Formation of nucleoprotein filaments by mammalian DNA methyltransferase Dnmt3a in complex with regulator Dnmt3L

Analytical ultracentrifugation

Analytical ultracentrifugation of the co-purified Dnmt3a-C/3L-C complex was done with an An50-Ti 8-place rotor in a Beckman-Coulter model XL-A centrifuge equipped with UV absorption optics. Sedimentation velocity experiments were run at speeds of 50 000 r.p.m. at 4°C in 20 mM Tris–HCl pH 8.0, 0.3 M KCl, 0.67 M glycerol, 16 mM β-mercaptoethanol (for details see Supplementary Text 2).

Mutagenesis, protein expression and protein purification

The sequence encoding C-terminal domains of the mouse Dnmt3a (residues 608–908) and Dnmt3L (residues 208–421) were subcloned as N-terminal 6-His fusion proteins into pET-28a vector (Novagen) (3,11). Selected Dnmt3a-C and Dnmt3L-C variants were generated using the megaprimer site-directed mutagenesis method as described (21). Introduction of the mutations was confirmed by restriction marker site analysis and DNA sequencing. Protein expression and purification was carried out as described (11,19) (for details see Supplementary Text 2).

Methylation kinetics

Time courses of DNA methylation were measured by the incorporation of titrated methyl groups from labeled S-[methyl-3H] adenosyl-methionine (Perkin Elmer) into biotinylated oligonucleotides (Bt-GAG AAG CTG GGA CTT CCG GGA GGA GAG TGC/GCA CTC TCC TCC CGG AAG TCC CAG CTT CTC) using the avidin-biotin methylation kinetic assay as described (22). The reactions were carried out in the methylation buffer [20 mM HEPES pH 7.2, 1 mM EDTA, 50 mM KCl, 25 μg/ml bovine serum albumin (BSA)] at 37°C, using 1 μM substrate DNA, 0.76 μM AdoMet and 2.5 μM Dnmt3a-C wild-type (wt) or mutant enzyme, 2.5 μM Dnmt3L-C wt or mutant protein or 1.25 μM co-purified tetramer. To determine relative catalytic activities of the wt and mutants, equal concentrations of enzymes were used in the reactions. To determine the initial slope, the data were fitted by linear regression of the initial part of the reaction progress curves. All kinetic analyses were carried our at least in triplicate.

AdoMet binding

The cofactor binding was analyzed by UV-crosslinking experiments basically as described (11) using 6 μM wt Dnmt3L-C, 3 μM Dnmt3a-C (wt or mutant) and 1.4 μM [methyl-3H]AdoMet in methylation buffer supplemented with 10% glycerol and 1% β-mercaptoethanol. An excess of Dnmt3L-C was used to overcome the potentially weakened binding of Dnmt3L-C to the Dnmt3a-C variants, because it was the aim of this experiment to determine the effect of the Dnmt3L-C-induced conformational change of Dnmt3a-C on AdoMet binding (for details see Supplementary Text 2).

Electrophoretic mobility shift assay (EMSA)

DNA binding by Dnmt3a-C wt and mutants (1–10 μM) after preincubation with equimolar amounts of Dnmt3L-C was investigated by EMSA using a 146-bp DNA (30 nM) in reaction buffer (20 nM HEPES pH 7.5, 1 mM EDTA, 100 mM KCl, 0.2 mM sinefungin, 0.5 mg/ml BSA) basically as described (19). Sinefungin is an analogon of AdoMet that does not allow DNA methylation but supports DNA binding of Dnmt3a-C (data not shown). The gel was then scanned with a phosphoimager system (Fuji). For quantitative analysis the intensities of the gel regions corresponding to free DNA, fully occupied DNA and binding intermediates were determined at different protein concentrations and fitted to noncooperative and to cooperative binding models (for details see Supplementary Text 2).

