Curcuma-derived nanovesicle-loaded ROS-responsive hydrogels reprogram iron metabolism to enhance cartilage regeneration after microfracture Authors: Yuhao Mu, Hao Li, Ruiyang Zhang, Runmeng Li, Ke Ma, Fakai Li, Yongkang Yang, Yiming Ren, Zheng Guo, Tianze Gao, Chao Wang, Jiafeng Miao, Qinlin Yan, Juan Guan, Shuyun Liu, Quanyi Guo Published: 2026 ABSTRACT Microfracture (MF) is widely used for cartilage repair, but it often yields limited clinical benefit because it predominantly induces fibrocartilage formation. However, the mechanisms underlying this fibrocartilaginous repair remain insufficiently defined. In this study, temporal histopathological profiling and integrative bioinformatic analyses of post-microfracture specimens revealed that early extracellular iron accumulation was positively associated with ferroptosis severity. Single-cell RNA sequencing further delineated a bifurcating differentiation trajectory of bone marrow mesenchymal stem cells (BMSCs) toward either hyaline-like chondrocytes or fibrocartilaginous chondrocytes, with ferroptosis acting as a critical regulator at the branch point. Given the antioxidant and ferroptosis-modulating activity of curcumin-derived components, we hypothesized that curcuma-derived extracellular vesicles (CDEVs) could suppress ferroptosis and bias the differentiation of BMSCs toward hyaline cartilage. Mechanistically, in vitro assays identified Pvu-miR-159 in CDEVs as a key functional cargo that attenuates ferroptosis via the PTPN12-ERK1/2-ATF4-GPX4 pathway. To enhance its translational potential, we developed an injectable reactive oxygen species (ROS)-responsive hydrogel enabling sustained CDEVs delivery, which effectively reduced ferroptosis and promoted cartilage regeneration in vivo. Together, these findings uncover a ferroptosis-driven mechanism contributing to suboptimal microfracture repair and support a plant-vesicle-based, ROS-responsive delivery strategy to reprogram BMSCs toward improved regenerative outcomes. Introduction Due to the limited intrinsic repair capacity of articular cartilage, achieving durable regeneration remains a significant clinical challenge. Clinically, mesenchymal stem cells (MSCs) are introduced or recruited to injured sites to facilitate cartilage repair; among current approaches, microfracture is the most widely used surgical intervention because it promotes the recruitment of bone marrow mesenchymal stem cells (BMSCs) to cartilage defects [1,2]. Although microfracture can relieve symptoms effectively in the short term, its long-term efficacy is often disappointing: the repair tissue is predominantly fibrocartilage, which has inferior biomechanical properties and durability compared with native hyaline cartilage, and it commonly undergoes progressive degeneration within 3-5 years postoperatively, with approximately 30% of patients requiring secondary surgical intervention within 5-10 years [[3], [4], [5]]. Because BMSCs are the principal cellular source for microfracture-mediated repair, their differentiation trajectory critically determines the histological phenotype of the regenerated cartilage [6]. However, following microfracture, recruited BMSCs frequently adopt a fibrotic/fibrocartilaginous differentiation program, directly compromising repair quality [7]. This process is tightly regulated by the tissue microenvironment. Prior studies have shown that excessive inflammatory response, oxidative stress, and related factors can induce aberrant the differentiation of MSCs, and promote fibrosis or heterotopic ossification [8,9]. Nevertheless, key research gaps remain regarding endogenous BMSCs after microfracture surgery, including the pathways through which recruited BMSCs transition into fibrochondrocytes within the defect niche and the core mechanisms governing this lineage bifurcation. From the perspective of the post-microfracture biochemical microenvironment, hemoglobin and heme degradation generate large amounts of free iron ions, which we hypothesize may create an acute, localized iron-overload milieu at the repair site. Iron is an essential trace element for numerous cellular physiological processes; however, under overload conditions, it catalyzes the production of reactive oxygen species (ROS) through the Fenton reaction [10,11], triggering intense oxidative stress [12]. This stress not only directly damages the cellular structures but also perturbs signal transduction and inhibits the normal chondrogenic differentiation of BMSCs. Using pseudotime analysis of single-cell sequencing data from microfracture-repaired tissues, we inferred divergent BMSC differentiation trajectories toward either fibrocartilage or hyaline cartilage, and identified ferroptosis as a critical pathway regulating the branch point of this process (Scheme 1a).Scheme 1This schematic illustrates ferroptosis as a central mechanism regulating the differentiation trajectory(leading to Hyalinosis and Fibrosis) of regenerated cartilage following microfracture surgery (a). The injectable ROS-responsive hydrogel HAMA-APBA/CDEVs (b) targets the PTPN12-ERK1/2-GPX4 (SLC7A11) axis via Pvu-miR-159, enhancing stem cell resistance to ferroptosis and thereby improving the quality of regenerated cartilage repair after microfracture surgery (c). This schematic illustrates ferroptosis as a central mechanism regulating the differentiation trajectory(leading to Hyalinosis and Fibrosis) of regenerated cartilage following microfracture surgery (a). The injectable ROS-responsive hydrogel HAMA-APBA/CDEVs (b) targets the PTPN12-ERK1/2-GPX4 (SLC7A11) axis via Pvu-miR-159, enhancing stem cell resistance to ferroptosis and thereby improving the quality of regenerated cartilage repair after microfracture surgery (c). Plant-derived extracellular vesicles (PDEVs) emerge as promising therapeutic agents, carrying diverse bioactive molecules (e.g., microRNAs, proteins, and metabolites) that exert anti-inflammatory, antioxidant, and anti-tumor activities in humans [13,14]. Their biocompatibility, low immunogenicity, and cost-effectiveness further support their potential for clinical translation [15]. Plant-derived compounds such as polyphenols and saponins can directly regulate changes in cellular metabolism [16,17]. Curcuma longa is a traditional Chinese medicinal herb with broad pharmacological activities [18]. Curcuma-derived extracellular vesicles (CDEVs) have been reported to exert anti-inflammatory and metabolic regulatory effects and show therapeutic potential for degenerative conditions such as osteoarthritis and intervertebral disc degeneration by attenuating inflammation and promoting extracellular matrix synthesis, thereby mitigating disease progression [[19], [20], [21]]. Moreover, CDEVs may modulate iron metabolism by regulating oxidative stress and glutathione metabolism pathways, thereby inhibiting chondrocyte ferroptosis [22]. However, existing studies have primarily focused on small-molecule metabolites (e.g., curcumin, turmerone), whereas the cross-species regulatory roles of plant-derived microRNAs remain unclear. Collectively, it is clinically important to examine whether CDEVs can regulate the ferroptosis cascade and restore the abnormal BMSCs differentiation trajectory. Here we demonstrate that CDEVs act as messengers, delivering diverse bioactive cargos (e.g., proteins, RNA, and plant metabolites) that confer ferroptosis resistance to BMSCs. Sequencing analyses further revealed that the CDEV-mediated ferroptosis-resistant imprint on differentiating BMSCs is mediated primarily by Pvu-miR-159. This miRNA suppresses PTPN12, a member of the protein tyrosine phosphatase (PTP) family, thereby restoring downstream ERK1/2 phosphorylation and activating the transcription factor 4 (ATF4) axis. Consequently, expression of solute carrier family 7 member 11 (SLC7A11) is upregulated, strengthening the capacity of the glutathione peroxidase 4 (GPX4)-glutathione (GSH) system to scavenge lipid peroxides and ultimately protecting BMSCs from ferroptosis. These findings provide a mechanistic basis for leveraging CDEVs to counter ferroptosis in chondral tissues during microfracture repair. To improve the clinical translational potential of CDEVs, we developed a ROS-responsive hydrogel composed of 3-aminophenylboronic acid (APBA)-modified hyaluronic acid methacrylate (HAMA) as a delivery platform (HAMA-APBA/CDEVs) (Scheme 1b) [23]. This system dynamically responds to oxidative stress in the pathological microenvironment [24]. The hydrogel exhibited excellent tolerability in surgically induced cartilage defects and enabled sustained CDEVs release, resulting in persistent suppression of ferroptosis in BMSCs. Consistent in vitro and in vivo evidence further confirmed that this CDEV-incorporated hydrogel is well-tolerated at the defect site after microfracture and restores redox homeostasis in ferroptotic BMSCs via the PTPN12-ERK1/2-ATF4-GPX4 axis (Scheme 1c), thereby promoting robust regeneration of high-quality cartilage. Thus, our study presents a clinically translatable strategy to improve the quality of cartilage repair after microfracture surgery. Results Aberrant differentiation of regenerated cartilage following microfracture surgery Repair of articular cartilage defects remains a major challenge in orthopedics because of the extremely limited intrinsic healing capacity of articular cartilage; untreated lesions often progress to debilitating osteoarthritis (OA). Microfracture is a first-line clinical intervention designed to recruit BMSCs and activate endogenous repair programs (Fig. S1) [25]. However, its clinical efficacy is frequently limited because the regenerated tissue is dominated by fibrocartilage, which has inferior biomechanical properties compared with native hyaline cartilage (Fig. S2) [26]. Although BMSCs are considered the principal reparative cell source, the molecular determinants driving their aberrant commitment to fibrocartilaginous repair after microfracture remain poorly understood. To address this question, we established a rat microfracture model (Fig. 1a). Safranin-O/Fast Green (SOG) staining revealed progressive fibrous scar–like repair within the regenerated cartilage matrix (Fig. 1b). Meanwhile, the peripheral cartilage adjacent to the defect exhibited reduced Safranin-O staining and loss of glycosaminoglycans, likely reflecting mechanical overload–induced degeneration secondary to poor-quality fibrocartilaginous repair (Fig. S3a).Fig. 1Abnormal cartilage differentiation and single-cell sequencing of microfracture surgery. a) Modeling and single-cell sequencing of microfractures in SD rats. b) SOG staining of regenerated cartilage at different time points (Control、3d、7d、14d、28d) following microfracture