Disrupted transcriptional networks regulated by CHD1L during neurodevelopment underlie the mirrored neuroanatomical and growth phenotypes of the 1q21.1 copy number variant

Abstract

Distal 1q21.1 deletions and duplications are associated with variable phenotypes including autism, head circumference and height defects. To elucidate which gene(s) are responsible for the 1q21.1 duplication/deletion-associated phenotypes, we performed gene manipulation in zebrafish and mice. We modeled 1q21.1 duplication by overexpressing the eight human protein-coding genes in zebrafish. We found that only overexpression of CHD1L led to macrocephaly and increased larval body length, whereas chd1l deletion caused opposite phenotypes. These mirrored phenotypes were also observed in mouse embryos. Transcriptomic, cistromic, and chromatin accessibility analyses of CHD1L knock-out hiPSC-derived neuronal progenitor cells revealed that CHD1L regulates the expression levels and chromatin accessibility of genes involved in neuronal differentiation and synaptogenesis, including autism genes. Moreover, we found that CHD1L favors telencephalon development during forebrain regionalization by facilitating chromatin accessibility to pioneer transcription factors, including SOX2 and OTX2, while simultaneously compacting chromatin through its interaction with the repressor NuRD complex. Overall, our data reveal a novel role for CHD1L as a master regulator of cell fate and its dosage imbalance contributes to the neuroanatomical and growth phenotypes associated with the 1q21.1 distal CNV.

Graphical Abstract

Graphical Abstract.

Introduction

Copy number variants (CNV) are enriched in neurocognitive disorders, such as intellectual disability, schizophrenia, and autism spectrum disorders (ASD) [1–9], with a documented co-morbidity between CNV underscoring neurocognitive traits and micro/macrocephaly [10, 11]. To date, a total of 17 autism-associated recurrent CNV loci are referenced [retrieved from the Simons Foundation Autism Research Initiative (SFARI); https://gene.sfari.org/database/cnv/]. Among them, reciprocal deletions and duplications of the 1q21.1 (breakpoints 3–4; BP3–BP4) region are particularly prevalent in individuals with ASD.

The 1q21.1 region includes two CNV-prone regions, delimited by breakpoint BP2–BP3 and BP3–BP4, respectively, as well as deletion and duplication encompassing both intervals (categorized as class II rearrangements by Brunetti-Pierri in 2008) [4]. Proximal BP2–BP3 microdeletions are known as susceptibility factor for thrombocytopenia-absent radius syndrome (MIM: 274 000) and brain anomalies have also been observed in some carriers. The recurrent reciprocal 1q21.1 BP3–BP4 deletions (MIM: 612 474) and duplications (MIM: 612 475) also known as class I rearrangements or distal microdeletion/microduplication syndromes, have been found in individuals with syndromic autism [4, 12]. Variable phenotypes have been reported, including congenital heart defects, autism, intellectual disability (ID), attention-deficit/hyperactivity disorder (ADHD), schizophrenia, increased/decreased head circumference, and anthropometric variations [4]. The 1q21.1 distal deletion is associated with microcephaly and short stature, whereas the reciprocal duplication is associated with increased risk of macrocephaly and carriers tend to be in the upper height percentiles [13, 14]; suggesting a possible undergrowth/overgrowth phenotype. The 1q21.1 distal region contains eight protein-coding genes, including PRKAB2, FMO5, CHD1L, BCL9, ACP6, GJA5, GJA8, GPR89B. The human-specific NOTCH2-paralogs NOTCH2NLA/B provide the breakpoints of the 1q21.1 distal region. Among these genes, GJA5 and GJA8 encode gap junction proteins and missense variants at these loci have been reported to cause autosomal dominant atrial fibrillation and standstill (MIM: 614 049 and 108 770) and cataract (MIM: 11 200), respectively [15–19].

A major challenge in the interpretation of CNV encompassing several genes is the identification of the loci whose dosage sensitivity drives the phenotype(s) [20]. Only a handful of functional studies focusing on the 1q21.1 distal region syndromes have been reported [21, 22]. First, with the advent of new technologies for generating cellular models from human differentiated cells, a recent study indicated that human cortical neurons harboring the 1q21.1 distal deletion or duplication exhibit differentiation defects and altered electrophysiological property [21]. Second, a mouse modeling 1q21.1 distal deletion presents schizophrenia-like behavior [22]. Last, the human-specific NOTCH2NLA and NOTCH2NLB paralogs have been shown to control brain size in vitro [23, 24]; however, their candidacy as 1q21.1 phenotypic gene drivers remains unclear since no causative pathogenic single nucleotide variants have been reported for these genes in ASD cohorts.

Here, we aimed to identify the causal gene(s) that drive neuroanatomical and growth defects associated with the 1q21.1 distal CNV. To this end, we performed dosage perturbation of each of the 1q21.1 distal human gene during zebrafish development and identified CHD1L as a phenotypic driver. We then conducted transcriptomic, cistromic and chromatin accessibility analyses upon suppression of CHD1L in human-induced pluripotent stem cell (hiPSC)-derived neuronal progenitor cells (hNPC), followed by cell fate assessment in human cerebral organoids (hCO) and validation of pathogenicity of genetic variants in zebrafish.

CHD1L is a poly (ADP-ribose) and ATP-dependent remodeler with a key role in chromatin relaxation, and is implicated in several types of cancer [25–27]. Structurally, it features a two-lobed catalytic Snf2-like ATPase domain linked to a C-terminal macrodomain via a linker region responsible for CHD1L binding to histones. CHD1L is widely recognized for its involvement in DNA-repair, particularly with PARP1 during base excision repair (BER) and nucleotide excision repair [28–35]. Recent studies further clarified its role in early cellular response to DNA damage, revealing an auto-inhibitory mechanism arising from interactions between its macrodomain and the bilobate ATPase module [32, 36]. In addition to its function in DNA repair, Chd1l has been shown to play a critical role in mouse embryo implantation [37] and to promote neuronal differentiation of cultured human embryonic stem cells [38]. However, CHD1L has never been associated to neurodevelopmental syndromes. Here, we describe novel findings indicating that CHD1L plays a critical role as a co-transcription factor (TF) of key pioneer TFs to promote telencephalon fate in humans.

Overall, we found that dosage imbalance of CHD1L contributes to the 1q21.1 distal CNV-associated neuroanatomical and growth phenotypes. Furthermore, our study uncovers a hitherto unknown function of CHD1L that does not require its remodeling capability during human forebrain cell-fate decision, expanding its portfolio of functions beyond DNA repair and its association with cancer.

Materials and methods

Zebrafish strains and husbandry

Zebrafish (Danio rerio) were raised and maintained as described [39]. Adult zebrafish were raised in 15 l tanks containing a maximum of 24 individuals, and under a 14–10 h light–dark cycle. The water had a temperature of 28.5°C and a conductivity of 200 μS and was continuously renewed. The fish were fed three times a day, with dry food and Artemia salina larvae. Embryos were raised in E3 medium, at 28.5°C, under constant darkness. AB strain obtained from the European Zebrafish Resource Center (EZRC) was used as wild-type (WT) for this study, except for the NOTCH2NLA/B overexpression experiments for which AB strain obtained from the Zebrafish International Resource Center (ZIRC) was used as WT. Of note, a difference in baseline head size measurements between the two WT lines was observed (i.e. mean = 120 μm for WT AB from EZRC and mean = 140 μm for WT AB from ZIRC), which could potentially be explained by differences in genetic background. The mutant line chd1l^sa14029, carrying a C > T point mutation at genomic location Chr 6: 36 844 273 (GRCz11) leading to a premature stop was obtained from ZIRC (ZIRC, ZL10486.01). F2 mutant embryos were obtained from an in vitro fertilization with heterozygous mutant F1 sperm (TL background) and AB eggs. F2 chd1l^+/− fish were then inbred to generate F3 populations. All fish lines reproduce normally. Developmental stages of zebrafish embryos and larvae are indicated in the text and figures. For zebrafish embryos and larvae, both males and females were used since the sex can only be determined at 2 months of age. Genotyping of the chd1l mutants was performed by digestion with Taq alpha I restriction enzyme of the polymerase chain reaction (PCR) product generated with the following primers: forward 5′-CAGCGTCAGTTTTGCTACCC-3′; reverse 5′-CACCTGGATTGTTCTTGAGC-3′. In figures, chd1l^+/− refers to heterozygous chd1l ^sa14029/+ and chd1l^−/− refers to homozygous chd1l^{sa14029/sa14029}. All animal experiments were carried out according to the guidelines of the Ethics Committee of IGBMC and ethical approval was obtained from the French Ministry of Higher Education and Research under the number APAFIS#15025-2018041616344504 and #49694-2024050316122698.

In vivo analysis of gene expression and fish embryo manipulations

For overexpression experiments, the human WT messenger RNA (mRNAs; NOTCH2NLA, ENST00000362074.8; NOTCH2NLB, ENST00000593495.4; PRKAB2, ENST00000254101.4; CHD1L-202, ENST00000369258.8, hereafter termed CHD1L-FL; CHD1L-203, ENST00000369259.4, hereafter termed CHD1L-ΔLobe1; BLC9, ENST00000234739.8; ACP6, ENST00000583509.7; GJA5, ENST00000579774.3; GJA8, ENST00000369235.1; GPR89B, ENST00000314163.12; FMO5, ENST00000254090.9) were cloned into the pCS2 vector and transcribed using the SP6 Message Machine kit (Ambion). Similarly, we cloned and transcribed two additional truncated forms of CHD1L, that were hereafter termed CHD1L-FL-ΔMacro, and CHD1L-ΔLobe1-ΔMacro using primers indicated in Table 7. We injected 1 nl of diluted mRNA (50, 100, 200, 250 pg) into WT AB zebrafish embryos at the one- to two-cell stage. Injected embryos were fixed at 2–3 days post-fertilization (dpf) to perform whole-mount immunostaining or scored at 5 dpf for the head size and body length. Head size was assessed using two distinct 2D measurement proxies at 5 dpf: [1] the distance between the eyes (distance between the convex tips of the eyecups) and [2] the distance from the anterior-most part of the forebrain to the hindbrain on a dorsal view. The body length was measured from the head to the tail following the vertebral column and using inter-6 somites distance as proxies. All the experiments were repeated at least three times and statistical analysis were performed as indicated to determine the significance of the observed phenotypes.

High-resolution episcopic microscopy (HREM) and 3D reconstruction

Five days post-fertilization WT AB and chd1l^{sa14029/sa14029} homozygous larvae were fixed in Bouin’s solution, then washed and dehydrated with graded ethanol solutions (70%, 90%, 95%, 100%). The larvae were incubated in JB4-resin mixture (containing acridine orange and eosin)/ethanol (1:1), followed by incubation in pure resin mixture. The samples were embedded in fresh activated JB4 resin left to polymerize overnight at 4°C and harden in oven at 95°C. The blocks were sectioned using the Histo 3D system to generate data by repeated removal of 5 μm thick sections as described [40]. Resulting HREM data with a voxel size of 4 × 4 × 5 μm³ were generated from ∼250 aligned images. HREM data were analyzed using the Amira-Avizo™ software (Thermo Fisher Scientific) in segmentation mode to define brain volumes. The experiments were repeated twice, and statistical analysis were performed as indicated to determine the significance of the mutant phenotype.

Whole-mount immunostaining on zebrafish larvae

Two days post-fertilization embryos were fixed in 4% paraformaldehyde (PFA) overnight and stored in 100% methanol at −20°C. After rehydration in phosphate buffered saline (PBS), PFA-fixed embryos were washed in IF buffer [0.1% Tween-20, 1% bovine serum albumin (BSA) in PBS 1×] for 10 min at room temperature. The embryos were incubated in the blocking buffer (10% fetal bovine serum (FBS), 1% BSA in PBS 1×) for 1 h at room temperature. After two washes in IF Buffer for 10 min each, embryos were incubated in the first antibody solution, 1:750 antiphospho-histone H3 (ser10)-R, (Santa Cruz, sc-8656-R), in blocking solution, overnight at 4°C. After two washes in IF Buffer for 10 min each, embryos were incubated in the secondary antibody solution, 1:1000 Alexa Fluor donkey antirabbit IgG, in blocking solution, for 1 h at room temperature. At least 15 larvae were imaged per condition and z-stacks were acquired. We used the ImageJ software to generate a “Maximum Intensity” projection and scored the number of fluorescent cells using the ITCN v1.6. plugin. The experiments were repeated three times, and statistical analysis were performed as indicated to determine the significance of the phenotypes.

Whole-mount TUNEL assay on zebrafish larvae

TUNEL assay on 2 dpf embryos was performed using the ApopTag^® Fluorescein In Situ Apoptosis Detection Kit (Merck Millipore, S7110) according to a modified protocol. Dechorionated embryos were fixed in 4% PFA at 4°C overnight and then stored in 100% methanol at −20°C for at least 2 h. After rehydration in PBS, embryos were permeabilized with proteinase K (10 μg/ml) in PBS for 5 min at room temperature and then washed three times in sterile water for 3 min each. Then, embryos were post-fixed with 4% PFA for 20 min at room temperature and followed by prechilled ethanol:acetic acid (2:1) for 10 min at −20°C. Embryos were washed in PBS-T (PBS 1×, 0.1% Tween-20) for 5 min, three times at room temperature. Incubation in equilibration buffer and next steps of the TUNEL were followed according to the manufacturer instructions. At least 15 larvae were imaged per condition and z-stacks were acquired. We used the ImageJ software to generate a “Maximum Intensity” projection and scored the number of fluorescent cells in defined regions of the zebrafish head (forebrain, midbrain, eyes) using the ITCN 1.6. plugin. The experiments were repeated three times, and statistical analysis were performed as indicated to determine the significance of the mutant phenotype.

Acridine orange whole-mount staining on zebrafish larvae

Acridine orange experiment was performed on 2 dpf embryos. The embryos were incubated at 28°C for 30 min in E3 embryo medium supplemented with 2 μg/ml Acridine Orange solution (Sigma, A231). After extensive washing, embryos were anesthetized with tricaine and imaged as Z-stacks with GFP green light excitation. Cell counting was performed with Fiji using the ITCN plugin coupled to manual counting. The experiments were repeated three times, and statistical analysis were performed as indicated to determine the significance of the mutant phenotype.

