Dlgap2 deficiency disrupts synaptic homeostasis by promoting ubiquitin-mediated Itsn1 degradation in a valproic acid-induced autism-like model

Abstract

Prenatal valproic acid (VPA) exposure increases the risk of neurodevelopmental disorders, though its synaptic mechanisms remain unclear. Using multi-omics analyses, we identified Dlgap2 as a consistently dysregulated protein in VPA models. Mice with Dlgap2 knockdown exhibited synaptic deficits and autism-like behaviors, including social and cognitive impairments. Proteomics of postsynaptic density following Dlgap2 knockdown revealed disruption of synaptic organization and a specific reduction in Intersectin-1 (Itsn1), which interacts with Dlgap2 and undergoes ubiquitin-mediated degradation upon Dlgap2 deficiency. Our study defines a Dlgap2-Itsn1 regulatory axis that underlies VPA-induced synaptic dysfunction.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-39099-x.

Introduction

Valproic acid (VPA), a widely prescribed antiepileptic and mood-stabilizing agent, has been associated with an elevated risk of neurodevelopmental disorders—including autism spectrum disorder (ASD)—when exposure occurs during pregnancy¹. Despite robust clinical and epidemiological evidence, the molecular mechanisms underlying VPA-induced synaptic dysfunction and behavioral impairments remain only partially understood². Previous research suggests that VPA may disrupt transcriptional regulation³, alter the expression of synaptic proteins⁴, and impair neural circuit stability⁵; however, a conserved pathophysiological pathway across species has yet to be clearly defined.

Synaptic scaffolding proteins are essential for maintaining the structural and functional integrity of synapses. Among these, Discs Large Associated Protein 2 (human DLGAP2, mouse Dlgap2) is a core component of the postsynaptic density, critically involved in synaptic signaling, plasticity, and stability⁶. Human genetic studies have linked DLGAP2 to several neuropsychiatric conditions, including ASD^7–9, yet its role in environmentally induced neurodevelopmental anomalies, such as those resulting from VPA exposure, has not been explored.

Another central player in synaptic regulation is Intersectin-1 (Itsn1), a multidomain scaffold protein predominantly expressed in the nervous system. Itsn1 is known to regulate endocytic processes, neurotransmitter receptor internalization, and actin cytoskeleton dynamics, each vital for synaptic transmission and plasticity¹⁰. Dysregulation of Itsn1 has been implicated in several neurological disorders¹¹, though its involvement in VPA-mediated synaptopathy remains unclear.

In this study, we leveraged an integrated multi-omics strategy incorporating human cortical organoids, Macaca fascicularis, and rodent models to identify evolutionarily conserved pathways disrupted by prenatal VPA exposure^12–14. We pinpointed Dlgap2 as a consistently dysregulated gene across models and validated its essential role in synaptic maintenance and behavior. Furthermore, we uncovered a novel mechanistic axis through which Dlgap2 deficiency triggers ubiquitin-mediated degradation of Itsn1, resulting in disrupted endocytic trafficking and loss of synaptic homeostasis. These findings not only elucidate a key molecular cascade in VPA-induced neurodevelopmental deficits but also highlight potential therapeutic targets for mitigating synaptic dysfunction in ASD and related disorders.

Result

Cross-species transcriptomic and proteomic profiling identifies Dlgap2 as a conserved VPA target

We performed comparative transcriptomic analysis of VPA-exposed human cortical organoids, Macaca fascicularis, and rat brains, which identified 23 consensus differentially expressed genes (DEGs) significantly associated with synaptic function and myelination processes (Fold Change > 1.2 or 0.83; p < 0.05) (Table S1; Fig. 1A, B)^12–14. Integrated analysis of transcriptomic and proteomic data¹⁵ from VPA-exposed mouse brains consistently identified Fmnl1 (Formin-like 1), Dlgap2 (Discs Large Associated Protein 2), and Slc7a5 (Solute Carrier Family 7 Member 5) as significantly downregulated at both mRNA and protein levels. Fmnl1 regulates actin cytoskeleton dynamics and synaptic structure integrity, with its dysregulation linked to impaired synaptic plasticity¹⁶. Dlgap2, a key postsynaptic scaffolding protein, is essential for maintaining excitatory synapse organization and stability, and has been strongly associated with neurodevelopmental disorders such as autism and schizophrenia, even in Alzheimer’s disease^17,18. Slc7a5, a critical neutral amino acid transporter, facilitates leucine uptake and mTOR signaling in the brain, thereby influencing protein synthesis and neuronal function; its deficiency may contribute to synaptic dysfunction and behavioral deficits¹⁹. The coordinated downregulation of these genes highlights disruptions in synaptic architecture, metabolic support, and signal transduction as pivotal mechanisms in VPA-induced neurodevelopmental impairment (Table S2; Fig. 1C, D). Further analysis of transcriptomic datasets revealed that while VPA exposure consistently altered the transcription of all three genes across species, a striking divergence emerged: primates showed an upward trend, whereas rodents exhibited marked downregulation—a pattern consistent with the proteomic changes observed in rodents (Fig. 1E and Supple Fig. 1A, B). This bidirectional regulation highlights species-specific synaptic vulnerability mechanisms. Subsequent interrogation of the Tabula Muris single-cell transcriptomic atlas confirmed predominant enrichment of Dlgap2 and Fmnl1 in neuronal populations, supporting their roles in neural development and synaptic integrity. In contrast, Slc7a5 expression was largely restricted to endothelial cells²⁰, suggesting a distinct functional context (Fig. 1F).

Fig. 1.

