RES E A R C H Open Access
CRISPR/Cas9-mediated heterozygous
knockout of the autism gene CHD8 and
characterization of its transcriptional
networks in cerebral organoids derived
from iPS cells
Ping Wang
1†
, Ryan Mokhtari
2†
, Erika Pedrosa
2
, Michael Kirschenbaum
2
, Can Bayrak
3
, Deyou Zheng
1,4,5*
and Herbert M. Lachman
1,2,5,6*
Abstract
Background: CHD8 (chromodomain helicase DNA-binding protein 8), which codes for a member of the CHD
family of ATP-dependent chromatin-remodeling factors, is one of the most commonly mutated genes in autism
spectrum disorders (ASD) identified in exome-sequencing studies. Loss of function mutations in the gene have also
been found in schizophrenia (SZ) and intellectual disabilities and influence cancer cell proliferation. We previously
reported an RNA-seq analysis carried out on neural progenitor cells (NPCs) and monolayer neurons derived from
induced pluripotent stem (iPS) cells that were heterozygous for CHD8 knockout (KO) alleles generated using CRISPR-
Cas9 gene editing. A significant number of ASD and SZ candidate genes were among those that were differentially
expressed in a comparison of heterozygous KO lines (CHD8
+/−
)vsisogeniccontrols(CHD8
+/−
), including the SZ and
bipolar disorder (BD) candidate gene TCF4, which was markedly upregulated in CHD8
+/−
neuronal cells.
Methods: In the current study, RNA-seq was carried out on CHD8
+/−
and isogenic control (CHD8
+/+
)cerebralorganoids,
which are 3-dimensional structures derived from iPS cells that model the developing human telencephalon.
Results: TCF4 expression was, again, significantly upregulated. Pathway analysis carried out on differentially expressed
genes (DEGs) revealed an enrichment of genes involved in neurogenesis, neuronal differentiation, forebrain
development, Wnt/β-catenin signaling, and axonal guidance, similar to our previous study on NPCs and monolayer
neurons. There was also significant overlap in our CHD8
+/−
DEGs with those found in a transcriptome analysis carried
out by another group using cerebral organoids derived from a family with idiopathic ASD. Remarkably, the top DEG in
our respective studies was the non-coding RNA DLX6-AS1, which was markedly upregulated in both studies; DLX6-AS1
regulates the expression of members of the DLX (distal-less homeobox) gene family. DLX1 was also upregulated in
both studies. DLX genes code for transcription factors that play a key role in GABAergic interneuron differentiation.
Significant overlap was also found in a transcriptome study carried out by another group using iPS cell-derived
neurons from patients with BD, a condition characterized by dysregulated WNT/β-catenin signaling in a subgroup of
affected individuals.
(Continued on next page)
* Correspondence: Deyou.Zheng@einstein.yu.edu;
Herb.Lachman@einstein.yu.edu
†
Equal contributors
1
Department of Genetics, Albert Einstein College of Medicine, 1300 Morris
Park Ave, Bronx, NY, USA
Full list of author information is available at the end of the article
© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Wang et al. Molecular Autism (2017) 8:11
DOI 10.1186/s13229-017-0124-1
(Continued from previous page)
Conclusions: Overall, the findings show that distinct ASD, SZ, and BD candidate genes converge on common molecular
targets—an important consideration for developing novel therapeutics in genetically heterogeneous complex traits.
Keywords: DLX6-AS1, Distal-less homeobox, Gabaergic, Cancer, Autism, Schizophrenia, Bipolar disorder, TCF4, HMGA2,
ZNF132, Wnt, Beta-catenin
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 muta-
tions in the gene have also been found in schizophrenia
(SZ) and intellectual disabilities [4–6]. CHD8 is a ubiqui-
tously 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 inter-
acting 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 can-
cer cells appears to be due, in part, to an interaction with
p53 [24]. CHD8 also recruits MLL histone methyltransfer-
ase 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 remo-
delers regulate embryonic stem cell transcriptional pro-
grams 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 differ-
ent 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 ex-
ample, the SZ and BD candidate gene MIR137 has been
found to target other candidates : CSMD1, C10orf26,
CACNA1C,andTCF4 [34]. In addition, clinically dis -
tinct disorders can be caused by the same risk genes ,
suggesting that therapies aimed at specific molecular
targets could have a therapeutic effe ct across diagnos-
tic categories [4, 35].
The molecular studies that have targeted CHD8 cer-
tainly support the con cept of converging molecular tar-
gets and pathways. In shRN A 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 candi-
date genes were found to be differentially expressed in
CHD8
+/−
NPCs and neurons, compared with isogenic
controls [8], and furthermore, DEGs were found to over-
lap 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 sign aling, and
WNT/β-catenin and PTEN signaling. In addition, genes
involved in head circumference were found to be differ-
entially expressed. This is notable because loss of func-
tion 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 neur-
onal differentiation methods have emerged that are more
suitable for SZ and ASD, one of which is the direct con-
version of iPS cells into 3-dimensional cerebral orga-
noids, 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
Wang et al. Molecular Autism (2017) 8:11 Page 2 of 17
demonstrated that the organoid system is ideal for study-
ing gene × environment interactions relevant to neuro-
psychiatric and neurodevelopmental disorders [39].
