In Vitro Modeling of the Bipolar Disorder and Schizophrenia Using Patient-Derived Induced Pluripotent Stem Cells with Copy Number Variations of PCDH15 and RELN

Visual Abstract


Introduction
Both bipolar disorder (BP) and schizophrenia (SCZ) are chronic and debilitating psychiatric disorders that affect ϳ1% of the worldwide population (McGrath et al., 2008;Grande et al., 2016). Although these disorders are highly heritable (Craddock and Sklar, 2013;Millan et al., 2016), the molecular mechanisms underlying the complex pathology of these disorders remain to be elucidated.
There are limitations to the recapitulation of clinical characteristics in animal models and postmortem brain studies because of genetic heterogeneity (O'Shea and McInnis, 2016;Prytkova and Brennand, 2017). Therefore, reliable models that functionally mimic live human brains are sought after. Induced pluripotent stem cells (iPSCs) are expected to become a promising tool for recapitulating disease-specific phenotypes in vitro (Okano and Yamanaka, 2014;O'Shea and McInnis, 2016;Watmuff et al., 2016;Prytkova and Brennand, 2017;Tobe et al., 2017). Although recent studies established iPSCs from BP and SCZ patients and induced neurons to analyze phenotypes (O'Shea and McInnis, 2016;Prytkova and Brennand, 2017;Wen, 2017), the maturity and subtype specificity of induced neurons remain to be considered. Thus, analysis of mature and subtype-specific neurons is required for further elucidation of the pathologies. It has been suggested that the collapse of the excitation-inhibition (E/I) balance plays key roles in BP and SCZ (Gao and Penzes, 2015;Lee et al., 2018). Therefore, it is important to focus on certain neurons that are the main players in the E/I balance, such as glutamatergic neurons and GABAergic neurons. Recent studies have shown that transcription factor overexpression enabled iPSCs to be differentiated into specific neurons, including glutamatergic neurons (Zhang et al., 2013) and GABAergic neurons (Colasante et al., 2015;Yang et al., 2017).
Many genetic mutations are associated with these disorders, especially copy number variations (CNVs), which are important contributive factors that affect the onset and treatment resistance of BP and SCZ (Georgieva et al., 2014;Green et al., 2016;Kushima et al., 2017). Thus, to analyze the pathologies, we used iPSC lines derived from patients who carried certain CNVs: two BP patients with PCDH15 exonic deletions and an SCZ patient who carried an RELN exonic deletion. Protocadherin 15 (PCDH15), encoded by PCDH15, is a member of the cadherin superfamily. PCDH15 mutations cause Usher syndrome, which results in hearing vision loss (Ahmed et al., 2001;Alagramam et al., 2001;Kim et al., 2011). A recent genome-wide association study suggested that PCDH15 is associated with psychiatric disorders (Lo et al., 2017). In addition, de novo or rare exonic deletions in PCDH15 were identified in BP patients (Georgieva et al., 2014;Noor et al., 2014). These studies suggested that PCDH15 is a risk gene for psychiatric disorders. Reelin, which is encoded by RELN, is an extracellular matrix protein that regulates brain developmental processes, such as neuronal migration and dendrite formation, and modulates synaptic functions in adult brains (Ishii et al., 2016;Lee and D'Arcangelo, 2016). Reelin is associated with various psychiatric disorders such as SCZ, autism spectrum disorders, and BP (Folsom and Fatemi, 2013;Ishii et al., 2016). Two SCZ patients with exonic deletions in RELN have been reported in previous studies (Costain et al., 2013;Kushima et al., 2017).
In this study, to recapitulate the pathologies in BP and SCZ in vitro, we differentiated patient-derived iPSCs into neurons and identified a set of similar phenotypes among induced neurons derived from patients with different genetic lesions, which potentially represents a cell biological basis for shared clinical features between these two different disorders.

Subjects
A human subject was recruited at Keio University, and another from Kyushu University. Written informed consent for the present study was obtained from this patient. All the experimental procedures for biopsy, genetic analysis, and iPSC production were approved by the Keio University School of Medicine Ethics Committee (approval #20080016), the Ethics Committee of Nagoya University (approval #2010-1033-3 and #2012-0184-14), and the Ethics Committee of Kyushu University (approval #IRB-30-537).
We recruited two patients with BP. The first patient (BP1) was a 43-year-old Japanese female with a family history of alcohol use disorder. Her developmental history was unremarkable. In her early thirties, she had depressive moods and suicidal thoughts. Although she was treated with antidepressant medications, a significant treatment effect was not observed. After 3 years, she received a diagnosis of bipolar I disorder because she had her first manic episode, which included an elevated mood, increased activity and talkativeness, and decreased need for sleep. In her late thirties, she became depressed again after she received a diagnosis of a malignant tumor. Her depressive symptoms were chronic and refractory to pharmacological treatments, requiring multiple hospitalizations. At the time of the study evaluation, she was receiving treatment with lithium and quetiapine.
The second patient (BP2) was a 57-year-old Japanese female with no remarkable developmental history and no family history of psychiatric disorders. She had been committed to various social activities and was sometimes hyperactive until her middle thirties (hypomanic episodes). Since her late thirties, she had experienced insomnia, depressive moods, and severe suicidal thoughts two or three times a year, requiring multiple hospitalizations. Despite treatment with antidepressant medications, a considerable treatment effect was not observed. After she had received a diagnosis of bipolar I disorder based on her life history and hyperactivity between depressive episodes, lithium was prescribed. However, the response was still poor. At age 50 years, lithium was completely replaced with valproate and lamotrigine. After that, her mood stabilized and no subsequent hospitalizations were necessary. At the time of the study evaluation, she was still receiving valproate and lamotrigine. Although she had once received a diagnosis of bipolar I disorder, the diagnosis was revised to bipolar II disorder because of the lack of severe manic episodes.

Identification and validation of an exonic deletion of PCDH15
Genomic DNA was extracted from blood or iPSC lines. To identify an exonic deletion of PCDH15, we performed array comparative genomic hybridization (aCGH) using an SurePrint G3 Human CGH Microarray, 2 ϫ 400K (Agilent). The experiment was performed following the manufacturer instructions. CNV calls were made with Nexus Copy Number software version 9.0 (BioDiscovery) using the Fast Adaptive States Segmentation Technique 2 algorithm, which is a hidden Markov model-based approach. The log2 ratio thresholds for copy number loss and copy number gain were set at Ϫ0.7 and 0.45, respectively. The significance threshold p value was set at 1 ϫ 10 Ϫ6 , and at least four contiguous probes were required for CNV calls. To validate the exonic deletion of PCDH15, quantitative real-time PCR was performed with TaqMan copy number assay (Hs03733966_cn for BP1 and Hs00117569_cn for BP2; Applied Biosystems). All genomic locations are given in NCBI36/hg18.

