Convergence of schizophrenia risk genes on CALN1 disrupts human forebrain development and drives core psychotic behaviors including hallucination-like perception

Abstract

Background

Schizophrenia, a highly heritable neurodevelopmental disorder characterized by core clinical symptoms such as hallucination, remains poorly understood its polygenic pathogenesis.

Methods

We selected 11 key schizophrenia risk genes and generated single-gene-knockout-precise-dorsal/ventral-forebrain-organoids via CRISPR-Cas9 system to investigate the polygenic effects of these schizophrenia risk genes on forebrain development via bulk and scRNA-sequencing, multi-electrode array recording and knockout mouse model.

Results

We found that knockout of 11 risk genes leads to different levels of deficits in human dorsal/ventral forebrain organoids. Notably, CALN1 knockout significantly impacted pathways involved in forebrain development and the set of prioritized schizophrenia risk genes. Importantly, Caln1-KO mice exhibited a broad spectrum of schizophrenia-like behaviors, encompassing deficits in spatial memory, social ability and pre-pulse inhibition, alongside spontaneous startle behavior, head-twitch response and hallucination-like perception. At the cellular level, Caln1-deficient mice exhibit abnormal spontaneous abrupt burst spiking in mature layer-5 pyramidal neurons of the prefrontal cortex, and snRNA-seq analysis of Caln1^−/− mice revealed severe transcriptional and gene network dysregulation in layer-5 excitatory neurons.

Conclusions

This study reveals that CALN1 as a pivotal role in schizophrenia risk genes robustly perturb human forebrain development and cause a broad spectrum of schizophrenia behaviors in mice.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12916-026-05039-9.

Background

Schizophrenia (SCZ) is a highly heritable neurodevelopmental disorder including impairments of cognition, perception and motivation that affects about 1% of the world’s population [1–3], and previous epidemiological studies have revealed that the heritability of this illness was estimated to be about 80% [4, 5]. Recent advances in psychiatric genetics have identified 287 significant genomic loci through genome-wide association study (GWAS) involving 76,755 schizophrenia patients and 243,649 controls [6]. Based on fine-mapping and functional genomic analysis for schizophrenia-associated common variants, 106 protein-coding genes were prioritized as schizophrenia risk genes and these genes were enriched expression in excitatory and inhibitory neurons, opening an avenue to understand schizophrenia pathogenic mechanisms. While some risk genes like GRIN2A have established roles in neural function [7, 8], but the neurobiological mechanisms underlying most risk genes remain elusive. Deciphering how these risk genes disrupt human brain function to drive SCZ pathogenesis remains a critical challenge in neuroscience.

The human forebrain, a critical mediator of complex cognitive functions, sensory processing and motor coordination, exhibits well-documented structural and functional abnormalities in schizophrenia patients [9–12]. While access to developing human brain tissue remains challenging, human pluripotent stem cells-derived brain organoids have emerged as powerful models for investigating neurodevelopmental processes and neuropsychiatric disease mechanisms [13–15]. Pioneering studies using patient-derived induced pluripotent stem cells (iPSCs) cerebral organoids by Kathuria et al. and Notaras et al. revealed schizophrenia-associated transcriptomic dysregulation [16, 17]. Besides, Sebastian et al. revealed abnormal developmental trajectories and transcriptomic signatures in forebrain organoids derived from schizophrenia patients carrying NRXN1 deletions [18], reinforcing the hypothesis that early neurodevelopmental disruptions underlie schizophrenia pathogenesis. Nevertheless, quantifying individual genetic variant contributions to schizophrenia remains problematic due to the polygenic architecture of the disorder [6, 19], and so we think that the parallel functional exploration of these candidate genes during human forebrain development will elucidate their distinct pathogenic mechanisms and collective disease liability.

Given polygenic architecture of schizophrenia, co-dysregulation of susceptibility genes drives neurodevelopmental pathology, but the degree of contribution of each gene to specific developmental processes may vary significantly. In this study, through CRISPR-Cas9, we generated single-gene-knockout-precise-dorsal/ventral-forebrain-organoids (SKOPOS, i.e., DFOs and VFOs) and performed systematic functional characterization of 11 schizophrenia risk genes, which are high expressed in human fetal forebrain, and provided evidence for a polygenic regulatory network underlying SCZ pathogenesis. Among them, CALN1 may be a key contributor to pathogensis, potentially involved in disrupting forebrain development and prioritized schizophrenia risk genes network, and associating with delayed neuronal maturation and impaired neural circuit in developing forebrain organoids. Further behavioral analyses revealed that the Caln1 KO mice not only exhibited the defects of spatial memory, social ability and pre-pulse inhibition, but also displayed spontaneous startle behavior, head-twitch response and hallucination-like perception. Electrophysiological deficits of abnormal spontaneous abrupt burst spiking in layer-5 pyramidal cells were observed. Finally, snRNA-seq analysis result indicated severe transcriptional and network dysregulation in layer-5 excitatory neurons and a large number of synapse-related genes show abnormally high expression. These genes may affect neuronal excitability by promoting the formation, stabilization, and functional potentiation of synapses.

Methods

Schizophrenia risk genes from GWAS summary statistics

The 11 tested risk genes were derived from the prioritized gene list reported by Trubetskoy et al. (Extended Data Table 12 of that study) [6]. For visualization purposes, we generated LocusZoom plots for each gene region using the summary statistics from the same schizophrenia GWAS. In each region, the most significant SNP was designated as the index SNP, and linkage disequilibrium (LD) of surrounding SNPs was calculated based on the European population genotype data from the 1000 Genomes Project.

CRISPR-editing human embryonic stem cell lines

Guide RNAs (gRNAs) were designed as described in Additional file 1: Supplementary Methods. Each gRNA was cloned into the LentiCRISPRv2 vectors including Cas9 sequence (Addgene #52961), and then were used to lentivirus package based on calcium phosphate method along with plasmids psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259). The 200 µl lentivirus supernatant was added to 50%-60% confluent hESCs and hESCs were selected by adding three-day 0.5 µg/ml puromycin after 48 h of infection. We picked more than 5 puromycin-resistant positive cell colonies per gene for proliferation, and Sanger sequence was used to confirm the genotype each colony. Immunofluorescence staining of OCT4 and SSEA4 were performed to test whether the pluripotency of hESCs with knock-out risk gene altered.

Generation of dorsal and ventral forebrain organoids

To generate DFOs and VFOs, we modified the previous methods in ref [12, 20]. Specifically, on day 0, hESCs (as negative control), hESCs with non-human gRNA (as positive control) and hESCs with knock-out risk gene were dissociated to single cell using Gibco™ TrypLE™ Express Enzyme (cat. 12604021) and 5,000 cells were reaggregated in U-bottomed 96-well plates with ultra-low-cell adhesion (Corning, cat. 7007) in dorsal or ventral forebrain differentiation medium (named dDM/vDM I). Basal medium of dDM/vDM I is DMEM containing 20% Knockout Serum Replacement (KSR) (Gibco), 1% GlutaMAX (Gibco, cat. #35050061), 10 µM β-mercaptoethanol, 1% penicillin-streptomycin (Gibco, cat. #15140122). For the first week, 10 µM TGFβ inhibitor SB431542 and 0.1 µM BMP inhibitor LDN193189 were added to the dDM/vDM I and additional 0.1 µM SAG and 5 µM IWP2 were added to vDM I. Besides, from day 0 to day 2, 10 µM ROCK inhibitor Y-27632 was added to the dDM/vDM I and cell was incubated in ultra-low 96-well plates in a static mode. After day 3, the aggregates were transfer to six-well plates (Corning) and cultured on the orbital shakers at 70 r.p.m. At the end of the first week, we embedded every organoid with 5 µl Matrigel (Corning, cat. #354230).

