Background

Chromodomain helicase DNA-binding protein 8 (CHD8) has emerged as a top ASD candidate gene from multiple exome-sequencing studies [1–4]. Loss of function mutations in the gene have also been found in schizophrenia (SZ) and intellectual disabilities [4–6]. CHD8 is a ubiquitously expressed member of the CHD family of ATP-dependent chromatin-remodeling factors that play important roles in chromatin dynamics, transcription, and cell survival [7–11]. Previous studies have shown that CHD8 protein negatively regulates Wnt signaling by interacting with β-catenin: Wnt/β-catenin signaling plays a critical role in normal brain development and has been implicated in bipolar disorder (BD), SZ, ASD, and cancer [4, 10, 12–23]. The effect of CHD8 on the growth of cancer cells appears to be due, in part, to an interaction with p53 [24]. CHD8 also recruits MLL histone methyltransferase complexes to regulate cell cycle genes [25] and binds to the chromatin insulator CTCF [25, 26]. Recent work also shows that CHD8 and other CHD chromatin remodelers regulate embryonic stem cell transcriptional programs by targeting specific nucleosomes that flank nucleosome-free promoter regions [9].

Based on these observations, we have been studying the effects of CHD8 on human neurons and neural progenitor cells (NPCs) using CHD8 +/− lines generated in isogenic-induced pluripotent stem (iPS) cells by CRISPR-Cas9 gene editing [8]. Other investigators have been studying the effect of CHD8 on neuronal cells using RNA interference (RNAi) [27–29]. These studies have focused primarily on analyzing downstream targets of CHD8 in order to identify differentially expressed genes (DEGs). This is a particularly useful strategy for studying ASD and SZ candidate genes that function as regulators of gene expression, in order to find converging pathways that could connect many different genetic risk factors into more manageable common molecular subgroups—an idea that could facilitate drug discovery. ASD and SZ candidate genes that code for gene expression regulators (e.g., transcription factors and chromatin-remodeling complexes) represent, along with genes that code for synaptic proteins, calcium channels, potassium channels, and the HLA (MHC) locus, the major categories of validated candidate genes in these conditions [4, 30–33]. Molecular genetic convergence has previously been demonstrated for some candidate genes. For example, the SZ and BD candidate gene MIR137 has been found to target other candidates: CSMD1, C10orf26, CACNA1C, and TCF4 [34]. In addition, clinically distinct disorders can be caused by the same risk genes, suggesting that therapies aimed at specific molecular targets could have a therapeutic effect across diagnostic categories [4, 35].

The molecular studies that have targeted CHD8 certainly support the concept of converging molecular targets and pathways. In shRNA knockdown studies and chromatin immunoprecipitation using NPCs, neural stem cells (NSCs), and SK-N-SH neuroblastoma cells, downregulation of CHD8 predicted a disruption of gene networks involved in neurodevelopment and resulted in altered expression of a significant number of other ASD-risk genes [3, 27–29].

Similarly, in our recently published study, a significant number of previously characterized ASD and SZ candidate genes were found to be differentially expressed in CHD8 +/− NPCs and neurons, compared with isogenic controls [8], and furthermore, DEGs were found to overlap with the downstream targets of several other SZ and ASD candidate genes that code for transcription factors or chromatin regulators, including TCF4, EHMT1, and SATB2 [6, 8, 36–38]. This suggests that CHD8 not only has a direct effect on gene expression but has indirect effects as well. We also found that DEGs were enriched for pathways that affect the extracellular matrix (ECM), cell adhesion, neuron differentiation, neuron projection, synaptic transmission, axonal guidance signaling, and WNT/β-catenin and PTEN signaling. In addition, genes involved in head circumference were found to be differentially expressed. This is notable because loss of function CHD8 mutations are associated with large head circumference, a finding that has been experimentally validated in a zebrafish model [2, 3].

Our previous study was carried out using NPCs and a monolayer neuronal culture system consisting of a fairly heterogeneous array of neurons expressing forebrain, midbrain, and hindbrain markers. Recently, several neuronal differentiation methods have emerged that are more suitable for SZ and ASD, one of which is the direct conversion of iPS cells into 3-dimensional cerebral organoids, which resemble a first trimester developing telencephalon [39–41]. This is particularly appropriate for studying neurodevelopmental disorders that are associated with cognitive dysfunction. We have also demonstrated that the organoid system is ideal for studying gene × environment interactions relevant to neuropsychiatric and neurodevelopmental disorders [39].

