Functional variants at 1p36.23 confer risk of schizophrenia through modulating RERE

Journal: 2026;17():1742  |  DOI: 10.1038/s41467-026-68449-6  |  PMID: 41580391  |  PMCID:

Authors: Yixing Liu; Junyang Wang; Hong Yang; Yifan Li; Changgai Mu; Xinglun Dang; Xinran Chen; Danyang Li; Mingna Li; Jiewei Liu; Shiwu Li; Jinfeng Yang; Xi Chen; Chen Zhang; Jin Feng; Xiong-Jian Luo

Abstract

Genome-wide association studies have identified 1p36.23 as a schizophrenia risk locus. However, the functional variants and genes driving the association remain unknown. Here, we identified two functional variants (i.e., rs159961 and rs301792) at the 1p36.23 risk locus. Both variants reside introns of RERE and exhibit allele-specific enhancer activity. Risk alleles of rs159961 and rs301792 increase enhancer activity by altering REST and POLR2A binding, leading to RERE upregulation. Consistently, RERE was significantly elevated in brains of schizophrenia cases. Functionally, RERE-overexpression impaired neurogenesis, altered dendritic spine density and dendritic complexity, and altered genes related to dendrite development and glutamatergic synapses. Through interacting with RARB and RXRA at the Grin2a promoter, RERE regulates the well-known schizophrenia risk gene Grin2a (encodes an NMDAR subunit), and RERE-overexpression impairs excitatory synaptic transmission. Our study indicates that functional variants rs159961 and rs301792 confer schizophrenia risk by upregulating RERE, which affects neuronal development and synaptic function.

Subject terms: Schizophrenia, Functional genomics, Genetics of the nervous system, Disease genetics

Introduction

Schizophrenia (SCZ) is a severe and heterogeneous psychiatric disorder^[1] that affects approximately 1% of the global population. Due to its high prevalence, mortality, and morbidity, SCZ imposes a substantial economic burden and ranks among the top ten causes of disability worldwide^[2]. Unfortunately, the etiology of SCZ remains incompletely elucidated. Accumulating evidence indicates that genetic factors play a major role in the development of the disease^[3]–[5]. The heritability of SCZ is about 81%^[3], underscoring the major contribution of genetic factors. So far, genome-wide association studies (GWASs) have identified more than 300 SCZ risk loci^[6]–[11]. However, translating these genetic findings into insights regarding the underlying pathomechanisms remains a major challenge. Identifying potential causal variants and genes (or variants and genes with functional consequences) from risk loci is the first step toward a mechanistic understanding of disease biology. However, this step is hampered by the linkage disequilibrium (LD) among risk variants and the complex gene regulation process. As each risk locus usually contains hundreds or thousands of risk variants that show high-level LD, prioritizing causal variants from risk loci is a daunting task. Besides, even if the causal variant has been pinpointed, identifying the target gene affected by the causal variant is difficult, as many genetic variants regulate the expression of distal genes rather than genes located nearest to the risk variants. Thus, there is a pressing need to identify functional (or potential causal) variants from the reported risk loci.

To prioritize potential functional or causal variants from SCZ risk loci, we previously applied a functional genomics approach and identified 132 functional (i.e., affecting transcription factor binding) single-nucleotide polymorphisms (SNPs)^[12]. Among them, two functional SNPs (i.e., rs159961 and rs301792) on chromosome 1p36.23 showed robust associations with SCZ. These two functional variants reside in the intronic regions of RERE, a gene which was prioritized as the credible causal gene underlying 1p36.23 through fine-mapping^[10], and affect the binding of transcription factors (TFs) REST and POLR2A. Although our functional genomics study has prioritized these two functional variants as likely causal variants at the 1p36.23 risk locus, the mechanisms by which they confer SCZ risk remain unknown. Specifically, the molecular mechanisms of their regulatory effects, the identity of their target gene, and the role of this gene in SCZ pathogenesis remain to be clarified.

In this study, we first demonstrated the regulatory mechanisms of the two variants with dual-luciferase reporter gene assays and electrophoretic mobility shift assays (EMSA). We then identified RERE as the target gene regulated by these variants through expression quantitative trait loci (eQTL) analysis, CRISPR/Cas9-mediated knockout assays, and TF knockdown assays, among others. Finally, we showed that RERE plays a critical role in regulating neurogenesis, dendritic spine morphogenesis, dendritic complexity, the expression of N-methyl-D-aspartate (NMDA) receptor subunit GluN2A, and excitatory synaptic transmission. Our study identified likely causal variants and genes at the 1p36.23 risk locus and elucidated their underlying mechanisms in SCZ. These findings translate GWAS results into disease biology and provide important insights into the pathogenesis of SCZ.

Results

rs159961 and rs301792 are functional SNPs at the 1p36.23 risk locus

GWASs have shown that 1p36.23 is a robust risk locus for SCZ^{[6]–[8],[10]}. To pinpoint the likely causal variants from this locus, we previously used a functional genomics approach to identify functional (i.e., TF binding-affecting) SNPs. Our functional genomics approach showed that rs159961 and rs301792 are functional variants at the 1p36.23 risk locus^[12]. To further explore whether rs159961 and rs301792 are potential causal variants, we performed fine-mapping using SuSiE^[13], CARMA^[14], and FINEMAP^[15]. Genetic associations from populations of European-ancestry^[11] were used as input, and the results indicated the presence of an independent causal signal at this locus. SuSiE analysis yielded a 95% credible set that includes 55 SNPs. CARMA fine-mapping identified 102 SNPs, which comprise the 99% credible set. The 92% credible set comprised 50 variants in FINEMAP analysis. Notably, both rs301792 and rs159961 were included in the credible sets in all three analyses, with low posterior inclusion probabilities (PIPs) (rs301792 (PIP of SuSiE: 0.0073, CARMA: 0.0052, FINEMAP: 0.0066) and rs159961 (PIP of SuSiE: 0.0066, CARMA: 0.0046, FINEMAP: 0.0058)). The full results of the fine-mapping are provided in supplementary material (Supplementary Data –).

SNPs rs159961 and rs301792 showed strong associations with SCZ in European populations^[10]. These associations were confirmed in both East Asian populations (29,519 cases, 44,392 controls) and cross-ancestry populations (96,806 cases and 492,818 controls) in our replication study^[11], with the same risk alleles as in Europeans (Fig. and Supplementary Fig.). These results indicate that these two variants are credible risk variants for SCZ. Both rs159961 and rs301792 are located in the intronic regions of RERE (Fig.), and the intronic sequences containing rs159961 and rs301792 show enrichment for REST and POLR2A binding motifs (Fig.). To validate the functionality of these two SNPs, we examined whether these two SNPs are located in active regulatory elements. Epigenomic data indicated that rs159961 and rs301792 are located in open chromatin regions (as detected by DNase-seq) that were marked with enhancer-associated histone modifications (H3K27ac, H3K4me1, and H3K9ac) in human neuronal cells and human brain tissues (Fig.). To further assess the functional consequences of rs159961 and rs301792, we conducted dual-luciferase reporter gene assays. Both risk alleles (rs159961-T and rs301792-G) exhibited increased enhancer activity in SH-SY5Y cells (a human neuroblastoma cell line) compared with protective alleles (rs159961-C and rs301792-A) (Fig.). In U251 cells (a human glioblastoma cell line), the rs159961-T allele also exhibited higher enhancer activity than the C allele, whereas rs301792 alleles showed no significant difference (Fig.). Notably, the allele-specific enhancer effects were absent in HEK-293T cells (Fig.), indicating the regulatory effects of these variants are neuronal-dependent. Collectively, these results support the regulatory effects of these two functional variants.

[Fig. 1] rs159961 and rs301792 are functional genetic variants at the SCZ risk locus 1p36.23.a LocusZoom plots depict the association between genetic variations near rs159961 and rs301792 (±300 kb) and SCZ across ancestries, including East Asian, European, Latino, and African-American populations (96,806 cases and 492,818 controls)^[11]. b rs159961 and rs301792 are located in the intron region of RERE. c, d Epigenomic data indicate that in human neuronal cell lines (including neural progenitor cells, neural stem progenitor cells, neuroepithelial stem cells, SK-N-SH, and neurons) and human brain tissue, there is evidence of chromatin opening and enhancer modification signals, including DNase-seq, H3K27ac, H3K4me1, and H3K9ac, as well as binding signals of REST and POLR2A, within the chromatin regions encompassing rs159961 and rs301792. Red lines indicate the genomic locations of rs159961 and rs301792. Data source: https://epigenomegateway.wustl.edu/browser/^[80]. e, f Position weight matrices (PWMs) show that rs159961 (in PFSK-1 cells) and rs301792 (in SK-N-MC cells) are located in REST and POLR2A binding motifs. Red boxes indicate the positions of rs159961 and rs301792 within the motifs. Data source: https://www.encodeproject.org/^[81]. g–i Dual-luciferase reporter gene assays detected the enhancer activity of rs159961 and rs301792 in SH-SY5Y (g) U251 (h), and HEK-293T (i) cells (n = 10 for (g), n = 16 for (h, i), biological replicates). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. rs159961-C vs rs159961-T: P = 9.32 × 10⁻⁶, t = 6.09, df = 18, 95% CI [−0.53, −0.26]; rs301792-A vs rs301792-G: P = 9.03 × 10^-4, t = 3.97, df = 18, 95% CI [−0.70, −0.22] for (g). rs159961-C vs rs159961-T: P = 3.88 × 10⁻⁷, t = 6.46, df = 30, 95% CI [−1.14, −0.59]; rs301792-A vs rs301792-G: P = 0.96, t = 0.04, df = 30, 95% CI [−0.20, 0.21] for (h). rs159961-C vs rs159961-T: P = 0.36, t = 0.93, df = 30, 95% CI [−0.11, 0.04]; rs301792-A vs rs301792-G: P = 0.35, t = 0.94, df = 30, 95% CI [−0.12, 0.05] for (i). j, k EMSA showed that rs159961 and rs301792 exhibited allele-specific binding ability to nucleoprotein. Representative images are shown from n = 3 independent experiments with similar results. l, m Super-shift and competitive EMSA experiments. The binding specificity of rs159961 and rs301792 to nucleoprotein was verified by competitive experiments using unlabeled probes (500 fmol). Anti-REST and anti-POLR2A antibodies were used to detect the binding of rs159961 to REST and the binding of rs301792 to POLR2A. 50 fmol of labeled probes were used per lane. Red arrows indicate protein–probe complexes, blue arrows indicate free probes, and black arrows indicate super-shift bands. Nuclear proteins were extracted from SH-SY5Y cells. Representative images are shown from n = 3 independent experiments with similar results. Source data and uncropped blots are provided as a Source Data file.

Molecular regulatory mechanisms of rs159961 and rs301792

Our reporter gene assays revealed the neuronal-dependent regulatory effects of rs159961 and rs301792. To dissect the molecular mechanisms underlying these neuronal-dependent regulatory effects, we conducted EMSA. The EMSA showed that different alleles of rs159961 and rs301792 affect the binding affinity of TFs (Fig.). The super-shift assay indicated that rs159961 binds to REST, as a super-shift band was observed when REST antibody was added (Fig.). The risk allele (T allele) of rs159961 attenuated binding to REST compared with the protective allele (C allele) (Fig.). Different alleles of rs301792 also exhibited differential binding to POLR2A (band 1 in Fig.), as the binding intensity was decreased when POLR2A antibody was added (Fig.). The risk allele (G allele) of rs301792 demonstrated stronger binding ability to POLR2A compared with the A allele (Fig.). Taken together, these results indicate that functional genetic variants rs159961 and rs301792 regulate the enhancer activity by affecting the binding affinity of REST and POLR2A, respectively.