In vitro three-germ differentiation via embryoid body formation
To check the pluripotency of iPSCs, iPSCs treated with TrypLE Select (Thermo Fisher Scientific) were dissociated into single cells and plated in low-cell adhesion 96-well plates with V-bottomed conical wells. The cells were cultured in embryoid body (EB) medium (DMEM/F-12 containing 5% KSR, 2 mM L-glutamine, 1% NEAAs, and 0.1 mM 2-ME). On day 7, EBs were plated on culture plates coated with 0.1% gelatin (Merck) and 10 g/ml fibronectin (Merck). The plated EBs were cultured for 3 weeks at 37°C in an atmosphere containing 5% CO 2 .

High-content image analysis
For morphologic analysis, 96-well plates with V-bottomed conical wells containing 1 EB in each well were imaged on an IN Cell Analyzer 6000 using the 2ϫ objective. EBs were identified based on the contrast with the background; from the EB images obtained, the area and form factor were calculated using IN Cell Developer Toolbox version 1.9 (GE Healthcare).
For fluorescence intensity analysis and neurite length analysis of the EBs, stained plates were imaged on an IN Cell Analyzer 6000 high-content cellular analysis system; only the field containing the EB was selectively collected from each well using the 2ϫ objective, resulting in 1 EB being scored per well. For cell population assays and dendrite length analysis, stained plates were imaged with an IN Cell Analyzer 6000; a set of 5 ϫ 5 fields was collected from each well using the 20ϫ objective. For quantitative analysis of the synaptic puncta, stained plates were imaged with an IN Cell Analyzer 6000; a set of 6 ϫ 6 fields was collected from each well using the 60ϫ objective.
Analysis using IN Cell Developer Toolbox version 1.9 (GE Healthcare) began by identifying intact nuclei stained by the Hoechst dye, which were defined as traced nuclei that were Ͼ50 m 2 in surface area and with intensity levels that were typical and lower than the threshold brightness of pyknotic cells. Each traced nuclear region was then expanded by 50% and cross-referenced with an endodermal marker (AFP), a mesodermal marker (␣SMA), and an ectodermal marker (␤III tubulin) for identification; from these images, the fluorescence intensity of each marker was calculated. Using the expanded nuclear region images, neuron markers (␤III tubulin and MAP2), a glutamatergic neuron marker (VGluT2), and a GABAergic neuron marker (GABA) were identified; from these images, the percentage of each marker was calculated (for neural subtype markers stained as granules, the number of ␤III tubulin ϩ cells containing one or more of the markers was quantified). Using the above-described traced neuronal images of each cell, neurite length, dendrite length, and the number of synaptic puncta (Homer I ϩ , Gephyrin ϩ , and/or Synapsin I ϩ puncta) were also analyzed.

Microarray analysis
RNA was extracted with the RNeasy Kit (QIAGEN), and the RNA quality and quantity were assessed using a bioanalyzer (model 2100, Agilent Technologies). Microar-ray analysis was performed using the Clariom S Assay for humans (Thermo Fisher Scientific). The accession number was referred later. Data were analyzed by Transcriptome Analysis Console 4.0 (Thermo Fisher Scientific). Gene ontology (GO) and pathway analyses were performed using DAVID 6.8 Bioinformatics Resources (https://david. ncifcrf.gov). The data shown in this publication have been deposited in the NCBI Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/). The accession number of our microarray analysis data was GSE116820.

Genetic engineering of PCDH15-mutant iPSCs
PCDH15-deleted iPSCs were established using the CRISPR/Cas9 system, as previously described . In brief, single-guide RNA (sgRNA) expression vectors were constructed using pHL-H1-ccdB-mEF1␣-RiH vector (catalog #60601, Addgene) with two oligos containing the sgRNA target site and a universal reverse primer (Table 1). The T7 endonuclease I (T7E1) assay was performed using genomic DNA from HEK293FT cells co-transfected with the sgRNA expression vector and Cas9 expression vector (Addgene ID: 60599) to assess sgRNA activity. The target region of sgRNA was amplified from the genomic DNA using a specific primer set (Table 1). The PCR products were digested using T7E1 (New England Biolabs) after denaturing and reannealing. To establish the isogenic PCDH15-deleted iPSC line, sgRNA and Cas9 expression vectors were cotransfected into healthy control iPSCs, NC1032-1-2 (Control 1) and cells were selected with puromycin.

Microelectrode array analysis
The microelectrode array (MEA) was assayed using the Maestro system (Axion Biosystems). Neuronal inductions from iPSCs were performed in 48-well MEA plates coated with poly-L-lysine, iMatrix-511, and laminin. For neuronal differentiation, iPSCs were cultured in induction medium for the first 5 d and then in function assay medium from day 5 until day 28 or day 42. Data were acquired using a sampling rate of 12.5 kHz and filtered using a 200 -3000 Hz Butterworth bandpass filter. The detection threshold was set to ϩ6.0 ϫ SD of the baseline electrode noise. Five minutes of activity was subsequently recorded at 37°C. Given the considerable variation in spike count between wells and plates, we focused on active electrodes (electrodes with an average of Ͼ5 spikes/min; Isoda et al., 2016), which were certain to make contact with neurons. The number of active electrodes and the mean spike rate per active electrode (spikes per second) was calculated using the spike count file generated by the Axion Integrated Studio program (Axion Biosystems). The spike raster plot was generated by Neural Metric Tool (Axion Biosystems). To determine the responses to the agonist or antagonists of glutamate or the GABA receptor, 50 M CNQX (Tocris Bioscience), 50 M AP-5 (Alomone Labs), or 10 M GABA (nacalai tesque) was applied as treatment, and the activity was recorded.