The few studies that have been carried out so far using cerebral organoids as a model system have been revealing. Mariani et al., for example, showed that genes involved in cell proliferation, neuronal differentiation, synaptic assembly, and GABAergic inhibitory neuron development were differentially expressed in an idiopathic ASD family [42]. And, using a somewhat different organoid differentiation protocol, Lancaster et al. showed that cerebral organoids derived from patients with CDK5RAP2 loss of function variants and microcephaly have premature neuronal differentiation [41].

Accordingly, we have expanded our transcriptome analysis of CHD8 target genes in cerebral organoids derived from CHD8 +/− iPS cells and isogenic controls. The DEGs reported here validate many of the findings in our previous analysis in NPCs and monolayer neuronal cultures. In particular, we show that CHD8 haploinsufficiency again leads to a substantial increase in TCF4 expression [8]. In addition, significant overlap was found with the DEGs previously identified in the Mariani et al. study, which was carried out using subjects with idiopathic ASD in whom the responsible genetic variant could not be unequivocally characterized [42]. The long non-coding antisense RNA DLX6-AS1, a regulator of GABAergic interneuron development [43], was the top DEG in both. Considering the genetic heterogeneity found in ASD and SZ, the molecular convergence on DLX6-AS1 between CHD8 and an uncharacterized ASD-causing genetic variant is striking.

Methods

Development of iPSCs from skin fibroblasts

We have been developing iPS cells from controls and patients with 22q11.2 del diagnosed with SZ or schizoaffective disorder [44]. One of the male control samples was used to generate the CHD8+/− lines. The control was recruited from the Albert Einstein College of Medicine (AECOM). The study and consent forms were approved by the AECOM Institutional Review Board (IRB). Consent was obtained by a skilled member of the research team who had received prior human subjects training. iPSC lines were generated from fibroblasts obtained from skin biopsies performed by board-certified physicians. The procedure for growing fibroblasts in preparation for reprogramming into iPS cells is detailed in Additional file 1: Supplemental methods.

Generating CHD8 KO lines

CHD8 +/− lines were developed by introducing a CRISPR-Cas9 vector containing CHD8 guide sequences into iPS cells by nucleofection [8]. The procedure is described in detail in Additional file 1: Supplemental methods.

Cerebral organoid differentiation

The protocol is adapted from Mariani et al. [40]. Briefly, iPS cell colonies were maintained on matrigel in mTesr1. To induce cerebral organoid differentiation, iPS cells were pretreated with 50 μM Y27632 in mTesr1 for 1 h at 37 °C. Wells were rinsed with DMEM/F12, and iPS cell colonies were dissociated with accutase for 10 min at 37 °C. Cells were rinsed with DMEM/F12 and collected and counted for aggregate formation. Following the Stem Cell Technologies protocol, 3.0 × 106 cells were used to create 10,000 cell aggregates using an AggreWell™ plate. For the first 6 days, aggregates were cultured in mTesr1 supplemented with 500 ng/ml DKK-1, 1.5 μg/ml BMPRIA-Fc, and 10 μM SB431542. On day 6, aggregates were removed from the AggreWell™ plate, according to the Stem Cell Technology protocol, and transferred to a 24-well ultra-low attachment plate. On day 18, 1% N2 supplement was added to the medium. On day 25, aggregates were plated onto a 4-well chamber slide coated with 10 μg/ml polyornithine, 2.5 μg/ml laminin, and 50 μg/ml fibronectin, and cultured in Neurobasal medium supplemented with 2% B27 and 2 mM l-glutamine until day 50. Organoids were detached, and RNA was extracted. Organoids are composed of a mixture of GABAergic and glutamatergic neurons, and radial glia progenitor cells, and have gene expression profiles that resemble a first trimester telencephalon (Additional file 2: Figure S1) [39, 40, 45]. For immunohistochemistry (IHC), samples were fixed with 4% paraformaldehyde and 25% sucrose, and then embedded in O.C.T (optimal cutting temperature) (see Additional file 1: Supplemental methods for IHS methodology).

RNA-seq

Total RNA was isolated using the miRNeasy kit (Qiagen) according to the manufacturer’s instructions. We obtained 101 bp paired-end RNA-seq reads from an Illumina HiSeq 2500 instrument. Adapters and low quality bases in reads were trimmed by trim_galore (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). We employed Kallisto (v0.42.5) [46] to determine the read count for each transcript and quantified transcript abundance as transcripts per kilobase per million reads mapped (TPM), using gene annotation in the GENCODE database (v18) [47]. Then we summed the read counts and TPM of all alternative splicing transcripts of a gene to obtain gene expression levels. We restricted our analysis to 12,898 expressed genes with an average TPM >1 in either wild type or CHD8 +/− samples. DESeq2 [48] was used to identify DEGs (false discovery rate (FDR) <0.05). The software DAVID (v6.8 Beta) [49, 50] was used for Gene Ontology (GO) analysis, with the 12,898 expressed genes as background. Ingenuity pathway analysis (IPA) (https://www.qiagenbioinformatics.com/) was used for canonical pathway analysis, using the ingenuity knowledge base (genes only) as background. The RNA-seq data have been deposited in Gene Expression Omnibus (GEO: accession number GSE85417).