Risk alleles of rs159961 and rs301792 confer SCZ risk by upregulating RERE expression

To further identify the target gene (s) regulated by these two functional variants, we examined their associations with gene expression in the human brain. The largest brain eQTL dataset, BrainMeta (v2), which was generated using human brain tissues, was explored^[16]. Both rs159961 and rs301792 showed significant associations with the expression of RERE (P = 1.39 × 10⁻⁴⁸ and P = 1.98 × 10⁻⁴⁹), RERE-AS1 (P = 4.56 × 10⁻⁸ and P = 4.72 × 10⁻⁸), and SLC45A1 (P = 1.51 × 10⁻⁴ and P = 1.64 × 10⁻⁴) (Fig.). The strongest associations were observed for RERE, suggesting that these two functional genetic variants may confer SCZ risk by regulating RERE expression. In addition, eQTL analysis using postmortem adult brain tissues also showed that the risk alleles of rs159961 and rs301792 were associated with higher expression of RERE (Fig.), consistent with the results of reporter gene assays (Fig.). To test the colocalization posterior probability (PP) between the 1p36.23 locus and the eQTLs for RERE, we conducted colocalization analysis. The results showed that the colocalization PP of the 1p36.23 locus with the eQTLs for RERE was high (0.854), supporting that genetic variants in this locus confer SCZ risk by regulating RERE expression. However, we noticed that the colocalization PP for SNP rs159961 and rs301792 was relatively low, SNP.PP.H4 = 0.09 for rs301792, and SNP.PP.H4 = 0.013 for rs159961 (Supplementary Data). These consistent results suggest that risk alleles of these two functional variants contribute to SCZ risk by upregulating RERE expression.

[Fig. 2] rs159961 and rs301792 are associated with RERE expression in the human brain.a eQTL analysis of rs159961 and rs301792 with genes within 2 Mb, based on the BrainMeta v2 data. cis-eQTL associations were tested using additive linear regression with two-sided tests on the genotype coefficient. b, c The CMC brain eQTL data showed that both variants were significantly associated with RERE expression in the human brain. The middle line of the boxplot represents the median, the box indicates the first and third quartiles, and the whiskers indicate the 1.5 interquartile range (IQR). eQTLs were identified using additive linear regression, with significance assessed by two-sided tests on the regression coefficient and FDR correction. d, e The LIBD brain eQTL data showed that both variants were significantly associated with RERE expression in the human brain. Data source: https://eqtl.brainseq.org/phase1/eqtl/^[72]. Box plots show the median and IQR; whiskers correspond to 2.5* IQR. eQTL associations were tested using additive linear regression, and statistical significance was assessed using two-sided tests. f–h Using CRISPR/Cas9-mediated editing in SH-SY5Y cells, 330 bp and 267 bp fragments containing rs159961 and rs301792 were knocked out. Uncropped blots are provided in Source Data. i–j qPCR showed that knockout of rs159961 and rs301792 resulted in significant downregulation of RERE expression in SH-SY5Y cells (n = 3, biological replicates). k–n The knockdown efficiency of REST and POLR2A, as well as RERE expression, was assessed using qPCR in SH-SY5Y cells (n = 3, biological replicates). o The top 20 positive and negative TWAS-significant genes and RERE among 243 SCZ-associated genes are shown, with RERE highlighted in red. TWAS was performed using FUSION with gene expression weights from the PsychENCODE project^[18]. Significance was defined using Bonferroni correction for multiple testing across 12,434 genes (P < 4.02 × 10⁻⁶). p RERE expression was significantly upregulated in the brains of SCZ patients compared with controls. Violin plots show the distribution of the data, with the width of the shape representing the kernel density estimation. The central line indicates the median, and the box represents the IQR. Differential expression was assessed using PsychENCODE^[18] results with covariates adjusted; significance was defined as FDR (Benjamini–Hochberg) < 0.05. For i–n, unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. Control vs rs159961-KO: P = 1.74 × 10⁻³, t = 7.45, df = 4, 95% CI [0.16, 0.34] for (i). Control vs rs301792-KO: P = 2.72 × 10⁻⁴, t = 12.17, df = 4, 95% CI [0.39, 0.62] for (j). Control vs REST-shRNA#1: P = 1.19 × 10⁻², t = 4.38, df = 4, 95% CI [0.15, 0.68]; Control vs REST-shRNA#2: P = 2.73 × 10⁻², t = 3.40, df = 4, 95% CI [0.08, 0.79] for (k). Control vs REST-shRNA#1: P = 6.98 × 10⁻³, t = 5.10, df = 4, 95% CI [0.14, 0.48]; control vs REST-shRNA#2: P = 2.69 × 10⁻², t = 3.42, df = 4, 95% CI [0.03, 0.34] for (l). Control vs POLR2A-shRNA#1: P = 1.24 × 10⁻⁵, t = 26.29, df = 4, 95% CI [0.56, 0.70]; Control vs POLR2A-shRNA#2: P = 1.64 × 10⁻⁵, t = 24.53, df = 4, 95% CI [0.48, 0.60] for (m). Control vs POLR2A-shRNA#1: P = 1.38 × 10⁻², t = 4.19, df = 4, 95% CI [0.08, 0.38]; Control vs POLR2A-shRNA#2: P = 2.98 × 10⁻², t = 3.31, df = 4, 95% CI [0.04, 0.46] for (n). Source data are provided as a Source Data file.

To further investigate whether rs159961 and rs301792 regulate RERE expression, we performed CRISPR/Cas9-mediated gene editing. We knocked out ~300 bp genomic sequences encompassing the rs159961 or rs301792 sites (Fig.), followed by quantification of RERE expression. The real-time quantitative PCR (qPCR) results showed that deletion of these genomic regions led to downregulation of RERE (Fig.), indicating that the genomic elements containing rs159961 and rs301792 function as enhancers to regulate RERE expression. In addition to RERE, rs159961 and rs301792 were also associated with SLC45A1 expression (P = 1.51 × 10⁻⁴ and P = 1.64 × 10⁻⁴). However, these associations were much weaker than those with RERE (P = 1.39 × 10⁻⁴⁸ and P = 1.98 × 10⁻⁴⁹) (Fig.). We further examined whether knockout of rs159961 or rs301792 affected SLC45A1 expression. The results showed that the rs159961 knockout significantly downregulated SLC45A1, but the rs301792 knockout had no significant effect (Supplementary Fig.). Together, these results suggest that RERE (rather than SLC45A1) is the main target gene regulated by these two functional variants.

To investigate whether rs159961 and rs301792 regulate RERE expression independently or synergistically, we performed conditional eQTL analyses and haplotype analyses. Conditional eQTL modeling indicated that rs159961 and rs301792 exert non-independent effects on RERE expression (Supplementary Fig.). Haplotype analysis further revealed that only the AC/GT diplotype (heterozygous combinations of both risk alleles) was significantly associated with increased RERE expression, suggesting a potential synergistic regulatory effect (Supplementary Fig. and Supplementary Tables and). rs159961 and rs301792 affect the binding of REST and POLR2A (Fig.), implying that these two variants interact with REST and POLR2A to regulate RERE expression. We thus knocked down REST and POLR2A in SH-SY5Y cells using short hairpin RNAs (shRNAs). REST and POLR2A knockdown resulted in significant down-regulation of RERE (Fig.). Furthermore, Pearson’s correlation analysis revealed a significant positive correlation between REST and RERE expression in the human brain (Supplementary Fig.). Collectively, these results indicate that these two TFs interact with rs159961 and rs301792 to regulate RERE expression.

rs159961 and rs301792 are located in intronic regions of RERE, suggesting that they may regulate RERE expression through long-range interactions. Hence, we examined whether the genomic regions containing rs159961 and rs301792 interact with the promoter region of RERE, using Hi-C data from Yang et al.^[17]. Hi-C data showed long-range interactions between the genomic regions containing these two functional variants, and the RERE promoter in human hippocampus (Supplementary Fig.). Together, these results indicate that the risk alleles of rs159961 and rs301792 upregulate RERE expression by affecting the binding of REST and POLR2A.

TWAS supports RERE as a SCZ risk gene

The above data suggest that RERE is the likely causal gene regulated by rs159961 and rs301792. To further validate this, we performed a transcriptome-wide association analysis (TWAS) by integrating eQTL data from postmortem adult brain tissues (from the PsychENCODE project^[18]) and SCZ GWAS^[10]. This analysis identified 234 genes that showed significant associations with SCZ risk (Supplementary Data). Notably, the TWAS supported that RERE is a SCZ risk gene whose genetically predicted upregulation is associated with SCZ risk (Z-score = 6.24, P = 4.46 × 10⁻¹⁰) (Fig.). The TWAS result is consistent with the reporter gene assays and eQTL data, which showed that the risk alleles of rs159961 and rs301792 were associated with higher RERE expression. The TWAS results provide robust support for the involvement of RERE in SCZ.

RERE was significantly upregulated in the brains of SCZ cases compared with controls

TWAS and eQTL analysis suggested that functional risk variants confer SCZ risk by upregulating RERE. If this conjecture is true, RERE expression should be higher in SCZ cases than in controls. To further test whether RERE upregulation is associated with increased SCZ risk, we examined RERE expression in the brains of SCZ cases and controls. Consistent with eQTL and TWAS results, RERE expression was significantly upregulated in the brains of SCZ cases compared with controls (P = 8.45 × 10⁻⁹) (Fig.). The significant upregulation of RERE in SCZ cases compared with controls provided further evidence supporting that functional risk alleles confer SCZ risk by upregulating RERE expression.

RERE affects the proliferation and differentiation of mNSCs

Our consistent and convergent results suggest that risk alleles of rs159961 and rs301792 upregulate RERE expression, which in turn contributes to SCZ susceptibility. To explore the potential role of RERE in SCZ pathophysiology, we examined its function in neurodevelopment using the NSC model, which is widely used in delineating SCZ pathogenesis^[19]. We first examined RERE expression in major cell types in the human and mouse brain using RNA-seq data from Zhang et al.^[20],[21] and Yao et al.^[22]. The results showed that RERE was broadly expressed, with relatively higher expression levels in neurons and astrocytes (Supplementary Fig.). Further spatiotemporal expression showed that RERE was highly expressed in the prenatal human brain^[23] (Supplementary Fig.), implying that this gene may have a role in neurodevelopment. The proliferation and differentiation of NSCs are crucial for brain development, and previous studies have shown that SCZ risk genes regulate these processes^[24]–[29]. We obtained mNSCs from the subventricular zone (SVZ) of mouse brains at E14.5 and validated their identity using three well-characterized markers, SOX2, PAX6, and NESTIN (Fig.). To mimic the effect of RERE upregulation on neurodevelopment, we overexpressed RERE in mNSCs (Fig.). Proliferation assays showed that RERE overexpression inhibited the proliferation capacity of mNSCs (Fig.). Furthermore, cell-cycle analysis revealed a significant increase in the proportion of cells in the G2/M-phase in the RERE-overexpression group compared with the controls, indicating that RERE-overexpression induced cell-cycle arrest at the G2/M-phase (Fig. and Supplementary Fig.).

