Cellular and molecular characterization of multiplex autism in human induced pluripotent stem cell-derived neurons

The daughter was born at term with no complications or dysmorphia. Her language and motor development were typical and she was able to read at an early age. By age five she was reading at a fifth-grade level. She has been described as talented in writing and drawing. According to her parents, she exhibited social oddities from an early age, mainly in communication, and has somewhat intense/restricted interests in fantasy games. She has strong language abilities, and currently attends a 4-year college, but at times uses odd phrases and the rhythm of her speech includes irregular pauses. She has described feeling alienated and “different”, and was the victim of bullying in middle school, with few close friends. In late adolescence, she developed major depressive disorder with moderate severity, which brought her to first psychiatric contact. She is cognizant of some degree of social awkwardness, which leads to feelings of anxiety and self-consciousness. The social anxiety inhibits her from activities such as eating in the cafeteria and pursuing job opportunities for which she is otherwise well-qualified. She has a history of becoming emotionally dysregulated and overwhelmed in times of stress, which has led to self-injurious behaviors. She has had ongoing struggles with depressive decompensation and suicidal ideation. She has above-average intelligence but has struggled academically in college due to depression and anxiety. She is medically healthy with the exception of supraventricular tachycardia secondary to atrioventricular node reentry, which was treated with ablation and resulted in subsequent normalization of her electrocardiogram.

The twin boys were born at 35 weeks, had breathing problems at birth, and spent 10 days in the newborn intensive care unit. Neither child has any dysmorphic features nor congenital medical abnormalities, and brain imaging studies were negative. Likewise, neither child has a history of confirmed seizures; however, there are concerns for possible absence epilepsy. There is no history of abnormal neurological examination or macro-/microcephaly. Development of both siblings was delayed, but neither had appreciable regression. The more severely affected twin (designated as the affected proband, AP, and from whom the induced pluripotent stem cell (iPSC) model of severe ASD affectation was acquired) began to exhibit delays in development by 9 months of age. He was speaking single words at 14 months and was ultimately diagnosed with autism at 3.5 years old. Research confirmation of the diagnosis was obtained using the Autism Diagnostic Observation Schedule. Compared to his twin brother, he has had more perseverative interests on odd objects. Psychological testing at the age of nine revealed an IQ of 65 using the Leiter International Performance Scale. Now in late adolescence, he has the ability to engage in reciprocal and meaningful verbal exchanges, although his language is often echolalic and repetitive. He is socially motivated and develops superficial friendships with peers. Functionally, he is able to complete most self-care, dress himself, prepare food and feed himself, and count money. He participates in a vocational program at school and is able to complete rudimentary tasks assigned to him. His monozygotic twin was also diagnosed with ASD at 3.5 years old and is less severely affected, but still requires significant support. Selected clinical characteristics of the family members studied are summarized in Table 1.

Standard trio ES was performed by the clinical diagnostic laboratory GeneDx for the unaffected mother (UM), the AP, and the unaffected father. As described by GeneDx, exons were captured and sequenced on an Illumina platform with at least 100 base pair read length, followed by alignment to human genome build GRCh37/hg19 and subsequent variant identification. GeneDx’s custom-developed tool, Xome Analyzer, was used for variant analysis, as described [32]. This process involves comparing the sequences of each individual to a number of resources, such as published reference sequences, other family members, and control individuals, including the 1000 Genomes database, NHLBI Exome Sequencing Project, ExAC, gnomAD, OMIM, PubMed, and Clinvar. Variant annotations include evolutionary conservation scores, results of in silico prediction tools, and references from the published literature. A phenotype-based approach is also used to generate candidate gene lists. This information was interpreted by GeneDx experts according to the American College of Medical Genetics and Genomics Guidelines. ES was performed by The Genome Technology Access Center (GTAC) at Washington University on the intermediate phenotype sister (IS), and variant-specific testing was performed by GeneDx on the trait-affected brother (TB) to confirm the presence of the identified GPD2 variant.

