and labeled AdoMet, respectively, in methylation buffer (20 mM HEPES, pH 7.0, 1 mM EDTA, 100 mM KCl) at room temperature using enzyme concentrations of 0.5-2 M. The amount of radioactivity incorporated into 2 pmol of oligonucleotide substrate DNA was analyzed. Methylation of -DNA was analyzed by the DEAE filter binding assay using 300 ng of -phage DNA substrate at an enzyme concentration of 1 M in 10 l of methylation buffer. The reaction mixes were spotted on DEAE filters (Whatmann), which were washed five times with chilled 0.3 M ammonium bicarbonate solution, followed by a wash with 98% ethanol to remove water. The filters were dried and placed in scintillation fluid for the measurement of bound radioactivity.
Methylation Assay by Restriction Protection Analysis-The methylation-dependent restriction protection analysis was performed using an 850-bp and a 430-bp fragment of the Dnmt3b gene as a target for methylation. These fragments contain 4 (3) HhaI restriction sites and a total of 27 (9) CpG sites. A 50 nM concentration of each PCR fragment labeled with [␣-32 P]ATP was incubated with enzyme in methylation buffer containing 100 M AdoMet (Sigma). Aliquots were removed at defined times, and the reaction was stopped by addition of 3 volumes of ethanol. After precipitation, the DNA was dissolved in HhaI cleavage buffer (50 mM) potassium acetate, 20 mM Tris acetate, 10 mM magnesium acetate, 1 mM dithiothreitol, 100 g/ml bovine serum albumin, pH 7.9), and the samples were incubated with 10 units of HhaI for 1 h at 37°C. Cleavage products were separated on a 6% polyacrylamide gel run in TPE buffer (80 mM Tris phosphate, 20 mM EDTA, pH 8.0). The radioactivity was analyzed using an Instant Imager (Canberra Packard).

Enzymatic Characterization of Dnmt3a and Dnmt3b
Catalytic Domains-Long and short versions of the Dnmt3a and Dnmt3b catalytic domains (CD-3a, CD-3b) were cloned as Nterminal His 6 tag fusion proteins (Fig. 1). The proteins were overexpressed and purified over Ni-NTA-agarose to Ͼ90% as estimated from Coomassie Blue-stained SDS gels (Fig. 2). The identities of the proteins were confirmed by anti-His 6 antibody staining. The catalytic activity was determined by incorporation of tritiated methyl groups from [methyl-3 H]AdoMet into the DNA using an unmethylated oligonucleotide substrate with a single CpG site. All the four proteins were enzymatically active. However, both Dnmt3a enzymes were about 2-fold more active their the Dnmt3b counterparts. This result is in agreement to the kinetic properties of the full-length enzymes (17). Since the small version of CD-3a and the long version of CD-3b showed relatively higher activities, these proteins were used for most further investigations. The catalytic activity of the enzymes was tested at different salt concentrations and different pH, and the optimum activity was found at pH 7.0 in 20 mM HEPES buffer and at low KCl concentration (data not shown). We chose to use 100 mM KCl in our methylation buffer to stay close to physiological conditions. The result that the catalytic domains of Dnmt3a and Dnmt3b are active DNA MTases suggests that these domains resemble prokaryotic cytosine-C5 Mtases, which in general do not have a large N-terminal part (reviewed in Refs. 3, 44, and 45). It is in sharp contrast to the observation that the Cterminal part of Dnmt1 was repeatedly found not to be an active methyltransferase, although it harbors all the catalytic amino acid motifs characteristic for cytosine-C5 MTases (21,29,30).
Recognition of Satellite 2 Sequence-Since the catalytic domains of Dnmt3a and Dnmt3b resemble prokaryotic DNA Mtases, which in general methylate DNA at defined recogni-tion sequences, the preference of Dnmt3b for methylation at satellite 2 sequences could be due to an intrinsic preference of its catalytic domain for the nucleotide sequence of the satellite repeat. To test this model, a 29-mer oligonucleotide substrate was designed to have a consensus satellite 2 sequence. This substrate carried two centrally placed CG sites and one right at the end. As control the 33_2CpG substrate was employed that also carries two centrally placed CpG sites but in a random sequence context. As shown in Fig. 3, the catalytic domain of Dnmt3b clearly shows no preference for methylation at the satellite 2 sequence. Therefore, simple recognition of the satellite sequence cannot explain the in vivo role of Dnmt3b.
