Introduction

The Chromodomain Helicase DNA-binding (CHD) protein family is a key class of ATP-dependent chromatin remodeling proteins, which utilize energy derived from ATP hydrolysis to regulate chromatin structure, thereby modulating gene expression1,2. CHD proteins are crucial for developmental processes1,3, with various members implicated in major neurodevelopmental disorders including CHD2 in epileptic encephalopathy4, CHD7 in CHARGE syndrome5, CHD8 in autism6,7, and more recently CHD4 and CHD1 in neurodevelopmental syndromes8,9. Three CHD proteins (CHD3, CHD4, and CHD5) can exert their chromatin remodeling activity by forming the core ATPase subunit of the NuRD complex1,10,11,12. The NuRD complex is associated with various fundamental cellular mechanisms, including genomic integrity and cell cycle progression13, and plays important roles in embryonic stem cell differentiation14. A recent study reports that the different CHD factors within the NuRD complex (CHD3, CHD4, and CHD5) are developmentally regulated in the mouse brain, each having distinct and mostly non-redundant functions during cortical development15. In particular, the CHD3 protein has been implicated in late neural radial migration and cortical layer specification.

In contrast to most other members of the CHD protein family, a specific syndrome associated with mutations in CHD3 (MIM 602120) has not yet been characterized. In this study, based on an index case from whole genome sequencing of children with rare speech disorders, we assemble a set of 35 probands carrying de novo mutations that disrupt CHD3. We characterize the overlapping phenotypic features of probands with CHD3 mutations, including intellectual disability (with a wide range of severity), developmental delays, macrocephaly, impaired speech and language skills, and characteristic facial features. We identify mainly missense mutations that cluster in and around the ATPase/helicase domain of the CHD3 protein, and are predicted to disturb function, based on three-dimensional modeling. We use functional assays to describe the effects of multiple different CHD3 mutations on ATPase activity and chromatin remodeling capacities. Taken together, our data demonstrate that de novo missense mutations in CHD3 disturb chromatin remodeling activities of the encoded protein, thereby causing a neurodevelopmental disorder.

Results

De novo CHD3 mutations cause a neurodevelopmental phenotype

During whole genome sequencing of a cohort of 19 unrelated children with a primary diagnosis of Childhood Apraxia of Speech (CAS)16, we discovered a de novo missense mutation in CHD3, predicted to disrupt the helicase domain of the encoded protein. CAS is a rare neurodevelopmental disorder characterized by impairments in learning to produce the coordinated sequences of mouth and face movements underlying fluent speech. Remarkably, the CHD3 protein is one of the few documented interaction partners of FOXP2 (see Supplementary Table S1 in ref. 17), a transcription factor that has been implicated in monogenic forms of CAS, accompanied by wide-ranging language problems, in multiple families and unrelated cases18,19,20.

Discovery of the CHD3 mutation (NM_001005273.2, p.Arg1169Trp) in our index case motivated a search for other de novo mutations in this gene. Studies of large numbers of simplex families with an autistic proband have documented just two single non-synonymous de novo variants in CHD3 in probands21,22, while eight additional non-synonymous variants were recently recorded in a study of thousands of children with unexplained developmental disorders from the UK23, with limited information on phenotypic profiles of carriers of CHD3 variants. Via GeneMatcher24 we assemble a cohort of 35 independently diagnosed probands with de novo mutations disrupting CHD3, to systematically assess the phenotypic consequences of damage to this gene.

