‌Dual-target hUCMSCs/EVs therapy for autism spectrum disorder: remodeling gut microbiota and modulating neuroimmune crosstalk in a valproic acid-induced C57BL/6 mice model Authors: Caixia Wu, Xianjie Li, Xiaoya Yang, Han Wang, Guanzhen Lin, Zhaoming Liu Published: 2025 ABSTRACT IntroductionAutism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by deficits in social communication and repetitive behaviors. Accumulating evidence implicates neuroimmune dysregulation and gut microbiota dysbiosis in its pathogenesis, yet effective therapies targeting these mechanisms are lacking. This study investigated the therapeutic efficacy of human umbilical cord-derived mesenchymal stem cells (hUCMSCs) and their extracellular vesicles (EVs), both alone and in combination, in a valproic acid (VPA)-induced ASD mouse model.MethodsVPA-exposed C57BL/6 mice were randomized into five groups: control, VPA, hUCMSCs alone, EVs alone, and hUCMSCs + EVs combination. Behavioral tests, biochemical analyses, 16 S rRNA sequencing, immunofluorescence, and transmission electron microscopy (TEM) were performedResultsWhile both hUCMSCs and EVs alone showed some beneficial effects on certain ASD-like symptoms, each exhibited limited efficacy in achieving comprehensive remediation. In contrast, the combined hUCMSCs + EVs therapy yielded the most robust improvements across multiple domains, including social interaction and repetitive behaviors. Furthermore, the combination therapy synergistically normalized neuroinflammatory cytokine levels and oxidative stress, restored synaptic and mitochondrial ultrastructure in key brain regions, and promoted gut microbiota homeostasis by enriching beneficial bacteria such as Lactobacillus and reducing pathogens.DiscussionThese results highlight that although individual treatments offer partial relief, only the combined strategy fully restores neuroimmune–microbiota homeostasis, demonstrating a complementary and synergistic therapeutic effect. This study establishes a novel dual-target approach leveraging systemic hUCMSCs and CNS-targeted EVs, providing a promising translational strategy for ASD through orchestrated regulation of the neuro-immune-microbiota axis.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13287-025-04829-x. Introduction Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by persistent deficits in social communication, restricted interests, and repetitive behaviors, affecting approximately 2.2% of children globally [6, 10, 11]. Its multifactorial pathogenesis involves intricate interactions between genetic susceptibility, neuroinflammation, oxidative stress, and environmental triggers [17, 22].Recent reviews highlight persistent limitations in monoamine-targeting pharmacotherapies (Ramit [38], while emerging evidence implicates dysregulated ERK signaling in synaptic pathology[49]. Notably, neuroimmune dysregulation has emerged as a central pathological feature, with post-mortem studies revealing microglial activation, elevated pro-inflammatory cytokines (IL-1β, IL-6, TNF-α), and oxidative damage in the prefrontal cortex and hippocampus of ASD patients [28, 45]. Concurrently, clinical and preclinical evidence highlights gut-brain axis disruption, where dysbiosis of the gut microbiota—characterized by elevated Firmicutes/Bacteroidetes ratios, reduced SCFA-producing bacteria (e.g., Ruminococcaceae), and increased pro-inflammatory taxa (e.g., Enterobacteriaceae)—exacerbates neuroinflammation via vagal afferents and systemic cytokine release [13, 52]. Importantly, sex-specific gut microbiota alterations have been shown to critically modulate neuroinflammatory responses in neurological disorders, highlighting the complexity of gut-brain crosstalk [31]. Despite decades of research, effective therapies remain scarce. Pharmacological interventions targeting neurotransmitters show limited efficacy [39], while behavioral therapies address symptoms but not underlying biology[7]. Conventional interventions fail to address core pathophysiological processes, including mitochondrial dysfunction [27] and myelin disruption [43]. Neuroprotective agents like Tanshinone-IIA show promise in modulating MAPK pathways [44], and magnesium demonstrates essential roles in neuronal bioenergetics[20], yet their clinical translation remains limited.This therapeutic gap underscores the need for strategies targeting ASD’s core pathophysiological networks. While natural compounds like α-Mangostin demonstrate neuroprotection in rodent ASD models [48] [50]), and molecular approaches show preclinical potential [1], these strategies lack comprehensive modulation of the gut-brain-immune axis. Notably, specific probiotic interventions (e.g., Lactobacillus helveticus) can remodel gut-brain mediators to ameliorate neurological deficits, providing proof-of-concept for microbiota-targeted therapies [32].This gap motivates our investigation into dual-target stem cell/EVs therapy. Cell-based therapies, particularly MSCs, have shown promise in preclinical and clinical ASD research[15, 16, 21, 36]. Umbilical cord-derived MSCs (UC-MSCs) exhibit superior immunomodulatory and neurotrophic properties compared to adult-derived MSCs, making them ideal for neurodevelopmental disorder treatment[4, 9]. In a Phase I clinical trial, UC-MSC transplantation improved social interaction and reduced repetitive behaviors in children with ASD, with no severe adverse events. EVs derived from MSCs (MSC-EVs) further enhance therapeutic efficacy by crossing the blood-brain barrier (BBB) and delivering bioactive cargo to CNS targets[26, 35]. Intranasal administration of MSC-EVs in a Shank3B knockout mouse model restored social behavior and reduced repetitive grooming, associated with increased GABAB1 receptor expression in the prefrontal cortex[35]. Beyond their immunomodulatory capacity, hUCMSCs exert multifaceted protective effects in neurological disorders: (1) Mitochondrial protection: Transfer of functional mitochondria to damaged neurons rescues bioenergetic deficits [57];(2) Oxidative stress mitigation: Upregulation of endogenous antioxidants (SOD, GSH) counteracts ROS-mediated neuronal damage[56];(3) Blood-brain barrier preservation: TSP-1 secretion enhances endothelial tight junction integrity [19]. These mechanisms position hUCMSCs as ideal candidates for ASD intervention, where mitochondrial dysfunction and oxidative stress are established pathophysiological features [53]. Recent advances in regenerative medicine highlight the synergistic potential of combining cellular and acellular therapies. Human umbilical cord-derived mesenchymal stem cells (hUCMSCs) demonstrate potent immunomodulatory properties through paracrine signaling and direct cell-contact mechanisms [54], while their derived exosomes (Exos) offer targeted neurotrophic support by traversing biological barriers such as the blood-brain barrier (BBB) [55]. Critically, the dual-target strategy leveraging hUCMSCs (intravenous) and EVs (intranasal) represents an innovative approach to concurrently address peripheral inflammation and central nervous system pathology—a paradigm distinct from monotherapies that target singular pathways[24]. This combinatorial regimen capitalizes on the complementary actions: MSCs systemically remodel gut microbiota and enhance gut barrier integrity, while EVs deliver neuroprotective cargo directly to synaptic sites, establishing a coordinated ‘neuro-immune-microbiota’ regulatory axis. Despite these promising developments, several challenges remain. First, most studies on MSCs and EVs have focused on their individual effects rather than their combined potential. Second, while the role of the gut-brain axis in ASD has been explored, there is limited understanding of how interventions targeting this axis can be integrated with neuroprotective strategies to achieve comprehensive therapeutic benefits. Third, the mechanisms underlying the interaction between MSCs/EVs and the gut-brain axis remain poorly defined, particularly in the context of ASD. Building on the dual-target rationale, we hypothesize that combined hUCMSCs and EVs will synergistically attenuate ASD-like phenotypes by concurrently restoring gut-brain axis homeostasis and suppressing neuroinflammation in a VPA-induced mouse model. Specific aims include: (1)Assess behavioral improvements using standardized tests (open field, Morris water maze).༈2༉Quantify neuroinflammatory cytokines (IL-1β, IL-6, TNF-α, IL-10) and oxidative stress markers (GSH, MDA) in the prefrontal cortex.༈3༉Characterize gut microbiota restructuring via 16 S rRNA sequencing and correlate taxa shifts (e.g., Firmicutes/Bacteroidetes ratio) with neural outcomes.༈4༉Validate synaptic ultrastructural restoration using TEM. To overcome these challenges, we pioneer a dual-target, non-invasive combinatorial therapy using human umbilical cord MSCs (hUCMSCs) and their EVs. Our approach innovates by: Simultaneously modulating neuroinflammation and gut dysbiosis: Unlike single-modality interventions (e.g., UCB alone in [46][46], we leverage the complementary actions of hUCMSCs (systemic immunomodulation) and EVs (CNS-targeted synaptic repair via intranasal delivery). Optimizing biodistribution: Intravenous hUCMSCs suppress peripheral inflammation and enhance gut barrier integrity, while intranasal EVs bypass the BBB to directly target microglial activation and synaptic dysfunction. Integrating multi-omics validation: We unify behavioral, 16 S rRNA sequencing, cytokine profiling, and transmission electron microscopy (TEM) to establish causal links between microbial shifts (e.g., Ruminococcaceae↑), neuroimmune markers (IL-10↑; TNF-α↓), and phenotypic recovery. This study bridges stem cell biology, neuroimmunology, and microbiome research to address ASD’s complexity. By demonstrating that hUCMSCs + EVs: Normalize excitatory-inhibitory balance via synaptic vesicle density restoration; Rebalance microbial networks to favor anti-inflammatory SCFA producers (Ruminococcaceae↑), and Suppress microglial activation via Iba1 downregulation. we establish a blueprint for multi-target therapies in neurodevelopmental disorders. The non-invasive delivery protocol (IV + intranasal) further enhances clinical feasibility, offering a scalable strategy beyond invasive intrathecal approaches [42]. Materials and methods Experimental design and animal procedures All experimental protocols were approved by the Institutional Animal Care and Use Committee (IACUC) of Guangzhou Institute of Biomedicine and Health (Approval No. IACUC:2023081) and conducted in accordance with AAALAC-accredited SPF facility standards. Animal housing and breeding Twenty-four female and twelve male C57BL/6 mice (initial body weight 20–25 g) were maintained under controlled environmental conditions (22 ± 1 °C, 55 ± 5% humidity) with a 12-h light/dark cycle (lights on 07:00–19:00). Animals received autoclaved water and standard rodent chow ad libitum. For timed mating, male and female mice were co-housed at 17:00, with vaginal plug confirmation performed at 09:00 the following morning. Plug-positive females were designated as gestational day (GD) 0.5. Prenatal VPA exposure Twenty-four confirmed pregnant dams underwent stratified randomization into two groups. From GD11.5 to GD13.5 (corresponding to murine peak neocortical neurogenesis), the VPA-exposed group (n = 16 dams) received daily intraperitoneal injections of sodium valproate (300, 400, and 300 mg/kg on GD11.5, 12.5, and 13.5, respectively) dissolved in endotoxin-free saline. Control dams (n = 8) received equivalent volumes of 0.9% NaCl. The sample size of pregnant dams (n = 24) was determined based on pilot studies which indicated that prenatal VPA exposure induces significant maternal toxicity and embryolethality. Specifically, VPA administration resulted in a proportion of dams experiencing mortality, miscarriage, fetal resorption, or stillbirth. Furthermore, litter sizes from the remaining VPA-exposed dams were consistently reduced (typically yielding 3–6 pups) compared to controls. To ensure the generation of a sufficient number of offspring for downstream experimental analyses while adhering to the 3R principles (Replacement, Reduction, Refinement) of animal welfare, a total of 24 confirmed pregnant dams were utilized for model establishment and subsequent interventions. Postnatal group allocation Male offspring underwent cross-fostering to eliminate litter effects and were randomly assigned at postnatal day 21 (P21) to five groups: Control group (n = 12, saline-treated offspring from control dams), VPA group (n = 12, VPA-exposed offspring), and MSC + Evs group (n = 12, VPA-exposed offspring receiving combined human umbilical cord mesenchymal stem cells and EVs), MSC group (n = 12, VPA-exposed offspring receiving human umbilical cord mesenchymal stem cells alone), Evs group(n = 12, VPA-exposed offspring receiving EVs alone). Experimenters remained blinded to group assignments during behavioral assessments and biochemical analyses. Behavioral testing preceded histological evaluations. This study complies with ARRIVE guidelines [33] and was