The medium used in dorsal forebrain organoids from week 2 to 3 was named dDM II and after 3 weeks was named dDM III. And the mediums used in ventral forebrain organoids include vDM II (for week 2), vDM III (for week 3) and vDM IV (after 3 weeks). vDM II was no longer added SB431542 and IWP2 compared to vDM I. Basal medium of dDM II, vDM III and vDM IV is DMEM/F12 (Gibco) supplemented with 2% B27 (Gibco). dDM II contained 20 ng/µl bFGF (PeproTech, cat. #AF-100-18B) and 20 ng/µl EGF (R&D Systems, Cat. #236-EG) and vDM III contained 0.1 µM SAG and 50 ng/µl FGF8 (PeproTech, AF-100-25). Content of dDM III and vDM IV were the same, that is, basal medium was supplemented with 10 ng/µl GDNF (PeproTech cat. # AF-450-10), 10 ng/µl BDNF (PeproTech cat. # AF-450-02) and 200 µM L-Ascorbic acid (Sigma-A8960).

Bulk RNA-sequencing

All organoid samples were derived from four independent induction batches. Each biological replicate was prepared by pooling two organoids from the same batch and from a single organoid type (DFOs or VFOs). For each knockout genotype, we obtained 3‑5 independent replicates for each organoid type and the control group comprised 8 independent replicates for separately DFOs and VFOs. Cortical tissues were collected from E17.5 littermate mice, including two wild-type (WT) and three Caln1^−/− mice. Total RNA was extracted using RNAiso Plus (TaKaRa, cat. #9109) according to the product manual. Quality control of RNA, the construction of cDNA library and sequence were conducted by Shanghai OE Biotech Co., Ltd. Company. The raw 150 bp paired-reads sequenced from the Illumina NovaSeq 6000 were preprocessed using Trimmomatic-0.36 and clean reads were mapped to reference genome GRCh38/GRCm38 using HISAT2 (v2.2.1) [21]. Based on gene annotation resource GRCh38.104/GRCm38.102, the reads count of every gene was calculated using featureCounts (v2.0.1) [22] and then counts were further normalized using transcripts per million (TPM). The Wald significance tests of R package DESeq2 (v1.40.2) [23] was used for differential expression analysis. Differentially expressed genes (DEGs) were defined as those with a fold change (FC) > 1.5 or < 1/1.5 (for DFOs and VFOs), or FC > 1.2 or < 1/1.2 (for cortical tissue of Caln1^−/− E17.5), and a p-value < 0.01. Raw bulk RNA-seq data of organoids have been deposited within the Genome Sequence Archive for Human (https://ngdc.cncb.ac.cn/gsa-human/) with the accession identifier HRA005215.

Disruption score of gene expression (DSGE)

For a given knockout vs. control comparison, only protein-coding genes with a DESeq2 baseMean > 0.1 were retained. For each retained gene, we obtained its log2FoldChange (effect size) and pvalue from the DESeq2 output and computed the z-score for each gene using z = abs (qnorm (pvalue/2, lower.tail=FALSE)). The weighted z-score contribution of all genes is: NormZ = sum (N*z) / sqrt (sum(N^2)), where N is the total number of biological replicates per gene. Consequently, the final DSGE becomes the mean Z-score over all analysed genes: DSGE = mean (NormZ). A higher DSGE indicates a stronger global transcriptional disruption.

To predict the transcriptional consequence of a homozygous mutation for each heterozygous knockout line, we performed an adjustment of the raw read counts. Firstly, using DESeq2 results comparing the heterozygous knockout (e.g., THOC7^+/−) against the control group, we selected all protein‑coding genes with FDR < 0.1 excluding the knocked‑out gene itself. These genes were prioritized by increasing FDR. For the i-th gene in this ranked list, we assigned a multiplicative factor beta[i] that decreases from 2 to 1 according to a linear sequence: beta = seq (2,1, -1/n), where n is the number of significantly dysregulated genes. If the gene was up-regulated in the heterozygote, the factor remains beta[i], and if the gene was down-regulated, the factor is inverted: 1/beta[i]. For each significantly dysregulated gene, we multiplied its raw read count in the heterozygous knockout samples by its corresponding adjusting factor beta, then rounded to the nearest integer to preserve count-based model assumptions. Genes that were not significant (FDR ≥ 0.1) were left unchanged. Finally, we replaced the original heterozygous knockout counts with the adjusted counts for the significant genes, while keeping the control samples untouched. This generated a new count matrix representing a predicted homozygous knockout (e.g., THOC7^−/−). This synthetic matrix was then used as input to DESeq2 to perform differential expression analysis (homozygous vs. control), from which we calculated the predicted DSGE as described above.

To assess whether the specific gene list (e.g., 379 forebrain genes or 106 risk genes) show a significantly higher DSGE score than expected by chance, we performed a permutation-based enrichment analysis. Firstly, we generated a null distribution of DSGE values by randomly sampling 379 genes or 106 genes from the entire gene set without replacement, repeating this procedure 10,000 times. For each random sample, we computed its DSGE value, denoted as x_i (i = 1,2,3…10,000), yielding a null distribution. Next, we calculated the observed DSGE value for the target gene list, denoted as x₀. To obtain a one-side empirical P-value testing whether x0 is larger than expected under the null, we constructed an empirical cumulative distribution function (ECDF) of the null distribution. The empirical P-value for enrichment (right-tail test) was defined as the proportion of null DSGE values greater than x₀ (i.e., 1-ECDF(x₀)).

Single-cell/nuclei RNA sequencing

For scRNA-seq of DFOs/VFOs, organoids were dissociated into the single-cell suspension at 37 ℃ for 30 min using enzyme stock solution (i.e., 10x EBSS (Sigma-Aldrich, cat. #E7510), 30% D (+)-Glucose (Sigma-Aldrich, cat. #G7021), 1 M NaHCO₃ (Sigma-Aldrich, cat. #S5761), 50 mM EDTA (Sigma-Aldrich, cat. #E8008), 20 unit/ml Papain (Worthington, cat. #LS03126), 10 µg/ml DNase Ⅰ (Sigma-Aldrich, cat. #11284932001) and 180 µg/ml L-cysteine hydrochloride monochloride (Sigma-Aldrich, cat. #C7880)). For snRNA-seq of frozen cortex tissue of adult mice, samples were minced in 1 ml of nuclear lysis buffer (0.1% NP-40, 10 mM Tris-HCl, 146 mM NaCl, 1 mM CaCl₂, 21 mM MgCl₂, and 1 U/µL RNase inhibitor) for 7 minutes. Subsequently, 1 mL of ST Wash buffer (10 mM Tris-HCl, 146 mM NaCl, 1 mM CaCl₂, 21 mM MgCl₂, 0.01% BSA) and 40 U/mL RNase inhibitor were added. After filtering with a 40 µm cell strainer (Miltenyi Biotec), the cell suspensions were centrifuged at 500 g for 5 min and resuspended in Neurobasal Medium. Further processing of samples, the construction of scRNA-seq libraries and sequence were conducted by Shanghai OE Biotech Co., Ltd. Company. Briefly, the final concentration of cell suspension was adjusted to 1,000 cells/µl and cells were loaded onto a Chromium Single Cell 3’ Chip (10x Genomics) to generate single cell Gel Beads in Emulsions (GEMs). The 10x Genomics Chromium Single Cell 3’ Library and Gel Bead Kit v3 were used to construct scRNA-seq libraries and sequencing was conducted on the NovaSeq 6000 instrument (Illumina). Raw scRNA-seq data of organoids have been deposited within the Genome Sequence Archive for Human (https://ngdc.cncb.ac.cn/gsa-human/) with the accession identifier HRA005215. Single-cell/nuclei RNA-seq analysis were described in Additional file 1: Supplementary Methods.

