Recombinant Human DNA (Cytosine-5) Methyltransferase

Steady-state kinetic analyses revealed that the methylation reaction of the human DNA (cytosine-5) methyltransferase 1 (DNMT1) is repressed by the N-terminal domain comprising the first 501 amino acids, and that repression is relieved when methylated DNA binds to this region. DNMT1 lacking the first 501 amino acids retains its preference for hemimethylated DNA. The methylation reaction proceeds by a sequential mechanism, and either substrate (S-adenosyl-l-methionine and unmethylated DNA) may be the first to bind to the active site. However, initial binding of S-adenosyl-l-methionine is preferred. The binding affinities of DNA for both the regulatory and the catalytic sites increase in the presence of methylated CpG dinucleotides and vary considerably (more than one hundred times) according to DNA sequence. DNA topology strongly influences the reaction rates, which increased with increasing negative superhelical tension. These kinetic data are consistent with the role of DNMT1 in maintaining the methylation patterns throughout development and suggest that the enzyme may be involved in the etiology of fragile X, a syndrome characterized byde novo methylation of a greatly expanded CGG·CCG triplet repeat sequence.

The mammalian genome is epigenetically modified by the transfer of methyl groups from S-adenosyl-L-methionine (AdoMet) 1 to acceptor bases in double-stranded DNA, mostly at the carbon 5 of cytosine when the base is part of a CG dinucleotide (1). Several DNA methyltransferases (Dnmt) have been identified, including Dnmt1 (2), Dnmt2 (3), Dnmt3A, Dnmt3B (4), a splice variant of Dnmt1 (Dnmt1b) (5), and an oocytespecific isoform of Dnmt1 that lacks the first 118 N-terminal amino acids (6). Targeted mutations of Dnmt1 (7), Dnmt3A, and Dnmt3B genes are recessive lethals (8) in mice attesting to their essential role in development. In humans, mutations in the DNMT3B gene have been associated with the (immunodeficiency, centromere instability, and facial abnormalities syndrome (ICF syndrome) (9), a recessive disorder characterized biochemically by hypomethylation of satellite 2 and 3 DNA from the juxtacentromeric heterochromatin of chromosomes 1 and 16 (10 -12) as well as other common non-satellite repeats (13).
Genetic studies suggest that Dnmt1 and Dnmt3 carry out two types of modifications as follows: both Dnmt3A and -B establish patterns of methylation early in embryogenesis by de novo methylation of cytosine residues, whereas Dnmt1 maintains such patterns throughout life by copying them onto newly synthesized DNA (4,14). Methylation studies in vitro do not reveal such distinction of functions. In fact, whereas kinetic determinations show that Dnmt1 utilizes hemimethylated DNA 5-10-fold better than unmethylated DNA (15)(16)(17)(18)(19), they also show that this enzyme methylates unmethylated DNA more efficiently than Dnmt3A and Dnmt3B (4). Therefore, it is unclear whether the de novo activity of Dnmt1 is repressed in vivo.
Most cases of fragile X syndrome, a relatively frequent (1: 4000 births) hereditary neurological disease that causes mental retardation (20), are associated with the expansion of a CGG⅐CCG repeat in the 5Ј-untranslated region of the FMR1 gene (21). The repeat is polymorphic in the general population, varying from 6 to 60 units. Occasional expansions from 61 to 200 units, in the so-called pre-mutation range, trigger further instability such that the tract expands up to thousands of repeats (full mutation) when transmitted to offspring (22). In individuals with full mutation, the disease state also requires the de novo methylation of the CGG⅐CCG tract and the proximal CpG island (23)(24)(25), which leads to silencing of FMR1 transcription. However, two important unknowns include the following: first, what mechanisms trigger this de novo activity, and second, which methyltransferase(s) are involved.
Herein, we present steady-state kinetic investigations on human recombinant DNMT1 (19) using DNA substrates of random sequence or substrates composed of CGG⅐CCG repeats. The data show that DNMT1 is under allosteric repression, a control mechanism that reconciles the results in vitro with the genetic studies in vivo. The binding of AdoMet and the DNA to the active site occurs by a random mechanism; however, the preferred pathway involves binding of AdoMet followed by DNA. The enzymatic activity depends on the topological state of the DNA and increases with negative supercoiling. These results indicate that although the methylation reaction is un-der tight control, factors such as CGG⅐CCG expansion and the accumulation of high levels of supercoiling stimulate de novo activity, which are consistent with the possibility that DNMT1 may be involved in the fragile X syndrome.

