Cross-kingdom miRNA delivery by Panax notoginseng-derived extracellular–like nanoparticles vesicles restores neuronal function after ischemic injury

Abstract

Blood-brain barrier (BBB) impermeability remains a major obstacle to the effective treatment of neurological disorders, particularly ischemic stroke. Here, we revealed that plant-derived extracellular vesicle-like nanoparticles (PEVs) offer a promising strategy to overcome this barrier. Using an optimized high-yield extraction protocol, we isolated PEVs from four medicinal plants: Panax ginseng, Panax notoginseng, Gastrodia elata, and Ligusticum chuanxiong. Among these, extracellular vesicles derived from Panax notoginseng (NotoEV, vesicle population) exhibited the strongest neuroprotective effects under hypoxic conditions in vitro and in vivo stroke models. Mechanistically, NotoEV delivered conserved plant microRNAs to recipient neurons, where they suppressed key stress granule nucleators GTPase-activating protein-binding protein 2 (G3bp2), Ubiquitin-associated protein 2 like (Ubap2l), and LSM14A mRNA processing body assembly factor (Lsm14a), activated mammalian target of rapamycin (mTOR) signaling, and promoted mitochondrial stabilization via the B-cell lymphoma 2 (Bcl-2)/ Translocase Of Outer Mitochondrial Membrane 20 (TOM20) axis. This cross-kingdom RNA delivery reprogrammed neuronal stress responses, reduced infarct volume, preserved neuronal morphology, and restored electrophysiological function. Collectively, our findings establish a scalable platform for plant-based nanotherapeutics and highlight the translational potential of NotoEV in treating ischemic stroke.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12951-026-04103-z.

Introduction

Ischemic stroke (IS) remains a major cause of death and disability worldwide [1], with 69.9 million cases reported in 2021 and an age-standardized incidence rate of 92.4/100,000 individuals [2]. Despite advances in acute interventions, including thrombolysis, therapeutic options remain limited by a narrow treatment window [3, 4] and hemorrhagic transformation risk [5, 6]. The limited permeability of the blood-brain barrier (BBB) further complicates effective drug delivery, restricting many promising neuroprotective agents [7, 8].

PEVs have recently emerged as potential therapeutic agents for overcoming these barriers [9]. Nanoscale vesicles (30–150 nm in diameter) are naturally enriched with proteins and RNAs and have established low immunogenicity and intrinsic tissue-targeting capabilities. Initial studies suggest that PEVs may confer neuroprotective effects in cerebrovascular and neurodegenerative diseases [10, 11]; however, low extraction yields and insufficient standardization hinder their clinical translation.

To address these limitations, we developed an optimized extraction strategy that combined differential ultracentrifugation with sucrose density gradient purification, achieving yields exceeding 10¹² particles/mL from fresh rhizome tissues. This advancement has enabled the isolation and functional characterization of a high-purity PEV population derived from Panax notoginseng.

In this study, we studied the NotoEV ability to promote neuronal recovery after ischemic injury. Using a combination of oxygen-glucose deprivation/reperfusion (OGD/R) and in vivo stroke models, we demonstrated that NotoEV preserved neuronal morphology, restored mitochondrial function, and enhanced synaptic plasticity. Mechanistically, we identified a critical role for plant-derived microRNAs (miRNAs) in modulating stress granule (SG) dynamics and activating mTOR-dependent pathways, revealing a previously unrecognized cross-kingdom regulatory mechanism underlying neuroprotection. This insight into PEV-associated miRNAs offers a new understanding of how plant vesicles confer resilience to neural cells, offering exciting possibilities for plant-based nanotherapeutics in stroke and other neurological disorders.

Results

Isolation, Characterization, and neuroprotective activity of NotoEV

We isolated PEVs from Panax ginseng, Panax notoginseng, Gastrodia elata, and Ligusticum chuanxiong using an optimized protocol (Fig. 1A). Nano-Coulter analysis revealed that NotoEV had the highest particle concentration (7.57 × 10¹² particles/mL), a uniform mean diameter of 94 nm, and a zeta potential of − 8.80 (Figs. 1B–C). Transmission electron microscopy (TEM) confirmed the presence of intact vesicles with characteristic morphologies (Fig. 1D). Western blotting verified the presence of EV markers Cluster of Differentiation 81 (CD81) and Tumor susceptibility gene 101 protein (TSG101) and the absence of cellular contaminants Golgi matrix protein 130 kD (GM130) and calnexin (Figure S1A).

Fig. 1.

Isolation optimization, physicochemical characterization, and cellular uptake of PEVs. (A) Schematic of the optimized PEV isolation protocol, with key modifications (dashed red boxes) to enhance the extraction efficiency. (B–D) Physicochemical properties of GtEV, CxEV, NotoEV, and GsEV: (B) Size distribution, (C) zeta potential, and (D) TEM images exhibiting characteristic cup-shaped morphology. (E) Confocal microscopy images of DiD-labeled PEVs in primary neurons and (F) quantitative fluorescence intensity analysis.