The 35 probands with de novo mutations in CHD3 show overlapping phenotypes, summarized in Table 1 and in more detail in Supplementary Data 1. All individuals have global developmental delays and/or intellectual disability, with a total IQ varying from 70–85 (borderline intellectual functioning) to below 35 (severe intellectual disability). Nine individuals (29%) show autism or autism-like features, including stereotypic and hand-flapping behavior. Interestingly, the majority of individuals (19 individuals; 58%) have macrocephaly, and in cases where neuroimaging has been performed, widening of cerebrospinal fluid spaces is noted in 10 out of 30 MRI reports (33%). One individual (individual 5) has microcephaly. Hypotonia is reported in 21 individuals (75%). The facial phenotype consists of widely spaced eyes, a broad and bossing forehead, periorbital fullness and narrow palpebral fissures, laterally sparse eyebrows, low-set and often simple ears with thick helices, and a pointed chin (Fig. 1). Joint dislocations and/or hyperlaxity are reported in 12 cases, and five individuals have inguinal or umbilical hernias. Five of the 21 male individuals have undescended testes. Vision problems are quite common and include hypermetropia (11 individuals), strabismus (10 individuals), and cerebral visual impairment (three individuals). One individual (individual 34) developed epilepsy, two additional individuals had neonatal convulsions. In many individuals an abnormal and often unsteady gait is reported, and one individual (individual 13) developed symptoms of Parkinsonism at a later age.

Table 1 Summary of phenotypes found in this cohort of probands with CHD3 mutations
Fig. 1
figure 1

Photographs of affected individuals. Facial photographs showing dysmorphisms in 18 individuals with de novo CHD3 mutations. The majority of individuals have macrocephaly with a prominent or bossing forehead, individual 5 has microcephaly. Hypertelorism or telecanthus is common, often accompanied by narrow palpebral fissures, deep-set eyes, peri-orbital fullness, and/or epicanthal folds. The combination of macrocephaly and deep-set eyes leads to a more prominent supra-orbital ridge. Some individuals show midface hypoplasia. Many individuals have low-set ears that can be posteriorly rotated, and sometimes simple with thick helices. A broad nasal base, prominent nose, a bifid nasal tip, and characteristic pointy chin is also frequently seen, as well as laterally sparse eyebrows

Given that our index case was ascertained on the basis of a formal diagnosis of CAS, we pay special attention to the association of CHD3 mutations with speech and language deficits. The index case was diagnosed with severe speech apraxia at the age of 3 years, and then used sign language to communicate effectively. He has severe problems with expressive speech, against relatively normal scores on language comprehension tests and a composite IQ (KBIT) of 72. In all 33 subjects that were at least 2 years old at the last evaluation, CHD3 disruptions are associated with delayed milestones in the speech and language domain. The average age for first spoken words in this cohort is 2 years and 10 months (range: 1.5–5.5 years, after excluding six individuals that were non-verbal at the last evaluation). Our data suggest that expressive language is more affected than receptive language, and intelligibility is often impaired. Speech-related problems identified in our cohort include dysarthria, speech apraxia, oromotor problems, and stuttering.

De novo CHD3 mutations cluster in the helicase domain

The 35 unrelated probands have 23 different de novo mutations in CHD3 (Fig. 2a, b). None of these mutations are present in the GnomAD database (http://gnomad.broadinstitute.com). Except for four individuals, all individuals have missense mutations. Interestingly, within our cohort there are multiple cases of recurrent identical de novo mutations, revealing mutational hotspots. The most striking is p.Arg985Trp, found in six children from five different families, while two additional individuals have a different substitution affecting the same residue (p.Arg985Gln).

Fig. 2
figure 2

Schematic view of CHD3 transcript and protein with de novo mutations. a Schematic view of CHD3 exons (transcript 1, NM_001005273.2) with the splice site mutation c.4073-2A>G shown that most likely leads to skipping of exon 27 (22 amino acids), while preserving the reading frame. Exon 27 is part of the beginning of the second DUF domain (DUF 1086). Colors of the domains in a match with colors of domains in b and c. Five different types of domains are specified: plant homeodomains (PHD), chromodomains (Chromo), a Helicase domain consisting of two parts (Helicase ATP-binding and Helicase C-terminal), domains of unknown function (DUF), and a C-terminal 2 domain. b Schematic view of linear CHD3 protein (transcript 1, NM_001005273.2) with all mutations, except for the splice site mutation that is shown in a, found in our cohort. Almost all missense mutations cluster in or around the Helicase domain of the CHD3 protein. c Overview of one of the two CHD3-models used in this study, based on the 3MWY protein structure. This figure shows the different domains of the protein in their three-dimensional conformation: chromo domain 1 494–595 (magenta), chromo domain 2 631–673 (red), helicase ATP binding domain (yellow), helicase C-terminal domain (green), ATP binding residues 761–768 (cyan). ATP is orange, and gray residues do not belong to an indicated domain. Colors of the domains in c match with colors of domains in a and b. d The same structure as c, but in this figure the positions of the mutated residues are indicated in red, the sidechains of these residues are shown as red balls. The ATP molecule is shown in yellow. This figure illustrates the clustering of mutations on specific sites within the Helicase ATP-binding domain and Helicase C-terminal domain. A more detailed analysis of the different missense mutations in our cohort can be found in Supplementary Note 1