iPSC lines were generated by the Genome Engineering and iPSC Center (GEiC) at Washington University. Biomaterials for reprogramming were only available from the UM, IS, and AP. Briefly, renal epithelial cells were isolated and cultured from fresh urine samples and were reprogrammed using a CytoTune-iPS 2.0 Sendai Reprogramming kit (Thermo Fisher Scientific), following the manufacturer’s instructions. At least three clonal iPSC lines were derived for each subject, and one or two of these clonal lines (clones 1 and 2) were used for all experimentation involving the UM, IS, and AP. The UC line was previously derived by the GEiC, and one clonal line was available for use in all experiments involving this cell line. All clones (clones 1 and 2) used in experiments were assessed for karyotypic abnormalities by the Washington University School of Medicine Cytogenetics and Molecular Pathology Laboratory, and were also characterized for pluripotency by immunocytochemistry (ICC) and RT-qPCR. Each statistically significant experimental finding reported here was made in experiments that used two different clonal lines per individual (except for the UC, where only one clonal line was available), with at least three independent biological replicate experiments performed per clonal line. Statistical comparisons were made by one-way ANOVA or unpaired t test. Documentation of the clone used for each replicate experiment, the replicates performed, and the statistics for each finding is detailed in Additional file 1: Table S1.

iPSC lines were grown under feeder-free conditions on Matrigel (Corning) in mTeSR1 (STEMCELL Technologies). cExN and cIN differentiation of iPSCs was performed using previously described protocols [33]. Briefly, for cExN differentiation, iPSCs were dissociated to single cells with Accutase (Life Technologies) and 40,000 cells were seeded in V-bottom 96-well non-adherent plates (Corning). Plates were spun at 200×g for 5 min to generate embryoid bodies (EBs) and were incubated in 5% CO₂ at 37 °C in cExN differentiation medium with 10 μM Y-27632 (Tocris Biosciences). cExN differentiation medium components include Neurobasal-A (Life Technologies), 1X B-27 supplement (without Vitamin A) (Life Technologies), 10 μM SB-431542 (Tocris Biosciences), 100 nM LDN-193189 (Tocris Biosciences). On day four of differentiation, EBs were transferred from V-bottom plates to Poly-l-Ornithine- (20 μg/ml) and laminin- (10 μg/ml) coated plates. Media (without Y-27632) was replenished every other day, and on day 12, Neural Rosette Selection reagent (STEMCELL Technologies) was used to select neural progenitor cells (NPCs) from within neural rosettes, per the manufacturer’s instructions. cExN NPCs were grown as a monolayer using cExN differentiation media for up to 15 passages. cIN differentiation media contained the same components as cExN differentiation media, while also including 1 μM Purmorphamine (Calbiochem) and 2 μM XAV-939 (Tocris Biosciences). EBs were generated as described for cExNs. At day four of differentiation, the EBs were transferred to non-adherent plates and were placed on an orbital shaker (80 rpm) in an incubator with 5% CO₂ at 37 °C. The media was replenished every other day and, at day ten, EBs were transferred to Matrigel- and laminin- (5 μg/ml) coated plates. Y-27632 was included in media until day eight of differentiation. On day 12 of differentiation, NPCs were dissociated with Accutase and maintained as a monolayer for up to 15 passages. For both cIN and cExN NPC growth analysis, an equal number of cells were seeded on Matrigel- and laminin- (5 μg/ml) coated plates and total cells were counted 4 days later when the cells reached 70–80% confluence.

For differentiation of cExN NPCs into neurons for maturation, 40,000 cells per well were seeded in V-bottom 96-well non-adherent plates. Plates were spun at 200×g for 5 min and incubated in 5% CO₂ at 37 °C in maturation medium with Y-27632. Maturation medium components include Neurobasal-A, 1X B-27 supplement (without Vitamin A), 200 μM cAMP (Sigma), 200 μM Ascorbic acid (Sigma), and 20 ng BDNF (PeproTech). After 2 days, EBs were transferred to Matrigel- and laminin- (5 μg/ml) coated plates and media was replenished every other day (without Y-27632). On day 12 of neuronal differentiation and maturation, cells were dissociated with Accutase and seeded in an eight-well chamber for ICC.

