Engineering integrin αvβ8-targeted extracellular vesicles to deliver BDNF mRNA for motor recovery in spinal cord injury

Abstract

Spinal cord injury (SCI) remains difficult to treat, and current interventions provide limited functional restoration and often require invasive procedures. Existing cell- or extracellular vesicles (EV)-based approaches are frequently administered alongside surgery, limiting therapeutic reach and overall efficacy. In this study, we developed an engineered extracellular vesicle (EV) platform by displaying a single-chain variable fragment (scFv) against integrin αvβ8 (αITGEV) and loading brain-derived neurotrophic factor mRNA (mBDNF). The construct maintained canonical EV identity and morphology, and showed predominant single particle co-positivity for targeting ligand and cargo. In neuron–microglia co-culture, mBDNF@αITGEV preferentially entered both cell types under injury-relevant stress, shifted microglia toward a repair-associated phenotype, reduced TNF-α and IL-1β, increased IL-4 and IL-10, and preserved neuronal architecture. Our results indicate that mBDNF@αITG-EVs significantly promote functional motor recovery by modulating the inflammatory microenvironment and inhibiting neuronal ferroptosis. Mechanistically, the delivery of BDNF mRNA bolstered GPX4 expression and stabilized mitochondrial dynamics, thereby mitigating secondary oxidative damage. This study provides a non-invasive strategy for precision nanomedicine in neuro-regeneration. Collectively, this study supports a non-invasive systemically administered, targeted EV–mRNA therapeutic strategy for spinal cord injury with translational potential.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12951-026-04222-7.

Introduction

Spinal cord injury (SCI) remains a devastating neurological condition, leading to catastrophic functional deficits and permanent disability [1]. The initial mechanical trauma triggers a cascade of secondary injuries including ischemia, oxidative stress, and chronic neuroinflammation, which in turn creates a hostile microenvironment severely impeding endogenous neural repair and regeneration. Consequently, current therapeutic strategies offer limited benefits, highlighting an urgent need for innovative interventions that precisely modulate the complex post-injury pathology. Current clinical treatments for spinal cord injury include surgery, medication, cell therapy, rehabilitation therapy, and synergistic therapy [2]. Current cell therapy must be performed alongside surgery and in situ of injury [3]. As a result, the lack of access to cell therapy for most patients hinders its widespread clinical reach. In recent years, numerous studies have demonstrated that the use of nanomaterials as a non-invasive treatment for spinal cord injury significantly differs from traditional in situ surgical approaches [4]. This is primarily due to their ability to effectively cross the blood-spinal cord barrier, their high transport efficiency, and their ability to carry drugs, offering significant advantages for the treatment of spinal cord injuries [5].

Extracellular vesicles (EV) have recently emerged as a powerful nanoscale platform for regenerative medicine [6–9]. These natural lipid bilayer nanovesicles, typically 90–150 nm in diameter, act as intercellular communicators by transferring a rich cargo of proteins, lipids, and nucleic acids to recipient cells [10–13]. In recent years, various studies have also demonstrated that EVs are highly beneficial for nerve regeneration, mainly by promoting axon regeneration, reducing inflammation, and maintaining nerve cell activity [14–16]. In the context of SCI, mesenchymal stem cell (MSC)-derived EV have shown considerable promise by attenuating neuroinflammation and promoting neuronal survival [17–20]. MSC-derived EVs have various advantages and flexible delivery methods that exert several effects to reduce apoptosis and inflammation, promote tissue regeneration, and increase neural plasticity after SCI [21–23]. In addition, EV play a critical role in SCI repair by regulating the immune microenvironment via their bioactive molecules, such as anti-inflammatory factors, miRNAs, and antioxidant enzymes [24]. Although various studies in the past have demonstrated that MSC-derived EVs have great advantages, their effect on spinal cord tissue repair remains limited even when injected in situ into the injury site [25–27]. However, clinically translating native EVs is critically hampered by their inherent limitations as therapeutic agents. Lacking target specificity, they suffer from rapid systemic clearance, off-target biodistribution, and an inability to efficiently cross the blood–spinal cord barrier (BSCB), which collectively diminish their therapeutic payload at the lesion site, curtailing their efficacy [28].

In order to enhance EVs’ targeting ability and bypass undesirable side effects, various studies focused on generating and decorating the surface modification of EVs [29–32]. Nanoengineering provides a transformative solution to enable target-specific modification [33–35]. By functionalizing the surface of EV with targeting ligands, creating sophisticated nanocarriers that navigate the systemic circulation, recognize specific molecular signatures within the injured tissue, and execute lesion-specific delivery becomes possible [31]. Integrins, receptors for the extracellular matrix (ECM), comprise of an α and β transmembrane subunits that can engage with the ECM and modulate intracellular signaling function [36]. The SCI microenvironment is characterized by significant integrin αvβ8 (ITG) upregulation, which plays a pivotal role in cellular adhesion and signaling [37]. The injury-specific molecular landscape presents a prime target for engineered EVs. Concurrently, brain-derived neurotrophic factor (BDNF) is a potent neurotrophin essential for neuronal survival and axonal regeneration; however, its therapeutic application is constrained by poor stability and inefficient delivery [38–40].

Herein, we develop, design, and synthesize a novel engineered EV-based therapeutic platform specifically tailored for SCI repair. We functionalized EV with an integrin antibody (αITG) to confer active targeting capabilities and co-loaded them with BDNF mRNA (mBDNF) to provide robust neurotrophic support (coded as mBDNF@αITG EV). Figure 1 depicts the overall concept and workflow, including αITG-displaying EV production, mBDNF loading, intravenous administration after contusive SCI, integrin-guided homing to the lesion, and uptake by neurons and microglia, followed by microglial polarization toward an M2 state, reduced ferroptotic stress, and enhanced neuronal differentiation. This study also demonstrated that mBDNF@αITG EV can target an injured spinal cord. Based on these considerations, we hypothesized that an engineered EV platform integrating lesion-associated targeting with therapeutic mRNA delivery would improve EV localization to the injured spinal cord and mitigate secondary injury cascades by suppressing neuroinflammation and repairing mitochondrial dysfunction, thereby supporting functional recovery. The objectives of this study were to (i) construct and characterize αITG surface-displayed EV, (ii) load EV with mBDNF mRNA and evaluate delivery and uptake in relevant cellular models, and (iii) assess biodistribution and therapeutic efficacy in a contusive spinal cord injury model using functional, histological, and molecular readouts.

Fig. 1.

Schematic overview of the mBDNF@αITG EV platform and study workflow for spinal cord injury (SCI). Engineered HEK293T producer cells express a CD63-based fusion displaying an anti-integrin (αvβ8) scFv on the EV surface to generate αITG EV, which are then loaded with mBDNF cargo. Following intravenous injection after contusive SCI, the EV home to the lesion where integrin αvβ8 is enriched, bind target cells, and deliver BDNF within the spinal parenchyma. The working model illustrates post-delivery actions at the injury site, including engagement of neurons and microglia, polarization of microglia toward an M2-like phenotype, mitigation of ferroptosis, and support of neural differentiation and neurite outgrowth

Materials and methods

Plasmid construction and engineered EV production

To generate integrin-targeting EV (αITG EV), a plasmid was constructed by fusing a scFv sequence specific for integrin into the second extracellular loop of the human CD63 gene. HEK293T cells (human embryonic kidney 293 T) were used as producer cells and cultured in Dulbecco’s modified Eagle’s medium (DMEM, high glucose) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin at 37 °C in a humidified incubator with 5% CO₂. Cells were transfected with the engineered plasmid using a standard lipofection protocol when reaching approximately 70–80% confluence.

After transfection, the medium was replaced with EV-depleted culture medium, and conditioned media were collected after 48 h. Conditioned medium was collected and clarified by low-speed centrifugation to remove cells and debris, followed by filtration. EVs were then concentrated and buffer-exchanged using tangential flow filtration (TFF) equipped with a MWCO 300 kDa membrane. The EV fraction was concentrated to 50-fold, followed by diafiltration against sterile PBS to remove soluble proteins and residual reagents. The final EV preparation was sterile-filtered and stored at −20℃ for subsequent experiments.

Binding affinity

The binding affinity of αITG EV to recombinant integrin protein was quantified using surface plasmon resonance (SPR). Various concentrations of EVs were flowed over a sensor chip immobilized with integrin, and the association and dissociation rates were recorded to calculate the equilibrium dissociation constant.

