The Replication Focus Targeting Sequence (RFTS) Domain Is a DNA-competitive Inhibitor of Dnmt1*

Dnmt1 (DNA methyltransferase 1) is the principal enzyme responsible for maintenance of cytosine methylation at CpG dinucleotides in the mammalian genome. The N-terminal replication focus targeting sequence (RFTS) domain of Dnmt1 has been implicated in subcellular localization, protein association, and catalytic function. However, progress in understanding its function has been limited by the lack of assays for and a structure of this domain. Here, we show that the naked DNA- and polynucleosome-binding activities of Dnmt1 are inhibited by the RFTS domain, which functions by virtue of binding the catalytic domain to the exclusion of DNA. Kinetic analysis with a fluorogenic DNA substrate established the RFTS domain as a 600-fold inhibitor of Dnmt1 enzymatic activity. The crystal structure of the RFTS domain reveals a novel fold and supports a mechanism in which an RFTS-targeted Dnmt1-binding protein, such as Uhrf1, may activate Dnmt1 for DNA binding.

DNA cytosine methylation, an epigenetic mark that occurs predominantly at CpG dinucleotides, is the principal eukaryotic DNA modification (1). DNA methylation is a key component of gene silencing, which makes significant contributions to cell phenotype. Establishment and maintenance of DNA methylation patterns are governed by three catalytically active DNA methyltransferases: Dnmt3a, Dnmt3b, and Dnmt1. The Dnmt3 isoforms are responsible for de novo methylation during germ cell and embryonic development and bind with high affinity to unmethylated DNA sequences (2). CpG methylation patterns are maintained in mammals by Dnmt1 with hemimethylated CpG dinucleotides serving as preferred substrates.
The CXXC domain is understood to contribute to catalytic activity by interacting with unmethylated CpG DNA substrates (6,8). This was observed in the recently solved crystal structures of Dnmt1, which encompass sequences from the CXXC domain to the C terminus (9). In the structures, the CXXC domain binds and holds unmethylated duplex CpG-containing DNA away from the active site, whereas the acidic linker between the CXXC and BAH1 domains is bound in the active site between the DNA segment and the S-adenosylhomocysteine product (9). This observation helps to explain the relationship between the CXXC and catalytic domains. However, it does not address the role of domains N-terminal to the CXXC domain, which are not present in the structures.
We set out to clarify the structure and function of the RFTS domain. The RFTS domain is conserved by sequence (supplemental Fig. S1) and contains the binding site for Uhrf1 (10), a Dnmt1-associated protein that recruits Dnmt1 to hemimethylated DNA (11)(12)(13)(14)(15). Despite the significance of the RFTS domain, progress in understanding its function has been limited by the availability of stable soluble protein fragments and robust DNA methyltransferase assays. Here, we generated soluble protein fragments of Dnmt1 and established activities for them. Strikingly, the binding of Dnmt1 to naked DNA oligonucleotides and native polynucleosomes was inhibited by the RFTS domain. Kinetic analysis established that Dnmt1 without the RFTS domain functioned with a K m of 1 nM for an internally quenched oligonucleotide substrate. This represents a Ͼ100fold binding advantage with respect to recent assays with hemimethylated oligonucleotide substrates (9,16

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Purification-Human Dnmt1 constructs schematized in Fig. 1 were cloned and purified with similar methods (17). Briefly, cDNA templates (MHS1768-98980929.pCR-XL-TOPO) from Open Biosystems were cloned into the pET28MHL or pNICCH vector using the In-Fusion CF Dry-Down PCR cloning kit (639605, Clontech), transformed into Escherichia coli BL21(DE3) cells, and grown in Terrific Broth in the presence of 50 g/ml kanamycin at 37°C. Selenomethionyl derivatives of the RFTS domain were expressed in M9 medium supplemented with glycerol using an M9 SeMet high yield growth media kit package (M 2 D045004 -50L, Medicilon) according to the manufacturer's instructions. After lysis, cell supernatants were subjected to metal affinity chromatography using TALON columns (BD Biosciences). Protein was further purified by gel filtration (HiLoad 16/60 Superdex 200 column, GE Healthcare) equilibrated with buffer A (20 mM Tris-HCl (pH 8.0), 0.5 M NaCl, 5% glycerol, and 2 mM DTT) and by ion exchange chromatography on a 5-ml HiTrap Q column using a gradient of buffer A to 50% buffer B (20 mM Tris-HCl (pH 8.0), 1 M NaCl, 5% glycerol, and 2 mM DTT). Full-length Dnmt1 was obtained from New England Biolabs.
