Loss of function of the zinc finger homeobox 4 gene, ZFHX4, underlies a neurodevelopmental disorder

Summary

8q21.11 microdeletions involving ZFHX4 have previously been associated with a syndromic form of intellectual disability, hypotonia, unstable gait, and hearing loss. We report on 63 individuals—57 probands and 6 affected family members—with protein-truncating variants (n = 41), (micro)deletions (n = 21), or an inversion (n = 1) affecting ZFHX4. Probands display variable developmental delay and intellectual disability, distinctive facial characteristics, morphological abnormalities of the central nervous system, behavioral alterations, short stature, hypotonia, and occasionally cleft palate and anterior segment dysgenesis. The phenotypes associated with 8q21.11 microdeletions and ZFHX4 intragenic loss-of-function (LoF) variants largely overlap, although leukocyte-derived DNA shows a mild common methylation profile for (micro)deletions. ZFHX4 shows increased expression during human brain development and neuronal differentiation. Furthermore, ZFHX4-interacting factors identified via immunoprecipitation followed by mass spectrometry (IP-MS) suggest an important role for ZFHX4 in cellular pathways, especially during histone modifications, protein trafficking, signal transduction, cytosolic transport, and development. Additionally, using CUT&RUN, we observed that ZFHX4 binds the promoter of genes with crucial roles in embryonic, neuronal, and axonal development. Moreover, we investigated whether the disruption of zfhx4 causes craniofacial abnormalities in zebrafish. First-generation (F0) zfhx4 crispant zebrafish, a (mosaic) mutant for zfhx4 LoF variants, have significantly shorter Meckel’s cartilage and smaller ethmoid plates compared to control zebrafish. Behavioral assays showed a decreased movement frequency in the zfhx4 crispant zebrafish in comparison with controls. Furthermore, structural abnormalities were found in the zebrafish hindbrain. In conclusion, our findings delineate a ZFHX4-associated neurodevelopmental disorder and suggest a role for zfhx4 in facial skeleton patterning, palatal development, and behavior.

Graphical abstract

We associate ZFHX4 loss of function with syndromic intellectual disability. While ZFHX4 deletions showed a mild methylation profile, protein-truncating variants did not. Multi-omics data implicate ZFHX4 interactors in transcriptional regulation and development and its target genes in neural development. Zebrafish zfhx4 crispants exhibit craniofacial/medulla oblongata defects and reduced movement frequency.

Introduction

Neurodevelopmental disorders (NDDs) affect approximately 2%–5% of children worldwide. In up to 40% of the affected individuals, a monogenic cause can be identified, but many gene-disease relationships in NDDs remain to be discovered.¹ In 2011, Palomares et al. reported eight unrelated individuals with a recognizable intellectual disability syndrome (MIM: 614230), all sharing a (sub)microscopic deletion in the 8q21.11 region encompassing ZFHX4.² Since then, others have reported individuals with 8q21.11 deletions ranging from 0.66 to 14.5 Mb in size, all with similar clinical characteristics.^3,4,5,6,7,8 In 2019, a loss-of-function (LoF) ZFHX4 (MIM: 606940) splice variant was reported in a 7-year-old male from a cohort of individuals with apraxia of speech.⁹ In 2020, ZFHX4 was identified as a key NDD candidate gene in a large-scale exome sequencing study.¹⁰ Hence, ZFHX4 haploinsufficiency might be one of the drivers of the 8q21.11 microdeletion phenotype. In addition, Bishop et al. identified ZFHX4 as a candidate gene for orofacial cleft.¹¹ In both exome-driven studies, de novo ZFHX4 LoF variants appeared to be enriched in the affected individuals. In alignment with these data, the latest gnomAD version (v.4.1.0) predicts ZFHX4 to be intolerant to protein-truncating variants (PTVs), with a pLI (probability of being LoF intolerant) score of 1 and a constraint score of observed/expected intolerance of 0.24.¹²

ZFHX4 comprises 12 exons and encodes a transcription factor that carries 22 zinc fingers and four homeodomains.^13,14 Expression of ZFHX4 is high in the developing brain of different species^1,3,4,5 and has been shown to decrease in adult mice.¹⁵ In addition, Zfhx4 is highly expressed in cartilage.¹⁶ In mice, Zfhx4 deficiency is associated with cleft palate, and its expression in the palatal shelves supports a role for Zfhx4 in palatal development.¹⁶ Homozygous Zfhx4 knockout mice die of neurogenic respiratory disorders within 24 h after birth.¹⁷ In humans, ZFHX4 expression has been detected in adult brain, muscle, and liver.¹³ Altogether, ZFHX4 has been identified as a transcription factor involved in neural cell maturation, region-specific brain differentiation,¹⁸ and skeletal development.¹⁶

Moreover, in accordance with a role in cell maturation, ZFHX4 appears to be involved in the biology of different types of cancer, such as glioblastoma,¹⁹ esophageal squamous cell carcinoma (ESCC),^20,21 and ovarian cancer.^22,23 Chudnovsky et al. showed that ZFHX4 regulates the therapy-resistant tumor-initiating (TIC) glioblastoma cell state by interacting with CHD4, a core member of the nucleosome remodeling and deacetylase (NuRD) complex.¹⁹ Additionally, several studies showed that in ESCC and ovarian cancer, ZFHX4 expression can be used as a prognostic biomarker for survival outcomes in individuals with metastasis.

Here, we report a cohort of 63 individuals with ZFHX4 aberrations. We evaluate the common features and differences between ZFHX4 PTVs and copy-number variants (CNVs), including ZFHX4, to inform clinical management. Our observations confirm that ZFHX4 LoF drives the microdeletion phenotype. Furthermore, we evaluated the protein and DNA interactome of ZFHX4, indicating an important role for ZFHX4 in normal cell and tissue maturation and neurodevelopmental pathways, including axon development. Finally, zebrafish zfhx4 crispants show craniofacial malformations and medullary and behavioral abnormalities. Overall, our data indicate that ZFHX4 is essential for neural and craniofacial development.

Subjects, material, and methods

Recruitment of affected individuals

Individuals with ZFHX4 alterations were identified through Clinical Genetics Services worldwide using web-based databases such as GeneMatcher^22,24 and DECIPHER. The clinical research activity relating to this report has been in accordance with the Declaration of Helsinki on the ethical principles for medical research involving human subjects. Genetic tests were performed clinically or using research protocols approved by each center’s ethics committee, and proper informed consent was obtained. Informed consent for genetic analyses, clinical data and biological sample collection, and publication of relevant findings were obtained for all individuals. Consent to publish photographs was provided for the individuals shown in Figure 1. We collected the detailed phenotype data through a standard form specific to ZFHX4 (Tables S1 and S2).

Figure 1.

Front profile pictures and clinical representation of individuals with a deletion or PTV affecting ZFHX4

(A) Rows 1 and 2: frontal photographs of probands with a deletion containing ZFHX4. ^∗Proband 1 has an insertion breakpoint within ZFHX4. Rows 3 and 4: frontal photographs of probands with a ZFHX4 protein-truncating variant (PTV). °Proband 57 has compound heterozygous PTVs affecting ZFHX4.

(B) Clinical presentation of individuals with ZFHX4 aberration. De-identified face mask of images. Top left: a total of 31 frontal photographs of 24 individuals with a deletion affecting ZFHX4. Bottom left: a total of 26 facial photographs of 18 individuals with predicted LoF variants in ZFHX4. Top right: matched healthy individuals. Note that no significant differences are shown in the facial features of both affected groups (i.e., ZFHX4 deletion vs. ZFHX4 PTV). The consent forms of the patients for the publication of the clinical photographs are available for this study.

Statistics

Statistical studies were performed in SAS 9.4 (SAS Institute, Cary, NC, USA). Comparative studies between females and males and between CNVs and LoF variants were calculated using a two-tailed Fisher’s exact test. p values <0.05 were considered statistically significant. Counts for clinical characteristics were given for those where the respective information was available; therefore, the total count can be lower than the total number of individuals.

Facial feature analysis

In order to make the description of facial features as homogeneous as possible, the facial features of all individuals were evaluated by two experienced dysmorphologists, F.S.-S. and S.G.-M. The DeepGestalt technology by Face2Gene (FDNA, Boston, MA, USA) was used to validate the presence of distinct facial patterns. The cohort was divided into two groups to analyze facial features. The first group included a total of 31 frontal photographs of 20 individuals with CNVs. The second group included 26 facial photographs of 18 individuals with predicted LoF variants in ZFHX4. In both cases, age-, sex-, and ethnicity-matched healthy individuals were used as a comparison cohort to generate a composite image of the affected individuals. The composite image was generated in both cases using DeepGestalt facial analysis, and an assessment of the binary comparison between controls and affected individuals was performed by measuring the receiver operating characteristic (ROC) curves and the corresponding area under the curve (AUC) as described by Mak et al.²⁵

Cell culture

Fibroblast culture, treatment, and RNA extraction

Fibroblasts derived from skin biopsies were cultured in DMEM supplemented with fetal calf serum (FCS), penicillin/streptomycin, kanamycin, non-essential amino acids (NEAAs), and Fungizone until we had at least 600,000 cells. Afterward, 300,000 cells were seeded in a P60 plate on day 1 for cycloheximide treatment. On day 2, the culture medium was replenished with or without cycloheximide (5 μg/mL, C7698, Sigma-Aldrich), and on day 3, RNA was extracted with the Maxwell. In short, the cells were loosened with a cell scraper and transferred to a 15 mL tube, which was then centrifuged for 3 min at 300 relative centrifugal force (rcf). Afterward, the supernatant was removed, and 200 μL 1-thioglycerol/homogenization solution (Promega) was added. The cells were then vortexed, and 200 μL lysis buffer (Promega) was added. Finally, the sample (400 μL) was transferred to a well from the Maxwell, and RNA was extracted with the RSC simplyRNA Tissue Kit (Promega) following the RSC simplyRNA Cells method.

Induced pluripotent cells

The induced pluripotent stem cell (iPSC) line was commercially obtained from Axol BioScience and maintained in mTESR Plus (STEMCELL Technologies, #100-0276) on coated cultured plates with Geltrex LDEV-free hESC-qualified reduced growth factor basement membrane matrix (Gibco, Thermo Fisher Scientific) according to the manufacturer’s instructions. The iPSCs were cultured at 37°C in a 5% CO₂ humid atmosphere. Double volume was replaced every 2 days as a fixed feeding interval.

Neural progenitor cells

The neural progenitor cells (NPCs) were differentiated starting with 0.5 × 10⁶ of the abovementioned human iPSCs (hiPSCs) on coated cultured plates with Geltrex LDEV-free hESC-qualified reduced growth factor basement membrane matrix (Gibco, Thermo Fisher Scientific), proceeding with the monolayer protocol using the STEMdiff SMADi neural induction kit (STEMCELL Technologies, #05835) according to the manufacturer’s instructions for 6–9 days and further cultured up to 18–21 days prior to freezing. The NPCs were cultured at 37°C in a 5% CO₂ humid atmosphere. Daily full-medium changes were applied at a fixed feeding interval until the cells were ready to be passaged.