surgery. c) t-distributed stochastic neighbor embedding (t-SNE) visualization of 49,095 cells using scRNA-seq. d) Visualization of the characteristic genes of the Chondrocytes cluster (Col2a1, SOX9, Acan, Comp) and Mesenchymal stem cells cluster (Cd44,Thy1) on the t-SNE map. e) t-distributed stochastic neighbor embedding (t-SNE) visualization of Chondrocytes cluster and Mesenchymal stem cells cluster reclustering using scRNA-seq. f) Dot plot of the top 4 marker genes in each cell subcluster. g) Proportion of each cell subcluster in different groups. Abnormal cartilage differentiation and single-cell sequencing of microfracture surgery. a) Modeling and single-cell sequencing of microfractures in SD rats. b) SOG staining of regenerated cartilage at different time points (Control、3d、7d、14d、28d) following microfracture surgery. c) t-distributed stochastic neighbor embedding (t-SNE) visualization of 49,095 cells using scRNA-seq. d) Visualization of the characteristic genes of the Chondrocytes cluster (Col2a1, SOX9, Acan, Comp) and Mesenchymal stem cells cluster (Cd44,Thy1) on the t-SNE map. e) t-distributed stochastic neighbor embedding (t-SNE) visualization of Chondrocytes cluster and Mesenchymal stem cells cluster reclustering using scRNA-seq. f) Dot plot of the top 4 marker genes in each cell subcluster. g) Proportion of each cell subcluster in different groups. We next performed single-cell RNA sequencing (scRNA-seq) on regenerated cartilage tissues collected at multiple postoperative time points (Control, Week 1, Week 2, Week 4; n = 10). Unbiased clustering identified 13 major cell populations across 49,095 single cells (Fig. 1c; Fig. S4). Using canonical markers, we focused on chondrocytes (COL2A1, SOX9, ACAN, COMP) and mesenchymal stem cell clusters (CD44, THY1) (Fig. 1d). Subclustering further resolved 7 transcriptionally distinct subpopulations (Fig. 1e; Fig. S5): mesenchymal stem cells, hyaline chondrocytes, fibrocartilage chondrocytes, hypertrophic chondrocytes, pre-fibrocartilage chondrocytes, pre-hypertrophic chondrocytes, and regulatory chondrocytes. We further characterized their signature genes (Fig. 1f) and temporal abundance dynamics (Fig. 1g). Iron-overloaded microenvironment after microfracture induces ferroptosis in BMSCs and drives aberrant fibrocartilage differentiation Pseudotime trajectory reconstruction of mesenchymal stem cells, hyaline chondrocytes, and fibrocartilage chondrocytes identified Node 3 as a key bifurcation point at which BMSCs diverge toward either hyaline chondrogenesis or a fibrocartilage fate (Fig. 2a; Fig. S6). Cell states along these two branches were defined by distinct gene signatures (Fig. 2b). KEGG enrichment analysis of Node 3–associated differentially expressed genes highlighted focal adhesion, Rap1 signaling, ferroptosis, apoptosis, and the MAPK pathway among the top pathways (Fig. 2c and d). Microfracture induces intra-articular bleeding, exposing the defect site to ruptured erythrocytes and hemoglobin degradation products, thereby releasing substantial iron (Fig. S7) [27]. This creates an iron-overloaded microenvironment that can drive Fenton chemistry, generate excessive ROS, and promote ferroptosis-driven lipid peroxidation, ultimately impairing cell viability and differentiation.Fig. 2Pseudotime analyses reveal ferroptosis as the core factor inducing the differentiation of MSCs trajectories after microfracture surgery. a) Pseudotime ordering (Node1,2,3-key nodes in the differentiation process, branch lines-differentiation trajectory) on the mean subclusters (Fibrocartilage Chondrocytes, Hyaline Chondrocytes, Mesenchymal stem cell) arranged them into a major trajectory (Hyalinosis and Fibrosis). b) The distribution of each subcluster at the Pseudotime map (Fibrocartilage Chondrocytes-red, Hyaline Chondrocytes-brown, Mesenchymal stem cell-green). c) Distribution of characteristic genes in the differentiation trajectory(MSCs to Fibrocartilage- Col1a1,Col1a2; MSCs to Hyaline cartilage- Col2a1,Col9a1) (Color of point-each differentiation time state). d) The top 20 pathways after KEGG enrichment analysis of the differentiated gene set at Node 3.(Ferroptosis, Apoptosis). e) The top 3 transcription factors of each sucluscter revealed by the SCHENIC transcription factor regulatory network. f) The distribution of the gene ACSL4 in the Pseudotime trajectory. g-h) Volcano maps of differentiated genes of BMSCs in different groups after microfracture surgery, 2W vs 1W(G) and 4W vs 2W(H). i) Immunohistochemical staining of GPX4 between different groups after microfracture surgery of SD rats. j) Ferroptosis-related gene expression of rat's regenerated cartilage after microfracture surgery. k) Iron content of regenerated cartilage tissue in different groups. l) Heat map of ROS-related genes expression. Data are represented as mean ± SD. Statistical significance was determined by a two-sided Student's t-test. Pseudotime analyses reveal ferroptosis as the core factor inducing the differentiation of MSCs trajectories after microfracture surgery. a) Pseudotime ordering (Node1,2,3-key nodes in the differentiation process, branch lines-differentiation trajectory) on the mean subclusters (Fibrocartilage Chondrocytes, Hyaline Chondrocytes, Mesenchymal stem cell) arranged them into a major trajectory (Hyalinosis and Fibrosis). b) The distribution of each subcluster at the Pseudotime map (Fibrocartilage Chondrocytes-red, Hyaline Chondrocytes-brown, Mesenchymal stem cell-green). c) Distribution of characteristic genes in the differentiation trajectory(MSCs to Fibrocartilage- Col1a1,Col1a2; MSCs to Hyaline cartilage- Col2a1,Col9a1) (Color of point-each differentiation time state). d) The top 20 pathways after KEGG enrichment analysis of the differentiated gene set at Node 3.(Ferroptosis, Apoptosis). e) The top 3 transcription factors of each sucluscter revealed by the SCHENIC transcription factor regulatory network. f) The distribution of the gene ACSL4 in the Pseudotime trajectory. g-h) Volcano maps of differentiated genes of BMSCs in different groups after microfracture surgery, 2W vs 1W(G) and 4W vs 2W(H). i) Immunohistochemical staining of GPX4 between different groups after microfracture surgery of SD rats. j) Ferroptosis-related gene expression of rat's regenerated cartilage after microfracture surgery. k) Iron content of regenerated cartilage tissue in different groups. l) Heat map of ROS-related genes expression. Data are represented as mean ± SD. Statistical significance was determined by a two-sided Student's t-test. SCENIC transcription factor network analysis further showed that transcription factors associated with hyaline chondrogenesis (ATF4, SOX9, Maff_extended) and those associated with fibrocartilage differentiation (Fosl1, Klf6_extended, Creb3l1) exhibited distinct cell state-specific activity (Fig. 2e; Figs. S8 and 9). Prior evidence indicates that ATF4 deficiency reduces Slc7a11 expression, impairs cystine uptake, lowers GSH levels, and increases susceptibility to ferroptosis [28,42], consistent with its enrichment in hyaline chondrocyte clusters (Fig. S10). Trajectory mapping of the ferroptosis marker ACSL4 revealed a pattern closely aligned with the MSC-to-fibrocartilage lineage, further supporting ferroptosis as a driver of aberrant differentiation (Fig. 2f). Differential expression analyses showed that during the early repair phase (Weeks 1–2), hyaline cartilage–related genes (e.g., Lamb1, Serpinh1) were significantly downregulated, whereas during 2–4 weeks, ferroptosis-protective genes (GPX4, FTH1) were markedly decreased and fibrosis markers (Col1a1, Col1a2) were elevated (Fig. 2g and h). Immunohistochemistry confirmed that GPX4 expression increased compensatorily in regenerated cartilage during postoperative days 0-14, but declined sharply thereafter (Fig. 2i, Fig. S3b). Reverse transcription quantitative real-time polymerase chain reaction (RT-qPCR)further showed early upregulation followed by late-stage collapse of anti-ferroptosis genes (SLC7A11, GPX4, FTH1), whereas ACSL4 remained persistently elevated throughout days 0-28 (Fig. 2j). Direct quantification of tissue iron demonstrated a pronounced time-dependent accumulation that correlated with impaired cartilage repair (Fig. 2k). ROS levels also peaked during postoperative days 0-7 (Fig. 2l), defining a potential therapeutic window for redox-responsive interventions. Furthermore single-cell pseudotime analysis combined with KEGG enrichment revealed that ferroptosis and apoptosis are both involved in the aberrant differentiation trajectory of stem cells after microfracture; through in vivo inhibitor screening and in vitro cell death pathway-specific models, we demonstrate that CDEVs exert a highly specific and potent inhibitory effect on ferroptosis rather than apoptosis, necroptosis, or pyroptosis. Ferroptosis was identified as a major contributor to poor cartilage regeneration and supporting the anti-ferroptosis therapeutic strategy of this study (Fig. S11). Then CellChat analysis of post-microfracture cartilage regeneration revealed that ferroptosis-related signaling pathways (TWEAK, TNF, MIF, and CCL) are substantially enriched, particularly involving BMSCs interacting with macrophages, osteoclasts, and chondrocytes, indicating that the iron overload microenvironment promotes ferroptosis and inhibits normal cartilage regeneration via dysregulated paracrine signaling (Fig. S12). Together, these results demonstrate that ferroptosis biases BMSC differentiation away from a hyaline-like chondrocyte fate and toward fibrocartilaginous repair by suppressing the antioxidant axis (GPX4, FTH1, SLC7A11) and upregulating fibrotic markers such as Col1a1 and Col1a2. The post-microfracture iron overload generates ROS via Fenton chemistry, while hyperactivated ferroptosis-related paracrine signals (TWEAK, TNF) create a feedback loop that further reinforces aberrant fibrocartilaginous differentiation. Isolation and characterization of CDEVs CDEVs were isolated from turmeric rhizomes via size-exclusion chromatography (Fig. 3a). TEM and NTA confirmed their spherical morphology and an EV-consistent size distribution (50–200 nm; median 107.6 nm) (Fig. 3b and c). Untargeted metabolomics identified 2905 bioactive compounds, predominantly lipid-related molecules (50.77%) (Fig. 3d). The top metabolites included L-glutamine, zedoarondiol, and curcumin (Fig. 3e), which are associated with glutathione metabolism, ROS scavenging, and cytoprotection. PKH26 labeling confirmed rapid internalization of CDEVs by rat BMSCs within 4 h (Fig. 3f). Cytocompatibility assays supported their biosafety and identified 25 μg/mL as an optimal working concentration (Fig. 3g and h). Functionally, CDEVs