Electrophoretic Mobility Shift Assays-DNA binding was analyzed using duplex DNA oligonucleotides of varying lengths, sequences, and methylation states (supplemental Table 1). Duplex DNA (100 ng/lane) was incubated overnight at 4°C with an appropriate concentration of protein (as indicated in the figure legends) in 10 mM Tris-HCl (pH 7.5), 150 mM NaCl, 0.5 mM DTT, and 0.1 mg/ml BSA and subjected to polyacrylamide gel electrophoresis (10% Tris borate/EDTA gels (Bio-Rad) run in 0.5ϫ Tris borate/EDTA at 100 V for 1.5 h on ice). Gels were stained with 10,000-fold diluted SYBR Green I (Invitrogen) for 5 min, washed twice with 0.5ϫ Tris borate/ EDTA, and imaged using an excitation wavelength of 633 nm.
Dnmt1 Activity Assay-Dnmt1 activity was measured using an assay related to that developed by Wood et al. (16) in which a hemimethylated, internally quenched, fluorescent hairpin DNA substrate is incubated with Dnmt1 and AdoMet in the presence of GlaI, which cleaves the fully methylated product. Assays (0.1 ml) were performed in triplicate in 96-well half-area black plates at 37°C. Triplicate reactions were prepared by making a solution of 10 mM Tris-HCl (pH 7.5), 5 mM MgCl 2 , 1 mM DTT, 100 mM potassium glutamate, 0.105 mg/ml BSA, 1.05 mM AdoMet (HPLC-purified; Sigma), and 1.05ϫ the desired concentration of oligonucleotide. For inhibition studies, 1.05ϫ the desired concentration of the RFTS domain (amino acids 351-600) was added to the solution. AdoMet was added freshly to limit conversion to S-adenosylhomocysteine, which inhibits the Dnmt1 reaction. Ninety-five microliters of this solution were added per well and brought to 37°C. A second solution was made containing Dnmt1 and GlaI (SibEnzyme, Novosibirsk, Russia) in 10 mM Tris-HCl (pH 7.5), 5 mM MgCl 2 , 1 mM DTT, and 100 mM potassium glutamate such that 0.8 units of GlaI would be added to each well. Coupled reactions were initiated by adding 5 l of enzyme solution, and the wells were covered with 5 l of mineral oil. Data were collected in a Wallac VICTOR 2 using excitation and emission wavelengths of 485 and 535 nm, respectively. A negative control containing GlaI without Dnmt1 was subtracted from each assay condition to correct for both substrate fluorescence and fluorescence from slow GlaI cleavage of the hemimethylated substrate. A standard curve was created using the fully cleaved oligonucleotide products to convert arbitrary fluorescence units to concentration. The oligonucleotide, custom-synthesized at Integrated DNA Technologies (Coralville, IA), was 5Ј-FAM-CCTATGCGm-CATCAGTTTTCTGATGmCGmCATAGG-3Ј-Iowa Black, where mC denotes 5-methyldeoxycytidylate residues and underlined residues form the GlaI restriction site. Initial rate data were fit either to the Michaelis-Menten equation or to the quadratic velocity equation (Equation 1) for tight-binding substrates (19,20) Data were fitted using either KaleidaGraph (Synergy, Inc.) or Prism (GraphPad Software, Inc). Crystallization-Selenomethionyl derivatives of the RFTS domain (amino acids 351-600) were mixed with detergent Anapoe-80 solution (10%, v/v) in a 5:1 ratio (final protein and detergent concentrations of 16.5 mg/ml and 1.7%, respectively), and crystals were grown at 298 K using the hanging drop method by mixing 1 volume of this solution with 1 volume of well solution consisting of 23% polyethylene glycol 3350, 0.1 M bis-Tris (pH 6.0), 0.3 M sodium acetate, and 5 mM tris(2-carboxyethyl)phosphine. The crystals were cryoprotected by immersion in the well solution mixed 1:1 with 20% (w/v) sucrose, 4% (w/v) glucose, 18% (v/v) glycerol, and 18% (v/v) ethylene glycol and placed in liquid nitrogen.
