Chaenomeles speciosa-derived extracellular vesicles accelerate fracture healing via orchestrating immuno-angiogenic coupling Authors: Jingxiang Chen, Kaihan Zhuang, Yongpeng Wang, Shengjin Yuan, Zikun Xie, Qinghua Fan, Tao Liu Published: 2026 ABSTRACT AbstractPurposeFracture nonunion remains a major orthopaedic challenge, and dysregulation of the bone immune microenvironment is increasingly recognized as a key pathological driver. Chaenomeles speciosa is a medicinal plant with documented anti-inflammatory properties, making it a promising candidate for immune-modulatory therapy. This study investigated whether extracellular vesicles derived from Chaenomeles speciosa could reshape the immune microenvironment to promote fracture healing.MethodsSingle-cell RNA sequencing was performed on human nonunion tissues to characterize cellular heterogeneity. Network pharmacology and molecular docking analyses were used to identify potential therapeutic targets of Chaenomeles speciosa. Extracellular vesicles were isolated by ultracentrifugation and characterized by transmission electron microscopy. Macrophage polarization, angiogenesis, and inflammatory signaling were assessed in vitro using RAW 264.7 macrophages and human umbilical vein endothelial cells. Bone regeneration was evaluated in a rat femoral fracture model using micro-computed tomography and histology.ResultsSingle-cell RNA sequencing revealed M1 macrophage dominance and enhanced M1-mediated communication with osteoblasts and endothelial cells in nonunion tissues. The IL-17 signaling pathway and matrix metalloproteinase-3 were identified as key therapeutic targets. Isolated extracellular vesicles exhibited excellent biocompatibility, promoted M2 macrophage polarization, attenuated inflammation, and enhanced angiogenesis in vitro. In the rat fracture model, treatment with these extracellular vesicles significantly accelerated bone regeneration, as evidenced by increased bone bridging and callus formation.ConclusionThese findings establish a macrophage-centric pathological network in nonunion and demonstrate that Chaenomeles speciosa-derived extracellular vesicles promote bone repair through immuno-angiogenic coupling, supporting their potential as a multi-targeted therapeutic strategy for fracture healing.Keywords: Fracture nonunion, Single-cell RNA sequencing, Network pharmacology, Chaenomeles speciosa, Extracellular vesicles Introduction Nonunion is a severe obstacle during fracture healing, characterized by incomplete or atrophic callus ossification (Avin et al., 2023). This disabling condition not only impairs patients' physiological and psychosocial functioning but may also increase mortality risk (Tian et al., 2020). At present, the treatment methods for nonunion mainly include surgical fixation and bone grafting. Surgical treatment aims to provide mechanical stability, but current techniques have high complication rates (up to 26%), reintervention rates (23%), and failure rates (8%) (Feltri et al., 2022). In recent years, immunotherapy has emerged as a novel option for treating nonunion, and several studies have reported encouraging therapeutic effects (Baht et al., 2018; Ono and Takayanagi, 2017). Meanwhile, traditional Chinese medicine (TCM) has been proven to possess the potential to modulate immune responses and influence disease progression (Zheng et al., 2025). However, so far no TCM-based immunotherapy has been applied to nonunion as its specific immunological mechanisms remain unclear. In Nonunion, the bone immune microenvironment (BIM) constitutes a complex system, including a variety of immune cells, stromal cells, extracellular matrix components, and cytokines, interacting to orchestrate repair (Takayanagi, 2015). Within this microenvironment, macrophages are pivotal regulators, demonstrating remarkable plasticity by polarizing into a spectrum of phenotypes, from pro-inflammatory to pro-regenerative. Pro-inflammatory macrophages secrete high levels of TNF-α, IL-1β, and IL-6, exacerbating inflammation and suppressing osteoblast differentiation while promoting osteoclast-mediated bone resorption. Conversely, pro-regenerative macrophages produce anti-inflammatory cytokines (IL-10, TGF-β) and growth factors (VEGF, IGF-1, BMP-2), fostering an anti-inflammatory milieu that promotes angiogenesis, osteogenesis, and tissue remodeling (Hu et al., 2023). An imbalance toward a persistent pro-inflammatory state is hypothesized to fuel chronic inflammation and impede bone healing (Ono and Takayanagi, 2017), yet the precise heterogeneity and functional dynamics of macrophage subsets in human nonunion remain inadequately delineated. The latest advancements in single-cell RNA sequencing (scRNA-seq) enable the comprehensive characterization of cell subsets existing in the BIM (Chen et al., 2019). This technique provides valuable insights into cellular heterogeneity and dynamics, and illustrates the specific role of BIM in the progression and treatment response of Nonunion (Song et al., 2022). At the same time, network pharmacology integrates biology pharmacology, and computational methods to comprehensively study the effect of therapeutic agents (Luo et al., 2020). Herbs used in traditional Chinese medicine, such as licorice, peony, and Chaenomeles speciosa, significant immunomodulatory properties (Ding et al., 2022; Peng et al., 2022; Zhang and Wei, 2020). Among them, Chaenomeles speciosa has garnered attention due to its good anti-inflammatory and cartilage-protective effects (Duan et al., 2023). Meanwhile, recent evidence highlights EVs as key mediators of intercellular communication, serving as natural nanocarriers that efficiently deliver bioactive molecules to recipient cells with high specificity and low immunogenicity (Jin et al., 2024). However, the specific mechanisms and immune targets of Chaenomeles speciosa in bone healing therapy remain unclear. Herein, we first employed scRNA-seq to construct a high-resolution transcriptomic atlas of nonunion, with a specific focus on deciphering macrophage heterogeneity, polarization trajectories, and their dysregulated communication with other cellular constituents of the BIM. Guided by these pathological insights, we then utilized network pharmacology to identify the potential active compounds and therapeutic targets of Chaenomeles speciosa. Finally, we isolated and characterized CsEVs, experimentally investigating their potential to modulate macrophage polarization, promote angiogenesis, and facilitate bone regeneration both in vitro and in vivo. Our work not only elucidates the central role of macrophage-driven immunopathology in nonunion but also establishes CsEVs as a potent, multi-targeted nanotherapeutic agent informed by and validated through a cutting-edge integrative biological framework. Materials and methods Data acquisition We analyzed the publicly available single-cell RNA sequencing (scRNA-seq) dataset GSE217792, downloaded from the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/). The original sequencing data were used directly without further preprocessing. Ethical approval was not required for this analysis as it utilized open-access data. Quality control and dimensionality reduction clustering Quality control and clustering were performed with Seurat (v4.0.6). Cells were filtered out if they expressed fewer than 250 genes, contained over 20% mitochondrial genes, or had a sequencing depth below 500 reads. The Harmony package (v1.0) was applied to correct for batch effects. We selected the top 2000 highly variable