Active site footprinting by hairpin bisulfite methylation analysis

DNA methylation by the Dnmt3a-C/3L-C co-purified complex was investigated by a hairpin bisulfite approach (23). Two sets of oligonucleotides were used as substrate containing 10 or 13 GAC repeats which form 9 or 12 CG sites in identical sequence context, respectively. 9xCG substrate: d(AAT TGA CGA CGA CGA CGA CGA CGA CGA CGA CGAC)/d(GAT CGT CGT CGT CGT CGT CGT CGT CGT CGT CGTC); 12xCG substrate d(AAT TGA CGA CGA CGA CGA CGA CGA CGA CGA CGA CGA CGA CGA C)/d(GAT CGT CGT CGT CGT CGT CGT CGT CGT CGT CGT CGT CGT CGTC. The annealed oligonucleotides (100 nM) were incubated with 2.5 μM Dnmt3a-C/3L-C complex for 2 h in methylation buffer supplemented by 0.32 mM AdoMet. The methylation reaction was stopped by freezing the sample in liquid nitrogen, followed by proteinase K digestion (New England Biolabs). Afterwards, the hairpin loop and the adaptors were ligated to the methylated substrates, DNA was subjected to bisulfite conversion essentially as described (24,25), cloned into Topo-TA vector (Invitrogen) and individual clones were sequenced.

SFM

SFM experiments were carried out using preincubated Dnmt3a-C/3L-C complex and a 509-bp DNA fragment, which can be visualized easily in SFM. The DNA was derived from the CG island upstream of the human SUHW1 gene and contains 58 CG sites roughly equally distributed over the entire DNA length. DNA–protein filaments were generated in a total volume of 30 μl by combining 12 nM DNA with 100 nM Dnmt3a-C/3L-C tetramer in 50 mM HEPES (pH 7.5), 250 mM NaCl, 1 mM EDTA and 100 μM of sinefungin. The two protein components (200 nM each) where preincubated in the binding buffer for 30 min at room temperature. After addition of DNA, samples were incubated for additional 30 min at room temperature to allow for DNA binding. The complex solution (1 μl) was mixed with 9 μl of 10 mM MgCl₂ solution and deposited on freshly cleaved mica (Plano, Wetzlar), allowed to adhere for 40 s and then washed with 1 ml of bi-distilled water. The sample was then dried using compressed air. Protein–DNA filaments where observed by tapping mode in air using a Multimode SFM with a Nanoscope III controller (Digital instruments, Santa Barbara, CA). We used NSTNCHF silicon cantilevers (Nascatec, Stuttgart) with a nominal spring constant of 50 N/m and a resonance frequency of ∼150 kHz. All images were obtained with a scanning speed of 0.75–1 Hz and a resolution of 512 × 512 pixels. To remove background slope, raw images were flattened using the Nanoscope software. DNA–protein complexes were regarded as a filament if the height exceeded 150% of the height observed for DNA molecule alone and were at least 20 nm wide. Filaments were evaluated using the section tool of the Nanoscope V6r12 software.

Compaction of the DNA in nucleoprotein filaments was calculated using Equation (1):

1

With C, compaction factor, L, length of the filament; L₀, length of the free DNA (170 nm); f: fraction of the molecule covered by protein.

Oligomeric state of Dnmt3a-C/3L-C

We investigated the oligomeric state of the Dnmt3a-C/3L-C complex in solution by analytical ultracentrifugation. The co-purified Dnmt3a-C/3L-C complex sedimented in a single boundary with s_20°C,water = 5.4 S (Figure 1C). SDS gel electrophoresis showed the complex to contain both proteins in a 1:1 stoichiometric ratio (Figure 1D). The frictional ratios calculated from the measured sedimentation rate constant would be 1.5 for a 2:2 and 0.93 for a 1:1 Dnmt3a-C/3L-C hydrated complex. Since a ratio < 1.0 is not possible, the measured sedimentation rate is too fast for a 1:1 heterodimer and is only compatible with a 2:2 heterotetramer.

Catalytic activity of Dnmt3a-C/3L-C and its interface mutants

In order to study the functional importance of Dnmt3a–3a interaction at the RD interface, we mutated R881 to alanine (R881A) in Dnmt3a-C. At the Dnmt3a–3L interface (the FF interface), we exchanged F728 in Dnmt3a-C to alanine (F728A) and to aspartic acid (F728D), assuming that the F to D exchange, which introduces a charged residue instead of a hydrophobic one, would result in a stronger phenotype than the F to A mutation. We also made the corresponding mutations in Dnmt3L-C at the FF interface (F297A and F297D).

