Nucleosome-CHD4 chromatin remodeller structure explains human disease mutations

Chromatin remodelling plays important roles in gene regulation during development, differentiation and in disease. The chromatin remodelling enzyme CHD4 is a component of the NuRD and ChAHP complexes that are involved in gene repression. Here we report the cryo-electron microscopy (cryo-EM) structure of Homo sapiens CHD4 engaged with a nucleosome core particle in the presence of the non-hydrolysable ATP analogue AMP-PNP at an overall resolution of 3.1 Å. The ATPase motor of CHD4 binds and distorts nucleosomal DNA at super-helical location (SHL) +2, supporting the ‘twist defect’ model of chromatin remodelling. CHD4 does not induce unwrapping of terminal DNA, in contrast to its homologue Chd1, which functions in gene activation. Our results also rationalize the effect of CHD4 mutations that are associated with cancer or the intellectual disability disorder Sifrim-Hitz-Weiss syndrome.


Introduction
In the nucleus of eukaryotic cells, DNA is compacted into chromatin. The fundamental building block of chromatin is the nucleosome, a complex of 146 base pairs (bp) of DNA wrapped around an octamer of histone proteins. The degree of chromatin compaction influences DNA replication, transcription, and repair. Maintenance of the appropriate chromatin state requires ATP-dependent chromatin-remodelling enzymes. These 'chromatin remodellers' are divided into four families, called CHD, SWI/SNF, ISWI, and INO80 (Clapier et al., 2017). All chromatin remodellers contain a conserved ATPase core that utilizes ATP hydrolysis to alter contacts between nucleosomal DNA and the histone octamer and to facilitate nucleosome assembly, sliding, ejection, or histone exchange. Members of the CHD (chromodomain helicase DNAbinding) protein family of chromatin remodellers all contain a central SNF2-like ATPase motor domain and a double chromodomain in their N-terminal region. The double chromodomain binds modified histones (Sims et al., 2005) and interacts with nucleosomal DNA (Nodelman et al., 2017). The interaction with DNA regulates and fine tunes ATPase activity. Recent structures of the yeast remodeller Chd1 in complex with a nucleosome uncovered the architecture of one subfamily of CHD remodellers (subfamily I) and its interactions with the nucleosome (Farnung et al., 2017;Sundaramoorthy et al., 2018). A unique feature of these structures is that Chd1 binding induces unwrapping of terminal DNA from the histone octamer surface at superhelical location (SHL) -6 and -7 (Farnung et al., 2017;Sundaramoorthy et al., 2018). However, the resolution of these studies was limited, such that atomic details were not resolved. The human CHD family member CHD4 (Woodage et al., 1997) shows nucleosome spacing activity (Silva et al., 2016). CHD4 is also known as Mi-2 in Drosophila melanogaster (Kehle et al., 1998) and together with CHD3 forms subfamily II, which differs in domain architecture from subfamily I. CHD3 and CHD4 contain two N-terminal plant homeodomain zinc fingers (Schindler et al., 1993) (PHD fingers 1 and 2), a DNA-interacting double chromodomain, and the ATPase motor. CHD4 contains an additional high mobility group (HMG) box-like domain in its N-terminal region (Silva et al., 2016) and two additional domains of unknown function located in the C-terminal region. CHD4 is implicated in the repression of lineage-specific genes during differentiation (Liang et al., 2017) and is required for the establishment and maintenance of more compacted chromatin structures (Bornelöv et al., 2018). CHD4 mutations have a high incidence in some carcinomas (Getz et al., 2013) as well as thyroid and ovarian cancers (Längst and Manelyte, 2015). Some mutations in CHD4 have also been implicated in intellectual disability syndromes (Sifrim et al., 2016;Weiss et al., 2016). CHD4 is part of the multisubunit Nucleosome Remodelling Deacetylase (NuRD) complex (Tong et al., 1998;Xue et al., 1998;Zhang et al., 1998). NuRD also contains the deacetylase HDAC1/2 and accessory subunits that serve regulation and scaffolding roles. NuRD is implicated in gene silencing, but also gene activation (Gnanapragasam et al., 2011). It is essential for cell cycle progression (Polo et al., 2010), DNA damage response (Larsen et al., 2010;Smeenk et al., 2010), establishment of heterochromatin (Sims and Wade, 2011), and differentiation (Bornelöv et al., 2018;Burgold et al., 2019). It was recently shown that CHD4 is also part of the heterotrimeric ChAHP complex that is also involved in gene repression (Ostapcuk et al., 2018). Thus far, structural studies of CHD4 have been limited to individual domains (Kwan et al., 2003;Mansfield et al., 2011). Here we report the cryo-electron microscopy (cryo-EM) structure of human CHD4 bound to a nucleosome at an overall resolution of 3.1 Å. CHD4 engages the nucleosome at SHL +2 and induces a conformational change in DNA at this location in the presence of the ATP analogue adenylyl imidodiphosphate (AMP-PNP). Structural comparisons show that CHD4, in contrast to Chd1, does not induce unwrapping of terminal DNA. Maintenance of the integrity of the nucleosome in the presence of CHD4 is consistent with the role of CHD4 in gene repression, and in heterochromatin formation and maintenance. Finally, the detailed nucleosome-CHD4 structure enables mapping of known human disease mutations (Kovač et al., 2018;Sifrim et al., 2016;Weiss et al., 2016) and indicates how these perturb enzyme function.