DiD-labeling revealed that NotoEV was taken up by primary neurons more effectively than other PEVs, a pattern replicated in SH-SY5Y cells. NotoEV also exhibited excellent biocompatibility, with minimal cytotoxicity and preserved viability (Figs. 1E–F and S1B–D). In the SH-SY5Y OGD/R model, NotoEV treatment increased Bcl-2 levels (p < 0.05), decreased Bcl-2-associated X protein (Bax) (p < 0.05) and cleaved caspase-3, and reduced apoptosis (Figures S1E–F), supporting its neuroprotective potential.

NotoEV promoted neuronal structural recovery and synaptic repair after OGD/R

Next, we assessed whether NotoEV supports structural recovery in a primary neuronal OGD/R model (Fig. 2A). NotoEV treatment improved neuronal viability (Fig. 2B) and mitigated ischemia-induced morphological damage. Live-cell imaging revealed improved dendritic and synaptic structure reconstruction during reperfusion (Video1, Figure S2).

Fig. 2.

NotoEV promotes neuroprotection in primary cortical neurons by enhancing viability, neurite outgrowth, and dendritic complexity (A) Workflow for the isolation and culture of primary cortical neurons. (B) CCK-8 assay exhibited dose-dependent neuroprotection by NotoEV (n = 9). (C) MAP2 immunostaining demonstrated neuronal morphology post-treatment. (D-E) Arivis-based neurite reconstruction: (D) Skeletonized projections and (E) structural parameterization. (F–I) Morphometric quantifications: (F) average branch length, (G) branch points, (H) terminal section length, and neuronal volume (I). (J-K) Sholl analysis of dendritic arborization complexity. Data are expressed as mean ± standard error of the mean (SEM); ns: p > 0.05, *p < 0.05, **p < 0.01, and ***p < 0.001 versus OGD/R group

Immunofluorescence demonstrated that OGD/R disrupted dendritic architecture in Microtubule-associated protein 2 (MAP2)-positive neurons, whereas NotoEV restored branching and synaptic integrity (Figs. 2C–D). Arivis 3D analysis exhibited significant improvements in neuronal complexity (Figs. 2E–K), including a longer average branch length (p < 0.001), recovery of branch point numbers to approximately 75% of control levels, increased terminal segment length (p < 0.05), and expanded neuronal volume (p < 0.01). These findings indicate that NotoEV promotes cytoskeletal reconstruction and synaptic regeneration following ischemic injury.

NotoEV restored neuronal electrical activity after OGD/R

To assess functional recovery, we used microelectrode arrays (MEA) to monitor neuronal electrophysiology (Fig. 3A). After OGD/R, active cell numbers decreased significantly (Figs. 3B–D). NotoEV treatment preserved spontaneous firing, maintaining activity at 50–100 Hz, compared with approximately 30 Hz in the untreated OGD/R group. NotoEV increased the spike number (p < 0.05), increased the weighted firing rate (p < 0.001), and prolonged the burst duration (p < 0.01). It also improved neuronal synchronization. Consistent with these functional gains, NotoEV restored synaptic proteins postsynaptic density protein 95 (PSD95) and synaptophysin, which were suppressed using OGD/R (p < 0.05) (Figs. 3E–F), indicating strengthened synaptic networks.

Fig. 3.

NotoEV restores neuronal activity and synaptic integrity post-OGD/R. (A) Schematic of the multielectrode array (MEA) setup. (B–D) Neuronal network recovery: (B) Electrode-neuron interface analysis, (C) spontaneous firing frequency, and (D) network-wide spike activity. (E-F) Synaptic protein expression: (E) PSD95 / synaptophysin immunofluorescence, and (F) Western blotting of protein levels. Data are expressed as mean ± SEM; ns: p > 0.05, *p < 0.05, **p < 0.01, and ***p < 0.001 versus OGD/R group

NotoEV protected neurons through miRNA-Driven regulation of SGs and the G3bp2-mTOR pathway

To explain the underlying mechanism, we performed proteomic profiling to study how NotoEV modulates SG dynamics (Fig. 4A). NotoEV disrupted pathological SG assembly by inhibiting the core scaffold protein G3bp2, reducing Lsm14a, and promoting the premature disassembly of immature SGs by Ubap2l regulation (Figs. 4B–C). Immunofluorescence confirmed a marked reduction in the G3bp2-positive SG area (Fig. 4D).

Fig. 4.