EXPERIMENTAL PROCEDURES
Human (Cytosine-5) DNA Methyltransferase Transfer Vector, Viral Transfection, and Recombinant Protein Expression-Full-length 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 (26). Baculovirus expression of the full-length DNMT1 was from clone HMT8. Transfer vector pVICHMT (19) has three BamHI sites, one before the DNMT1 translation initiation (in the vector) and the other two in the DNMT1 open reading frame at nucleotide positions 1704 and 2784 (GenBank TM accession number X63692). For the preparation of the Nterminal 501 amino acid deletion mutant, DNMT1 ⌬501 , partial digestion of pVICHMT was done with BamHI. A 14.5-kilobase pair band corresponding to the DNA fragment lacking the first 1704 nucleotides of the DNMT1 cDNA was excised from the gel, religated, and transformed into Escherichia coli. The ligated junctions in the final pVICD501HMT construct were verified by DNA sequencing. The enzyme was purified from clone 21 (HMT21). A pupal ovarian cell line (SF9) from the worm Spodoptera frugiperda was used for co-transfection and expression of the human DNMT1 and DNMT1 ⌬501 as described previously (27) with the following modifications. SF9 cells were maintained as a suspension culture in TNM-FH media (JRH Biosciences, Lenexa, KS) supplemented with 10% v/v fetal calf serum and an antibiotic/antimycotic solution at a final concentration of 5 units of penicillin, 50 g/ml streptomycin, and 0.125 g/ml amphotericin B at 27°C on a Bellco steering platform at 70-80 rpm. Co-transfection of a monolayer of SF9 insect cells was carried out using BaculoGold DNA, a modified, linearized Autographa californica nuclear polyhedrosis virus DNA (PharMingen, San Diego, CA), and the transfer vector pVICD501HMT. Screening for recombinant clones and plaque purification was described previously (28). For routine protein expression, SF9 cells were grown in spinner culture flasks. SF9 cells at a density of 1.2-1.5 ϫ 10 6 ml Ϫ1 were infected at a multiplicity of infection between 7 and 10 with HMT8/ HMT21. The cells were kept at 27°C and stirred at 60 rpm. Cells were harvested 48 h post-infection, pelleted, and washed with phosphate-buffered saline. Protein purification and quantitation were described previously (19). Enzyme preparations were Ͼ95% pure (19).
Determination of the Kinetic Constants-We previously developed a velocity equation for DNMT1 (29), based on the King-Altman and Cleland methods (30), that considered the enzyme species bound to the AdoMet and unmethylated DNA substrates but not to the AdoHcy and methylated DNA products (initial forward velocity equation). These conditions are met experimentally when the amount of substrate utilized is negligible compared with its total concentration. The velocity equation in the presence of products would be quite complex. In our velocity equation, the term associated with K ia was arbitrarily chosen to be AdoMet because the order for substrate addition in the methyl transfer reaction was unknown. Because evidence is now presented that the reaction order for DNMT1 is random, we re-write the velocity equation in the form shown in Equation 1: ) as outlined in Fig. 1. This modification of the velocity equation does not alter the estimation of the Michaelis constants and k cat ; however, it enables the calculation of the dissociation constants for AdoMet and DNA as follows. In Equation 1 when [A] is the variable substrate the reciprocal of the velocity is as shown in Equation 2.
When the slopes in Equation 2, which are obtained by varying the fixed concentrations of [B], are replotted as a function of 1/[B], they yield the following Equation 3, Alternatively, when [B] is the variable substrate in Equation 1, the reciprocal of the velocity is given in Equation 4.
The reaction conditions for the methyl transfer reaction with recombinant human DNMT1 ⌬501 and DNMT1 with all of the DNA substrates used and the calculation of the steady-state kinetic constants were as described (29). The concentrations of the reactants and those of the products analyzed were always converted to nanomoles or micromoles per liter and therefore are expressed in nM or M, respectively. Also, the given concentrations are always those in the final reaction mixture.