Collapse of synaptic homeostasis is a conserved mechanism of transcriptional dysregulation across species following VPA exposure. (A) Comparative transcriptomics of VPA-exposed human organoids, Macaca fascicularis and rat identified 23 consensus DEGs (FC > 1.2 or < 0.83, p < 0.05). (B) Cytoscape–String network places the 23 genes in postsynaptic receptor signaling, excitatory synaptic homeostasis, protein trafficking and myelin maintenance. (C–D) Overlay with VPA mouse brain proteomics (same cut-offs) yields Fmnl1, Dlgap2 and Slc7a5 as concordantly down-regulated at both RNA and protein levels. (E) Dlgap2 expression across three published RNA-seq datasets, including two distinct organoid models (U1M and U2F), Macaca fascicularis and rat, with the X-axis representing Log₂(fold change) and the Y-axis showing -Log₁₀(P-Value). (F) Tabula Muris single-cell atlas: Fmnl1/Dlgap2 enriched in neurons; Slc7a5 predominates in endothelial cells.

Prenatal VPA exposure disrupts Dlgap2, leading to synaptic and behavioral deficits

In order to verify whether Dlgap2 is dysregulated in the VPA model, we further conducted modeling experiments in mice. Prenatal VPA exposure induced significant cognitive deficits in mice as assessed by Morris water maze test, including prolonged escape latency, reduced time in target quadrant, and fewer platform crossings (Supple Fig. 2A-E). These animals also exhibited pronounced deficits in social behavior in the three-chamber test, demonstrating significantly impaired sociability and a reduced preference for social novelty (Supple Fig. 2F–H), consistent with previous reports²¹. Molecular profiling revealed a consistent downregulation of cortical Dlgap2 in VPA-exposed mice at both mRNA and protein levels (Fig. 2A-C). This downregulation was also evident in isolated synaptoneurosomes²², demonstrating the specific synaptic vulnerability to the insult and corroborating the established postsynaptic role of Dlgap2. Immunostaining revealed substantially decreased Dlgap2 expression in cortical neurons at postnatal days 7 and 21 (Fig. 2D, E). The treatment of primary neurons with VPA resulted in a phenotype characterized by shortened neurites, reduced postsynaptic density, and downregulated Dlgap2 expression (Fig. 2F-I), pointing to potential structural correlates of synaptic dysfunction.

Fig. 2.

Prenatal exposure to VPA leads to downregulation of Dlgap2 expression in the neonatal mouse brain. (A) The expression of Dlgap2 was significantly reduced in the cortex of VPA-treated model mice, n = 9 samples per group. (B-C) Western blot analysis revealed a marked decrease in Dlgap2 expression in VPA-exposed mice in both whole-brain lysates and synaptoneurosome-derived fractions, n = 4 samples per group in brain and n = 6 samples per group in isolated synaptoneurosomes. (D-E) Immunostaining of Dlgap2 in the mouse cortex at postnatal 7 days (P7) and postnatal 21 days (P21) demonstrated significantly lower expression in the VPA group compared to controls. (E) Quantification of staining intensity is presented, n = 3 mice (3 slices per mouse) in two groups of P7 and P21. (F-I) Following transfection of pCAG-tdTomato plasmids into primary cortical neurons (DIV 3), 2 µM VPA was administered for 24 h to assess synaptic morphology in DIV 9. (G) Total neurite length was quantified (n = 25 cells per group), (H) the number of postsynapses (the co-localized signals of PSD95 and td-Tomato were defined as postsynapse) and (I) the intensity of Dlgap2 within the postsynaptic puncta were statistically analyzed (n = 35 branches per group). (Data represent mean ± SEM. ns: not significant. p < 0.05 indicates significance between the two indicated groups. C, E: Two-way ANOVA with post hoc Tukey’s test. A, G, H: Two-sided unpaired t-test)

Dlgap2 knockdown disrupts synaptic structure and function, and recapitulates autism-like behaviors in mice

Knockout of Dlgap2 has been shown to impair neuronal morphology, disrupt synaptic formation, and subsequently cause cognitive deficits and social impairments in mice²³. To further investigate the underlying molecular mechanisms, we developed and optimized a lentivirus-mediated shRNA system targeting Dlgap2 (termed shDG2) (Fig. 3A). Among the three candidate constructs, shDlgap2-2 (shDG2-2) was validated as the most effective, exhibiting superior knockdown efficiency as confirmed through both mRNA and protein expression analyses (Fig. 3B-D). Then, lentivirus-mediated knockdown of Dlgap2 in primary neurons led to reduced neurite length and synaptic puncta density (Fig. 3E-G).

Fig. 3.

Knockdown of Dlgap2 disrupts synaptic homeostasis in neurons. (A) To minimize off-target effects, three shDlgap2 (shDG2) sequences and a scrambled control sequence were designed, with the corresponding expression constructs shown, n = 9 samples per group. (B) Lentiviral delivery of these shDG2 constructs into primary cortical neurons was followed by RT-PCR analysis, revealing that shDG2-2 achieved the most efficient knockdown, n = 4 samples per group. (C-D) Western blot analysis further confirmed the superior knockdown efficiency of shDlgap2-2 compared to other constructs, n = 4 samples per group. (E-G) Primary neurons were transduced with shDG2-2 or control lentivirus at DIV 3. By DIV 9, immunostaining for PSD95 demonstrated a marked decrease in both total neurite length (F) (n = 14 cells per group) and synaptic puncta density (G) (n = 28 branches per group). (Data represent mean ± SEM. ns: not significant. p < 0.05 indicates significance between the two indicated groups. B, D: One-way ANOVA with post hoc Tukey’s test. F, G: Two-sided unpaired t-test.)

