Preclinical efficacy of extracellular vesicle therapy in peripheral nerve injury: a comprehensive systematic review and network meta-analysis

Abstract

Background

Peripheral nerve injuries (PNIs) result in significant motor and sensory deficits, with autologous nerve grafting as the gold standard but limited by donor morbidity and incomplete recovery. Extracellular vesicles (EVs) have emerged as a promising cell-free therapy, offering potent paracrine signaling, safety, and immunological compatibility. This study systematically evaluated the preclinical efficacy of EV-based therapies for PNI.

Methods

A systematic review and network meta-analysis were conducted according to PRISMA guidelines, searching PubMed, Scopus, and Web of Science to July 1, 2025. Eligible in vivo studies investigated EVs from any cell source in animal PNI models, compared with controls, and reported functional, structural, electrophysiological, or sensory outcomes. Risk of bias was assessed using the CAMARADES checklist. Three hierarchical meta-analyses compared EVs, structural grafts, cellular therapies, and combination strategies. The primary endpoint was sciatic functional index (SFI), and secondary outcomes included muscle preservation, sensory recovery, and electrophysiology.

Results

Thirty-five studies met inclusion, with 32 included in quantitative synthesis. EV therapy significantly improved SFI versus control (MD: 21.80; 95% CI 7.20–36.39; p = 0.0034), with greater effects in crush injuries and when combined with scaffolds, conduits, or autografts (up to MD: 54.41). Secondary outcomes showed improved gastrocnemius weight, muscle fiber area, nerve conduction velocity, and mechanical withdrawal thresholds. MSC-derived EVs, particularly scaffold-delivered, yielded the most consistent gains, while Schwann cell-derived EVs were effective as monotherapy. Mechanistically, EVs enhanced axonal growth, remyelination, angiogenesis, and M2 macrophage polarization while reducing inflammation.

Conclusions

EV-based therapies significantly enhance functional, structural, and electrophysiological recovery in preclinical PNI models, with synergistic benefits when combined with structural supports. Standardization of EV production and delivery, and validation in clinically relevant models, are essential for translation into human therapy.

Background

Peripheral nerve injuries (PNIs) remain a significant clinical challenge, often resulting in debilitating loss of sensory and motor function [1]. Despite the intrinsic regenerative capacity of the peripheral nervous system, spontaneous recovery is frequently incomplete, particularly in large-gap or delayed repair scenarios [2]. Autologous nerve grafting is considered the gold standard for bridging nerve defects, yet it is limited by donor site morbidity, tissue availability, and suboptimal functional outcomes [3]. These limitations have catalyzed the search for biologically active, cell-free alternatives capable of enhancing and accelerating nerve regeneration.

In recent years, extracellular vesicles (EVs) have emerged as a promising acellular therapeutic platform in regenerative medicine. EVs are nano-sized, membrane-enclosed vesicles secreted by virtually all cell types, carrying diverse molecular cargo, including proteins, lipids, and RNAs that mediate intercellular communication [4, 5]. Their role in neural repair has garnered particular attention, with accumulating evidence demonstrating that EVs derived from mesenchymal stem cells (MSCs), Schwann cells, and other progenitor sources can recapitulate many of the beneficial effects of their parent cells [6]. Unlike cell-based therapies, EVs offer superior safety profiles, reduced immunogenicity, and lower risk of tumorigenicity, while retaining robust paracrine signaling capacity [6]. Mechanistically, EVs exert their neuroregenerative effects through multiple converging pathways, such as axonal elongation and remyelination, changing macrophage polarization, contributing to angiogenesis, and supporting Schwann cell dedifferentiation [6].

Preclinical studies have demonstrated that EVs can enhance both motor and sensory recovery, improve nerve conduction, and support muscle preservation [7]. These findings are further potentiated when EVs are integrated with biomaterial scaffolds or autografts, suggesting a synergistic relationship between bioactive signaling and structural guidance. However, despite these advances, a comprehensive synthesis of preclinical data is lacking, leaving open questions regarding optimal EV sources, delivery strategies, and comparative efficacy. This comprehensive network meta-analysis aimed to systematically evaluate the preclinical evidence supporting the efficacy of EV-based therapies in PNI. By incorporating functional, structural, electrophysiological, and sensory outcomes, and stratifying results based on injury model, EV origin, and delivery modality, we provide a translational framework to guide future development of EV therapies in peripheral nerve repair.

Methods

Search strategy and study selection

This systematic review and meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines. A comprehensive literature search was performed across PubMed, Scopus, and Web of Science databases to identify eligible studies published up to 1 July 2025, investigating the efficacy of EV-based therapies in animal models of PNI. The search included various combinations of the following terms: “exosome,” “microvesicle,” “secretome,” “extracellular vesicle,” “peripheral nerve injury,” “nerve injury,” “nerve crush,” “nerve transection,” “nerve defect,” “nerve regeneration,” “nerve repair,” “sciatic nerve injury,” “median nerve injury,” “facial nerve injury.” Additional studies were retrieved by manually screening references of relevant reviews and included articles.

The eligibility criteria were defined according to the PICOS framework as follows: Population: in vivo animal models of peripheral nerve injury; Intervention: extracellular vesicle-based therapies; Comparator: sham, vehicle, scaffold, graft, cellular therapy, or untreated controls; Outcomes: functional, electrophysiological, histological, and sensory recovery outcomes; Study design: preclinical in vivo studies. Eligible studies met the following inclusion criteria: (1) employed in vivo animal models of PNI; (2) evaluated the therapeutic effect of EVs derived from any cell type; (3) included a control group receiving vehicle or no treatment; (4) reported at least one quantitative outcome relevant to nerve regeneration (e.g., functional, structural, electrophysiological, or sensory metrics); and (5) were published in English. Studies were excluded if they met any of the following criteria: (1) non-original publications, including reviews, editorials, conference abstracts, or protocols without primary data; (2) non-preclinical studies, such as clinical trials, in vitro experiments, or ex vivo models; (3) models not involving PNI, including central nervous system injuries; (4) interventions not involving EVs; (5) absence of an appropriate control group for comparative analysis; (6) articles published in languages other than English.

Data extraction and outcome measures

Two independent reviewers extracted data on study characteristics, including animal species, injury model (crush vs. transection), EV source (MSC-derived, Schwann cell-derived, or other sources), EV type (exosome or non-specified EVs), delivery route or combination with scaffolds or grafts, and outcome measures. Discrepancies were resolved by consensus or third-party adjudication. The primary outcome of this review was the Sciatic Functional Index (SFI), a widely used and validated measure of motor function recovery in rodent models of PNI. The SFI is a quantitative gait-based assessment derived from hind paw footprint parameters, including print length, toe spread, and intermediary toe spread. Values near 0 indicate normal function, whereas values approaching − 100 represent severe sciatic dysfunction. SFI was selected as the primary endpoint due to its frequent reporting and its ability to quantitatively reflect gait coordination and locomotor improvement. Secondary outcomes included gastrocnemius muscle weight ratio, muscle fiber cross-sectional area, compound muscle action potential (CMAP), nerve conduction velocity (NCV), mechanical withdrawal threshold, and thermal withdrawal latency, which reflect structural, sensory, and electrophysiological recovery processes. These outcomes were included in the quantitative meta-analysis only when data were available from at least three independent studies. When multiple time points were reported within a study, the latest time was extracted to best capture long-term functional recovery.

In addition, several outcomes that were reported inconsistently or lacked sufficient quantitative data across studies were included in the qualitative systematic review but excluded from meta-analysis. These included measures such as histological assessments (e.g., axon density, myelin thickness), inflammatory markers, and expression of regeneration-associated markers (e.g., GAP-43, MBP, NF200). These outcomes were synthesized narratively to complement the quantitative findings and provide broader insight into the regenerative mechanisms associated with EV therapy.