DNA Substrates and Products-The DNA substrates used for calculating the kinetic constants were the polymer poly(dI-dC)⅐poly(dI-dC) (Amersham Pharmacia Biotech, average length 7000 bp); the oligonucleotides (CGG⅐CCG) 12 , (m 5 CGG⅐CCG) 12 , (CGG⅐Cm 5 CG) 12 , and (CGG⅐CCG) 73 ; SNRPN (exon 1 of the small nucleoriboprotein N gene); SNRPN methylated on the upper strand; SNRPN methylated on the lower strand; and plasmid DNA that included linear, relaxed circular, and negatively supercoiled pRW3602. The synthesis and characterization of these substrates were reported (19,29). Negatively supercoiled pRW3602 (32) was isolated from E. coli strain HB101 and purified by CsCl banding (33). Its average negative superhelical density (ϪϽϾ ϭ 0.045) was determined from a series of agarose gel electrophoreses performed in the presence of chloroquine using known populations of topoisomers as reference (32). Two oligonucleotides were employed for product inhibition studies as follows: Me CG 20 , a chemically synthesized 40-bp duplex oligonucleotide containing 5-methylcytosine at all 20 CG steps, and F/Me CG 12 , a chemically synthesized 36-bp duplex oligonucleotide containing the FMR1-associated CGG⅐CCG triplet repeat substituted with 5-fluorocytosines on the upper strand and 5-methylcytosines on the lower strand at the CG steps as described (29).
Methylation Reaction on Topological Isomers of pRW3691-Sixteen g of CsCl-purified pRW3691 (29,34) were treated with chicken erythrocyte topoisomerase I in the presence of 0, 0.5, 1.0, 1.5, 2.0, and 3.0 g/ml ethidium bromide, purified, and analyzed for their average number of superhelical turns as described (32). The concentration of CG steps for each of the six purified topoisomer populations was obtained from the DNA concentration measured spectrophometrically at 260 and 280 nm. For each topoisomer population, an amount of pRW3691 corresponding to 25 M CG steps in the final reaction mixture was incu-  Co.), and counted on a liquid scintillation analyzer (Packard Instrument Co.). Control vials containing free [ 3 H]AdoMet, but no melted gels slices, or free [ 3 H]AdoMet plus a melted gel slice of comparable weight (ϳ0.8 g), but no DNA, indicated that the melted gel matrix did not quench the radioactive signals. The concentration of [ 3 H]CH 3 incorporated in each DNA fragment was obtained from the ratio of the counts/ min of the DNA samples minus the counts/min of the blank, divided by the counts/min given by 2 l of free [ 3 H]AdoMet. The blank consisted of an average counts/min value obtained from four melted gel slices of 0.8 g each that were excised from the DNA lanes of the agarose gels used to separate the restriction fragments. This value was 590 cpm. The values for the DNA samples ranged from 3,788 to 2,306,472 cpm.

RESULTS
An N-terminal Peptide Regulates the Reaction Rate through DNA Binding-Previous studies with full-length recombinant human DNMT1 revealed the linear and intersecting velocity patterns for the majority of the DNA substrates used (19,29). However, complex patterns were observed with two DNA substrates, poly(dI-dC)⅐poly(dI-dC) (35) and hemimethylated (CGG⅐Cm 5 CG) 12 , which revealed both inhibition and activation of the reaction. Furthermore, inhibition studies with fully methylated random sequence DNA (29) showed that this product stimulated methylation. These data suggested that the reaction was normally repressed and that de-repression could be achieved once the DNA substrate was bound to an allosteric site.