genes (HVGs) for principal component analysis (PCA). Optimal clustering parameters were determined via elbow plots and dendrograms. Graph-based clustering was executed using the FindNeighbors and FindClusters functions, and results were visualized in two dimensions using Uniform Manifold Approximation and Projection (UMAP) (Butler et al., 2018). Cell annotation Cell clusters were annotated by cross-referencing their expression profiles with established canonical marker genes from the literature. The expression of these markers was assessed across all clusters to assign definitive cell type identities, with no clusters remaining unannotated. Differentially expressed gene analysis Differentially expressed genes (DEGs) in Nonunion M1 macrophages versus controls were identified at the cell level using the FindMarkers function in Seurat (v4.3.0) and the Wilcoxon rank-sum test. Genes were required to be expressed in at least 10% of cells in either group (min.pct = 0.1). Significance was defined by an adjusted p-value <0.05 (Bonferroni correction) and an absolute log2-fold change >0.5. Pseudotime trajectory analysis Macrophage differentiation trajectories were reconstructed using Monocle2 (v2.30.0). Genes for ordering the trajectory were selected by identifying macrophage subcluster marker genes (Seurat's FindAllMarkers, logfc.threshold = 0.25(Trapnell et al., 2014)) and filtering for those significantly associated with cell state (Monocle2's differentialGeneTest, fullModelFormulaStr = “∼stim”, qval <0.01). The top 2500 genes by q-value were used for trajectory construction with the DDRTree algorithm (max_components = 2). Genes with significant expression changes along pseudotime were identified (differentialGeneTest, fullModelFormulaStr = “∼sm.ns (Pseudotime)”, qval <0.1) and clustered into modules. Expression dynamics of nine key M1 and M2 markers were also visualized along the trajectory. Cell-to-cell communication analysis Intercellular communication networks were inferred for Control and Nonunion samples separately using CellChat (v1.6.0). CellChat objects were created from normalized expression data and cell type labels, utilizing the full CellChatDB.human ligand-receptor database (Efremova et al., 2020). Over-expressed ligands and receptors were identified by default methods. Communication probabilities and p-values for ligand-receptor pairs were computed (computeCommunProb, raw.use = TRUE) with permutation testing, applying a filter for interactions (min.cells = 10). Networks were aggregated at the pathway level (computeCommunProbPathway, aggregateNet) and compared across groups using mergeCellChat. Drug component screening and target identification Active compounds of Chaenomeles speciosa (Sweet) Nakai were retrieved from the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP, https://www.tcmsp-e.com) (Ru et al., 2014). Compounds were screened based on oral bioavailability (OB ≥ 30%) and drug-likeness (DL ≥ 0.18) to identify bioactive components with therapeutic potential (Xu et al., 2012). Subsequently, the SwissADME tool (https://SwissADME), a widely validated platform in the field of natural product drug research and development, was used to comprehensively evaluate the pharmacokinetic and drug-likeness characteristics of the screened compounds. For pharmacokinetic evaluation, only compounds classified as “High” in the “Human Oral Absorption” parameter (corresponding to in vivo oral absorption rate > 80% in preclinical studies) were retained. This step further ensured the favorable in vivo bioavailability of the compounds on the basis of the initial OB ≥ 30% threshold. In the drug-likeness evaluation, four classic evaluation rules were applied: Lipinski's Rule of Five (RO5), Ghose Rule, Veber Rule, and Egan Rule. A compound was determined to be “drug-like” only if it met at least 2 of these rules (each rule was marked as “Yes” when satisfied). Three complementary tools with strict parameter settings were used to predict the potential molecular targets of the finally screened bioactive compounds to ensure high reliability: when using the SwissTargetPrediction tool, the target species was set to “Homo sapiens (human)”; when using the PharmMapper tool, the “Human Protein Target Database (2021 version)” was used as the target library; when using the Comparative Toxicogenomics Database (CTD), predictions were made by searching for “compound-disease-target” associations. All predicted targets were cross-validated and standardized using the UniProt database (https://www.uniprot.org), and only “Reviewed (Swiss-Prot)” entries were retained (excluding “Unreviewed (TrEMBL)” entries). This step ensured the accuracy of target information, excluded non-human targets or hypothetical proteins, and finally obtained a reliable set of compound-related molecular targets. After the initial acquisition of targets, the canonical SMILES (Simplified Molecular-Input Line-Entry System) of each compound was cross-validated in the PubChem database (https://pubchem.ncbi.nlm.nih.gov/) to eliminate structural errors (such as incorrect atom connectivity or abnormal charge states). Intersection of drug targets and disease genes To identify potential therapeutic targets of Chaenomeles speciosa (Sweet) Nakai in nonunion, the human nonunion-related differentially expressed genes (DEGs) were directly intersected with the documented molecular targets of Chaenomeles speciosa. The overlap was computed and visualized with the Venny R package (v2.1.0), yielding the common genes that may mediate the pro-osteogenic effect of Chaenomeles speciosa on nonunion. Drug-compound-target-disease network construction An interactive network was constructed using Cytoscape (v3.10.1) to display the relationships between Chaenomeles speciosa (Sweet) Nakai, its bioactive compounds, molecular targets, and related diseases. In this network, nodes represent entities such as bioactive compounds and genes, and edges represent the interactions between these entities. This network provides a comprehensive map of this complex system and can highlight key therapeutic targets and their potential roles in the treatment of nonunion. Gene ontology and pathway enrichment analysis Functional annotation of the intersecting genes was performed using the clusterProfiler R package (v3.14.3). Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways with significant enrichment (adjusted p-value <0.05) were screened out (Huang et al., 2009). Bubble charts were used to visualize the results of the enrichment analysis to highlight the most critical biological processes, molecular functions, and pathways related to the therapeutic effect of Chaenomeles speciosa (Sweet) Nakai on nonunion. Molecular docking and visualization analysis Molecular docking simulations were used to predict the interactions between the bioactive compounds of Chaenomeles speciosa (Sweet) Nakai and key target proteins. After obtaining compound structures from the TCMSP database, they were processed using ChemOffice software: first, the 2D structures were converted to 3D structures, then the structures were optimized via the Chem3D module using the MM2 molecular mechanics force field to eliminate torsional stress and steric hindrance, and subsequently, Gasteiger-Hückel charges were assigned to the compounds using AutoDockTools (v1.5.6); after adding polar hydrogens, conformational clustering was performed, and finally, the conformations with the lowest energy (in PDBQT format) were selected for docking. Protein structures were downloaded from