Quantitative real-time PCR (qPCR)

qPCR was carried out on reverse transcribed PCR using the 2−ΔΔCt method as previously described [51, 52]. A detailed description and the primers used for this analysis can be found Additional file 1: Supplemental methods.

ASD/SZ-risk gene sets

For ASD, we compared our DEG list with the following ASD gene sets: SFARI [https://gene.sfari.org/autdb/GS_Home.do] (genes scored as high confidence, to minimal evidence and syndromic); AutismKB (core dataset) [53]; a set of high-confidence ASD genes (Willsey_ASD) [54]; genes predicted by whole exome sequencing and co-expression network analysis (Liu_ASD) [55]; candidate genes with de novo mutations from massive whole exome sequencing (Iossifov_ASD) [56]; and candidates from the same dataset focusing on a combination of de novo and inherited mutations resulting in a high-confidence list (FDR < 0.1) (DeRubeis_ASD) [57]. The two SZ gene lists were from the SZ gene database [58] and a recent genome-wide association study (GWAS) report (SZC GWAS) [33]. These gene lists can be obtained from our previous publication [8].

Comparison of CHD8 +/− DEGs with idiopathic ASD organoids

The DEG list from CHD8 +/− organoids was compared to the DEG lists generated from idiopathic autism patient-specific organoids described by Mariani et al. [42]. The latter were obtained from two developmental stages, after 11 and 31 days of terminal differentiation (TD11 and TD31).

Comparison of CHD8 +/− DEGs with BD patient-derived neurons

DEG lists from Mertens et al. [59] were derived from the file “GSE58933_Jun_All_Data.txt.gz” in the GEO “GSE58933” record. For a comparison with our CHD8 KO samples, we applied the same criteria used in the original study for identifying DEGs (log2 (fold change) ≥1 and p ≤ 0.05).

Statistics

To determine if DEGs overlapped with or were significantly enriched with a specific gene set, 12,893 expressed genes in our samples were used as background for Fisher’s exact test. Statistics tests were conducted in R (http://www.R-project.org/). Common genes between two gene lists were input to DAVID (beta 6.8) for GO term analysis.

Results

RNA-seq was carried out on cerebral organoids derived from CHD8 KO iPS cells; two isogenic controls (CHD8 +/+) and four heterozygotes (CHD8 +/−). The CHD8 +/− samples contain a CHD8 KO allele with either a 10-base pair deletion (clones A, B, and C) or a 2-base pair deletion (D), both of which lead to frameshift mutations and premature stop signals in exon 1 [8]. The KO lines were derived from CHD8 +/+A; the other control, CHD8 +/+B, was a different iPS cell clone from the same subject. We previously showed that heterozygous KO leads to a ~50% reduction in CHD8 protein [8]. Similarly, quantitative immunohistochemistry showed a 54% decrease in CHD8 immunoreactivity in CHD8 +/− compared with CHD8 +/− organoids (analyzed in 15 random fields, p = 7.2E-13) (Additional file 2: Figure S1).

The RNA-seq data quality is shown in Additional file 3: Table S1. A total of 12,893 expressed genes were detected, and DESeq2 was used to identify DEGs, as described in detail in the “Methods” section. Using a cutoff of FDR < 0.05, there were 559 DEGs when the CHD8 +/+ organoids were compared with CHD8 +/−; 288 genes increased in the KO, 271 decreased. The DEGs separated our sample into two groups, as seen in the heat map shown in Fig. 1a. The entire list of DEGs is in Additional file 4: Table S2. CHD8 mRNA itself was not significantly differentially expressed based on our RNA-seq analysis. The KO allele, however, showed a much lower level of expression than the WT allele in the organoids (Additional file 1), probably due to nonsense mediated decay. Overall, though, the decrease in CHD8 mRNA was not proportional to the decrease in CHD8 protein, similar to our observations in NPCs [8]. The relatively imprecise correlation between mRNA and protein levels is found for many genes and can be due to a number of factors [60]. However, the mechanism of the discrepancy between CHD8 mRNA and protein is not known and will require further investigation.

