Adipose-derived extracellular vesicles miR-423-5p targets SMAD3 to reverse endometrial fibrosis in a porcine model of intrauterine adhesion: insights from single-cell sequencing Authors: Xiao Xu, Xiaomin Wang, Quan Tian, Junlin Lv, Nailong Pan, Zhen Shang, Yiwei Xu, Xiaotong Li, Xiaoyan Li, Teng Wang, Yanjun Liu, Liang Zhang, Xiaoying Kong, Xiaodan Hao, Dan Han, Wenhua Xu Published: 2026 ABSTRACT AbstractIntrauterine adhesion (IUA), a fibrotic cause of female infertility, lacks effective therapies due to high recurrence following surgical intervention. Extracellular vesicles derived from mesenchymal stem cells (MSC-EVs) show promise in small-animal models, yet clinical translation is hindered by poor retention and the absence of large-animal validation. To address this, we developed a thermosensitive chitosan hydrogel (GelCT) as an EV carrier for autologous adipose MSC-EV delivery in a porcine IUA model. EVs-GelCT significantly attenuated fibrosis, restored endometrial architecture, and enhanced vascularization. Single-cell transcriptomics revealed that EVs remodel fibroblast subpopulations, specifically suppressing pro-fibrotic F2 (ECM-reparative) and F3 (inflammation-associated) clusters. Integrated miRNA profiling identified miR-423-5p as a functional EV cargo that directly targets SMAD3 3′UTR, thereby inhibiting TGF-β/Smad signaling. This first-in-porcine demonstration establishes the miR-423-5p/SMAD3 axis as a therapeutically actionable mechanism for IUA, advancing EV-based strategies from proof of concept to clinical translation.Keywords: Intrauterine adhesions, Adipose-derived mesenchymal stem cells, Extracellular vesicles, Thermosensitive hydrogel, Single-cell RNA sequencing Introduction Intrauterine adhesions (IUA), a predominant etiology of female infertility, arise from aberrant endometrial repair after injury which triggers fibrosis and culminates in dense scar tissue [1,2]. Its incidence increases with the number of uterine procedures, particularly dilation and curettage after miscarriage or abortion, ranging from 2.8% to 45.5% [3,4]. Fibrosis plays a central role in this pathology. When the endometrial basal layer is injured, fibroblasts are activated and subsequently over-synthesize the extracellular matrix (ECM), especially collagen [[5], [6], [7], [8], [9]]. This process results in partial or complete obliteration of the uterine cavity and compression of glands and blood vessels, thereby interfering with embryo implantation and pregnancy maintenance. The presence of fibrosis not only complicates treatment but also leads to high postoperative recurrence rates and reduced pregnancy rates [[5], [6], [7], [8], [9]]. Clinically, most IUA cases are treated via transcervical resection of adhesions (TCRA). Postoperatively, adjunctive measures such as physical barriers (e.g., Cook balloon) or hormonal therapy (estrogen/progesterone) are employed to prevent re-adhesion; however, outcomes remain suboptimal [6,[10], [11], [12]]. Consequently, inhibiting fibrosis is crucial for effective IUA treatment. Extracellular vesicles (EVs), key secretory factors of mesenchymal stem cells (MSCs), play a vital role in intercellular signaling and tissue repair [13,14], and are critical components of IUA therapy. EVs derived from placental MSCs and bone marrow MSCs have been demonstrated to promote IUA repair [15,16]. Nevertheless, despite their notable efficacy, challenges persist. Direct vaginal administration of EVs fails to achieve prolonged retention due to fluid flow and smooth muscle peristalsis, resulting in low bioavailability and limited therapeutic effects16. Consequently, enhancing EV retention and efficacy within the uterine microenvironment is essential. Among the biomolecules carried by EVs, microRNAs (miRNAs) are the most abundant, participating in regulating downstream target genes to promote tissue repair [16,17]. However, the anti-fibrotic role of miRNAs encapsulated in autologous adipose-derived MSC-EVs (ADMSC-EVs) has not been validated in large animal models of IUA. Thermosensitive hydrogels, characterized by their excellent biocompatibility and biodegradability, possess a three-dimensional structure that mimics the ECM and supports cell proliferation. They are widely used in tissue engineering scaffolds, wound dressings, and drug delivery carriers [7]. Compared to other biomaterials, their morphology is temperature-regulated, enabling a sol-gel transition that provides a foundation for sustained drug release. In previous research, we successfully synthesized