Structure Determination-Diffraction data from a crystal of the selenomethionyl derivative of the Dnmt1 RFTS domain were collected at beamline 19-ID at the Advanced Photon Source. The data set, collected at the selenium peak wavelength, was integrated and scaled using the HKL2000 program suite (21). The structure was solved by single wavelength anomalous diffraction techniques using the program SOLVE/ RESOLVE (22). A second data set collected at beamline 23-ID-B was used for refinement of the structure to 2.3-Å resolution. Iterative model building using the graphics program Coot (23) and maximum likelihood and translation/liberation/screw refinement with the program REFMAC5 (24) led to the final model. REFMAC5 parameters were generated using the TLS Motion Determination (TLSMD) web server. Statistics of data collection, processing, and refinement are provided in Table 1.

RESULTS AND DISCUSSION
The RFTS Domain Inhibits DNA Binding Independently of DNA Sequence, Length, Methylation Status, or CpG Content-It has been proposed that the RFTS domain of Dnmt1 tunes the catalytic activity of the conserved DNA methyltransferase domain toward specific DNA substrates (6). To test this hypothesis, we titrated constructs containing the RFTS domain (amino acids 351-1616) or lacking the RFTS domain (amino acids 621-1616) against 12-bp hemimethylated DNA oligonucleotides and analyzed DNA binding using an EMSA ( Fig. 2A). Surprisingly, at a range of Dnmt1 concentrations from 0.5 to 10 M, the RFTS domain-lacking 621-1616 construct bound the DNA, but the RFTS domain-containing 351-1616 construct showed no detectable binding even at high protein concentrations ( Fig. 2A). Moreover, DNA binding to the RFTS domainlacking 621-1616 construct was inhibited by the RFTS domain (amino acids 351-600) added in trans (Fig. 2B). These data suggested that the RFTS domain might inhibit the ability of the Dnmt1 catalytic domain to bind particular naked DNA sequences. To determine whether the RFTS domain might be responsible for substrate selectivity and permit the catalytic domain to bind some but not all DNA sequences, we surveyed a series of different DNA substrates of various lengths, methylation states, CpG content, and sequences. As shown in supplemental Fig. S2, the 621-1616 construct had its electrophoretic mobility shifted by all DNA constructs, regardless of methylation status, length, or sequence, whereas the 351-1616 construct, containing the RFTS domain, showed impaired binding of all oligonucleotides investigated. Even non-CpG oligonucleotides were bound by the 621-1616 Dnmt1 construct in a manner that was inhibited in trans by the addition of the RFTS domain (supplemental Fig. S3).