Multi-electrode array (MEA) recording and analysis

The 12-week dorsal or ventral forebrain organoids were plated in 6-well MEA plates (Axion Biosystems) and each well contains 64 low-impedance platinum microelectrodes. Spontaneous neural spikes were recorded using the AxIS Navigator system and spikes were detected with an adaptive threshold crossing of > 6 standard deviation (s.d.) for each electrode. The organoids were first stayed for 5 min in the device, and then 3 min of data were recorded. We quantified the knock-out effects of gene on dorsal or ventral forebrain organoids based on MEA recording from three levels, including spike, electrode and network. Firstly, we used the empirical cumulative distributions to visualize the distribution of inter-spike intervals (ISIs) of different groups and then used the one-side Kolmogorov-Smirnov test to compare ISIs distributions of knock-out and wild-type group. Besides, we counted the firing rates of every electrode, i.e., the number of spikes per minute. At neuron level, shorter ISIs and higher firing rates indicate stronger neuron activity. The network index was used to estimate strength of neuron-neuron connections, which is calculated as (n is the sum of the number of activated electrodes per second, N is the number of electrodes activated per organoids, T is the time of recoding (s)). The t-test was used to compare the knock-out effects of gene on firing rates of electrodes and network index.

Caln1 knock-out mice

The C57Bl/6J Caln1^+/− mice, constructed by Cyagen Biosciences Company (Suzhou), were used as parent mice to generate offspring including WT, Caln1^+/− and Caln1^−/−. To knock out mice Caln1, two gRNAs are used, i.e., 5’-GCCATGGAGAGTGCTACTTA-3’ (GGG) and 5’-CATAGGTGACCCCGCCAAAG-3’(GGG). Three primers are used to verify knock-out mice, i.e., 5’-TTAGAATATGCCAATGAAAGGGCTG-3’ (F1), 5’-ACCCCTATGACAAACACACATAAGA-3’ (R1) and 5’-GTTTGCTGCTGCTGTTATTATTGTG-3’ (R2).

Electrophysiological recording

Mice neocortical slices were prepared according to Additional file 1: Supplementary Methods. After slice preparation, the prefrontal cortical slices were transferred into a recording chamber and perfused with ACSF. The solution temperature was maintained at 34.5–35 °C. Cortical neurons were visualized with an upright infrared differential interference contrast microscope (BX51WI; Olympus) equipped with a water-immersed objective (40×, numerical aperture 0.8). We performed whole-cell current-clamp recording from the soma of layer-5 pyramidal cells to investigate their intrinsic membrane properties and excitability. The patch pipettes had an impedance of 5–6 MΩ when filled with a K⁺-based internal solution containing (mM) 140 K-gluconate: 3 KCl, 2 MgCl₂, 10 Hepes, 0.2 ethylene glycol tetraacetic acid (EGTA), and 2 Na₂ATP (290 mOsm, pH 7.23 with KOH). We also added 0.2% biocytin to the pipette solution for cell staining after the experiments. Electrical signals were acquired using a MultiClamp 700B amplifier (Molecular Devices), digitized and sampled by Micro 1401 mkII (Cambridge Electronic Design) at 25 kHz using Spike2 software. The analysis of electrophysiological and morphological data is provided in Additional file 1: Supplementary Methods.

Animal behavior test

All animal experiments were approved by the Laboratory Animal Science of Fudan University and conducted in accordance with their guidelines. C57BL/6J mice were housed under specific pathogen-free conditions. A battery of behavioral tests was performed to assess the phenotypic consequences of gene manipulation, including Y-maze, three-chamber social test, open-field test, elevated plus maze, tail suspension test, prepulse inhibition (PPI), head-twitch and startle reflex evaluation, and a hallucination-like perception test. Detailed procedures, apparatus settings, and data analysis methods are provided in Additional file 1: Supplementary Methods.

Results

Knockout of schizophrenia risk genes in human embryonic stem cell lines

To discover the roles of schizophrenia risk genes in neurodevelopment [6], we first generated knockout models of 11 candidate genes (i.e., 5 FINEMAP genes and 6 SMR genes) in human embryonic stem cells (hESCs) using CRISPR-Cas9 (Fig. 1a). These genes, prioritized based on their elevated transcriptional activity in two critical schizophrenia-relevant brain regions (i.e., the fetal dorsolateral and ventrolateral prefrontal cortex, DLPFC/VLPFC) [24–26], include CLCN3 (encoding a voltage-gated channel), CALN1 (encoding a calcium-binding protein), CRHR1 (encoding a hormone receptor), GRM1 (encoding a glutamate receptor), GPM6A (encoding a glycoprotein), MAPT (encoding a microtubule protein), SF3B1 (encoding a splicing factor), THOC7 (encoding a RNA-binding protein) and three genes encoding metabolic enzymes: PLCH2, PDE4B and PCCB (Additional file 2: Figure S1a). According to the location on the genome, they span ten different genome loci which are genome-wide significantly associated with schizophrenia (P < 5 × 10⁻⁸) (Additional file 2: Figure S1b-k). To establish individual knockouts for these 11 genes, we isolated and expanded a total of 115 hESC clonal lines (Fig. 1b, c). By the verification of Sanger sequencing, these hESC lines (H9 WiCell) include 13 homozygous mutations, 65 heterozygous mutations and 37 wild types (WT), of which homozygous mutation were achieved for 8 of the 11 risk genes (Fig. 1c). For the remaining three genes, we only obtained their heterozygous mutant hESCs, which is speculated that their homozygous mutants may not survive easily, considering their high expression in hESCs [27] (Additional file 2: Figure S2a). Next, except naïve (normal hESCs H9) and negative control (hESCs H9 carrying sgRNA of non-target human gene), we selected one knock-out hESC line per gene including CLCN3^−/−, CALN1^−/−, CRHR1^−/−, GRM1^−/−, GPM6A^−/−, MAPT^−/−, PLCH2^−/−, PDE4B^−/−, PCCB^+/−, SF3B1^+/− and THOC7^+/− for downstream experiments in this study (Additional file 2: Figure S2b). For each selected KO hESC line, using off-targeting assessment CRISPRoff [28] and the Sanger sequencing, we furtherly verified that there were no off-target in the top ten predicted genome sites (Additional file 3). And then we also confirmed that the pluripotency of all knock-out hESC lines are maintained, with highly expressing pluripotent markers OCT4 and SSEA4 (Additional file 2: Figure S2c), and normal karyotypes of control and knock-out CALN1 hESC lines were furtherly confirmed (Additional file 2: Figure S2d).

Fig. 1.

