Human knee single-cell RNA sequencing (scRNA-seq) data, accessible under accession number GSE220243, were sourced from the Gene Expression Omnibus (GEO) database^[20]. The dataset includes tissues from healthy and OA knees, specifically healthy articular cartilage (n = 6), OA articular cartilage (n = 6), healthy meniscus (n = 7 donors), and OA meniscus (n = 6 donors). A total of 26 scRNA-seq libraries are included in the GEO series, with the vascular and avascular regions of the meniscus from one healthy donor analyzed separately. No synovial tissue samples are included in GSE220243.

The analytical workflow was performed according to established procedures as previously described. Quality control measures excluded low-quality cells with feature counts outside 200-6,000 or mitochondrial/ribosomal transcript proportions over 10%.To minimize technical variability from sequencing depth and cellular heterogeneity, the retained cells underwent normalization and scaling using the NormalizeData and ScaleData functions. Principal component analysis (PCA) was performed on the identified highly variable features to reduce dimensionality. To minimize inter-sample batch effects, data integration was performed using the RunHarmony function implemented in the Harmony package. The integrated dataset was visualized using Uniform Manifold Approximation and Projection (UMAP), and major cell populations were annotated based on the expression of established lineage-specific marker genes, referencing the CellMarker database (http://bio-bigdata.hrbmu.edu.cn/CellMarker/).

Monocle2 (version 2; http://cole-trapnell-lab.github.io/monocle-release/) was employed for single-cell trajectory reconstruction. Dimensionality reduction was performed to capture major transcriptional variation among cells, followed by cell ordering along the inferred pseudotime trajectory. This analysis enabled the characterization of transcriptional state transitions and branch points associated with differentiation or activation processes.

Differential expression analysis utilized the ‘limma’ R package, applying a significance threshold of P < 0.05^[21]. The biological functions of identified differentially expressed genes (DEGs) in cell clusters were analyzed using the ‘org.Hs.eg.db’ and ‘clusterProfiler’ packages. A Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was conducted to identify significantly enriched pathways, applying a P value threshold of 0.05.

Single-cell macrophage efferocytosis-related gene-set activity was quantified using AUCell, which employs area-under-the-curve (AUC) enrichment scores. Gene sets were sourced from the Hallmark collection in the Molecular Signatures Database (MSigDB; https://www.gsea-msigdb.org/). M1- and M2-associated transcriptional programs were similarly evaluated across macrophage subtypes using the corresponding gene signatures, and their relative activities were compared on the basis of AUC scores.

Samples of cartilage, synovial tissue, and synovial fluid were collected from 10 healthy controls and 10 osteoarthritis patients [Supplementary Table 1]. Normal-control specimens were collected during arthroscopic procedures for acute anterior cruciate ligament rupture or meniscal injury, while OA specimens were obtained from total knee arthroplasty patients. Subjects were excluded before experimental allocation if they had a history of bone fracture, previous knee surgery, including arthroplasty, ligament reconstruction, or cartilage repair, or inadequate clinical imaging quality. Following confirmation of eligibility, specimens were coded and allocated to the planned downstream assays. When only five specimens were required for a particular assay, five cases were randomly selected from the eligible cohort before experimental analysis. The study involving human specimens was reviewed and approved by the Human Ethics Committee of Changzheng Hospital (Approval No. SL202410-7A). Written informed consent was obtained from all participants before sample collection.

Primary BMSCs were obtained and cultured as previously described^[22]. C57BL/6 mice, aged one month, were euthanized through cervical dislocation and surface-sterilized by immersing them in 75% ethanol for 5 min. The sex of the donor mice was not specified. Femurs and tibias were aseptically dissected and transferred to ice-cold phosphate-buffered saline (PBS, G4202, Servicebio) supplemented with 1% penicillin-streptomycin (P/S; G4003, Servicebio). The bone epiphyses were excised to reveal the medullary cavities, which were then flushed with Dulbecco’s Modified Eagle Medium (DMEM)/F-12 medium (Gibco, 12634028) supplemented with 20% fetal bovine serum (FBS) (ZETAlife, Z7010FBS) and 1% P/S, using a 1 mL syringe equipped with a 25-gauge needle. The bone marrow was collected, homogenized by pipetting, and filtered through a 75 μm cell strainer to create a single-cell suspension. Cells were resuspended and seeded into T25 culture flasks, then incubated at 37 °C with 5% CO₂. After 24 h, non-adherent cells were removed by aspiration, and fresh complete medium was replenished. The medium was first replaced at 72 h and then every 48 h thereafter. Cells at around 80% confluency were detached with 0.25% Trypsin-ethylenediaminetetraacetic acid (EDTA) (G3440, Servicebio) for subculturing. Phenotypic surface biomarkers were identified using flow cytometry (CytoFlex, A00-1-1102, XItogen Biotechnologies Co., Ltd., China). According to the manufacturer’s protocols, BMSCs exhibited positive biomarkers CD44 (949601, BioLegend), CD29 (102201, BioLegend), and CD90 (105201, BioLegend), while negative biomarkers included CD11b (101203, BioLegend) and CD45 (157202, BioLegend). Third-passage cells were utilized for subsequent experiments.

BMDMs were extracted from the long bones of one-month-old C57BL/6 mice post-euthanasia. The marrow cavity was flushed under sterile conditions with a syringe after excising the epiphyses of the tibiae and femurs. The collected cells were resuspended and seeded into culture flasks. Bone marrow cells were cultured in 10-cm dishes using DMEM (Gibco, 11965092) with 10% FBS, 1% penicillin-streptomycin, and 25 ng/mL macrophage colony-stimulating factor (M-CSF; GB310046, Servicebio). Cultures were maintained for 6 days, with medium replacement every 48 h.

Primary chondrocytes were extracted from neonatal C57BL/6 mouse knee articular cartilage using enzymatic digestion, following established methods^[23]. The sex of the neonatal donor mice was not specified. Cartilage tissues were minced into ~1 mm³ fragments and digested sequentially with 0.25% trypsin for 30 min, then 0.2% type II collagenase for 4 h in a shaking incubator at 90 rpm. The isolated cells were cultured in DMEM/F12 medium (Gibco, 12634028) with 10% FBS (ZETAlife, Z7010FBS) and 1% P/S (Servicebio, G4003) at 37 °C in a 5% CO₂ humidified atmosphere. To preserve the chondrogenic phenotype, only cells within two passages were used for experiments.

Cultured cells were regularly screened for mycoplasma contamination using a PCR-based detection assay (C0301S, Beyotime Biotechnology, China) following the manufacturer’s guidelines. Cells were tested before their use in downstream experiments and periodically during culture. Only mycoplasma-negative cultures were used for experimental analyses, as all cultures tested negative for mycoplasma.

