Subjects were recruited through the UCSD Autism Center of Excellence from a pool of volunteers formerly included in previous brain imaging studies. Control subjects were selected randomly from lists of typically developing individuals who had had an MRI scan when they were toddlers. ASD subjects were selected from lists of ASD subjects who had been identified and diagnosed with ASD and MRI scanned when they were toddlers; from among these potential ASD subjects, we selected those with larger than normal average total brain volume as compared to typically developing toddlers. The ASD subjects demonstrated a behavioral presentation consistent with autism as defined by the criteria set forth in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV; APA)²⁵. Assessment of the history and presence of the disorder was achieved via standardized behavioral, cognitive, and functional assessments, including the appropriate Wechsler Intelligence Scale, the Autism Diagnostic Observation Schedule (ADOS), the Autism Diagnostic Interview, Revised (ADI-R), and Vineland Adaptive Behavior Scales (VABS). Participants in the control group had no history of psychological, genetic, or other disorder. This project was approved by the Ethics Committees of the institutes at which the study was conducted. After a complete description of the study, written informed consent was provided by the parents.

High-confidence CNV calls were made using the CNVision pipeline as previously described ^{7, 26}. In brief, patients and controls were genotyped using the Illumina 660W array. Probe intensity signals were analyzed to make CNV calls using e types of software PennCNV, QuantiSNP and GNOSIS. High-confidence calls were made by merging the prediction output of these 3 different algorithms. Rare events were defined based on the database of genomic variants (DGV) (http://dgv.tcag.ca/).

Fibroblasts derived from the 8 ASD individuals were cultured. DNA was extracted and exome libraries were generated for each line using Nextera Rapid Capture Exome kit from Illumina. Illumina Hi-Seq 2500 Pair-end sequencing was conducted (read length=100bp), resulting in 308,082,362 reads in average for each library. Short reads were aligned to the human genome (hg19) using BWA ²⁷ and Samtools ²⁸. Mutations were first filtered based on the reading depth (D>=10), alignment quality (q>=20) and base quality (Q>=30), and 4712088 variants passed filtering. Next, we used two different software programs, ANNOVAR³¹ and IlluminaVariantStudio (https://www.illumina.com/content/dam/illumina-marketing/documents/products/datasheets/datasheet_illumina_variantstudio_software.pdf) to annotate the variants. The results from ANNOVAR ²⁹ and IlluminaVariantStudio were very similar (Supplementary Table S1A), so we further analyzed variants found in the IlluminaVariantStudio only. There were a total of 485 stop-gains and 103 stop losses found in all ASD lines, but several variants were found in multiple lines. Therefore, a total of 275 unique stop gain/loss mutations were detected, of which 222 were listed in dbSNP while 53 rare mutations were not (Supplementary Table S1B). We further filtered the 58278 missense variants found in IlluminaVariantStudio by PolyPhen and SIFT scores. Of these, 3403 variants were considered both “deleterious” in SIFT and “probably damaging” in PolyPhen. Again, several of these variants were found in multiple patients, so the total number of unique “deleterious” and “probably damaging” SNVs was 1829. Of these, 1564 were found in dbSNP while 265 rare variants were not (Supplementary Table S1C).

IPSCs were obtained from skin fibroblasts of ASD patients and controls, collected by the UC San Diego Autism Center of Excellence. Briefly, fibroblasts were transduced with retroviruses containing OCT4, SOX2, KFL4, and MYC to induce overexpression of these genes ³⁰. Two days after transduction, the cells were transferred to a co-culture system with murine embryonic fibroblasts (mEFs) maintained with DMEM/F12 (Invitrogen, CA), 20% Knockout Serum Replacement (Invitrogen, CA), 1% non-essential amino acids, and 100 μM beta-mercaptoethanol and treated with 1 mM valproic acid (Sigma) for 5 days. The iPSC colonies were identified after approximately 2 weeks in this culture system, transferred to Matrigel (BD Biosciences)-coated plates, and maintained in mTeSR media (Stem Cell Technologies). To avoid potential variability caused by retroviral insertions in the genome, the experiments were performed with at least 2 independent iPSCs clones.

Standard G-banding karyotype analyses were performed by Cell Line Genetics (Madison, WI).

The iPSC colonies were plated on Matrigel (BD Biosciences)-coated plates and maintained for 5 days in mTSeR medium (Stem Cell Technologies). On the fifth day, the medium was changed to N2 media [DMEM/F12 medium supplemented with 1× N2 supplement (Invitrogen) and 1× B27 supplement (Invitrogen). After 2 days, the colonies were removed from the plate and cultured in suspension as EBs for 2 weeks using NPC medium with dorsomorphin during the entire procedure. The EBs were then gently dissociated with accutase (Gibco), plated on Matrigel-coated dishes, and maintained in NBF medium (DMEM/F12 medium supplemented with 1× N2, 1× B7 supplements, 20 ng/mL FGF2 and 1% penicillin/streptomycin). The rosettes that emerged after 3 or 4 days were manually selected, gently dissociated with accutase, and plated in dishes coated with 10 μg/mL poly-ornithine and 5 μg/mL laminin. This NPC population was expanded using NBF medium. To differentiate the NPCs into neurons, the cells were re-plated in the absence of FGF2, with regular medium changes every 3 or 4 days.

