Crystal structure of human nucleosome core particle containing enzymatically introduced CpG methylation

Cytosine methylation, predominantly of the CpG sequence in vertebrates, is one of the major epigenetic modifications crucially involved in the control of gene expression. Due to the difficulty of reconstituting site‐specifically methylated nucleosomal DNA at crystallization quality, most structural analyses of CpG methylation have been performed using chemically synthesized oligonucleotides, There has been just one recent study of nucleosome core particles (NCPs) reconstituted with nonpalindromic human satellite 2‐derived DNAs. Through the preparation of a 146‐bp palindromic α‐satellite‐based nucleosomal DNA containing four CpG dinucleotide sequences and its enzymatic methylation and restriction, we reconstituted a ‘symmetric’ human CpG‐methylated nucleosome core particle (NCP). We solved the crystal structures of the CpG‐methylated and unmodified NCPs at 2.6 and 3.0 Å resolution, respectively. We observed the electron densities of two methyl groups, among the eight 5‐methylcytosines introduced in the CpG‐fully methylated NCP. There were no obvious structural differences between the CpG‐methylated ‘symmetric NCP’ and the unmodified NCP. The preparation of a crystallization‐grade CpG‐methylated NCP provides a platform for the analysis of CpG‐methyl reader and eraser proteins.

Cytosine methylation, predominantly of the CpG sequence in vertebrates, is one of the major epigenetic modifications crucially involved in the control of gene expression. Due to the difficulty of reconstituting site-specifically methylated nucleosomal DNA at crystallization quality, most structural analyses of CpG methylation have been performed using chemically synthesized oligonucleotides, There has been just one recent study of nucleosome core particles (NCPs) reconstituted with nonpalindromic human satellite 2-derived DNAs. Through the preparation of a 146-bp palindromic a-satellite-based nucleosomal DNA containing four CpG dinucleotide sequences and its enzymatic methylation and restriction, we reconstituted a 'symmetric' human CpG-methylated nucleosome core particle (NCP). We solved the crystal structures of the CpG-methylated and unmodified NCPs at 2.6 and 3.0 A resolution, respectively. We observed the electron densities of two methyl groups, among the eight 5-methylcytosines introduced in the CpG-fully methylated NCP. There were no obvious structural differences between the CpG-methylated 'symmetric NCP' and the unmodified NCP. The preparation of a crystallization-grade CpGmethylated NCP provides a platform for the analysis of CpG-methyl reader and eraser proteins.
Eukaryotic DNA methylation is one of the key chromatin modifications that play crucial roles in the regulation of the epigenetic status [1]. Cytosine methylation at the C5 position (5-methylcytosine, 5mC) of CpG dinucleotides is common in eukaryotes. In vertebrates, methylation only occurs at this site. In the human genome, the CpG dinucleotides are clustered in short CpG-rich DNA stretches called 'CpG islands' and regions with repetitive DNA sequences, such as pericentromeric satellite regions [2,3]. The CpG islands exist in over half of the human gene promoters [4] and are usually kept unmethylated during development and differentiation [2], whereas the methylation of the CpG islands generally leads to long-term gene silencing. CpG island methylation is involved in a variety of biological phenomena, including X chromosome inactivation, genomic imprinting, genome stability, and embryogenesis [2,3]. Furthermore, aberrant CpG methylation is a characteristic observed in several human diseases, such as cancers [5], atherosclerosis [6], and schizophrenia [7]. The CpG methylation status is dynamically modulated during development, through a combination of active modification by cytosine methyltransferases and 5mC oxidases and passive dilution [8]. The methylated CpG sites (i.e., 5 0 -mCG-3 0 /5 0 -mCG-3 0 where mC indicates 5mC) are typically recognized by MBD (methyl CpG-binding domain)-containing proteins [9]. The MBD proteins recruit additional proteins, such as histone deacetylases and chromatin remodeling factors, thereby leading to transcriptionally repressed chromatin. This connection between the DNA methylation state and the chromatin status is critically involved in the regulation of several diseases. As an example, the loss of MeCP2, one of the MBD proteins, is reportedly correlated with Rett syndrome [10,11]. In addition, MBD2 is known to mediate the transcriptional silencing of hypermethylated genes in cancer [12].
Biochemical CpG methylation, for the reconstitution of a CpG-methylated nucleosome, has been performed, using one of the de novo CpG methyltransferases from Spiroplasma, M.SssI [13,14]. For example, Choy et al. [15] reported the CpG methylation of 5S rDNA in a reconstituted nucleosome by M.SssI. In addition, a nucleosome containing hemimethyl CpG was reconstituted, using nucleosome array DNA methylated by M.SssI [16]. Furthermore, Osakabe et al. [17] recently reconstituted nucleosome core particles (NCPs) containing M.SssI-methylated CpG dinucleotide sites, using a 160-bp human satellite 2 DNA.
Although a variety of NCP structures have been solved thus far, the tertiary structure of the NCP containing CpG-methylated DNA still remains poorly understood. This is probably due to the difficulty in reconstituting a crystallization quality CpG-methylated NCP in a milligram-scale quantity. Recently, the crystal structures of CpG-methylated human satellite 2 NCPs were reported [17]. The structures of two NCPs containing nonpalindromic DNAs with six CpG dinucleotides, derived from a 160-bp human satellite 2 DNA, were solved. In the report, the overall structures of the CpG-methylated and unmodified NCPs, both composed of one of the satellite 2-derived DNAs, are essentially the same, and no structural effect of the CpG methylation on the NCP DNA was observed. As the NCPs composed of nonpalindromic DNAs might have been packed in opposite orientations in the crystals, the structure determination was not simple. In contrast, the structure determination is expected to be straightforward with a 'symmetric' CpG-methylated NCP with a palindromic DNA. Therefore, in this study, we reconstituted human NCPs with a 146-bp palindromic a-satellite-based DNA containing four methylated CpG sites, and examined whether a different CpG methylation pattern in a different DNA sequence of the NCP affects the nucleosomal structure. The CpG sites were enzymatically fully methylated, as examined by digestions using the methylation-sensitive restriction endonuclease Eco72I and the methylationdependent restriction endonuclease MspJ1. We solved the crystal structure of the 'symmetric' CpG-fully methylated NCP at 2. 6 A resolution, and compared it with that of the unmodified NCP solved at 3.0 A. We confirmed that the CpG full methylation does not essentially affect the nucleosome structure, regardless of the base sequence of the DNA.