Calcium imaging
Neurons were induced from iPSCs by culture in induction medium for 5 d followed by function assay medium from day 5. Calcium imaging analyses were performed on days 28 -30. Cells were loaded with 1 g/ml fluo-8 AM (AAT Bioquest) in recording medium (20 mM HEPES, 115 mM NaCl, 5.4 mM KCl, 0.8 mM MgCl 2 , 1.8 mM CaCl 2 , and 13.8 mM glucose; Dojindo) containing 0.02% Cremophor EL (Dojindo) and incubated for 20 min at 37°C and 5% CO 2 . After washing with PBS, the medium was changed to the recording medium. Changes in fluorescence intensity were measured using an IX83 inverted microscope (Olympus) equipped with an Electron Multiplying CCD Camera (Hamamatsu Photonics) and LED illumination system pE-4000 (CoolLED). We recorded 3500 frames (1 frame: 31-32 ms) per well using the stream acquisition mode. MetaMorph Image Analysis Software (Molecular Devices) was used to analyze the live cell calcium traces. Regions of interest (ROIs) were drawn on cells based on time projection images of the recordings. ROI traces of the time course of changes in fluorescence intensity were generated and used as substrates for subsequent analyses. To adjust for photobleaching, the difference in intensity between the first frame and last frame was calculated and subtracted from the raw intensity. The change in fluorescence intensity over time was normalized as ⌬F/F ϭ (F Ϫ F 0 )/F 0 , where F 0 is fluorescence at the starting point of exposure (time 0). ⌬Fmax was defined as the difference of the largest change of ⌬F/F associated with each calcium spike.

Statistical analysis
Statistical analyses were performed using a Dunnett's or Tukey's test for multiple comparisons. For comparison between two groups, an unpaired Student's t test or a paired t test was used. All statistical analyses were performed using SAS 9.2 (SAS Institute). Probability values (p value) Ͻ0.05 were considered to be statistically significant. Statistical test results are included in Table 2.

Results
Generation of iPSCs from BP patients BP and SCZ are distinct neuropsychiatric diseases but share a subset of similar pathologies and symptoms (Maggioni et al., 2017). To recapitulate the shared BP and SCZ pathologies in vitro, we prepared iPSCs derived from BP and SCZ patients to induce neurons (Fig. 1A). We focused on two BP patients (BP1 and BP2), who were evaluated in a high-resolution CNV analysis study (our unpublished observation). The patient characteristics are listed in Figure 1B (see also Materials and Methods). Using aCGH, we identified an exonic deletion of PCDH15 in each patient with BP (Fig. 1C). The coordinate of this deletion of BP1 was chr10:56,149,683-56,924,932, and the deletion of BP2 was chr10:54,162,953-55,694,398. The deletion of BP1 included exon 1 of most of the PCDH15 transcripts, and the deletion of BP2 included PCDH15 exon 9 to the last exon and MBL2 (mannose binding lectin 2). These PCDH15 deletions were validated by the TaqMan copy number assay (Fig. 1D). We generated two iPSC clones (BP1-1, BP1-2) from BP1 and one iPSC clone (BP2-1) using episomal plasmids (see Materials and Methods), and we detected the same exonic deletions of PCDH15 as those detected in blood cells (Fig.  1C,D). Immunocytochemical analysis demonstrated that our newly established BP-iPSCs expressed the pluripotent markers Oct4, Nanog, SSEA4, and Tra1-60 (Fig. 1E).
In vitro three-germ layer differentiation analysis via EBs also showed the differentiation potentials of the iPSCs into three germ layers (Fig. 1F). These results suggest that the BP-iPSCs conserved the patient's genetic background and had adequate pluripotency.

Ectoderm-enriched embryoid body and neuron formation by chemical treatment
There are two major methods for neural differentiation from iPSCs or embryonic stem cells (ESCs). One is the chemical induction method, in which iPSCs are treated with several pharmacological agents or bioactive proteins for differentiation into neurons via EB or neurosphere formation (Han et al., 2011;Espuny-Camacho et al., 2013;Matsumoto et al., 2016). The other is the direct conversion method, involving the overexpression of specific transcription factors (Zhang et al., 2013;Colasante et al., 2015;Sun et al., 2016;Yang et al., 2017). To select the best method for recapitulating the disease-specific mature neuron phenotypes, we used both methods to differentiate iPSCs into neurons (Fig. 1A).
First, we performed chemical treatment. In previous studies, SCZ or BP patient-derived iPSCs exhibited ab-normal phenotypes of early-stage neural differentiation when this type of method was used (Madison et al., 2015;Toyoshima et al., 2016). We used BP-iPSCs (BP1-1, BP1-2) and SCZ-iPSCs (SCZ1-1, SCZ1-2) as patientderived iPSCs. To confirm whether such phenotypes occurred in our iPSC lines, neurons were induced via EB formation. First, we checked the potential of the two types of EB formation. Normal EBs were formed in EB medium ( Fig. 2A). Another type of EB is induced by dual SMAD inhibition to facilitate neural differentiation (Chambers et al., 2009). DSi-EBs were induced in MHM with two DSis, namely, SB431542 and LDN193189 (Imaizumi et al., 2015;Matsumoto et al., 2016). The Wnt inhibitor IWP-2 was also used for DSi-EB formation to induce anterior regionality of the brain ( Fig. 2A; Imaizumi et al., 2015;Nehme et al., 2018). Six clones of iPSCs (1210B2, 201B7, eKA3, eTKA4, WD39, and 414C2) were used as healthy control iPSCs. Each group of iPSCs formed EBs by both methods (Fig. 2B). To assess EB development, we examined morphometric parameters including size and form factor. The patient-derived iPSCs tended to form smaller and more distorted EBs than the control iPSCs, although there were no significant differences in size and form factor among EBs derived from each group of iPSC lines (Fig. 2C,D). In addition, the PCDH15 and RELN gene expression levels of DSi-EBs were analyzed by qRT-PCR. The expression of both genes did not change significantly between the control and BP or SCZ, although PCDH15 tended to exhibit low expression in BP samples and RELN exhibited a similar trend in SCZ samples (Fig. 2E).
To investigate whether neurons from the patientderived iPSCs exhibited aberrant phenotypes, we induced neural differentiation via DSi-EB formation. DSi-EBs were plated on culture plates on day 7. Plated DSi-EBs were cultured for up to an additional 15 d in MHM with DSi and IWP-2 (Fig. 2F). We used 1210B2 and 201B7 as control iPSCs, and BP-iPSCs (BP1-1, BP1-2) and SCZ-iPSCs (SCZ1-1, SCZ1-2) were used as patientderived iPSCs. Immunocytochemical analysis showed that the expression levels of markers of the three germ layers, namely, ␤III tubulin (ectoderm), ␣SMA (mesoderm), and AFP (endoderm), in the patient-derived cells did not differ significantly from those in the control cells (Fig. 2G). Thus, the capacity of differentiation into the three germ layers among the control and both patient-derived iPSCs was likely to be the same. We then measured the length of neurite extension from the cell clusters. To visualize the neurites, cells were immunostained for ␤III tubulin. Binarized images were formed from stained images to quantify the total fiber length of each cell cluster (a) and fiber numbers of each cell cluster (b) of the ␤III tubulin ϩ cells (Fig. 2H). Average neurite length was defined as (a)/(b). The neurite lengths of patient-derived cells were significantly shortened from day 13 onward (Fig. 2I,J). We found that this method was not suitable for the analysis of mature neurons because of early-stage neuron abnormalities, although these phenotypes may reflect developmental abnormalities of the pathologies of BP and SCZ. Thus, we decided not to perform further analysis using this method.