Physicochemical EV characterization

The morphology of the engineered EV was examined using transmission electron microscopy (TEM). Vesicles were adsorbed onto formvar–carbon-coated copper grids and negatively stained with 2% uranyl acetate. The particle size distribution and concentration were determined by nanoparticle tracking analysis (NTA) using a ZetaView^® instrument. Protein expression was confirmed by western blot analysis. EV lysates were probed with primary antibodies against the αITG scFv, canonical EV markers (CD63, CD81, CD9, Alix), and a negative control cellular marker (calnexin).

Loading of BDNF mRNA into EV

mBDNF (6 µg) was added to EV samples containing 1 × 10¹¹ particles, and the volume was adjusted to 50 µL using PBS filtered through a 0.22-µm membrane. Subsequently, 50 µL of Lonza Solution was added and mixed thoroughly. The mixture was transferred to a Nucleocuvette^™, and electroporation was performed according to the experimental parameters. The samples were then incubated at 37 °C for 1 h. After incubation, the samples were transferred to 3 K centrifugal filters and centrifuged at 12,000 × g for 15–30 min at 4 °C. The remaining supernatant in the upper chamber was collected and designated as mBDNF@αITG EV. The encapsulation efficiency of mBDNF within αITG EVs was determined using high-performance liquid chromatography (HPLC). The co-localization of the αITG ligand and mBDNF cargo on a single-vesicle basis was analyzed using a nano-flow cytometer after fluorescently labeling each component.

EV-associated RNA was extracted from mBDNF-loaded EVs using the Total Exosome RNA and Protein Isolation Kit. Absolute quantification of mBDNF was performed via a one-step RT-qPCR process using PrimeTime^™ One-Step RT-qPCR Master Mix with mBDNF-specific primers. The mBDNF amounts were determined by interpolating Ct values into a standard curve generated from serial dilutions of mBDNF standards.

EV storage conditions and stability assessment

Purified EVs were stored as single use aliquots to minimize freeze thaw stress. EVs were mixed with a pre prepared cryopreservation solution containing Polysorbate 80 0.1 mg/mL, Histidine Hydrochloride Monohydrate 10 mM, and Sodium Chloride 4.38 mg/mL, adjusted to pH 7.4. Samples were frozen using a programmed cooling procedure at a rate of 1 °C per minute until reaching −20 °C, and then stored at −20 °C until use. Stability against freeze thaw stress was evaluated by subjecting aliquots to three freeze thaw cycles, followed by nanoparticle tracking analysis to quantify residual particle percentage.

High resolution imaging

EV characterization was performed using the EV profiler 2 kit (Oxford Nanoimaging, Oxford, UK) on functionalized ONI assay chips, which are compatible with downstream super-resolution imaging using the ONI Nanoimager system. To encourage EV binding, 10 µL of the EV suspension was then added to each lane. After being fixed for 10 min with 20 µL of fixative, the collected EVs were washed thoroughly. Prior to immunolabeling, 10 µL of staining buffer was added to each lane following fixation, and they were incubated for 10 min. Primary antibodies against the ScFv, conjugated to Alexa Fluor 488, were used for detection. The dSTORM imaging buffer was made fresh right before imaging by mixing 1 µL of reconstituted Part B with 99 µL of thawed Part A. To allow for oxygen depletion, a 20 µL aliquot of the prepared imaging buffer was added to each lane and incubated for 10 min. After the lanes were sealed, the ONI Nanoimager was used for imaging. To visualize mitochondrial morphology, cells were immune-stained for TOMM20. Cells were fixed with 4% paraformaldehyde for 15 min at room temperature, followed by permeabilization with 0.1% Triton X-100 for 15 min to facilitate antibody and dye penetration. After blocking, the cells were incubated with primary antibodies against TOMM20 diluted in blocking buffer overnight at 4 °C. Following PBS washes, cells were incubated with a goat anti-mouse IgG/IgM (H + L) secondary antibody conjugated to Alexa Fluor^™ 647 for 1 h at room temperature in the dark.

Flow cytometry analysis

To ensure the reproducibility of our flow cytometry findings, all data were analyzed through a standardized hierarchical gating strategy. Initial events were identified via forward scatter (FSC-A) and side scatter (SSC-A) characteristics to isolate the target cell populations while excluding cellular debris. For immunomodulatory analysis, microglial polarization was quantified by assessing the expression of M1 (CD86) and M2 (CD206) markers using specific fluorescently conjugated antibodies. Intracellular reactive oxygen species (ROS) levels within co-cultures were evaluated following incubation with a ROS-sensitive fluorescent probe. Additionally, mitochondrial membrane potential was monitored via JC-1 dye staining to assess cellular metabolic status. For the characterization of engineered vesicles, the co-localization of the αITG ligand and mBDNF cargo was analyzed on a single-vesicle basis using a CytoFLEX Nano Flow Cytometer (Beckman Coulter, Indianapolis, IN, USA). All data acquisition and post-hoc analyses, including doublet discrimination and gate boundary settings, were performed using FlowJo^™ software (version 10), ensuring statistical precision across all experimental replicates.

Microglia differentiation

The human microglial cell line was cultured in Micorglia Medium (MM) (Sciencell, CA, USA) supplemented with 10% heat-inactivated Fetal Bovine Serum (FBS) (Gibco) and 1% Penicillin-Streptomycin (Gibco). Cells were maintained in a humidified incubator at 37 °C with 5% CO₂. For polarization experiments, microglia cells were seeded into 6-well plates at a density of 5 × 10⁵ cells/well and allowed to adhere overnight. The following day, the medium was replaced with fresh complete medium containing specific polarizing stimuli: M0 (Resting State) cells were treated with fresh medium only or vehicle control (e.g., sterile PBS); M1 Polarization (Classical Activation) was induced by stimulating with 100 ng/mL Lipopolysaccharide (LPS, Sigma-Aldrich, St. Louis, MO, USA) and 20 ng/mL recombinant murine Interferon-gamma (IFN-γ, PeproTech, Cranbury, NJ, USA); and M2 Polarization (Alternative Activation) was induced with 20 ng/mL recombinant murine Interleukin-4 (IL-4, PeproTech) and 20 ng/mL recombinant murine Interleukin-13 (IL-13, PeproTech). Following a 24 h incubation with the polarizing stimuli, cells were harvested for flow cytometry analysis.

Generation of iPSC-derived neurons and neuron–microglia co-culture

Human-induced pluripotent stem cells (iPSCs) were used to generate neuronal cultures for subsequent co-culture experiments. iPSCs were maintained on culture plates coated with Geltrex^® LDEV-Free, hESC-qualified, Reduced Growth Factor Basement Membrane Matrix (Cat# A1413302), which was diluted 1:100 in DMEM/F-12 medium. The complete culture medium used for routine maintenance was StemFlex^™ medium, consisting of StemFlex^™ basal medium and StemFlex^™ supplement. For neural induction, iPSCs were seeded in a 6-well Geltrex-coated plate, and the StemFlex^™ medium was replaced with PSC neural induction medium, which is composed of neurobasal medium and neural induction supplement after 24 h. The cells were cultured for 7 days to facilitate differentiation into neural stem cells (NSCs), with fresh medium changes every 2 days. On day 7, the newly generated NSCs were harvested and subsequently expanded in StemPro^™ NSC SFM complete medium. This expansion medium consisted of KnockOut^™ DMEM/F-12 medium and StemPro^™ neural supplement. For terminal differentiation into mature neurons, expanded NSCs were cultured in a differentiation medium composed of neurobasal medium, B-27^™ supplement, and GlutaMAX^™ supplement for 4 weeks to ensure neuronal maturation. The human microglia cell line was used for co-culture experiments. To investigate paracrine signaling, a non-contact Transwell co-culture system was established. The monolayer of microglia was seeded onto 0.4 μm pore size polycarbonate membrane inserts, which were then placed into 24-well plates containing a pre-seeded iPSC-derived neurons.

Cell culture under hypoxic conditions

To establish an in vitro SCI model, hypoxic conditions were applied to simulate the oxygen-deprived microenvironment that occurs after SCI. For hypoxia treatment, all cells were transferred to a hypoxia chamber and exposed to a gas mixture composed of 92% N₂, 3% O₂, and 5% CO₂ at 37 °C for the indicated time intervals. Following exposure, the cells were either immediately harvested for subsequent analyses or returned to normoxic conditions (21% O₂, 5% CO₂) for recovery experiments, depending on the experimental design.