Results
Design for the preparation of CpG-methylated nucleosomal DNA To solve the crystal structure of a CpG-methylated NCP, we first designed the construction of the 146-bp a-satellite-based nucleosomal DNA (NCP146; [18]) containing CpG dinucleotide sequences (Fig. 1A), because the a-satellite sequence has a strong tendency for nuclesome formation, and is palindromic. Therefore, we regard it as one of the best nucleosomal DNAs for an X-ray crystallographic analysis to examine whether the CpG methylation affects the nucleosomal structure. Since there is no CpG dinucleotide sequence in the original NCP146, we decided to introduce a couple of CpG dinucleotides into NCP146, at SHLs À1 and À6 in the half unit of the palindromic NCP146 sequence. For the positions, we selected one from a SHL near the DNA end (i.e., À1), and the other from a SHL near the axis (i.e., À6), to investigate the effects of CpG methylation on the NCP structure at different SHLs. Accordingly, the designed nucleosomal DNA, designated as CpG146, has four CpG dinucleotide sequences at around SHLs À6, À1, +1, and +6 (Fig. 1A,B).
To create the CpG sequence in CpG146, we introduced the CACGTG sequence, so the unmethylated nucleosomal DNA could be digested by the methylation-sensitive restriction endonuclease Eco72I, after CpG methylation by the de novo CpG methyltransferase M.SssI. This digestion method would allow us to purify the CpG146 nucleosomal DNA, in which four CpG dinucleotides are fully methylated, by its size, if Eco72I digests not only the CpG-nonmethylated DNA but also the CpG-hemimethylated DNA. We thus biochemically examined the ability of Eco72I to digest the non-, hemi-and full-methylated DNAs at its recognition sequence, CACGTG. As shown in Fig. 1C, the two CACGTG-containing double-stranded (ds) DNAs of CpG146 were almost completely resistant to the digestion by Eco72I at a concentration of 1.8 unitsÁpmol À1 DNA, when both strands have 5mC modifications (i.e., full-methyl DNA; lane 5). At this concentration, the nonmethylated (lane 2) and hemimethylated (lanes 3 and 4) DNAs were completely and partially digested by Eco72I, respectively. Therefore, we assumed that the purification of CpG-fully methylated DNA is possible by the utilization of Eco72I at this concentration.
Preparation of CpG146 DNA, de novo CpG methyltransferase, and CpG-methylated CpG146 DNA The designed CpG146 nucleosomal DNA was purified on a large-scale, using the E. coli HB101 strain carrying the plasmid pWMD01 [19], containing 16 tandem copies of the 73-bp CpG146 unit sequence. From 10.8 L of the TB culture of this strain, we purified 95 mg of the plasmid DNA. The 73-bp CpG146 unit DNA was excised from the plasmid, and self-ligated to yield CpG146, essentially as previously described [20]. Finally, 24 mg of the ligated CpG146 DNA was purified to near homogeneity and used for subsequent CpG methylation.
To prepare a larger quantity of the methylated CpG146 DNA with quality suitable for crystallization, we chose the Spiroplasma M.SssI enzyme as a de novo CpG methyltransferase. We subcloned the ORF of M.SssI into the bacterial expression vector pET15b, for expression of the enzyme on a milligram-scale. We purified approximately 8 mg of the M.SssI enzyme, with a total of 96 000 units of enzymatic activity. The specific activity of the purified M.SssI was calculated in comparison with the M.SssI commercially available from New England Biolabs (NEB, Ipswich, MA, USA; cat. M0226M), by comparatively analyzing the Eco72I digestion efficiency toward the, respectively, methylated CpG146 DNA ( Fig. 2A). One unit of the specific activity is defined by NEB as the amount of enzyme required to protect 1 lg of k DNA, in a total reaction volume of 20 lL in 1 h at 37°C, against cleavage by the BstUI restriction endonuclease. Assuming that the amounts of digested DNA in lane 4 (3 unitsÁlg À1 DNA; equivalent to 0.27 unitsÁpmol À1 DNA) and lane 9 (0.25 lgÁlg À1 DNA; equivalent to 0.023 lgÁpmol À1 DNA) are nearly the same (i.e., bands shown by the gray square bracket), we calculated that our purified M.SssI has a specific activity of 12 000 unitsÁmg À1 .
Using the purified M.SssI enzyme, 6 mg of CpG146 nucleosomal DNA was methylated at the enzyme concentration of 6 unitsÁlg À1 DNA, which is 0.54 unitsÁpmol À1 DNA and two-fold higher than the methylation conditions in lanes 4 and 9 in Fig. 2A. After the methylation reaction, we obtained 5.6 mg of the CpG146 nucleosomal DNA, by phenol-chloroform extraction. The yield of the DNA purification was 93%.
Validation of the CpG full methylation of the CpG146 DNA Using a portion of the methylated CpG146 DNA, we examined whether the DNA has CpG full methylation as designed, by the usage of two restriction endonucleases and one MBD protein. First, we used the Eco72I enzyme, which digests non-and hemimethylated DNA, but does not digest full-methylated DNA (Fig. 2B). As expected, the digestion of the methylated CpG146 DNA by Eco72I was negligibly detected (lane 4) under the conditions, where the unmethylated CpG146 was completely digested (lane 3). The efficiency of the CpG146 DNA methylation was estimated to be more than 95%, by a densitometric analysis. We also validated the methylation of the CpG146 DNA, by using the methylation-dependent restriction endonuclease, MspJI (Fig. 2C). MspJI digests DNA at the positions 9-and 13 bases after the sequence 5 0 -mCNNR-3 0 , where mC, N, and R indicate 5mC, A/G/ C/T, and A/G, respectively. Thus, the four CG dinucleotide-containing sequences around SHLs À6, À1, +1, and +6 in CpG146 can be digested by MspJI, when the DNA is CpG-methylated ( Fig. 1). As expected, the methylated CpG146 DNA was completely digested by  [18] was mutated to contain four CpG dinucleotide-containing CACGTG sequences, which are recognized by the methylation-sensitive restriction enzyme Eco72I. The DNA designated as CpG146 is biochemically methylated by M.SssI, and then digested by Eco72I, which is a CpG methylation-sensitive restriction endonuclease for the CACGTG sequence. The full-length 146-bp nucleosomal DNA is methylated by M.SssI, and a portion of the methylated DNA is examined by a digestion with Eco72I before the NCP reconstitution. (B) Sequences of nucleosomal DNAs used for crystal structure analyses. NCP146, 146-bp a-satellite nucleosomal DNA [18]; and CpG146, CpG dinucleotide sequence-introduced 146-bp nucleosomal DNA (this study). Four CpG dinucleotide-containing Eco72I recognition sequences, created by the site-directed mutagenesis of NCP146, are shown in green. The positions of the CpG dinucleotides are underlined. Sat2R: the 145-bp satellite 2 derivative right nucleosomal DNA [17]. Sat2L: the 146-bp satellite 2 derivative left nucleosomal DNA [17]. The relative positions of DNA bases from the dyad axis (0) are indicated from À70 to +70 at the top. Minor groove-inward facing regions, as reported by Chua et al. [31], are boxed within blue squares. The major grooves of the boxed DNA sequences are outwardfacing, and CpG-methyl reader and/or eraser proteins can access 5mC. (C) Eco72I digestion patterns of double-stranded CpG-containing oligonucleotides. Oligonucleotides MspJI (lane 4 in Fig. 2C), whereas the unmodified CpG146 DNA remained undigested (lane 3). As judged from the digestion assays, we concluded that the purified CpG146 DNA was fully CpG methylated.
Next, we examined whether the purified CpG146 DNA with the CpG full methylation is biochemically functional, by assessing the DNA-binding ability of MBD2 (Fig. 3A). MBD2 is one of the MBD-containing proteins that specifically recognize the full-methyl CpG sequence, and it shows higher affinity for methylated DNA than other MBD domains [21]. We examined MBD2 binding to three nucleosomal DNAs: M.SssItreated NCP146 DNA containing no CpG dinucleotide site, M.SssI-untreated CpG146 DNA containing four CpG dinucleotide sites, and M.SssI-treated CpG146 DNA. MBD2 did not bind to either the M.SssI-treated NCP146 DNA (lane 5 in Fig. 3B) or the M.SssIuntreated CpG146 DNA (lane 5 in Fig. 3C), as expected from their lack of CpG sites and CpG methylation, respectively. In contrast, MBD2 bound to the M.SssItreated CpG146 DNA (lane 5 in Fig. 3D), validating the biochemical functionality of the M.SssI-treated, CpG-methylated CpG146 DNA.