Neuronal differentiation of iPSCs by overexpression of transcription factors
We next tested the direct method for neuronal conversion of iPSCs by transcription factor overexpression. This method can induce specific neuron subtypes. Previous studies have shown that Neurog2 overexpression induced glutamatergic neuron differentiation (Zhang et al., 2013), and co-overexpression of ASCL1 and DLX2 induced GABAergic neuron differentiation (Yang et al., 2017). Although these studies used lentivirus as a tool for introducing transcription factors, use of the PiggyBac transposon vector system enabled the establishment of transgenic cells that can be easily stored and used for experiments. One report using the PiggyBac system for the introduction of transcription factors into human ESCs resulted in highly efficient neural differentiation (Matsushita et al., 2017). Therefore, we established transgenic iPSC lines in which transcription factors can be induced by Dox. We used 1210B2 and 201B7 as healthy control iPSCs, while we used BP-iPSCs (BP1-1, BP1-2, and BP2-1) and SCZ-iPSCs (SCZ1-1, SCZ1-2) as patientderived iPSCs. NEUROG2 expression vector-transduced iPSCs were established for glutamatergic neuronal induction, and Ascl1-and DLX2-transduced iPSCs were established for GABAergic neuronal induction (see Materials and Methods). The iPSCs were treated with Dox for 5 d and cultured in MHM for 28 d (Fig. 3A,E). Immunocytochemical analysis showed that the neuronal marker ␤III tubulin was expressed in most of the cells induced from control iPSCs (control neurons), BP-iPSCs (BP neurons), Figure 1. Generation and characterization of iPSCs derived from a BP patient. A, Schematic diagram of the strategy to explore the phenotypes of BP and SCZ in vitro. B, Subjects list containing basic information. C, CNVs in chromosome 10 were detected in blood and iPSCs derived from two BP patients by aCGH. D, The exonic deletions of PCDH15 identified in this study were validated by the TaqMan copy number assays. Bars indicate copy numbers predicted by the TaqMan copy number assays. Capped bars indicated the minimum and maximum copy number calculated for the sample replicate group (n ϭ 4). Controls carried no aCGH-detected CNVs of PCDH15 (copy number ϭ 2) and were used to calibrate the assays. E, The generated iPSCs expressed the pluripotent markers Oct4, Nanog, SSEA4, and Tra1-60. Scale bar, 100 m. F, Representative images of immunocytochemical analysis for in vitro three-germ layer differentiation. ␤III-Tubulin, ␣SMA, and AFP are pluripotent markers of the ectoderm, mesoderm, and endoderm, respectively. Blue indicates Ho, and green indicates the pluripotent marker. Scale bar, 100 m. Values were normalized to that of the control, which was considered to be 1.0. One sample of 1210B2-derived DSi-EBs, in which PCDH15 expression was under the detection limit, was removed from the analysis. F, Overview of the protocol for neuron differentiation via EB formation. DSi represents SB431542 and LDN193189. G, Intensity levels of three germ layer markers, namely, ␤III-tubulin, ␣SMA, and AFP. Intensity levels were normalized to that of the control, which was considered to be 1.0 (n ϭ 3 independent experiments; mean Ϯ SD; Dunnett's test among each group, no significant differences were observed). and SCZ-iPSCs (SCZ neurons; Fig. 3B,C,F,G). The conversion efficiency of iPSCs into ␤III tubulin ϩ neurons was close to 100% in all lines, except line BP2-1 which still showed Ͼ80% efficiency (Fig. 3C). ␤III tubulin ϩ neurons derived from NEUROG2-transduced iPSCs expressed the mature neuronal marker MAP2 and the glutamatergic neuronal marker VGLUT2 at similar ratios in each line (Fig.  3B,C). Next, the PCDH15 and RELN gene expression levels of NEUROG2-transduced cells (iPSCs and induced glutamatergic neurons) were analyzed by qRT-PCR. We used the primer sets targeting the deletion of PCDH15 in BP1 or RELN in SCZ1. Both PCDH15 and RELN were substantially upregulated on neuronal induction but were not differentially expressed between control, BP, SCZ iPSCs, or neurons (Fig. 3D). Induced neurons derived from ASCL1-and DLX2-transduced iPSCs expressed MAP2 and the GABAergic neuronal marker GABA at similar ratios in each line (Fig. 3F,G). The PCDH15 and RELN gene expression levels of ASCL1-and DLX2-cotransduced cells (iPSCs and induced GABAergic neurons) are shown in Figure 3H. The expression of both genes was upregulated in GABAergic neurons compared with undifferentiated iPSCs. Neither PCDH15 nor RELN expression changed significantly in patient-derived cells. These results showed that this method enabled both healthy and patient-derived iPSCs to efficiently differentiate into subtype-specific and mature neurons. Thus, we chose this method for further analyses.