BMSCs were maintained in 10 cm dishes with exosome-depleted FBS (System Biosciences, USA, 50A-1) at 37 °C in a normoxic environment (21% O₂).MSC-EVs were extracted from conditioned medium using differential ultracentrifugation, following established protocols. Supernatants were centrifuged sequentially at 300 × g for 10 min, 2,000 × g for 10 min, and 10,000 × g for 30 min at 4 °C to eliminate cellular debris. The clarified medium underwent two rounds of ultracentrifugation at 100,000 × g for 70 min each at 4 °C. The resulting pellet was resuspended in PBS to obtain BMSC-derived MSC-EVs. Transmission electron microscopy (TEM) (HT7800/HT7700, Hitachi) was utilized for morphological characterization. Nanoparticle tracking analysis (NTA) with a Flow NanoAnalyzer N30E (China) was employed to assess the particle size distribution and concentration of MSC-EVs. The expression of exosome markers CD81, CD63, and TSG101 was assessed using Western blot analysis.

An efferocytosis assay was conducted following the method outlined in a previousstudy^[6]. BMDM were cultured in 6-well plates at 1 × 10⁶ cells per well and incubated overnight in DMEM with 10% FBS. Cells were subsequently treated for 24 or 48 h with lipopolysaccharide (LPS, 100 ng/mL), apoptotic chondrocytes, BMSCs, MSC-EVs (104 particles /cell), miR-21a-5p mimics, or Yes1-associated transcriptional regulator (Yap1)-OE. Chondrocytes underwent apoptosis induction via 24-h exposure to tumor necrosis factor alpha (TNF-α) (500 ng/mL), following the methodology outlined in a prior study^[24]. Apoptotic chondrocytes labeled with carboxyfluorescein succinimidyl ester (CFSE) were co-cultured with macrophages at a 10:1 ratio for 120 min. After incubation, cells were washed three times with PBS to eliminate non-phagocytosed chondrocytes. Immunofluorescence (IF) microscopy was utilized to evaluate efferocytosis. The efferocytosis index was calculated by determining the percentage of macrophages with apoptotic bodies in relation to the total number of macrophages, then multiplying by 100%.

To characterize the miRNA cargo of BMSC-derived MSC-EVs and screen for highly abundant candidate miRNAs, one purified MSC-EV sample (n = 1) was subjected to high-throughput small RNA sequencing. RNA extraction from purified MSC-EVs was performed using TRIzol reagent following the manufacturer’s guidelines. RNA quality and concentration were assessed before library preparation. Genekinder Medicaltech (Shanghai, China) provided technical support for small RNA library construction and high-throughput sequencing. Raw sequencing reads were subjected to quality control, including removal of adaptor sequences, low-quality reads, and other invalid reads. Clean reads were aligned to the mouse reference genome, and known miRNAs were identified by comparing them with annotated sequences in the miRBase database. miRNA levels were normalized to reads per million (RPM) by dividing individual miRNA reads by the total mapped miRNA reads and multiplying by 10⁶. The miRNAs were then ranked by normalized abundance, and the top ten were chosen for further validation via quantitative real-time polymerase chain reaction (RT-qPCR). Because only one MSC-EV sample was subjected to miRNA sequencing, this analysis was used exclusively for descriptive miRNA abundance profiling and candidate screening, rather than for differential-expression analysis or statistical inference.

In recipient-macrophage gain-of-function experiments, primary BMDMs were transfected with miR-21a-5p mimic or a non-targeting control (mimic-NC) using Lipofectamine^TM 3000, following the manufacturer’s protocol. Both the miR-21a-5p mimic and mimic-NC were sourced from Shanghai BioTNT Biotechnology Co., Ltd. 4 h post-transfection, the medium was refreshed with new growth medium, and the cells were subjected to the specified treatments involving LPS and/or Yap1 overexpression for further functional assays.

In the miR-21a-5p inhibition study, BMSCs were transfected with either a miR-21a-5p inhibitor or a non-targeting control (inhibitor-NC) using Lipofectamine^TM 3000 for 24 h. Post-transfection, the cells were washed and maintained in an EV-depleted medium. The conditioned medium was collected, and EVs were isolated using differential ultracentrifugation as previously detailed. EV preparations derived from inhibitor-NC- or miR-21a-5p inhibitor-transfected BMSCs were subsequently applied to recipient BMDMs.

The complete coding sequence of the Yes1-associated transcriptional regulator (Yap1; Gene ID 22601, NM_001171147.1) was synthesized and inserted into a Flag-tagged vector (PGMLV-6751) by BioTNT Biotechnology Co., Ltd. (Shanghai, China). Mouse primary BMDMs were transfected with the constructed plasmid using Lipofectamine^TM 3000 reagent (Invitrogen) following the manufacturer’s protocol. Cells were seeded at 6 × 10⁴ per well in 6-well plates a day before transfection. The culture medium was replaced when the cells reached 80% confluency. A transfection mixture was prepared with 125 µL Opti-MEM® medium, 7.5 µL Lipofectamine3000^TM, 2,500 ng plasmid DNA, and 5 µL P3000^TM reagent, then added to each well and incubated for 24 h. Finally, overexpression was assessed using RT-qPCR [Supplementary Figure 1].

Three siRNA sequences were designed in collaboration with BioTNT Biotechnology Co., Ltd. (Shanghai, China) to silence mouse Yap1 expression. Mouse primary BMDMs were transfected at 80% confluency in 12-well plates using Lipofectamine^TM 3000 Reagent (Invitrogen) following the manufacturer’s guidelines. 24 h post-transfection, total RNA was extracted, and the silencing efficiency of each siRNA was evaluated via RT-qPCR. The siRNA with the most effective Yap1 mRNA knockdown was chosen for further experiments. Finally, Yap1-si-3 was chosen as the best gene-silencing siRNA: ACUUGGAGGCGCUCUUCAAUGTT (sense), CAUUGAAGAGCGCCUCCAAGUTT (antisense).

Primary BMDMs were isolated and seeded in 6-well plates at a density of 5 × 10⁶ cells per well, followed by a 24-h culture period. Macrophages with deficient efferocytosis were generated by treating them with 100 ng/mL of LPS. The intervention group received concurrent incubation with MSC-EVs at a concentration of 10⁴ particles per cell. Cells from both groups were collected after 24 h, and total RNA was extracted using TRIzol reagent, then stored at -80 °C. mRNA sequencing was subsequently performed with technical support from Genekinder Medicaltech (Shanghai, China). The R package DESeq2 was utilized to identify differentially expressed genes (DEGs)^[25]. mRNAs with |log₂(Fold Change)| > 1.5 and false discovery rate (FDR) < 0.05 were selected for further analysis. Expression patterns of genes associated with efferocytosis and inflammation across experimental groups were visualized using heatmaps, created with the ggplot2 and heatmap R packages.

Total RNA was extracted from BMDMs, chondrocytes, or mouse knee joint tissues using Magen Biotechnology Co., Ltd. (R4122-02). RNA purity and concentration were assessed using a NanoDrop spectrophotometer (Thermo Fisher Scientific, USA), employing the A260/A280 ratio for quality evaluation. RNA (1 µg) was converted to cDNA using the HiScript® III RT SuperMix for quantitative PCR (qPCR) Kit (Vazyme, R323-01) in a thermal cycler. qPCR was conducted with SYBR qPCR Master Mix (Q711-02, Vazyme) on an Applied Biosystems Real-Time PCR system. The expression levels of the target genes Gas6, Mertk, Stab2, Lamp1, SIRPa, Cd300a, Inos, Il1b, Il10, Arg1, Yap1, Arhgap24, Nfia, Nfib, Smad7, Tgfbi, Bax, Bcl2, Col2a1, and MMP3 were analyzed using GAPDH as the endogenous control. Relative mRNA expression was quantified using the 2-ΔΔCt method, with target gene threshold cycle (Ct) values normalized to GAPDH and fold changes assessed against the control group. All primers were designed and synthesized by Shanghai BioTNT Co., Ltd.