Neurons (iPSC-derived NPCs)

The chemically induced neurons were differentiated starting with the above-mentioned 1 × 10⁶ NPCs on coated cultured plates with Geltrex LDEV-free hESC-qualified reduced growth factor basement membrane matrix (Gibco, Thermo Fisher Scientific), proceeding with the monolayer protocol using the BrainPhys neuronal medium (STEMCELL Technologies, #05795) according to the manufacturer’s instructions. The neurons were cultured at 37°C in a 5% CO₂ humid atmosphere. Daily full-medium changes were applied at a fixed feeding interval until the cells were ready to be passaged.

Hi-C library preparation and analysis

A Hi-C library was generated according to the in situ Hi-C protocol adopted by the 4D Nucleome consortium,²⁶ with a few adaptations. Briefly, five million fibroblasts were crosslinked using 2% formaldehyde, lysed, snap frozen in liquid nitrogen, and stored at −80°C until further processing. Pre-lysed, crosslinked nuclei were then digested overnight using 250 U DpnII restriction enzyme (New England Biolabs, R0543L). DNA ends were marked by incorporating biotin-14-dATP (Life Technologies, 19524-016) and ligated for 4 h using 2,000 U T4 DNA ligase (New England Biolabs, M0202L). Subsequently, crosslinks were reversed overnight using proteinase K (Qiagen, 19131), and Hi-C template DNA was purified using 1× AMPure XP beads (Beckman Coulter, A63881) and stored at 4°C until library preparation. Hi-C template DNA was sheared to a size of 300–500 bp using microTUBE snap caps (Covaris, 520045) in a Covaris M220 sonicator, and MyOne streptavidin T1 beads (Life Technologies, 65601) were used to pull down biotinylated ligation junctions. Next, the sample was split into 5 μg aliquots for sequencing library preparation using the NEBNext Ultra II DNA Library Prep Kit (New England Biolabs, E7645L) and NEBNext multiplex oligos (New England Biolabs, E7335L). The amplified library was purified and size selected using 0.55× and 1.2× AMPure XP beads (Beckman Coulter, A63881). Pooled libraries were sequenced on an Illumina NovaSeq 6000 using 100 bp paired-end reads to a depth of ∼500 million reads.

FASTQ files containing raw sequencing data were processed into a Hi-C contact matrix containing both raw and normalized counts using the Juicer pipeline (v.1.6)²⁷ with BWA-MEM mapping (v.0.7.17)²⁸ to the hg38 reference genome. A processed and normalized Hi-C matrix for control fibroblasts was obtained from Melo et al.²⁹ We used FAN-C³⁰ to visualize normalized Knight-Ruiz (KR) and observed/expected affected individual vs. control Hi-C matrices for the region of interest.

Expression profiling

The RNA of cultured cells was isolated using the Direct-zol RNA MiniPrep Kit (Zymo Research) according to the manufacturer’s instructions and the concentration determined by NanoDrop (Thermo Fisher Scientific). cDNA synthesis was performed using the iScript Advanced cDNA Synthesis Kit (Bio-Rad) according to the manufacturer’s instructions. Subsequently, RT-qPCR was performed with a PCR mix containing 5 ng cDNA, 2.5 μL SsoAdvanced SYBR qPCR mix (Bio-Rad), and 0.25 μL forward and reverse primers (250 μM concentration, Integrated DNA Technologies [IDT]). RT-qPCR was performed on an LC-480 device (Roche), and gene expression levels were analyzed using the qBase+ software 3.2 (Biogazelle). The software qBase+ uses a generalized model of the delta-delta-Ct approach, thereby supporting the use of gene-specific amplification efficiencies and normalization with multiple reference genes, with up to 204 genes. All formulas of this model are detailed in Hellemans et al.³¹ The RT-qPCR primers used in this study can be found in Table S15. Three biological and two technical replicates were collected per time point.

Cell harvesting and protein extraction

The cells were scraped in PBS and collected in 15 mL falcons. The collected cells were centrifuged at 4°C for 5 min at 2,000 rpm. The radioimmunoprecipitation assay (RIPA) buffer (250 mg sodium deoxycholate, 40.5 mL water, 5 M NaCl, 1 M Tris-HCl, 10% SDS solution, and 10% NP-40) is used to lyse the cells for 45 min, and following centrifugation (10 min, 4°C, 8,000g), the proteins were extracted. The concentration was determined using the BCA protein assay kit (Thermo Fisher Scientific, catalog no. 23225).

Immunoprecipitation followed by mass spectrometry

Protein from 15–20 million cells was extracted as mentioned above (cell harvesting and protein extraction). Next, 2 μg antibody (anti-ZFHX4 antibody, HPA023837, rabbit [DA1E] monoclonal antibody [mAb] immunoglobulin G [IgG] XP isotype control #3900S for wild-type ZFHX4 immunoprecipitation sample [WTIP] vs. immunoglobulin G immunoprecipitation sample [IGGIP] sample groups, respectively) was added, and after 4 h, 20–40 μL extensively washed protein UltraLink resin beads (Thermo Fisher Scientific, catalog no. 53139) were added and further rotated overnight. The samples were centrifuged and washed five times with 50 mM digestion buffer (50 mM Tris-HCl [pH 8] and 2 mM CaCl₂) to remove unbound proteins. Beads containing the antibody and proteins of interest were resuspended in 175 μL digestion buffer. For immunoprecipitation (IP), 25 μL of samples was incubated at 95°C for 10 min with 6.25 μL of 5× Laemmli buffer at 1,000 rpm to elute the proteins. Supernatants containing the proteins of interest were used for western blot analysis. For IP followed by mass spectrometry (IP-MS), the remaining 150 μL of the samples was incubated for 4 h with 1 μg trypsin (Promega) at 37°C while shaking. Beads were removed by centrifugation at maximum speed, and proteins were digested again with 1 μg of trypsin overnight at 37°C.

Peptides were re-dissolved in 20 μL loading solvent A (0.1% trifluoroacetic acid in water/acetonitrile [ACN] [98:0.5, v/v]), of which 2 μL was injected for liquid chromatography-tandem MS (LC-MS/MS) analysis on an Ultimate 3000 ProFlow nanoLC system in-line connected to an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific). Trapping was performed at 20 μL/min for 2 min in loading solvent A on a 5 mm trapping column (Thermo Fisher Scientific, 300 μm internal diameter [I.D.], 5 μm beads). The peptides were separated on a 250 mm Aurora Ultimate, 1.7 μm C18, 75 μm inner diameter (Ionopticks), kept at a constant temperature of 45°C in a butterfly heater (Phoenix S&T). Peptides were eluted by a non-linear gradient starting at 1% MS solvent B and reached 26.4% MS solvent B (0.1% formic acid [FA] in ACN) in 30 min, 44% MS solvent B in 38 min, and 56% MS solvent B in 40 min, followed by a 5 min wash at 56% MS solvent B and re-equilibration with MS solvent A (0.1% FA in water), all at a flow rate of 250 nL/min. The mass spectrometer was operated in data-independent mode, automatically switching between MS and MS/MS acquisition. Full-scan MS spectra ranging from 390 to 910 m/z with a target value of 4 × 10⁵, a maximum fill time of 50 ms, and a resolution of 60,000 were followed by 30 quadrupole isolations with a precursor isolation width of 10 m/z for higher-energy collisional dissociation (HCD) fragmentation at a nominal collision energy (NCE) of 34% after filling the trap at a target value of 4 × 10⁵ for a maximum injection time of 54 ms. Mass spectrometry second stage (MS2) spectra were acquired at a resolution of 30,000 at 200 m/z in the Orbitrap analyzer without multiplexing. The isolation intervals were set from 400 to 900 m/z with a width of 10 m/z using window placement optimization.

QCloud^32,33 was used to control instrument longitudinal performance during the project.

MS data analysis

LC-MS/MS runs of all samples were searched together using the Data-Independent Acquisition-Neural Network (DIA-NN) algorithm (v.1.9) library free. Spectra were searched against the human protein sequences in the Swiss-Prot database (database release v.2022_01), containing 20,588 sequences (www.uniprot.org) and the imbedded protein contaminant database. Enzyme specificity was set C-terminal to arginine and lysine, also allowing cleavage at proline bonds with a maximum of two missed cleavages. Variable modifications were set to oxidation of methionine residues and acetylation of protein N termini. Mainly default settings were used, except for the addition of a 400–900 m/z precursor mass range filter, the match between runs (MBR) option. Further data analysis of the shotgun results was performed with an in-house script in the R programming language, v.4.2.2. Protein expression matrices were prepared as follows: the DIA-NN main report output table was filtered at a precursor and protein library q value cutoff of 1%, and proteins identified by at least one proteotypic peptide were retained. After pivoting into a wide format, iBAQ intensity columns were then added to the matrix using the DIAgui’s R package get_IBAQ function (https://rdrr.io/github/mgerault/DIAgui/man/DIAgui.html). PG.MaxLFQ intensities were log2 transformed, and replicate samples were grouped. Proteins with less than three valid values in at least one group were removed, and missing values were imputed from a normal distribution centered around the detection limit (package DEP),³⁴ leading to a list of 2,795 quantified proteins in the experiment, which were used for further data analysis. To compare protein abundance between pairs of sample groups (WTIP vs. IGGIP sample groups), statistical testing for differences between two group means was performed using the package limma.³⁵ Statistical significance for differential regulation was set to a false discovery rate (FDR) of <0.05 and a fold change (FC) of >4- or <0.25-fold (|log2FC| = 2). Results are presented in Tables S8–S11. Z scored PG.MaxLFQ intensities from significantly regulated proteins were plotted in a heatmap after non-supervised hierarchical clustering. We analyzed the significantly enriched genes from the MS data analysis with thousands of annotated gene set libraries retrieved from high-quality projects with the EnrichR tool,^36,37,38 and the results are visualized with a horizontal bar graph, where the bars plotted horizontally show the gene sets found in different libraries and their corresponding pathways along the x axis with the combined score with the measured p values, q values, and Z scores, which represents the measure of significance of enrichment level of the associated pathways.

Western blot

Western blot was performed according to an in-house protocol. Either protein lysates or eluted proteins after IP were subjected to NuPage gel (Thermo Fisher Scientific, catalog no. EA0375BOX) electrophoresis, subsequently transferred to a nitrocellulose membrane, and detected using the appropriate antibodies: anti-ZFHX4 antibody (Sigma-Aldrich, HPA023837), anti-ZFHX4 antibody (Abnova, H00079776-M02), and anti-vinculin (Merck Life Sciences, V9131). Blots were stripped twice. Imaging was performed with an Amersham Imager 600 CCD camera (GE Healthcare, Chicago, Illinois, USA).