Reconstitution and crystallization of CpGmethylated NCP
To understand the influence of CpG methylation on the nucleosome structure, we reconstituted the NCP using the CpG-fully methylated CpG146 DNA and the four human core histone proteins, H2A, H2B, H3, and H4, essentially as previously described [19,20]. For a structural comparison, we also reconstituted the NCP using the unmodified CpG146 DNA. We found that the CpG-methylated NCP (lane 5 in Fig. 4A,B) was reconstituted apparently as efficiently as the unmodified NCP (lane 1 in Fig. 4A,B), as shown by the ethidium bromide-stained (i.e., DNA-stained; Fig. 4A) and CBB-stained (i.e., protein-stained; Fig. 4B) native PAGE images. The reconstituted NCPs were then subjected to heat shifting (lanes 2 and 6 in Fig. 4A,B), and the free nucleosomal DNAs were removed by the magnesium-dependent condensation of the NCPs (lanes 3 and 7 in Fig. 4A). The purified NCP samples, which are predominantly composed of NCPs with a minor portion of free nucleosomal DNAs (lanes 4 and 8 in Fig. 4A), were subjected to crystallization.
We obtained crystals from both the CpG-full-methylated ( NCPs, under similar crystallization conditions. From the single crystals of the CpG-methylated and unmodified CpG146 NCPs, we collected X-ray diffraction data sets at BL32XU of SPring-8. To exclude the possibility that the methylated CpG sites were demethylated or degraded during the crystallization, we collected crystallization droplets that yielded crystals of the NCPs, and examined them by the Eco72I digestion to determine whether the designed CpG methylation remained intact in the NCP. From three crystallization droplets of both the CpG-methylated and unmodified CpG146 NCPs, we extracted the nucleosomal DNAs by proteinase K treatment and phenol-chloroform extraction, and digested them with Eco72I (Fig. 5C). We confirmed that the CpG full methylation in the CpG-methylated NCPs remained intact for several months after the crystallization (Fig. 5C).