[Fig. 3] RERE-overexpression significantly inhibited the proliferation of mNSCs.a Three well-characterized markers, NESTIN (purple), PAX6 (cyan), and SOX2 (yellow), confirmed the identity of the isolated mNSCs through immunofluorescence (IF) staining. DAPI was used to stain the nuclei (blue). Representative images are shown from n = 3 independent experiments with similar results. b Validation of RERE-overexpression in mNSCs with reverse transcription-PCR (RT-PCR). PCR was performed for 26 cycles (lanes 1–2, and 7-8), 28 cycles (lanes 3–4, and 9-10), and 30 cycles (lanes 5–6, and 11-12). Uncropped blots are provided in Source Data. c, d IF staining and quantification of the EdU incorporation assay (n = 3, biological replicates). The EdU-positive cells undergoing DNA amplification are shown in red and DAPI stains nuclei (blue). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. Control vs RERE-OE: P = 0.36, t = 1.03, df = 4, 95% CI [−2.72, 5.94] for (c). e CCK-8 assay, in which CCK-8 was added at 0, 24, 48, and 72 h after cell plating, and absorbance at 450 nm was detected 2 h later (n = 10, biological replicates). Two-way ANOVA followed by Sidak’s multiple comparisons test; Data are presented as mean ± SD. 24 h: P = 7.60 × 10⁻³, t = 3.63, df = 17.97, 95% CI [0.01, 0.10]; 48 h: P = 2.24 × 10⁻³, t = 4.49, df = 13.47, 95% CI [0.06, 0.28]. f–h Flow cytometry analysis of the cell cycle in mNSCs showed that cells were arrested in the G2/M phase upon RERE-overexpression (n = 3, biological replicates). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. Control-S vs RERE-OE-S: P = 2.14 × 10⁻², t = 3.67, df = 4, 95% CI [0.91, 6.57]; Control-G2/M vs RERE-OE-G2/M: P = 3.09 × 10⁻², t = 3.27, df = 4, 95% CI [−5.0, −0.40] for (h). Source data are provided as a Source Data file.

We also investigated the effect of RERE-overexpression on the differentiation of mNSCs. We used markers for astrocytes (GFAP), mature neurons (MAP2), oligodendrocytes (CLDN11), and immature neurons (TUBB3). After three days of spontaneous differentiation, the protein levels of GFAP and MAP2 were measured using IF (Fig.), and the expression levels of Gfap, Map2, Cldn11, and Tubb3 were assessed using qPCR (Fig.). The proportion of GFAP-positive cells did not differ significantly between the RERE-overexpression and control groups (Fig.), indicating that overexpression of RERE did not affect astrocytic differentiation of mNSCs. However, the proportion of MAP2-positive cells was significantly reduced in the RERE-overexpression group compared with controls (Fig.), indicating that RERE-overexpression inhibited neurogenesis. Consistent with IF results, qPCR analysis also revealed that Map2 and Tubb3 expression were significantly downregulated in the RERE-overexpression group compared with controls (Fig.). By contrast, the expression of Gfap and Cldn11 was unchanged (Fig.). Western blot results further confirmed that MAP2 protein levels were significantly reduced, whereas GFAP levels did not differ between the RERE-overexpression and control groups (Fig.). Taken together, these results indicate that RERE-overexpression impairs neural development by inhibiting the proliferation and differentiation of mNSCs.

[Fig. 4] RERE-overexpression significantly affected the differentiation of mNSCs.a, b IF staining of astrocytes and neurons for 3 days after spontaneous differentiation of mNSCs. GFAP-positive cells are shown in green, MAP2-positive cells in red, and nuclei were stained with DAPI (blue). c, d Quantification of the ratios of GFAP- and MAP2-positive cells (n = 3, biological replicates). e–h qPCR analysis of Gfap, Cldn11, Map2, and Tubb3 expression (n = 3, biological replicates). i–l Western blot analysis and quantification of GFAP and MAP2 protein expression. Uncropped blots are provided in Source Data. (n = 3, biological replicates). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. Control vs RERE-OE: P = 0.46, t = 0.81, df = 4, 95% CI [−4.08, 2.24] for (c). P = 3.12 × 10⁻², t = 3.26, df = 4, 95% CI [0.78, 9.76] for (d). P = 0.53, t = 0.69, df = 4, 95% CI [−0.23, 0.14] for (e). P = 5.10 × 10⁻³, t = 5.57, df = 4, 95% CI [0.13, 0.38] for (f). P = 0.46, t = 0.81, df = 4, 95% CI [−0.10, 0.19] for (g). P = 1.29 × 10⁻², t = 4.27, df = 4, 95% CI [0.11, 0.53] for (h). P = 0.85, t = 0.20, df = 4, 95% CI [−0.28, 0.33] for (j). P = 2.60 × 10⁻², t = 3.45, df = 4, 95% CI [0.04, 0.33] for (l). Source data are provided as a Source Data file.

RERE regulates pathways related to dendrite development and glutamatergic synapse

It has been reported that RERE acts as a nuclear receptor modulator to regulate the expression of specific genes^[30],[31]. Moreover, RERE also regulates gene expression by acting as a coregulator of the retinoic acid (RA) signaling pathway^[32]. To elucidate the biological processes and signaling pathways regulated by RERE, we conducted transcriptome sequencing using RERE-overexpressing and control mNSCs. A total of 213 genes showed differential expression (|log₂ (Fold Change)| > 1 and P-adj < 0.05) in RERE-overexpressing and control mNSCs (Fig. and Supplementary Data). Among them, 118 genes were downregulated, and 95 were upregulated (Fig.). Several differentially expressed genes are related to neurodevelopment. For instance, Tnn is associated with neuronal development^[33], Sema5a is involved in neurodevelopment and axonal projection^[34], and has been linked to autism risk^[35], and Erbb4, an important SCZ risk gene, is involved in central nervous system development and synaptic transmission^[36],[37]. We validated the RNA-seq results with qPCR. The qPCR results of four selected genes (Erbb4, Sema5a, Fbln2, and Mest were selected as they exhibited highly significant P-values, i.e., ranked among the top 20%) and were consistent with RNA-seq (Supplementary Fig.), indicating the reliability of the transcriptomic analysis. Intriguingly, KEGG analysis revealed that the differentially expressed genes were predominantly enriched in glutamatergic, cholinergic, and GABAergic synapses (Fig.), three important neurotransmitter systems that have been reported to be involved in SCZ^[38]–[40]. GO analysis showed that the differentially expressed genes were enriched in pathways related to endomembrane system organization, vesicle organization, dendrite development, and related processes (Fig.). These results suggest that RERE plays an important role in regulating pathways related to dendrite development and neurotransmitter systems.

[Fig. 5] RERE affects signaling pathways involved in neurodevelopment.Differentially expressed genes identified by comparing RNA-seq data from RERE-overexpressing and control mNSCs. a Heatmap of the top 20 differentially expressed genes (DEGs, |log₂ (Fold Change)|>1 and P-adj < 0.05). b Volcano plot of 213 DEGs (95 upregulated and 118 downregulated; |log₂ (Fold Change)| > 1 and P-adj < 0.05). Red dots represent upregulated genes, blue dots represent downregulated genes, and gray dots represent non-differentially expressed genes. The top significantly differentially expressed genes were marked. c KEGG pathway analysis of DEGs (P-adj < 0.05). Pathways related to neurodevelopment are highlighted in red. d GO biological process analysis of DEGs (P-adj < 0.01). Pathways related to neurodevelopment are highlighted in red.

Overexpression of RERE affects the density of dendritic spines and dendritic complexity

Dendritic spine alterations have been widely reported in SCZ^[41],[42]. Accumulating evidence supports the dendritic spine pathology of SCZ. For example, one study showed that excessive pruning of dendritic spines in late childhood or adolescence may contribute to SCZ risk^[41]. Our previous studies also showed that dysregulation of the SCZ risk genes NAGA and NMRAL1 led to altered dendritic spine density^[24],[27]. These findings suggest that risk genes may contribute to SCZ risk by affecting dendritic spines. Consistent with these observations, our RNA-seq analysis also suggested that RERE may have a pivotal role in dendritic spine morphogenesis. Differentially expressed genes were significantly enriched in pathways related to dendrite development, GABAergic, glutamatergic, and cholinergic synapses (Fig.). We thus investigated whether RERE has a role in regulating dendritic complexity and dendritic spine density. Dendritic complexity analysis showed that RERE-overexpression resulted in a significant increase in the number of neurites in rat cortical neurons (Fig.). Dendritic spine density analysis further revealed that the densities of thin spines and the total spines were significantly reduced in RERE-overexpressing neurons compared with controls (Fig.). In addition, compared with controls, the proportion of stubby spines was significantly increased in RERE-overexpressing neurons, while the proportions of other spine types (including long thin, filopodia, and branched spines) were significantly reduced (Fig.). These results demonstrate the important role of RERE in regulating and maintaining dendritic spine density and complexity.

[Fig. 6] Overexpression of RERE affects neuronal development and dendritic spine formation.a Representative IF staining image showing neuronal morphology in RERE-overexpression neurons and control neurons. b qPCR analysis of RERE-overexpression efficiency in rat neurons (n = 3, biological replicates). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. Control vs RERE-OE: P = 6.49 × 10⁻⁵, t = 17.34, df = 4, 95% CI [−8.07, −5.84]. c Quantification of neuronal complexity following RERE-overexpression (n = 41 for control, n = 39 for RERE-OE, biological replicates). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SEM. Control vs RERE-OE, 20 µm: P = 1.90 × 10⁻⁴, t = 3.92, df = 78, 95% CI [−2.95, −0.96]; 30 µm: P = 3.00 × 10⁻³, t = 3.06, df = 78, 95% CI [−2.53, −0.54]; 40 µm: P = 3.38 × 10⁻⁴, t = 3.75, df = 78, 95% CI [−2.78, −0.85]; 50 µm: P = 5.44 × 10⁻⁴, t = 3.61, df = 78, 95% CI [−2.50, −0.72]; 60 µm: P = 1.94 × 10⁻³, t = 3.21, df = 77, 95% CI [−2.10, −0.49]; 70 µm: P = 2.34 × 10⁻², t = 2.31, df = 76, 95% CI [−1.48, −0.11]. d Representative IF staining images showing dendritic spine density and morphology in RERE-overexpression neurons and control neurons. e, f Quantification of dendritic spine density and the proportion of spine types in rat neurons after RERE-overexpression (n = 20 for control, n = 20 for RERE-OE, biological replicates). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. Control vs RERE-OE, Thin: P = 1.52 × 10⁻², t = 2.54, df = 38, 95% CI [0.09, 0.83]; Total: P = 4.71 × 10⁻², t = 2.05, df = 38, 95% CI [0.007, 1.003] for (e); Control vs RERE-OE, Stubby: P = 7.88 × 10⁻³, t = 2.81, df = 38, 95% CI [−7.40, −1.20]; Total: P = 4.34 × 10^⁻8, t = 6.82, df = 38, 95% CI [3.44, 6.35] for (f); Source data are provided as a Source Data file.