CRISPR/Cas9 genome engineering in zebrafish

CRISPR (clustered regularly interspaced short palindromic repeats) guide RNA (gRNA) was designed with ChopChop online tool. The gRNA targets exon 3 of D. rerio chd1l: 5′-ACTGGAGGCAAGAGTTGGAACGG-3′. The oligo was cloned into the pT7 vector and gRNAs was synthesized using the Mega transcript kit (Thermo Fisher) and purified as described [41]. To assess CRISPR efficiency, the targeted region was amplified by specific primers from total DNA as described [41] from 2 dpf F0 injected zebrafish embryos (F: 5′-ATATTGCTTGCTTTTAAGCCGA-3′; R: 5′-GGGGCCAAACATTTATGAATTA-3′). PCR products were denaturated and re-annealed as follows: 95°C for 2 min, cooling progressively to 85°C (−2°C per second) and then 25°C (−0.1°C per second to 25°C) and finally 16°C. The PCR products were run on Criterion precast PAGE gels (Bio-Rad) to visualize homo- and heteroduplexes, indicative of small indels. To estimate the fraction of CRISPR mosaicism, one control and six injected embryos were picked to be Sanger sequenced. From the PCR step above, DNA was purified by gel extraction (QIAquick gel extraction kit) and the fragments were cloned into TOPO4 vector (Invitrogen). Plasmid DNA was extracted from 10 bacterial clones per embryo and the inserts were Sanger sequenced with M13 primers. Fertilized eggs were injected at one-cell stage with a cocktail of 150 pg chd1l gRNA and 150 pg Cas9 protein. Mosaicism percentage was determined as the number of sequences carrying an indel at the targeted guide locus sequence over the total number of sequences read.

esiRNA knockdown validation

Neuro-2A cells were plated at a quantity of 10⁵ cells per well in a 12-well plate. Twenty-four hours later, cells were transfected with 3 μl of lipofectamine RNAiMAX (invitrogen) and 30 pmol of esi Rluc or esi Chd1l. Medium was changed 24 h after transfection. The next day, cells were lysed in Lysis Buffer [RadioImmunoPrecipitation Assay buffer (RIPA), 50 mM Tris–HCl (pH 8), 150 mM NaCl, 0.5% Triton X-100, and 1× Protease Inhibitor Cocktail (PIC)]. Proteins were purified and were diluted 1 × final with Laemmli buffer and dithiothreitol (DTT) 0.1 M, boiled for 10 min and then separated by sodium dodecyl sulphate–polyacrylamide gel electrophoresis (SDS–PAGE). A total of 30 μg of protein per lane were loaded on a 10% acrylamide gel. Resolved proteins were transferred to nitrocellulose membranes and blocked in 10% milk 1 × Tris Buffered Saline (TBS)-Tween 0.1% for 1 h at room temperature prior to incubation with mouse anti-Chd1l (1 μg/ml, in house) or mouse anti-GAPDH (1/1000, Abcam) antibodies. Membranes were washed and incubated in goat antimouse peroxidase secondary antibody (1:1000, Invitrogen). Blots were developed using SuperSignal West Pico PLUS (Fisher Scientific) according to the manufacturer’s instructions. Experiment was performed three times.

Mouse in utero electroporation

In utero electroporation was performed using adult CD1 mice (Charles River) as previously described [42]. Animal experimentations were performed at the IGBMC animal facilities in accordance with European Union and French legislation (APAFIS#17876–2018112916498297). The light/dark cycle for animals housed was 12 h and all experiments were conducted during the light cycle. Timed pregnant mice (E13.5) were anesthetized with isoflurane (2 l per min of oxygen, 4% isoflurane during sleep and 2% isoflurane during surgery operation; Tem Sega). The uterine horns were exposed, and a lateral ventricle of each embryo was injected using pulled glass capillaries (Phymep, 30-0016) and microinjector femtojet 4i (Eppendorf) with Fast Green (2 μg/ml; Sigma) combined with the following Endofree DNA plasmid solution: 1 μg of p-CIG-eGFP together with (i) 1 μg of pCAGGs-IRES-GFP empty or (ii) 1 μg of WT mouse Chd1l complementary DNA (cDNA; ENSMUST00000029730.4) cloned in pCAGGs-IRES-GFP or (iii) 100 ng/μl of either control Rluc esi RNA (Sigma–Aldrich, EHURLUC) or CHD1L targeting esi RNA (Sigma–Aldrich, EMU070191). Plasmids were further electroporated into the neuronal progenitors adjacent to the ventricle by discharging five electric pulses at 30 V for 50 ms at 950 ms intervals using an ECM 830 electroporator (BTX) and 3 mm electrodes. After electroporation, embryos were placed back in the abdominal cavity and development was allowed to continue until E14.5. Embryos brains were dissected and fixed in 4% PFA in PBS overnight. Brains were then rinsed in 20% sucrose to cryopreserve the tissues and embedded in sakura tissue-tek and stored at −80°C. Sectioning was performed using cryostat (Leica) at 14 μm. Immunostainings were done using following antibodies concentrations: Stainings were performed using Anti-GFP (Abcam, ab6673, 1:500) + Donkey antigoat 488 (Thermo Fisher, A-11055, 1:1000) and Anti-TBR1 (Abcam, ab31940, 1:250) + Donkey antirabbit 555 (Thermo Fisher, A-31572, 1:1000). Image acquisition was done with Leica confocal microscope SP5. Analysis was done by manual counting using Fiji. Electroporated and imaged embryos were obtained from at least three different litters, and statistical analysis were performed as indicated to determine the significance of the phenotypes.

Cell culture, DNA transfection and single cell isolation by fluorescence-activated cell sorting

Control hiPSCs (hiPSC GM8330-8) derived from adult fibroblasts, were kindly provided by Prof. M.E. Talkowski. The cells were maintained on Matrigel-coated dish (Corning) with mTESR™ (StemCell) and incubated at 37°C in a humidified atmosphere with 5% CO₂. We transfected the human iPSCs with the pSpCas9 (BB)-2A-GFP gRNA plasmid using Lipofectamine™ Stem Reagent, adapting the protocol described in [43]. At 48 h post-transfection, the hiPSCs were dissociated into a single cell suspension with Accutase and resuspended in PBS with 10 μM ROCK inhibitor (Santa Cruz). All samples were filtered through 5 ml polystyrene tubes with 35 μm mesh cell strainer caps (BD Falcon, 352 235) immediately before being sorted. After adding the viability dye 4′,6-diamidino-2-phenylindole (DAPI) (BD Bioscience), single GFP + DAPI- cells were isolated by fluorescence-activated cell sorting (FACS) gated for a high level of GFP expression and sorted, with one cell placed into each well of Matrigel-coated 96-well plates by BD FACS Aria II with 100-mm nozzle under sterile conditions. The medium was supplemented with CloneR™ (StemCell) from day 0 to day 4 according to the manufacturer’s instructions.

hiPSC-editing gRNA design and preparation

We used the CRISPR MIT tool (http://crispr.mit.edu) to generate a gRNA targeting the exon 1 of CHD1L (5′-TCATACTGAGGGCCGAGCCGAGG-3′, chr1: 147 242 763–147 242 785, GRCh38). The gRNA was cloned into pSpCas9 (BB)-2A-GFP (Addgene, PX458) plasmid. Validation of the guide sequence in the gRNA vector was confirmed by Sanger Sequencing. Before transfection, all plasmids were purified from PureLink™ HiPure Plasmid Midiprep Kit according to the manufacturer’s instruction (Thermo Fisher Scientific).

Colony screening and western blot validation

Genomic DNA from two thirds of each hiPSC colony (obtained ∼14 days after sorting) were extracted by using Quick-DNA 96 kit (Zymo research) and screened by PCR (using the indicated primers: forward 5′-GGAAGTTGGGAGGGAGGT-3′ and reverse 5′-GCTGATCTCACCACGTTTCC-3′) followed by Sanger sequencing. For hiPSC screening validation, 100 μg of total iPSC protein lysate was prepared in RIPA buffer and Protease Inhibitor Cocktail from control and two CHD1L-edited iPSCs lines, diluted 1 × final with Laemmli buffer and DTT 0.1 M, boiled for 5 min, and then separated by SDS–PAGE on 10% polyacrylamide gels. Resolved proteins were transferred to nitrocellulose membranes and blocked in 3% milk 1 × TBS for 1 h at room temperature prior to incubation with either anti-CHD1L (2170C3a) antibody (Santa Cruz, sc-81065, 1:200) or anti-β-Tubulin (1:10 000, produced in house). Membranes were washed and incubated in goat antimouse peroxidase secondary antibody (Jackson Immuno Research, 1:10 000). Blots were developed using Immobilion Western (Millipore, France) according to the manufacturer’s instructions.

Karyotyping

Karyotypic analysis was performed by Cell Guidance Systems (Cambridge, UK) following standard methods.

Differentiation of hiPSCs into neural progenitor cells (hNPC)

Cells were differentiated into hNPC utilizing the StemXVivo Neural Progenitor Differentiation Kit (R&D systems). The expression of CNS-type hNPCs markers was validated by immunocytochemistry, as described in the Fixing and Staining procedure of the StemXVivo datasheet, using anti-Human SOX1 included in the kit and Sheep Anti-Human Pax6 Polyclonal Antibody (R&D Systems, AF8150) (10 μg/ml in PBS) followed by incubation with Chicken anti-Goat IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 488 and Donkey anti-Sheep IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 594 (Thermo Fisher Scientific, 1:200), respectively.

RNA-seq library preparation and bioinformatic analyses

Total RNA was extracted from three biological replicates per condition (control, Mutant Line 1 and Mutant Line 2) using RNA Plus Mini Kit (Macherey Nagel). Libraries were prepared using the Illumina TruSeq stranded mRNA protocol and sequenced on an Illumina HiSeq 4000 (single-end 50 bp reads). Reads were preprocessed using cutadapt (v1.10) [44] in order to remove adapter, polyA and low-quality sequences (Phred quality score below 20) and reads shorter than 40 bases were discarded for further analysis. Reads mapping to rRNA were also discarded. This mapping was performed using bowtie (v2.2.8) [45]. Reads were then mapped onto the hg38 assembly of Homo sapiens genome using STAR (v2.5.3a) [46]. Gene expression quantification was performed using htseq-count (v0.6.1p1) [47], with annotations from Ensembl version 98. Normalization of read counts and differential expression analysis between controls and CHD1L-edited samples were performed using the method proposed by Love et al. [48] and implemented in the Bioconductor package DESeq2 (v1.16.1), adjusting for batch effect. For comparison among datasets, transcripts with >50 raw reads were considered. The common differentially expressed genes (DEG) discussed in this manuscript are those that show differential expression in both CHD1L mutant lines compared to the control line, and exhibit the same direction of change (either upregulated or downregulated) in both mutants. For STRING analysis, the online tool was used (string-db.org). The database DECIPHER was also used for gene annotations (https://www.deciphergenomics.org/). The top 500 DEG (both upregulated and downregulated and filtered based on False Discovery Rate (FDR)) were entered into Online tool Lisa.cistrome for Transcription Regulators prediction (lisa.cistrome.org). IGV v2.17.4 was used to generate sashimi plot. For Gene Ontology terms analysis, the online tool PANTHER was used (geneontology.org), and illustrated using R package ggplot2.

ATAC-seq library preparation

Assay for transposase accessible chromatin was performed on hiPSC-derived hNPC using a derived protocol from Buenostro et al. [49]. A total of 50 000 cells from GM8330-8 (control line CHD1L^+/+), Line 1 and Line 2 (isogenic CHD1L mutant lines), lines (a total of four biological replicates were used per condition) were resuspended in resuspension buffer [10 mM Tris–HCl (pH 7.5), 10 mM NaCl, 3 mM MgCl₂] and then lysed in lysis buffer (Resuspension buffer + 0.1% NP-40 + 0.1% tween-20 + 0.01% Digitonin) and incubated 3 min on ice. Then, 1 ml of wash buffer was added (resuspension buffer + 0.1% Tween-20). Cells were centrifugated 10 min at 500 × g at 4°C and supernatant was discarded. Transposition reaction mix (Illumina, Tagment DNA Enzyme and Buffer Small Kit, 20 034 197) was added to pellet and incubated 30 min at 37°C in a thermomixer at 1000 rpm. DNA fragments were isolated using Qiagen MinElut Reaction Cleanup kit. DNA fragments were amplificated and libraries were generated by PCR using appropriate primers as described in^{54, 55} and NEBNext High-Fidelity 2× PCR Master mix (NEB, M0541S). Finally, libraries were purified using SRIselect beads (Beckman Coulter) by a one-sided purification to remove primers. Sequencing was performed using Illumina HiSeq 4000 with 100 bp paired-end sequencing.

ATAC-seq bioinformatic analyses

Data analysis was performed using the Encode ATAC-seq pipeline (v1.4.2). Adapter sequences were removed and low-quality ends were trimmed. Reads were mapped onto the hg38 assembly of Homo Sapiens genome using Bowtie2 (v2.2.6) [45] choosing the zero multi-mapping option. Mitochondrial reads were removed. The Peak calling was performed using MACS2 (v2.1.1.20160309). Finally, the optimal overlap peaks were used for downstream analyses (total number of peaks: 161, 696). Peaks from different conditions were merged to form a consensus peak set. The peaks were annotated using annotatePeaks.pl script in HOMER program [50] and with Ensembl 98 database. The read coverage for each sample was calculated with multicov function from bedtools program (v2.26.0). Differential analyses of Control GM8330-8 versus either Line 1 or Line 2 isogenic mutants were performed using the Bioconductor package DESeq2 (v1.16.1) [48]. The common differentially accessible peaks discussed in this manuscript are those that show differential accessibility in both CHD1L mutant lines compared to the control line, and exhibit the same direction of change (either more accessible or less accessible) in both mutants. Comparisons between differentially accessible region (DARs) and DEGs were performed using gene set enrichment analyses (GSEA; R package clusterProfiler_4.14.6). Four gene sets were constructed by selecting the top 100 genes with the lowest P-values from either upregulated or downregulated genes in RNA-seq data comparing Mutant Line 1 versus WT and Mutant Line 2 versus WT. For the ATAC-seq data, one peak per gene was selected based on the closest distance to the gene’s transcription start site (TSS). Peaks were pre-ranked using the formula sign (logFC) × −log₁₀ (P-value), following the DESeq2 analysis. For the TOBIAS analysis (Transcription factor occupancy prediction by investigation of ATAC-seq signal) of enriched motif elements, the pipeline snakemake (v0.12.11) was used [51]. ATACseq datasets were compared to H3K4me2 peaks from CUT&RUN analysis performed in this study using seqMINER [52]. Representative traces were generated using Figeno [53].