Overall structures of CpG-methylated and unmodified NCPs
The crystal structures of the CpG-methylated and unmodified NCPs were refined to 2.6 and 3.0 A resolution, respectively, by molecular replacement as described in the Materials and methods. To exclude model bias, we used the auto-modeling program Auto-Build (PHENIX; www.phenix-online.org). Consequently, the R work /R free decreased by several percentage points. The final refinement statistics are summarized in Table 1. We solved the structures of the CpG-methylated and unmodified CpG146 NCPs (Fig. 6A,B). In the electron density map, both ends of the 146-bp DNA were clearly observed in the CpG-methylated NCP (data not shown), as also observed for the unmodified NCP in the present and previous studies [22,23] using the 146-bp DNA. The histone-fold core regions were also clearly visible in both NCPs (Fig. 6A,B). The two overall NCP structures were nearly identical, as demonstrated by superimposition (Fig. 6B). In the superimposition of the atomic models of the CpG-methylated and unmodified NCP structures, the global RMSD (rootmean-square deviation) values of all of the protein Ca atoms and the DNA sugar-phosphate backbones were as small as 0.27 and 0.46 A, respectively. Between the CpG-methylated NCP and the modeled NCP146 (PDB: 1KX3) [22] structures, the RMSD values of all of the Ca atoms and the DNA sugar-phosphate backbone were 0.35 and 0.93 A, respectively.