Integrated multi-omics identifies NotoEV-mediated suppression of SGs through dual miRNA-mRNA and mTOR activation. (A) Proteomic workflow for differential protein analysis. (B) Volcano plot of significantly altered proteins (|log2FC|>1, p < 0.05, n = 3). (C) Gene enrichment of biological processes. (D) Decreased G3bp2 immunofluorescence intensity. (E-F) Western blotting of mTOR pathway-related proteins. (E) Analysis of G3bp2, p-mTOR, p-S6K, and p-4EBP1 in OGD/R-injured samples (NotoEV alone/combined with Compound C108, 4 µM). (F) Assessed p-mTOR, p-S6K, p-4EBP1 in OGD/R-injured samples (NotoEV alone/combined with rapamycin, 20 nM). (G) miRNA sequencing pipeline. (H–J) Validation of PC-3p-385136_6-G3bp2 interaction: (H) Targetscan score and binding energy for PC-3p-385136_6 targeting G3bp2. (I) Schematic of the dual-luciferase reporter vector (harboring G3bp2 sequence) and aligned sequences of G3bp2 3’UTR wild-type (WT1), miRNA PC-3p-385136_6, and G3bp2 3’UTR mutant (MUT1; mutated sites indicated). (J) Relative luciferase activity in the dual-luciferase reporter assay groups. Data are expressed as mean ± SEM; ns: p > 0.05, *p < 0.05, **p < 0.01, and ***p < 0.001

G3bp2 downregulation reduced TSC recruitment to lysosomes, thereby relieving Rheb inhibition and activating mammalian target of rapamycin complex 1 (mTORC1) [12]. Western blotting revealed reduced G3bp2 (but not G3bp1) and increased phosphorylated mTOR (p-mTOR) levels after NotoEV treatment (Figure S3B). Pharmacological experiments supported this pathway: The G3bp2 inhibitor C108 reduced G3bp2 levels dose-dependently, whereas rapamycin blocked mTORC1 activation (Figs. 4E–F and S3C–D).

Deep sequencing identified multiple plant miRNAs enriched in NotoEV that targeted SG-related genes (Fig. 4G; Table S3). Eleven miRNAs targeted the 3′UTR of G3bp2. Quantitative PCR confirmed that NotoEV reversed the OGD/R-induced G3bp2 mRNA upregulation (Figure S3A). Dual-luciferase assays demonstrated the direct binding of miRNA PC-3p-385136_6 to G3bp2 site 1 (Figs. 4H-J and S3E-F). Together, these results establish that NotoEV activates the G3bp2-mTOR signaling axis through cross-kingdom miRNA regulation.

NotoEV restored mitochondrial function and reduced apoptosis through the Bcl-2/TOM20 pathway

Given the strong mitochondrial phenotype observed in vitro, we assessed mitochondrial function. OGD/R caused reactive oxygen species accumulation, membrane potential loss, and cytochrome c release. NotoEV reversed these changes by upregulating TOM20, shifting the Bcl-2/Bax balance toward survival and reducing caspase-3 activation (Figs. 5A–C). Blocking Bcl-2 abolished these benefits, confirming its essential role.

Fig. 5.

NotoEV inhibits mitochondrial apoptosis in OGD/R-injured neurons. (A) Schematic representation of the mitochondrial protective mechanism. (B-C) Analysis of apoptosis-related proteins: (B) Western blotting of Bcl-2, Bax, and cleaved-caspase3 in OGD/R-injured samples (NotoEV treatment following 1.5 µM ABT-199 inhibition, or ABT-199 treatment alone). (C) Corresponding quantification of the above Western blotting bands. (D) TEM depicting mitochondrial ultrastructural preservation (M: mitochondria; arrows indicate crista integrity). (E-F) Bcl-2 mitochondrial translocation: Mitotracker colocalization analysis. (G-H) JC-1 assay demonstrated the preserved mitochondrial membrane potential. Data are expressed as mean ± SEM; ns: p > 0.05, *p < 0.05, **p < 0.01, and ***p < 0.001

TEM demonstrated that NotoEV restored cristae density and reduced mitochondrial swelling (Fig. 5D). Immunofluorescence revealed increased Bcl-2 localization to the outer mitochondrial membrane and colocalization with TOM20 (Figs. 5E–F). JC-1/mitotracker assays further confirmed restored mitochondrial potential and morphology, effects eliminated by the Bcl-2 inhibitor ABT-199 (Figs. 5G–H and S4A–B). These results demonstrate that NotoEV maintains mitochondrial homeostasis by Bcl-2/TOM20 axis.

NotoEV targeted ischemic lesions and promoted neurological recovery in vivo

In the middle cerebral artery occlusion (MCAO) model, DiD-labeled NotoEV accumulated in the ischemic core and penumbra and reduced the infarct volume by approximately 75% (p < 0.001) (Figs. 6A–C). This protection corresponded with an increase in NeuN-positive neurons (Fig. 6D). Notably, we observed no significant effects of NotoEV on astrocytes or microglia in this region, indicating that the protective action of NotoEV primarily targets neurons rather than glial cells (Figure S5A–B). NotoEV increased TOM20 and Bcl-2 colocalization (Fig. 6E), consistent with restored mitochondrial stability.

Fig. 6.