Processivity of DNA Methylation by CD-3a and CD-3b-The processivity of enzymatic turnover is an important mechanistic parameter for enzymes like DNA Mtases, which act on linear substrates containing several target site for the enzyme. Processive and distributive modes of DNA methylation can be distinguished on the basis that they result in a completely different distribution of unmodified, partially modified, and fully modified substrate molecules during the reaction. Processive methylation converts the unmodified substrate directly into fully modified products without allowing the intermediates to populate, whereas a distributive methylation leads to an accumulation of partially methylated DNA molecules as reaction intermediates. We checked the processivity of DNA methylation by CD-3a and CD-3b using two substrates varying in length, one 850-mer and one 430-mer. We investigated the protection of a these DNA fragments against HhaI digestion by CG methylation. The 430-bp substrate DNA contains three HhaI sites. HhaI cleavage of the 430-mer can result in up to 10 different fragments (four substrate cleavage fragments, five intermediates, one full-length fragment), eight of which can be observed in polyacrylamide gels.
As shown in Fig. 4, CD-3a methylates the 430-mer in a distributive fashion. There is a continuous increase in the amount of the partially methylated DNA during the time course of the reaction, and fully protected DNA is only observed after almost complete disappearance of the substrate cleavage fragments. This reaction profile clearly shows that CD-3a treats each target site independently and is mechanistically distributive as already shown by us for the Dnmt3a full-length protein (12). In contrast, not many intermediates appear after methylation of the same fragment by CD-3b, and large amounts of fully protected DNA and substrate cleavage fragments are present at the same time in the reaction mixture. Therefore, CD-3b appears be much more processive than CD-3a. Similar results were obtained with an 850-mer that contains 5 HhaI and 27 CG sites (data not shown). This difference in the reaction mechanism of CD-3a and CD-3b was observed  with different enzyme preparations and also with a buffer containing no KCl. It is not due to the fact that we used the long version of the catalytic domain of Dnmt3b and the short version of Dnmt3a for these experiments, because similar experiments with the longer version of Dnmt3a also showed a distributive reaction mechanism (data not shown), and also the full-length Dnmt3a enzyme methylates DNA in a distributive fashion (12).
To analyze the processivity of DNA methylation of both enzymes in more quantitative terms, we determined the relative amounts of all fragments in methylation reactions with CD-3a and CD-3b ( Fig. 4 and Supplemental Fig. 1) (in Supplemental Fig. 1 the relative amounts of all the different species are shown, in Fig. 4 the amounts of all substrate cleavage fragments and all intermediates are added to obtain a graph the is less complicated). The data were fitted to a model, which uses six different rate constants to describe the reactions: three distributive rate constants (k 1 , methylation at site 1; k 2 , methylation at site 2; and k 3 , methylation at site 3), two partially processive ones (k 12 , coupled methylation at sites 1 and 2; k 23 , coupled methylation at sites 2 and 3), and one fully processive one (k 123 , coupled methylation at all three sites). It should be noticed that this model does not make any initial assumptions on the degree of processivity of the methylation reaction. Both reaction profiles could be nicely fitted ( Fig. 4 and Supplemental Fig. 1) yielding the following rate constants for CD-3a: k 1 , 3.1 ϫ 10 Ϫ2 min Ϫ1 ; k 2 , 8.6 ϫ 10 Ϫ3 min Ϫ1 ; k 3 , 2.4 ϫ 10 Ϫ2 min Ϫ1 ; k 12 , 0 min Ϫ1 ; k 23 , 6.2 ϫ 10 Ϫ4 min Ϫ1 ; k 123 , 0 min Ϫ1 . This result shows that CD-3a methylates DNA in an almost completely distributive reaction, because 99% of the activity is assigned to k 1 , k 2 , and k 3 and only 1% to a coupled methylation of two sites (k 23 ). This outcome confirms the qualitative interpretation of the gel figures that CD-3a works in an almost purely distributive manner. For CD-3b a completely different result was obtained: k 1 , 0 min Ϫ1 ; k 2 , 0 min Ϫ1 ; k 3 , 6.3 ϫ 10 Ϫ3 min Ϫ1 ; k 12 , 0 min Ϫ1 ; k 23 , 2.9 ϫ 10 Ϫ3 min Ϫ1 ; k 123 , 0.12 min Ϫ1 . Here 93% of the total activity is assigned to k 123 , which corresponds to a processive methylation of all three sites. Similar results were obtained in repeated analyses, where processive activity of CD-3a never went beyond 5%, and processivity of CD-3b never was below 80%. As the 430-mer contains nine CG sites, and the three HhaI sites are a representative subset of all the CG sites, our results mean that CD-3b methylates all nine CG sites on the 430-mer processively. Since almost complete processivity of DNA methylation also was observed with the 850-mer, we conclude that CD-3b is able to methylate at least 27 CG sites without dissociating from the DNA. These analyses also show that the total activity of CD-3b (0.36 sites/min) is much higher than that of CD-3a (0.06 site/min) with the macromolecular substrate, although with oligonucleotides that contain only one target site, CD-3a is about 2-fold more efficient. This result clearly demonstrates that a processive reaction mechanism accelerates methylation of macromolecular substrates, which contain more than one target site.