Fig. 1
figure 1

Heat map and summary of GO terms and pathways. a The heat map shows differentially expressed genes between controls (CHD8 +/+) and heterozygous knockouts (CHD8 +/−). Enriched GO terms by DAVID (top) and pathways by IPA (bottom) for upregulated (b) and downregulated (c) genes in CHD8 +/− organoids. P values were corrected by the Benjamini method [147]

Full size image

We should also point out that among the three CHD8 alternatively spliced transcripts in the GENCODE annotation, the two containing the exon 1 accounted for 70 ~ 80% of the CHD8 transcripts in the WT organoids and 60–70% in the CHD8 +/−, based on our RNA-seq data (Additional file 1).

Of the 559 DEGs, 203 have CHD8 binding sites in their promoters, using data from a ChIP-seq study carried out on NPCs by Sugathan et al. (see Additional file 4: Table S2, column I) [28]. The finding that such a large fraction of DEGs are direct targets of CHD8 confirms the validity of our RNA-seq findings. However, it also shows that many downstream genes are indirect targets of CHD8, most likely through the actions of other genes coding for transcription factors and chromatin-remodeling proteins that are directly affected by CHD8, such as TCF4, POU3F2, SMARCA4, SOX2, and PAX6. This result is consistent with our previous report [8].

The software DAVID was used to identify enriched GO pathways in DEGs using 12,893 expressed genes as background [49]. IPA was used for canonical pathways and disease association. The top GO terms (Biological Process, (BP)) for genes that were upregulated in the CHD8 KO organoids were nervous system development, neurogenesis, neuron differentiation ,and forebrain development; the top GO:BP terms for downregulated genes were nervous system development, generation of neurons, and neuron differentiation (Fig.1b, c; Additional file 5: Table S3). Genes coding for components of the ECM were the top cellular component (CC) GO terms for upregulated DEGs, and among the top eight for downregulated DEGs, similar to our previous findings using monolayer neurons [8]. The top enriched IPA canonical pathways were Wnt/β-catenin signaling and axonal guidance for upregulated genes and axonal guidance for downregulated genes. An enrichment of DEGs involved in Wnt/β-catenin signaling is similar to that found in our previous transcriptome analysis on CHD8 +/− NPCs and neurons [8], as well as findings by other investigators [10, 20, 24], firmly establishing that altered expression of CHD8 disrupts this critical signaling pathway.

As a complementary analysis, we also applied TopHat and DESeq2 for aligning the RNA-seq reads and for DEG analysis, respectively, as we previously carried out [8]. This resulted in 811 DEGs (Additional file 4: Table S2 sheet 2), 534 of which were included in the DEG list from Kallisto/DESeq2 analysis. GO analysis showed an enrichment of similar GO terms in the two DEG lists, with “neuron system development” being the top term for both upregulated and downregulated genes (Fig. 1).

Overall, the findings show that CHD8 directly, or indirectly through effects on other transcription factors and chromatin regulators, regulates a program of gene expression that affects critical aspects of brain development (e.g., neurogenesis, neuron differentiation, and axonal guidance).

Comparison between organoid data and NPCs and monolayer neurons

We compared current transcriptome data with our previous study using NPCs and monolayer neurons [8]. There is a significant overlap between the studies, with nearly 50% of DEGs in organoids showing differential expression in NPCs and neurons (neurons odds ratio [OR] = 2.88, p < 2.2E-16; NPCs OR = 4.44, p < 2.2E-16, Fisher’s test) (Fig. 2; see Additional file 4: Table S2 for overlapping genes). The top GO terms for overlapping genes were neuron differentiation and neurogenesis, respectively, which is consistent with the main pathway findings in organoids described above.

Fig. 2
figure 2

Overlapping DEGs in organoids compared with NPCs and monolayer neurons from previous study [8]

Full size image

qPCR validation

We validated several RNAs of interest by qPCR using the 2−ΔΔCt relative expression method: SOX2, PAX6, TCF4, CNTNAP2, HMGA2, RELN, MEG3, DLX6-AS1, and CRNDE (Fig. 3). These genes were chosen because of their known importance in brain development and disease. SOX2 and PAX6 code for transcription factors that influence neural stem cell growth and brain development [60, 61]. TCF4 codes for a transcription factor; both common and rare variants have been implicated in the etiology of SZ, BD, ASD, and developmental delay [34, 36, 37, 62]. CNTNAP2 codes for a member of the neurexin family of presynaptic proteins; it too has been implicated in the pathogenesis of SZ and ASD [63–67]. HMGA2 codes for a non-histone DNA-binding protein that has been implicated in regulating brain growth and head circumference; increased expression was found in our previous transcriptome analysis [8, 68]. RELN codes for reelin, a key secreted ECM protein involved in neuronal migration during brain development [69–72]. Altered expression has been found in SZ and ASD [72–77]. As seen in Fig. 4, reelin is expressed throughout the organoids, in fields of neurons as well as in the zone of proliferating radial glia progenitors found in these structures [39, 40, 42]. MEG3 is a maternally expressed imprinted gene that acts as a tumor suppressor gene in a number of malignancies [21, 78–82]. DLX6-AS1 and CRNDE will be discussed below. qPCR analysis validated the RNA-seq findings for each of these genes.