CUT&RUN library preparation

Protocol was adapted from Hainer and Fazzio [54]. For each condition [GM8330-8 control CHD1L^+/+ line, Mutant Line 1 and Mutant Line 2 (isogenic CHD1L mutant lines) lines], hNPC were harvested and pooled from three culture wells per condition. Total of 1.10⁵ hNPC cells were washed with in cold PBS 1× and resuspended in cold Nuclear Extraction Buffer (20 mM HEPES–KOH, pH 7.9; 10 mM KCl; 0.5 mM spermidine; 0.1% Triton X-100; 20% glycerol; PIC 1×) and incubated for 5 min at 4°C. The nuclei obtained were pelleted and recovered in cold NEB. Concanavalin A magnetic beads (BioMagPlus, 86057-3, 25 μl bead slurry/sample) were washed twice in cold Binding Buffer (20 mM HEPES–KOH, pH 7.9; 10 mM KCl; 1 mM CaCl₂; 1 mM MnCl₂; 1× PIC) and recovered in Binding Buffer. The beads were added to the nuclei with a gentle vortexing and incubated for 10 min on the wheel at 4°C. Bead-bound nuclei were incubated for 5 min at room temperature in Blocking Buffer [1 ml Wash Buffer (20 mM HEPES–KOH, pH 7.5; 150 mM NaCl; 0.5mM Spermidine; 0.1% BSA; PIC 1×), 4 μl 0.5M ethylenediaminetetraacetic acid (EDTA)] and washed in Wash Buffer. Nuclei were re-suspended in 250 μl of cold Wash Buffer per condition. An antibody solution (250 μl) containing the primary antibodies anti-H3K4me2 or anti-CHD1L diluted at 1:100 was added to the bead-bound nuclei with a gentle vortexing. Samples were incubated overnight at 4°C on a wheel. The beads were then washed with cold Wash Buffer and protein A–micrococcal nuclease recombinant protein (pA–MN) enzyme (0.7ng/ml; 200 μl per condition) was added by gentle vortexing. Samples were incubated with rotation at 4°C for 1 h. The pA–MN was produced in-house according to the protocol described by Schmid et al. [55] and using the pK19pA–MN plasmid (RRID: Addgene_86 973; http://n2t.net/addgene:86973). The nuclei were then washed and re-suspended in Wash Buffer. The samples were incubated for 10 min in equilibrated water at 0°C. Cleavage was initiated by the addition of 100 mM CaCl₂ with gentle vortexing and incubated for 30 min. The reaction was stopped by adding 2× Stop Buffer (5 M NaCl; 0.5 M EDTA; 0.2 M Ethylene glycol tetraacetic acid (EGTA); 50 μg/μl RNAseA; 40 μg/ml glycogen) and DNA fragments were released by passive diffusion during incubation at 37°C for 20 min. After centrifugation for 5 min at 16 000 × g at +4°C to pellet cells and beads, 3 μl 10% sodium dodecyl sulphate and 2.5 μl proteinase K 20 mg/ml were added to the supernatants, and samples were incubated 10 min at 70°C. DNA purification was done with phenol/chloroform/isoamyl alcohol extraction followed by a second chloroform extraction. DNA was precipitated with ethanol after addition of 20 μg glycogene and resuspended in 0.1× Tris EDTA. DNA fragments were amplificated and library was generated by PCR using MicroPlex Library Preparation Kit v3 (Diagenode, C05010001). Finally, libraries were purified using SRIselect beads (Beckman Coulter) by a one-sided purification to remove primers. Sequencing was performed using NextSeq 2000 with 50 bp paired-end sequencing.

CUT&RUN bioinformatic analyses

Paired-end reads were mapped to Homo Sapiens genome (assembly hg38) using Bowtie2 (v2.3.4.3, parameters: -N 1 -X 1000). Reads overlapping with ENCODE hg38 blacklisted region V2 were removed using bedtools. Bigwig tracks were generated using bamCoverage. Tracks were normalized with RPKM method. The ≤120 bp fragments were used for samples obtained with anti-CHD1L. Peak calling was performed with the Sparse Enrichment Analysis for CUT&RUN (SEACR v1.3) tool using a numeric threshold of 0.002 for CHD1L (https://seacr.fredhutch.org), and with MACS2 for H3K4me2 (broad peaks, q-value 0.05, other parameters as default). Peak annotation and genomic features [promoter/TSS, 5′ UTR, exon, intron, 3′ UTR, transcription termination site (TTS), and intergenic regions] were defined and calculated using Refseq and annotatePeaks.pl script in Homer program according to the distance to the nearest TSS (default range from −1 kb to +100 bp) and default settings were applied. A single nearest gene was assigned per peak. HOMER and RSAT were used for motif search [56]. Heatmaps were generated with deeptools [57]. ChIP-Atlas was used for TF enrichment (chip-atlas.org) [58] using the following parameters: ChIP TFs and others for Experiment type, Neural for Cell type Class, 50 for Threshold for significance, and random permutation. CHD1L CUT&RUN peaks were compared to the following ChIP-seq data: SOX2 peaks from human neural progenitor cells (SRX330107), CHD7 peaks from human cerebellum (SRX9795022), HDAC2 (SRX19212560) and RBBP4 (SRX19212562) peaks from human neuroblastoma. Representative peaks were generated using Figeno [53] on the reference human genome hg38. Intersections between datasets were performed with Venny 2.1 (https://bioinfogp.cnb.csic.es/tools/venny/). WebGestalt was used to run DisGeNET enrichment analysis for CHD1L and SOX2 common gene targets (WebGestalt.org).

Immunoprecipitation and Liquid Chromatography Tandem Mass Spectrometry (LC-MS/MS)

Control CHD1L^+/+ and mutant CHD1L^−/− (Mutant Line 1) hNPCs were harvested in PBS and centrifugated at 300 × g for 5 min. Pellets were lysed for 30 min under agitation in lysis buffer [Tris (pH 7.5) 50 mM, NaCl 150 mM, glycerol 5%, NP-40 1%, PIC 1× in H₂0]. Immunoprecipitation were performed using Slurry Dynabeads G coupled to 5 μg of Anti-CHD1L antibody for 1 h at 4°C under agitation. On ice, 100 μg of proteins were added on antibody-coupled beads and agitated at 4°C overnight. After three washes in lysis buffer, bound proteins were eluted in 1× blue Laemmli and 0.1M DTT and boiled. Experiment was repeated three times for each genotype. The isolated proteins were subjected to an overnight trypsin digestion at 37°C. The peptides were separated on a C18 column using an Ultimate 3000 nano-RLSC (Thermo Fisher Scientific) coupled to an LTQ-Orbitrap Elite mass spectrometer (Thermo Fisher Scientific) for peptide identification. Proteins were identified with Proteome Discoverer 2.4 software (Thermo Fisher Scientific, PD2.4 [59] from Homo Sapiens proteome database (Swissprot, reviewed, release 2021_06_03, 20 309 entries), and proteins were quantified with a minimum of two unique peptides based on the XIC (sum of the Extracted Ion Chromatogram). Partners of CHD1L were considered if enriched (fold-change >1) in control condition compared to the protein listed in the mutant condition to exclude nonspecific bindings. Protein identified were filtered for nuclear protein only using BioDBnet.org [60] db2db tool. For visual representation and Gene Ontology terms analysis, the online tool STRING.db [61] was used, and illustrated using R packages GGplot2.

Co-immunoprecipitation in HEK cells

CHD1L and SOX2 cDNA were cloned in pCDNA3.1 plasmids with either N-terminal hemagglutinin (HA) or FLAG tag sequence, respectively. Plasmids coding for HA-CHD1L or FLAG-SOX2 were transfected in HEK293-T using lipofectamine 2000 (0.5 μg alone or 1.5 μg when co-expressed). The next day, cells were lysed in Lysis Buffer [RIPA, 50 mM Tris–HCl (pH 8), 150 mM NaCl, 0.5% Triton X-100, and 1× PIC]. Samples were sonicated and centrifuged for protein purification. Proteins were incubated with anti-HA magnetic beads overnight at 4°C with agitation. After washes, bound proteins were diluted 1 × final with Laemmli buffer and DTT 0.1 M, boiled for 10 min and then separated by SDS–PAGE. Resolved proteins were transferred to nitrocellulose membranes and blocked in 5% milk 1 × TBS-Tween 0.1% for 1 h at room temperature prior to incubation with rabbit anti-HA and rat anti-FLAG antibodies. Membranes were washed and incubated in goat antirabbit and Goat antirat peroxidase secondary antibody (1:10 000, Jackson Immuno Research, U.K.). Blots were developed using Immobilon Western (Millipore, France) according to the manufacturer’s instructions. Experiment was performed three times.

Generation of hCO and immunostaining

hCO were generated using the STEMdiff Cerebral Organoid Kit (STEMCELL, 08 570) and following the manufacturer’s instructions, and matured until day 60 in vitro. For immunostainings, cerebral organoids were fixed in 4% PFA overnight at 4°C, after PBS washes, cerebral organoids were incubated in PBS-sucrose 30% until sedimentation. The hCO were embedded in 1:1 OCT and sucrose 30% and stored at −80°C. Cryosections were performed at 14 μm. Sections were hydrated in PSB 1× for 10 min then permeabilized with 0.1% Triton in PBS for 20 min at room temperature (RT) and were incubated 1 h with Blocking Solution (10% fetal calf serum (FCS) + PBS-Triton 0.3%). Sections were incubated overnight at 4°C with primary antibodies anti-SOX2 (Santa Cruz, sc-365823, 1:200) and anti-TUJ1 (D71G9) (Cell Signaling technology, 5568S, 1:500) in blocking solution. After extensive washing, sections were incubated in blocking solution with secondary antibodies antimouse 488 (Thermo Fisher, A1101, 1:500) and antirabbit 647 (Thermo Fisher, A21244, 1:500). Images were acquired using Leica Image acquisition Leica confocal microscope SP8X. Image reconstruction was done using FiJi.

Western blot for CHD1L and TUJ1 in cerebral organoids

hCO were snap frozen using liquid N2 and lysed in 200 μl of RIPA buffer supplemented with Protease Inhibitor Cocktail. Proteins were purified and quantified, then resuspended in 1 × final with Laemmli buffer and DTT 0.1 M, boiled for 5 min and 15 μg of proteins was separated by SDS–PAGE on 10% polyacrylamide gels. Resolved proteins were transferred to nitrocellulose membranes and blocked in 3% milk 1 × TBS for 1 h at room temperature prior to incubation with either anti-CHD1L (2170C3a) antibody (Santa Cruz, sc-81065, 1:200), anti-β tubulin (1:10 000, produced in house), and anti-TUJ1 (D71G9) (Cell Signaling Technology, 5568S, 1:1000). Membranes were washed and incubated in goat antimouse peroxidase secondary antibody (Jackson Immuno Research, 1:10 000) or goat antirabbit (for TUJ1) peroxidase secondary antibody (Jackson Immuno Research, 1:10 000). Blots were developed using Immobilon Western (Millipore, France) according to the manufacturer’s instructions.

Single-nuclei multiome on hCO

hCO were maintained until day 60 in vitro. Nuclei isolation was followed the protocol “Nuclei Isolation from Complex Tissues for Single Cell Multiome ATAC + Gene Expression Sequencing” (10Xgenomics, protocol CG000375 Rev C) without the cell sorting step. Specifically, a total of six cerebral organoids from three independent culture wells per condition, control CHD1L^+/+ and CHD1L^−/- (Mutant Line 1), were washed in PBS 1×. Lysis were performed using 1× NP-40 Lysis Buffer with mechanical dissociation of tissues. Samples were filtered through 70 μm strainer and centrifugated for 5 min at 4°C 500 rcf. Pellets were re-suspended in PBS + 1% BSA + 1U/μl RNAse Inhibitor and incubated for 5 min. After a second centrifugation, pellet was re-suspended in 0.1× Lysis Buffer and incubated 5 min on ice. Wash buffer was added, and samples were centrifugated before a final resuspension in Diluted Nuclei Buffer. Libraries were generated following the 10X Genomics Multiome Guidelines (protocol vA.01) and the protocol “Chromium Next Gem Single Cell Multiome ATAC + Gene expression User guide” (CG000338 Rev E). Sequencing was performed using Illumina NextSeq 2000 with 50 bp paired-end sequencing for ATAC-seq and 28 + 85 bp paired-end sequencing for RNA-seq. BCLconvert (v3.8.4), Cell Ranger ARC (v2.0.2) and the human reference cellranger-arc-GRCh38-2020-A-2.0.0 were used for demultiplexing, alignment, barcode and UMI filtering, and counting. Control CHD1L^+/+ and CHD1L^−/− samples were aggregated using cellranger-arc aggr function without performing the depth normalization.

Seurat analysis of snMultiome

10x Genomics multiomic data were processed using Seurat (v4.3) and Signac (v1.1) [62]. The output of the Cell Ranger arc pipeline was read using the Read10X function of Seurat (R, v4.2.2) to obtain a matrix of the number of UMIs of each gene detected in each cell. A total of 4019 cells, including 2385 cells from CHD1L^+/+ organoids and 1634 cells from CHD1L^−/− organoids, were retained after filtering based on RNA-assay metrics (total counts >1000 and <50 000, percent.mt < 20) and ATAC-assay metrics (total counts >2000 and <500 000). We applied SCTransform to normalize RNA counts and TFIDF to normalize ATAC peaks. Dimensionality reduction was performed using Latent Semantic Indexing (LSI) for ATAC data and principal component analysis (PCA) for RNA data. We constructed a weighted nearest neighbor (WNN) graph with Seurat (v4.0) [63] using 2–50 ATAC LSI dimensions and 1–50 RNA PCA dimensions. We used the FindClusters function with the SLM algorithm and a resolution of 0.8 to identify the clusters. DEG for RNA-seq analysis were calculated using the FindAllMarkers function with the Wilcoxon test. For the Sankey diagram, a threshold of read number >1 was used to count the number of positive nuclei expressing FOXG1 and SIX3. Representative figures for genotype or marker gene expressions were generated using R and ggplot2.

SCENIC + analysis of snMultiome

Multiomic datasets were subjected to the SCENIC + workflow (v0.1.dev456 + g9662363) as described by Bravo Gonzalez-Blas et al., 2023 [64]. Topic modeling of scATAC-seq data was carried out using pycisTopic (v1.0.2.dev15 + g242c2a4). Consensus peaks were called using Macs2 (total of 362, 610 peaks), and topic modeling was performed using Latent Dirichlet Allocation with the collapsed Gibbs sampler. A model of 16 topics was selected based on stabilization of metrics. Motif enrichment analysis was conducted using pycisTarget (v1.0.2.dev11 + g7daf370). The hg38 v10 motif collection was downloaded from the cistarget resources website (https://resources.aertslab.org/cistarget/). Motif enrichment was performed using both the cisTarget and DEM algorithms on cell line-based DARs with default thresholds. The analysis was run both including and excluding promoters, defined as regions within 500 bp upstream or downstream of the TSS of each gene. The SCENIC + workflow was executed using default parameters. A search space of a maximum between either the boundary of the closest gene or 150 kb and a minimum of 1 kb upstream of the TSS (ensembl release 98) or downstream of the end of the gene was considered for calculating region–gene relationships using gradient-boosting machine regression. TF–gene relationships were calculated using gradient-boosting machine regression between all TFs and all genes. Genes were considered as TFs if they were included in the TF list available on http://humantfs.ccbr.utoronto.ca/ (v1.01) [65]. Final eRegulons were constructed using the GSEA approach, and only eRegulons with a minimum of ten target genes were retained. For each eRegulon, cellular enrichment scores [area under the recovery curve (AUC)] of target genes and regions were calculated using the AUCell algorithm [66]. eRegulons with correlation coefficients between pseudobulked per cell type TF expression and region enrichment AUC scores > 0.6 or ←0.4 were considered high quality and used for downstream analysis. This resulted in 63 regulons, with a median of 408 genes and 719 regions per regulon. The eRegulon enrichment scores for regions and genes were normalized for each cell and used as input into t-distributed stochastic neighbor embedding (t-SNE) from the Python package fitsne (v.1.2.1). eRegulon specificity scores were calculated, per cell type and eRegulon, using the eRegulon specificity score (RSS) algorithm as described in [67], using target region or target gene eRegulon enrichment scores as input.