To test this hypothesis, deletions were made in the N-termi-  using poly(dI-dC) ⅐ poly(dI-dC) and FMR1 locus as substrates The standard deviations from two determinations ranged from 0.6 to 14% of the mean values. The ratio between the hemimethylated and the unmethylated FMR1 sequence is in parentheses. nal portion of DNMT1, which was believed not to be involved in catalysis (2), and the velocity curves were re-evaluated at various concentrations of AdoMet and DNA (poly(dI-dC)⅐poly(dI-dC) and hemimethylated (CGG⅐Cm 5 CG) 12 ). Herein, we report the results with DNMT1 ⌬501 , a catalytically active enzyme that lacks the first 501 N-terminal amino acids (19) (Fig. 2). Fig. 3 shows the velocity patterns for (CGG⅐CCG) 12 , as control, and for (CGG⅐Cm 5 CG) 12 12 with the truncated enzyme are identical to those obtained with the full-length enzyme (29), i.e. they are linear and intersect to the left of the vertical axis. This result indicates that the truncation did not alter the bi-reactant sequential mode of substrate binding characteristic of DNMT1 and gives confidence that kinetic studies carried out with DNMT1 ⌬501 contribute to elucidating the properties of the full-length enzyme. The patterns for the hemimethylated (CGG⅐Cm 5 CG) 12 are also linear and intersecting (C and D), contrary to those observed with DNMT1 (29), which were curved and not intersecting. This result indicates that a region within the 501 N-terminal amino acids was involved in generating the complex kinetic pattern (29) for (CGG⅐Cm 5 CG) 12 with DNMT1.
The velocity curves with full-length DNMT1 showed substrate inhibition with poly(dI-dC)⅐poly(dI-dC), particularly at low AdoMet concentrations (19). The patterns obtained with DNMT1 ⌬501 are shown in Fig. 4. At the identical AdoMet concentrations, substrate inhibition is no longer present (A), confirming a role for the N-terminal domain in regulating the turnover rate. When AdoMet is the variable substrate (B) the velocity responses are linear from 1.74 to 20 M (1/AdoMet of 0.05-0.57). The loss of linearity at 1.02 M (1/AdoMet ϭ 0.98) probably reflects a departure from steady-state conditions.
In summary, this transformation from curved velocity patterns with DNMT1 (19,29) into linear and intersecting patterns with DNMT1 ⌬501 (Figs. 3 and 4) shows that the N-terminal domain of the enzyme is involved in the regulation of the reaction rate through binding of the DNA.  (Fig. 5). These results enable the evaluation of the Michaelis constants for DNA and AdoMet as well as the turnover numbers (Table I) (see "Experimental Procedures" and Ref. 29). A comparison of the k cat and k cat /K m DNA values for the unmethylated and the hemimethylated FMR1 sequence indicates that preference for the hemimethylated template is retained in DNMT1 ⌬501 , showing that the residues involved are downstream of the N-terminal regulatory domain. Table II shows the ratios of the steady-state kinetic parameters between DNMT1 ⌬501 and DNMT1. Both the values of k cat and k cat /K m DNA are greater for the truncated enzyme, indicating that the N-terminal peptide represses the enzymatic activity. Inspection of the ratios for the substrate dissociation constants (K i(app) DNA and K i(app) AdoMet ) indicates that the truncation increases the apparent affinity of DNMT1 ⌬501 for the hemimethylated FMR1 sequence. Furthermore, because the K i(app) AdoMet and K i(app) DNA ratios are 0.1 and 0.3 for the unmethylated sequence, respectively, we conclude that the truncation increases the apparent affinity for both AdoMet and the FMR1 sequence. Thus, we propose that the N-terminal domain of DNMT1 interferes with the accessibility of the substrates to the catalytic center, therefore reducing the overall reaction rate.
AdoMet and DNA Bind Randomly to the Active Site-An aspect that remained unresolved during the previous study (29) was the order in which the two substrates bound to the active center, i.e. whether either AdoMet or DNA can be the first to bind or whether their addition must follow a defined order. The difficulty in establishing this kinetic question was the presence of the N-terminal regulatory site, which yielded  stimulatory effects in the presence of fully methylated DNA, thus obscuring the inhibition patterns. If the absence of such a domain in DNMT1 ⌬501 confines binding of methylated DNA to the active site and thereby eliminates activation, this would reveal whether substrate addition is random or ordered. For a bireactant ordered reaction, the inhibition pattern for the first substrate that binds is competitive versus its product, whereas that of the second substrate is non-competitive versus its product. Alternatively, the inhibition patterns for both substrates are competitive versus their respective products for a random reaction (36). Our previous experiments (29) showed that the inhibition pattern between AdoMet and its product AdoHcy was competitive. The inhibition pattern of fully methylated DNA versus unmethylated DNA for DNMT1 ⌬501 is shown in Fig. 6. Fig. 6A shows the reciprocal of the incorporation of [ 3 H]CH 3 by supercoiled pRW3602 (vertical axis) as a function of DNA (horizontal axis) at changing fixed concentrations of methylated DNA and a constant AdoMet concentration of 6.67 M. The convergence of the curves on the vertical axis was based on the replot of their slopes and intercepts (Fig. 6B), which indicated that the slopes increased as a function of methylated DNA, whereas the 1/V max(app) values remained constant. This pattern shows competitive inhibition between fully methylated and unmethylated DNA and, together with the patterns obtained with AdoHcy, indicates that both AdoMet and DNA add randomly to the catalytic center. Thus, we conclude that the methyl transfer reaction for recombinant human DNMT1 occurs by a bireactant sequential random mechanism.