Fig. 3
figure 3

Validation of selected DEGs by qPCR. The RNA samples used in the RNA-seq were used for this analysis. Samples were analyzed in triplicate as described in “Methods” section and Additional file 1: Supplemental methods. Significant differences between control and KO are denoted by an asterisk (*). The p values derived by Student’s t test were as follows: SOX2, 0.047; TCF4, 0.004; HMGA2, 0.00005; PAX6, 0.02; RELN, 0.02; CNTNAP2, 0.04; DLX6-AS1, 0.0001; MEG3, 0.00002

Full size image
Fig. 4
figure 4

Immunohistochemistry. RELN and the neuronal marker NeuN were visualized as described in Additional file 1: Supplemental methods. The DAPI+ tubular structures are zones of proliferating radial glia progenitors. NeuN+ cells are fields of neurons surrounding the radial glia progenitors

Full size image

Top DEGs and overlap with organoid transcriptome in idiopathic ASD

Two of the top three DEGs in CHD8 +/− cerebral organoids were DLX6-AS1 and DLX1, which increased ~39 and 13-fold respectively (Additional file 4: Table S2). They were hardly expressed in controls. DLX6-AS1 (also known as EVF2) forms a complex with DLX1 and DLX2 proteins that subsequently regulates GABAergic interneuron development by increasing DLX5 and DLX6 gene expression [43, 83–86]. Strikingly, in a study by Mariani et al., DLX6-AS1 was also the top DEG in a transcriptome analysis carried on day 11 cerebral organoids derived from members of a family with idiopathic ASD, and the 6th top DEG in day 31 organoids [42]. DLX1 was differentially expressed at both time points as well.

In addition to DLX6-AS1, among the top 15 DEGs in CHD8 +/− cerebral organoids, 11 were also DEGs in the Mariani et al. study in either or both of their day 11 and day 31 organoids, and the direction of change was the same (Table 1). This included genes involved in brain development, several of which have been implicated in ASD, including FZD8, PAX6, SLC1A3, EOMES (TBR2), and MPPED1.

Table 1 Top 15 DEGs in CHD8+/− vs idiopathic ASD cerebral organoids
Full size table

Aside from the top DEGs, overall, there was a significant overlap in DEGs between our CHD8 +/− organoids and idiopathic ASD day 11 and day 31 organoids in the Mariani et al. study (Fig. 5; Additional file 6: Table S4). Similarly, significant overlap was detected in a comparison of DEGs from our previous study on NPCs and monolayer neurons with idiopathic ASD organoids. The most significant overlap was found in the comparison between CHD8 +/− organoids and day 31 organoids from the Mariani et al. study, which showed that 23% of DEGs were the same (131/560; OR = 5.04; p = 1.34E-40). The top GO terms for overlapping genes were nervous system development, neuron differentiation, and neurogenesis for day 11 and day 31 organoids (Additional file 6: Table S4).

Fig. 5
figure 5

Overlap in DEGs between CHD8 and idiopathic ASD organoids. CHD8 KO DEGs were compared with DEGs from day 11 and day 31 organoids (TD11, TD31) derived from individuals with idiopathic ASD [42]. The number in each panel shows the number of overlapping genes, which can be seen in Additional file 5: Table S4. The numbers in parentheses are the odds ratios. Color represents p value from Fisher’s test

Full size image

Although the degree of overlap is impressive, a key difference in our respective studies is that FOXG1 was a top, upregulated DEG in the idiopathic ASD organoids, but not in the CHD8 +/− samples; reducing FOXG1 by RNAi rescued the over-abundance of GABAergic interneurons found in the idiopathic ASD organoids [42]. Conversely, TCF4, a major DEG in all of our CHD8 +/− samples, was not differentially expressed in the idiopathic ASD organoids. This suggests that unique expression changes can occur in key genes despite the extensive overlap in transcriptomes, which could conceivably limit the full therapeutic impact of novel drugs that target common pathways.