Results
Nucleosome-CHD4 complex structure. To investigate how the human chromatin remodeller CHD4 engages a nucleosome and to understand the structural basis of cancerrelated mutations in CHD4, we determined the structure of H. sapiens CHD4 bound to a Xenopus laevis nucleosome core particle in the presence of the ATP analogue AMP-PNP. We recombinantly expressed and purified full-length CHD4 and reconstituted a complex of CHD4 with a pre-assembled nucleosome core particle. The nucleosome comprised 145 base pairs (bp) of DNA, corresponding to the Widom 601 sequence (Lowary and Widom, 1998) with additional 4 and 30 bp of extranucleosomal DNA on the entry and exit side of the nucleosome, respectively. The nucleosome-CHD4 complex was purified by size exclusion chromatography (Supplementary Fig. 1).
To determine the structure of the nucleosome-CHD4 complex, we collected single particle cryo-EM data on a Titan Krios (FEI) microscope equipped with a K2 direct electron detector (Gatan) (Methods). We obtained a cryo-EM reconstruction of the nucleosome-CHD4 complex at an overall resolution of 3.1 Å (FSC 0.143 criterion) (Supplementary Fig.  2-4). The nucleosome was resolved at a resolution of 3.0-4.5 Å, whereas CHD4 was resolved at 3.1-5.0 Å depending on the protein region. The register of the DNA was unambiguously determined based on distinct densities for purine and pyrimidine nucleotides around the dyad axis ( Supplementary  Fig. 3h). Well-defined density was also obtained for AMP-PNP and a coordinated magnesium ion in the CHD4 active site ( Supplementary Fig. 3i). The model was locally adjusted and real-space refined, leading to very good stereochemistry (Methods) ( Table 1). CHD4 architecture. The CHD4 ATPase motor binds the nucleosome at SHL +2 (Fig. 1). Binding at this location has also been observed for the chromatin remodellers Chd1 (Farnung et al., 2017;Sundaramoorthy et al., 2018), Snf2 (Liu et al., 2017), and Swr1 (Willhoft et al., 2018. The ATPase motor is in a closed, post-translocated state with AMP-PNP bound in the active site. A similar state was observed for Chd1 when bound to ADP·BeF 3 (Farnung et al., 2017;Sundaramoorthy et al., 2018Sundaramoorthy et al., , 2017. The double chromodomain is located at SHL +1 and contacts the nucleosomal DNA phosphate backbone via electrostatic interactions, in a fashion similar to that observed for S. cerevisiae Chd1 (Fig. 1) (Farnung et al., 2017;Nodelman et al., 2017). The CHD4 domain PHD finger 2 is located near SHL +0.5 and the double chromodomain. This location is consistent with NMR studies that predicted binding of this PHD fin-ger close to the dyad axis and the H3 tail (Gatchalian et al., 2017). Additionally, we observe parts of the C-terminal bridge (Hauk et al., 2010), an amino acid segment that follows the ATPase lobe. Part of the C-terminal bridge docks against ATPase lobe 2 and extends towards the first ATPase lobe (Fig. 1, Supplementary Figure 3j). This region was not resolved in the nucleosome-Chd1 structures but was observed in a previously published crystal structure of autoinhibited Chd1 (Hauk et al., 2010). Taken together, CHD4 and Chd1 share a core architecture that involves the ATPase motor and the double chromodomain but differ in their peripheral subfamily-specific protein features.