NotoEV stimulates functional neurorestoration through mTOR activation in an MCAO mouse model. (A-B) Biodistribution of DiD-labeled NotoEV in vivo: (A) Whole-body imaging; (B) Brain section fluorescence. (C) MRI quantification of infarct volume. (D-E) Immunofluorescence analysis: (D) NeuN images; (E) Bcl2/TOM20 colocalization. Boxed areas: Infarct core and penumbra (white dashed line). (F–H) Molecular analyses: (F) Representative Western blotting images of G3bp2 and mTOR pathway proteins (p-mTOR, p-S6K, p-4EBP1); (G) representative Western blotting images of apoptotic proteins (Bcl-2, Bax, cleaved caspase-3) and synaptic protein (PSD95); (H) Quantitative analysis of Western blotting-detected protein levels. Data are expressed as mean ± SEM; ns: p > 0.05, *p < 0.05, **p < 0.01, and ***p < 0.001 versus MCAO group

Western blotting confirmed decreased G3bp2 expression and enhanced mTORC1 activation and its downstream effectors, eukaryotic translation initiation factor 4E binding protein 1 (p-4EBP1) and phosphorylated S6 Kinase (p-S6K) (Fig. 6F). These molecular changes reduced Bax and cleaved caspase-3 levels, increased Bcl-2, and elevated PSD95 expression (p < 0.01), indicating reduced apoptosis and improved synaptic maintenance (Figs. 6G–H). Behaviorally, NotoEV-treated mice revealed improved motor coordination and fewer neurological deficits (Figures S6A–E).

Finally, liquid chromatography-mass spectrometry (LC-MS) demonstrated that classical saponins (notoginsenoside R1 and ginsenoside Rg1) were depleted to < 0.1 mg/L during EV extraction (Figures S7A–E). Nevertheless, NotoEV outperformed these monomers therapeutically (Figures S7F–G), confirming that neuroprotection is driven by miRNA-based mechanisms rather than residual small molecules.

Discussion

PEVs are emerging as scalable biocompatible platforms for therapeutic applications, offering distinct advantages in drug delivery and regenerative medicine. Despite their promise, the low productivity and inconsistent purity of isolated PEVs limit their application in preclinical and clinical settings. In this study, we developed an optimized isolation strategy that integrated ultracentrifugation with sucrose density gradient purification, producing PEVs at concentrations exceeding 10¹² particles/mL across four different plant species. Among these, NotoEV exhibited the highest yield and most stable cellular uptake, enabling detailed functional studies.

Using an OGD/R model, a well-established in vitro system for ischemic injury, NotoEV demonstrated superior neuroprotective effects compared to EVs from other plant sources. Consistent with previous studies on the neuroprotective properties of Panax notoginseng extracts [13], NotoEV significantly enhanced neuronal survival by upregulating anti-apoptotic proteins and suppressing pro-apoptotic pathways. Morphological analyses revealed that NotoEV treatment preserved dendritic complexity and promoted synaptic reconstruction, corroborated by electrophysiological assays demonstrating restored spontaneous firing and improved network synchronization.

A central mechanistic insight of this study lies in SGs and mTOR pathway regulation by plant-derived miRNAs delivered using NotoEV. SGs form in response to cellular stress, transiently sequestering mRNAs and proteins; however, persistent SGs contribute to neuronal dysfunction and cell death [14]. However, G3BP1 is traditionally recognized as the dominant isoform associating lysosomal TSC complex tethering to mTORC1 inhibition [12, 15]. Our findings highlight the unique pathological significance of G3BP2 in ischemic injury. Studies have highlighted that G3BP2 functions as a non-redundant nucleator of SG formation under environmental stressors. Unlike G3BP1, which often operates in a constitutive capacity, G3BP2 is dynamically recruited to RNA processing sites during cellular distress, including hypoxia or oxidative stress [16–19].

In the context of ischemic stroke, our data revealed that OGD/R injury specifically elevated G3BP2 levels, driving the aberrant SG assembly that sequesters the translation machinery and inhibits neuronal recovery. By targeting G3BP2, NotoEV efficiently dismantles these pathological condensates, liberating TSC complex from the lysosomal surfaces and reactivating mTORC1-dependent protein synthesis. This mechanism emphasizes a previously underappreciated therapeutic window in which modulating SG-specific nucleators can restore metabolic homeostasis without disrupting the baseline mTOR regulation governed by G3BP1.

NotoEV disrupts pathological SG assembly by downregulating the core nucleating proteins G3bp2, Ubap2l, and Lsm14a through a cross-kingdom RNA interference mechanism. This disruption improves mTORC1 inhibition, reactivates protein synthesis pathways, and enhances neuronal recovery. The direct mammalian SG-related gene regulation by plant miRNAs represents a novel therapeutic paradigm that expands cross-species RNA communication scope [20]. Notably, this miRNA-driven mechanism differentiates NotoEV from mammalian EVs, which primarily exert their effects through protein cargo modulation [21].

Beyond cytoskeletal repair, NotoEV confers mitochondrial protection, which is a crucial determinant of neuronal fate following ischemic injury. NotoEV treatment preserved mitochondrial membrane potential, restored cristae structure, and inhibited apoptosis by upregulating TOM20 and Bcl-2. These mitochondrial effects translated into reduced infarct volumes and enhanced neuronal survival in a mouse model of MCAO, confirming the therapeutic relevance of NotoEV in vivo.