Quality assessment and risk of bias

The methodological quality and risk of bias for included studies were assessed using a modified CAMARADES checklist, which evaluates internal validity across ten domains, including randomization, blinded assessment of outcomes, sample size calculation, compliance with animal welfare regulations, and conflict of interest disclosure [8]. Each item was scored as “Yes,” “No,” or “Unclear,” and a cumulative quality score was calculated. This tool scores methodological quality across 10 domains, with scores categorized as follows: 0–3 low quality, 4–6 moderate quality, and > 6 high quality.

Statistical analysis

A structured, three-level meta-analytical approach was adopted to comprehensively evaluate the efficacy of EV-based therapies in PNI models. Pairwise meta-analysis was first performed to compare EV-treated groups versus controls, using a random-effects model. Continuous outcomes were pooled using mean differences (MDs) with 95% confidence intervals (CIs). Analyses were conducted using the “meta” and “metafor” packages in “R” version 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria). A second-tier network meta-analysis compared EV-based strategies in a stepwise therapeutic hierarchy, specifically including EV alone, scaffold alone, EV/scaffold, cells, and sham-operated controls. The third level involved a comprehensive network meta-analysis incorporating all reported treatment combinations. Network meta-analyses were performed using the netmeta package, applying a frequentist random-effects model. Network geometry plots and contribution matrices were used to visualize the comparative landscape. Between-study heterogeneity was quantified using the I² statistic, with values > 75% interpreted as substantial. Subgroup analysis based on PNI model, cell source, and EV type was also performed to address the potential existing heterogeneity among the included studies. To assess the influence of study quality on effect estimates, sensitivity analyses were conducted by stratifying studies according to their risk of bias, as determined by the CAMARADES checklist. Pooled estimates were reanalyzed within each quality subgroup to identify potential bias or instability in the overall findings. P values less than 0.05 were considered statistically significant.

Results

Study characteristics

A total of 524 records were identified through the initial systematic literature review. After removing duplicates, 521 unique records were screened by title and abstract. A total of 60 full-text articles were assessed for eligibility, and 35 studies met the inclusion criteria (Table 1) [9–43]. Of these, 32 were included in the meta-analysis. The study selection process is shown in the PRISMA flow diagram (Fig. 1).

Table 1.

