Schizophrenia is defined by cell-specific neuropathology and multiple neurodevelopmental mechanisms in patient-derived cerebral organoids

Due to an inability to ethically access developing human brain tissue as well as identify prospective cases, early-arising neurodevelopmental and cell-specific signatures of Schizophrenia (Scz) have remained unknown and thus undefined. To overcome these challenges, we utilized patient-derived induced pluripotent stem cells (iPSCs) to generate 3D cerebral organoids to model neuropathology of Scz during this critical period. We discovered that Scz organoids exhibited ventricular neuropathology resulting in altered progenitor survival and disrupted neurogenesis. This ultimately yielded fewer neurons within developing cortical fields of Scz organoids. Single-cell sequencing revealed that Scz progenitors were specifically depleted of neuronal programming factors leading to a remodeling of cell-lineages, altered differentiation trajectories, and distorted cortical cell-type diversity. While Scz organoids were similar in their macromolecular diversity to organoids generated from healthy controls (Ctrls), four GWAS factors (PTN, COMT, PLCL1, and PODXL) and peptide fragments belonging to the POU-domain transcription factor family (e.g., POU3F2/BRN2) were altered. This revealed that Scz organoids principally differed not in their proteomic diversity, but specifically in their total quantity of disease and neurodevelopmental factors at the molecular level. Single-cell sequencing subsequently identified cell-type specific alterations in neuronal programming factors as well as a developmental switch in neurotrophic growth factor expression, indicating that Scz neuropathology can be encoded on a cell-type-by-cell-type basis. Furthermore, single-cell sequencing also specifically replicated the depletion of BRN2 (POU3F2) and PTN in both Scz progenitors and neurons. Subsequently, in two mechanistic rescue experiments we identified that the transcription factor BRN2 and growth factor PTN operate as mechanistic substrates of neurogenesis and cellular survival, respectively, in Scz organoids. Collectively, our work suggests that multiple mechanisms of Scz exist in patient-derived organoids, and that these disparate mechanisms converge upon primordial brain developmental pathways such as neuronal differentiation, survival, and growth factor support, which may amalgamate to elevate intrinsic risk of Scz.

To confirm the reproducibility of our cerebral organoid cultures within a larger cohort (n = 25 unique biological donors), we employed new analytical tandemmass tag (TMT) chemistry that allowed multiplexing of 16 samples (16-plex) with high-resolution quantitative mass spectrometry. Because of our large cohort size and the need to include internal reference pools (see below), we generated organoids from n = 25 lines in two distinct batches that could be tested for 1) batch effects, 2) inter-sample variance, and 3) overall reproducibility between individual iPSC donors within a given group. This allowed for a high-throughput and statistically unbiased evaluation of variation within our organoid cultures across lines and independently generated batches. For each TMT set (16-plex) we applied batch-specific normalization using global external pooled reference standards so that all n = 25 iPSC donors could be combined and examined together. Each 16-plex experiment contained a pooled reference label (134 isobaric label) that consisted of the same peptide digest from organoid lysates. Proteomic analysis of organoids from n = 25 iPSC lines (n = 8 Ctrl iPSC lines and n = 17 Scz iPSC lines) yielded 33,711 identified fragments and 3,910 quantifiable data points that could be used as an unbiased sample for computational analysis of organoid variance and reproducibility. No imputation of missing values was performed in any cohort, statistical q values were generated for each LC/MS data point, and all p values were corrected for multiple hypothesis testing. a, Orthogonal LC/MS analysis split by iPSC line supports organoid reproducibility.
To provide an unbiased evaluation of LC/MS intensities, data distributions are provided for each individual iPSC line. Likewise, to evaluate coefficients of variance between all lines, we generated correlation matrices that revealed a high degree of similarity between all samples within two independent batches of organoids and iPSC donors. This confirms that organoids exhibited comparably low variance between individual iPSC donors. b, Orthogonal LC/MS analysis revealed no major evidence of batch effects. Between batches, there was a high degree of similarity in the LC/MS intensity maps and their density distributions. Critically, the generation of coefficients of variance for LC/MS data supported this interpretation. Namely, box-and-whisker plots revealed that coefficients of variation were less than 2% between batches (black bars denote median; boxes denote upper and lower quartiles; whiskers denote extremes). Analysis of individual batch variance consequently revealed variance of less than 2%. Together, these unbiased computational data confirm that there was no evidence of batch effects between organoids or within groups. c, Orthogonal LC/MS analysis also confirms inter-group reproducibility. Correlation analysis revealed that each batch following pooled reference normalization exhibited similar group variance. In sum, this analytical and unbiased approach to defining the variance within our cerebral organoid cultures is consistent with our ability to replicate and reproduce core phenotypes multiple times throughout our manuscript. This analysis therefore provided an unbiased, and computationally-derived, validation for baseline reproducibility of organoids between-and within-lines, which is consistent with prior analysis that cerebral organoids hold suitable construct-validity as a model of in utero human brain development [1][2][3]. In the main manuscript, we replicated an increase in progenitor cell death via colocalization with cleaved CAS3 ( fig. 1, 4, and 5). However, we also sought to validate the increase in apoptosis via an unbiased, DNA-based, methodology to ensure that apoptosis induction in Scz organoids could be orthogonally validated.