CHD4 binding does not detach exit side nucleosomal DNA. In contrast to the nucleosome-Chd1 structure (Farnung et al., 2017), we did not observe unwrapping of nucleosomal DNA from the histone octamer on the second DNA gyre at SHL -6 and -7 (Fig. 2). This major difference between these complex structures may be due to a lack of a DNA-binding region in CHD4. Chd1 uses its DNA-binding region to interact extensively with terminal DNA on the exit side at SHL -7, and such contacts are absent in the nucleosome-CHD4 structure (Fig. 2). It is likely that other CHD family members such as CHD3 and CHD5, which also lack a DNA-binding region, will also not induce unwrapping of terminal DNA.
CHD4-DNA interactions. The high resolution of our nucleosome-CHD4 structure enables a detailed description of the interactions of the ATPase motor with nucleosomal DNA. CHD4 contacts the phosphate backbone of the tracking and guide strand via electrostatic interactions that are mostly mediated by lysine and arginine residues (Fig. 3). These interactions with the DNA phosphate backbone are formed by residues in the canonical ATPase motifs Ia, Ic, II, IV, IVa, V, and Va and by residues present in non-canonical motifs (e.g. Lys810) (Fig. 3, Supplementary Fig. 5). We also observe that residues Asn1010, Arg1127, and Trp1148 insert into the DNA minor groove over a stretch of seven base pairs (Fig. 3c). Asn1010 is not part of a canonical ATPase motif and inserts into the DNA minor groove around SHL +2.5. Arg1127 (motif V) is universally conserved in all CHD chromatin remodellers and inserts into the DNA minor groove at SHL +2. Our density is consistent with two alternative conformations of the Arg1127 side chain, with the guanidinium head group pointing either towards the tracking or the guide strand of DNA. Trp1148 is located in motif Va, inserts into the minor groove near the guide strand, and plays a critical role in coupling ATPase hydrolysis and DNA translocation (Liu et al., 2017). We further observe a contact between a negatively charged loop in ATPase lobe 1 (residues 832-837) and the second DNA gyre at SHL -6. This loop is present in CHD3, CHD4, and CHD5, but not in Snf2 or ISWI remodellers ( Supplementary Fig. 5  , dyad (c), and side (d). Histones H2A, H2B, H3, H4, tracking strand, guide strand, CHD4 PHD finger 2, double chromodomain, ATPase lobe 1, and AT-Pase lobe 2 are coloured in yellow, red, light blue, green, blue, cyan, pink, purple, orange, and forest green, respectively. Colour code used throughout. The dyad axis is indicated as a black line or a black oval circle. Magnesium and zinc ions shown as pink and grey spheres, respectively. AMP-PNP shown in stick representation. motor engages its DNA substrate (SHL +2) (Fig. 3d). The high resolution of the nucleosome-CHD4 structure shows that 5 DNA base pairs between SHL +1.5 and SHL +2.5 are pulled away from the octamer surface by up to 3 Å. This distortion does not include the previously observed 'bulging' or a 'twist defect' that is characterized by a 1 bp local underwinding of the DNA duplex and observed when the AT-Pase motor adopts the open/apo or ADP-bound states (Li et al., 2019). In contrast, the DNA distortion observed in our AMP-PNP bound state is an intermediate between the bulged and the canonical DNA conformation (Fig. 3d). This AMP-PNP bound intermediate DNA state was predicted based on biochemical experiments (Winger et al., 2018). This observation demonstrates that the extent of DNA distortion at SHL +2 depends on the functional state of the ATPase motor and is consistent with the proposed twist defect propagation model of chromatin remodelling (Winger et al., 2018).