the RCSB Protein Data Bank (RCSB PDB) and preprocessed using PyMOL software—removing solvents and organic small-molecule impurities—before conducting docking simulations using AutoDockTools (v1.5.6) and AutoDock Vina (v4.2). The PyMOL software (the same version as mentioned above) was used to visualize and analyze the docking results, evaluate the binding affinity between compounds and target proteins, and verify whether the ligands were located in the active pockets and formed specific interactions (Shamsol Azman et al., 2023). Isolation and purification of CsEVs Fresh Chaenomeles speciosa fruit was washed with deionized water, soaked in cold PBS, and homogenized. The juice was sequentially filtered through a 212 μm mesh and centrifuged at 1000 ×g for 20 min, 3000 ×g for 20 min, 10,000 ×g for 30 min, and 70,000 ×g for 1 h to remove large particles. The supernatant was then ultracentrifuged at 150,000 ×g for 90 min, and the pellet was resuspended in PBS to obtain a crude EVs sample. Further purification was achieved by sucrose density gradient centrifugation (8%, 30%, 45%, and 60% w/v) at 150,000 ×g for 90 min. The fraction between the 30% and 45% layers was collected as purified Chaenomeles speciosa EVs. The morphology of the isolated EVs was examined by transmission electron microscopy (TEM). In vitro studies Cell cytoskeletal staining RAW 264.7 macrophages were seeded on coverslips at 1 × 105 cells per well in 12-well plates and cultured for 3 days. Cells were exposed to different concentrations of EVs (0, 105, and 106 particles/mL) to assess biocompatibility. Cells were fixed with 4% paraformaldehyde, stained with Phalloidin (green) for actin filaments and Hoechst (Sigma) for nuclei, and imaged under a laser confocal microscope (ZEISS LSM 980). Cell spreading area was quantified using ImageJ software. Live/dead cell viability assay RAW 264.7 macrophages were seeded at 5 × 105 cells per sample. After 24 h, cells were exposed to different EV concentrations (0, 105, and 106 particles/mL) and then incubated with a solution containing 2 μM Calcein-AM (live cells, green) and 4 μM propidium iodide (PI, dead cells, red) (both from Invitrogen) in PBS for 30 min at 37 °C in the dark. Viability was assessed using laser scanning confocal microscopy (ZEISS LSM 980), and cell viability was quantified using ImageJ software. EV uptake and cellular internalization assay CsEVs were labeled with PKH26 (Sigma-Aldrich) according to the manufacturer's protocol. Briefly, EVs were resuspended in Diluent C and mixed with PKH26 dye (final concentration 2 μM) for 5 min at room temperature. The labeling reaction was stopped by adding an equal volume of 1% BSA, followed by ultracentrifugation at 100,000 ×g for 1 h to remove unbound dye. The pellet was washed with PBS and resuspended in culture medium. RAW 264.7 macrophages were incubated with PKH26-labeled EVs, and cellular uptake of EVs was assessed by confocal microscopy with red fluorescence labeling. The internalization efficiency was quantified by measuring the fluorescence intensity using ImageJ software. Gene expression profiling Total RNA was extracted with an RNA kit (Omega), and cDNA was synthesized using a reverse transcription kit (Takara). Quantitative RT-PCR was performed in triplicate with SYBR Green Master Mix (Takara) on a LightCycler 480 system to assess the expression of MMP3, iNOS, and Arg. Gene expression levels were calculated by the 2^–ΔΔCt method, using primers listed in Table 1. Table 1Displays the primers used in the present study.TargetForwardReverseGAPDHCCTCGTCCCGTAGACAAAATGTGAGGTCAATGAAGGGGTCGTMMP3TTCCTGATGTTGGTGGCTGTAGGGATGTAAGGGATGTAAGGGiNOSCAGCTGGGCTGTACAAACCTTCTCTCCTCGCTTCAATGCTCAArg-1GTTCCCAGATGTGCCCTAATGAGCTGGTTGTCAGTGGCTTTA Displays the primers used in the present study. Western blot (WB) assay Cells or tissues were lysed in RIPA buffer (CWBIO) containing protease and phosphatase inhibitors (Thermo Fisher). Protein concentration was determined by a BCA assay kit (Beyotime). Proteins (40 μg per lane) were separated by SDS-PAGE, transferred to PVDF membranes (Thermo Fisher), and blocked with a high-performance blocking reagent (Genefist). Membranes were incubated with primary antibodies against iNOS, Arg, MMP3, and GAPDH (Table 2) overnight at 4 °C, followed by incubation with secondary antibodies. Protein bands were visualized using an ECL substrate (Thermo Fisher) and imaged with a GelView 6000 Pro system (BLT). Band intensities were quantified using ImageJ software. Table 2Related information about primary antibodies.AntibodiesSpeciesTypeDilution (IF/WB/IHC)SourceAnti-GAPDHRabbitMonoclonal lgG1:2000ProteinTech, ChinaAnti-iNOSRabbitMonoclonal lgG1:1000/1:200/1:150ProteinTech, ChinaAnti-Arg-1RabbitPolyclonal lgG1:1000/1:200/1:100ProteinTech, ChinaAnti-MMP3RabbitMonoclonal lgG1:1000/1:200/1:150ProteinTech, ChinaAnti-OPNRabbitPolyclonal lgG1:1000/1:200/1:100ProteinTech, China Related information about primary antibodies. Immunofluorescence staining assay For EV uptake analysis, RAW 264.7 macrophages were seeded on coverslips and incubated with PKH26-labeled CsEVs (0, 105, and 106 particles/mL) for 4 h. Cells were then fixed with 4% paraformaldehyde for 30 min, permeabilized with 0.2% Triton X-100 for 10 min, and blocked with 3% BSA for 30 min. Actin filaments were stained with Phalloidin-FITC (green) and nuclei with Hoechst 33342 (blue). Images were acquired using a laser scanning confocal microscope (ZEISS LSM 980). Tube formation assay The pro-angiogenic ability of EVs was evaluated using an in vitro tube formation assay with HUVECs. Briefly, 50 μL of Matrigel (BD Biosciences) was polymerized in 96-well plates at 37 °C for 30 min. RAW 264.7 macrophages (2 × 104 cells/mL) were seeded in 24-well plates and exposed to Control, LPS, and LPS + EVs conditions for 24 h. The conditioned medium was then collected and applied to HUVECs (2 × 104 cells/mL) seeded on Matrigel. After 3 h and 6 h of incubation, tubular structures were imaged with an inverted microscope (Leitz MM6). Total tubule length was quantified using ImageJ software, with four replicates per condition. Construction of rat femoral fracture model Sprague-Dawley rats (male, aged 8–10 weeks, weighing 280–320 g) were anesthetized with an intraperitoneal injection of ketamine (75 mg/kg) and xylazine (15 mg/kg). A medial incision was made on the right knee, the patella was dislocated laterally, and a 0.5 mm hole was drilled into the femoral condyle. Kirschner wire fixation was applied to stabilize the fracture. Then, after exposing the femur, a fracture model was created in the mid-femur using a rongeur. Rats received injections of either PBS (control) or Chaenomeles speciosa-derived EVs at the fracture site, which were sterilized by UV irradiation or filtration prior to use. Animals were monitored postoperatively until fully recovered. Bone healing was assessed at 4 and 8 weeks post-injury.Tramadol hydrochloride was administered in drinking water for perioperative analgesia. Micro-CT The right femora were harvested at 4 and 8 weeks post-surgery and scanned using a SkyScan CT scanner (SkyScan 1276, Bruker, Belgium) with the following settings: source voltage 50 kV, source current 200 μA, pixel size 10 μm, rotation step 0.5°, and frame averaging of 3. NRecon software (version 1.7.4.6) was used for 3D reconstruction, and CTAn (version 1.18.8.0) was used for quantitative analysis. Parameters including bone volume fraction (BV/TV), trabecular thickness (Tb·Th), and bone mineral density (BMD) were analyzed with the manufacturer's