Patient recruitment

Recruitment of the affected individual to this study was initiated by clinicians, in the context of care, in accordance with the Montpellier Hospital ethics committee, followed by written informed consent from his legal representatives. The patient genetic variants were determined with exome sequencing followed by CGH array. The exome sequencing was performed with a NextSeq 550 sequencer (Illumina) in paired end 2 × 75 base-pairs. A bioinformatics filter based on the individual’s symptoms is applied to prioritize already known genes or those expressed in the brain. Three heterozygous variants were identified in the individual: NM_004284.6 (CHD1L): c.1929del, p.Arg643Serfs*16; NM_001346810.2 (DLGAP2): c.1696C > T, p.Arg566*; NM_017990.5 (PDPR):c.1147G > T, p.Gly383*.

SDS–PAGE analysis of p.Arg392His CHD1L

The cDNA coding for the various constructs of CHD1L WT and CHD1L carrying the p.Arg392His mutation were cloned into a pET-MCN vector [68] coding for an N-terminal 6xhistidine affinity purification tag. Escherichia coli BL21 (DE3) cells were transformed with the respective plasmids and the colonies having incorporated the plasmids were selected by ampicillin resistance. These colonies were used for inoculating 4 ml liquid cultures. The cultures were left to grow upon shaking during 4 h at 37°C. The cultures were then cooled down at 20°C and protein expression was induced by adding 0.5 mM final of Isopropyl-β-d-thiogalactopyranoside (IPTG) (Euromedex). Expression was achieved overnight upon shaking at 20°C. The cultures were centrifuged (2500 rpm, 20 min) and the supernatant discarded. All cultures were resuspended in a purification buffer composed of 10 mM Tris (pH 8.0) and 200 mM NaCl. A small sample of each resuspended cultures was collected, mixed with Laemmli buffer and boiled at 95°C for 5 min prior to analysis of the total expression levels by SDS–PAGE. The resuspended cultures were lysed by sonication. After centrifugation (2500 rpm, 20 min), the supernatants were incubated with 20 μl of Talon affinity beads (Clonetech) for 1 h at 4°C. The supernatants were then discarded, and the beads washed twice with the purification buffer. The beads were then resuspended in Laemmli buffer for analysis of the soluble levels by SDS–PAGE.

Statistical analysis

Number of measures (n), statistical tests used, and associated P-values are indicated in the figures or in the legends of each figure. Graph representation and statistical analyses were performed using GraphPad Prism 10 (v10.2.0). All measurements were performed on distinct samples. Rout test was performed to remove outliers. Graphs are represented as mean ± standard error of the mean (SEM). To ease the visualization of the data, the nonsignificant P-values are not showed on the graphs.

Results

In vivo testing of human 1q21.1 distal region genes

Manipulation of zebrafish embryos is an attractive method to discover human dosage-sensitive genes. This is particularly useful when the CNV under investigation has mirrored anatomical phenotypes detectable during early development, allowing for assays using a combination of gene suppression and overexpression experiments [41, 69, 70]. Given the association between the 1q21.1 distal CNV and changes in head circumference and anthropometrics traits, we proposed that (i) systematic overexpression of each of the ten genes in the region might induce macrocephaly as described in 1q21.1 distal duplication syndrome and (ii) reciprocal suppression of these candidate genes might yield the microcephalic phenotype seen in the 1q21.1 distal deletion syndrome.

To test this possibility, we first queried the zebrafish genome by reciprocal BLAST (Basic Local Alignment Search Tool) for each of the eight single copy protein coding genes within the 1q21.1 distal region (Fig. 1A). All genes but FMO5 have Ensembl-annotated orthologues [CABZ01083448.1/prkab2 (ENSDARG00000067817); chd1l (ENSDARG00000015471); bcl9 (ENSDARG00000036687); gpr89b (ENSDARG00000077983); acp6 (ENSDARG00000040064)]; for two of them, GJA5 and GJA8, two copies each were mapped (gja5a (ENSDARG00000040065), gja5b (ENSDARG00000069450) and gja8a (ENSDARG00000069451), gja8b (ENSDARG00000015076), respectively). NOTCH2-derived NOTCH2NLA and NOTCH2NLB paralogs, located at the BP3–BP4 breakpoints of the 1q21.1 distal region, are human specific genes [23, 24].

Figure 1.

In vivo overexpression of the genes from the 1q21.1 distal region in zebrafish. (A) Schematic representation of the human 1q21.1 BP3–BP4 distal region showing eight genes comprised in the region and NOTCH2-paralogs NOTCH2NLA and NOTCH2NLB located at the breakpoints. (B) Schematic representation of the experimental procedure including micro-injection of mRNA in one- to two-cell stage zebrafish embryos and morphometric analyses at 5 dpf for the head size and the body length of zebrafish larvae. (C) Dorsal views of the head of control larvae and larvae injected with NOTCH2NLA mRNA (50 pg), NOTCH2NLB mRNA (50 pg) or NOTCH2NLA+ NOTCH2NLB mRNA (50 + 50 pg) at 5 dpf. Double-ended arrows indicate distance between the eyes. (D) Dot plot showing the distance between the eyes (head size) of control larvae and larvae injected with NOTCH2NLA mRNA (50 pg), NOTCH2NLB mRNA (50 pg), or NOTCH2NLA+ NOTCH2NLB mRNA (50 + 50 pg) at 5 dpf. Data shown as mean ± SEM of a least triplicate batches; Kruskal–Wallis test. (E) Lateral view of control larvae and larvae injected with NOTCH2NLA mRNA (50 pg), NOTCH2NLB mRNA (50 pg) or NOTCH2NLA+ NOTCH2NLB mRNA (50 + 50 pg) at 5 dpf. (F) Dot plot showing the body length measurements of control larvae and larvae injected with NOTCH2NLA mRNA (50 pg), NOTCH2NLB mRNA (50 pg), or NOTCH2NLA+ NOTCH2NLB mRNA (50 + 50 pg) at 5 dpf. Data shown as mean ± SEM of at least triplicate batches per condition; ordinary one-way analysis of variance (ANOVA) test. (G) Dot plot showing the distance between the eyes (head size) of 5 dpf control larvae and larvae injected with each of the eight human genes from the 1q21.1 distal region. Data shown as mean ± SEM of a least triplicate batches per gene; Kruskal–Wallis test. (H) Dorsal views of control and larvae injected with CHD1L mRNA (100 pg) at 5 dpf. Double-ended arrows indicate distance between the eyes. (I) Dot plot showing the distance between the eyes (head size) of control and CHD1L mRNA-injected larvae (100 pg) at 5 dpf. Data shown as mean ± SEM of a least triplicate batches; Mann–Whitney test. (J) Lateral views of control and CHD1L mRNA-injected larvae (100 pg) at 5 dpf. (K) Dot plot showing the body length measurements of control or injected larvae with CHD1L mRNA (100 pg) at 5 dpf. Data shown as mean ± SEM of triplicate batches; Mann–Whitney test. To simplify the visualization of the data, the nonsignificant P-values are not shown on the graphs. Scale bars: 200 μm [panels (C) and (H)] and 500 μm [panels (E) and (J)].

Using a similar strategy to our previous studies [41, 69, 70], we first generated capped mRNA for all 10 human genes and injected zebrafish embryos at one-cell stage with a dose of 50 or 100 pg of mRNA per embryo. Expression levels exceeding those of endogenous transcripts were achieved by injecting 0.25%–0.5% of the total embryonic polyadenylated mRNA. Previous study indicated that the injected human mRNAs persist in the larvae up to 4.25 dpf [69] (Fig. 1B).

First, we determined whether NOTCH2NLA and NOTCH2NLB overexpression might affect brain size and/or body length of zebrafish larvae. Despite the absence of NOTCH2NLA/B in zebrafish, their overexpression led to relevant phenotypes. More specifically, the overexpression of each of the two genes led to opposite effects on head size: NOTCH2NLA overexpression led to microcephaly (−6.3%) whereas NOTCH2NLB overexpression led to macrocephaly (+7.9%) in overexpressant larvae compared to controls (Fig. 1C and D). Overexpression of both NOTCH2NLA and NOTCH2NLB did not lead to head size variations, suggesting a compensatory effect (Fig. 1C and D). In addition, we observed no physiologically relevant effect on the body length upon injection of NOTCH2NLA, NOTCH2NLB, or both transcripts (−1.5%) in zebrafish (Fig. 1E and F). Despite their critical role during primate brain development [23], our in vivo data indicated that the opposite effects mediated by the two NOTCH2NL paralogs counteract each other under control conditions. Moreover, we exclude a role for these genes in the growth phenotypes observed in individuals with 1q21.1 distal syndromes. Therefore, we postulated that other genes comprised in the 1q21.1 distal region contribute to the 1q21.1 distal syndromes head size and growth phenotypes.

We next overexpressed each of the remaining eight genes comprised in the 1q21.1 distal region in zebrafish. We did not observe toxicity, lethality, or gross morphological defects upon injection of 100 pg of mRNA, except for GJA5, for which the maximum dosage that could be tested was 50 pg. GJA5 is a member of the connexin gene family, whose mutations have been associated with atrial fibrillation and arteriovenous malformations and the duplication of GJA5 leads to tetralogy of Fallot [15–17]. The gene is likely responsible for the congenital heart diseases associated with the 1q21.1 distal deletion. Of the remaining seven genes, none induced macrocephaly, even when injected at higher mRNA doses (250 pg); only overexpression of the chromatin remodeler CHD1L produced a significant increase in head size at 5 dpf (Fig. 1G and Supplementary Fig. S1A–E). Specifically, we measured the distance between the eyes (Fig. 1H and I) and the distance between the anterior-most part of the forebrain and the hindbrain (Supplementary Fig. S2A and B) in CHD1L-injected larvae, revealing a brain overgrowth phenotype when CHD1L is overexpressed. We further evaluated whether CHD1L could be implicated in the general height abnormalities observed in 1q21.1 distal CNV carriers. An effect on height has been reported for the distal deletion, with 25%–50% of the carriers having short stature, whereas the duplication carriers tend to be in the upper percentiles of height [4, 13, 14]. CHD1L overexpression led to a significant increase of the body length (Fig. 1J and K) and inter-somite distance (Supplementary Fig. S2C and D) at 5 dpf in zebrafish larvae, supporting the possibility that CHD1L could broadly impact anthropometric traits and growth.

To validate further the specificity of the phenotypes driven by CHD1L and to investigate whether it was possible to simulate the reciprocal phenotypes seen in individual carriers of 1q21.1 distal deletion, we sought to determine the effect of a loss of the sole ortholog of CHD1L in zebrafish. chd1l is ubiquitously expressed as early as 3 hours post-fertilization (hpf) in blastomeres with an expression peak between 3 and 72 hpf (Daniocell database; Supplementary Fig. S3A and B). More specifically, chd1l is expressed in all brain vesicles including telencephalon, diencephalon, midbrain, midbrain-hindbrain boundary and neural crest cells between 12 and 48 hpf. These expression data confirmed the relevance of a zebrafish model to study the developmental role of CHD1L. We thus obtained an N-ethyl-N-nitrosourea (ENU) mutagenized stable line for the sole chd1l zebrafish orthologue (chd1l^sa14029 line carrying a p.Arg6X stop mutation) and determined the total brain volume of 5 dpf chd1l^+/+ and chd1l^−/− larvae using High-Resolution Episcopic Microscopy [40]. We detected a ∼25% reduction of the brain in chd1l^−/− larvae compared to WT larvae (Fig. 2A and B). Bi-dimensionally measurements confirmed a significant decrease of head size (Fig. 2C and Supplementary Fig. S2E) and body length (Fig. 2D and Supplementary Fig. S2F) in homozygous chd1l mutant larvae at 5 dpf. Importantly, these phenotypes are specific to chd1l; both morphometric phenotypes could be fully rescued upon injection of 100 pg of wildtype human CHD1L mRNA in the chd1l^−/− homozygous embryos at 1–2 cell stage (Fig. 2C and D and Supplementary Fig. S2E and F). There was also no defect in other structures, including the heart and the swim bladder, indicating the absence of global developmental delay of the chd1l mutant larvae. Notably, the reduction of the head size and body length measurements at 5 dpf was also observed in chd1l^+/− heterozygous mutant larvae, which more closely mimics the gene dosage of heterozygous human 1q21.1 distal deletion (Supplementary Fig. S4A–D).

Figure 2.

Loss of chd1l recapitulates mirrored neuroanatomical phenotypes of the 1q21.1 distal deletion in zebrafish. (A) 3D reconstitution of 5 dpf control and chd1l^−/− zebrafish larvae acquired by HREM. The brain volume is shown in yellow. (B) Dot plot showing the quantification of brain volume of 5 dpf control and chd1l^−/− zebrafish larvae. Data shown as mean ± SEM of a representative batch; Student’s t-test. (C, D) Dot plot showing the rescue of chd1l^−/- phenotypes for head size and body length upon injection of CHD1L mRNA (100 pg) in chd1l ^−/− larvae compared to control larvae at 5 dpf, respectively. Data shown as mean ± SEM of triplicate batches; ordinary one-way ANOVA. (E) Lateral views of 2 dpf zebrafish heads for mutant chd1l^−/−, control and CHD1L mRNA-injected larvae stained with phospho-histone H3 (pHH3; M-phase marker). Dotted yellow lines indicate quantified head areas. (F) Dot plot showing the pHH3 ratio of proliferating cells in the heads of 2 dpf chd1l^−/− and CHD1L mRNA-injected larvae normalized to control. Data shown as mean ± SEM of triplicate batches; ordinary one-way ANOVA. (G) Dorsal views of 2 dpf heads of control and chd1l^−/- mutant larvae stained with Acridine Orange (AO) or TUNEL. Dotted yellow line indicates quantified head area. (H) Dot plot showing the number of AO-positive cells in the heads of 2 dpf control and chd1l^−/− mutant larvae. Data shown as mean ± SEM of triplicate batches; Mann–Whitney test. (I) Dot plot showing the number of TUNEL-positive cells in the heads of 2 dpf control and chd1l^−/− mutant larvae. Data shown as mean ± SEM of triplicate batches; Mann–Whitney test. To simplify the visualization of the data, the nonsignificant P-values are not showed on the graphs. Scale bars: 100 μm.