Negative Supercoiling Increases Reaction Rates-The interest in determining whether negative supercoiling influences the rate of methylation was largely stimulated by the clinical observation that the number of CGG⅐CCG repeats in the FMR1 locus expands to more than 200 copies, from 6 to 60 of the general population, and becomes de novo methylated in fragile X patients (22). Because the bending restoring force, but not the torsional modulus, for CGG⅐CCG DNA is ϳ40% lower than for random DNA (34), we hypothesized that an expanded CGG⅐CCG tract would represent a genomic site for the preferential partitioning of negative supercoiling (37), which is known to unwind stretches of DNA (38). The free energy of supercoiling associated with unwinding would consequently generate alternative DNA structures such as slipped structures, which could increase the rates of the methylation reaction. Thus, the CG steps within an expanded CGG⅐CCG would become kinetically favorable substrates for de novo methylation.
Six families of topological isomers of pRW3691, which contains an insert with 73 consecutive CGG⅐CCG triplet repeats, with increasing average negative supercoil densities were methylated with DNMT1 and [ 3 H]AdoMet under conditions where no more than 5-7% of CG sites were modified. The insert harboring the triplet repeat tract, with a total of 164 CG steps (146 CG steps from the triplet repeat plus 18 CG steps from its flanking sequence), and the vector harboring 170 CG steps were separated by 1% agarose gel electrophoresis following restriction enzyme cleavage, and the extent of methylation was determined for the insert and the vector (Fig. 7).
The turnover rates increased with negative supercoiling, about 2-fold for the vector and 7-fold for the triplet repeat tract, indicating that the free energy of supercoiling plays a significant role in methylation. The concentration of [ 3 H]CH 3 incorporated into the triplet repeat-containing fragment were 5-10fold lower than those incorporated in the vector. These differences are not justified by the k cat values for plasmid DNA (Table III) and the (CGG⅐CCG) n triplet repeats (Table V). Instead, because these two DNAs differ mostly in their K m DNA and K i(app) DNA values (Table IV), it is possible that DNA sequencespecific interactions with DNMT1 and/or differences in DNA binding at the catalytic versus regulatory domain dominate the apparent turnover rates (i.e. at unsaturating substrate concentrations). A second possibility is that the CGG⅐CCG repeats display substrate inhibition [like poly(dI-dC)⅐poly(dI-dC)] at high concentrations. In fact, because of the low K m DNA values, the concentrations of CG steps ranged between 0.1 and 5 M in the velocity studies with (CGG⅐CCG) n , whereas they ranged between 3 and 40 M for this plasmid DNA.
A second analysis was conducted by evaluating the steadystate kinetic parameters with DNMT1 for pRW3602 under three different topological conditions, supercoiled (Ϫ ϽϾ ϭ 0.045), relaxed circular, and linear (Table III) (29). The k cat values were 5-and 3-fold higher for the supercoiled and linear conformers, respectively, than for relaxed DNA. The most dramatic difference was the 7.7-fold reduction in K m DNA for relaxed DNA, suggesting that DNA topology influenced the rate constants for substrate binding and/or product release. In summary, these data show that the rates of methylation with the human recombinant DNMT1 are accelerated by the free energy of supercoiling and that such increases are sensitive to DNA sequence.