conducted under ethical approval CAS (IACUC:2023081) from Guangzhou Institute of Biomedicine and Health, Chinese Academy of Sciences. Procedures adhered to the Guide for the Care and Use of Laboratory Animals (8th ed., 2011). Euthanasia protocols followed 2020 AVMA Guidelines, excluding prohibited methods (chloral hydrate, ether, chloroform). Analgesic regimens and humane endpoints were implemented to minimize animal distress. Isolation and culture of umbilical cord mesenchymal stem cells (UC-MSCs) Human umbilical cords from full-term cesarean deliveries were collected aseptically in ice-cold saline containing 1% penicillin/streptomycin (HyClone). After repeated saline rinsing to remove blood components, vascular remnants were excised and Wharton’s jelly dissected into 3–4 cm segments. Tissue digestion employed 0.1% collagenase IV (Sigma-Aldrich)/trypsin (1:1) for 1–2 h at 37 °C. The digested suspension was filtered through 100-µm strainers and centrifuged (300 × g, 5 min). Pelleted cells were resuspended in DMEM/F12 (HyClone) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin, then plated in 10-cm culture dishes. Non-adherent cells were removed after 72 h. Primary cultures (P0) at 80–90% confluence were passaged using 0.05% trypsin-EDTA (Thermo Fisher) and expanded at 5,000 cells/cm². Cryopreservation and thawing Cells were cryopreserved in 90% FBS/10% DMSO at 2 × 10⁶ cells/mL using programmed freezing containers (Nalgene). Thawing involved 37 °C water bath immersion, centrifugation (300 ×g, 5 min), and resuspension in complete medium prior to culture. Phenotypic characterization Passage 3 cells were immunophenotyped using BD FACSCanto II flow cytometry (BD Biosciences, San Jose, CA) with antibodies against CD105-PE/Cy7 (1:100, BioLegend 344016), CD90-FITC (1:200, BioLegend 328108), CD73-APC (1:150, BioLegend 323208), and negative markers CD34-PerCP (1:100, BioLegend 343608), CD45-PE (1:200, BioLegend 202208), HLA-DR-APC/Cy7 (1:150, BioLegend 307616). Data analysis used FlowJo v10.8.1 (BD Biosciences). EVs isolation and characterization EVs were isolated from human umbilical cord mesenchymal stem cell (hUC-MSC) conditioned media using differential ultracentrifugation in accordance with MISEV2018 guidelines. Cell culture supernatants underwent sequential centrifugation (Eppendorf 5910 R) under the following conditions: 300 ×g for 10 min (4 °C) to remove residual cells, 2,000 ×g for 10 min (4 °C) to eliminate cellular debris, and 10,000 ×g for 30 min (4 °C) to pellet microvesicles. The clarified supernatant was filtered through 0.22-µm polyethersulfone (PES) membranes (Millipore) before final ultracentrifugation at 100,000 ×g for 2 h (4 °C; Optima XPN-100, Beckman Coulter). Purified EVs pellets were resuspended in sterile phosphate-buffered saline (PBS) for subsequent applications. Combined hUCMSCs and EVs transplantation protocol The MSC + Evs group received combinatorial therapy via three administration routes. Systemic delivery involved weekly intravenous injections of 2 × 106 hUCMSCs (passages 3–5; >95% viability confirmed by Trypan Blue exclusion) suspended in 200 µL PBS through the tail vein for four consecutive weeks. Concurrently, local intranasal administration of 50 µg EVs in 20 µL PBS was performed weekly using a calibrated micropipette tip inserted 2 mm into the nasal cavity. At 4 weeks after the final systemic injection, an intracerebroventricular (ICV) boost was administered under stereotactic guidance (RWD Life Science) with 3% isoflurane anesthesia. Bilateral ICV injections delivered 1 × 105 cells/µL hUCMSCs at 0.4 µL/min (2 µL per ventricle) using the following stereotactic coordinates relative to bregma: anterior-posterior, − 0.3 mm; medial-lateral, ± 0.8 mm; dorsal-ventral, − 1.5 mm from the skull surface. Injection needles remained in place for 10 min post-infusion to minimize backflow. Prophylactic gentamicin (0.3%) was administered for three consecutive postoperative days to prevent infection. The MSC alone group received systemic therapy consisting of weekly intravenous injections of 2 × 10⁶ hUCMSCs (passages 3–5; >95% viability confirmed by Trypan Blue exclusion) suspended in 200 µL PBS via the tail vein for four consecutive weeks.The EVs alone group was treated with local intranasal administration of 50 µg EVs in 20 µL PBS weekly using a calibrated micropipette tip inserted 2 mm into the nasal cavity, also for a total of four weeks.Behavioral assessments commenced 7 days after the final intervention (initiated at postnatal day 50, PND50) to ensure acute procedural effects had subsided. Behavioral assessments All behavioral testing was conducted in a dedicated experimental room under standardized conditions (23 °C, 50 lx illumination), with mice acclimatized for 60 min prior to each assay. Testing commenced at postnatal day 50 (PND50) and proceeded sequentially over 17 days according to the established timeline (Figure S1). Three-chamber social interaction test‌ The social interaction test was performed over a 7-day habituation period. During days 1–4, mice received daily 2-hour exposures to the testing apparatus to establish environmental familiarity. Progressive adaptation occurred on days 5–7, with days 5–6 dedicated to 3-minute central compartment exploration sessions, followed by full arena access on day 7 for 3 min. The experimental protocol consisted of three sequential phases: (1) a 5-minute baseline period with empty wire cages in both lateral chambers; (2) a 10-minute sociability test phase introducing an age- and sex-matched C57BL/6 stranger mouse (Stranger 1) in one chamber versus an empty cage; and (3) a 10-minute social novelty preference phase where a novel mouse (Stranger 2) replaced the empty cage while Stranger 1 remained. To ensure experimental consistency, stranger mice were individually housed for 48 h prior to testing to maintain olfactory consistency. Social interaction was quantified using EthoVision XT 15.0 (Noldus IT, Netherlands), with positive interaction defined as snout proximity < 2 cm from the cage. Self-grooming analysis‌ Individual mice were placed in sanitized polycarbonate cages under 50 lx illumination for a 10-minute habituation period followed by 10-minute video recording sessions using Logitech C920 HD Pro cameras. Three blinded observers analyzed grooming parameters including bout frequency (discrete grooming initiations) and cumulative duration (total time spent in facial/body/genital cleaning behaviors). Inter-rater reliability was maintained through consensus scoring with frame-by-frame video analysis for discrepant observations. Marble-burying evaluation‌ The marble-burying test was conducted in specialized chambers containing 5 cm depth corncob bedding with sixteen 1.6-mm black glass marbles arranged in a 4 × 4 grid (3 cm spacing). After a 3-minute acclimatization period, mice underwent a 10-minute testing session under 50 lx illumination. Burial events were scored when ≥ 75% of the marble surface was covered by bedding material, with final determinations requiring ≥ 2/3 observer agreement. Discrepancies were resolved through frame-by-frame video review. Open field test‌ Exploratory behavior and anxiety-like responses were evaluated in a 40 × 40 × 30 cm square arena with a defined 20 × 20 cm central zone. Following a 10-minute habituation period for free exploration, 10-minute test sessions were recorded and analyzed using EthoVision XT (Noldus Information Technology). Key parameters included central zone entries (all limbs crossing boundary) and vertical activity (forelimb elevation maintained for ≥ 2 s). All behavioral testing was conducted between 09:00–12:00 h to control for circadian influences. Morris water maze Spatial learning and memory were assessed using the Morris water maze during postnatal days 60–67 (PND60-67). Mice were trained in a black circular pool (120 cm diameter) filled with opacified water (23.0 ± 0.5 °C; nontoxic white tempera) containing a submerged platform (8 cm diameter, 1.5 cm depth). During the 5-day acquisition phase, animals underwent four daily trials (90 s maximum/trial) with randomized entry points (cardinal directions) and 25-min inter-trial intervals. Mice failing to locate the platform within 90 s were manually guided to it for 10 s of orientation training. Twenty-four hours after the final training session (Day 6), a 60-s probe trial was conducted with the platform removed. Spatial navigation cues consisted of four high-contrast geometric patterns positioned uniformly around the arena (50 ± 5 lx illumination). Camera calibration was performed using a chessboard pattern to ensure < 2% positional error. All sessions were recorded and analyzed using EthoVision XT v15.0 (Noldus Information Technology). Tissue collection and processing Euthanasia was performed on PND68-70 following behavioral testing completion. Animals received intraperitoneal anesthesia (ketamine 100 mg/kg + xylazine 10 mg/kg) followed by transcardial perfusion with ice-cold 0.9% NaCl (200 mL) and 4% paraformaldehyde (PFA; Merck, Germany; 300 mL) via the ascending aorta. Brains were post-fixed in 4% PFA for 24 h at 4 °C, then dissected to isolate the prefrontal cortex and hippocampus. Tissues were paraffin-embedded using standard protocols (Merck, Germany), and coronal sections. (5 μm thickness) were prepared using a rotary microtome (Microm HM335E, Walldorf, Germany). Biochemical assays Prefrontal cortical homogenates were analyzed for oxidative stress markers and inflammatory cytokines using commercial kits according to manufacturers’ protocols: Oxidative stress: Glutathione (GSH; Nanjing Jiancheng A006-2-1), malondialdehyde (MDA; A003-1–2), total superoxide dismutase (T-SOD; A001-3-2), catalase (CAT; A007-1-1), glutathione peroxidase (GSH-Px; A005-1–2), total nitric oxide synthase (T-NOS; A014-2-1), and nitric oxide (NO; A012-1–2). Cytokines: IL-1β (Fankel EM001), IL-6 (EM002), TNF-α (EM003), and IL-10 (EM004) mouse ELISA kits. Optical density measurements were performed using a SpectraMax M5 microplate reader (Molecular Devices). For immunofluorescence, sections were stained with anti-Iba1 antibody (Sigma-Aldrich SAB2702364; 1:400) and DAPI nuclear counterstain (Beyotime C1005; 1 µg/mL). 16 S rRNA sequencing and bioinformatics analysis Fecal samples were collected aseptically at postnatal day 59 (PND59) between 09:00–11:00 to minimize circadian variation. Defecation was induced via gentle abdominal pressure using sterile finger cots, with expelled pellets immediately transferred into pre-labeled 2 mL tubes (Axygen PCR-02-C, USA) using sterile forceps. Samples were standardized to 3–5 intact pellets (≈ 100 mg), flash-frozen in liquid nitrogen within 5 min, and stored at − 80 °C until processing. Genomic DNA was extracted using mechanical lysis with 0.1 mm zirconium beads (BioSpec 11079101z, USA) in a FastPrep-24 homogenizer (MP Biomedicals), followed by a modified CTAB protocol. DNA purity was verified via NanoDrop 2000 (Thermo Fisher), requiring A260/A280 = 1.8–1.9 and A260/A230 > 2.0. Purified DNA was normalized to 1 ng/µL in ultrapure water (Millipore Milli-Q). The V3-V4 hypervariable regions were amplified using primers 341 F (5’-CCTAYGGGRBGCASCAG-3’) and 806R (5’-GGACTACNNGGGTATCTAAT-3’) with Phusion® High-Fidelity Master Mix (NEB M0531) under the following conditions: 98 °C for 1 min; 30 cycles of 98 °C for 10 s, 50 °C for 30 s, 72 °C for 30 s; and final extension at 72 °C for 5 min. Amplicons were size-verified on 2% agarose gels (1× TAE buffer) and purified using the TIANgel Midi Kit (Tiangen DP209, China). Libraries were constructed with the NEBNext Ultra II DNA Library Prep Kit (Illumina E7645S), with quality control confirming insert sizes of 450–650 bp (Agilent 5400 Bioanalyzer) and concentrations ≥ 20 nM (Qubit 4.0 Fluorometer, Thermo Fisher Q33238). Sequencing was performed on the Illumina NovaSeq 6000 platform (paired-end 250 bp mode; 50,000 reads/sample). Bioinformatic analysis employed QIIME2 (v2023.2) with DADA2 for denoising and the SILVA 138.1 database for taxonomic assignment. α-diversity (Chao1, Shannon, Faith’s PD) and β-diversity (Bray-Curtis, weighted UniFrac with PCoA/NMDS) were calculated via QIIME2 core metrics. Differential taxa abundance was determined using LEfSe (LDA score > 2.0), ANCOM (W > 0.7), and DESeq2 (FDR < 0.05). Functional potential was predicted via PICRUSt2 (v2.4.1) for KEGG pathways, with STAMP implementing Welch’s t-test (FDR-corrected). Co-occurrence networks (Gephi v0.10.1; |ρ| >0.6, P < 0.01) and multivariate analyses (RDA: vegan package; PLS-DA: mixOmics) completed the workflow. Immunofluorescence Staining‌ For immunofluorescence analysis, brain sections were processed through sequential steps to ensure optimal antigen preservation and antibody binding. Following deparaffinization with xylene and graded ethanol series, heat-induced antigen retrieval was performed in citrate buffer (pH 6.0) at 95 °C for 20 min. After cooling to room temperature, sections underwent three 5-minute washes in phosphate-buffered saline (PBS, pH 7.4) to remove residual buffer components. To minimize nonspecific binding, sections