The binding sequences between miR-21a-5p and Yap1 were initially predicted using bioinformatic tools. BioTNT Co., Ltd. synthesized DNA fragments of the wild-type (WT) and mutant (MUT) 3’-untranslated region (3’-UTR) of mouse Yap1, which include potential miR-21a-5p binding sites. These fragments were cloned into the pmirGLO dual-luciferase vector (Promega), featuring both firefly and Renilla luciferase reporter genes. The MUT construct involved mutating the seed region of the miR-21a-5p binding site to eliminate the interaction. Using Lipofectamine^TM 3000 (Invitrogen), HEK293T cells were co-transfected with the constructs and miR-21a-5p mimics. Luciferase activity was assessed after 48 h using the Dual-Luciferase® Reporter Assay System (Promega), with Renilla luciferase values serving as normalization controls.

Human synovial fluid-derived extracellular vesicles (SF-EVs) were isolated using differential and ultracentrifugation methods as previously outlined^[26]. Fresh synovial fluid samples were centrifuged sequentially at 3,000 × g and 10,000 × g to eliminate cells, debris, and particulates. The clarified supernatant was filtered through a 0.22 μm membrane and then ultracentrifuged at 100,000 × g for 120 min. The EV pellet was resuspended in sterile PBS for immediate downstream analysis or stored at -80 °C for later use. The isolated vesicles were referred to as SF-EVs.

Total miRNA was extracted from the isolated SF-EVs using a Magnetic EV miRNA Kit (TIANGEN Biotech, Beijing, China) following the manufacturer’s instructions. miRNAs were reverse-transcribed with the Evo M-MLV RT Premix for qPCR, followed by RT-qPCR using the SYBR Green Premix Pro Taq HS qPCR Kit, both from Accurate Biology, China.U6 served as the internal reference for normalizing miRNA levels. The 2-ΔΔCt method was employed to determine the relative expression of miR-21a-5p.

Servicebio (Wuhan, China) designed and synthesized miR-21a-5p probes. Mouse knee joint tissue sections underwent fluorescence in situ hybridization (FISH) using a commercial kit (GF002-50T, Servicebio) following the manufacturer’s instructions. In human synovial tissue, miR-21a-5p was identified with a FAM-labeled probe, resulting in a green fluorescence signal. For mouse knee synovial tissue and subchondral bone marrow tissue, a Cy3-labeled miR-21a-5p probe was used, producing a red fluorescence signal. Servicebio (Wuhan, China) designed and synthesized the probes. Imaging was performed using a Nikon A1Si laser scanning confocal microscope (Nikon Instruments, Japan), and images were analyzed with Image-Pro Plus 6.0 software (NIH, USA). Results are expressed as relative fluorescence intensity, as described in previous study^[27].

Protein extraction from chondrocytes, BMSCs, and MSC-EVs under different treatments was performed using radioimmunoprecipitation assay (RIPA) lysis buffer (G2002, Solarbio) with added phosphatase inhibitor cocktail (G2007, Bimake, USA) and protease inhibitor (G2008, Solarbio). Western blotting was performed following established protocols^[22]. Primary antibodies included: CD81 (GB111073, Servicebio), CD63 (GB115712, Servicebio), TSG101 (GB150022, Servicebio), Calnexin (GB111369, Servicebio), Collagen II (GB11021, Servicebio), SOX9 (GB11208, Servicebio), MMP13 (#AF5355, Affinity), TRPV1 (680834, Zenbio), MMP3 (#AF0217, Affinity), BAX (#AF0120, Affinity), Bcl2 (#AF6139, Affinity), YAP1 (#AF6328, Affinity), and GAPDH (#AF7021, Affinity).Protein bands were visualized using a ChemiDoc XRS+ Imaging System (Tanon, China) following incubation with horseradish peroxidase-conjugated goat anti-rabbit or anti-mouse secondary antibodies (#BL003A / #BL001A, Biosharp). Quantification was performed with ImageJ software (v1.8.0). All antibodies used were commercially available.

All animal procedures were reviewed and approved by the Animal Ethics Committee of Wuhan Servicebio Animal Experiment Center (Approval No. 2024112) and were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

C57BL/6 mice, sourced from GemPharmatech Co., Ltd., were utilized in this study. An OA model was induced in 6-8-week-old mice, weighing approximately 20 g, through the destabilization of the medial meniscus (DMM) surgical method, as previously described^[28,29]. After administering isoflurane anesthesia, a longitudinal incision was performed on the skin from the distal patella to the proximal tibial plateau of the right knee. The joint capsule medial to the patellar tendon was carefully opened, and the medial meniscus and medial meniscotibial ligament (MMTL) were visualized. The MMTL was identified and transected using microsurgical scissors, resulting in destabilization of the medial meniscus. The medial meniscus was retained in situ and was not excised or resected. The surgical site was disinfected with povidone-iodine after closing the joint capsule and skin. Sham-operated mice experienced identical surgical exposure and arthrotomy, except that the MMTL remained unaltered. Postoperative mice had unrestricted cage activity and free access to food and water. Four independent in vivo experimental designs were performed as follows.

In the experiment assessing MSC-EVs’ role in BMSCs’ therapeutic effects, mice were randomly divided into four groups (n = 5 per group): (1) Sham; (2) DMM + PBS; (3) DMM + BMSCs; and (4) DMM + BMSCs treated with GW4869.Mice in the DMM + PBS group received 10 μL PBS by intra-articular injection. Mice in the BMSC treatment groups received 2 × 104 BMSCs suspended in 10 μL PBS per injection. To pharmacologically reduce neutral sphingomyelinase-sensitive EV release, BMSCs were pretreated with GW4869 (10 μM) for 24 h before administration; the cells were subsequently collected, washed, and resuspended at the same cell dose as that used in the DMM + BMSC group. Treatments commenced three days post-surgery and were administered every four weeks until the experiment concluded at eight weeks. Eight weeks post-DMM surgery, mice were euthanized to collect knee-joint specimens. Samples for histological analysis underwent a 4-week fixation and decalcification process before being embedded in paraffin, sectioned, and stained for further examination.

An additional independent in vivo experiment was performed to directly evaluate the effect of local GW4869 administration on OA progression. Mice were randomly assigned into three groups (n = 5 per group): (1) Sham + Vehicle; (2) DMM + Vehicle; and (3) DMM + GW4869. GW4869 was dissolved in vehicle solution consisting of sterile PBS containing the appropriate solvent concentration, and 10 μL of vehicle or GW4869 solution at a final concentration of 20 μM was administered by intra-articular injection into the knee joint. Treatment was initiated 3 days after DMM surgery and repeated once weekly for 4 consecutive weeks. At 8 weeks after DMM surgery, mice were euthanized, and knee joint specimens were collected for subsequent micro-computed tomography (micro-CT), histological, and molecular analyses. This experiment was performed independently of the BMSC pretreatment with GW4869 experiment described above.