For neurosphere size measurement analysis, P values *P < 0.05, **P < 0.01, ***P < 0.001 were determined by one-way ANOVA.

For ICC, cells were plated on eight-well chamber slides coated with Matrigel and laminin (5 μg/mL). After 1 day, cells were washed once with PBS without calcium and magnesium (PBS - Ca²⁺/Mg²⁺) and fixed in 4% paraformaldehyde for 20 min, followed by washing with PBS + Ca²⁺/Mg²⁺. Cells were blocked with blocking buffer (10% donkey serum, 1% BSA, and 0.1% TritonX-100 in PBS + Ca²⁺/Mg²⁺) for at least 1 h and incubated with primary antibodies overnight (Additional file 1: Table S1) in antibody dilution buffer (1% donkey serum, 1% BSA, and 0.1% TritonX-100 in PBS + Ca²⁺/Mg²⁺). After overnight incubation, cells were washed three times with wash buffer (0.1% Triton X-100 in PBS + Ca²⁺/Mg²⁺). Cells were incubated with corresponding secondary antibodies (Additional file 1: Table S1), along with DAPI (1 mg/mL; ThermoFisher Scientific), diluted in antibody dilution buffer for 1 h. Following secondary antibody incubation, cells were washed twice with wash buffer and once with PBS + Ca²⁺/Mg²⁺. Slides were mounted with Prolong Gold anti-fade agent (Life Technologies). Images were obtained using a spinning-disk confocal microscope (Quorum) with MetaMorph software and were processed using ImageJ. For immunoblotting, cell lysate was extracted and 30 μg of protein was used per lane. Antibodies used are listed in Additional file 1: Table S1.

For each experiment, approximately one million cells were pelleted, washed with PBS – Ca²⁺/Mg²⁺, resuspended in PBS – Ca²⁺/Mg²⁺ and fixed by adding 70% ice-cold ethanol dropwise while vortexing. Cells were stained with 10 μg/mL propidium iodide (Sigma) and 200 μg/mL RNase A (Fisher Scientific) in FACS buffer (PBS – Ca²⁺/Mg²⁺, 0.2% BSA, 1 mM EDTA). FACS was performed on single-cell suspensions and the cell cycle analysis function of FlowJo was used to analyze cell cycle composition for each sample, based on propidium iodide staining to detect DNA content in each cell. P values *P < 0.05, **P < 0.01, ***P < 0.001 were determined by one-way ANOVA.

Total RNA was collected from iPSC-derived day 12 cExN and cIN NPCs using the NucleoSpin RNA II kit (Takara) per the manufacturer’s instructions. RNA was quantified using a NanoDrop ND-1000 spectrophotometer (Thermo Scientific), and the integrity of RNA was confirmed with an Agilent Bioanalyzer 2100 to ensure a RIN value above eight. RNA-sequencing (RNA-seq) library preparation and Illumina sequencing were performed by the GTAC at Washington University. Single-end 50 base pair reads were obtained using an Illumina HiSeq 3000 sequencer, obtaining an average of ~30 million uniquely aligned reads per sample. For RT-qPCR, 1 μg total RNA was reverse transcribed using iScript Reverse Transcription Supermix (Bio-Rad). Equal quantities of cDNA were used as a template for RT-qPCR, using the Applied Biosystems Fast Real-Time quantitative PCR system. RPL30 mRNA levels were used as endogenous controls for normalization. P values *P < 0.05, **P < 0.01, ***P < 0.001 were determined by an unpaired t test.

RNA-seq data analysis was performed as described in [33] to curate differentially expressed gene (DEG) lists. In summary, RNA-seq reads were aligned to the human genome (assembly hg38) with STAR version 2.5.4b [34]. Gene counts were derived from the number of uniquely aligned unambiguous reads by Subread:featureCount, version 1.6.3, with GENCODE gene annotation (V27) [35, 36]. All gene-level transcript counts were then imported into the R/Bioconductor package DESeq2 [37]. Genes expressed below a CPM of 1.0 in more than half the samples were excluded from further analysis. DEG cutoffs were set at a log2-fold change of > 1.0 and a Benjamini and Hochberg FDR of < 0.05.