significantly restored BMSC migration (scratch assay; Fig. 3i–l), proliferation (EdU; Fig. 3j–m), and transwell migration (Fig. 3k–n), indicating strong reparative potential under ferroptotic stress.Fig. 3Isolation, characterization, and biocompatibility of CDEVs. a) Schematic illustration of CDEVs collection isolation from Curcuma longa (turmeric). b) Representative TEM images revealing nanovesicular morphology. c) Size distributions analysis of CDEVs. d) Pie charts showing the biomolecular compositions of CDEVs. e) The top 10 metabolites identified in CDEVs. f) Cellular uptake of PKH26-labeled CDEVs (red) by BMSCs (nuclei: DAPI, blue). g) BMSCs viability after 24 h exposure to increasing CDEVs concentrations (0, 5, 10, 25, 50, 75, and 100 μg/mL, respectively) for 24 h (n = 4). h) BMSCs treated with PBS or CDEVs(100 μg/mL) for different time periods(0H,24H; 72H) and stained with calcein-AM (living cells, green) and propidium iodide (dead cells, red). i) Wound healing capacity of BMSCs measured using a scratch wound healing test in different groups (Control:PBS; Fe:FeCl3-100μM; CDEVs: FeCl3-100μM, CDEVs-25 μg/mL). j) EDU proliferation assay (red) of BMSCs in different groups. Nuclei: DAPI (blue). k) Migration capacity of BMSCs measured using the transwell assay. l-n) Quantified scratch wound healing rate (l), proliferation rate of EDU staining (m) and migrated cell count/HPF in Transwell (n) (n = 6). Isolation, characterization, and biocompatibility of CDEVs. a) Schematic illustration of CDEVs collection isolation from Curcuma longa (turmeric). b) Representative TEM images revealing nanovesicular morphology. c) Size distributions analysis of CDEVs. d) Pie charts showing the biomolecular compositions of CDEVs. e) The top 10 metabolites identified in CDEVs. f) Cellular uptake of PKH26-labeled CDEVs (red) by BMSCs (nuclei: DAPI, blue). g) BMSCs viability after 24 h exposure to increasing CDEVs concentrations (0, 5, 10, 25, 50, 75, and 100 μg/mL, respectively) for 24 h (n = 4). h) BMSCs treated with PBS or CDEVs(100 μg/mL) for different time periods(0H,24H; 72H) and stained with calcein-AM (living cells, green) and propidium iodide (dead cells, red). i) Wound healing capacity of BMSCs measured using a scratch wound healing test in different groups (Control:PBS; Fe:FeCl3-100μM; CDEVs: FeCl3-100μM, CDEVs-25 μg/mL). j) EDU proliferation assay (red) of BMSCs in different groups. Nuclei: DAPI (blue). k) Migration capacity of BMSCs measured using the transwell assay. l-n) Quantified scratch wound healing rate (l), proliferation rate of EDU staining (m) and migrated cell count/HPF in Transwell (n) (n = 6). Furthermore, in vitro and in vivo biosafety assessments demonstrated that CDEVs exhibit a wide safety window, with no cytotoxicity even at 1000 μg/ml and no toxicity to major organs or vital signs in rats, leading to the selection of 25 μg/ml as the optimal therapeutic dose (Fig. S13). Injectable ROS-responsive HAMA-APBA hydrogel loaded with CDEVs Histological staining and PCR analyses showed that ferroptosis-responsive markers increased progressively from day 0 to day 14 after microfracture and then declined thereafter. In parallel, ROS levels rose during days 0–7 and decreased after day 7. Leveraging this temporal pattern, we incorporated CDEVs into 3-aminophenylboronic acid (APBA)-modified hyaluronic acid methacrylate (HAMA) to fabricate an injectable, ROS-responsive hydrogel (HAMA-APBA/CDEVs). The hydrogel undergoes accelerated degradation at the ROS peak (∼day 7), enabling controlled CDEVs release into the iron-overloaded microenvironment before BMSCs reach ferroptosis decompensation (∼day 14). 3-aminophenylboronic acid-modified hyaluronic acid methacrylate (HAMA-APBA) was synthesized in two steps. First, methacrylic acid was grafted onto hyaluronic acid via esterification to obtain HAMA. Second, APBA was conjugated to HAMA through amidation to generate HAMA-APBA. Guided by controlled-release design principles for biomaterial-based EV delivery, CDEVs were mixed with thermosensitive HAMA-APBA to obtain HAMA-APBA/CDEVs. The boron atom in APBA forms reversible covalent boronate ester bonds with the cis-diol structures enriched on glycoproteins and glycolipids of EV membranes [23,29,30]. Dual staining with PKH-26 and FITC-PEG-PBA confirmed effective binding between CDEVs and the boronic groups while preserving cellular uptake capability (Fig. S14). Because boronate ester bonds are ROS-labile, high ROS induces bond cleavage, allowing CDEVs to detach from their chemical anchoring; concurrent ROS-accelerated hydrogel degradation synergistically further promotes responsive EV release at the lesion site (Fig. 4a).Fig. 4Synthesis and Characterization of HAMA-APBA/CDEVs. a) Illustration of the stepwise preparation of HAMA-APBA/CDEVs. b) Nuclear magnetic resonance hydrogen spectrum (1H NMR) of HAMA modified with APAB group or not. c) Morphology of hydrogels in different groups (HAMA, HAMA-APBA, HAMA-APBA/CDEVs). d) Viscosity test of hydrogels in different concentrations (Control, 2%, 4%, 6%). e) Representative tensile stress-strain curves of hydrogels in different groups at < 50% strain. f) The axial compressive cyclic stress-strain curves of hydrogels under 1000 loading-unloading cycles (1st, 5th, 10th,20th, 50th, 100th, 500th, and 1000th cycles) at 20% strain per time. g) Inner structure of hydrogels in different groups by Scanning Electron Microscope (SEM). h) BMSCs loaded with hydrogels in different groups and stained with calcein-AM (living cells, Green) and propidium iodide (dead cells, red). i) Degradation of hydrogels of different groups in the ROS environment. j) Release rates of exosomes from different scaffold groups. Synthesis and Characterization of HAMA-APBA/CDEVs. a) Illustration of the stepwise preparation of HAMA-APBA/CDEVs. b) Nuclear magnetic resonance hydrogen spectrum (1H NMR) of HAMA modified with APAB group or not. c) Morphology of hydrogels in different groups (HAMA, HAMA-APBA, HAMA-APBA/CDEVs). d) Viscosity test of hydrogels in different concentrations (Control, 2%, 4%, 6%). e) Representative tensile stress-strain curves of hydrogels in different groups at < 50% strain. f) The axial compressive cyclic stress-strain curves of hydrogels under 1000 loading-unloading cycles (1st, 5th, 10th,20th, 50th, 100th, 500th, and 1000th cycles) at 20% strain per time. g) Inner structure of hydrogels in different groups by Scanning Electron Microscope (SEM). h) BMSCs loaded with hydrogels in different groups and stained with calcein-AM (living cells, Green) and propidium iodide (dead cells, red). i) Degradation of hydrogels of different groups in the ROS environment. j) Release rates of exosomes from different scaffold groups. 1H NMR spectra confirmed successful HAMA modification (Fig. 4b), with the proton signal at 7.71 ppm corresponding to the phenyl ring of the APBA moiety. HAMA, HAMA-APBA, and HAMA-APBA/CDEVs formed transparent hydrogels with robust photocuring performance (Fig. 4c). Rheological assessment (shear rate: 10 s−1, 25 °C) revealed a concentration-dependent increase in viscosity (Fig. 4d): 0.5604 Pa s (2%), 1.416 Pa s (4%), and 7.457 Pa s (6%), all substantially higher than baseline control (0.00793 Pa s). Balancing syringe injectability and post-extrusion structural stability, the 4% formulation was selected for subsequent experiments. Uniaxial compression testing showed favorable compressive properties across all hydrogels (Fig. 4e; Fig. S15a–c), with a Young's modulus of 31.8 ± 2.3 kPa for 4% HAMA (n = 3). To simulate cyclic loading in the joint cavity, hydrogels underwent 1000 loading–unloading cycles at 20% strain (Fig. 4f; Fig. S15d). The compressive modulus decreased from 32.3 kPa (cycle 1) to 27.4 kPa (cycle 100) and 18.85 kPa (cycle 1000), indicating controlled modulus attenuation with preserved mechanical integrity and supporting the hydrogel's fatigue resistance and suitability for intra-articular application. SEM analysis of freeze-dried hydrogels revealed uniform porous structures (Fig. 4g). Pore diameters were 128.7 ± 22.97 μm (HAMA), 133.4 ± 23.06 μm (HAMA-APBA), and 114.4 ± 31.68 μm (HAMA-APBA/CDEVs), with no significant between-group differences, indicating that APBA modification and EV loading did not alter hydrogel microarchitecture. To evaluate biocompatibility, BMSCs were encapsulated within HAMA, HAMA-APBA, and HAMA-APBA/CDEVs hydrogels (Fig. 4h). Live/dead staining confirmed excellent biocompatibility in all groups. ROS-responsive degradation and release profiles were assessed by incubating hydrogels in Phosphate-Buffered Saline (PBS, 4%) or H2O2 solutions (200 μM, 400 μM). Hydrogels degraded markedly faster under ROS stimulation (Fig. 4i), with complete degradation by day 10 (200 μM) and day 7 (400 μM). Correspondingly, cumulative CDEVs release increased significantly in a dose-dependent manner (Fig. 4j), reaching 73.3% ± 6.6% at 200 μM H2O2 and 88.3% ± 3.8% at 400 μM. Furthermore, immunofluorescence staining revealed that intra-articular ROS levels after microfracture surgery peaked at day 7 and remained elevated through day 14, during which the HAMA-APBA hydrogel underwent endogenous ROS-responsive degradation and enabled on-demand CDEV release within this optimal therapeutic window (days 7–14), in contrast to the rapid clearance of free CDEVs or the slow sustained release from non-responsive HAMA hydrogel, as confirmed by in vivo live imaging and residual hydrogel mass measurement (Fig. S16). In Vitro ferroptosis model of BMSCs under high-iron conditions To recapitulate the iron-overloaded, pro-ferroptotic microenvironment observed in vivo, we exposed BMSCs to 100 μM FeCl3 (Fe), a concentration optimized using proliferative and cytotoxicity assays (Fig. 5a). Tert-butyl hydroperoxide (TBHP, 100 μM) served as a positive control for ferroptosis induction. FerroOrange staining confirmed increased intracellular iron in both TBHP- and Fe-treated cells, with Fe exposure causing more pronounced accumulation (Fig. 5b and c). Transmission electron microscopy (TEM) revealed classic ferroptotic mitochondrial features, including condensation, loss of cristae, and shrinkage (Fig. 5d). Immunofluorescence staining showed reduced GPX4 and FTH1 expression after TBHP and Fe treatment (Fig. 5e; Fig. S17), consistent with ferroptotic dysregulation. In contrast, the expression of COL1 increased progressively during ferroptosis induction, suggesting a concurrent fibrotic response (Fig. S18). JC-1 staining indicated collapse of the mitochondrial membrane potential (Fig. 5f–j), while BODIPY and DCFH-DA assays confirmed enhanced lipid peroxidation and ROS generation (Fig. 5g–k; Fig. S19). MitoTracker imaging further demonstrated marked mitochondrial