In the experiment assessing the therapeutic impact of miR-21a-5p mimics, mice were randomly divided into four groups (n = 5 each): (1) Sham + negative control (NC); (2) Sham + miR-21a-5p mimic; (3) DMM + NC; and (4) DMM + miR-21a-5p mimic. An intra-articular administration of 10 μL per knee was performed using either a miR-21a-5p mimic or a negative-control oligonucleotide (BioTNT, Shanghai) at a concentration of 2 nM. The first injection was administered on postoperative day 3 and was repeated once weekly. Eight weeks post-DMM surgery, mice were euthanized to collect knee-joint specimens. Samples for histological analysis underwent a 4-week fixation and decalcification process before being embedded in paraffin, sectioned, and stained for further examination.

An experiment was conducted to assess if miR-21a-5p inhibition affects the therapeutic efficacy of MSC-EVs in vivo. Mice were divided into four groups (n = 5 each): (1) Sham+MSC-EVs; (2) Sham+MSC-EVs with miR-21a-5p inhibitor; (3) DMM + MSC-EVs; and (4) DMM+MSC-EVs with miR-21a-5p inhibitor. Purified MSC-EVs were administered at a dose of 1 × 10⁸ particles per knee. Where indicated, miR-21a-5p inhibitor (Antagomir-miR-21a-5p, 2 nM, BioTNT, Shanghai) was co-administered. The preparations were diluted to a 10 μL total volume using sterile PBS and delivered via intra-articular injection. Treatment was initiated on postoperative day 3 and repeated every four weeks. Eight weeks post-DMM surgery, mice were euthanized to collect knee-joint specimens. Samples for histological analysis underwent a 4-week fixation and decalcification process before being embedded in paraffin, sectioned, and stained for further examination.

A formal a priori power calculation was not conducted. Sample sizes were determined based on prior studies using similar experimental OA and extracellular-vesicle interventions, as well as our previous experimental experience^[26,30,31]. A minimum of three independent biological replicates was used for molecular experiments, whereas five mice per group were used for the principal in vivo radiological, histological, and imaging analyses. Each mouse was treated as a separate experimental unit.

Animals were eligible for analysis if they completed the assigned DMM or sham procedure and the scheduled treatment regimen and provided samples of sufficient technical quality for the prespecified endpoint. No animal attrition or exclusion occurred during the study; all animals allocated to the four in vivo experiments survived to the planned Week-8 endpoint and were accounted for in the analyses.

Animals in all in vivo experiments were randomly assigned to experimental groups using a random-number sequence, ensuring equal numbers in each group before treatment began. The individual mouse was considered the independent experimental unit. Because the administered formulations differed among treatment groups, the investigator responsible for preparing and administering the intra-articular treatments was aware of group allocation but did not participate in subsequent outcome assessment. Following completion of the animal experiments, knee specimens, histological sections, and corresponding imaging datasets were assigned coded identification numbers that concealed surgical and treatment-group information. Investigators responsible for Micro-CT quantification, histological scoring, and quantitative image analysis were blinded to group allocation until all measurements had been completed and recorded. Group allocation was revealed only after completion of the quantitative analyses.

Eight weeks post-DMM surgery, the mice were euthanized and analyzed using the MX-20 Cabinet X-ray System (Faxitron, USA). The scan was conducted with a 10-s exposure time at a voltage of 32 kV. The articular space width in mice was assessed using X-ray radiographs. Subsequently, arthrography of the knee joint samples was performed utilizing a high-resolution micro-CT imaging system (Bruker SkyScan1276, USA). Micro-CT scanning and reconstruction results indicate measurements of articular space width, relative osteophyte volume, bone mineral density (BMD), bone volume to total volume ratio (BV/TV), and trabecular thickness (Tb.Th), trabecular number (Tb.N), and trabecular separation (Tb.Sp) were analyzed.

Hematoxylin and Eosin (H&E) and Safranin O-Fast Green (SO/FG) staining were utilized to assess tissue morphology and cartilage structural integrity. In this study, H&E staining was applied to assess human synovium tissue, providing detailed visualization of cellular and extracellular matrix (ECM) architecture and facilitating the identification of pathological alterations. SO/FG staining was employed to assess structural alterations in knee joint articular cartilage and osteophytes. Fixed tissue sections underwent Safranin O staining to label glycosaminoglycans in the cartilage matrix (red), followed by Fast Green counterstaining to visualize collagenous and bony elements (green). This differential staining enables clear distinction between cartilaginous and non-cartilaginous components.

For immunohistochemical and IF analyses, human synovium and mouse knee joint tissue sections were treated with 5% bovine serum albumin (BSA) for 30 min at room temperature, followed by an overnight incubation at 4 °C with primary antibodies: CD68 (GB113150, Servicebio), iNOS (GB11119, Servicebio), TREM2 (#DF12529, Affinity), interleukin-1 beta (IL-1β) (BF8021, Affinity), MMP3 (340612, ZenBio), Collagen II (DF7561, Affinity), and YAP1 (Kerafast, #ENH001). BMDMs and chondrocytes were fixed with 4% paraformaldehyde for 10 min, permeabilized using 0.15% Triton X-100 for 5 min, and washed thrice with PBS before incubation with primary antibodies targeting Arg1 (GB111841, Servicebio), iNOS (AF5484, Affinity; R23896, Zenbio), IL-1β (DF12084, Affinity), MMP3 (#AF0217, Affinity), Collagen II (GB11021, Servicebio), and YAP1 (#AF6328, Affinity). After washing with TBST (G0004, ServiceBio), samples were incubated with species-appropriate secondary antibodies conjugated to FITC or Cy3 (GB21401, GB22301, ServiceBio). Imaging utilized a Nikon A1Si Laser Scanning Confocal Microscope (Nikon Instruments Inc., Japan), with semi-quantitative analysis performed using ImageJ software.

In vivo apoptotic cells were identified using a TUNEL assay kit (G1504, G1509, Servicebio) following the manufacturer’s guidelines. Joint sections embedded in paraffin were deparaffinized using xylene and then rehydrated with a graded series of ethanol. Sections were fixed in 4% paraformaldehyde for 20 min at room temperature and then permeabilized using 0.1% Triton X-100. For chondrocyte samples, permeabilization was conducted following fixation with 4% paraformaldehyde. All sections underwent a 15-min proteinase K treatment, followed by a 60-min incubation with the TUNEL reaction mixture, and were then counterstained with 4’,6-diamidino-2-phenylindole (DAPI) for 5 min. A Nikon A1Si Laser Scanning Confocal Microscope (Nikon Instruments Inc., Japan) was used for imaging.

