1 Introduction

Schizophrenia is a severe neuropsychiatric disorder that affects about 1% of the world's population.[1] This disorder imposes an enormous economic and mental burden on patients, their families, and society.[2, 3] Understanding the pathophysiology of schizophrenia is important as it can provide new therapeutic targets for drug development and treatment. Unfortunately, so far we know little about the pathogenesis of schizophrenia. The interactions between genetic and environmental factors are thought to be involved in schizophrenia. The heritability of schizophrenia was estimated as high as 80%,[4, 5] indicating the dominant role of genetic components in this disorder. Lee et al. showed that single nucleotide polymorphisms (SNPs) captured ≈23% schizophrenia heritability (SNP heritability), indicating the pivotal role of common variants in schizophrenia.[6] To identify common risk variants for schizophrenia, multiple genome-wide association studies (GWASs) have been conducted in different continental populations and over 200 schizophrenia risk loci have been reported.[7-15]

Although GWASs have provided important insights into the genetic architecture of schizophrenia, challenges and obstacles remain in deciphering the underlying genetic mechanisms. For example, the lead (or index) variants reported by GWASs are usually not the causal variants. Genetic variants in linkage disequilibrium (LD) often show similar P values, impeding the identification of causal variants. Moreover, most of the risk SNPs identified by GWASs are located in non-coding regions, suggesting that modulating gene expression is a primary manner that the risk variants exert their biological effects on schizophrenia susceptibility. Finally, gene regulation is a complex process and gene (or genes) nearest to the reported risk variants are not necessarily the target genes of the risk variants (risk variants may confer disease risk by modulating distal gene).[16-21] To pinpoint the plausible causal variants, we used the functional genomics approach to identify the schizophrenia risk variants that affected binding of transcription factors (TFs) recently.[22] By integrating chromatin immunoprecipitation followed by sequencing (ChIP-seq) and DNA binding motifs (position weight matrix, PWM), we identified 132 TF binding-disrupting SNPs from the reported schizophrenia risk loci (Figure S1, Supporting Information).[22] Of note, a CTCF binding-disrupting functional variant (rs2535629, located on 3p21.2) showed robust association with schizophrenia (P = 9.85 × 10−13)[23] and bipolar disorder (P = 4.93 × 10−7).[24] More importantly, we conducted an independent genetic association study in Chinese population (N = 4291 cases and 7847 controls) to replicate the association between rs2535629 and schizophrenia, and the result showed that rs2535629 was robustly associated with schizophrenia in Chinese population (P = 1.86 × 10−9), with the same risk allele as in European populations (Table 1). These lines of evidence suggest that rs2535629 is a potential causal variant with regulatory effect. However, the regulatory mechanisms and biological effects of rs2535629 in schizophrenia remain unclear. Functional understanding of this TF binding-disrupting variant will not only provides important insights into the genetic etiology and biological mechanisms of schizophrenia, but also brings new opportunities to elucidate the pathogenesis of schizophrenia and potential therapeutic targets.

Table 1. rs2535629 is significantly associated with schizophrenia in Chinese, meta-analysis of rs2535629 in Europeans and Chinese populations SNP ID Sample Cases Controls Allele (A1/A2) P value ORa) rs2535629 Chinese sample 4291 7847 A/G 1.86 × 10−9 0.8461 PGC2+EAS 56418 78818 A/G 1.87 × 10−9 0.9466 Combined 60709 86665 A/G 2.90 × 10−14 0.9364

In this study, we systematically investigated the regulatory mechanisms and biological effects of rs2535629. We first showed that rs2535629 is located in an active regulatory element bound with CTCF (in neuronal cells)[22] and confirmed that CTCF could bind to the genomic sequence containing rs2535629. We then validated the regulatory effect of rs2535629 with reporter gene assays. Interestingly, although rs2535629 is located in the intron 7 of ITIH3 gene, expression quantitative traits loci (eQTL) analysis showed that this functional SNP was associated with expression of three distal genes (GLT8D1, NEK4, and SFMBT1) in the human brain. CRISPR-Cas9-mediated genome editing further validated the regulatory effect of rs2535629 on GLT8D1, NEK4, and SFMBT1. Of note, gene expression showed that SFMBT1 was dysregulated in neurons induced from fibroblasts of schizophrenia cases compared with controls, and rs2535629 physically interacts with SFMBT1, supporting that this functional variant confers risk of schizophrenia by modulating SFMBT1 expression. Finally, we explored the potential roles of Sfmbt1 in schizophrenia pathogenesis by using the neural stem cell model (including proliferation and differentiation of neural stem cells) and dendritic spine density analysis. Our study identified rs2535629 as a likely causal variant (for schizophrenia) at the 3p21.2 risk locus and demonstrated that this functional variant regulates SFMBT1 expression by affecting CTCF binding, offering pivotal insights into genetic mechanisms and pathogenesis of schizophrenia.