Zebrafish experiments

Breeding

Zebrafish lines were housed at the zebrafish facility in Ghent in semi-closed recirculating housing systems (ZebTEC and WTU systems, Tecniplast) at a constant temperature (27°C–28°C), pH (∼7.5), conductivity (∼550 μS), and light-dark cycle (14/10 h). Zebrafish were fed every day with dry food (Gemma Micro) and with artemia (Ocean Nutrition). Zebrafish lines were maintained according to standard protocols.^39,40 Embryos/larvae were raised in E3 medium⁴¹ in a Petri dish and placed in an incubator at 28°C until 5 days post-fertilization (dpf), the age at which experiments were performed.

zfhx4 crispant RNP complex generation and injections

We followed the protocol from Kroll et al.⁴² for the generation of first-generation (F0) knockout larvae for a single gene. Following the proper guidelines and by using CRISPOR,⁴³ we were able to find the best corresponding gRNAs for exons 2, 3, and 4 of zfhx4. The CRISPR RNAs (crRNAs) were designed and produced by IDT and checked in InDelphi.⁴⁴ The possible off targets with the highest probability found by CRISPOR were assessed via PCR and sequencing on a Miseq instrument (Illumina). Sequences of the zfhx4 crRNAs, scrambled crRNAs, and the corresponding primers used for genotyping of the injected embryos are mentioned in Table S16.

To assemble the ribonucleoprotein (RNP), 2 nmol crRNA and 5 nmol trans-activating CRISPR RNA (tracrRNA) pellets were resuspended in 10 or 25 μL, respectively, of the corresponding duplex buffer (IDT) to obtain a 200 μM stock of each. These were stored at −20°C and aliquoted to protect the RNA from the continuous thaw-freeze cycles and longer storage time. The crRNA was annealed with the tracrRNA to form the gRNA by mixing 1 μL crRNA with 1 μL tracrRNA and 1.51 μL duplex buffer. To obtain the duplex crRNA:tracrRNA (61 μM), this mix was heated to 95°C for 5 min and then cooled on ice. Each crRNA:tracrRNA complex was individually mixed with an equal molar amount of Cas9. Each RNP complex was incubated for 5 min at 37°C and had a concentration of 30.5 μM. The three RNP solutions were pooled in equal amounts (10.1 μM and 2 μL), making a total mix of 6 μL. The RNPs were stored at −20°C before the injections.

Around 1 nL (experiment one) and 1 nL (repeat experiment) of the three-RNP pool were injected into the single-cell yolk before cell inflation. This amounted to around 5,029/800 pg of Cas9 and 1,070/1,000 pg of total gRNA. Each unique RNP was present in equal amounts in the pool. Therefore, in the case of three RNPs, 10.1/9.5 fmol of each RNP was injected in total with the FemtoJet injector. The solution contained a dye to follow the successful injections.

Behavioral studies

Locomotor activity in response to light and dark stimuli was analyzed in zebrafish larvae (5d pf) using the DanioVision observation chamber (Noldus, the Netherlands) equipped with a Basler GenICam camera capturing 30 fps for image acquisition. Data were recorded and analyzed using the EthoVision XT 15 software (Noldus, the Netherlands). At 5 dpf, zfhx4 crispants and scrambled zebrafish were transferred individually to a 24-well plate containing 1 mL of E3 medium. Plates were subsequently placed in the DanioVision observation chamber with a protocol consisting of 20 min of 5% light (corresponding to 139 lux) and 20 min dark, followed by alternating periods of 5 min 5% light and 5 min dark, for a total of 4 cycles. During the experiment, the temperature was kept constant at 28°C using a temperature control unit (Noldus). The first 10 min of the experiment were considered an acclimatization period and were not used for analyses. For each zebrafish, center-point detection was used. Movement frequency is depicted in time intervals of 1 min. This behavior experiment was repeated twice with n = 60/genotype.

Genotyping

From each experimental and control condition, 30 5 dpf larvae were collected, and DNA was extracted. The PCR reactions were performed on the exons of the sequence of interest (Table S16) following paired-end 2 × 150 bp sequencing on a Miseq instrument (Illumina) to determine the knockout efficiency, following the protocol by Clement et al.⁴⁵ The primers were used in the following PCR reaction: 2 min at 94°C, 12× (30 s at 94°C, 30 s at 60^∗°C, and 1 min at 72°C), and 25× (40 s at 94°C, 40 s at 48°C, 30 s at 72°C, and 10 s at 72°C). The symbol “^∗” indicates the lowering of the temperature by 1°C in each cycle. The primers, target sequences, and knockout efficiencies can be found in Table S16.

Alcian blue stainings and imaging

The following protocol is based on the published work from NeuHauss et al.⁴⁶ and Delbaere et al.⁴⁷ Image analysis was performed to confirm the phenotype-genotype of crispants in comparison with scrambled zebrafish based on measuring different structures on the head of the 5 dpf injected larvae with Leica fluorescent microscope M165 FC and a microscope light source. The magnification of the images was 12×. These stainings and measurements were repeated twice. We evaluated the length of the Meckel’s cartilage and the ethmoid plate, the distance between the middle point of the Meckel’s cartilage to the frontal point on the basihyal cartilage and the angle, the length of each ceratohyal cartilage, and the angle between them with Fiji (ImageJ). Each measurement was normalized to the length of the middle point of both Meckel’s structures to the parachordal structure.

mesoSPIM light-sheet microscopy, sample preparation, and imaging

Zebrafish larvae were prepared and imaged with mesoscale selective plane illumination microscopy (mesoSPIM) light-sheet microscopy adapted from previously described protocols.⁴⁸ Zebrafish larvae, 5 dpf, were fixed for 40 min in 4% PFA and dehydrated to 100% MeOH. Bleaching was performed under strong light in 10% H2O2/23% H2O/66% methanol to reduce the formation of air bubbles until depigmentation (3 days). Embryos were rehydrated in steps and permeabilized in 1× PBS with 0.1% Triton X-100 (PBT). Embryos were blocked for 2–3 h in 3% natural goat serum with 2% BSA (blocking buffer; Sigma, S26-100ML) and incubated overnight at 4°C with primary Col2a1 antibody (1:250, DSHB, II-II6B3), with the addition of DAPI counterstain (20 μg/mL, Thermo Fisher Scientific, D21490). Embryos were washed for 3 × 20 min with PBT, blocked again for 3 h, and incubated with goat anti-mouse IgG (H+L) cross-adsorbed secondary antibody, DyLight 633 (1:250, 35513, Thermo Fisher Scientific) overnight at 4°C. After washing in PBT 3× for 1 h and PBS 1× for 1 h, larvae were embedded in 2% low-melting agarose and dehydrated as follows: 25% methanol/75% 1× PBS (15 min), 50% methanol/50% 1× PBS (15 min), 75% methanol/25% 1× PBS (15 min), and three times in 100% methanol, including once overnight. Clearing was performed in BABB (benzyl alcohol/benzyl benzoate 1:2) overnight. Imaging was performed on a benchtop mesoSPIM platform.⁴⁹ Fluorophores were excited with the appropriate laser lines, and a quadband emission filter (ZET405/488/561/640, AHF) was employed. Embryos were imaged at either a voxel size of 0.48 × 0.48 × 2 μm³ (X×Y×Z) for Col2a1 or 0.21 × 0.21 × 1 μm³ for DAPI using a Mitutoyo M Plan Apo 10×/0.28 (Edmund Optics; 59-877) or Mitutoyo G Plan APO 20×/t3.5 (Edmund Optics; 58-175), respectively. Data were further processed using Fiji⁵⁰ or Napari.⁵¹

DNA methylation analysis

DNA was obtained from the peripheral blood of individuals with ZFHX4-related syndrome and control subjects. Subsequently, a bisulfite-based technique was employed, and the levels of methylation were evaluated utilizing Illumina Infinium EPIC bead chip arrays (San Diego, CA, USA) following the procedures prescribed by the manufacturer. The assessment of sample quality, as well as the importation of normalized methylation and non-methylation signal intensity data, was conducted using the R minfi package (v.1.44.0).⁵²

The methylation data underwent analysis via a standard pipeline, outlined by Aref-Eshghi et al.,^53,54,55 within the R environment (v.4.2.2). To ensure that disparities noted between the ZFHX4 deletion and control cohorts were solely due to alterations in methylation rather than other potentially influencing factors, probes located on the X and Y chromosomes, probes prone to cross-reactivity with other genomic regions, and those with a detection p value > 0.01 were removed from the analysis.

Probes exhibiting beta values of 0 and the top 1% most variable probes in terms of variance within either the ZFHX4 deletion or control samples were excluded. Methylation levels, represented as beta values, underwent conversion into M-values through logit transformation. These transformed values were then employed for linear regression modeling utilizing the limma package (v.3.54.2).³⁵

The control samples were chosen from the EpiSign Knowledge Database (EKD)⁵³ via the R software package MatchIt (v.4.5.2).⁵⁶ Controls were matched by age, sex, and array type, resulting in the identification of 56 samples.

The model matrix was refined by incorporating estimated proportions of diverse blood cell types as covariates to address potential confounding influences.⁵⁷ Following this adjustment, p values were refined for statistical significance using the eBayes function within the limma package.

The parameters for probe selection underwent adjustment to enhance the discrimination between the ZFHX4 deletion and control samples. This evaluation utilized hierarchical clustering, employing Ward’s method based on Euclidean distance, and multidimensional scaling (MDS), which entailed scaling the pairwise Euclidean distances between samples for visualization purposes.

We employed the R package e1071 (v.1.7-13) to train a support vector machine (SVM) and construct a binary prediction model, following the methodology outlined in the referenced article.⁵⁸ The training process involved ZFHX4 deletion samples, matched control samples, 75% of the remaining control samples, and other rare genetic disorders from the EKD (datasets of control samples and other rare genetic disorders can be found in Pérez Baca et al.⁵⁹), while the remaining 25% of samples were reserved for testing purposes.

For each sample, the methylation variant pathogenicity (MVP) score, ranging from 0 to 1, was computed to designate the similarity between its methylation pattern and the identified methylation profile for ZFHX4 deletion samples. The conversion of SVM decision values into MVP scores was accomplished using the Platt scaling method.