Furthermore, we selected the construct with the highest knockdown efficiency (shDlgap2-2) and packaged it into an AAV-based interference vector (termed AAV-shDG2). This recombinant AAV was then injected into the lateral ventricles of newborn mice²⁴, successfully recapitulating key autism-like behaviors, including spatial learning deficits and impaired social novelty (Fig. 4A-H)²³. Furthermore, immunofluorescence analysis performed after behavioral testing confirmed robust viral transduction, as demonstrated by intense fluorescent signals present throughout both the cerebral cortex and hippocampus (Supple Fig. 3A). These results confirm the successful in vivo expression of both AAV-shDG2 and AAV-shCon, while also highlighting a clear correlation between Dlgap2 dysregulation and ASD-related behaviors.

Fig. 4.

Knocking down Dlgap2 can induce autistic-like behaviors in mice. (A) Schematic diagram depicting the stereotactic injection strategy of AAV-shDG2 or AAV-shCon into the lateral ventricle region of postnatal day 0 (P0) mice, along with the test procedures. (B-E) Morris water maze test. (B) Learning curve across six days of training; (C) Time spent by mice in the target quadrant after the platform was removed on the test day; (D) Number of times mice crossed the platform location on the test day; (E) Latency of mice to first reach the platform area on the test day. (F-H) Three-chamber social interaction test. (F) The habituation period; (G) The social preference test and (H) The social memory test. (All the mice used in behavior tests are male and 1.5 months old. Mouse number used in behavior tests: AAV-shCon: 10 mice; AAV-shDG2: 10 mice. Data represent mean ± SEM. p < 0.05 indicates significance between the two indicated groups. C, D, E: Two-sided unpaired t-test. F, G, H: Two-sided paired t-test.)

Dlgap2 knockdown disrupts the synaptic proteome and specifically reduces Itsn1 abundance

To elucidate the downstream mechanisms underlying Dlgap2 deficiency, we performed quantitative proteomic profiling of postsynaptic density (PSD) fractions purified from Dlgap2-knockdown cortical neurons, using scrambled RNA-treated cortical neurons as controls^24,25. This analysis identified 241 differentially expressed proteins, comprising 189 downregulated and 52 upregulated proteins (Table S2; Fig. 5A, B). Bioinformatic analysis revealed significant enrichment of pathways related to synaptic organization, vesicle trafficking, and endocytosis (Fig. 5C, D). Among these, a heatmap was generated to visualize the expression levels of key synaptic proteins—including postsynaptic density components, receptors, endocytosis regulators, and kinases—implicated in synaptic homeostasis pathways. The analysis revealed pronounced and consistent downregulation of Itsn1, a pivotal endocytic regulator involved in synaptic vesicle recycling (Fig. 5E)^10,26.

Fig. 5.

Proteomic analysis of PSD component reveals that knockdown of Dlgap2 in neurons affects events such as synapse formation and neuronal endocytosis. (A) We isolated postsynaptic density (PSD) protein and conducted a proteomic analysis in Dlgap2-KD primary cortical neurons. (B) Volcano plot shows differentially expressed proteins (DEPs) in PSD components. In Dlgap2-KD group, 189 proteins were down-regulated and 52 up-regulated. Down-regulated proteins included Dlgap2, validating knockdown effectiveness. (C) Gene Ontology (Cellular Component) analysis reveals that DEPs are predominantly localized in synapses and have a significant impact on synaptic functions. And the red boxes represent the synapse-related annotation items. (D) The functional enrichment network diagram reveals that the DEPs are predominantly correlated with a series of biological processes, such as endosomal transport, regulation of endocytosis, regulation of synaptic organization, and autophagy. (E) Heat map shows the changes in the expression levels of crucial synaptic proteins (including postsynaptic density, receptors, endocytosis regulatory, and kinase). Through this visualization, it was evident that synapses suffered substantial damage. A remarkable downregulation of Intersectin 1 (Itsn1), a key neuronal endocytosis regulator, was observed.

We further compared the 241 differentially expressed proteins with 913 risk genes from the SFARI ASD database²⁷ and found only 11 overlapping candidates. Although this overlap is limited, it is notable that both Dlgap2 and Itsn1 were among the enriched proteins, underscoring the potential significance of the Dlgap2–Itsn1 axis dysregulation in ASD pathology (Supple Fig. 4A).

LC-based co-immunoprecipitation confirms that Dlgap2 and Itsn1 are present in the same protein complex in synaptoneurosomal extracts (Table S3; Fig. 6A, B). Gene Ontology and KEGG analyses of Dlgap2-binding proteins highlighted significant enrichment of synapse and endocytic pathways (Fig. 6C, D). Identification of Itsn1, a known ASD gene, as a key convergent node emerged from an overlap analysis between the Dlgap2 interactome and the proteome dysregulated upon Dlgap2 knockdown, narrowing 23 candidate proteins under its regulatory influence (Table S4; Fig. 6E, F). These data demonstrate a biochemical interaction between Dlgap2 and Itsn1.

Fig. 6.

By analyzing the interaction network of Dlgap2 within the synaptoneurosomes, endocytosis was identified as a significantly enriched process, with Itsn1 being one of the associated proteins. (A) Schematic diagram of the proteomic analysis of Dlgap2 antibody-enriched complexes from whole-brain synaptoneurosomes. (B) 853 Dlgap2 potential specific-binding proteins were identified. (C) Gene Ontology (GO) Cellular Component analysis identified 853 Dlgap2-specific binding proteins, with a significant enrichment in synaptic components. (D) KEGG pathway analysis revealed 853 Dlgap2-specific binding proteins significantly correlated with neuronal endocytic processes. (E-F) Overlap between Dlgap2-binding proteins and proteins dysregulated by Dlgap2 knockdown revealed 23 DEPs, visualized by heatmap in F (including Itsn1).