For this correlation analysis, the read counts matrix for DEGs in either the cIN or cExN samples was log2-CPM transformed. A similarity matrix for these genes was created by calculating the Euclidian distance among the genes from the log2-CPM matrix in R. The WGCNA R package “adjacency.fromSimilarity” function with arguments power=12, type=’signed’ was used to create an adjacency matrix from the similarity matrix [38]. Hierarchical clustering was performed using the hclust function in R on the distance matrix transformation of the adjacency matrix using the ward.D2 method. The dendrogram of this gene tree was split into k = 3 clusters using the cutree function in the stats package of R. The read counts matrix for all genes in either the cIN or cExN samples was CPM transformed. Genes expressed at < 1 CPM in over half of the samples were excluded. Average CPM among the four sample types was calculated for each gene, and genes with expression ratios of < 1.25 between the highest and lowest expressed sample type were excluded from further analysis. The filtered gene CPM matrix was log2 transformed, and the Pearson correlation matrix for the remaining genes was calculated using the cor function from the stats package in R. A gene was considered to be correlated to a DEG cluster if it had a Pearson correlation coefficient > 0.7 for over half the genes in the DEG cluster. Genes with expression correlated to a cluster of genes from the DEG clustering were merged with those genes and fed into the ToppFun GO analysis of ToppGene [39]. For each correlated cluster, up to five terms with the most significant Benjamini and Hochberg FDR values were retained for each biological process, cellular component, and molecular function GO terms. The top three pathway and disease terms were also retained for each cluster. To assess the impact of different covariates on expression, the R/Bioconductor package variancePartition was utilized [40]. The parameters assessed were cell type (cExN vs. cIN), subject (UC, UM, IS, AP), age (young vs. old), or sex (male vs. female).

The multiplex autism spectrum disorder (ASD) pedigree selected for study (Fig. 1; Table 1) underwent clinical phenotyping and genotyping (see Methods). From this pedigree, the individuals selected for iPSC line derivation and modeling included the affected proband (AP), his sister, who has an intermediate phenotype (IS), and their unaffected mother (UM) (indicated in Fig. 1). As described in the Methods, a non-synonymous single nucleotide variant in the GPD2 gene was identified in the UM and all of the children (chr2:157352686 (hg19) G>A, NM_001083112.2 c.233G>A, p.G78E). This variant is not present in the father. The variant is in exon three of the GPD2 gene, within the region encoding the flavin adenine dinucleotide (FAD)-binding domain of the GPD2 protein (Additional file 2: Figure S1A). The GeneDx interpretation of this variant states that it was not observed in approximately 6500 individuals of European and African American ancestry in the NHLBI Exome Sequencing Project and that it is evolutionarily conserved. In addition, in silico analysis predicts that this variant is probably damaging to the protein structure and function. Overall, GeneDx designated this as a variant of uncertain significance (VUS) following the American College of Medical Genetics criteria. Subsequent mutation-specific testing and Sanger sequencing identified this same variant in the AP, IS, and UM, while it was absent in an unrelated, unaffected control (UC) (Additional file 2: Figure S1B). Given the differential affectation of members of this pedigree carrying this variant, it was apparent that this inherited VUS was insufficient to cause ASD by itself, but may have contributed to polygenic risk/liability in this family.

Two clonal iPSC lines were derived from the UM, IS, and AP, and were characterized in parallel with a single, clonal iPSC line derived from the UC. When grown in stem cell maintenance media, there were minimal differences observed in the expression of pluripotency markers between these iPSC lines, as assessed by RT-qPCR (Additional file 2: Figure S1C) and immunocytochemistry (ICC) (Additional file 2: Figure S1D). In addition, no differences in GPD2 protein levels were detected in iPSCs by ICC (Additional file 2: Figure S1D) or Western blotting (Additional file 2: Figure S1E). All cell lines used in this study were also shown to be karyotypically normal (Figure S1F). Finally, we used FACS analysis of propidium iodide (PI)-stained iPSCs to assess cell cycle progression and detected no observable differences between these iPSC lines, which had similar percentages of cells in each stage of the cell cycle (Additional file 2: Figure S1G-H).