Compared to previous plant- or mammal-derived EV therapies, NotoEV has several key advantages. Their plant origin ensures scalability and reduces immunogenic risks [22, 23], thereby overcoming the production and safety challenges associated with mammalian EVs. Furthermore, the specific disruption of pathological SGs by miRNA-mediated regulation offers higher precision than traditional pharmacological SG inhibitors. These findings align with recent advances in plant EV engineering, including structural droplet drug development that enhances BBB transcytosis [24]. In contrast to hydrophilic herbal monomers, which often exhibit poor BBB permeability [25, 26], PEVs provide efficient BBB penetration and enhanced bioavailability, highlighting their superiority in neurological applications [27–29].

Although this study established NotoEV as promising candidates for IS therapy, several limitations exist. Functional validation was restricted to acute ischemic models. Future studies using chronic stroke models characterized by persistent neuroinflammation and glial scar formation are required to assess long-term efficacy. Moreover, expanding studies into other neurological disorders, including traumatic brain injury, Alzheimer’s and Parkinson’s diseases, where mitochondrial dysfunction and proteostasis impairment are prominent, could further establish the breadth of therapeutic potential of NotoEV. Clinical translation requires rigorous evaluation of pharmacokinetics, dosing, and complex and human-relevant system immunogenicity.

Conclusions

This study demonstrates that NotoEV represents a novel, scalable, and mechanistically distinct neuroprotective platform. NotoEV offers a transformative approach for IS treatment and other neurodegenerative disorders by integrating morphological, functional, and molecular restoration through miRNA-driven regulation of SGs and mitochondrial homeostasis. These findings lay the groundwork for the future application of PEVs as next-generation therapeutics in materials science and regenerative medicine.

Materials and Methods

Materials

DiD perchlorate (DiD) was obtained from GLPBIO (Montclair, CA, USA). Poly-L-lysine hydrobromide was bought from Sigma-Aldrich (St. Louis, MO, USA). Neurobasal Medium SFM was provided by GIBCO, Thermo Fisher Scientific (Grand Island, NY, USA). DNase I was procured from Roche Diagnostics (Basel, Switzerland). The JC-1 mitochondrial membrane potential assay kit was acquired from Solarbio (Beijing, China). MitoTracker Deep Red was obtained from MedChemExpress (Monmouth Junction, NJ, USA). Fetal bovine serum (FBS) was sourced from Procell Life Science & Technology (Wuhan, China).

Animals

Fifty male C57BL/6 mice (6–8 weeks, 22–25 g) were bought from Zhuhai Bestest Biotechnology Co., Ltd. (Zhuhai, China). Animals were housed under controlled temperature and humidity with ad libitum access to food and water. Group allocation followed a randomization sequence generated using statistical software, and allocation concealment was maintained by a researcher not involved in downstream experiments.

All procedures complied with the National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals (NIH Publication No. 8023, revised 1978) and the guidelines of Jinan University. Every effort was made to minimize animal use and reduce pain or distress. All protocols were approved by the Institutional Animal Care and Use Committee of Jinan University (Approval ID: 20210702-16).

Methods

PEV extraction and characterization

Fresh roots and rhizomes of Panax notoginseng, Panax ginseng, Ligusticum chuanxiong, and Gastrodia elata were washed, ground with phosphate-buffered saline (PBS), and filtered through 200-mesh gauze to collect the plant juices. The juices were subjected to gradient centrifugation at 1000 × g (15 min), 2500 × g (15 min), 5000 × g (30 min), and 10,000 × g (60 min), discarding the precipitates each time. The supernatant was filtered through a 0.22 μm membrane, concentrated using a 100 kD ultrafiltration tube, and ultracentrifuged at 120,000 × g for 1 h. The precipitate was resuspended in 1 mL PBS.

Sucrose cushions (8%, 30%, 45%, and 60% in 20 mM Tris-hydrogen chloride) were layered sequentially. Briefly, 1 mL of the resuspended precipitate was added to the top and centrifuged at 100,000 × g for 1 h. Sucrose bands at 30%, 45%, and 60% were collected, mixed, and centrifuged at 100,000 × g for 1.5 h. The final precipitate was dissolved in 500 µL of PBS to obtain an exosome solution.

For characterization, Coulter technology with a chip was used to detect the exosome concentration and particle size from the four plants. The zeta potential was measured using electrophoresis. Transmission Electron Microscopy (TEM) combined with phosphotungstic acid staining was used to characterize the morphological features of the four EV types. Additionally, EV verification was confirmed using Western blotting.

Notoginsenoside R1 and ginsenoside Rg1 quantification in NotoEV

To exclude the potential influence of residual saponin on NotoEV’s efficacy, three distinct fractions were analyzed for ginsenoside content: Plant Juice (original supernatant), post-purification supernatant (supernatant following isolation), and NotoEV (final pellet). Quantification was performed using the optimized LC-MS protocol described below.