EV uptake visualization

αITG EV were fluorescently labeled using the ExoGlow-Protein EV Labeling Kit (System Biosciences). Briefly, the isolated EV pellet was dissolved in 500 µL of PBS, and 1 µL of 500× ExoGlow-Protein dye was added. The suspension was incubated at 37 °C with gentle agitation for 20 min to allow efficient labeling. Afterward, the labeled vesicles were recovered by centrifugation for 10 min and resuspended in 200 µL of sterile PBS. For cellular uptake experiments, 1 × 10⁴ cells were seeded onto µ-Slide 8-well chambers (ibidi, Gräfelfing, Germany) and treated with the fluorescently tagged EV. The cells were subsequently fixed with 4% paraformaldehyde, counterstained with DAPI to visualize nuclei, and labeled with Alexa Fluor^™ 594 phalloidin to reveal the actin cytoskeleton at specific time intervals. Confocal imaging was conducted using a Leica TCS SP8 X microscope (Leica, Wetzlar, Germany).

intracellular uptake and lysosomal colocalization assay

To investigate the internal trafficking of EVs, mBDNF-loaded EVs were first fluorescently labeled with ExoGlow-Protein EV Labeling Kit according to the manufacturer’s instructions. The were plated onto µ-Slide 8-well chambers (ibidi) and allowed to reach 70% confluency. The labeled EVs were then added to the culture medium and incubated with the cells for 16 h at 37 °C. To visualize lysosomal localization, the cells were treated with 50 nM LysoTracker^™ Deep Red (Thermo Fisher Scientific) for 30 min prior to the end of the incubation period. Following incubation, the cells were washed three times with PBS to remove unbound EVs, fixed with 4% paraformaldehyde (PFA) for 15 min, and counterstained with DAPI to identify the nuclei. The spatial distribution and colocalization of EVs within the lysosomal compartments were analyzed using ONI Nanoimager.

In vitro immunomodulation and neuroprotection assays

To assess immunomodulatory effects, the neuron–microglia co-cultures were treated with different EV formulations. After treatment, microglial polarization was analyzed by flow cytometry using fluorescently conjugated antibodies against M1 (CD86) and M2 (CD206) markers. Culture supernatants were collected to quantify secreted cytokine levels, including TNF-α, IL-1β, IL-4, and IL-10, using enzyme-linked immunosorbent assay (ELISA) kits. A broader cytokine profile (IFN-γ, IL-1β, IL-6, IL-8, IL-18, BDNF, NGF, GDNF, VEGFA, and TGF-β) was assessed using a commercial cytokine antibody array. Neuronal health was evaluated by immunofluorescence staining for neuronal markers MAP2 and βIII-tubulin, followed by imaging with a confocal microscope to assess neurite network density and morphology.

Mitochondrial homeostasis and oxidative stress analysis

Intracellular reactive oxygen species (ROS) levels in co-cultures were measured by flow cytometry after incubation with a ROS-sensitive fluorescent probe. Lipid peroxidation was quantified by measuring malondialdehyde (MDA) content using a colorimetric assay kit. Mitochondrial respiration was analyzed using the Seahorse XF Pro Analyzer (Agilent Technologies, Santa Clara, CA, USA). The Mito Stress Test was conducted to comprehensively evaluate mitochondrial performance by determining parameters such as basal respiration, ATP-linked oxygen consumption, proton leak, maximal respiratory capacity, and spare respiratory capacity. To achieve this, sequential injections of oligomycin, FCCP, and a mixture of rotenone and antimycin A were administered to the cells during the assay. These compounds selectively inhibit or uncouple components of the electron transport chain, thereby allowing quantitative assessment of individual aspects of mitochondrial bioenergetics. The mitochondrial membrane potential (MMP) was assessed using JC-1 dye staining and flow cytometry. For ultrastructural analysis, cells were fixed, processed, and sectioned for TEM to observe mitochondrial morphology, including swelling and cristae disruption. The expression of key antioxidative proteins (Nrf2, HO-1, GPX4) was determined using western blotting.

Contusive spinal cord injury model

All animal experiments were conducted in accordance with the guidelines of the Animal Research: Reporting of In Vivo Experiments and were approved by the Institutional Animal Care and Use Committee of China Medical University (CMUIACUC-2025-195). Female Sprague–Dawley rats (8–10 weeks old) were purchased from BioLASCO Taiwan Co., Ltd. Before surgery, animals were anesthetized via intraperitoneal injection with a mixture of Zoletil 50^™ (35 mg/kg) and Xylazine (2.5 mg/kg). A dorsal midline incision was made over the thoracic vertebrae (T9–T10), and the paraspinal muscles were bilaterally retracted to expose the vertebral column. A laminectomy was performed at the T9 vertebral level to expose the dura mater of the spinal cord. A contusion injury was then induced using an impactor device (RWD Life Science, Guangdong, China) with parameters set at 250 kilodyne force, 2.5 m/s velocity, and 1.7 mm displacement. Following the injury, EVs were administered via tail vein injection at 1-, 24-, 48-, and 72-h post-injury (5 × 10⁹ EV in 100 µL of saline per injection). All drug administration and subsequent behavioral assessments were conducted under a double-blind protocol.

In vivo biodistribution and target engagement

To track vesicle distribution, EV were labeled with a near-infrared lipophilic dye, 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindotricarbocyanine iodide (DiR). At designated times after intravenous injection, animals were euthanized, and major organs (spine, heart, kidneys, lungs, liver, pancreas, and spleen) were harvested for ex vivo imaging using an IVIS. For microscopic analysis, spinal cord tissue was sectioned and imaged via confocal microscopy to co-localize the DiR-labeled EVs and Cy3-labeled mBDNF cargo within the lesion parenchyma. To quantify local drug delivery, intracellular BDNF concentration in spinal cord tissue homogenates was measured by ELISA at various time points post-injection.

Behavioral assessment

Locomotor performance was assessed weekly from week 1–4 following SCI. During the first three weeks, video recordings were obtained by a single-blind operator as the animals moved freely in an open field. The Basso–Beattie–Bresnahan rating scale was used to evaluate joint movement, paw placement, weight support, and coordination. To further evaluate hindlimb motor coordination during walking, animals were placed on a horizontal grid runway approximately 1 m in length, with regularly spaced metal bars (3 cm intervals). Video recordings were used to assess foot-faults across the central 60 cm segment of the runway. The error rate was calculated as: error rate (%) = (number of hindlimb errors/total number of hindlimb steps) × 100. A foot placement was considered an error if the limb slipped, dragged, or completely missed the bar. Each animal performed at least three walking trials.

Moreover, to obtain a detailed analysis of post-injury gait, the CatWalk XT system (Noldus, VA, USA) was used. Animals were habituated in the behavioral testing room for at least 2 h prior to data collection. Each animal was required to complete at least three uninterrupted runs that met software validation criteria (run variation ≤ 60% within a 5-s run). The detection settings were configured with a camera gain of 24.4 dB and a green intensity threshold of 0.1 arbitrary units. The following gait features were extracted for analysis: regularity index, swing speed (hindlimb/forelimb), mean intensity (hindlimb/forelimb), and stride length (hindlimb/forelimb).

Histological and immunohistochemical analysis

At the study endpoint, animals were euthanized, and spinal cords were harvested for histological analysis. Tissue sections were stained with hematoxylin and eosin (H&E) to assess lesion volume and tissue preservation. Apoptosis was detected using the TUNEL assay. Neuroinflammation and glial scarring were evaluated by immunofluorescence staining for microglia/macrophage markers (CD80 for M1, CD163 for M2), astrocytes (GFAP), and inflammatory cytokines (TNF-α). Neuronal preservation and axonal integrity were assessed by staining for βIII-tubulin and neurofilament.

Statistical analysis

All quantitative data are presented as mean ± standard deviation (SD). For comparisons at a single time point, statistical significance between groups was determined using one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test for multiple comparisons. For longitudinal outcomes measured repeatedly over time (e.g., BBB scores), data were analyzed using two-way repeated-measures ANOVA followed by an appropriate multiple-comparisons post hoc test. A p-value of less than 0.05 was considered statistically significant. All analyses were performed using GraphPad Prism software.