RERE regulates GluN2A

Transcriptome analysis revealed that the differentially expressed genes between RERE-overexpressing and control mNSCs were significantly enriched in glutamatergic and GABAergic synapses (Fig.). Additionally, dysfunction of glutamatergic receptors has been widely reported in SCZ^[43],[44]. To further explore whether RERE contributes to SCZ by affecting the glutamatergic system, we conducted western blots and qPCR analyses (Fig. and Supplementary Fig.). The expression of NMDA receptors (NMDARs) and AMPA receptors (AMPARs) was quantified. The results revealed a significant decrease in the expression of NMDAR subunits (Grin1, Grin2a, and Grin2b) and AMPAR subunits (Gria2) (Fig. and Supplementary Fig.). We further examined the expression levels of GABAergic subunits (Gaba-br2 and Gabrg2), the dopaminergic receptor Drd2, and Synaptophysin. The results showed that the RERE-overexpression significantly upregulated Gaba-br2 and Gabrg2, while Synaptophysin expression was significantly downregulated. In contrast, Drd2 expression remained unchanged (Supplementary Fig.). Western blot further revealed that GluN2A (encoded by Grin2a) was significantly reduced (Fig.). GluN2B expression also showed a trend toward a decrease; however, this change did not reach statistical significance (Fig.). Moreover, the levels of GluA4, GluK1, and PSD95 did not show significant variations, but synaptophysin expression was increased (Fig.). Taken together, these results indicate that RERE regulates GluN2A, an important subunit of NMDARs, suggesting that RERE may confer SCZ risk by modulating the glutamatergic system.

[Fig. 7] RERE modulates NMDAR-mediated miniature excitatory postsynaptic currents (mEPSCs).a–g Western blot analysis and quantification of GluN2A, GluN2B, GluA4, GluK1, PSD95, and synaptophysin expression. Uncropped blots are provided in Source Data. h, i qPCR analysis of Grin2a and Grin2b expression. Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD; (n = 3, biological replicates) (b–i). Control vs RERE-OE: P = 3.74 × 10⁻³, t = 6.06, df = 4, 95% CI [0.09, 0.23] for (b). P = 0.06, t = 2.55, df = 4, 95% CI [−0.01, 0.13] for (c). P = 0.06, t = 2.64, df = 4, 95% CI [−0.24, 0.01] for (d). P = 0.79, t = 0.29, df = 4, 95% CI [−0.36, 0.30] for (e). P = 0.09, t = 2.24, df = 4, 95% CI [−0.17, 0.02] for (f). P = 0.01, t = 4.60, df = 4, 95% CI [−0.14, −0.04] for (g). P = 2.27 × 10⁻², t = 3.60, df = 4, 95% CI [0.05, 0.37] for (h). P = 2.68 × 10⁻³, t = 6.63, df = 4, 95% CI [0.13, 0.32] for (i). j Representative traces of NMDAR-mEPSCs recorded in the presence of 1 µM TTX, 10 mM bicuculline, and 10 µM CNQX. k Summary graphs of NMDAR-mEPSC peak amplitude (Control: mean ± SEM, 9.655 ± 0.3163, n = 57 cells/3 cultures; RERE-OE: mean ± SEM, 8.071 ± 0.2161, n = 58 cells/3 cultures, two-tailed Mann–Whitney U-test. U = 988, P = 1.61 × 10⁻⁴, Hodges–Lehmann median difference = −1.387, 95% CI [−2.089, −0.655]). l Summary graphs of NMDAR-mEPSC frequency (Control: mean ± SEM, 0.8482 ± 0.05262, n = 57 cells/3 cultures; RERE-OE: mean ± SEM, 0.8778 ± 0.05055, n = 58 cells/3 cultures, two-tailed Mann–Whitney U-test. U = 1537, P = 0.52, Hodges–Lehmann median difference = 0.044, 95% CI [−0.087, 0.177]). m–o Cumulative probability plots of NMDAR-mEPSC peak amplitude. Thick yellow (Control) and blue (RERE-OE) lines represent the pooled distribution of all events in each group, while thin gray lines represent the individual traces. Source data are provided as a Source Data file.

RERE modulates NMDAR-mediated miniature excitatory postsynaptic currents (mEPSCs)

Given the significant downregulation of GluN2A observed upon RERE-overexpression, we performed whole-cell recordings to examine NMDAR-mediated mEPSCs in primary rat frontal cortical neurons (Fig.). mEPSCs offer a refined approach to study synaptic mechanisms, as each event reflects the quantal release of neurotransmitters and activation of receptors at individual synapses. At a holding potential of -70 mV, mEPSCs are mainly mediated by AMPARs because NMDARs are blocked by Mg²⁺. To relieve the voltage-dependent block of NMDARs, we used an Mg²⁺-free extracellular solution. We recorded NMDAR-mediated mEPSCs from rat frontal cortex neurons at -70 mV, in the presence of tetrodotoxin (TTX, blocking action potentials), bicuculline (blocking inhibitory inputs), and 6-cyano-7-nitroquinoxaline-2, 3-dione (CNQX, blocking AMPA/kainate currents). Quantitative analysis revealed that RERE-overexpression reduced peak amplitude of NMDAR-mEPSCs by 16% (to ~84% of control values; Fig.), suggesting impaired synaptic efficacy and reduced functional receptor density. Conversely, no significant alteration was observed in event frequency between groups, indicating that presynaptic glutamate release remains unaffected (Control: 0.8482 ± 0.05262 Hz vs RERE-OE: 0.8778 ± 0.05055 Hz; Fig.). Cumulative probability analysis demonstrated a pronounced leftward shift in amplitude distribution for RERE-overexpressing neurons (Fig.), indicating an increased proportion of small-amplitude synaptic events. Collectively, these electrophysiological findings demonstrate that RERE-overexpression impairs NMDAR-dependent synaptic transmission, which correlates mechanistically with the observed GluN2A subunit depletion.

RERE-RXRA/RARB complex regulates Grin2a via promoter binding

To investigate how RERE-overexpression in neurons leads to the downregulation of Grin2a, we first performed TF binding predictions to explore if the promoter region (~2000 bp upstream of the transcription start site, TSS) of Grin2a contains binding sites for RERE-related TFs. This analysis revealed the presence of motifs for RARB and RXRA (Fig. and Supplementary Fig.). Within this region, we identified two fragments, designated F1 and F2, both of which contained putative binding motifs for RARB and RXRA (Supplementary Fig.). RERE has been reported to function as a nuclear receptor modulator that interacts with the RAR/RXR heterodimer to regulate gene expression through retinoic acid response elements (RAREs)^[32]. We therefore hypothesized that RERE represses Grin2a expression via this pathway. To test this, we first performed a dual-luciferase reporter assay and found that both F1 and F2 were transcriptionally active (Fig.). However, their regulatory activity was attenuated upon RERE-overexpression (Fig. and Supplementary Fig.), suggesting that RERE acts as a repressor. Co-immunoprecipitation (Co-IP) experiments further demonstrated that RERE interacted with RXRA and RARB in rat neurons (Fig.). To confirm the direct binding of RARB and RXRA to these promoter fragments, we performed Chromatin immunoprecipitation (ChIP)-qPCR, which verified their enrichment at fragments F1 and F2 (Fig.). Finally, knocking down Rxra or Rarb in rat neurons resulted in a significant downregulation of Grin2a expression (Fig.). Together, these findings indicate that RERE interacts with RARB and RXRA at the promoter region of Grin2a to repress Grin2a expression (Fig.).

[Fig. 8] RERE represses Grin2a expression through interacting with RARB/RXRA.a, b PWMs of TFs RARB and RXRA. Data were obtained from JASPAR^[82]. c, d Dual-luciferase reporter gene assays validated the transcriptional activity of F1 and F2 in B104 cells (n = 6, biological replicates). The genomic sequences containing RARB and RXRA binding motifs exhibited enhancer activity compared with the control vector in B104 cells. Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. Control vs Grin2a-F1: P = 6.27 × 10⁻¹⁴, t = 57.39, df = 10, 95% CI [−2.51, −2.32] for; Control vs Grin2a-F2: P = 1.01 × 10⁻¹³, t = 54.69, df = 10, 95% CI [−1.92, −1.77] (c). RERE-overexpression suppresses the transcription activity of the genomic sequences containing RARB and RXRA binding motifs. Two-way ANOVA followed by Sidak’s multiple comparisons test; Data are presented as mean ± SD. Control: P = 7.42 × 10⁻⁶, t = 10.48, df = 8.98, 95% CI [0.68, 1.21]; Grin2a-F1: P = 1.38 × 10⁻⁴, t = 8.28, df = 7.60, 95% CI [1.23, 2.66]; Grin2a-F2: P = 1.16 × 10⁻⁵, t = 9.29, df = 9.71, 95% CI [1.57, 2.97] (d). e Co-IP analysis showed that RERE interacts with RARB and RXRA in rat neurons. Uncropped blots are provided in Source Data. f–g ChIP-qPCR analysis confirmed enrichment of RARB and RXRA at the F1 and F2 fragments of the Grin2a promoter in rat neurons (n = 3, biological replicates). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. IgG vs RARB: P = 7.55 × 10⁻⁴, t = 9.26, df = 4, 95% CI [−1.22, −0.66]; IgG vs RXRA: P = 9.59 × 10⁻⁵, t = 15.71, df = 4, 95% CI [−1.36, −0.95] for (f). IgG vs RARB: P = 1.44 × 10⁻⁴, t = 14.16, df = 4, 95% CI [−0.59, −0.40]; IgG vs RXRA: P = 1.19 × 10⁻³, t = 8.23, df = 4, 95% CI [−0.97, −0.48] for (g). h–k The knockdown efficiency of Rxra and Rarb, as well as the expression of Grin2a, was assessed using qPCR in rat neurons (n = 3, biological replicates). Unpaired two-tailed Student’s t-test; Data are presented as mean ± SD. Control vs Rxra-shRNA#1: P = 1.64 × 10⁻⁵, t = 24.51, df = 4, 95% CI [0.69, 0.86]; Control vs Rxra-shRNA#2: P = 9.73 × 10⁻⁶, t = 27.96, df = 4, 95% CI [0.76, 0.93] for (h). Control vs Rxra-shRNA#1: P = 3.04 × 10⁻³, t = 6.41, df = 4, 95% CI [0.18, 0.45]; Control vs Rxra-shRNA#2: P = 1.17 × 10⁻², t = 4.40, df = 4, 95% CI [0.08, 0.35] for (i). Control vs Rarb-shRNA#1: P = 1.57 × 10⁻⁴, t = 13.86, df = 4, 95% CI [0.38, 0.57]; Control vs Rarb-shRNA#2: P = 2.50 × 10⁻⁴, t = 12.31, df = 4, 95% CI [0.28, 0.45] for (j). Control vs Rarb-shRNA#1: P = 2.44 × 10⁻², t = 3.53, df = 4, 95% CI [0.03, 0.24]; Control vs Rarb-shRNA#2: P = 1.57 × 10⁻³, t = 7.65, df = 4, 95% CI [0.20, 0.43] for (k). l Schematic model illustrating that RERE cooperates with RARB and RXRA to repress Grin2a transcription. Source data are provided as a Source Data file.