Characteristics of included studies

Author (year)	Animal	Injury model	Injured nerve	Gap length (mm)	EV source	EV type	Timing	Frequency	Route	Isolation method	Outcomes
Mou et al. (2023)	Sprague-Dawley rats	Chronic constriction injury	Sciatic	NA	HU-MSC	Exosome	3 days before	Single	Intrathecal catheter	Ultracentrifugation	Iba-1, CCL2, IL-6, IL-1B, MWT, TWT
Ock et al. (2023)	C57BL/6 mice	Crush	Cavernous	NA	Mouse cavernous pericyte	Non-specified	3 days before	2	Intracavernous injection	EXOCET exosome isolation solution	ICP/MSBP, quantitative analysis of endothelial cell PECAM-1, pericyte NG-2, neuronal cell NF, number of TUNEL-positive or PH3-positive pericytes, signaling explorer antibody assays
Hu et al. (2023)	Sprague-Dawley rats	Transection	Sciatic	5	Schwann cell-like	Exosome	Immediately	Single	Conduit	Ultracentrifugation	SFI, relative wet weight of gastrocnemius muscle, cross-sectional area of individual muscle fibers, diameter of axon, number of NF200 positive axons, myelin density Myelin thickness, myelin G-ratio, vessel density based on positive CD31 staining
Chen et al. (2023)	Sprague-Dawley rats	Crush	Sciatic	NA	HU-MSC	Exosome	1 and 7 days after	2	Intramuscular injection	Ultrafilteration	Gastrocnemius muscle weight, gastrocnemius muscle Index, mean area of muscle fibers, muscle fiber density, mean number of slow and fast fibers, mean area of slow and fast muscle fibers, Fbxo32 and Trim63 level in muscle, relative protein expression
Yu et al. (2021)	Sprague-Dawley rats	Transection	Sciatic	10	Schwann cell	Non-specified	Immediately	Single	Conduit	exoEasy Maxi Kit	Distribution and cellular uptake of EVs in vivo, gait analysis and SFI, TWT, CMAP amplitude and latency, motor nerve conduction velocity, percentage of FG-labeled sensory neurons in DRGs, total number of FG-labeled MNs in spinal cord, the average distance of regenerated axons, fluorescence intensity of NF200- and S100- positive nerve fibers Thickness of myelin sheath, the diameter of myelinated axons, the G-ratio, wet weight ratio of gastrocnemius and Soleus and tibialis anterior muscles, cross-sectional area of muscle fibers
Sun et al. (2024)	Sprague-Dawley rats	Crush	Sciatic	NA	Schwann cell	Exosome	Immediately	Single	Direct injection	Ultracentrifugation	SFI, tube formation, transwell invasion, EdU assay, percentage of NF200 positive, percentage of MBP positive, lactate Production, pyruvate Production, Acetyl-CoA Production, overall levels of GLUT1, HK2, PFKFB3, and LDHA at the injury site, PDH-E1α expression
Errante et al. (2025)	Rattus Fisher	Transection	Sciatic	13	Schwann cell	Exosome	Immediately	Single	Conduit	Ultracentrifugation	Hargreaves test, Von fray filaments test, gastrocnemius muscle amplitude, gastrocnemius muscle latency, tibialis anterior muscle amplitude, tibialis anterior muscle latency, percent weight difference between injured and uninjured gastrocnemius muscles, autotomy score, epineurium-graft ratio, myelinated axon count, G ratio
Zhang et al. (2023)	Sprague-Dawley rats	Transection	Sciatic	10	BMSC	Non-specified	Immediately	Single	Conduit	Ultracentrifugation	SFI, CMAP amplitude, Motor nerve conduction velocity, gastrocnemius muscle wet weight ratio, gastrocnemius muscle cross-sectional area, density of myelinated nerve fibers, myelin thickness, diameter of myelinated axons, density of microvessels, percentage of positive NF200, percentage of positive S100
Li et al. (2022)	Sprague-Dawley rats	Transection	Sciatic	2	BMSC	Exosome	Immediately	Single	Conduit	Ultracentrifugation	SFI, CMAP amplitude, CMAP latency, gastrocnemius muscle wet weight ratio, gastrocnemius muscle cross-sectional area muscle fibers, density of myelinated axons, Thickness of myelin sheath, diameters of the regenerated axons
Liu et al. (2024)	C57BL/6 mice	Crush	Sciatic	NA	Schwann cell-like	Exosome	Immediately	Single	Intramuscular injection	Ultracentrifugation	Von Frey test, perpendicular anterior extreme position, perpendicular posterior extreme position, foot pressure, stance duration, swing duration, gastrocnemius muscle wet weight ratio, the retention time of the exosomes, biodistribution of the exosomes, nerve fiber area, axon area, myelin area, G-ratio, CD68 positive area, CSF-1R positive area
Chai et al. (2024) (1)	Sprague-Dawley rats	Crush	Sciatic	NA	Dental pulp	Exosome	Immediately	Single	Direct injection proximal to the injured site	Ultracentrifugation	The expression levels of proteins (LC3-I/II, Beclin-1, and P62), phosphorylation of proteins (PI3K, AMPK, ERK, and P53), mRNA levels of TP53
Wang et al. (2024)	Sprague-Dawley rats	Crush	Sciatic	NA	Schwann cell	Exosome	Immediately	Single	Direct injection proximal and distal to the injured location	NA	SFI, MBP and NF-200 fluorescence intensity
Chai et al. (2024) (2)	Sprague-Dawley rats	Crush	Sciatic	NA	Dental pulp	Exosome	Immediately	Single	Direct injection proximal to the injured site	Ultracentrifugation	SFI, myelin sheath thickness, myelin sheath diameter
Fan et al. (2025)	Sprague-Dawley rats	Transection	Sciatic	12	HU-MSC	Non-specified	Immediately	Single	Conduit	Ultracentrifugation	SFI, diameter of muscle fiber, wet weight ratio of gastrocnemius muscle, cross-sectional area of laminin in muscle fibers, thickness of myelin sheath, diameter of myelinated axons, number of myelin lamellar, G ratio, fluorescence intensity in NF200- and S100β-positive nerve fibers, expression of regeneration-associated genes (BDNF, GAL, EGR1, CNTF, ARG1, GDNF, ATF3, FOS, Nestin, S100Beta, VEGF, NF200), p-PI3 K, PI3 K, p-AKT, AKT, p-mTOR, mTOR, p-p38, p38, p-JNK, JNK, p-p70S6 K, and p70S6 K protein expression
Avites et al. (2022)	Sprague-Dawley rats	Transection	Sciatic	9–10	Olfactory MSC/OEC	Non-specified	Immediately	Single	Intraneural Injection	Centrifugation	Motor deficit, WRL time, SFI, SSI, muscle weight, fibers area, minimum Feret’s diameter
Aldali et al. (2025)	Sprague-Dawley rats	Crush	Sciatic	NA	iPSC	Exosome	Immediately	Single	Injection under the epineurium at both proximal and distal ends	Ultracentrifugation	SFI, grip strength, PWTs, body weight, muscle wet weight
Yang et al. (2023)	Sprague-Dawley rats	Crush	Sciatic	NA	BMSC	Exosome	Immediately	Single	Composite hydrogel implantation	Ultracentrifugation	Arg, INOS, PWT, TWL, SFI, muscle weight, muscle fiber, collagen deposit
Cui et al. (2023)	Sprague-Dawley rats	Transection	Sciatic	5	ADSC	Exosome	Immediately	Single	Intramuscular injection	Ultracentrifugation	Free walking pattern, SFI, PWTL
Pan et al. (2021)	Sprague-Dawley rats	Transection	Sciatic	15	iPSCs	Exosome	Immediately	Single	Incorporated into acellular nerve allograft	Ultracentrifugation	SFI, CMAP, NCV, MTRR, muscle Wet weight, fiber diameter, area of muscle fibers
Zeng et al. (2023)	Sprague-Dawley rats	Crush	Sciatic	NA	MDSC	Exosome	Immediately	Single	Peri-neural injection	Ultracentrifugation	SFI, gastrocnemius mass, axon/myelin staining (NF, S100), PTEN, JAK2/STAT3
Demyanenko et al. (2022)	Wistar rats	Crush	Sciatic	NA	Placental-derived MSC	Non-specified	1 h after injury	Single	Topical application in collagen gel	Ultracentrifugation	Muscle atrophy, fiber diameter, functional deficit (postural thrust), apoptosis, NF-200, GAP-43
Shen et al. (2024)	Sprague-Dawley rats	Transection	Sciatic	10	Skin precursor-derived Schwann cell	Non-specified	Immediately	Single	Conduit	Ultracentrifugation	Blood perfusion (Laser Doppler), CMAP, SFI, angiogenesis quantification, withdrawal time (thermal nociception)
Zhang et al. (2024)	Sprague-Dawley rats	Crush	Sciatic	NA	Platelet-rich plasma	Exosome	Immediately, 3 and 7 days after	3	Local injection under the epineurium	Ultracentrifugation	SFI, CMAP, muscle weight and stiffness, histology
Yi et al. (2023)	Sprague-Dawley rats	Crush	Sciatic	NA	Platelet-rich plasma	Exosome	1,7, and 14 days after	3	Intravenous (tail vein)	Ultracentrifugation	SFI, CMAP, muscle analysis
Wang et al. (2021)	Lewis rats	Transection	Sciatic	10	Purified exosome product	Exosome	Immediately	Single	Local application via fibrin glue	Freeze-thaw, filtration, and lyophilization cycles	CMAP, ITF, muscle weight, ankle contracture angle, histology
Wang et al. (2023)	Sprague-Dawley rats	Crush	Sciatic	NA	Skin precursor-derived Schwann cell	Non-specified	Immediately	Single	Local injection into proximal site of crushed sciatic nerve	Ultracentrifugation	SFI, plantar test, muscle wet weight, immunofluorescence
Sun et al. (2022)	Sprague–Dawley rats	Crush	Sciatic	NA	Schwann cell	Exosome	Immediately	Single	Local injection at the injury site	Ultracentrifugation	SFI, immunofluorescence, axon/myelin morphology
Pan et al. (2023)	Sprague–Dawley rats	Transection	Facial	5	EPI-NCSC	Exosome	Immediately	Single	Incorporated into acellular nerve allograft	Ultracentrifugation	CMAP, blink reflex, tentacle movement
Yeo et al. (2025)	Balb/C mice	Chronic constriction injury	Sciatic	NA	iPSC	Non-specified	3 days after	Single	Injection into the site of sciatic nerve injury	Ultracentrifugation	Von-Frey test, Rotarod test, wet muscle evaluation
Roballo et al. (2019)	FVB mice	Transection	Sciatic	NA	Neuron	Non-specified	Immediately	Single	Injection into the site of sciatic nerve injury	ExoQuick and ultracentrifugation	SFI, gastrocnemius wet weight, sciatic nerve histomorphometry, IVIS imaging, SNAP25, TUBIII, Map2 and Snap25, miR-184, miR-137, miR-124, miR-132
Chen et al. (2019)	Sprague-Dawley rats	Transection	Sciatic	10	ADSC	Exosome	Immediately	Single	Conduit	Ultracentrifugation	Wet weight Ratio of gastrocnemius, cross sectional area of gastrocnemius fiber, myelination assessment (NGF, BDNF, and CNTF levels), DRG Neurite outgrowth assay and histological assessment, mRNA expression of Ccnd1, Ccna2, Rac1, Cthrc1, and Mmp9
Xia et al. (2019)	Sprague-Dawley rats	Transection	Sciatic	5	OEC	Non-specified	Immediately	Single	Conduit	Ultracentrifugation	SFI, muscle weight recovery, average percentage of muscle fiber area, peak amplitude of CMAP, NCV, latency of CMAP, histomorphometric analysis of nerve, Alix, TSG101, CD63, calnexin
Rao et al. (2019)	Sprague-Dawley rats	Transection	Sciatic	10	Gingiva-derived MSC	Exosome	Immediately	Single	Conduit	Ultracentrifugation	SFI, muscle weight, muscle fiber remodeling, CMAP amplitude and latency, morphological analyses
Bucan et al. (2018)	Wistar rats	Crush	Sciatic	NA	ADSC	Exosome	Immediately	Single	Direct injection proximal and distal to the injured location	Ultracentrifugation	SFI, axonal regeneration (NF immunohistochemistry), Schwann cell proliferation (Ki67 immunofluorescence), DRG neurite outgrowth assay, growth factor analysis
Ma et al. (2018)	Sprague-Dawley rats	Transection	Sciatic	5	HU-MSC	Non-specified	1 day after	Single	Intravenous (tail vein)	Ultracentrifugation	SFI, muscle weight ratio, immunohistochemical analysis (Myelinated axon density and myelin thickness), MBP, NF-200, S100, IL-6, IL-1β, IL-10

ADSC Adipose-derived mesenchymal stem cell, BMSC Bone marrow-derived mesenchymal stem cell, CMAP Compound muscle action potential, DRG Dorsal root ganglion, EPI-NCSC Epidermal neural crest stem cell, EV Extracellular vesicle, HU-MSC Human umbilical cord-derived mesenchymal stem cell, iPSC Induced pluripotent stem cell, ITF Isometric tetanic force, MDSC Muscle-derived stem cell, MSC Mesenchymal stem cell, MTRR Maximum tension recovery ratio, MWT Mechanical withdrawal test, NCV Nerve conduction velocity, OEC Olfactory ensheathing cell, PWT Paw withdrawal test, PWTL Paw withdrawal latency, SFI Sciatic functional index, SSI Static sciatic index, TWT Thermal withdrawal test

Fig. 1.