2 Results 2.1 Functional Genomics Identified rs2535629 as a Plausible Causal Variant at the 3p21.2 Locus

Our functional genomics analysis showed that rs2535629 is a functional variant at the 3p21.2 schizophrenia risk locus.[22] SNP rs2535629 resides in a genomic region with multiple SNPs showing significant associations with schizophrenia (Figure 1a). The association significance between rs2535629 and schizophrenia in Europeans (40 675 cases and 64 643 controls) is P = 1.48 × 10−7 (Figure 1a).[25] However, rs2535629 showed robust association with schizophrenia (P = 9.85 × 10−13) in a recent larger GWAS (67 390 cases and 94 015 controls).[23] If rs2535629 is a bona fide causal SNP, it may also be associated with schizophrenia in non-European populations such as the Chinese population. To further verify the association between rs2535629 and schizophrenia, we conducted a replication study in Chinese population (4291 cases and 7847 controls). We found that rs2535629 was also significantly associated with schizophrenia in the Chinese cohort (P = 1.86 × 10−9) (Table 1), with the risk allele as reported in the Europeans. We further investigated the association between rs2535629 and schizophrenia by combining results from European GWAS (56 418 cases and 78 818 controls)[9] and our cohort. Meta-analysis (a total of 60 709 cases and 86 665 controls) showed a strong association between rs2535629 and schizophrenia (P = 2.90 × 10−14) (Table 1). Of note, rs2535629 has a similar minor allele frequency (MAF) in European (MAF, 37%, minor allele: A) and southern Han Chinese (MAF, 37%) populations (Figure S2, Supporting Information). In northern Han Chinese (Han Chinese in Beijing, CHB), the MAF is 47%. In line with the similar MAF in European and Chinese populations, LD analysis also showed a similar LD structure of the genomic region surrounding rs2535629 in European (EUR) and Chinese populations (Figure S3, Supporting Information), supporting that the genomic region surrounding rs2535629 may be a common risk locus for schizophrenia. rs2535629 lies in the seventh intron of ITIH3 gene on chromosome 3 (Figure 1b). Our previous functional genomics showed that rs2535629 is located in a well-defined CTCF binding motif, with different alleles of rs2535629 affected CTCF binding affinities (Figure 1c). The strong CTCF ChIP-Seq signal covering rs2535629 indicated that CTCF bound to the genomic region containing rs2535629 in neuronal cells and tumor cell lines from the human brain (including neuroblastoma and medulloblastoma cell lines) in vivo (Figure 1d).

rs2535629 is a functional variant that disrupts CTCF binding at the 3p21.1 risk locus. a) Locuszoom plot showed that rs2535629 is located in a genomic region with multiple SNPs showing significant associations with schizophrenia.[25] b) rs2535629 is located in the seventh intron of ITIH3. c) rs2535629 is located in the binding motif of CTCF and different alleles of rs2535628 altered CTCF binding. d) Chromatin immunoprecipitation and sequencing (ChIP-Seq) data showed that CTCF preferentially bound to genomic sequence containing rs2535629 in neuronal cells and tumor cell lines from the human brain (including neuroblastoma and medulloblastoma cell lines) in vivo. The heights of the colored graphs reflect the ChIP-Seq signal intensities, and the location of rs2535629 is highlighted with the red line. 2.2 Reporter Gene Assays Validated rs2535629 as a Regulatory Variant

