Acutely denervated muscle EVs reshape neuronal mitochondrial metabolism via retrograde signaling to rescue peripheral nerve injury

Summary

Peripheral nerve injury causes muscle atrophy due to slow axonal regeneration, highlighting an unmet therapeutic need. Although neuromuscular interactions are classically viewed as unidirectionally nerve dominated, we show that acutely denervated muscle (adMu) regulates nerve regeneration via a retrograde signaling pathway. adMu initiates a trans-tissue regulatory mechanism through extracellular vesicles (EVs^adMu) that orchestrate neural energy homeostasis to accelerate regeneration. Functional profiling identifies IDH2 and CS as key metabolic enzymes within EVs^adMu. Neurons treated with EVs^adMu exhibit a 1.39-fold increase in NADPH/NADP⁺ ratio via IDH2, along with a 1.18- and 1.27-fold increase in NADH/NAD⁺ and FADH₂/FAD ratios via CS, fueling the tricarboxylic acid (TCA) cycle to enhance mitochondrial bioenergetics. This restores redox balance and energy supply, driving axonal regeneration. In sciatic nerve injury models, EV^adMu-microneedle conduits significantly promote energy metabolism and functional recovery. Together, our findings position adMu as a metabolic signaling center enabling retrograde regulation for nerve regeneration, offering potential for clinical translation.

Graphical abstract

Highlights

• •Acutely denervated muscle promotes nerve repair via retrograde signaling
• •EVs^adMu enhance axonal growth by improving energy metabolism
• •EVs^adMu deliver IDH2 and CS, boosting mitochondrial metabolism and redox balance
• •Targeting the muscle-nerve metabolic axis offers a promising therapy for PNI

Liu et al. find a retrograde signaling axis wherein acutely denervated muscle rescues peripheral nerve injury by delivering IDH2 and CS to injured neurons. These metabolic enzymes are shuttled via extracellular vesicles to boost neuronal mitochondrial metabolism and restore redox balance, thereby driving axon regeneration.

Introduction

Peripheral nerve injury (PNI) refers to traumatic damage to peripheral nerves, causing prolonged denervation of distal target organs.^1,2 Given the slow regeneration rate of peripheral nerves (1–3 mm/day), denervated muscle undergoes progressive atrophy and irreversible functional impairment.^3,4 As a core functional unit of the motor system, muscle and nerve exhibit coordinated interactions during development and regeneration.^5,6 Whereas prior research predominantly focused on unidirectional neural dominance over muscle, the regulatory role of denervated muscle in nerve regeneration remains poorly defined.^7,8 Elucidating this bidirectional crosstalk may provide important insights for optimizing PNI repair.

Muscle has long been regarded as a purely effector organ, functioning primarily as a passive contractile apparatus under efferent neural control.⁹ This traditional paradigm emphasizes its mechanical roles in locomotion and postural maintenance. However, emerging evidence has demonstrated that muscle serves not only as a contractile apparatus but also as an active participant in tissue repair, energy metabolism regulation, and systemic homeostasis maintenance.^10,11,12,13 A recent study revealed that folliculin-interacting protein 1 ablation in myofibers enhances functional revascularization of ischemic muscle by orchestrating macrophage recruitment, uncovering a critical myofiber-macrophage crosstalk mechanism in vascular regeneration.¹⁴ This exemplifies muscle’s key role as a signaling hub coordinating tissue repair.

The critical dependence of muscle on neural input for its survival and function is well established.^15,16,17 However, recent advances reveal that muscle itself plays an active role in this neuromuscular interdependence, exhibiting potent intrinsic plasticity.^18,19,20 This capacity holds considerable therapeutic potential for treating PNI. Studies have demonstrated that inducing NANOG expression reprograms adult muscle into a developmentally plastic state, promoting functional recovery through preserved neuromuscular junction receptivity and enhanced synaptic reintegration after PNI.²¹ These findings challenge the classical unidirectional paradigm of neuromuscular signaling, suggesting instead a sophisticated bidirectional communication system. However, the reverse active regulatory role of denervated muscle and its impact on nerve regeneration remain an unresolved scientific challenge.

Following denervation, muscle undergoes profound molecular reprogramming marked by extensive transcriptomic alterations in protein-coding genes.^22,23,24 This includes epigenetic remodeling such as HDAC4-mediated elevated histone acetylation, which helps maintain neuromuscular endplates.²⁵ Denervated muscle-derived exosomal miR-206 promotes the compensatory regeneration of neuromuscular synapses to slow ALS progression.²⁶ These findings reveal the epigenetically retained pro-regenerative potential within muscle, establishing denervation-induced muscle reprogramming as a potent regulator of cross-tissue repair. Nevertheless, whether this genomic-scale muscle reprogramming actively participates in nerve regeneration remains unexplored. Resolving this gap may uncover a different regulatory model of reverse nerve-muscle control, where the denervated target organ actively controls neural repair through muscle-driven molecular mechanisms, establishing a new regulatory dimension that expands the canonical view of unidirectional neural dominance.

Herein, this study demonstrates that denervation initiates a trans-tissue regulatory mechanism, whereby acutely denervated muscle (adMu) orchestrates axonal energy homeostasis via the muscle-nerve axis to drive regeneration. This represents a shift in understanding wherein the denervated muscle actively contributes to neural repair, refining the classical view of unidirectional neural dominance in neuromuscular interactions. Specifically, this effect is mediated through acutely denervated muscle-derived extracellular vesicles (EVs^adMu), representing a functionally active EV-dependent metabolic regulatory axis in nerve regeneration. We further identified IDH2 and CS as critical components of EVs^adMu that are specifically delivered to injured neurons. Within these neurons, the delivered metabolic enzymes modulate tricarboxylic acid (TCA) cycle activity and redox balance to enhance mitochondrial bioenergetics, thereby fueling the high-energy demands of axonal sprouting and elongation. These findings reposition denervated muscles as pivotal regulatory nodes in PNI, analogous to the emerging recognition of target organs as active participants in neuro-regeneration. Furthermore, we highlight the therapeutic promise of this pathway by establishing EV^adMu as a potential intervention, providing a translatable strategy to accelerate nerve repair and functional recovery.

Results

Denervation triggers an enhanced secretory reprogramming in muscle

To investigate the pro-regenerative potential of muscle following denervation, we established a unilateral sciatic nerve transection model in C57BL/6J mice and systematically tracked adaptive responses. Specifically, bilateral gastrocnemius samples were harvested at six time points: baseline (0 day) and 1, 3, 5, 7, and 14 days post-denervation (DPD) for atrophy progression analysis (Figure 1A). Progressive gastrocnemius wet weight loss was observed, reaching 49.78% ± 1.00% reduction vs. baseline by 14 days (Figure 1B), and corresponding myofiber diameter degeneration was confirmed (Figure 1C). Intriguingly, while progressive muscle atrophy was observed following denervation, the gastrocnemius exhibited a relatively slow atrophic rate during the initial 3 days, which is much slower than that of days 3–14 (Figure 1D), suggesting that there may exist intrinsic protective mechanisms in muscle that transiently counteract atrophy during the acute phase of denervation.

Figure 1.

Hypersecretory state of acutely denervated muscle

(A) Schematic of gastrocnemius collection workflow and downstream analysis. 5-mm sciatic nerve transection model for observing morphological and molecular changes in denervated gastrocnemius.

(B) Changes in gastrocnemius wet weight ratio (vs. the contralateral side) at predetermined time points (n = 10).

(C) WGA staining for gastrocnemius atrophy at predetermined time points. (Upper) Representative images, (bottom) myofiber diameter pseudocolor. Scale bars, 150 μm.

(D) Statistical evaluation of changes in the mean diameter of myofibers (n = 5).

(E) PCA of gastrocnemius gene expression at predetermined time points (n = 4).

(F) Multi-group volcano plot of DEGs in gastrocnemius at predetermined time points.

(G) DEG trend analysis in gastrocnemius at predetermined time points, with two muscle atrophy-associated genes highlighted.

(H) UpSet plot of shared DEGs in gastrocnemius at predetermined time points.

(I) GO biological process enrichment analysis of 1,312 persistently altered genes post-denervation.

(J) Heatmap of Z score-normalized ssGSEA scores for secretion-associated gene sets in gastrocnemius at predetermined time points. Rows, gene sets; columns, individual samples.

(K) Mean ssGSEA scores of secretion-related gene sets reflecting gastrocnemius secretory capacity post-denervation.

(L) Heatmap of secretion-related gene expression (CON vs. 3 DPD groups).

(M) Rab21-WGA co-staining of gastrocnemius cross-sections. (Left) Representative images. (Right) Rab21 fluorescence intensity statistics (n = 3). Scale bar, 150 μm.

(N) TEM assessment of MVB morphology and EV-like intraluminal vesicle number per MVB in gastrocnemius. (Left) Representative images. (Right) EV count per MVB statistics (n = 6). Red, myofibers; blue, MVB. Scale bar, 250 μm.

Statistical analysis: one-way ANOVA with Tukey’s multiple comparisons test (K); two-tailed Student’s unpaired t test (M and N). Data: mean ± SD. ∗p < 0.05, ∗∗∗p < 0.001.

To elucidate the molecular mechanism of this atrophy trajectory, we performed bulk RNA sequencing (RNA-seq) of gastrocnemius samples across all time points, and integrated public triceps surae transcriptomic data (GSE44259, GEO) to complement our findings (Figures S1 and S2). Unsupervised principal-component analysis (PCA) revealed distinct segregation of denervated and innervated groups along PC1 (78.3% variance), indicative of systemic transcriptomic reprogramming (Figures 1E and S1A). Temporal analysis of differentially expressed genes (DEGs) via multi-group volcano plots showed progressive expression changes (log2 fold-change), reflecting dynamic transcriptional reprogramming during atrophy progression (Figures 1F and S1B). Fuzzy c-means clustering of Z score-normalized data identified temporally co-regulated gene modules; cluster 1 included key atrophy regulators (MuRF-1 and atrogin-1) (Figure 1G), with expression trajectories strongly correlated with muscle atrophy progression. These data prompted functional investigation of these expression patterns. UpSet and Venn analyses identified genes with persistent expression changes across time points, representing a core transcriptional signature of denervation (Figures 1H and S1C). Subsequent Gene Ontology (GO) enrichment analysis of these DEGs revealed significant enrichment in biological processes of secretory pathway regulation and inflammatory response activation (Figures 1I and S1D). These results suggest a potential link between early secretory remodeling and biological transition of denervated muscle.

We next analyzed the secretory pathway dynamics in denervated muscle. Single-sample gene set enrichment analysis (ssGSEA) revealed peak secretory pathway activity in denervated gastrocnemius and triceps surae, both at 3 DPD (Figures 1J, 1K, S1E, and S1F). Notably, this secretory peak coincided with the “slow” atrophic phase, indicating a close link between secretory activation and delayed atrophy progression. We then detected marked upregulation of secretory pathway-related genes during acute denervation (Figures 1L and S1G), a finding corroborated by elevated Rab21 protein expression in gastrocnemius (Figure 1M). Multivesicular bodies (MVBs), precursors of EV biogenesis, are established indicators of secretory pathway activation.^27,28 Transmission electron microscopy (TEM) showed a marked increase in EV-like intraluminal vesicles within MVBs during acute denervation vs. controls (Figure 1N). These molecular and ultrastructural findings collectively suggest that denervated muscle actively engages in the denervation process via enhanced secretory activity, rather than mounting a passive response.