Electron density map around CpG-methylated DNA regions
There are four CpG-fully methylated dinucleotide sites (i.e., 5 0 -mCG-3 0 /5 0 -mCG-3 0 ), containing eight CpGmethyl groups, in the present CpG-methylated NCP structure (Fig. 1B). The 2mF o -DF c electron density maps with stick models are shown for the 5mCs at positions À61, À5, +5, and +61 of chain I and the 5mCs at À62, À6, +4, and +60 of chain J (Fig. 7A-D). In the 2mF o -DF c electron density maps, we observed the methyl groups of 5mC at positions À5 of chain I (Fig. 7B) and À6 of chain J (Fig. 7A), among the eight 5mC positions. The electron densities for the 5mC methyl groups were not observed at the other six positions, at À61, +5, and +61 of chain I and À62, +4, and +60 of chain J (Fig. 7A-D). In general, it is difficult to distinguish the existence of one methyl group, which is located in a DNA region with high B-factors, from the 2mF o À DF c electron density map at 2.6 A resolution in the NCP structures. We placed all eight 5mC methyl groups in the final coordinates of the CpG-methylated CpG146 NCP.

Local structural differences between the CpG-fullmethylated and unmodified NCPs
The overall structures of the CpG-methylated and unmodified NCPs are basically the same. To examine if there are any local structural differences, we plotted the RMSD differences in all of the nucleotides along the nucleosomal DNA chains (Fig. 8A) and all of the modeled amino acid residues in the histone proteins ( Fig. 8B-E). These RMSD differences are mapped onto the CpG-methylated NCP structure, with color coding according to the RMSD differences (Fig. 8F). In the nucleosomal DNA, the most prominent RMSD difference is observed around SHL À2.5, with the peaks of À28I and +24J (Fig. 8A,F). The positions of the backbone phosphate groups of the four CpGmethylated dinucleotide sites (i.e., À61/+60, À5/+4, +5/ À6, and +61/À62 in chains I/J) are basically the same (Fig. 6B), and the differences in their RMSDs between the two NCPs are not significant (Fig. 8A,F). In the histone proteins, the structures of histones H2A, H2B, and H4 are basically the same, and they do not prominently deviate between the two NCPs (Figs 6B and 8B-F). In histone H3, the largest RMSD difference is observed around the a1 helix (residues 70-78 of 64-78) of molecule A (Fig. 8B,F), where the a1 helix of histone H3 and SHL À2.5 of the DNA are closer to each other in the nucleosome (Fig. 8F).

Reconstitution of the CpG-fully methylated NCP
Cytosine methylation of CpG dinucleotides in eukaryotes is one of the key epigenetic modifications regulating a variety of biological phenomena, such as gene silencing, genomic imprinting, and embryogenesis. From both biological and structural points of view, all of the possible statuses of CpG methylation, full-, hemi-, and nonmethyl CpG, are critically involved in the control of these phenomena. The difference in the methylation statuses is precisely recognized by a series of DNA-binding domains, such as the MBD domain (full-methyl CpG reader) [24], the SRA domain (hemi-methyl CpG reader) [25][26][27], and the CXXC domain (nonmethyl CpG reader) [28], which subsequently trigger different biochemical reactions. Therefore, the precise reconstitution of a nucleosome with a designed pattern of CpG methylation, strictly discriminating full-methyl from hemimethyl CpG and vice versa, was eagerly awaited for biochemical and structural studies in epigenetics and chromatin biology. Here, we biochemically addressed this issue, and obtained the crystallization-grade NCP containing four CpG-fully methylated dinucleotide sites.
As for the restriction specificity of Eco72I and its isozymes recognizing the CACGTG sequence (Fig. 1A), the methylation-sensitivity of BbrPI, which is one of the isozymes, is validated using 5mC- substituted DNA [29] or M.SssI-methylated DNA [30]. However, it is unclear whether Eco72I or its isozyme is sensitive to the hemimethylation of the CpG dinucleotide in the CACGTG sequence. In this study, we found that Eco72I can digest the CpG-hemimethylated CACGTG sequence about half as efficiently as the CpG-nonmethylated one (Fig. 1C). Therefore, the present method enabled us to prepare a nucleosome with designed CpG full methylation, using a portion of the methylated nucleosomal DNA and Eco72I (Fig. 2B), which will be applicable to the preparation of NCPs with CpG-fully methylated CACGTG sites.