CUT&RUN sequencing and data analysis

Cleavage Under Targets and Release Using Nuclease (CUT&RUN) was performed using CUTANA pA/G-MNase (EpiCypher, 15-1016) according to the manufacturer’s protocol. Briefly, cells (0.5 M cells/sample) were washed, NE buffer was added to extract the nuclear protein, and the cells were incubated with activated concanavalin A beads for 10 min at room temperature. Cells were then resuspended in antibody buffer containing 0.5 M EDTA, a 1:100 dilution of each antibody (anti-ZFHX4 antibody, HPA023837; rabbit [DA1E] mAb IgG XP isotype control #3900S; anti-CTCF, Merck 07-729; H3K27ac antibody) was added to individual cell aliquots, and the tubes were rotated at 4°C overnight. The following day, targeted chromatin digestion and release were performed with 2.5 mL CUTANA pA/G-MNase and 100 mM CaCl₂. The genomic DNA was retrieved with the CUTANA DNA Purification Kit. Sequencing libraries were prepared with the NEBNext End Prep, followed by the adaptor ligation and after size selection with Ampure beads, the indexes were added, and the fragments were amplified. After the PCR amplification, clean up with the Ampure beads predominantly elutes mostly the fragments between 150 and 500 bp. The sequencing libraries were then sequenced with Illumina Nextseq 2000 using 2 × 150 bp paired-end reads to obtain 10 million read pairs per sample.

Read processing and peak calling was performed using a dedicated nextflow CUT&RUN pipeline available through nf-core⁶⁰ (https://doi.org/10.5281/zenodo.5653535). Briefly, raw reads were trimmed, quality control (QC) filtered, and aligned to the hg38 reference genome using Bowtie2 (v.2.4.4).⁶¹ Upon Q-score filtering, duplicate removal and counts per million (CPM) normalization, peaks in target samples were called with SEACR (v.1.3)⁶² using corresponding IgG samples as controls. A consensus peak list per sample was obtained by requiring the presence of peaks in both replicates. Downstream annotation of consensus peaks was performed using the ChIPseeker R package.⁶³

The top 20 enriched Gene Ontology (GO) terms (q < 0.05) are visualized as a dot plot in Figure 7B. The Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis provides the annotation of biological pathways, which is represented in Figure 7C. The top 10 enriched pathways (q < 0.05) are visualized as a dot plot. Homer37 (v.4.11) was used to perform motif enrichment analysis, with 200 bp around the peak summit as input. The top 20 de novo and known motifs that were found from the analyses are summed up in Table S14.

Figure 7.

CUT&RUN sequencing data show ZFHX4 binding to genes involved in axonogenesis, the regulation of the nervous system development, neurogenesis, and neural migration

(A) Annotation of CUT&RUN consensus peaks reveals ZFHX4 binding to promoter regions. Left: distribution of the ZFHX4 consensus peaks relative to transcription start sites (TSSs). Half of the peaks are found between 0 and 1 kb of the TSS. Right: pie chart visualizing the genomic annotation and the percentage of the peaks that overlap a TSS, 5′ UTR, 3′ UTR, or exonic, intronic, downstream, or distal intergenic regions. This shows that 55.62% of ZFHX4 consensus peaks are found near gene promoters.

(B) ZFHX4 binds to genes involved in axonogenesis, forebrain, and nervous system development, including the regulation of the neurogenesis and synapse organization and cell and neuron differentiation. The Gene Ontology (GO) dot plot displays the top 20 enriched biological processes (BPs) ranked by gene ratio (the number of genes related to GO term/total number of significant genes) and the p-adjusted values for these terms (color, p = 1 × 10⁻⁷–1 × 10⁻⁵). The size of the dot represents the gene counts per BP.

(C) ZFHX4 plays a key role in embryonic, neuron, and axon developmental pathways. The Kyoto Encyclopedia of Genes and Genomes (KEGG) dot plot provides all enriched biological pathways ranked by gene ratio (# of genes related to GO term/total number of significant genes) and the p-adjusted values for these terms (color). The size of the dot represents the gene counts per pathway.

Human expression datasets

We examined the expression of ZFHX4 and its different isoforms in 54 non-diseased adult tissues (see Figure S1). To this end, the ZFHX4 expression levels in terms of TPM (transcripts per million) were retrieved from GTEx (https://www.gtexportal.org/home/). Furthermore, the ZFHX4 expression levels in terms of RPKM (reads per kilobase million) during neurodevelopment and adulthood were retrieved from BrainSpan (http://www.brainspan.org; Figure 4). The brain-tissue-specific bulk RNA sequencing (RNA-seq) data were downloaded from their website (“RNA-Seq Gencode v10 summarized to exons”). The brain regions were matched between development stage 2A and the later ones based on table 2 in their technical white paper.⁶⁴ Read counts of the human embryonic craniofacial tissue bulk RNA-seq by Yankee et al. were accessed through their website (http://cotneyweb.cam.uchc.edu/craniofacial_bulkrna/).⁶⁵ The ZFHX4 expression levels in human embryonic craniofacial tissue⁶⁵ are represented in terms of RPKM from neurocranial crest cells to craniofacial tissue in embryonic periods.

Figure 4.

Deletion of a ZFHX4 region is associated with a mild methylation profile

(A) The Euclidean clustering plot depicts ZFHX4 deletion samples in red, and the matched controls are shown in blue. The rows represent selected probes for the identified methylation pattern, while the columns represent training ZFHX4 deletion samples and controls. A clear separation between ZFHX4 deletion samples and controls was observed. Methylation levels are color coded: blue for 0 beta values and red for 1 beta values, with intensity indicating methylation level intensity.

(B) The MDS plot demonstrates the differentiation between ZFHX4 deletion and control samples based on selected probes. The graph shows two dimensions of multidimensional scaling plots (x axis = coordinate 1, y axis = coordinate 2) representing the pairwise distance across the ZFHX4 deletion samples with controls. ZFHX4 deletion samples are represented by red circles and controls by blue circles. The plot shows individuals with ZFHX4 deletion cluster away from the controls.

(C) Utilizing SVM classification trained with ZFHX4 deletion samples against matched controls, as well as 75% of other control samples and samples from other disorders (blue), in addition to the remaining 25% of the database samples used for testing (gray), produced significant findings. Except for the ZFHX4 deletion samples (training), all other conditions and ZFHX4 testing samples (ZFHX4_truncating) exhibited significantly low MVP scores, indicating an enhancement in the specificity of the model. The ZFHX4 testing samples, which encompass truncating variants (ZFHX4_truncating), a large duplication variant (ZFHX4_duplication), an inversion variant (ZFHX4 _inversion), and an intragenic deletion (ZFHX4_in_frame_del), were included, enabling the model to predict the class probability for ZFHX4 samples. ADCADN, cerebellar ataxia, deafness, and narcolepsy, autosomal dominant; ARTHS, Arboleda-Tham syndrome; ATRX, alpha-thalassemia/intellectual development syndrome, X-linked; AUTS18, autism, susceptibility to 18; BEFAHRS, Beck-Fahrner syndrome; BFLS, Börjeson-Forssman-Lehmann syndrome; BIS, blepharophimosis intellectual disability SMARCA2 syndrome; CdLS, Cornelia de Lange syndrome1–4; CHARGE, CHARGE syndrome; 16p11.2del, 16p11.2 deletion syndrome; CSS_c.6200, Coffin-Siris syndrome (c.6232G>A [GenBank: NM_006015.4 (ARID1A)] [p.Glu2078Lys]) (c.6254T>G [GenBank: NM_006015.4 (ARID1A)] p.Leu2085Arg) (c.6133T>C [GenBank: NM_017519.2 (ARID1B)] [p.Cys2045Arg]); CSS4_c.2650, Coffin-Siris syndrome (c.2656A>G [GenBank: NM_001128849.1 (SMARCA4)] [p.Met886Val]); CSS9, Coffin-Siris syndrome-9; Down, Down syndrome; Dup7, Williams-Beuren duplication syndrome (7q11.23 duplication syndrome); DYT28, Dystonia-28, childhood onset; EEOC, epileptic encephalopathy, childhood onset; FLHS, Floating-Harbour syndrome; GADEVS, Gabriele de Vries syndrome; GTPTS, Genitopatellar syndrome; HMA, HVDAS_C, Helsmoortel-Van der Aa syndrome (ADNP syndrome [central]); HVDAS_T, Helsmoortel-Van der Aa syndrome (ADNP syndrome [terminal]); ICF1 immunodeficiency, centromeric instability, facial anomalies syndrome 1; ICF2_3_4, immunodeficiency, centromeric instability, facial anomalies syndrome 2, 3, and 4; IDDSELD, intellectual developmental disorder with seizures and language delay; Kabuki, Kabuki syndromes 1 and 2; KDM2B, KDM2B-related syndrome; KDM4B, KDM4B-related syndrome; KDVS, Koolen de Vries syndrome; Kleefstra, Kleefstra syndrome 1; LLS, Luscan-Lumish syndrome; MKHK_ID4, Menke-Hennekam syndrome-1, 2; MLASA2, myopathy, lactic acidosis, and sideroblastic anemia-2; MRD23, intellectual developmental disorder, autosomal dominant 23; MRD51, intellectual developmental disorder, autosomal dominant 51; MRX93, intellectual developmental disorder, X-linked, XLID93; MRX97, intellectual developmental disorder, X-linked 97, XLID97; MRXSA, Armfield syndrome; MRXSCJ, syndromic X-linked intellectual disability, Claes-Jensen type; MRXSN, syndromic X-linked intellectual disability, Nascimento type; MRXSSR, syndromic X-linked intellectual disability, Snyder-Robinson type; PHMDS, Phelan-McDermid syndrome; PRC2, RENS1, Renpenning syndrome; RMNS, Rahman syndrome; RSTS, Rubinstein-Taybi syndrome-1, 2; RSTS1, Rubinstein-Taybi syndrome-1; RSTS2, Rubinstein-Taybi syndrome-2; SBBYSS, Say-Barber-Biesecker-Young-Simpson syndrome; Sotos, Sotos syndrome; TBRS, Tatton-Brown-Rahman syndrome; VCFS_comp, velocardiofacial syndrome; VCFS_core, velocardiofacial syndrome; WDSTS, Wiedemann-Steiner syndrome; WHS, Wolf-Hirschhorn syndrome; Williams, Williams-Beuren deletion syndrome (7q11.23 deletion syndrome).

The single-cell RNA-seq (scRNA-seq) neural organoid datasets⁶⁶ (23–180 days of age) were accessed as serialized R (RDS) objects accessed through PsychENCODE on synapse (syn36139275), harmonized, and integrated into a high-performance computing unit (R v.4.1.0) using the “harmony” package (v.1.2.0). Plots were created using the ggplot2 package (v.3.5.2) in R (v.4.3.2).^67,68 Lines in the plots were fitted using the ggplot2 function geom_smooth (method = "loess") on the normalized counts (Figure 5C) or RPKM (Figure 5C). The plot in Figure 5D was created using the ggplot2 function "geom_bar()", with the error bars made with stat_summary (fun.data = "mean_sdl").

Figure 5.

ZFHX4 expression increases during human neuronal differentiation and brain and craniofacial development

(A) ZFHX4 expression levels during in vitro neural differentiation. RNA was sampled at the following stages: iPSCs, NSC (neural stem cells), neural progenitor cells (NPCs), and direct differentiated neurons (Ns). An increasing expression is observed upon differentiation. The median expression value is indicated by the middle horizontal line. The whiskers indicate the minimum and maximum values. CNRQ, calibrated normalized relative quantity.