ICF Mutations-Specific point mutations have been mapped in the DNMT3B gene of ICF patients, which are located in or close to the catalytic domain of the Dnmt3b protein (34 -36). We wanted to check the ICF mutant forms of CD-3b for catalytic activity and carried out site directed mutagenesis to create six of them. The proteins showed different levels of overexpression, but all of them could be purified from BL21(DE3, pLysS) E. coli cells and the methylation activity tested using the DEAE filter binding assay. We first analyzed the total incorporation of radioactivity into -DNA after 1-h incubation with the enzymes in the presence of labeled [methyl-3 H]AdoMet (Fig. 5A). To check for background methylation, protein purification was carried out from BL21(DE3, pLysS) cells, and the same volumes were used for control methylation reactions. The total counts observed in the control reactions never went beyond 200 cpm. However, five of the six variants result in significantly higher incorporation of radioactivity in the DNA, demonstrating that these mutants are active MTases. To determine the rates of DNA methylation by the ICF variants, methylation kinetics were measured and compared with reactions with control preparations. Examples of the time courses obtained are shown in Fig. 5B, and the averaged results of at least three independent experiments are compiled in Fig. 5C. In these gels the intensities of all bands were determined (see Supplemental Fig. 1) and the data fitted to a model that allows distributive and processive methylation of DNA. In the lower panel the quantitative results are shown. Here the intensities of all substrate cleavage fragments and of all intermediates are combined, such that only three curves must be compared: fully protected DNA (f), intermediates (OE), and substrate cleavage fragments (ࡗ). The lines show fits of the data to a model that considers six rate constants for DNA methylation: k 1 , methylation only at site 1; k 2 , methylation only at site 2; k 3 , methylation only at site 3; k 12 , simultaneous methylation at sites 1 and 2; k 23 , simultaneous methylation at sites 2 and 3; k 123 , simultaneous methylation at sites 1, 2, and 3. k 1 , k 2 , and k 3 describe a distributive reaction; k 12 and k 23 are partially processive; and k 123 is fully processive.
The slopes of the reaction progress curves observed in the control reactions always was Ͻ1% of those obtained with the wild type CD-3b preparations. We, therefore, conclude that catalytic activities Ͼ1% of CD-3b can be detected in this assay. As shown in Fig. 5, five of the mutants show a clear catalytic activity that is between 1 and 10% of the wild type activity with A609T being the lowest with a relative activity of 1.8% of CD-3b. We could not detect activity that was higher than the background with the V726G. However, given the limited sensitivity of the assay, this result does not mean that this variant is catalytically inactive. We conclude form our data that all ICF mutants have a significantly reduced catalytic activity, suggesting that decreased Dnmt3b activity causes ICF. However, ICF variants cannot be considered as being catalytically inactive in general. DISCUSSION In vertebrates the pattern of DNA methylation is used to transmit epigenetic information that has numerous important biological functions. The methylation pattern of the DNA is generated during embryogenesis by de novo methylation (reviewed in Ref. 16). The Dnmt3a and Dnmt3b enzymes participate in this process whose regulation and control is still enigmatic. One interesting example of de novo methylation is the methylation of the pericentromeric satellite 2 sequences. These repeats are unmethylated in germ cells but methylated in adult tissues (46,47). Satellite 2 sequences are undermethylated in patients with ICF syndrome that have a mutated DNMT3B gene (34 -36, 40) and in Dnmt3b knock-out mice (34). Therefore, Dnmt3b most likely is responsible for methylation at these sites during embryogenesis, and Dnmt3a cannot efficiently replace Dnmt3b in this function.