Characterization of gene expression in neurons induced from iPSCs
To identify the characteristics of neurons induced from patient-derived iPSCs, the global gene expression profiles of neurons induced from control-iPSCs (1210B2 and 201B7) and patient-derived iPSCs (BP1-1, BP1-2, SCZ1-1, and SCZ1-2) were examined by a microarray analysis. Principal component analysis (PCA ; Fig. 4A,E) and hierarchical analysis (Fig. 4B,F) showed that the control, BP, and SCZ neurons were placed in different groups. In particular, the profiles of the BP and SCZ neurons were closer to each other than to that of the control neurons (Fig. 4B,F). We next identified the differentially expressed genes (DEGs) in neurons induced from patient-derived iPSCs compared with control neurons (fold change, Ͼ2.0). We identified 1812 genes of BP glutamatergic neurons and 2382 genes of SCZ glutamatergic neurons that were upregulated or downregulated compared with the control neurons (Fig. 4C). For the GABAergic neurons, the expression levels of 1703 genes of BP neurons and 2357 genes of SCZ neurons were changed (Fig. 4G). To explore the cellular functions associated with the DEGs in the BP and SCZ neurons, GO analysis and pathway analysis were performed. Featured GO terms and pathways of common DEGs between BP and SCZ are shown in Figure 4, D and H. In both glutamatergic and GABAergic neurons, multiple GO terms and pathways related to cell adhesion and neural function were enriched compared with control neurons (Fig. 4D,H).

Neurons induced from patient-derived iPSCs exhibited shorter dendrites and reduced formation of excitatory and inhibitory synapses
Microarray and GO analyses suggested that genes involved in cell adhesion may contribute to pathogenetic mechanisms (Fig. 4). Cell adhesion is a key event for neurite extension and synapse formation (Kiryushko et al., 2004;Missler et al., 2012). In addition, previous studies using postmortem human brains showed reduced spine density and dendrite length in the brains of BP and SCZ patients (Glantz and Lewis, 2000;Konopaske et al., 2014). To investigate whether BP and SCZ neurons have similar phenotypes, we determined the dendrite lengths and synaptic densities of induced neurons by immunocytochemical analysis with an IN Cell Analyzer 6000. Dendrite length was measured as the length of the region immunostained with the dendrite marker MAP2 (Fig. 5A). Then, to detect the number of morphological synapses in glutamatergic neurons, we counted the puncta of Synapsin I and Homer I as presynaptic or postsynaptic markers. For GABAergic neurons, we determined the number of puncta positive for Synapsin I and Gephyrin, which is a postsynaptic scaffolding protein found in GABAergic synapses (Fig. 5A). MAP2 ϩ dendrites of BP and SCZ neurons were significantly shorter than those of control neurons in both glutamatergic and GABAergic neurons (Fig. 5B,D,F,H). In addition, fewer puncta of Homer I and Synapsin I on glutamatergic neurons were detected in BP and SCZ neurons than in control neurons (Fig. 5C,E). The number of puncta of GABAergic Synapsin I and Gephyrin were also reduced in BP and SCZ neurons compared with control neurons (Fig. 5G,I). Moreover, the number of colocalized puncta between presynaptic and postsynaptic markers was also significantly decreased in BP and SCZ neurons (Fig. 5E,I). Thus, the number of synapses decreased in both BP and SCZ neurons. These results suggest that neurons induced from patient-derived iPSCs exhibit synapse-and dendrite-related phenotypes that are consistent with the phenotypes often observed in patients with psychiatric disorders and can thus serve as disease models of psychiatric disorders in vitro.

Establishment of isogenic PCDH15-deleted iPSCs
It is difficult to perform large-scale analyses because of the infrequency of patients carrying particular CNVs. Moreover, line-to-line variability caused by different genetic backgrounds complicate genotype-phenotype correlation. Thus, we sought to establish mutant and control iPSCs by targeted genome editing using the CRISPR/ continued H, Schematic diagram of the analysis protocol. Fiber length and number of ␤III tubulin ϩ cells were quantified from binarized image data obtained from stained images. I, Representative images of ␤III-tubulin ϩ neurons. Scale bar, 300 m. J, Time-dependent changes of ␤III-tubulin ϩ mean neurite length per neurite fiber (n ϭ 3 independent experiments; mean Ϯ SD; ‫ء‬p Ͻ 0.05, ‫‪p‬ءء‬ Ͻ 0.01; Dunnett's test among each group). Mean neurite length is shown as the mean length of ␤III-tubulin ϩ neurite fiber. Figure 3. Neuron differentiation by transcription factor overexpression. A, Overview of the protocol for differentiation into glutamatergic neurons. B, Immunocytochemical analysis of neuronal markers (VGluT2, MAP2, and ␤III tubulin). Scale bar, 40 m. C, Ratio of positive cells for each marker; ␤III tubulin ϩ cells/all cells (␤III tubulin ϩ /Ho ϩ ), MAP2 ϩ cells/␤III tubulin ϩ cells (MAP2 ϩ / ␤III-tubulin ϩ ), and VGluT2 and ␤III tubulin double ϩ cells/␤III tubulin ϩ cells (VGLUT2 ϩ ␤III tubulin ϩ /␤III tubulin ϩ ). (n ϭ 3-6 independent experiments; mean Ϯ SD; ‫ء‬p Ͻ 0.05; Tukey's test, the population of ␤III tubulin ϩ cells in BP2-1-derived neurons was significantly lower than those in 1210B2-, 201B7-, BP1-1-, and BP1-2-derived neurons). D, Relative gene expression levels of PCDH15 (primer set1) and RELN in NEUROG2-transduced iPSCs and induced neurons (n ϭ 3 independent experiments; mean Ϯ SD; Dunnett's test among iPSCs and among neurons, no significant differences were observed). Values were normalized to that for 1210B2-derived neurons, which was considered to be 1.0. Some iPSC samples, in which PCDH15 expression was under the detection limit, were removed from the analysis. Specifically, two samples of BP1-1, and one sample of 1210B2, 201B7, BP2-1, and SCZ1-1 were excluded from the analysis. E, Overview of the protocol for differentiation into GABAergic neurons. F, Immunocytochemical analysis of neuronal markers (MAP2, GABA, and ␤III tubulin). Scale bar, 40 m. G, Ratio of positive cells for each marker; ␤III tubulin ϩ cells/all cells (␤III tubulin ϩ /Ho ϩ ), MAP2 ϩ cells/␤III tubulin ϩ cells (MAP2 ϩ /␤III tubulin ϩ ), and GABA and ␤III-tubulin double-positive cells/␤III tubulin ϩ cells (GABA ϩ ␤III tubulin ϩ /␤III tubulin ϩ ). (n ϭ 3-6 independent experiments; mean Ϯ SD; Tukey's test; no significant differences were observed). H, Relative gene expression levels of PCDH15 and RELN in ASCL1-and DLX2-transduced iPSCs and induced neurons. (n ϭ 3 independent experiments; mean Ϯ SD; Dunnett's test among iPSCs and among neurons; no significant differences were observed). Values were normalized to that for 1210B2-derived neurons, which was considered to be 1.0. Some iPSC samples, in which PCDH15 expression was under the detection limit, were removed from the analysis. Specifically, each one sample of 1210B2 and BP2-1 was excluded from the analysis. Cas9 system. To generate PCDH15-deleted iPSCs, sgRNAs were constructed in exon 9 (Fig. 6A). According to the results of the T7E1 assay (see Materials and Methods), sgRNA#3 showed the strongest cleavage activity among the five sgRNA candidates (Fig. 6B). We obtained an isogenic line with the homozygous PCDH15 deletion (PCDH15del) from healthy control NC1032-1-2 (Control 1; Fig. 6C). PCDH15del mutant iPSCs showed the capacity experiments; mean Ϯ SD; ‫‪p‬ءء‬ Ͻ 0.01; Dunnett's test). E, Quantitative analysis of the number of synaptic marker puncta in glutamatergic neurons (n ϭ 3-6 independent experiments; mean Ϯ SD; ‫ء‬p Ͻ 0.05, ‫‪p‬ءء‬ Ͻ 0.01; Dunnett's test). Homer I puncta, Synapsin I puncta and Homer-Synapsin I colocalized puncta on ␤III-tubulin ϩ cells were counted. F, G, Representative images of immunocytochemical analysis of a dendrite marker (MAP2; scale bar, 60 m; F) and synaptic markers (Gephyrin, Synapsin I; scale bar, 10 m; G) in GABAergic neurons. H, Quantitative analysis of dendrite length in GABAergic neurons (n ϭ 3-6 independent experiments; mean Ϯ SD; ‫‪p‬ءء‬ Ͻ 0.01, Dunnett's test). I, Quantitative analysis of the number of synaptic marker puncta in glutamatergic neurons (n ϭ 3-6 independent experiments; mean Ϯ SD; ‫ء‬p Ͻ 0.05, ‫‪p‬ءء‬ Ͻ 0.01, Dunnett's test). Gephyrin puncta, Synapsin I puncta, and Gephyrin-Synapsin I colocalized puncta on ␤III tubulin ϩ cells were counted. to differentiate into three germ layers as well as Control 1 (Fig. 6D). Then, we analyzed the expression level of PCDH15 by qRT-PCR using the primer set targeting the deleted region. The mRNA level of PCDH15 was decreased by about 50% in PCDH15del-induced neurons (PCDH15del neurons) when normalized to Control 1-induced neurons (Fig. 6E). We also prepared an isogenic line with the homozygous RELN deletion (RELNdel) that was established from 201B7 (Control 2) in a previous study Fig. 6F). Remarkably, the expression level of RELN in RELNdel-induced neurons (RELNdel neurons) was much lower than that in Control 2-induced neurons (Fig. 6G).