Cells were washed once with PBS and then permeabilized with 0.5% Triton X-100 in PBS for 5 minutes. After washing with PBS and blocking with 10% Donkey serum, 0.5% Triton X-100 in PBS (blocking solution), cells were incubated with the primary antibodies diluted in blocking solution for 16 hours at 4°C. Primary antibodies used were: anti-Pax6 (Covance PRB-278P); anti-Sox1 (R&D, AF3369); anti-Otx2 (Milipore, AB9566); anti-Vimentin (abcam, RV202) anti-phospho-Histone H3 (Sigma, H0412); anti-Ki67 (Novocastra NCL-Ki67p); Anti-Brn2 (Santa Cruz, cs6029) anti-Brn2 (2R2) ³¹; anti-Ngn2 (Santa-cruz, sc-19233); anti-Mash1/Ascl1 (BD 556604 and Chemicon AB5696s); anti-NKX2.1 (Novacastra, TTF-1-L-CE); anti-Olig2 (Millipore, AB9610); anti-Tuj1 (Covance, MMS-435P) and anti-Doublecortin (Santa Cruz, cs-8066); anti-GABA (Sigma, A2052), anti-Map2AB (Abcam, ab5392). After PBS washes, cells were incubated with labeled secondary antibodies for 60 minutes. Nuclei were stained after incubation with 1 μg/ml DAPI for 5 minutes. The slides were mounted using prolong-gold (Invitrogen, P36930).

After NPCs grew close to confluence (4~9 days), the cell number in each well was counted using a cell counter (Invitrogen countess automated cell counter). The population doubling time was inferred for each well based on the culture time and the ratio between input and harvest. Differences in doubling time between LiCl-treated and untreated conditions in each NPC line were compared using 2-tail t test. Also, doubling time differences between LiCl-treated and untreated conditions were tested for all 5 autistic lines (treated=27, non-treated=28) and all 5 control lines (treated=11, non-treated=10) using the two-tail t test.

Monolayer-adherent NPCs were harvested, washed with PBS, resuspended in PBS and fixed for 1 hour in 4°C in 100% ethanol (−20°C). Cells were collected, resuspended in PBS containing 100 μg/ml RNase A for 30 minutes in 37°C, and stained with 1 mg/ml propidium iodide solution (10 mg / 1×10^6 cells stained) for 2 hours at 4°C. For each sample, 50,000 cells were collected in the appropriate gate for analysis on FACSCalibur (BD Biosciences, Mountain View, CA) using FlowJo software (Tree Star, Ashland, OR).

NPCs cultured as monolayers were washed with cold PBS. Protein was extracted with 50 mM Tris pH 7.6, 500 mM NaCl, 1 mM EDTA, 1% NP-40, 1% deoxycholate, 0.1% SDS and protease inhibitor cocktail (Roche, # 1 836 153). Lysates were centrifuged for 25 minutes at 15,000 rpm at 4°C. The resulting supernatant was collected and the total protein content of the samples was determined by the BCA Protein Assay Kit (Pierce Biotechnology, #23225). SDS gel electrophoresis was performed utilizing 15 μ5 protein per lane. Gels were transferred to nitrocellulose membranes (BioRad, 9004-70-0). Proteins were detected with anti-Pax6 (Covance PRB-278P), anti-Sox1 (R&D, AF3369), anti-Otx2 (Milipore, AB9566), anti-Vimentin (abcam, RV202), anti-Brn2 (2R2) ³¹, anti-β-III-tubulin (Covance, MMS-435P) and anti-GAPDH (Acris, ACR001P). All membranes were visualized using ECL (Thermo Scientific, 34080) and exposure to autoradiography film (Denville Scientific Inc. E3018).

Total cellular RNA was extracted with TRIzol (Invitrogen) and purified on a column using the RNeasy Mini Kit. DNAase treatment was performed on the column. Extracted RNA (1 μg) samples were reverse transcribed into cDNA according to the manufacturer’s protocol for RT-PCR (BioRad, 170-8891). RT-qPCR was performed by using SsoFast EvaGreen Supermix (Biorad) and run on the Eppendorf Mastercycler EP Realplex 2 thermal cycler. The specific primers used for RT-qPCR were: HPRT1 realtime primers.com, VHPS-4263, Pax6 realtime primers.com, VHPS-6650, Otx2 realtime primers.com, VHPS-6546, Sox1 Left primer: GGGAAAACGGGCAAAATAAT; Right primer: TTTTGCGTTCACATCGGTTA. Brn2 Left primer: TGACGATCTCCACGCAGTAG; Right primer: GAGGGTGTGGGACCCTAAAT.

For neurotransmitter receptors analysis, iPSC-derived neurons from ASD and related controls were rapidly frozen and placed in TRIzol (Life Technologies) for subsequent RNA isolation and purification. One microgram of RNA of each individual/clone were converted into cDNA using the RT² First Strand Kit (SABiosciences). Quantitative RNA expression analysis of 84 neurotransmitter receptors-related genes was simultaneously analyzed using human RT² Profiler PCR Array (#PAHS-060Z; SABiosciences) on a Bio-Rad CFX96 following the manufacturer’s instructions. Data normalization was based on correcting all Ct values for the average Ct values of five constantly expressed housekeeping genes present on the array, and reported as fold changes over similar values from controls samples.