were incubated with blocking solution containing 4% (w/v) bovine serum albumin (BSA, Sigma-Aldrich A7906) and 0.3% Triton X-100 (Sigma-Aldrich T9284) in PBS for 1 h at room temperature. Primary antibody incubation was carried out using rabbit anti-Iba1 antibody (1:400 dilution, Sigma-Aldrich SAB2702364) in blocking solution at 4 °C for 16 h in a humidified chamber. After primary antibody incubation, sections were thoroughly washed with PBS containing 0.1% Tween-20 (PBST) through three 10-minute washes to remove unbound antibodies. Secondary staining was performed using Alexa Fluor 488-conjugated goat anti-rabbit IgG (1:500 dilution, HUABIO HA1125) for 1 h at room temperature under light-protected conditions. Nuclear counterstaining was achieved with 4’,6-diamidino-2-phenylindole (DAPI, 1 µg/mL, Sigma-Aldrich D9542) for 10 min in darkness. Confocal imaging was conducted using a TCS SP8 STED 3X system (Leica Microsystems) with standardized acquisition parameters: 12% laser power at 488 nm excitation wavelength, detector gain maintained between 650 and 750 V, and z-stack intervals of 0.5 μm to ensure optimal optical sectioning. Integrated fluorescence density (IntDen) quantification was performed using ImageJ software (v1.53k, NIH) with threshold-optimized regions of interest drawn consistently across all samples. Transmission electron microscopy‌ For ultrastructural analysis, tissue preparation began with terminal anesthesia of postnatal day 70 (PND70) mice using intraperitoneal injection of ketamine/xylazine (75:5 mg/kg). Transcardiac perfusion was performed through the ascending aorta, beginning with a vascular flush using 200 mL ice-cold PBS (0.01 M sodium/potassium phosphate buffer, pH 7.4) followed by primary fixation with 300 mL of mixed aldehyde solution containing 2% paraformaldehyde and 2.5% glutaraldehyde (Sigma-Aldrich G5882) in 0.1 M sodium cacodylate buffer (pH 7.4). Following perfusion, brains were carefully extracted and microdissected to obtain 1 mm³ tissue samples from the prefrontal cortex and dorsal hippocampal CA1 region (n = 3 animals per group, 6 regions per animal). Post-processing protocols included secondary fixation in primary fixative for 20 h at 4 °C, followed by 2 h in 1% osmium tetroxide (OsO₄) with 0.8% potassium ferrocyanide [K₄Fe(CN)₆] in 0.1 M cacodylate buffer at room temperature. Samples were then dehydrated through a graded ethanol series (50% to 100%) with 10-minute intervals at each concentration, followed by transition in propylene oxide. Embedding was performed using EPON™ 812 resin (Electron Microscopy Sciences 14900) with 48-hour polymerization at 60 °C. For imaging, semithin Sect. (1 μm) were first cut and stained with toluidine blue (Sigma-Aldrich 198161) for anatomical orientation. Ultrathin Sect. (60 nm) were collected on 200-mesh copper grids and subjected to double contrast staining: 15 min in uranyl acetate (Electron Microscopy Sciences 22400) followed by 5 min in Reynold’s lead citrate. Imaging was performed using a JEOL JEM-1200EX transmission electron microscope operated at 80 kV acceleration voltage. Digital image capture was achieved using a MORADA CCD camera (Olympus iTEM v5.2) with 3280 × 2464 pixel resolution and 8-bit grayscale depth, ensuring consistent imaging parameters across all samples. Tumor formation assay of hUCMSCs To examine the tumorigenicity of cultured hUCMSCs, 1*107 cells of hUCMSCs were transplanted into the left quadriceps muscle of the thigh or subcutaneously into the abdomen of tenweek old NOD-SCID mice. Six months later, the mice were sacrificed, and the muscles, skin, liver, kidneys, myocardium, Small intestine, brain, spleen and lung were histologically examined. A graphical abstract summarizing the experimental rationale, therapeutic mechanisms, and key findings was created using BioRender.com and included as Figure S2 in supplementary materials. Statistical analysis All quantitative data are expressed as mean ± standard deviation (SD). Between-group comparisons were performed using one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test for multiple comparisons. Normality was confirmed via Shapiro-Wilk tests (p > 0.05), and homogeneity of variance was verified with Levene’s test. Statistical significance was defined as P < 0.05. All analyses utilized GraphPad Prism v9.4.1. Results Characterization of EVs and hUCMSCs, and tumor formation assay hUCMSCs exhibited typical morphology (Fig. 1I). Multilineage differentiation potential was confirmed by Toluidine Blue staining (chondrocytes, Fig. 1K), Oil red O staining (adipocytes, Fig. 1L), and Alizarin red S staining (ARS; osteocytes, Fig. 1J). Flow cytometry analysis identified the characteristic expression profile of mesenchymal stem cell markers. Cells derived from umbilical cord showed minimal expression of hematopoietic progenitor markers CD45 (4.15%; Fig. 1D), CD34 (0.12%; Fig. 1E), and HLA-DR (0.15%; Fig. 1F). In contrast, they exhibited high expression of MSC markers CD73 (99.8%; Fig. 1A), CD90 (99.9%; Fig. 1B), and CD105 (99.9%; Fig. 1C). These results indicate negligible contamination by hematopoietic cells. G-band chromosomal analysis confirmed that hUCMSCs possessed normal karyotypes (Fig. 1H). hUCMSCs were cultured, and EVs were isolated from the supernatant via differential ultracentrifugation. Following isolation of sufficient EVs, they were characterized and analyzed. Transmission electron microscopy (TEM) revealed that EVs derived from hUCMSCs displayed typical cup-shaped structures with diameters ranging from approximately 30 to 150 nm (Fig. 1G). To assess the tumorigenicity of cultured hUCMSCs, 1 × 10⁷ cells were transplanted into the left quadriceps muscle of six-week-old NOD-SCID mice. Histological analysis performed 6 months post-transplantation detected no tumor formation in the myocardium, liver, spleen, lung, kidney, brain, stomach, small intestine, pancreas, or muscle (Fig. 1M–V). Fig. 1Characterization of EVs and hUCMSCs as well as tumor formation assay. A–F Surface markers profiling of hUCMSC. The cells were highly positive for CD105, CD90 and CD73 and were negative for HLA-DR, CD34 and CD45. Control cells (blue), Sample cells (red). G By TEM, purified hUCMSC-Exo exhibit cup-like morphologies. H Chromosomal analysis judged by G-band test revealed that the hUCMSCs had normal karyotypes, 46XY. I Morphology of hUCMSC. Scale bar = 100 μm. J–L Multiple differentiation potential of hUCMSC. Scale bar = 200–100 μm. M–V Histological analysis of Myocardium, liver, spleen, lung, kidney, brain, stomach, small intestine, pancreas, muscle. Scale bar = 50 μm Characterization of EVs and hUCMSCs as well as tumor formation assay. A–F Surface markers profiling of hUCMSC. The cells were highly positive for CD105, CD90 and CD73 and were negative for HLA-DR, CD34 and CD45. Control cells (blue), Sample cells (red). G By TEM, purified hUCMSC-Exo exhibit cup-like morphologies. H Chromosomal analysis judged by G-band test revealed that the hUCMSCs had normal karyotypes, 46XY. I Morphology of hUCMSC. Scale bar = 100 μm. J–L Multiple differentiation potential of hUCMSC. Scale bar = 200–100 μm. M–V Histological analysis of Myocardium, liver, spleen, lung, kidney, brain, stomach, small intestine, pancreas, muscle. Scale bar = 50 μm Open field test analysis To assess spontaneous exploratory behaviors and anxiety-like responses in novel environments, we performed the open field test. As shown in Fig. 2, valproic acid (VPA) exposure significantly suppressed exploration and increased anxiety-like behaviors in C57BL/6 mice, while combined hUC-MSC/EVs (hUCMSCs + EVs) intervention mitigated these deficits. Quantitative analysis demonstrated that mice exposed to VPA exhibited a significantly lower number of central zone crossings compared to the control group (P < 0.0001; Fig. 2A). Administration of hUCMSCs alone significantly increased the number of central zone crossings relative to the VPA group (P < 0.01). Similarly, treatment with EVs alone and the combination of hUCMSCs and EVs resulted in a significant increase in central crossings (P < 0.0001).Similarly, treatment with EVs alone resulted in a significant increase in central zone entries (P < 0.0001). Similarly, the total movement distance within the central zone was significantly shorter in VPA-exposed mice compared to control animals (P < 0.0001; Fig. 2B). Administration of hUCMSCs alone significantly increased the central movement distance relative to the VPA group (P < 0.01). Likewise, treatment with EVs alone also led to a notable increase in locomotor activity within the central zone (P < 0.001). Most remarkably, the combination of hUCMSCs and EVs produced the strongest restorative effect, markedly reversing the deficit and substantially increasing the central movement distance compared to the VPA group (P < 0.0001). Anxiety-related metrics demonstrated parallel alterations: Consistent with these findings, VPA exposure significantly reduced the duration spent in the central zone compared to the control group (P < 0.0001; Fig. 2C). Treatment with hUCMSCs alone significantly increased the center time compared to the VPA group (P < 0.01). Similarly, administration of EVs alone also led to a pronounced prolongation in the central zone duration (P < 0.001). Notably, the combination of hUCMSCs and EVs produced the most substantial effect, markedly prolonging the central time relative to the VPA group (P < 0.0001), suggesting a synergistic therapeutic benefit. Vertical exploration activity was significantly reduced in the VPA group compared to the control animals (P < 0.001; Fig. 2F). Administration of hUCMSCs alone resulted in a slight increase in vertical exploration, although this difference did not reach statistical significance compared to the VPA group (P > 0.05). Similarly, treatment with EVs alone also showed an increasing trend, but again, the difference was not statistically significant (P > 0.05). In contrast, the combined treatment with hUCMSCs and EVs produced a statistically significant improvement in vertical exploration scores relative to the VPA group (P < 0.05). Representative movement trajectories (Figs. 2G–K) corroborated these findings. VPA-exposed mice displayed restricted exploration with peripheral thigmotaxis (Fig. 2H), whereas the hUCMSCs + EVs group, hUCMSCs alone group and EVs alone group (Fig. 2I–K) exhibited increased center exploration resembling control patterns (Fig. 2G). Fig. 2Behavioral analysis of C57BL/6 mice in the open field and repetitive stereotypic behaviors. A Cross-center grid; B Inner area distance; C Inner area time; D Self-grooming; E Buried marbles; F Vertical score; G Representative open field trajectory map of the control group; H Representative open field trajectory diagram of the ASD model; I Representative open field trajectory maps of the hUCMSCs + EVs stem cell treatment group; J Representative open field trajectory maps of the hUCMSCs treatment grou. K Representative open field trajectory maps of the EVs treatment group. The data are presented as the mean ± SD (x̅±s, n = 12). Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, compared with the VPA-exposed ASD model group Behavioral analysis of C57BL/6 mice in the open field and repetitive stereotypic behaviors. A Cross-center grid; B Inner area distance; C Inner area time; D Self-grooming; E Buried marbles; F Vertical score; G Representative open field trajectory map of the control group; H Representative open field trajectory diagram of the ASD model; I Representative open field trajectory maps of the hUCMSCs + EVs stem cell treatment group; J Representative open field trajectory maps of the hUCMSCs treatment grou. K Representative open field trajectory maps of the EVs treatment group. The data are presented as the mean ± SD (x̅±s, n = 12). Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, compared with the VPA-exposed ASD model group Analysis of repetitive stereotypic behavior Marble-burying test Marble-burying test Marble-burying test The marble-burying test was employed to evaluate repetitive and stereotypic behaviors in rodents. As shown in Fig. 2E, VPA-exposed C57BL/6 mice exhibited a significant increase in marble-burying behavior compared to control animals (P < 0.0001). Administration of hUCMSCs alone significantly reduced this repetitive behavior, with treated mice burying fewer marbles than the VPA group (P < 0.05). Similarly, treatment with EVs alone also led to a marked reduction in marble-burying (P < 0.001). Most notably, the combined administration of hUCMSCs and EVs produced the strongest therapeutic effect, markedly attenuating repetitive behavior and resulting in significantly fewer buried marbles compared to the VPA model group (P < 0.0001). This partial restoration of normal exploratory patterns suggests a synergistic beneficial effect of the combined treatment. (2)Self-grooming assessment (2)Self-grooming assessment Self-grooming assessment Consistent with the marble-burying results, VPA-exposed mice exhibited significantly increased self-grooming frequencies compared to control animals (P < 0.0001; Fig. 2D). Treatment with hUCMSCs alone did not result in a statistically significant reduction in self-grooming behavior relative to the VPA group (P > 0.05). In contrast, administration of EVs alone led to a significant decrease in grooming frequency (P < 0.05). Most importantly, the combined intervention of hUCMSCs and EVs produced the most substantial effect, significantly reducing this repetitive behavior compared to the untreated VPA-exposed mice(P < 0.01). Our findings demonstrate that VPA exposure induces robust repetitive stereotypic behaviors in C57BL/6 mice, as evidenced by elevated marble-burying (Fig. 2E) and self-grooming (Fig. 2D) measures. These behavioral alterations may reflect neurodevelopmental disturbances in basal ganglia-cortical circuitry. The observed attenuation of repetitive behaviors following hUCMSCs + EVs treatment suggests potential for targeting circuit-specific pathophysiology in ASD-like phenotypes. Three-chamber social behavioral analysis The three-chamber test was conducted to assess social motivation (0–10 min) and social novelty preference (10–20 min) by quantifying exploration time toward unfamiliar conspecifics (Stranger 1/2) versus nonsocial stimuli (empty cages/objects). I.Social motivation phase (sociability) I.Social motivation phase (sociability) Social motivation phase (sociability) VPA-exposed C57BL/6 mice exhibited significant deficits in social motivation compared to controls (Fig. 3A). hUCMSCs + EVs treatment ameliorated these behavioral abnormalities as follows: Spatial exploration preference In the social preference test, VPA-exposed mice spent significantly less time exploring the Stranger 1 cage compared to control animals (P < 0.0001) and exhibited a heightened preference for the empty cage (P < 0.0001). Treatment with hUCMSCs alone significantly increased the time spent exploring the Stranger 1 cage relative to the VPA group (P < 0.01); however, it did not significantly reduce the duration of empty cage exploration (P > 0.05). Similarly, administration of EVs alone also resulted in a significant increase in Stranger 1 cage investigation time (P < 0.001), accompanied by a significant reduction in empty cage exploration (P < 0.001). Most importantly, the combined treatment with hUCMSCs and EVs produced the most pronounced reversal of these aberrant behaviors, significantly increasing time spent exploring Stranger 1 cage (P < 0.0001 vs. VPA group) while markedly reducing empty cage fixation (P < 0.0001vs. VPA group; Fig. 3A). Social interaction behavior Consistent with the observed social interaction deficits, VPA-exposed mice exhibited significantly reduced sniffing time toward Stranger 1 compared to control animals (P < 0.0001), as well as a reduction in object sniffing duration (P < 0.05). Treatment with hUCMSCs alone did not result in a statistically significant increase in sniffing time directed toward Stranger 1 relative to the VPA group (P > 0.05). In contrast, administration of EVs alone significantly increased the time spent sniffing Stranger 1 (P < 0.01). Most notably, the combined treatment with hUCMSCs and EVs produced the most substantial improvement, significantly enhancing social investigation by increasing Stranger 1 sniffing time compared to the VPA-exposed group (P < 0.0001; Fig. 3A). Trajectory analysis VPA-exposed mice predominantly occupied the empty cage quadrant (Fig. 3C), whereas hUCMSCs + EVs-treated animals exhibited spatial patterns comparable to controls (Figs. 3D–E), demonstrating restored proximity to social stimuli. ‌This study demonstrates: VPA exposure impaired sociability, as evidenced by social avoidance (Stranger 1) and nonsocial preference (empty cage; P < 0.0001). hUCMSCs + EVs intervention normalized exploratory balance toward social contexts. II.Social novelty preference phase II.Social novelty preference phase Social novelty preference phase Spatial exploration patterns VPA-exposed mice displayed aberrant novelty-seeking behavior, characterized by significant hyperfixation on the Stranger 1 cage (P < 0.0001 vs. controls) and pronounced avoidance of the Stranger 2 cage (P < 0.0001 vs. controls; Fig. 3B). Treatment with hUCMSCs alone significantly reduced the time spent exploring the Stranger 1 cage (P < 0.05 vs. VPA group) and markedly increased exploration of the Stranger 2 cage (P < 0.0001). Similarly, administration of EVs alone also significantly decreased Stranger 1 cage exploration (P < 0.05) while substantially enhancing investigation of the Stranger 2 cage (P < 0.0001). Most effectively, the combined treatment with hUCMSCs and EVs produced the most pronounced reversal of these abnormalities, significantly reducing Stranger 1 cage perseveration (P < 0.0001 vs. VPA group) while concurrently enhancing Stranger 2 cage exploration (P < 0.0001 vs. VPA group). Social interaction disparities The VPA group exhibited significantly excessive interaction time with Stranger 1 compared to control animals (P < 0.01) and markedly suppressed sniffing behavior toward Stranger 2 (P < 0.0001). Treatment with hUCMSCs alone did not produce statistically significant alterations in the time spent exploring Stranger 1 (P > 0.05 vs. VPA group) or Stranger 2 (P > 0.05). Similarly, administration of EVs alone also failed to induce significant changes in the duration of exploration toward either Stranger 1 (P > 0.05) or Stranger 2 (P > 0.05). In contrast, the combined administration of hUCMSCs and EVs effectively modulated these social behaviors, significantly attenuating the excessive fixation on Stranger 1 (P < 0.05 vs. VPA group) while concurrently increasing novel social exploration toward Stranger 2 (P < 0.05 vs. VPA group; Fig. 3B). Trajectory analysis VPA-exposed mice persistently clustered near Stranger 1 (Fig. 3F), whereas hUCMSCs + EVs-treated mice showed dynamic exploration patterns resembling controls (Figs. 3G–H), with balanced attention to both strangers. ‌This study demonstrates: VPA exposure induced restricted social novelty preference, manifesting as Stranger 1 perseveration and Stranger 2 avoidance. hUCMSCs + EVs treatment restored cognitive flexibility, promoting appropriate novelty-seeking behavior (P < 0.05 for Stranger 2 metrics). Fig. 3Behavioral analysis of three-chamber social behavior in C57BL/6 mice. A Social ability; B Social novelty; C Representative heat map of social ability (0–10 min) in the ASD model group; D Representative trajectory map of social ability (0–10 min) in the control group; E Representative trajectory map of social ability of hUCMSCs + EVs stem cell treatment group (0–10 min); F Representative heat map of the social novelty preference phase (10–20 min) in the ASD model group; G Representative heat map of the social novelty preference phase (10–20 min) in the control group; H Representative trajectory map of the social novelty preference phase (10–20 min) in the hUCMSCs + EVs stem cell treatment group.The data are presented as the mean ± SD (x̅±s, n = 12). Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, compared with the VPA-exposed ASD model group. Stranger 1 cage: Time spent in the stranger 1 cage; Empty cage: Time spent in the object cage; Stranger 1: Interaction time with stranger 1; Object: Interaction time with the object; Stranger 2 cage: Time spent in the stranger 2 cage; Stranger 2: Interaction time with stranger 2 Behavioral analysis of three-chamber social behavior in C57BL/6 mice. A Social ability; B Social novelty; C Representative heat map of social ability (0–10 min) in the ASD model group; D Representative trajectory map of social ability (0–10 min) in the control group; E Representative trajectory map of social ability of hUCMSCs + EVs stem cell treatment group (0–10 min); F Representative heat map of the social novelty preference phase (10–20 min) in the ASD model group; G Representative heat map of the social novelty preference phase (10–20 min) in the control group; H Representative trajectory map of the social novelty preference phase (10–20 min) in the hUCMSCs + EVs stem cell treatment group.The data are presented as the mean ± SD (x̅±s, n = 12). Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, compared with the VPA-exposed ASD model group. Stranger 1 cage: Time spent in the stranger 1 cage; Empty cage: Time spent in the object cage; Stranger 1: Interaction time with stranger 1; Object: Interaction time with the object; Stranger 2 cage: Time spent in the stranger 2 cage; Stranger 2: Interaction time with stranger 2 Assessment of learning and memory capacity To evaluate spatial learning and memory retention, we conducted the Morris water maze (MWM) test during postnatal days 60–67 (PND60-67). As illustrated in Fig. 4, valproic acid (VPA)-exposed mice exhibited significantly prolonged escape latencies compared to controls during acquisition trials (P < 0.01; Fig. 4A), indicating impaired spatial learning. The hUCMSCs + EVs group, hUCMSCs alone group and EVs alone group treatment reduced these latencies to near-control levels (P < 0.01 vs. VPA group; Fig. 4A), suggesting enhanced task acquisition. Spatial navigation patterns differed markedly between groups (Figs. 4F–J). VPA-exposed mice displayed undirected, random swimming paths (Fig. 4F), whereas controls showed goal-directed navigation with direct platform approaches (Fig. 4G). The hUCMSCs + EVs group, hUCMSCs alone group and EVs alone group demonstrated improved search efficiency through reduced circling and targeted trajectories resembling controls (Fig. 4H–J). During probe trials, VPA-exposed mice spent significantly less time in the target quadrant compared to control animals (P < 0.0001; Fig. 4B). Treatment with hUCMSCs alone did not significantly alter the time spent in the target quadrant (P > 0.05), whereas administration of EVs alone significantly increased target quadrant duration (P < 0.05). The combined treatment with hUCMSCs and EVs produced the most substantial effect, significantly increasing target quadrant duration compared to the VPA group (P < 0.001; Fig. 4B).Similarly, VPA-exposed mice exhibited significantly fewer platform crossings than controls (P < 0.0001; Fig. 4D). Both hUCMSCs alone and EVs alone significantly increased the number of platform crossings (P < 0.05 and P < 0.01, respectively). Most effectively, the combined treatment with hUCMSCs and EVs restored crossing frequency to control-equivalent levels (P < 0.0001 vs. VPA; Fig. 4D).While VPA-exposed mice showed no significant target quadrant preference (P > 0.05), the combined hUCMSCs and EVs treatment group exhibited a significant spatial bias (P < 0.05 vs. VPA; Fig. 4E). Trajectory analysis revealed disorganized exploration across quadrants in VPA mice (Fig. 4K), contrasted with focused navigation in controls (Fig. 4L). The hUCMSCs + EVs group, hUCMSCs alone group and EVs alone group showed localized search patterns similar to controls (Fig. 4L). No intergroup differences in swimming distance were observed (P > 0.05; Fig. 4C), confirming preserved locomotor function. This study demonstrates VPA exposure induced significant spatial memory deficits, characterized by prolonged escape latencies, random search strategies (Figs. 4A, F), and impaired target quadrant preference (Figs. 4B, E). hUCMSCs + EVs treatment ameliorated these deficits, enhancing both spatial learning efficiency and memory retention, independent of locomotor effects. Fig. 4Behavioral analysis of the Morris water maze. A Escape latency to locate the hidden platform. B Time spent in the target quadrant. C Total swimming distance. D Number of platform crossings. E Time distribution across the four quadrants during the spatial probe trial. F Representative swimming paths of the group of C57BL/6 mice exposed to VPA during spatial acquisition trials. G Representative swimming paths of control mice during spatial acquisition trials. H Representative swimming paths of the hUCMSCs + EVs stem cell treatment group mice during spatial acquisition trials. I Representative swimming paths of the hUCMSCs alone treatment group mice during spatial acquisition trials. J Representative swimming paths of the EVs alone treatment group mice during spatial acquisition trials. K Representative exploration trajectories of the group of C57BL/6 mice exposed to VPA during spatial probe trials. L Representative exploration trajectories of control mice during spatial probe trials. M Representative exploration trajectories of the hUCMSCs + EVs stem cell treatment group mice during spatial probe trials. N Representative exploration trajectories of the hUCMSCs alone treatment group mice during spatial probe trials. O Representative exploration trajectories of the EVs alone treatment group mice during spatial probe trials.The data are presented as the mean ± SD (x̅±s, n = 12). Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, Compared with the VPA-exposed ASD model group Behavioral analysis of the Morris water maze. A Escape latency to locate the hidden platform. B Time spent in the target quadrant. C Total swimming distance. D Number of platform crossings. E Time distribution across the four quadrants during the spatial probe trial. F Representative swimming paths of the group of C57BL/6 mice exposed to VPA during spatial acquisition trials. G Representative swimming paths of control mice during spatial acquisition trials. H Representative swimming paths of the hUCMSCs + EVs stem cell treatment group mice during spatial acquisition trials. I Representative swimming paths of the hUCMSCs alone treatment group mice during spatial acquisition trials. J Representative swimming paths of the EVs alone treatment group mice during spatial acquisition trials. K Representative exploration trajectories of the group of C57BL/6 mice exposed to VPA during spatial probe trials. L Representative exploration trajectories of control mice during spatial probe trials. M Representative exploration trajectories of the hUCMSCs + EVs stem cell treatment group mice during spatial probe trials. N Representative exploration trajectories of the hUCMSCs alone treatment group mice during spatial probe trials. O Representative exploration trajectories of the EVs alone treatment group mice during spatial probe trials.The data are presented as the mean ± SD (x̅±s, n = 12). Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, Compared with the VPA-exposed ASD model group Analysis of neuroinflammatory markers in the prefrontal cortex ELISA quantification demonstrated that VPA exposure significantly altered the prefrontal cortical inflammatory profile in C57BL/6 mice. hUCMSCs + EVs administration effectively restored cytokine homeostasis, as evidenced by the following: As demonstrated in Fig. 5, valproic acid (VPA) exposure significantly altered the neuroinflammatory profile in the PFC of C57BL/6 mice, whereas combined hUCMSCs and EVs (hUCMSCs + EVs) intervention effectively restored cytokine homeostasis. VPA-exposed animals exhibited substantially elevated IL-1β levels compared to controls (P < 0.0001; Fig. 5A). Administration of hUCMSCs alone significantly reduced IL-1β concentrations (P < 0.01 versus VPA group), and treatment with EVs alone also resulted in a marked decrease (P < 0.001). Most effectively, the combined administration of hUCMSCs and EVs normalized IL-1β levels (P < 0.0001 versus VPA group). Similarly, IL-6 levels were markedly elevated in the VPA cohort (P < 0.0001; Fig. 5B). hUCMSCs alone significantly attenuated IL-6 concentrations (P < 0.05), and EVs alone also produced a significant reduction (P < 0.001). Again, the combined hUCMSCs and EVs treatment most substantially decreased IL-6 levels (P < 0.0001 versus VPA group). TNF-α concentrations were significantly increased following VPA exposure (P < 0.0001; Fig. 5C). hUCMSCs alone significantly lowered TNF-α levels (P < 0.01), as did EVs alone (P < 0.001). The combined treatment with hUCMSCs and EVs returned TNF-α concentrations to control-equivalent ranges (P < 0.0001 versus VPA group). Conversely, anti-inflammatory IL-10 production was significantly suppressed in VPA-exposed mice (P < 0.0001; Fig. 5D). hUCMSCs alone significantly elevated IL-10 levels (P < 0.05), and EVs alone also markedly increased IL-10 production (P < 0.001). Most notably, the combined administration of hUCMSCs and EVs restored IL-10 to physiological concentrations (P < 0.0001 versus VPA group). Fig. 5Analysis of neuroinflammatory and oxidative stress levels in the prefrontal cortex. A IL-1β B IL-6 C TNF-α D IL-10 E GSH-Px activity; F GSH content; G SOD activity; H MDA content; I T-NOS activity; J NO content; K CAT activity. The data are presented as the mean ± SD (x̅±s, n = 12). Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, compared with the VPA-exposed ASD model group Analysis of neuroinflammatory and oxidative stress levels in the prefrontal cortex. A IL-1β B IL-6 C TNF-α D IL-10 E GSH-Px activity; F GSH content; G SOD activity; H MDA content; I T-NOS activity; J NO content; K CAT activity. The data are presented as the mean ± SD (x̅±s, n = 12). Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, compared with the VPA-exposed ASD model group Assessment of oxidative stress in the prefrontal cortex ELISA analyses revealed that VPA exposure significantly disrupted redox homeostasis in the prefrontal cortex of C57BL/6 mice. hUCMSCs + EVs administration restored oxidative balance through the following mechanisms: As shown in Fig. 5, VPA-exposed animals exhibited markedly reduced catalase (CAT) activity compared to controls (P < 0.0001; Fig. 5K). Administration of hUCMSCs alone significantly increased CAT activity (P < 0.05 versus VPA group), and treatment with EVs alone also produced a significant elevation (P < 0.01). Most effectively, the combined administration of hUCMSCs and EVs normalized CAT activity to control levels (P < 0.0001 versus VPA group).Similarly, glutathione peroxidase (GSH-Px) activity was significantly suppressed in the VPA group (P < 0.0001; Fig. 5E). hUCMSCs alone significantly enhanced GSH-Px activity (P < 0.05), and EVs alone also markedly increased its activity (P < 0.001). The combined treatment with hUCMSCs and EVs effectively restored GSH-Px activity (P < 0.0001 versus VPA group).Furthermore, glutathione (GSH) content was substantially diminished following VPA exposure (P < 0.0001; Fig. 5F). hUCMSCs alone significantly elevated GSH concentrations (P < 0.05), and EVs alone also resulted in a significant increase (P < 0.01). The combined intervention of hUCMSCs and EVs significantly elevated GSH levels (P < 0.001 versus VPA group).Superoxide dismutase (SOD) activity was also significantly impaired in the VPA group (P < 0.0001; Fig. 5G). Both hUCMSCs alone and EVs alone significantly enhanced SOD activity (P < 0.01for both treatments). Most substantially, the combined treatment with hUCMSCs and EVs recovered SOD activity to normal levels (P < 0.0001 versus VPA group). Concurrently, oxidative damage markers were significantly elevated in VPA-exposed mice. Malondialdehyde (MDA) levels increased substantially (P < 0.0001; Fig. 5H), indicating enhanced lipid peroxidation. Treatment with hUCMSCs alone significantly reduced MDA levels (P < 0.01 versus VPA group), and administration of EVs alone also resulted in a marked decrease (P < 0.001). Most effectively, the combined treatment with hUCMSCs and EVs substantially attenuated MDA elevation (P < 0.0001 versus VPA group).Similarly, total nitric oxide synthase (NOS) activity rose significantly in the VPA group (P < 0.0001; Fig. 5I). hUCMSCs alone significantly reduced NOS activity (P < 0.05), and EVs alone also produced a significant decrease (P < 0.01). The combined administration of hUCMSCs and EVs effectively mitigated NOS activity (P < 0.001 versus VPA group).Nitric oxide (NO) levels were significantly elevated in the VPA group (P < 0.0001; Fig. 5J). While hUCMSCs alone did not result in a statistically significant reduction in NO levels (P > 0.05), treatment with EVs alone significantly decreased NO concentrations (P < 0.05). Notably, the combined intervention with hUCMSCs and EVs significantly reduced NO levels (P < 0.01 versus VPA group). Microglial activation and neuroinflammatory response To evaluate neuroinflammatory responses, we performed Iba1 immunofluorescence staining on hippocampal CA1 and prefrontal cortical tissues collected at postnatal day 70 (PND70). As shown in Figs. 6D–F, control mice exhibited resting microglia with low Iba1 expression in the hippocampal CA1 region. In contrast, VPA-exposed mice demonstrated significantly activated microglia characterized by hypertrophic morphology and elevated Iba1 fluorescence intensity compared to control animals (P < 0.0001; Fig. 6A–C and G). Treatment with hUCMSCs alone significantly decreased Iba1 fluorescence intensity versus the VPA cohort (P < 0.05; Fig. 6K–M and G). Similarly, administration of EVs alone also produced a substantial reduction in Iba1 intensity relative to the VPA group (P < 0.001; Fig. 6O–Q and G). Most effectively, the combined treatment with hUCMSCs and EVs resulted in the most pronounced reduction in microglial activation, showing significantly decreased Iba1 fluorescence intensity compared to the VPA group (P < 0.0001; Fig. 6H-J and G). Similar alterations in microglial activation were observed in the prefrontal cortex (Figs. 6R–B3). Control animals exhibited quiescent microglia with minimal Iba1 immunoreactivity (Figs. 6U–W), whereas VPA exposure induced intense Iba1 staining, indicating pronounced proinflammatory activation (P < 0.0001 vs. controls; Fig. 6R–T and N). Treatment with hUCMSCs alone significantly decreased Iba1 fluorescence intensity compared to the VPA group (P < 0.05; Figs. 6A1–A3, 6 N). Similarly, administration of EVs alone also markedly reduced Iba1 expression relative to the VPA group (P < 0.01; Figs. 6B1–B3, 6 N). Most effectively, the combined administration of hUCMSCs and EVs substantially attenuated this inflammatory response, resulting in significantly lower Iba1 expression compared to the VPA group (P < 0.0001; Fig. 6X–Z N). Fig. 6Immunofluorescence of Iba1 in the prefrontal cortex (PFC) and hippocampal CA1 region of control and ASD model mice A–Q Representative immunofluorescence images of Iba1 expression in the hippocampal CA1 region in the VPA-exposed (A–C) and Control (D–F) groups and hUCMSCs + EVs stem cell treatment group (H–J), hUCMSCs alone treatment group (K–M), EVs alone treatment group (O–Q). G Mean fluorescence intensity of Iba1, expressed in arbitrary units. (R-B3) Representative Iba1 immunofluorescence in the prefrontal cortex of the VPA-exposed (R–T) and Control (U‒W) groups and hUCMSCs + EVs stem cell treatment group (X–Z), hUCMSCs alone treatment group (A1–A3), EVs alone treatment group (B1–B3). N Shows the mean fluorescence intensity of Iba1 in the cerebral cortex, also expressed in arbitrary units. The data are expressed as arbitrary units (AUs) and were analyzed via ImageJ. The values represent the mean ± SD (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, Compared with VPA group; n = 6/group). Integrated density quantification (Figs. 6G, N) based on n = 6 animals per group, with 3 sections analyzed per animal.For all the immunofluorescence images, Iba1 is stained green, and DAPI (nuclei) is stained blue. The scale bar in all the images represents 50 μm Immunofluorescence of Iba1 in the prefrontal cortex (PFC) and hippocampal CA1 region of control and ASD model mice A–Q Representative immunofluorescence images of Iba1 expression in the hippocampal CA1 region in the VPA-exposed (A–C) and Control (D–F) groups and hUCMSCs + EVs stem cell treatment group (H–J), hUCMSCs alone treatment group (K–M), EVs alone treatment group (O–Q). G Mean fluorescence intensity of Iba1, expressed in arbitrary units. (R-B3) Representative Iba1 immunofluorescence in the prefrontal cortex of the VPA-exposed (R–T) and Control (U‒W) groups and hUCMSCs + EVs stem cell treatment group (X–Z), hUCMSCs alone treatment group (A1–A3), EVs alone treatment group (B1–B3). N Shows the mean fluorescence intensity of Iba1 in the cerebral cortex, also expressed in arbitrary units. The data are expressed as arbitrary units (AUs) and were analyzed via ImageJ. The values represent the mean ± SD (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, Compared with VPA group; n = 6/group). Integrated density quantification (Figs. 6G, N) based on n = 6 animals per group, with 3 sections analyzed per animal.For all the immunofluorescence images, Iba1 is stained green, and DAPI (nuclei) is stained blue. The scale bar in all the images represents 50 μm Ultrastructural pathology and restoration in PFC and hippocampal CA1 neurons To evaluate synaptic and mitochondrial ultrastructure, we performed transmission electron microscopy (TEM) on prefrontal cortical and hippocampal CA1 tissues collected at postnatal day 70 (PND70). Control group ultrastructure Transmission electron microscopy (TEM) of prefrontal cortex (PFC) and hippocampal CA1 neurons in control mice revealed preserved ultrastructural integrity (Fig. 7). Synapses displayed narrow synaptic clefts with well-defined, electron-dense postsynaptic densities (PSDs). Presynaptic terminals contained abundant synaptic vesicles (SVs) clustered near active zones (Fig. 7D). Axonal terminals exhibited intact membranes without swelling, while mitochondria maintained organized cristae and intact membranes (Fig. 7A). Pathological alterations in VPA-exposed mice VPA-exposed mice exhibited marked ultrastructural abnormalities in PFC and hippocampal CA1 neurons (Fig. 7E). Presynaptic terminals showed significant SV depletion, with complete absence of vesicles in localized regions. Synaptic membranes displayed disorganization, including blurred and thickened synaptic clefts, alongside indistinct pre-/postsynaptic boundaries. Abnormal PSD thickening with reduced electron density was observed in select synapses. Mitochondrial pathologies included cristae disorganization and membrane rupture, with alternating swelling and shrinkage (Fig. 7B). Concurrent astrocytic swelling suggested neuroinflammatory involvement. hUCMSCs + EVs-mediated ultrastructural restoration As shown in Fig. 7C and F, hUCMSCs + EVs administration