fragmentation and loss of network integrity in the Fe and TBHP groups (Fig. 5h and i; Fig. S20). Collectively, these morphological, molecular, and functional readouts confirm the successful establishment of an in vitro high-iron ferroptosis model in BMSCs.Fig. 5Ferroptosis model constructed by high iron concentration in vitro. a) BMSCs activity under different iron concentration(0 μM,5 μM,25 μM,50 μM,75 μM,100 μM,150 μM,200 μM) detected by CCK8 assay. b-c) Ferro Orange staining of BMSCs in different groups(Control, TBHP:100 μM, Fe:100 μM)(b) and semi-quantitative visual analysis(c). d) Representative TEM images of mitochondria (blue arrows) in BMSCs treated by different groups. e) Immunofluorescence staining images of ferroptosis markers (GPX4, FTH1) of BMSCs. The nuclei were stained with DAPI, and the cytoskeletons were stained with TRITC Phalloidin. f) JC-1 staining images (DAPI-blue, JC-1 Aggregate-red, JC-1 Monomeric-green) of BMSCs in different groups. g) Bodipy staining images(DAPI-blue, Neutral-red, Oxidized-green) of BMSCs in different groups. h-i) Mito tracker red staining images of BMSCs in different groups (h) and quantitative analysis of mitochondrial morphology (i). j-k) Quantitative analysis of JC-1 staining (j) and Bodipy staining (k).Data are represented as mean ± SD. Statistical significance was determined by a two-sided Student's t-test (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001). Ferroptosis model constructed by high iron concentration in vitro. a) BMSCs activity under different iron concentration(0 μM,5 μM,25 μM,50 μM,75 μM,100 μM,150 μM,200 μM) detected by CCK8 assay. b-c) Ferro Orange staining of BMSCs in different groups(Control, TBHP:100 μM, Fe:100 μM)(b) and semi-quantitative visual analysis(c). d) Representative TEM images of mitochondria (blue arrows) in BMSCs treated by different groups. e) Immunofluorescence staining images of ferroptosis markers (GPX4, FTH1) of BMSCs. The nuclei were stained with DAPI, and the cytoskeletons were stained with TRITC Phalloidin. f) JC-1 staining images (DAPI-blue, JC-1 Aggregate-red, JC-1 Monomeric-green) of BMSCs in different groups. g) Bodipy staining images(DAPI-blue, Neutral-red, Oxidized-green) of BMSCs in different groups. h-i) Mito tracker red staining images of BMSCs in different groups (h) and quantitative analysis of mitochondrial morphology (i). j-k) Quantitative analysis of JC-1 staining (j) and Bodipy staining (k).Data are represented as mean ± SD. Statistical significance was determined by a two-sided Student's t-test (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001). CDEVs suppress ferroptosis and restore normal differentiation of BMSCs Our findings indicate that the iron-overloaded microenvironment generated after microfracture induces substantial iron accumulation in recruited BMSCs, leading to dysregulated iron metabolism and ferroptosis-associated lipid peroxidation. These pathological events impair BMSC function and disrupt chondrogenic commitment, ultimately biasing differentiation toward fibrocartilage (i.e., a mechanically inferior and physiologically abnormal phenotype) rather than hyaline cartilage. Therefore, counteracting ferroptosis in BMSCs and restoring a normal chondrogenic trajectory are critical for improving cartilage regeneration after microfracture. To evaluate the ability of CDEVs to antagonize BMSC ferroptosis, deferoxamine (DFO), a well-established ferroptosis inhibitor, was used as a positive control. DFO alleviates intracellular iron overload by chelating excess iron, thereby suppressing ferroptotic processes. BMSCs were subjected to Fe-induced ferroptosis and then treated with DFO or CDEVs, respectively. Transmission electron microscopy revealed that, compared with the Fe group, DFO-treated BMSCs showed partial restoration of mitochondrial cristae with lighter staining and a loosened ultrastructure, albeit with mild swelling. CDEVs-treated cells similarly exhibited clear recovery of mitochondrial structure (Fig. 6a). FerroOrange staining further confirmed that both DFO and CDEVs significantly reduced intracellular Fe2+ levels (Fig. 6b; Fig. S21). Consistently, immunofluorescence staining showed that ferroptosis-related markers FTH1 and GPX4 were markedly restored following DFO or CDEVs treatment (Fig. 6c; Fig. S22), indicating effective ferroptosis suppression. JC-1 and BODIPY staining further showed that CDEVs treatment restored mitochondrial membrane potential and reduced lipid peroxidation (Fig. 6d and e; Fig. S23). DCFH-DA staining additionally confirmed that CDEVs significantly reduced intracellular ROS levels in BMSCs (Fig. S24). MitoTracker Red staining demonstrated improved mitochondrial morphology after CDEVs treatment (Fig. S25a). Compared with the fragmented mitochondria in the Fe group, both DFO and CDEVs groups exhibited elongated mitochondrial networks with more uniform fluorescence intensity and preserved membrane potential. Quantitative mitochondrial morphology metrics all showed a consistent recovery, supporting the role of CDEVs in rescuing mitochondrial structural and functional integrity by suppressing ferroptosis (Fig. S25b), including mean area (0.2265 μm2; 0.575 μm2; 0.5293 μm2), mean branch length (0.5048 μm; 0.8692 μm; 0.7406 μm), mean perimeter (1.191 μm; 2.567 μm; 2.137 μm), and total branch length/mitochondrion (0.892 μm; 2.056 μm; 1.446 μm).Fig. 6Antagonistic effect of CDEVs on ferroptosis in vitro. a) Representative TEM images of mitochondria (blue arrows) in BMSCs treated by different groups(Fe:FeCl3-100μM; TBHP: FeCl3-100μM, DFO-10μM; CDEVs: FeCl3-100μM, CDEVs-25 μg/mL). b) Ferro Orange staining of BMSCs in different groups. c) Immunofluorescence staining images of ferroptosis markers (GPX4, FTH1) of BMSCs. The nuclei were stained with DAPI, and the cytoskeletons were stained with TRITC Phalloidin. d-e) JC-1 and Bodipy staining images(merged) of BMSCs in different groups (d) and their quantitative analysis (e). f) qRT-PCR analysis of ferroptosis markers (GPX4, FTH1), hyalinosis markers (COL2A1, ACAN) and fibrosis markers (COL1A1, TNC) (n = 4). g) Immunofluorescence and pathological staining of 3D cultured cartilage organoids in different groups: Immunofluorescence staining (DAPI-blue, Col2a1-green, Mmp13-red); Safranin O staining; Alcian blue staining and Immunohistochemical staining of Col1a1. Data are represented as mean ± SD. Statistical significance was determined by a two-sided Student's t-test. Antagonistic effect of CDEVs on ferroptosis in vitro. a) Representative TEM images of mitochondria (blue arrows) in BMSCs treated by different groups(Fe:FeCl3-100μM; TBHP: FeCl3-100μM, DFO-10μM; CDEVs: FeCl3-100μM, CDEVs-25 μg/mL). b) Ferro Orange staining of BMSCs in different groups. c) Immunofluorescence staining images of ferroptosis markers (GPX4, FTH1) of BMSCs. The nuclei were stained with DAPI, and the cytoskeletons were stained with TRITC Phalloidin. d-e) JC-1 and Bodipy staining images(merged) of BMSCs in different groups (d) and their quantitative analysis (e). f) qRT-PCR analysis of ferroptosis markers (GPX4, FTH1), hyalinosis markers (COL2A1, ACAN) and fibrosis markers (COL1A1, TNC) (n = 4). g) Immunofluorescence and pathological staining of 3D cultured cartilage organoids in different groups: Immunofluorescence staining (DAPI-blue, Col2a1-green, Mmp13-red); Safranin O staining; Alcian blue staining and Immunohistochemical staining of Col1a1. Data are represented as mean ± SD. Statistical significance was determined by a two-sided Student's t-test. RT-qPCR analysis of key ferroptosis- and chondrogenesis-related genes further supported these findings (Fig. 6f). The ferroptosis-related markers GPX4 and FTH1 were significantly downregulated in TBHP- and Fe-treated groups, but were restored after DFO or CDEVs treatment. Hyaline cartilage markers (COL2A1, ACAN) were markedly decreased under ferroptotic conditions but recovered following DFO or CDEVs treatment. Conversely, fibrocartilage-associated genes (COL1A1, TNC) were upregulated in ferroptotic cells but returned to baseline after CDEVs or DFO intervention. These results indicate that ferroptosis suppresses hyaline chondrogenesis while promoting fibrocartilaginous differentiation, whereas CDEVs reverse this pathological shift. Immunofluorescence staining of COL2A1, SOX9, and COL1A1 (Fig. S26) further corroborated the corrective effect of CDEVs on differentiation fate. To assess whether CDEVs rescue chondrogenic differentiation under iron-overloaded conditions, we performed a 3D chondrogenic organoid assay. After 21 days of induction, COL2A1/MMP13 immunofluorescence showed that CDEVs attenuated ferroptosis-associated matrix degradation and restored type II collagen deposition. Safranin O and Alcian Blue staining demonstrated recovery of extracellular matrix and glycosaminoglycan content following CDEVs treatment, while COL1A1 immunohistochemistry confirmed reduced fibrosis (Fig. 6g, Fig. S3c–f). Together, the organoid model supports that CDEVs suppress ferroptosis and restore the chondrogenic differentiation trajectory of BMSCs. Pvu-miR-159 in CDEVs inhibits ferroptosis by targeting the PTPN12–ERK1/2–ATF4–GPX4/SLC7A11 axis To elucidate the molecular mechanism by which CDEVs inhibit ferroptosis in iron-overloaded BMSCs, we performed mRNA transcriptome sequencing (n = 3) on Fe-induced ferroptosis cultures with or without CDEVs treatment. Principal component analysis showed clear separation between the two groups (Fig. 7a and b). Compared with the Fe-only group, CDEVs treatment upregulated 1560 genes and downregulated 1591 genes (Figs. S27 and 28). KEGG pathway enrichment analysis of differentially expressed genes (DEGs) (Fig. 7c) revealed significant enrichment of pathways associated with ferroptosis regulation, including the MAPK signaling pathway, ferroptosis, and glutathione metabolism. A volcano plot further showed that GPX4, a key antioxidant enzyme that blocks lipid peroxidation, was markedly upregulated after CDEVs treatment (Fig. 7d). GO enrichment analysis indicated that CDEVs modulated biological processes related to multicellular regulation, intercellular communication, ERK signaling, transcriptional regulation, and bioactive molecule biosynthesis (Fig. 7e). Gene set enrichment analysis (GSEA) confirmed that CDEVs counteracted ferroptosis primarily by restoring glutathione metabolism and regulating MAPK signaling (Fig. 7f).Fig. 7CDEVs reverse high Fe level-induced ferroptosis in BMSCs through Pvu-mir159 by regulating the MAPK pathway and reshape Glutathione metabolism. a-b) PLS-DA score plot (a) and Pearson correlation (b) between different groups (Fe-treated BMSCs ± CDEVs). c) KEGG enrichment of DEGs, which are mainly associated with