CHD1L dosage perturbation affects cell proliferation and cell death

To investigate the mechanism leading to the head size differences, we examined the developing brain in both CHD1L overexpressant and chd1l^−/− mutant embryos. First, we determined the number of mitotic cells by phospho-histone H3 immunostaining of chd1l^−/−, control and CHD1L mRNA-injected zebrafish heads at 2 dpf (Fig. 2E). A significantly decreased number of proliferating cells was observed in the head at 2 dpf in chd1l^−/− mutant larvae compared to controls. Conversely, an increased cell proliferation in the head at 2 dpf was observed upon CHD1L overexpression (Fig. 2F). Second, we assessed cell viability by performing Acridine Orange staining and TUNEL assay on control and chd1l^−/− larvae at 2 dpf (Fig. 2G). Both staining revealed that loss of chd1l increases apoptotic level in the head of the larvae, supporting the microcephalic phenotype observed at 5 dpf (Fig. 2H and I). Such imbalance of cell proliferation and apoptosis have been previously seen in zebrafish and mice modeling micro/macrocephaly phenotypes [41, 69–71]. To validate our data obtained with our chd1l mutant stable line, we generated microdeletions in chd1l exon 3 using CRISPR/Cas9 genome editing. We confirmed the presence of genetic editing in 82.6% of injected embryos (founders, F0) by polyacrylamide gel electrophoresis and Sanger sequencing (Supplementary Fig. S5A). F0 zebrafish CRISPR larvae showed decreased head size and body length at 5 dpf (Supplementary Fig. S5B–E) and phospho-histone H3 immunostaining also showed decreased cell proliferation in the brain at 2 dpf (Supplementary Fig. S5F and G).

Although the zebrafish brain bears several similarities with the mammalian brain in terms of developmental programming, we examined whether our findings might be relevant to cortical development in a mammalian system. In both human and mouse, CHD1L/Chd1l is ubiquitously expressed; we noted similar profiles in both species in the whole brain with an expression peak at 4 weeks post-conception in human and murine E11.5 (Supplementary Fig. S3C and D) [72]. We performed in utero electroporation (IUE) to overexpress or deplete Chd1l in mouse embryonic cortices and investigated the potential of apical progenitors lining the lateral ventricle to generate TBR1-positive cortical neurons. We performed IUE of plasmids expressing mouse Chd1l transcripts or small interfering RNA (siRNA) directed against mouse Chd1l together with a pCAGGS-GFP reporter construct, allowing the expression of GFP specifically in electroporated apical progenitors and their progeny, in WT mouse cortices at E13.5 (Fig. 3A and Supplementary Fig. S6A). One day after electroporation, we evaluated the number of TBR1-positive neurons produced from the electroporated progenitors. When compared to respective controls (empty vector, or siRNA directed against luciferase), overexpression of Chd1l induced a significant increase (+19.7%) of TBR1-positive neuronal progeny (Fig. 3B and C) whereas depletion of Chd1l impaired the generation of TBR1-positive cortical neurons (−24%) (Fig. 3D and E). These mirrored phenotypes in mouse cortices indicated that function of Chd1l during neuronal development is conserved between nonmammalian vertebrates and mammals and that the dosage of Chd1l is critical for neuronal production.

Figure 3.

Dosage changes of Chd1l control neuronal production in transient mouse model. (A) Schematic of in utero electroporation. Mouse embryos are electroporated at E13.5, mouse brain are dissected at E14.5 and stained for Tbr1 neuronal marker. Double GFP-positive and Tbr1-positive cells representing the newborn neurons are counted in a defined area in the VZ/SVZ. (B) E14.5 mouse brain slices for control and Chd1l overexpressant conditions. (C) Dot plot showing the percentage of Tbr1-positive cells among the electroporated GFP-positive cells from the defined area in E14.5 mouse brains from control or electroporated with Chd1l expression vector. Data shown as mean ± SEM, electroporated and imaged embryos were from five different litters for control condition and from six different litters for Chd1l overexpression condition; Student’s t-test. (D) E14.5 mouse brain slices for control and knockdown Chd1l conditions. (E) Dot plot showing the percentage of Tbr1-positive cells among the electroporated GFP-positive cells in E14.5 mouse brains from control or electroporated with esiChd1l RNA. Data shown as mean ± SEM, electroporated and imaged embryos were from three different litters for esiRluc condition and from four different litters for esiChd1l condition; Student’s t-test. CP, cortical plate ; IZ, intermediate zone ; VZ/SVZ, ventricular/subventricular zones. Scale bars: 50 μm.

Taken together, our zebrafish and mouse data suggested that CHD1L is a major contributor of the mirrored body size and head size phenotypes associated with the 1q21.1 distal CNV. However, our data do not exclude the possibility that other loci within the region also have an independent contribution to the 1q21.1 distal deletion or duplication anatomical phenotypes. Therefore, we performed the combinatorial overexpression of CHD1L with each of the remaining 1q21.1 distal transcripts to evaluate whether other genes within the CNV might also be relevant to the head size phenotypes through additive or multiplicative interactions with CHD1L. None of the other transcripts exacerbated nor alleviated the macrocephaly phenotype driven by CHD1L (Supplementary Fig. S7A), indicating that none of the other genes within this interval act as modifier for the head size phenotype, at least as determined by our assays.

Loss of CHD1L disrupts neurogenic program in human neuronal progenitor cells

To our knowledge, CHD1L has never been shown to control human neurogenesis. Therefore, to investigate CHD1L’s potential gene targets and associated pathways in the neuro-developmental context, we designed a gRNA targeting exon 1 and performed CRISPR-Cas9 editing of CHD1L in hiPSCs (hiPSC GM8330-8, used in [73] adapting the protocol described in [74]). We selected two isogenic hiPSCs lines carrying the following homozygous frameshift mutations, c.72_76delCCGAG (hiPSC CHD1L^−/- line 1) and c.75_76delAG (hiPSC CHD1L^−/− line 2) (Supplementary Fig. S8A), respectively, and confirmed the absence of further mutations in the Top-20 predicted off-targets (Supplementary Table S1). Karyotype analysis confirmed the genomic integrity of CRISPR-generated lines as well as control hiPSCs and allowed us to exclude the presence of aneuploidies and large chromosomal rearrangements that could have been introduced during the gene-editing process (Supplementary Fig. S9A–C). In parallel, we confirmed the absence of the CHD1L protein in both lines by western blot (Fig. 4A). Given that recent studies on brains from ASD individuals highlight transcriptional changes occurring primarily in neuronal lineages, we chose to focus our analyses on neuronal progenitor cells, where these changes are first observed [75, 76]. The control and the CHD1L-edited hiPSC lines were thus differentiated into human neural progenitor cells (hNPC) (Fig. 4B). After seven days of differentiation, formation of rosettes was observed. Positive expression of hNPC markers SOX1 and PAX6 confirmed the lineage commitment (Fig. 4C) and validated the successful differentiation into hNPC. RNA-seq analysis of the derived cells confirmed that both mutant lines shared similar expression profiles (Fig. 4D). Further, we intersected the list of DEGs from both mutant lines and we found a total of 857 common DEG in hNPC lacking CHD1L compared to control hNPC, including 563 downregulated and 294 upregulated genes (FDR < 0.05; Fig. 4E and Supplementary Table S2).

Figure 4.

Loss of CHD1L perturbs expression and chromatin accessibility of neurogenesis-associated genes in hNPC. (A) Western blot analysis of CHD1L protein level in control hiPSC CHD1L^+/+ and in the two isogenic CHD1L^−/− cell lines generated by CRISPR-Cas9 editing showing the absence of CHD1L protein in mutant lines. (B) Schematic representation of human neural progenitor cells (hNPC) generation from hiPSC and subsequent analysis pipeline. (C) Immunostainings of hiPSC-derived hNPC with PAX6 and SOX1 neural progenitor markers. Scale bars: 100 μm. (D) Heatmap of the DEGs in hNPC Mutant Line 1 compared to Mutant Line 2 showing similar expression profiles between both mutants. (E) Pie chart of DEGs in CHD1L^−/− hNPC compared to control hNPC (FDR < 0.05). (F, G) Gene Ontology Biological Processes enrichment of the downregulated and upregulated genes in mutant hNPC, respectively. (H) Dot plot of referenced genes in indicated databases associated with human disorders and synaptogenesis function among the 52 most DEG (|log₂FC|> 1; FDR < 0.05) in CHD1L mutant hNPC. (I) Average tag density profiles and corresponding heatmap representation of tag density map of H3K4me2 CUT&RUN peaks, ± 2 kb from peak center of ATAC-seq peaks from control hNPC. (J) Pie chart of differentially accessible chromatin sites in CHD1L^−/− mutant hNPC (Line 1 and Line 2 combined) compared to control hNPC. (K) Tag density map of control, Mutant Line 1 and Mutant Line 2; ± 2 kb from peak center for 3723 less accessible and 2539 more accessible peaks in CHD1L^−/− hNPC and corresponding average tag density profiles. Line 1 and Line 2 were combined and the same direction of change for both mutants was selected; Mixed refers to peaks that exhibit opposite changes in Mutant Line 1 compared to Mutant Line 2 and were discarded from the subsequent analyses. (L) Venn diagram showing 74 downregulated genes that are also less accessible in absence of CHD1L. (M) Venn diagram showing 35 upregulated genes that are also more accessible in absence of CHD1L. (N) Chromatin accessibility and level of expression for UNC5D and NRP2 in control and mutant hNPC.

To interpret the transcriptomic changes, we conducted a pathway analysis on the list of the 563 downregulated genes using a PANTHER overrepresentation test (geneontology.org) for biological processes. The analysis revealed significant enrichment of terms associated to neurodevelopment and maturation in downregulated genes: “Neurogenesis” (GO:0 022 008, FDR = 3.16E-32), “Forebrain development” (GO:00 309 005, FDR = 1.02e-10); “Forebrain regionalization (GO: GO:0 021 871, FDR = 1.94e-2)”; “Synapse assembly” (GO:0 007 416, FDR = 2.94e-9); “Developmental growth” (GO:0 048 589, FDR = 1.34e-4) and terms associated to renal development “Metanephric nephron development” (GO: GO:00 722 721, FDR = 3.07e-2) (Fig. 4F). Conversely, upregulated genes are associated to various translational processes such as “Translation” (GO:0 006 412, FDR = 3.04e-19) and “RNA processing” (GO:0 006 396, FDR = 1.83e-6) in addition to cell death processes highlighted by “Regulation of programmed cell death” (GO:0 043 067, FDR = 1.8e-5), and “Regulation of apoptotic signaling by p53” (GO:1 902 253, FDR = 8.64e-5). We also noted association to “Post-embryonic camera-type eye development” (GO:0 031 077, FDR = 3.09e-2) as the most enriched term (Fig. 4G and Supplementary Table S2). Further, analysis of reactome pathways confirmed the association of downregulated genes to the nervous system and nephrotic development whereas upregulated genes tended to be associated to DNA damage response and cell cycle in absence of CHD1L (Supplementary Fig. S8B and C and Supplementary Table S2).

To determine the gene sets that are the most affected by CHD1L loss, we applied an absolute value of |log₂FC|> 1 and we obtained 52 DEG including 45 downregulated and 7 upregulated genes between control and both mutant hNPC lines (Supplementary Table S2). We constructed a functional protein–protein interaction network of the identified 52 DEG using STRING [61] (string-db.org), which highlighted the presence of a significantly greater number of protein–protein interactions than expected (P= 1.10⁻¹⁶) with three functional clusters associated to brain regionalization/nervous system development, regulation of neurotransmitter levels and axonogenesis-associated terms (Supplementary Fig. S8D). Furthermore, 8/52 genes (DCC, DPP6, EBF3, ELAVL3, NRP2, SERPINE1, SYP, UNC5D) were found in the SFARI gene list for their association to ASDs (enrichment P= 0.02), and a significant enrichment was also identified for the developmental disorder genes (11/52 genes: CPLX2, DCC, EBF3, EGR2, ELAVL3, EOMES, FGFR3, SVOP, SYP, SYT4, TFAP2A) in the Decipher Database (P= 6e-4). A total of 9/52 genes are referenced in the SynGO database recapitulating genes associated to synapse (ARC, CPLX2, DCC, MDGA1, NRP2, PRPTRN2, SVOP, SYP, and SYT4) (P= 0.001) (Fig. 4H). One more autism risk DE gene (THSD7A) was annotated by crossing our list with the de novo “high confidence” ASD genes carrying likely gene disrupting mutations [77]; interestingly, two pleiotropic genes (DCC, UNC5D) with a prominent role in different psychiatric disorders [78] were also identified among the 52 DEG. A trend for enrichment was found when we intersected the 52 DEG with a list of manually curated growth genes obtained from two publications [79, 80] (four genes: DPP6, FGFR3, TFAP2A, IGFBP3). Remarkably, the long noncoding RNA PAX8-AS1 associated with hypothyroidism [81] was one of the most downregulated genes in absence of CHD1L (log₂FC←4). Moreover, other genes associated with hypothyroidism and short stature were found downregulated (e.g. SOX3, GLI2, KMT2D, and FGD1) [82–88] (Supplementary Table S2). Of note, individuals with 1q21.1 deletion often exhibit short stature [13, 14] and one case has been reported with congenital hypothyroidism [89].

Chromatin accessibility of neurodevelopmental-associated genes is altered in CHD1L-depleted human neuronal progenitor cells

CHD1L is a known ATP-dependent chromatin remodeler [28–34]. To determine whether the transcriptomic changes are due to perturbed chromatin accessibility, we performed Assay for Transposase-Accessible Chromatin with high throughput sequencing (ATAC-seq) on control and CHD1L^−/− mutant hNPC lines [49]. We validated that the detected ATAC-seq peaks from control cell line colocalized with the presence of dimethylated lysine residue at position 4 of histone H3 (H3K4me2), a hallmark for active promoters and enhancers (Fig. 4I and Supplementary Fig. S10A and B). In the absence of CHD1L, we found a total of 3723 peaks for which the chromatin was less accessible and 2539 peaks that were more accessible in hNPC (P-value ≤0.05; Fig. 4J and K and Supplementary Fig. S10C and D), further demonstrating that CHD1L plays a role in chromatin remodeling in neuronal progenitors (Fig. 4K and Supplementary Table S3). Although the distribution of more and less accessible peaks is similar across intergenic regions, introns, exons, and TTS, a notable difference was observed in promoter regions, with 21.5% of more accessible peaks and only 7.8% of less accessible peaks in CHD1L-/- hNPC (Supplementary Fig. S10E and F).