All statistical analyses were performed using GraphPad Prism version 10.1 (GraphPad Software, USA). In vitro cell-based experiments report the sample size (n) as the count of independent biological experiments, unless noted otherwise. In vivo experiments treated each animal as an independent experimental unit. Technical subsamples, consisting of multiple tissue sections or microscopic fields from the same animal, were averaged to produce a single value per animal for statistical analysis. Accordingly, the reported n for animal experiments represents the number of independent animals rather than the number of sections or images. Continuous data are expressed as mean ± standard error of the mean (SEM). The Shapiro-Wilk test and graphical inspection were used to assess normality, while homogeneity of variance was evaluated prior to conducting parametric analysis. An unpaired two-tailed Student’s t-test was applied for continuous outcomes between two independent groups when parametric test assumptions were met. A one-way analysis of variance (ANOVA) followed by Tukey’s multiple-comparison test was employed for comparisons among more than two independent groups lacking a factorial structure. Non-parametric statistics were used to analyze ordinal histopathological scores, such as Osteoarthritis Research Society International (OARSI), synovitis, and osteophyte scores. The Mann-Whitney U test compared two groups, while the Kruskal-Wallis test with Dunn’s post hoc analysis was applied for comparisons among multiple groups. Continuous outcomes in factorial animal experiments were analyzed using two-way ANOVA, followed by Šidák’s multiple-comparison test. Summary of experimental replication and statistical analyses was presented in Supplementary Table 2. A P value below 0.05 was deemed statistically significant.

To further characterize macrophage populations associated with OA progression, we reanalyzed the publicly available GSE220243 single-cell transcriptomic dataset comprising 26 samples^[20]. Based on established marker-gene expression profiles^[3,4,32], cells from control and OA samples were classified into 14 major cell populations, comprising a total of 182,943 cells [Supplementary Figure 2A]. The relative distribution of these cell populations between the control and OA groups was also examined [Supplementary Figure 2B]. The identified cell clusters including seven chondrocyte subgroups, endothelial cells, epithelial cells, granulocyte-monocyte progenitors (GMPs), macrophages, natural killer T cells (NKT), smooth muscle cells (SMCs), and T cells, and their cell markers were also visualized [Supplementary Figure 2C]. We next examined the pseudotemporal trajectory of the chondrocyte subpopulations. Pseudotime analysis showed that Chon2 and Chon7 were predominantly distributed toward the terminal region of the inferred trajectory [Supplementary Figure 2D]. KEGG pathway analysis identified enrichment of apoptosis-related signaling pathways in the two clusters [Supplementary Figure 2E and F], aligning with previous findings that associate chondrocyte apoptosis with OA progression^[33-35].

Efferocytosis, the process by which macrophages clear apoptotic cells, plays a crucial role in resolving inflammation^[7]. Inefficient efferocytosis hinders apoptotic-cell clearance, leading to sustained inflammatory signals and synovial inflammation^[36], while effective efferocytosis reduces pro-inflammatory cytokine production and aids in resolving inflammation^[37]. Given the prominent role of macrophages in the synovial microenvironment and their close association with OA progression, we next focused on macrophage heterogeneity and functional states. OA samples exhibited a higher proportion of macrophages compared to control samples [Supplementary Figure 2B].

Unbiased clustering identified seven macrophage subclusters (Mac0-Mac6) with distinct transcriptional profiles [Figure 1A-C].Particularly, these cell clusters were:Mac0 (LGMN, LYVE1, SLC40A1, LGI2, GABRB2), Mac1 (TIMD4, XPO7, CEPT1, ZNF117, MYO1G), Mac2 (FOXB, DDIT3, NXF1, PROX1, LDLR), Mac3 (COL8A1, TSHZ2, IL17D, TMEM196, CCN5), Mac4 (TNFAIP3, IL1B, IL6, CXCL2, CCL4), Mac5 (PLCG2, CSF3R, S100A9, MMP9, S100A8), and Mac6 (CLEC10A, FCER1A, ZFAND2A, CD1C, LGALS2).To explore the transcriptional relationships among these subsets, we performed pseudotime trajectory analysis using Monocle.Mac0 cells were predominantly positioned toward the beginning of the inferred trajectory, whereas Mac4 cells, characterized by high expression of inflammatory genes, were enriched toward the opposite terminal region [Figure 1D-F]. Gene-signature scoring revealed stronger M2-like transcriptional features in Mac0 and more pronounced M1-like features in Mac4 [Figure 1G]. Consistently, Mac0 showed higher expression of phagocytosis- and efferocytosis-related genes, whereas Mac4 exhibited greater expression of pro-inflammatory genes [Figure 1H]. Efferocytosis scoring and KEGG pathway analysis further supported the enrichment of efferocytosis-related programs in the Mac0 cluster [Supplementary Figures 3 and 4].

The synovium is the most important macrophage-resident tissue in the knee joint^[38]. Initially, we investigated synovial inflammation and macrophage phenotypes in human osteoarthritis synovial tissue. In line with earlier research^[39-41], human OA synovial tissues showed pronounced synovial hyperplasia, extensive cellular infiltration, and notably higher synovitis scores than normal controls [Figure 1I and J]. We conducted a detailed analysis of macrophage accumulation and phenotypic alterations in human osteoarthritis synovial tissue. OA synovium exhibited a significant rise in CD68-positive macrophages and a greater proportion of iNOS-positive cells, primarily located in the intimal lining layer [Figure 1K-M], compared to controls. In contrast, the proportion of TREM2-positive macrophages, associated with an efferocytosis-related phenotype, was significantly reduced in OA synovium [Figure 1N and O]. Macrophage infiltration in the synovium was quantified, and the phenotypic profile of synovial tissue near the patellofemoral joint was characterized in the DMM mouse model^[28]. Histological analysis revealed significant synovial thickening and increased cellularity in DMM-induced OA mice compared to sham-operated controls, along with a notably higher synovitis score [Supplementary Figure 5A and B]. IF staining demonstrated increased F4/80-positive macrophage accumulation and a higher proportion of iNOS-positive pro-inflammatory macrophages in OA synovium [Supplementary Figure 5C and D], suggesting enhanced macrophage infiltration and inflammatory activation post-DMM surgery.

Efferocytosis, the process by which macrophages clear apoptotic cells, plays a crucial role in maintaining tissue homeostasis and resolving inflammation^[37]. We investigated the impact of efferocytosis on macrophage phenotype by isolating murine BMDMs and primary chondrocytes, following established protocols^[35,42]. An in vitro co-culture system was set up using apoptotic chondrocytes, with efferocytosis either permitted or inhibited [Supplementary Figure 6A]. Apoptosis was induced in chondrocytes by TNF-α treatment.TUNEL staining indicated a dose-dependent rise in apoptotic chondrocytes, with 500 ng/mL TNF-α for 24 h chosen to create the apoptotic chondrocyte model [Supplementary Figure 6B and C]. Co-culturing with apoptotic chondrocytes led to a macrophage phenotype with decreased Il1b, Inos, and Tnfα expression and enhanced interleukin-10 (IL-10) production [Supplementary Figure 6D and E]. In contrast, blockade of efferocytosis with Annexin V^[24] increased pro-inflammatory macrophage responses, including higher Il1b, Inos, and Tnfα expression and increased IL-1β production [Supplementary Figure 6D and E]. These findings suggest that effective clearance of apoptotic chondrocytes is linked to a pro-resolving macrophage phenotype, while impaired efferocytosis promotes inflammatory activation.