Transfection was performed in NPC medium, which was replaced 24 hours following transfection. Transfection efficiency was measured by GFP expression and was ~45%. Super Top-Flashx8 (50 ng), super Fop-Flashx8 (50 ng) and Renilla (10 ng) plasmids were transiently transfected into NPCs (40,000 cells per well of a 96-well plate) using Lipofectamine LTX+ in NPC medium, which was replaced 24 hours following transfection. After 48 hours, cells were lysed in the 96-well plates with passive lysis buffer (Promega). Firefly luciferase and Renilla luciferase activity in the lysates were measured on a Synergy 2 microplate reader (BioTek Instruments, Winooski, VT) in replicates of 3, using the Dual-Luciferase reporter assay system (Promega). The ratios for firefly luciferase:Renilla luciferase were determined and expressed as relative luciferase activity.

Neuronal tracing was performed on neurons in which the shortest dendrite was at least 3 times longer than the cell soma diameter, as determined using a semi-automatic ImageJ plug-in (NeuroJ). Spines and VGLUT1 puncta were quantified after 3-dimensional reconstruction of z-stack confocal images. The same density of neurons was plated in each condition. Final cell density was confirmed by DAPI. Only Tuj1- or Map2-positive neurons were scored. Images were taken randomly for each individual and from 2 different experiments, using at least 2 different clones. Quantification was performed blind to the cell genotype. No distinction was made between different types of spines due to the unviability of this assessment using the present method. All experiments were performed with independent clones and different controls. All analyses were performed double blinded to avoid bias.

The immunofluorescent slides were viewed and photographed with 40× or 60× objectives using a Nikon C1si microscope. Analysis of the thickness and/or cell counts of immunohistochemical staining were captured and quantified across the entire image taken by the Image-Pro Plus system (version 5.1; Media Cybernetics).

Whole-cell patch clamp recordings were performed on neurons co-cultured with rat hippocampal astrocytes for 8 weeks. The same density of neurons was plated in each condition. Neurons were labeled with a synapsin-GFP lentiviral marker. Cells were perfused with ACSF solution containing (in mM): 119 NaCl, 2.5 KCl, 1 NaH₂PO₄, 26 NaHCO₃, 2 CaCl₂, 1.3 MgSO₄, 11 glucose, 0.1 picrotoxin. 2-5 MΩ pipettes were pulled on a horizontal micropipette puller (Sutter P-97). For sEPSPs and Na+/K+ currents, pipettes were filled with a cesium-gluconate internal solution that consisted of (in mM): 130 D-Gluconic Acid, 130 CsOH, 5 NaCl, 12 Phosphocreatine, 10 HEPES, 10 EGTA, 3 adenosine triphosphate, 0.2 guanosine triphosphate (pH 7.3). For spontaneous and evoked action potentials, cells were recorded with a potassium-gluconate internal solution that consisted of (in mM): 120 K-gluconate, 15 KCl, 4 MgCl2, 0.1 EGTA, 10 Hepes, 4 MgATP, 0.3 Na2GTP, 7 phosphocreatine (pH, 7.2). The mean access resistance was 7.5 ± 1.0 MΩ. Recordings were collected using an Axoclamp-200B amplifier (Axon Instruments) and analyzed with Clampfit 10 software (Molecular Devices). Signals were low-pass filtered at 2kHz and digitized at 5kHz. Na+/K+ currents were recorded by stepping cells in +5-mV increments. For evoked action potentials, current injections were applied to hold cells at approximately −50 mV. A series of 300-ms current steps were injected into cells, starting at −0.2 nA, and increasing by +0.1 nA increments. Statistical comparisons of control and autism cells were made using two-tailed Student t-tests with a significance threshold of p = 0.05.

Using the 12-well MEA plate from Axion Biosystems, we plated cells derived from 5 control individuals and 7 ASD patients in triplicate. Each well was seeded with 10,000 NPCs that were induced into neuronal differentiation as described above. Each well was coated with poly-ornithine and laminin prior to cell seeding. Cells were fed once a week and measurements were taken before media changes. Recordings were performed in a Maestro MEA system and AxIS software (Axion Biosystems) using a bandwidth with a filter for 10Hz to 2.5kHz cutoff frequencies. Spike detection was performed using an adaptive threshold set to 5.5 times the standard deviation of the estimated noise on each electrode. Each plate rested for 5 minutes in the Maestro Instrument, and then was recorded for 10 minutes to calculate the spike rate per well. Multi-electrode data analysis was performed using the Axion Biosystems Neural Metrics Tool, and an active electrode was considered once 5 spikes occurred over the length of 1 minute (5 spikes/min). Bursts were identified in the data recorded from each individual electrode using an adaptive Poisson surprise algorithm. Network bursts were identified for each well; using a non-adaptive algorithm requiring a minimum of 10 spikes with a maximum inter-spike interval of 100ms. Only the wells that exhibited bursting activity were included in this analysis. For the rescue experiments, 10ng/ml of recombinant human IGF1 (PeproTech) was added to the neuronal cultures on day 35 of differentiation. Media was changed every week (with or without the drug for the blank wells) for the duration of the experiment.