The CHD3 protein is characterized by a SNF2-like ATPase/helicase domain, together with two plant homeodomain (PHD) fingers and two chromodomains (Fig. 2b, c)1,11, which mediate chromatin interactions and nucleosome remodeling1. The overwhelming majority of missense mutations (17/19) cluster within and around the ATPase/helicase motif, a functional domain that consists of two subdomains: a Helicase ATP-binding lobe and a Helicase C-terminal lobe. This domain provides energy for nucleosome remodeling through its ATPase activity. All missense mutations affect amino acids that are highly conserved, both in different species and also in the other CHD proteins that can be part of the NuRD complex (Supplementary Fig. 1), and clearly cluster in and around highly conserved SF2-family helicase motifs (Supplementary Fig. 1). All are predicted to be pathogenic by Polyphen-2 and/or SIFT, and have CADD scores above 24 (Supplementary Data 1).

The identified de novo mutations also include one in-frame deletion of one amino acid (p.Gly1109del) and two truncating mutations (p.Glu457* and p.Phe1935Glufs*108), although the latter causes a frameshift at the very end of the protein, leading to a stop codon after 108 amino acids. RNA sequencing of transcripts with and without cycloheximide showed that this mutation escapes nonsense-mediated decay (Supplementary Fig. 2). Finally, one case has a splice-site mutation (c.4073-2A>G) which is predicted to yield skipping of exon 27, while preserving the reading frame (Fig. 2a). Data from the ExAC database (http://exac.broadinstitute.com) indicate that CHD3 is extremely intolerant for loss-of-function mutations (loss-of-function intolerance score of 1.0) and highly intolerant for missense mutations (Z-score of +7.15)25, supporting the pathogenicity of the mutations that we found.

All CHD3 mutations were determined to be the most likely causal variant contributing to the disorder of the proband. In proband 15 who has a de novo CHD3 p.Asp1120His mutation, a de novo truncating mutation in CIC was also identified (NM_015125.3:c.1444G>T; p.Glu482*). Since truncating mutations in CIC were recently suggested as a potential cause of intellectual disability (ID)26, both mutations might be involved in the phenotype of this proband.

A subset of CHD3 mutations directly affects ATP hydrolysis

The striking clustering of almost all missense mutations in the ATPase/helicase domain of the CHD3 protein led us to hypothesize that disturbance of ATPase and/or chromatin remodeling activities of CHD3 could be potential pathogenic mechanisms. Three-dimensional modeling and mutation analysis of all missense mutations, including analysis of the conserved SF2-characteristic helicase motifs, demonstrates clear clustering of mutations and disturbance of important binding and interaction domains (Fig. 2d and Supplementary Note 1). Direct fluorescence imaging of mCherry-tagged CHD3 mutations in cellular models revealed no differences in subcellular localization for the mutated proteins as compared to wild-type CHD3 (Supplementary Fig. 3).