Acute denervation-reprogrammed muscle drives axon growth via retrograde signaling

Transcriptomic profiling revealed enhanced secretory activity in acutely denervated muscle, prompting investigation into the biodistribution and functions of secreted factors. GO enrichment analysis identified three core biological processes among DEGs (innervated muscle vs. adMu): secretory regulation, inflammatory regulation, and regeneration regulation (Figures 2A and 2B, and S2A). GSEA confirmed enrichment of pathways including positive regulation of exosomal secretion (NES = 1.43, p < 0.05) and acute inflammatory response (NES = 2.17, p < 0.0001) in acutely denervated gastrocnemius, with consistent patterns in triceps surae (Figures 2C and S2B). Combined with morphological findings, these imply that EVs are essential for denervation-initiated muscle adaptive reprogramming. We then conducted in-depth research on muscle-derived extracellular vesicles (Mu-EVs).

Figure 2.

Acutely denervated muscle drives nerve growth via retrograde signaling

(A) Schematic of muscle collection workflow and downstream analysis.

(B) GO enrichment analysis of DEGs (innervated vs. acutely denervated gastrocnemius).

(C) GSEA of DEGs (innervated vs. acutely denervated gastrocnemius).

(D) Schematic of AAV transduction in gastrocnemius and downstream analysis. 5-mm sciatic nerve transection model for evaluating in vivo Mu-EV distribution post-denervation.

(E) Representative images of EGFP-labeled Mu-EVs internalized by DRG (upper) and sciatic nerve (bottom). Scale bars: 100 and 200 μm.

(F and G) EGFP-EV fluorescence intensity statistics in DRG and sciatic nerve (n = 6).

(H) Schematic of transwell assay for gastrocnemius effects on DRG neurons and explants.

(I) Immunofluorescence of DRG neurons (upper) and explants (bottom) co-cultured with gastrocnemius for 24 h (βⅢ-tubulin, S100 staining) Scale bars: 100 μm and 1 mm.

(J) Sholl analysis of neurite branching complexity in DRG neurons treated with different gastrocnemius groups. (Upper) Representative normalized Sholl plots (Vehicle control, n = 5).

(K) Quantification of total neurite length in co-cultured DRG neurons (n = 5).

(L) Average length of top 5 longest axons in co-cultured DRG explants (n = 5).

(M) Ratio of total neurite area to explant body area in co-cultured DRG explants (n = 5).

(N–P) WB analysis of axonal regeneration markers (GAP43 and NF200) in co-cultured DRG neurons (n = 3).

Statistical analysis: two-way ANOVA-Bonferroni (F and G); one-way ANOVA with Tukey’s multiple comparisons test (K–M, O, and P). Data: mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; ns, not significant.

To trace in vivo Mu-EV biodistribution, we constructed two adeno-associated viral (AAV) vectors: AAV9-MHCK7-CD63-EGFP (specific Mu-EV labeling)²⁹ and AAV9-MHCK7-shRNA-Rab27a (Mu-EV secretion inhibition)³⁰ (Figure S3A). Both were locally injected into the left gastrocnemius of 6-week-old C57BL/6J mice (Figures 2D and S3B). Western blot (WB) confirmed Rab27a knockdown in gastrocnemius (Figures S3C and S3D). Quantitative analysis showed that 41.60% of isolated Mu-EVs were CD63-EGFP⁺ post-AAV9-MHCK7-CD63-EGFP injection, while AAV9-MHCK7-shRNA-Rab27a reduced Mu-EV secretion by 58.37% (Figures S3E–S3G). Decreased serum CD63-EGFP⁺ Mu-EVs (0.44%–0.17%) further validated vector efficacy (Figures S3H and S3I). Confocal microscopy of functionally connected tissues (sciatic nerve, DRG) at 3 DPD (Figure 2D) detected CD63-EGFP⁺ Mu-EVs in all non-control tissues (Figure 2E). Notably, EGFP intensity was significantly enhanced in ipsilateral DRG and sciatic nerve during acute denervation, indicating Mu-EV accumulation in neural tissues (Figures 2F and 2G). Comparable EGFP intensity in contralateral tissues suggested systemic Mu-EV transport. Collectively, denervation enhances Mu-EV secretion and their systemic trafficking to sciatic nerve and DRG.

To functionally validate the role of Mu-EVs in axonal regeneration, we established a transwell co-culture system with gastrocnemius samples in upper chambers and DRG explants in bottom chambers (Figure 2H). Acutely denervated gastrocnemius significantly enhanced axonal outgrowth compared to innervated controls after 24-h co-culture. This effect was substantially attenuated by pretreatment with AAV9-MHCK7-shRNA-Rab27a (Figure 2I). Quantitative analysis revealed that adMu increased neurite branching complexity by 1.26-, 1.46-, and 1.58-fold compared to Mu, adMu+shRab27a, and Vehicle groups, respectively (Figure 2J). Total axon length of neurons was significantly increased, representing 1.55-, 2.43-, and 3.03-fold of Mu, adMu+shRab27a, and Vehicle groups (Figure 2K). The DRG explant model recapitulated these findings, with the average length of the five longest axons maximal in the adMu group (2.06 ± 0.26 mm), followed by the Mu (1.53 ± 1.00 mm), adMu+shRab27a (1.12 ± 0.09 mm), and Vehicle groups (1.21 ± 1.28 mm) (Figure 2L). Furthermore, adMu (2.60 ± 0.09) significantly increased the neurite-to-explant area ratio, compared to the other groups (Mu: 2.40 ± 0.07; adMu+shRab27a: 2.03 ± 0.08; Vehicle: 1.95 ± 0.10) (Figure 2M). WB analysis confirmed upregulated axonal regeneration markers (GAP43 and NF200)³¹ in the adMu group relative to Vehicle (Figures 2N–2P). Collectively, these findings demonstrate that adMu potently stimulated axonal growth through an EV-mediated mechanism.

Mu-EVs mediate acutely denervated muscle-driven nerve regeneration promotion

To investigate the potential role of Mu-EVs in nerve regeneration following denervation, we isolated Mu-EVs from innervated (EV^Mu) and adMu (EV^adMu) groups (Figure 3A). TEM revealed that both Mu-EV types had characteristic cup-shaped morphology and homogeneous size distribution (Figure 3B). WB confirmed canonical EV markers (HSP90 and CD63) and absence of calnexin (used to exclude non-vesicular contaminants; Figure 3C). NTA showed monodisperse size distribution with a predominant peak at ∼100 nm (Figures 3D and 3E). Notably, Mu-EV production was significantly increased post-denervation (Figures 3F and S4A). These results comply with the Minimal Information for Studies of Extracellular Vesicles (MISEV2023) guidelines, confirming the validity and purity of the isolated Mu-EVs.³²

Figure 3.

Mu-EVs released by acutely denervated muscle promote axonal growth and nerve regeneration

(A) Schematic of Mu-EV isolation and characterization.

(B) TEM micrographs of EV^Mu and EV^adMu. Scale bar, 250 nm.

(C) WB analysis of EV markers (CD63 and HSP90) and negative control calnexin in EV^Mu and EV^adMu.

(D) Particle size distribution of EV^Mu and EV^adMu (NTA).

(E) Quantification of EV^Mu and EV^adMu particle size (NTA, n = 3).

(F) Quantitation of EV^Mu and EV^adMu particle concentrations (normalized to tissue protein content, NTA, n = 3).

(G) Schematic of EV stimulation assay for Mu-EV effects on DRG neurons.

(H) Immunofluorescence of PKH67-labeled Mu-EVs internalized by DRG neurons. (Left) Low magnification; (right) high magnification of boxed area. Scale bars: 50 and 25 μm.

(I) Immunofluorescence of DRG neurons (upper) and explants (bottom) treated with EV^Mu and EV^adMu for 48 h (βⅢ-tubulin, S100 staining). Scale bars: 200 μm (upper) and 1 mm (bottom).

(J) Sholl analysis of neurite branching complexity in DRG neurons (EV^Mu and EV^adMu treatment). (Upper) Normalized representative Sholl plots (Vehicle control, n = 5).

(K) Total neurite length in EV^Mu- and EV^adMu-treated DRG neurons (n = 5).

(L) Average length of top 5 longest axons in EV^Mu- and EV^adMu-treated DRG explants (n = 5).

(M) Ratio of total neurite area to explant body area in EV^Mu- and EV^adMu-treated DRG explants (n = 5).

(N) Schematic of EV stimulation for transcriptomic analysis.

(O) RNA-seq analysis of regulated pathways in EV^Mu/EV^adMu-treated neurons. (Left) DEG heatmap (dashed line: EV^adMu vs. EV^Mu gene expression trend). (Right) Functional enrichment of DEGs.

(P) Bean plot of upregulated critical pathway genes in EV^adMu-treated neurons.

(Q) GSEA of DEGs (EV^Mu vs. EV^adMu-treated neurons).

(R) Schematic of in vivo EV^adMu-mediated axonal growth assessment. 5-mm sciatic nerve crush model for evaluating the nerve regeneration efficacy.

(S) Immunofluorescence of sciatic nerve longitudinal sections (SCG-10 staining, 72 h post-crush, EV^Mu/EV^adMu treatment). Scale bars, 500 μm.

(T) Regenerating axon percentage beyond crush site (normalized to lesion center) (n = 3).

(U) TEM micrographs of sciatic nerve cross-sections (EV^Mu/EV^adMu treatment). (Upper) Low magnification; (bottom) high magnification of boxed region. Scale bars: 10 μm and 1 μm.

(V–Y) Quantification of myelinated axon number (V), myelin sheath thickness (W), and G-ratio (X and Y) in sciatic nerve (n = 6).

Statistical analysis: two-tailed Student’s unpaired t test (E and F); one-way ANOVA with Tukey’s multiple comparisons test (K–M, V, and X); two-way ANOVA with Sidakʼs test (T). Data: mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. ns: not significant. (W and X) Large solid dots, biological replicates (individual samples); small translucent dots, technical replicates (same sample repeated measurements).

We next evaluated the functional impact of EV^adMu on axonal growth (Figure 3G). PKH26-labeled EV^Mu and EV^adMu exhibited strong colocalization within neurons, indicating efficient internalization (Figure 3H). Quantitative analysis revealed that EV^adMu significantly enhanced neurite branching complexity (1.64-fold vs. EV^Mu, 4.81-fold vs. Vehicle) (Figures 3I and 3J) and increased total axonal length (1.54-fold vs. EV^Mu, 2.35-fold vs. Vehicle) (Figure 3K). In DRG explants, the EV^adMu treatment promoted robust axonal growth, with the average length of the longest five axons averaging 1.53 ± 0.05 mm, exceeding both EV^Mu (1.17 ± 0.09 mm) and Vehicle (0.88 ± 0.05 mm) (Figure 3L). The neurite-to-explant area ratio was similarly enhanced in the EV^adMu group (1.24-fold vs. EV^Mu; 1.63-fold vs. Vehicle) (Figure 3M).

To investigate the molecular mechanisms of EV^adMu-mediated axonal growth, we performed RNA-seq on EV^Mu- vs. EV^adMu-treated DRG neurons (Figure 3N). PCA revealed clear segregation between groups, indicating extensive transcriptomic alterations induced by EV^adMu (Figure S4B). Comparative transcriptome analysis identified 3,663 DEGs (1,355 upregulated, 2,308 downregulated; |log2(FC)|>0.585, p < 0.05) between the EV^Mu and EV^adMu groups (Figures 3O and S4C). GO enrichment analysis of upregulated DEGs revealed significant enrichment in key neuro-regenerative pathways (axonogenesis, axon guidance, and axon regeneration) and other neural development/repair processes (Figure 3O), with higher expression of pathway-related genes in EV^adMu vs. EV^Mu (Figure 3P). GSEA uncovered notable enrichment in synapse organization, cell migration regulation, cell projection morphogenesis, and axon ensheathment pathways (Figure 3Q). These findings confirm EV^adMu potential to promote axonal growth.