The PRISMA workflow of the study

The 35 included studies were published between 2018 and 2025 and conducted across 7 countries, including China (n = 28) [9–11, 13–17, 19–25, 27–33, 36, 37, 39–41, 43], United States (n = 2) [12, 18], Portugal (n = 1) [35], Russia (n = 1) [34], Brazil (n = 1) [38], Germany (n = 1) [42], and South Korea (n = 1) [26]. Rats were the most commonly used species (n = 31) [9–17, 19–25, 27–36, 39–43], followed by mice (n = 4) [18, 26, 37, 38]. The majority of studies merely used male animals (n = 24) [9–11, 13–22, 24, 26, 28–30, 34–36, 39, 40, 43], while 7 studies [23, 25, 27, 31, 33, 37, 42] only included female animals and one [12] had a 48% male population. A wide array of outcomes were reported across the 35 studies, reflecting the multifactorial processes of nerve regeneration. Outcomes were categorized into five primary domains, including functional recovery, electrophysiology, histology and morphometric analysis, molecular and cellular responses, as well as angiogenesis and vascularization.

Functional recovery

Functional recovery was evaluated in 30 studies [9–12, 14, 15, 17–21, 23–25, 27–29, 31–43]. The Sciatic Functional Index (SFI) was most the commonly reported functional outcome (n = 25) [9–11, 14, 15, 17, 19–21, 23, 25, 27–29, 31–33, 35, 36, 38–43], with consistent improvements in EV-treated groups. Additional functional assessments included grip strength (n = 1) [9], Rotarod test (n = 1) [37], postural thrust and weight-bearing asymmetry (n = 1) [34], pain-related behaviors (n = 11) [9, 11, 12, 18, 20, 23, 24, 27, 35–37], such as Von Frey test for mechanical sensitivity and Hargreaves test or thermal withdrawal latency (TWL), as well as gait analysis (n = 3) [18, 34, 36], free walking pattern (n = 1) [20], and autotomy score (n = 1) [12] were also reported.

Electrophysiology

Electrophysiological evaluation was performed in 12 studies [11, 12, 19, 21, 22, 30–33, 36, 40, 41]. The reported outcomes in this category included CMAP amplitude and latency (n = 12) [11, 12, 19, 21, 22, 30–33, 36, 40, 41], motor NCV (n = 4) [31, 32, 36, 40], neurophysiological reflexes like blink reflex or tentacle movement (n = 1) [30]. These results consistently showed improved nerve conduction and functional reinnervation in EV-treated animals.

Histology and morphometric analysis

Histological findings from 27 studies supported structural regeneration [9, 10, 12, 16–19, 21–23, 25, 27–29, 31–43]. Reported outcomes included myelinated axon count/diameter (n = 12) [10, 12, 17, 18, 28, 29, 31–34, 36, 43], G-ratio (n = 5) [10, 12, 18, 28, 36], myelin thickness or lamellae (n = 6) [10, 28, 31, 33, 36, 43], cross-sectional area and diameter of muscle fibers (gastrocnemius, tibialis anterior, soleus)(n = 17) [9, 10, 16, 19, 21, 23, 28, 31–37, 39–41], and muscle weight (n = 24) [9, 10, 12, 16, 18, 19, 21–23, 25, 27, 28, 31–41, 43].

Molecular and cellular responses

Molecular outcomes (n = 18) also showed enhanced regeneration and reduced inflammation in EV-treated animals [10, 13–16, 18, 23–26, 28, 31, 34, 38–40, 42, 43]. These included neuroregenerative markers, such as BDNF, GAP-43, NF200, S100β, VEGF, CNTF, EGR1, ARG1, GDNF, ATF3, FOS, Nestin, GAL (n = 10) [10, 14, 25, 28, 31, 34, 38, 39, 42, 43], signaling pathways, such as PI3K/AKT/mTOR, JAK2/STAT3, PTEN, p38/JNK, p70S6K (n = 3) [10, 13, 25], microRNA and mRNA expression (n = 2) [38, 39], autophagy and apoptosis markers (e.g., LC3-I/II, Beclin-1, P62, TUNEL, TP53, PH3)(n = 3) [13, 26, 34], pro- and anti-inflammatory cytokines, such as IL-6, IL-1β, IL-10, CCL2, iNOS, ARG1 (n = 3) [23, 24, 43], and metabolic enzymes, such as GLUT1, HK2, LDHA, PFKFB3, PDH-E1α (n = 1) [15]. Besides, three studies visualized EV biodistribution via fluorescent labeling and in vivo imaging system (IVIS) [18, 36, 38].

Angiogenesis and vascularization

Angiogenesis and vascularization were evaluated in 4 studies [11, 26, 28, 31] using CD31+, PECAM-1 + endothelial cells, and NG2 + pericyte labeling, as well as microvessel density, blood perfusion (Laser Doppler), and tube formation assays.

Risk of bias assessment

Risk of bias was assessed using the CAMARADES 10-item checklist, adapted for preclinical in vivo studies. Each study was evaluated on ten criteria, including randomization, blinding, sample size calculation, and animal welfare compliance. The mean CAMARADES total score across all 35 studies was 6.4 ± 1.4 (range: 4–9). A total of 22 studies [9–25, 30, 34, 36, 40, 43] reported random allocation to treatment groups, while only 9 reported blinded outcome assessment [10–12, 15–17, 23, 26, 34]. No study reported sample size calculations. Animal welfare compliance and conflict of interest declarations were stated in all the included studies. The overall quality of the studies ranged from moderate to high. Accordingly, a total of 21 studies were deemed moderate-quality [9, 19–22, 25, 27–29, 31–33, 35–43], and a total of 14 studies were considered as high-quality [10–18, 23, 24, 26, 30, 34]. A summary of the CAMARADES scores is provided in Table 2.

Table 2.

CAMARADES 10-item checklist for risk of bias

Author (year)	Peer-reviewed publication	Temperature control	Random allocation	Blinded induction of injury	Blinded outcome assessment	Appropriate anesthesia	Appropriate animal model	Sample size calculation	Animal welfare compliance	Conflict of interest statement	Total (0–10)
Mou et al. (2023)	Yes	Yes	Yes	No	No	No	No	Yes	Yes	Yes	6
Ock et al. (2023)	Yes	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Yes	8
Hu et al. (2023)	Yes	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes	9
Chen et al. (2023)	Yes	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes	9
Yu et al. (2021)	Yes	Yes	Yes	No	No	No	Yes	Yes	Yes	Yes	7
Sun et al. (2024)	Yes	Yes	Yes	No	No	No	Yes	Yes	Yes	Yes	7
Errante et al. (2025)	Yes	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Yes	8
Zhang et al. (2023)	Yes	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes	9
Li et al. (2022)	Yes	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes	9
Liu et al. (2024)	Yes	Yes	Yes	No	No	No	Yes	Yes	Yes	Yes	7
Chai et al. (2024) (1)	Yes	Yes	Yes	No	No	No	No	Yes	Yes	Yes	6
Wang et al. (2024)	Yes	Yes	Yes	No	No	No	No	Yes	Yes	No	5
Chai et al. (2024) (2)	Yes	Yes	Yes	No	No	No	No	Yes	Yes	Yes	6
Fan et al. (2025)	Yes	Yes	Yes	No	No	No	No	Yes	Yes	Yes	6
Avites et al. (2022)	Yes	No	Yes	Yes	Yes	No	No	Yes	Yes	Yes	7
Aldali et al. (2025)	Yes	Yes	Yes	No	No	No	Yes	Yes	Yes	Yes	7
Yang et al. (2023)	Yes	No	Yes	No	No	No	Yes	Yes	Yes	Yes	6
Cui et al. (2023)	Yes	No	No	Yes	Yes	No	Yes	Yes	Yes	Yes	7
Pan et al. (2021)	Yes	No	No	No	No	No	No	Yes	Yes	Yes	4
Zeng et al. (2023)	Yes	Yes	No	No	No	No	No	Yes	Yes	Yes	5
Demyanenko et al. (2022)	Yes	Yes	No	No	No	No	Yes	Yes	Yes	Yes	6
Shen et al. (2024)	Yes	Yes	Yes	No	No	No	Yes	Yes	Yes	Yes	7
Zhang et al. (2024)	Yes	No	No	No	No	No	Yes	Yes	Yes	Yes	5
Yi et al. (2023)	Yes	No	No	No	No	No	Yes	Yes	Yes	Yes	5
Wang et al. (2021)	Yes	No	No	No	No	No	Yes	Yes	Yes	Yes	5
Wang et al. (2023)	Yes	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes	9
Sun et al. (2022)	Yes	Yes	No	No	No	No	Yes	Yes	Yes	Yes	6
Pan et al. (2023)	Yes	No	Yes	No	No	No	No	Yes	Yes	Yes	5
Yeo et al. (2025)	Yes	Yes	No	No	No	No	Yes	Yes	Yes	Yes	6
Roballo et al. (2019)	Yes	Yes	No	No	No	No	No	Yes	Yes	Yes	5
Chen et al. (2019)	Yes	No	No	No	No	No	No	Yes	Yes	Yes	4
Xia et al. (2019)	Yes	No	Yes	No	No	No	Yes	Yes	Yes	Yes	6
Rao et al. (2019)	Yes	No	No	No	No	No	Yes	Yes	Yes	Yes	5
Bucan et al. (2018)	Yes	No	No	No	No	No	No	Yes	Yes	Yes	4
Ma et al. (2018)	Yes	No	Yes	No	No	No	No	Yes	Yes	Yes	5