software. Histological examination Tissue samples were fixed in 10% neutral buffered formalin for 48 h, dehydrated through a graded series of ethanol, cleared in xylene, and embedded in paraffin. Serial sections were cut at 5 μm thickness using a rotary microtome. For morphological evaluation, sections were stained with hematoxylin and eosin (H&E) to assess general tissue architecture, cellular infiltration, and callus formation. For collagen deposition analysis, sections were stained with Masson's trichrome to visualize collagen fibers (blue) and distinguish them from muscle fibers and cytoplasm (red). Stained sections were observed under a light microscope (Leica DM6000) and images were captured for quantitative analysis. Immunohistochemical At 4 and 8 weeks post-treatment, femurs were harvested, fixed in 4% PFA, decalcified in 10% EDTA, and paraffin-embedded. Sections (5 μm) were subjected to antigen retrieval, blocked with 5% normal goat serum, and incubated overnight at 4 °C with primary antibodies against OPN (cat. no. 22952-1-AP, Proteintech, China), iNOS (cat. no. 18985-1-AP, Proteintech, China), and MMP3 (cat. no. 17873–1-AP, Proteintech, China), all at a dilution of 1:200 for OPN and 1:100 for iNOS and MMP3. After incubation with HRP-conjugated secondary antibodies, immunoreactivity was visualized using DAB and counterstained with hematoxylin. The distribution of these markers was observed under a light microscope (Leica DM6000) as previously described. Statistical evaluation Data are presented as mean ± SD from at least three independent experiments. For the in vivo fracture model, six rats were used per group (n = 6 per group) at each time point (4 and 8 weeks). Group comparisons were analyzed by one-way ANOVA with Bonferroni post-hoc test using GraphPad Prism 9.0 and Origin 2019b. Significance was defined as *p < 0.05, **p < 0.01, and ***p < 0.001; ns, not significant (p ≥ 0.05). Result and discussion Single cell transcriptome map of nonunion The overall experimental strategy and proposed mechanistic framework of this study are illustrated in Fig. 1. Single-cell RNA sequencing of the GSE217792 dataset (n = 9 samples from Control and Nonunion groups) yielded a transcriptome map of nonunion. Rigorous quality control, filtering on genes, UMIs, and mitochondrial content, resulted in 76,162 high-quality cells for analysis. We employed an integration strategy to minimize batch effects before performing PCA for dimensionality reduction, using the first 30 principal components for all further steps. Cell clustering at a resolution of 0.3 (as determined by clustree analysis) identified 23 distinct clusters (Fig. 2A). Annotation based on established lineage markers enabled the definition of ten principal cell types (Fig. 2B): Macrophages (CD14, CD68), B cells (MS4A1, CD79B), Endothelial cells (VWF, PECAM1), T cells (CD3D, CD3E), Osteoblasts (COL1A1, COL3A1), Dendritic cells (CD1C), Smooth muscle cells (MYH11), Megakaryocyte-erythroid progenitors (CD38), Neutrophils (MPO, ELANE), and Erythrocytes (GATA1, GYPA) (Fig. 2D, F). Analysis of cellular distribution demonstrated the presence of all ten lineages in both groups, but with markedly altered relative abundances in Nonunion compared to Control (Fig. 2C, E). A key finding was a significant increase in the proportion of macrophages within the Nonunion samples. Schematic illustration of mechanism exploration and therapeutic strategy for fracture nonunion. Single-cell profiling of control and nonunion tissues. (A) UMAP visualization showing 23 distinct cell clusters identified by unsupervised clustering. (B) Annotation of ten major cell types based on the expression of canonical marker genes. (C) UMAP plots comparing the cellular distribution between Control and Nonunion groups. (D) Dot plot displaying the expression and prevalence of key marker genes for each cell type. (E) Bar graph illustrating the proportional abundance of each cell type in Control and Nonunion samples. (F) Feature plots highlighting the spatial expression patterns of 10 definitive marker genes across the UMAP. Changes of macrophage subtypes during nonunion To delineate macrophage heterogeneity in nonunion, we first isolated macrophage clusters based on canonical M1 and M2 markers. Graph-based clustering resolved 12 distinct macrophage subpopulations (Fig. 3A), whose transcriptomic profiles across Control and Nonunion tissues are visualized in Fig. 3C. Subsequent annotation using specific marker genes categorized these populations into two primary subtypes (Fig. 3B). M1-like macrophages were defined by upregulated expression of pro-inflammatory genes (IL1B, CCL3, CXCL8, TIMP1), whereas M2-like macrophages exhibited elevated levels of genes associated with immunoregulation and repair (CSF1R, CD163, MRC1, ACP5, SELENOP) (Fig. 3D, F). Comparative analysis revealed a significant expansion of the M1-like macrophage fraction in Nonunion tissues relative to controls (Fig. 3E). Given that an imbalanced M1/M2 ratio is implicated in driving pathogenesis, our findings point to a shift toward a pro-inflammatory microenvironment in nonunion, which may impede bone healing. This M1/M2 imbalance creates a pathological microenvironment where M1-derived TNF-α and IL-1β directly suppress osteoblast differentiation and promote osteoclast-mediated bone resorption, while the deficiency of M2-secreted VEGF and BMP-2 deprives the fracture site of essential signals for angiogenesis and osteogenesis, ultimately arresting the healing process at the inflammatory stage (Zhou et al., 2023). Identification and characterization of macrophage subpopulations. (A) UMAP plot revealing 12 distinct macrophage clusters. (B) Annotation of M1-like and M2-like macrophage subsets based on established polarization markers. (C) UMAP visualization comparing the distribution of macrophage subtypes in Control and Nonunion tissues. (D) Dot plot visualizing the expression levels of signature genes for M1-like and M2-like macrophages. (E) Proportional abundance of M1-like and M2-like macrophages in Control and Nonunion groups. (F) Feature plots showing the expression distribution of key M1 and M2 marker genes on the UMAP. Pseudotime analysis and assessment of macrophages subtype dynamics in nonunion We applied pseudotime analysis to reconstruct the maturation trajectory of macrophages in the nonunion microenvironment. Using the Monocle2 algorithm, which employs DDRTree dimensionality reduction to preserve both local and global data structure while reconstructing developmental trajectories (Feng et al., 2022), we resolved 12 distinct cellular states across the differentiation continuum (Fig. 4C). The algorithm's ability to infer pseudotemporal ordering from static snapshot data enabled us to identify state 1 as the trajectory origin (Fig. 4B). The analysis outlined a clear progression from M2-like to M1-like phenotypes, indicating a dynamic shift toward a pro-inflammatory macrophage population in nonunion (Fig. 4A). Pseudotime trajectory analysis of macrophage polarization. (A) Pseudotime trajectory with cells colored by their respective macrophage subtype. (B) Developmental trajectory of macrophages inferred by the Monocle2 algorithm. State 1 was defined as the root. (C) Pseudotime analysis projecting the 5 identified macrophage states onto the differentiation path. (D) Heatmap of 2500 trajectory-dependent genes, clustered into four modules. Significantly enriched Gene Ontology (GO) terms for each module are annotated (q-value < 10−10). (E) Expression kinetics of nine