We compared the catalytic activity of wt Dnmt3a-C and its variants using roughly equal concentrations of the enzymes and an unmethylated oligonucleotide substrate with one centrally placed CG site. All the Dnmt3a-C variants turned out to be catalytically inactive (Figure 2A and Supplementary Figure 1). To ensure that the lack of activity was not caused by protein misfolding, we determined the near UV circular dichroism (CD) spectra of the Dnmt3a-C wt and its mutants. As shown in Supplementary Figure 2, the wt enzyme and R881A mutant CD spectra superimpose, indicating that the mutant protein is folded as the wt protein. The CD spectra of the F728A and F728D mutants also demonstrate that the proteins are folded. However, both showed a small, but reproducible shift in the shorter wavelength region indicating a change in the secondary structure composition.

Next, we compared the activity of the wt Dnmt3a-C/3L-C complex and its variants. Since the co-purified Dnmt3a-C/3L-C complex showed similar activity to the complex formed by preincubation of equimolar amounts of purified Dnmt3a-C and Dnmt3L-C for 30 min (Figure 2A), in all subsequent experiments the complexes formed by preincubation of both proteins were used. After incubation with Dnmt3L-C, the activity of F728A reached 70% of the level observed for the wt complex, but F728D showed almost no activity (Figure 2A). The R881A variant was completely inactive (Figure 2A) even after incubation with excess of Dnmt3L-C (data not shown). Mutations at the FF interface in Dnmt3L-C (F297A and F297D) reduced the activity of Dnmt3a-C/3L-C by 50% and 75%, respectively (Figure 2A). In addition, the activity of Dnmt3a-C F728A variant was nearly abolished when incubated with the F297A mutants of Dnmt3L-C (Supplementary Figure 1B), because two exchanges at the FF interface (one in Dnmt3a-C and one in Dnmt3L-C) disturb the interface more strongly or prevent complex formation. Hence all the three interface residues (R881 and F728 of Dnmt3a and F297 of Dnmt3L)—though located distant from the catalytic center of the enzyme—are important for the catalytic function of Dnmt3a-C and Dnmt3a-C/3L-C complexes.

AdoMet binding by Dnmt3a-C and its interface mutants

AdoMet binding to Dnmt3a-C can be detected in a UV crosslinking assay (11), whereas Dnmt3L-C does not bind to AdoMet (11,15,19). In the absence of Dnmt3L-C, Dnmt3a-C has reduced AdoMet binding (Figure 2B) (11,15), and its three mutants showed a strong reduction in AdoMet binding in comparison to the wt protein. When preincubated with Dnmt3L-C, the AdoMet binding of the Dnmt3a-C F728A mutant was efficiently rescued, but little rescue was seen for F728D (Figure 2B). This result indicates that binding of Dnmt3L-C to the Dnmt3a-C F728A variant can recover its AdoMet-binding ability, presumably by stabilizing the FF interface, but the interaction is weak or lost between Dnmt3L-C and F728D variant. Addition of Dnmt3L-C only partially rescued the AdoMet binding by the Dnmt3a-C R881A, indicating that the mutational defect in the RD interface cannot be overcome by binding of Dnmt3L at the FF interface. Reduction of AdoMet binding by disruption of either of the two interfaces can be explained structurally by interaction networks which connect residues important for AdoMet binding with interface structural elements [Supplementary Figure 3 and (19)].

DNA binding by Dnmt3a-C/3L-C and its interface mutants

Using EMSAs, we have previously shown that Dnmt3a-C and Dnmt3a-C/3L-C complexes multimerize on the DNA, and that Dnmt3L-C binds very weakly to DNA (19). The binding process shows the co-existence of DNA fully occupied by protein (band 1 in Figure 3A) and free DNA with low amounts of incompletely occupied complexes over a range of protein concentrations (Figure 3A). We determined the amount of the free DNA, as well as DNA partially and fully occupied by protein by densitometry. Fitting of the binding profile was not possible with a non-cooperative binding model suggesting a cooperative binding mechanism (Figure 3B and C).