Dendritic and synaptic characterization of isogenic PCDH15-or RELN-deleted neurons
To investigate whether our isogenic gene-edited iPSCderived neurons recapitulate patient iPSC-derived neu-rons, we performed phenotypic analyses focusing on their dendrite length and synapse formation using the same method as performed in Figure 5. Both glutamatergic and GABAergic neurons were efficiently induced from PCDH15del and RELNdel (Fig. 7A,B,E,F). While some differentiation characteristics were confirmed, Ͼ60% of cell populations constantly differentiated into our target neurons in all cell lines we tested (Fig. 7B,F).
Regarding glutamatergic neurons, MAP2 ϩ dendrites of PCDH15del neurons were significantly shorter than those of Control 1 neurons, while RELNdel neurons tended to form shorter dendrites than Control 2 neurons, although the difference was not statistically significant (p ϭ 0.063; Fig. 7C). Unlike patient-derived neurons, the number of Homer I puncta was increased in both PCDH15del neurons and RELNdel neurons compared with Control 1 or 2 Figure 6. Generation of isogenic PCDH15 deletion iPSCs by targeted genome editing. A, The target sites of CRISPR-sgRNAs in exon 9 of the PCDH15 gene (NM_001142763). Red bars represent PAM (protospacer adjacent motif) sequences. B, Analysis of CRISPR-sgRNA activity by the T7E1 assay using HEK293FT. NTC, No transfection. Among the constructed sgRNAs, sgRNA#3 showed the strongest cleavage activity. C, Indel pattern of the isogenic line using CRISPR-sgRNA#3. Blue letters represent stop codon. D, Representative images of immunocytochemical analysis for in vitro three-germ layer differentiation. Blue indicates Ho and green indicates the pluripotent marker (␤III-tubulin, ␣SMA, and AFP). Scale bar, 100 m. E, Relative gene expression levels of PCDH15 (primer set2) in glutamatergic or GABAergic neurons on day 28 (n ϭ 3 independent experiments; mean Ϯ SD; ‫ء‬p Ͻ 0.05, Student's t test). F, Information of isogenic RELN-deleted iPSCs generated in the previous study . G, Relative gene expression levels of RELN in induced glutamatergic or GABAergic neurons on day 28 (n ϭ 3 independent experiments; mean Ϯ SD; ‫‪p‬ءء‬ Ͻ 0.01, Student's t test).  (Fig. 7D). The number of Synapsin I puncta and colocalized puncta of Homer I and Synapsin I tended to be lower in isogenic lines, although the changes were not significant (Fig. 7D). Regarding GABAergic neurons, MAP2 ϩ dendrite shortening was observed only in PCDH15del neurons (Fig. 7G), while both Gephyrin and Synapsin I puncta were significantly decreased only in RELNdel neurons (Fig.  7H). The number of colocalized puncta was not significantly different in either isogenic line (Fig. 7H). Based on these findings, isogenic gene-edited lines partially recapitulated the structural phenotypes observed in patient-derived lines.