CHD4 binds the histone H4 tail. As observed for S. cerevisiae Chd1 (Farnung et al., 2017), H. sapiens CHD4 contacts the histone H4 tail with its ATPase lobe 2. The H4 tail is located between ATPase lobe 2 and the nucleosomal DNA at SHL +1.5. The conformation of the H4 tail differs from that observed in structures of the free nucleosome where the tail makes inter-nucleosomal contacts with the 'acidic patch' of a neighbouring nucleosome. It also differs from the H4 position observed in a higher-order structure where the H4 tail extends over the DNA interface between two nucleosomes (Schalch et al., 2005). A loop in lobe 2 of the ATPase (CHD4 residues 1001-1006) replaces the H4 tail in this position, apparently inducing H4 positioning that allows ATPase lobe 2 binding (Fig. 4a). ATPase lobe 2 contains a highly acidic cavity formed by Asp1080, Glu1083, Asp1084, and Glu1087 (Fig. 4a). This acidic cavity is conserved across all CHD family members. The basic side chain of the H4 histone tail residue Arg17 inserts into this acidic cavity (Fig. 4a). Similar interactions with the H4 tail have also been reported for Snf2 and ISWI remodellers (Armache et al., 2019;Yan et al., 2019). The side chain of H4 Lys16 also points towards the acidic cavity and is positioned in close proximity to residues Asp1080 and Glu1083. Acetylation of H4 Lys16 is therefore predicted to weaken these charge-based interactions and to reduce the affinity of chromatin remodellers for the H4 tail, as noted before (Yan et al., 2016).
CHD4 interacts with histone H3. The ATPase lobe 2 also contacts the core of histone H3 (alpha helix 1, Gln76 and Arg83) via CHD4 residues Asn1004 and Leu1009, respectively (Fig. 4b). This contact is critical for chromatin remodelling. Deletion of the homologous region in Chd1 leads to abolishment of chromatin remodelling activity (Sundaramoorthy et al., 2018). However, it remains unclear if these contacts are required for proper substrate recognition and positioning or whether they are also necessary to generate the force required for DNA translocation. Low-pass filtering of our map further shows the H3 N-terminal tail trajectory, which extends to the double chromodomain ( 4c). The contact between the H3 tail and the double chromodomain could target CHD4 to nucleosomes methylated at Lys27 of H3 (Kuzmichev et al., 2002), a classical mark for gene repression.
Two CHD4 molecules can engage with the nucleosome. During 3D classification of our cryo-EM dataset we observed a distinct class of particles that showed two CHD4 molecules bound to the same nucleosome (Fig. 5, Supplementary Fig. 2-4). Refinement of this class of particles yielded a reconstruction at an overall resolution of 4.0 Å (FSC 0.143 criterion) ( Table 1). A model of this nucleosome-CHD42 complex was obtained by docking the refined nucleosome-CHD4 model into the density and then placing another CHD4 molecule into the additional density observed on the opposite side. The resulting nucleosome-CHD42 complex structure shows pseudo-twofold symmetry with CHD4 molecules bound at SHL +2 and SHL -2 (Fig. 5). The second CHD4 molecule uses its double chromodomain and PHD finger 2 to contact nucleosomal DNA at SHL +1 and +0.5, respectively. Binding of the second CHD4 molecule also did not lead to unwrapping of terminal DNA. Binding of two chromatin remodellers to a single nucleosome was previously observed for S. cerevisiae Chd1 (Sundaramoorthy et al., 2018) and H. sapiens SNF2H (Armache et al., 2019). However, in contrast to the structure of the nucleosome-SNF2H2 complex, we do not observe a distortion in the histone octamer due to the presence of the chromatin remodellers. Binding of two remodeller molecules could allow for higher efficiency in positioning the nucleosome at a precise location but necessitates coordination of the remodellers. A possible mechanism for coordination could be that twist defects that are introduced by remodeller binding are propagated from the entry SHL 2 into the exit side SHL 2 (Brandani et al., 2018;Brandani and Takada, 2018). Presence of the twist defect at the second remodeller binding site could interfere with the translocation activity of the second remodeller (Sabantsev et al., 2019).

Cancer-related CHD4 mutations.