Structural comparison between the CpG-fullmethyl and unmodified NCPs
We designed and prepared a 146-bp a-satellite-based DNA containing four CpG dinucleotide sequences (CpG146), using enzymatic methylation and digestion, and reconstituted human NCPs with site-specific CpG full methylation. Furthermore, we have solved the crystal structures of the CpG-methylated and unmodified NCPs at 2.6 and 3.0 A resolution, respectively. We found that the overall structure of the CpGmethylated NCP is basically the same as that of the unmodified NCP (Fig. 6B). The CpG full methylation, verified by the digestion with two different endonucleases (Fig. 2B,C), does not affect the overall structure of the histone octamer and the conformations of the DNA in the NCP. We observed the electron densities corresponding to the 5mC methyl groups at two positions among the eight 5mC positions: the À6 5mC of chain J (SHL +0.5; Fig. 7A) and the À5 5mC of chain I (SHL À0.5; Fig. 7B). The two CpG dinucleotide sites are both located within the minor groove-inward center, which is stabilized by the DNA-binding loops of histones (i.e., L2 of histone H3 and L1 of histone H4) [31]. The other two CpG dinucleotide sites (i.e., À61/+60 and +61/À62 in chains I/J) are not located at such histone-stabilized DNA regions, and no clear electron densities for their 5mC methyl groups were observed (Fig. 7C,D). Similarly, the electron densities of the thymine methyl groups were observed for 30 among the 90 thymidine positions in the 2.6 A CpG-methylated NCP structure. Therefore, the partial observation of the 5mC methyl groups is reasonable, considering the 2. 6 A resolution. In this context, no electron density was observed for the 5mC positions in the previous CpG-methylated NCP structures [17].
The effect of the CpG full methylation on the NCP structure In spite of the very small chemical change, the addition of the methyl group to the C5 position of cytosine can influence the major groove readout, using hydrophobic contacts and shape readout in the minor groove [32]. Do the eight CpG methylations affect the local structures of the DNA and the histone octamer? The positions of the four CpG-methylated dinucleotide sites (i.e., À61/+60, À5/+4, +5/À6, and +61/À62 in chains I/J) are shown in Fig. 8A. From the RMSD plots between the CpG-methylated and the unmodified NCPs, the most prominent difference is observed around SHL À2.5, with peaks at À28I/+24J (Fig. 8A,  F). This shift may cause a structural change around the nearby a1 helix of the molecule-A histone H3. All four CpG dinucleotide sites (i.e., À61/+60, À5/+4, +5/ À6, and +61/À62 in chains I/J) are far away from SHL À2.5 (Fig. 8F), and SHL À2.5 is located around one of the most flexible SHLs in this crystal packing [31,33]. Therefore, we conclude that the slight local structural shift between the CpG-methylated and unmodified NCPs is basically attributable to variations in the analyzed crystals.

Comparison with the CpG-methylated human satellite 2 NCPs
The present CpG-methylated NCP structure was compared with the recently reported structures, at 2.63 and 3. 15 A resolution, of the human satellite 2 NCPs prepared with human nonpalindromic satellite 2 derivative sequences (i.e., Sat2R and Sat2L) (PDB IDs: 5CPJ and 5CPK) [17]. Between our structure and the Sat2R and Sat2L NCP structures, the RMSD values of all Ca atoms and the DNA sugar-phosphate backbone are 0.48/0.38 and 1.98/1.94 A, respectively. As shown in Fig. 9A,B, the histone octamers fit well, even though the nucleosomal DNAs do not. This is probably because the DNA sequences of CpG146 and Sat2R/L are quite different. Therefore, it has now been clarified that the CpG methylation does not detectably affect the NCP structure, regardless of the DNA sequence.

Perspective for structural analysis of NCP complexes
The present methodology for the preparation of crystallization quality CpG-fully methylated NCPs can facilitate structural analyses, such as that of a CpGmethyl-containing NCP in complex with MBD proteins [9]. Two major groove-exposed regions (i.e., À5/ +4 and +5/À6 of chains I/J; see Figs 1B and 8F) among the four CpG dinucleotide sites in the CpG146 sequence are potential binding sites for a CpG-binding protein, such as an MBD protein. Since the current crystal packing cannot accommodate the CpG-methylated NCP in complex with an MBD domain, different crystallization conditions with new crystal packing will be required for the structural analysis of the MBDbound, CpG-methylated NCP complex. The present methodology will be useful to understand whether and how the CpG methylation of nucleosomes, and the binding of CpG-methyl reader and/or eraser proteins, influences nucleosomal assembly, positioning, and remodeling.