(B) Endogenous ZFHX4 (399 kDa) levels in HEK293T, iPSCs, and NPCs as detected with western blot using the ZFHX4 antibody (top) and a vinculin antibody as a loading control (bottom).

(C) ZFHX4 levels in developing human brain. ZFHX4 expression in the developing human brain (normalized RPKM data) shows higher expression during early prenatal development followed by decreased expression upon brain maturation. Data were obtained from BrainSpan: http://www.brainspan.org.

(D) Human embryonic craniofacial tissue bulk RNA-seq is visualized in a plot, showing an increase in ZFHX4 in craniofacial samples from five Carnegie stages (CS13, CS14, CS15, CS17, and CS22).⁷⁴

Results

Individuals harboring a ZFHX4 deletion or PTV present with a clinically distinctive NDD

Demographic features

Data on 63 individuals (57 probands and 6 family members) with an aberration affecting ZFHX4 (i.e., inversions, CNVs, and single-nucleotide variants [SNVs]) are summarized in Table S3 and more comprehensively documented in Tables S1 and S2. There are 38 males and 25 females (Table S4). Ages ranged between 5 months and 54 years. The median age at the last examination was 8 years.

Genotype

Twenty-one probands present with a structural variant affecting ZFHX4, including one chromosomal inversion and 20 CNVs. The inversion between genomic positions chr8:69,893,659 and chr8:76,806,725 (hg38) in proband 1 results in a breakpoint at chromosome 8q21.11, disrupting the ZFHX4 intron between exons 4 and 5 (Figure 2A). The CNVs are between 1.80 kb and 12.92 Mb in size and comprise 13 multigenic deletions, eight intragenic ZFHX4 deletions (less than 90 kb in size), and one intragenic duplication that potentially causes a frameshift. The smallest region of overlap between the multigenic deletions is 361.9 kb (chr8:76,444,811–76,806,725 [hg38]) containing ZFHX4 and PEX2 (Figure 2A). Detailed information can be found in Table S1.

Figure 2.

Overview of ZFHX4 aberrations in 57 probands and 6 affected family members

(A) (Micro)deletions identified in probands 1–21 (#) are represented by the red bars and the duplication in proband 15 by the yellow bar, all affecting ZFHX4 (light blue box). RefSeq coding genes are indicated in blue.⁶⁹ Genomic positions are according to hg38; ZFHX4 is located on the reverse strand. The lightning bolts indicate the breakpoint positions of the inversion.

(B) Nonsense, frameshift, and splice variants in ZFHX4 are identified in probands 22–57 and visualized with IBS 2.0.⁷⁰ Protein and gene structure of the canonical transcript (GenBank: NM_024721.5) are shown on the top and bottom, respectively. The majority of PTVs are located in exons 2 and 10. The variants in probands 22–34, 36, 37, and 39–57 (Table S2) are shown from the N to the C terminus of the protein: p.Glu84^∗, p.Ser319Profs^∗3, p.Pro331Leufs^∗36, p.Ser341Ilefs^∗30, p.Ser341Ilefs^∗30, p.Leu408Argfs^∗7, p.Leu411Alafs^∗3, p.Leu411Alafs^∗3, p.Gly630Aspfs^∗16, p.Tyr655^∗, p.Gln681^∗, p.Gln681^∗, p.Trp767^∗, p.Leu903Tyrfs^∗18, p.Cys962Alafs^∗27, p.Arg1078Profs^∗19, p.Gly1114Glufs^∗14, p.?, p.Arg1384Glyfs^∗57, p.His1443Thrfs^∗8, p.Gln1712Valfs^∗64, p.Phe1739Serfs^∗37, p.Gln1797^∗, p.Gln1836Argfs^∗4, p.Lys1886^∗, p.Lys1886Glufs^∗33, p.Arg1899^∗, p.Cys1951Trpfs^∗2, p.Cys1951Trpfs^∗2, p.Gln2258Serfs^∗69, p.Thr2349Glnfs^∗43, p.Thr2349Glnfs^∗43, p.Gly2549Thrfs^∗35, p.Gly2549Thrfs^∗35, p.(Gly2549Thrfs^∗35, p.Gln2664^∗, p.His2874Serfs^∗30, p.His2874Serfs^∗30, p.His2874Serfs^∗30, and p.Trp2176^∗/p.Asn2188Lysfs^∗12. As the effect on the protein is not known for the splice variants in probands 35 and 38, only the c. notation is mentioned. Top: LoF variants are present in the ZFHX4 cohort and located in different domains of ZFHX4. Blue dots represent the frameshift variants, the yellow dots represent the nonsense variant, and the brown-gray dot represents the variants of proband 57, which contains a nonsense p.Trp2176^∗ and a frameshift p.Asn2188^∗fs mutation. The dot has been placed on the position of the nonsense variant p.Trp2176^∗. Bottom: coding exons are indicated as light blue boxes, and the UTRs are represented with white boxes.

All c. notations can be found in Table S2.

Heterozygous frameshift, nonsense, and canonical splice site variants were identified in 29, 10, and 2 probands, respectively (Table S2). These PTVs were found in exon 2 (probands 22–32), exon 3 (probands 33, 34, and 36), exon 5 (proband 37), and penultimate exon 10 (probands 39–57) and are all predicted to result in nonsense-mediated decay (i.e., >55 bp away from last exon-exon boundary) (Figure 2B). All nonsense variants for which a CADD (combined annotation-dependent depletion) score was available have a score of 35 or higher, indicating that they are within the 0.1% most deleterious possible substitutions in the human genome. Probands 35 and 38 harbor a ZFHX4 splice donor variant (ZFHX4; c.3093+1G>T [GenBank: NM_024721.5]) and a splice acceptor variant (ZFHX4; c.3965−1G>A [GenBank: NM_024721.5]), respectively. According to SpliceAI,^71,72 the c.3093+1G>T variant results in the loss of a canonical donor sequence (Figure S2), possibly leading to skipping of exon 3, while the c.3965−1G>A variant results in the loss of the canonical acceptor sequence at exon 10 and the gain of an acceptor, causing a frameshift in exon 10 (Figure S2).

Additionally, we identified one individual (proband 57) who is compound heterozygous for a nonsense (ZFHX4: c.6528G>A [GenBank: NM_024721.5] [p.Trp2176^∗]) and a frameshift variant (ZFHX4: c.6564_6567del [GenBank: NM_024721.5] [p.Asn2188Lysfs^∗12]).

Although we collected their clinical characteristics, the phenotypic consequences of the variants in probands 1 and 57 might be different, and we excluded both individuals from the clinical characteristics analysis to determine the main phenotype traits associated with alterations in ZFHX4.

Fifteen probands (4, 5, 8, 10, 15, 25, 31, 36, 48, 50, 51, 52, 55, 56, and 57) inherited the variant from a similarly affected parent, although the mothers of proband 8 (chr8:74,953,019–77,754,408 x1 [hg38]), proband 25 (ZFHX4: c.1020dup [GenBank: NM_024721.5] [p.Ser341Ilefs^∗30), and proband 50 (ZFHX4: c.7045del [GenBank: NM_024721.5] [p.Thr2349Glnfs^∗43) had learning difficulties and/or reading disabilities without a formal diagnosis. Proband 31 and his mother (ZFHX4: c.2041C>T [GenBank: NM_024721.5] [p.Gln681^∗), who were 7 and 42 years old, respectively, had normal intelligence. Also, proband 48, 5 years old, and his mother (ZFHX4: c.5850_5851delAT [GenBank: NM_024721.5] [p.Cys1951Trpfs^∗2]), 38 years old at the last examination, had mild intellectual disability. The mothers of probands 8, 25, 31, and 48 showed facial characteristics similar to those of their affected children. The facial features of proband 50 were not assessable, but those of his mother resembled those of the overall cohort (Figure 1A). Both parents of compound heterozygote individual 57 each carried a truncating variant in ZFHX4, were reportedly healthy, and lived independently. In the absence of formal neuropsychological testing or facial photographs, we cannot rule out subtle features that remain unnoticed.

The inheritance patterns of the ZFHX4 variant in probands 17, 22, 23, mother of proband 25, mother of proband 31, mother of proband 50, proband 53, and father of proband 56 could not be determined. In the remaining 39 probands, the aberration occurred de novo (i.e., 16 de novo CNVs and 24 de novo SNVs) (Tables S1 and S2).

Four variants, identified in probands 25, 27, 28 (same variant as proband 27), and 47, were found in the public variant database gnomAD (release v.4.1.0). Twenty missense variants present in gnomAD affect the same codon as the truncating variant in probands 33, 34, 36, 37, 41–44, 47, and 49–57 (Table S5). While ZFHX4 LoF variants appear to be under high constraint (as indicated by a pLI score of 1 and a loss-of-function observed/expected upper bound fraction [LOEUF] score of 0.24), missense variants are not (as indicated by a Z score of 1.38), thereby suggesting that these missense variants do not result in LoF. Nevertheless, the presence of pathogenic variants in gnomAD does indicate that phenotypic variability may be wide and manifestations of the phenotype may be mild or absent.

Facial features

Dysmorphic facial features were reported in all individuals (n = 63/63; 100%). Frequently recurring facial characteristics were a broad lower half of the face (64%), a prominent forehead (67%), laterally sparse eyebrows (56%), eyelid ptosis (62%), low-set ears (59%), prominent ear helices (59%), a wide nasal bridge (66%), and underdeveloped nasal alae (57%). Facial characteristics observed in less than half of the individuals were biparietal narrowing (43%), hypertelorism (30%), upslanted palpebral fissures (41%), short palpebral fissures (34%), periorbital fullness (49%), epicanthal folds (41%), posteriorly rotated ears (30%), a narrow nasal tip (38%), flared nostrils (48%), abnormal upper lip vermillion (43%), a thin upper lip (34%), and micrognathia (39%). Less frequently observed characteristics were microphthalmia (7%), enlarged nares (26%), short philtrum (20%), smooth philtrum (26%), an everted upper lip (16%), downturned corners of the mouth (16%), and cleft lip and/or palate (5%) (Figure 1A; Table S3). A computationally based facial analysis using DeepGestalt facial analysis on 20 individuals with a ZFHX4 deletion and 18 individuals with a LoF SNV in ZFHX4 (Figure 1B) vs. age- and sex-matched controls indicates a highly specific and sensitive gestalt with average AUCs of 0.988 and 0.986 for individuals with a deletion or SNV, respectively (p < 0.005). A comparison of the facial features of both affected groups (i.e., ZFHX4 deletion vs. ZFHX4 LoF variants) showed no significant difference (AUC = 0.656, p = 0.131).