RNAs were isolated from 88 cell lines (31 iPSCs, 29 NPCs and 28 neurons) derived from idiopathic ASD patients with macrencephaly and non-autistic controls using the RNeasy Mini kit (QIAGEN). A total of 1,000 ng of RNA was used for library preparation using the Illumina TruSeq RNA Sample Preparation Kit. The RNAs were sequenced on Illumina HiSeq2000 with 50bp paired-end reads, generating 50 million high quality sequencing fragments per sample on average. We first utilized TopHat2 aligner to align the reads against the human reference genome from the 1000 Genomes Project (i.e., human_g1k_v37), and then summarized the read counts to Genecode11 union-exon gene models using htseq-count (http://www-huber.embl.de/users/anders/HTSeq/doc/count.html). Of the reads, 85% could be uniquely mapped, and 74% of those uniquely aligned reads fell into exonic and UTR regions. Next, we processed the expression data for each experimental stage separately. At every differentiation stage, outlier samples were defined by low inter-sample correlations as previously described ³² and subsequently excluded in the follow-up expression analysis. Eighty-three samples were retained for further analysis, including 28 iPSCs, 30 NPCs, and 25 neuronal samples. Gene-based read counts were further median-normalized and log₂-transformed to bring their distribution to normality. Batch effects were corrected afterwards using R ComBat package with the non-parametric model option. Lastly, we filtered expressed genes requiring count per million (cpm) values larger than 1 in at least 1 group of the samples (ASD or controls). Consequently, 14,291 genes in iPSCs, 14,413 in NPCs, and 14,974 in neurons were identified as expressed and were retained for further analysis. All gene expression data is deposited at GEO (GSE67528).

DE analysis at static experimental stage was performed using the linear mixed effect model from R limma package. Samples derived from the same subject were considered as replicates, where their variations were accounted for the random effects in the model. RNA qualities may greatly influence the DE analysis. Therefore, we included RNA integrity numbers (RIN) as a covariate in our regression model. DE genes were detected based on the significance threshold of p-value <0.005, unless we specified otherwise. To identify the genes showing differentiation-induced expression changes, we first paired progenitors with neurons if they were from the same iPSC clone and plated for differentiation in 1 experiment. Next, we computed the gene-based ratios of progenitor expression over neuronal expression for every differentiation pair and we used the log₂-transformed ratios as the input for the follow-up regression model. The comparison was conducted using the linear mixed model via the lme function provided by R package nlme. Ratios computed for the same subject were considered as replicates. Significant events were detected if p-value <0.005.

Gene ontology (GO) enrichment analysis was assessed using the GO-Elite Pathway Analysis tool. GO-Elite performs permutations to obtain an over-representation Z score and enrichment p-value for each GO term. We used the default permutations settings in GO-Elite package (see supplementary materials for details).

We conducted signed co-expression network analysis on 25 neuronal samples using R WGCNA package, as previously described ^{6, 8}. Briefly, 14,974 expressed protein-coding genes were used in our analysis, and the soft thresholding power was chosen to be 16 based on the approximate scale-free topology criterion. We set the minimum module size to 100 genes and required the dissimilarities among modules to be at least 15%. Each identified module was summarized by the module eigengene (first component of the standardized expression patterns) and was thereafter correlated with ASD disease status. The disease association significance was evaluated by a linear mixed effect model via R lme function. Four modules (brown, tan, magenta, and purple) were considered significantly associated with ASD status with an association of FDR<0.1. Furthermore, genes were prioritized based on their correlations with the module eigengene, named kME. The top 200 connected genes ranked by kME values were used to generate the network plots via the R igraph package.

All statistical analyses were performed using SPSS version 12. The control and ASD groups were compared by Student’s t-test and, for the doubling time, we used the repeated measurement analysis. For the pCAG-Brn2 over-expression and LiCl experiments two-way ANOVA was used. Where appropriate, Student’s t-test post-hoc analysis was performed with Bonferroni correction for multiple comparisons, and the p-values presented are those of the corrected values.

A summary Table containing number of patients and control cell lines used of each experiment can be found on Supplementary Table S7.

We recruited 8 ASD Individuals with quantitative MRI-validated early brain enlargement ranging from mild brain enlargement to macrencephaly and 5 age/gender-matched control individuals for skin biopsies and phenotypic characterization (Supplementary Table S2). The ASD donors displayed larger brain size compared to the normal average brain size of typically developing control subjects at any given age (Figure 1A, B). Copy Number Variation (CNV) analysis, using DNA extracted from the donors’ whole blood did not show the presence of any rare structural variant known to be associated with ASD (Supplementary Table S3). A small duplication involving intron 2 of the NLGN1 gene was detected in one ASD case; however, this event has also been reported in typically developing controls (http://dgv.tcag.ca/). Exome sequencing was performed on DNA from the 8 ASD fibroblasts. Variants were called by two pipelines (lluminaVariantStudio and ANNOVAR) that allowed for the detection of variant alleles with higher frequency, since we reasoned that both rare and common variants may contribute to the ASD phenotypes. Using these criteria, more than 50,000 total variants (58866 and 53101 from each of the two pipelines) in all 8 libraries were found (Supplementary Table S1A). Since both pipelines gave similar results, we focused our analysis on stop-gain, stop-loss and non-synonymous variants identified by the lluminaVariantStudio pipeline. A total of 275 unique stop gain/loss mutations were detected, of which 222 were listed in dbSNP while 53 rare mutations were not (Supplementary Table S1B). Filtering the non-synonymous variants that were considered both “deleterious” in SIFT and “probably damaging” in PolyPhen, a total of 1829 unique variants were detected (Supplementary Table S1C). Of these, 1564 were found in dbSNP (common) while 265 rare variants were not in dbSNP (Supplementary Table S1C).