a, Single-cell DNA content supports increased cell death in Scz organoids.
To provide unbiased validation of increased apoptosis in Scz, we subjected a cross-sectional cohort of Ctrl and Scz organoids to a cellular DNA content analysis using Propidium Iodide (PI), flow cytometry, and FlowJo modeling. This allows live and dead cells to be detected based on their DNA content. Notably, when cells undergo apoptosis, DNA becomes fragmented. To reduce sample heterogeneity while promoting power, approximately 10 organoids were pooled per iPSC line (n = 9 total lines, 3 Ctrls and 6 Scz), dissociated into a single-cell suspension, and subjected to single-cell DNA content analysis via highthroughput flow cytometry. This analysis revealed that there was an approximately 3-fold increase in DNA fragmentation in Scz organoid samples (~7% in Ctrls vs ~21% in Scz organoids; see red dotted box in panel a). Additionally, Scz organoids also tended to exhibit a slightly elevated proportion of S-phase cells based on cellular DNA content (see black arrows in panel a). When coupled with an increase in DNA fragmentation, an alteration in S-phase may be indicative of a potential DNA-related mechanism that leads to cell cycle abortion and increased death of proliferating progenitors within Scz organoids. b, Increased pH2AX recruitment is linked to cell death in Scz organoids. H2AX is involved in DNA damage [4] and DNA double-strand break repair [5], and is commonly induced during cellular division if DNA damage/breaks have occurred [6]. Phosphorylation of H2AX (pH2AX, gate p6) indicates the presence of DNA damage machinery [7], and if pH2AX is co-expressed with a cell death factor (e.g. cleaved PARP, gate p4) then DNA damaged dying cells are deemed to have been targeted for death (gate p5). Therefore, based on our neural progenitor cell death and single-cell DNA content analyses, we hypothesized that there would be an increased number of cells exhibiting pH2AX+ (p6) as well as pH2AX+ and PARP+ (p5) expression within Scz organoids. Organoids were sampled as described above from a pseudorandomly selected cross-section comprising n = 7 lines (3 Ctrls and 4 Scz). As expected, we detected an increased percentage of pH2AX+ and pH2AX+/cleaved PARP+ double-positive cells in Scz organoids. This data therefore provides yet further validation of increased cell death in Scz organoids, and suggests that cell death in Scz organoid is related to DNA damage.  Shown are ratio Log2 against Log2 intensity (a), a volcano plot parsing ratio Log2 by -1*Log10 (p-value) (b), and differentially expressed proteins mapped by protein identifiers (c). As can be seen across these panels, differential expression within the Scz organoid proteome was limited to relatively few targets, suggesting specificity in those proteins that were significantly differentially expressed. The two proteins which with high ratio Log2 values relative to Log2 intensity were WASL (N-WASP) and MPV-17, which were the only two unique proteins detected in Scz organoids. Of note, transcripts for these factors were detected in Ctrl and Scz organoids in single-cell sequencing, but post-translational detection via TMT-LC/MS was restricted to Scz organoid samples. This suggests that N-WASP and MPV17 may be translationally repressed in Ctrl but not Scz organoids. While the functions of MPV17 to Scz remain unclear, the localization of this protein to mitochondria may be indicative of mitochondrial alterations in Scz organoids. Indeed, potential mitochondrial dysfunction in Scz postmortem brains has been reported [8], and mtDNA depletion has been reported in a family with psychosis [9]. Our other novel factor, N-WASP (WASL), has a role in cytoskeletal actin assembly that supports dendritic spines [10]. Dendritic spines are typically disrupted in pyramidal layer 3 cells in postmortem Scz tissue [11,12], where expression of WASL also happens to be depleted in postmortem Scz tissue [13]. The post-translational expression of N-WASP (WASL) in Scz organoids therefore suggests that this molecule may play a role in early Scz neuropathology in the developing cerebral cortex. d, High reproducibility of whole-organoid proteomes across groups. Shown is a heat map exhibiting all group and replicate samples, which exhibited both 1) global, 2) inter-group, and 3) inter-line reproducibility per coefficient of variance metrics. This once more provides validation of the high reproducibility and low variance of our experimental organoids. Individual protein variation across the proteome was similar across groups and is provided in Figure 2c. e-f, Gene ontology and KEGG pathway analysis of the organoid proteome. Shown are up-regulated (e) and down-regulated (f) protein sets (f; defined as ±1.0 Log2, p < 0.05) in Scz patient-derived organoids. Y-axes represent gene ontology (biological process, molecular function, and pathway), and X-axes represents enrichment factor (rich factor = amount of differentially expressed proteins enriched/amount of all proteins in background gene set). Size and color of individual bubbles represents amount of differentially expressed proteins enriched in pathway and enrichment significance, respectively. Ctrl: Control, Scz: Schizophrenia.