Many studies have reported mutations in CHD4 that are related to human diseases, in particular cancer (Xia et al., 2017). Mutations involved in various cancer phenotypes have been observed in the PHD finger 2, the double chromodomain, and both lobes of the ATPase motor. To elucidate effects of such mutations on CHD4 activity, the D. melanogaster CHD4 homologue Mi-2 has been used as a model protein for functional analysis (Kovač et al., 2018). CHD4 mutations have been found to fall in two categories. Whereas some mutations influence ATPase and DNA translocation activity (Arg1162, His1196, His1151 and Leu1215), other mutations seem to change protein stability (Leu912, and Cys464) or disrupt DNA binding (Val558 and Arg572). To rationalize these findings, we mapped known CHD4 mutations on our high-resolution structure (Fig. 6, Table 2). Selected sites of mutation are described below. Mutation of residue His1151 to arginine results in a significant reduction of ATPase activity and abolishes chromatin remodelling activity (Kovač et al., 2018). The close proximity of this residue to motif Va (CHD4 residues 1147-1150) makes it likely that the mutation disrupts motif Va function, leading to an uncoupling of the ATPase activity from chromatin remodelling. Similar findings were made for Snf2 where mutation of the tryptophan residue in motif Va resulted in an uncoupling phenotype (Liu et al., 2017). The most frequently mutated residue in endometrial cancer, arginine 1162, is located in the ATPase motif VI. It forms an 'arginine finger' that directly interacts with AMP-PNP in our structure. Mutation of Arg1162 to glutamine impairs ATP hydrolysis as suggested by biochemical data (Kovač et al., 2018).
Other disease-related CHD4 mutations. De novo missense mutations in CHD4 are also associated with an intellectual disability syndrome with distinctive dysmorphisms (Sifrim et al., 2016;Weiss et al., 2016). Mutations observed in patients with this syndrome are located in PHD finger 2 (Cys467Tyr) and predominantly in ATPase    lobe 2 (Ser851Tyr, Gly1003Asp, Arg1068His, Arg1127Gln, Trp1148Leu, Arg1173Leu, and Val1608Ile). We mapped the sites of these mutations onto our structure (Fig. 6) and predicted the effects of the mutations as far as possible ( Table  2). The Cys467Tyr mutation disrupts coordination of a zinc ion in PHD finger 2. Gly1003 in ATPase lobe2 is located in a loop near H3 alpha helix 1. Deletion of this loop in Chd1 results in a loss of chromatin remodelling activity (Sundaramoorthy et al., 2018). Residue Arg1068 forms a hydrogen bond network with the side chain of Thr1137 and the main chain carbonyl groups of Phe1112 and Gln1119. The Arg1068Cys mutation disrupts this network and is predicted to impair the integrity of the ATPase fold. Mutation of Arg1127 disrupts its interactions with the DNA minor groove (Fig. 2c). The equivalent arginine residue in SMARCA4, which is one of the catalytic subunits of the BAF complex, has been implicated in the rare genetic disorder Coffin-Siris syndrome (Tsurusaki et al., 2012). Trp1148, which is part of ATPase motif Va, contacts the guide strand in a fashion similar to Chd1 and Snf2 (Farnung et al., 2017;Liu et al., 2017) (Fig. 2c). Mutation of this residue uncouples ATP hydrolysis and chromatin remodelling (Liu et al., 2017). Arg1173 inserts into an acidic pocket formed by residues Glu971, Asp1147, and Asp1153. Mutation of the arginine residue to leucine is likely to destabilize ATPase lobe 2 folding.