As rs2535629 resides in the intronic region, we hypothesized that this TF binding-disrupting SNP may be located in a regulatory element to modulate its target gene (or genes). To test the regulatory activity of the genomic sequence containing rs2535629, we conducted dual-luciferase reporter gene assays. Reporter gene assays showed that rs2535629 is located in a repressive element, as the luciferase activity of the fragments containing rs2535629 was lower than controls (Figure 2a–d). In addition, reporter gene assays also revealed that different alleles of rs2535629 affected luciferase activity significantly (Figure 2a–d). The G allele (risk allele) of rs2535629 conferred significantly lower luciferase activity than A allele in all the tested cells (Figure 2a–d). These results validated the regulatory effect (or functional consequences) of rs2535629 and suggested that rs2535629 may exert its biological effect on schizophrenia by modulating the transcription activity of the regulatory element it located.

Validation of the regulatory effects of rs2535629. a–d) Dual-luciferase reporter gene assays (enhancer assay, the cloned fragments were cloned into pGL3-promoter vector, which is used to test the enhancer activity of the cloned fragments) showed that the G allele (risk allele) of rs2535629 conferred significant lower luciferase activity than A allele in HEK-293T cells (a), SK-N-SH cells (b), SH-SY5Y cells (c), and U251 cells (d). Data represent mean ± SD, n = 8 for control group, n = 16 for each experimental group in HEK293T, and n = 24 for each group in SK-N-SH, SH-SY5Y, and U251 cells. e) Significant enrichment of CTCF on genomic sequence containing rs2535629. DNA templates from cross-linked chromatins of cells (without CTCF immunoprecipitation) were used in INPUT group, and DNA templates from cross-linked chromatins of cells (with CTCF immunoprecipitation) were used in CTCF group. n = 3 for each group. f) ChIP-AS-qPCR revealed that A allele of rs2535629 was preferentially bound to CTCF compared with G allele. DNA templates from cross-linked chromatins of cells (with CTCF immunoprecipitation (left panel) and IgG immunoprecipitation (left panel)) were used. Different alleles of rs2535629 had different CTCF enrichment efficiencies and the results reflected that the binding affinity (to CTCF) of A allele was higher than G. n = 3 for each group. g) ChIP-Sanger sequencing showed stronger CTCF binding to A allele of rs2535629 than G allele. P values were calculated using the Student's t-test (two-tailed) in (a–f). **P <0.01, ***P < 0.001.

2.3 rs2535629 Affects CTCF Binding

To further verify if different alleles of rs2535629 alter CTCF binding affinity, we conducted ChIP-Allele-Specific-qPCR (ChIP-AS-qPCR) in SH-SY5Y (a neuroblastoma cell line) cells. We first carried out ChIP-qPCR and found significant enrichment of CTCF binding on the sequence containing rs2535629 compared with the control, supporting that CTCF preferentially binds to the genomic region containing rs2535629 (Figure 2e). ChIP-AS-qPCR further revealed that CTCF preferentially binds to the A allele of rs2535629 compared with G allele (Figure 2f). Consistent with ChIP-AS-qPCR results, ChIP-Sanger sequencing also indicated that CTCF prefers to bind the A allele than G allele at rs2535629 (Figure 2g). These results demonstrated that CTCF bound to the genomic sequence containing rs2535629 in vivo and revealed that different alleles of rs2535629 altered the binding affinity of CTCF.

2.4 rs2535629 is Associated with SFMBT1 Expression in the Human Brain

As rs2535629 is located in the intronic region and is a regulatory variant, we hypothesized that this functional variant may confer schizophrenia risk by modulating gene expression. We therefore examined if rs2535629 is associated with gene expression in the human brain by using eQTL data from the Common Mind Consortium (CMC).[26] eQTL analysis showed that rs2535629 was significantly associated with the expression of three genes (GLT8D1, NEK4, and SFMBT1) in the human brain (corrected P < 0.05, Figure 3a). Although rs2535629 is located in intronic region of ITIH3, no significant association between rs2535629 and ITIH3 expression was detected in the human brain. Of note, the risk allele (G) of rs2535629 is associated with a higher expression level of NEK4 (Figure 3b), and a lower expression level of SFMBT1 and GLT8D1 (Figure 3c,d). These eQTL results indicated that rs2535629 may confer schizophrenia risk by regulating these three genes.