CHD8 haploinsufficiency and WNT-β-catenin signaling

WNT/β-catenin signaling is a key pathway in the developing brain that is dysregulated in neuropsychiatric disorders, as well as in various cancers [4, 10, 12–23]. Thus, the finding that CHD8 binds to β-catenin, inhibiting its transcriptional effects, [10, 20] is relevant to the role of CHD8 in both neuropsychiatric and neurodevelopmental disorders, as well as cancers. Recently, however, CHD8 was shown to be a positive regulator of WNT/β-catenin signaling in human NPCs [87]. Among the neuropsychiatric disorders, WNT/β-catenin is particularly relevant to BD because lithium salts, which are used to treat the condition, inhibit GSK3β, which would be expected to result in an increase in β-catenin levels (constitutive GSK3β activity leads to β-catenin degradation) [88–92]. With these considerations in mind, as well as our finding that WNT/β-catenin signaling is the top pathway for upregulated DEGs in CHD8+/− organoids (Fig. 1b, c), we compared our DEG list with those found in a recent study by Mertens et al. in which transcriptome analyses were carried out in iPS cell-derived neurons from BD patients who were clinically responsive or not responsive to lithium [59]. In addition, we also evaluated the DEG list derived from NPCs and monolayer neurons from our previous study, which also showed that WNT/β-catenin was a top pathway among DEGs in monolayer neurons [8]. Using the same criteria for defining DEGs in the Mertens et al. study (log2 fold change ≥1 and p ≤ 0.05), significant overlap was found in each comparison (Fig. 6; Additional file 7: Table S5). Note that only 30–50% of DEGs in the Mertens et al. study were expressed in our organoids, which is probably due to differences in the differentiation protocols used. However, the most significant overlap (p ≤ 5.73E-23; OR = 4.64) occurred in the comparison between our CHD8 +/− DEGs and the lithium non-responders in Mertens et al. (Interestingly, DLX6-AS1 and DLX1 were in this group of overlapping DEGs; Additional file 7: Table S5).

Fig. 6
figure 6

Overlap in DEGs between CHD8 KO and BD neuronal cells. CHD8 KO DEGs from the current study (organoids) and our previous study (NPCs, neurons) [42] were compared with DEGs found in neurons derived from lithium responder and lithium non-responder patients with BD, as described in “Methods” section. The number in each panel shows the number of overlapping genes, which can be seen in Additional file 6: Table S5. The numbers in parentheses are the odds ratios

Full size image

The top GO terms for overlapping genes were similar for both the Li non-responder and Li responder vs CHD8+/− organoid DEGs; nervous system development, neuron differentiation, and neurogenesis (Additional file 7: Table S5). However, one GO term found exclusively in the former was axonogenesis. Although lithium is extremely useful in a substantial proportion of BD patients, it is also used on occasion to treat patients diagnosed with SZ and ASD, especially as adjunctive therapy for those with a mood component and for refractory patients to augment the effect of anti-psychotic medications [88, 93–98]. Although there was a greater association to the lithium non-responder group, considering that there is significant overlap with the lithium-responder group as well could have therapeutic implications.

ASD and SZ candidates in CHD8 +/− DEGs

The protein-coding DEGs were considered for their over-representation of SZ and ASD candidate genes using a variety of sources, as described in the “Methods” section and our previous study [8]. As seen in Fig 7, among the CHD8 +/+ vs CHD8 +/− DEGs, there was significant enrichment of ASD candidate genes in the SFARI, AutismKB, and Willsey ASD datasets, and an even more significant enrichment of SZ candidates in the SZGene and SZ GWAS lists (see Additional file 8: Table S6 for complete list). Enriched GO terms were identified for the overlapping genes in three of these datasets: SFARI, AutismKB, and SZ GWAS. The top GO terms for SFARI and AutismKB were similar; forebrain development, telencephalon development, and pallium development were the most significant (Additional file 8: Table S6). For the SZ GWAS data set, the top GO terms for overlapping genes were somatodendritic compartment, synapse, neuron projection, cell body, and axon. The differences between the ASD and SZ sets of overlapping genes reflect the observation that CHD8 haploinsufficiency is a risk factor for both groups of conditions and suggest that disruption of different molecular pathways is involved in the increased risk and differences in clinical presentation.