Dnmt1 Binding to Polynucleosomes Is Inhibited by the RFTS Domain-The CXXC and catalytic domains of Dnmt1 are both known to possess DNA-binding activities (8,25,26). To deter-   CXXC domain, were unsuccessful, suggesting that CXXC sequences may be required for folding and/or stability of the catalytic domain. Because naked DNA bound very poorly to CXXC domain constructs, we purified native polynucleosomes, which consist of epigenetically modified histone octamers and segments of DNA in various methylation states. EMSAs indicated that the 351-1616 construct was inhibited with respect to the polynucleosome-binding ability of the 621-1616 construct (Fig. 3A). In addition, the CXXC domain possesses a polynucleosome-binding capacity (Fig.  3B, lanes 6 and 7), which was inhibited by the RFTS domain both in cis (B, lane 4; and C, lane 2) and in trans (C, lanes  4 -6). However, the binding to polynucleosomes by the 621-1616 construct, containing both the CXXC and catalytic domains, was not disrupted in trans by the addition of the RFTS-only domain (Fig. 3D). These data suggest that the polynucleosome-binding activity of the CXXC domain is additive with the DNA-binding activity of the catalytic domain and that simultaneous engagement of polynucleosomes by both domains renders the complex relatively resistant to exclusion of DNA by the RFTS domain.
The RFTS Domain Is a Potent Inhibitor of Dnmt1 Enzymatic Activity-Data from EMSAs suggested that the RFTS domain is an intrinsic inhibitor of DNA binding by Dnmt1 sequences C-terminal to the RFTS domain. To test this hypothesis, we synthesized an internally quenched, fluorescent hairpin DNA substrate (16). In this restriction enzyme-coupled assay, methylation of a hemimethylated CpG site forms a GlaI site and generates fluorescence in real time (Fig. 4A). Analysis of the RFTS domain-lacking protein construct (amino acids 621-1616) with 20 nM DNA substrate indicated striking fluorescence-generating activity (Fig. 4B). Initial attempts to determine steady-state kinetic parameters with this enzyme and oligonucleotide indicated that the K m was in the low nanomolar range. Because assays were conducted with 2 nM Dnmt1, there was a need to test for the complication that the enzyme may titrate a significant proportion of substrate from the free form to an E-S complex at low substrate concentrations (19). To test for this behavior, substrate-dependent initial rates were measured at enzyme concentrations of 1-5 nM. As the enzyme concentration increased, data deviated significantly from the hyperbolic behavior of Michaelis-Menten enzyme kinetics. However, at all enzyme concentrations examined, the data fit well to the tight-binding equation (19,20), giving essentially the same K m (1 nM) and k cat (ϳ0.1 min Ϫ1 ) values.
Remarkably, Dnmt1 activity could not be detected with 2 nM RFTS domain-containing construct (amino acids 351-1616) and 20 nM DNA substrate (Fig. 4C). This led us to assay the RFTS domain-containing construct at higher enzyme and DNA substrate concentrations to test the hypothesis that the RFTS domain inhibits DNA binding and not some other step such as binding AdoMet. As shown in Fig. 5A, the second-order rate constant with respect to the DNA substrate was 1.87 ϫ 10 6 M Ϫ1 s Ϫ1 for the RFTS domain-lacking construct. Inclusion of the RFTS domain increased the K m for DNA by ϳ15-fold while depressing k cat by ϳ40-fold ( Fig. 5A and supplemental Fig. S4). Similar kinetic constants were determined for the full-length protein. Thus, the RFTS domain functions as an ϳ600-fold inhibitor of binding and methyl group transfer to the DNA substrate. This finding is strikingly different from recently published results from a study that monitored transfer of a tritiated methyl group from AdoMet to DNA (9). In that study, the form of Dnmt1 with the first 645 amino acids deleted was only 2-fold more active than full-length Dnmt1 in k cat /K m (9). The greater activity in this study may be attributed to an improved assay and/or the greater activity of the Dnmt1 construct beginning at residue 621. Because full-length Dnmt1 has been assayed using an internally quenched fluorogenic substrate and found not to be saturated with the DNA substrate at 240 nM (16), we argue that the 621-1616 Dnmt1 construct is strongly derepressed for DNA binding and catalysis. To further examine the inhibitory function of the RFTS domain, we titrated the RFTS polypeptide in trans to the RFTS domain-lacking Dnmt1 construct (amino acids 621-1616) and determined kinetic parameters. Plotting the data according to the Lineweaver-Burk method (Fig. 5B) clearly indicated that inhibition was competitive with the DNA substrate. As the concentration of the RFTS domain was increased, the K m(app) was increased without alteration of V max . By nonlinear regression, the RFTS domain-dependent effects on K m(app) indicated a K i of 103 Ϯ 11 nM. Thus, the RFTS domain is a previously unrecognized endogenous inhibitor of DNA binding within Dnmt1.