hydrogels using natural polymers [18]. The porous 3D structure of the hydrogel can adsorb and encapsulate EVs, prolonging their in vivo residence time, improving bioavailability, and providing mechanical support for endometrial repair, akin to balloon dilation therapy [7,19]. For IUA repair, thermosensitive hydrogels offer a promising platform for the delivery and retention of therapeutic EVs. In this study, we established an IUA animal model using large animals (Rongchang pigs) via transvaginal injury, aiming to develop an autologous porcine ADMSC-EVs (pADMSC-EVs) complex loaded on chitosan-based thermosensitive hydrogel (GelCT) to increase the thickness of the damaged endometrium and the number of glands, reduce fibrosis, and promote angiogenesis, thereby providing a basis for improving poor TCRA prognosis in clinical practice. In addition, we further analyzed the impact of the complex on endometrial regeneration and repair by combining single-cell RNA sequencing (scRNA-seq) and miRNA sequencing (miRNA-seq) technologies to explore the potential anti-fibrotic mechanisms. Materials and methods Animals Rongchang pigs aged 8-10 weeks (weighing 25 ± 5 kg) were purchased from the Institute of Biotechnology, Chongqing Academy of Animal Sciences. They were housed in the SPF-grade barrier facilities of the Sichuan Institute of Laboratory Animal Science, with an adequate supply of food. All animal experiments were conducted in accordance with the Guidelines for the Management of Laboratory Animals and approved by the Animal Care and Use Committee of the Sichuan Institute of Laboratory Animal Science (Approval No.: IACUC-2024-009). Isolation and identification of pADMSCs Sexually mature, unmated female pigs weighing approximately 25 kg and of SPF grade were selected. After anesthesia through intramuscular injection of Zoletil 50 (2-3 mL per pig), the pigs were fixed in the supine position. They were then maintained under continuous isoflurane inhalation anesthesia, and their vital signs were monitored. Under sterile conditions, the groin area of the lower abdomen was disinfected, and the inguinal fat pads were excised and placed in pre-cooled normal saline containing a 10% penicillin-streptomycin (PS) solution. After the operation, the incisions were sutured, and cefoperazone sodium and sulbactam sodium (1.0 g per pig) were injected for preventive infection treatment. The excised adipose tissue was rinsed with normal saline containing 10% penicillin - streptomycin solution, then minced into a paste with pieces smaller than 1 mm3. The tissue was centrifuged at 1500×g for 5 min at 22°C in 45 mL of normal saline with 5% PS solution to remove the oil film. The tissue was digested with an equal volume of 0.1% type I collagenase (0.05 g/50 mL basal medium) for 1 h at 37°C with shaking at 150 rpm. Digestion was terminated by adding complete medium, followed by centrifugation at 1500×g for 10 min. The pellet was resuspended and seeded into a T25 flask and cultured at 37°C, 5% CO2 atmosphere. The cells obtained were used to identify the cell-surface molecular markers of pADMSCs by flow cytometry. Trilineage differentiation (adipogenic, osteogenic, and chondrogenic) was performed to verify the staining of calcium nodules, lipid droplets, and chondrosphere sections. All the procedures were carried out according to the instructions of OriCell® Adipose MSC Osteogenic (GUXMD-90021), Adipogenic (GUXMD-90031), and Chondrogenic (GUXMD-90041). They were characterized based on morphology, multi-differentiation potential, and surface marker expression to confirm successful isolation. Isolation and characterization of pADMSC-EVs and evaluation of their sustained release from GelCT hydrogel Isolation and characterization of pADMSC-EVs When the cell confluence of the P2–P3 generations reached 80%, they were transferred to a serum-free medium. The supernatant was harvested after 48–72 h, centrifuged at 3000 g for 10 min at 4°C, filtered through a 0.22 μm filter, and then centrifuged again. The supernatant was transferred to a pre-wetted 100 kDa ultrafiltration tube (Millipore, MWCO 100 kDa) and centrifuged at 3500 g for 30 min at 4°C. After centrifuging the concentrated solution at 2000 g for 10 min at 4°C, the supernatant was collected and then centrifuged at 10000 g for 30 min at 4°C. Ultra-centrifugation (120000 g, 4°C, 70 min) was performed twice. The precipitate was then resuspended in pre-cooled PBS. The crude extract was passed through an EV Purification Filter Column (Yumeibo, UR90102) and