We next assessed the therapeutic potential of EV^adMu in a sciatic nerve crush model (Figure 3R). In vivo imaging detected DiR-labeled EV^adMu at the nerve injury site (Figure S4D), and ex vivo confocal imaging confirmed its uptake into neurons (Figure S4E). EV^adMu significantly enhanced nerve regeneration in C57BL/6J mice vs. Vehicle and EV^Mu controls (Figures 3S and 3T). Fluoro-gold (FG) retrograde tracing showed that EV^adMu increased FG-labeled sensory neurons by 67.67% (Vehicle) and 29.03% (EV^Mu) and motor neurons by 146.67% (Vehicle) and 54.42% (EV^Mu) (Figures S4G), indicating robust sensory and motor axon regeneration across the injury site. Furthermore, EV^adMu treatment exhibited superior remyelination compared to controls (Figure 3U). Quantitative analysis revealed a 29.01% increase in myelinated fiber density relative to Vehicle and 12.4% relative to EV^Mu (Figure 3V) and enhanced myelin sheath thickness by 2.50-fold (Vehicle) and 1.57-fold (EV^Mu) (Figure 3W). Notably, EV^adMu normalized the G-ratio to near-physiological values (Figures 3X and 3Y). Additionally, axon diameters were significantly larger in the EV^adMu group (3.38 ± 0.16 μm), representing 33.1% and 12.7% increases over Vehicle (2.54 + 0.34 μm) and EV^Mu (3.00 ± 0.11 μm) groups (Figure S4H). These findings demonstrate that acute denervation triggers EV^adMu release, which potently enhances axonal regeneration and remyelination.

To determine if structural regeneration translated to functional recovery, we performed behavioral and electrophysiological analyses. The sciatic functional index (SFI), assessed by longitudinal CatWalk gait,³³ revealed a 53.86% improvement in locomotor function in EV^adMu-treated mice at 21 days post-injury (p < 0.001), alongside restored paw placement patterns (Figures S5A and S5B). Electrophysiological analysis showed that EV^adMu treatment significantly increased compound muscle action potential (CMAP) amplitude (45.82%) and nerve conduction velocity ([NCV] 38.63%), while reducing CMAP latency (31.90%) compared to Vehicle (p < 0.001) (Figures S5C–S5F), indicating improved neuromuscular signal transmission. These data confirm that EV^adMu not only promotes structural regeneration but also underpins functional reconnection of severed nerves and neuromuscular junctions.

EV^adMu orchestrates mitochondrial bioenergetic reprogramming in neurons

To investigate the mechanisms of EV^adMu-mediated axonal growth, we performed comprehensive proteomic profiling via liquid chromatography-tandem mass spectrometry (Figure 4A). PCA of proteomic profiles revealed pronounced inter-group heterogeneity but high intra-group homogeneity in protein cargo between EV^Mu and EV^adMu (Figure S6A). We identified 3,658 proteins in EV^Mu and EV^adMu, including canonical EV markers and muscle-specific markers, validating Mu-EV isolation specificity (Figure S6B). To pinpoint functionally relevant molecules driving EV^adMu’s pro-axonal activity, we focused on the 140 proteins that were significantly upregulated in EV^adMu compared to EV^Mu and subjected them to enrichment analyses (Figures 4B and 4C). GO enrichment analysis revealed a unique molecular signature with pronounced enrichment in ATP biosynthesis, oxidative phosphorylation, TCA cycle, and other related pathways (|log₂(FC)|>1, p < 0.05) (Figure 4D). KEGG enrichment analysis showed pathway clustering in TCA cycle, glycolysis/gluconeogenesis, oxidative phosphorylation, and other related pathways (Figure 4E). GSEA network analysis further corroborated these findings, with predominant involvement in aerobic respiration, carboxylic acid catabolic process, and mitochondrial translation (Figure 4F). These findings indicate that EV^adMu possesses the capacity to regulate mitochondrial metabolism.

Figure 4.

Validation of EV^adMu-mediated regulation of energy metabolism in neurons

(A) Schematic of Mu-EV proteomic analysis and neuronal energy metabolomics post-Mu-EV stimulation.

(B) Volcano plots of DEPs (EV^Mu and EV^adMu).

(C) Heatmap of upregulated DEPs (EV^adMu vs. EV^Mu).

(D and E) GO biological process and KEGG pathway enrichment of upregulated DEPs (EV^adMu vs. EV^Mu).

(F) GSEA for DEPs (EV^adMu vs. EV^Mu).

(G) Heatmap of TCA cycle-associated DEMs in EV^Mu/EV^adMu-treated neurons.

(H) Expression profiles of TCA cycle-associated DEMs.

(I) Neuronal OCR (EV^Mu/EV^adMu treatment). Quantification of ATP production and basal and maximal respiration (n = 5).

(J) JC-1 staining for mitochondrial membrane potential (EV^Mu/EV^adMu-treated neurons). (Upper) Representative images; (bottom) red/green fluorescence ratio. Scale bars, 50 μm.

(K) Quantification of JC-1 red/green fluorescence ratio (n = 3).

(L) MitoTracker Green staining for mitochondrial visualization (EV^Mu/EV^adMu-treated neurons). (Upper) Representative images; (bottom) mitochondrial size pseudocolor. Scale bars: 10 and 3 μm.

(M and N) Mitochondrial count per cell (n = 5) and mitochondrial area (n = 3).

(O) TEM micrographs of neuronal mitochondria (EV^Mu/EV^adMu treatment). (Upper) Low magnification; (bottom) high magnification (arrows: mitochondrial swelling). Yellow, nuclei; blue, cytoplasm; magenta, mitochondria. Scale bars: 1 μm and 250 nm.

(P–S) Quantification of mitochondrial count (n = 5), size (n = 3), cristae number (n = 3), and total cristae length/mitochondrial area (n = 3).

Statistical analysis: one-way ANOVA with Tukey’s multiple comparisons test (I, K, M, N, and P–S). Data: mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; ns, not significant. (K, N, and Q–S) Large solid dots, biological replicates (individual samples); small translucent dots, technical replicates (same sample repeated measurements). Each color represents one sample.

Energy-targeted metabolomics identified differentially expressed metabolites (DEMs) in EV^adMu-treated neurons. PCA showed complete separation of metabolic profiles among Vehicle, EV^Mu, and EV^adMu groups (Figure S6C). Notably, EV^adMu elevated TCA cycle-related metabolites (alpha-ketoglutarate, cis-aconitic acid) vs. EV^Mu and Vehicle, indicating robust mitochondrial TCA cycle activation (Figures 4G and 4H).

Given the central role of mitochondria in cellular metabolism, particularly TCA cycle regulation within their matrix,³⁴ we assessed their functional response to EV^adMu. Extracellular flux analysis (Seahorse XFe96 Extracellular Flux Analyzer) revealed that EV^adMu significantly enhanced ATP production vs. both controls (Figure 4I). Further mitochondrial function assessment via JC-1 staining (a sensitive oxidative metabolism indicator)³⁵ showed higher aggregate/monomer ratios in EV^adMu-treated neurons, reflecting elevated Δψm, enhanced ATP synthesis, and improved bioenergetics. Quantitatively, EV^adMu increased this ratio by 1.20-fold (vs. EV^Mu) and 1.58-fold (vs. Vehicle) (Figures 4J and 4K).

MitoTracker analysis revealed distinct mitochondrial size distributions across treatment groups, visualized via thermal intensity gradients (Figure 4L). EV^adMu-treated neurons showed 1.21-fold (vs. EV^Mu) and 1.62-fold (vs. Vehicle) increases in mitochondrial density (Figure 4M). Mean mitochondrial cross-sectional areas were significantly larger in the EV^adMu group vs. EV^Mu and Vehicle controls (Figure 4N), suggesting enhanced mitochondrial biogenesis and function. TEM further corroborated these findings, revealing improved mitochondrial ultrastructure in EV^adMu-treated neurons (Figure 4O), including increased numerical density (Figure 4P), larger cross-sectional area (Figure 4Q), greater cristae number (Figure 4R), and expanded cristae surface area (Figure 4S). Collectively, these data demonstrate that EV^adMu enhances neuronal bioenergetics by increasing mitochondrial quantity, optimizing ultrastructure, and potentiating TCA cycle flux.

EV^adMu-delivered IDH2 and CS boost neuronal bioenergetics via mitochondrial regulation

With mitochondrial metabolism confirmed as the central mechanism, we identified key protein mediators in EV^adMu responsible for this effect. Systematic analysis of the 140 upregulated proteins revealed 10 candidates significantly enriched in EV^adMu and functionally linked to TCA cycle regulation (Figure 5A). Notably, four rate-limiting TCA cycle enzymes (IDH2, IDH3α, IDH3γ, and CS) were selectively enriched in EV^adMu, as further validated by WB analysis (Figures 5B and 5C). Given their pivotal role in controlling TCA cycle flux, we prioritized these four enzymes for subsequent mechanistic investigation.

Figure 5.

Identification of critical proteins in EV^adMu-mediated orchestration of energy metabolic reprogramming and axonal growth

(A) Heatmap of TCA cycle-associated upregulated DEPs (EV^adMu vs. EV^Mu).

(B and C) WB analysis of 4 candidate enzymes in EV^Mu/EV^adMu (n = 3).

(D) Schematic of Mu-EV supplementation for downstream analysis.

(E and F) WB analysis of 4 candidate enzymes in EV^Mu/EV^adMu-treated neurons (n = 3).

(G) mRNA levels of 4 candidate enzymes in EV^Mu/EV^adMu-treated neurons (n = 4).

(H) Schematic of target protein knockdown and validation. 5-mm sciatic nerve transection model for denervated gastrocnemius Mu-EV isolation.

(I) Neuronal OCR post-target protein-knockdown EV^adMu treatment. Quantification of ATP production and maximal respiration (n = 5).

(J) NADPH/NADP⁺ ratio and ROS levels in treated neurons (n = 5); schematic of EV^adMu-derived IDH2-mediated neuronal antioxidant effect.

(K) NADH/NAD⁺ ratio, FADH2/FAD ratio and ATP levels in treated neurons (n = 5); schematic of EV^adMu-derived CS promoting neuronal ATP production.

(L) Schematic of different supplementation experiments for downstream analysis.

(M) WB analysis of IDH2 and CS in IDH2-knockdown EV^adMu-treated neurons.

(N) ATP levels in treated neurons (n = 5).

(O) Immunofluorescence of DRG neurons (upper) and explants (bottom) (βⅢ-tubulin, S100 staining). Scale bars: 100 μm and 1 mm.

(P) Sholl analysis of DRG neuronal neurite branching complexity. (Upper) normalized representative Sholl plots (Vehicle control, n = 5).

(Q) Average length of top 5 longest axons in treated DRG explants (n = 5).

(R) Schematic of in vivo EV^adMu-mediated metabolic reprogramming assessment. 5-mm sciatic nerve crush model for assessment of nerve regeneration efficacy.

(S) NADPH/NADP⁺ ratio and ROS levels in treated sciatic nerves (n = 5).

(T) NADH/NAD⁺ ratio, FADH₂/FAD ratio and ATP levels in treated sciatic nerves (n = 5).

(U) Immunofluorescence of sciatic nerve longitudinal sections (72 h post-crush, EV^adMu and shIdh2/Cs-EV^adMu treatment; SCG-10 staining). Scale bars, 500 μm.

(V) Regenerating axon percentage beyond crush site (normalized to lesion center) (n = 3).

(W) Schematic of EV^adMu-mediated metabolic reprogramming.

Statistical analysis: two-way ANOVA with Sidak’s test (C, F, and G); one-way ANOVA with Tukey’s multiple comparisons test (I, J, K, N, Q, S, and T). Data: mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; ns, not significant.