Results

Engineering and physicochemical characterization of mBDNF@αITG EV

To generate a lesion-targeting carrier, an anti-integrin single-chain variable fragment (scFv) was inserted into the large extracellular loop of CD63 to produce αITG-displaying EVs (Fig. 2A). Surface plasmon resonance confirmed high-affinity integrin binding (KD ≈ 3.6 × 10⁻⁹ M), consistent with effective biological targeting (Fig. 2B). Immunoblotting verified the engineered scFv and canonical EV markers (CD63, CD81, CD9, and Alix), whereas the absence of calnexin confirmed the lack of cellular contamination (Fig. 2C). Flow cytometry corroborated robust CD63 and αITG scFv surface expression. The selected ITG-HEK293T clone showed a larger CD63-positive fraction (92.6% vs. 34.7%) than that of the parental HEK293T cells as well as strong scFv display (74.6% vs. 0.60%) (Fig. S1). mBDNF loading into αITG EV yielded ~55–65% encapsulation efficiency across independent preparations (Fig. 2D). Nano-flow cytometry revealed that most particles were co-positive for αITG and mBDNF, with ~68.9% double-positive, ~ 15.7% αITG-only, and minor BDNF-only or double-negative fractions (Fig. 2E). As shown in Fig. 2F, the mBDNF mRNA content was expressed as ng mBDNF mRNA per 10¹⁰ EVs, providing a particle-normalized metric for cross-batch comparison of loading levels. Nanoparticle tracking analysis revealed a unimodal size distribution centered at 130–150 nm with narrow polydispersity and particle concentrations in the 10⁶–10⁷/mL range (Fig. 2G). TEM showed intact, spheroidal vesicles with a typical cup-shaped profile at low and high magnification, with no morphology change after surface engineering or mBDNF loading (Fig. 2H). Super-resolution imaging detected discret mBDNF signals within single-vesicle and αITG at the surface, consistent with luminal mBDNF and external ligand display (Fig. 2I).

Fig. 2.

Engineering and physicochemical characterization of mBDNF@αITG EV. (A) Schematic of the αITG scFv–CD63 fusion construct used for engineering EV. (B) SPR analysis of αITG EV binding to integrin proteins at different concentrations. (C) Western blot analysis of ScFv, CD63, CD81, CD9, Alix, and calnexin in EV and αITG EV. Representative images shown from three independent experiments. (D) Encapsulation efficiency of mBDNF in αITG EV. Data are representative of n = 32 independent experiments with similar results. (E) Nano-flow cytometry analysis of mBDNF@αITG EV. Representative plots shown from three independent experiments. (F) mBDNF mRNA payload normalized to EV particle number, expressed as ng mBDNF mRNA per 10¹⁰ EVs. Data are representative of n = 15 independent experiments with similar results. (G) NTA of αITG EV size distribution. Data are representative of n = 6 independent experiments with similar results. (H) TEM images of αITG EVs at low and high magnifications. Scale bars: 1 μm (left) and 100 nm (right). Representative images shown from three independent experiments. (I) Super-resolution microscopy images of single vesicles labeled for αITG (green) and mBDNF (red). Scale bar: 50 nm. Representative images shown from three independent experiments. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test

Supplementary characterization (Fig. S2) focuses on side-by-side comparisons of EV properties before and after mBDNF loading. Representative TEM images of αITG EV and mBDNF@αITG EV at low and high magnification are provided in Fig. S2A. Overlay NTA profiles comparing αITG EV and mBDNF@αITG EV are shown in Fig. S2B. Immunoblotting confirmed retention of canonical EV markers (CD63, CD81, CD9, and Alix) with no detectable Calnexin, and the ScFv signal verified αITG surface display after mBDNF loading (Fig. S2C). In HPLC traces, mBDNF@αITG EV exhibited a single peak that co-eluted with the mBDNF standard, supporting successful and intact mBDNF loading (Fig. S2D). Single-vesicle quantification further showed larger mean spot areas for the αITG module than those for mBDNF mRNA (Fig. S2E). In freeze–thaw stability assay,, residual particle levels decreased gradually over time but remained close to baseline, with > 85% particles retained after 7 days (Fig. S2F).

Microglial polarization modulation by mBDNF@αITG EV in co-culture system

Primary microglia and neurons were cultured in a transwell system that allowed exchange of soluble factors without direct contact (Fig. 3A). Target availability and microglial uptake were confirmed first. Under hypoxia, integrin positivity increased relative to normoxia (63.9% vs. 49.1% by flow cytometry) with stronger ITG immunofluorescence (Fig. S3A–B), and αITG EVs entered microglia more efficiently than unmodified EVs (86.0% vs. 27.1% by flow cytometry; Fig. S3C). We next evaluated neuronal targeting and protection. Under the same hypoxic stress used in the co-culture system, neurons also upregulated the intended target and showed efficient engagement by engineered vesicles. Flow cytometry demonstrated a higher ITG-positive fraction under hypoxia than normoxia (72.4% vs. 44.7%; Fig. S4A), consistent with stronger ITG immunofluorescence (Fig. S4B). Accordingly, αITG EVs were internalized by neurons more efficiently than unmodified EVs (85.8% vs. 15.1% by flow cytometry), and confocal imaging confirmed abundant ExoGlow-labeled vesicles within the neuronal arbor (Fig. S4C). Enhanced neuronal delivery translated into elevated high levels of intraneuronal BDNF. mBDNF@αITG EV yielded the greatest increase in mBDNF and the highest BDNF concentration in the neuronal lysates, significantly exceeding mBDNF@EV and all non-mBDNF controls (Fig. S4D–E). This preferential uptake was further tested using an anti-ITG blocking antibody, which reduced αITG EV internalization in both microglia and neurons (Fig. S5A–B). Confocal imaging then examined the intracellular fate of the delivered mBDNF signal and showed co-localization with lysosomes after uptake (Fig. S6). Together with the microglial data, these findings verify target expression and preferential uptake in both cellular compartments and provide a mechanistic basis for improved immunomodulation and neuroprotection, which was also observed in the following co-culture experiments.

Fig. 3.

Evaluation of mBDNF@αITG EV uptake and neuroprotective effects in a neuron–microglia co‑culture. (A) Schematic illustration of the Transwell co‑culture system used for EV treatment. (B) Flow cytometry analysis of microglial polarization markers CD86 and CD206 after treatment with different EV formulations. (C) Western blot analysis of NF-κB, p-NF-κB, and PPARγ in microglia after the indicated treatments; β-actin served as a loading control. ELISA quantification of (C) TNF‑α, (D) IL‑1β, (E) IL‑4, and (F) IL‑10 in co‑culture supernatants. (G) Immunofluorescence staining of MAP2 (green), βIII‑tubulin (red), and nuclei (blue) in neurons from the indicated groups. Scale bars: 50 μm. (H) Cytokine antibody array-based analysis and densitometric quantification of secreted inflammatory and neurotrophic factors across groups. Data are representative of n = 6 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one‑way ANOVA with Tukey’s post hoc test. Different letters (a–c) indicate statistically significant differences among groups. Asterisks (*) indicate statistically significant differences between the compared groups. Significance levels: * < 0.05, ** < 0.01, *** < 0.005, **** < 0.001

Engineered EV consistently shifted microglial polarization toward a reparative phenotype. Specifically, the CD86⁺ fraction decreased whereas CD206⁺ cells increased, with changes larger than those seen with unmodified EVs or mBDNF-lacking αITG EV (Fig. 3B). Consistent with this phenotypic shift, western blotting revealed reduced NF-κB phosphorylation together with increased PPARγ in microglia treated with mBDNF@αITG EV compared with EV, αITG EV, and mBDNF@EV (Fig. 3C). At similar doses, ELISA of co-culture supernatants showed stepwise reductions in TNF-α and IL-1β together with higher IL-4 and IL-10 in the mBDNF@αITG EV group than that in SCI, EV, and αITG EV controls (Fig. 3D–G). Neuronal morphology tracked with these changes. MAP2 and βIII-tubulin staining revealed denser neurite networks, longer processes, and more intact somatodendritic architecture after mBDNF@αITG EV treatment than in the control groups (Fig. 3H). In parallel, iNOS, IL-1β, and TNF-α mRNA levels decreased after EV treatment and were lowest in the mBDNF@αITG EV group, whereas Arg-1, CD206, and IL-10 mRNA levels increased across EV-treated groups and were highest in the mBDNF@αITG EV group (Fig. S7). To assess temporal dynamics and TrkB dependence, key readouts were monitored over a 3-day treatment course in the co-culture system. BDNF levels increased over time and were highest in the mBDNF@αITG EV group, whereas TrkB inhibition attenuated this elevation (Fig. S8A). In parallel, the reduction in CD86 positivity and the increase in CD163 positivity observed with mBDNF@αITG EV were blunted by TrkB inhibition across the same time window (Fig. S8B–C). Consistently, mBDNF@αITG EV decreased TNF-α and increased IL-4 in a time-dependent manner, and these cytokine changes were diminished in the presence of a TrkB inhibitor (Fig. S8D–E). In line with the ferroptosis-related protection observed in the injury model, a pharmacological rescue experiment further supported this mechanism. Ferrostatin-1 attenuated the injury-associated ferroptosis marker changes and partially reversed the NRF2 GPX4 and 4-HNE profile (Fig. S9). Meanwhile, immunoblotting showed that mBDNF@αITG EV increased NRF2 and GPX4 levels and reduced 4-HNE, whereas TrkB inhibition attenuated NRF2 and GPX4 induction and blunted the reduction of 4-HNE (Fig. S8F). In addition, along with immunomodulatory shift, secreted trophic factors also increased. Microglia exposed to mBDNF-loaded EVs released more VEGF, with the highest levels in the mBDNF@αITG EV group (Fig. S10). Reactive astrocyte markers were further examined in injured spinal cords. Western blot analyses of spinal lysates showed decreased A1 markers (C3, GBP2) and increased A2 markers (PTX3, S100A10) following treatment with mBDNF-loaded EVs, with the strongest changes in the mBDNF@αITG EV group and intermediate changes in the mBDNF@EV group (Fig. S11A). ELISA showed higher BDNF levels in the conditioned medium after mBDNF-loaded EV treatment (Fig. S11B). Densitometry values were background-subtracted and normalized to on-membrane positive controls before aggregation across replicates. The resulting quantification (Fig. 3I; S12) showed reduced IFN-γ, IL-1β, IL-6, IL-8, IL-18, and TNF-α together with increased BDNF, NGF, GDNF, IL-10, VEGFA, and TGF-β.