Figure S4. Further computational analysis of single-cell transcriptomes. a, Down-regulated expression of developmental factors in Scz organoids.
Shown are a selection of four noteworthy developmental factors found to be unbiasedly depleted in Scz organoids. Each of these factors has known roles in brain development and/or neuronal maturation. This is consistent with the remodeling of Scz progenitors away from neuron production ( fig. 1-3). b, Additional KEGG plot of pathway analyses of Scz neurons. Because only our GO pathway analysis of Scz neurons were not shown in fig. 3 for consistency sake, here we present the outcome of an alternative KEGG pathway analysis in neurons. Note enrichment for factors that belong to Pl3K-Akt signaling (a primary neurotrophin/growth factor signaling cascade; see red box).
Note, enrichment does not necessarily depict up-or down-regulation, rather just factor-counts belonging to these biological processes (thus, individual factors in this enriched pathway may be differentially up-or down-regulated, indicating the potential for broader pathway dysfunction). Ctrl: Control, Scz: Schizophrenia.

Figure S5. Cell-specific analysis of proteome targets in Scz organoids. a, Global gene-expression of proteome targets in Ctrl and Scz organoids.
Unbiased analysis of the Scz organoid proteome unveiled that Scz organoids principally differed from Ctrl organoids in the quantity of differentially expressed molecular factors ( fig. 2e). This included novel targets that regulate development (e.g. POU-domain transcription factors, e.g. POU3F2/BRN2) or were implicated in genetic risk for Scz in large-scale GWAS and/or meta-analysis (e.g. PTN, PLCL1, COMT, and PODXL, see main text). We cross-referenced expression of these factors in a global analysis of single-cell transcriptomes and found that gene-expression of the forebrain-specific neuronal transcription factor POU3F2 (BRN2) was down-regulated. Similarly, gene expression of Scz risk candidates PTN, COMT, and to a lesser extent, PLCL1, were also down-regulated. Consistent with proteomics, expression of PODXL remained up-regulated in global analysis of Scz transcriptomes. b, Cell-specific analysis of proteome targets reveals nuanced expression. Next, we sought to resolve cell-specific contributions to expression differences as well as gene-expression levels in Scz progenitors and neurons. Thus, we analyzed our scRNA-Seq dataset by cell type ( fig. 3h-I and S4). This analysis revealed a striking depletion of POU3F2 (BRN2) and PTN in Scz progenitor celltypes, as well as a reduction in Scz neurons (as shown in fig. 3). Similarly, COMT appeared to be more prominently altered in Scz progenitors, relative to Scz neurons. PLCL1 expression was more prominently reduced in Scz neurons relative to Scz progenitors. PODXL gene-expression was up-regulated in both progenitors and neurons. Ctrl: Control, Scz: Schizophrenia.