MAPK, Ferroptosis, and Glutathione metabolism. d) Volcano maps of BMSCs in different groups. e) Analysis of functional clustering enrichment for positive regulation of multicellular organismal process, positive regulation of signal transduction, positive regulation of ERK cascade, and regulation of DNA-templated transcription between Fe and CDEVs groups. f) GSEA of the MAPK pathway, Ferroptosis, and Glutathione metabolism. g) PLS-DA score plot of differentially abundant metabolites in Fe-treated BMSCs ± CDEVs. (h) KEGG topology analysis, which is mainly associated with Linoleic acid metabolism, Alanine, aspartate, glutamate metabolism, and Glutathione metabolism. (i) Multi-dimensional enrichment circle map of metabolites. (j) Distribution rate of the top 20 miRNAs in CDEVs. k) Venn analysis of all predicted genes for Pvu-miR-159 potential targets. l) miRNA-mRNA interaction of Pvu-miR-159 and the following pathway network on gene sets of Fibrosis, Glutathione metabolism, MAPK pathway, and Ferroptosis constructed by Cytoscape software. CDEVs reverse high Fe level-induced ferroptosis in BMSCs through Pvu-mir159 by regulating the MAPK pathway and reshape Glutathione metabolism. a-b) PLS-DA score plot (a) and Pearson correlation (b) between different groups (Fe-treated BMSCs ± CDEVs). c) KEGG enrichment of DEGs, which are mainly associated with MAPK, Ferroptosis, and Glutathione metabolism. d) Volcano maps of BMSCs in different groups. e) Analysis of functional clustering enrichment for positive regulation of multicellular organismal process, positive regulation of signal transduction, positive regulation of ERK cascade, and regulation of DNA-templated transcription between Fe and CDEVs groups. f) GSEA of the MAPK pathway, Ferroptosis, and Glutathione metabolism. g) PLS-DA score plot of differentially abundant metabolites in Fe-treated BMSCs ± CDEVs. (h) KEGG topology analysis, which is mainly associated with Linoleic acid metabolism, Alanine, aspartate, glutamate metabolism, and Glutathione metabolism. (i) Multi-dimensional enrichment circle map of metabolites. (j) Distribution rate of the top 20 miRNAs in CDEVs. k) Venn analysis of all predicted genes for Pvu-miR-159 potential targets. l) miRNA-mRNA interaction of Pvu-miR-159 and the following pathway network on gene sets of Fibrosis, Glutathione metabolism, MAPK pathway, and Ferroptosis constructed by Cytoscape software. Given the critical roles of lipid peroxidation and glutathione imbalance in ferroptosis, we next performed metabolomic profiling of BMSCs treated with CDEVs (n = 6). Distinct clustering was observed (Fig. 7g), with 202 metabolites significantly upregulated and 271 downregulated (Fig. S29). KEGG topology analysis and metabolite classification (Fig. 7h and i) highlighted enrichment in linoleic acid metabolism, alanine/aspartate/glutamate metabolism, and glutathione metabolism, suggesting that CDEVs alleviate ferroptosis by reprogramming cellular metabolism, particularly glutathione homeostasis. These findings are consistent with the transcriptomic results and indicate metabolic remodeling as a mechanism underlying CDEVs-mediated ferroptosis suppression. Because miRNAs are key functional cargos of extracellular vesicles, we performed high-throughput miRNA sequencing on CDEVs. Among the highly expressed miRNAs, Pvu-miR-159 was the most abundant (44.1%) (Fig. 7j). Integration of five prediction platforms (TargetFinder, miRanda, miRDB, Bibiserv, psRobot) identified 19 high-confidence target genes (Fig. 7k). To pinpoint mechanistically relevant candidates, we intersected these predicted targets with DEGs from transcriptomic and metabolomic analyses related to fibrosis, ferroptosis, glutathione metabolism, and MAPK signaling. Multi-omics network analysis using GeneCards and STRING highlighted the PTPN12–ERK1/2–ATF4–GPX4/SLC7A11 axis as the most plausible pathway through which CDEVs exert their anti-ferroptotic activity (Fig. 7l). Mechanistic validation of Pvu-miR-159–mediated suppression of ferroptosis via PTPN12–ERK1/2–ATF4–GPX4 (SLC7A11) Based on integrated miRNA–mRNA network predictions, we hypothesized that PTPN12, a classical protein tyrosine phosphatase that dephosphorylates and inactivates downstream targets, is a direct target of Pvu-miR-159. Dual-luciferase reporter assays confirmed this interaction: Pvu-miR-159 significantly decreased luciferase activity of the wild-type PTPN12 3′UTR construct, but not the mutant construct, indicating sequence-specific binding (Fig. 8a and b). Western blotting showed that Pvu-miR-159–mediated suppression of PTPN12 increased ERK1/2 phosphorylation, activated ATF4, and upregulated GPX4 and SLC7A11,key ferroptosis-protective genes (Fig. 8c, d, Fig. S30). To validate the role of PTPN12 in this axis, we generated a PTPN12-knockdown BMSC model using siRNA. Loss of PTPN12 markedly elevated ERK1/2 phosphorylation (Fig. 8e and f). Under ferroptotic stress, ATF4, GPX4, and SLC7A11 expression were significantly reduced; notably, the protective effect of CDEVs was largely abolished in PTPN12-deficient cells, indicating that CDEVs exert anti-ferroptotic activity predominantly via PTPN12 repression. The RT-qPCR reveals the importance and necessity of Pvu-miR-159 as well (Fig. S31).Fig. 8Pvu-mir159 in CDEVs inhibits ferroptosis by binding to PTPN12 and regulating the downstream ERK-ATF4-GPX4 (SLC7A11) axis. a) Sequence of PTPN12 binding sites of wild type (WT) and mutant (Mut) Pvu-miR-159. b) Quantitative analysis of luciferase activity. c-d) Western blot analysis and quantitative analysis of PTPN12, P-ERK, ERK, ATF4, GPX4 and SLC7A11 in BMSCs transfected with Pvu-miR-159 mimics and inhibitor (MC: mimics-control, MM: mimics-micro, IC: inhibitor-control, IM: inhibitor-micro) (n = 4). e-f) Western blot analysis and quantitative analysis of PTPN12, P-ERK, ERK, ATF4, GPX4, and SLC7A11 in BMSCs which is transfected with si-PTPN12 or not (n = 4). g) Immunofluorescence staining images of PTPN12, GPX4, and SLC7A11, the nuclei were stained with DAPI, and the cytoskeletons were stained with phalloidin. h) Glutathione metabolism level identified by the concentration of GSH and GSSG in different groups. Data are represented as mean ± SD. Statistical significance was determined by a two-sided Student's t-test. Pvu-mir159 in CDEVs inhibits ferroptosis by binding to PTPN12 and regulating the downstream ERK-ATF4-GPX4 (SLC7A11) axis. a) Sequence of PTPN12 binding sites of wild type (WT) and mutant (Mut) Pvu-miR-159. b) Quantitative analysis of luciferase activity. c-d) Western blot analysis and quantitative analysis of PTPN12, P-ERK, ERK, ATF4, GPX4 and SLC7A11 in BMSCs transfected with Pvu-miR-159 mimics and inhibitor (MC: mimics-control, MM: mimics-micro, IC: inhibitor-control, IM: inhibitor-micro) (n = 4). e-f) Western blot analysis and quantitative analysis of PTPN12, P-ERK, ERK, ATF4, GPX4, and SLC7A11 in BMSCs which is transfected with si-PTPN12 or not (n = 4). g) Immunofluorescence staining images of PTPN12, GPX4, and SLC7A11, the nuclei were stained with DAPI, and the cytoskeletons were stained with phalloidin. h) Glutathione metabolism level identified by the concentration of GSH and GSSG in different groups. Data are represented as mean ± SD. Statistical significance was determined by a two-sided Student's t-test. Immunofluorescence staining corroborated these findings: CDEVs decreased PTPN12 expression while restoring SLC7A11 and GPX4 levels (Fig. 8g). Glutathione quantification further revealed that ferroptosis induction caused marked accumulation of oxidized glutathione (GSSG) and disruption of the GSH/GSSG redox ratio, which was reversed by CDEVs (Fig. 8h). These data support that PTPN12–ERK-ATF4 signaling is essential for restoring glutathione-dependent antioxidant defenses in iron-overloaded BMSCs. Collectively, these results identify Pvu-miR-159 as a key functional miRNA in CDEVs that suppresses ferroptosis by targeting PTPN12 and activating the ERK1/2–ATF4–GPX4/SLC7A11 cytoprotective axis. Furthermore, through co-culture experiments, RT-qPCR, multi-platform in silico target prediction, and integrated transcriptomic and metabolomic analyses, we confirmed that Pvu-miR-159 is efficiently delivered into BMSCs via CDEVs (peaking at 6-12 h) and acts primarily by targeting PTPN12 with high affinity (−21.9 kcal/mol), thereby modulating the transcription of the downstream axis, without significant off-target effects (Fig. S32). In Vivo Evaluation of the therapeutic effect of CDEVs on cartilage regeneration after microfracture In a full-thickness femoral trochlear cartilage defect model in Sprague–Dawley (SD) rats, we evaluated four treatment regimens (i.e., microfracture alone [MF], HAMA-APBA hydrogel, CDEVs, and the composite HAMA-APBA/CDEVs hydrogel) to determine their efficacy in improving cartilage regeneration following microfracture (Fig. 9a). We next examined the retention of DiD-labeled CDEVs within the articular cartilage defect site across three delivery groups (CDEVs injection, HAMA/CDEVs hydrogel, HAMA-APBA/CDEVs hydrogel) at various time points (0, 3, 7, 10, and 14 days) using an in vivo fluorescence imaging system. HAMA/CDEVs primarily exhibit sustained passive release of CDEVs, whereas HAMA-APBA/CDEVs enable ROS-triggered release via the dynamic phenylboronate ester bond. Both hydrogel groups showed greater stability and more sustained fluorescence signals than direct CDEVs injection. Notably, the HAMA-APBA/CDEVs hydrogel exhibited a more pronounced ROS-responsive release profile than the HAMA/CDEVs hydrogel (Fig. S16).Fig. 9HAMA-APBA/CDEVs facilitated the original differentiation of regenerated cartilage after microfracture surgery. a) Schematic diagram of the rat microfracture surgery model and treatment with HAMA-APBA/CDEVs. b) Macroscopic observation of cartilage defects with different implants at 6 and 12 weeks post-operation. c) Micro-CT images showing 2D and 3D reconstruction of the repaired cartilage at 6 and 12 weeks postoperation. d) H&E staining, e) Safranin O/Fast Green staining, and f) Type II collagen immunohistochemical staining of the repaired tissue from different groups at 6 and 12 weeks postoperation. g) BMD and BV/TV in Micro-CT 3D reconstruction, h) ICRS histological total score and Mankin histological total score, i) O'Driscoll histological total score and the mean optical density of type II collagen immunohistochemical staining. The data are presented as the mean ± SD (n = 8). HAMA-APBA/CDEVs facilitated the original differentiation of regenerated cartilage after microfracture surgery. a) Schematic diagram of the rat microfracture surgery model and treatment with HAMA-APBA/CDEVs. b) Macroscopic observation of cartilage defects with different implants at 