centrifuged at 3000 g for 10 min at 4°C for purification. The liquid was collected at the bottom of the tube, aliquotted (50–100 μL), and stored at - 80°C to avoid repeated freeze–thaw cycles. The protein concentration in the exosomes was measured using a BCA protein quantification kit. The morphology and particle size of the particles were observed using TEM and NTA (ZetaView PMX 110). The surface markers CD63 (25682-1-AP, dilution 1:1000, Proteintech), CD81 (27855-1-AP, dilution 1:2000, Proteintech), and Calnexin (2679, dilution 1:1000, Cell Signaling Technology) were identified by Western blotting (WB). The exosomes obtained were stored at −80°C and used for experiments within one month. The EV suspension was quantified by BCA assay, and the final EV concentration was adjusted according to the required experimental dosage before administration. In vitro EV release kinetics from GelCT hydrogel Briefly, EVs were mixed with GelCT solution at a concentration of 100 μg/mL and allowed to form a hydrogel. The EVs-GelCT hydrogels were immersed in 2 mL of sterile PBS and incubated at 37°C. At predetermined time points (day 2, 4, 6, and 8), 200 μL of the release medium was collected and replaced with an equal volume of fresh PBS to maintain a constant volume. The concentration of released EVs in the collected supernatants was quantified using an EV-specific ELISA kit according to the manufacturer's instructions. The cumulative release percentage of EVs was calculated based on the initial amount of EVs incorporated into the GelCT hydrogel. Preparation of the GelCT hydrogel system The GelCT hydrogel system was prepared using solutions of chitosan (CTS), carboxymethyl chitosan (CMCTS), and β-glycerophosphate (GP). The hydrogel precursor was freeze-dried for storage and rehydrated prior to wound repair experiments [18]. Detection of cytocompatibility To assess the cytocompatibility of GelCT with pADMSCs, cells were seeded in 6-well plates and cultured at 37°C for 24 h. For the experimental group, the culture medium was replaced with GelCT extract (prepared by incubating GelCT in a cell culture medium for 48 h), while the control group used standard medium. The live/dead cell staining and cell scratch assays were carried out following the experimental procedures described in the literature [18]. Scanning electron microscopy (SEM) analysis of EVs-GelCT The GelCT and EVs-GelCT samples were frozen at −80°C for 12 h and then freeze-dried to a constant weight. The samples were quenched in liquid nitrogen and cryo-fractured. The cross-sections were fixed on the SEM stage with conductive adhesive and sputtered with a 5 nm gold/platinum coating to eliminate charging. The cross-sectional morphology was imaged using a Tescan field-emission SEM (Tescan, Czech Republic). The porosity was measured by ImageJ using the following formula: Fourier transform infrared spectroscopy (FTIR) spectroscopy of EVs-GelCT The procedures for the FTIR detection of EVs-GelCT were carried out as reported in the literature [20]. Characterization of the mechanical properties of EVs-GelCT Compression tests of GelCT and EVs-GelCT were conducted using a computer-controlled universal testing machine (Model CMT4503, SANS/MTS Systems (China) Co., Ltd., Shenzhen, China). Stress-strain curves were recorded to calculate compressive modulus. where, F = the stress at which the hydrogel sample sustains 80% damage. Shydrogel = the hydrogel sample's area of stress. Antibacterial evaluation of EVs-GelCT against Staphylococcus aureus (S. aureus) and Escherichia coli (E. coli) (colony counting method) The experimental steps for the antibacterial evaluation of EVs-GelCT against S. aureus and E. coli using the colony counting method were carried out as previously reported [21]. The antibacterial rate was calculated according to the following formula: C0 and Cs represent the colony-forming unit (CFU) counts of the control group and the EVs-GelCT group, respectively. Antibiofilm performance evaluation of EVs-GelCT against S. aureus and E. coli S. aureus and E. coli (at a concentration of 1 × 106 CFU/mL) were cultured at 37°C in 12-well plates for 24 h to form mature biofilms. After discarding the culture medium, fresh medium (control), CTS, CTH, or EVs-GelCT were added according to the experimental design respectively, and then co - incubated at 37°C for another 24 h to evaluate the ability of the materials to disrupt the existing biofilms. Subsequently, the suspended bacterial solution was gently aspirated, and the biofilms were