Assessment of neuronal function of patient-derived and isogenic iPSCs-derived neurons
We showed that patient-derived neurons or isogenic gene-edited neurons displayed structural phenotypes. However, it remains unknown whether these neurons show differences in functional activity. Thus, we investigated their neuronal function by assaying spontaneous neuronal activity. To clearly assess neuronal activity, we evaluated the effect of BrainPhys medium, which was developed to support neurophysiological activity (Bardy et al., 2015) on neuronal phenotypes we detected (Fig.  8A). Even by culturing with BrainPhys, BP and SCZ neurons showed the recapitulated phenotypes of dendrite shortening and decreased synapse number and for both glutamatergic and GABAergic neurons (Fig. 8B). Based on these results, we used BrainPhys as the neuron culture media for functional analysis.
For MEA analysis, we induced neurons from iPSCs directly on the MEA plates at a high density (Fig. 8C). Spontaneous activity was detected by 16 electrodes per well in the MEA plate, and the activities of neurons were recorded at each electrode (Fig. 8D). In glutamatergic neurons, active electrodes were stably detected on day 28 after induction, while few active electrodes appeared in GABAergic neurons on day 28 (data not shown). In spike frequency of glutamatergic neurons on day 28 and day 42, there were no significant differences in spike frequency among control, BP, and SCZ groups, although the time-dependent increase in spike frequencies of each group (Fig. 8E). Isogenic PCDH15-deleted or RELNdeleted neurons also showed no significant differences compared with Control 1 or Control 2 both on day 28 and day 42, except for the time-dependent increase in spike frequencies (Fig. 8F). These results indicate that spontaneous neuronal activity was maintained at a comparable level among glutamatergic neurons whether they were control neurons, patient-derived neurons, or isogenic neurons.
Next, we investigated receptor reactivity by treatment with a glutamate receptor antagonist (AMPA receptor antagonist CNQX and NMDA receptor antagonist AP-5) or GABA receptor agonist (GABA). By using induced glutamatergic neurons, we measured changes in the total number of spikes per well before and after treatment. We found that the spike number was significantly decreased both in BP and SCZ neurons after CNQX treatment, but not after AP-5 treatment compared with Control neurons (Fig. 8G). Similarly, GABA treatment resulted in a significant decrease in spike number in SCZ neurons compared with Control neurons (Fig. 8G). These results suggest that BP and SCZ neurons have higher sensitivities in AMPA receptor and GABA receptor stimulation, which might lead to the maintenance of spontaneous activity of neurons.
In GABAergic neurons, 6 weeks of differentiation enabled neurons to have active spikes in most patientderived and isogenic lines. However, some cell lines did not show sufficient neuronal activity to be analyzed (data not shown). Then, we used a calcium imaging system to detect the GABAergic neuron activity with higher sensitivity. We used 1210B2 as a healthy control for comparison with patient-derived lines. We recorded spontaneous calcium spikes from GABAergic neurons on day 28 and focused on the following parameters: ⌬Fmax and spike number ( Fig. 8H; see also Materials and Methods). There were no significant differences both in ⌬Fmax and spike frequency among control-and patient-derived lines (Fig.  8I). Isogenic neurons also showed no significant differences in these parameters compared with controls (Fig.  8J). These results of calcium imaging thus suggest that the spontaneous activity of GABAergic neurons is comparable among control, patient-derived, and isogenic neurons, indicating some compensatory mechanism for structural abnormalities.