6 and 12 weeks post-operation. c) Micro-CT images showing 2D and 3D reconstruction of the repaired cartilage at 6 and 12 weeks postoperation. d) H&E staining, e) Safranin O/Fast Green staining, and f) Type II collagen immunohistochemical staining of the repaired tissue from different groups at 6 and 12 weeks postoperation. g) BMD and BV/TV in Micro-CT 3D reconstruction, h) ICRS histological total score and Mankin histological total score, i) O'Driscoll histological total score and the mean optical density of type II collagen immunohistochemical staining. The data are presented as the mean ± SD (n = 8). At 6 weeks postoperatively, the Control group exhibited severe loss of cartilage matrix within the defect region, forming an irregularly margined cavity. In the MF group, although granulation tissue partially filled the defect, the surface remained rough and was predominantly fibrous. Both the HAMA-APBA hydrogel group and the HAMA-APBA/CDEVs composite group showed more apparent reparative tendencies; however, newly formed tissue still displayed a clear boundary with the host cartilage, and surface smoothness remained suboptimal. By 12 weeks, partial fibrous scarring tissue was observed within the defect in the MF group, but distinct fissures between tissues and poor integration with surrounding cartilage persisted. The HAMA-APBA and CDEVs groups showed further coverage of the defect by cartilage-like tissue; however, regeneration remained predominantly fibrocartilaginous, with a clear demarcation from the adjacent normal cartilage and limited integration. Notably, the HAMA-APBA/CDEVs group displayed regenerated tissue closely resembling hyaline cartilage at this time point. The defect was almost completely repaired, with a smooth surface comparable to adjacent native cartilage, and the newly formed tissue integrated seamlessly with the host cartilage, with blurred boundaries indicative of superior morphological repair and integration (Fig. 9b). Reconstructed 2D and 3D micro-CT images showed that at 6 weeks, both the MF and HAMA-APBA groups exhibited subchondral bone defects of varying severity, with the MF group showing the largest defects. Bone filling was improved in the CDEVs group, although boundaries with surrounding bone remained apparent. These pathological features were substantially ameliorated in the HAMA-APBA/CDEVs group. At 12 weeks, the MF group continued to exhibit pronounced osseous defects, whereas the HAMA-APBA group showed markedly increased bone filling compared with 6 weeks. Both the CDEVs and HAMA-APBA/CDEVs groups demonstrated robust bone regeneration and seamless integration with surrounding tissue, without visible boundaries (Fig. 9c). Furthermore, the HAMA-APBA/CDEVs group provides better strength than the MF group (Fig. S33). Histological analyses using H&E, Safranin-O/Fast Green, and type II collagen (COL2) immunohistochemistry showed minimal cartilage tissue filling in the MF group, with little to no COL2 production or deposition, indicating poor cartilage regeneration. At 12 weeks, the HAMA-APBA and CDEVs groups showed partial tissue filling, with tissue thickness approaching that of normal cartilage and a relatively flattened surface; however, integration remained poor, COL2 deposition was limited, and Safranin-O staining was weak. Most samples exhibited pronounced fibrosis with minimal type II collagen. In contrast, the HAMA-APBA/CDEVs group exhibited a smoother surface and thicker regenerated tissue, with uniform and intense Safranin-O staining and abundant COL2 deposition, indicating superior cartilage repair quality (Fig. 9d, e, f, Fig. S3g–h). To assess biosafety, H&E staining of major organs (heart, liver, spleen, lung, kidney) at 12 weeks revealed no significant histopathological abnormalities, supporting excellent biocompatibility of the nanomaterial system (Fig. S34). According to International Cartilage Repair Society (ICRS) macroscopic scoring, the CDEVs group achieved markedly improved repair compared with the MF and HAMA-APBA hydrogel groups at corresponding time points, and the HAMA-APBA/CDEVs group showed further improvement at 12 weeks relative to CDEVs alone. Quantitative analysis of bone volume/tissue volume (BV/TV) and bone mineral density (BMD) showed significant increases in the HAMA-APBA/CDEVs group at both 6 and 12 weeks, indicating more favorable subchondral bone reconstruction. Histological evaluation using Mankin scoring and the modified O'Driscoll scoring demonstrated lower (i.e., better) scores in the HAMA-APBA/CDEVs group, with more complete surface architecture, deeper matrix staining, and more orderly cellular organization. Semi-quantitative analysis of COL2 immunostaining further confirmed that COL2 deposition in the HAMA-APBA/CDEVs group at both 6 and 12 weeks was the most pronounced and closely resembled normal cartilage, supporting the conclusion that CDEVs loaded ROS-responsive hydrogels promote a hyaline cartilage phenotype and enhance cartilage regeneration after microfracture (Fig. 9g–i). Discussion In this study, we sought to elucidate the mechanisms underlying aberrant chondral differentiation during cartilage regeneration following microfracture. We first established a rat in vivo microfracture model, in which regenerated tissue within the defect exhibited disorganized collagen fiber alignment and a fibrotic scar–like repair pattern (Fig. 1b). Notably, abundant BMSCs accumulated within the defect region (Fig. S1). We then performed scRNA-seq across multiple postoperative time points, and, together with pseudotime trajectory analysis, constructed the differentiation landscape of endogenous BMSCs. This analysis revealed two major differentiation fates of BMSCs after microfracture: an aberrant fibrocartilage lineage and a normal hyaline cartilage lineage, and enabled precise localization of the critical nodes governing lineage bifurcation (Fig. 2a). Pathway enrichment analysis of trajectory-associated nodal genes highlighted ferroptosis, a form of programmed cell death driven by iron-dependent lipid peroxidation and modulated by intracellular antioxidant systems (e.g., GPX4), as a key regulatory mechanism influencing BMSC fate determination. In light of local pathophysiological changes after microfracture, this finding is biologically plausible. Although microfracture promotes stem-cell migration, it also profoundly alters the intra-articular microenvironment. Surgical bleeding introduces blood components into the defect site, where hemolysis and hemoglobin degradation release large amounts of iron ions, creating an iron-overload pathological milieu. While iron is essential for cellular metabolism, excess iron catalyzes ROS generation via the Fenton reaction [27,31], triggering oxidative stress that damages cellular structures, disrupts signaling pathways, and impairs the chondrogenic differentiation of BMSCs. Dynamic monitoring of tissue iron concentrations revealed, for the first time, a pronounced postoperative increase in local iron levels that was positively correlated with the progression of cartilage repair failure(Fig. 2k). This finding clarifies, at the pathophysiological level, how an iron-overload microenvironment develops and provides direct experimental evidence supporting an iron-dependent basis for ferroptosis. Collectively, these results establish ferroptosis as a central mechanism underlying impaired cartilage regeneration observed after microfracture. Plant-derived bioactive compounds, owing to their antioxidant, metabolic regulatory, and tissue-repair properties, have long been explored for regenerative applications. Curcuma plants contain abundant bioactive constituents with demonstrated efficacy in counteracting ferroptosis in cardiomyocytes and neurons [32,33]. Plant-derived extracellular vesicles are an emerging therapeutic nanomedicine that harbors diverse bioactive cargos, including plant metabolites, proteins, microRNAs, and lipids, and exhibits multiple biological activities, such as anti-inflammatory, antioxidant, and antitumor effects [34,35]. In our study, integrated transcriptomic and metabolomic analyses showed that CDEVs modulate the downstream PTPN12–ERK1/2–ATF4–GPX4 (SLC7A11) signaling axis via their key cargo Pvu-miR-159 (Fig. 7j), thereby restoring endogenous antioxidant defense, reprogramming glutathione metabolism, suppressing iron overload-induced ferroptosis, and reinstating the hyaline chondrogenic differentiation trajectory of BMSCs to improve the quality of regenerated cartilage. Furthermore, nanoparticle tracking and component analysis confirmed that the peak particle diameters of CDEVs extracted from different batches, the particle concentrations obtained, and the levels of representative functional Pvu-miR-159 were comparable across independent batches. Together, these findings provide evidence for the cross-species regulatory potential of plant-derived EVs. Immunogenicity of plant-derived vesicles in humans. As xenogeneic bioactive substances, the immunogenicity of plant-derived exosome-like nanoparticles in mammals is a primary safety concern for clinical translation. To evaluate CDEV immunogenicity, we systematically assessed the effects of CDEVs on major organs in our rat in vivo toxicity experiments. H&E staining of the heart, liver, spleen, lungs, and kidneys revealed no obvious inflammatory cell infiltration, tissue structural disruption, or necrotic changes, indicating that CDEVs did not induce significant acute immunotoxicity or organ damage (Figs. S34 and S13). We acknowledge that inter-individual genetic variability may lead to heterogeneous immune responses, and future preclinical studies using more human-relevant animal models (e.g., non-human primates) will be necessary to further validate immunogenicity. Furhermore, the loading and release process during HAMA-APBA hydrogel fabrication did not compromise the structural integrity, cellular uptake, or anti-ferroptosis bioactivity of CDEVs (Fig. S35). However, the long-term immunogenicity, potential adverse responses to repeated dosing, and risks associated with systemic exposure remain to be systematically evaluated in future preclinical and clinical studies. Protein tyrosine phosphatases (PTPs), classical regulators of the MAPK pathway, function by dephosphorylating downstream targets and suppressing their activity. Prior studies have reported elevated expression of PTPN family members in degenerative chondrocytes in OA models [36]. ERK1/2, a central MAPK node inhibited by PTPN12 [37], regulates ATF4, a transcription factor identified by our SCENIC regulon analysis as essential for specifying the hyaline cartilage phenotype. The increased ATF4 expression observed after CDEVs treatment is