washed slowly with PBS three times to remove planktonic bacteria while keeping the biofilms intact. The biofilm layers were fixed with 95% formalin for 15 min, stained with 0.1% crystal violet in the dark for 20 min, washed gently with PBS three times again, air -dried at room temperature for 1 h, and then photographed. Rheological characterization of EVs-GelCT The rheological properties of EVs-GelCT composites were characterized as previously [20]. In vivo organ toxicity evaluation After completion of the treatment, the pigs in each group were humanely euthanized. Immediately thereafter, the heart, liver, lungs, kidneys, and spleen were harvested and fixed in 4% paraformaldehyde. Subsequently, the tissues were routinely embedded in paraffin, sectioned continuously, and subjected to hematoxylin and eosin (H&E) staining to observe the histological morphology. Construction of the IUA pig model Rongchang pigs aged 6-8 weeks were housed under specified conditions (21-24°C, relative humidity 40%-60%, 12 h light/12 h dark cycle) with free access to food and water. Then, the pigs were randomly divided into 5 groups–sham group, IUA group, GelCT alone group, EVs alone group, and EVs-GelCT group. The wounds in the sham group were left untreated, and the IUA pig model was established during the estrous phase. Only animals confirmed to be in the estrous phase, without evidence of reproductive tract injury or infection, were included and randomly assigned to the experimental groups to minimize cycle-related variability and ensure the reliability of subsequent analyses. For anesthesia, pigs received an intramuscular injection of Zoletil 50 and were subsequently maintained under anesthesia with isoflurane during surgery. Blood oxygen saturation and blood pressure were continuously monitored using auricular monitoring devices. Subsequently, 4% phloroglucinol was administered intramuscularly to relax the cervix and facilitate the transvaginal procedure. After positioning on the operating table, mechanical injury was induced by gently scraping the uterine endometrium with a sterile curette. Subsequently, 95% ethanol was injected into the uterine cavity and retained for 5 min, followed by thorough irrigation with sterile normal saline (Fig. 3B–C). PADMSC-EVs promote endometrial repair and diminish fibrosis in injured endometrium. (A) Operational procedure for porcine uterine tissue collection. (B) Establishment of the porcine IUA model and an observation equipment. (C) The tool for modeling injury—a curette. (D) The morphological assessment of the endometrium at PWD60. (E) The uterine tissues as indicated were performed for HE staining (the solid arrows pointed to the gland; the star pointed to the edge lines of the muscular layer and the endometrial layer; the wedges pointed to the blood vessels). (F) Quantitative analysis of relative endometrial thickness (G) Quantitative analysis of the relative number of endometrial glands. (H) Uterine tissues as indicated were subjected to Masson staining. (I) Quantitative analysis of endometrial fibrosis. n = 5, ***p < 0.001, ****p < 0.0001, ns means non-significant. After successful model establishment, the EVs group directly received an 800 μL injection of EVs (1 mg/mL) into the uterine body, while the IUA group received an equal volume of saline. The GelCT group received a 1 mL direct injection. The EVs-GelCT group received a well-mixed injection of 800 μL EV (1 mg/mL) and 1 mL GelCT. After PWD45, peripheral venous blood samples were collected from the pigs for routine blood and biochemical examinations to evaluate the safety of the EVs. After two estrous cycles (PWD60), phloroglucinol was injected intramuscularly to relax the cervix, and endometrial morphology was observed through a transvaginally inserted bronchoscope. Fifteen Rongchang pigs (n = 3) were euthanized, and wound tissues were collected according to the procedure shown in Fig. S1 for the subsequent evaluation of the treatment effect. Tissue acquisition After morphological observation, tissues were obtained through surgery. Under anesthesia, a mid - lower abdominal incision (approximately 6 cm) was made to expose the pig's uterus and locate the modeled area (cervical region) (Fig. S1). H&E and Masson trichrome staining The collected tissues were fixed with 4% paraformaldehyde, then dehydrated and embedded. The sections were cut to a thickness of 3 μm and stained according to the reported procedures. A panoramic scanner was used for photographing and observation [19]. H&E staining