Conclusions
Through the preparation of a 146-bp a-satellite-based palindromic nucleosomal DNA, in which four CpG sequences are generated, and its enzymatic methylation and restriction, we reconstituted a 'symmetric' human CpG-full-methylated nucleosome core particle (NCP), and solved the crystal structures of the CpG-methylated and unmodified NCPs at 2.6 and 3.0 A resolution, respectively. The electron densities for two methyl groups among the eight 5mCs were observed in the CpG-methylated NCP structure. The CpG-fully methylated NCP resembled the unmodified NCP, except for the local shift in the region composed of the DNA around SHL À2.5 and the a1 helix of the molecule-A histone H3, which are both far from the four CpG-methylated sites. The present methodology to prepare a crystallization-grade, CpG-fully methylated NCP will facilitate future structural analyses of NCPs in complexes with CpG-methyl reader and eraser proteins.

Construction of CpG dinucleotide-containing nucleosomal DNA
The nucleotide sequence of the 146-bp palindromic DNA fragment of human a-satellite DNA region, a kind gift from T. J. Richmond, (Fig. 1B; sequence in the first row) and designated as NCP146 in this study, was modified in order to introduce four Eco72I-sensitive CpG dinucleotide sites (i.e., CACGTG) in the nucleosomal DNA. The constructed DNA is designated as CpG146 in this study ( Fig. 1B; sequence in the second row). The DNA sequence of CpG146 is as follows: 5 0 -ATCAA TATCC ACGTG CCAGT TATAC CAAAA GTGTA TTTGG AAACT CCTAA CTGAA AAGGC ATGTT CACGT GAATT CACGT GAACA TGCCT TTTCA GTTAG GAGTT TCCAA ATACA CTTTT GGTAT AACTG GCACG TGGAT ATTGA T-3 0 . In this sequence, the underlined nucleotides are the positions that were modified from the original a-satellite sequence of NCP146, and the bold CpG dinucleotide sequences are the methylation target sites. Sixteen copies of the half unit of the palindromic CpG146 DNA sequence were tandemly subcloned into the plasmid pWMD01 [19]. Large-scale preparation of the CpG146 nucleosomal DNA was performed essentially as previously described [20].

Eco72I digestion assay with non-, hemi-and fullmethyl DNAs
In order to examine the sensitivity of Eco72I toward hemi CpG-methylated DNA, we prepared double-stranded oligonucleotides containing 5-methylcytosine (5mC) in one strand (i.e., hemimethylated DNA) or both strands (i.e., fully methylated DNA), as well as that without 5mC (i.e., in E. coli, was custom-synthesized by Eurofins Genomics (Tokyo, Japan). The cDNA was subcloned into pET15b at the NdeI and XhoI restriction digestion sites. For bacterial expression of the M.SssI protein, T7 Express lysY Competent E. coli (High Efficiency) cells from NEB (cat. C3010I), harboring the pET15b-M.SssI plasmid, were grown at 37°C in LB medium containing 100 lgÁmL À1 ampicillin and 2% glucose. When the cells reached an OD 600 of approximately 0.6, M.SssI expression was induced by the addition of 50 lM IPTG, and the cells were further cultured at 16°C for 16 h. Purification of the M.SssI protein was performed essentially as described previously [34]. We used a final dialysis buffer composed of 10 mM Tris-HCl (pH 8.0) buffer, containing 500 mM NaCl, 2 mM DTT, 100 lM EDTA, and 50% glycerol. The specific activity of the purified M.SssI enzyme was determined using CpG146 DNA as substrate, in comparison with the activity unit of the NEB M.SssI enzyme (cat. M0226M), by quantifying the amounts of the Eco72I-digested DNA fragments.