Meta-analysis

Three distinct meta-analyses were performed to evaluate the efficacy of EV therapies in animal models of PNI, each including a progressively broader set of treatment comparisons. The first meta-analysis compared EVs to the control. The second network meta-analysis incorporated studies using control, sham, cells, scaffold, EV/scaffold, and EV alone. The third detailed network included all available intervention combinations, including autograft, allograft, conduit, EV/autograft, EV/allograft, EV/conduit, cell therapy, sham, EV alone, and control.

Primary outcome

In the first meta-analysis (Fig. 2), comparing EV versus control across 18 studies (n = 311, I² = 99.8%), EV treatment significantly improved the SFI (MD: 21.80; 95% CI 7.20 to 36.39; p = 0.0034)(Table 3) [9–11, 14, 15, 17, 19–21, 23, 25, 27–29, 31–33, 35, 36, 40, 41, 43]. Regarding sensitivity analysis, this effect remained significant in both high-quality studies (8 studies, n = 127; MD: 25.71; 95% CI 5.11 to 46.32; p = 0.0145) and moderate-quality studies (10 studies, n = 184; MD: 18.60; 95% CI 7.57 to 29.63; p = 0.0009). Subgroup analysis by injury model showed greater benefit in crush injury models (MD: 23.62; 95% CI 11.17 to 36.08; p = 0.0002) compared to transection (MD: 19.51; 95% CI 2.57 to 41.60; p = 0.0834) in terms of SFI improvement. When stratified by cell source, both MSC-derived EVs (MD: 30.75; 95% CI 6.33 to 55.17; p = 0.0136) and Schwann cell-derived EVs (MD: 18.31; 95% CI 13.36 to 23.27; p < 0.0001) showed significant improvements in SFI compared with the control group. Subgroup analysis by EV type also showed that both exosomes (12 studies, n = 198; MD: 24.96; 95% CI 5.56 to 44.37; p = 0.0117) and non-specified EVs (6 studies, n = 113; MD: 14.73; 95% CI 11.63 to 17.83; p < 0.0001) significantly improved SFI compared to control.

Fig. 2.

Funnel plot for the meta-analysis of effects of EV therapy on sciatic functional index (SFI)

Table 3.

Results of meta-analyses of the sciatic functional index (SFI) as the primary outcome measure

Intervention (vs. control)	No. of studies	N	Effect size (MD)	P-value	Heterogeneity (I2)
First meta-analysis	18	311			99.8%
EV therapy			21.7953 [7.2013; 36.3892]	0.0034
Second network meta-analysis	22	452			99.3%
EV alone			20.5897 [ 11.3196; 29.8598]	< 0.0001
EV/scaffold			28.8440 [ 13.4436; 44.2443]	0.0002
Scaffold alone			14.2496 [ -1.8721; 30.3712]	0.0832
Cell therapy			9.5649 [-18.9630; 38.0927]	0.5111
Sham			52.0027 [ 37.9181; 66.0874]	< 0.0001
Third network meta-analysis	22	505			99.4%
Allograft			33.8751 [ -5.3303; 73.0805]	0.0904
Autograft			35.6551 [ 17.6528; 53.6575]	0.0001
Cell			10.2759 [-19.0930; 39.6448]	0.4929
Conduit			13.3661 [ -3.1831; 29.9153]	0.1134
EV			20.9323 [ 11.3890; 30.4757]	< 0.0001
EV/allograft			28.3951 [-10.7233; 67.5135]	0.1548
EV/autograft			54.4072 [ 24.9331; 83.8814]	0.0003
EV/conduit			30.3597 [ 14.6257; 46.0937]	0.0002
Sham			53.7502 [ 39.3608; 68.1396]	< 0.0001

EV Extracellular vesicle

In the second network meta-analysis (Fig. 3), comparing control, sham, cells, scaffold, EV/scaffold, and EV, 22 studies (n = 452, I² = 99.3%) were included (Table 3) [9–11, 14, 15, 17, 19–21, 23, 25, 27–29, 31–33, 35, 36, 40, 41, 43]. Sham group showed the highest functional scores (MD: 52.00; 95% CI 37.91 to 66.09; p < 0.0001), followed by EV/scaffold (MD: 28.84; 95% CI 13.44 to 44.24; p = 0.0002) and EV alone (MD: 20.58; 95% CI 11.32 to 29.86; p < 0.0001). Scaffold alone (MD: 14.25; 95% CI 1.87 to 30.37; p = 0.0832) and cell therapy (MD: 9.56; 95% CI 18.96 to 38.09; p = 0.5111) were not significantly different from control. In high-quality studies (19 studies, n = 380, I² = 99.4%), EV/scaffold (MD: 29.48; 95% CI 9.68 to 49.28; p = 0.0035) and EV alone (MD: 23.93; 95% CI 9.45 to 38.41; p = 0.0012) remained significantly effective and demonstrated functional scores approaching those observed in sham-operated animals (MD: 69.44; 95% CI 46.70 to 92.17; p < 0.0001), while scaffold (MD: 7.84; 95% CI 13.97 to 29.66; p = 0.4809) did not reach significance. In moderate-quality studies (12 studies, n = 265, I² = 97.4%), EV/scaffold (MD: 39.92; 95% CI 23.90 to 55.95; p < 0.0001), scaffold (MD: 31.10; 95% CI 15.08 to 47.13; p = 0.0001), and EV (MD: 18.43; 95% CI 10.77 to 26.10; p < 0.0001) were all significantly effective in improving SFI scores (Fig. 3).

Fig. 3.