Dlgap2 stabilizes Itsn1 by preventing ubiquitin-mediated degradation

We first investigated the expression changes of Itsn1 in three reanalyzed RNA-seq datasets. The results showed that significant alterations occurred only in the Organoid-U1M and rat models following VPA exposure, with opposing directional changes observed between these two systems. In contrast, no significant change was detected in the whole cortex of VPA-exposed mice (Supple Fig. 5A). Interestingly, further examination of Itsn1 in synaptoneurosomal fractions revealed a pronounced decrease, suggesting that protein-level fluctuations in subcellular compartments—particularly the synapse—may be functionally linked to ASD pathogenesis and merit further investigation (Supple Fig. 5B, C).

We further demonstrated that Dlgap2 knockdown leads to a pronounced decrease in Itsn1 protein levels within postsynaptic compartments (Fig. 7A–C). Mechanistic studies revealed that Dlgap2 loss promotes the degradation of Itsn1 by K48-mediated ubiquitin-proteasome-dependent pathway (Fig. 7D, E)²⁸. Furthermore, treatment with MG132²⁹—a canonical inhibitor of the ubiquitin-proteasome pathway—effectively restored Itsn1 levels in Dlgap2 knockdown cortical neurons exposed to cycloheximide (CHX)³⁰ (Fig. 7F, G), further supporting the critical involvement of ubiquitin-proteasome-mediated degradation in this regulatory process (Fig. 7D, E). These results establish a previously unrecognized regulatory axis in which Dlgap2 post-translationally stabilizes Itsn1, thereby maintaining synaptic endocytic function and overall synaptic homeostasis (Fig. 7H, I).

Fig. 7.

Knockdown of Dlgap2 triggers ubiquitin-mediated degradation of Itsn1. (A) An interaction between Dlgap2 and Itsn1 was detected by co-immunoprecipitation using anti-Dlgap2 antibody on whole-brain synaptoneurosomal lysates. n = 2 samples per group. (B-C) Knockdown of Dlgap2 leads to a decrease in Itsn1 levels within the postsynaptic compartment (PSD fraction) from DIV 9 primary cortical neurons. n = 6 samples per group. (D-E) Knockdown of Dlgap2 reduces protein levels of Itsn1, PSD95, and Homer1 in synaptoneurosomes from shDG2-injected mice compared with shCon controls. n = 4 samples per group. (F-G) Dlgap2 downregulation promotes the K48-mediated ubiquitin-proteasome-mediated degradation of Itsn1 in primary cortical neurons (DIV 9). n = 4 samples per group. (H-I) Treatment with MG132 restored Itsn1 expression in CHX-treated Dlgap2 knockdown cortical neurons. n = 4 samples per group. (Data represent mean ± SEM. ns: not significant. p < 0.05 indicates significance between the two indicated groups. C, E, G: Two-way ANOVA with post hoc Tukey’s test. I: One-way ANOVA with post hoc Tukey’s test)

Discussion

This study establishes that VPA-mediated Dlgap2 deficiency disrupts synaptic homeostasis by promoting ubiquitin-mediated degradation of Itsn1, ultimately leading to synaptic dysfunction and manifestations of autism-like behaviors. The dysregulation of Dlgap2 across human, non-human primate, and rodent models underscores its fundamental role in mediating neurodevelopmental impairments induced by prenatal VPA exposure. These findings provide crucial mechanistic insights into how dysregulated protein turnover in synapses may contribute to neurodevelopmental disorders, and highlight the Dlgap2-Itsn1 axis and the ubiquitin-proteasome system as potential therapeutic targets for ameliorating VPA-induced neurological deficits.

Advances in sequencing technologies have revealed strong associations between both Dlgap2 and Itsn1 and autism spectrum disorder^8,31. The identification of the Dlgap2-Itsn1 regulatory axis offers a unified mechanistic framework for understanding the synaptopathology observed in VPA-exposed animals, bridging genetic susceptibility and environmental insult^32,33. The involvement of ubiquitin-dependent protein degradation not only clarifies the post-translational mechanisms underlying synaptic instability but also positions protein quality control as a central hub in environmentally triggered neurodevelopmental disorders³⁴. This raises the broader hypothesis that aberrant ubiquitin-proteasome activity may be a common pathway through which diverse risk factors—both genetic and environmental—converge to disrupt neurodevelopment.

Moreover, the precise regulatory mechanisms governing Dlgap2 transcription following VPA exposure remain an important area for further investigation. As a class I histone deacetylase (HDAC) inhibitor³⁵, prenatal VPA treatment may disrupt epigenetic regulation by altering histone acetylation patterns, potentially resulting in the dysregulated expression of synaptic genes such as Dlgap2³⁶. Future studies should prioritize evaluating whether pharmacological or genetic inhibition of Itsn1 ubiquitination can rescue synaptic integrity and behavioral phenotypes in VPA-exposed or Dlgap2-deficient models. In addition, predicting and validating the specific interaction sites between Dlgap2 and Itsn1 represents a critical next step. It is plausible that in the absence of Dlgap2 expression, lysine residues at key Itsn1 interaction sites may become accessible to ubiquitinating enzymes, leading to targeted ubiquitination and subsequent proteasomal degradation of Itsn1.