In the cortex, as a result of their abnormal development, imbalances in glutamatergic excitatory neurons (cExN) and GABAergic inhibitory interneurons (cINs) are thought to contribute to neurodevelopmental disorders including ASD [3, 5, 41]. We, therefore, differentiated iPSCs derived from the UC, and from three family members (the UM, IS, and AP), in parallel into either cExN or cIN neural progenitor cells (NPCs) and/or neurons, to determine if we observed any alterations in the in vitro development of either or both of these neural cell types. We performed 12 days of cExN differentiation (4 days as embryoid bodies (EBs) in V-bottom plates, followed by 8 days with the EBs plated for two-dimensional (2-D) culture; Fig. 2a). At all time points assessed during this differentiation, the IS and AP lines generated significantly smaller neurospheres than the UC and UM. The UM neurospheres were also slightly smaller than those of the UC (Fig. 2b, c).

To identify whether an increase in apoptosis and/or a decrease in proliferation could be contributing to these differences in neurosphere size, we performed FACS analysis of PI-stained cells at day four of differentiation and found that the IS and AP neurospheres had a significantly higher fraction of sub-G1 (apoptotic) cells, compared to neurospheres derived from the UM and UC lines (< 2 N DNA content; Fig. 2d, e, g). There was a corresponding decrease in the percentage of cells in the G1 phase of the cell cycle in the IS and AP neurospheres (2 N DNA content; Fig. 2d, f–g). However, neurospheres from all lines had similar percentages of cells in the S and G2/M phases of the cell cycle, suggesting that their cell cycle characteristics and rates of progression were otherwise similar (S phase and 4 N DNA content; Fig. 2d, g). To determine whether induction of neural differentiation was a stressor that was contributing to this increase in apoptosis in the IS and AP line-derived neurospheres, we compared sphere size after culturing spheres from each line either in stem cell maintenance media (mTeSR) or in neural induction media. In general, sphere size was larger for all cell lines when kept in mTeSR media rather than neural induction media, while the differences in sphere size for the IS and AP versus the UC and UM was much less pronounced in mTeSR, relative to differences seen under neural induction conditions (Fig. 2h, i). These data suggest that, by comparison with the UC and UM, the IS and AP lines have a slightly elevated propensity to undergo apoptosis upon dissociation and sphere formation, while this is exacerbated by induction of neural differentiation.

We next maintained these four lines as NPCs after neural rosette selection at day 12 and then subjected them to PI staining and FACS analysis (Additional file 2: Figure S2A). Unlike the results from earlier time points, the cExN NPCs showed little difference in cell cycle across the four lines (Additional file 2: Figure S2B-C), nor in the rate of growth or apoptosis over the course of culture for 4 days (Additional file 2: Figure S2D). However, morphological analysis by bright-field imaging indicated a possible adhesion defect in the IS NPCs, as indicated by uneven growth on the cell culture plate surface (Additional file 2: Figure S2E). At the NPC stage, GPD2 protein levels remained similar across the four lines, as was shown for iPSCs (Additional file 2: Figure S2F).

Finally, to determine if the NPCs derived from the affected individuals exhibited an altered capacity to differentiate into cExN neurons, NPCs from the four lines were further differentiated for 12 days as shown (Additional file 2: Figure S3A) and subjected to ICC. No apparent differences between the four lines were observed in the expression of NPC markers (PAX6, NESTIN, and SOX1) or markers of immature (TUJ1) and mature cExN neurons (VGLUT, MAP2) (Additional file 2: Figure S3B). Furthermore, there were no observable differences between the lines in the fraction of cells expressing Ki-67, a marker of cell proliferation, or cleaved Caspase-3, a marker of apoptosis (Additional file 2: Figure S3B).