key M1 and M2 marker genes along the pseudotime trajectory. Each point represents a single cell, colored by subcluster, with the y-axis indicating normalized expression levels. We identified 2500 trajectory-dependent genes, which were categorized into four distinct modules, as shown in the heatmap (Fig. 4D). Analysis of these genes delineated a coordinated transcriptional program, wherein the early-activated Module 1 initiated the pro-inflammatory microenvironment through antigen presentation and immune activation (Hartmann and Bendall, 2020). This was followed by the transient states of Modules 2 and 4, which were linked to a reactive oxygen species (ROS) burst and glycolytic metabolic switching, serving to fine-tune the inflammatory response (Hashimoto et al., 2018). Subsequently, the late-upregulated Module 3 was induced, potentially in response to glucose starvation, and acted to sustain chronic inflammation and osteogenic inhibition via enhanced T cell activation and intercellular adhesion (Mohanan et al., 2023). Collectively, this cascade establishes a three-stage model of M1 macrophage polarization that progressively locks in a pro-inflammatory state, positioning it as a core brake on the healing process. Visualization of key M1/M2 marker expression confirmed that M1-related markers (IL1B, CCL3, CXCL8, TIMP1) were consistently expressed at higher levels than M2-related markers (CSF1R, CD163, MRC1, ACP5, SELENOP) throughout the trajectory, corroborating our findings (Fig. 4E). A complete gene list and functional enrichment data are provided in Supplementary data sheets 1 and 2. Intercellular communication between macrophages and other cell types in nonunion To decipher the intercellular signaling network in nonunion, we systematically analyzed ligand-receptor interactions from our scRNA-seq data. This revealed a global enhancement in both the frequency and intensity of cellular communication in nonunion tissues relative to controls (Fig. 5A). A prominent finding was the specifically elevated interaction between macrophages and key cellular constituents of the bone immune microenvironment (BIM), suggesting that active macrophage-centric crosstalk may perturb BIM function (Fig. 5B, C). Analysis of cell-cell communication in the bone immune microenvironment. (A) Comparison of the total number and interaction strength of inferred ligand-receptor pairs between Control and Nonunion groups. (B) Comprehensive cellular interaction networks for Control and Nonunion tissues. (C) Specific interaction networks between macrophages and all other cell types in Control and Nonunion conditions. (D) Quantification of communication number and strength among M1 macrophages, M2 macrophages, endothelial cells, and osteoblasts. (E) Interaction networks highlighting the communications between M1/M2 macrophages and endothelial cells/osteoblasts. (F) Expression levels of key ligand-receptor pairs mediating communication between M1-like macrophages and other cell types in Control versus Nonunion tissues. We then focused on the impact of inflammatory macrophages on osteogenesis and healing by dissecting the interactions of M1 and M2 subtypes with osteoblasts and vascular endothelial cells. The communication frequency and intensity were markedly heightened in nonunion compared to the control group (Fig. 5D). Specifically, the interactions between M1 macrophages and both osteoblasts and endothelial cells were substantively enhanced (Fig. 5E). This shift likely stems from substantial alterations in specific ligand-receptor pairs between M1 macrophages and these adjacent cell populations (Fig. 5F). Specifically, upregulated TNF-α-TNFR1 and IL-1β-IL1R1 pairs transmit inhibitory signals to osteoblasts, while CCL2-CCR2 interactions disrupt endothelial barrier integrity and prevent vascular maturation, collectively establishing a microenvironment hostile to both bone formation and angiogenesis (Gomez et al., 2017; Lencel et al., 2011). A comprehensive list of all ligand-receptor pairs is provided in Supplementary data sheets 3 and 4. Collectively, these findings underscore the role of M1 macrophages as a central mediator of cell-cell communication within the bone microenvironment and suggest potential therapeutic strategies aimed at targeting M1 macrophage signaling. Identification of potential targets of Chaenomeles speciosa in treating nonunion Differential expression analysis using edgeR identified 322 genes with significant differences in expression between the nonunion group and the control M1 (cells) group (moderated F-test; Fig. 6A). These genes are key pathogenic genes associated with M1 macrophage dysregulation in nonunion. The complete list of these genes is provided in Supplementary data sheet 5. Screening of shared targets and functional enrichment in Chaenomeles speciosa for nonunion treatment. (A) The volcano map showed the differential gene expression between M1 groups. (B) Venn diagram, showing the intersection between Chaenomeles speciosa component targets and Nonunion related significant difference genes. (C) The network diagram of “drugs - active ingredients - Cross genes - diseases”, in which the yellow arrow represents Chaenomeles speciosa, the blue circle represents active ingredients, the purple regular octagons represent cross genes, and the orange square rectangle represents Nonunion. (D) Gene Ontology (GO) enrichment analysis showed that the top six representative biological processes (BPs), molecular functions (MFs) and cellular components (CCs) with the lowest p-value (p < 0.05). (E) KEGG pathway enrichment analysis found 18 important pathways related to cross genes (p < 0.05). (F) The two target pathways of signal transduction were analyzed by chord diagram. (G) “Drug-pathway-target-disease” interaction network. Green diamond: Chaenomeles speciosa; blue rectangles: IL-17 and TNF signaling pathways; purple octagons: cross-target genes; pink circle: Nonunion. Network pharmacology analysis of Chaenomeles speciosa showed that two bioactive compounds could act on 4556 unique proteins (Supplementary data sheet 6). The above datasets were integrated by cross-comparing the human orthologs of nonunion-related DEGs (converted via the biomaRt R package) with the molecular targets of Chaenomeles speciosa (Sweet) Nakai, and finally 114 potential therapeutic targets were obtained (Fig. 6B). An interactive “drug-compound-target-disease” network was constructed to visually present these relationships (Fig. 6C), which can highlight the direct associations between the active components of Chaenomeles speciosa (Sweet) Nakai and the pathogenic genes of nonunion. This association confirms the therapeutic potential of Chaenomeles speciosa (Sweet) Nakai, as it can directly act on the molecular nodes driving the pathological process of nonunion. Functional enrichment of Chaenomeles speciosa targets: GO and KEGG pathway analysis Functional annotation of the 114 intersecting genes via the clusterProfiler R package showed that these genes were significantly enriched in biological processes (BP), cellular components (CC), and molecular functions (MF) (adjusted p-value <0.05; Fig. 6D). Key biological processes included myeloid leukocyte migration, leukocyte chemotaxis, and monocyte chemotaxis—these processes are closely related to the persistent inflammatory microenvironment at the fracture site of nonunion. Cellular components were mainly localized in collagen-containing extracellular matrix and collagen trimers, which reflects the structural dysregulation of the