We experimentally assessed ATPase activity of six representative mutations, selected to include one mutation in the Helicase ATP-binding lobe and several mutations in the Helicase C-terminal lobe. FLAG-tagged full-length wild-type CHD3 protein and each of the six mutant proteins were transiently expressed in mammalian HEK293 cells and purified (Supplementary Fig. 4). Radiometric ATPase assays were performed to assess the activity of these mutant proteins relative to wild-type, in the presence of dsDNA (Fig. 3), recombinant nucleosomes (Fig. 3), or in the absence of DNA substrates as a control (Supplementary Fig. 5). ATPase activities of p.Arg1121Pro and p.Arg1172Gln were significantly lower than wild-type for both substrate conditions. These findings are consistent with the modeling data, since p.Arg1121Pro is predicted to disrupt a helix integral to motif V, while p.Arg1172Gln is located in helicase motif VI, and both motifs are known to be critical in ATP hydrolysis. The activity of p.Asn1159Lys was significantly lower only in the presence of dsDNA, although the reason for the different activity depending on the substrate is currently unknown. The protein with the p.Leu915Phe mutation, located in conserved SNF2-motif III, is significantly hyperactive under both conditions. The p.Arg1187Pro and p.Trp1158Arg mutations do not show statistically significant differences from the wild-type protein in these ATPase assays. According to the three-dimensional structure, the location of p.Arg1187Pro is not close to the ATP-binding or interaction surface. To assess whether mutant protein could impact activity of wild-type enzyme, we mixed wild-type protein with equimolar amounts of several mutant proteins, finding no biochemical evidence in this assay for interference (Supplementary Fig. 6).

Fig. 3
figure 3

ATPase assays. Radiometric ATPase assays were performed to assess the activity of the mutant proteins relative to wild-type, in the presence of recombinant nucleosomes (blue), dsDNA (green), or in the absence of DNA substrates as a control (Supplementary Fig. 4). Released phosphate was separated from unhydrolyzed ATP by thin layer chromatography, and detected by exposure to a phosphorimager. The experimental values (percentage hydrolyzed ATP) for the different mutant conditions were normalized to values for the wild-type condition within the experiment, to derive a normalized ATPase activity. The experimental data are presented as means ± standard deviation, individual data points are shown as red triangles. Three independent experiments from two individual purifications (wild-type, p.Leu915Phe, p.Arg1121Pro, p.Asn1159Lys, p.Arg1172Gln, and p.Arg1187Pro) (N = 6) or one purification (p.Trp1158Arg) (N = 3) were performed. Raw values from the individual experiments can be found in Supplementary Data 2. Asterisk (*) indicates significant difference for mutant values compared to wild-type values (unpaired t-test, P < 0.05) within the same substrate condition

CHD3 mutations disturb chromatin remodeling capacities

We measured the effects of six mutations on the chromatin remodeling activity of CHD3, by assessing restriction enzyme accessibility to nucleosomal DNA27. Consistent with its reduced activity in the ATPase assays, the p.Arg1172Gln mutant was partially, but not fully, active at chromatin remodeling (Fig. 4). p.Arg1121Pro, which showed severely reduced ATPase activity, was highly compromised in the chromatin remodeling assay. Moreover, p.Leu915Phe demonstrated hyperactivity in this assay, mirroring its elevated ATPase activity. Crucially, chromatin remodeling assays can also detect functional defects beyond ATP hydrolysis27. Two of the mutant proteins, p.Trp1158Arg and p.Asn1159Lys, exhibited severely compromised ability to remodel chromatin (Fig. 4) against a background of some preserved ATPase activity (c.f. Fig. 3). In sum, with the sole exception of p.Arg1187Pro, all the mutant versions of CHD3 that we tested differ from wild-type protein in their ability to remodel chromatin, with some mutants exhibiting decreased activity while one shows increased activity.

Fig. 4
figure 4

Restriction enzyme accessibility assay. a Restriction enzyme accessibility analysis of CHD3 wild-type and mutant proteins. 3.125, 6.25, or 12.5 nM of CHD3 proteins were incubated with 347 bp mono-nucleosomes. Digested fragments were analyzed by native polyacrylamide gel. b Quantitative analysis of restriction enzyme accessibility. Three individual experiments from two individual purifications (wild-type, p.Leu915Phe, p.Arg1121Pro, p.Asn1159Lys, p.Arg1172Gln, and p.Arg1187Pro) (N = 6) or one purification (p.Trp1158Arg) (N = 3) were conducted. The experimental data are presented as means with standard deviations

Discussion

In this study, we show that de novo CHD3 mutations cause a neurodevelopmental disorder. We demonstrate defining clinical features of this syndrome. The characteristic phenotype of individuals with CHD3 mutations overlaps with that reported for de novo mutations in CHD4, in which intellectual disability, macrocephaly, ventriculomegaly, undescended testes, and similar facial features have been reported. However, comparisons to the CHD4-related syndrome are currently limited because so far only five individuals with CHD4 mutations have been clinically characterized. Also interesting in this context is the fact that four of the six recently described patients with missense mutations in CHD1 have a diagnosis of speech apraxia9, a relatively rare condition. Although CHD1 does not function in the same protein complex as CHD3 and has different expression patterns9, there might be shared pathogenic mechanisms leading to speech problems in patients with mutations in these chromatin remodelers.