The identification of the specific E3 ubiquitin ligase responsible for Itsn1 degradation upon Dlgap2 deficiency remains an important open question. Several promising candidates warrant further investigation, particularly those implicated in synaptic regulation and neurodevelopmental disorders. Notably, Nedd4-family HECT-domain E3 ligases are critically involved in synaptic protein turnover through the ubiquitination of key scaffolds, positioning them as central regulators of neurodevelopment and synaptic function^37,38. Similarly, select TRIM proteins, such as TRIM67, which critically regulate neuronal development and connectivity by targeting cytoskeletal components³⁹, represent another plausible category of E3 ligases that may mediate Itsn1 ubiquitination. Systematic evaluation of these and other candidate ligases will be essential to fully elucidate the upstream regulatory mechanism controlling Itsn1 stability. Addressing this gap represents a key objective for our future research, as it will provide a more complete understanding of the ubiquitin-dependent pathway linking synaptic scaffolding deficiency to neurodevelopmental impairment. Ultimately, these findings advocate for further investigation into targeted modulation of the ubiquitin-proteasome system as a promising therapeutic strategy for synaptic disorders associated with autism spectrum conditions.

Limitation

One notable limitation of our study lies in the interpretation of cross-species transcriptomic data. In identifying differentially expressed genes (DEGs) across rat, monkey, and human organoid models, we applied a uniform statistical threshold to define changes without fully distinguishing the direction of regulation. This approach revealed an important nuance: Dlgap2 was significantly downregulated in rats but upregulated in the other two species following VPA exposure. This divergence suggests a model of convergent vulnerability of synaptic scaffolding pathways that manifests through species-specific regulatory patterns, including the bidirectional effects of VPA on neuronal gene expression observed across models¹.

This observation aligns with existing literature reporting that VPA can paradoxically either exacerbate synaptic deficits or promote synaptogenesis, depending on the experimental or pathological context. Furthermore, studies have documented distinct, even opposing, synaptic responses to VPA across different brain regions within the same animal. Consequently, our findings highlight the necessity for future investigations to move beyond a binary classification of DEGs and to systematically delineate the critical parameters—such as exposure time window, dosage, and brain region-specific circuits—that govern these bidirectional outcomes. A more refined analysis is essential to unravel the precise conditions under which VPA modulates neural circuits toward adaptive or maladaptive plasticity.

Our proteomic analysis identified 241 differentially expressed proteins in the PSD fraction following Dlgap2 knockdown. Among these, only 11 overlapped with the SFARI ASD gene set, which notably included both Dlgap2 and Itsn. This limited yet specific overlap underscores the unique role of the Dlgap2-Itsn1 axis within the broader landscape of ASD-associated synaptic pathways, reinforcing its relevance to human disease mechanisms despite the polygenic complexity of the disorder.

Notwithstanding these complexities, VPA-induced dysregulation of Dlgap2 appears to be a convergent pathway in the emergence of autism-like phenotypes across species. Fully elucidating its function, especially the Dlgap2-Itsn1 axis, is therefore imperative.

Methods

Ethical approval

All animal experiments were conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee at Xiamen University under approval number XMULAC20190144. All methods were performed in accordance with the local relevant guidelines and regulations, including animal ethics established by Institutional Animal Care and Use Committee at Xiamen University. All methods involving animal experimentation are reported in agreement with ARRIVE guidelines (https://arriveguidelines.org).

Animal models and VPA exposure

The mice were reared on a 12/12 light/dark cycle. Pregnant C57BL/6J mice were intraperitoneally injected with valproic acid (VPA; 500 mg/kg) or saline control on embryonic day 12.5 (E12.5). Offspring were weaned on postnatal day 21 and group-housed under standard conditions. For behavioral studies, 1.5-month-old male mice were used. Sample sizes were as follows: saline group, n = 16; VPA group, n = 9 for Supple Fig. 2; AAV-shCon, n = 10; AAV-shDG2, n = 10 for Fig. 4. For stereotactic injection, newborn mice were anesthetized with isoflurane (R510-22, RWD, China) (2% for induction and 1% for maintenance). Subsequently, the mice were euthanized by an isoflurane overdose. Brain tissues were then rapidly dissected on ice and immediately flash-frozen on dry ice for subsequent biochemical analyses. To establish a stable behavioral baseline, all experiments were conducted using age-matched male mice, thereby excluding the influence of female estrous cycle fluctuations.

Behavioral tests

Morris water maze was conducted using a computerized system (Clever Sys Inc.). The water maze comprised a circular tank with a diameter of 120 cm, filled with tap water at a temperature of 20 ± 2 ℃. Spatial reference cues in the form of different shapes were placed along the walls of the tank. A camera was mounted above the maze to record the swimming traces. During the acquisition trials, mice were placed into the maze at one of four points (N, S, E, W) facing towards the wall. The platform was submerged 1–2 cm below the water surface, allowing mice to search for it within 60 s. If they failed to find it, they were guided to and kept on the platform for 10 s. Two trials were conducted each day with an intermission of at least one hour between them. Escape latency indicating spatial memory acquisition, was recorded for each trial. On day 6–7, the platform was removed and a probe test was conducted. The percentage time spent in each of the four quadrants as well as the number of target (platform) area crossings, mean speed and total distance were recorded. Data and video were recorded, using automated software (Clever Sys Inc.).

Three-chamber sociability test apparatus consisted of three chambers: a middle chamber (40 × 20 cm²) and two other chambers (40 × 20 cm²) in which wire cages containing stranger mice were placed. For the sociability test, the mouse was first placed in the center chamber and allowed to acclimatize to the entire three-chamber apparatus for 5 min with free movement. Then, a wire cage containing a novel juvenile male mouse was placed in one of the side chambers, and an empty wire cage was placed in the other. The test animal was placed in the middle chamber and allowed to acclimatize to the entire three-chamber apparatus for 10 min with free movement. Lastly, the two side chambers were equipped with identical wire cages, one housing the familiar mouse (from the social preference test) and the other housing a novel, unfamiliar mouse. The test animal was placed in the middle chamber and allowed to acclimatize to the entire three-chamber apparatus for 10 min with free movement. The time spent in each chamber and the time spent in the interaction zone (defined as a 2-cm perimeter surrounding each wire cage) were analyzed using SMART v.3.0 software.