Notably, novel stop-gains were found in both pipelines in the canonical Wnt pathway genes CTNNB1, coding for β-catenin, and FZD6, coding for Frizzled 6 (Table 1A). Additionally, two more common missense variants listed in dbSNP in FRZB (rs288326), coding for a secreted Frizzled antagonist, and WNT10B (rs121908120), coding for a WNT ligand, were also detected (Table 1A). These three missense variants were predicted to be damaging by both SIFT and PolyPhen algorithms. No mutations were found in 2 alleles of a single gene in any patient (i.e., all variants were heterozygous). In addition, stop-gain or damaging mutations by both SIFT and PolyPhen algorithms of 97 known autistic genes found in SFARI Gene (https://sfari.org/resources/sfari-gene) were detected (Supplementary Table S4), including one mutation found in PTEN (rs181113165), previously implicated in ASD brain growth trajectory³³. Of the mutations found in these 97 genes, 78 were found in dbSNP and 19 were novel. Gene ontogeny (GO) analysis revealed that 11 GO terms were significantly represented among the variants detected in these ASD samples after correction for multiple testing (Table 1B, Supplementary Table S5). These GO terms included sensory perception, cognition, neural systems, adhesion terms, as well as G-protein and cell surface receptor signaling terms. Pathway analysis using DAVID revealed that two pathways, olfactory transduction and ABC transporters, were significantly represented among the variants detected in these ASD samples after correction for multiple testing (Table 1C, Supplementary Table S6).

ASD and non-affected control fibroblasts were transduced with 4 retroviral reprogramming vectors (Sox2, Oct4, c-Myc and Klf4), as described elsewhere ³⁰. Following 2 to 3 weeks of culture in human embryonic stem cells (hESC)-supporting conditions, compact refractile ESC-like colonies emerged from a background of fibroblasts. iPSC colonies were then manually picked and cultured under feeder-free conditions. Cells were mechanically expanded for at least 10 passages and tested for the expression of pluripotent markers. We obtained several clones that continuously expressed pluripotent markers, such as Nanog, Lin28, Tra-1-81 and Sox2, from each control wild-type (WT)-iPSCs and ASD-iPSCs. We excluded all karyotypically unstable clones from further experiments (Figure S1).

Our recent analyses showed that iPSC-derived neural cell types could recapitulate human fetal cortical development and thus could be used to model this developmental stage in ASD³⁴. We used our previous published protocol to generate NPCs from iPSCs⁴ in the presence of Noggin. Briefly, we initiated neural differentiation by plating 1-week-old embryoid bodies (EBs) treated with Noggin. After a week in culture, EB-derived rosettes became apparent in the dish. Rosettes were then manually collected, dissociated and re-plated. The NPCs derived from rosettes formed a homogeneous population after a few passages from ASD individuals and controls, and continued to proliferate in the presence of FGF2 as adherent monolayers.

We hypothesized that an alteration of the rates of NPC proliferation could result in early brain overgrowth. Proliferation was measured in (2 clones per line) by calculating the population doubling time from plating at passages 3 to 7 (P3-7) in continuous culture, and the doubling times of all ASD and control lines were pooled to calculate averages. From P4, the population doubling time decreased in ASD NPCs from all 8 Individuals compared to NPCs from all 5 controls, reaching statistical significance at P6 (Figure 1C and Figure S2A). Surprisingly, all ASD NPC clones proliferated faster than all control NPC clones. Cell cycle analysis at P6 revealed that shortening of G1 and S phases was the main reason for the decrease in the population doubling time, with no change of G2-M phase length (Figure 1D and Figure S2B). Double labeling for Ki67 and pHH3 revealed an increased percentage of Ki67⁺ (cycling cells) in ASD relative to control NPCs, whereas the percentage of pHH3⁺Ki67⁺ (G2-M phase mitotic cells) was unaffected in autistic NPCs (Figure 1E, F and Figure S2C). After patient and control age-matching correction, we performed pair-wise correlations between brain size (volume) and NPC proliferation and doubling time. We observed a positive correlation with NPC proliferation and an inverse correlation with NPC doubling time across all samples (ASD and controls) (Figure 1G and Figure S2D). Strikingly, proliferation (represented by percentage of Ki67 positive cells) is robustly significant in its correlation with brain volume, suggesting that it may be proliferation that specifically contributes to the large brain early on in ASD. These findings demonstrate that iPSC-derived NPCs from ASD Individuals with macrencephaly proliferated faster than those derived from controls and NPC proliferation is correlated increase in brain volume. To our knowledge, this is the first time that clinical data is significantly correlated to an iPSC-neural phenotype and open the doors for potential use of vitro assays as diagnostic tools for complex multigenic disorders such ASD.