Forest plot, network diagram, and funnel plot for the second network meta-analysis of effects of EV therapy combined with scaffolds on sciatic functional index (SFI). The control group was the reference. MD Mean difference

The third and most detailed network (Fig. 4) included 22 studies (n = 505, I² = 99.4%) comparing all relevant combinations: EV, EV/autograft, EV/allograft, EV/conduit, autograft, allograft, conduit, cell therapy, sham, and control (Table 3) [9–11, 14, 15, 17, 19–21, 23, 25, 27–29, 31–33, 35, 36, 40, 41, 43]. EV/autograft showed the largest improvement in SFI (MD: 54.41; 95% CI 24.93 to 83.88; p = 0.0003), followed by EV/conduit (MD: 30.36; 95% CI 14.63 to 46.09; p = 0.0002) and EV alone (MD: 20.93; 95% CI 11.39 to 30.48; p < 0.0001). Sham remained significantly different from the control group (MD: 53.75; 95% CI 39.36 to 68.14; p < 0.0001), and autograft alone also showed substantial benefit (MD: 35.66; 95% CI 17.65 to 53.66; p = 0.0001). EV/allograft produced a large but statistically non-significant effect (MD: 28.40; 95% CI 10.72 to 67.51; p = 0.1548). According to sensitivity analysis, in high-quality studies within this network (n = 217), EV/autograft (MD: 54.17; 95% CI 22.69 to 85.65; p = 0.0007), EV/conduit (MD: 31.03; 95% CI 10.00 to 52.06; p = 0.0038), and EV (MD: 24.45; 95% CI 8.91 to 39.99; p = 0.0020) remained significant. In moderate-quality studies (n = 288), EV/conduit (MD: 41.69; 95% CI 26.45 to 56.94; p < 0.0001), EV/allograft (MD: 40.92; 95% CI 14.26 to 67.59; p = 0.0026), and EV (MD: 18.39; 95% CI 11.16 to 25.62; p < 0.0001) were also significantly effective (Fig. 4).

Fig. 4.

Forest plot, network diagram, and funnel plot for the third comprehensive network meta-analysis of effects of all treatment combinations among the included studies on sciatic functional index (SFI). The control group was the reference. MD Mean difference

Subgroup analysis by injury model in the second network meta-analysis showed that in crush models (10 studies, n = 239), EV/scaffold (MD: 40.78; 95% CI 17.47 to 64.13; p = 0.0006) and EV alone (MD: 20.67; 95% CI 12.10 to 29.22; p < 0.0001) are significantly associated with improved SFI scores. Similarly, in transection injuries (12 studies, n = 213), EV/scaffold (MD: 40.02; 95% CI 23.34 to 56.71; p < 0.0001) and EV alone (MD: 15.91; 95% CI 1.55 to 30.27; p = 0.0299) significantly improved SFI scores. In the third network, EV-based treatments were effective in both crush and transection injuries. In crush injuries (10 studies, n = 239, I² = 98.1%), EV/conduit (MD: 40.80; 95% CI 17.47 to 64.13; p = 0.0006) showed notable improvements, and EV alone also remained significantly effective (MD: 20.67; 95% CI 12.11 to 29.22; p < 0.0001). In transection models (12 studies, n = 266, I² = 98.0%), EV/autograft yielded the strongest results (MD: 65.17; 95% CI 46.90 to 83.43; p < 0.0001), followed by EV/conduit (MD: 44.34; 95% CI 31.31 to 57.37; p < 0.0001), EV/allograft (MD: 44.02; 95% CI 19.56 to 68.49; p = 0.0004), and EV alone (MD: 15.91; 95% CI 4.47 to 27.34; p = 0.0064).

In addition, network subgroup analysis based on EV source was also performed. In this regard, MSC-derived EV combined with scaffold (MD: 32.15; 95% CI 16.20 to 48.10; p < 0.0001) showed the strongest effect on SFI improvement, followed by other source-derived EV alone (MD: 26.27; 95% CI 11.96 to 40.58; p = 0.0003) and Schwann cell-derived EV alone (MD: 20.35; 95% CI 7.27 to 33.42; p = 0.0023). Based on the third network, MSC-derived EV/autograft (MD: 51.02; 95% CI 19.06 to 82.99; p = 0.0018) had the strongest improving effect, followed by autograft alone (MD: 30.44; 95% CI 12.38 to 48.50; p = 0.0010), MSC-derived EV/conduit (MD: 27.23; 95% CI 10.85 to 43.61; p = 0.0011), other sources-derived EV alone (MD: 27.06; 95% CI 10.56 to 43.56; p = 0.0013), and Schwann cell-derived EV alone (MD: 20.47; 95% CI 5.38 to 35.57; p = 0.0078). Based on EV type, both exosomes (MD: 46.83; 95% CI 29.51 to 64.15; p < 0.0001) and non-specified EVs (MD: 27.4; 95% CI 19.89 to 34.93; p < 0.0001) showed significant improvements in SFI when used with a scaffold.

Secondary outcomes

In terms of secondary outcomes, gastrocnemius muscle weight ratio (%) was significantly improved (MD: 17.56; 95% CI 14.86 to 20.27; p < 0.0001) by EV therapy compared to the control group (Fig. 5). Similar to Sham (MD: 54.50; 95% CI 45.23 to 63.78; p < 0.0001), EV/scaffold (MD: 21.02; 95% CI 10.68 to 31.36; p < 0.0001), EV alone (MD: 16.56; 95% CI 8.82 to 24.30; p < 0.0001), and cell therapy (MD: 15.61; 95% CI 0.85 to 30.37; p = 0.0382) were significantly associated with improved gastrocnemius muscle weight ratio (Fig. 6). Besides, EV/allograft (MD: 44.00; 95% CI 14.96 to 73.05; p = 0.0030), EV/conduit (MD: 22.83; 95% CI 10.41 to 35.25; p = 0.0003), and EV alone (MD: 16.84; 95% CI 8.92 to 24.76; p < 0.0001) were associated with substantial improvements in gastrocnemius muscle weight ratio (17 studies, n = 466, I² = 98.0%)(Fig. 7).

Fig. 5.

Funnel plot for the meta-analysis of effects of EV therapy on gastrocnemius muscle weight ratio

Fig. 6.

Forest plot, network diagram, and funnel plot for the second network meta-analysis of effects of EV therapy combined with scaffolds on gastrocnemius muscle weight ratio (percentage). The control group was the reference. MD Mean difference

Fig. 7.

Forest plot, network diagram, and funnel plot for the third comprehensive network meta-analysis of effects of all treatment combinations among the included studies on gastrocnemius muscle weight ratio (percentage). The control group was the reference. MD Mean difference

Gastrocnemius muscle fiber cross-sectional area (µm²) also significantly improved (MD: 426.12; 95% CI 276.53 to 573.70; p < 0.0001) with EV therapy compared to the control group (Fig. 8). Based on network meta-analysis (14 studies, n = 396, I² = 99.3%), the increase in fiber area was highest with EV/allograft (MD: 1018.48; 95% CI 451.50 to 1585.46; p = 0.0004), followed by EV alone (MD: 443.00; 95% CI 293.52 to 592.49; p < 0.0001) and EV/conduit (MD: 363.74; 95% CI 118.53 to 608.96; p = 0.0036)(Figs. 9 and 10). Sensitivity analyses by quality were not feasible for all muscle-related outcomes due to the limited number of studies per subgroup.

Fig. 8.

Funnel plot for the meta-analysis of effects of EV therapy on gastrocnemius muscle fiber area

Fig. 9.

Forest plot, network diagram, and funnel plot for the second network meta-analysis of effects of EV therapy combined with scaffolds on gastrocnemius muscle fiber area. The control group was the reference. MD Mean difference

Fig. 10.

Forest plot, network diagram, and funnel plot for the third comprehensive network meta-analysis of effects of all treatment combinations among the included studies on gastrocnemius muscle fiber area. The control group was the reference. MD Mean difference

Mechanical withdrawal thresholds improved significantly (MD: 4.19; 95% CI 1.33 to 7.04; p = 0.0041) with EV therapy compared with the control group. Moreover, network meta-analysis (5 studies, n = 176, I² = 99.0%) showed that EV/conduit (MD: 5.89; 95% CI 0.09 to 11.69; p = 0.0467) and EV alone (MD: 3.62; 95% CI 0.03 to 7.21; p = 0.0479) could significantly improve mechanical withdrawal thresholds (Figs. 11 and 12). Nevertheless, thermal withdrawal latency (7 studies, n = 166, I² = 99.9%) did not show statistically significant effects for any EV-based treatment (p > 0.05) (Figs. 13 and 14). Electrophysiological outcomes showed that EVs significantly increased (MD: 11.70; 95% CI 4.07 to 19.32; p = 0.0026) nerve conduction velocity (4 studies, n = 54, I² = 98.1%). Overall, the CMAP recovery index (4 studies, n = 74, I² = 96.9%) showed a significant improvement (MD: 17.65; 95% CI 1.32 to 33.99; p = 0.0341) using EV therapy compared with the control group. According to the network meta-analysis (5 studies, n = 207, I² = 98.6%), EV alone (MD: 26.57; 95% CI −2.11 to 55.25; p = 0.0694) and EV/allograft (MD: 36.19; 95% CI 14.96 to 87.34; p = 0.1655), although improved CMAP recovery index, their effect was not statistically significant (Fig. 15). Sensitivity analyses by study quality as well as subgroup analyses were not conducted for these outcomes due to insufficient data.