Discussion
It is challenging to elucidate the molecular mechanisms and general pathologies of BP and SCZ for the following two reasons: the genetic background complexity of these disorders and the difficulty of pathological recapitulation. However, neurons induced from patient-derived iPSCs can overcome these problems. In particular, the use of iPSCs derived from patients with disease-associated CNVs is a promising strategy (Flaherty et al., 2017;Rutkowski et al., 2017). In the present study, we focused on the following two CNVs: the heterozygous deletion of PCDH15 and of RELN. We used newly generated iPSCs derived from BP patients with PCDH15 deletion and differentiated the BP-iPSCs (PCDH15-deleted) and SCZ-iPSCs (RELN-deleted) into neurons efficiently. We identified several neuronal abnormalities in neurons induced from patient-derived iPSCs.
In this study, we used the following two types of methods for neural differentiation: induction by chemical treatment or overexpression of transcription factors. Neurons induced from patient-derived iPSCs using the chemical treatment method exhibited abnormal phenotypes of neurite extension in the early differentiation stage. Although continued length in GABAergic neurons (n ϭ 4 -6 independent experiments; mean Ϯ SD; ‫‪p‬ءء‬ Ͻ 0.01, Student's t test). H, Quantitative analysis of the number of synaptic marker puncta in GABAergic neurons (n ϭ 4 -6 independent experiments; mean Ϯ SD; ‫ء‬p Ͻ 0.05, Student's t test). Figure 8. Spontaneous activity of neurons induced from patient-derived or isogenic iPSCs. A, Overview of the protocol for neuronal differentiation for functional analysis. B, Quantitative analysis of dendrite length and synaptic markers puncta in neurons cultured in BrainPhys for differentiation on day 28 (n ϭ 3-4 independent experiments; mean Ϯ SD; ‫‪p‬ءء‬ Ͻ 0.01, Dunnett's test among control, BP, and SCZ neurons; BP: BP1-2, SCZ: SCZ1-2). C, Overview of MEA plate and representative images of neurons induced from iPSCs on the 48-well MEA plates. Bright-field image and immunocytochemical images of neuron markers. Scale bar, 200 m. D, Representative image of raster plot and definition of active electrodes. E, Spike frequency of control or patient-derived glutamatergic neurons on day 28 and day 42 (n ϭ 4 -6 independent experiments; mean Ϯ SD; Dunnett's test among Control, BP, and SCZ neurons, no significant differences were observed; paired t test between day 28 and day 42, ‫ء‬p Ͻ 0.05). F, Spike frequency of isogenic iPSC-derived glutamatergic neurons (n ϭ 3-4 independent experiments; mean Ϯ SD; Dunnett's test among Control, BP, and SCZ neurons, no significant differences were observed; paired t test between day 28 and day 42, ‫ء‬p Ͻ 0.05). G, Relative change in the total number of spikes after drug treatment in glutamatergic neurons on day 42 (n ϭ 3 independent experiments; mean Ϯ SD; ‫ء‬p Ͻ these phenotypes may support the neurodevelopmental hypothesis for BP and SCZ (O'Shea and McInnis, 2016;Owen and O'Donovan, 2017), this differentiation method is not suited for the analysis of adult and specific pathologies for medical treatment. On the other hand, we showed that transcription factor overexpression is suitable for the preparation of specific types of mature neurons and analysis of the phenotypes of clinical pathologies. NEUROG2 overexpression induced glutamatergic neurons, and ASCL1 and DLX2 co-overexpression induced GABAergic neurons, from iPSCs efficiently, which enabled analysis of the phenotype of each type of neuron. In addition, this type of method is expected to be applicable for electrophysiological analysis of specific neurons or functional analysis of neural networks by coculturing different types of neurons (Sun et al., 2016). However, the induction of specific subtypes of GABAergic neurons is also needed for further understanding of the pathologies. It was reported that ASCL1 and DLX2 overexpression induced various subtypes of GABAergic neurons, such as calretinin ϩ , somatostatin ϩ , and parvalbumin ϩ neurons (Yang et al., 2017). Although parvalbumin ϩ neurons have important roles in BP and SCZ (Wang et al., 2011;Marin, 2012), the existing method can induce parvalbumin ϩ neurons at only low levels (Sun et al., 2016;Wen, 2017;Yang et al., 2017). Therefore, an efficient method to induce parvalbumin ϩ neurons is required for further investigation of the roles of GABAergic neurons in BP and SCZ.
Interestingly, we identified similar phenotypes of decreased MAP2 ϩ dendrite length and synapse number between BP and SCZ neurons and between glutamatergic and GABAergic neurons. These phenotypes seem to be associated with the common and general pathologies in BP and SCZ in terms of neurite extension and synapse formation, similar to the phenotypes observed in the brains of patients. Our microarray analysis suggested that these common phenotypes were associated with cell adhesion. PCDH15 participates in cell-cell adhesion of hair cells at the inner ear by forming tip-link filaments (Jaiganesh et al., 2018), while there are no reports on the precise role of PCDH15 in neural adhesion. Reelin is also involved in cell adhesion. Reelin-dependent signaling pathways regulate neural adhesion (Sekine et al., 2012;Hirota and Nakajima, 2017). Thus, PCDH15 and Reelin are expected to play key roles in neural adhesion to regulate dendrite or synapse formation. Indeed, it is well known that cell adhesion plays important roles in synapse formation and neurite extension (Missler et al., 2012). Previous studies have suggested that cell adhesion molecules are associated with BP and SCZ (O'Dushlaine et al., 2011;Sakurai, 2017). In addition, if such a basic system of neural development is associated with these phenotypes, the phenotypes may be associated with a broader range of psychiatric disorders. Although, based on these observations, cell adhesion is expected to play important roles in the pathologies of BP and SCZ, it remains unknown whether these phenotypes reflect common pathologies in general patients or are specific to certain genetic backgrounds because of two limitations. First, the number of donors was limited in this study, and further studies are warranted to confirm our findings. Second, the PCDH15 or RELN mRNA levels did not decrease significantly in neurons induced from patient-derived iPSCs according to our qRT-PCR data in contrast to isogenic homozygous PCDH15-deleted or RELN-deleted neurons. A previous report using NEUROG2 overexpression-induced neurons from iPSCs of SCZ patients with heterozygous gene deletion obtained a similar result (Flaherty et al., 2017). Therefore, heterozygous gene deletion does not always indicate the downregulation of gene expression in neurons. Considering that these genes tended to be downregulated in patient-derived DSi-EBs, although these changes were not significant, it is possible that gene expression in patient-derived cells was different from that in the control cells during differentiation.
For further insights into the relationship between CNVs and phenotypes, we established and analyzed isogenic PCDH15-deleted or RELN-deleted iPSCs using genomeediting technology. Although the isogenic lines showed dendritic or synaptic phenotypes partially correlated with those of patient iPSC-derived neurons, these phenotypes of isogenic neurons were weaker. These results suggest that CNVs as well as other factors contribute to these phenotypes. We also confirmed that other clinically important CNVs were not present in the patients, although it is still possible that an unknown genetic background affected the phenotypes or clinical pathologies because BP and SCZ are highly polygenic disorders. If large-scale genome-editing technologies that allow for the rescue of CNVs become available in the future, the contributions of CNVs can be clearly and directly proved. Under these conditions, patient-derived iPSCs may be a better tool than isogenic iPSCs for modeling BP and SCZ.
In addition, we performed functional analysis by measuring spontaneous neural activity to investigate whether the observed structural phenotypes led to functional abnormalities. However, we did not identify any functional differences in spontaneous activity in patient or isogenic neurons. Although our neurons seemed to be immature compared with in vivo neurons, which might make it difficult to detect functional differences, our data suggest that the electrophysiological functions of patient-derived glutamatergic neurons were compensated by the enhancement of sensitivity in their receptors. Although structural abnormalities do not always cause functional defects because a compensatory mechanism might con-continued 0.05, ‫‪p‬ءء‬ Ͻ 0.01, Dunnett's test). H, Representative image of calcium spikes and display of parameters (⌬Fmax and calcium spike numbers). I, ⌬Fmax and calcium spike frequency in control or patient-derived GABAergic neurons (n ϭ 3-6 independent experiments; mean Ϯ SD; Dunnett's test among each line, no significant differences were observed). J, ⌬Fmax and calcium spike frequency in control or patient-derived GABAergic neurons (n ϭ 4 -6 independent experiments; mean Ϯ SD; Student's t test, no significant differences were observed). tribute to functional maintenance, detailed electrophysiological analyses are warranted in future studies. We also showed that our neurons were electrically active to some extent. Therefore, our in vitro model is suitable for further electrophysiological analyses to elucidate the functional significance of CNVs. Also, coculture of both types of our neurons (glutamatergic and GABAergic) or coculture with glial cells, which is expected to promote further functional maturation and neural network formation, may support this approach in the future studies.
For various diseases, potential therapeutic drugs were identified from phenotypic screening using induced tissues from patient-derived iPSCs (Okano and Yamanaka, 2014;Wainger et al., 2014;Hino et al., 2017;Hosoya et al., 2017;Imamura et al., 2017;Kondo et al., 2017;Fujimori et al., 2018;Tabata et al., 2018). Thus, phenotype-recapitulated in vitro model using iPSC technology is available for exploring new therapeutic agents. Our model showing disorder-related phenotypes can be a useful tool for identifying novel drug candidates.
Overall, our approach for the induction of specific neurons from patient-derived iPSCs with disease-related genetic polymorphisms enabled us to analyze pathologyassociated phenotypes. Our in vitro model, which may show general phenotypes of psychiatric disorders, is expected to be applicable for further elucidation of molecular mechanisms or drug discovery.