Analysis of the Dissociation Constants for AdoMet and DNA-The dissociation constants for AdoMet and DNA were obtained from the slope replots (see "Experimental Procedures"). These equilibria (K i(app) AdoMet and K i(app) DNA ) represent apparent values that contain an additional factor, i.e. the ratio between the free enzyme and the enzyme bound to DNA at the allosteric site, K ia(app) ϭ K ia (1 ϩ [E]/ [DE]). If the binding affinity of the DNA for the regulatory site is high, the term in parentheses contributes little, whereas if binding is weak the value increases significantly.  [E]/[DE]) since the regulatory site is no longer present. Hence, the range from 2 to 4 M K i AdoMet measured for poly(dI-dC)⅐poly(dI-dC), (CGG⅐CCG) 12 , and (CGG⅐Cm 5 CG) 12 should also represent the value for DNMT1. The values for K i DNA ranged between 0.1 and 1 M for the same DNA substrates. Contrary to K i AdoMet , the dissociation constant for the DNA is not expected to have a constant value since sequences flanking the CG recognition step may contribute to the binding affinity. Thus, the 10-fold variation between (CGG⅐CCG) 12 and (CGG⅐Cm 5 CG) 12 shows that hemimethylated (CGG⅐Cm 5 CG) 12 has a greater affinity for the catalytic center than unmethylated (CGG⅐CCG) 12 , as expected (19).
The dissociation constants for both AdoMet and DNA with DNMT1 were lowest with poly(dI-dC)⅐poly(dI-dC). These values were similar to those obtained with DNMT1 ⌬501 , indicating that the (1 ϩ [E]/[DE]) term did not contribute substantially. Because these data show that this DNA has a strong affinity for the catalytic site, the [DE] complex may represent DNMT1 associated with DNA at the regulatory or at the catalytic site.

The K i(app)
AdoMet values for the unmethylated (CGG⅐CCG) 12 and the SNRPN duplex DNA with DNMT1 were substantially higher than for the corresponding hemimethylated substrates (m 5 CGG⅐CCG) 12 , (CGG⅐Cm 5 CG) 12 , SNRPN-methylated upper strand, and SNRPN-methylated lower strand. This result indicates that the (1 ϩ [E]/[DE]) term contributes significantly only for unmethylated DNA, confirming the data from inhibition studies (29) that hemimethylation increases the affinity of the DNA for the allosteric site. The same conclusion was obtained from the comparison of the K i(app) AdoMet values for (CGG⅐CCG) 12 and (CGG⅐Cm 5 CG) 12 between DNMT1 ⌬501 and DNMT1, which shows an 8-fold increase for the unmethylated FMR1 duplex, but identical values for the hemimethylated (CGG⅐Cm 5 CG) 12 .
The value of K i(app) AdoMet decreased 5-fold for the supercoiled pRW3602 compared with the relaxed circular form showing that the free energy of supercoiling increases the affinity of the DNA for the enzyme. The low value of K i(app) AdoMet for the linear pWR3602 was unexpected, and it is unclear what contributes to the difference between relaxed circular and linear pRW3602; obviously, these two substrates differ for the presence of free ends in the linear molecule.
The range of K i(app) DNA values for DNMT1 varied by 140-fold. These variations likely reflect differences in the affinities of each DNA molecule for the catalytic and the regulatory sites and the competition for binding at both locations. The 10-fold decrease in K i DNA for (CGG⅐Cm 5 CG) 12 ) relative to (CGG⅐CCG) 12 with DNMT1 ⌬501 indicates that methylation increased the affinity of the DNA for the catalytic site. Comparison of the increases in K i(app) DNA obtained with DNMT1 relative to DNMT1 ⌬501 for (poly(dI-dC)⅐poly(dI-dC), (CGG⅐CCG) 12 , and (CGG⅐Cm 5 CG) 12

) with those for K i(app)
AdoMet indicates that neither K i(app) AdoMet nor K i(app) DNA increased for poly(dI-dC)⅐poly(dI-dC) and (CGG⅐Cm 5 CG) 12 . For (CGG⅐CCG) 12

, the K i(app)
AdoMet value increased about 8-fold with DNMT1; however, K i(app) DNA only increased 3-fold. This difference suggests that if a duplex DNA has a low affinity for the regulatory site but a high affinity for the catalytic site, then this DNA may bind to the catalytic site prior to or without binding to the regulatory site. This would then decrease the contribution of the apparent term to K i(app)

DNA but not to K i(app)
AdoMet . Thus, it is possible that three pathways lead to the catalytically competent central complex as follows: one in which DNA binds to the regulatory site followed by AdoMet and then DNA to the catalytic site; a second in which DNA binds to the regulatory site followed by DNA and AdoMet to the catalytic site; and a third in which allosteric binding is bypassed and DNA binds to the catalytic center followed by AdoMet (Fig. 1).