We next intersected the DEG with the list of differentially accessible genes detected in CHD1L^−/− mutant hNPC. We found a total of 74 genes that were coordinately less expressed and showed closed chromatin sites at their loci as exemplified by the ASD and ADHD-susceptibility gene UNC5D [90–94], and 35 genes that were coordinately more expressed and were associated to newly accessible chromatin sites, including NRP2 that regulates dendritic growth of adult-born neurons in the dental gyrus [95, 96] (Fig. 4L–N, Supplementary Fig. S11A–F, and Supplementary Table S3).

Next, TOBIAS ATAC-seq footprinting analysis [51] indicated that 19 clusters of TFs had favorized binding abilities when CHD1L was lost, whereas 8 clusters were more able to bind chromatin when CHD1L was expressed (FDR < 0.01, Supplementary Fig. S12A and B). The 27 clusters included 551 individual TFs (Supplementary Table S3). In absence of CHD1L, we noted that the binding motifs of POU5F1, SOX2, CTCF, FOXO1, FOXP2 were more accessible whereas the binding motifs for GLI2, GLI3, JUN, SMAD2::SMAD3::SMAD4, TP53 were less accessible (Supplementary Fig. S12C and Supplementary Table S3). Most of those factors are found to be dysregulated in ASD brain tissues [75].

Taken together, the transcriptomic data and footprinting analysis collectively revealed that the transcriptional perturbation in the CHD1L-edited hNPC lines specifically impairs cell identity, brain regionalization, neurogenesis, synaptogenesis, and growth-associated pathways. Notably, the most affected genes are linked to ASD and developmental disorders. These disrupted pathways align with known signatures of ASD identified through both bulk and single-cell genomic studies on tissues from ASD individuals, reinforcing the notions that CHD1L plays a critical role in neurodevelopment and that its dosage imbalance likely underlies the cognitive impairments associated with 1q21.1 CNV.

CHD1L acts as a co-TF interacting with NuRD complex and the pioneer TF SOX2

To identify direct CHD1L targets, hNPC chromatin was subjected to CUT&RUN using an anti-CHD1L antibody [97] (Supplementary Fig. S13A). Bioinformatics analysis uncovered 5070 CHD1L peaks, mainly located at intronic and intergenic regions (Fig. 5A and Supplementary Table S4). These peaks were assigned to 4002 unique genes using HOMER. Amongst these genes with CHD1L peaks, we found 102 genes in our transcriptome data that were downregulated including key neurodevelopmental genes such as ONECUT2 and ONECUT3 and 55 genes that were upregulated (Fig. 5B and C and Supplementary Table S4). An analysis of genomic distribution revealed that CHD1L peaks did not correlate with H3K4me2 peaks and were present in far upstream and downstream regions (−100 to −50 kb and 50–100 kb), intermediate locations (−50 to −20 kb and 20–50 kb) and DNA segments in the vicinity of the TSS (−10 to 10 kb) (Fig. 5D and E).

Figure 5.

Characterization of CHD1L binding in hNPC. (A) Pie chart of the genomic distribution of CHD1L CUT&RUN peaks in control hNPC. (B) Examples of CHD1L-bound sites located upstream of downregulated genes ONECUT2 and ONECUT3. (C) Venn diagram showing CHD1L-bound genes that are downregulated or upregulated in CHD1L mutant hNPC. (D) CHD1L Peak distribution from TSS. (E) Tag density map of CHD1L and H3K4me2 CUT&RUN peaks, ± 2 kb from the CHD1L peak center and corresponding average tag density profiles. (F, G) Lisa results showed as pairwise scatterplots for comparison of the two gene sets (downregulated and upregulated genes in absence of CHD1L in hNPC) based on TF ChIP-seq and TF motifs respectively. The top 10 TFs are shown in yellow. (H, I) RSAT-based motif enrichment analysis and ChIP-Atlas comparison of CHD1L CUT&RUN data, respectively. (J) Venn diagram showing common TFs predicted by Lisa, ChIP-Atlas, RSAT, and HOMER analyses.

To identify putative TFs that may be involved in the recruitment of CHD1L during neurogenesis, we predicted the transcriptional regulators of the DEG in CHD1L^−/− hNPC by inference through integrative modeling of public chromatin accessibility and ChIP-seq data (http://lisa.cistrome.org/) [98]. Notably, the Top-10 predicted regulators of the upregulated genes in CHD1L^−/− hNPC included TFs involved in DNA damage response and repair (XRN2, ERCC3, SOX4, BRD4) and in early morphogenesis of the central nervous system (OTX2). We also noted a strong enrichment of NKX2.2 binding motif, a factor known to control ventral neuronal patterning [99] and to regulate oligodendrocyte differentiation [100]. For the downregulated gene set, cistromic analysis highlighted the EGR family of transcription regulatory factors (EGR1, EGR2), which is implicated in orchestrating the changes in gene expression that underlie neuronal patterning and plasticity [101, 102], the architectural protein CTCF that changes higher-order chromatin structure and controls the distance between associating domains within and among chromosomes, and several TFs controlling pluripotency and differentiation such as SOX2, GLIS1, ZFP281, and ERG [103–106] (Fig. 5F and G and Supplementary Table S2). We next performed RSAT and HOMER motif analyses as well as ChIP-Atlas comparisons on CHD1L CUT&RUN data, and found a strong enrichment for SOX/FOX, EGR2, ZNF449, and PRDM/STAT binding sequences in hNPC (Fig. 5H and I and Supplementary Table S4). By intersecting the lists of TFs from the different prediction tools utilized on our transcriptomic and cistromic data, we identified 68 candidate TFs including SOX2, OTX2, MECP2, CTCF, CHD8, EGR2, and GLI3 (Fig. 5J and Supplementary Table S4). This list suggested the implication of CHD1L in major neuro-developmental pathways controlling correct nervous system development (SOX2, OTX2, EGR2, GLI3, CTCF) [102, 107–112] and with genes mutated in neurodevelopmental disorders (CHD8, MECP2) [41, 113].

To identify protein partners of CHD1L in hNPC and whether the aforementioned predicted TFs interact physically with CHD1L, we performed affinity purification with CHD1L antibody followed by mass spectrometry (AP-MS) on both wildtype and CHD1L^−/− mutant lines. By applying stringent thresholds, we identified a total of 286 candidate partners (Supplementary Table S5). To test whether some TFs were present in the AP-MS dataset, we filtered the candidates based on their cellular localization and retrieved 104 nuclear partners of CHD1L. STRING analysis revealed four functional clusters including DNA-repair proteins (Cluster 1), components of the NuRD complex (Cluster 2), Arp2/3 complex proteins (Cluster 3), and Histone domains (Cluster 4). More specifically, Cluster 2 included core components of the NuRD complex (GATAD2, MBD3, HDAC1, HDAC2, MTA2, and RBBP4), the ATP-dependent remodeling factor CHD7, and the TFs SOX2, OTX2, and PAX6 known for their role in cell-fate and brain regionalization [107, 114, 115] (Fig. 6A and B and Supplementary Table S5). Clusters 2 and 4 also includes candidate protein partners associated with cell cycle and cell survival (“G1/S specific transition”, “Apoptosis induced DNA fragmentation”) (Fig. 6B and Supplementary Table S5).

Figure 6.

Immunoprecipitation and mass-spectrometry define CHD1L interactome and reveal developmental partners in hNPC. (A) STRING analysis of 104 nuclear proteins co-immunoprecipitated with CHD1L in hNPC extracts. Protein–protein interactions are shown (Protein-Protein interaction (PPI) enrichment P-value < 1e⁻¹⁶) and annotated according to functional clusters: Cluster 1 (63 proteins), Cluster 2 (29 proteins), Cluster 3 (7 proteins), Cluster 4 (3 proteins). (B) Gene ontology analyses showing reactome pathways enriched for each cluster. (C) Representative co-localization of CHD1L, HDAC2 (SRX19212560), RBBP4 (SRX19212562), SOX2 (SRX330107), CHD7 (SRX9795022), and H3K4me2 peaks on neuro-developmental genes (NFIA, JARID2, SLIT1) in human neural cells. (D) Venn diagram of SOX2 target genes (SRX330107) and CHD1L target genes showing an overlap of 1347 genes. (E) DisGeNET analysis of the common 1347 SOX2 and CHD1L target genes showing enrichment in human neurodevelopmental disorders. (F) Representative western blot of CHD1L and SOX2 co-immunoprecipitation in transfected HEK cells.

We next explored whether CHD1L CUT&RUN peaks significantly overlap with publicly available ChIP-seq datasets for SOX2, HDAC2, RBBP4, and CHD7 on neural cell types using the ChIP-Atlas enrichment tool. We obtained a significant enrichment of overlapping CHD1L peaks with SOX2 peaks (log P-value = −11; SRX330107), HDAC2 peaks (log P-value= −3.3; SRX7643943), and a trend with RBBP4 peaks (log P-value = −1.2; SRX19212562). In contrast, there was no significant enrichment between CHD1L and CHD7 peaks despite the fact that some peaks were in common for a discrete number of neural genes (NFIA, JARID2, and SLIT1) when CHD7 acts in concert with the NuRD complex (Fig. 6C). By examining the genes assigned to SOX2 ChIP-seq peaks (SRX330107), we found that one third of genes bound by CHD1L were also bound by SOX2 in human neuronal progenitor cells (Fig. 6D and Supplementary Table S4). DisGeNET analysis on the 1347 common genes revealed an enrichment for terms including schizophrenia, autistic disorder, obesity and seizures (Fig. 6E). Co-immunoprecipitation experiments further confirmed the physical interaction between SOX2 and CHD1L (Fig. 6F). Overall, our data suggested that CHD1L regulates gene expression through its direct or indirect interaction with pioneer TFs and NuRD regulatory complex during human neurodevelopment and brain regionalization.

CHD1L loss perturbs cell-fate decision during forebrain regionalization

Self-organizing cerebral organoids grown from pluripotent stem cells combined with single-cell genomic technologies provide opportunities to examine gene regulatory networks (GRNs) underlying human brain development and diseases [116–118]. To better characterize the role of CHD1L during cerebral development and to test the possibility that CHD1L plays a role during brain regionalization as indicated by our omics data, we derived cerebral organoids for 60 days in vitro (DIV) from CHD1L^+/+ and CHD1L ^−/− hiPSC (Mutant Line 1). We generated organoids through a self-organization process by providing a permissive environment with minimal external cues, which allowed us to determine the intrinsic capacity of the control and CHD1L mutant hiPSC to undergo in vivo-like morphogenesis [117] (Fig. 7A). CHD1L^−/− (Mutant Line 1) exhibited similar morphology to CHD1L^+/+ control organoids during the in vitro maturation (Fig. 7B). At 52 DIV, control and mutant organoids were positive for both SOX2 (neural progenitor cells marker) and TUJ1 (neural cytoskeleton marker) (Fig. 7C). We noted typical structures of cerebral organoids including rosettes (SOX2+) and cortical plate layers (TUJ1+) [118]. However, we observed a decreased level of TUJ1 protein in CHD1L^−/− mutant organoids compared to controls at 60 DIV (Fig. 7D and Supplementary Fig. S14A), suggesting impaired neurogenesis in organoids lacking CHD1L.

Figure 7.

Loss of CHD1L perturbs forebrain cell-fate determination in hCO. (A) Schematic of hCO generation from hiPSC. (B) Pictures of 60 DIV hCO derived from CHD1L^+/+ and the two CHD1L^−/− cell lines. Scale bars: 1 mm. (C) Immunostaining against TUJ1 (neuronal cytoskeleton marker) and SOX2 (neural progenitor cells) on 52 DIV hCO either CHD1L^−/− (Mutant Line 1) or control. Scale bars: 500 μm. (D) Semi-quantitative analysis of TUJ1 protein expression in CHD1L^−/− (Mutant Line 1) and control hCO normalized with β-Tubulin. Data shown as mean ± SEM of 8 hCO per condition; representative of N = 2 batches of generated organoids; Student’s t-test. (E) Uniform Manifold Approximation and Projection (UMAP) projection of CHD1L^+/+ and CHD1L^−/- hCO cells colored by condition. N = 6 organoids per condition, 2385 cells CHD1L^+/+ and 1634 cells CHD1L^−/− sequenced. (F) UMAP representation of CHD1L^+/+ and CHD1L^−/− (Mutant Line 1) multiome analysis combined and integrated on the same graph. N = 12 organoids, 4019 cells. (G) UMAP representation of CHD1L^+/+ and CHD1L^−/− (Mutant Line 1) hCO multiome analysis on the same graph. Clusters based on cell identities are indicated. (H) UMAP representation of CHD1L^+/+ and CHD1L^−/− (Mutant Line 1) hCO multiome analysis combined and integrated on the same graph. Clusters based on cell identities are indicated. (I) Dot plot representing key cell type markers and their levels of gene expression within cerebral organoids’ cell clusters. Dot size indicates the proportion of cells per cluster expressing the corresponding gene and color is associated to the average expression level of the corresponding gene per cluster. (J) Sankey diagram visualization of the proportion of nuclei positive for either the ventral telencephalic marker FOXG1, or the eye territory marker SIX3 (normalized expression >1) in CHD1L^+/+ and CHD1L^−/− hCO. Percentage of nuclei is indicated for each category.

From our transcriptomic and cistromic data, we reasoned that if CHD1L cooperates with proteins involved in brain development and regionalization, organoids lacking CHD1L should exhibit forebrain cell-fate determination defects. To test this possibility and to determine which cell populations are the most affected by the absence of CHD1L, we employed Single Nuclei Gene Expression and ATAC-seq Multiome (referred as snMultiome) analysis to simultaneously profile the transcriptional and chromatin states of control and CHD1L mutant (Mutant Line 1) 60 DIV organoids. A total of 4019 nuclei were individually sequenced, including 2385 nuclei extracted from control organoids, and 1634 nuclei from CHD1L^−/− mutant organoids (Fig. 7E). To our surprise, we observed almost no overlap of the nuclei identity clusters from control and CHD1L^−/− mutant organoids showing a dramatic effect of the absence of CHD1L on cell identity (Fig. 7E and F).

The Seurat WNN method was used to compute a neighbor graph which was visualized with UMAP. In control human organoids, a total of 16 clusters were annotated based on expression of marker genes (Fig. 7G–I and Supplementary Fig. S14B and C). We found groups with telencephalic identity (FOXG1, PAX6, SOX2), including radial glial cells, cycling radial glial cells, intermediate progenitor cells (IPC), and two groups of neurons including immature and excitatory neurons. We also identified groups with retinal identity (SIX3, RORB, VSX2, OTX2), including Retinal Progenitor Cells, Cycling Retinal Progenitor Cells, Transient Retinal Precursor Cells, Retinal Ganglionic Cells, Amacrine/Horizontal cells, and Photoreceptors/Cones. We further found cell clusters from Forebrain Telencephalic/Diencephalic boundary, Mesenchyme, Hindbrain, and Choroid Plexus (Fig. 7G–I).