Mitochondrial homeostasis restoration and antioxidative signaling activation by mBDNF@αITG EV

Flow cytometry showed persistently high ROS levels in the SCI, EV, and αITG EV groups, a partial reduction with mBDNF@EV, and near-baseline levels with mBDNF@αITG EV (Fig. 4B). Lipid peroxidation followed the same hierarchy, with malondialdehyde significantly reduced only in the mBDNF@αITG EV group (Fig. 4C). ATP assays indicated higher total ATP and a larger mitochondrial ATP fraction after mBDNF@αITG EV treatment than those in all the controls, which concurs with the improved oxidative phosphorylation that was measured independently (Fig. 4D; Fig. S13). Ultrastructural analyses supported these functional gains. TEM revealed swollen, vacuolated mitochondria with disrupted cristae in SCI and EV conditions, partial correction with BDNF@EVs, and the most preserved cristae and matrix density with mBDNF@αITG EV (Fig. 4E, upper). JC-1 flow cytometry quantified membrane potential recovery, with ΔΨm-high cells at 15.8% in SCI, 17.9% in EV, 16.4% in αITG EV, 35.7% in BDNF@EV, and 58.2% in mBDNF@αITG EV cultures (Fig. 4E, lower). Mitochondrial content and network organization were visualized using TOMM20. Signals were weak and fragmented in SCI, EV, and αITG EV groups, increased with mBDNF@EV, and were strongest after mBDNF@αITG EV treatment, where TOMM20 outlined a continuous and reticulated network along neurites (Fig. 4F). Consistently, neuronal immunofluorescence showed higher GPX4 together with preserved βIII-tubulin architecture in the mBDNF@αITG EV group than those in all the other treatments (Fig. 4G). Following treatment, mitochondrial organization was further assessed by TOMM20 staining and autophagy association by co-staining TOMM20 with LC3. Compared with SCI, EV, and αITG EV groups, mBDNF@αITG EV yielded stronger TOMM20 signals together with increased TOMM20–LC3 co-localization, consistent with enhanced mitochondrial turnover and quality-control engagement under the treated condition (Fig. 4H). Western blotting detected clear induction of endogenous antioxidant pathways, with NRF2, HO-1, and GPX4 elevated most in mBDNF@αITG EV samples (Fig. 4I). In addition, immunoblotting showed reduced DRP1 and increased OPA1 levels after mBDNF@αITG EV treatment, together with elevated PINK1 and Parkin expression, consistent with enhanced mitochondrial quality control (Fig. 4J). Taken together, BDNF@αITG EVs reduced oxidative burden, restored mitochondrial polarization, improved bioenergetic capacity, and activated endogenous antioxidant pathways.

Fig. 4.

Effects of mBDNF@αITG EV on mitochondrial oxidative stress and bioenergetics in SCI neurons. (A) LDH release quantified in neuron–microglia co-culture supernatants after the indicated treatments. (B) Flow cytometry analysis of intracellular ROS in SCI neurons after the indicated treatments. (C) Quantification of malondialdehyde (MDA) content in SCI neurons. (D) ATP production in SCI neurons following different EV treatments. (E) TEM images of mitochondrial ultrastructure (upper row). Scale bar: 1 μm. JC‑1 flow cytometry analysis of mitochondrial membrane potential (ΔΨm) (lower row). (F) Immunofluorescence staining of TOMM20 to visualize mitochondrial networks in neurons (magenta). Scale bars: 10 μm. (G) Immunofluorescence staining of GPX4 (green), βIII‑tubulin (red), and nuclei (blue). Scale bars: 50 μm. (H) Immunofluorescence co-staining of TOMM20 (green), LC3 (red), and nuclei (blue). Scale bars: 50 μm. (I) Western blot analysis of HO‑1, GPX4, and NRF2 in SCI neurons; β‑actin served as a loading control. (J) Western blot analysis of mitochondrial dynamics and mitophagy-related markers (DRP1, OPA1, PINK1, and Parkin); β-actin served as a loading control. Data are presented as mean ± SD. P values were calculated using one‑way ANOVA with Tukey’s post hoc test. Different letters (a–c) indicate statistically significant differences among groups

mBDNF@αITG EV selectively target the injured spine and enhance local BDNF delivery in vivo

ITGs were first verified as an accessible target at the lesion. Compared with healthy cords, the injured spine displayed a high fraction of ITG⁺ cells and strong signal across the parenchyma, providing a molecular address for αITG engagement (Fig. 5A). Building on this targeting rationale, ex vivo biodistribution showed that αITG EVs concentrated along the spine substantially more than unmodified EVs, yielding roughly four to fivefold higher signal than EV controls, whereas heart, kidney, lung, liver, and spleen tissues remained similar between groups, which supports lesion-focused homing rather than global organ trapping (Fig. 5B). At the tissue scale, confocal images aligned with this pattern. In animals receiving mBDNF@αITG EV, DiR-labeled vesicles were abundant within the spinal parenchyma and overlapped with Cy3-BDNF, whereas mBDNF@EV produced sparse puncta with limited overlap, indicating efficient co-delivery of carrier and cargo to the target site rather than vascular confinement (Fig. 5C). Quantitative measurements of spinal BDNF closed the loop. To quantify delivered cargo beyond qualitative imaging, Cy3 labeled mBDNF mRNA associated with EVs was measured by HPLC, providing a quantitative readout that complements the spinal section fluorescence in Fig. 5C. The mBDNF mRNA signal was higher in the mBDNF@αITG EV group than in the mBDNF@EV group across the measured time points (Fig. 5D). To specifically assess the exogenous translation product in vivo, human BDNF protein was quantified in rat spinal cord homogenates using a human specific ELISA. Human BDNF levels increased over time and remained higher in animals receiving mBDNF@αITG EV compared with mBDNF@EV (Fig. 5E). Together these linked readouts move from target presence to selective homing to cellular co-localization and finally to pharmacodynamic exposure, establishing that integrin targeting improves on-target delivery of BDNF without increasing off-target burden.

Fig. 5.

In vivo targeting and local cargo delivery of mBDNF@αITG EV. (A) Immunofluorescence analysis of ITG in spinal cord sections from healthy and SCI animals, with ITG (red) and nuclei (blue) and quantification of ITG⁺ cells. (B) IVIS imaging of major organs after systemic administration of EV or αITG EV, with quantitative region‑of‑interest analysis normalized to the EV group. (C) Confocal microscopy of spinal cord sections showing DAPI (blue), mBDNF‑Cy3 (red), and DiR‑labeled EV (green). Scale bars: 100 μm. (D) HPLC quantification of mBDNF-Cy3. (E) Human specific ELISA measurement of hBDNF in rat spinal cord homogenates at the indicated time points after administration. Data are representative of n = 6 independent experiments with similar results. Data are presented as mean ± SD. P values were calculated using one‑way ANOVA with Tukey’s post hoc test. Asterisks (*) indicate statistically significant differences between the compared groups. Significance levels: ** < 0.01, *** < 0.005, **** < 0.001

Systemic mBDNF@αITG EV treatment improves locomotor recovery and gait performance in vivo