The few studies that have been carried out so far using
cerebral organoids as a model system have been reveal-
ing. Mariani et al., for example, showed that genes
involved in cell proliferation, neuronal differentiation,
synaptic assembly, and GABAergic inhibitory neuron de-
velopment were differentially expressed in an idiopathic
ASD family [42]. And, using a somewhat differ ent orga-
noid differ entiation 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 de-
rived 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 cul-
tures. In particular, we show that CHD8 haploinsuffi-
ciency 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 idio-
pathic ASD in whom the responsible gen etic 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 unchara cterized 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 schizoaf-
fective 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 Medi-
cine (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 ob-
tained from skin biopsies performed by board-certified
physicians. The procedure for growing fibroblast s 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 wer e 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. Follow-
ing the Stem Cell Technologies protocol, 3.0 × 10
6
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 cham-
ber slide coated with 10 μg/ml polyornithine, 2.5 μg/ml
laminin, and 50 μg/ml fibronectin, and cultured in Neu-
robasal 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 mix-
ture 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 immuno-
histochemistry (IHC), samples were fixed with 4% para-
formaldehyde 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_gal ore/). We employed
Kallisto (v0.42.5) [46] to determine the read count for each
transcript and quantified transcript abundance as tran-
scripts 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 al-
ternative splicing transcripts of a gene to obtain gene ex-
pression 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 iden-
tify DEGs (false discovery rate (FDR) <0.05). The software
DAVID (v6.8 Beta) [49, 50] was used for Gene Ontology
Wang et al. Molecular Autism (2017) 8:11 Page 3 of 17
(GO) analysis, with the 12,898 expressed genes as back-
ground. 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 de-
tailed 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 con fidence, to min-
imal 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]; candi-
date 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) re-
port (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 lat-
ter 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” re cord. 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 signifi-
cantly enriched with a specific gene set, 12,893
expressed genes in our samples were used as back-
ground for Fisher’s exact test. Statistics tests wer e con-
ducted in R (http://www.R-project.org/). Common genes
between two gene lists were input to DAVID (beta 6.8)
for G O 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 dele-
tion (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 immunohis-
tochemistry showed a 54% decrease in CHD8 im mu-
noreactivity 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” sec tion. Using a cutoff of
FDR < 0.05 , there were 55 9 DEGs when the CHD8
+/+
organoids were compared with CHD8
+/−
; 288 genes
increased in the KO, 271 decrea sed. The DEGs sepa-
rated our sample into two groups , as seen in the heat
map shown in Fig. 1a. The entire list o f DEGs is in
Additional file 4: Table S2. CHD8 mRNA it self 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 rela-
tively imprecise correlation between mRNA and protein
levels is found for many genes and can be due to a num-
ber of factors [60]. However, the mechanism of the dis-
crepancy between CHD8 mRNA and protein is not
known and will require further investigation.
We should also point out that among the three CHD8
alternatively spliced transcripts in the GENCODE anno-
tation, 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 car-
ried 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
Wang et al. Molecular Autism (2017) 8:11 Page 4 of 17
CHD8, most likely through the actions of other genes cod-
ing for transcription factors and chromatin-remodeling
proteins that are directly affected by CHD8, such as TCF4,
POU3F2, SMARCA4, SOX2,andPAX6. This result is con-
sistent with our previous report [8].
The software DAVID was used to identify enriched GO
pathways in DEGs using 12,893 expressed genes as back -
ground [49]. IP A was used for canonical pathways and dis-
ease association. The top GO terms (Biological Process,
(BP))forgenesthatwereupregulatedintheCHD8 KO
organoids were nervous system development, neurogenesis,
neuron differentiation ,and forebrain development; the top
GO:BP terms for downregulated genes were nervous sys-
tem development, generation of neurons, and neuron dif-
ferentiation (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]. Th e
top enriched IP A 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], a s well a s findings
by othe r investigators [10, 20, 24], firmly establishing
that altered e xpression of CHD8 disr upts 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 upregu-
lated and downregulated genes (Fig. 1).
Overall, the findings show that CHD8 directly, or in-
directly through effects on other transcription factors
and chromatin regulators, regulates a program of gene
expression that affects critical aspects of brain develop-
ment (e.g., neurogenesis, neuron differentiation, and
axonal guidance).
Comparison between organoid data and NPCs and
monolayer neurons
We compared current transcriptome data with our pre-
vious 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, re-
spectively, which is consistent with the main pathway
findings in organoids described above.
Organoids
Neurons
NPCs
409
1875
369
291
49
139
80
Organoids vs NPCs
OR = 4.44, p < 2.2e-16
Top GO term:
Neuron differentiation
(35 genes of 129 genes)
Organoids vs Neurons
OR = 2.88, p < 2.2e-16
Top GO term:
Neurogenesis
(74 genes of 219 genes)
Fig. 2 Overlapping DEGs in organoids compared with NPCs and
monolayer neurons from previous study [8]
ab c
Fig. 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]
Wang et al. Molecular Autism (2017) 8:11 Page 5 of 17