To study the role of the RD and FF interfaces in DNA binding, we performed gel retardation experiments with the interface variants as well. After preincubation with Dnmt3L-C, the F728A variant bound DNA slightly better than the wt complex (Figure 3A), whereas Dnmt3a-C F728D/3L-C show increased amount of smear and incompletely occupied complexes (Figure 3A). In contrast, the R881A mutant only weakly bound DNA and formed smaller complexes, as shown by an appearance of a ladder of bands, corresponding to one, two or even more protein complexes bound to the DNA (Figure 3A, bands 2 and 3). These results demonstrate that the RD interface is required for DNA binding and the FF interface is supportive but not essential.

Protein–DNA filament formation by Dnmt3a-C/3L-C complex

Dnmt3a-C/3L-C protein–DNA complexes were studied by SFM experiments using a CG rich substrate. DNA bound by protein could be distinguished from free DNA by the change in height and width (Figure 4A). The results showed that Dnmt3a-C/3L-C complexes polymerize on the DNA forming a nucleoprotein filament (Figure 4B–E). We observed many complexes in which the DNA was fully occupied by protein, although the protein was used in substoichiometric amounts with respect to binding sites (100 nM tetramer, 12-nM DNA corresponding to ∼600 nM binding sites, when assuming that one tetramer binds to ∼10 bp, as suggested by the hairpin bisulfite experiments described below). Similarly as in the gel retardation assay, we observed the co-existence of free DNA and DNA covered completely with Dnmt3a-C/3L-C. Furthermore, multimeric binding of complexes was predominantly observed rather than binding of single tetramers. This is in contrast to results obtained with the EcoRV MTase where mostly single complexes were found using the same technique (26). These observations suggest cooperative binding of Dnmt3a-C/3L-C to DNA.

For each DNA molecule the length of the nucleoprotein complex region and the length of the unbound region were determined and the factor of shortening of the DNA length calculated (Figure 4F). Naked DNA (194 molecules) had a length of 171 ± 18 nm (standard deviation) (Supplementary Figure 4). The filaments fully occupied with protein (33 examples) were 117 ± 11 nm in length, which corresponds to a shortening of the DNA length by ∼1.5-fold (average 1.47). For 116 examples, combining all partially occupied filaments with more than 20% protein coverage and all fully occupied filaments, the shortening factor shows a distribution that peaks at 1.5 as well (Figure 4F).

Product pattern of methylated DNA generated by Dnmt3a-C/3L-C

To determine the binding of several Dnmt3a-C/3L-C tetramers to the DNA and map the positions of the individual active sites with respect to the DNA, we used two substrates having either 9 or 12 CG sites next to each other. Methylation was investigated by the hairpin bisulfite approach that allows analyzing the methylation of both DNA strands in a single substrate molecule. The distribution of the methylation marks in the 9xCG and 12xCG substrates (with 5′ overhangs of four nucleotides) was investigated after sequencing of 260 and 169 clones, respectively. The results showed a pronounced, nonrandom distribution of methylation events (Figure 5). To assess the statistical significance of the preferential methylation at particular sites, P-values were calculated assuming a binomial distribution of the methylation events (Supplementary Text 1). For the 9xCG substrate we observed statistically significant preferential methylation at CG2 and CG3 in the upper strand, CG5 in both strands and CG7 and CG8 in the lower strand (Figure 5A). Similarly, the 12xCG substrate was preferentially methylated to a significant degree at CG2 and CG3, CG5 and CG6 and CG8 in the upper strand and CG5, CG8, CG10 and CG11 in the lower strand (Figure 5B). Hence, the multimeric complex positions its active sites such that it methylates the DNA in distances of 8–10 bp in both strands of the DNA. Both substrates show 3–4 CG sites at the 3′ parts of each strand that were hardly methylated. This suggests binding of the tetramer to DNA in an orientation such that two CG sites are methylated in a distance of 8–10 bp with both strands being methylated towards the 5′ end.