We have shown that the C-terminal catalytic domains of the Dnmt3a and Dnmt3b DNA MTases are enzymatically active independent of their N-terminal parts. This implies that Dnmt3a and Dnmt3b, unlike Dnmt1, do not essentially require any part of N-terminal domain for catalytic activity. Therefore, the catalytic domains of Dnmt3a and Dnmt3b resemble prokaryotic DNA Mtases, which in general specifically methylate DNA at short recognition sequences (reviewed in Refs. 3, 44, and 45). We have studied the catalytic activities of six ICF mutants and show that five of them have 1-10% of the wild type activity, and one mutant has Ͻ1% activity. In a previous study the D823G variant did not show catalytic activity in an in vivo assay (35), although this is only reduced 12-fold according to our data. This difference might be due to different expression levels of the wild type and mutant in vivo and/or the sensitivity of the in vivo assay. These results confirm that a reduced catalytic activity of Dnmt3b leads to hypomethylation at satellite 2 sequences and causes ICF. Moreover, the residual activity of the ICF variants may explain why a knock-out of Dnmt3b in mice is lethal during embryogenesis, whereas ICF patients are viable. This observation is in agreement with the finding that in ICF patients methylation of the satellite repeats is not completely lost but only reduced 2-10-fold with large individual differences (48).
Our results show that the catalytic domains of Dnmt3a and Dnmt3b methylate DNA following a very different kinetic mechanism; whereas the catalytic domain of Dnmt3a, like fulllength Dnmt3a (12), methylates DNA in a distributive reaction, the catalytic domain of Dnmt3b is processive. Given the fact that the Dnmt3a and Dnmt3b catalytic domains are about 84% identical in amino acid sequence and in addition share a very high level of homology, it is rather puzzling to find such a clear mechanistic difference between these two proteins. However, among the 44 amino acid residues that are not identical between human and murine Dnmt3a and Dnmt3b in the Cterminal 283-amino acid residues of the proteins, 15 include charged residues. The exchanges observed among these residues are highly biased such that finally Dnmt3b carries six more positive charges than Dnmt3a. Therefore, Dnmt3b has a much more positively charged DNA binding cleft than Dnmt3a (Supplemental Fig. 2), which could explain why Dnmt3b methylates DNA in a processive reaction, whereas Dnmt3a is distributive.
The difference in the kinetic mechanisms of the catalytic domains of Dnmt3a and Dnmt3b could be related to the distinct functions the enzymes in the cell, because satellite 2 repeats are among the most CG-rich sequences in the genome: human methylcytosine in methylated CG sequences has a higher propensity for deamination and conversion to T, which leads to a depletion of CG sequences in the genome (reviewed in Ref. 51). Satellite 2 DNA is not CG depleted, because it is unmethylated in the germ line. For comparison, so called CpG islands are defined as having a value of observed/expected of Ͼ0.6 (52), and only very few regions in the human genome (0.03% of all nucleotides) have observed/expected values of Ͼ0.8 for the CG dinucleotide (50). Therefore, the satellite repeats are exceptionally rich in CG sites when compared with the rest of the genome. The high processivity of Dnmt3b makes it well suited to modify these regions, because after targeting to the DNA it can methylate several cytosine residues in a processive reaction. In contrast, Dnmt3a methylates DNA in a distributive fashion and dissociates from the DNA after each turnover, which could explain why Dnmt3a cannot replace Dnmt3b at satellite repeats. This model is directly based on our experimental results and does not rule out that Dnmt3b could be targeted to satellite sequences by interaction with other cellular proteins. In fact both of these mechanisms might work synergistically, because after targeting Dnmt3b to satellite DNA, the processive methylation would promote methylation of the satellite sequences.
Finally, the question on the biological relevance of our results obtained with isolated domains of Dnmt3a and Dnmt3b has to be addressed. The interaction between different domains in multidomain proteins can vary between the two different extremes that individual domains can be completely independent of each other or that the activity of one domain is under strict control of another domain. We show here that the Dnmt3a and Dnmt3b proteins do not follow the latter mode, since the isolated catalytic domains are catalytically active. This observation justifies studying the enzymatic properties of these proteins by using their catalytic domains as model systems. This approach allows us to investigate the enzymology of the catalytic part of the MTases in the absence of the possible influence of the N-terminal domains and of other proteins. Thereby, the fundamental properties of the enzymes can be studied, which represent the starting point for any kind of modulation by interactions with the N-terminal part and other cellular proteins. Therefore, the results of our study represent an essential basis for further investigation, because any influence of the N-terminal part or other proteins on the properties of the Dnmt3a and Dnmt3b can only be defined by comparison with the properties of the individual domains. Moreover, our results lead to conclusions that most likely are also applicable to the full-length enzyme. If the catalytic domain of Dnmt3b is processive it is very likely that the full-length enzyme retains this property, because processivity is a highly evolved enzymatic property that is unlikely to be generated by truncation of a protein. In contrast, Dnmt3a is distributive both as a fulllength protein and truncated protein. This observation does not exclude that Dnmt3a could acquire processivity by interaction with other proteins. Finally, if Dnmt3b ICF variants are catalytically active even as isolated domains it is very likely that the full-length variants will also display some catalytic activity.