The study design and dosing schedule are shown in Fig. 6A, with contusion SCI followed by serial intravenous administrations and longitudinal behavioral testing up to day 28. Open-field locomotion was tracked with the BBB score across the 28-day window. Both mBDNF@EV and mBDNF@αITG EV groups scored higher than controls from day 7 to day 28 (two-way ANOVA, p < 0.001), and mBDNF@αITG EV exceeded mBDNF@EV overall (p < 0.01). On day 21, all control animals remained below 14, whereas 12.5% of BDNF@EV and 22.2% of mBDNF@αITG EV animals surpassed 14, indicating coordinated forelimb–hindlimb locomotion. BBB scores increased over time in all groups, with mBDNF@αITG EV outperforming mBDNF@EV from day 7 onward and remaining clearly above αITG EV, EV and SCI controls through day 28 (Fig. 6B). Stepping precision and bilateral coordination were captured by the grid-walk success rate, defined as the proportion of correct paw placements. Motor coordination and strength were assessed next. Stepping precision on the regular rung test improved most in the mBDNF@αITG EV, with a higher success rate than both controls and mBDNF@EV (p < 0.001) (Fig. 6C, movie 1). Animals receiving mBDNF@αITG EV also sustained greater maximum platform angles on the inclined plane, indicating better hindlimb support and trunk stability than all other groups (p < 0.001) (Fig. 6D). Quantitative gait parameters were quantified with the CatWalk system, and the analysis corroborated these findings. Regularity index approached a near‑normal pattern after mBDNF@αITG EV (p < 0.001 vs. controls), contrasting with irregular stepping in SCI, EV and αITG EV groups and an intermediate response with mBDNF@EV (Fig. 6E, movie 2). Dynamic limb power and range improved as well, with higher swing speed (p < 0.01 vs. controls and greater than mBDNF@EV), longer stride length (p < 0.001), and increased hindlimb mean intensity (p < 0.001) in the mBDNF@αITG EV cohort (Fig. 6F–G). Across assays, the targeted, mBDNF‑loaded EV produced the strongest functional gains, consistent with enhanced on‑target delivery and neuroprotection observed in preceding figures.

Fig. 6.

Behavioral assessments following SCI and EV treatments. (A) Schematic of contusion induction, dosing schedule, and behavioral assessment timeline. (B) Basso–Beattie–Bresnahan locomotor scores over the 28-day period post-injury. (C) Grid-walk success rate, (D) CatWalk-derived angle of incline for hindlimb paw placement, (E) CatWalk regularity index, (F) CatWalk hindlimb swing speed, and (G) CatWalk hindlimb stride length. Data are representative of n = 6 animals per group with similar results. Data are presented as mean ± SD. For BBB scores in (B), P values were calculated using two-way ANOVA with treatment group and time as factors, followed by Tukey’s multiple-comparison test. For endpoint measurements in (C–G), P values were calculated using one-way ANOVA with Tukey’s post hoc test. Different letters (a–c) indicate statistically significant differences among groups. Significance levels: * < 0.05, *** < 0.005

mBDNF@αITG EV suppress neuroinflammation and limit tissue loss after SCI

Macrophage and microglial phenotypes were evaluated in the injured cord. Immunofluorescence demonstrated lower CD80 and higher CD163 signals with mBDNF@αITG EV than those with the other formulations, indicating a shift away from an M1-like state toward an M2-like profile (Fig. 7A). Quantification confirmed this pattern, with CD80 markedly reduced and CD163 strongly increased in the targeted mBDNF group, whereas unmodified EV or αITG EV alone produced modest changes and mBDNF@EV showed an intermediate response (Fig. 7B–C). Neuroinflammatory burden and axonal integrity were assessed in parallel. Neuronal preservation and local cytokine burden were examined next. Sections stained for βIII and TNF-α showed denser neuronal fibers together with weaker TNF-α immunoreactivity in the mBDNF@αITG EV cohort relative to all controls, consistent with greater structural protection and lower inflammatory tone in the lesion area (Fig. 7D). Longitudinal cytokine measurements aligned with these histological findings. IL-1 and TNF-α rose early after injury and gradually declined in all groups, yet the decrease was most pronounced under mBDNF@αITG EV treatment across the 21-day window, with mBDNF@EV again showing a partial effect and the remaining groups clustering near SCI levels (Fig. 7E–F). Tissue repair markers were then examined to capture regenerative and remodeling outcomes. GAP43 staining increased most prominently in the mBDNF@αITG EV group, consistent with enhanced axonal sprouting or regeneration potential (Fig. 7G), and MBP signals were also higher in the targeted group, consistent with improved remyelination-associated features (Fig. 7H). In contrast, markers associated with reactive gliosis and fibrotic remodeling were reduced. Vimentin staining decreased most in the mBDNF@αITG EV group (Fig. 7I), and α-SMA signals were similarly lower, consistent with attenuation of fibrotic scar-associated components (Fig. 7J). Therefore, mBDNF@αITG EV most effectively dampened pro-inflammatory responses, promoted pro-repair polarization, and preserved neural tissue architecture in vivo.

Fig. 7.

Neuroinflammation, and neuronal markers in spinal cord sections. (A) Immunofluorescence staining for CD80 (red), CD163 (green), and nuclei (blue). (B) Quantification of CD80-positive area normalized to control. (C) Quantification of CD163-positive area normalized to control. (D) Immunofluorescence staining for βIII-tubulin (green), TNF-α (red), and nuclei (blue). Temporal profile of (E) IL-1 and (F) TNF-α levels over the experimental period. (G) Immunofluorescence staining for GAP43 (green) and nuclei (blue). (H) Immunofluorescence staining for MBP (red) and nuclei (blue). (I) Immunofluorescence staining for vimentin (red), GFAP (green), and nuclei (blue). (J) Immunofluorescence staining for α SMA (red), GFAP (green), and nuclei (blue). Scale bars are 100 μm. Data are presented as mean ± SD. P values for (B) and (C) were calculated using one way ANOVA with Tukey’s post hoc test. P values for time course data in (E) and (F) were calculated using two way ANOVA with treatment group and time as independent factors, followed by Tukey’s multiple comparison test. Different letters (a–c) indicate statistically significant differences among groups

mBDNF@αITG EV attenuate ferroptotic stress and reinforce endogenous antioxidant programs

Given the prominent oxidative stress after SCI, ferroptosis was examined as a secondary injury component. Lipid peroxidation quantification showed that malondialdehyde was elevated in SCI and remained high with EV or αITG EV, declined with mBDNF@EV, and reached the lowest values with mBDNF@αITG EV, indicating the most effective membrane lipid damage suppression (Fig. 8A). Tissue Fe²⁺ displayed the same hierarchy, with pronounced iron accumulation in SCI that was only partially lowered by mBDNF@EV and most strongly reduced by mBDNF@αITG EV (Fig. 8B). Consistent with reduced lipid peroxidation and iron dysregulation, GPX4 activity was also restored following treatment (Fig. 8C). These biochemical indices point to a substantive reduction of ferroptotic pressure when mBDNF is delivered by the integrin‑targeted carrier. Cellular readouts converged on improved neuronal status. Neun and Ki67 staining showed scarce proliferating neuronal profiles in SCI that changed little with EV or αITG EV, increased with mBDNF@EV, and were most frequent with mBDNF@αITG EV, consistent with enhanced neuronal maintenance and turnover in the injured parenchyma (Fig. 8D). βIII‑tubulin and HO‑1 immunostaining further illustrated this improvement. Discontinuous axonal bundles and weak HO‑1 signal dominated in SCI and vehicle groups, whereas mBDNF@EV yielded partial restoration of axonal continuity and stronger HO‑1, and mBDNF@αITG EV produced the most continuous βIII‑tubulin tracts together with robust HO‑1 across the peri‑lesional cord (Fig. 8E). Consistently, Nrf2, HO 1, and GPX4 mRNA levels were increased by mBDNF loaded EVs and peaked with mBDNF@αITG EV, which aligns with activation of endogenous antioxidant programs (Fig. 8F–H). Across biochemical, histological, and transcriptional measures, the integrin‑targeted mBDNF formulation most effectively decreased MDA and Fe²⁺ burden, stabilized neuronal cytoarchitecture, and upregulated the Nrf2–HO‑1–GPX4 axis, indicating superior control of ferroptotic susceptibility at the lesion site.

Fig. 8.