Construction of knock-out hESC lines and dorsal and ventral forebrain organoids. a, Base on BrainSapn datasets, 11 schizophrenia risk genes expressed in fetal DLPFC and VLPFC were selected for further functional studies in this study. RPKM, Reads Per Kilobase per Million mapped reads. DLPFC, dorsolateral prefrontal cortex. VLPFC, ventrolateral prefrontal cortex. ExNs, excitatory neurons. InNs, inhibitory neurons. b, Schematic illustration for construction of KO hESC lines. hESC, human embryonic stem cells. DFOs, dorsal forebrain organoids. VFOs, ventral forebrain organoids. c, The number of gene-edited hESC lines verified by Sanger sequence. WT, wild type. d, The schematics of generating DFOs and VFOs. LDN, LDN193189. SB, SB431542.Vc, Vitamin C. e, The expression of dorsal and ventral forebrain markers in DFOs and VFOs. f, Comparison of gene expression modules between DFOs/VFOs and fetal DLPFC/VLPFC based on WGCNA analysis. WGCNA, weight gene co-expression network analysis. The modules with Z-score > 5 were considered as preservative

Generation of dorsal and ventral forebrain organoids as 3D models of schizophrenia risk genes

Genes with high expression specificity for both excitatory neurons and inhibitory neurons are enriched for genetic associations with schizophrenia [6, 29], which are, in vivo, respectively generated from the dorsal and ventral telencephalon during human early developmental brain [30]. To fully investigate how these schizophrenia risk genes affect early forebrain development across excitatory and inhibitory neuronal systems, we individually generated dorsal forebrain organoids (DFOs) and ventral forebrain organoids (VFOs) induced from those 11 KO hESC lines and two wild-type hESC lines (Fig. 1d). To minimize pluripotency variability, all KO hESC lines used in this experiment were maintained at early passages 3–20. Bulk RNA-seq confirmed that DFOs (n = 8) have high expression of PAX6, NEUROD1, NEUROD2 and NEUROG2, the dorsal forebrain marker, while VFOs (n = 8) expressed ventral markers (NKX2-1, DLX2, DLX5 and DLX6), validating that regional specificity of the DFOs/VFOs (Fig. 1e). Bulk RNA-seq or qPCR showed the RNA levels of the target genes in 6-week-old KO organoids, are significantly reduced (Additional file 4: Figure S3a), and we furtherly confirmed that CALN1 loss at protein level in both CALN1^−/− DFOs and VFOs (Additional file 4: Figure S3b). By Western blot, we observed silencing of Cas9 in 6-week-old organoids, which preclude the possibility of off-target editing during organoids differentiation (Additional file 4: Figure S3c). Morphometric analysis at 4 weeks revealed that DFOs with CALN1, GRM1, PDE4B or PCCB knockouts, as well as CALN1^−/− VFOs, were significantly larger than wild-type organoids (Additional file 4: Figure S3d, e). To benchmark the organoid models, we compared their transcriptome to human fetal DLPFC and VLPFC. Weight gene co-expression network analysis (WGCNA) of 47 DFOs and 43 VFOs RNA-seq datasets (6 weeks) identified 22 and 19 gene co-expression modules (except for the gray module), respectively. Using BrainSpan database, we analyzed transcriptomes of fetal 17 DLPFC and 16 VLPFC. Z-summary statistics demonstrated most of DFOs and VFOs gene modules are preserve in fetal DLPFC and VLPFC (Z-score > 5) (Fig. 1f). Collectively, these data indicated the successful generation of schizophrenia risk gene KO organoids that recapitulate fetal brain transcriptional profiles, establishing a robust platform for modeling neurodevelopmental deficits of schizophrenia.

Schizophrenia risk genes disrupt forebrain gene expression network

To investigate the gene regulatory network of these schizophrenia risk genes, we performed bulk RNA-seq on dorsal and ventral forebrain organoids at 6 weeks, which comprised 90 independent biological experiments. Specifically, for each of the 11 knockout lines, we prepared 3–5 biological replicates separately for DFOs and VFOs; the wild-type control group (pooled from naïve WT and non-targeting sgRNA WT) consisted of 8 replicates for each organoid type (Fig. 2a and Additional file 5). On average, only 11.64% (95% confidence interval [CI], 8.21%-15.06%) of genes were co-upregulated and 16.55% (95% CI, 11.93%-21.16%) were co-downregulated across both DFOs and VFOs, which indicated divergent roles for these 11 schizophrenia risk genes in excitatory and inhibitory neuronal lineages (Additional file 4: Figure S4a, b and Additional file 5). Rank-Rank Hypergeometric Overlap (RRHO) analysis further validated distinct dysregulation patterns in the transcriptional profiles of DFOs and VFOs harboring the same KO risk gene (Additional file 4: Figure S4c, d).

Fig. 2.

Schizophrenia risk genes regulate forebrain gene expression network. a, Top three PCs of gene expression of DFOs and VFOs at 6 weeks (90 independent biological experiments). For the knockout groups, n=3-5; for the control group, n=8 (pooled from two wild‑type lines). PCs, principal components. b, The disruption score of gene expression (DSGE) for each mutation based on z-score of differentially expressed analysis of 6-week organoids at genome-wide level, of which, including the predicted DSGE for homozygous state of three heterozygous mutations (i.e., PCCB+/-, SF3B1+/- and THOC7+/-). c, The Rank-Rank Hypergeometric Overlap (RRHO) maps between any two genes in DFOs or VFOs. Signals in top right quadrant and bottom left quadrant indicate an overlap of down-regulated and up-regulated genes found in both KO genes, respectively. d, MixDFOs (n=2) and MixVFOs (n=3) based on pooled KO cells were constructed and compared to individual knockout DFOs or VFOs. e, The correlations of log2FC (left) and RRHO maps (right) between the MixDFOs and DFO with single gene knockout. f, The correlations of log2FC (left) and RRHO maps (right) between the MixVFOs and VFO with single gene knockout. g, h, The permutation‑based enrichment analysis of 106 schizophrenia risk genes in each KO DFOs (g) and VFOs (h). The null distribution was generated by 10,000 permutations of randomly sampled 106 genes without replacement from the entire gene set. A p-value < 0.05 indicated significant enrichment.

To quantify the functional impact of individual gene knockout, we computed a disruption score of gene expression (DSGE) for every mutation. This score was derived from the mean of z-score of protein-coding genes with a baseMean greater than 0.1 based on differentially expressed analysis in 6-week-old organoids. For three heterozygous mutations (i.e., PCCB^+/−, SF3B1^+/− and THOC7^+/−), we also predicted the DSGE for their hypothetical homozygous counterparts (Fig. 2b). Genome-wide DSGE analysis revealed that CALN1 KO organoids exhibited the strongest transcriptional disruption, particularly in VFOs (Fig. 2b). RRHO analysis demonstrated significant concordance in transcriptional profiles across DFOs harboring diverse mutations including CRHR1^−/−, GRM1^−/−, MAPT^−/−, PDE4B^−/−, PCCB^+/−, SF3B1^+/− and THOC7^+/−, indicating shared dysregulation patterns (Fig. 2c). In contrast, VFOs showed no such global similarity, only CLCN3^−/− exhibited significant overlap with GPM6A^−/− and THOC7^+/− (Fig. 2c). WGCNA module analysis further highlighted divergent regulatory effects: 7 out of 22 gene expression modules in DFOs (e.g., darkgrey, floralwhite, lightcyan, lightsteelblue1, plum2, salmon, skyblue) are significantly associated with more than three risk genes, whereas only 1 out of 19 expression modules (darkgrey) in VFOs (Additional file 4: Figure S4e, f). These results collectively suggest that these 11 schizophrenia risk genes exert more convergent transcriptional effects in DFOs compared to VFOs.

To better evaluate the polygenic effects of these schizophrenia risk genes on brain development, we generated a pooled co-culture system (village-in-a-dish) containing equal proportions of validated knockout hESCs clones for each gene. These were differentiated into mixed DFOs and VFOs (MixDFOs and MixVFOs) to model combinatorial genetic perturbations (Fig. 2d) [31]. At week 6, based on bulk RNA-seq analysis of mixed and single-gene KO organoids revealed distinct transcriptional dynamics. Log2 fold-change (Log₂FC) correlation and RRHO analysis demonstrated that the dysregulation in MixDFOs was driven by synergistic interaction among multiple risk genes including CALN1, CRHR1, GRM1, MAPT, PDE4B, PCCB, SF3B1 and THOC7 (Fig. 2e). In contrast, transcriptional changes in MixVFOs were predominantly mediated by CALN1 KO, consistent with its effect observed in single-gene KO VFO models (Fig. 2f).