Fig. 11.

Forest plot, network diagram, and funnel plot for the second network meta-analysis of effects of EV therapy combined with scaffolds on mechanical withdrawal test threshold (g). The control group was the reference. MD Mean difference

Fig. 12.

Forest plot, network diagram, and funnel plot for the third comprehensive network meta-analysis of effects of all treatment combinations among the included studies on mechanical withdrawal test threshold (g). The control group was the reference. MD Mean difference

Fig. 13.

Forest plot, network diagram, and funnel plot for the second network meta-analysis of effects of EV therapy combined with scaffolds on thermal withdrawal test (s). The control group was the reference. MD Mean difference

Fig. 14.

Forest plot, network diagram, and funnel plot for the third comprehensive network meta-analysis of effects of all treatment combinations among the included studies on thermal withdrawal test (s). The control group was the reference. MD Mean difference

Fig. 15.

Forest plot, network diagram, and funnel plot for the third comprehensive network meta-analysis of effects of all treatment combinations among the included studies on compound muscle action potential (CMAP) recovery index. The control group was the reference. MD Mean difference

Discussion

This systematic review and network meta-analysis comprehensively evaluated the preclinical efficacy of EV-based therapies in PNI models, synthesizing data from 35 studies. EVs significantly improved functional recovery as measured by the SFI, with further enhancement observed when EVs were combined with biomaterial scaffolds, nerve conduits, or autologous grafts. EV/autograft and EV/conduit combinations ranked highest for SFI improvement, consistently outperforming EV alone, scaffold alone, and standard surgical repairs. These benefits were echoed across structural, sensory, and electrophysiological outcomes. Subgroup analyses revealed that EV treatment was effective in both crush and transection injury models, although greater benefits were typically observed in crush injuries. Sensitivity analyses confirmed that the beneficial effects of EVs were robust in studies with lower risk of bias. Subgroup analyses revealed that EVs derived from MSCs and Schwann cells both yielded significant improvements, with MSC-EVs producing slightly more consistent effects across outcomes. Specifically, MSC-derived EVs showed the largest estimated effect when administered in combination with scaffolds. In contrast, Schwann cell-derived EVs and EVs from other sources (e.g., dental pulp, iPSCs) were significantly effective when delivered alone, but not when combined with scaffolds or conduits.

Functional recovery remains the primary endpoint in assessing therapeutic efficacy following PNI, as it directly correlates with real-world outcomes such as limb mobility and quality of life. Among preclinical models, the SFI is widely used due to its non-invasive, reproducible nature and its ability to reflect gait dynamics and neuromuscular coordination. In the present meta-analysis, EV-based therapies produced a statistically significant improvement in SFI compared to controls, and the most substantial effect was observed in groups receiving EVs in combination with autografts or conduits. This suggests that EVs can enhance not only intrinsic regeneration but also augment the reparative microenvironment provided by structural grafts [44–46]. The functional benefit of EVs alone was also notable, which emphasizes their standalone therapeutic potential. Although SFI remains one of the most commonly utilized functional outcomes in rodent sciatic nerve injury models, important limitations should be recognized, particularly in neurotmesis and long-gap injury paradigms [47–49]. Walking track analysis may be influenced by compensatory gait adaptation, chronic flexion contractures, altered ankle biomechanics, and muscle stiffness, potentially confounding interpretation of functional recovery independently of true axonal regeneration [48].

Moreover, subgroup analyses revealed superior efficacy in crush injury models versus transection models. These divergent regenerative responses between the two models in our analysis underscore the critical influence of injury type on therapeutic outcomes. Crush injuries retain endoneurial tubes and the extracellular matrix framework, which serve as potential conduits for regenerating axons and preserve Schwann cell alignment [50]. In this permissive environment, EVs can effectively modulate the local microenvironment by enhancing axonal sprouting, promoting Schwann cell activity, and reducing inflammation [42]. Conversely, transection injuries are characterized by a complete loss of continuity in nerve architecture, often requiring surgical reconstruction to re-establish physical alignment [50]. Here, EV monotherapy yielded less consistent results, likely due to the absence of natural scaffolds and the need for longer-distance axonal bridging. However, when delivered in combination with biomaterial conduits or autografts, EVs significantly improved outcomes even in transection models. This suggests that EVs function optimally when paired with structural supports that localize and retain their bioactive cargo, facilitate sustained release, and provide mechanical guidance to regenerating axons [6, 44–46].

The regenerative potential of EVs is inherently influenced by their cellular origin, which determines the composition and specificity of their bioactive cargo [4, 51–53]. The subgroup analysis also revealed that MSC-derived EVs provided the most robust improvement in functional and histological outcomes, particularly when combined with scaffolds or autografts. MSC-EVs are known to carry a diverse repertoire of therapeutic molecules, including immunomodulatory cytokines, pro-angiogenic factors, and regenerative miRNAs (e.g., miR-21, miR-17-92, miR-135b), which enable them to modulate multiple aspects of the repair process [4, 51–53]. Their synergistic performance with biomaterials likely stems from their ability to enhance vascularization, suppress fibrosis, and create a pro-regenerative milieu within the scaffold [54]. In contrast, Schwann cell-derived EVs, while less versatile in cargo, are specialized for neural repair. They are enriched in neurotrophic factors and axon-guidance molecules that directly support axonal regrowth, growth cone stabilization, and remyelination [55]. Interestingly, SC-EVs performed better as monotherapy than when combined with scaffolds, suggesting that their activity may rely on direct interaction with injured axons and glial cells. This may also reflect a temporal window of activity or sensitivity to scaffold composition, underscoring the importance of tailoring EV source and delivery context to the specific injury environment.

Beyond gross functional restoration, EV therapy yielded significant improvements in structural integrity, electrical conduction, and sensory modulation. Increases in muscle weight and fiber cross-sectional area observed in EV-treated animals are indicative of enhanced and timely reinnervation, which prevents muscle atrophy and maintains neuromuscular junction (NMJ) stability [23, 28]. Electrophysiologically, elevated CMAP and NCV values reflect restored axonal continuity, improved myelination, and effective signal propagation. These outcomes are underpinned by EV-mediated upregulation of neurofilament proteins (e.g., NF200), myelin-associated markers (e.g., MBP), and Schwann cell activation. Sensory recovery, as evidenced by improved mechanical withdrawal thresholds, is equally critical yet often underreported. Our analysis supports the role of EVs in attenuating neuropathic pain and sensory hypersensitivity, likely through modulation of inflammatory cascades, stabilization of ion channel expression, and repair of damaged dorsal root ganglia (DRG) axons [23, 42].