extracellular matrix (ECM) in impaired bone healing. Molecular functions focused on growth factor binding, extracellular matrix structural constituent, collagen binding, and platelet-derived growth factor binding, which highlights the role of Chaenomeles speciosa (Sweet) Nakai targets in regulating ECM integrity and growth factor signaling pathways—both processes are crucial for bone repair. KEGG pathway analysis identified 18 significantly enriched pathways (adjusted p-value <0.05; Fig. 6E), and chord diagrams were used to visualize the key target pathways (Fig. 6F). The comprehensive “drug-pathway-target-disease” network highlights the regulatory interactions among these pathways and targets (Fig. 6G). The IL-17 signaling pathway emerged as a core pathogenic mechanism, which is consistent with previous research results (Liu et al., 2024). Six up-regulated targets were identified in the IL-17 signaling pathway: CCL2, MMP3, MAPK13, IL6, CXCL10, and CXCL1. A schematic diagram illustrating the regulation of IL-17 signal transduction by Chaenomeles speciosa in the treatment of nonunion is presented in Fig. 7A. Docking results of active ingredients and nonunion target molecules of Chaenomeles speciosa. (A) IL-17 signaling pathway, with up-regulated genes highlighted in purple. (B) Thermogram of molecular docking results. (C) Results of molecular docking between ent-epicatechin and MMP3 (affinity −8.80). (D) Molecular docking results of quercetin and MMP3 (affinity −9.50). The IL-17 signaling pathway promotes the occurrence of nonunion through two synergistic pathogenic mechanisms: on the one hand, it maintains a persistent pro-inflammatory microenvironment by sustaining M1 macrophage polarization and preventing its transition to the M2 repair phenotype; on the other hand, it directly down-regulates the expression of vascular endothelial growth factor (VEGF) and angiopoietin-1, induces endothelial cell apoptosis, and inhibits the construction of functional vascular networks systemically. This dual effect not only hinders inflammation resolution but also inhibits angiogenesis—both processes are prerequisites for effective bone repair, indicating that the IL-17 signaling pathway is a key therapeutic target of Chaenomeles speciosa (Sweet) Nakai. Molecular docking verification of Chaenomeles speciosa's active compounds Molecular docking simulations were used to verify the interactions between the bioactive compounds of Chaenomeles speciosa (Sweet) Nakai and the 6 up-regulated targets (CCL2, MMP3, MAPK13, IL6, CXCL10, CXCL1) in the IL-17 pathway. The binding affinity thresholds were defined as follows: potential binding (< −4.25 kcal/mol), good binding (< −5.0 kcal/mol), and strong binding (< −7.0 kcal/mol) (Hsin et al., 2013). The strongest binding interactions were observed between ent-epicatechin and MMP3 (−8.80 kcal/mol; Fig. 7C) and between quercetin and MMP3 (−9.50 kcal/mol; Fig. 7D). This calculation result is consistent with previous studies, confirming the inhibitory effect of Chaenomeles speciosa on the MMP3 target (Thakore et al., 2026). In addition, ent-epicatechin also showed strong binding with CCL2 and MAPK13, while quercetin also had strong interactions with CCL2, IL6, and MAPK13—this screened out new potential targets for Chaenomeles speciosa (Sweet) Nakai in the treatment of nonunion. The multi-target binding profile of quercetin and ent-epicatechin aligns with their established anti-inflammatory and antioxidant activities: quercetin is a well-documented flavonoid that inhibits NF-κB and MAPK signaling pathways to suppress pro-inflammatory cytokine production (Cheng et al., 2019), while ent-epicatechin, a major catechin in Chaenomeles speciosa, exhibits similar anti-inflammatory effects through modulation of iNOS and COX-2 expression (Savina et al., 2023). These pharmacological properties support their role as key effectors in CsEVs-mediated immunomodulation and bone regeneration. Complete docking parameters, control ligand information, and score matrices are provided in Supplementary data sheet 7. In the IL-17 signal cascade, MMP3 is a key downstream effector molecule that amplifies the inflammatory response by degrading extracellular matrix components and activating pro-inflammatory mediators (Singh et al., 2018). The down-regulation of MMP3 by the active components of Chaenomeles speciosa (Sweet) Nakai can disrupt this inflammatory amplification loop, weaken the IL-17-driven cascade (Duan et al., 2023), and at the same time retain the extracellular matrix scaffold necessary for VEGF-mediated angiogenesis. This process helps blood vessels invade the fracture callus and accelerate bone healing, which is consistent with the results of pathway enrichment analysis and also verifies the therapeutic mechanism of Chaenomeles speciosa (Sweet) Nakai. Direct binding studies, such as surface plasmon resonance or isothermal titration calorimetry, would provide more definitive evidence for the interaction between quercetin and MMP3, though such experiments are technically demanding and beyond the current scope. Nevertheless, several lines of evidence support the biological relevance of this interaction. Quercetin has been extensively documented as a natural inhibitor of matrix metalloproteinases, including MMP3, in various inflammatory and degenerative conditions (Lu et al., 2026; Tepedelenlioglu et al., 2026). Moreover, the anti-inflammatory and bone-protective effects of quercetin have been demonstrated in preclinical models of osteoarthritis and periodontitis, where its therapeutic effects were associated with downregulation of MMP3 expression (Afzal et al., 2025; Arumuganainar et al., 2025). Our in vivo data further corroborate this functional relevance, as CsEVs treatment significantly reduced MMP3 expression in fracture callus tissues (Fig. 11C) while promoting bone regeneration. Taken together, the combination of molecular docking predictions, literature evidence, and our in vivo observations provides a coherent body of evidence supporting the role of quercetin as a key bioactive compound in CsEVs-mediated MMP3 modulation. Immunohistochemical analysis of fracture nonunion tissues. Staining for key markers in the Control and EVs groups at 4 and 8 weeks post-treatment. (A) Osteogenic markers OPN. (B) M1 macrophage marker iNOS. (C) Matrix remodeling enzyme MMP3. Male Sprague-Dawley rats were used. Scale bar: 200 μm. Isolation, characterization, and biocompatibility of Chaenomeles speciosa EVs Plant-derived EVs have emerged as superior therapeutic agents compared to crude plant extracts due to their phospholipid bilayer structure, which protects and facilitates the efficient delivery of bioactive cargo to target cells (Ilahibaks et al., 2023). To investigate the therapeutic potential of Chaenomeles speciosa, we isolated its derived EVs (CsEVs) using sequential centrifugation and sucrose density gradient ultracentrifugation. Transmission electron microscopy confirmed the presence of homogeneous, cup-shaped, membrane-bound vesicles with a mean particle size of approximately 168.5 nm (Fig. 8A), which is characteristic of plant EVs. Crucially, cellular uptake experiments demonstrated that CsEVs were efficiently internalized by RAW 264.7 macrophages at different EV concentrations (0, 105, and 106 particles/mL) (Fig. 8B), a prerequisite for their subsequent biological functions. Quantification of EV uptake confirmed concentration-dependent internalization efficiency (Fig. 8D). Characterization and biocompatibility of CsEVs. (A) Workflow for