Based on the molecular and phenotypic data of individuals in our cohort, there is no obvious correlation between the precise type or location of the mutation, and the severity of the variable features of the resulting syndrome. However, the only individual in our cohort with epilepsy is also the only case with a missense mutation in the C-terminal domain of the protein. Future identification of more individuals with missense mutations in this region of the protein will help resolve whether this reflects a phenotype–genotype correlation.

In addition to defining the phenotype associated with CHD3 mutations, we aimed to characterize the effects of CHD3 mutations at a molecular and functional level. ATPase assays with six different mutant CHD3 constructs showed a clearly decreased ATPase activity for two mutations (p.Arg1121Pro and p.Arg1172Gln) and increased ATPase activity for one mutation (p.Leu915Phe). The disturbed ATPase activities are associated with corresponding effects on chromatin remodeling capacities for these three mutants, as shown by the restriction enzyme accessibility assays. It is currently unclear how deactivating and activating mutations can both yield similarly disruptive effects on neurodevelopmental outcomes. However, a recent study of cancer-specific mutations in the chromatin remodeling ATPase SMARCA4 concluded that mutations in the ATPase core of this enzyme had dominant-negative impacts on the global chromatin landscape regardless of whether they displayed increased or decreased dynamic recovery in fluorescence after photobleaching28. By analogy, it seems plausible that perturbed chromatin remodeling activity of CHD3, whether by gain or loss of activity relative to wild-type or by affecting its interactions, might likewise alter chromatin landscapes, to contribute to a neurodevelopmental phenotype.

Two mutations (p.Trp1158Arg and p.Asn1159Lys) show severely decreased chromatin remodeling capacities, despite unaffected ATPase activity in the presence of recombinant nucleosomes. In line with these findings, the highly conserved tryptophan residue at a position analogous to CHD3 residue 1158 has recently been shown to be critical for chromatin remodeling, but not for ATP hydrolysis, in the context of yeast SNF227. Interestingly, the mutation in our cohort affecting this amino acid (p.Trp1158Arg) directly matches the position of a previously published mutation in CHD48 (Supplementary Fig. 1), while the other previously published de novo missense mutations in CHD4-related syndrome are also mainly affecting the orthologous Helicase domain of CHD4 (Supplementary Fig. 1)8,29.

To systematically assess whether the distribution of the missense mutations in CHD3 reflects mutational hotspots in the gene, we performed a formal clustering analysis based on mutual distances, as previously described30. This analysis revealed significant clustering within the transcript (P = 0.0017), a finding that argues against simple haploinsufficiency as an underlying molecular mechanism. The paucity of patients with truncating mutations compared to the 31 patients with missense mutations in our cohort also supports this view, although the precise mechanistic effects of CHD3 mutations during neurodevelopment are a topic for future study.

Taken together, with our research we identify a recognizable neurodevelopmental disorder. We define the phenotypic spectrum associated with mutations in CHD3, and show the effects of several different mutations on ATPase activity and chromatin remodeling capacities. Our findings highlight the importance of chromatin remodeling factors, and specifically the CHD3 protein, in human brain development.

Methods

Individuals and consents

The authors affirm that (the legal representatives of) all human research participants provided informed consent for publication of the images in Fig. 1. Informed consent was also derived for the use of biological materials from all individuals or their legal representative. Genetic testing and research were performed in accordance with protocols approved by the local Institutional Review Boards where the patients were followed. Specifically, research exomes were performed after informed consent on protocols approved by the Institutional Review Boards of the following institutions: University of British Columbia, Augustana College, CHU Dijon, Mass General Hospital for Children, University of Erlangen-Nuremberg, Hamburg Chamber of Physicians, Cambridge South—UK Research Ethics Committee, University of Wisconsin-Madison Social & Behavioral Sciences.