The therapeutic efficacy of EVs arises from their role as biologically active vectors that orchestrate intercellular communication across multiple cell types in the injured peripheral nerve microenvironment. EVs carry a complex cargo of signaling molecules, such as proteins, lipids, mRNAs, and regulatory non-coding RNAs that influence cell behavior in a context-specific manner [4, 51–53]. Upon release, EVs interact with target cells via surface receptor engagement, membrane fusion, or endocytosis, initiating intracellular cascades that modulate regeneration [4, 51–53]. A central mechanism involves the delivery of pro-regenerative microRNAs such as miR-21, miR-132, and miR-17-92, which inhibit intrinsic inhibitors like PTEN and activate downstream effectors in the PI3K/Akt/mTOR, JAK/STAT3, and Wnt/β-catenin pathways [25, 53, 54, 56]. These cascades promote neuronal survival, axonal growth, and cytoskeletal remodeling. Simultaneously, EVs induce Schwann cell dedifferentiation into a repair phenotype, which enhances their migration, proliferation, and secretion of neurotrophic factors such as NGF and BDNF [42, 54]. On the immune front, EVs downregulate inflammatory cytokines (e.g., TNF-α, IL-6) and promote M2 macrophage polarization, which is essential for debris clearance and matrix remodeling [57]. In addition, EVs facilitate revascularization by stimulating endothelial proliferation via pro-angiogenic miRNAs (e.g., miR-126, miR-135b) and VEGF delivery [58].

Despite promising preclinical evidence, the successful clinical translation of EV-based therapies for PNI requires overcoming several logistical and scientific hurdles. Foremost among these is the standardization of EV production, including donor cell selection, isolation protocols, storage conditions, and potency assays. Current heterogeneity across studies impairs reproducibility and complicates regulatory approval. Furthermore, the development of clinically adaptable, scaffold-compatible delivery platforms is essential to maximize therapeutic bioavailability and retention at the injury site. Innovative strategies such as EV-functionalized hydrogels, aligned nanofiber conduits, or encapsulated release systems should be prioritized to enable controlled, localized delivery [23]. Parallel efforts must focus on dose optimization, pharmacokinetics, and long-term safety profiling, particularly in large animal models that better replicate human anatomy and pathology. Additionally, sex as a biological variable and chronic or comorbid injury models remain understudied and should be systematically incorporated into future designs. Regulatory readiness can be enhanced by establishing GMP-compliant manufacturing pipelines, leveraging off-the-shelf MSC sources, and integrating EV characterization standards such as those outlined in the MISEV guidelines [59]. Finally, next-generation ‘designer EVs’ engineered for cargo enrichment or cell-specific targeting hold immense promise for precision nerve regeneration and merit active exploration in translational pipelines [60].

This meta-analysis represents the most comprehensive synthesis of preclinical data on EV-based interventions in PNI, combining both pairwise and network meta-analytical approaches to rank therapeutic efficacy across delivery modes, cell sources, and injury types. A key strength lies in the stratified subgroup analyses, which allow nuanced interpretation of context-specific effects and offer translational guidance for therapy optimization. Furthermore, the inclusion of diverse outcome domains, including functional, structural, electrophysiological, and sensory measures, provides a multidimensional assessment of therapeutic impact. Nevertheless, limitations exist. First, substantial heterogeneity across included studies, in terms of animal models, EV preparation methods, and outcome metrics, may introduce bias or inflate effect estimates. The exceptionally high heterogeneity observed across most analyses (I² frequently > 90%) substantially limits the precision and interpretability of pooled effect estimates. Therefore, the quantitative findings should primarily be interpreted as indicating an overall therapeutic trend favoring EV-based interventions rather than providing definitive estimates of treatment magnitude. The observed heterogeneity likely reflects major methodological and biological variability across studies, including differences in animal species, injury paradigms, defect length, timing of repair, EV isolation techniques, characterization protocols, dosing strategies, administration routes, scaffold composition, and follow-up duration. Accordingly, treatment rankings derived from network meta-analysis should be interpreted cautiously and viewed as hypothesis-generating rather than conclusive comparative evidence. Second, reporting quality was inconsistent, with many studies lacking sample size justification, randomization procedures, or blinding, as factors that compromise internal validity. The moderate methodological quality of many included studies may have contributed to effect size inflation, a phenomenon widely recognized in preclinical biomedical research. In particular, limited use of blinded outcome assessment, absence of sample size calculations, and incomplete reporting of randomization procedures increase susceptibility to performance and detection bias. These methodological limitations reinforce the need for cautious interpretation of pooled efficacy estimates. Third, most data are derived from young, male rodents in acute injury settings, limiting generalizability to aged, chronic, or comorbid patient populations. Importantly, most included studies relied on acute rodent sciatic nerve injury models, particularly short-gap crush or transection paradigms, which only partially replicate the complexity of clinically relevant human peripheral nerve injuries. Large-gap defects, delayed surgical repair, chronic denervation, scar formation, and age-related regenerative decline remain underrepresented in the current literature despite their major clinical relevance. Therefore, caution is warranted when extrapolating these findings toward human therapeutic application, and future studies should prioritize clinically representative large-animal and chronic injury models. Fourth, the absence of long-term outcome data and lack of in vivo tracking of EV biodistribution also leaves open questions about persistence, off-target effects, and integration. Addressing these gaps is essential for building the robust evidence base needed to support regulatory advancement and clinical trial design. Fifth, a major challenge across the included literature is the lack of methodological standardization in EV production and characterization. Studies varied substantially in EV isolation techniques, purification methods, particle characterization, dose quantification, administration timing, and delivery route, thereby limiting direct comparability between interventions. Importantly, inconsistent adherence to MISEV characterization recommendations may affect reproducibility, biological potency, and translational scalability. Further, the term ‘EV therapy’ should not be interpreted as a single uniform intervention, but rather as an umbrella category encompassing biologically diverse extracellular vesicle products derived from multiple cellular origins and processed using heterogeneous methodologies. Besides, the aggregation of heterogeneous EV formulations into broad therapeutic categories such as ‘EV,’ ‘EV/scaffold,’ or ‘EV/conduit’ inevitably simplifies substantial biological and methodological variability among interventions. Consequently, these treatment nodes should not be interpreted as representing mechanistically equivalent therapies. The validity of network meta-analysis relies on assumptions of transitivity and comparability between treatment nodes. However, given the marked heterogeneity, these assumptions may only be partially satisfied in the present analysis. Accordingly, indirect comparisons between treatment categories should be interpreted cautiously and viewed as exploratory rather than definitive comparative evidence. Future studies should prioritize standardized EV characterization pipelines, dose normalization strategies, and reporting frameworks to improve reproducibility and facilitate clinical translation. Sixth, additional subgroup analyses according to EV isolation technique, characterization strategy, administration route, treatment timing, and dose normalization were not consistently feasible due to limited subgroup sizes and incomplete reporting across studies.

Conclusions

EV-based therapies represent a promising experimental strategy for peripheral nerve repair, offering a cell-free, immunologically inert, and biologically active platform capable of modulating multiple facets of regeneration. The present meta-analysis confirmed that EVs, particularly those derived from MSCs and used in combination with structural supports, significantly improve motor, sensory, and electrophysiological outcomes in preclinical models. Their pleiotropic mechanisms of action, including immunomodulation, trophic support, axon guidance, and remyelination, allow them to interface effectively with the complex pathophysiology of PNI. As the field advances, efforts must focus on standardizing production, optimizing delivery, and validating efficacy in clinically relevant models. EVs may contribute to future regenerative strategies pending further validation in standardized and clinically relevant models.

Abbreviations

CI

Confidence interval

BCM

Biosynthesized cellulose membrane

CMAP

Compound muscle action potential

DLS

Dynamic light scattering

DMEM

Dulbecco’s Modified Eagle Medium

DRG

Dorsal root ganglion

EV

Extracellular vesicle

iPSC

Induced pluripotent stem cell

MD

Mean difference

MSC

Mesenchymal stem cell

NCV

Nerve conduction velocity

NGC

Nerve guidance conduit

NMJ

Neuromuscular junction

NTA

Nanoparticle tracking analysis

PBS

Phosphate-buffered saline

PNI

Peripheral nerve injury

SFI

Sciatic function index

TEM

Transmission electron microscopy

TWL

Thermal withdrawal latency

Author contributions

RT: Conceptualization, formal analysis, methodology, and writing-original draft; AA & AS: Data curation and writing-original draft. MC, MG, SOY, AR, ANA: Validation and writing-review & editing; RH: Conceptualization, validation, project administration, and writing-review & editing.

Funding

Not applicable.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.