extracellular vesicle (EV) isolation from Chaenomeles speciosa, including ultra-centrifugation, sucrose gradient centrifugation, and TEM visualization. (B) Confocal microscopy images showing cellular localization of EVs (red) with Hoechst nuclear staining (blue) and actin filaments (green) in cells at different EV concentrations (0, 105, and 106 particles/ml). (C) Phalloidin staining (green) showing cytoskeleton organization in RAW 264.7 macrophages exposed to different EV concentrations. (D) Quantification of EV uptake/cellular internalization at different particle concentrations (n = 3). (E) Live/Dead staining (green: live cells; red: dead cells) of RAW 264.7 macrophages after EV treatment at different concentrations. (F) Quantification of cell spreading area from phalloidin staining (n = 3). (G) Statistical analysis of cell viability from Live/Dead staining (n = 3). Data are mean ± SD; significance was determined by one-way ANOVA with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001). Scale bar: 100 nm (A), 200 μm (B, C, E). Given that bone regeneration involves complex interactions between endothelial cells (Mani et al., 2019), assessing the biocompatibility of CsEVs is essential. Phalloidin staining of RAW 264.7 macrophages exposed to different EV concentrations (0, 105, and 106 particles/mL) for 3 days showed well-spread actin filaments and normal morphology (Fig. 8C), with quantification of cell spreading area confirming no adverse effects on cytoskeletal organization (Fig. 8F). Live/Dead staining after exposure to different EV concentrations revealed a vast majority of viable cells (green) with minimal cell death (red) across all groups (Fig. 8E), and statistical analysis of cell viability confirmed that CsEVs were non-cytotoxic (Fig. 8G). These results collectively establish the excellent in vitro biocompatibility of CsEVs, warranting their further investigation for therapeutic applications. Compared to mammalian cell-derived EVs, plant EVs offer scalable production without ethical constraints, intrinsic stability during storage, and natural targeting capacity toward inflammatory cells due to their unique membrane composition, making them particularly suitable for clinical translation in bone defect therapy (Alzahrani et al., 2023). CsEVs attenuate inflammation and promote angiogenesis in vitro To investigate the immunomodulatory potential of CsEVs, we assessed the expression of macrophage polarization markers in LPS-stimulated cells. To determine the baseline effects of CsEVs, we included a group treated with CsEVs alone in the absence of LPS stimulation. Heatmap visualization revealed differential expression patterns of MMP3, iNOS, and Arg across treatment groups (Fig. 9A). RT-qPCR analysis quantified these changes, revealing that treatment with CsEVs significantly downregulated the expression of the M1 marker iNOS while upregulating the M2 marker Arg-1 at the mRNA level in the LPS + EVs group compared to the LPS group (Fig. 9B). CsEVs alone did not significantly alter the expression of these markers compared to the untreated control. Western blot analysis confirmed these findings at the protein level (Fig. 9C), with densitometric quantification showing decreased iNOS and increased Arg-1 expression in the LPS + EVs group compared to the LPS group (Fig. 9D). This shift toward an anti-inflammatory M2 phenotype is conducive to tissue repair (Franzini et al., 2023). Concurrently, we observed a modulation in the expression of MMP3, a matrix remodeling enzyme, suggesting that CsEVs may also influence extracellular matrix reorganization during healing (Fig. 9A–D). Given the intimate coupling between angiogenesis and osteogenesis, we next evaluated the pro-angiogenic capacity of CsEVs. RAW 264.7 macrophages were treated with CsEVs under LPS stimulation, and the conditioned medium was collected and applied to endothelial cells for tube formation assays. Representative fluorescent microscopy images showed that endothelial cells treated with the conditioned medium from the LPS + EVs group developed more extensive and intricate tubular networks compared to controls (Control, EVs, and LPS groups) (Fig. 9E), indicating that CsEVs promoted angiogenic activity in vitro. These findings reveal that CsEVs effectively promote angiogenesis in vitro while simultaneously reshaping the inflammatory landscape by driving macrophage polarization toward the pro-regenerative M2 phenotype. This coordinated action simultaneously attenuates inflammation and promotes vascular growth, strongly suggesting that CsEVs operate through an “immuno-angiogenic coupling” mechanism. This coupling is mechanistically supported by the M1-to-M2 transition: M1-derived TNF-α and IL-1β directly inhibit endothelial proliferation and tube formation, whereas M2-secreted VEGF and FGF-2 stimulate angiogenesis (Fang et al., 2020). Our observation that CsEVs rescue angiogenesis by switching macrophage phenotype aligns with recent studies on Chaenomeles speciosa extracts in osteoarthritis and rheumatoid arthritis models, where similar anti-inflammatory effects (iNOS suppression, IL-6 and TNF-α inhibition) were reported (Duan et al., 2023; Hsieh et al., 2023), supporting the conserved immunomodulatory activity of this herb across inflammatory bone diseases. Assessment of anti-inflammatory and pro-angiogenic activities. (A) Heatmap of relative mRNA expression levels of MMP3, iNOS, and Arg in Control, EVs, LPS, and LPS + EVs groups. (B) Quantification of relative mRNA expression levels by RT-qPCR analysis (n = 3). (C) Western blot analysis showing protein levels of iNOS, Arg, MMP3, and GAPDH (loading control) in Control, EVs, LPS, and LPS + EVs groups. (D) Densitometric quantification of protein bands shown in bar graphs (n = 3). (E) Representative fluorescent microscopy images of tube formation by endothelial cells treated with conditioned medium from RAW 264.7 macrophages in Control, EVs, LPS, and LPS + EVs groups. Data are mean ± SD; significance was determined by one-way ANOVA with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001). The “EVs” group represents macrophages treated with CsEVs alone in the absence of LPS. Scale bar: 100 μm. CsEVs accelerate bone regeneration in a fracture model Building on the promising in vitro results, we evaluated the therapeutic efficacy of CsEVs in an established fracture healing model. Micro-CT 3D reconstructions of femoral defects at 4 and 8 weeks post-treatment showed enhanced bone repair in the CsEVs-treated group compared to the control (Fig. 10B). Quantification of micro-CT parameters including bone volume fraction (BV/TV), trabecular thickness (Tb·Th), and bone mineral density (BMD) confirmed significantly improved bone regeneration in the CsEVs group (Fig. 10C). The control group exhibited delayed bone healing with minimal callus formation, whereas the CsEVs group showed substantial formation of new bony callus, bridging the osteotomy gap. Therapeutic efficacy of CsEVs in a fracture model. (A) Schematic diagram of the experimental timeline for nonunion treatment. (B) Representative 3D micro-CT reconstructions of femoral defects from Control and EVs groups at 4 and 8 weeks post-treatment. (C) Quantification of micro-CT parameters including bone volume fraction (BV/TV, %), trabecular thickness (Tb·Th, 1/mm), and bone mineral density (BMD, mgHAp/cm3) (n = 6). (D) H&E staining showing bone tissue morphology and callus formation at the defect site in Control and EVs groups at 4 and 8 weeks, with magnified views of the boxed regions shown