rs2535629 is associated with the expression of GLT8D1, NEK4, and SFMBT1 in human brain tissues. a) eQTL analysis of rs2535629 and its surrounding genes (within 1M bp window, 500 kb upstream, and 500 kb downstream of rs2535629, respectively) in CMC brain eQTL dataset. Expression of GLT8D1, NEK4, and SFMBT1 showed the most significant associations with rs2535629. The horizontal red line shows the corrected P value (-log10 (Pvalue/number of genes)). b–d) eQTL analysis demonstrated the correlations between three rs2535629 genotypes and the expression of NEK4, GLT8D1, and SFMBT1. e) SFMBT1 was significantly down-regulated in neuron (induced from hiPSs) of schizophrenia cases compared with controls in GSE25673 dataset[28] (P = 0.001). 2.5 rs2535629 Regulates SFMBT1 Expression via CTCF

To test if rs2535629 regulates the expression of its eQTL genes by interacting with CTCF, we knocked down CTCF expression in SH-SY5Y (a neuroblastoma cell line) and U251 (a human glioblastoma cell line) cells (Figure 4a,b). CTCF knockdown affected the expression of NEK4, GLT8D1, and SFMBT1, indicating that CTCF regulates these three genes (Figure 4a,b). We also examined the effect of CTCF knockdown on Sfmbt1 expression in GEO dataset (GSE99230). In GSE99230 dataset, CTCF knocked-down up-regulated Sfmbt1 expression by 25%, which is consistent with our observation. Of note, all three genes showed significant up-regulation in CTCF knockdown cells, suggesting a repressive effect of CTCF on these three genes.

Regulatory effects of CTCF and rs2535629 on GLT8D1, NEK4, and SFMBT1. a) CTCF knockdown in U251 cells affected GLT8D1, SFMBT1, and NEK4 expression significantly. b) CTCF knockdown in SH-SY5Y cells up-regulated expression of GLT8D1, SFMBT1, and NEK4 significantly. c–f) rs2535629 knockout up-regulated expression of GLT8D1, SFMBT1, and NEK4 in SH-SY5Y cells. (c) Schematic diagram of rs2535629 knockout. The fragment length of wildtype cells is 527 bp. However, the fragment length of the genomic sequence containing rs2535629 in knocked-out cells is 135 bp (as a 392 bp sequence containing rs2535629 was deleted by CRISPR-Cas9). n = 3 for each group. Two-tailed Student's t-test was used for detection of statistical significance, *P < 0.05, **P <0.01, ***P < 0.001.

To further verify if the regulatory effects of CTCF on NEK4, GLT8D1, and SFMBT1 are mediated by rs2535629, we knocked out a 392 bp genomic sequence containing rs2535629 in SH-SY5Y cell line (Figure 4c). Knockout of the genomic region containing SNP rs2535629 up-regulated the expression of NEK4, GLT8D1, and SFMBT1 (Figure 4d–f), corroborating that the sequence containing rs2535629 acts as a repressive element for these genes (which was consistent with reporter gene assays). Considering that CTCF preferentially binds to the A allele rs2535629 and CTCF knockdown resulted in significant up-regulation of NEK4, GLT8D1, and SFMBT1, these data collectively suggested that rs2535629 may regulate the expression of NEK4, GLT8D1, and SFMBT1 via interacting with CTCF.

2.6 Dysregulation of SFMBT1 in Schizophrenia Cases

To further test if rs2535629 conferred schizophrenia risk by modulating the expression of NEK4, GLT8D1, and SFMBT1, we examined the expression level of these three genes in brains of schizophrenia cases and controls using data from the PsychENCODE[27] (559 cases and 936 controls). None of them showed significant expression alteration in brains of schizophrenia cases compared with controls. We thus further examined the expression of these three genes in neurons (differentiated from human induced pluripotent stem cells, hiPSCs) induced from hiPSCs of schizophrenia cases compared with controls (GSE25673).[28]GLT8D1 and NEK4 did not show significant expression change in neurons induced from schizophrenia cases compared with controls. However, SFMBT1 was significantly down-regulated in neurons induced from schizophrenia cases compared with controls (P = 8.90 × 10−4, FDR<0.05) (Figure 3e). These data suggested that dysregulation of SFMBT1 might have a role in schizophrenia. Of note, eQTL analysis predicted down-regulation of GLT8D1 and SFMBT1 in schizophrenia cases compared with controls as the risk allele (G) was associated with lower expression of GLT8D1 and SFMBT1 (Figure 2c,d), which was concordant with the observation of significant down-regulation of SFMBT1 in neurons differentiated from hiPSCs of schizophrenia cases. Taken together, these data suggested that the functional regulatory variant rs2535629 might confer schizophrenia risk by regulating SFMBT1.