Based on the genomic analyses of the ASD individuals, we hypothesized that the alterations in the canonical Wnt pathway might govern the aberrant proliferation in ASD-derived NPCs. To test our hypothesis, we examined Wnt/β-catenin transcriptional activity using TOP-flash assays and found that β-catenin transcriptional activity was indeed reduced in ASD NPCs compared to control NPCs (Figure 2A and Figure S2E). Activation of canonical Wnt signaling with 5 mM LiCl, which inhibits GSK3β-mediated degradation of β-catenin, elevated β-catenin transcriptional activity in both ASD and control NPCs. However, β-catenin transcriptional activity was significantly reduced in LiCl-treated ASD NPCs compared to control NPCs, suggesting that the cause of reduced β-catenin transcriptional activity is downstream of GSK3β. While the elevation of β-catenin transcriptional activity with LiCl treatment in ASD NPCs is modest, LiCl treatment was able to rescue the overproliferation defects on ASD NPCs. Importantly, since the level of Wnt transcriptional activity required for normal proliferation is currently unknown, it is conceivable that a moderate increase of Wnt is sufficient to rescue abnormal over proliferation in ASD. In our experiments, LiCl treatment increased Wnt transcriptional activity by over 2 fold on ASD NPCs and that is potentially sufficient to elicit normal proliferation. Activation of canonical Wnt signaling with Wnt3A resulted in a marked but non-significant elevation of β-catenin transcriptional activity in both control and ASD NPCs, with a similar trend of reduced β-catenin transcriptional activity in the ASD NPCs (Figure 2A and Figure S2E). Notably, NPCs derived from the patient with the heterozygous stop-gain mutation in CTNNB1, Arch, displayed the weakest responses to LiCl and Wnt3a stimulation (Figure S2E). To determine if this reduced β-catenin transcriptional activity was responsible for the growth defects displayed, we repeated the proliferation experiment in ASD-derived and control NPCs in the presence and absence of LiCl. In the absence of LiCl, all ASD-derived NPC lines displayed faster growth (reduced doubling times) than all control NPC lines. LiCl treatment slowed the growth of ASD-derived NPCs close to doubling times of control NPCs, whereas the growth of control NPCs was unaffected (Figure 2B and Figure S2F). These findings indicate that the reduced β-catenin transcriptional activity played a functional role in the accelerated proliferation found in ASD-derived NPCs.

Dvl1^−/−3^+/− mutant mice display reduced social interaction and transient embryonic brain overgrowth, the result of reduced β-catenin transcriptional activity and reduced levels of the transcription factor Brn2 or Pou3f2³⁵. To test if BRN2 levels were similarly affected in NPCs from ASD patients with early brain overgrowth, we performed immunocytochemistry for BRN2 on ASD and control NPCs. The percentage of BRN2⁺ cells was in fact reduced in ASD NPCs compared to controls (Figure 2C, D and Figure S2G), Western blot analysis confirmed that BRN2 protein levels were reduced in ASD NPCs compared to controls (Figure 2E, F and Figure S2H). To examine whether exogenous BRN2 rescued the increased rate of proliferation observed in ASD NPCs, we transfected CAG-BRN2 into control and ASD NPCs and performed immunocytochemistry for BRN2 and Ki67. Transfection efficiency was similar between control and ASD NPCs (around 45%). The number of DAPI⁺ cells (as an index of proliferation) in ASD NPCs was similar to controls after BRN2 overexpression (Figure 2G, H and Figure S2I, J). The expression pattern of BRN2 and Ki67 in the BRN2 transfected ASD NPCs resembled that of control NPCs, providing additional support for the direct regulation of proliferation of autistic NPCs by BRN2. These data also suggest that partially conserved pathways regulate proliferation of NPCs in a mouse model of social behavior and ASD-derived NPCs from patients with early brain overgrowth (Figure S2K).

Evidence from the literature suggests an imbalance in excitatory versus inhibitory signals in developing ASD^{36, 37}. In addition, analysis of Brn1/Brn2 double mutant mice^{31, 38} demonstrated that these genes regulate neurogenesis and proliferation in the cerebral cortex of glutamatergic precursor fields. Therefore, we tested whether glutamatergic and GABAergic cell fate determination were affected in ASD NPCs. Indeed, a reduction in the percentage of glutamatergic NGN2⁺ NPCs in ASD NPCs compared to controls was observed (Figure 2I, J and Figure S2L). In contrast, markers of GABAergic inhibitory precursors present in the subpallium (MASH1, DLX2 and NKX2.1) were up-regulated in ASD compared to control NPCs, whereas OLIG2, another subpallium marker, was unchanged between control and ASD NPCs (Figure 2J and Figure S2L). In light of these findings, we tested whether this change in fate determination had an impact on neurogenesis and synaptogenesis. We did not detect a significant alteration in ASD neuronal survival after differentiation when compared to controls, as measured by MAP2 staining.