Fig. 7
figure 7

Overlap between CHD8 KO DEGs and ASD and SZ candidate genes. CHD8 KO DEGs were evaluated for enrichment of ASD and SZ candidate genes. Bars represent p value (Fisher’s exact test, one-tailed), while red dots represent odds ratio of overlap. The number above each red dot shows the number of overlapping genes, which can be found in Additional file 7: Table S6

Full size image

DEGs involved in head circumference/brain volume

Patients with loss of function CHD8 mutations typically have large head circumferences, a finding confirmed in a zebrafish model [2, 3]. In our previous study, 7 out of 12 genes (TESC, DDR2, HMGA2, SBNO1, FAT3, BCL2L1, and MSRB3) implicated in brain size in genome-wide association studies were differentially expressed in CHD8 +/− NPCs and neurons [8, 68, 99, 100]. However, among these genes, only HMGA2 (high mobility group AT-hook 2) expression was similarly affected in CHD8 KO organoids. In addition, one gene, DCC (deleted in colorectal carcinoma), which was not differentially expressed in the previous study, was significantly decreased in CHD8 +/− organoids. Both the HMGA2 and DCC gene loci have CHD8 binding sites [28]. The findings show unequivocally that HMGA2 expression is regulated by CHD8.

ncRNAs

In addition to DLX6-AS, there were 19 other non-coding RNAs (ncRNAs) that were differentially expressed in CHD8 +/− organoids (Table 2), of which 9 are CHD8 targets based on published ChIP findings [28]. Several of the differentially expressed ncRNAs we detected have been implicated in neuropsychiatric disorders. One is the RMST locus, which contains a microRNA involved in forebrain development through its modulation of WNT/β-catenin signaling and has been found to regulate neurogenesis through an interaction with SOX2 [13, 101]. Another is MIAT (also known as GONAFU); RNA levels were found to be expressed at higher levels in parvalbumin GABAergic interneurons in SZ subjects [102] and have been found to regulate fear-related anxiety traits in mice [103]. MIAT also binds to several splicing factors and may affect splicing of the SZ-associated ERBB4 and DISC1 genes in response to neuronal activation [104]. Finally, CRNDE expression is significantly elevated in iPS cell-derived neurons from patients with SZ who have 22q11.2 deletions [105]. Although the molecular effects of most ncRNAs have not been determined, many function at the gene expression level through their regulation of chromatin architecture and nuclear organization, which, if disrupted, could potentially alter neurodevelopmental molecular programs [43, 82, 104, 106–111].

Table 2 Differentially expressed ncRNAs
Full size table

Discussion

One of the most interesting characteristics of CHD8 is its diverse effect on several neuropsychiatric and neurodevelopmental disorders and cancer. Although ASD and cancer differ fundamentally in a key aspect regarding loss of function CHD8 mutations and disease in that the former are due to germline mutations, while the latter are usually somatic, it is not surprising that a chromatin and transcriptional regulator like CHD8 would play a role in both types of conditions. Indeed, in recently published pathway network analyses and sequencing studies, overlap was found for several ASD candidate genes and cancer [30, 112]. Correspondingly, germline mutations in NF1, which cause neurofibromatosis type I, often display autistic-like behaviors [113, 114]. The molecular genetic overlap suggests that some novel cancer therapies currently being developed, especially those that target the epigenome might be beneficial in treating subgroups of individuals with neurodevelopmental disorders [112, 115].

The mechanisms by which loss of function CHD8 mutations increase cancer risk are likely to be multifactorial. Part of the effect appears to be due to the direct interaction between CHD8 protein with β-catenin and p53, and an effect on the cell cycle [2, 24, 25]. An effect mediated by β-catenin is given additional support by our transcriptome analysis. In addition, based on our findings, CHD8 may also contribute to malignant transformation indirectly through its effects on other genes, which were found to be differentially expressed in this study and have been implicated in malignant transformation, such as SMARCA4, POU4F1, ARMCX1, HMAG2, DCC, and ZNF132 (Additional file 4: Table S2) [116, 117]. Several of the differentially expressed ncRNAs we show on Table 2 have also been found to be associated with cancer development, including TERC, which codes for the RNA component of telomerase; telomere shortening is a feature of malignant transformation and aging [118]. MEG3, as noted above, CRNDE, LINC00340, and RMST (rhabdomyosarcoma 2 associated transcript) is also found in various cancers [118–124]. These findings suggest that the effect of CHD8 on malignant transformation is multifactorial and not simply due to a direct interaction with the Wnt/β-catenin signaling pathway.

Similarly, the role of CHD8 on neuropsychiatric and neurodevelopmental disorders is multifactorial, with both direct effects on downstream targets, such as β-catenin, and indirect effects mediated by dysregulated expression of other transcription factors and chromatin remodelers. The best example of this is the SZ and BD candidate gene TCF4, which codes for a basic helix-loop-helix transcription factor [125]. CHD8 haploinsufficiency leads to a ~2-fold increase in TCF4 expression in cerebral organoids and NPCs and neurons (Additional file 4: Table S2) [8]. In addition, pathway analysis showed extensive overlap with TCF4 targets, and CHD8 binds to the TCF4 gene locus [8, 28]. An increase in TCF4 expression has also been found in iPS cell neurons and fibroblasts derived from SZ patients [126, 127]. In addition, TCF4 is upregulated by loss of function mutations in the SZ candidate MIR137, and overexpression in the forebrain of mice leads to cognitive impairments and deficits in pre-pulse inhibition [128–130]. Overall, the findings strongly suggest that TCF4 and CHD8 cooperate to influence neuronal differentiation and brain development, and that TCF4 overexpression is a key feature in SZ and BD.