Discussion

SCZ is a complex polygenic disorder. The neurodevelopmental hypothesis of SCZ posits that SCZ is caused by a combination of environmental and genetic factors, which impact normal brain development^[45]. Genetic factors have a dominant role in the pathophysiology of SCZ^[3]. Over the past two decades, genetic studies of SCZ have advanced rapidly. To date, GWASs have identified over 300 risk loci for SCZ^[6]–[11]. Although these findings provide a pivotal starting point for delineating the pathogenesis of SCZ, major obstacles still exist. Identifying potential causal variants and genes from the reported risk loci is the first step in translating GWAS findings into disease biology. However, this step is hampered by complex LD between risk variants and the complicated regulatory mechanisms of risk variants. In this study, we systematically delineate the molecular mechanisms of the 1p36.23 risk locus in SCZ.

Accumulating evidence indicates that 1p36.23 is an important risk locus for SCZ. In 2014, the Psychiatric Genomics Consortium (PGC) identified 108 risk loci, including the locus at 1p36.23^[6]. The association between the 1p36.23 locus and SCZ was replicated in subsequent studies by Li et al.^[7] and Pardinas et al.^[8], respectively. In 2022, an expanded PGC GWAS (PGC-SCZ3) again confirmed the important role of the 1p36.23 locus in SCZ^[10]. Furthermore, fine-mapping combined with summary-data-based Mendelian randomization (SMR) analysis nominated RERE at 1p36.23 as one of 120 candidate genes^[10]. Consistent with these reports, our study confirmed the association of rs159961 and rs301792 with SCZ in an East Asian population, with the same risk alleles^[11]. These consistent genetic findings strongly suggest that 1p36.23 is a pivotal risk locus for SCZ.

We first characterized the genetic regulatory mechanisms of the 1p36.23 risk locus. We identified two functional variants (rs159961 and rs301792) and demonstrated their molecular regulatory mechanisms. The risk and protective alleles of these variants exhibited differential binding affinities to REST and POLR2A, leading to differential enhancer activity. The risk allele rs159961-T attenuated the binding of the transcriptional repressor REST, whereas the risk allele rs301792-G enhanced the binding of POLR2A. Consistent with these effects, both rs159961-T and rs301792-G exhibit increased enhancer activity. Reporter gene assays demonstrated that the two variants exhibit cell–type–specific enhancer activity in SY5Y cells (Fig.). Analysis of single-cell eQTL data from Fujita et al.^[46] and Bryois et al.^[47] further showed significant associations between these two variants and RERE expression in astrocytes (Supplementary Table), supporting cell-type-specific regulatory effect of rs159961 and rs301792. These results indicate that the two variants are functional and their regulatory effects are cell-type-specific. Interestingly, both variants are located in intronic regions of RERE and show robust associations with RERE expression in the human brain, suggesting that RERE is the target gene through which these variants influence SCZ risk. Consistent with this, knockout of the genomic elements containing these variants provided further evidence for their role in regulating RERE. These consistent results indicate that rs159961 and rs301792 contribute to SCZ risk by regulating RERE expression. In line with these findings, TWAS also supports RERE as an SCZ risk gene, with genetically predicted upregulation associated with increased SCZ risk. Moreover, our lines of evidence not only demonstrate the regulatory mechanisms of rs159961 and rs301792 but also prioritize RERE as the likely causal gene at 1p36.23 risk locus. Of note, rs159961 is in high LD with rs301792 (r² > 0.8), indicating the presence of two highly linked functional variants at the 1p36.23 risk locus. Indeed, a previous study has shown that a proportion of risk loci identified by GWAS contain more than one causal variant^[48].

In addition to elucidating the molecular regulatory mechanisms of the 1p36.23 risk locus, we also investigated the role of RERE in the pathophysiology of SCZ. Both reporter gene assays and eQTL data suggested that RERE upregulation is associated with increased risk of SCZ. We therefore mimicked RERE elevation by overexpression assays. We found that RERE-overexpression led to abnormal neurodevelopment, including impaired proliferation of mNSCs and neurogenesis. Furthermore, RERE-overexpression affected dendritic complexity and dendritic spine density, indicating its pivotal role in dendrite morphogenesis. Our results indicate that RERE may confer SCZ risk by affecting neurodevelopment and dendritic spine density. Though the pathophysiology of SCZ remains unclear, multiple lines of evidence support the neurodevelopmental hypothesis of SCZ^[45]. To explore if rs159961 and rs301792 are associated with RERE expression in prenatal brains, we examined prenatal brain datasets^[49]. Neither rs159961 nor rs301792 showed significant eQTL effects on RERE expression in prenatal stages (rs159961: beta = 0.044, P = 0.596; rs301792: beta = 0.134, P = 0.117). Owing to the very low cis-heritability of RERE in fetal brain samples (h² = −0.002, P = 0.498), we could not perform a meaningful TWAS in prenatal stages. Similarly, Hall et al.^[50] did not report RERE in their fetal TWAS, likely reflecting the limited sample sizes and low heritability of currently available prenatal resources. While these constraints preclude firm conclusions from fetal datasets, our serial cellular experiments, including proliferation and differentiation of mNSCs, dendritic morphogenesis, and transcriptomic analysis, consistently support RERE as a gene involved in neurodevelopmental processes relevant to SCZ.

RERE encodes an arginine-glutamic acid dipeptide (RE) repeat protein, which functions as a nuclear receptor coregulator and plays a crucial role in neurodevelopmental processes^{[30]–[32],[51]}. Rere influences the target gene Hes5 in the Notch signaling pathway, thereby affecting the progenitor proliferation and neuronal differentiation^[52]. RERE interacts with the RAR/RXR heterodimer of the RA receptor and binds to the RAREs to regulate the expression of downstream genes^[32]. The RA signaling pathway has been implicated in SCZ^[53]. Patients with RERE deficiency typically exhibit ventriculomegaly, aberrant cerebellar development, diminished white matter volume, and thinning of the corpus callosum, among other associated abnormalities^[54]. The majority of these individuals also exhibit clinical signs of mental illnesses like autism spectrum disorder, intellectual disability, and developmental delay^[54]. Consistent with human phenotype, Rere-deficient mice exhibit reduced hippocampal neurons, cerebellar maldevelopment, decreased branches of Purkinje cells within the cerebellum, and decreased brain size and weight^[55],[56]. Notably, most previous studies have investigated the function of RERE using loss-of-function strategies, which do not model the upregulation of RERE observed in SCZ cases. However, previous studies have provided evidence associating RERE-overexpression with an increased susceptibility to SCZ^[18],[57]. Therefore, we used an overexpression strategy in this study to explore the role of RERE in SCZ. Our results indicate that RERE may confer SCZ risk by disrupting neurodevelopment.

An intriguing finding of our study is that RERE regulates the expression of the NMDAR subunit GRIN2A. Both common and rare variants in GRIN2A have been significantly associated with SCZ, underscoring the pivotal role of GRIN2A in SCZ^[10],[39]. The Schizophrenia Exome Sequencing Meta-Analysis (SCHEMA)^[39] identified GRIN2A as one of ten high-confidence risk genes for SCZ in a large-scale exome sequencing study. In mice, reduced NMDAR expression causes SCZ-like behavior that can be ameliorated with antipsychotics such as haloperidol or clozapine^[58]. Research shows that individuals with SCZ exhibit specific changes in gene expression related to NMDAR subunits in the prefrontal cortex, including down-regulation of NR2A (GRIN2A) expression^[59]. NMDARs are crucial for synaptic transmission and plasticity^[60]. Furthermore, phosphorylation of the NR1 subunit serine(S)897 of NMDAR is notably reduced in patients with SCZ, impacting synaptic transmission and plasticity^[61]. Accumulating evidence indicates that dysfunction of the glutamatergic system plays an important role in SCZ. Our study shows that RERE contributes to SCZ risk through regulating the expression and function of GluN2A.

There are several limitations of this study. First, our fine-mapping and colocalization analyses did not provide strong causal evidence for rs301792 and rs159961. Nevertheless, it should be noted that fine-mapping primarily relies on statistical inference; hence, fine-mapping strategies inherently remain correlative in nature. In consequence, it remains unclear which of the fine-mapped variants are indeed functional and which are merely statistical bystanders. In fact, recent massively parallel reporter assays (MPRA) studies^[62],[63] showed that many variants inferred by fine-mapping were not supported by functional experimental validation. MPRA findings showed that posterior inclusion probabilities (PIPs) showed no significant difference between functional and non-functional variants, and statistical fine-mapping alone cannot effectively discriminate MPRA-positive (i.e., functional) and MPRA-negative (i.e., non-functional) variants. These findings suggest that statistical fine-mapping may help to identify potential causal variants for loci with small credible sets and simple LD patterns, but may be less effective for loci characterized by complex LD structures and larger credible sets^[62],[63]. Thus, variants with low statistical fine-mapping posterior inclusion probabilities are not necessarily non-causal functional variants. These observations are consistent with our comprehensive functional analyses and highlight the limitations of relying solely on statistical fine-mapping or colocalization analyses to identify causal variants. These data also underscore the importance of integrating functional genomics and experimental validation to elucidate the biological mechanisms underlying genetic associations. However, we acknowledge that other functional variants within this region may contribute to SCZ risk and warrant further investigation. Second, while we have provided lines of evidence supporting that rs159961 and rs301792 confer SCZ risk by modulating RERE expression, our genome editing assays involved deleting about 300 bp regions containing rs159961 or rs301792. Single-base deletion or editing is needed to confirm whether RERE is the only target of these variants. Third, our conditional eQTL analyses suggest that rs159961 and rs301792 do not act independently in regulating RERE expression. Haplotype analysis indicates that only the diplotype AC/GT at rs159961 and rs301792 is associated with increased RERE expression, implying a potential synergistic effect. Further experimental validation is needed to investigate whether rs159961 and rs301792 act independently or synergistically to regulate RERE. Fourth, neuronal physiological analyses in this study were performed in rat primary neurons rather than mouse NPC-derived neurons. RERE is conserved across human and rodent species, and rat primary neurons reliably achieve functional maturation in vitro^[25],[26], allowing reproducible assessment of neuronal morphology and synaptic function following RERE overexpression. Although mouse NPC-derived neurons are valuable for modeling early neurodevelopment, their increased heterogeneity and incomplete maturation can complicate physiological analyses^[64],[65]. Several limitations should be considered: species-specific differences between rat and mouse may influence phenotypic outcomes; mature primary neurons may preferentially reflect later-stage neuronal functions rather than early developmental processes; overexpression relies on exogenous constructs with potential variability in transcript usage and expression efficiency; and precise control of gene dosage remains challenging, with supraphysiological expression potentially deviating from endogenous regulation. Despite these limitations, this validated primary neuron model provides a controlled system to investigate the impact of RERE on neuronal structure and synaptic function. Finally, while we used in vitro models (neural stem cells and primary neurons), future studies employing genetically modified animal models are necessary to fully understand the role of RERE in SCZ pathophysiology in vivo.

In summary, our study provides consistent and convergent lines of evidence that rs159961 and rs301792 confer SCZ risk by regulating RERE, a gene that plays an important role in neurodevelopment and dendrite morphogenesis (Fig.). Our study not only elucidates the biological mechanisms of the 1p36.23 risk locus in SCZ but also provides a framework for translating GWAS findings into disease biology.