Upon neuronal differentiation of these NPCs, we detected a significant reduction in inhibitory excitatory neurotransmitters GABA (γ-amino butyric acid) ASD neurons compared to controls (Figure 3A, B and S3A) while the percentage of neurons (assessed by Map2 staining) and astrocytes (assessed by GFAP staining) was unchanged (Figure S3 B, C). We also found a clear reduction in the density of both Synapsin and VGLUT1 (vesicular glutamate transpoter-1) puncta from ASD neurons, suggesting a specific defect in vesicular glutamate transport in ASD cultures (Figure 3C-E and Figure S3D-G). To confirm the specificity of glutamatergic neurons in our cultures, we demonstrated that VGLUT1 puncta were mostly adjacent to the postsynaptic density-95 (PSD95) protein, a postsynaptic glutamatergic marker³⁹ (Figure 3C). The distribution of puncta densities for both Synapsin and VGLUT1 displayed a distinctly increased number of ASD neurites with lower densities and an increased number of control-derived neurites with higher densities, corroborating the defect in glutamatergic synapses. We also found significant differences in the area size of synaptic puncta between the 2 groups (Figure S3H, I).

A decrease in the number of excitatory synapses may directly affect other neurotransmission systems in ASD cultures⁴⁰. To investigate whether the expression of neurotransmitters was affected in ASD-derived neurons, we performed a qPCR array of 84 neurotransmitter receptors-related genes in both ASD and control neurons. Interestingly, we detected a tendency for GABA-related and neuropeptide Y receptors to be down-regulated in ASD-derived neurons (Figure 3F and Supplementary Table S8). In this experiment, only the subunits GABRQ and NPY2R were significantly down-regulated in ASD compared to control neurons. We then investigated the expression of 86 different gene indicators for different developmental stages of neural differentiation, neurotransmitters maturation and cortical layer formation using the Fluidigm Biomark qPCR Dynamic Array (Figure 3G and Supplementary Table S8). We detected down-regulation of the GABA receptor GABRR1, plus other relevant neurotransmitters and synapse-related genes, such as KCNA1, TH, PHOX2A and DRD3, in ASD neurons. In contrast, genes related to early stages of neural differentiation, such as NEUROG3, FOXG1, SOX10 and P2RY2, were significantly up-regulated in ASD. Together the results described above indicate that, while ASD neurons initiate an earlier differentiation program, ASD neurons do not mature as extensively as controls.

Next, electrophysiological recordings using single-cell patch clamp were used to investigate the functional properties of iPSC-derived neurons 6 weeks after differentiation. Spontaneous action potentials were observed in control and ASD-derived neurons. However, no difference was observed in the frequency of spontaneous action potentials between iPSC-derived neurons from control (4.3 ± 1.2Hz, n = 21) and ASD (3.0 ± 0.7Hz, n = 27, p = 0.3) cultures. iPSC-derived neurons further exhibited spontaneous excitatory postsynaptic currents (sEPSCs), with comparable amplitudes (p = 0.7) between control (26.5 ± 2.6pA, n = 20) and ASD (27.7 ± 2.6pA, n = 32) cultures. Likewise, no significant difference was found in terms of mean sEPSC frequency (control: 7.6 ± 1.4Hz, ASD: 5.3 ± 1.0Hz, p = 0.2, Figure S4A-D). Despite the lack of significant sEPSC frequency differences between control and ASD neurons, there was a trend of lower frequencies in ASD cultures (Figure S4E). Both control and ASD-derived cells exhibited characteristics of functional neurons, including transient voltage-gated sodium and potassium currents evoked by voltage-step depolarizations and action-potentials generated following depolarizing current injections.

Next, we analyzed spike activity within the neuronal networks by multielectrode arrays (MEA). Connectivity defects have been commonly observed in postmortem ASD brains, but connectivity cannot be studied during early stages of embryogenesis^{41, 42}. Unlike single-cell patch clamp recordings, MEA allow for the dynamic interrogation of how human-derived neurons behave in circuits. We plated paired control and ASD NPCs to mature in MEA dishes and recorded their activity over time. After 30 days, the total number of spikes in control and ASD neuronal networks was similar. As the cultures matured, the number of spikes increased in control neurons but the numbers of spikes in ASD did not, leading to a significant difference between ASD and control networks (Figure 4A-C). After 50 days, we also detected a reduction in the number of synchronized bursts, i.e., spikes that were not singular or random (sequential 10 spikes over 100 ms) in the ASD cultures (Figure 4D). While the number of network bursts was variable at 30 days of maturation, with a tendency toward more synchronized events in control neurons, within 50 days of maturation this difference was 6 times higher in control compared to ASD neurons and highly significant (Figure 4C, D).