consistent with restored hyaline chondrogenic differentiation (Fig. 2e). The GPX4–glutathione (GSH) axis and System Xc-mediated cystine uptake represent two major intracellular defense mechanisms against ferroptosis. GPX4 is the only enzyme capable of directly reducing phospholipid hydroperoxides to alcohols, thereby terminating lipid peroxidation chain reactions [38,39]. System Xc-imports cystine, which is rapidly reduced to cysteine for GSH synthesis, the essential cofactor for GPX4 [40]. Extensive evidence indicates that ATF4 directly binds to the SLC7A11 promoter to activate its transcription [[41], [42], [43]] and regulates both GSH synthesis and GPX4 stabilization [39,44,45], thereby restoring cellular redox balance and protecting cells from ferroptotic injury. As a widely applied biomaterial, hydrogels have been increasingly used in cartilage regeneration. In the iron-overload microenvironment after microfracture, excess iron ions and ROS create conditions that favor ferroptosis [46,47]. BMSCs entering the defect zone are therefore highly susceptible to ferroptotic injury, which can promote aberrant differentiation and compromise regenerative capacity. Using histological staining and qPCR, we found that ferroptosis-antagonizing markers increased during days 0–14 postoperatively and then declined, whereas ROS levels rose during days 0–7 and declined thereafter (Fig. 2l). These dynamics indicate a critical “therapeutic window.” Leveraging this feature, we developed an injectable, tissue-adhesive hydrogel (HAMA-APBA/CDEVs) by combining CDEVs with 3-aminophenylboronic acid-modified hyaluronic acid methacrylate (HAMA-APBA) [48,49]. Boronate ester-based crosslinking enables high EV loading, injectability, shape adaptability, and photocrosslinkable mechanical integrity. Importantly, HAMA-APBA/CDEVs exhibit ROS-responsive degradation and enables sustained release of CDEVs, better than the non-ROS-responsive hydrogel HAMA/CDEVs (Fig. S36). Around postoperative day 7, when ROS levels peak, the hydrogel degrades responsively and releases CDEVs into the iron-overloaded microenvironment, enabling intervention before ferroptotic decompensation (day 14). Regarding the translational potential of the HAMA-APBA/CDEVs hydrogel, we propose strategies to enhance scalability and clinical applicability by focusing on (1) large-scale production of high-quality CDEVs, and (2) optimization of the efficacy, safety, and patient compliance associated with ROS-responsive HAMA-APBA/CDEVs hydrogel administration. Specifically, EVs derived from medicinal plants such as Curcuma longa represent a promising therapeutic modality because of their abundant sources, high yield, and favorable biocompatibility. In addition, owing to their injectability, biocompatibility, and biodegradability, hydrogels can be delivered via direct intraoperative injection at the surgical site, simplifying the treatment procedure and improving patient compliance. We anticipate that these advances will facilitate large-scale manufacturing and clinical translation of bioactive hydrogels, thereby supporting improved cartilage regeneration after microfracture surgery. Conclusion This study systematically elucidates the core mechanism underlying impaired cartilage regeneration following microfracture surgery and proposes a plant extracellular vesicle-based targeted therapeutic strategy. Using a rat microfracture model, we delineated, for the first time at single-cell resolution, the fate landscape of endogenous BMSCs and identified critical bifurcation nodes governing differentiation into either normal hyaline cartilage or aberrant fibrocartilage lineages. Through integrated pathophysiological assessment and dynamic monitoring of local tissue iron levels, we provide evidence that postoperative iron overload is a key pathological driver of ferroptosis in BMSCs, thereby promoting aberrant commitment toward a fibrotic phenotype. Building on these mechanistic insights, we developed a ROS-responsive CDEVs-loaded hydrogel delivery system (HAMA-APBA/CDEVs). This platform is designed to release CDEVs during the postoperative ROS peak (approximately day 7), enabling intervention before ferroptotic decompensation (before day 14). Mechanistic studies show that CDEVs, through their key cargo Pvu-miR-159, modulate the PTPN12–ERK1/2–ATF4 signaling axis and increase the expression of antioxidant genes such as GPX4 and SLC7A11. This reprograms glutathione metabolism and restores the antioxidant defenses, thereby suppressing ferroptosis and directing BMSCs toward a hyaline chondrogenic differentiation trajectory. Histological and functional evaluations confirmed substantial structural repair and biomechanical restoration, supporting the therapeutic efficacy of HAMA-APBA/CDEVs. By integrating the metabolic regulatory properties of plant-derived nanovesicles with a stimulus-responsive hydrogel platform, HAMA-APBA/CDEVs address key barriers to cartilage regeneration after microfracture, including limitations in small-molecule pharmacokinetics, challenges in targeted delivery, and persistent metabolic dysregulation. This work establishes a plant-inspired, biomaterials-based paradigm for metabolic intervention in regenerative medicine. Future studies should evaluate chronic degenerative models and downstream molecular cascades to accelerate the clinical translation of this targeted metabolic rehabilitation strategy for correcting aberrant BMSC differentiation after microfracture surgery. Experiment section Establishment of the Rat Microfracture Model. Fifty male Sprague–Dawley rats (220–240 g) were randomly assigned to a sham-operated group (n = 10) and a microfracture (MF) defect group (n = 40; n = 10 per time point). Briefly, rats were anesthetized, the knee joint was exposed under sterile conditions, and the patella was identified as the anatomical landmark. A medial parapatellar incision was made, the patella was dislocated laterally, and the trochlear groove was exposed. A ∼2-mm diameter ring drill was used to remove the superficial cartilage layer. A sterile micro–Kirschner wire was inserted approximately 1 mm into the subchondral bone until bone marrow exudation was visible. The joint capsule and skin were closed with absorbable sutures. Samples were harvested at 3, 7, 14, and 28 days postoperatively following euthanasia. Single-Cell RNA Sequencing (scRNA-seq). Microfracture models were established in SD rats (n = 10 per group) with four sampling time points: control, postoperative day 7, day 14, and day 28. Tissue digestion was conducted under sterile conditions using 2 mg/mL collagenase II + 0.1 mg/mL DNase I at 37 °C for 1.5–2 h, followed by filtration through a 40-μm strainer. Cell viability (>90%) was verified using trypan blue. Libraries were constructed using the 10x Genomics Chromium X platform (5000–10,000 cells captured per sample, Agilent 2100 QC: 200–300 bp fragments) and sequenced on an Illumina NovaSeq 6000 (PE150, ≥50,000 reads/cell). Data processing was performed using Seurat 4.0 (filtering criteria: mitochondrial genes <15%, UMI counts 200–5000), Harmony for batch correction, UMAP visualization (resolution = 0.6), DEGs identified using DESeq2 (|log2FC| > 1, adjusted p < 0.05), and functional enrichment by GO/KEGG analysis. Data matrices were uploaded to the Majorbio Cloud Platform (cloud.majorbio.com) for downstream analyses. Histological and Immunohistochemical Staining. Samples collected at 3, 7, 14, and 28 days were fixed in formalin for 1 week, decalcified in EDTA for 2 months, dehydrated, embedded in paraffin, and sectioned at 6 μm using a Leica SM2000R microtome. Safranin-O and Fast Green staining (Solarbio, G1371) was performed per manufacturer's instructions. For immunohistochemistry, sections were deparaffinized, treated with hydrogen peroxide for 30 min to quench endogenous peroxidase, subjected to pepsin antigen retrieval (30 min), blocked for 30 min, and incubated with primary antibody GPX4 (1:200, Signalway Antibody, SAB 32506, USA) overnight at 4 °C. Secondary antibody incubation and DAB visualization were subsequently performed. Isolation of Rat BMSCs. Adult SD rats were euthanized, and bilateral femurs and tibias were harvested. After removing attached soft tissues, epiphyses were excised under aseptic conditions, and the marrow cavity was flushed with α-MEM (Gibco, C12571500BT). Cell suspensions were filtered and seeded at appropriate densities in complete medium with10% FBS (Oricell, FBSST-01033, from Cyagen Biosciences (Guangzhou) Inc.) and 1% penicillin-streptomycin (Beyotime, C0222). After 24–48 h, non-adherent hematopoietic cells were removed, and the medium was refreshed until spindle-shaped adherent BMSCs were observed on the 25 cm2 cell culture flask(H805002, BDBIO, BaiDi Biotechnology Co., Ltd.). RT-qPCR of Cartilage Tissue and Cultured Cells. Fresh regenerated cartilage was snap-frozen in liquid nitrogen and ground. Cultured cells were collected after digestion. Total RNA was extracted using FastPure RNA Kit (Vazyme, RC112-01). cDNA synthesis was performed with HiScript II RT SuperMix (Vazyme, R32301). qPCR was conducted using Taq Pro Universal SYBR Master Mix (Vazyme) on a QuantStudio™ 3 system. GAPDH served as the internal reference. Relative expression was calculated using the 2−ΔΔCt method. Primer sequences are listed in Table S1. Measurement of Tissue Iron Content. Iron concentrations in regenerated cartilage tissues were quantified using the Total Iron Colorimetric Assay Kit (E-BC-K772-M, Elabscience Biotechnology Co.,Ltd.) following manufacturer instructions. Isolation of CDEVs. CDEVs were isolated via gel-filtration chromatography. Fresh turmeric rhizomes were homogenized, centrifuged at 5000×g for 10 min and 12,000×g for 10 min at 4 °C, filtered (0.45 μm), and concentrated using 100-kDa ultrafiltration tubes (Pall MAP100C3). Final purification was performed using EV extraction columns (Aidimai, EVScore-05). The eluate (1 mL) contained enriched EVs. EV morphology was examined by TEM (JEOL JEM-2100LPUS) after staining with 2% phosphotungstic acid. Size distribution was determined by NTA (Particle Metrix PMX-120). Protein concentration was measured via BCA assay (Beyotime, P0012S). For uptake assays, EVs were labeled with PKH26 (Sigma). Metabolomic profiling of CDEVs was conducted using LC-MS (UHPLC-Q Exactive HF-X). Data processing (baseline correction, peak alignment, RT correction) and metabolite annotation were performed using Progenesis QI and matched against HMDB and Metlin databases. Data matrices were uploaded to the Majorbio Cloud Platform for extended analysis. Determination of Optimal Fe and CDEVs Concentrations. CCK-8 assays (HY-K0301, MedChemExpress, USA) were performed to