Figure S6. Orthogonal validation of BRN2 and PTN disease relevance via informatic mining of human Scz studies and Scz database repositories.
To provide further insight into the disease relevance of our candidate rescue factors, PTN and BRN2, we consulted large-scale data repositories that have indexed, pooled, or compiled the results of human brain tissue arrays, GWAS, and other commonly employed techniques in Scz association studies. a, BRN2 and PTN are enriched during prenatal human cortex development.
To orthogonally validate BRN2 and PTN as embryonic active factors in the developing cortex of humans, we mined the BrainCloud TM (dbGaP Accession: phs000417.v2.p1; [14]) dataset that has quantified gene expression in the human brain across the lifespan. This yielded data congruent with enrichment patterns observed in our organoid cultures, whereby BRN2 (ProbeID: hHR008743, EntrezID: 5454) and PTN (ProbeID: hHC023376, EntrezID: 5764) were expressed in the embryonic human PFC. Additionally, relative to expression patterns across the lifespan, BRN2 and PTN are also enriched during human prenatal development (n = 269 human sample time points, including n = 38 human fetal samples; dotted lines reflect an approximate line of best fit). These data thus provide orthogonal validation that PTN and BRN2 are transcriptionally enriched during embryonic development of the human cortex. b-c, An index SNP in proximity to PTN exhibits gene association in Scz. As discussed in the main manuscript, PTN has been associated with Scz. Here we sought to explore this relationship further by leveraging the Schizophrenia Database (SZDB) [15] and Schizophrenia Gene Resource (SZGR) v2.0 [16,17]. Both of these references function as multi-omics data repositories that aggregate data from genetic, transcriptome, and epigenetic Scz association studies. Searching these databases and datasets yielded several positive associations of PTN and BRN2 to Scz. This included an index SNP (rs3735025) in proximity to PTN that exhibited genome-wide association within two of the largest Scz cohorts ever studied (total n = 35,802 [18] and n = 150,064 [19] humans, respectively; see tabulation in b). Additionally, analysis also revealed positive association of both BRN2 and PTN in differential methylation of samples from Scz patients (see [20][21][22] and tabulation in c for dataset summary). When combined, these data provide orthogonal validation that both of our candidate targets maintain disease relevance to Scz in human association studies. PFC: Prefrontal cortex. Figure 5. Our BRN2-Virus is a self-regulating construct designed to be translational suppressed when infected cell-types differentiate into post-mitotic neurons. Thus, sustained overexpression can be induced transiently without continuing to influence neurons. This allowed us to determine in figure 4 that BRN2 influences neuronal differentiation but not survival in Scz organoids. The BRN2-Virus construct was adapted from a previously validated construct, but was modified to contain an EGFP reporter construct for visualization (see technical details in Materials and Methods). Our simplified experimental timeline (a) from Figure 5 is provided along with viral construct (b) for comparison and additional details not included in main text for both brevity and simplicity. Also shown is a highmagnification image from within a 3D organoid infected with BRN2-Virus that visualizes self-regulated translational suppression via loss of EGFP signal in a differentiating progenitor (arrows) and a new-born neuron (arrowheads; c). As expected, BRN2-Virus expression was only observed in progenitors and differentiating cells, but not in mature postmitotic neurons.

d, Confirmation that BRN2-Virus regulates neuron numbers in Scz organoids when data are analyzed by the average of iPSC lines.
For transparency, here we provide an alternative visualization of BRN2-Virus rescue data whereby raw data have been collapsed into the average of iPSC line (each data point here reflects the average of quantifications from an iPSC line). As can be seen in the leftmost panels, BRN2-Virus significantly increased neuron numbers when two independent neuronal antigens (BRN2 and MAP2) were utilized for immunohistochemistry and neuron quantifications. Consistent with the data shown in figure 4, neither progenitor nor neuronal survival were rescued by BRN2-Virus Scz organoids when the average of iPSC lines was analyzed. Thus, as discussed in the main text, BRN2 was shown to regulate neuron numbers but not cell death phenotypes in Scz organoids. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Scale: c = 20µm. Error bars reflect Standard Error of the Mean. Ctrl: Control, Scz: Schizophrenia, EGFP = Enhanced green-fluorescent protein.