To compare the impact of different schizophrenia risk genes on forebrain development, we firstly established the null distribution of DSGE values from 10,000 random samples of 379 genes (matching the size of the forebrain development gene set, GO: 0030900, Additional file 5). And then the observed DSGE value x₀ for the 379 forebrain development genes was compared against this null distribution. The empirical P-value calculated as the proportion of random samples with DSGE ≥ x₀. This indicates whether the forebrain gene set exhibits a statistically significant enrichment of high DSGE scores relative to random gene sets of the same size. Notably, the DSGE values of forebrain development genes were significantly higher than random expectations in multiple gene-knockout organoids (P < 0.05), including CALN1^−/−, PCCB^+/−, THOC7^+/−, GRM1^−/−, MAPT^−/− DFOs, and CALN1^−/−, THOC7^+/−, PDE4B^−/−, CLCN3^−/−, MAPT^−/−, GRM1^−/−, GPM6A^−/− VFOs (Additional file 4: Figure S4g, h), indicating that knockout of these genes potentially mediate early human forebrain development and influence the occurrence of schizophrenia.

Expanding our analysis to schizophrenia risk genes, we found that after knocking out these multiple risk genes, the DSGE of the overall set of 106 schizophrenia risk genes was also higher than random DSGE, in same cases in THOC7^+/−, SF3B1^+/−, PLCH2^−/−, CRHR1^−/− DFOs and CALN1^−/−, CLCN3^−/−, MAPT^−/−, SF3B1^+/− VFOs (P < 0.05, Fig. 2g, h).

Taken together, these results indicated that knockout of 11 schizophrenia risk genes lead to different degrees of developmental impairments in both human DFOs and VFOs, and more heterogeneous effects were observed on VFOs compared with DFOs. Notably, CALN1 knockout significantly impacted pathways involved in forebrain development and the set of prioritized schizophrenia risk genes.

Schizophrenia risk genes disturb specific cell population and developmental trajectories

To further discover the cell type-specific effects of these risk genes, we profiled 56 DFOs/VFOs with wild type (WT) and 11 KO genes at 12 weeks using single-cell RNA-sequencing. A total of 249,430 qualitied cells (WT: 44,899 cells; KO: 204,531 cells) were obtained (Fig. 3a, c). Using known marker genes, we identified distinct 14 distinct cell populations in DFOs and 13 in VFOs based on the wild-type cellular architecture (Additional file 6: Figure S5a-d). We then computationally mapped the cell identity of each KO organoids (Fig. 3b, d). For all knockout lines tested, the overall repertoire of major neuronal and glial cell classes was largely preserved (Fig. 3b, d). This observation is consistent with the notion that these genes may exert more subtle effects on cell state, differentiation timing, or gene expression rather than causing wholesale elimination of specific cell types.

Fig. 3.

Schizophrenia risk genes affect specific cell population in forebrain organoids at 12 weeks. a, c, The UMAP plots of scRNA-seq for each KO DFO (a) and VFO (c) at 12 weeks. Each experimental group consisted of 1-3 biological samples representing 1-2 independent batches, with scRNA-Seq performed on pooled RNA extracted from 3-4 organoids per sample. UMAP, Uniform Manifold Approximation and Projection. b, d, The proportion of main cell population in scRNA-seq for each KO DFO (b) and VFO (d). e, g, Reproducible dysregulated genes in specific neuron clusters in KO DFOs (e) and VFOs (g). f, h, Top 20 GO terms related with biological process enriched by dysregulated genes in DFO neurons (f) and VFO neurons (h).

Given that major cell populations were largely preserved, we next asked whether more subtle transcriptional changes could explain the abnormal development of KO neurons. We therefore performed differential expression analysis across neuronal lineages and found that, when compared against 2–3 independent controls, 70 genes in KO DFOs and 73 genes in KO VFOs were reproducibly abnormally expressed (Fig. 3e, g). GO enrichment analysis found that these differentially expressed genes (DEGs) are key genes related to forebrain development, neuron differentiation, gliogenesis and so on (Fig. 3f, h and Additional file 7).

We also reconstructed cellular developmental trajectories for every DFO/VFO based on pseudotime using radial glial cells as original cell population. The diffusion maps showed bifurcating differentiation pathways into neuronal and glial lineages in both DFOs and VFOs (Fig. 4a, c). The branching dendrogram structures based on cell pseudotime further depicted temporal maturation dynamics of individual cell populations (Fig. 4b, d). Compared with wild type, we found that significant lineage-specific developmental deviations in KO organoids versus WT controls (Fig. 4e, g). The ratio of pseudotime difference in area under the curve (AUC) in KO organoids compared to control was used to quantify deviations in maturity for each cell population. Accelerated maturation (AUC ratio > 0.3) occurs in 5/121 (~ 4%) of DFOs cell population and 14/110 (~ 13%) of VFOs cell population, while maturational delay (AUC ratio < -0.3) in 35/121 (~ 29%) of DFOs cell population and 9/110 (~ 8%) of VFOs cell population (Fig. 4f, h). Strikingly, compared to the control, CALN1^−/− DFOs retain more EarlyExN cells with less maturation (Figs. 3b and 4f), and CALN1^−/− VFOs have more InN1 of SST⁺ with higher maturation (Figs. 3d and 4h). Taken together, these results indicated that knockout of these schizophrenia risk genes have significant effects on the pathways associated with forebrain development, neuronal subtype specification and temporal maturation programs, of which CALN1^−/− leads to delayed neuronal maturation of DFOs and gliogenesis of VFOs.

Fig. 4.

Schizophrenia risk genes affect cell development trajectories in forebrain organoids at 12 weeks. a, c, The diffusion maps of different cell clusters for DFOs (a) and VFOs (c). RGs, radial glial cells. b, d, The branching dendrogram structures of neurons and glial cells for DFOs (b) and VFOs (d). e, g, The density plots of pseudotime for each cell cluster in KO DFOs (e) and VFOs (g). f, h, The AUC ratio of pseudotime in different cell clusters of KO DFOs (f) and VFOs (h). AUC ratio=(AUC[KO]-AUC[WT])/AUC[WT]. AUC, area under the curve.

Schizophrenia risk genes affect spontaneous neural circuit activity

The observed abnormalities through integrated analysis of gene expression network, specific cell population and developmental trajectories in these KO organoids prompted further investigation of their neural network activity. We analyzed spontaneous neuronal activity in 145 DFOs/VFOs (12 weeks) encompassing wild-type and 11 KO models by multielectrode array (MEA) recordings. There were 61,654 and 88,850 spikes generated by 962 and 893 activated electrodes from DFOs and VFOs, respectively (Fig. 5a, b and Additional file 8: Figure S6a). For these activated electrodes, we analyzed their inter-spike intervals (ISIs), firing rates and network index (NI) (Fig. 5c) and found that the spontaneous neuronal activities are heterogeneous in KO DFOs and VFOs. In addition to MAPT and PDE4B, significantly longer ISIs were observed in other KO DFOs, while shorter ISIs in multiple KO VFOs apart from CALN1, PLCH2 and SF3B1 (Fig. 5d, e). At the level of activated electrodes, 7/11 KO DFOs (i.e., CALN1^−/−, GRM1^−/−, GPM6A^−/−, PLCH2^−/−, PDE4B^−/−, PCCB^+/− and THOC7^+/−) exhibited significant firing rate reductions, which indicate weaker neuron activity (Fig. 5f). Notably, knockout of GRM1, results in weaker neuronal activity in DFOs (Fig. 5f), while abnormally higher network firings were detected in VFOs (Fig. 5i and Additional file 8: Figure S6a). The significantly longer ISIs, lower firing rates and weaker network index were observed in both DFOs and VFOs with CALN1^−/− at 12 weeks, indicating both neuronal activity and connections are affected by CALN1 (Fig. 5d-i). Taken together, these results indicated that most of these risk genes attenuate excitatory circuit dynamics in DFOs and the potential inhibitory network activity in VFOs. Among them, CALN1 exhibited the strongest alterations in neuronal activity across both DFOs and VFOs.