Strikingly, we observed major differences between control and CHD1L mutant organoids. We detected only 11 clusters in mutant organoids mainly due to the absence of the telencephalic clusters (i.e. radial glial cells, cycling radial glial cells, IPC, immature and excitatory neurons). More specifically, we observed that among the 2385 sequenced nuclei from CHD1L^+/+ cerebral organoids, a total of 1436 nuclei were FOXG1-positive (60.2%) whereas 169 nuclei were SIX3-positive (7.1%). On the contrary, we found only three FOXG1-positive nuclei (0.1%) among the 1634 nuclei and a total of 809 SIX3-positive nuclei (49.5%) for CHD1L^−/− (Line 1) cerebral organoids (Fig. 7J). These findings exemplified the profound cell fate difference between control and CHD1L mutant organoids, the latter expressing retinal-specific genes and thus resembling mature hiPSC-derived retinal organoids at 60 DIV [119, 120] (Fig. 7B).

Based on our snMultiomic data, we thus speculated that CHD1L could be a driver of cell-fate decisions in the forebrain. To test this possibility, we performed a SCENIC + workflow analysis to catalog the set of enhancer-driven regulons that form GRNs in both wildtype and CHD1L mutant organoids [64] (Fig. 8A and B). SCENIC + identified 55 activator and 12 repressor eRegulons (Fig. 8C). SCENIC + recovered well-known master regulators of excitatory neurons (NEUROD2, MEF2C, TBR1, and FOXG1), retinal ganglionic cells (RAX, CRX, RXRG, NEUROD1), transient retinal precursor cells (RFX2, MITF) and photoreceptors/cones (ISL1, POU2F2, EBF1/3). The majority of the top five cell-type-specific TFs showed co-binding to shared enhancers (Fig. 8D and Supplementary Fig. S14D and E). Individual eRegulon visualization further confirmed the presence of TFs favorizing telencephalic fate [FOXG1 (+), NEUROD6 (+)] and the subsequent positive regulation of their associated genes and regions in CHD1L^+/+ organoids whereas TFs associated with retinal fate specification [OTX2 (+), VSX2 (+), SIX3 (+)] were found in CHD1L mutant organoids. Of note, the eRegulon SOX2 (+) was found in both control and CHD1L mutant organoids as expected (Fig. 8E and F and Supplementary Fig. S14F). Taken together, these data revealed that CHD1L acts as a master regulator of cell-fate decisions and promotes telencephalon fate during forebrain regionalization.

Figure 8.

SCENIC + analysis reveals CHD1L function in forebrain cell-fate determination. (A) t-SNE dimensionality reduction of 4019 cells based on gene and target region enrichment scores of eRegulons. Cells are colored according to their genotypes CHD1L^+/+ and CHD1L^−/− and were analyzed for gene expression and chromatin accessibility in 60-day in vitro hCO. (B) t-SNE dimensionality reduction of organoid cells based on gene and target region enrichment scores of eRegulons. Cells are colored according to their identity. (C) Heatmap/dot-plot showing transcription factor (TF) expression of the eRegulon on a color scale and cell-type specificity (RSS) of the eRegulon on a size scale. Cell types are ordered on the basis of their gene expression similarity. (D) Overlap of target genes of eRegulons. The overlap is divided by the number of target genes of the eRegulon in each row. (E) Representative t-SNE dimensionality reduction of multiome dataset colored based on TF expression, target gene and region activity of eRegulons for telencephalic markers (FOXG1, SOX2, NEUROD6). (F) Representative t-SNE dimensionality reduction of multiome dataset colored based on TF expression, target gene and region activity of eRegulons for retinal markers (OTX2, VSX2, SIX3). AUC, area under the recovery curve; g, gene; r, region; RSS, eRegulon specificity score; TF, transcription factor.

ATPase activity of CHD1L is not required to rescue neuroanatomical and body length phenotypes

Human CHD1L has two protein-coding isoforms characterized by alternative N-termini: CHD1L-202 (ENST00000369258.8; 897 amino-acids), encoding a full-length (FL) protein isoform (hereafter referred as CHD1L-FL), and CHD1L-203 (ENST00000369259.4, 693 amino-acids), lacking the first ATPase lobe (hereafter referred as CHD1L-ΔLobe1), whose relevance to our study is due to its higher expression in brain-related tissues, as reported in GTEX (https://gtexportal.org/home/gene/CHD1L) and its expression in our hiPSC-derived hNPC (Fig. 9A and B). A functional role for CHD1L-ΔLobe1 isoform is unexpected since a construct of CHD1L lacking the first ATPase lobe has been shown to be devoid of an ATPase activity [32].

Figure 9.

CHD1L function is mediated through its macrodomain and is implicated in neurodevelopmental disorders. (A) Sashimi plot of CHD1L mRNA transcripts expressed in hNPC showing two main isoforms corresponding to CHD1L-202 (ENST00000369258.8) and CHD1L-203 (ENST00000369259.4). (B) Schematic representation of the human protein isoforms of CHD1L including CHD1L-FL (CHD1L-202), CHD1L-ΔLobe1 (CHD1L-203), and two truncated forms of CHD1L (CHD1L-FL-ΔMacro and CHD1L-ΔLobe1-ΔMacro). The homozygous variant p.Arg392His is shown on the protein. (C) Dot plot showing the distance between the eyes (head size) of 5 dpf control larvae and larvae injected with mRNA for each of the two isoforms and the two truncated forms of CHD1L. Data shown as mean ± SEM of triplicate batches; Kruskal–Wallis test performed between chd1l^−/- alone versus all the other conditions. (D) Dot plot showing the body length of 5 dpf control larvae and larvae injected with mRNA for each of the two isoforms and the two truncated forms of CHD1L. Data shown as mean ± SEM of triplicate batches; ordinary one-way ANOVA performed between chd1l^−/- alone versus all the other conditions to validate phenotypic rescue. (E) Dot plot showing the distance between the eyes (head size) of 5 dpf control larvae or larvae injected with human CHD1L-FL mRNA either WT or carrying the mutation p.Arg392His. Data shown as mean ± SEM of triplicate batches; ordinary one-way ANOVA. (F) Dot plot showing the distance between the eyes (head size) of 5 dpf control, noninjected chd1l mutant larvae and chd1l mutant larvae injected with the human CHD1L-FL mRNA either WT or carrying the mutation p.Arg392His. Data shown as mean ± SEM of triplicate batches; ordinary one-way ANOVA. (G) Ribbon representation of the structure of the self-inhibited form of CHD1L (PDB entry 7epu) in full (left panel) and with a closeup on the region harboring Arg392 residue (boxed right panel). Arg392 structures a specific region of CHD1L ATPase lobe 2 (orange) by interacting with main chain carbonyls and aspartate side chains. This region is key in interacting with the CHD1L macrodomain (blue) in the inhibited form. (H) Total protein expression levels in E. coli of the three constructs used in this study. No significant changes are observed between the WT and p.Arg392His constructs. (I) Levels of the soluble CHD1L constructs from panel (H). The p.Arg392His mutation causes a major decrease of the solubility of CHD1L constructs, irrespective of the construct considered.

We thus sought to determine whether the ATPase activity of CHD1L is necessary during neurogenesis by injecting the CHD1L-ΔLobe1 mRNA into zebrafish embryos. Similar to the effect of overexpression of the CHD1L-FL mRNA, we observed macrocephaly and increased body length upon injection of the CHD1L-ΔLobe1 mRNA (Supplementary Fig. S15A–D). Likewise, overexpression of the CHD1L-ΔLobe1 isoform was also able to rescue both reduced head size and body length of chd1l^−/− mutant zebrafish larvae (Fig. 9C and D). In addition, we explored the consequences of the ablation of the macrodomain in both CHD1L isoforms: CHD1L-FL-ΔMacro and CHD1L-ΔLobe1-ΔMacro (Fig. 9B). Neither of these truncated forms was able to rescue the smaller head and body size of the chd1l^−/− mutant zebrafish larvae (Fig. 9C and D). These in vivo data indicated that the macrodomain but not the ATPase activity is essential to CHD1L’s role during brain development and growth in a vertebrate organism, further supporting the possibility that the remodeling activity of CHD1L is not required during these developmental processes.

CHD1L contribution to neurodevelopmental and growth phenotypes

Finally, we examined whether loss of CHD1L in humans might be sufficient to cause some of the commonly observed phenotypes associated with the 1q21.1 distal region deletion. During our analyses, a maternal inherited 260 kb deletion in 1q21.1 only encompassing CHD1L was discovered in a large autism cohort study (SSC proband 12719.p1). Although inherited from an apparent unaffected mother, this 260 kb deletion was not observed in 2090 control individuals [121]. By interrogating the Decipher database (deciphergenomics.org), we found two heterozygous atypical deletions and one heterozygous atypical duplication affecting CHD1L only in individuals with autism and other neuropsychiatric traits. Furthermore, three additional atypical duplications encompassing either CHD1L and FMO5 or CHD1L and BCL9 were identified in probands with complex neurodevelopmental disorders with unknown parental status (Supplementary Fig. S16A and Supplementary Table S6).

In ClinVar database, three unrelated individuals were reported as carriers of a heterozygous CHD1L mutation that introduces a stop codon at the protein’s C-terminus (c.1929del; p.Arg643Serfs*16). This variant is classified as “pathogenic” in one individual presenting with short stature, while it is listed as “of uncertain significance” in the other carriers, for whom no clinical information is available. Moreover, we have identified the same heterozygous truncating variant in CHD1L in a 44-year-old male with ID and ASD, motor delay, speech delay, seizures, facial features, and normal growth parameters. This individual also carries two other variants of unknown significance in DLGAP2 (c.1696C > T) and PDPR (c.1147G > T). While rare, the p.Arg643Serfs*16 variant has an allele count of 67 in gnomAD which complicates its pathogenic interpretation. Of note, a deleterious nonsense de novo variant (p.Gln600*) has been found in a proband with sporadic ADHD [122]. These two truncating variants are predicted to result in truncation of CHD1L’s macrodomain reinforcing the idea that the macrodomain is necessary for CHD1L’s function during neurogenesis.

Although these genetic findings support the possibility that dosage changes of CHD1L might contribute to the neurocognitive and growth phenotypes associated with 1q21.1 distal deletions and duplications, we interpret data from atypical, shorter 1q21.1 CNVs and heterozygous deleterious CHD1L variants with caution. Population genetic data suggest that heterozygous loss-of-function in CHD1L is generally tolerated [pLI = 0; Genome Aggregation Database (gnomAD)] (Supplementary Table S6). Therefore, we propose that in the context of 1q21.1 distal deletion/duplication syndromes, some phenotypic features may be influenced or modified by additional genetic or environmental factors.

We next investigated whether CHD1L could act as a susceptibility gene for neurodevelopmental disorders in a homozygous hypomorphic state. To test this possibility, we searched for CHD1L homozygous variants in published exome data. A study focusing on the diagnostic yield of exome sequencing to identify disease genes in consanguineous families described the case of a female patient (ID: ER100167) with mild intellectual disability, microcephaly, muscular hypotonia, rigidity, ataxia, intention tremor, hypopigmented macules, electroencephalogram abnormalities [123]. The patient carries a homozygous variant in the ATPase Lobe 2 of CHD1L (NM_004284.4, c.1175G > A, p.Arg392His in CHD1L-FL protein; this position corresponds to p.Arg188His in the CHD1L-ΔLobe1 protein). The variant was predicted as “probably damaging” in PolyPhen-2 (score = 1.00 in HumDiv and score = 0.99 HumVar) and as “affecting protein function” in SIFT (score of 0.00, median protein conservation = 3.01). Furthermore, the p.Arg392His variant has not been reported in homozygous form in gnomAD.

We examined whether the p.Arg392His variant could be pathogenic by overexpressing either CHD1L-FL WT or p.Arg392His mutant mRNA in WT zebrafish embryos. We found that injection of the mutant mRNA leads to macrocephaly and increased body size; however, the amplitude of the phenotypes was significantly lower compared to the WT mRNA effect (Fig. 9E and Supplementary Fig. S16B). We next investigated whether the mutant p.Arg392His mRNA could rescue the phenotypes observed in chd1l^−/− mutants. In this context, the mutant mRNA partially rescued the head size and body measurements of the ENU mutants compared to the full rescue of both phenotypes upon injection of the WT mRNA (Fig. 9F and Supplementary Fig. S16C).

The arginine residue at position 392 (Arg392) belongs to the ATPase Lobe 2 of CHD1L and contributes to the structural organization of this second lobe. Specifically, Arg392 participates in the structuring of a specific region of the second lobe which interacts with the CHD1L macrodomain when CHD1L is in its auto-inhibited form (Fig. 9G) [36]. Arg392, however, is not in direct contact with the macrodomain, suggesting that it mostly plays a structuring role for CHD1L. Importantly, although the replacement of Arg392 by a histidine will keep the positive charge of this residue, the shorter histidine side chain will not be able to retain all interactions made by Arg392, which could alter the stability of this region, or of CHD1L.

We further investigated the impact of the p.Arg392His variant by introducing this mutation into CHD1L-FL, CHD1L-ΔLobe1, CHD1L-FL-ΔMacro constructs. We examined the expression and solubility levels of these mutants compared to the WT proteins upon expression in E. coli. Analysis of the total expression levels showed no significant differences between the WT and any of the mutant proteins (Fig. 9H). Analysis of the soluble fractions revealed that the p.Arg392His mutation caused a major decrease of the quantity of soluble proteins, irrespective of the construct considered (Fig. 9I). These data showed the important role of Arg392 not only in structuring the ATPase lobe domain but also at the full CHD1L protein level. That a fraction of p.Arg392His mutant CHD1L protein remains soluble is in accordance with our zebrafish functional data showing partial phenotypic rescue upon injection of this variant. Overall, these in vitro findings strongly indicated that a small amount of p.Arg392His CHD1L protein could still be functional explaining the hypomorphic deleterious effect seen in zebrafish. Our functional investigations underlined the fact that homozygous CHD1L mutations found in cohorts with neurodevelopmental disorders should not be discarded during variant filtration and prioritization, as they may represent candidate risk alleles that require further functional validation.

Discussion

Investigating the biology of rare but relatively penetrant CNV provides a unique opportunity to understand brain development and cellular mechanisms underlying increased susceptibility to autism. We and others have found some phenotypic drivers for several pathogenic CNV such as 15q13.3 deletions and duplications [124–128], 16p11.2 deletions and duplications (BP4–BP5 and BP2–BP3) [69, 70]; and 17q12 deletions and duplications [129] all of which are associated with increased risk for developmental and neuropsychiatric disorders.