An optimized LC-MS protocol was used in this study. A Quadrupole Time-of-Flight (Q-TOF) LC/MS system (Synapt G2 Si, Waters) was used for detection. Chromatographic separation was achieved using an ACQUITY UPLC BEH C18 column (2.1 × 100 mm, 1.7 μm) maintained at 35 ℃. The mobile phase consisted of 0.1% formic acid in water (solvent A) and 0.1% formic acid in acetonitrile (solvent B). A gradient elution program was applied as follows: 0–1 min, 10% B; 1–10 min, 10%–95% B; 10–12 min, 95% B; 12–12.1 min, 95%–10% B; 12.1–15 min, 10% B. The flow rate was set to 0.3 mL/min. The mass spectrometry parameters were as follows: Positive ion mode (+ ESI), capillary voltage (2.5 kV), sampling cone voltage (35 V), source temperature (120 ℃), desolvation temperature (350 ℃), cone gas flow (40 L/h), and desolvation gas flow (800 L/h). The standard stock solutions of notoginsenoside R1 and ginsenoside Rg1 were prepared at 10 ppm for quantification.

Cell uptake assay

EVs from Panax notoginseng, Panax ginseng (GsEV, vesicle population), Ligusticum chuanxiong (CxEV, vesicle population), and Gastrodia elata (GtEV, vesicle population) were labeled with DiD. EVs were incubated with DiD (5 μM) at 4 ℃ for 30 min with gentle inversion every 5 min. After labeling, the samples were divided into four ultracentrifuge tubes, brought to volume with PBS, and ultracentrifuged at 100,000 × g for 3 h to remove unbound dye. The EV pellet was resuspended in PBS.

The primary cortical neurons were seeded in confocal dishes and maintained in neurobasal medium supplemented with 2% B27 and 1% GlutaMAX at 37 ℃ and 5% CO₂. Half-medium changes were performed every three days. DiD-labeled NotoEV, GtEV, GsEV, or CxEV were added at 10⁶ particles/cell. After 6 h, the cells were fixed with 4% paraformaldehyde (PFA), counterstained with 4′,6-diamidino-2-phenylindole (DAPI), and imaged using confocal microscopy.

Cytotoxicity assay

The SY5Y cells were seeded in 96-well plates at an 8,000 cells/well density and cultured at 37 ℃ under 5% CO₂ for 24 h until the cells were fully adherent. After 24 h, the medium was replaced with a PEV-containing medium. The PEV administrations were set to 10⁶, 10⁵, 10⁴, 10³, and 10² particles/cell, respectively, with five replicate wells prepared for each concentration. After continuous culturing for another 24 h, cell counting kit 8 (CCK-8) solution was added, and the absorbance at 450 nm was measured to calculate cell viability and determine the appropriate administration concentration.

MEA recording

Primary neurons (8 × 10⁴ cells/well) were plated in 10 µL droplets on poly D-lysine/laminin-coated CytoView 48-well MEA plates (Axion Biosystems). After 1 h of attachment, neurobasal medium was added to a final volume of 200 µL, with half-media changes every three days. At 8 days in vitro, the baseline neuronal activity was recorded for 30 min.

OGD was induced by replacing the culture medium with glucose-free Dulbecco’s modified Eagle medium (DMEM) and exposing the cells to a gas mixture of 94% N₂, 1% O₂, and 5% CO₂ for 45 min. The medium was then replaced with neurobasal, and NotoEV was applied at 2,000 particles/cell during reperfusion. Untreated cultures served as the OGD/R control. Each group included 7–10 wells.

After 24 h of reperfusion, neuronal activity was recorded. Action potentials were detected using a threshold of six standard deviations above the background noise. The Axion Integrated Studio software was used for quantitative analysis, including spike rate, burst rate, network burst activity, and synchrony index.

Immunofluorescence and neuronal morphology

The cells were fixed with 4% PFA for 15 min, washed with PBS, permeabilized with 0.5% Triton X-100 for 15 min, and blocked with 2% bovine serum albumin for 30 min. Primary antibodies were applied overnight at 4 ℃. After washing, fluorescent secondary antibodies were added for 1 h at room temperature. The samples were counterstained with DAPI and imaged using Olympus or Zeiss microscopes.

G3bp2/MAP2 co-stained neurons were imaged on a Zeiss high-resolution system. MAP2, PSD95, synaptophysin, TOM20, and Bcl-2 staining were imaged using an Olympus system. For cryosectioned brain tissue, sections were permeabilized for 30 min, incubated with primary antibodies (Bcl-2, TOM20, and NeuN) overnight at 4 ℃, washed, incubated with secondary antibodies, and imaged using fluorescence microscopy.

Live-Cell imaging

Neurons (3.5 × 10⁶ cells/35 mm dish) were cultured for 5 days before OGD induction. During reperfusion, NotoEV (2,000 particles/cell) was added, and neurons were imaged every 10 min for 24 h using a live-cell imaging system at 37 ℃ and 5% CO₂.