BMSCs exhibit potent immunomodulatory properties and can promote macrophage polarization toward a reparative phenotype^[43,44]. However, their effects on macrophage efferocytosis remain poorly understood. Given their role in inflammation resolution and tissue homeostasis, BMSCs may provide a means of restoring impaired efferocytosis under inflammatory conditions.

We isolated BMSCs from 1-month-old C57BL/6 mice to study their impact on macrophage efferocytosis, following established methods^[45] [Figure 2A]. Microscopic examination showed that the cells displayed a characteristic spindle-shaped morphology [Figure 2B]. Flow cytometry analysis confirmed the BMSC phenotype, revealing high expression levels of CD44 (99.9%), CD29 (99.8%), and CD90 (96.9%), alongside minimal expression of CD45 (0.37%) and CD11b (0.24%) [Figure 2C]. We then established a BMSC-macrophage co-culture model with or without LPS stimulation6^[6] [Supplementary Figure 7A]. RT-qPCR analysis indicated that LPS decreased the expression of efferocytosis-related genes such as Gas6, Mertk, Stab2, Lamp1, Sirpα, and Cd300a; however, co-culture with BMSCs mitigated these effects [Supplementary Figure 7B].

Treatment with GW4869 attenuated the effects of BMSCs on efferocytosis-related gene expression [Supplementary Figure 7B]. Phagocytosis assays demonstrated a consistent trend [Supplementary Figure 7C and D]. Furthermore, we evaluated the inflammation modulation of BMSCs on macrophages. As expected, LPS also induced a pro-inflammatory macrophage phenotype, characterized by increased iNOS-positive cells and elevated Il1b, Tnfα, and Inos expression, together with reduced Arg1 and Il10 expression [Supplementary Figure 7E-G]. Co-culture with BMSCs significantly promoted anti-inflammatory polarization, and restored the down-regulation of anti-inflammatory factors Il10 and Arg1 induced by LPS in BMDMs [Supplementary Figure 7E-G]. However, administration of GW4869 evidently reversed the therapeutic effects of BMSCs [Supplementary Figure 7E-G].

We next isolated MSC-EVs from BMSC-conditioned medium by ultracentrifugation. TEM revealed a typical vesicular morphology [Figure 2D]. Western blot analysis verified the presence of exosomal surface markers CD81, CD63, and TSG101 in both MSC-EVs preparations [Figure 2E]. NTA indicated that the isolated MSC-EV preparations primarily exhibited a particle-size distribution between 30 and 200 nm [Figure 2F]. Then, we evaluated the effects of MSC-EVs on macrophage efferocytosis and inflammation [Figure 2G]. RT-qPCR analysis demonstrated that MSC-EVs, akin to BMSCs, reinstated the expression of efferocytosis-related genes such as Gas6, Mertk, Stab2, Lamp1, Sirpα, and Cd300a following LPS induction [Figure 2H and Supplementary Figure 8]. As expected, MSC-EVs significantly enhanced macrophage efferocytosis under LPS stimulation, as demonstrated by an increased clearance of fluorescent apoptotic cells [Figure 2I and J]. IF staining demonstrated that LPS resulted in decreased expression of Arg1 and increased expression of Inos, whereas MSC-EVs restored the expression of Arg1 [Figure 2K and L]. In vitro experiments with LPS-polarized M1 macrophages demonstrated that MSC-EVs modulate macrophage plasticity by downregulating pro-inflammatory genes and upregulating anti-inflammatory genes, promoting a resolution-prone phenotype [Figure 2M].

These experiments demonstrate that purified BMSC-derived EVs are sufficient to enhance macrophage efferocytosis and promote a pro-resolving inflammatory phenotype in vitro. MSC-EVs may promote inflammatory resolution by restoring efferocytosis-related gene expression and rebalancing macrophage polarization, indicating a novel mechanism of BMSC action. This suggests that macrophages undergo a phenotypic shift from a pro-inflammatory to an anti-inflammatory state after efferocytosis. In addition, this MSC-EV-associated pathway may be exploited therapeutically in inflammatory diseases where impaired efferocytosis perpetuates inflammation, such as OA.

To investigate whether MSC-EVs could mitigate chondrocyte damage through enhancing macrophage efferocytosis, we isolated murine primary chondrocytes from neonatal C57BL/6 mice^[46], and co-cultured them with macrophages under varying efferocytosis conditions [Figure 3A]. Gene expression analysis revealed that chondrocytes co-cultured with macrophages with impaired efferocytosis had reduced expression of anabolic genes Sox9 and Col2a1, alongside elevated expression of catabolic genes Mmp3 and Mmp13 [Figure 3B], suggesting disrupted cartilage homeostasis. Notably, LPS-stimulated macrophages markedly upregulated Il1b expression in chondrocytes, an effect substantially reversed by MSC-EVs treatment [Figure 3C and D], underscoring the role of MSC-EVs in suppressing pro-inflammatory signaling. Further supporting this, chondrocytes co-cultured with LPS-stimulated macrophages exhibited degenerative phenotypes, including reduced Col2a1 and elevated Mmp3 expression. The LPS+MSC-EVs group showed significantly reduced effects [Figure 3E and F], indicating that MSC-EVs may preserve chondrocyte integrity by influencing macrophage behavior. Western blot analysis corroborated these findings, showing downregulation of Col2a1 and Sox9 and upregulation of Mmp3 and Mmp13 under impaired efferocytosis, which were effectively rescued by MSC-EVs [Figure 3G and H]. The restoration of anabolic-catabolic balance indicates that enhanced efferocytosis may promote a pro-resolutive macrophage phenotype capable of supporting chondrocyte viability.

Apoptosis assays further demonstrated that exposure to LPS-stimulated macrophages elevated chondrocyte apoptosis, whereas MSC-EVs treatment markedly reduced cell death [Figure 3I and J]. This protective effect underscores the therapeutic potential of MSC-EVs-mediated efferocytosis enhancement in interrupting the apoptotic-inflammation loop characteristic of OA. To delineate the role of efferocytosis in a more physiologically relevant context, we established an indirect co-culture model^[24]. The experiment consisted of four groups. The lower compartment was seeded with either normal or apoptotic chondrocytes, while the upper compartment contained macrophages with normal or impaired efferocytosis [Figure 3K]. After 24 h of co-culture, apoptotic chondrocytes exhibited decreased Bcl2 and Col2a1 and increased Bax and Mmp3 expression, which was exacerbated by co-culture with efferocytosis-impaired macrophages [Figure 3L and Supplementary Figure 9]. Conversely, macrophages with enhanced efferocytosis capacity induced by MSC-EVs significantly ameliorated these apoptotic and catabolic responses [Figure 3L and Supplementary Figure 9]. Western blot demonstrated consistent results [Figure 3M and N]. The consistent findings from molecular and functional assays demonstrate that MSC-EVs restore macrophage efferocytic function, disrupting the cycle of inflammation and cartilage degradation crucial to OA progression.