Here, we investigated the 1q21.1 CNV-prone locus, the second most common region associated with ASD but for which the contribution of genes within the distal locus remained to be explored. Using a combination of in vivo modeling in zebrafish, mouse, human organoids along with omics analyses, our data support a major role for CHD1L in the neuroanatomical and growth phenotypes associated with the 1q21.1 distal CNV. This conclusion is supported by four key lines of evidence. First, of the eight genes within the 1q21.1 interval only CHD1L yielded macrocephaly and increased body size in the in vivo overexpression screen. Second, the reciprocal suppression of this gene mirrored the corresponding human 1q21.1 deletion phenotypes. Third, omics analyses established neurogenesis impairment that was consistent across species. Fourth, combined GWAS data and animal modeling strongly suggest its contribution to growth defects of the 1q21.1 distal deletion/duplication syndrome.

In addition to its driver role for brain and growth phenotypes of the 1q21.1 distal CNV, we found that CHD1L acts as a co-transcriptional factor for the pioneer factor SOX2, and that CHD1L loss perturbs the neurogenic program in human neuronal progenitor cells. Lastly, modeling the loss of CHD1L in self-organizing cerebral organoids revealed that CHD1L is a master regulator of cell-fate decision, specifically favoring telencephalon fate during forebrain regionalization. This work uncovers a novel function for CHD1L beyond its well-characterized role in DNA repair and its association with cancer.

Recently, the NOTCH2NL paralogs (NOTCH2NLA and NOTCH2NLB) have been proposed as candidates contributing to the neurocognitive phenotypes of 1q21.1 distal deletion/duplication syndrome. Deletion of these genes leads to premature neuronal maturation whereas ectopic expression leads to a delay in the differentiation of radial glial cells [23, 24]. Here, we confirmed that overexpression of NOTCH2NLB but not NOTCH2NLA leads to macrocephaly in zebrafish. In contrast, the simultaneous overexpression of both NOTCH2NLA/B did not affect zebrafish head size. This observation prompts us to speculate that NOTCH2NLB contributes to the brain size phenotypes associated with 1q21.1 distal CNV, but only when its copy number is increased relative to NOTCH2NLA. Reported atypical 1q21.1 CNV support this possibility; Fiddes et al. found that NOTCH2NLB was completely duplicated in all atypical duplication cases (which were all macrocephalic) and entirely deleted in all atypical deletion cases (which were all microcephalic) [23]. Further, we observed that overexpression of either NOTCH2NLA or NOTCH2NLB had no impact on body size. Taken together, we suggest that the 1q21.1 distal CNV follows a cis-epistasis complex CNV model [20] for which multiple primary drivers are sufficient to cause independent or similar phenotypes and modifiers are present to modulate the expressivity or penetrance of the phenotypes. Specifically, CHD1L and NOTCH2NLB dosage changes cause head size phenotypes and autistic traits; CHD1L dosage changes cause growth phenotypes; GJA5 overexpression is responsible for cardiac defects [130]; and GJA8 is responsible for eye defects [18, 19] with CHD1L acting as a possible modifier according to our data obtained in hCO.

Considering the variable penetrance and phenotypic expressivity, we are under no illusion that other genes outside the 1q21.1 distal region might also contribute to the associated phenotypes. For instance, it has been proposed that 3D proximity of gene loci supports the notion of a co-regulation mechanism of genes with related function [131]. One can think about possible co-regulation of autism genes or CNV in -cis and -trans at the chromatin level. Of note, recent chromatin conformation analysis showed that the 16p11.2 phenotypic modifiers MVP and MAPK3 have long-range chromatin interactions with PTEN and CHD1L, respectively [132]. Analogous to 1q21.1 distal CNV, deletions and duplications of the 16p11.2 BP4–BP5 interval are linked to macro- and microcephaly, respectively [69]. Regulatory chromatin loops between ASD susceptibility loci/regions should be thus further explored to characterize better the relationship between ASD driver genes, their cellular functions in time and space, and how their modulation affects the penetrance and expressivity of the human phenotypes.

Human growth is a highly complex and multifactorial trait. Rare CNV including the 1q21.1 deletions are a relatively common cause of short stature [14]. Our zebrafish data indicate that larval body size can be modulated by CHD1L dosage changes only and we found that several genes associated with short stature in humans are downregulated in CHD1L mutant hNPC. Of note, Chd1l mutant mice also exhibit decreased body size and decreased body weight [133]. Moreover, the height associated SNP rs6658763 has been found in linkage disequilibrium with two CHD1L nonsynonymous variants (r²= 0.907) in a large genome wide association study on human adult height [80] which further supports the possibility that CHD1L influences metabolism and growth. Recently, we interrogated the most comprehensive database of human genome wide association studies (GWAS), the NHGRI-EBI GWAS catalog [134]. We found four additional large GWAS that support a link between CHD1L and human height [135–138]. Moreover, the musculoskeletal Knowledge Portal (MSK-KP) indicates that CHD1L common variants are strongly associated with height (P= 4.50e-36; Sample size: 2248 846 individuals; https://msk.hugeamp.org/). Additional experiments will be required to determine how CHD1L controls these biological processes, and which cells are sensitive to CHD1L dosage during development.

CHD1L belongs to the SNF2 superfamily [139] and is a poly (ADP-ribose) and ATP-dependent remodeler, with a role in chromatin relaxation. It is composed of a two-lobed catalytic Snf2-like ATPase domain, which is connected through a linker region to a C-terminal macrodomain that mediates PARP1 activity-dependent chromatin-targeting [28–32]. Its involvement in the early cellular response to DNA damage was elucidated, providing evidence of the presence of an auto-inhibitory mechanism regulated by the interaction of the CHD1L’s macrodomain and the bilobate ATPase module [28–35]. DNA-damage-mediated PARP1 induction suppresses this interaction, allowing release from auto-inhibition, activation and ultimately chromatin decompaction [36, 139]. Our proteomic data confirmed that BER factors and PARP1 are protein partners of CHD1L in hNPC. Published data on adult Chd1l^−/− mice does not support DNA repair defects as a potential cause for the increased apoptosis observed in our zebrafish experiments; Chd1l^−/− mice exhibit neither elevated levels of DNA damage across a variety of tissues nor a shortened lifespan. This suggests that DNA repair deficiencies are not responsible for the apoptotic changes we observed in zebrafish larvae [133]. The fact that a Chd1l mutant murine model is not lethal contradicts a previous study reporting that Chd1l was necessary for embryo preimplantation [37]. Another study reports that Chd1l has a role during cell reprogramming during the initial step of this process in mice, and that this requires both its macrodomain and DNA helicase activity [140]. Strikingly, our study contradicts the importance of CHD1L’s DNA helicase activity. The presence of a functional CHD1L isoform lacking the first ATPase lobe in humans, along with our zebrafish data, strongly suggests that CHD1L’s remodeling activity is dispensable for brain development and body growth in vivo.

To date, CHD1L is best known for his role during tumorigenesis [140, 141]. More recently, CHD1L has been associated with HIV-1 replication rate [142]. To our knowledge, single nucleotide variants of CHD1L have not been linked to rare neurodevelopmental disorders, possibly due to its high tolerance to heterozygous loss-of-function that have prevented clinicians from considering CHD1L heterozygous variants as relevant in complex neurodevelopmental disorders. Based on our modeling data, we conclude that CHD1L homozygous missense variants should be considered as susceptibility alleles in the future. Heterozygous missense mutations in CHD1L have been associated with congenital anomalies of the kidneys and urinary tract also known as CAKUT [143]; this is in accordance with our transcriptomic data indicating that CHD1L regulates genes involved in kidney/nephron development. In addition, we recently showed that hypermethylation at 1q21.1 locus influences CHD1L dosage and constitutes a susceptibility to multiple sclerosis; the loss of chd1l leads to oligodendrogenesis impairment and reduced axonal tract projections in zebrafish [144], which reflects that CHD1L has the potential to control cell fate of various neural lineages.

A common feature seen in cellular models of autism is synaptic defects (e.g. SHANK3 and FMR1) [145–148]. Several studies looking at cellular phenotypes of other CNV such as 2p16.3/NRXN1, 15q13.3, 16p11.2, and 22q11.21 have also shown synaptic dysfunction [128, 149, 150]. Our transcriptomic data revealed that loss of CHD1L leads to decreased expression of genes involved in biological processes such as “neuron differentiation” and “synapse assembly” in hNPC suggesting that CHD1L dosage affects synaptic function. In fact, through knockdown experiments in hiPSC-derived neurons, we recently demonstrated that CHD1L-deficient neurons exhibited branching abnormalities whereas synaptic density appeared unaffected. Additionally, CHD1L knockdown led to reduced calcium signal intensity in neurons, which suggests diminished electrical activity in neurons with lower CHD1L expression [144]. Taken together, these in cellulo data confirm that CHD1L expression is necessary to generate electrically competent neurons and point to impairment of calcium dynamic in CHD1L-deficient neurons. Chapman and co-workers showed that human iPSC-derived neurons carrying 1q21.1 distal deletion or duplication also exhibit altered synaptogenesis, differential expression of calcium channels, and aberrant neural network activity [21]. It is thus reasonable to speculate that cognitive impairment in 1q21.1 distal CNV carriers is due, in part, to altered synaptogenesis and synaptic dysfunction mediated by CHD1L dosage imbalance.

Finally, we explored the notion of regionalization during brain development. During our investigation, several lines of evidence pointed to a critical role of CHD1L during brain regionalization. First, loss of CHD1L impaired the expression of genes controlled by NKX2-2, ERG, OTX2, BRD4, and SOX2, known transcriptional factors regulating cell fate and regionalization of the brain [107, 108, 114, 115, 151, 152]. Second, CHD1L binds the SOX and EGR2 motifs in hNPC and our CUT&RUN data are similar to OTX2 and SOX2 ChIP-seq data. Third, proteomic data showed that CHD1L interacts directly or indirectly with the pioneer TFs SOX2, OTX2, PAX6, and components of the NURD complex. Notably, the possibility that CHD1L acts during brain regionalization is supported by the observation of structural differences in brain architecture in individuals with 1q21.1 CNV [153]. Combining self-organizing cerebral organoids with single-nuclei genomic technologies, we determined that CHD1L-deficient cerebral organoids exhibit cell fate anomalies including enrichment of SIX3- and OTX2-positive nuclei and detection of nuclei clusters such as retinal precursor cells and photoreceptors at the expense of FOXG1-positive cells such as excitatory neurons normally expected in cerebral organoids. Of note, loss of function of SOX2, OTX2, PAX6, leads to eye abnormalities in humans such as microphthalmia (reduced eye size), anophthalmia (no eye formation at all), and coloboma [154]. Similarly, some individuals with 1q21.1 distal deletion exhibit abnormally small eyes, coloboma and lens abnormalities. These eye abnormalities could be thus attributed to CHD1L based on its physical interaction with SOX2, OTX2, PAX6 but also to GJA8, for which heterozygous mutations have already been associated with eye phenotypes [18, 155]. To our knowledge, interaction between CHD1L and SOX2 in hNPC is novel. Notably, Liu and co-workers have shown that Parp1 stabilizes Sox2 binding to nucleosome which facilitates its pioneer activity at less accessible regions of the genome [156]. Therefore, it is reasonable to speculate that a “ménage à trois” involving SOX2, PARP1, and CHD1L exists at some specific SOX2 motifs, that activates the neurogenic program. Further experiments will be needed to test this possibility and to determine whether the ability of CHD1L to bind nucleosomes is necessary in the reported Parp1–Sox2 protein complex in mice.

Overall, we showed that integrating in vivo and in cellulo modeling of CNVs across multiple vertebrate species with single-cell technologies is crucial for pinpointing phenotypic driver genes and unraveling the complex interplay between phenotypic drivers and modifiers. Our functional pipeline is particularly effective for exploring numerous genomic regions implicated in a wide range of human genomic disorders. Furthermore, this study highlights CHD1L as a master regulator of cell-fate in the forebrain, along with its identified transcription partners, exemplifying how autism and morphometric traits can result from diverse developmental insults. This underscores the critical importance of characterizing the developmental function of autism susceptibility genes in order to guide future therapeutic approaches.

Limitations of the study

Our study aimed to identify gene(s) within the 1q21.1 region whose dosage variations contribute to mirrored phenotypes observed in individuals with 1q21.1 syndromes. We acknowledge that loss-of-function in other genes within the 1q21.1 region may also lead to developmental phenotypes, which were not investigated in this study. We also acknowledge that transient, ectopic gene expression in zebrafish embryos may not fully model certain endogenous gene functions, but it still provides valuable insights. We did not examine the effects of heterozygous loss of CHD1L in hiPSC-derived tissues. Finally, our comparison of CHD1L-bound regions with publicly available ChIP-seq data is subject to limitations inherent in cross-study analyses, including differences in methodology, tissue origin, and data processing pipelines.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was funded by Agence Nationale de la Recherche under the projects JCJC-ANR-17-CE12-0006 and ANR-22-CE12-0011 and Fondation de France (WB-2022–45 868) (C.G.). This work of the Interdisciplinary Thematic Institute IMCBio+, as part of the ITI 2021–2028 program of the University of Strasbourg, CNRS and Inserm, was supported by IdEx Unistra (ANR-10-IDEX-0002), and by SFRI-STRAT’US project (ANR-20-SFRI-0012) and EUR IMCBio (ANR-17-EURE-0023) under the framework of the France 2030 Program. This work was funded by the French National Research Agency (ANR) through the Programme d’Investissement d’Avenir under contract ANR-10-LABX-0030-INRT grant under the frame programme Investissement d’Avenir ANR-10-IDEX-0002–02 and ITMO Cancer of Aviesan within the framework of the 2021–2030 Cancer Control Strategy on funds administered by Inserm. We acknowledge the IGBMC Imaging Facility, member of the national infrastructure France-BioImaging supported by the French National Research Agency (ANR-10-INBS-04). We are grateful to the staff of the IGBMC PluriCell East Facility, Flow Cytometry Facility, the Mass-Spectrometry Facility, the GenomEast Platform, member of the “France Génomique” consortium (ANR-10-INBS-0009). GM8330-8 hiPSC line was kindly provided by Michael E. Talkowski. C.G. is a permanent INSERM investigator. M.V.L. is a doctoral fellow supported by EUR IMCBio funds and Fondation de France (WB-2022-45 868). M.N.L. is a SNSF Swiss Postdoctoral Fellow and a FRM Postdoctoral Fellow (SPF20170938810). Funding to pay the Open Access publication charges for this article was provided by Agence National de la Recherche.

Data availability

RNA-seq, ATAC-seq, CUT&RUN, single-nuclei multiome fastq files have been uploaded to GEO with the accession number GSE266032. All reagent information can be found in Supplementary Table S7. Supplementary Figs S1–16 and Supplementary Tables S1–S7 are available.