ameliorated pathological ultrastructural features in VPA-exposed mice, with treated animals exhibiting: (1) narrow synaptic clefts with sharply defined postsynaptic densities (PSDs), (2) presynaptic terminals containing clustered synaptic vesicles (SVs) adjacent to membranes (Fig. 7F), and (3) mitochondria with restored cristae organization. Quantitative synaptic vesicle analysis Quantitative analysis of synaptic vesicle (SV) density revealed significant depletion in the hippocampal CA1 region of VPA-exposed mice compared to controls (P < 0.0001; Fig. 7G). Treatment with hUCMSCs alone did not result in a statistically significant increase in SV density relative to the VPA group. Similarly, administration of EVs alone also failed to produce a significant improvement in this parameter. In contrast, the combined treatment with hUCMSCs and EVs partially restored SV density, showing a significant increase compared to the VPA group (P < 0.01).Similarly, a significant reduction in SV counts was observed in the prefrontal cortex of VPA-exposed mice (P < 0.0001 vs. controls; Fig. 7H). Again, neither hUCMSCs alone nor EVs alone resulted in a significant increase in SV counts compared to the VPA model. However, the combined intervention with hUCMSCs and EVs effectively normalized SV counts, demonstrating a significant increase relative to the VPA group (P < 0.01). These ultrastructural alterations suggest three pathophysiological implications: synaptic dysfunction characterized by impaired vesicular release and disrupted postsynaptic signaling; mitochondrial compromise indicating bioenergetic deficits; and neuroinflammatory crosstalk evidenced by astrocytic swelling. Fig. 7Ultrastructural pathology and restoration in PFC and hippocampal CA1 neurons. A Representative ultrastructural features of mitochondria in the control group. B Representative ultrastructural features of mitochondria in the ASD model group. C Representative ultrastructural features of mitochondria in the hUCMSCs + EVs stem cell treatment group. D Representative synaptic ultrastructure in the control group. E Representative synaptic ultrastructural morphology in the ASD model group. F Representative synaptic ultrastructural in the hUCMSCs + EVs stem cell treatment group. Representative pictures from n = 3 independent experiments for the control and ASD model animals are presented. G The effect of prenatal exposure to VPA and hUCMSCs + EVs stem cell treatment on the number of synaptic vesicles (SVs) in the CA1 region of the hippocampus was analyzed at postnatal. H The effect of prenatal exposure to VPA and hUCMSCs + EVs stem cell treatment on the number of synaptic vesicles (SVs) in the Prefrontal cortex was analyzed at postnatal day 70 (PND70). Synaptic vesicles counted in 30 nerve terminals per animal (n = 3 per group) using ImageJ particle analysis. The data represent the mean ± standard deviation (SD) from n = 3 independent experiments. Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, compared with the VPA-exposed ASD model group Ultrastructural pathology and restoration in PFC and hippocampal CA1 neurons. A Representative ultrastructural features of mitochondria in the control group. B Representative ultrastructural features of mitochondria in the ASD model group. C Representative ultrastructural features of mitochondria in the hUCMSCs + EVs stem cell treatment group. D Representative synaptic ultrastructure in the control group. E Representative synaptic ultrastructural morphology in the ASD model group. F Representative synaptic ultrastructural in the hUCMSCs + EVs stem cell treatment group. Representative pictures from n = 3 independent experiments for the control and ASD model animals are presented. G The effect of prenatal exposure to VPA and hUCMSCs + EVs stem cell treatment on the number of synaptic vesicles (SVs) in the CA1 region of the hippocampus was analyzed at postnatal. H The effect of prenatal exposure to VPA and hUCMSCs + EVs stem cell treatment on the number of synaptic vesicles (SVs) in the Prefrontal cortex was analyzed at postnatal day 70 (PND70). Synaptic vesicles counted in 30 nerve terminals per animal (n = 3 per group) using ImageJ particle analysis. The data represent the mean ± standard deviation (SD) from n = 3 independent experiments. Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, compared with the VPA-exposed ASD model group Gut microbiota compositional shifts across taxonomic levels Comparative analysis of gut microbiota composition revealed distinct taxonomic profiles among control, VPA-exposed ASD, and hUCMSCs + EVs-treated groups (Figs. 8A–E). Class level: The VPA_Mice group showed a pronounced reduction in Clostridia (critical for SCFA production) and Bacilli, alongside a marked increase in Gammaproteobacteria (a class linked to intestinal inflammation).The MSC + Evs group demonstrated partial restoration of Clostridia abundance and reduced Gammaproteobacteria compared to VPA_Mice (Fig. 8A). Family level: The VPA_Mice group exhibited a marked increase in Bacteroidaceae (a family linked to gut inflammation) and Tannerellaceae (associated with dysbiosis), alongside a severe depletion of beneficial SCFA-producing families (Lachnospiraceae, Ruminococcaceae) and probiotic Lactobacillaceae. The MSC + Evs group showed intermediate trends: partial recovery of Lachnospiraceae and Ruminococcaceae compared to VPA_Mice (Fig. 8B). Genus level: The VPA_Mice group exhibited a dominant abundance of Bacteroides and Parabacteroides (genera linked to intestinal inflammation and ASD pathology), alongside marked depletion of beneficial genera (Lactobacillus, Ruminococcus, and Alistipes).Critically reduced levels of Lactobacillus (a probiotic genus essential for gut-brain axis regulation) were observed. The MSC + Evs group demonstrated intermediate characteristics: moderate recovery of Ruminococcus and Lactobacillus compared to VPA_Mice (Fig. 8C). Order/phylum levels: The VPA_Mice group displayed a dominant abundance of Enterobacterales (order containing E. coli and other pro-inflammatory taxa) and Bacteroidales, alongside severe depletion of beneficial orders (Lachnospirales: SCFA producers; Lactobacillales: probiotics). The MSC + Evs group showed partial recovery: increased Lachnospirales versus VPA_Mice (though below Control levels) and reduced Enterobacterales (Fig. 8D). The VPA Group exhibited a dominant abundance of Bacteroidia (class associated with gut inflammation in ASD) and Pseudomonadales (opportunistic pathogens), alongside depletion of beneficial classes (Bacilli: probiotics; Verrucomicrobiae: mucin degraders for barrier integrity).Elevated Deferribacteria (linked to mucosal irritation) further indicates pathological imbalance. The MSC + Evs Group displayed intermediate characteristics: moderate recovery of Bacilli and Verrucomicrobiae compared to VPA Group (Fig. 8E). Microbial community structure: Venn diagram analysis identified 1880 control-specific, 427 ASD-specific, and 523 treatment-specific bacterial species, with 51 core species conserved across groups (Fig. 8F). Fig. 8Gut microbiota profiling in control, VPA-induced ASD model, and MSC + Evs group. A Class-level composition. B Family-level composition. C Genus-level composition. D Order-level composition. E Phylum-level composition. F Venn diagram of unique and shared bacterial species among groups Gut microbiota profiling in control, VPA-induced ASD model, and MSC + Evs group. A Class-level composition. B Family-level composition. C Genus-level composition. D Order-level composition. E Phylum-level composition. F Venn diagram of unique and shared bacterial species among groups Taxonomic shifts in gut microbiota composition To characterize genus-level microbiota alterations in autism spectrum disorder (ASD) pathology and therapeutic intervention, we conducted 16 S rRNA sequencing on fecal samples from Control, VPA_Mice, and 3KOhiPSCs groups. The results demonstrated distinct taxonomic shifts across groups (Figs. 9A–D). Family-level alterations: The gut microbiota composition at the family level revealed severe dysbiosis in the VPA_Mice group, characterized by a pronounced enrichment of pathobionts including Bacteroidaceae (associated with ASD-related inflammation), Enterobacteriaceae (pro-inflammatory Gammaproteobacteria), and Desulfovibrionaceae (known mucosal irritants). Concurrently, this group exhibited a marked reduction in commensal taxa critical for gut homeostasis, such as Ruminococcaceae (short-chain fatty acid producers), Lachnospiraceae (butyrate synthesizers), and Lactobacillaceae (neuroprotective probiotics). In contrast, the MSC + Evs group demonstrated partial restoration of microbial balance, evidenced by moderate recovery of Lachnospiraceae and Lactobacillaceae abundance relative to the VPA_Mice group (Fig. 9A). Genus-level variations: The VPA_Mice group exhibited pronounced dysbiosis, characterized by significant enrichment of Bacteroides (pro-inflammatory pathobiont) and Parabacteroides, alongside marked depletion of neuroprotective genera including Lactobacillus, Limosilactobacillus, Ligilactobacillus (probiotics), Ruminococcus (SCFA producer), and Akkermansia (mucosal barrier enhancer). This profile aligns with established ASD-associated dysbiosis, suggesting impaired gut-brain axis communication. The MSC + Evs group displayed partial restoration of microbial balance, evidenced by moderate recovery of Lactobacillus and Ruminococcus compared to VPA_Mice (Fig. 9B). Order-level dynamics: The VPA_Mice group exhibited severe dysbiosis, characterized by substantial enrichment of Enterobacterales (pro-inflammatory taxa) and Bacteroidales, alongside marked depletion of beneficial orders including Lachnospirales and Oscillospirales (short-chain fatty acid producers critical for gut barrier integrity), Lactobacillales (probiotics), and Verrucomicrobiales (mucin degraders). This profile aligns with established ASD-associated dysbiosis, suggesting impaired microbial metabolite production and heightened intestinal inflammation. The MSC + Evs group demonstrated partial restoration of microbial balance, evidenced by moderate recovery of Lachnospirales and Lactobacillales compared to VPA_Mice (Fig. 9C). Phylum-level composition: The VPA Group exhibited a dysbiotic signature characterized by substantial enrichment of Deferribacterota (associated with mucosal irritation), alongside marked depletion of beneficial Bacillota (encompassing SCFA-producing taxa) and Verrucomicrobiota (mucin degraders critical for barrier integrity). This profile aligns with established ASD-associated dysbiosis, suggesting compromised microbial metabolite production and impaired gut homeostasis. The MSC + Evs group displayed partial restoration of microbial balance, evidenced by increased Bacillota abundance compared to the VPA Group (Fig. 9D). The ASD microbiota profile featured dysbiotic characteristics including reduced beneficial taxa (Lactobacillus). Enriched pro-inflammatory suggested gut inflammation. hUCMSCs + EVs intervention restored symbiotic genera while suppressing inflammatory taxa, aligning with evidence linking microbiota homeostasis to neurological function. Fig. 9Hierarchical alterations in gut microbiota composition across taxonomic levels. Heatmaps illustrate relative bacterial abundance at A Family, B Genus, C Order, and D Phylum levels for Control (normal mice), VPA-ASD (VPA-induced ASD model), and Treatment (hUCMSCs + EVs intervention) groups Hierarchical alterations in gut microbiota composition across taxonomic levels. Heatmaps illustrate relative bacterial abundance at A Family, B Genus, C Order, and D Phylum levels for Control (normal mice), VPA-ASD (VPA-induced ASD model), and Treatment (hUCMSCs + EVs intervention) groups Gut microbiota compositional shifts at family and genus levels The cladogram (Fig. 10A) demonstrated that Tannerellaceae (node c) was predominantly associated with the VPA_Mice group, consistent with its established role as a dysbiosis-associated family linked to intestinal inflammation in autism spectrum disorder (ASD) models. Conversely, Bifidobacteriaceae (node a) and its higher taxonomic order Bifidobacteriales (node b) showed strongest affiliation with the MSC + Evs group, aligning with their known functions as anti-inflammatory commensals and producers of neuroprotective metabolites. As shown in (Fig. 10B), the results demonstrated that Tannerellaceae exhibited a significantly higher abundance in the VPA_Mice group (positive LDA score) compared to the MSC + Evs group (negative LDA score). This finding aligns with Tannerellaceae’s established association with gut dysbiosis in autism spectrum disorder (ASD) models, where it promotes intestinal inflammation and barrier disruption through LPS production. The pronounced enrichment of Tannerellaceae in VPA_Mice further confirms its pathological dysbiotic state, consistent with core features of ASD gut pathology. Conversely, the reduced abundance of this family in MSC + Evs suggests partial remediation through stem cell-derived EVs intervention, potentially via immunomodulatory pathways that suppress