2.7 Sfmbt1 Regulates Proliferation of NSCs

Our above data suggest that the gene (or genes) regulated by rs2535629 may be involved in the pathogenesis of schizophrenia. To test this hypothesis, we investigated the role of genes regulated by rs2535629 in neurodevelopment. As we have demonstrated the potential role of GLT8D1 in schizophrenia in our previous study[29] and expression analysis showed no significant change of NEK4 and GLT8D1 in schizophrenia cases compared with controls, we focused on SFMBT1 in this study.

To date, the pathogenesis of schizophrenia remains largely unknown. However, accumulating evidence supports the neurodevelopmental hypothesis of schizophrenia (which posits that schizophrenia is mainly attributed to abnormal brain development). Consistent with the neurodevelopmental hypothesis, the neural stem cell model (a model that was widely used to explore the function of SCZ-associated genes)[30-34] also has revealed the pivotal of some schizophrenia risk genes in neurodevelopment (including proliferation, migration, and differentiation). We thus used the neural stem cell model to investigate the role of SFMBT1 in neurodevelopment. Of note, 84% coding sequence (https://blast.ncbi.nlm.nih.gov/Blast) and 89% protein sequence (https://www.uniprot.org/align) of SFMBT1 were identical between humans and mice, suggesting that the function of SFMBT1 is relatively conserved in humans and mice. The isolated mouse neural stem cells (mNSCs) expressed the three well-characterized markers for NSCs (SOX2, NESTIN, and PAX6) (Figure 5a), indicating that these cells were NSCs. We then knocked-down Sfmbt1 expression in mNSCs using shRNAs (Figure 5). Both BrdU (5-bromodeoxyuracil nucleotide) incorporation and CCK8 assays showed that Sfmbt1 knock-down impaired the proliferation of NSCs significantly (Figure 5b–d). These results demonstrated the important role of Sfmbt1 in regulating the proliferation of NSCs.

Sfmbt1 knockdown affected proliferation of NSCs. a) Co-labeling (Immunofluorescence) with three well-characterized markers for NSCs validated the identity of isolated mouse neural stem cells. b,c) BrdU assays showed that Sfmbt1 knockdown impaired the proliferation abilities of NSCs. n = 3 (3 independent biological replicates) for each group and five technical replicates were performed for each biological replicate. c) Quantification data for (b). d) Relative expression of Sfmbt1 in control and shRNA-mediated knocked-down mNSCs (measured by real-time qPCR) (n = 3). e) CCK8 assays revealed that Sfmbt1 knockdown affected proliferation of NSCs significantly. n = 10 for each group. P values were calculated using the Student's t-test (two-tailed) in (c–e). *P < 0.05, **P <0.01, ***P < 0.001.

2.8 Sfmbt1 Regulates Differentiation of NSCs

In addition to proliferation (which occurs during the early developmental stage), differentiation also plays an important role in neurodevelopment as functionally distinct neurons are generated in the process of differentiation. To further investigate the role of Sfmbt1 in neuronal differentiation, we differentiated the mNSCs into neurons and glia cells. The results showed that the proportion of GFAP positive glia cells was significantly reduced in Sfmbt1 knocked-down cells compared with controls (Figure 6a,b). However, Sfmbt1 knock-down resulted in a significant increase of MAP2 positive neurons (Figure 6c,d). We also examined the mRNA expression level of Gfap (a marker for glia cells), Map2 (a marker for mature neurons), and Tuj1 (a marker for newly generated immature post-mitotic neurons) with qPCR. Consistent with immunofluorescence results, the mRNA expression level of Gfap was significantly decreased in Sfmbt1 knocked-down cells compared with controls (Figure 6e). By contrast, the mRNA expression level of Map2 and Tuj1 was significantly increased (Figure 6f,g). These results indicate that Sfmbt1 plays a role in the differentiation of NSCs.