[Fig. 9] The working model of rs159961 and rs301792 in SCZ pathogenesis.The genetic variants rs159961 and rs301792 are located within the 1p36.23 region, which is associated with the susceptibility of SCZ. These variants are situated within the introns of the RERE gene. This genomic region exhibits an open chromatin state and displays enhancer activity marked by H3K27ac, H3K9ac, and H3K4me1 histone modifications. It also contains a DNA motif that influences the binding of TFs REST and POLR2A. The risk alleles of rs159961 and rs301792 (rs159961-T and rs301792-G) upregulate RERE expression by modulating the interaction between these transcription factors and the DNA motif. The overexpression of RERE inhibits the proliferation and differentiation of mouse neural stem cells (mNSCs) into neurons. Furthermore, RERE-overexpression in rat neurons increases dendritic complexity, reduces total spine and thin spine density, and decreases the expression of Grin2a, which encodes a subunit of NMDAR. Additionally, RERE-overexpression affects NMDAR-mediated mEPSCs in primary rat prefrontal cortical neurons.

Methods

Functional genomics analysis

Detailed information on the functional genomics analysis has been described in our previous study^[12]. Briefly, we downloaded 34 ChIP-Seq datasets generated from brain tissues or neuronal cells from ENCODE. The peaks of the ChIP-Seq datasets were called, and binding motifs (i.e., position weight matrix, PWM) of each TF were derived. We then integrated binding motifs derived from ChIP-Seq data and risk variants from SCZ GWAS to identify functional variants. If a risk variant is located in a TF binding motif and its different alleles affect the TF’s binding affinity, then this SNP is prioritized as a functional variant.

Statistical fine-mapping

To identify putative causal variants within the 1p36.23 locus, we conducted fine-mapping using three methods: SuSiE^[13], CARMA^[14], and FINEMAP^[15]. Summary statistics from populations of European ancestry (from the largest SCZ GWAS^[11]) were used as input. Genotype data of European individuals (i.e., reference panel) were extracted from the 1000 Genomes project^[66], and the LD matrix was calculated using PLINK (v.1.9)^[67]. For SuSiE, we performed fine-mapping using the function susie_rss()^[13] implemented in the R package susieR (v.0.14.2). For CARMA, we calculated the Z score and conducted fine-mapping using the R package CARMA (v.1.0)^[14]. For the FINEMAP method, we utilized FINEMAP software (v1.4.2) to perform fine-mapping by using the Z file, LD file, and master file^[15].

Colocalization analysis

To explore whether the same variants drive the expression of RERE and SCZ risk at the 1p36.23 locus, we conducted a colocalization analysis. The eQTL dataset was obtained from BrainMeta, which was constructed by using RNA-seq data of 2865 brain cortex samples^[16]. The SCZ GWAS dataset was obtained from Dang et al.^[11]. We extracted the cis-eQTL of RERE in the 1p36.23 locus and the GWAS summary statistics of SCZ, and performed Bayesian colocalization analysis utilizing the R package coloc (v.5.2.3)^[68] with default parameters.

Investigation of the regulatory effects of rs159961 and rs301792 on RERE

To investigate whether rs159961 and rs301792 regulate RERE independently or synergistically, we performed both conditional eQTL and haplotype analyses. For conditional eQTL analyses, we applied the linear regression model implemented in R (lm() function) using the genotype and expression data from CMC^[69], in which RERE expression was regressed on each SNP's allele either individually or jointly, thereby evaluating whether the effect of one SNP remained significant after conditioning on the other. We then phased the genotypes of rs159961 and rs301792 using Eagle2^[70] to infer haplotypes and construct diplotypes for each individual, which resulted in a total of seven distinct diplotypes. To assess the impact of these diplotypes on RERE expression, we conducted linear regression analyses with diplotype groups included as categorical predictors. Setting individuals carrying the non-risk alleles at both SNPs on both chromatids (AC/AC) as the reference group, and subsequently estimating the association of RERE expression for each diplotype relative to this reference.

Transcriptome-wide association study

To translate the findings of SCZ GWAS into potential causal genes, we performed a TWAS to identify genes whose genetically regulated expression levels were associated with SCZ risk. Specifically, we integrated gene expression weights from the PsychENCODE project^[18] with SCZ GWAS^[10] summary statistics using FUSION software^[71] (http://gusevlab.org/projects/fusion/) under default settings. The expression weights in the PsychENCODE dataset (N = 1321) were calculated from genotype data and gene expression data with covariates adjusted. Further details are available in the original PsychENCODE publication^[18] and on the PsychENCODE website (http://resource.psychencode.org/). For TWAS results, only genes that pass the Bonferroni-corrected P threshold (0.05/12434 = 4.02 × 10⁻⁶) are considered as TWAS significant genes.

Differential expression analysis

To identify differentially expressed genes between SCZ patients and controls, we utilized differential expression results from the PsychENCODE project^[18], which quantified gene expression in postmortem human brain tissues. In brief, PsychENCODE collected RNA-seq data from postmortem human brains of SCZ patients and non-psychiatric controls across eight studies. They conducted strict quality control on all samples, discarding those with an ambiguous diagnosis, age, sex, or fewer than 10 million total reads. Outlier samples were also removed. After rigorous quality control, a total of 1442 high-quality RNA-seq data sets were obtained (including 899 non-psychiatric controls and 543 SCZ patients), and all samples were from the frontal or temporal cortex. In differential gene expression analysis, the authors considered the impact of all potential covariates on the results (including age, sex, and sequencing platform, among others) and incorporated these covariates into the differential gene expression analysis model. Finally, they identified 4821 genes that showed differential expression (FDR < 0.05) in SCZ cases and controls. More details on sample collection, expression quantification, and quality control can be found in the original article^[18] and on Synapse (doi.org/10.7303/syn12026837.1). The normalized expression values of RERE in SCZ cases and controls were downloaded from the PsychENCODE website (http://resource.psychencode.org/), and visualizations were generated using ggplot2 in R (v4.1.1).

Cell culture

HEK-293T, SH-SY5Y, and U251 cell lines were purchased from the cell bank of Kunming Institute of Zoology, Chinese Academy of Sciences. B104 cells were purchased from Procell, Wuhan, China. HEK-293T, U251, and B104 cells were cultured in Dulbecco’s Modified Eagle medium (DMEM) (Gibco, C11995500BT), supplemented with 10% fetal bovine serum (FBS) (Gibco, 10091148) and 100 units/mL penicillin-streptomycin (Gibco, 15140-122). SH-SY5Y cells were cultured in DMEM (Gibco, C12430500BT) supplemented with 10% FBS, 1 mM sodium pyruvate solution (Gibco, 11360070), 1× MEM non-essential amino acids solution (Gibco, 11140050), and 100 units/mL Penicillin-Streptomycin. All the cells were cultured at 37 °C in a humidified atmosphere containing 5% CO₂. PCR-based detection was conducted periodically to ensure that there was no mycoplasma contamination in the cultured cells.

Plasmid construction and generation of engineered cells

The full-length human RERE coding sequence (ENST00000377464.5) was cloned into the pCDH vector containing a 3× FLAG tag and GFP reporter. All constructs were verified by Sanger sequencing. For mNSCs, RERE overexpression was achieved by lentiviral transduction, followed by puromycin selection to establish stable overexpression prior to downstream experiments. Primary neurons used for assessing RERE overexpression efficiency and synapse-associated protein expression were transduced with lentivirus. For other neuronal experiments, RERE overexpression was achieved by transient transfection. RERE-overexpressing B104 cells were generated by lentiviral transduction and selected with puromycin to obtain stable cell populations before subsequent analyses. Other plasmids and engineered cell lines used in this study are described in detail in the corresponding Methods sections below. Plasmids were obtained from commercial sources or collaborators, where indicated.

Dual-luciferase reporter gene assays

Dual-luciferase reporter gene assays were performed as previously described^[24]–[26]. As the two functional variants are not located in the promoter region, we assessed their enhancer activity. DNA sequences (about 600 bp) containing the functional variants were cloned into the pGL3-Promoter vector (Promega, E1761). HEK-293T, SH-SY5Y, and U251 cells were seeded in 96-well plates at densities of 3.5 × 10⁴, 1 × 10^5, and 3 × 10⁴ cells per well, respectively. After 24 h of plating, 100 ng of recombinant vector and 20 ng of the pRL-TK vector (internal control; Promega, E2241) were transfected into HEK-293T cells. For SH-SY5Y and U251 cells, 150 ng of recombinant vector and 30 ng pRL-TK vector were transfected using the Lipofectamine^TM 3000 transfection reagent (Invitrogen, L3000-015). Twenty-four hours post-transfection, the Dual-Luciferase® Reporter 1000 Assay System (Promega, E1980) was used to quantify luciferase activity in HEK-293T cells. For SH-SY5Y and U251 cells, luciferase activity was measured 48 h post-transfection. Luminoskan Ascent chemiluminescence analyzer (Thermo Scientific, Thermo Luminoskan Ascent) was used to detect fluorescence activity. All sequences were validated by Sanger sequencing, and primer sequences are listed in Supplementary Table.

To assess the activity of the Grin2a promoter region, two predicted motif-containing fragments, F1 and F2 (± 500 bp around the putative RXRA and RARB binding motifs), were cloned into the pGL3-Promoter vector. These vectors were transfected into B104 cells, as well as B104 cells overexpressing RERE, and luciferase activity was measured 48 h post-transfection. The sequences of the promoter region of Grin2a, F1, and F2, are shown in Supplementary Fig..

EMSAs

EMSAs were carried out as previously described^[24]–[26]. The EMSA Probe Biotin Labeling Kit (Beyotime, GS008) was used to label the synthesized oligonucleotide strands. The labeled oligonucleotides were then annealed. The nuclear extracts from SH-SY5Y cells were prepared using the Nuclear and Cytoplasmic Protein Extraction Kit (Beyotime, P0028). EMSA experiments were performed using the Chemiluminescent EMSA Kit (Beyotime, GS009). Nuclear proteins were first incubated without labeled probes for 10 min, then labeled probes were added, and the mixture was incubated for 20 min before electrophoresis. For supershift experiments, the nucleoprotein was incubated with the corresponding TF antibody and unlabeled probes for 10 min, followed by the addition of labeled probes and incubation for 20 min. The sequences of all oligonucleotides are listed in Supplementary Table.

eQTL analysis

In order to explore whether rs159961 and rs301792 were associated with gene expression in the human brain, we examined their association with gene expression in three brain eQTL datasets, including the Common Mind Consortium (CMC)^[69], the Lieber Institute for Brain Development (LIBD)^[72], and BrainMeta v2^[16]. Briefly, the CMC dataset contains eQTL data from brain tissues of 467 post-mortem human brain samples, the LIBD dataset includes eQTL data from brain tissues of 412 post-mortem human brain samples, and the BrainMeta v2 dataset integrates brain sample genotype and RNA-seq data from seven cohorts (N = 2443). Detailed information on sample collection, data processing, and eQTL generation can be found in the original studies^{[16],[69],[72]}.