We also characterized cellular phenotypes of these four lines during differentiation into cIN NPCs, to define any differences between the development of this neural cell type in lines derived from affected and unaffected individuals. The differentiation scheme to produce cIN NPCs is outlined in Fig. 3a. On day five of differentiation in this scheme, neurospheres derived from the IS line were smaller than those of the UM. Conversely, the AP line-derived neurospheres were slightly larger than the UM line neurospheres (Additional file 2: Figure S4).

After dissociation on day 12 of differentiation, we assessed the cell cycle of the cIN NPCs using FACS of PI-stained cells (Fig. 3b–e). The IS and AP cIN NPCs had an increased sub-G1 cell population, compared to the UC and UM NPCs, an indication of increased apoptosis in the cells from the affected individuals (Fig. 3c). Correspondingly, there was a decrease in the proportion of cells in the G1 phase of the cell cycle (Fig. 3d). However, no differences were observed in frequencies of cells in the S and G2/M phases of the cell cycle between lines, suggesting that these lines had similar proliferation rates (Fig. 3b, e). This result was supported by analysis of NPC cell counts after 4 days of growth, which revealed a significant reduction in the number of AP NPCs, as well as a slight reduction in the number of IS NPCs, compared to the UM NPCs (Fig. 3f). These reductions in NPC number may have resulted from the increased NPC apoptosis detected in our PI FACS analysis. The AP NPCs also exhibited altered morphology that could indicate impaired adhesion capacity relative to the control UM/UC lines, which could also have contributed to the reduction in the number of AP NPCs persisting in the culture after 4 days of growth (Fig. 3g).

To investigate which classes of genes could be differentially expressed in neural cells from the affected individuals, by comparison with the unaffected controls, we performed RNA-seq analysis on both cExN and cIN NPCs on day 12 of differentiation for all four subject-derived lines. Four biological replicates were analyzed for each sample type and were clustered by principal component analysis (PCA) of processed reads (Additional file 2: Figure S5A). We defined genes that were significantly differentially expressed genes (DEGs) in pairwise comparisons of these four sample types for either cExN or cIN NPCs, selecting DEGs with a log2-fold difference between sample types of > 1 and a Benjamini and Hochberg FDR of < 0.05 (Additional file 3: Table S2). In a within-family comparison of the UM, IS, and AP samples, greater numbers of DEGs were obtained in the cIN NPC pairwise comparisons versus the numbers of DEGs obtained for cExN NPC pairwise comparisons (Additional file 2: Figure S5B-C). These data indicate that the cIN samples from the affected individuals (IS/AP) exhibit more transcriptomic differences from the UM control than the affected individual-derived cExN samples.

We focused first on identifying classes of genes that were differentially expressed in NPCs derived from the affected individuals, by comparison with unaffected controls. To do this, we defined the subset of DEGs that were similarly expressed in samples from both affected individuals (AP/IS) but that differed in expression by comparison with the unaffected mother (UM) sample. Relative expression is also shown for the UC, for a full cross-sample comparison. Four hundred fifty-two and 437 DEGs for the cExN and cIN NPC samples met these criteria, respectively. Hierarchical clustering and visualization of the relative expression of these DEGs across the four sample types are shown for the cExN NPCs (Fig. 4a, Additional file 4: Table S3). We next used the Ingenuity Pathway Analysis (IPA) to assess the potential biological significance of these genes. For the 452 DEGs in the cExN NPCs described above, the most significant function- and disease-related gene ontology (GO) terms included ‘behavior’, ‘neurological disease’, and ‘embryonic development’ (Fig. 4b). Network analysis using IPA revealed several interesting networks of DEGs related to these GO terms, including networks related to ‘locomotion’ (from DEGs within the ‘behavior’ GO term) and ‘behavior and developmental disorder’ (Fig. 4c, d). Within the ‘locomotion’ network, most genes were upregulated in the affected individuals compared to the controls, including genes relating to neural adhesion and ion channels (Fig. 4c and Additional file 5: Table S4). Genes with known roles in NPCs or neurons, as well as stress-related genes were present in the larger ‘behavior and developmental disorder’ network (Fig. 4d and Additional file 5: Table S4). Interestingly, another network comprising genes from the GO term ‘neurological disease’ is related to ‘inflammation of central nervous system’ (Additional file 2: Figure S6A and Additional file 5: Table S4).