Mitochondrial analysis

Neurons (5 × 10⁵ cells/confocal dish) were assigned to Ctrl, OGD/R, or OGD/R + NotoEV groups. After 24 h of reperfusion, the cells were stained with MitoTracker Deep Red (100 nM, 30 min) and JC-1 (20 min) [32, 33]. The cells were then washed and imaged using a super-resolution microscope. ImageJ software was used for the colocalization analysis of MitoTracker, JC-1 monomers, and JC-1 aggregates.

Proteomics

Neurons from Ctrl, OGD/R, and NotoEV groups (n = 3/group) were collected for protein extraction. For every 5 µL sample, 30 µL lysis buffer (6 M urea and 2 M thiourea) was added. Proteins were denatured, reduced with TCEP, alkylated with IAA, digested, desalted, and analyzed using liquid chromatography-tandem mass spectrometry.

Peptides were analyzed using ultra-high-performance liquid chromatography (Bruker Daltonics, Germany) coupled to a timsTOF HT mass spectrometer. Data processing and protein identification were performed using FragPipe software.

Inhibitor treatments

The primary cortical neurons (6 × 10⁵ cells/well) were cultured for five days before pharmacological treatment. Cells were treated with ABT-199 (1.5 µM, 48 h), Compound C108 (4 µM, 48 h) [34], or rapamycin (20 nM, 24 h) [35]. OGD/R was induced after the drug treatment.

The experimental groups included Ctrl, OGD/R, OGD/R + NotoEV, ABT-199, ABT-199 + NotoEV, C108, C108 + NotoEV, rapamycin, and rapamycin + NotoEV. After 24 h of reperfusion, the cells were lysed using radioimmunoprecipitation assay buffer(RIPA buffer) supplemented with protease/phosphatase inhibitors for Western blotting.

Luciferase reporter assay

Wild-type (WT) and mutant (MUT) G3BP2 3′-UTR (seed region substituted) were cloned into luciferase vectors and verified using Sanger sequencing. HEK293T cells were cultured in DMEM containing 10% FBS.

Cells were seeded in 24-well plates (5 × 10⁴ cells/well) and cultured for 24 h prior to co-transfected with 1 µg per well of G3BP2-3′UTR-WT/MUT and 50 nM PC-3p-385136_6 or negative control mimic using Lipofectamine 3000. After 48 h, firefly and Renilla luciferase activities were quantified using the Dual-Luciferase Reporter Assay System (Promega), and the ratio of firefly to Renilla luciferase activity was calculated.

TEM

The neurons were fixed in TEM fixative for 30 min, washed, and embedded in agarose. The samples were post-fixed in 1% OsO₄ for 2 h, dehydrated through ethanol and acetone gradients, and embedded in Embed 812 resin. Ultrathin Sects. (60–80 nm) were mounted on formvar-coated copper grids and stained with uranyl acetate (8 min) and lead citrate (8 min). After drying, the samples were imaged using a Hitachi HT7800 transmission electron microscope.

MCAO

The mice were anesthetized with isoflurane (5% induction, 1.5% maintenance). After cervical midline incision and isolation of the carotid arteries, a silicone-coated nylon filament (model 2000AAA, Jialing) was inserted to occlude the middle cerebral artery for 60 min. Reperfusion was initiated by withdrawing the filaments. Mice received 1 mL saline intraperitoneally and recovered in a 37 ℃ chamber before being returned to their cages.

In vivo uptake of NotoEV

Mice received intranasal NotoEV (30 µL, 7.31 × 10¹² particles/mL; n = 4) or PBS (30 µL; n = 4). MCAO was induced 20 min later. After MRI acquisition, the mice were perfused with saline and fixed with 4% PFA. Brains were cryoprotected in 15%–30% sucrose, embedded in optimal cutting temperature compound, and sectioned at 10 μm.

MRI acquisition

MRI was performed using a 9.4 T Bruker PharmaScan scanner. The mice were anesthetized with 2% isoflurane, and their vital signs were continuously monitored. T2-weighted images were obtained 24–72 h after MCAO using a fast spin-echo sequence (TR/TE = 3500/33 ms; FOV 20 × 20 mm; 17 slices; slice thickness 1 mm; matrix 256 × 256). Infarct volumes were quantified using a 3D Slicer, with reconstruction from T2WI and threshold-assisted segmentation.

In vivo tracing of NotoEV

Mice (n = 3/group) received DiD-labeled NotoEV (1.5 × 10¹²particles/mL, 150 µL) or PBS via the tail vein injection after MCAO. Three hours later, whole-body fluorescence imaging was performed under anesthesia using an IVIS Lumina III. After euthanasia, ex vivo imaging of the brain sections was performed to assess blood-brain barrier penetration and parenchymal distribution.

Mice behavioral tests

After the mice were numbered, they were randomly divided into three groups (n = 5 per group) : Sham, MCAO (PBS, 30 µL), and MCAO+NotoEV (30 µL, 7.31 × 10¹² particles/mL). PBS or NotoEV was administered at 60 min after MCAO model reperfusion, once daily for 3 consecutive days. Behavioral assessments were conducted by two researchers, one of whom was blinded to the group allocation to minimize observer bias.