showed the number of glands and the endometrial thickness. Masson staining showed a fibrotic area. ImageJ software (Rawak Software, Germany) was used for the numerical analysis. Sirius red staining The prepared paraffin sections were dewaxed and dehydrated in 100%, 90%, 80%, and 70% ethanol, respectively, and then stained with Sirius red (Solarbio, G1472, China). After staining, the sections were sealed with neutral gum. After the sections were air - dried, a panoramic scanner was used for photographing and observation. Immunohistochemistry staining The prepared paraffin sections were dewaxed and hydrated, and the antigens were retrieved. They were incubated with a 3% hydrogen peroxide solution at 37°C for 30 min to block endogenous peroxidase activity and remove non-specific binding. The primary antibodies α-MA (67735-1-Ig, dilution 1:3000; Proteintech), TGF-β (26155-1-AP, dilution 1:500; Proteintech), and CXCL10 (DF6417, dilution 1:100; Affinity) were added, and the samples were incubated overnight at 4°C. Subsequently, they were incubated with a ready-to-use secondary antibody for 30 min. Color development was carried out with diaminobenzidine (DAB) (Solarbio), and the cell nuclei were stained with hematoxylin (Solarbio). The sections were sealed with neutral gum. After air - drying, a panoramic scanner was used for photographing and observation. The immunoreactivity was quantified using Image-Pro Plus 6.0 software. Immunofluorescence staining Immunofluorescence staining was performed according to previous studies. For paraffin sections, dewaxing and rehydration steps were carried out. Subsequently, the cell culture slides were incubated with 5% BSA for 30 min and then stained with the antibody against CD31 (28083-1-AP, dilution 1:500; Proteintech) as indicated. DAPI was used for nuclear staining. Immunofluorescence staining images were obtained using a fluorescence microscope. Western blotting analysis Reduced sodium dodecyl sulfate–polyacrylamide gel electrophoresis was performed according to the experimental procedures reported in the literature [22]. Finally, protein bands were detected using the Omni-ECL™ ultrasensitive chemiluminescence detection kit (Vazyme, China) and visualized using the Amersham ImageQuant™ 800 WB imaging system (Cytiva, Sweden). Single-cell RNA sequencing The sequencing samples were derived from fresh uterine tissues obtained from the porcine IUA group and the EVs-GelCT group (n = 3). The tissues were cut into small pieces of 0.5 mm2, washed with 1×PBS, and then digested in a mixed enzymatic solution containing collagenase IV, papain, and DNase I on a shaker at 37°C for 20 min. After terminating the reaction, the cells were collected by filtration and centrifugation. Red blood cells and dead cells were removed using red blood cell lysis buffer and Dead Cell Removal MicroBeads. The cells were resuspended in PBS and purified by centrifugation. Finally, a cell suspension with cell viability > 85% and a concentration of 700–1200 cells/μL was obtained, which was detected by trypan blue staining and an automatic cell counter. Subsequently, library construction and sequencing were carried out. The dissociated cell suspension was subjected to single-cell capture using a 10x Chromium instrument (10x Genomics Chromium Single-Cell 3’ Kit v3) (expected to capture 10,000 cells). After cDNA amplification and library construction, paired-end 150 bp sequencing was performed on the NovaSeq 6000 platform (sequencing depth: 20,000 reads/cell). The raw data were converted to the FASTQ format by bcl2fastq and then aligned to the reference genome using Cell Ranger (https://www.10xgenomics.com/support/software/cell-ranger, version 7.2.0). After quality control, 42,914 high-quality cells were obtained from all groups (thresholds: gene expressed in ≥ 3 cells, number of genes per single cell ≥ 500, mitochondrial gene proportion < 25%). Seurat (v3.1.1) was used for log normalization, PCA dimensionality reduction (top 10 principal components), and SNN clustering analysis. Differentially expressed genes (DEGs) were screened by FindAllMarkers (logFC ≥ 0.26, expression proportion ≥ 10% of cells). The experiment and analysis were completed by Hangzhou Lianchuan Biotechnology Co., Ltd. Pathway enrichment analysis The “cluster profiler” R package (version 3.16.1) was used to explore the potential functions of DEGs through KEGG and GO analyses [23]. Pathways characterized by a p_adj value of less than 0.05 were considered to be significantly enriched. The important pathways