Quantitative analysis post-EV^adMu treatment showed significantly increased protein expression of IDH2 and CS, while IDH3α and IDH3γ subunits remained unchanged (Figures 5D–5F). This differential regulation might be attributed to the inherent structural instability of IDH3α/γ heterodimers, potentially limiting their post-delivery accumulation in neurons. Notably, quantitative reverse-transcription PCR confirmed that these changes occurred post-transcriptionally, as mRNA levels of all four enzymes were unaltered (Figure 5G). These results indicate direct mediation of neuronal IDH2 and CS accumulation by EV^adMu delivery.

To validate IDH2 and CS as energy metabolic regulators in EV^adMu, we engineered two tissue-specific AAV constructs (AAV9-MHCK7-shRNA-Idh2 and AAV9-MHCK7-shRNA-Cs) for gastrocnemius-targeted knockdown of Idh2 and Cs, respectively (Figures 5H and S7A–S7D). WB analysis confirmed significant target protein reductions (55.19% for IDH2 and 57.86% for CS; Figures S7E and S7F). Post-AAV transduction, EVs retained typical morphology and size distribution (predominant peak at ∼100 nm) with unaltered particle concentration (Figures S7G–S7I). Importantly, EV^adMu from Idh2/Cs knockdown gastrocnemius showed corresponding IDH2 and CS content decreases (Figures S7J and S7K), establishing an effective model to investigate functional contributions of IDH2 and CS ablation in EV^adMu in modulating neuronal mitochondrial metabolism.

Intriguingly, simultaneous knockdown (both IDH2 and CS) profoundly attenuated the positive regulatory effects of EV^adMu on neuronal energy metabolism compared to individual knockdown of either protein alone (Figure 5I). Furthermore, combined depletion of IDH2 and CS induced a more pronounced reduction in mitochondrial membrane potential (Δψm) than individual knockdown, as demonstrated by a decreased aggregate/monomer ratio (Figures S8A and S8B). MitoTracker staining-based mitochondrial quantification further revealed that dual ablation caused greater declines in mitochondrial number and area than single-protein depletion (Figures S8C–S8E). Consistent with these findings, TEM analysis showed simultaneous IDH2/CS loss caused more severe mitochondrial defects (reduced organelle number, size, cristae density, and surface area) than single-protein depletion (Figures S8F–S8J). These results collectively identify both IDH2 and CS as pivotal regulatory molecules in EV^adMu-mediated metabolic reprogramming, underscoring their essential roles in sustaining neuronal energy homeostasis.

We next investigated how IDH2 and CS regulate neuronal energy metabolism post-EV^adMu treatment. IDH2 converts NADP⁺ to NADPH, playing a key role in mitochondrial antioxidant defense.³⁶ IDH2-knockdown EV^adMu (6.05 ± 0.33) markedly reduced neuronal NADPH/NADP⁺ ratio compared to EV^adMu (8.16 ± 1.02), with comparable values to the Vehicle group (5.84 ± 0.84). Furthermore, mitochondrial reactive oxygen species (ROS) analysis revealed that IDH2-knockdown EV^adMu (1,551 ± 110 RFU) significantly increased the neuronal mitochondrial ROS levels vs. EV^adMu (1,221 ± 105 RFU), an effect reversed by NADPH supplementation (1,294 ± 77 RFU). These results indicate that IDH2 is crucial for EV^adMu-mediated maintenance of mitochondrial homeostasis, generating NADPH to counteract oxidative stress (Figure 5J).

As the first and rate-limiting TCA cycle enzyme, CS initiates the cycle to support subsequent NADH/FADH₂ generation for ATP production. CS knockdown significantly reduced neuronal NADH/NAD⁺ (0.07 ± 0.01 vs. 0.08 ± 0.01) and FADH₂/FAD (0.35 ± 0.04 vs. 0.48 ± 0.05) ratios in EV^adMu-treated neurons Additionally, CS-knockdown EV^adMu significantly reduced ATP production compared to EV^adMu (3.48 ± 0.55 vs. 4.32 ± 0.32 pmol/μg protein, p < 0.05), which was rescued by citrate supplementation, demonstrating CS’ essential role in maintaining TCA cycle-driven ATP generation (Figure 5K). Collectively, these findings identify IDH2 and CS as key regulators through which EV^adMu enhances neuronal bioenergetics.

EV^adMu-mediated energy metabolic reprogramming promotes axonal growth

Given previous reports linking enhanced mitochondrial metabolism to axon regeneration,^37,38,39 we investigated whether EV^adMu promotes axonal growth through IDH2/CS-dependent energy metabolic reprogramming (Figure 5L). To specifically verify the functional requirement of mitochondrial ATP production, we used oligomycin (Oli) to inhibit mitochondrial ATP synthase. WB analysis confirmed efficient knockdown of IDH2 and CS in recipient neurons treated with IDH2/CS-knockdown EV^adMu (Figures 5M and S9A). Both genetic (IDH2/CS knockdown) and pharmacological (Oli treatment) inhibition of mitochondrial energy production significantly attenuated EV^adMu-mediated axonal growth promotion (Figures 5N and 5O). Quantitative assessment showed that Oli treatment reduced neurite branching complexity by 63.27% compared to the EV^adMu group (Figure 5P). Additionally, Oli treatment decreased total axonal length in DRG neurons by 49.87% (Figure S9B) and the average length of the five longest axons in DRG explants by 40.03% vs. the EV^adMu group (Figure 5Q). The neurite-to-explant area ratio in the EV^adMu+Oli (3.19 ± 0.14) group was significantly lower than that in the EV^adMu (6.21 ± 0.51) group (Figure S9C). Notably, Oli’s inhibitory effects closely mimicked those of IDH2/CS knockdown, demonstrating that EV^adMu promotes axonal growth primarily through IDH2/CS-dependent ATP production.

To validate the pathophysiological relevance of these findings, we conducted in vivo studies to confirm the functional significance of EV^adMu-mediated metabolic reprogramming in nerve regeneration (Figure 5R). Targeted metabolite analysis revealed that IDH2/CS-knockdown EV^adMu-treated animals significantly compromised mitochondrial redox homeostasis, evidenced by a 27% reduction in NADPH/NADP⁺ ratio (5.08 ± 0.38 vs. 7.00 ± 1.18, p < 0.01) and a 29% increase in ROS levels (1,000 ± 188 vs. 1,410 ± 133 RFU, p < 0.01) compared to EV^adMu controls (Figure 5S). Similarly, mitochondrial energy metabolism was markedly impaired, with significant decreases in both NADH/NAD⁺ (38% reduction; 0.05 ± 0.01 vs. 0.08 ± 0.01) and FADH₂/FAD ratios (39% reduction; 0.11 ± 0.01 vs. 0.18 ± 0.01) (Figure 5T). Furthermore, IDH2/CS knockdown was associated with significantly reduced ATP production (Figure 5T), which is consistent with the in vitro neuronal findings. Notably, IDH2/CS knockdown EV^adMu significantly compromised nerve regeneration compared to the EV^adMu treatment (Figures 5U and 5V). Collectively, these findings indicate that EV^adMu reprograms neuronal mitochondrial metabolism to drive nerve regeneration, with IDH2 and CS playing pivotal roles in EV^adMu-mediated metabolic reprogramming, which in turn promotes axonal growth (Figure 5W).

EV^adMu-loaded microneedle nerve guidance conduits promote nerve regeneration and remyelination

Next, the effect of enhancing the muscle-to-nerve signaling axis in PNI repair was systematically evaluated in vivo. To overcome limitations associated with systemic Mu-EV delivery (insufficient biodistribution to target tissues, transient therapeutic effects, and suboptimal efficacy), microneedle (MN)-incorporated nerve guidance conduits (MN-NGCs) were developed. This integrated delivery platform simultaneously increases conduit-regenerative tissue contact area and enables controlled, sustained Mu-EV release, thus ensuring prolonged pharmacodynamic activity and superior regenerative outcomes^40,41 (Figure 6A).

Figure 6.

Fabrication, characterization, and functional validation of MN-NGCs for promoting nerve regeneration and remyelination

(A) Schematic of MN patch fabrication.

(B) Optical microscopy images of MN patch. Scale bar, 500 μm.

(C) Scanning electron microscopy images of MN patch. Scale bar, 500 μm.

(D) Fluorescence images of PKH26-labeled Mu-EV-loaded MN patch. Scale bar, 200 μm.

(E) Daily protein release profiles of MN patches (n = 5).

(F) Schematic of MN-NGCs bridging 10-mm sciatic nerve defect and validation. Rat 10-mm sciatic nerve resection model for MN-NGC therapeutic efficacy evaluation.

(G) WB analysis of IDH2 and CS in MN-NGC-treated sciatic nerves.

(H) ROS levels in MN-NGC-treated sciatic nerves (n = 6).

(I) ATP levels in MN-NGC-treated sciatic nerves (n = 6).

(J) Immunofluorescence of MN-NGC cross-sections (12 weeks post-implantation; NF160, S100β staining). Scale bars: 500 and 50 μm.

(K and L) Quantification of NF160 and S100β expression (n = 6).

(M) Methylene blue staining and TEM micrographs of MN-NGC cross-sections (12 weeks post-implantation). (Upper) methylene blue staining; (middle) low-magnification TEM; (bottom) high-magnification TEM. Scale bars: 100, 10, and 1 μm.

(N–P) Quantification of myelinated axon density, myelin sheath thickness, and G-ratio (12 weeks post-implantation, n = 6).

Statistical analysis: one-way ANOVA with Tukey’s multiple comparisons test (H, I, K, L, N, and O). Data: mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; ns, not significant. (O) Large solid dots, biological replicates (individual animals); small translucent dots, technical replicates (same sample repeated measurements).

The MN patch features an ordered 11 × 11 array of conical needles (base diameter: 200 μm, height: 600 μm, inter-needle spacing: 300 μm), each loaded with therapeutic Mu-EVs (Figures 6B and 6C). Confocal microscopy confirmed uniform spatial distribution of PKH26-labeled Mu-EVs within the MN matrix (Figure 6D). Total protein release (a proxy for Mu-EVs release; Figure 6E) and IDH2/CS enzymatic activity measurements (Figures S10A and S10B) verified continuous Mu-EV release from MN patches for ∼40 days. MN-NGC degradation was assessed in PBS and trypsin solution; under simulated physiological conditions, the scaffolds showed controlled degradation with 36.33% cumulative mass loss by day 28 (Figure S10C), confirming biodegradability. These results indicate that MN patches enable controlled, long-term Mu-EV delivery suitable for in vivo applications.

To evaluate in vitro biocompatibility of the MN-NGCs, we cultured Schwann cells with the material and assessed cell viability. LIVE/DEAD staining at 24 and 72 h post-treatment showed robust green fluorescence and minimal red fluorescence across all groups (Figures S10D and S10E), indicating high cell survival. CCK-8 assays quantitatively confirmed no significant reduction in cell viability vs. controls (Figure S10F). Collectively, these in vitro results confirm that MN-NGCs have no detectable cytotoxicity and support Schwann cell proliferation, providing a safety foundation for subsequent in vivo studies.

Subsequently, we bridged 10-mm sciatic nerve defects in Sprague-Dawley rats using MN-NGCs (Figure 6F). At 12 weeks post-operation, the EV^adMu-MN group exhibited significantly higher IDH2, CS levels, and ATP production, along with reduced ROS accumulation in regenerated nerves, compared to other MN-NGC groups (Figures 6G–6I, S10G, and S10H). These findings indicate that EV^adMu-MN exerted a regulatory role in mitochondrial metabolism in vivo. Immunofluorescence analysis further revealed superior axonal regeneration (NF160) and Schwann cell myelination (S100β) in the EV^adMu-MN group, with significantly higher densities than the EV^Mu-MN, shIdh2/Cs-EV^adMu-MN and MN groups (all p < 0.05) (Figures 6J–6L). TEM morphological assessment showed that the EV^adMu-MN group had significantly more myelin sheaths, thicker myelin, and larger axon diameters than the other three MN-NGC groups (all p < 0.05) (Figures 6M–6O and S10I). Furthermore, G-ratio analysis confirmed superior remyelination in the EV^adMu-MN group compared to the other MN-NGC groups, with efficacy comparable to the Autograft group (all p < 0.05) (Figures 6P and S10J).