Mitochondrial/ferroptosis-related indices and neuroregeneration markers in vivo. (A) MDA levels, (B) Fe²⁺ concentration, and (C) GPX4 activity in spinal cord tissue. (D) Immunofluorescence of NeuN (green), Ki67 (red), and nuclei (blue). Scale bars: 100 μm. (E) Immunofluorescence of βIII-tubulin (green), HO-1 (red), and nuclei (blue). Scale bars: 100 μm. qRT-PCR analysis of (F) NeuN, (G) HO-1, and (H) GPX4 mRNA ratio. Data are representative of n = 6 per group with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. Different letters (a–c) indicate statistically significant differences among groups

mBDNF@αITG EV reduce glial scarring and preserve axonal tracts in vivo

Histology showed smaller lesion cavities and better-preserved parenchyma after mBDNF@αITG EV treatment than that with SCI and mBDNF@EV groups. TUNEL labeling revealed a few apoptotic profiles in the same animals, consistent with reduced secondary damage (Fig. 9A). Systemic tolerability was evaluated by HE staining of the heart, liver, spleen, lungs, and kidneys at the endpoint. These in vivo efficacy findings were obtained under a dosing regimen that was well tolerated at the organ level (Fig. S14). Whole-cord immunostaining showed attenuated astrocyte reactivity and axonal continuity after mBDNF@αITG EV treatment. Compared with SCI, the targeted formulation displayed weaker GFAP bands around the lesion and longer, continuous neurofilament signals across the injured segment, producing an extensive yellow merge (Fig. 9B–C). Quantification supported these observations. Neurofilament coverage increased with EV administration but was maximal with mBDNF@αITG EV, whereas mBDNF@EV produced an intermediate improvement (Fig. 9D). Cytokine profiling of spinal tissue aligned with the structural readouts. mBDNF@αITG EV most effectively reduced the pro-inflammatory mediators IFN-γ, IL-1β, IL-6, IL-8, and TNF-α while elevating trophic and anti-inflammatory factors BDNF, NGF, GDNF, IL-10, and VEGF-A (Fig. 9E; Fig. S15). The combined pattern indicates a microenvironment with low scarring propensity and support for axonal maintenance.

Fig. 9.

Regeneration-related changes in the injured spinal cord after mBDNF@αITG EV treatment. (A) HE and TUNEL staining of lesion sites across the indicated groups. (B) Whole-mount immunostaining of spinal cord segments for GFAP (green) and neurofilament (red) across the indicated groups. Merged images are shown below. (C) Lesion area quantification in spinal cord tissue. Area of lesion (mm²) measured from cross-sections of spinal cord across the indicated groups. (D) Quantification of neurofilament coverage (%) within the lesion region. (E) Densitometric quantification of the cytokine antibody array highlighting the shift from a pro-inflammatory to a neurotrophic microenvironment in spinal cord tissue. Data are representative of n = 6 animals per group with similar results. Data are presented as mean ± SD. P values were calculated using one-way ANOVA with Tukey’s post hoc test. Different letters (a–c) indicate statistically significant differences among groups. Asterisks (*) indicate statistically significant differences between the compared groups

Discussion

SCI involves a cascade of secondary events, including neuroinflammation, oxidative stress, mitochondrial dysfunction, glial activation, and cell death, which collectively amplify the initial injury and limit functional recovery. Addressing these interlocking processes requires therapeutic exposure at the lesion coupled to mechanisms that concurrently re-balance immunity and cellular energetics. In recent years, various literature have also pointed out that the EV platform, as a gene therapy delivery technology, has great potential for treating CNS diseases [41–43]. In this study, an engineered EV platform displaying an α‑integrin scFv on CD63 and carrying mBDNF (mBDNF@αITG EV) achieved lesion-centric accumulation, enhanced intralesional BDNF exposure, and coordinated modulation of microglial state and mitochondrial/ferroptotic axes, collectively aligning with tissue preservation and improved locomotor outcomes. These findings support a testable working model that connects lesion-centric delivery with coordinated immunometabolic and cytoprotective changes that align with functional recovery, addressing limitations of non-engineered EV or single-agent strategies with short tissue residence [44]. Recent studies have also indicated that treatment strategies for nanoparticles still require surface modification to achieve their target function and maximize effectiveness [45–47]. We summarize the mechanistic interpretation as a working hypothesis rather than a closed causal chain. The data support a hierarchical model in which lesion-enriched αvβ8 engagement improves local EV association and delivery, leading to increased BDNF availability and TrkB engagement at the injury site. Within this context, the observed microglial shift toward a reparative state coincides with reduced p-NF-κB and increased PPARγ, and this immunologic rebalancing occurs in parallel with recovery of mitochondrial homeostasis markers and induction of NRF2-associated antioxidant programs. These coordinated changes align with attenuation of lipid peroxidation readouts and reduced ferroptosis-associated stress. Because most links are inferred from convergent readouts and TrkB-level inhibition, dedicated pathway-specific gain and loss studies will be required to resolve directionality and interdependence among microglial polarization, mitochondrial remodeling, and NRF2–GPX4 signaling (Fig. 10).

Fig. 10.

Proposed mechanism of mBDNF@αITG-EVs delivery for SCI therapy. The engineered mBDNF@αITG-EVs act through a multi-stage process to restore spinal cord function. First, the surface-displayed scFv ensures high-affinity binding to upregulated integrin αvβ8 at the injury epicenter. Second, the encapsulated BDNF mRNA is successfully delivered, translated into functional protein within host cells, and secreted. Finally, the BDNF enters cells via TrkB to mitigate neuronal ferroptosis and inflammatory damage by modulating the NRF2/GPX4 signaling axis and restoring mitochondrial homeostasis, eventually leading to axonal sprouting and motor recovery

At the delivery–pharmacology interface, SCI increases the accessibility of integrin-related epitopes across neurons, glia, and the perivascular compartment, thereby creating a window for αITG-guided uptake [48, 49]. Consistently, mBDNF@αITG EV display preferential internalization in neuron–microglia co-cultures and high intralesional accumulation in vivo, as supported by DiR/in vivo imaging system (IVIS) imaging and histological co-localization (Fig. 5B–C). ELISA further demonstrated a higher and more sustained BDNF concentration in injured cords than with non-engineered EVs or free BDNF (Fig. 5D). Single-vesicle analytics, including nano-flow cytometry and super-resolution microscopy, corroborated the co-occurrence of the αITG display with mBDNF cargo, whereas HPLC quantified loading (Fig. 2; Fig. S2). Across these layers of target presence, cellular uptake bias, tissue distribution, and local concentration, the evidence coherently supports a delivery–pharmacology link that underpins downstream biology [50].

Within the neuroimmune milieu, mBDNF@αITG EV shift microglia from a pro-inflammatory toward a repair-associated phenotype, evidenced by reductions in TNF-α/IL-1β/iNOS and gains in IL-10/Arg1/CD206 in vitro and in vivo (Fig. 3A–E; Fig. S5). When microglia polarize into M2 type, they can repair damaged motor neurons and promote neural tissue regeneration [51, 52]. Consistent with the microglial M1-to-M2 switch, C3 and GBP2 suppression together with PTX3 and S100A10 enrichment in vivo (Fig. S11) suggests remodeling of the glial milieu that curbs neurotoxic A1 astrocytes while favoring an A2 program supportive of tissue repair and axonal regrowth [53]. In addition, A2 astrocyte also increases the release of BDNF (Fig. S11B), which helps protect damaged nerve cells and enhances the regenerative capacity of neural stem cells [54]. In fact, astrocytes are uniquely affected in key neuronal processes but are inclined to assume neuroinflammatory phenotypes in SCI [55]. These immune changes paralleled improved neuronal morphology in the co-culture (MAP2/βIII-tubulin; Fig. 3G) and reduced apoptosis in vivo (TUNEL; Fig. 9A), supporting a model in which localized neurotrophin exposure interrupts the self-reinforcing loop between inflammation and neurotoxicity [56]. In addition to early immunologic rebalancing, the tissue-level profiles in Fig. 7 support a broader concept of microenvironment remodeling that links inflammation resolution to structural repair in the injured cord. Across the peri-lesional parenchyma, mBDNF@αITG EV reduced CD80-positive inflammatory microglia/macrophages while increasing CD163-positive repair-associated populations (Fig. 7A–C), together with lower local TNF-α immunoreactivity and a sustained reduction of IL-1 and TNF-α over the subacute time window (Fig. 7D–F). Within this calmer inflammatory context, regenerative markers were elevated: GAP43 signals increased and MBP coverage improved (Fig. 7G–H), consistent with enhanced axonal sprouting and remyelination-associated features. Notably, components of fibroglial remodeling were concurrently attenuated, as reflected by reduced vimentin and lower α-SMA signals (Fig. 7I–J), a pattern compatible with limitation of reactive gliosis and fibrotic scar-associated elements. Collectively, these coordinated readouts align with a lesion targeted shift toward a repair permissive microenvironment shaped by immune cells, glial responses, and fibrotic remodeling rather than an isolated anti inflammatory effect. Mechanistically, this coordinated pattern is consistent with localized BDNF availability and TrkB-related pro-survival/pro-repair signaling operating within an inflammatory milieu that is shifted away from NF-κB–dominant programs; however, cell-type–resolved gain/loss studies will be required to determine the relative contributions of microglia, astrocytes, OPC-lineage cells, and perivascular fibrotic sources to the observed reduction in scar markers and the increase in remyelination-associated features.