below each image. (E) Masson's trichrome staining showing collagen deposition (blue) and bone formation at the defect site in Control and EVs groups at 4 and 8 weeks, with magnified views of the boxed regions shown below each image. Male Sprague-Dawley rats were used. Scale bar: 200 μm. Histological analysis further substantiated the micro-CT findings. Hematoxylin and eosin (H&E) staining demonstrated a significantly larger volume of callus tissue at the defect site in the CsEVs-treated group at both 4 and 8 weeks (Fig. 10D). Masson's trichrome staining revealed more abundant collagen deposition (stained blue) in the CsEVs group, indicating advanced matrix remodeling and osteoid formation (Fig. 10E). Masson's trichrome staining of collagen fibers provides critical insight into the quality of bone repair: the transition from disorganized fibrous tissue (predominant in nonunion) to organized collagen deposition with proper alignment precedes mineralization and determines the mechanical competence of the healed bone (Long et al., 2025). The enhanced collagen deposition observed in the CsEVs group suggests that the treatment not only accelerates bone formation but also promotes the structural organization necessary for functional load-bearing capacity, a feature often compromised in conventional bone grafting procedures (Yu and Wei, 2021). These results collectively demonstrate that CsEVs effectively promote callus formation, matrix deposition, and bone bridging in fracture healing. To elucidate the molecular mechanisms underlying the observed therapeutic effects in vivo, we performed immunohistochemical (IHC) analysis on fracture callus tissues at 4 and 8 weeks post-treatment. Staining for the osteogenic marker OPN revealed markedly stronger positive signals in the CsEVs-treated group (Fig. 11A), confirming enhanced osteogenic activity at the injury site. IHC analysis of macrophage phenotypes showed that CsEVs treatment significantly reduced the number of iNOS-positive M1 macrophages (Fig. 11B). This in vivo validation of macrophage polarization aligns perfectly with our in vitro data, strongly supporting that the anti-inflammatory effect of CsEVs is a key mechanism of action. Furthermore, IHC staining for MMP3, a marker of matrix remodeling, showed regulation in the CsEVs group (Fig. 11C), suggesting a role in extracellular matrix reorganization. These integrated in vivo findings demonstrate that CsEVs reprogram the pathological state of fracture into one that is primed for healing through coordinated immunomodulatory and pro-regenerative mechanisms. Beyond efficacy, a potential concern regarding the translational application of plant-derived EVs is their immunogenicity in mammalian hosts. Unlike mammalian cell-derived EVs, which may carry xenogeneic or allogeneic surface proteins that trigger immune rejection, plant-derived EVs are composed of phospholipid bilayers enriched with plant-specific lipids and membrane-associated proteins that are generally well-tolerated in vivo. Previous studies have demonstrated that plant EVs, including those derived from edible plants such as grapefruit, ginger, and lemon, exhibit low immunogenicity and do not elicit significant inflammatory responses upon systemic or local administration (Alzahrani et al., 2023; Jin et al., 2024). Moreover, our in vivo data showed no signs of local or systemic adverse reactions in CsEVs-treated rats, as evidenced by normal postoperative recovery and the absence of abnormal inflammatory infiltration in histological sections. These observations, together with the excellent biocompatibility we observed in vitro (Fig. 8E–G), support the safety profile of CsEVs for bone repair applications. Nevertheless, future studies with long-term follow-up and comprehensive immunological evaluations in larger animal models would further clarify the biosafety and immunogenicity of plant-derived EVs before clinical translation. While direct in vivo administration of purified quercetin or ent-epicatechin would provide more definitive evidence for their individual contributions to the observed effects, such long-term studies are beyond the scope of the present investigation. We therefore consider the current work as a foundational step, with future studies planned to dissect the specific roles of these bioactive compounds in macrophage polarization and fracture healing using purified compound administration in vivo. Conclusion This study establishes a comprehensive mechanistic framework for the efficacy of Chaenomeles speciosa in fracture nonunion. By integrating single-cell sequencing and network pharmacology, we identified the IL-17 signaling pathway and a dominant M1 macrophage phenotype as central pathological drivers. We further demonstrated that CsEVs function as the key therapeutic effectors, effectively breaking the cycle of inflammation and impaired healing. CsEVs orchestrate bone repair through a coordinated “immuno-angiogenic coupling” mechanism, simultaneously polarizing macrophages toward a regenerative M2 state and promoting functional angiogenesis. This multi-targeted action was validated to successfully bridge bone defects in vivo. Our findings position CsEVs as a potent, natural nanotherapeutic strategy for fracture healing, bridging the gap between traditional medicine and modern targeted therapy. Abbreviations Single-cell RNA sequencing Chaenomeles speciosa-derived extracellular vesicles Extracellular vesicles Traditional Chinese Medicine Gene Expression Omnibus Traditional Chinese Medicine Systems Pharmacology Differentially expressed genes Gene Ontology Kyoto Encyclopedia of Genes and Genomes Bone immune microenvironment Pro-inflammatory macrophage Anti-inflammatory/regenerative macrophage Matrix metalloproteinase-3 Extracellular matrix Reactive oxygen species Phosphate-buffered saline Transmission electron microscopy Micro-computed tomography Hematoxylin and eosin Immunohistochemistry Western blot Reverse transcription quantitative PCR Standard deviation Analysis of variance CRediT authorship contribution statement Jingxiang Chen: Data curation, Formal analysis, Investigation, Methodology, Writing – original draft. Kaihan Zhuang: Investigation, Software, Validation, Writing – original draft. Yongpeng Wang: Investigation, Methodology, Visualization, Writing – original draft. Shengjin Yuan: Formal analysis, Investigation, Software, Writing – original draft. Zikun Xie: Conceptualization, Methodology, Supervision, Writing – review & editing. Qinghua Fan: Project administration, Resources, Supervision, Writing – review & editing. Tao Liu: Conceptualization, Funding acquisition, Project administration, Supervision, Writing – review & editing. Funding This study was supported by the Scientific Research Project of Guangdong Provincial Administration of Traditional Chinese Medicine (No. 20242134). Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgments Not applicable. Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.bonr.2026.101951. Contributor Information Jingxiang Chen, Email: 530290112@qq.com. Kaihan Zhuang, Email: 13643037126@163.com. Yongpeng Wang, Email: 675393063@qq.com. Shengjin Yuan, Email: yuanshenjin2020@qq.com. Zikun Xie, Email: xiezikun77@smu.edu.cn. Qinghua Fan, Email: doc.fan@163.com. Tao Liu, Email: liutao5411@126.com. Supplementary data Supplementary material Data availability Data will be made available on request. References Associated Data Supplementary Materials Supplementary material Data Availability Statement Data will be made available on request.