Collectively, these results suggest that EV^adMu-MN treatment effectively promotes nerve regeneration and remyelination, providing a structural basis for functional recovery. Significantly, its therapeutic efficacy is critically mediated by IDH2 and CS components.

EV^adMu-loaded MN-NGCs delay muscle atrophy and promote functional recovery

The ultimate goal of PNI treatment is functional recovery.^42,43 To evaluate the therapeutic efficacy of EV^adMu, functional and structural recovery of the gastrocnemius muscle post-operation was systematically analyzed (Figure 7A). At 12 weeks post-operation, gross morphological examination revealed that EV^adMu-MN treatment most effectively attenuated denervation-induced muscle atrophy among all MN-NGC groups, with muscle architecture preservation comparable to the Autograft (Figure 7B). Quantitative analysis demonstrated that the EV^adMu-MN group achieved a gastrocnemius wet weight ratio (injured/contralateral) of 56.17% ± 3.31%, significantly outperforming other treatment groups (EV^Mu-MN: 50.83% ± 2.04%; shIdh2/Cs-EV^adMu-MN: 45.00% ± 2.61%; MN: 38.00% ± 2.61%) and approaching Autograft efficacy (62.17% ± 4.36%) (Figure 7C). Additionally, Masson’s trichrome staining showed the EV^adMu-MN group exhibited a significantly lower collagen fiber proportion (11.82% ± 2.36%) than the EV^Mu-MN (16.75% ± 2.49%), shIdh2/Cs-EV^adMu-MN (20.25% ± 1.03%), and MN groups (23.62% ± 1.66%) (Figures 7D and 7E). Concurrently, PGP9.5 immunostaining-based epidermal reinnervation assessment showed markedly enhanced nerve fiber density in the EV^adMu-MN group vs. other MN-NGC groups, supporting sensory restoration capacity of EV^adMu-MN treatment (Figures 7F–7H).

Figure 7.

Evaluation of muscle atrophy and functional recovery after MN-NGC implantation

(A) Schematic of muscle atrophy and functional recovery evaluation.

(B) Gross morphology of gastrocnemius (12 weeks post-implantation).

(C) Gastrocnemius muscle weight ratio (n = 6).

(D) Masson staining of gastrocnemius transverse sections (12 weeks post-implantation). Scale bar, 100 μm.

(E) Collagen fiber area percentage (n = 6).

(F) Immunofluorescence of hind paw interdigital skin (12 weeks post-implantation, PGP9.5 staining). (Upper) Low magnification; (bottom) high magnification of boxed area (dashed lines, epidermis-dermis boundary). Scale bars: 50 and 20 μm.

(G and H) Intra-epidermal (IENF)/dermal (DNF) nerve fiber ratio and IENF count per mm interdigital skin (12 weeks post-implantation, n = 6).

(I) 3D stress diagrams and left hindlimb footprint (12 weeks post-implantation).

(J) SFI quantification (n = 6).

(K) Representative motor-evoked potential waveforms.

(L–N) CMAP peak amplitude, conduction velocity, and CMAP onset latency (n = 6).

(O) Paw withdrawal latency (Hargreaves thermal nociception test, n = 6, seconds).

(P) Mechanical withdrawal threshold (von Frey test, n = 6, grams).

Statistical analysis: one-way ANOVA with Tukey’s multiple comparisons test (C, E, G, and H); two-way ANOVA with Tukey’s multiple comparisons test (J and L–P). Data: mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. (J, L–N, O, and P) ∗ denotes statistical comparison with EV^adMu-MN group.

SFI assessment of motor function recovery confirmed superior efficacy of EV^adMu-MN treatment (Figure S11A). Specifically, 3D imaging showed significant improvement in toe extension in the EV^adMu-MN group at 12 weeks post-operation, outperforming other MN-NGC groups and being comparable to the Autograft group (Figure 7I). Quantitatively, the EV^adMu-MN group exhibited a significantly higher score (−52.00 ± 1.41) than the EV^Mu-MN (−56.00 ± 2.10), shIdh2/Cs-EV^adMu-MN (−60.67 ± 1.21), and MN groups (−64.83 ± 1.47) (Figure 7J). These results indicate that EV^adMu-MN treatment significantly promotes motor function recovery following PNI.

Electrophysiological nerve conduction analysis revealed superior recovery in the EV^adMu-MN group at 12 weeks post-operation (Figure 7K). Specifically, the EV^adMu-MN group achieved significantly higher CMAP amplitude (15.66 ± 0.85 mV), faster NCV (21.40 ± 1.13 m/s), and shorter CMAP latency (1.53 ± 0.02 ms) than the EV^Mu-MN (13.86 ± 0.28 mV, 19.55 ± 0.57 m/s, 1.70 ± 0.03 ms), shIdh2/Cs-EV^adMu-MN (12.92 ± 0.54 mV, 17.85 ± 0.66 m/s, 1.82 ± 0.03 ms), and MN groups (11.89 ± 0.80 mV, 15.14 ± 1.26 m/s, 1.94 ± 0.03 ms) (Figures 7L–7N). Notably, all parameters in the EV^adMu-MN group approached those in the Autograft group (amplitude: 17.06 ± 0.86 mV; NCV: 24.14 ± 1.23 m/s; latency: 1.37 ± 0.04 ms). Importantly, sensory recovery assessments showed the EV^adMu-MN group exhibited accelerated thermal nociception recovery (Figures 7O and S11B) and alleviated mechanical allodynia (Figures 7P and S11C) compared to other MN-NGC groups.

Long-term biosafety of EV^adMu-MN was evaluated in vivo. Histological examination of major organs at 3 months post-treatment revealed no significant pathological changes in the heart, liver, spleen, lung, or kidney (Figure S12A). Consistent with this, hematological and serum biochemical analyses—including complete blood count (Figures S12B–S12D) and levels of ALT, AST, ALB, UREA, and CHO (Figures S12E–S12I)—showed no significant abnormalities. These results collectively demonstrate that EV^adMu-MN treatment has no adverse effects on major organ functions or systemic physiology.

In sum, these integrated analyses suggest that EV^adMu-MN, with IDH2 and CS as principal therapeutic components, effectively mitigated denervation-induced muscle atrophy, restored neural electrical signal conduction, and accelerated sensorimotor functional recovery, highlighting its therapeutic potential for PNI repair.

Discussion

PNI frequently causes denervation atrophy of target muscles due to inadequate regenerative potential, which prevents timely reinnervation and ultimately leads to irreversible functional deficits. Functioning as the fundamental unit for movement, the neuromuscular system exhibits profound structural and functional interdependence between its neural and muscular components. Traditionally conceptualized as a passive effector organ receiving neural input for contraction, muscle is now recognized to actively regulate multisystem physiological processes and maintain metabolic homeostasis across tissues.^10,11,12,13 Therefore, advancing the understanding of bidirectional nerve-muscle crosstalk axis holds great potential for promoting nerve regeneration.

Here, we identified a novel retrograde signaling axis wherein acutely denervated muscle initiated a reverse regulatory mechanism promoting nerve regeneration. Our findings demonstrate that acute denervation triggered a profound transcriptional reprogramming of muscle, which reshaped its secretome to enhance axonal growth via an EV-mediated mechanism. Further, integration of multi-omics and functional studies shows that EV^adMu-delivered IDH2 and CS reprogram neuronal mitochondrial metabolism by boosting ATP and safeguarding mitochondrial homeostasis, thereby rectifying bioenergetic deficits to further drive axonal growth.

Muscle’s systemic endocrine signaling capacity exhibits functional plasticity, dynamically regulated by its physiological state.^44,45,46 We demonstrate here that parallel to exercise activation, denervation triggers adaptive reprogramming of muscle, enhancing its secretory activity and driving a transition to a pro-regenerative phenotype. In our study, RNA-seq indicated a potential link between early-stage secretory remodeling and the biological transition of muscle following denervation. Additionally, the number of EV-like intraluminal vesicles in MVBs is significantly increased, with these secretory activities closely linked to neuro-regenerative processes. We further demonstrated that acutely denervated muscle promoted axonal growth through an EV-mediated mechanism. These results suggest that denervated muscle may orchestrate neural repair through dynamically controlled EV-mediated mechanisms, thereby establishing its role from a passive atrophy target to an active architect of neuro-regeneration.

In general, the energy-demanding nature of successful tissue repair requires vital local metabolic adaptations, notably during axonal regeneration, which imposes immense energetic demands.^37,38 Recently, nerve injury-induced adipocyte-derived adipokine leptin regulates glial metabolic adaptation and thereby facilitates nerve regeneration.⁴⁷ Here, we demonstrate that muscle remodeled secretome and EV^adMu potently enhances axonal regeneration. We identified IDH2 and CS as key regulators in EV^adMu that operate through complementary mechanisms: IDH2 by maintaining mitochondrial antioxidant defense to counteract oxidative stress and CS by driving the TCA cycle to promote energy synthesis. This synergistic action rescues the bioenergetic deficit caused by PNI, thereby robustly enhancing nerve regeneration—a finding that aligns with and extends previous studies on these metabolic enzymes.^48,49,50,51

Sufficient retention time critically determines the therapeutic efficacy of EV-based therapeutics, particularly in the context of PNI characterized by inherently chronic regenerative processes.^52,53,54 To prolong the residence time and achieve continuous release of EV^adMu, the MNs were developed to encapsulate EV^adMu for PNI repair. In contrast to conventional delivery strategies (such as systemic or local injection), MN technology facilitates precise, localized, and sustained release of EVs through conformal tissue adhesion, demonstrating significant therapeutic potential across various tissues.^40,55,56,57 Our study indicated that MN patches enable up to 40 days of sustained EV^adMu release and exhibit a controlled degradation profile, fully covering the therapeutic window for PNI. This approach not only maintains the bioactivity of EV^adMu, preventing enzymatic degradation, but also effectively reduces drug dosage and the inconvenience of repeated injections. These results present a promising and potentially translational platform for the clinical treatment of PNI.

In summary, this study provides a reverse muscle-nerve axis regulation paradigm. Far from being a passive atrophic target, skeletal muscle triggered by denervation actively secretes EV^adMu to deliver metabolic enzymes IDH2 and CS to neurons, thereby addressing neuronal energy deficits and driving axonal regeneration. Our findings complement the established view of unidirectional nerve-to-muscle regulation, revealing that muscle actively coordinates nerve regeneration through metabolic reprogramming. Building on this, muscle-specific secretion profiles could be leveraged to offer a perspective and therapeutic potential for nerve repair.

Limitations of the study

The full relevance of our findings regarding EV^adMu-delivered IDH2 and CS in enhancing TCA cycle activity and mitochondrial respiration in neurons post-acute nerve injury requires further investigation. Although several authoritative studies have addressed the general impact of energy metabolism on nerve repair, none have focused on the specific disruption of energy metabolism regulation from skeletal muscle after nerve injury,^52,53,54 leaving unestablished the extent to which EV^adMu-delivered IDH2 and CS-mediated energy production contributes to nerve repair. Moreover, elaborating the precise nature of the skeletal muscle response to nerve injury (beyond IDH2 and CS delivery) demands additional research, which may uncover further pro-regenerative mechanisms. While our study provides comprehensive in vivo validation, its scope is currently limited to small animal models; subsequent research should validate these findings in large animal models to improve clinical translatability.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed and will be fulfilled by the lead contact, Bing Xia (xiabing8807@fmmu.edu.cn).