Sfmbt1 knockdown affected differentiation of NSCs. a,b) Sfmbt1 knockdown inhibited the differentiation of NSCs into GFAP positive astrocytes. n = 3 (3 independent biological replicates) for each group and four technical replicates were performed for each biological replicate. c,d) Sfmbt1 knockdown promoted the differentiation of NSCs into MAP2 positive neurons. n = 3 (3 independent biological replicates) for each group and four technical replicates were performed for each biological replicate. e–g) qPCR showed the expression change of Gfap, Map2, and Tuj1 in Sfmbt1 knockdown cells. P values were calculated using the Student's t-test (two-tailed) in (b,d) and (e–g). **P <0.01, ***P < 0.001.

2.9 Sfmbt1 Regulates Neurodevelopment-Associated Pathways

To further investigate the pathways (or biological processes) regulated by Sfmbt1, we conducted transcriptome analysis using RNA sequencing (RNA-Seq). A total of 3092 genes showed significant expression change (Padj<0.01, |Log2 (fold change)|>0.5) in Sfmbt1 knocked down NSCs compared with controls (Figure S4, Supporting Information). The top 30 differentially expressed genes (DEGs) are shown in Figure 7a. Gene Ontology (GO) analysis showed that the DEGs were mainly enriched in neurodevelopment-associated pathways, including negative regulation of nervous system development, negative regulation of neurogenesis, axonogenesis, glial cell differentiation, and gliogenesis (Figure 7b). Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis showed that the DEGs were mainly enriched in the pathways related to the neuronal synapse (including glutamatergic, dopaminergic, and cholinergic) (Figure 7c), suggesting that Sfmbt1 has an important function in the central nervous system (CNS) and synaptic transmission. In addition, DEGs were also enriched in cell cycle-related pathways (Figure 7c) such as p53 signaling pathway, which is consistent with the results of BrdU and CCK8 proliferation assays. Of note, previous studies have shown that schizophrenia risk genes were enriched in pathways related to neurodevelopment and synaptic transmission.[35-38] Besides, spatio-temporal expression pattern analysis showed that SFMBT1 expression level was higher in the developing brain (prenatal stages) than adulthood brain (postnatal stages) (Figure S5, Supporting Information),[39] suggesting this gene may have a pivotal role in human brain development. Collectively, these results indicate that Sfmbt1 regulates important pathways related to neuronal-related cell components and neurological diseases, further indicating the important role of Sfmbt1 in the CNS.

Sfmbt1 regulates neurodevelopment-associated pathways and synaptic organization. a) The top 30 genes that showed the most significant expression differences in Sfmbt1 knocked-down and control cells. b) GO analysis of the DEGs. The DEGs were enriched in schizophrenia-associated pathways, including negative regulation of nervous system development, negative regulation of neurogenesis, axonogenesis, glial cell differentiation, and gliogenesis. The black vertical dashed bar indicates adjusted P<0.05. Neurodevelopment-associated pathways were highlighted with red. c) KEGG analysis showed enrichment of DEGs in neuronal synapse (including Glutamatergic synapse, Dopaminergic synapse, MAPK signaling pathway, Cholinergic synapse, and Calcium signaling pathways) and signaling pathways, indicating the pivotal role of Sfmbt1 in the development and function of synapse. The black vertical bar indicates adjusted P<0.05.