Large-scale methylation of CpG146 DNA and its verification
For the preparation of CpG-methylated nucleosomal DNA, 50 lgÁmL À1 of the CpG146 DNA was incubated at 37°C for 16 h with M.SssI (6 unitsÁlg À1 DNA), in 10 mM Tris-HCl buffer (pH 8.0), containing 50 mM NaCl, 2.5 mM EDTA, and 640 lM SAM. To separate the reacted nucleosomal DNA from the M.SssI, the reaction was extracted with phenol-chloroform-isoamyl alcohol (25 : 24 : 1), and the DNA was ethanol-precipitated. A small portion of the reacted DNA was digested with Eco72I (Thermo Fisher  (F) Representation of the RMSD difference between the two structures on the NCP structure. The RMSD difference between the CpG-methylated and unmodified CpG146 NCPs is colored on the CpG-methylated NCP structure. The color bar represents the RMSD difference from 0 A 2 (bottom) to 1.34 A 2 (top). The regions with the most prominent RMSD difference in the nucleosomal DNA and histone H3 are shown by the orange and yellow circles, respectively. The positions of the four CpG dinucleotide sites are shown by arrows. The +61I/+62I CpG dinucleotide, located behind the first turn of the DNA duplex, is denoted in gray. The major groove-exposed or -unexposed positions of the methylated CpG dinucleotides are shown.
Scientific; cat. ER0361) or MspJI (NEB, cat. R0661S), at 1.8 unitsÁpmol À1 DNA and 0.45 unitsÁpmol À1 DNA, respectively. The CpG-methylated CpG146 DNA was also used in the binding assay with MBD2, using an EpiXplore methylated DNA enrichment kit (Takara Bio, Shiga, Japan; cat. 631963). In each MBD2-binding reaction, the histidinetagged MBD2 protein (5 lg) from the kit, which was immobilized onto 4 lL of TALON resin, was incubated with 500 ng of the 146-bp nucleosomal DNA at 25°C for 1 h. The incubated samples were washed four times, using the 19 binding/washing buffer of the kit, and the nucleosomal DNAs were eluted with elution buffer (high). The nucleosome DNA of each fraction was ethanol-precipitated and suspended in 20 lL of 10 mM Tris-HCl buffer (pH 8.0) containing 100 lM EDTA. One quarter (5 lL) of each fraction was loaded onto a 5% polyacrylamide gel and analyzed by electrophoresis.

Reconstitution of the CpG-methylated NCP for crystal structure analysis
Reconstitution of the NCPs was performed by the salt gradient method, essentially as described previously [19,20], using bacterially expressed, purified human histones H2A, H2B, H3, and H4 [20], and M.SssI-treated or -untreated CpG146 DNA. The reconstituted NCPs were heat-shifted at 42°C for 1 h, and precipitated by the addition of MgCl 2 to a final concentration of 12 mM. The reaction solutions were incubated for 15 min at 25°C, and the NCPs were precipitated by centrifugation at 17 500 g for 10 min at 4°C. Then, the precipitated NCPs were separated from the supernatant containing the free nucleosomal DNA. For crystallization, the precipitated NCPs were resolved in CCS buffer (20 mM potassium cacodylate buffer [pH 6.0] containing 1 mM ethylenediamine tetraacetic acid) [20]. Crystallization was performed by the hanging-drop vapordiffusion method at 20°C. The drops were prepared by mixing 1.0 lL of potassium cacodylate buffer (pH 6.0), containing 85-110 mM manganese(II) chloride and 110-140 mM potassium chloride, with 1.0 lL of the NCP solution (4-6 mgÁmL À1 ) in CCS buffer, and were equilibrated against 500 lL of the reservoir solution (20 mM potassium cacodylate buffer [pH 6.0], containing 45 mM manganese (II) chloride and 35 mM potassium chloride). Single crystals typically appeared within 2-7 days. The crystals were transferred to 20 mM potassium cacodylate buffer (pH 6.0), containing 37 mM manganese(II) chloride, 40 mM potassium chloride, 2% trehalose, and 24% MPD, and then flashed cooled with liquid nitrogen.
Data collection, structure determination, and refinement Diffraction data were collected at 100 K on the BL32XU beam line at SPring-8 (Hyogo, Japan). Diffraction images were processed with XDS [35] and HKL2000 [36]. The structures were solved by molecular replacement (MR) with PHASER [37], using the structural coordinates of the 146-bp DNA in NCP146 (PDB: 1KX3) [22] and the histone octamer in the human NCP (PDB: 2CV5) [38] as the search models. Structural refinement was accomplished with the PHENIX suite [39], and manual model building in Coot [40]. PYMOL (The PyMOL Molecular Graphics System) was used to render the structural figures and for general manipulations. The final refinement statistics are summarized in Table 1.

PDB accession numbers
The structural coordinates have been deposited in the Protein Data Bank, under the accession codes PDB: 5B2J (the CpG-methylated CpG146 NCP) and PDB: 5B2I (the unmodified CpG146 NCP), respectively.