Fig. 5.

The multielectrode array (MEA) recording and analysis of knock-out dorsal and ventral organoids. a, b, Example of electrode-activity heatmaps in DFOs (a) and VFOs (b) with knock-out gene and wild type. An average of 6 organoids were tested in each group from 3 batches for MEA recording. The number of organoids, activated electrodes and spikes were labeled in turn above each heatmap. Scale bar, 500 μm. c, Schematic illustration for Inter-Spike Intervals (ISIs) and network index (NI). d, e, Empirical cumulative distribution plots of Inter-Spike Intervals (ISIs) for DFOs (d) and VFOs (e). The one-side Kolmogorov-Smirnov test to compare ISIs distributions of knock-out and wild-type group. f, g, Firing rates of each electrode in DFOs (f) and VFOs (g). h, i, Network indexes of each DFO (h) and VFO (i). The two-tail t-test was used in f-i. j, CALN1 loss at protein level in CALN1^-/-rep DFOs and VFOs. k, l, The MEA analysis of Inter-Spike Intervals, firing rates and network indexes for CALN1^-/-rep DFOs (k) and VFOs (l). Data are presented as mean ± SE. ***, p-value < 0.001. **, p-value < 0.01. *, p-value < 0.05.

To further confirm the role of CALN1 knockout on brain development, we generated an independent CALN1^−/− hESCs line (CALN1^−/−rep) using alternative sgRNA (Additional file 3). No detectable off-target effect and normal karyotypic integrity were confirmed for the CALN1^−/−rep hESC line (Additional file 3 and Additional file 2: Figure S2d), and Western blot confirmed complete protein ablation in CALN1^−/−rep DFOs and VFOs (Fig. 5j). MEA analysis of CALN1^−/−rep DFOs and VFOs recapitulated CALN1^−/− phenotypes including more prolonged ISIs, reduced firing rates and impaired network synchrony (Fig. 5k, l and Additional file 8: Figure S6b, c). To further examine whether developmental defects underlie the electrophysiological hypoactivity in these organoids, we analyzed the correlation of gene expression levels between CALN1 and PAX6, DCX and RBFOX3 (NeuN) at 6-week DFOs based on bulk RNA-seq data. Correlation analysis showed that the gene expression of CALN1 and PAX6 is significantly negatively correlated (Additional file 8: Figure S7a), and positively correlated with DCX and RBFOX3 (Additional file 8: Figure S7b, c). Immunofluorescence staining further confirmed knockout of CALN1 results in the retention of more neural progenitor cells (PAX6⁺) and a significant reduction in generation of new neurons (DCX⁺) in 6-week DFOs (Additional file 8: Figure S7g, h), and this delay in neural development lasts up to 12-week DFOs with lower expression of NeuN (Additional file 8: Figure S7i). Notably, there was no significant expression correlation between CALN1 and NKX2-1 in 6-week VFOs (Additional file 8: Figure S7d) and knockout of CALN1 did not show increase of NKX2-1 in consistency between two different CALN1^−/− VFOs (Additional file 8: Figure S7j). However, S100B⁺ and MBP⁺ glial cells were significantly decreased in CALN1^−/− VFOs (Additional file 8: Figure S7e, f, k and l). Taken together, developmental abnormalities in CALN1^−/− DFOs/VFOs may attenuate neural circuit activity in CALN1^−/− forebrain organoids.

Caln1^−/− mice display schizophrenia-like behavior and hallucination-like perception

Given the remarkable dysregulation of key developing genes and neuronal functional abnormality in CALN1 KO forebrain organoids, we next sought to investigate whether the impact of CALN1 on forebrain development is conserved across species. To this end, we constructed Caln1^−/− transgenic mice (Fig. 6a) and performed bulk RNA-seq on cortical tissues from E17.5 Caln1^−/− mice (N_WT=2, N_KO=3). Differential expression analysis revealed 154 down-regulated and 411 up-regulated genes (Fig. 6b, Additional file 5). These genes were enriched in pathways associated with sensory perception, forebrain regionalization, E/I balance (including gamma-aminobutyric acid signaling pathway and glutamate receptor signaling pathway), adult behavior and learning or memory (Fig. 6c, Additional file 5). The results further support the conclusions from DFOs/VFOs that CALN1 affects early forebrain development. To further assess how this gene knockout affects adult mouse behavior, we carried out a series of behavioral assays, using WT, and Caln1^−/− mice at 8–11 weeks. In the three-chamber social test, Caln1^−/− mice appeared weaker social preference between stranger 1 mice and empty cage than WT mice in session 1, along with spending more time in the middle area, and in session 2 Caln1^−/− mice showed no significantly social tendency to stranger 2 mice compared with stranger 1 mice, which indicated an impaired sociability and a decreased social motivation in Caln1^−/− mice (Fig. 6d-f). In the Y-maze test, comparing with wild type, Caln1^−/− mice have significantly lower percentage of spontaneous alternation, displaying significant deficiency of short-term spatial working memory and cognition compared with wild type mice (Fig. 6g). Meanwhile, Caln1^−/− mice showed a remarkable increase of the total number of arm entries but normal average moving speed (Fig. 6h and Additional file 9: Figure S8a). In addition, in the pre-pulse inhibition test (PPI), the Caln1^−/− mice showed the significantly PPI deficits (Fig. 6i). In the open-field, elevated plus maze and tail suspension test, there were no significant difference between Caln1^−/− and WT mice (Additional file 9: Figure S8b-e).

Fig. 6.

Caln1-/- mice displayed schizophrenia-like behaviors including hallucination-like perception. a, The generation of Caln1^-/- mice. b, Volcano plot of differentially expressed genes from bulk RNA-seq of Caln1^-/- mice cortex at E17.5 (NWT=2, NKO=3). c, Pathway enrichment analysis of DEGs in the cortical tissue of Caln1^-/- mice at E17.5. d-f, The three-chamber social test. The heatmaps of social time between WT and Caln1^-/- mice (d), social duration with stranger mouse and empty cage in session 1 (e), social duration with stranger 1 and stranger 2 in session 2 (e) and duration in the middle area (f) were showed. g, h, The Y-maze test. Comparing with wild type, Caln1^-/- mice have significantly lower the percentage of spontaneous alternation (g) and more total numbers of arm entries (h). i, The 120dB startle stimulus and three different pre-pulse stimuluses (i.e., 72, 76 and 80 dB) were conducted in the pre-pulse inhibition (PPI) test. j, k, The counts of spontaneous startle behavior (j) and head-twitch response (k) within a random 15 minutes from 10 Caln1^-/- mice. l-r, The hallucination-like perception test. In the test phase, the number of trials (m), accuracy (n), hit rate (o), false alarm (FA) rate (p), the density of FA rate (q) and FA invested time (r) between Caln1^-/- mice and WT mice. Data are presented as mean ± SE. The two-tail t-test was used to calculate p-value.