Materials availability

This study did not generate new reagents.

Data and code availability

• •Transcriptomic data of triceps surae (previously published) are accessible via GEO: GSE44259.
• •Transcriptomic data of gastrocnemius are accessible via GEO: GSE312514.
• •Transcriptomic data of DRG neurons are accessible via GEO: GSE312113.
• •Proteomic data of Mu-EVs are accessible via PXD071459.
• •Metabolomic data of DRG neurons are accessible via MTBLS13442.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We acknowledge the experimental assistance from the Institute of Orthopedic Surgery, Xijing Hospital, Fourth Military Medical University, Xi’an. This work was supported by the National Natural Science Foundation of China grants 82372404 (J.H.), 82122043 (J.H.), 82430077 (Z.L.), 82201537 (B. Xia), and 82472418 (B. Xia) and the National Key R&D Program of China grants 2024YFA1802502 (J.H.) and 2022YFB3808000 (J.H.).

Author contributions

Conceptualization, Q.L., B. Xia, and J.H.; methodology, Q.L., B. Xue, Y.Z., and Z.W.; investigation, Q.L., Z.Z., H.L., S.F., and M.S.; visualization, Q.L., S.Y., A.Q., H.S., and X.G.; software, Q.L., M.Q., and Q.W.; data curation, Q.L., S.N., and L.G.; supervision, B. Xia, J.H., and Z.L.; writing – original draft, Q.L., B. Xue, and Y.Z.; writing – review & editing, Q.L., S.L., B. Xia, T.M., and J.H.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rab 21 Monoclonal Antibody	Santa Cruz Biotechnology, Inc.	Cat#sc-81917; RRID: AB_2253236
Mouse monoclonal anti-beta III Tubulin	Abcam	Cat#ab78078; RRID: AB_2256751
Rabbit monoclonal anti-S100 beta	Abcam	Cat#ab52642; RRID: AB_882426
Stathmin-2/STMN2 Antibody	Novus biologicals	Cat#NBP1-49461; RRID: AB_10011569
Mouse Anti-Neurofilament (Medium, 160 kD) Monoclonal Antibody, Unconjugated, Clone NF-09	Abcam	Cat#ab7794; RRID: AB_306083
anti-PGP9.5 antibody	Abcam	Cat#ab108986; RRID: AB_10891773
Mouse monoclonal anti-Neurofilament 200	Sigma	Cat#N0142; RRID: AB_477257
anti-HSP90 rabbit antibody	Cell Signaling Technology (CST)	Cat#4874S; RRID: AB_2121214
anti-CD63 mouse antibody	Proteintech	Cat# 67605-1-Ig; RRID: AB_2882811
anti-Calnexin rabbit antibody	Proteintech	Cat# 10427-2-AP; RRID: AB_2069033
anti-IDH2 rabbit antibody	Cell Signaling Technology (CST)	Cat# 56439; RRID: AB_2799511
anti-Citrate rabbit antibody	Abcam	Cat# ab129095; RRID: AB_11143209
anti-IDH3A rabbit antibody	Proteintech	Cat# 15909-1-AP; RRID: AB_2123282
anti-IDH3G rabbit antibody	Proteintech	Cat# 25848-1-AP; RRID: AB_2880267
anti-beta actin rabbit antibody	Servicebio	Cat# GB11001; RRID: AB_2801259
Cy3 conjugated Goat Anti-Rabbit IgG (H + L)	Servicebio	Cat# GB21303; RRID: AB_2861435
FITC conjugated Goat Anti-Mouse IgG (H + L)	Servicebio	Cat# GB22301; RRID: AB_3096857
FITC conjugated Goat Anti-Rabbit IgG (H + L)	Servicebio	Cat# GB22303; RRID: AB_2904189
HRP-conjugated Goat Anti-Rabbit IgG H&L	Abcam	Cat# ab6721; RRID: AB_955447
HRP-conjugated Goat Anti-Mouse IgG H&L	Abcam	Cat# ab6789; RRID: AB_955439
Bacterial and virus strains
AAV9-MHCK7-mCD63-EGFP	HanBio Tech	N/A
AAV9-MHCK7-mRab27a-ZsGreen	HanBio Tech	N/A
AAV9-MHCK7-mIdh2-ZsGreen	HanBio Tech	N/A
AAV9-MHCK7-mCs-ZsGreen	HanBio Tech	N/A
Biological samples
Primary DRG neuron	This paper	N/A
Primary Schwann cells	This paper	N/A
muscle-derived extracellular vesicles	This paper	N/A
Tissues from mice	This paper	N/A
Chemicals, peptides, and recombinant proteins
iF488-Wheat Germ Agglutinin	Servicebio	Cat#G1730-100UL
FluoroGold	Uelandy	Cat#F404
DiR lodide (DilC18(7))	Maokang Bio Biotech	Cat#100068-60-8
Mito-Tracker Green	Beyotime	Cat#C1048
4% paraformaldehyde	Biosharp	Cat# BL539A
EDTA	Sigma-Aldrich	Cat# ED2P
Fetal bovine serum	Gibco	Cat# 26400044
Penicillin/streptomycin	Gibco	Cat# 2321125
DMEM/F12	Gibco	Cat# C11995500BT
TRizol™	Invitrogen	Cat# 15596026
Critical commercial assays
PKH26 Red Fluorescent Cell Linker Mini Kit	Sigma-Aldrich	Cat# AE20204M
Mitochondrial membrane potential assay kit with JC-1	Beyotime	Cat# C2006
CCK-8 Assay Kit	Dojindo Laboratories	Cat# CK04
LIVE/DEAD Cell Imaging Kit	Thermo Fisher Scientific	Cat# R37601
NADP+/NADPH detection kit	Beyotime	Cat# S0179
NAD+/NADH detection kit	Beyotime	Cat# S0175
BCA assay kit	Beyotime	Cat# P0010
Enhanced ATP Assay Kit	Beyotime	Cat# S0027
GelMA Hydrogel	Engineering for life	Cat#EFL-GM-60
Deposited data
RNA-seq data	This study	GEO: GSE312514, GEO: GSE312113
RNA-seq data	Gene Expression Omnibus	GEO: GSE44259
Metabolomics data	This study	MTBLS13442
Proteomics sequencing data	This study	PXD071459
Experimental models: Organisms/strains
Mouse: C57BL/6	Fourth Military Medical University	N/A
Rat: SD (Sprague-Dawley)	Fourth Military Medical University	N/A
Software and algorithms
ImageJ software	NIH	https://imagej.nih.gov/ij/
GraphPad Prism 9.0	GraphPad	https://www.graphpad-prism.cn/
R Studio	R Studio	https://rstudio.com
Adobe Illustrator	Adobe	N/A
BL-420 N bio-signal acquisition system	Chengdu Taimeng	N/A
FlowJo 10.8	TreeStar	N/A
Illumina NovaSeq 6000 sequencing	Illumina	https://www.illumina.com.cn/systems/sequencing-pl atforms/novaseq.html
Zen blue (version 3.5)	Zeiss	https://www.zeiss.com/corporate/en/home.html
Other
Transmission electron microscopy	HITACHI	HT7800
Laser-scanning confocal microscope	Zeiss	Zeiss LSM 900, Germany
Tissue-Tek O.C.T. Compound	Sakura Finetek	Cat#4583
FV31S-DT software	Olympus	N/A
CatWalk XT	Noldus	N/A
Seahorse XFe96 Extracellular Flux Analyzer	Agilentek	N/A
Nanoparticle Tracking Analysis System (ZetaView)	Particle Metrix GmbH, Germany	N/A
IVIS Spectrum imaging system	PerkinElmer	N/A

Experimental model and study participant details

Animals

All animal experiments were reviewed and approved by the Animal Experiment Ethics Committee of Air Force Medical University (Approval No.20250051), with all procedures performed in accordance with current guidelines for laboratory animal care. The study strictly adhered to international standards for laboratory animal welfare, implementing the 3Rs principles (Replacement, Reduction, and Refinement).

All experiments were conducted with simple randomization performed by an independent researcher prior to the experiments. During group assignment, animals were randomly allocated to each cohort, and the groupings were kept blinded to the therapists, sample collectors, and data analysts.

After the experimental interventions were administered, data collection was carried out by personnel unaware of both the treatment regimens and group assignments. Likewise, data analysis was conducted by analysts who remained uninformed of the specific treatment details.

All experimental mice were obtained from the Animal Center of the Air Force Medical University (Xi’an, China) and allowed to acclimatize for 72 h before the experiment. Animals were housed in a specific pathogen-free facility under controlled conditions, constant temperature (24°C–26°C), relative humidity (40–50%), and a regulated 12-h light/dark cycle (lights on 08:00-20:00). Standard rodent chow and sterilized water were available ad libitum for the entire study duration.

Healthy adult male wild-type C57BL/6J mice (6–8 weeks old, 18–23 g) were utilized to investigate the molecular alterations in the gastrocnemius, harvest Mu-EVs, and perform functional analyses on these Mu-EVs.

Healthy adult male wild-type Sprague–Dawley (SD) rats (6–8 weeks old, 220–240 g) were utilized to validate the feasibility of our innovative EV-based PNI repair strategy.

DRG neurons and explants culture

Postnatal day 1–3 Sprague–Dawley rat pups were euthanized for DRG isolation. Following head removal, bilateral paravertebral incisions were made using microscissors to expose spinal DRGs, and then individual ganglia were dissected with ophthalmic forceps. After carefully removing the epineurial sheath, the ganglia were minced into fragments for subsequent processing. Isolated ganglia fragments were enzymatically dissociated in a solution containing Collagenase type IV and 0.25% trypsin. The digestion was carried out at 37°C for 60 min to achieve complete single-cell dissociation, then terminated by adding 4 mL of DF-12 medium containing FBS. The cell suspension was filtered through 70 μm meshes and plated on poly-L-lysine/laminin-coated dishes in Neurobasal-A medium supplemented with B-27 and 50 ng/mL NGF. For explant cultures, intact ganglia were maintained in Matrigel-embedded cultures. The DRG neurons and explants were treated with different conditions. Neurite complexity was assessed through semi-automated Sholl profiling (NeuronJ plugin) with 10 μm radial increments, followed by two-way ANOVA with Tukey’s post-hoc testing.

Method details

Animals models and treatment

All experimental animals were anesthetized by intraperitoneal injection of a mixture containing ketamine (87.5 mg/kg) and xylazine (12.5 mg/kg) to ensure adequate analgesia. Following standard surgical site preparation and disinfection, the left sciatic nerve was exposed through gluteal muscle dissection.

For mouse sciatic nerve injury models, two distinct protocols were established based on research objectives: (1) Sciatic nerve transection: a 5.0 mm nerve segment was resected 3 mm distal to the sciatic notch; this model ensures sustained, complete denervation (eliminating spontaneous reinnervation interference) and was used to investigate temporal molecular changes in the gastrocnemius muscle and isolate Mu-EVs. (2) Sciatic nerve crush: the nerve was clamped for 10 s using 5-mm-wide sterile forceps positioned 3 mm distal to the sciatic notch; this model preserves the nerve sheath and was employed to evaluate the therapeutic efficacy of EV^adMu in promoting nerve regeneration.

For rat sciatic nerve repair experiments, a 10-mm nerve segment was excised to create a standardized defect—one that does not heal spontaneously, serving as a stringent platform to test MN-NGCs constructs. Rats were randomly allocated to five treatment groups (n = 7/group): Autograft (gold standard, nerve segment reversed), empty MN-NGCs (control), EV^Mu-MN NGCs, EV^adMu-MN NGCs, and shIdh2/Cs-EV^adMu-MN NGCs.