Crystal Structure of the Dnmt1 RFTS Domain and Model of Autoinhibition-Having established discrete assays for the function of the RFTS domain, we aimed to determine whether the physical characteristics of the domain might account for inhibition of DNA binding. A selenomethionyl form of the RFTS domain crystallized in space group P2 1 and diffracted to 2.3-Å resolution. A refined model of the crystal structure is depicted in Fig. 6 (see Table 1 for refinement statistics). The RFTS domain consists of a zinc-binding motif, followed by a ␤-barrel and a helical bundle, which are closely associated. Using the secondary structure matching algorithm (27), the closest match to the ␤-barrel is the structure of the Src family kinase HCK Src homology 3 domain (Q score of 0.25; root mean square deviation of 2.36 Å for 46 equivalent C␣ atom pairs). The six-stranded ␤-barrel lobe of the RFTS domain extends from residues 401 to 502, capped on one end by an ␣-helix (residues 496 -499, ␣2) and on the other end by a helical bundle (residues 495-600). The nearest structural homolog of the helical bundle, as calculated by the DALI server (28), is the winged helix B lobe of the cullin protein Cdc53 (root mean square deviation of 1.7 Å across 68 backbone atoms). The role of the winged helix B lobe in these E3 ubiquitin ligases, which has recently been elucidated, involves regulating the enzyme via an autoinhibitory mechanism (29). The winged helix B lobe in the inhibited state binds to the RING (really interesting new gene) domain of the ligase, preventing the RING domain from reaching the substrate. Neddylation of the winged helix B relieves its interaction with the RING domain, allowing the RING domain to access substrate (29).
In recent crystal structures, the acidic linker between the CXXC and BAH1 domains associates with the DNA-binding site of the methyltransferase domain and prevents productive access of DNA to the active site, where S-adenosylhomocysteine is bound (Fig. 7A) (9). Deletion of the CXXC domain and linker region increased k cat /K m by 10-fold on unmethylated DNA substrates and increased activity on hemimethylated substrates by 20% (9). However, the 600fold effect of the RFTS domain suggests that, were it present, the RFTS inhibitory activity would predominate over the mild effect of the CXXC-BAH1 linker in excluding DNA from the active site.  We sought to develop a model for Dnmt1 autoinhibition that would account for the 600-fold inhibition by the RFTS domain and account for how a positive regulator of Dnmt1 might relieve this inhibition. As shown in Fig. 7B, the RFTS domain has a size and electrostatic properties compatible with occlusion of the DNA substrate-binding site of Dnmt1. Three acidic loops, each disordered in our crystal structure of the RFTS domain, are positioned on the side of the RFTS domain we predict to associate with three basic patches of the methyltransferase domain of Dnmt1. We postulate that Uhrf1, a multidomain E3 ubiquitin ligase that is required for maintenance of DNA methylation and Dnmt1 localization in embryonic stem cells (11,12), is the key mediator of Dnmt1 derepression. According to our model, Uhrf1 binding to the RFTS domain would release the inhibitory domain from the active site and allow DNA to bind productively. This model is attractive for two reasons. First, Uhrf1 has been reported to associate with Dnmt1 within the RFTS domain (10). Second, Uhrf1 is thought to function by binding hemimethylated DNA and recruiting Dnmt1 for methyl transfer to the non-modified cytosine of a hemimethylated site (13) or to nearby CpG dinucleotides (14,15). If Uhrf1 binding to the RFTS domain is required to relieve inhibition of DNA bind-ing, a mechanism is provided to activate Dnmt1 for processive methylation of hemimethylated regions of DNA identified by the essential activator Uhrf1.