Annotation of mutations

All mutations in this report are annotated in GRCh37 (hg19) and CHD3 transcript variant 1 (NM_001005273.2).

Next-generation sequencing

For the index case (individual 22), whole genome sequencing was performed using Illumina’s HiSeq X Ten technology, the Burrows–Wheeler Aligner (BWA) software version 0.7.8-r45531 and GATK v.3.432. In other individuals, exome or genome sequencing and data analysis were performed as previously described33,34,35,36,37,38,39,40,41,42,43,44.

Expression and purification of FLAG-CHD3

CHD3 proteins were prepared as previously described45, with the following modifications. FLAG-CHD3 constructs were cloned into expression vectors (kindly provided by Guang Hu) using Gateway Cloning technology. Primer sequences are provided in Supplementary Table 1. HEK293-f (ThermoFisher, FreeStyle™ 293-F Cells) were grown in suspension culture using FreeStyle™ 293 Expression Medium (ThermoFisher) in optimum growth flasks (Thomson) using a shaking incubator set at 8% CO2, 80% humidity, and 150 rpm shaker rate. The cell count was 106 cells/ml on the day of transfection. Cells were transfected with 1 mg of expression vector using PEI max (Polysciences). Cells were harvested 48 h after transfection by centrifugation at 400 × g for 6 min. Cells were washed once with phosphate buffer saline solution prior to storage at −80 °C or protein purification.

The cell pellet was resuspended in lysis buffer (20 mM HEPES, 1.5 mM MgCl2, 10 mM KCl, 1 mM DTT, 1 mM PMSF, and 1× cOmplete® protease-inhibitor EDTA-free (Roche), pH 7.6). Cells were incubated on ice for 30 min, vortexed briefly, and nuclei were collected by centrifugation (5 min, 3300 × g, 4 °C). The supernatant was discarded and the nuclear pellet was resuspended in nuclear extraction buffer (20 mM HEPES, 0.5 M KCl, 1.5 mM MgCl2, 0.2 mM EDTA, 20% glycerol, 0.2% NP-40, 1 mM DTT, 1 mM PMSF, and 1× cOmplete® protease-inhibitor EDTA-free (Roche), pH 7.6). The nuclear pellet was homogenized using a Dounce homogenizer, incubated on ice for 30 min, and insoluble material was removed by centrifugation (20 min, 110,000 × g, 4 °C). The supernatant (nuclear extract) was incubated with α-FLAG M2 affinity gel (Sigma-Aldrich) and rotated overnight at 4 °C. The α-FLAG beads were then washed twice with nuclear extraction buffer, followed by 2 additional washes with wash buffer (20 mM HEPES, 0.1 M KCl, 0.2% NP-40, 20% glycerol, and 1 mM DTT, pH 7.6). The FLAG-CHD3 protein was eluted with 0.3 mg/ml 3XFLAG peptide (in 20 mM HEPES, 0.1 M KCl, 0.05% NP-40, 20% glycerol, and 1 mM DTT, pH 7.6). Wild-type and mutant protein samples were analyzed by SDS-PAGE and stained with Coomassie Brilliant Blue (Supplementary Fig. 4). The concentration of the CHD3 proteins was estimated from BSA standards in SDS-PAGE gels stained with Coomassie Brilliant Blue.

Radiometric ATPase assay

Each ATPase reaction (10 μL) contained 20 mM Tris–HCl, pH 7.5, 1 mM MgCl2, 0.1 mg/ml BSA, 1 mM DTT, 100 μM ATP, 1 μCi of [γ-32P]ATP as a tracer. 25 nM of each CHD3 purified protein was incubated with 70 nM of recombinant nucleosomes or naked dsDNA. Nucleosome was reconstituted by the salt gradient dialysis method using recombinant histone octamer and 201 bp 601 DNA fragment46. The reactions were initiated by the addition of nucleosome or DNA substrate and incubated at 37 °C for 40 min. The reaction was quenched by the addition of EDTA to a final concentration of 100 mM. Aliquots (2.5 μL) were removed and spotted on PEI-cellulose thin-layer chromatography plates and developed in 1 M formic acid and 0.5 M LiCl. ATP hydrolysis was quantified using a Phosphorimager with Image Quant Software. For the mixing experiment, all reaction components except for CHD3 protein were incubated for 10 min at 37 °C, and the CHD3 protein mixture was added last to start the reaction. This experiment was performed 3 times per condition (N = 3) for all conditions, except for the conditions “no CHD3”, “WT 12.5 nM” and “WT 25 nM” (N = 2).