All animals received consistent postoperative care: layered wound closure (muscle fascia sutured first, then skin with absorbable sutures) and prophylactic administration of gentamicin sulfate (5 mg/kg/day via intraperitoneal injection for 3 days) to prevent infection. Postoperatively, all animals were kept under standard housing conditions (12h light/dark cycle, constant temperature of 24°C–26°C, relative humidity 40–50%) with free access to food and water. Functional recovery of both mice and rats was systematically evaluated at predefined intervals using a combination of behavioral tests, electrophysiological measurements, and histological assessments.

RNA-seq analysis

Total RNA was extracted from gastrocnemius muscle and primary DRG neurons using TRIzol reagent (Invitrogen). For muscle, connective tissue was microdissected to reduce contamination; for DRG neurons, cells were trypsinized and PBS-washed to remove medium residues.

Eukaryotic mRNA was enriched with Oligo(dT) beads, and eukaryotic rRNA was removed with the Ribo-Zero Magnetic Kit (Epicentre). Subsequently, fragmented mRNA was reverse-transcribed into cDNA with the NEBNext Ultra RNA Library Prep Kit (NEB), followed by library construction procedures such as end repair, adapter ligation, and PCR amplification. The final cDNA libraries were sequenced on an Illumina NovaSeq 6000 platform (GeneDenovo Biotechnology).

Raw reads were filtered (Trimmomatic v0.39) and mapped to reference genomes (Mus musculus GRCm39 for muscle; Homo sapiens GRCh38 for DRG) via HISAT2 (v2.2.1). Transcript levels were quantified as TPM (StringTie v2.2.1). Differentially expressed genes (DEGs) were identified (DESeq2 v1.34.0) with |log₂ fold change (FC)|>0.585 (FC > 1.5) and adjusted p < 0.05.

For Gene Set Enrichment Analysis (GSEA), GSEA v4.3.2 software (Broad Institute) with MSigDB C5 (GO Biological Processes, v7.5.1) was used. Genes were ranked by log₂FC from DEG analysis; significance was determined via 1000 permutations, with adjusted p < 0.05 and FDR q-value<0.25 as thresholds. Other downstream analyses were done via OmiCSmart (https://www.omiCSmart.com/RNAseq/home.html).

AAV9 transduction of gastrocnemius

Hanbio Biotechnology was engaged in the synthesis of plasmids and AAV vectors, adhering to a pre-established protocol. cDNA encoding murine CD63, Rab27a, Idh2, and Cs was generated via reverse transcription from mRNA, with each fused to the N-terminus of enhanced EGFP or ZsGreen reporters. These fusion constructs were inserted into AAV backbones under the transcriptional control of the synthetic muscle-specific MHCK7 promoter, yielding AAV9-MHCK7-mCD63-EGFP, AAV9-MHCK7-mRab27a-ZsGreen, AAV9-MHCK7-mIdh2-ZsGreen, and AAV9-MHCK7-mCs-ZsGreen vectors. Detailed information on these AAV vectors is available in Figures S3 and S7. For viral packaging, 293T cells were co-transfected with 10 μg of the transfer plasmid (from the AAV9-MHCK7 series), 10 μg of pAAV-RC, and 20 μg of pHelper. The culture medium was replaced 6 h after transfection, and AAV vectors were collected 72 h post-transfection. To transfect the gastrocnemius muscle, 6-week-old male C57BL/6J mice received injections of 100 μL AAV9 vectors (titer: 1.3–1.8×10¹² vg/ml) or control virus into the left gastrocnemius. The total volume was administered as 10 separate injections (10 μL per injection) at distinct, evenly spaced sites across the muscle to ensure uniform distribution. Subsequent experiments were performed four weeks post-injection.

Coculture experiments

To evaluate muscle-derived effects on neuronal behavior, a transwell coculture system was established using 0.4 μm pore inserts. Primary DRG neurons or explants were plated in the lower chamber and left to adhere overnight. Four distinct types of treated groups were subsequently set up for the upper chamber: 1. PBS-only group, with an equal volume of sterile PBS added to the upper chamber (as a blank control); 2. Innervated gastrocnemius, dissected from naive non-denervated mice; 3. Acutely denervated gastrocnemius, dissected from mice with 3-day prior denervation; 4. AAV-shRab27a + acutely denervated gastrocnemius, from mice pre-injected with AAV9-MHCK7-mRab27a-ZsGreen 4 weeks earlier, then denervated 3 days before collection. The coculture was maintained for 48 h at 37°C under 5% CO₂. Following incubation, neurons were immunostained, imaged, and quantitatively analyzed.

Immunofluorescence

The harvested cell samples or tissue sections were fixed with 4% PFA for 20 min. Subsequently, sections were stained as follows: after blocking with 5% BSA for 1 h at room temperature, they were incubated overnight at 4°C with primary antibodies, including anti-beta ΙΙΙ Tubulin mouse antibody (1:500, Abcam), anti-S100 beta rabbit antibody (1:50, Abcam), anti-STMN2 rabbit antibody (1:500, Novusbio), anti-160 kD Neurofilament mouse antibody (1:500, Abcam), anti-PGP9.5 rabbit antibody (1:500, Abcam). After three washes with PBS, secondary antibodies were applied for 1 h at room temperature in the dark: Cy3 conjugated Goat Anti-Rabbit IgG (H + L) (1:100, Servicebio), FITC conjugated Goat Anti-Mouse IgG (H + L) (1:100, Servicebio), and FITC conjugated Goat Anti-Rabbit IgG (H + L) (1:100, Servicebio). Then, all samples were sealed with anti-fluorescence quenching sealer containing DAPI. After staining, the sections were observed under confocal laser scanning microscopy (Olympus, Japan). ImageJ was used for quantitative analysis of the acquired images.

Mu-EVs isolation and characterization

Mu-EVs were isolated using differential centrifugation procedures.⁵⁸ Gastrocnemius muscles were dissected from 6-week-old male C57BL/6J mice, minced into 0.2 to 0.4 cm³ fragments, and washed with cold PBS. The tissue was digested with collagenase D and DNase I at 37°C, and the supernatant was then filtered through a 70-μm nylon mesh to remove large debris. To isolate Mu-EVs, the filtrate was sequentially centrifuged: first at 750×g for 20 min (4°C) to eliminate cellular debris, then at 2,000×g for 30 min (4°C) to remove residual fragments. The supernatant was ultracentrifuged at 100,000×g for 70 min (4°C) to pellet Mu-EVs. The pellet was washed by resuspension in cold PBS and centrifuged again under the same conditions. Purified Mu-EVs were resuspended in PBS, with storage at −80°C.

EGFP-positive Mu-EVs were quantified using a calibrated FlowJo analysis system (v10.8.1, BD Biosciences). Mu-EV samples were diluted in cold PBS to an optimal concentration, and signals were acquired using 488 nm excitation. Fluorescence thresholds were set using PBS-only controls, with positive events defined as those exceeding the 99th percentile of control sample intensity.

To verify the quality of EV isolation, the size distribution and particle concentration of Mu-EVs were assessed via nanoparticle tracking analysis (ZetaView, Particle Metrix, Germany), and total protein content was quantified using a BCA assay kit (Beyotime Biotechnology, Shanghai, China). Western blot analysis confirmed the presence of EV markers CD63 and HSP90, with Calnexin as a negative control to exclude cellular contamination. Ultrastructural features of Mu-EVs were observed via TEM (HITACHI, Japan) following negative staining with uranyl acetate.

In vivo imaging and ex vivo fluorescence quantification

To evaluate the internalization of Mu-EVs by neurons, EV^Mu and EV^adMu were labeled with the near-infrared fluorescent dye DiR (Shanghai Maokang Biotech, 100068-60-8). Mice with sciatic nerve injury received a treatment of DiR-labeled vesicles (1×10⁸ particles per mouse). After 12 h, whole-body fluorescence imaging was performed using an IVIS Spectrum imaging system (PerkinElmer). Fluorescence signals were analyzed using Living Image software. Three days after treatment, the injured sciatic nerve segment and its corresponding lumbar DRG (L4-L6) were harvested. These tissues were subjected to ex vivo fluorescence quantification to assess whether Mu-EVs were internalized by neurons.

Proteomics analysis

Mu-EV proteins were extracted with urea/thiourea lysis buffer (6 M/2 M) containing protease inhibitors. Protein solubilization was enhanced by 90 cycles of high-pressure treatment (45,000 psi, 30°C). Sequential digestion employed trypsin/rLys-C (5 μg/1.25 μg) at pH 8.0 with 120 pressure-assisted cycles (20,000 psi, 30°C). Peptides were acidified (TFA to pH 2–3) and desalted via C18 columns (methanol-activated, 80% ACN/0.1% TFA-equilibrated, washed with 2% ACN/0.1% TFA, 40% ACN/0.1% TFA-eluted). Purified peptides were vacuum-concentrated and reconstituted to 0.1 μg/μL for LC-MS/MS analysis. For proteomic analysis of Mu-EVs, we employed an Orbitrap ExplorisTM 480 mass spectrometer (Thermo Scientific, San Jose, USA) linked to a Vanquish Neo UHPLC system. The setup included a FAIMS Pro (Thermo Scientific). In data-dependent acquisition mode, we used Buffer A (2% ACN, 98% H₂O with 0.1% FA) and Buffer B (80% ACN in water with 0.1% FA), both of MS grade. Peptides were first loaded onto a precolumn at 6 μL/min for 4 min, then separated on an analytical column using a 60-min LC gradient (8%–35% Buffer B) at 300 nL/min. The MS1 resolution was set at 60,000 with a mass range of 375–1800 m/z. Statistical and bioinformatics analyses were conducted using R (version 2024.12.1 + 563). Missing values were imputed by the median method. PCA, using the “stats” package, visualized group separation. Volcano plots were utilized to identify significant features using the “EnhancedVolcano” package. Heatmaps were drawn with “pheatmap” to show expression patterns. To explore biological functions, GO and KEGG enrichment analyses were conducted using “clusterProfiler”. Network analysis, using “igraph”, explored entity relationships.

Western blot

WB analysis was carried out following standard protocols.⁵⁹ In a nutshell, 15 μg of protein per lane was separated by 10% SDS-PAGE and transferred to PVDF membranes for 2 h. After blocking with 5% non-fat milk (1 h, RT), membranes were incubated overnight at 4°C with primary antibodies: anti-HSP90 rabbit antibody (1:1000, CST), anti-CD63 mouse antibody (1:5000, proteintech), anti-Calnexin rabbit antibody (1:5000, proteintech), anti-IDH2 rabbit antibody (1:1000, CST), anti-Citrate rabbit antibody (1:1000, Abcam), anti-IDH3A rabbit antibody (1:5000, proteintech), anti-IDH3G rabbit antibody (1:500, proteintech), and anti-beta actin rabbit antibody (1:1000, Servicebio). After washing, membranes were incubated with a secondary antibody, HRP-conjugated Goat Anti-Rabbit IgG H&L (1:3000, Abcam) and HRP-conjugated Goat Anti-Mouse IgG H&L (1:3000, Abcam) for 1 h at room temperature. Protein bands were visualized using ECL substrate (BIO-RAD, CA, USA) on GE AI600 equipment (General Electric Company, Boston, MA, USA).