2.10 Sfmbt1 Knockdown Affects the Dendritic Spine Density of Neurons

Synapses are connections between axons and dendrites of different neurons, which are the basis of information processing and storage in the brain.[40] Previous studies have shown that dysfunction of synaptic transmission had a critical role in schizophrenia.[41-44] Of note, morphological alterations of synapses, including the number, density, size, and shape of dendritic spines have been frequently reported in schizophrenia.[45-49] It has been reported that the density of spine decreased in schizophrenia due to excessive spine pruning during late childhood or adolescence,[49] indicating the pivotal role of synaptic dysfunction in schizophrenia. Dendritic spines can be divided into mushroom, stubby, thin, and branched subtypes based on their shape.[50, 51] Mushroom spines are regarded as long-term stable dendritic spines, which mainly correspond to stable and mature synaptic connections. They have larger spine heads, and the size of their spine heads is positively correlated with the size of the excitatory synaptic post-synaptic density (PSD) and the strength of the synapse.[52, 53] Thin and stubby are regarded as transient, immature dendritic spines with the property of quick formation and/or elimination.[54, 55] The dynamic changes of dendritic spines are closely related to learning and memory at the synaptic level. Newly formed dendritic spines (most of them are thin spines) are the structural basis for memory acquisition, and stable dendritic spines (most of them are mushroom spines) are responsible for memory storage. We therefore investigated the role of Sfmbt1 in spine morphogenesis by knocking-down Sfmbt1 expression in primary neurons (Figure 8). We found that Sfmbt1 knockdown resulted in a significant decrease in the density of stubby, thin, and mushroom spines (Figure 8f). In addition, the proportion of thin spines was also significantly decreased in Sfmbt1 knocked-down neurons (Figure 8g). These results recapitulated the synaptic abnormality observed in schizophrenia, indicating that Sfmbt1 may confer risk of schizophrenia by regulating spine structure and function.

Sfmbt1 knockdown affected the density and proportion of the dendritic spines. a–c) Confocal images of whole neurons transfected with control and Sfmbt1 knockdown vectors (scale bars represent 20 µm). Dendritic branches were captured from each corresponding neuron, respectively (scale bars represent 2 µm). Green fluorescence is the result of immunofluorescence of GFP protein expressed by co-transfected GFP vector. d) A schematic diagram to show different types of spines. e) Relative expression of Sfmbt1 in control and Sfmbt1 knocked-down neurons (measured by real-time qPCR). (n = 3). f) Density analysis of each dendritic spine subtype. g) Percentage analysis of each dendritic spine subtype. n = 33 for control group, n = 30 for shRNA#1 group, and n = 27 for shRNA#2 group. P values were calculated using the Student's t-test (two-tailed) in (e–g). *P < 0.05, **P <0.01, ***P < 0.001.

3 Discussion

So far, GWASs have identified multiple risk loci that showed robust associations with schizophrenia.[7-15] Despite the fact that these large-scale GWASs have provided important insights into the genetic architecture of schizophrenia, how to translate the GWAS findings into biological mechanisms and potential therapeutic targets remain major challenges in the post-GWAS era. The first step toward mechanistic understanding is to pinpoint the functional (or potential causal) variants from the reported risk loci, a daunting task that is usually impeded by the complex LD pattern and gene regulation. We conducted a functional genomics study on schizophrenia and identified 132 TF binding-disrupting SNPs in our previous study.[22] SNP rs2535629 represents a promising causal variant among the 132 TF binding-disrupting SNPs. We thus systematically investigated the regulatory mechanism of rs2535629 in this study. We provided convergent and consistent results to prove the functionality of rs2535629 in this study. First, ChIP-seq and PWM data indicated that rs2535629 is located in a CTCF binding motif with CTCF binding in neuronal cells (Figure 1c,d). Second, FIMO analysis suggested that rs2535629 affected CTCF binding (Figure 1c). Third, we showed that CTCF bound to the sequence containing rs2535629 in vivo and different alleles of rs2535629 altered the binding affinity of CTCF (Figure 2e–g). Fourth, we validated the regulatory effect of rs2535629 using reporter gene assays (Figure 2a–d). These results collectively confirmed that rs2535629 is a functional variant with regulatory effect.

In addition to elucidating the regulatory mechanism of rs2535629, we also identified the potential target genes regulated by rs2535629. Our eQTL analysis showed that rs2535629 was associated with the expression of GLT8D1, NKE4, and SFMBT1 in the human brain (Figure 3), suggesting that rs2535629 may confer schizophrenia risk by modulating these genes. We further verified the regulatory effect of CTCF and rs2535629 on these three genes (Figure 4). Of note, we previously showed that GLT8D1 and NEK4 might have a role in schizophrenia.[29, 56] These results indicated that SFMBT1 might also be one of the potential causal genes (at this risk locus) regulated by rs2535629. We thus characterized the function of Sfmbt1 in the CNS. We found that Sfmbt1 not only regulated proliferation and differentiation of NSCs, but also affected the dendritic spine density of neurons, implicating that this gene may be involved in schizophrenia pathophysiology by affecting neurodevelopment and synaptic transmission. It should be noted that we only investigated the effect of SFMBT1 on the differentiation of NSCs into GFAP positive glia cells and MAP2 positive neurons. Considering there are many cell types (including oligodendrocytes and microglia cells) in the brain, more work is needed to explore if SFMBT1 affects the differentiation of NSCs into other cell types.