SCI causes an increase in ROS, leading to severe oxidative damage in cells [57]. This is closely related to changes in various antioxidant enzymes and mitochondrial function. Additionally, it causes damage to the blood-spinal cord barrier, hemolysis of red blood cells, and the release of iron stores from microglia [58]. These factors increase iron ion concentrations in the local microenvironment, promote lipid peroxidation, and exacerbate cellular ferroptosis [24]. In addition to immunomodulation, mBDNF@αITG EV restores key nodes in the mitochondrial–oxidative stress axis. We observed recovery of mitochondrial membrane potential and ATP generation, attenuation of ROS and lipid peroxidation, and induction of NRF2 HO 1 GPX4 associated antioxidant defenses (Fig. 4B–E and I; Fig. S13). These changes coincided with reduced cell death in the injured cord (Fig. 9A), assembling a continuum from immune relief to energetic stabilization [59, 60].

Consistent with the behavioral gains, tissue-level pathology also improved. Lesion area was reduced and neuronal and axonal markers were better preserved, aligning with higher BBB scores and improved spatiotemporal gait parameters (Figs. 6B–G and 9A–D). In preclinical SCI, BBB and quantitative gait measures are accepted functional endpoints, and their improvement alongside convergent cellular and molecular readouts strengthens both interpretability and reproducibility of efficacy claims [61]. In our study, BBB curves separated early and remained divergent, mirrored by gait domains capturing coordination, propulsion, timing, and stability. The regularity index increased toward intact values; stride length and swing speed rose; base of support narrowed and paw print area stabilized. The largest effects occurred with mBDNF@αITG EV and were consistent across time. These behavioral improvements relate to functions that matter in rehabilitation, including coordination and balance control during overground walking and gait stability on uneven surfaces. Stride length and swing speed provide proxy measures for locomotor efficiency and walking capacity in preclinical settings, while base of support reflects postural stability and fall-related control [62]. Consistent with these functional readouts, tissue-level analyses showed reduced lesion burden together with better preservation of neuronal and axonal markers and lower reactive gliosis, supporting improved movement quality as well as quantity after targeted EV treatment (Figs. 6B–G and 9A–D). Target engagement and local exposure measurements further complement this outcome profile by showing preferential spinal cord accumulation and increased BDNF levels at the lesion site (Fig. 5B–D). Conceptually, the scFv–CD63 display strategy is modular. Targeting ligands can be exchanged to match disease- or stage-associated epitopes, and the EV lumen can accommodate nucleic acids or other therapeutic payloads depending on the application (Fig. S1). Prior work in CNS injury and degeneration suggests that engineered EVs can converge on autophagy–lysosome regulation, mitochondrial homeostasis, and immune tone, and the coordinated patterns observed here align with this broader framework [63].

Some limitations warrant emphasis in this study. First, the mechanistic attribution in this study mainly focuses on the target display and exogenous BDNF mRNA delivery axis, while a comprehensive mechanism centered on RNA load remains an important direction for future research. Furthermore, many studies have indicated that the cargo carried by HEK293T-related EVs is relatively simple and has little impact on cell physiology [64, 65]. Therefore, we did not conduct a systematic analysis of endogenous EV RNA load, nor did we perform pathway targeting loss-of-function experiments to distinguish its contribution from the delivered mRNA load. Due to the fact that integrins are broadly expressed across neural and vascular compartments, the absence of overt off-target accumulation in the present datasets does not preclude longer-term or higher-resolution pharmacokinetic and tissue analyses to exclude unintended vascular or parenchymal retention [66]. Systemic administration raises translational questions that require further study. This work focuses on efficacy in the early post-injury phase and does not address long-term biodistribution or chronic vascular/parenchymal retention. Future studies should extend time-course biodistribution with higher-resolution tissue mapping and evaluate repeated-dose vascular and immune interactions, together with dose-escalation and delayed-start designs to define the therapeutic window and safety margins for neurotrophin signaling. In this context, uptake enhancement was supported by competitive inhibition: pretreatment with a function-blocking anti-αvβ8 antibody reduced αITG-EV uptake toward the non-targeted EV baseline, consistent with an αvβ8-dependent contribution to cellular interaction. However, the specific endocytic routes responsible for internalization (e.g., clathrin-mediated endocytosis, caveolin-dependent uptake, and macropinocytosis) were not dissected in this study and warrant dedicated investigation in future work. In addition, although higher intralesional BDNF exposure was associated with functional benefit, neurotrophins can elicit ectopic plasticity and pain sensitization; accordingly, local PK and PD boundaries must be defined with explicit safety margins.

From a translational standpoint, clinical translation will require further process development built on the TFF-based workflow used in this study, including fit-for-purpose release specifications that cover EV identity and purity, αITG display density, mBDNF payload, and a potency assay, as well as demonstration of batch-to-batch consistency toward GMP readiness. In parallel, defining systemic and local exposure–response relationships and safety margins for neurotrophin signaling will be essential for dose selection and risk management in subsequent translational studies [67–69]. Furthermore, to improve the therapeutic efficacy of EVs, our group will also attempt to establish stem cells capable of expressing high levels of BDNF and increase the number of αITG on the surface of EVs through various treatments such as viral infection and click chemistry. In the future, prioritizing increasing specificity while maintaining exposure should be considered. Examples include targeting or inducible/switchable displays to refine cellular selectivity, barcoded quantitative tracing and spatial transcriptomics or proteomics to resolve in situ receptor-ligand interactions across neurons, oligodendrocytes, astrocytes, and microglia, as well as cell-type-specific analyses of lipid metabolism and ferroptosis programs across both acute and chronic phases. In addition, an in vitro hypoxia-only condition does not fully reproduce the complex secondary injury milieu that includes ATP depletion, excitotoxicity, and Ca²⁺ overload. Future studies will incorporate complementary paradigms (e.g., OGD or excitotoxic stressors) to further dissect these components [70]. Finally, randomized, blinded, and dose-escalation studies that integrate diffusion tensor imaging, gait analytics, and blood EV biomarkers as pharmacodynamic and prognostic endpoints will help align mechanistic readouts with clinically meaningful function.

Conclusion

In this study, we report the mBDNF-loaded integrin αvβ8-targeted-EV platform that connects molecular targeting to cellular mechanisms and behavioral function in SCI. mBDNF@αITG EV establish a direct connection between molecular targeting, local pharmacology, and functional recovery in spinal cord injury. mBDNF@αITG EV concentrate within the injured cord, elevate and sustain local mBDNF exposure, and reprogram the lesion milieu by coupling microglial polarization with mitochondrial restoration and antioxidant defenses. By concentrating BDNF within the injured cord, the platform reprograms the neuroimmune milieu toward a repair-associated state and restores bioenergetic and antioxidant defenses, yielding consistent gains from cellular and tissue metrics to locomotor performance. Single vesicle characterization, quantitative biodistribution, and intralesional pharmacokinetics collectively support a mechanism in which ITG-guided uptake increases the magnitude and duration of neurotrophin action at the lesion. The scFv CD63 architecture provides a modular scaffold that can accommodate alternative ligands and cargos, enable disease and phase-specific tailoring and create opportunities for combination with rehabilitation or neuromodulation. Collectively, this study supports a systemically administered, lesion-targeted EV–mRNA platform for spinal cord injury that combines enhanced lesion association with functional recovery in vivo. These efforts will help advance engineered EV therapeutics toward clinically actionable strategies for spinal cord injury.

Author contributions

M.Y.S., C.D.C., and Y. C. designed and carried out the experiments, acquired and analysed the data, and wrote the manuscript. Y.H.L., M.H.Y., Y. P. C., P. F. C., C.Y.C., Y.W.C., M.C.C., and S.C.C. carried out the experiments. D.Y.C. supervised the entire project.

Funding

The authors acknowledge receipt of grants from the National Science and Technology Council of Taiwan (NSTC 112-2628-E-039-001-MY3, 114-2811-E-039-001, 114-2314-B-039-044-MY3, and 114-2327-B-039-001), China Medical University Hospital (EXO-112-005 and EXO-113-007), and China Medical University (CMU114-MF-16). Experiments and data analyses were performed in part by the Medical Research Core Facilities, Office of Research & Development, China Medical University, Taichung, Taiwan.

Data availability

All data generated or analyzed during this study are included in this published article.

Declarations

Ethics approval and consent to participate

All animal procedures were conducted in accordance with the guidelines of the Institutional Animal Care and Use Committee at the University of China Medical University, and study protocols were reviewed and approved prior to performing the experimental procedures described.

Consent for publication

All authors have approved the contents of this manuscript and provided consent for publication.

Competing interests

The authors declare no competing interests.