In addition to these classic behavioral paradigms, we also observed that Caln1^−/− mice from about 4 weeks appeared spontaneous startle behavior (Fig. 6j and Additional file 10: Video S1) and spontaneous head-twitch response (Fig. 6k and Additional file 11: Video S2). And both specific behaviors occur up to 45 and 39 times within 15 min during the onset of the abnormality (n = 10, Fig. 6j, k). The head-twitch response (HTR) observed in mice has been proposed as a behavioral correlate of hallucination-like states in humans [32–37]. To further investigate whether Caln1^−/− mice have hallucination-like perception, we subjected water-deprived wild-type and Caln1^−/− mice to an 8-week auditory discrimination task with water reinforcement, followed by a 7-day testing phase (Fig. 6l) [38]. 6 wild type and 7 Caln1^−/− mice successfully completed training and proceeded to the 7-days test phase (Fig. 6l). Based on the statistical results in the test phase, we found that the number of trials and hit rate have no significant difference between Caln1^−/− and WT mice (Fig. 6m, o). It is worth noting that besides lower correct rate (Fig. 6n), the Caln1^−/− mice showed significantly greater error rate when there is no sound, that is, the wrong belief that there is a sound, which is called false alarm (FA) here, and it is accompanied by a longer FA invested time (Fig. 6p-r), indicating Caln1^−/− mice may display hallucination-like perception.

Collectively, our behavioral analyses demonstrate that Caln1 deficiency recapitulates core schizophrenia-related phenotypes, including spatial memory deficits, cognitive impairment, social dysfunction, disrupted sensorimotor gating (pre-pulse inhibition), and hallucination-like percepts.

Caln1^−/− layer-5 pyramidal mature neurons exhibit spontaneous abrupt depolarization burst firing

The activity of cortical layer-5 pyramidal neurons has been causally implicated in supporting conscious perception in mice [39]. Intriguingly, this key neuronal population appears to be a specific target for three antipsychotic drugs, which are known to selectively reduce the activity of layer-5 neurons in cortical circuits [40]. And so, we investigated electrophysiological properties of cortical mature neurons in Caln1^−/− mice through whole-cell current-clamp recordings from layer-5 pyramidal cells (PCs) in acute slices of the prefrontal cortex of adult mice (Fig. 7a-h). Remarkably, 40.4% of recorded Caln1^−/− PCs (23/57, N = 7 mice) showed abrupt burst spiking (ABS) from their resting membrane potential (V_m) (Fig. 7b-d), a phenomenon that does not occur in PCs of WT mice (n = 47, N = 6 mice). The ABS occurred randomly during the recordings, with duration ranging from 0.03 s to 725 s (Fig. 7f). Close examination of the abrupt burst revealed a train of action potentials (with a frequency of 3.0 ± 0.36 Hz, n = 71 abrupt bursts) riding on top of a depolarization plateau. To examine whether ABS neurons possess distinct intrinsic membrane properties, we compared the electrophysiological properties of ABS and non-ABS neurons of Caln1^−/− mice with those of WT mouse neurons. A significant increase in the neuronal excitability was observed in ABS neurons as compared to those of non-ABS neurons and WT neurons, as reflected by the upward shift of the input-output (I-F) curve of ABS cells (Fig. 7e). In consistent, we also found a decrease in the threshold current (rheobase) for action potential generation (WT 119.7 ± 5.7 vs. ABS 96 ± 8.9 pA, P = 0.04, Fig. 7e), and a slight difference in their resting V_m (WT − 66.6 ± 0.5 vs. ABS − 64.7 ± 1.1 mV, P = 0.07, Fig. 7h). We found no significant difference in the input resistance (WT, 132.7 ± 8.8; ABS, 129.7 ± 9.8; non-ABS, 143.3 ± 9.8 MΩ) and the membrane time constant (WT, 20.7 ± 1.2; ABS, 20.2 ± 1.3; non-ABS, 20.2 ± 1.5 ms, Fig. 7h). Close examination of APs revealed differences in their waveforms between WT neurons and ABS cells of Caln1^−/− mice, including significant changes in AP amplitude, half-width, maximal depolarizing and repolarizing slopes (Fig. 7g-h). The voltage threshold of APs and the resting membrane potential showed a slight but non-significant change (Fig. 7h). Similar abrupt-burst firing was observed in layer-2/3 PCs of Caln1^−/− mice, but the percentage of ABS cells (25%, n = 5/20) was less than that of layer-5 PCs (Additional file 12: Figure S9a-e). In summary, the loss of Caln1 destabilizes the excitability of PFC layer-5 pyramidal mature neurons, driving a significant proportion into a hyperexcitable, burst-firing state with altered action potential dynamics. These cell-autonomous electrophysiological deficits likely constitute a fundamental mechanism contributing to the behavioral impairments observed in Caln1^−/− mice.

Fig. 7.

The layer-5 pyramidal neurons of Caln1^-/- mice displayed abrupt depolarization burst firing. a, Projection of confocal z-stack images of pyramidal cells (PCs) that has been filled with biocytin during whole-cell recording in a prefrontal cortical slice. b, An example of continuous recording from a Caln1^-/- cortical neuron showing membrane potential responses (bottom) to intracellular injections of step currents (top) and the occurrence of abrupt burst spiking (ABS) events from the resting membrane potential. c, Representative traces showing spiking responses to a current step of 150 pA (top) and ABS events (bottom). d, Percentage of ABS (dark blue) and non-ABS neurons (light blue) from Caln1^-/- mice. e, Comparison of I-F curves of WT neurons, Caln1^-/- ABS and non-ABS neurons in layer 5. Inset, comparing the threshold current (rheobase) for AP generation. f, Counts of abrupt burst events in individual ABS neurons (top) and their duration (bottom). g, Representative AP waveforms (up) and their phase plots (bottom) from a WT neuron (gray) and a Caln1^-/- ABS neuron (dark blue). h, Comparison of AP parameters among WT neurons, Caln1^-/- ABS, and non-ABS neurons.

Severe transcriptional and network dysregulation in cortical layer-5 excitatory neurons of Caln1^−/− mice cortex

To elucidate the molecular mechanisms underlying the observed behavioral and electrophysiological abnormalities in Caln1 knockout mice, we performed single-nucleus RNA sequencing of the cortex (Fig. 8a). Using the reference mouse cell atlas, we annotated all single cells into 16 distinct cell clusters (Fig. 8b), comprising 5 excitatory neuron clusters (i.e., Ex-L2/3, Ex-L4, Ex-L5, Ex-L6, Ex-PIR-Ndst4), 6 inhibitory neuron clusters (i.e., In-Drd2, In-Hap1, In-Pvalb, In-Sst, In-Tac1, In-Vip/Lamp5), and 5 other cell clusters (i.e., Astro, OPC, Oligo, Misc, VLMC) [41]. Upon conducting differential expression analysis (FDR < 0.05) on each major neuronal cell cluster, we discovered that the Ex-L5 neurons exhibited the most severe gene expression dysregulation (Fig. 8c). The dysregulated genes in Ex-L5 cell population were enriched in biological pathways related to synapse organization, learning or memory, neuron migration and so on (Fig. 8d). Further analysis revealed that the differentially expressed genes in the synapse organization pathway (GO: 0050807) formed a tightly interconnected protein-protein interaction network (Fig. 8e). Notably, the hub genes (i.e., Dlg4, Nrxn1, Nlgn1, Nrxn2, Grin2b, Cdh2, Lrfn5, Ntrk2, Ptprs and Il1rapl1) within this network were significantly upregulated in the Ex-L5 excitatory neurons of Caln1 knockout mice (Fig. 8f). Their coordinated activities span synaptic adhesion (Nrxn1/2, Nlgn1, Cdh2) [42–44], glutamate receptor signaling and trafficking (Dlg4, Grin2b) [45, 46], neurotrophic signaling (Ntrk2) [47], and synaptic organization/regulation (Lrfn5, Ptprs, Il1rapl1) [48–50], which defines a precise molecular substrate underlying the excitability of L5 pyramidal neurons and indicated underpins the observed behavioral deficits in Caln1 KO mice.