In addition to GLT8D1[29] and NEK4,[56] our study also suggests that SFMBT1 may be a potential risk gene at this locus, and functional SNP rs2535629 may confer schizophrenia risk by regulating SFMBT1. We noticed that rs2535629 is located in the intron 7 of ITIH3 gene and the distance between rs2535629 and SFMBT1 is about 104 kb (Figure 9), suggesting that rs2535629 may regulate SFMBT1 through long-range chromatin interaction. We thus examined if rs2535629 interacts with SFMBT1 in neuronal cells using the 3D-genome Interaction Viewer and database.[57] Intriguingly, rs2535629 physically interacts with SFMBT1 (but not NEK4 and GLT8D1) in the dorsolateral prefrontal cortex (human) (Figure S6, Supporting Information). More importantly, a recent study by Jung et al. also showed that rs2535629 physically interacts with SFMBT1 promoter.[58] Of note, rs2535629 disrupted the binding of CTCF, a key regulator which mediates chromatin interaction.[59, 60] These data suggested that rs2535629 regulates SFMBT1 expression by affecting long-range chromatin interaction. Consistent with this, our recent transcriptome-wide association also showed that SFMBT1 is a schizophrenia risk gene whose expression level change may have a role in schizophrenia.[61] These convergent lines of evidence suggest that SFMBT1 is a potential schizophrenia risk gene at this locus.

The working model of rs2535629 in schizophrenia pathogenesis. Locuszoom plot revealed the association significances between SNPs located in the 3p21.2 region and schizophrenia. Of note, rs2535629 is located in the seventh intron of ITIH3 and there are multiple SNPs that showed a high degree of LD with rs2535629. These SNPs had similar association P values, making it difficult to pinpoint the causal variant. By using functional genomics, we identified rs2535629 as a regulatory variant in the 3p21.2 risk locus. We validated the regulatory mechanism of rs2535629 with serial experiments, including reporter gene assays, ChIP-AS-qPCR, TFs knockdown, and genome editing. In addition, an independent genetic association study further confirmed that rs2535629 was associated with schizophrenia in the Chinese population. rs2535629 physically interacts with SFMBT1 (by long-range chromatin interaction) and the schizophrenia risk allele of rs2535629 (G) regulates SFMBT1 expression by altering CTCF binding, which resulted in lower SFMBT1 expression. Perturbation of SFMBT1expression inhibits proliferation of NSCs and differentiation of NSCs into glial cells, as well as promotes differentiation of NSCs into neurons. Sfmbt1 expression perturbation also affected genes and pathways associated with neurodevelopmental and synaptic transmission. These results demonstrated that the functional variant rs2535629 in ITIH3 conferred schizophrenia risk through regulating (via long-range chromatin interaction) expression of SFMBT1, a gene whose expression perturbation affected neurodevelopment.

SFMBT1 encodes Scm-like with four malignant brain tumor repeat (MBT) domains 1, which is postulated to be a histone-binding protein.[62]SFMBT1 mediates the recruitment of transcriptional inhibition complexes to target genes to repress transcription via compressing chromatin.[63, 64]SFMBT1 regulates important cellular functions, including cell differentiation,[63] migration[65] and gene transcription. Expression analysis showed that SFMBT1 is widely expressed in human tissues and many brain regions (Figures S7 and S8, Supporting Information).[66, 67] However, SFMBT1 expression in brain tissues is relatively low compared with other tissues (Figure S7, Supporting Information). In addition, cell-type-specific expression analysis (using the expression data from the UCSC cell browser (https://cells.ucsc.edu/?ds=autism&gene=SFMBT1)) showed that SFMBT1 expression is relatively high in L2/3 (upper-layer excitatory neurons), L5/6 (deep-layer cortico-cortical excitatory projection neurons), VIP interneurons and parvalbumin interneurons (Figure S9, Supporting Information). Of note, SFMBT1 copy number abnormality has been identified in patients with brain diseases[68,