Figure S8. Enlarged whole-organoid images of BRN2-Virus modulation of neuron numbers within Scz organoids.
To provide additional visualization, we sought to provide additional wholeorganoid images of the BRN2-Virus rescue effects on neuron numbers in Scz organoids. In the main figures, individual fields were quantified and shown which are representative of both sampling and reported phenotypes. Here we provide whole-organoid images to complement these visualizations, and to exemplify the extent of rescue achieved at the whole-organoid level. Two independent organoids are shown for each treatment group for confirmation of phenotype reproducibility. Note, MAP2 has been presented as an isolated channel for one organoid set for clarity and transparency so that the extent of MAP2 depletion can be observed at baseline in Scz organoids that received Ctrl-Virus and the extent of rescue achieved in Scz organoids following infection with BRN2-Virus. Note that multiple independent organoids have been selected in the manuscript here to provide numerous independent visualizations of core phenotypes and their rescue in Scz organoids. Scale bar: 60µm.

Figure S9. Whole-organoid images of PTN rescue effects in Scz samples.
To provide additional visualization, we sought to provide additional wholeorganoid images of PTN rescue effects in Scz organoids. In the main figures, individual fields were quantified and shown which was representative of both sampling and reported phenotypes. Here we provide whole-organoid images to complement these visualizations, and to exemplify the extent of rescue achieved at the whole-organoid level.

a, Whole-organoid visualization of progenitor cell death in Scz organoids at baseline and following PTN treatment.
Two independent organoids are shown for each treatment group for confirmation of phenotype reproducibility. Note, cleaved CAS3 has been presented as an isolated channel for one organoid set for additional transparency. b, Whole-organoid visualization of neurons in Scz organoids at baseline and following PTN treatment. Once more, two independent organoids are shown for each treatment group for confirmation of phenotype reproducibility. MAP2 has been presented as a split channel for one organoid set for transparency. Note that multiple independent organoids, and fields within them, were shown across the manuscript to provide numerous independent visualizations of core phenotypes and their rescue in Scz organoids. Scale bar: 60 µm

Figure S10. PTN improves new-born cell survival in Scz organoids.
In figure 5, we show that PTN-supplemented Scz organoids exhibited increased progenitor survival and neuronal differentiation. In addition to these phenotypes, we also assessed the overall survival of new-born cells labeled with BrdU as per our pulse-chase paradigm (see fig. 1 and 5, as well as our schematic above). PTN-supplemented Scz organoids exhibited significantly decreased new-born cellular death (Vehicle Ctrl n = 71 fields, n = 24 organoids, from n = 4 independent Ctrl iPSC lines, PTN-Supplemented Ctrl n = 95 fields, n = 30 organoids, from n = 4 independent Ctrl iPSC lines; Vehicle Scz n = 69 fields, n = 24 organoids, n = 4 independent Scz iPSC lines, PTN-Supplemented Scz n = 89 fields, n = 26 organoids, from n = 4 independent Scz iPSC lines). All graphed data represent unique, non-overlapping, fields. Together with data in fig. 5, these data indicate that in Scz organoids PTN exerts putative neurotrophic-like effects by promoting cellular survival. Thus, in Scz organoids, reconstitution of PTN as an exogenous morphogen facilitated progenitor survival and neurogenesis that resulted in the rescue of neuron numbers in Scz cortical fields. ****p < 0.0001. Scale: 60µm, Error bars reflect Standard Error of the Mean. Ctrl: Control, Scz: Schizophrenia.

Figure S11. Analysis of PTN rescue phenotypes by the average of iPSC lines within groups, including graphing of raw data split by iPSC line.
For transparency, visualization of Scz core phenotypes (both at baseline and following PTN treatment) have been provided here as an average of lines. This allows the relative consistency of Scz phenotypes to be assessed by the average of each patient iPSC line both with and without PTN treatment. As can be seen, phenotypes were broadly reproducible across individual lines, and all patient lines included in the PTN rescue experiment exhibited consistency in phenotype via restoration and/or improvement following PTN treatment. a, Donor-split analysis of progenitor death in Scz organoids. Analysis of progenitor death confirmed that Scz organoids exhibited increased SOX2+ cell death in ventricular zones, and that PTN treatment could negatively regulate SOX2+ cell death in Scz organoids. In the iPSC donor-split analysis, each data point represents the average of each single iPSC donor.  provided from the NIMH beyond the disclaimer that these individuals comprise more than one race. Mixed races were further categorized as "Hispanic" vs. "Non-Hispanic" or "Null" (missing or unknown).