pathobiont proliferation. The cladogram (Fig. 10C) demonstrated that dysbiotic taxa (nodes m, o, p: Tannerellaceae, Desulfovibrionaceae, Desulfovibrionales) exhibited strongest phylogenetic association with the VPA Group. This finding aligns with their established roles in promoting intestinal inflammation and impaired gut barrier function in ASD models through mechanisms involving LPS production and sulfate reduction. Conversely, commensal taxa (nodes a, b, c: Bifidobacterium, Bifidobacteriaceae, Bifidobacteriales) showed predominant association with the MSC + Evs group, consistent with their neuroprotective functions via butyrate production and anti-inflammatory signaling. As shown in Fig. 10D, the results demonstrated significant enrichment of pathobionts in the VPA Group (f. Tannerellaceae; c__Desulfovibrionia; o__Desulfovibrionales) and commensal taxa in the MSC + Evs group (f. Bifidobacteriaceae; g. Bifidobacterium). The MSC + Evs group exhibited intermediate biomarker profiles, with partial reduction in pathobionts and moderate recovery of select commensals. The selective enrichment of Bifidobacterium—a genus linked to gut barrier maintenance, anti-inflammatory effects, and neuromodulatory metabolite production—in the treatment group suggests microbiota-mediated therapeutic mechanisms. Conversely, the expansion of Tannerellaceae in ASD mice aligns with reported associations between these taxa and dysbiosis-related inflammation. These findings demonstrate the microbiota-modulating capacity of hUCMSCs + EVs administration, potentially contributing to ASD pathophysiology mitigation. Fig. 10Gut microbiota profiling across experimental groups at family and genus levels. A Phylogenetic cladogram (family level): taxonomic nodes are colored by group-specific enrichment (red: MSC + Evs; blue: VPA-induced ASD model; green: normal control). B LDA scores (family level).C Phylogenetic cladogram (genus level) D LDA scores (genus level) Gut microbiota profiling across experimental groups at family and genus levels. A Phylogenetic cladogram (family level): taxonomic nodes are colored by group-specific enrichment (red: MSC + Evs; blue: VPA-induced ASD model; green: normal control). B LDA scores (family level).C Phylogenetic cladogram (genus level) D LDA scores (genus level) Comparative analysis of gut microbiota composition Differential abundance analysis (DESeq2; |log2FC| >1, P < 0.05) revealed distinct microbial signatures among control, VPA-exposed ASD (VPA_Mice), and hUCMSCs + EVs-treated (MSC + EVs) groups at genus and phylum levels. Control vs. VPA_Mice Control vs. VPA_Mice Control vs. VPA_Mice To identify differentially abundant bacterial genera associated with autism spectrum disorder (ASD) pathology, we performed differential abundance analysis of 16 S rRNA sequencing data. As shown in the volcano plot (Fig. 11A), the results demonstrated significant enrichment of pro-inflammatory genera (Bacteroides, Escherichia_Shigella, Parabacteroides, Desulfovibrio, Staphylococcus) in the VPA_Mice group, alongside marked depletion of beneficial taxa (Bifidobacterium, Lachnospiraceae groups, Marvinbryantia) compared to Control. This dysbiotic signature aligns with established ASD gut pathology, characterized by elevated pathobionts that promote intestinal inflammation through LPS production and impaired short-chain fatty acid metabolism.Concurrently, the Control group exhibited inverse abundance patterns, with commensal genera demonstrating significantly higher abundance. The analysis (Fig. 11B) revealed that the VPA_Mice group exhibited a dysbiotic signature characterized by substantial enrichment of Pseudomonadota (containing pro-inflammatory Proteobacteria), alongside marked depletion of beneficial Bacillota (encompassing SCFA-producing taxa) and Actinomycetota (anti-inflammatory commensals). This profile aligns with established ASD-associated dysbiosis, suggesting compromised gut barrier function and impaired neuroimmune signaling. In contrast, the Control group maintained a homeostatic microbiota dominated by Bacillota and Actinomycetota, with minimal colonization by Pseudomonadota. This balanced composition supports gut-brain axis homeostasis through enhanced short-chain fatty acid production and anti-inflammatory activity. (2)MSC+Evs vs. VPA_Mice (2)MSC+Evs vs. VPA_Mice MSC+Evs vs. VPA_Mice As shown in the volcano plot (Fig. 11C), the results demonstrated significant enrichment of pro-inflammatory genera (Escherichia_Shigella, Parabacteroides, Desulfovibrio, Helicobacter, Mucispirillum) in the VPA_Mice group, alongside marked depletion of beneficial taxa (Bifidobacterium, Limosilactobacillus, Ruminococcus, Marvinbryantia) relative to MSC + Evs group. This dysbiotic signature aligns with established ASD gut pathology, characterized by elevated pathobionts that disrupt gut barrier integrity through LPS production and impaired short-chain fatty acid metabolism.Concurrently, commensal genera exhibited significantly higher abundance in the MSC + Evs group, indicating partial remediation through stem cell intervention. The pronounced enrichment of Desulfovibrio (sulfide producer) and depletion of Limosilactobacillus (GABA synthesizer) in VPA_Mice suggest specific mechanisms of gut-brain axis disruption. As shown in Fig. 11D, the results demonstrated significant enrichment of Pseudomonadota in the VPA_Mice group, alongside marked depletion of beneficial Bacillota and Actinomycetota compared to MSC + Evs group. This dysbiotic signature aligns with established ASD gut pathology, characterized by elevated pro-inflammatory phyla that disrupt gut barrier integrity through LPS production and impaired short-chain fatty acid metabolism. Concurrently, the MSC + Evs group exhibited inverse abundance patterns, with commensal phyla showing significantly higher abundance, indicating partial remediation through stem cell intervention. The depletion of Bacillota (essential for butyrate synthesis) in VPA_Mice suggest specific mechanisms of gut-brain axis disruption. Fig. 11Gut microbiota compositional changes in VPA-induced ASD mice and hUCMSCs + EVs intervention. A Genus-level comparison (DESeq2) between control and VPA-induced ASD mice B Phylum-level comparison (DESeq2) between control and VPA-induced ASD mice. C Genus-level comparison (DESeq2) between hUCMSCs + EVs-treated mice (MSC + EVS) and VPA-induced ASD mice.D Phylum-level comparison (DESeq2) between hUCMSCs + EVs-treated mice (MSC + EVS) and VPA-induced ASD mice Gut microbiota compositional changes in VPA-induced ASD mice and hUCMSCs + EVs intervention. A Genus-level comparison (DESeq2) between control and VPA-induced ASD mice B Phylum-level comparison (DESeq2) between control and VPA-induced ASD mice. C Genus-level comparison (DESeq2) between hUCMSCs + EVs-treated mice (MSC + EVS) and VPA-induced ASD mice.D Phylum-level comparison (DESeq2) between hUCMSCs + EVs-treated mice (MSC + EVS) and VPA-induced ASD mice Gut microbial community diversity analysis Alpha-diversity analysis revealed distinct microbial community structures among the control, VPA-exposed ASD model (VPA_Mice), and hUCMSCs + EVs-treated (MSC + EVS) groups (Fig. 12). The Shannon Index indicated comparable diversity trends across all groups as sequence coverage increased (Fig. 12A). In contrast, the Chao1 Index demonstrated significantly higher species richness in the control group relative to VPA_Mice (Fig. 12B). Phylogenetic diversity assessed by the Faith_pd Index was notably elevated in the MSC + EVS group compared to VPA_Mice (Fig. 12C), while observed operational taxonomic unit (OTU) counts were increased in MSC + EVS relative to VPA_Mice (Fig. 12D). These patterns were further corroborated by the Shannon entropy index (Fig. 12E) and Simpson index (Fig. 12F). Collectively, these findings indicate that hUCMSCs + EVs administration enhances microbial community richness (Chao1, Observed Features) and phylogenetic diversity (Faith_pd) in ASD model mice, suggesting therapeutic modulation of gut ecosystem complexity. This restoration warrants further investigation into microbiota-mediated mechanisms underlying the observed effects. Fig. 12Comparative analysis of gut microbial alpha-diversity across experimental groups. A Rarefaction curves of Shannon diversity index showing convergence at high sequencing depths. B–D Boxplots of richness estimators: B Chao1 index, C Faith’s phylogenetic diversity, and D Observed features.E Shannon_entropy_index,FSimpson_index. control: normal control mice; VPA_Mice: VPA-induced ASD model mice; hUCMSCs + EVs: stem cell therapy group Comparative analysis of gut microbial alpha-diversity across experimental groups. A Rarefaction curves of Shannon diversity index showing convergence at high sequencing depths. B–D Boxplots of richness estimators: B Chao1 index, C Faith’s phylogenetic diversity, and D Observed features.E Shannon_entropy_index,FSimpson_index. control: normal control mice; VPA_Mice: VPA-induced ASD model mice; hUCMSCs + EVs: stem cell therapy group Beta-diversity patterns across experimental groups Beta-diversity analysis using Bray-Curtis, unweighted UniFrac, and weighted UniFrac metrics revealed structural dissimilarities in gut microbial communities among experimental groups (Fig. 13A–F). Bray-Curtis analysis: To assess beta-diversity differences in gut microbiota composition among experimental groups, we performed distance-based analysis of 16 S rRNA sequencing data. As shown in Figs. 13A–B, the results demonstrated distinct microbial community separation, revealing significant dysbiosis in the VPA Group relative to Control. The VPA Group exhibited the largest phylogenetic distance from the Control group, indicating profound compositional shifts consistent with autism spectrum disorder (ASD)-like dysbiosis. Conversely, the MSC + Evs group displayed intermediate phylogenetic distance from Control—significantly smaller than the VPA Group but larger than Control—suggesting partial microbial restoration following therapeutic intervention. Unweighted UniFrac: To quantify beta-diversity differences in gut microbiota composition, we performed phylogenetic distance analysis of 16 S rRNA sequencing data. As shown in Fig. 13C–D, the results demonstrated significant microbial community divergence, where the VPA Group exhibited the largest beta-diversity distance from the Control group, indicating severe dysbiosis consistent with autism spectrum disorder (ASD) pathology. Conversely, the MSC + Evs group displayed an intermediate distance from Control—significantly smaller than the VPA Group yet larger than Control—suggesting partial microbial restoration. Crucially, when measuring distances to the VPA Group, the MSC + Evs group showed closer phylogenetic proximity to VPA than Control did, further confirming its therapeutic role in modulating dysbiosis toward the pathological baseline. Weighted UniFrac: As shown in Figs. 13E–F, the results demonstrated significant microbial community divergence, where VPA Group exhibited a significantly greater distance from the Control group compared to other groups, indicating severe dysbiosis consistent with autism spectrum disorder (ASD) pathology. Conversely, MSC + Evs displayed an intermediate distance from Control—smaller than VPA Group but larger than Control—suggesting partial microbial restoration. Crucially, when measuring distances to VPA Group (Figure F), The MSC + Evs Group, positioned between the VPA and Control Groups, suggests its microbiota profile is indicative of stem cell-treated mice. The conserved structural homogeneity in controls aligns with healthy gut ecosystem expectations. Conversely, VPA-induced ASD modeling increased microbial community instability across all metrics (Bray-Curtis; unweighted UniFrac; weighted UniFrac), suggesting pathological disruption of microbiota configuration. These findings highlight distinct microbiota signatures among different mouse groups, offering insights into the potential role of gut microbiota in ASD modeling and stem cell therapy. Fig. 13Beta-diversity analysis of gut microbial community structure. A–F boxplots comparing intra-group dissimilarity using (A–B) Bray-Curtis, C–D Unweighted UniFrac, and E–F Weighted UniFrac distance metrics. The VPA-induced ASD model group demonstrates significantly broader dispersion in pairwise distances compared to the Control group, indicating elevated microbial community heterogeneity. Control: Normal control mice; VPA-induced ASD model: Valproic acid-induced autism spectrum disorder model mice Beta-diversity analysis of gut microbial community structure. A–F boxplots comparing intra-group dissimilarity using (A–B) Bray-Curtis, C–D Unweighted UniFrac, and E–F Weighted UniFrac distance metrics. The VPA-induced ASD model group demonstrates significantly broader dispersion in pairwise distances compared to the Control group, indicating elevated microbial community heterogeneity. Control: Normal control mice; VPA-induced ASD model: Valproic acid-induced autism spectrum disorder model mice Hylogenetic and compositional shifts in gut microbiota Figure 14 delineates