IPA analysis of the cIN DEGs also revealed several interesting classes of genes that were differentially expressed between the affected participants and unaffected control NPC-derived samples. Hierarchical clustering and visualization of the relative expression of DEGs across the four sample types for the cIN NPCs are shown in Fig. 4e (Additional file 4: Table S3). The top GO terms included ‘developmental disorder’, ‘behavior’, ‘nervous system development and function’, ‘psychological disorders’, and ‘neurological disease’ (Fig. 4f). Within the term ‘behavior’, a network that includes ‘learning’-, ‘cognition’-, and ‘behavior’-related genes was identified (Fig. 4g, Additional file 5: Table S4). The network related to the GO term ‘psychological disorder’ includes genes related to ‘anxiety disorders’, ‘mood disorders’, and ‘depressive disorder’ (Fig. 4h). The ‘nervous system development and function’ network includes genes involved in the ‘quantity of neurons’ and ‘quantity of synapse’, as well as cell adhesion genes (Additional file 2: Figure S6B and Additional file 5: Table S4). Finally, a ‘neurological’ network included a number of genes also present in the other networks (Additional file 2: Figure S6C). We further assessed these affectation-linked DEGs by gene co-expression analysis and hierarchical clustering, as described in the methods (Additional file 2: Figure S7). Gene ontology analysis of co-expressed gene clusters identified neuron-related (e.g. neuron projection, axon, synapse) and neurological disease-related terms (e.g. mental depression, autistic disorder) as top GO terms (Additional file 6: Table S5). Taken together, this analysis of DEGs in both cExN and cIN NPCs shows evidence of altered expression of a number of neurological and psychological disease-relevant gene classes in the AP- and IS-derived lines, relative to lines derived from the UM and/or UC.

As differences in genetic background can confound differential gene expression analysis [42], we also performed a pairwise, within-family data comparison of DEGs that distinguish the UM-, IS-, and AP-derived samples, focusing on DEGs specific to the AP that could contribute to the greater degree of affectation observed. Using pairwise comparisons of DEGs, we defined 190 genes which were uniquely differentially expressed in cExN NPCs from the AP (Fig. 5a). The top GO terms associated with these AP-specific DEGs included ‘psychological disorders’, ‘behavior’, ‘nervous system development and function’, ‘developmental disorder’, and ‘neurological disease’ terms (Fig. 5b). Within the ‘behavior’ term, network analysis showed genes related to ‘memory’ and ‘learning’ to be dysregulated (Fig. 5c and Additional file 5: Table S4). Within the ‘nervous system development and function’ term, a network of dysregulated genes related to ‘differentiation of neurons’ was identified (Fig. 5d, Additional file 5: Table S4).

A similar analysis was performed on the cIN samples, revealing 384 DEGs unique to the AP samples in the within-family comparison (Fig. 5e). IPA analysis identified classes of DEGs related to the GO terms ‘psychological disorders’, ‘developmental disorder’, ‘neurological disease’, ‘behavior’, and ‘nervous system development and function’ (Fig. 5f). Within the ‘behavior’ disease term, a network of genes related to ‘behavior’ and ‘cognition’ was identified (Fig. 5g and Additional file 5: Table S4). Within the ‘nervous system development and function’ term, a network of genes related to ‘development of neurons’ and ‘synaptic transmission’ had altered expression in the AP versus the IS/UM-derived samples (Fig. 5h and Additional file 5: Table S4). Together, this analysis identified AP-unique DEGs in both cExN and cIN NPCs, many of which are broadly related to neural development, as well as to specific aspects of ASD, such as behavioral alterations. These gene expression changes, therefore, correlate with and may contribute to, the severity of affectation in the AP.