A comparison of the dissociation constants for the (CGG⅐CCG) 12 and (CGG⅐CCG) 73 DNA substrates reveals that although the K i(app) AdoMet values did not change, K i(app) DNA decreased about 10-fold for the longer molecule. Comparison of the steady-state kinetic parameters for the (CGG⅐CCG) 12 and (CGG⅐CCG) 73 (Table V) indicates that k cat values doubled for (CGG⅐CCG) 73 and that the k cat /K m DNA ratio increased five times. Together with the results from the topoisomer distribution, these results show that both length and negative supercoiling contribute to increasing the methylation rate of the CGG⅐CCG triplet repeat sequence involved in the fragile X syndrome.
Finally, the K i(app) DNA values for supercoiled, relaxed closed, and linear pRW3602 were the highest of all the substrates evaluated. Together with the K i(app) AdoMet values, these results indicate that plasmid DNAs had a weak affinity for both the regulatory and catalytic sites.
AdoMet Binding Is Preferred as the First Substrate-The results obtained with DNMT1 and DNMT1 ⌬501 indicate that the turnover rates were not limited by the chemical steps of the methyl transfer reaction but rather by substrate binding or product release. Therefore, if one of the three pathways shown in Fig. 1 dominates over the other two, a correlation may unfold between the k cat and K ia (app) values. Fig. 8 shows the relationship between k cat and K i (app) DNA (open circles) or between k cat and K i(app) AdoMet (filled circles) for the 11 DNA templates shown in Table  IV with DNMT1. A significant correlation (r 2 ϭ 0.862) was found between log k cat and log K i(app) AdoMet , as shown by the 95% confidence interval, whereas there was only poor correlation (r 2  12 33 3 (CGG ⅐ Cm 5 CG) 12 2 0.2 (m 5 CGG ⅐ CCG) 12 11 1 (CGG ⅐ CCG) 73 12 4 1 (CGG ⅐ Cm 5 CG) 12 2 0.1 ϭ 0.250) between the log k cat and log K i(app) DNA . This result indicates that, subsequent to DNA binding to the regulatory site, AdoMet binding to the catalytic site followed by DNA was the most common pathway. DISCUSSION Steady-state kinetics revealed that the methyl transfer reaction with the recombinant human DNMT1 (19,29) involves control mechanisms that act through DNA sequence, methyl groups, and DNA topology and that these properties are consistent with the maintenance methylation role of the enzyme in development.
Allosteric Repression and Activation-The N-terminal region comprising the first 501 amino acids of DNMT1 inhibits enzymatic activity. This region also binds methylated DNA, which partially relieves the inhibition. Therefore, methylated DNA is both an inhibitor and an allosteric activator for the methyl transfer reaction. Previous inhibition studies (29) indicated that two DNA molecules may bind simultaneously to DNMT1, a methylated substrate at the regulatory site and an unmethylated DNA at the catalytic site, suggesting that binding at the regulatory site exposes the catalytic center, otherwise occluded, to the AdoMet and DNA substrates. Because the N-terminal region is also known to bind unmethylated DNA (39), activation requires the presence of methyl groups.
The protein domain involved in allosteric regulation also mediates the interaction of DNMT1 with other cellular factors (Fig. 2) responsible for targeting the enzyme to the nucleus (2) and to replication foci 2 (40 -42). This domain is also involved in transcriptional repression. Therefore, because of the dual properties of catalytic repression and protein binding, the "maintenance methylation" role of DNMT1, and therefore the silencing of its de novo activity in vivo, may be achieved by stable inhibition of the catalytic center through the interaction of the N-terminal domain with binding proteins. This interaction may either interfere with DNA binding at the regulatory site or suppress its activation. The loss of DNMT1 activity following truncations in the N-terminal domain (43,44) or the binding with specific antibodies (2) suggests that the catalytic center is in proximity to N-terminal regions and that its function is sensitive to their folding and/or dynamic properties.