Clinical findings

Global developmental delay and/or intellectual disability were present in most individuals (81%, n = 42/52). For those with a documented IQ score (n = 33), most had mild intellectual disability (n = 16), two had borderline intellectual disability, six had moderate intellectual disability, and one had severe intellectual disability. Interestingly, a statistical difference was found in the levels of intellectual disability between individuals with a CNV affecting ZFHX4 and those with ZFHX4 PTVs (100% vs. 69%). In 8/33 individuals (24%), intellectual functioning was reportedly normal; however, none were in the CNV group, but 38% were in the PTV group. In both categories, the most frequent degree of intellectual disability was mild or borderline (67% and 48%, respectively). In Table S6, we included other features that were observed in our cohort. Muscular hypotonia was present in 21 individuals (n = 21/63). Two-thirds had behavioral differences (n = 35/54), with stereotypies (n = 7) and autism (n = 10) being the most frequently reported. Morphological abnormalities of the central nervous system were identified in 38% of the individuals, mostly ventriculomegaly (n = 5), aplasia/hypoplasia of the corpus callosum (n = 3), and widened subarachnoid space (n = 3). Ophthalmological abnormalities included strabismus (48%, n = 25/52), refraction abnormalities (44%, n = 20/45), either myopia (n = 8) or hypermetropia (n = 5), nystagmus (n = 1), and amblyopia (n = 2). Notably, 9 individuals had an abnormal anterior eye segment morphology, with astigmatism (n = 4), sclerocornea (n = 2), microcornea (n = 2), anterior segment dysgenesis (n = 2), corneal opacity (n = 1), and/or persistent pupillary membrane (n = 1). Musculoskeletal manifestations included digital abnormalities in 43% (n = 23/54), with clinodactyly (n = 8) and camptodactyly (n = 6) being the most frequently reported. Growth abnormalities were described in 39% (n = 20/31) of the probands. Prenatally, three individuals presented with intrauterine growth restriction and two with fetal macrosomia. Five individuals showed postnatal growth retardation, and short stature was documented in 26% (n = 15/57). Genital abnormalities were found in 26% (n = 12/46), including cryptorchidism (n = 6), hypogenitalism (n = 4), and abnormality of the scrotum (n = 3). Congenital heart defects were reported in 19% (n = 9/48) of the probands, mainly pulmonic stenosis (n = 1), ventricular septal defect (n = 1), atrial septal defect (n = 2), aortic arch hypoplasia (n = 1), and right aortic arch (n = 1). About one-third of the children (n = 16/51) experienced feeding difficulties in infancy. Recurrent infections were noted in 17% (n = 8/47).

Genotype-phenotype correlation

We compared the frequencies of 27 facial features and 15 clinical features between individuals with CNVs (group 1, n = 21) and individuals with PTVs (group 2, n = 41) in ZFHX4. Underdeveloped alae nasi (86% in CNVs and 46% in PTVs), abnormality of refraction (77% in CNVs and 31% in PTVs), and morphological abnormalities of the central nervous system (70% in CNVs and 26% in PTVs) were more frequent in group 1 (p = 0.0046, p = 0.0081, and p = 0.0023, respectively), while micrognathia was frequent in group 2 (p = 0.005). Intellectual disability is generally more severe in individuals with a multigenic deletion (p < 0.05, Table S3).

Penetrance of the clinical features did not differ between both sexes (p < 0.05, Table S4).

A 6.9 Mb de novo inversion involving chromosome bands 8q13–8q21.13 results in changes in 3D chromatin structure and disruption of ZFHX4

Individual 1 presented with a compatible but more severe phenotype (severe intellectual disability, significant dysmorphic facial features, abnormal muscle tone, and abnormal heart morphology). We delineated the underlying 6.9 Mb de novo inversion spanning chromosome bands 8q13–8q21.13 at bp resolution via genome sequencing (GS). This revealed an intergenic breakpoint on the centromeric end of the inversion and a breakpoint within intron 4 of ZFHX4 on the telomeric right end (Figure 3). We next employed Hi-C to evaluate topologically associating domain (TAD) shuffling, revealing the typical bow tie configuration when mapped onto the reference genome (Figure 3). On the centromeric end, the inversion breakpoint resides within a TAD boundary (chr8:69,875,002–69,900,000 [hg38]), thereby possibly affecting two small (sub)TADs containing the SLCO5A1 (MIM: 613543) and PRDM14 (MIM: 611781) genes, respectively. On the telomeric end of the inversion, the affected TAD contains ZFHX4 and PEX2 (MIM: 170993) (Figure 2A).

Figure 3.

Hi-C on proband 1 fibroblasts reveals disrupted chromatin structure at inversion breakpoints

Hi-C contact frequency matrix for control and proband 1 fibroblasts (inversion indicated by red arrows). Close-up of left and right breakpoint regions, including observed/expected (O/E) contact frequencies for control and proband 1.

Individuals with a deletion encompassing a “critical region” seem to show a mild methylation profile

Since we previously identified a methylation pattern associated with LoF of ZFHX3,⁵⁹ a paralog of ZFHX4, we also evaluated this for ZFHX4 LoF. To this end, we performed an unbiased methylome analysis on blood-derived DNA samples of 26 individuals, including nine individuals with a CNV of ZFHX4 (probands 4 and 5 and their mothers, proband 10 and her father, and probands 12–14), eleven individuals with a ZFHX4 PTV (proband 23 and his mother, probands 24–25, 29, 30, 33, 35, and 38, and proband 48 and his mother), proband 15 with an intragenic ZFHX4 duplication, proband 21 with an intragenic in-frame ZFHX4 deletion, and proband 1 with an inversion affecting ZFHX4, as well as three individuals from the cohort published in 2011² (all listed in Table S7). No specific methylation difference was observed when comparing all ZFHX4 LoF variants or ZFHX4 truncating variants with matched controls. However, DNA methylation analysis identified 227 CpG probes displaying notable methylation disparities between individuals with deletions that included a minimal region (of 53 kb; deletion observed in proband 14) containing the ZFHX4 promoter and matched controls, as evidenced by hierarchical clustering and MDS analyses (Figure 4A).

The SVM classifier underwent training utilizing signature probes to precisely categorize individuals with ZFHX4-related disorder, as observed in Figure 4B. It comprised 75% of control samples and other NDDs (blue dots). The remaining 25% of control samples and other disorders were allocated for testing (gray dots). The ZFHX4 samples with truncating variants (ZFHX4_truncating), the intragenic ZFHX4 duplication (ZFHX4_duplication), the inversion (ZFHX4_inversion), and the in-frame deletion (ZFHX4_in_frame_del) exhibited low MVP scores, which differed from the methylation pattern observed in the ZFHX4 deletion samples (Figure 4B). Overall, we observed a distinct methylation profile for individuals with a deletion overlapping the deletion of proband 14 but not in individuals with intragenic ZFHX4 defects or PTVs.

ZFHX4 expression increases during human neuronal differentiation and brain and craniofacial development

GTEx v.8 (https://www.gtexportal.org/home/) indicates abundant, ubiquitous ZFHX4 expression throughout the adult human body (Figure S1) except for several brain regions. However, expression profiles across seven organs (including the brain and heart) throughout 23 developmental stages show the highest ZFHX4 expression during the embryonic and fetal periods, especially in the human brain and cerebellum⁷³ (Figure S1). In line with these observations, we confirmed increased expression of ZFHX4 in NPCs and early developing neurons upon in vitro neural differentiation starting from iPSCs (Figure 5A). This increased expression was also confirmed at the protein level for NPCs in comparison to iPSCs (Figure 5B).

Furthermore, reanalysis of transcriptomic data within the BrainSpan⁶⁴ atlas of the developing human brain also confirmed the highest ZFHX4 expression in the brain between 8 and 16 weeks post-conception, more specifically in the mediodorsal nucleus of the thalamus, the striatum, and the amygdaloid complex (left, Figure 5C). Through PsychENCODE, we retrieved scRNA-seq datasets from developing neural organoids.⁶⁶ In the early differentiation stages between days 23 and 45, the highest ZFHX4 expression is observed in the subcortical neurons (left, Figure 5C), while in day 120–150 neural organoids, this shifts to the outer radial glia and intermediate progenitors (right, Figure 5C). In addition, a clear increase in ZFHX4 expression is observed in later embryonic stages of the developing facial bones⁶⁵ (Figure 5D), in line with observed facial skeletal changes in individuals with ZFHX4 deficiency.

ZFHX4 interacts with proteins involved in neural tube morphology and cellular organization in NPCs

Since NPCs showed increased ZFHX4 mRNA expression and increased ZFHX4 protein levels, we performed IP-MS in NPCs to discover potential interaction factors of ZFHX4. Forty-seven significantly enriched proteins were identified on NPCs compared to an IgG isotype control (t test FDR < 0.05) (Figures 6A and 6B; Tables S8 and S9). Biological process and pathway enrichment with EnrichR (https://maayanlab.cloud/enrichr-kg)^37,38 revealed that ZFHX4-interacting proteins play a role in biological processes associated with histone modifications, development, and cytosolic transport (Figures 6C and S3). Furthermore, ZFHX4 interacts significantly with proteins encoded by genes associated with NDDs (see Table S12 for a full list of associated disorders and corresponding MIMs): ZMIZ1 (MIM: 607159), ZEB1 (MIM: 189909), WDR81 (MIM: 614218), KCNQ2 (MIM: 602235), KMT2D (MIM: 602113), NUP107 (MIM: 607617), IMPA2⁷⁵ (MIM: 605922), ZMYND8⁷⁶ (MIM: 615713), NACC1 (MIM: 610672), and WASHC4 (MIM: 615748). Additionally, we also performed IP-MS in differentiated neurons, revealing 36 interacting factors compared to the IgG isotype control (t test FDR < 0.05), of which 12 overlap with those identified in NPCs (Tables S10 and S11). These 12 enriched proteins mostly play an important role in transcription regulation, signal transduction, and protein trafficking. Three of them are associated with NDDs: WDR81, KCNQ2, and NACC1. The neuron-enriched interacting factors (IFs) are mainly involved in signal transduction and transport (Tables S10 and S11; Figure S4). IFITM1 (MIM: 604456) and IFITM3 (MIM: 605579), mainly involved in viral replication and stress, were removed for this functional enrichment analysis. Five of these proteins are encoded by genes associated with neurological syndromes (Table S12): SLC1A3 (MIM: 600111), huntingtin interacting protein-1 (HIP1; MIM: 601767),^77,78,79 KLC2 (MIM: 611729), SLC9A6 (MIM: 300231), and WBP2 (MIM: 606962).

Figure 6.

ZFHX4 interacts with 46 enriched proteins involved in axonogenesis, cell differentiation, and neuron differentiation

(A) The enriched ZFHX4-interacting proteins in NPCs shown in a volcano plot. For each detected protein, the log2 fold change (FC) between the ZFHX4 IP-enriched samples and the IgG controls is shown on the x axis, while the −log10 of the p value is shown on the y axis. The horizontal dashed line indicates the −log10 value of a p value equal to 0.05. Proteins with an FDR < 0.05, a log2 FC > 2, and a log2 FC < −2 are, respectively, highlighted in blue and red.