On the other hand, loss of function TCF4 mutations have been found in patients with Rett syndrome-like phenotypes [131, 132], ASD [133] and Pitt–Hopkins syndrome, which is characterized by ASD, intellectual disabilities, and microcephaly [134, 135]. Thus, TCF4 gene dosage in either direction adversely affects brain development.

Other transcription factors and chromatin regulators that are significantly affected by CHD8 haploinsufficiency that are also ASD and SZ candidate genes include POU3F2 and AUTS2. POU3F2 expression is significantly decreased in CHD8 +/− in cerebral organoids, as well as in NPCs and neurons, while AUTS2 expression is significantly decreased in CHD8 +/− in cerebral organoids and neurons, but not NPCs (Additional file 4: Table S2) [8]. POU3F2 codes for a member of the POU family of transcription factors, and has been implicated in SZ, and more recently in BD, developmental delay, and intellectual disability [136, 137]. Its critical role in neuronal differentiation is highlighted by the finding that it is one of the three factors, along with MYT1L and ASCL1, used in the direct reprogramming of fibroblasts into neurons [138]. AUTS2 codes for a chromatin-remodeling protein that functions as a component of the polycomb repressive complex 1 (PRC1); loss of function mutations can lead to ASD, SZ and intellectual disabilities [139–141]. Also, it is interesting to note, in the context of the ASD/cancer connection, that AUTS2 is part of a translocation commonly found in childhood B cell precursor acute lymphoblastic leukemia [142].

One of the most interesting downstream targets of CHD8 we found regarding neurodevelopmental and neuropsychiatric disorders is the non-coding RNA, DLX6-AS1, which was the top DEG in our CHD8 +/− organoids, as well as organoids derived from a family with idiopathic ASD [42]. DLX6-AS1 overlaps with DLX6 and is expressed in the opposite transcriptional orientation. In mice, a splice variant of Dlx6-as1 called Dlx6-as2 (evf2) cooperates with Dlx2 to increase the transcriptional activating function of the Dlx5/6 enhancer [143].

These findings suggest that CHD8 affects GABAergic interneuron development, by modulating DLX gene expression. Consistent with this idea is the finding that several genes, in addition to DLX1, involved in cerebral cortex GABAergic interneuron differentiation, FEZF2, ARX, and CNTN2, were differentially expressed in CHD8 +/− organoids (Additional file 4: Table S2). Statistically, the enrichment of genes involved in cerebral cortex GABAergic interneuron differentiation only achieved a trend toward significance (p = 3.6E-3; padj = 5.8E-2), which could be due to the relatively small sample size. Abnormalities in cortical GABA interneuron function, in particular, parvalbumin positive and somatostatin positive interneurons have been found in SZ and ASD [144–146].

It should be noted that a limitation of this study is that it is based on a knockout carried out on a single iPS cell line, so replication in other lines is critical. Nevertheless, our findings strongly support a major role of CHD8 on Wnt/β-catenin signaling, and a connection between CHD8 and TCF4, and several other genes that have been implicated in neuropsychiatric and neurodevelopmental disorders, in particular, members of the DLX gene family. In addition, the overlap we detected between the CHD8 +/− transcriptome and the transcriptomes obtained in idiopathic ASD and BD shows that common molecular pathways exist in different clinical conditions caused by seemingly disparate candidate genes. Identifying such common pathways will facilitate drug discovery in these genetically heterogeneous disorders.

Conclusions

CHD8, which codes for a chromatin-remodeling factor, is mutated in a subgroup of patients with ASD and SZ. RNA-seq analysis of cerebral organoids derived from iPS cells that are heterozygous for a CHD8 knockout allele, and isogenic controls, shows that CHD8 regulates the expression of other genes implicated in ASD and SZ, notably TCF4 and AUTS2. In addition, extensive overlap was observed for differentially expressed genes (DEGs) found in another study using organoids derived from a family with idiopathic autism, especially for genes involved in GABAergic interneuron development. These findings show molecular convergence of disparate genes involved in the development of ASD and SZ, an observation that will facilitate drug discovery. In addition, pathway analysis of DEGs revealed an enrichment of genes involved in regulating Wnt/β-catenin signaling, a druggable pathway.