RT-PCR (Real-time PCR)

Total RNA was extracted from the tissues using Trizol (Invitrogen, Carlsbad, CA, USA) according to the manual instruction. Approximately 100 mg of tissue was directly homogenized in a 1.5 ml microcentrifuge tube with Trizol reagent using a homogenizer to create a uniform lysate for subsequent RNA extraction and rested horizontally for 15 min. The mixture was centrifuged for 10 min at 12,000 × g at 4 °C, and then the supernatant was transferred into a new 1.5 ml microcentrifuge tube with 0.2 ml chloroform/isoamyl alcohol (5:1). The mixture was vigorously shaken for 15 s, and centrifuged at 12,000 × g for 10 min at 4 °C. After centrifugation, the upper aqueous phase containing RNA was transferred into a new tube with an equal volume of isopropyl alcohol supernatant and then centrifuged again at 12,000 × g for another 20 min at 4 °C. After carefully removing the supernatant, the RNA pellet was washed three times with 1 ml of 75% ethanol, then the mixture was centrifuged at 12,000 × g for 3 min at 4℃ to collect residual ethanol, followed by the pellet air drying for 5–10 min in the biosafety cabinet. Finally, DEPC-treated water (25 µl ~ 40 µl) was added to dissolve the RNA. Subsequently, total RNA was qualified and quantified using a NanoDrop and Agilent 2100 bioanalyzer (Thermo Fisher Scientific, MA, USA). Total RNA was reverse-transcribed in a reaction volume of 10 µl using ReverTra Ace^® qPCR RT Kit (FSQ-101). cDNA was amplified by real-time quantitative RT-PCR using SYBR Green (Roche) reagent. Samples were assayed in triplicate and β-actin was used as internal control. The primer sequences are as follows:

Dlgap2-PP1: AAGCATTGCTGCATCTTACCA;

Dlgap2-PP2: GTGCGGGTGAGTACCTCTCT;

β-actin-PP1: GGCTGTATTCCCCTCCATCG.

β-actin-PP1: CCAGTTGGTAACAATGCCATGT.

Tissue preparation and sectioning

Male mice were deeply anesthetized with sevoflurane and transcardially perfused with saline followed by 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer. Brains were promptly dissected and post-fixed in the same fixative for 24 h at 4 °C. Subsequently, tissues were cryoprotected by sequential immersion in 20%, 25%, and 30% sucrose solutions until sinking, embedded in OCT compound (SAKURA, 4583), and coronally sectioned at a thickness of 15 μm using a Leica cryostat. The resulting sections were mounted on glass slides and stored at -20 °C for long-term preservation.

Immunofluorescence staining

Mouse brain sections or cultured cells were fixed in 4% PFA and washed three times with phosphate-buffered saline (PBS). For brain sections, antigen retrieval was performed using citrate buffer (pH 7.0). All samples were then permeabilized and blocked in PBS containing 0.5% Triton X-100 and 10% normal goat serum for one hour at room temperature. Subsequently, samples were incubated with primary antibodies diluted in blocking buffer overnight at 4 °C. The following primary antibodies were used: anti-NeuN (1:1000, abcam, ab177487), anti-PSD95 (1:100, CST, #3450), anti-Dlgap2 (1:100, Bioss, bs-12139R). After washing, samples were probed with corresponding Alexa Fluor-conjugated secondary antibodies (Invitrogen) for one hour at room temperature. Following a final wash, nuclei were stained with DAPI.

Image acquisition and analysis

Images were acquired using a Zeiss LSM 880 laser scanning confocal microscope equipped with Airyscan and analyzed with ImageJ software. Confocal images were acquired as Z-stacks spanning the entire thickness of the stained structures. The maximum intensity projections of these Z-stacks are presented in the figures to represent the overall fluorescence distribution.

For quantitative fluorescence intensity analysis using ImageJ, maximum intensity projections were first generated. Regions of interest (ROIs) were then manually delineated around relevant cellular structures, such as dendritic segments. Within these ROIs, we measured both the mean fluorescence intensity and the density of postsynaptic puncta, with background signal (assessed from an adjacent non-fluorescent region) systematically subtracted for each imaging channel.

Western blotting

For analysis of cultured cells and mouse brain tissues, samples were homogenized in RIPA lysis buffer on ice for 40 min (following an initial mechanical disruption using a homogenizer for brain tissue, with intermittent vortexing every 5 min during incubation). Homogenates were centrifuged at 12,000 rpm for 10 min at 4 °C. The supernatant was collected, and total protein concentration was determined; for synaptosome-related experiments, protein was measured from the initial whole-brain lysate to ensure consistency. Equal amounts of protein were separated on 8–15% SDS-PAGE gels and transferred to Immobilon-P PVDF membranes (Millipore). Membranes were blocked with 5% non-fat milk and incubated with primary antibodies overnight at 4 °C. The following primary antibodies were used: anti-Dlgap2 (1:1,000, GeneTex, GTX124115), anti-Itsn1 (1:1,000, Santa Cruz, sc-136242), anti-PSD95 (1:1,000, CST, #3450), anti-Homer1 (1:500, Santa Cruz, sc-17842), anti-α-tubulin (1:1,000, Abclonal, A6830), anti-ubiquitin K48 (1:1,000, MCE, YA6171) and anti-ubiquitin (1:1,000, Selleck, F0109). After three washes with 1×TBST, membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies (diluted 1:5,000 for target proteins and 1:10,000 for α-tubulin) for 1 h at room temperature. Following final washes, immune-detected proteins were visualized using enhanced chemiluminescence (ECL) in a traditional film-based darkroom system. Band intensities were quantified by densitometric analysis using ImageJ software (NIH, Bethesda). Target protein levels were normalized to α-tubulin, and results were expressed relative to the control group (set to 1). Specific brain regions, animal ages, and sexes used in each experiment are detailed in the corresponding figure legends. All Western blot data in this study were derived from mouse tissues.