(B) Hierarchical clustering of potentially interesting proteins. The proteins shown to be differentially abundant between groups are visualized in a heatmap after non-supervised hierarchical clustering of Z scored log2 PG.MaxLFQ intensities. The heatmap color scale from blue to red represents the level of abundance. The interacting factors that are absent in the IgG control, significant but not absent in the IgG control, and significantly enriched in the protein and absent in the IgG control are listed in Tables S8 and S9.

(C) ZFHX4-interacting factors associated to their biological processes with EnrichR.

ZFHX4 binds to the promoter of genes involved in axonogenesis, regulation of the nervous system development, neurogenesis, and neural migration

To assess the downstream target genes of the transcription factor ZFHX4, we performed CUT&RUN for ZFHX4 on two NPC replicates, retrieving around 1,640 consensus peaks mainly consisting of promoter regions (55.62%) (Figure 7A; Table S13). Furthermore, GO analysis of these peaks showed that ZFHX4 binds to the promoter of genes involved in (the regulation of) axonogenesis, forebrain development, cell differentiation nervous system development, and neuron differentiation (Figure 7B). KEGG pathway analysis indicated that the proteins encoded by these genes are involved in axon guidance and the Wnt, Rap1, and other signaling networks regulating the pluripotency of stem cells. All pathways are strongly involved in cell organization and embryonic and nervous system development (Figure 7C).^{80,81,82,83,84,85,86} Motif enrichment analysis of the consensus peaks revealed an enrichment of homeobox, transcription factors, SOX-related genes, and zinc finger motifs (Table S14).

zfhx4 F0 zebrafish crispant larvae show craniofacial and behavioral phenotypes

The zebrafish zfhx4 gene (ENSDARG00000075542) resides on chromosome 24, contains 10 exons, and has 71.37% sequence identity with the human ZFHX4 (Figure 8A). In 2021, Kroll et al. showed that F0 mosaic crispant zebrafish recapitulate the phenotype of germline knockouts.⁴² We generated zfhx4 crispants by injection of a mixture of three crRNAs, obtaining a zfhx4 insertion or deletion (indel) efficiency of approximately 79% (see Table S16).

Figure 8.

Zebrafish zfhx4 crispants showed both craniofacial and behavioral changes

(A) Gene structure of the zebrafish zfhx4 and ZFHX4. Top: the coding exons of ZFHX4 are shown as light gray boxes and the CDS (coding sequence) as light blue boxes, and untranslated regions are in red (5′ UTR) and green (3′ UTR). Bottom: the coding exons of zfhx4 are shown as light gray boxes and the CDS as light blue boxes, and untranslated regions are in red (5′ UTR) and green (3′ UTR).

(B) Representative craniofacial images for zfhx4 and scrambled zebrafish crispants are shown. Left: microscopic ventral view scrambled (top) and zfhx4 crispants (bottom) of Alcian blue-stained craniofacial structures at 5 dpf reveal smaller ethmoid plates and shorter Meckel’s cartilage in the F0 zfhx4 crispant larvae. Right: adapted image⁴⁶ from the representation of the cartilage structures of the zebrafish neurocranium (dark blue) at 5 dpf by lines. The red lines and circle show the measurements taken and normalized with the length of the head (purple line).

(C) Left: the surface area of the ethmoid plate was measured and normalized to the head length of the larvae. Right: the length of the individual Meckel’s cartilage structures was measured and normalized to the head length of the larvae. The ethmoid plate and the Meckel’s cartilage showed significant differences between crispant larvae compared to scrambled larvae (^∗∗∗p ≤ 0.001 and ^∗∗∗∗p ≤ 0.0001).

(D) zfhx4 crispants show a consistently lower movement frequency compared to scrambled control larvae during the complete behavior experiment, but the trends during light and dark cycles remain similar. Dark periods are indicated with gray boxes. Data are represented as mean ± SEM (n = 60/genotype).

To determine craniofacial cartilage defects, we stained neural crest-derived chondrocytes in 5 dpf zfhx4 crispant larvae and scrambled crRNA control larvae with Alcian blue (Figure 8B). zfhx4 F0 crispant larvae had a significantly smaller ethmoid plate and shorter Meckel’s cartilage in comparison with scrambled larvae (^∗∗∗p ≤ 0.001 and ^∗∗∗∗p ≤ 0.0001, Figures 8C and S5), thereby confirming that ZFHX4 LoF results in craniofacial anomalies.

Next, locomotor behavior was assessed in response to light/dark transitions using the DanioVision observation chamber. Analysis of the swimming pattern of scrambled zebrafish showed a higher movement frequency during the dark cycles. Zfhx4 crispant zebrafish showed a movement pattern comparable to scrambled zebrafish during the light and dark cycles but with a consistently lower movement frequency (Figures 8D, S6, and S7, ^∗p ≤ 0.05 and ^∗∗∗∗p ≤ 0.0001). This could indicate potential neurological involvement in the zebrafish model, consistent with our clinical data and previously reported mouse models, which exhibit subtle medullary defects leading to abnormal respiratory rhythmogenesis.¹⁷ To investigate the structural abnormalities further, we employed tissue clearing and in toto light-sheet microscopy using COL2A1 staining to visualize the cartilage. This approach validated specific abnormalities in craniofacial cartilage, particularly in the ethmoid plate and Meckel’s cartilage (Figure S8). To assess potential neurological features underlying the observed reductions in locomotor behavior and movement frequency, we next performed whole-mount mesoSPIM imaging on DAPI-stained and cleared zebrafish larvae to investigate structural abnormalities (Figure S8). No overt neural defects were identified except for notable effects localized to the rhombencephalon (hindbrain), specifically in the zebrafish medulla oblongata, which serves as the functional counterpart to the medulla in mammals.

Discussion

In this study, we delineated a clinically distinctive neurodevelopmental phenotype caused by pathogenic variants in ZFHX4. Most individuals presented with mild to moderate intellectual disability and behavioral problems, often associated with anatomical brain anomalies. All assessed individuals showed a recurrent facial gestalt. Short stature, digital anomalies, genital defects, cleft lip/palate, ophthalmological anomalies, hypotonia, and feeding difficulties were present in 25%–50% of the affected individuals, while congenital heart defects were occasionally observed. The high phenotypic similarity, including the facial characteristics, between individuals with CNVs, including ZFHX4, and individuals with intragenic ZFHX4 variants indicates that ZFHX4 haploinsufficiency is the mechanism underlying the common phenotype. However, individuals with CNVs appear to have an increased risk for intellectual disability, behavioral abnormalities, anatomical brain anomalies, and refraction problems, while the prevalence of micrognathia was lower compared to individuals with PTVs in ZFHX4. Furthermore, the methylation profile identified in individuals with 8q21.11 (micro)deletions encompassing ZFHX4 could not be replicated in individuals with PTVs. Although other genes contained in the 8q21.11 (micro)deletions (Table S17) might contribute to or modify the methylation pattern, the smallest region of overlap among all deletion samples tested for a methylome signature is the size of the deletion in individual 14 (53 kb). This deletion partially deletes ZFHX4 and also removes a non-coding region containing the ZFHX4 promoter. In the other tested individuals with a ZFHX4 PTV, an intragenic duplication, an intragenic deletion, or an inversion, the aberration does not result in the removal of this critical region. While ZFHX4 haploinsufficiency is the main contributor to the observed phenotype in all individuals, this suggests that the deletion of this critical region might be the main contributor to the methylation pattern of persons with deletions affecting ZFHX4. To confirm a possible role for this critical region as one of the drivers for the methylation profile, analysis of additional individuals is necessary.

Of note, 12 variants (i.e., six CNVs and six SNVs) were inherited from healthy parents, of whom three also had learning difficulties in childhood, although without a formal diagnosis. Although ZFHX4 is very intolerant for LoF variants (pLI = 1; LOEF = 0.24), 172 LoF variants (excluding low-coverage ones) are present in gnomAD (v.4.1.0), including three variants present in our cohort, albeit all with a very low frequency in the general population. As we observed variable phenotype feature expressivity in individuals with ZFHX4 CNVs and PTVs, analysis of ZFHX4 should not be limited to the identification of de novo aberrations. Moreover, careful recording of the phenotypic data of the parents remains important. Additionally, modifying genetic and/or environmental factors might also influence the observed phenotypic variability.

Individual 1 harbors a de novo inversion disrupting ZFHX4 and presented with a typical but more severe phenotype. Using Hi-C, we observed that the inversion affects two small (sub)TADs on the centromeric end, containing the SLCO5A1 and PRMD14 genes. At the telomeric end, the affected TAD contains ZFHX4 and PEX2 (Figure 2A). SLCO5A1 and PEX2 have been previously linked to disease. PEX2 is associated with peroxisome biogenesis disorders 5A and 5B, both autosomal recessive multiple congenital anomaly disorders (MIM: 614866 and 614867). SLCO5A1, on the other hand, has been linked to the autosomal dominant mesomelia-synostoses syndrome (MSS; MIM: 600383).^87,88 MSS is characterized by mesomelic shortening of the limb, acral synostoses, and multiple congenital malformations. In 2010, Isidor et al. reported 8q13 deletions associated with MSS.⁸⁹ Although these deletions varied in size, they always encompassed SULF1 and SLCO5A1. Furthermore, Roshandel et al. generated a small interfering RNA (siRNA)-mediated knockdown of Oatp30B (i.e., the Drosophila ortholog of SLCO5A1) and observed a startling and seizure-like phenotype.⁹⁰ Although SLCO5A1 is not directly affected in individual 1, the inversion might exert an effect on the expression of SLCO5A1, modifying the phenotype due to ZFHX4 disruption (including vertebral anomalies and progressive contractures). However, further functional studies are necessary to unravel this.

Although previous reports suggested ZFHX4 as a candidate gene for non-syndromic orofacial clefting,¹¹ we observed orofacial clefting in about 25% of the individuals with ZFHX4 LoF variants. We investigated the craniofacial cartilage development in mosaic knockout zfhx4 zebrafish and observed significant differences between Meckel’s cartilage and the ethmoid plate structures that correspond to the human jaw and palate, respectively. A homozygous Zfhx4 knockout mice model further supports this hypothesis, presenting cleft palate at 100% penetrance.¹⁶ In addition to the craniofacial differences in our zebrafish model, we also observed behavioral changes. The decreased movement frequency could be attributed to (a combination of) neurological, skeletal, and/or muscle defects in the zfhx4 crispant zebrafish larvae, given the role of ZFHX4 in different biological processes of these tissues.^13,16,18 Previous studies on Zfhx4 knockout mice revealed medullary abnormalities,¹⁷ and consistent with these findings, we identified structural abnormalities in the zebrafish hindbrain, the functional equivalent of the mammalian medulla. These abnormalities appear to be localized specifically to the hindbrain medulla oblongata, a region critical for autonomic functions and motor coordination.⁹¹ Such structural disruptions may underlie the behavioral phenotypes observed in our study. Interestingly, the International Mouse Phenotyping Consortium (IMPC) applied a panel of phenotyping screens on Zfhx4 homozygous and heterozygous knockout mice. Similar to what was observed by Nakamura et al., the homozygous Zfhx4-deficient mice showed full penetrant preweaning lethality. However, the heterozygous Zfhx4-deficient mice were viable (11 females and 13 males) but did not present with a cleft palate and showed decreased exploration in a new environment. Taken together, our in vivo zebrafish results, as well as the previous studies in mice, clearly indicate a relationship between ZFHX4, clefting, brain morphology changes, and behavioral abnormalities.