CRISPR/Cas9-mediated genome editing

CRISPR/Cas9-mediated gene editing was used to knock out the segments containing the rs159961 and rs301792. Two sgRNAs (located upstream and downstream of the target SNP) were designed using the tool developed by the Zhang lab (https://zlab.bio/guide-design-resources). The sgRNAs were ligated into the pX459 vector (provided by Mr Yan Ruiqing), and the recombinant vector (3 μg) was transfected into SH-SY5Y (cultured in a 6-well plate) using the Lipofectamine^TM 3000. The transfected cells were selected using puromycin (1 μg/mL), and puromycin-resistant cells were seeded into a 10 cm dish at a very low density for monoclonal growth. After several days of culture, monoclones were selected and plated onto 48-well plates. The knockout efficiency of each single clone was determined by PCR. The sequences of sgRNAs are listed in Supplementary Table, and the primer information for knockout efficiency verification is listed in Supplementary Table.

Real-time qPCR

Total RNA was extracted using TRIzol Reagent (Invitrogen, 10296028CN). Following the manufacturer’s protocol, 1 μg total RNA was reverse transcribed into complementary DNA (cDNA) using the PrimeScript™ RT reagent Kit (Takara, RR047A). The expression of the target gene was determined using TB Green^® Premix Ex Taq™ II (Tli RNaseH Plus) (Takara, RR820A). ACTB (for human) or Actb (for mouse or rat) was used as the internal control, and the results were analyzed using the 2^−ΔΔCt method^[73]. All qPCR experiments were performed with three independent biological replicates, each measured in three technical replicates. qPCR was carried out using a CFX96 or CFX Opus 96 Real-Time PCR System (Bio-Rad), with data acquisition performed using Bio-Rad CFX Maestro and Bio-Rad CFX Manager software. qPCR primer sequences are listed in Supplementary Data.

Knockdown assays

The shRNAs were designed using the online tool at https://rnaidesigner.thermofifisher.com and were cloned into the pLKO.1-EGFP-Puro vector (from XIAMEN Anti-hela Biological Technology Trade Company) or the pSicoR-Ef1a-mCh-puro vector (Addgene, 31845). The lentiviral packaging plasmids psPAX2, pMD2.G, and the constructed vector were co-transfected into HEK-293T cells (cultured in a 10 cm dish) using polyethyleneimine (PEI). Forty-eight hours post-transfection, the packaged viral particles were collected through a 0.22 μm filter and were used to infect SH-SY5Y cells or rat primary cortical neurons (cultured in a 6-well plate). qPCR was used to determine the expression levels of REST, POLR2A, RERE, Rxra, Rarb, and Grin2a. All shRNA sequences are listed in Supplementary Table.

Isolation and characterization of mNSCs

Isolation and characterization of mNSCs were performed as previously described in refs. ^[24]–[29]. Briefly, the brains of embryonic day 14.5 (E14.5) both male and female mice (C57BL/6J) were dissected, and tissues from the subventricular zone (SVZ) were obtained. The tissues were placed into pre-cooled proliferation medium (DMEM/F-12, Gibco, C11330500BT; 1× N2, Gibco, 17502048; 1×B27, Gibco, 12587010; basic fibroblast growth factor (bFGF) (20 ng/mL), STEMCELL, 78003; epidermal growth factor (EGF) (20 ng/mL), STEMCELL, 78006; and heparin, STEMCELL, 7980) and were dissociated with a 1 mL pipette (by repeated pipetting). The resultant cells were filtered through a 70 μm cell strainer and transferred into a fresh proliferation medium. After culturing for 2–3 days, neurospheres were selected and plated onto culture dishes (pre-coated with laminin, 3 μg/mL, Sigma, L2020-1mg). The identity of the isolated mNSCs was characterized with three neural stem cell markers, including SOX2 (R&D Systems, AF2018), PAX6 (Merck, AB2237), and NESTIN (Merck, MAB353).

Fluorescent-activated cell sorting

The 1.0 × 10⁶ mNSCs were plated onto 6-well plates, which were pre-coated with laminin (3 µg/mL). After culturing for 48 h, the mNSCs were harvested using StemPro Accutase (Gibco, A1110501). The dissociated cells were collected into a 15 mL centrifuge tube and centrifuged at 1000×g for 3 min at 4 °C. The cell precipitation was gently rinsed with precooled phosphate buffer saline (PBS). The cells were then resuspended in 1 mL precooled PBS, and 3 mL absolute ethanol was added dropwise to achieve a final concentration of 75% ethanol. The mixture was thoroughly mixed and incubated at a temperature of 4 °C for 24 h. The cell precipitation was collected by centrifugation at 1000×g for 5 min at 4 °C and rinsed with precooled PBS. The cells were resuspended in 500 μL of propidium iodide (Beyotime, C1052) and incubated for 30 min at 37 °C, shielding the samples from light. Red fluorescence was detected using a flow cytometer (BD, LSRFortessa™), and the cell cycle was analyzed with FlowJo software version 10.8.1.

Proliferation of mNSCs

Cell Counting Kit-8 (CCK-8) and 5-Ethynyl-2’-deoxyuridine (EdU) assays were carried out as previously described^[24]–[29]. In brief, for the CCK-8 assay, 1 × 10⁴ cells were plated onto 96-well plates (pre-coated with laminin, 3 μg/mL). After culturing for 0, 24, 48, and 72 h, the absorbance at 450 nm was measured to estimate the cell viability. For the EdU assay, 2 × 10⁵ cells were plated onto 24-well plates (pre-coated with laminin, 3 μg/mL). After culturing for 48 h, 20 μM EdU (RIBO-Bio, C00054) was added into the medium and incubated for 45 min at 37 °C (to label proliferating cells). EdU assays were conducted with the Cell-Light^TM EdU Apollo567 In Vitro Kit (RIBO-Bio, C10310-1) to detect EdU-labeled cells. Images were captured with a laser scanning confocal microscope (OLYMPUS, FV1000). Image analysis was performed using ImageJ software on three independent biological replicates, with five images acquired for each replicate.

Differentiation of mNSCs

Differentiation assays were performed as previously described^[24]–[29]. Briefly, 2 × 10⁵ cells and 1 × 10⁶ cells were plated onto 24-well and 6-well plates (pre-coated with laminin, 3 μg/mL), respectively. After culturing for 24 h in proliferation medium, the medium was replaced with differentiation medium (proliferation medium without bFGF and EGF). After differentiating for 3 days, cells were fixed for immunofluorescence (IF) staining. GFAP (Sigma, G9269) and MAP2 (Merck, AB5622) antibodies were used to label the astrocytes and neurons, respectively. Images were captured with a laser scanning confocal microscope (OLYMPUS, FV1000), with a total of 15 images obtained from three biological replicates. Image J software was used to count the GFAP-positive and MAP2-positive cells. qPCR was performed to detect the expression levels of Gfap, Map2, Tubb3, and Cldn11.

Western blot

Western Blot was carried out to determine protein expression levels. Cells were lysed in cold RIPA Lysis Buffer (Beyotime, P0013B) supplemented with PMSF (Beyotime, ST507) for 30-60 min. Homogenates were centrifuged at 12,000×g for 10 min at 4 °C, and the supernatants were collected. After the protein concentration was determined using the BCA Protein Assay Kit (Beyotime, P0010S), protein samples were mixed with loading buffer and heated in a dry bath heater at 95 °C for 5-10 min. Equal amounts of protein were separated by SDS-PAGE and transferred to PVDF membranes. The membranes were blocked for 2 h at room temperature in 5% non-fat milk in Tris-buffered saline with 0.1% Tween-20 (TBST). Subsequently, the membranes were incubated overnight at 4 °C in primary antibody dilution prepared with 5% bovine serum albumin (BSA) in TBST. After washing the membrane with TBST six times (5 min each), the membrane was incubated for 1 h at room temperature in HRP-conjugated secondary antibody diluent prepared with 5% BSA in TBST. ECL substrate was used to detect the immunoblot signals. Protein band intensity was quantified using ImageJ software.

IF staining

After removing the medium and rinsing with PBS, the cells were fixed with 4% paraformaldehyde (PFA) (the mNSCs and the rat primary cortical neurons were fixed with 4% PFA containing 4% sucrose) for 15 min. Following three 5-min rinsing sessions in PBS, the cells were permeabilized with PBS containing 0.3% Triton X-100 (PBST) for 20 min. Next, the cells were blocked using the Immunol Staining Blocking Buffer (Beyotime, P0102) for 1 h at room temperature. Following this, the primary antibody was diluted in PBST containing 2% BSA and incubated at 4 °C overnight. After several rinses with PBS, the cells were further incubated with the fluorescent secondary antibody (diluted in PBST containing 2% BSA) for 1 h at room temperature. Subsequently, the samples were mounted and imaged under a fluorescence microscope.

Transcriptome analysis

Differential expression analysis between RERE-overexpressed mNSCs and the control group was conducted following the widely used RNA-seq analysis pipeline^[74]. RNA sequencing was performed by Novogene Co. Ltd. After adapter trimming and filtering out low-quality reads, FastQC v0.11.9 was used to further confirm the high quality of reads. Clean reads were aligned to the grcm38 reference genome(downloaded from: https://cloud.biohpc.swmed.edu/index.php/s/grcm38_tran/) using HISAT2 v2.2.1^[75]. The SAM files were then sorted and converted to BAM format using SAMtools v1.9^[76]. Subsequently, StringTie v2.1.7 and PrepDE.py scripts^[77] were utilized for gene-level reads assembly and expression quantification. After filtering low-expressed genes (only genes with >10 counts in more than four samples were retained), differential expression analysis was performed on the quantified count matrix using DESeq2 v 1.42.0^[78]. Genes with |Log₂FoldChange | >1 and a Benjamini-Hochberg (BH) method adjusted P-value < 0.05 were considered to be differentially expressed.

Isolation of the rats' primary cortical neurons

The brains of embryonic day 17.5–18.5 (E17.5–18.5) Sprague Dawley rats (both males and females) were dissected, and tissues from the frontal cortex were isolated. The tissues were cut into small pieces and rinsed with pre-cooled Hank’s Balanced Salt Solution (HBSS) (Gibco, 14175079) supplemented with 1% HEPES (Solarbio, H8090-100g) and 200 units/mL of penicillin-streptomycin. They were then digested in a pre-prepared digestion solution (Neurobasal, Gibco, 21103049; 2% B27, Gibco, 17504044; 1% GlutaMAX, Gibco, 35050061; 2 mg/mL Papain, Worthington, LS003119; 5 units/mL DNase I, Sigma, D4263) at 37 °C for 20 min. The digestion was quenched with 5 mL of neuronal medium containing FBS (Neurobasal; 2% B27; 1% GlutaMAX; 1% Penicillin-Streptomycin; 2.5% FBS, Gibco, 10091148). The digested tissues were transferred to a 15 mL centrifuge tube, and the supernatant was aspirated after standing for 30 s. The remaining digested tissues were resuspended with 1.5 mL FBS-containing neuronal medium and slowly pipetted with the pre-sterilized pasteurized tube 8–9 times. After pipetting, the suspension was allowed to stand for 2–3 min to permit the undigested tissue mass to settle. The supernatant cell suspension was filtered through a 70 μm strainer and plated onto pre-coated 6-well plates. After 6-8 h of plating, half of the medium was aspirated from each well, and FBS-free neuronal medium (Neurobasal, 2% B27, 1% GlutaMAX) was added. The neurons were cultured for about two weeks, and the medium was changed every three days.