Fig. 8.

Severe transcriptional and network dysregulation in cortical layer-5 excitatory neurons of Caln1^-/- mice cortex. a, snRNA-seq for WT and Caln1^-/- mice cerebral cortex. b, 16 distinct cell clusters were annotated for WT and Caln1^-/- mice. c, The number of genes differentially expression (FDR < 0.05) on each major neuronal cell cluster. d, GO enrichment analysis for the dysregulated genes in Ex-L5 cell population. e, f, The dysregulated genes in GO:0050807 term formed a tightly interconnected protein-protein interaction network (e), in which the hub genes were significantly upregulated in the Ex-L5 excitatory neurons of Caln1 knockout mice (f). The Wilcoxon rank-sum test was applied to compare gene expression between the two groups.

Discussion

Schizophrenia is a polygenic psychiatric disorder characterized by clinical symptoms such as hallucinations, delusions, and disorganized thinking and behavior [51, 52]. Previous studies have implicated some risk genes like PCDHA [12], C4A [53, 54], NRXN1 [18], DISC1 [55] and MAD1 [56] in disrupting neuronal and glial development in schizophrenia. A recent GWAS identified 106 protein-coding schizophrenia risk genes by fine-mapping and functional genomic analysis [6]. However, disentangling the contribution of individual risk genes to disease pathogenesis within the polygenic architecture of schizophrenia remains challenging. To address this, we established the strategy of single-gene-knockout-precise-dorsal/ventral-forebrain-organoids (SKOPOS), enabling systematic investigation of multiple schizophrenia risk genes across excitatory and inhibitory neuronal lineages (Additional file 12: Figure S10). Based on knock-out excitatory dorsal and inhibitory ventral forebrain organoids, our study simultaneously and comprehensively explored the impacts of 11 schizophrenia risk genes on developing human brain from gene expression profiles, specific cell population and neural functional activity. According to bulk RNA-seq data, the disruption score of gene expression at both genome-wide and forebrain development levels revealed that although these risk genes disrupt the various degrees of gene expression network, most of them have significant impact on forebrain development. While no single cell population showed consistent change across all knockouts, dysregulated genes in knock-out neurons were enriched in forebrain development pathways. Besides, MEA functional analysis suggested that inhibitory neural activity is enhanced in most of these knock-out organoids. Remarkably, among those multiple schizophrenia risk genes, CALN1^−/− induces the most severe transcriptomic, cellular, and functional perturbations (Additional file 12: Figure S10). The collected results indicate that different risk genes may lead to similar functional consequences through different molecular pathways and cell populations, potentially explaining heterogeneous patient response to the same antipsychotics [57, 58].

To investigate the impact of Caln1 on postnatal mature neurons and to shed light on its clinical relevance to schizophrenia, we generated Caln1 knock-out mice. Caln1 knock-out mice not only displayed the defects of spatial memory, cognition, social ability and pre-pulse inhibition, but displayed spontaneous startle behavior and head-twitch response, which has been reported as a hallucination-like behavior in human [32–37]. And auditory hallucinations were furtherly confirmed by hallucination-like perception test. Whole-cell current-clamp recordings revealed that Caln1 loss destabilizes the excitability of PFC layer-5 pyramidal neurons-a population critically involved in perception [39]. Corroborated by snRNA-seq, Ex-L5 neurons in Caln1^−/− cortex exhibited the most severe gene network dysregulation, spanning synaptic adhesion, glutamate receptor signaling, neurotrophic signaling and synaptic organization.

Notably, while a previous study reported that Caln1 overexpression also induces schizophrenia‑like phenotypes [59], our findings uniquely establish that CALN1 loss‑of‑function exerts the most severe impact among a panel of risk genes during early human brain development, suggesting that dosage‑dependent or bidirectional dysregulation of CALN1 may converge on common pathogenic pathways.

Several limitations of this study should be acknowledged. First, due to the limited sample size in scRNA-seq analysis, we were unable to perform statistical testing of cell type proportion changes, and future studies with a larger number of independent organoids per genotype will be necessary to rigorously quantify perturbation-induced shifts in cell composition. Second, our knockout strategy uniformly reduces gene expression. However, for some of the 11 risk genes examined, the disease‑associated risk allele may lead to increased rather than decreased expression. Therefore, we cannot exclude the possibility that the phenotypic effects observed in our knockouts do not fully recapitulate the pathogenic mechanism for genes where gain‑of‑function or altered splicing is the primary risk driver. Future work employing allele‑specific expression modulation, overexpression, or knock‑in approaches will be required to determine the causal direction of effect for each individual gene.

Conclusions

In the present study, we systematically dissected the contributions of multiple schizophrenia risk genes during early human forebrain development. Among 11 risk genes examined, CALN1 knockout induced the most pronounced transcriptomic, cellular, and functional perturbations in human dorsal and ventral forebrain organoids, highlighting CALN1 as a particularly potent modulator of neurodevelopmental processes relevant to schizophrenia. Complementary Caln1 knockout mouse models recapitulated core behavioral features of schizophrenia, including cognitive deficits, social impairments, sensorimotor gating deficits, and hallucination-like responses, with layer‑5 pyramidal neurons in the prefrontal cortex exhibiting marked excitability dysregulation and gene network disruptions. By integrating human organoid and mouse models, our work provides convergent evidence that CALN1 critically influences both early corticogenesis and postnatal neuronal function.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Z.S. and HJ.L. designed the experiments. HJ.L., T.C. and J.C. performed sgRNAs design and CRISPR-Cas9 plasmid construction. HJ.L. constructed knock-out cell lines. HJ.L. induced organoids. HJ.L. and J.X. performed scRNA-seq experiments. HJ.L. performed bulk RNA-seq and scRNA-seq analysis. HJ.L. and X.Y. performed organoids immunostaining. HJ.L. performed MEA experiments and analysis. X.Y., HJ.L. and X.C. performed animal experiments and analysis. X.L. performed animal electrophysiological studies, supervised by Y.S. Z.S., HJ.L., X.Y., X.L., S.C. and Y.S. wrote the manuscript and data interpretation. All authors read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (Z.S., 32450530, 32070956; HJ.L., 82101574; Y.S., 32130044, T2241002; X.L., 32200807), The Joint Project of the Yangtze River Delta Science and Technology Innovation Community (Z.S., 2024CSJZN00600), Brain Science and Brain-like Intelligence Technology-National Science and Technology Major Project (Z.S., 2021ZD0202500), National Key Research and Development Program of China (Z.S., 2024YFA1108000), China Postdoctoral Science Foundation (HJ.L., 2021TQ0073, 2021M700837), Program of Shanghai Academic/Technology Research Leader (Y.S., 21XD1400100), MOE Frontiers Center for Brain Science fund and starting fund from Fudan Universities (Z.S.).

Data availability

All data are available in the main text or the supplementary materials. All sequence data in this study have been deposited within the Genome Sequence Archive for Human (https://ngdc.cncb.ac.cn/gsa-human/) with the accession identifier HRA005215.

Declarations

Ethics approval and consent to participate

The human embryonic stem cell line WA09 used in this study was obtained following a formal Material Transfer Agreement (WiCell Agreement No. 22.W0510). Animal experiments were conducted according to the Guidelines for the Care and Use of Laboratory Animals of Fudan University and approved by Laboratory Animal Science of Fudan University (Approval No. 2020ITBR-JS-002), China.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.