The complex structure of ZFHX4, containing 4 homeodomains and 22 zinc fingers, suggests dynamic functions in diverse biological processes. Our IP-MS screen in NPCs and neurons identified ZFHX4 protein interactors involved in the regulation of histone modifications, development, signal transduction, transport, gene expression, and cell division. Most of the ZFHX4 IFs in NPCs are associated with histone modifications and development. YEATS2 (appears in the neurons as well), ZZZ3, KMT2D, and ZMYND8 play a crucial role in chromatin remodeling and histone modifications, such as acetylation.^{92,93,94,95,96,97,98} Although CHD4, a crucial member of the NuRD complex, was previously identified as a ZFHX4-interacting protein in glioblastoma cells, it was not identified as an interaction partner in our NPC or neuron screen. However, in the NPCs, we identified the CHD4-interacting multivalent histone reader ZMYND8⁹² as an interacting factor of ZFHX4. Furthermore, in 2022, Dias et al. reported disruption of ZMYND8 as a cause for syndromic intellectual disability.⁷⁶ Moreover, several of the other ZFHX4-interacting proteins have also been associated with NDDs: YEATS2 (identified in both NPCs and neurons) plays an important role in histone acetylation and has been involved with epilepsy (MIM: 615127; epilepsy, myoclonic, familial adult).^93,94 ZEB1 has been associated with corneal dystrophy.^99,100 Another interacting factor (in both NPCs and neurons) is KCNQ2, encoding a potassium voltage-gated channel, which plays a key role in the regulation of neuronal excitability.^101,102 Mutations in KCNQ2 have been associated with self-limited familial neonatal epilepsy (SelFNE) and other forms of epilepsy (early-onset developmental and epileptic encephalopathy).^103,104 Additionally, mutations in KMT2D, encoding a histone methyltransferase, give rise to Kabuki syndrome,^105,106 associated with neural crest cell differentiation and facial morphology.^107,108 IMPA2 plays an important role in the phosphatidylinositol signaling pathway and has been associated with susceptibility to bipolar disorder.^75,109 Furthermore, GAMT is crucial for creatine biosynthesis and is associated with cerebral creatine deficiency syndrome 2 (MIM: 612736).^110,111,112 Other interacting factors, such as NPTX2 and ZNF592, are involved in synaptic plasticity and neuronal development.^{113,114,115,116}

In neurons, most of the IFs are involved in signal transduction and transport. WDR81,^117,118 SLCA3,^119,120 NOS1AP,¹²¹ HIP1,^77,78,79 KCNQ2,^101,102 KLC2,^122,123 and SLC9A6^124,125 are involved in protein trafficking and signal transduction, including in the brain, where their dysfunction can lead to neurological syndromes. WDR81 is found in the brain, plays an important role in endolysosomal trafficking, and is associated with cerebellar ataxia and intellectual disability.¹¹⁸ KLC2, involved in axoplasmatic transport and highly expressed in neurons,¹²³ and HIP1, important for endocytosis, have both been associated with Huntington disease. The overexpression of KLC2 has been associated with the spastic paraplegia, optic atrophy, and neuropathy (SPOAN) syndrome (MIM: 609541), a neurodegenerative disorder recognized by spastic paraplegia, optic atrophy, and neuropathy.¹²³ Both overexpression and knockdown caused a curly-tail phenotype in the zebrafish, indicating its association with a neuromuscular disorder.¹²³ Both SLCA3 and SLC9A6 are solute carriers, and SLC9A6 variants have been involved with Christianson syndrome (MIM: 309620), a developmental disorder with a wide range of neurological, somatic, and behavioral symptoms,¹²⁵ and an in-frame deletion of this carrier was associated with mental retardation with tau deposition.¹²⁴ Lastly, NOS1AP, which regulates neuronal nitric oxide synthase, has been associated with schizophrenia.¹²¹ Although we identified several overlapping interacting factors in NPCs and neurons, to explicitly confirm them as interaction partners of ZFHX4, coIP or other orthogonal validation techniques are still required.

Furthermore, in NPCs, ZFHX4 mainly binds to the promoter regions of genes that are significantly enriched for homeobox, SOX-related genes, and zinc finger motifs according to de novo and known HOMER motif analyses (Table S14). Via Simple Enrichment Analysis (SEA) motif enrichment, the same consensus peaks seem to show ERF (ETS domain-containing transcription factor), KLF (Krüppel-like factor), and SP (specificity protein) motif enrichment (Table S14). These transcription factors are crucial for transcriptional regulation and cell proliferation.^126,127 Mutations in ERF genes have been associated with craniosynostosis with facial dysmorphism,^63,128 while KLF and SP genes are involved in embryonic development and tissue differentiation.¹²⁶ In this study, we focused solely on NPCs and early differentiated neurons. We recognize the limitations of not investigating other potentially relevant cell types and emphasize the importance of addressing this in future research.

In conclusion, we have delineated a ZFHX4-associated NDD and identified ZFHX4 haploinsufficiency as the major driver of the clinical manifestations observed in 8q21.11 microdeletion syndrome (MIM: 614230). ZFHX4 should be included in NDD/intellectual disability panels to enable the diagnosis of additional affected individuals. Our initial findings indicate an important role for ZFHX4 in craniofacial and brain development through interaction with proteins involved in chromatin organization and signal transduction.

Data and code availability

Tables S8–S11 of the published article contain the analyzed IP-MS data. Additionally, the ZFHX4 consensus peaks of the CUT&RUN sequencing, as well as the results of the de novo and known motif analyses, can be found in Tables S13 and S14. All genotypic data of the individuals presented in this study can be found in Tables S1 and S2.

Consortia

The members of the ZFHX4 consortium are Abbey Scott, Alejandro Iglesias, Allyn McConkie-Rosell, Amy Kritzer, Ana Boto, Andi M. Lewis, Andreas Tzschach, Annarita Nicosia, Anne O’Donnell-Luria, Angela Peron, Ann-Charlotte Thuresson, Anneke Vulto-van-Silfhout, Antonio Novelli, Astrid S. Plomp, Balram Gangaram, Belén Gil-Fournier, Boris Keren, Casie Genetti, Chiara De Luca, Christian P. Schaaf, Claire Beneteau, Daniela Graziani, Daryl A. Scott, David A. Koolen, David Geneviéve, Dorine Bax, Elizabeth J. Leslie, Elizabeth Mizerik, Elizabeth Roeder, Emily Magness, Esther Nibbeling, Gaetan Lesca, Göran Annerén, Gregory M. Enns, Hilde Van Esch, Himanshu Goel, Inés Quintela, Jan Cobben, Jennifer L. Morrison, Jesus Eiris, Joan Stoler, Josephine Gao, Katharina Steindl, Katrina Dipple, Konstantinos Kolokotronis, Laurie H. Seaver, Lynda Pollack, Mahdi Motazacker, Manuela Gasparri, Margje Sinnema, Maria Cristina Digilio, Marie T. McDonald, Marisol Ibarra Ramírez, Martine Doco-Fenzy, Marzena Kucharczyk, Mathilde Nizon, Megan Boothe, Melanie C. O’Leary, Melissa T. Carter, Meredith Ross, Meredith Wilson, Merel Klaassens, Monia Ginevrino, Monica Hsiung Wojcik, Nathaniel Robin, Nicolas Chatron, Odile Boute, Pankaj Agrawal, Paolo Fontana, Paolo Zanoni, Pascal Joset, Pernille M. Torring, Ping Yee Billie Au, Servi Stevens, Soraya Ramiro León, Stephanie Baskin, Stephanie Parker DiTroia, Sue-Faye Siow, Tanguy Demaret, Vasilica Plaiasu, Virginia Clowes, Xi Luo, and Yvonne Hilhorst-Hofstee

Acknowledgments

We extend our gratitude to the participating families for their contribution in this study. We also appreciate the collaboration with the Flemish Institute for Biotechnology (VIB) Proteomics Core. We are particularly grateful for the data analysis conducted by the bioinformatics team, specifically Simon Devos and Sara Dufour. Additionally, the zebrafish project was made possible with the support of the Zebrafish Facility Ghent Core Facility at the Center for Medical Genetics Ghent (Ghent University), with special thanks to Andy Willaert and Hanne De Saffel. We would also like to extend our gratitude to Suzan Demunck, who helped prepare the zebrafish samples for mesoSPIM imaging. Financial support has been provided by grants 1520518N and G055422N from the Research Foundation – Flanders (FWO) and BOF/STA/201909/009 from the Special Research Fund (BOF) from Ghent University. M.d.R.P.B. is supported by a doctoral grant from the Marguerite-Marie Delacroix Foundation. L.H. is supported by a doctoral fellowship from the FWO (FWO 11I2525N) and was previously supported by a Marguerite-Marie Delacroix doctoral fellowship. E.D. is supported by a postdoctoral grant from the FWO (FWO 12D8523N) and was previously supported by Marguerite-Marie Delacroix and FWO doctoral fellowships. E.Z.J. was previously supported by an FWO doctoral fellowship. B.C. is a senior clinical investigator of the FWO. T.N. received a senior postdoctoral fellowship from the FWO and is further supported by the Polycystic Kidney Disease Foundation. Further, research was supported by the Concerted Research Actions from Ghent University (BOF20/GOA/023 to K.V. and BOF23/PDO/063 to T.N.) Extended acknowledgments can be found in the supplemental information.

Declaration of interests

B.S. is a shareholder in Episign, Inc.

Published: May 13, 2025

Web resources

• BrainSpan Atlas of the Developing Human Brain, http://www.brainspan.org

• Evo-devo Mammalian Organs, https://apps.kaessmannlab.org/evodevoapp/

• GenBank, https://www.ncbi.nlm.nih.gov/genbank/

• gnomAD v.4.1.0, https://gnomad.broadinstitute.org/about

• GTEx Portal, https://www.gtexportal.org/home/

• GTEx v.8, https://www.gtexportal.org/home/

• Human Embryonic Craniofacial Tissue Bulk RNA-seq, http://cotneyweb.cam.uchc.edu/craniofacial_bulkrna/

• Human Organoid Single-Cell Browser (Shcheglovitov Lab), https://shcheglovitov.shinyapps.io/u_brain_browser/gnomAD

• OMIM, https://www.omim.org

• UniProt, www.uniprot.org