phylogenetic and compositional shifts in gut microbiota at the phylum level across experimental groups, visualized through hierarchical clustering heatmaps integrating taxonomic abundance and evolutionary relationships (Figs. 14A–B). The analysis reveals distinct microbial restructuring patterns associated with ASD pathology and therapeutic intervention. In the Control group (Figs. 14A–B), Bacteroidota dominated the microbial profile, while Firmicutes and Proteobacteria exhibited suppressed abundance. This configuration aligns with established signatures of healthy gut microbiota. Strikingly, the VPA-induced ASD model group displayed a pathological inversion of this profile, marked by pronounced expansion of Firmicutes and Proteobacteria concurrent with depletion of Bacteroidota. The Firmicutes/Bacteroidota ratio elevation mirrors clinical observations in ASD-associated dysbiosis, while Proteobacteria enrichment suggests potential inflammatory niche formation. Notably, hUCMSCs + EVs administration reversed these dysbiotic trends, partially restoring the Control-like phylum distribution. Bacteroidota abundance rebounded to near-Control levels, accompanied by suppression of Firmicutes and Proteobacteria. This therapeutic recalibration indicates stem cell-derived EVs may counteract ASD-associated microbial perturbations, favoring re-establishment of taxa critical for mucosal barrier integrity and anti-inflammatory signaling. The Firmicutes/Proteobacteria surge in the ASD model group (Fig. 14A–B) recapitulates human ASD dysbiosis patterns, supporting gut-brain axis involvement in neurodevelopmental disorders.hUCMSCs + EVs-driven recovery of Bacteroidota (Fig. 14B) suggests EVs factors may promote growth of short-chain fatty acid producers and immunoregulatory taxa. Fig. 14Phylogenetically resolved gut microbiota composition at the phylum level. A Phylogenetic_tree_heatmap_ID. B Phylogenetic_tree_heatmap Phylogenetically resolved gut microbiota composition at the phylum level. A Phylogenetic_tree_heatmap_ID. B Phylogenetic_tree_heatmap Network correlation analysis Co-occurrence network analysis (Figs. 15A–D) revealed complex taxonomic interactions across bacterial families, with nodes color-coded by phylum. Node sizes represented family abundance, while edge thickness/color indicated interaction strength and type (positive/negative correlations). Key observations showed higher interaction frequencies in Actinomycetota and Bacteroidota versus Patescibacteria and Pseudomonadota, with Bacteroidota and Bacillota families forming core network hubs. Biologically, VPA-ASD mice exhibited depleted immunomodulatory phyla (Bacillota, Pseudomonadota) and suppressed gut barrier-critical orders (Enterobacteriales, Bifidobacteriales), indicating metabolic-immune axis disruption. hUCMSCs + EVs intervention enhanced metabolic stability-associated phyla (Bacillota, Pseudomonadota), restored symbiotic Bifidobacterium and pathogen-suppressive Enterobacteriaceae, and partially recovered microbial network connectivity. These findings align with microbiota-gut-brain axis involvement in ASD, suggesting microbiome modulation contributes to hUCMSCs + EVs therapeutic efficacy. Fig. 15Network correlation analysis. A Family-level network correlation analysis. B Genus-level network correlation analysis. C Order-level network correlation analysis. D Phylum-level network correlation analysis Network correlation analysis. A Family-level network correlation analysis. B Genus-level network correlation analysis. C Order-level network correlation analysis. D Phylum-level network correlation analysis Discussion Our study demonstrates that combined intravenous hUCMSCs and intranasal hUC-MSC-derived EVs significantly ameliorate ASD-like phenotypes in VPA-exposed mice. The dual-target approach restored social behavior (increased Stranger 1 interaction time↑, p < 0.0001; Fig. 3A) and reduced repetitive behaviors (marble-burying decreased↓, p < 0.01; Fig. 2E), aligning with synaptic ultrastructural restoration in the prefrontal cortex (PFC) and hippocampal CA1 regions (synaptic vesicle density increased↑, p < 0.01; Fig. 7G–H). Notably, this therapy normalized neuroimmune dysregulation, suppressing pro-inflammatory cytokines (IL-1β↓, IL-6↓, TNF-α↓, p < 0.0001; Fig. 5A–C) while elevating anti-inflammatory IL-10↑ (p < 0.0001; Fig. 5D). Concurrently, gut microbiota remodeling was observed, characterized by enrichment of beneficial taxa (e.g., Prevotella ↑) and depletion of pathobionts (e.g., Escherichia-Shigella ↓; Fig. 9B). These findings suggest a synergistic “neuro-immune-microbiota” regulatory loop, wherein systemic MSCs improve gut barrier integrity and EVs directly target CNS inflammation via intranasal delivery. Our study investigates the therapeutic effects of human umbilical cord-derived mesenchymal stem cells (hUCMSCs) and their EVs in a valproic acid (VPA)-induced mice model of autism spectrum disorder (ASD). The results demonstrate that the combined administration of hUCMSCs and their EVs significantly alleviates autism-like phenotypes, including improved social interaction and reduced repetitive behaviors. These findings align with previous studies on MSC-based therapies in neurological disorders, which have shown that MSCs can exert neuroprotective effects by reducing inflammation and promoting neuronal repair [16, 37, 40]. Similarly, the use of MSC-derived EVs has been reported to deliver therapeutic molecules across biological barriers, such as the blood-brain barrier (BBB), further enhancing their potential for treating central nervous system (CNS) disorders [8, 34, 35].However, our study extends these findings by specifically targeting both neuroinflammation and gut- brain axis dysregulation in an ASD model. Previous studies have highlighted the role of the gut microbiome in ASD pathogenesis, with altered microbial composition and reduced diversity observed in individuals with ASD [2, 29, 41]. Our results demonstrate that hUCMSCs and their EVs modulate the gut microbiome by increasing the abundance of beneficial bacteria (Prevotella and Lactobacillus) while reducing potential pathogens (Enterococcus). This observation is consistent with a recent study showing that probiotic interventions can improve behavioral outcomes in preclinical ASD models[29]. Intriguingly, the microbiota shifts we observed—increased Lactobacillus and Ruminococcaceae—correlate with elevated anti-inflammatory SCFAs, a mechanism previously linked to ASD symptom mitigation [23, 29]. However, unlike Sun et al [47], who focused solely on intravenous UC-MSCs, our combinatorial regimen enhanced biodistribution: EVs bypassed the BBB to suppress microglial activation (Iba1↓, p < 0.0001; Fig. 6G, N), while MSCs modulated gut dysbiosis. This multi-target approach may explain the superior rescue of spatial memory deficits (Morris water maze escape latency reduced to control levels, p < 0.01; Fig. 4C). Notably, our synaptic ultrastructural restoration (vesicle density↑; Fig. 7H) extends Perets’findings by providing direct TEM evidence for synaptic repair—a gap in prior literature[35]. Our observed synaptic repair (Fig. 7F) aligns with the hypothesis that MSC-EVs deliver neurotrophic miRNAs (e.g., miR-21) to suppress neuroinflammation [24]. The reduction in microglial activation (Iba1 ↓; Fig. 6N) supports Cheng et al.’s finding that hUC-MSC-EVs inhibit microglial NF-κB signaling, explaining TNF-α suppression (Fig. 5C) [5]. Based on our findings, we propose a potential tripartite mechanism through which hUCMSCs may remodel gut microbiota to mitigate ASD pathogenesis: MSC-secreted antimicrobial peptides (e.g., LL-37) could selectively suppress pathobionts while promoting obligate anaerobes through oxygen tension modulation [12], thereby enriching SCFA-producing taxa (e.g., Ruminococcaceae) that may generate butyrate capable of crossing the blood-brain barrier. This microbial metabolite might subsequently activate microglial GPCR43 to suppress NLRP3 inflammasome activation [18],while potentially enhancing mitochondrial biogenesis in prefrontal neurons [51]. Concurrently, MSC-induced Bifidobacterium enrichment could shift tryptophan metabolism toward neuroprotective kynurenic acid production rather than neurotoxic quinolinic acid [30], potentially normalizing excitatory/inhibitory balance via AMPAR trafficking. Collectively, this gut-initiated cascade—complementing direct neuroimmune modulation—might establish a self-reinforcing therapeutic loop. The observed gut microbiota remodeling by intranasal EVs likely operates through three interconnected pathways: (1) Vagal modulation: Exosomal miRNAs (e.g., miR-21-5p) delivered to the nucleus tractus solitarius regulate gut motility and permeability via vagal efferents, creating a favorable niche for commensal bacteria [14]; (2) Systemic immunoreprogramming: Exosome-carried TGF-β and IL-10 reprogram circulating monocytes, which migrate to gut-associated lymphoid tissue (GALT) and shift macrophage polarization toward anti-inflammatory phenotypes [25]; (3) Metabolite shuttling: EVs transport mitochondrial-derived metabolites (succinate, itaconate) that directly inhibit pathogenic Enterobacteriaceae growth while promoting SCFA producers [3].This multi-tiered crosstalk—though requiring further validation—explains how CNS-targeted EVs indirectly reshape gut ecology. Critically, our dual-target approach differs fundamentally from monotherapies: MSC-only regimens improve social behavior but show limited synaptic restoration and negligible gut microbiota modulation [15];Exosome-only treatments enhance neuronal repair but exhibit restricted immunomodulation [34].Our combinatorial strategy concurrently achieved: neuroinflammatory suppression (Fig. 5); synaptic ultrastructural restoration (Fig. 7). Significant microbiota rebalancing (↑SCFA producers, ↓pathogens; Fig. 9).This multi-dimensional efficacy supports our hypothesis that bridging peripheral (MSC) and central (EVs) targets creates synergistic “neuro-immune-microbiota” crosstalk—a paradigm shift from single-pathway interventions. This study extends existing literature by demonstrating that combining hUCMSCs and EVs creates a “neuro-immune-microbiota” regulatory loop, surpassing the limitations of single-modality approaches. While prior studies have focused on either neural repair (genetic models) or microbiota modulation (FMT trials), our work integrates both pathways in an environmentally induced ASD model, providing mechanistic insights and a translational blueprint for multi-target therapy. In conclusion, our study demonstrates that hUCMSCs and EVs administration in a VPA model of autism can effectively alleviate behavioral deficits, reduce microgliosis in the hippocampus and Prefrontal cortex, and modulate the expression of Pro- and anti-inflammatory cytokines in favor of therapy. These findings suggest the therapeutic potential of hUCMSCs and EVs in treating autism spectrum disorders and shed light on the importance of understanding the immunomodulatory properties of stem cell-based therapies in neurodevelopmental disorders. While our data establish significant correlations between microbiota remodeling, neuroinflammation attenuation, and behavioral recovery, the precise causal mechanisms remain undetermined due to technical constraints. Specifically, we acknowledge that: (i) Fecal microbiota transplantation (FMT) studies were not performed to validate gut-brain axis causality; (ii) Metabolomic profiling of microbially derived molecules (e.g., SCFAs, tryptophan metabolites) was beyond this study’s scope. Conclusion and perspectives This study demonstrates that dual hUC-MSC/EVs therapy synergistically ameliorates ASD-like behaviors by simultaneously restoring gut microbiota balance and repairing neuroimmune dysregulation. Although monotherapies demonstrated partial and limited efficacy, the combined strategy achieved comprehensive recovery across multiple pathological domains, including synaptic ultrastructure, microglial homeostasis, and microbial diversity. Our work provides the first preclinical evidence for a multi-target ASD therapy that integrates gut–brain axis modulation, offering novel mechanistic insights into microbiota–neuroimmune crosstalk. From a clinical perspective, we propose a translatable protocol combining intravenous MSCs (systemic immunomodulation) and intranasal EVs (CNS-targeted delivery), leveraging both cellular and cell-free components. However, we emphasize that this combinatorial strategy remains at the preclinical validation stage, and further safety and efficacy studies are essential before clinical translation. Although this study prioritized demonstrating the feasibility and synergistic efficacy of the combined treatment—rather than dissecting individual contributions—future work should systematically elucidate mechanism-specific effects through multi-omics profiling and functional validations, such as metabolomic screening of neuroactive metabolites and fecal microbiota transplantation experiments. Supplementary Information Supplementary Material 1. Supplementary Material 1.