We then treated the cells with IGF1, a drug that has been previously shown to rescue synaptic deficits in other ASD monogenetic syndromes such as Rett and Phelan-McDermid Syndromes, and is currently in clinical trials for idiopathic ASD^{4, 21}. When IGF1 was used during the differentiation process, we observed an increase in the number of GABAergic neurons in all ASD-derived cultures, but no effect in controls (Figure S4 F-I). We next used IGF1 to treat mature neuronal cultures after differentiation, starting on day 35 of differentiation. Collectively, ASD patients displayed a tendency to improve neuronal spike number that became significant at day 72 of differentiation (Figure 4E). Interestingly, when we looked at individual patients’ response to the drug, while most patients improved the neuronal activity, Apex did not show any response (Figure S4J). Further investigation should clarify the mechanistic reasons for such drug response variation, but nonetheless, the assay may have the sensitivity to pre-screen patients for future clinical trials. To address the involvement of stabilization of Wnt-β-cat pathway in NPCs and the consequence effect on neuronal function, we treated NPCs with LiCl and assessed the differentiation by quantifying the number of GABAergic neurons after 4 weeks of differentiation and also analyzed neuronal functional maturation using Multi Electrode Array (MEA) technology. Interestingly, treatment of NPCs with LiCl did not improve the percentage of GABA positive cells 4 weeks after differentiation or improved MEA activity on ASD neurons (Figure S4I and S4K). Hence, contrary to IGF1 treatment, LiCl treatment (and consequent stabilization Wnt-β-catenin pathway and rescue of NPC proliferation) does not affect neuronal function significantly, suggesting that there are multiple pathways involved in the ASD phenotypical deficit or that stabilization of the Wnt-β-catenin pathway is also required at the early stages of differentiation³⁵. These results demonstrate an impairment followed by a partial rescue with IGF1, but not LiCl of network formation in ASD-derived neuronal cultures.

Strong evidence implicates a convergence of pathways at the transcriptome level in postmortem brains from ASD patients^{8, 32}. Therefore, we performed RNA sequencing analysis in iPSCs and in iPSC-derived NPCs and neurons to evaluate whether any convergent transcriptional alterations could be identified to be associated with early brain overgrowth in ASD during development in this in vitro model. At the transcriptome level, samples were explicitly separated into 3 distinct clusters by cell-type differences based on the top 2 principal components, strongly supporting the reproducibility of our iPSC generation and neural differentiation procedures (Figure S5A, B).

We first conducted a standard differential expression (DE) analysis to evaluate individual gene expression pattern changes at the NPC and neuronal stages. We identified 71 genes that were significantly differentially expressed in ASD compared to controls at the NPC stage and 154 genes at the neuronal stage, which by hierarchical clustering clearly distinguished(the ASD cell lines from controls (Figure 5A and Supplementary Table S9). The expression fold changes of the top dysregulated genes are shown in Figure 5B. Of note, while the expressions of both CTNNB1 (coding for β-catenin) or POU3F2 (coding for BRN2) mRNAs were reduced, neither was significantly altered in ASD-derived NPCs, although β-catenin activity in the canonical Wnt pathway is primarily regulated at the post-transcriptional level. To reveal the biological pathways associated with the differentially expressed genes, we performed GO enrichment analysis. We found that the up-regulated genes in ASD progenitors were significantly enriched for GO category of brain development, which agrees with the observations of the high NPC proliferation rate and differentiation defects in ASD cultures (Figure S5C, D). At the neuronal stage, the up-regulated genes in ASD were enriched for the GO categories related to extracellular matrix, whereas the down-regulated genes were significantly enriched for the GO categories of cilium and axoneme, consistent with the observed synaptic dysregulation (Figure S5E, F).

To explore the network organization of the transcriptome^{6, 43, 44}, we next applied a Weighted Gene Co-expression Network Analysis (WGCNA) to provide a higher order view of the biological processes altered in the patient cells. Given the substantial transcriptomic alterations in the patient-derived neurons, we mainly focused on determining the gene co-expression organizations at the neuronal stage. We identified 17 gene co-expression modules in the assigned co-expression network (each labeled in a different color). We also observed a striking clustering pattern of the case-associated genes and known ASD susceptibility genes (curated from SFARI database), respectively, upon the identified co-expression structure by plotting the module assignment under the gene clustering dendrogram (Figure 5C, D and Figure S5E, F).

Comparing the expression patterns of the modules between ASD and controls (see Methods), we found 4 modules that were significantly associated with ASD status (FDR <0.1): brown, tan, purple, and magenta (Supplementary Table S10). Remarkably, known autism candidate genes were significantly enriched in these autism-correlated modules (Supplementary Table S11). Genes that were commonly affected by ASD-associated CNVs were overrepresented in the brown module (OR=1.5, enrichment p-value=0.02), whereas the tan module was enriched for genes known to be affected by ASD-associated protein-disrupting and missense rare de novo variations (RDNVs) (OR=1.6, p-value=0.02). Because co-expressed genes imply co-regulation, our results provide further evidence for the suggested convergent transcriptional regulation in ASD⁸.

Since ASD NPCs exhibited abnormal differentiation phenotypes, we evaluated whether there were dynamic expression changes in patient cells upon differentiation. Our analysis identified 35 genes showing significant differences in their expression trajectories between ASD NPCs and controls in the progenitor to neuron transition (p<0.005) (Figure 5E). These genes were significantly enriched for voltage-gated cation channels (Figure 5F), including CACNG5, KCNA6, KCNC2, and KCNIP2. As shown in Figure 5G and H, those channel genes were up-regulated in control cells, but this increase was attenuated in the ASD cells. These data further implicate voltage-gated cation channel dysregulation in ASD neural differentiation, consistent with decreased excitatory glutamatergic synapses, likely affecting synaptic transmission.