Discussion
Here we provide the 3.1 Å resolution cryo-EM structure of human CHD4 engaged with a nucleosome and the 4.0 Å resolution structure of a nucleosome-CHD42 complex that contains two molecules of CHD4. Our structure of the nucleosome-CHD4 complex reveals how a subfamily II CHD remodeller engages with its nucleosomal substrate. We observe a distortion of nucleosomal DNA at SHL +2 in the presence of AMP-PNP. Similar observations were previously made for the Snf2 chromatin remodeller (Li et al., 2019;Liu et al., 2017) in its apo and ADP-bound states. Our high-resolution structure elucidates the mechanism of chromatin remodelling by capturing an additional enzymatic state. The DNA distortion at SHL +2 that we observed in the AMP-PNP bound state differs from distortions observed previously in the apo and ADP bound state that involved a twist distortion (Li et al., 2019;Winger et al., 2018). This is consistent with a proposed 'twist defect' mechanism for chromatin remodelling (Li et al., 2019;Sabantsev et al., 2019). In this model, binding of the ATPase motor at SHL ±2 induces a twist defect in the DNA. Subsequent ATP binding, captured by AMP-PNP and ADP·BeF 3 structures, then induces closing of the ATPase motor and leads to propagation of the twist defect towards the dyad. It is possible that previous nucleosome-Chd1 structures with ADP·BeF 3 (Farnung et al., 2017;Sundaramoorthy et al., 2018) also contained the DNA distortion observed here but that their lower resolution prevented its detailed observation. Finally, ATP hydrolysis would reset the remodeller and the enzymatic cycle can resume at the next DNA position. A major difference between the subfamily II remodeller CHD4 and the subfamily I remodeller Chd1 is that Chd1 induces unwrapping of the terminal nucleosomal DNA, whereas CHD4 does not change the DNA trajectory between SHL -7 to -5. This is likely related to a striking difference in function. Whereas Chd1 functions in euchromatic regions of the genome during active transcription (Skene et al., 2014), CHD4 plays a central role in the establishment and maintenance of repressive genome regions. Consistent with these findings, DNA unwrapping should be prevented in stable heterochromatic regions. It is likely that the evolution of auxiliary domains in different CHD subfamilies led to these different functionalities. In particular, the DNA-binding region in Chd1 or the PHD fingers in CHD4 alter the functional properties of these chromatin remodellers, with the former working on active genes, and the latter often functioning in gene repression. Our structure also helps to define how causative disease mutations impair CHD4 function. Mutations in disease phenotypes are able to disrupt DNA binding, impede ATP hydrolysis, or uncouple ATP hydrolysis and DNA translocation. The structure rationalizes the effects of CHD4 mutations in cancer and intellectual disability syndromes on chromatin remodelling. It also helps in understanding disease phenotypes of other chromatin remodellers such as the BAF complex that shows a related domain architecture for its ATPase motor. Due to its high resolution, the structure may also guide drug discovery using chromatin remodellers as targets.
The authors declare no competing financial interests.
Nucleosome Preparation. Xenopus laevis histones were expressed and purified as described (Dyer et al., 2003;Farnung et al., 2017). DNA fragments for nucleosome reconstitution were generated by PCR essentially as described (Farnung et al., 2018). A vector containing the Widom 601 sequence was used as a template for PCR. Super-helical locations are assigned based on previous publications (Farnung et al., 2018(Farnung et al., , 2017Kujirai et al., 2018;Sundaramoorthy et al., 2018), assuming potential direction of transcription from negative to positive SHLs. Large-scale PCR reactions were performed with two PCR primers (forward primer: TGT  TGG ATG TTT TAT AAT TGA GTG GGT TCC TGT TAT  TCC TAG TAA TCA ATC AGT GCC TAT CGA TGT ATA TAT CTG ACA CGT GCC T, reverse primer: CCC CAT CAG AAT CCC GGT GCC G) at a scale of 25 mL. Nucleosome core particle reconstitution was performed using the salt-gradient dialysis method (Dyer et al., 2003). Quantification of the reconstituted nucleosome was achieved by measuring absorbance at 280 nm. Molar extinction coefficients were determined for protein and nucleic acid components and were summed to yield a molar extinction coefficient for the reconstituted extended nucleosome.

Reconstitution of nucleosome-CHD4 complex.
Reconstituted nucleosome core particles and CHD4 were mixed at a molar ratio of 1:2. AMP-PNP was added at a final concentration of 1 mM and the sample was incubated for 10 minutes on ice. After 10 minutes compensation buffer was added to a final buffer concentration of 30 mM NaCl, 3 mM MgCl 2 , 20 mM Na·HEPES pH 7.5, 4% (v/v) glycerol, 1 mM DTT. The sample was applied to a Superose 6 Increase 3.2/300 column equilibrated in gel filtration buffer (30 mM NaCl, 3 mM MgCl 2 , 20 mM Na·HEPES pH 7.5, 5% (v/v) glycerol, 1 mM DTT). The elution was fractionated in 50 µL fractions and peak fractions were analysed by SDS-PAGE. Relevant fractions containing nucleosome core particle and CHD4 were selected and cross-linked with 0.1% (v/v) glutaraldehyde. The crosslinking reaction was performed for 10 min on ice and subsequently quenched for 10 min using a final concentration of 2 mM lysine and 8 mM aspartate. The sample was transferred to a Slide-A-Lyzer MINI Dialysis Unit 20,000 MWCO (Thermo Scientific), and dialysed for 4 h against 600 ml dialysis buffer (30 mM NaCl, 3 mM MgCl 2 , 20 mM Na·HEPES pH 7.4, 20 mM Tris·HCl pH 7.5, 1 mM DTT). The sample was subsequently concentrated using a Vivaspin 500 ultrafiltration centrifugal concentrator (Sartorius) to a final concentration of 200-300 µM.