determine appropriate Fe and CDEVs doses. BMSCs (2 × 103 cells/well) were stimulated for 24 h, incubated with CCK-8 for 2 h, and the absorbance was measured to assess cell viability. Intracellular Iron Detection. BMSCs were seeded into 24-well plates (1 × 104 cells/well) and treated with Control, TBHP (100 μM), Fe (100 μM), DFO (10 μM), or CDEVs (25 μg/mL). FerroOrange (1 μM; Dojindo, F374) was used to detect intracellular Fe2+ following 30-min incubation at 37 °C. Fluorescence images were captured using a Nikon fluorescence microscope. Transmission Electron Microscopy (TEM). Cells fixed in 2.5% glutaraldehyde for 24 h were dehydrated through ethanol gradients, embedded in resin (SPI, 90529-77-4), polymerized at 65 °C, sectioned at 70 nm (Leica UC7), stained with uranyl acetate and lead citrate, and imaged using TEM (Hitachi HT7800) to assess mitochondrial morphology. Cellular Immunofluorescence. Cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, blocked (Beyotime P0102), incubated with primary antibodies overnight, and with fluorophore-conjugated secondary antibodies for 1 h. Nuclei were stained with DAPI; F-actin with phalloidin. Fluorescence intensity and colocalization were analyzed using Fiji. Primary antibodies included GPX4 (SAB, 32506), FTH1(SAB, 32180-1), COL1A1(SAB, 28307), COL2A1 (abcom, ab34712), MMP13(SAB, AP74527), SOX9(SAB, 43125), SLC7A11(abcom, ab307610), and PTPN12(abcom, ab289859). Mitochondrial Membrane Potential (JC-1 Staining). Cells were incubated with JC-1 probe (10 μg/mL; Abbkine) for 30 min, counterstained with Hoechst, and imaged by fluorescence microscopy. Green/red fluorescence ratios were quantified using Fiji. Lipid Peroxidation Detection (BODIPY C11). Cells were incubated with BODIPY 581/591C11 (10 μM, Thermo Fisher) for 30 min, stained with Hoechst, and imaged. Green/red ratios were used to assess lipid peroxidation. MitoTracker Red Staining. Cells were labeled with Mito tracker Red CMXRos (100 nM, 15 min, M19979, Abmole, TX, USA), counterstained with Hoechst, and imaged by confocal microscopy (ZEISS). Mitochondrial morphology was analyzed via Fiji mitochondrial plugins. Intracellular ROS Detection. Cells were incubated with DCFH-DA probe (1:1000, Beyotime) at 37 °C for 30 min, washed, and imaged under a Nikon fluorescence microscope. EV Uptake Assay. PKH26-labeled EVs were incubated with BMSCs for 24 h. Cells were fixed and stained with FITC-phalloidin (cytoskeleton) and DAPI (nuclei), and imaged using confocal microscopy. EV Biocompatibility Assay. Live/dead staining was performed at 0, 24, and 72 h after stimulation. Fluorescence images were acquired using confocal microscopy. Scratch Wound Healing Assay. BMSCs were seeded into 6-well plates (Jet Biofil) and cultured until reaching over 90% confluence. Then a linear scratch was created using a 1-mL pipette tip. Then cells were washed with PBS, and treated with serum-free medium, Fe (100 μM), or Fe + CDEVs (25 μg/mL). Images were taken at 0, 12, and 24 h. Migration rate (%) was calculated using Fiji. EdU Proliferation Assay. Cell proliferation was assessed using an EdU assay kit (Alexa Fluor 594, Beyotime). EdU-positive cells were imaged and quantified. Transwell Migration Assay. BMSCs (5 × 104 per chamber) in serum-free medium were seeded into Transwell chambers (8 μm pore size, Jet Biofil) positioned in 24-well plates. Lower chambers contained medium per treatment group. After 24 h, migrated cells were fixed, stained with crystal violet (Solarbio, C8470, BejingSolarbio Science & Technology Co., Ltd), and counted from three random fields. In Vitro 3D Chondrogenic Organoid Culture. BMSCs (2 × 105 cells/well) were seeded in 96-well plates covered with OrganoGel (10 μl, abs9495, absin) and induced with DMEM + CHIR (10 μM), LDN (250 nM), and A8301 (1 μM). Media were replaced every 2 days. On day 9, cells were switched to chondrogenic medium by 5% FBS, TGF-β3 (10 ng/mL, CA59, Novoprotein, Shanghai, China) and BMP (20 ng/mL). Organoids were harvested on day 21. Groups included Fe (100 μM) and Fe + CDEVs (25 μg/mL). Transcriptomic Analysis. Total RNA (1 μg/sample) was extracted using Vazyme R401. Libraries were prepared following Illumina protocols and sequenced on NovaSeq X Plus. Samples meeting the criteria OD260/280 = 1.8–2.2, OD260/230 ≥ 2.0, RIN ≥6.5, 28S:18S ≥ 1.0 were used. DEGs were identified using DESeq2 (|log2FC| ≥ 1, FDR ≤0.05) and DEGseq (FDR ≤0.001). Enrichment analyses were performed using Goatools and KOBAS (p ≤ 0.05). KEGG visualization was conducted using OmicShare tools. Metabolomic Analysis. UHPLC–FTMS (UHPLC-Q Exactive HF-X, Thermo) with an HSS T3 column (100 mm × 2.1 mm, 1.8 μm) was used. Following extraction in pre-cooled methanol/water, samples were separated and analyzed in positive and negative ion modes. PCA and OPLS-DA identified differential metabolites. KEGG pathway enrichment elucidated biological relevance. miRNA Sequencing of CDEVs. Purified CDEVs were sent to Majorbio (Shanghai, China) for small RNA sequencing and analysis. Dual-Luciferase Reporter Assay. Bioinformatic prediction identified putative binding sites between Pvu-miR-159 and PTPN12. Wild-type (WT) and mutant (MUT) PTPN12 3′UTR luciferase reporters were co-transfected with Pvu-miR-159 mimics or negative controls into HEK293T cells (C5004, BDBIO, BaiDi Biotechnology Co., Ltd.). After 48 h, firefly and Renilla luciferase activities were measured (Beyotime). Western Blotting. Cells were lysed in RIPA buffer containing PMSF. Protein concentration was determined by the BCA assay. Equal amounts of protein were subjected to SDS-PAGE, transferred to PVDF membranes (KGC4805-5, KeygenBioTECH), blocked in 5% milk, and incubated with primary antibodies overnight at 4 °C before secondary incubation. Bands were visualized using ECL reagents. Primary antibodies: PTPN12, ERK1/2, p-ERK1/2, ATF4, GPX4, SLC7A11, GAPDH. Glutathione Metabolism Assays. GSH and GSSG levels were quantified using Beyotime S0054 kits to assess metabolic regulation across treatment groups. Preparation of HAMA-APBA Hydrogels. 2-Aminophenylboronic acid (43 mg) was added to 20 mL of HA solution (5 mg/mL), followed by the addition of EDC (48 mg) and HOBt (34 mg) in DMSO. After 48 h reaction at RT, the mixture was dialyzed (MWCO 8 kDa, 3 days) and lyophilized to obtain HA-PBA. Functional group modification was verified by 1H NMR (Bruker AVANCE III 600M). Hydrogel morphology and pore structure were analyzed via SEM (Hitachi S-4800). Then measure the appropriate amount and dissolve it with LAP (4%, EFL-LAP, Yongqinquan Intelligent EquipmentCo.,Ltd,Suzhou, China) before using. Mechanical Characterization of Hydrogels. Viscosity was measured using a rheometer (DHR-2, TA Instruments) at shear rate = 10 s−1, 25 °C. Compression properties were tested using a mechanical tester (Zwick Z0.5, strain = 50%). Fatigue resistance was assessed via DMA Q800 (strain = 20%). Hydrogel Biocompatibility. BMSCs were seeded onto hydrogels and cultured for 7 days. Live/dead staining was performed and imaged via confocal microscopy. Hydrogel Degradation and CDEVs Release. Hydrogels were incubated in PBS, 200 μM H2O2, or 400 μM H2O2. Mass loss was measured every 24 h. CDEVs’ release was quantified using the BCA assay. In Vivo Evaluation of HAMA-APBA/CDEVs. Thirty male rats (6–8 weeks, 250–300 g) were randomly divided into four groups (MF, HAMA-APBA, CDEVs and HAMA-APBA/CDEVs). Macroscopic cartilage assessment was performed (Table S2) Micro-CT analysis. The femoral samples were scanned using the General Electric (GE) Explorer Locus SP system (GE, Boston, MA, USA). Subsequently, we performed three-dimensional reconstruction of the images and calculated the bone mineral density (BMD) and bone volume/tissue volume (BV/TV). At 6 weeks and 12weeks post-surgery, all the rats were euthanized, and their femurs were harvested. Micro-computed tomography (micro-CT; IMAGING100pro; Raycision Medical Technology Co., Ltd., China) was performed to assess bone formation at the regenerated sites. This part was processed under these scanparameters: Raycision Imaging 100 scanner, 65 kV/180 A, 360° rotation with 4° step, 800x800x1000 px. A cylindrical area of subchondral bone directly beneath the cartilage defect, with a diameter of 2 mm and a height of Imm, was selected as ROI. Reconstruction via Imaging100, 3D quantification of bone volume fraction (BV/TV), separation (Tb.Sp) and number (Tb.N)using adaptive thresholding. Histological and Immunohistochemical Analysis (6 and 12 Weeks). Samples were fixed for 1 week, decalcified for 2 months, paraffin-embedded, sectioned at a 6 μm thickness using a microtome (2065A, Dakewe, China), and stained with H&E (Solarbio G1120) or Safranin-O/Fast Green. IHC was performed as described above. Scoring details are provided in Tables S3–S4. Statistical Analysis: Data were processed with GraphPad Prism 9.0 software (San Diego, CA) and expressed as mean ± standard deviation (SD) from at least three independent replicates. Intergroup comparisons between two groups were analyzed by unpaired Student's t-test, whereas multiple-group comparisons were performed via one-way ANOVA followed by Tukey's post-hoc test. Statistical significance was evaluated using Prism, with asterisks indicating significance levels: ∗p < 0.05, ∗∗p < 0.01 and ∗∗∗p < 0.001. CRediT authorship contribution statement Yuhao Mu: Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Conceptualization. Hao Li: Writing – review & editing, Validation, Project administration, Investigation, Conceptualization. Ruiyang Zhang: Supervision, Resources, Methodology, Investigation. Runmeng Li: Writing – review & editing, Supervision, Formal analysis. Ke Ma: Software, Resources, Formal analysis, Data curation. Fakai Li: Investigation, Data curation. Yongkang Yang: Resources, Methodology, Investigation. Yiming Ren: Writing – review & editing, Supervision, Investigation. Zheng Guo: Data curation, Conceptualization. Tianze Gao: Validation, Methodology. Chao Wang: Validation, Supervision, Data curation. Jiafeng Miao: Software. Qinlin Yan: Software. Juan Guan: Writing – review & editing, Supervision, Project administration, Funding acquisition. Shuyun Liu: Writing – review & editing, Supervision. Quanyi Guo: Project administration, Funding acquisition, Conceptualization. Ethics approval and consent to participate This study was performed in accordance with the Declaration of Helsinki and relevant national regulations. Ethical approval was granted by the Ethical Review Committee of General Hospital of PLA (Approval No. SQ20251016). All animal experiments were conducted in compliance with the ARRIVE guidelines and institutional animal care protocols. 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.