The Simons Foundation Autism Research Initiative (SFARI) [43] maintains a database of genes that are mutated to cause or that contribute to ASD risk. We compared our DEGs to these ASD-related genes, to assess whether their dysregulated expression could contribute to affectation in these individuals. Of the 584 unique DEGs in the cExN differentiation scheme that were either specific to the AP (Fig. 5a) or that had similar expression in the AP and IS that differed from that seen in the UM (Fig. 4a), 30 (5.1%) were SFARI ASD genes (Fig. 6a; Additional file 6: Table S5). For the corresponding cIN NPC comparison, 48 of 692 unique DEGs (6.9%) were ASD genes in the SFARI Gene database (Fig. 6b; Additional file 6: Table S5). Based upon the 1019 genes present in the SFARI Gene database [43] and the total of 27,731 genes with > 0.1 RPKM average expression across all cExN and cIN samples, the number of AP- and IS-specific DEGs that are ASD genes is significantly greater than would be expected by chance (hypergeometric distribution, P = 3.67 × 10⁻³ and 3.49 × 10⁻⁷ for cExN and cIN data, respectively). A subset of these are associated with syndromic ASD (NTNG1, ALDH1A3, DMD, EBF3, PRODH, and RNF135) and/or are linked with ASD with the highest confidence (SFARI gene scores 1–2: KATNAL2, MYT1L, CACNA2D3, GRIA1, SCN9A, and CNTN4). Comparison with the 465 genes in the Geisinger Developmental Brain Disorder Gene Database also revealed recurrent association of some of these genes with ASD and/or intellectual disability, but infrequent association with other neurological disorders (Additional file 6: Table S5) [44]. Interestingly, only one (for cExN) or none (for cIN) of the DEGs overlapped with high-confidence genes unique to adult-onset psychiatric disorders (e.g., not also associated with neurodevelopmental disorders) identified in the PsyGeNET database (Additional file 6: Table S5) [45, 46]. Therefore, it is possible that misregulated expression and consequently the function of ASD genes contributed to the disruption of neural development and/or continues to contribute to altered neurological function in the IS and AP.

We validated the differential expression of a subset of the DEGs described above by RT-qPCR analysis, isolating RNA from NPCs derived from the second set of iPSC clones that differed from those used for the RNA-seq experiments. Expression changes of DEGs selected from the cExN NPC RNA-seq data (Fig. 7a) were robustly recapitulated in these experiments (Fig. 7b). Genes from the ‘behavior and developmental disorder’, from other identified networks, and genes involved in neurodevelopment were evaluated (Additional file 5: Table S4). We also derived cIN NPC RNA from the second set of iPSC clones and validated the corresponding RNA-seq data for a subset of the DEGs (Fig. 7c). Differential expression was assessed for SFARI ASD genes [43], and for genes encoding transcription factors, ion channels, and cell adhesion molecules (Fig. 7d, e and Additional file 5: Table S4). In addition, we validated a subset of DEGs in cINs which also had differential expression in cExN NPCs (Fig. 7e).

Differential gene expression in iPSC disease modeling studies involving female iPSC lines can be confounded by erosion of X-inactivation, including alteration of sex chromosome-linked gene expression [47‐50]. Therefore, we also assessed XIST expression levels, as an indicator of potential X-erosion in our female lines. We observed that XIST RNA expression was reduced in the UC but not in either clonal line derived from the UM or IS, evidence of potential X-erosion in the UC line (Additional file 2: Figure S8A-B). However, as few of our DEGs are X-linked (Additional file 4: Table S3), this does not appear to be a major contributor to the differential gene expression identified in this study. Finally, we performed variancePartition analysis on these RNA-seq data to quantify contributions of multiple sources of variation to the differential gene expression obtained. The cell type (cExN or cIN) or subject from whom the sample was derived were the main contributors to differential gene expression, while the age and sex of the subject were only minor contributors to the total variance observed in both cExN and cIN RNA-seq datasets (Additional file 2: Figure S8C). Together, these analyses revealed that relative to unaffected individuals, samples from affected individuals exhibited altered expression of classes of genes involved in behavior, learning, cognition, mood disorders, and neurodevelopment, including perturbed ASD gene expression, suggesting that these differences could contribute to aberrant neural development or function in the affected individuals.