Primary cortical neurons culture

Primary cortical neurons maintained in culture were obtained from E16.5 mouse embryos from timed-pregnant females. Briefly, brain cortices were dissected from pups of C57BL6/J, where the meninges were removed from the cortical tissue and dissociated by enzymatic digestion. Isolated primary neurons were plated on poly-D-lysine-coated dishes and cultured in Neurobasal medium (Gibco, 10888022) supplemented with B27 (Gibco, A3582801), GlutaMAX™100x (Gibco, 35050061) and 1% penicillin/ streptomycin (Gibco, 15140163), and maintained in a 5% CO2 incubator at 37 ℃. The cell plating densities have been specified for each culture format: cells were plated at a density of 1 × 10⁷ cells per 10-cm dish, 1.2 × 10⁶ cells per well in 6-well plates, and 1 × 10⁵ cells per well in 24-well plates. The media was changed every three days. For pharmacological treatment, cultured neurons were exposed to 2 µM VPA for 24 h on day 9 in vitro (DIV 9). For Western blot analysis, primary neurons were rapidly lysed to assess the protein levels.

Neurons were transfected on day 3 in vitro (DIV 3) using the Lipofectamine 3000 reagent according to the manufacturer’s instructions. Briefly, the pCAG-tdTomato (Addgene, #83029) plasmid DNA was complexed with the Lipofectamine 3000 reagent in Opti-MEM medium and added dropwise to the neuronal cultures. The transfection mixture was replaced with fresh, pre-warmed conditioned medium after an incubation period of 120 min. Neurons were typically fixed for immunocytochemistry on DIV 9.

Lentivirus and adeno-associated virus

All the viral vectors were packaged by BrainVTA (BrainVTA, Co., Ltd., Wuhan, China). The lentiviral transfer plasmid containing the target gene and packaging plasmids were constructed and purified. 293T cells, selected for high transfection efficiency, were cultured, passaged, and transfected using a plasmid-transfection reagent complex in Opti-MEM medium. Viral supernatants were collected at 48 and 72 h post-transfection, clarified by centrifugation and filtration, and concentrated via ultracentrifugation. Viral titer was determined by quantitative PCR targeting the WPRE sequence in infected cell genomes and calculated based on integrated vector copies relative to a reference gene. The following is the information of the interfering fragment:

shDlgap2-1: gcactacagctcacactatga;

shDlgap2-2: gagctccctgaacttggacaa;

shDlgap2-3: gatggagagagaagcagagga.

AAV: these expression vectors contain a transgene expression cassette that carries hSyn promoter, EGFP, and shDlgap2-2 which were transfected using the calcium phosphate method. Recombinant AAV was harvested at the post transfection time of 60 h. The AAV serotypes 2/9 were purified using anion exchange chromatography, followed by an iodixanol purification procedure. The AAV purification was completed and used to stereotactic injection.

Stereotactic injection

P0 pups were cryo-anesthetized and injected with AAV-shDlgap2 or AAV-scramble (1 µL per ventricle) into the lateral ventricles using a stereotactic frame. Behavioral tests were conducted at 1.5 months of age as described before.

Preparation of PSD fractions

PSD fractions were prepared as previously described^24,25. Briefly, shDlgap2 knockdown and scramble RNA-treated primary cortical neurons (DIV 9) were collected in cold HEPES-buffered sucrose (HBS = 0.32 M sucrose and 25 mM HEPES, pH = 7.4). The homogenates were centrifuged at 500 × g for 5 min to separate the supernatant (S1) from the nuclei and large debris fraction. The S1 fraction was centrifuged at 10,000 × g for 12 min to separate the supernatant (S2, light membrane and cytosolic fraction) and the pellet (P2, crude synaptosomal fraction). The P2 fraction was re-suspended in cold HBS buffer (25 mM HEPES, pH = 7.4, and 150 mM NaCl) to obtain the synaptosomal fraction. The PSD fraction was prepared by solubilizing the synaptosomal fraction in 1% Triton HBS buffer and centrifuging at 10,000 × g for 20 min. Solubilized pellets were re-dissolved in 3% SDS-containing HBS buffer and used for subsequent analyses. Subsequent mass spectrometry analysis was conducted on the PSD fraction.

Purification of synaptoneurosomes

Synaptoneurosomes and total mouse forebrain samples were prepared as previously described²². Briefly, the forebrain was isolated and rapidly cooled to 4 °C. washed in ice-cooled sucrose buffer (320 mM sucrose, 5 mM HEPES, pH 7.4), and homogenized using a Teflon-glass tissue grinder with a motor-driven pestle while maintaining sample cooling. To isolate synaptoneurosomes, each forebrain homogenate was centrifuged at 1200 × g for 10 min. Two milliliters of the supernatant were loaded onto discontinuous Percoll gradients (3%, 10%, and 23% Percoll in sucrose buffer) and centrifuged at 31,000 × g for 5 min. The fractions between the interfaces of the 3–10% and 10–23% layers were collected and further centrifuged at 20,000 × g to pellet synaptoneurosomes. All centrifugation steps were performed at 4 °C. All solutions were supplemented with a complete protease and phosphatase inhibitor cocktail (Roche). Subsequent mass spectrometry analysis was conducted on the synaptoneurosomes.