For the quantification analysis, we performed 3 individual experiments for each of the two biological replicates (total N = 6), except for the p.Trp1158Arg mutant (one biological replicate, total N = 3). An unpaired t-test was used to determine whether the activity of the mutant proteins differed significantly from wild-type protein activity.

Restriction enzyme accessibility assay

Remodeling activities were measured with a restriction enzyme accessibility assay as previously described27. 12.5 nM nucleosomes (347 bp) were incubated with the indicated amounts of CHD3 proteins at 37 °C for 60 min in the remodeling buffer (20 mM Tris–HCl pH 7.5, 1 mM DTT, 1 mM MgCl2, 1 mM ATP, 0.1 mg/ml BSA, and 5 U HhaI). The reactions were stopped by adding 2 µL of proteinase K buffer containing 6.7 mg/ml proteinase K, 167 mM EDTA, and 1.7% SDS. After incubation at 50 °C for 10 min, the DNAs were analyzed by 6% native polyacrylamide gel electrophoresis. The separated DNA fragments were visualized with UV light on the ChemiDox XRS system (BIO-RAD). The band intensities were quantified by ImageJ.

Cloning constructs for immunofluorescence

Wild-type CHD3 (NM_001005273.2) was amplified by PCR and cloned into pCR2.1-TOPO (Invitrogen) as described47. CHD3 mutation constructs were generated using the QuikChange II Site-Directed Mutagenesis Kit (Agilent), primer sequences are provided in Supplementary Table 1. CHD3 cDNAs were subcloned using BamHI/NheI restriction sites into a modified pmCherry-C1 vector (Clontech). All constructs were verified by Sanger sequencing.

Immunofluorescence

HEK293 cells were obtained from ECACC (Catalogue number 85120602) and grown in Dulbecco’s modified Eagle’s medium (Invitrogen), supplemented with 10% fetal bovine serum (Invitrogen). Transfection was performed using GeneJuice (Merck-Millipore). The cells were seeded onto coverslips coated with poly-L-lysine (Sigma). At 36 h post-transfection, cells were fixed using 4% paraformaldehyde solution (Electron Microscopy Sciences) for 10 min at room temperature. The mCherry fusion proteins were visualized by direct fluorescence, nuclei were visualized with Hoechst 33342 (Invitrogen). Fluorescence images were obtained using an Axiovert A-1 fluorescent microscope with ZEN Image Software (Zeiss).

Three-dimensional modeling

As no experimentally solved 3D-structure of CHD3 exists, we performed homology modeling using the modeling option with standard parameters in the YASARA48 & WHAT IF49 twinset. Several models of the ATPase/helicase domain were created. The best scoring model was based on template PDB-file 5JXR (sequence identity 41% over the aligned residues). We also studied the model based on PDB-file 3MWY (sequence identity 45%), which shows a more open conformation and contains an ATP substitute. These two models provided information about the relative position of the mutated residues in the different conformation of the protein complex.

Clustering analysis of missense mutations

The locations of observed de novo missense mutations were permutated 1,000,000 times over the cDNA of the CHD3 gene (RefSeq transcript: NM_001005273.2). The distances between missense mutations were adjusted to take into account the total size of the coding region of CHD3 (6003 bp). Then, the geometric mean (the nth root of the product of n of all distances separating the mutations) was calculated, giving an index of clustering, as previously described30. To circumvent a mean distance of 0 as the result of recurrent mutations, pseudocount (adding 1 to all distances and 1 to the gene size) was used. To avoid artificial deflation of the clustering P-value, only one of the recurrent mutations present in the sibling-pair (individuals 7 and 8) and twin-pair (individuals 20 and 21) were included for the analysis.