Cellular metabolic analysis

The metabolic function of neurons was assayed using a Seahorse XFe96 Extracellular Flux Analyzer as previously described. Briefly, 1×10⁵ neurons were seeded in minimal, pH 7.4 unbuffered medium in XFe96 plates and adhered with Cell-Tak to ensure stable attachment during the assay. On the assay day, the medium in the cell culture plate was replaced with detection medium. Subsequently, the plate was incubated within a CO₂-free incubator for 60 min. The cartridge, which had been immersed in pre-heated XF calibrant (Agilent, 100840), was then loaded with drugs. The treatment involved the application of 1.5 μM oligomycin, followed by 1 μM carbonyl cyanide p-trifluoromethoxy phenylhydrazone (FCCP), and finally 1 μM rotenone/antimycin A. Cell counts per well were obtained for data normalization.

Mitochondrial morphology analysis

Neurons subjected to different treatments were labeled with Mito-Tracker Green (Beyotime, C1048, 100 nM) according to the manufacturer’s protocol. The incubation was performed at 37°C for 15 min, followed by three washes with pre-warmed (37°C) PBS. And then, images were acquired using a laser scanning confocal microscope (Olympus FV3000) with a 60×oil-immersion objective.

TEM

For animal-derived samples, sciatic nerve tissues were harvested and trimmed into 1–2 mm³ cubes, which were then fixed in 2.5% glutaraldehyde prepared with 0.1 M sodium cacodylate buffer (pH 7.2) at 4°C for 4 h. For cultured DRG neurons, cells were first rinsed with PBS, fixed in the same 2.5% glutaraldehyde solution at 4°C for 1.5 h, scraped, and centrifuged at 1000 rpm for 5 min to form pellets. After fixation, all samples were washed 3 times with 0.1 M sodium cacodylate buffer, post-fixed in 1% osmium tetroxide at room temperature for 2 h, and dehydrated via a graded ethanol series (50%, 70%, 90%, 100%) with 10–15 min per step. Next, the dehydrated samples were infiltrated with epoxy resin (first a 1:1 mixture of propylene oxide and epoxy resin, then pure epoxy resin overnight), embedded in silicone molds, and polymerized at 65°C for 48 h. The polymerized resin blocks were trimmed, cut into ultra-thin sections (70–90 nm thick) using an ultramicrotome, and collected on copper grids. Finally, the sections were stained sequentially with uranyl acetate and lead citrate, observed using a TEM (HITACHI, Japan), and images were captured with the microscope’s built-in digital camera.

JC-1 staining

JC-1 staining was performed using the Mitochondrial Membrane Potential Assay Kit with JC-1 (C2006, Beyotime). DRG neurons were seeded on glass coverslips in culture plates and incubated at 37°C with 5% CO₂ for 24 h, then subjected to different treatment conditions; a positive control group was treated with 20 μM FCCP for 1 h prior to staining. After treatment, cells were washed twice with supplemented buffer, incubated with pre-warmed 2 μM JC-1 working solution at 37°C with 5% CO₂ for 25 min in the dark, then washed twice with JC-1 assay buffer containing 25 mM HEPES. Images were acquired immediately using a confocal laser scanning microscopy (Olympus, Japan) with consistent exposure settings. Quantitative analysis and visualization were performed using ImageJ to assess the red/green fluorescence intensity ratio, reflecting ΔΨ_m status.

RT-qPCR

Total RNA was isolated from primary neurons utilizing Trizol reagent. Subsequently, cDNA synthesis was performed using HiScript II Q RT SuperMix (Vazemy, R223-01, Nanjing, China). For quantitative PCR (qPCR) analysis, the synthesized cDNA was amplified with Hieff qPCR SYBR Green Master Mix (Yeasen, 11201ES08, Shanghai, China) on a StepOnePlus Real-Time PCR System. To quantify gene expression, the threshold cycle (Ct) values of target genes were normalized to the corresponding Ct values of β-Actin, which served as an internal reference. All calculated fold changes in gene expression were expressed relative to either the untreated group or the specified control group to ensure standardized comparisons. The primer sequences used for the target genes were as follows: CCCACCAGTACCAACCCTATTG (forward), GGTGTTCAGGAAGTGCTCATTCA (reverse) for Idh2, GCTTGGCAGAGACATGTATGAGA (forward), ATGCTGCTGTCTGAAGGTCTTAA (reverse) for Cs, CCCATCTATGAGGGTTACGC (forward), TTTAATGTCACGCACGATTTC (reverse) for β-Actin, TGGGTGTCCAAGGTCTCTC (forward), CTCCCACTGAATAGGTGCTTTG (reverse) for Idh3a, GGTGCTGCAAAGGCAATGC (forward), TATGCCGCCCACCATACTTAG (reverse) for Idh3g.

NADPH/NADP⁺, NADH/NAD⁺ and measurement

According to the manufacturer’s instructions, the intracellular NADPH/NADP⁺ ratio and NADH/NAD⁺ were determined using the NADP⁺/NADPH detection kit (Beyotime, S0179) and the NAD⁺/NADH detection kit (Beyotime, S0175, with WST-8), respectively. Briefly, treated neurons or sciatic nerves were collected, lysed in ice-cold buffer via homogenization or thorough shaking, and 20 μL of each lysate was incubated with 20 μL of the corresponding reaction mixture at 60°C for 30 min. Absorbance at 450 nm was detected with a microplate reader (BioTek) to quantify NADPH/NADP⁺ and NADH/NAD⁺.

Preparation and characterization of MN-NGCs

The preparation of MN patches was strictly carried out according to previous research experience.^60,61 Briefly, Mu-EVs were homogenized with GelMA prepolymer in deionized water, and the mixture was cast into a polydimethylsiloxane (PDMS) mold. After desiccation at 4°C for >30 h, the 10 × 10 square microneedle array (needle dimensions: 600 μm height, 200 μm base diameter; 300 μm inter-needle spacing) was photocrosslinked under 405-nm UV light to solidify the structure. The cured MN patch was then gently peeled off the PDMS mold and stored dry at 4°C until use. For MN-NGCs fabrication, patches were rolled into cylindrical configurations.

The bright-field views and morphology of MN patches were observed using a stereomicroscope (Zeiss, Germany). The surface morphology of MN patches was meanwhile analyzed via a Scanning Electron Microscope (SEM, Zeiss, Germany). After mixing PKH26-labeled EVs with GelMA to prepare MN patches, fluorescent microneedle images were captured via a Laser Scanning Confocal Microscope (Olympus, Japan), revealing the distribution of EVs within the MN patches.

To assess the EV release capacity of MN patches, the quantity of protein released from them and the activity of IDH2 and CS were analyzed. Briefly, the MN patches were placed in culture dishes with serum-free medium. The supernatant was collected regularly and analyzed using the BCA Protein Assay kit, IDH2 activity detection kits, and CS activity detection kits. All release experiments were replicated three times to guarantee consistent and reliable results.

CCK-8 assays

Cell viability and cytotoxicity were assessed using the CCK-8 assay (Dojindo, Kumamoto, Japan) following the manufacturer’s protocol. Schwann cells in 96-well plates were co-cultured with different MN-NGCs and incubated for 24 or 72 h. After incubation, cells were rinsed three times with sterile PBS (HyClone). Each well then received 100 μL of fresh medium containing 10 μL CCK-8 reagent, followed by incubation for 4 h at 37°C in a 5% CO₂ humidified atmosphere. Absorbance at 450 nm was measured using a microplate reader (BioTek).

LIVE/DEAD assays

LIVE/DEAD staining was performed in 24-well plates using a LIVE/DEAD Cell Imaging Kit (Thermo Fisher Scientific, R37601) according to the manufacturer’s instructions, with live cells stained green and dead cells red. Briefly, Schwann cells were co-cultured with various MN-NGCs and incubated for 24 or 72 h. After incubation, cells were washed three times with PBS (HyClone) before adding the staining solution. Following 15 min of incubation at 37°C, cell viability was determined by counting live (green) and dead (red) cells using ImageJ software.

Evaluation of sensory and motor functional recovery

We conducted a comprehensive assessment of functional recovery at predetermined time points (4-, 8-, and 12-week post-operation) through the following experimental paradigms: Gait Analysis, Electrophysiological Assessment, Von Frey filament testing, and Hargreaves test.

Gait analysis was conducted via the CatWalk XT system (Noldus Information Technology) to quantitatively assess locomotor recovery at predetermined post-treatment time points. Before analysis, all animals underwent a one-week acclimatization period to familiarize them with the testing apparatus. During testing sessions, rats voluntarily traversed an illuminated glass walkway toward a darkened goal chamber. At the same time, the system captured high-resolution digital footprints, with a minimum of three complete, uninterrupted runs (meeting strict quality criteria) analyzed per animal. The acquired gait parameters were processed to calculate the SFI using the Bain-Mackinnon-Hunter formula.

Electrophysiological evaluation was conducted at predetermined post-treatment intervals using the BL-420 N bio-signal acquisition system (Chengdu Taimeng Software) to assess sensorimotor signal conduction quantitatively. Under standardized anesthesia, the regenerating sciatic nerve was surgically exposed. A bipolar stimulating electrode was positioned proximal to the nerve regeneration site, while a recording electrode was placed distally along the nerve trajectory to record CMAP.

For mechanical sensitivity evaluation, rats were acclimated for 30 min in individual acrylic chambers with wire-mesh floors before applying von Frey filaments (0.008–50g range) to the plantar hindpaw; positive responses (paw withdrawal, licking, or jumping) were documented, with the 50% withdrawal threshold determined using the up-down method across 3–5 trials per animal at 5-min intervals.

Thermal sensitivity was measured using a Hargreaves apparatus (45% intensity, 30s cutoff) by recording paw withdrawal latency to radiant heat stimulation, with six trials conducted per rat at 10-min intervals.

All behavioral tests were carried out during the light phase from 8:00 to 16:00 under the above strictly controlled experimental environment. These tests were conducted by investigators who were blinded to the experimental groups. All procedures adhered to standardized protocols to ensure the reliability and reproducibility of the results.

Biosafety evaluation

We evaluated the biological safety of MN-NGCs therapy by assessing major organ histology and key physiological indicators at 3 months post-injury using the following experimental approaches: major organ histological staining, complete blood count analysis, and serum biochemical analysis.

For major organ histological analysis, key organs (heart, liver, spleen, lungs, and kidneys) were rapidly harvested from each treatment group, then subjected to standard histological processing: first fixed in 4% paraformaldehyde solution, followed by dehydration, paraffin embedding, and sectioning into thin slices. The sections were stained with hematoxylin and eosin (H&E) to visualize tissue structure, then observed and imaged using a light microscope.

For complete blood count analysis, blood samples were collected from each animal via retro-orbital plexus bleeding immediately after euthanasia. Fresh blood was transferred to EDTA-anticoagulant tubes and gently inverted to prevent coagulation. Complete blood count parameters were measured using an automated hematology analyzer to quantify three key indicators: WBC, RBC, and PLT.

For serum biochemical analysis, after collecting blood for the complete blood count, the remaining blood samples were transferred to non-anticoagulant tubes and centrifuged at 3000 rpm for 15 min at 4°C to separate serum. The upper serum layer was carefully aspirated, aliquoted, and stored at −80°C until analysis. Serum biochemical indicators, including ALT, AST, ALB, UREA, and CHO, were detected using an automatic biochemical analyzer to assess organ function and metabolic status.

Quantification and statistical analysis

All data were presented as means ± standard deviation, with statistical analyses carried out using GraphPad 9.0 software (San Diego, CA). For comparisons between independent groups, an unpaired Student’s t test was employed. When dealing with comparisons among multiple groups, one-way analysis of variance (ANOVA) was performed. Specifically, after one-way ANOVA, pairwise comparisons were carried out using either the least significant difference method for general pairwise comparisons or Tukey’s test. All detailed statistical information, including the specific statistical tests used for each analysis, the definition of “n” (biological/technical replicates), and exact p-values, can be found in the corresponding figure legends. p < 0.05 was considered statistically significant.

Published: February 3, 2026