Co-IP

Mice were euthanized by carbon dioxide asphyxiation, and fresh brain tissues were rapidly collected. Tissue samples were homogenized in ice-cold IP lysis buffer and incubated on ice for 1 h with gentle resuspension every 10 min. The lysates were centrifuged at 12,000 × g for 15 min at 4 °C, and the supernatants were collected. For each immunoprecipitation, 5 µg of anti-Dlgap2 or anti-Itsn1 antibody was added to the lysate and incubated with rotation overnight at 4 °C. The following day, protein A/G magnetic beads were added and incubated for 2 h at 4 °C. The beads were then washed three times with 1 mL of ice-cold IP buffer, and bound proteins were eluted by boiling in 1× SDS loading buffer for 10 min. Eluted proteins were analyzed by immunoblotting or mass spectrometry. For immunoblotting analysis, primary antibodies raised in species different from the immunoprecipitation antibody were selected to avoid cross-reactivity.

Liquid chromatography-tandem mass spectrometry (LC-MS)

Synaptoneurosomal lysate was performed and the LC-MS experiment was conducted in double-blind conditions. Briefly, please refer to the Co-IP procedure mentioned above, where Protein G-bound immune complexes were analyzed by SDS-PAGE/Immunoblotting. The gel was cut into small pieces with a size of 1.5 mm and placed in centrifuge tubes (volume ≤ 300 μL in each tube). The gel was washed with ultra-pure water (DDW) for more than three times and then decolorized with a decolorizing solution [40% acetonitrile (ACN), 50 mm ammonium bicarbonate (ABC) in H2O] until it became colorless. The sample was finally stored in the decolorizing solution. The fractions were dried for LC-MS/MS analysis using a Bruker timsTOF Pro instrument. The resulting raw files were imported into MaxQuant software for data interpretation and protein identification against the database. This experiment received significant support from the Core Facility of Biomedical Sciences at Xiamen University.

Proteomics

Proteomic profiling of PSD fraction was performed by BGI (Shenzhen, China). PSD fraction from primary cortical neurons were lysed, digested with trypsin and analyzed by Astral on an Easy nLC1000 system coupled to a Q Exactive mass spectrometer. Data were processed in MaxQuant against the Mus musculus UniProt database with FDR < 1%. Statistical and functional analyses, including t-tests, hierarchical clustering, and GO/KEGG enrichment based on Fisher’s exact test, were conducted using Perseus and OmicShare Tools.

Cycloheximide chase assays

Protein degradation was assessed using a CHX chase assay (Cell Signaling Technology, #2112). shDlgap2 interfering primary cortical neurons were treated with cycloheximide (150 ng/µl) and MG132 (10 µM) to block de novo protein synthesis and inhibited the ubiquitination-mediated degradation. Primary neurons were collected at 6 h post-treatment. Protein levels were analyzed by Western blotting.

Bioinformatic analysis

All data represent mean ± SEM. Statistical analyses were performed using GraphPad Prism 9 statistical software. All sequencing and bioinformatics analysis were performed using the Omicsmart online platform (http://www.omicsmart.com). Protein-protein interaction networks were constructed using STRING and visualized in Cytoscape. GO and KEGG enrichment analyses were performed using DAVID^40,41, with the entire set of proteins identified in the specific fraction (e.g., synaptoneurosome or PSD) serving as the custom background. Single-cell data were retrieved from Tabula Muris. The schematic illustration was drawn using BioRender software.

Prior to statistical analysis, all quantitative datasets were assessed for normality using the Shapiro-Wilk test. Parametric tests were applied only to data that conformed to the assumption of normality (all p > 0.05 in the Shapiro-Wilk test); otherwise, non-parametric alternatives were used (e.g., Mann-Whitney U test for two-group comparisons, Kruskal-Wallis test followed by Dunn’s post hoc test for multi-group comparisons). For comparisons involving two groups under a single condition, an unpaired two-tailed Student’s T-test was used. For comparisons among more than two groups or across multiple factors, one-way or two-way ANOVA was employed, followed by appropriate post hoc tests where applicable. The specific statistical test used for each experiment is detailed in the corresponding figure legend. To ensure transparency, we report both the exact p values and the sample size (n) for each group.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

**X.G.** and **L.Z.** contributed equally to this work; **K.Z.** conceptualized the study; **X.G.** prepared and maintained the mice; **K.Z.** and **X.G.** designed and performed morphological analysis and biochemical assays; **L.Z.** constructed related plasmids; **X.G.** and **L.Z.** performed experiments on the isolation of synaptoneurosomes, postsynaptic densities (PSD), and immunopurification; **K.Z.** reanalyzed the data obtained from the public database; **K.Z.** conducted on data visualization; **L.Z.** performed behavior tests; **K.Z.** wrote the manuscript; **K.Z.** supervised the project and acquired funding; All authors reviewed and gave final approval to the manuscript.

Funding

This work was supported by National Natural Science Foundation of China grant 82101242 and 32371505 (K.Z.); China postdoctoral science foundation grant 2020M682094 (K.Z.).

Data availability

Reagents generated in our laboratory in this study or previous studies are available upon request. Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Kai Zhuang (xmuzhuangkai@xmu.edu.cn).

Competing interests

The authors declare no competing interests.