Building on promising in vitro results where MSC-EVs improved macrophage efferocytosis and reduced chondrocyte damage, we next evaluated their therapeutic effects in vivo using a murine DMM-induced OA model^[30] to investigate whether the observed cytoprotective and anti-inflammatory effects of MSC-EVs could translate into functional preservation of joint architecture and reduction in disease progression [Figure 4A]. Sham mice underwent exposure of the joint capsule only. The mice were allocated to four groups: Sham, DMM + PBS, DMM + BMSC, and DMM + GW4869-treated BMSC. At 8 weeks after surgery, micro-CT analysis revealed marked structural deterioration in the DMM + PBS group, including a significant increase in osteophyte volume and calcified tissue formation [Figure 4B and C, Supplementary Figure 10]. In contrast, BMSC administration markedly mitigated these OA-associated structural changes, resulting in reduced osteophyte formation and ectopic calcification. These findings indicate that BMSC treatment partially preserves joint structural integrity during DMM-induced OA progression. However, pretreatment of BMSCs with GW4869, a neutral sphingomyelinase inhibitor commonly used to reduce the release of certain EV populations, significantly attenuated the structural benefits associated with BMSC administration [Figure 4B and C]. Histopathological evaluation by H&E and SO/FG further supported the micro-CT findings [Figure 4D and E]. Compared with the DMM + PBS group, mice receiving BMSCs exhibited better-preserved cartilage architecture, with reduced surface discontinuity, fewer vertical fissures, less cartilage erosion, and attenuated structural deformation. Consistently, semiquantitative assessments showed significantly lower OARSI, osteophyte, and synovitis scores in the DMM + BMSC group. These histopathological improvements were attenuated when BMSCs were pretreated with GW4869. These findings are consistent with earlier research highlighting the significance of MSC-derived MSC-EVs in cartilage regeneration and subchondral bone remodeling^[47-49]. Inhibition of exosome secretion by GW4869 diminished the arthroprotective effects of BMSCs, leading to the reappearance of pathological features similar to those in OA controls. Together with the direct MSC-EV treatment experiments, the attenuation of BMSC-mediated therapeutic effects following GW4869 pretreatment supports an important role of EV-mediated paracrine signaling in the protective effects of BMSCs^[50].

IF staining of synovial tissue revealed markedly increased iNOS expression in DMM-induced OA mice [Figure 4F and G], indicating enhanced synovial inflammatory activation. BMSC administration significantly reduced iNOS expression, whereas this effect was attenuated by GW4869 pretreatment. TUNEL staining further showed increased chondrocyte apoptosis in OA cartilage, which was reduced by BMSC treatment but partially restored following GW4869 pretreatment [Figure 4H and I]. Evaluation of ECM homeostasis revealed that OA mice exhibited decreased Aggrecan alongside elevated Mmp3 expression [Figure 4J and K], indicating a catabolic shift and loss of matrix integrity. BMSC administration effectively restored this imbalance, promoting an anabolic environment conducive to cartilage preservation. The reversal of this effect with GW4869 supports that MSC-EVs are pivotal in mediating ECM remodeling. Consistent with the results of H&E and SOFG staining, the administration of GW4869 suppressed the therapeutic effects of BMSCs [Figure 4J-L].

Furthermore, gene expression analysis of synovial tissue revealed reduced expression of the efferocytosis-related genes Gas6, Mertk, and Stab2 in OA mice [Figure 4M], consistent with impaired apoptotic-cell clearance and sustained inflammation^[7]. These changes were accompanied by increased Inos and Il1b expression and reduced Il10 expression, indicating a pro-inflammatory synovial environment. BMSC treatment partially restored the expression of efferocytosis-related and inflammatory genes, whereas these effects were attenuated by GW4869 pretreatment. Together, these findings support a contribution of EV-associated paracrine signaling to BMSC-mediated regulation of macrophage efferocytosis and inflammatory phenotype in OA.

miR-21a-5p is prevalent in BMSC-EVs and linked to macrophage changes related to OA. After confirming the protective effects of BMSCs and BMSC-derived EVs in prior in vitro and in vivo studies, we explored the molecular mechanisms responsible for these effects. Given the critical role of miRNAs as key functional cargo in MSC-EV-mediated intercellular communication and regulatory functions^[51], we aimed to profile the miRNA expression in EVs from BMSCs using miRNA sequencing [Figure 5A]. The data analysis identified the ten most abundant miRNAs, as depicted in Figure 5B. RT-qPCR analysis verified that miR-21a-5p, miR-5119, and miR-17-3p were the most prevalent miRNAs in MSC-EVs [Figure 5C]. miR-21a-5p was increased in macrophages following MSC-EV treatment but not following LPS stimulation alone, indicating that this increase was associated with MSC-EV exposure rather than being a nonspecific response to inflammatory stimulation [Figure 5D and E]. The findings indicated that macrophage miR-21a-5p expression was elevated due to MSC-EV exposure, not inflammatory stimuli. The significant downregulation of miR-21a-5p in synovial fluid-derived extracellular vesicles and synovial tissues from OA patients highlights its clinical relevance and suggests a potential deficiency in miR-21a-5p-mediated regulatory mechanisms in OA [Figure 5F-H].

Co-culture with BMSCs increased miR-21a-5p abundance in macrophages, whereas this increase was attenuated by GW4869 pretreatment of BMSCs [Figure 5I and J]. The enrichment of miR-21a-5p in BMSC-EVs supports the role of EV-associated intercellular signaling. The enrichment of miR-21a-5p in MSC-EVs suggests a link between EV-related intercellular signaling and elevated miR-21a-5p levels in recipient macrophages, despite the absence of direct evidence for EV-mediated transfer in this study. In vivo, suppression of MSC-EVs production via local administration of GW4869 (20 μM, 10 μL per joint, once a week for four weeks)^[26] significantly reduced miR-21a-5p levels in mouse synovial tissues [Figure 5K and L], and in situ hybridization revealed its prominent expression in subchondral bone marrow, a major niche of BMSCs, and cartilage tissue with notable reduction in OA mice [Figure 5M and N]. Collectively, these findings support an association between MSC-EV signaling and increased miR-21a-5p availability in recipient macrophages. Functional experiments further indicate that miR-21a-5p contributes to the regulation of macrophage efferocytosis and inflammatory phenotype. Its deficiency in OA synovium may thus partly explain the observed impairment in efferocytosis and sustained inflammation. The ability of MSC-EVs to replenish miR-21a-5p levels provides a plausible mechanism for their therapeutic effects in recipient macrophages. In fact, previous studies have reported that MSC-derived EV/exosomal miR-21a-5p exerts protective effects in inflammatory and tissue-injury settings, including regulation of macrophage polarization and attenuation of tissue damage^[13,52].

Collectively, these results position exosomal miR-21a-5p as a functional contributor through which MSC-EVs recalibrate macrophage behavior and mitigate OA-related tissue damage, providing novel insights into OA pathogenesis and opening avenues for miRNA-based therapeutics.