Cryo-EM analysis and image processing. The nucleosome-CHD4 sample was applied to R2/2 gold grids (Quantifoil). The grids were glow-discharged for 100 s before sample application of 2 µl on each side of the grid. The sample was subsequently blotted for 8.5 s (Blot force 5) and vitrified by plunging into liquid ethane with a Vitrobot Mark IV (FEI Company) operated at 4°C and 10% humidity. Cryo-EM data were acquired on a Titan Krios transmission electron microscope (FEI/Thermo) operated at 300 keV, equipped with a K2 summit direct detector (Gatan) and a GIF Quantum energy filter. Automated data acquisition was carried out using FEI EPU software at a nominal magnification of 130,000x in nanoprobe EF-TEM mode. Image stacks of 40 frames were collected in counting mode over 10 s. The dose rate was 4.3-4.5 eper Å 2 per s for a total dose of 43-45 eper Å 2 . A total of 3,904 image stacks were collected. Micrograph frames were stacked and processed. All micrographs were CTF and motion corrected using Warp (Tegunov and Cramer, 2018). Particles were picked using an in-house trained instance of the neural network BoxNet2 of Warp, yielding 650,598 particle positions. Particles were extracted with a box size of 300 2 pixel and normalized. Image processing was performed with RELION 3.0-beta 2 (Zivanov et al., 2018). Using a 30 Å low-pass filtered ab initio model generated in cryoSPARC from 1,679 particles ( Supplementary Fig.  2c) we performed one round of 3D classification of all particle images with image alignment. One class with defined density for the nucleosome-CHD4 complex was selected for a second round of classification. The second round of classification resulted in two classes with one copy of CHD4 bound to the nucleosome. The respective classes were selected and 3D refined. The refined nucleosome-CHD4 model was subsequently CTF refined and the beam tilt was estimated based on grouping of beam tilt classes according to their exposure positions. The CTF refined particles were submitted to one additional round of masked 3D classification without image alignment. The mask encompassed CHD4. The most occupied class from this classification was subsequently CTFrefined. The final particle reconstruction was obtained from a 3D refinement with a mask that encompasses the entire nucleosome-CHD4 complex. The nucleosome-CHD4 reconstruction was obtained from 89,623 particles with an overall resolution of 3.1 Å (goldstandard Fourier shell correlation 0.143 criterion). The final map was sharpened with a B-factor of -36 Å 2 . Additionally, the second round of 3D classification yielded a class with a nucleosome-CHD4 2 complexes. The particles were subsequently classified and refined. The resulting reconstruction with 40,233 particles had an overall resolution of 4.0 Å (goldstandard Fourier shell correlation 0.143 criterion). The final map was sharpened with a B-factor of -86 Å 2 . Local resolution estimates for both structures were determined using the built-in RELION tool.  Fig. 3).

Model building.
Additional structural elements such as the H4 tail, the Cterminal bridge and loop regions of CHD4 were built using COOT. AMP-PNP and a coordinated Mg2+ ion were placed into the corresponding density. AMP-PNP was derived from the monomer library in COOT. The high resolution of our reconstruction enabled us to model several DNA-interacting side chains in two alternative conformations.      (Tegunov and Cramer, 2018). Scale bar with a length of 500 Å is shown. b, 2D classes of single copy CHD4 bound to a nucleosome. Scale bar with a length of 200 Å is shown. c, Classification tree employed to obtain cryo-EM density of CHD4 bound to a nucleosome. Particle numbers and class distribution percentages are indicated. Final reconstructions are highlighted.