Reaction Order-Our inhibition studies show that the reaction order is random. However, because the turnover numbers correlate with the apparent dissociation constants for AdoMet, binding of AdoMet as the first substrate occurs more frequently. DNA substrates with high affinity for the catalytic center, such as poly(dI-dC)⅐poly(dI-dC) and (CGG⅐Cm 5 CG) 12 , may bind first without necessitating prior allosteric binding (18).
Preference for Hemimethylated DNA-It has long been recognized that hemimethylated DNA serves as a better substrate than unmethylated DNA for mammalian Dnmt1 (16 -18, 45). We find that this property is independent of the presence of the first 501 N-terminal amino acids, which rules out a role for the allosteric control region. Since the velocity patterns for poly(dI-dC)⅐poly(dI-dC) and (CGG⅐Cm 5 CG) 12 were curved and not intersecting with DNMT1, whereas they were linear and intersecting with DNMT1 ⌬501 , we propose that the preference for hemimethylated DNA is an intrinsic property of the catalytic center.
Role of Negative Supercoiling, Triplet Repeat Length, and Hemimethylation in Methylation of the Fragile X (CGG⅐CCG) n Sequence-The methylation rates increased with increasing negative supercoiling up to 2-fold for DNA of random sequence and up to 7-fold for the (CGG⅐CCG) 73 triplet repeat, whereas the turnover number was higher (10 h Ϫ1 ) for the supercoiled form of the plasmid than for the relaxed and linear forms. These data clearly show that DNA topology plays a significant role in methylation. The finding is particularly relevant to the fragile X syndrome because the (CGG⅐CCG) n repeat has a low persistence length (34) that enables higher levels of superhelical stress to be present in the repeat than in the flanking sequences (37). Negative supercoiling is known to unwind and destabilize duplex DNA and promote alternative DNA structures (38,46), which for (CGG⅐CCG) n may consist of hairpin loops (47). Such hairpin loops are preferred substrates for DNMT1 (48).
The turnover numbers and specificity constants (k cat /K m DNA ) also increased when the number of (CGG⅐CCG) repeats increased, whereas the apparent dissociation constant decreased. Our interpretation is that, due to the high affinity of the repeat for the catalytic pocket, DNMT1 will "stick" in the vicinity of (CGG⅐CCG) n molecules by non-elastic collisions (49) and will associate with molecules with more triplet repeats for longer times than with molecules with fewer repeats.
We also show that an m 5 CG step stimulates methylation both at the complementary unmethylated CG dinucleotide as well as at flanking CG steps. Therefore, sporadic methylation in an otherwise unmethylated (CGG⅐CCG) n tract may act as a catalyst for further methylation and thereby elicit a cooperative methylation process.
Model for Involvement of DNMT1 in Fragile X Syndrome-  Table IV for DNMT1 were plotted versus the log of their respective k cat values (19). The shaded area shows the 95% confidence interval for the correlation with the K i(app) AdoMet data. Although the mechanism for the methylation of (CGG⅐CCG) n tracts that lead to the fragile X syndrome is not understood, two consequences of large repeat expansions are consistent with the involvement of DNMT1. First, a delay in replication was observed. The timing of DNA replication for the FMR1 locus and its flanking sequences (ϳ1 megabase pair total) is lengthened in fragile X males (50,51), in concert with transcriptional silencing (52). A delay in replication is also observed in chromosomes of carrier individuals (53)(54)(55) where the CGG⅐CCG tract is only moderately expanded and methylation is not observed (22). This suggests that repeat expansion leads to delay in replication irrespective of its methylation status. The compact, higher order structure of the transcriptionally silent heterochromatin requires the interactions between heavily methylated DNA (56), methylated DNA-binding proteins (57,58), and deacetylated histones (59,60). It is possible that the late replication-specific interaction of DNMT1 with deacetylated histones (40,42,61) activates the enzyme activity to sustain the heavy methylation patterns. Second, DNA structure and topology may play a critical role in the methylation process. Transient surges in negative supercoiling may arise during replication due to the disassembly of nucleosomes (62)(63)(64) and the helicase activity of the polymerase complex (65). An expanded CGG⅐CCG tract may form hairpin structures (66) both under the influence of negative supercoiling and the strand separation that accompanies replication (67,68). Therefore, negative supercoiling, hairpins, and the high concentration of CG steps may trigger sporadic methylation events that will catalyze further methylation in the following rounds of replication, eventually leading to extensive methylation.