Grid-Walking test

The grid-walking test was used to evaluate the motor coordination and gait deficits. A raised wire grid apparatus was used (grid size: 12 mm squares; dimensions: 32 × 20 × 50 cm; length × width × height). Each mouse was placed individually on the grid and allowed to move freely for 3 min. A camera was positioned beneath the grid to record forelimb and hindlimb placement. The number of foot faults and non-fault steps for each limb was quantified from the video recordings. The percentage of foot faults was calculated as follows: Number of foot faults/(number of foot faults + number of non-faults) × 100%. Higher ratios indicate more severe motor impairments.

Cylinder test

The cylinder test was used to assess forelimb use asymmetry. Mice were placed individually in a transparent glass cylinder (height, 15 cm; diameter, 10 cm) and videotaped for 5 min. The number of contacts made with the cylinder wall using the left forelimb, right forelimb, or both forelimbs simultaneously was measured. Forelimb asymmetry was calculated using the following formula: (Number of left hand − number of right hand)/ (number of right hand + number of left hand + number of both hands). Positive values indicated a preference for left forelimb use, whereas negative values indicated right-sided preference.

Modified neurological severity score (mNSS) scoring

Neurological function was assessed using mNSS. The mNSS included assessments of motor function, sensory function, reflexes, and balance, with scores ranging from 0 to 18. Higher scores reflected more severe neurological deficits. Evaluations were performed by trained personnel at the predetermined time points.

In vivo administration of ginsenoside Rg1 and notoginsenoside R1

Saponins were intraperitoneally administered as a single dose immediately after reperfusion in the MCAO model, using a solution which comprised ginsenoside Rg1 and notoginsenoside R1 (both at 168 mg/kg) dissolved in saline [36, 37].

Western blotting analysis

Cells or tissue samples were harvested, and total protein was isolated following the protocol described in a prior study [38]. Protein concentrations were quantified using a BCA protein assay kit (Thermo Fisher Scientific, USA). Aliquots of 5–15 µg total protein were resolved via SDS-PAGE electrophoresis, after which the separated proteins were transferred onto nitrocellulose (NC) membranes. The membranes were blocked in 5% bovine serum albumin (BSA) solution, then probed with specific primary antibodies at 4 °C overnight. Following three 10-minute rinses with TBST buffer, the membranes were incubated with secondary antibodies for 1 h at ambient temperature. After final washing steps, protein band signals were detected and densitometrically analyzed to quantify protein expression levels.

Data analysis

All statistical analyses were conducted using R software (version 4.2.1). Visualization was performed using ggplot2 (version 3.4.4). Two-group comparisons were performed.

• Welch’s t-test (normal distribution + unequal variance).

• Wilcoxon rank-sum test (non-normal distribution).

For ≥ 3 groups:

• One-way analysis of variance (ANOVA) (normal distribution + equal variance).

• Welch’s ANOVA (normal distribution + unequal variance).

• Kruskal–Wallis test (non-normal distribution).

Sholl analysis was performed using two-way repeated-measures ANOVA. Representative images were selected based on a typical and unbiased presentation of experimental trends.

Author contributions

Hongcheng Mai and Dan Lu conceived the idea and designed the experiment. Yuanyuan Yu, Na Tan, Zhifeng Xu performed experiments. Zhijian Tan, Tao Wang, Huimin Liu analysed the data. Yamei Tang improved the manuscript.

Funding

This work was supported by grants from National Natural Science Foundation of China (82503549)，Guangdong Basic and Applied Basic Research Foundation (2025B1515020086, China), Noncommunicable Chronic Diseases-National Science and Technology Major Project (2024ZD0533700, China)，Sun Yat-sen University Hundreds of Talent Program (1320324001, China) to Hongcheng Mai; This work was supported by grants from the STI 2030 Major Projects (2022ZD0211603), National Natural Science Foundation of China (82330099, 82530100), the Key Area Research and Development Program of Guangdong Province (2023B0303040003), and Science and Technology Program of Guangzhou (2023A03J0708) to Yamei Tang; the “Guangdong Special Support Program” (referred to as the “Guangdong Tezhi Plan”) from the Provincial Health Commission (0720240214), the National Natural Science Foundation of China (82271304, 81801150, 81971121, 82171316 and 81671167) to Dan Lu.

Data availability

The datasets generated and analyzed during this study are available in the ProteomeXchange Consortium repository via the iProX [30, 31] partner repository with the dataset identifier PXD064044, and the miRNA sequencing data are available in the NCBI Sequence Read Archive repository under the accession number PRJNA1290867 (https://dataview.ncbi.nlm.nih.gov/object/PRJNA1290867?reviewer=p8bli46sga7jukrut5s3s1eai3).

Declarations

Ethics approval

All animal procedures were approved by the Institutional Animal Care and Use Committee of Jinan University (approval ID: 20210702-16). The study protocol conformed to the ethical guidelines of the NIH Guide (NIH Publications No. 8023, revised 1978) for the Care and Use of Laboratory Animals.

Competing interests

The authors declare no competing interests.