We conducted bulk RNA sequencing on LPS-stimulated macrophages treated with MSC-EVs to clarify the molecular mechanisms by which MSC-EVs influence macrophage function [Figure 6A]. KEGG pathway enrichment analysis revealed significant modulation of efferocytosis and inflammation-related pathways, including TNF and IL-17 signaling [Figure 6B], consistent with established pathways involved in OA progression and macrophage polarization^[39,40]. Subsequent analysis demonstrated that MSC-EVs downregulated pro-inflammatory cytokines Il1b and Il6, while upregulating key efferocytosis-related genes, such as Gas6^[6], Stab1^[53], Trem2^[53], and Cd36^[53] [Figure 6C]. These results indicated a shift toward a pro-resolutive phenotype that aligns with recent findings on efferocytosis-mediated inflammation resolution^[6,53]. The macrophage response to MSC-EVs diminished when BMSCs producing these EVs were transfected with a miR-21a-5p inhibitor, indicating that donor-cell miR-21a-5p functionally contributes to the MSC-EV-associated effects [Figure 6D].

To identify downstream targets of miR-21a-5p, we integrated DEGs with predictions from Elmmo, Diana_microT, and TarBase databases, identifying six candidate genes, including Smad7, Nfla, Yap1, Nflb, Arhgap24, and Tgfbi [Figure 6E]. Among the six candidate target genes (Smad7, Nfla, Yap1, Nflb, Arhgap24, and Tgfbi), only Yap1 expression was significantly reversed by miR-21a-5p inhibition following MSC-exosome treatment, supporting Yap1 as a potential downstream target of miR-21a-5p [Supplementary Figure 11]. Yap1 exhibited a significant decrease in mRNA expression after MSC-EV treatment, which was counteracted by inhibiting miR-21a-5p. Western blot analysis showed a concordant, although more modest, reduction in YAP1 protein abundance [Figure 6F and G, Supplementary Figure 12]. The difference in the magnitude of mRNA and protein changes may reflect differences in transcript and protein turnover kinetics as well as post-translational regulation of YAP1. Bioinformatic analysis predicted miR-21a-5p binding sites in the 3’UTR of Yap1, as illustrated in Figure 6H. This was corroborated by luciferase reporter assays, which demonstrated decreased activity with the WT 3’UTR upon miR-21a-5p overexpression, as shown in Figure 6I. The inverse relationship between miR-21a-5p and YAP1 was further validated by IF staining, which showed LPS-induced YAP1 upregulation was suppressed by miR-21a-5p mimics [Supplementary Figure 13A and B]. To further clarify the relationship between Yap1-related efferocytosis and inflammation phenotype in macrophages, we performed Yap1 overexpression using plasmids in macrophages. Functional studies demonstrated that while miR-21a-5p mimics restored efferocytosis in LPS-treated macrophages, overexpression of Yap1 completely abolished this effect [Figure 6J and K], indicating that YAP1 suppression is essential for miR-21a-5p-mediated functional recovery in macrophages. The miR-21a-5p/Yap1 axis was shown to regulate macrophage efferocytosis and inflammation resolution, as confirmed by efferocytosis assays and RT-qPCR analysis [Figure 6J and K, Supplementary Figure 14]. This finding is particularly significant given emerging evidence that YAP/TAZ signaling promotes pro-inflammatory macrophage polarization^[54,55]. Further, we evaluated the effects of Yap1 on macrophage efferocytosis and inflammation resolution [Figure 6L]. siRNA-mediated silencing of Yap1 in inflammatory macrophages enhanced efferocytosis capacity and reduced Il1b expression while increasing Mertk and Gas6 levels [Figure 6M and N], phenocopying the effects of MSC-EVs treatment and providing compelling evidence that YAP1 acts as a critical negative regulator of efferocytosis.

These findings underscore the significance of the miR-21a-5p/YAP1 axis in MSC-EV-mediated regulation of macrophage efferocytosis and inflammatory phenotype, suggesting its potential as a therapeutic target for restoring immune homeostasis in OA.

Building on our in vitro findings that highlight the role of the miR-21a-5p/YAP1 axis in controlling macrophage efferocytosis and inflammatory responses, we proceeded to investigate the impact of miR-21a-5p supplementation in vivo. The experimental design is shown in Figure 7A. Mice were assigned to four groups: Sham + NC, Sham + miR-21a-5p mimic, DMM + NC, and DMM + miR-21a-5p mimic. Eight weeks post-surgery, micro-CT analysis revealed no notable differences between the two sham groups. DMM mice treated with the miR-21a-5p mimic showed a significant reduction in OA-related structural changes, such as decreased osteophyte volume and less calcified tissue formation, compared to DMM + NC mice [Figure 7B and C, Supplementary Figure 15]. Histopathological analysis further showed better-preserved cartilage architecture, reduced surface fibrillation, and lower OARSI, osteophyte, and synovitis scores in the DMM + miR-21a-5p mimic group than in the DMM + NC group [Figure 7D and E].

At the molecular level, miR-21a-5p mimic treatment reduced synovial iNOS expression and decreased chondrocyte apoptosis [Figure 7F-I]. Analysis of ECM homeostasis showed reduced Aggrecan and increased MMP3 expression in DMM mice, whereas miR-21a-5p mimic administration partially restored these changes [Figure 7J-L]. In the synovium, the proportion of YAP1-positive macrophages was significantly reduced following miR-21a-5p mimic treatment [Figure 7M and N]. Consistently, gene expression analysis of synovial tissue showed increased expression of the efferocytosis-related genes Gas6, Mertk, and Stab2, together with reduced Yap1, Inos, and Il1b expression and increased Il10 expression in miR-21a-5p mimic-treated mice [Figure 7O and Supplementary Figure 16]. Collectively, these findings indicate that miR-21a-5p supplementation is associated with improved efferocytosis-related and inflammatory molecular profiles in DMM-induced OA.

To further assess the functional contribution of miR-21a-5p to the in vivo effects associated with MSC-EV treatment, purified MSC-EVs were co-administered intra-articularly with antagomir-miR-21a-5p [Supplementary Figure 17]. At 8 weeks after surgery, micro-CT analysis revealed no significant difference between the Sham + MSC-EVs and Sham + MSC-EVs + antagomir-miR-21a-5p groups. In DMM-induced OA mice, however, co-administration of antagomir-miR-21a-5p attenuated the structural protection associated with MSC-EV treatment, as reflected by increased osteophyte volume and greater calcified tissue formation compared with the DMM + MSC-EVs group [Supplementary Figure 17B and C]. SO/FG revealed increased cartilage degeneration in mice treated with MSC-EVs and antagomir-miR-21a-5p, characterized by significant articular surface disruption and diminished cartilage integrity [Supplementary Figure 17D and E]. These findings support a functional contribution of miR-21a-5p to the protective effects associated with MSC-EV administration in vivo. The experiment does not confirm the EV-specific origin of the inhibited miR-21a-5p, as antagomir-miR-21a-5p was co-administered locally with MSC-EVs instead of selectively depleting miR-21a-5p from the EV preparations.