Analysis of the density and morphology of dendrite spines

After culturing the isolated rat primary neurons for 14–15 days, 3 μg pCDH-RERE-3×flag vector and 1 μg Venus plasmid (expressing GFP) were co-transfected into rat cortical neurons using Lipofectamine^TM 3000. Three days post-transfection, IF staining was performed. The dendritic spines of neurons were captured with a laser scanning confocal microscope (ZEISS 880). NeuronStudio and ImageJ were used to analyze the density and morphology of neuronal dendritic spines.

Analysis of the complexity of neuronal dendrites

The isolated rat primary neurons were cultured for 3–4 days, then a co-transfection was performed using Lipofectamine^TM 3000 to transfect 3 μg pCDH-RERE-3× flag vector and 1 μg Venus plasmid (expressing GFP) into rat cortical neurons. Neuronal morphology was captured using a ZEISS 880 confocal microscope, and the complexity of neurons was assessed using ImageJ software by Sholl analysis (https://imagej.net/Sholl)^[79].

Electrophysiological recordings

Recordings were obtained using an EPC 10 patch clamp (HEKA). The data were digitized at 10 kHz. The holding potential of primary rat frontal cortex neurons was set at −70 mV, and series resistance compensation was adjusted to 60–80%. Recordings with series resistances > 20 MΩ or the leak current (I_leak) > −200 pA were rejected. The cells were recorded in an extracellular solution containing 140 mM NaCl (Sigma, S3014), 5 mM KCl (Sigma, P9541), 2.5 mM CaCl₂ (Sigma, C7902), 10 mM HEPES (Sigma, H4034), and 10 mM D-glucose (Sigma, D9434) at pH 7.4 and 315 Osm. Electrodes with a resistance of 3–5 MΩ were pulled from borosilicate glass capillaries (WPI). The intracellular solution contained 120 mM K-Gluconate (MCE, HY-Y0569C), 5 mM NaCl (Sigma, S3014), 1 mM MgCl₂ (Sigma, M2670), 0.2 mM EGTA (Sigma, E43T8), 2 mM MgATP (Sigma, A9187), 0.1 mM Na₂GTP (Sigma, G8877), 10 mM HEPES (Sigma, H4034), and 10 mM phosphocreatine disodium (Sigma, HY-D0885B), adjusted to pH 7.2. When recording NMDAR-mEPSCs, 1 µM TTX (Hellobio, HB1034), 10 mM bicuculline (MCE, HY-N0219), and 10 µM CNQX (MCE, HY-15066) were added to the extracellular solution. Primary rat frontal cortex neurons from embryonic day 17.5–18.5 (E17.5–18.5) Sprague-Dawley rats were cultured on poly-D-lysine-coated (Sigma, P6407) glass coverslips and maintained at 37 °C in the presence of 5% CO₂ for 16–18 days before the experiment. Cells were transfected with 3 µg of plasmid per well in six-well plates, and electrophysiological recordings were performed 48–72 h after transfection. Data were analyzed using Clampfit 9.02 (Molecular Devices), Igor Pro 6.0 (WaveMetrics), and Prism 8 (GraphPad Software). Cumulative probability plots of NMDAR-mEPSC were plotted in Origin 2021. The normality of data distribution was assessed using the D’Agostino-Pearson omnibus normality test. Statistical differences were evaluated using the two-tailed Mann–Whitney U-test, with a significance threshold set at P < 0.05.

Co-IP

We performed Co-IP using the Immunoprecipitation Kit with Protein A + G Magnetic Beads (Beyotime, P2179). Primary neurons from the rat prefrontal cortex were collected and lysed in lysis buffer supplemented with protease inhibitors. After complete lysis, the samples were centrifuged at 10,000–14,000×g for 3–5 min at 4 °C, and the supernatant was collected for subsequent immunoprecipitation assays. The RERE antibody (Bioss, bs-7077R) was diluted in 1× TBS and incubated with pre-washed Protein A + G magnetic beads on a rotator at room temperature for 15 min to 1 h. After washing the beads three times with 1× TBS, the collected protein lysates were added to the antibody-conjugated Protein A + G beads and incubated overnight at 4 °C on a rotator. Following incubation, the supernatant was discarded, and the beads were washed three times with 1× TBS. The bound proteins were eluted by adding 100 μL of 1× SDS-PAGE Sample Loading Buffer, followed by heating at 95 °C for 5 min. Finally, the samples were separated on a magnetic rack for 10 s, and the supernatant was collected for Western blot.

Chromatin immunoprecipitation (ChIP)-qPCR

We performed ChIP experiments using the SimpleChIP® Enzymatic Chromatin IP Kit (CST, 9003) according to the manufacturer’s instructions. Primary rat prefrontal cortical neurons were cross-linked with 1% formaldehyde at room temperature for 10 min, followed by quenching with glycine solution for 5 min at room temperature. After removing the supernatant, cells were washed twice with ice-cold PBS and collected in ice-cold PBS containing protease inhibitors. The cells were pelleted by centrifugation at 2000×g for 5 min at 4 °C for subsequent experiments. Chromatin was digested with micrococcal nuclease and incubated at 37 °C for 20 min. Cross-linked chromatin fragments were then incubated overnight at 4 °C with the indicated antibodies, including IgG, RXRA (CST, 3085S), and RARB (Abcam, ab53161). For each IP reaction, 30 µL of ChIP-grade protein G magnetic beads were added and incubated for 2 h at 4 °C. Enriched chromatin complexes were recovered by magnetic separation and washing, and the binding of RXRA and RARB was assessed by RT-qPCR. qPCR primer sequences are listed in Supplementary Table.

Ethics statement

All animal experiments were conducted in accordance with institutional guidelines and approved by the Institutional Animal Care and Use Committees of Southeast University (SEU-IACUC-20250226017) and Kunming Institute of Zoology, Chinese Academy of Sciences (SMKX-20200701-35). Primary rat neurons and primary mouse neural stem cells (mNSCs) were prepared from embryonic rats and mice, respectively, following approved protocols. This study also used publicly available datasets, and no additional ethical approval was required for the analysis of these data. No human participants or clinical samples were involved in this study.

Reporting summary

Further information on research design is available in the linked to this article.

Supplementary information

Source data

Acknowledgements

This study was equally supported by the Youth Science Fund Project (Category A) of the National Natural Science Foundation of China (Grant No. 82525026 to X.-J.L.) and the Brain Science and Brain-like Intelligence Technology-National Science and Technology Major Project (2025ZD0216000 to X.-J.L). It was also supported by grants from the National Key Research and Development Program of China (2024YFF0728700 to C.Z., 2023YFF0724802 to C.Z.), Beijing Natural Science Foundation (F251014 to C.Z.), the National Natural Science Foundation of China (82301690 to Y.F.L, 82401767 to C.G.M, U2102205 to X.-J.L., U22A20304 to X.-J.L., 81925011 to C.Z., 92149304 to C.Z., 82401765 to J.W., 32450728 to C.Z., 92468302 to C.Z.), the startup funds from Southeast University (RF1028623032 to X.-J.L.), the Postdoctoral Fellowship Program of CPSF (GZB20230136 to Y. Li., GZB20240678 to J.W.), and Chinese Institutes for Medical Research, Beijing (CX23YZ03 to C.Z). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We thank Misses Zhongli Ding, Qian Li, and Yinuo Cao for their technical assistance.

Author contributions

X.-J.L. conceived, designed, and supervised the whole study. YXL performed most of the experiments and analyzed data under the supervision of X.-J.L. J.W. performed REST TF knockdown experiments, mNSCs cycle experiments, and dendritic spine analysis of rat neurons. H.Y., X.C., J.F., and C.Z. performed electrophysiological recordings and conducted subsequent data processing and analysis. Y. Li conducted dendritic spine analyses, complexity experiments of dendritic spines, and ChIP qPCR experiments. Statistical fine-mapping, colocalization analysis, haplotype-based analysis, TWAS, eQTL, transcriptome, and differential expression analyses were all carried out by X.D., C.M., J.L., and J.Y. D.L. and M.L. performed plasmid and DNA extraction, and isolated primary rat cortical neurons. S.L. performed dual luciferase reporter gene assays. X.C. performed a Western blot. X.-J.L. and Y. Liu interpreted the results and drafted the manuscript. X.-J.L. oversaw the project and finalized the manuscript. All authors revised the manuscript critically and approved the final version.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Materials availability

Materials generated in this study are available from the corresponding author upon reasonable request.

Data availability

Transcriptomics data generated in this study (i.e., from RERE-overexpressing and control mNSCs) are available from the Genome Sequence Archive (CNCB-NGDC) under accession number CRA025203, [https://ngdc.cncb.ac.cn/gsa/browse/CRA025203]. The data that support the findings of this study are available in the following public databases: the WashU Epigenome Browser [https://epigenomegateway.wustl.edu/]; the ENCODE database [https://www.encodeproject.org/]; the JASPAR database [https://jaspar.elixir.no/]; the CMC eQTL database [https://www.synapse.org/CMC]; the LIBD eQTL database [https://eqtl.brainseq.org/phase1/eqtl/]; [https://eqtl.brainseq.org/phase2/eqtl/]; BrainMeta v2 eQTL database [https://yanglab.westlake.edu.cn/software/smr/#eQTLsummarydata]; PsychENCODE [http://resource.psychencode.org/]; 3DIV; Brain RNA-Seq [http://www.brainrnaseq.org/]; Allen Institute for Brain Science, ABC Atlas [https://portal.brain-map.org/atlases-and-data/bkp/abc-atlas]; and BrainSpan atlas [http://www.brainspan.org/]. Source Data supporting the figures in this study are provided with this paper. Other data are available within the Article and Supplementary Files. are provided with this paper.

Competing interests

The authors declare no competing interests.

Footnotes

Contributor Information

Chen Zhang, Email: czhang@188.com.

Jin Feng, Email: fengjin@ccmu.edu.cn.

Xiong-Jian Luo, Email: luoxiongjian@seu.edu.cn.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-68449-6.

References

Associated Data

Supplementary Materials

Data Availability Statement

Materials generated in this study are available from the corresponding author upon reasonable request.

Transcriptomics data generated in this study (i.e., from RERE-overexpressing and control mNSCs) are available from the Genome Sequence Archive (CNCB-NGDC) under accession number CRA025203, [https://ngdc.cncb.ac.cn/gsa/browse/CRA025203]. The data that support the findings of this study are available in the following public databases: the WashU Epigenome Browser [https://epigenomegateway.wustl.edu/]; the ENCODE database [https://www.encodeproject.org/]; the JASPAR database [https://jaspar.elixir.no/]; the CMC eQTL database [https://www.synapse.org/CMC]; the LIBD eQTL database [https://eqtl.brainseq.org/phase1/eqtl/]; [https://eqtl.brainseq.org/phase2/eqtl/]; BrainMeta v2 eQTL database [https://yanglab.westlake.edu.cn/software/smr/#eQTLsummarydata]; PsychENCODE [http://resource.psychencode.org/]; 3DIV; Brain RNA-Seq [http://www.brainrnaseq.org/]; Allen Institute for Brain Science, ABC Atlas [https://portal.brain-map.org/atlases-and-data/bkp/abc-atlas]; and BrainSpan atlas [http://www.brainspan.org/]. Source Data supporting the figures in this study are provided with this paper. Other data are available within the Article and Supplementary Files. are provided with this paper.