identified were visually presented through bar charts. In the context of GO, the reference gene sets included categories described as CC, BP, and MF. MiRNA microarray analysis The QIAseq miRNA library kit (Qiagen, Germany) was used to construct the sequencing library. Then, each miRNA was amplified by cDNA synthesis and PCR. On average, each sample was sequenced 10 million times on the Illumina platform. The raw data were converted to the FASTQ format and verified using FastQC. Trimmomatic was used to filter the low-quality reads. The data were matched with the miRBase database to identify known miRNAs and then cross-referenced with the Rfam database to analyze the distribution of ncRNAs in the small RNAs obtained. The sequence data were compared with the whole-genome sequence of the species, and a folding model was used to predict potential novel miRNAs. Finally, the microRNAs were analyzed. In vivo validation of miR-423-5p function in a rat IUA model Thirty female Sprague-Dawley (SD) rats (8–10 weeks old) with normal estrous cycles were randomly divided into five groups (n = 5/group): Sham, IUA, IUA + NC-agomir/NC-antagomir + GelCT, IUA + miR-423-5p-agomir + GelCT, and IUA + miR-423-5p-antagomir + GelCT. The IUA model was established by mechanical injury of the endometrium under anesthesia. Briefly, a ∼3 mm incision was made in each uterine horn, followed by endometrial scraping in four directions and subsequent closure. After modeling, GelCT formulations containing miR-423-5p agomir, antagomir, or corresponding controls were locally administered into the uterine horns. Ten days after treatment, uterine tissues were collected for experimental evaluation. Statistical analysis Quantitative data are presented as the mean ± standard deviation. Comparisons among multiple groups were performed using the Kruskal–Wallis test followed by Dunn's multiple comparisons test. Statistical analyses were performed using GraphPad Prism 8.0 software. A P-value < 0.05 was considered statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; ns indicates no significant difference). Results Characterization of pADMSC-EVs and in vivo biosafety of the GelCT for intrauterine therapy Porcine ADMSCs were isolated from subcutaneous adipose tissue in the porcine inguinal region (Fig. 1A). They were characterized based on morphology, multi-differentiation potential, and surface marker expression to confirm successful isolation. Passage 1 (P1) cells exhibited adherent growth and displayed the typical spindle-shaped morphology characteristic of pADMSCs (Fig. 1B). Subsequently, compared with non-induced cells, Oil Red O staining showed that following induction, large and round positive lipid droplets with strong refractivity and red spherical shapes appeared (Fig. 1C). Alizarin Red staining results indicated that red staining points, which were calcium nodules, appeared inside the cells (Fig. 1D). Alcian Blue staining results showed that after approximately 24 days, the induced cartilage balls were paraffin-sectioned and stained blue, indicating the formation of mucinous polysaccharides in the cartilage tissue (Fig. 1E). These results confirmed the adipogenic, osteogenic, and chondrogenic differentiation abilities of the extracted stem cells. The P2–P3 cells were analyzed by flow cytometry for surface marker expression. The results showed positive expression for markers CD44 and CD105, while the expression of hematopoietic stem cell marker CD34 and macrophage marker CD11b was negative (Fig. 1F). In addition, immunofluorescence staining confirmed the expression of CD73 and CD90, further supporting the MSC phenotype of the isolated pADMSCs (Fig. S2A). Together, these findings confirmed the identity of the isolated pADMSCs. PADMSCs-EVs were identified using transmission electron microscopy (TEM), particle size analysis (NTA), and WB. The TEM results showed that EVs were in a “cup - shaped” round or oval form (Fig. 1G). NTA revealed that the particle size was approximately 153 nm (Fig. 1H). WB analysis showed significant expression of the typical markers CD81 and CD63 of pADMSCs-EVs (Fig. 1I). Characterization and evaluation of pADMSC-derived EVs and EVs-GelCT delivery system. (A) The operating procedures of EVs-GelCT. (B) Morphological observation of P1 pADMSCs. Scale bar: 200 μm (bottom), 50 μm (up). (C–E) The multidirectional differentiation potential of pADMSCs. The representative images of the non-induced (left) and induced pADMSCs (right) differentiated into adipocytes and osteocytes using Oil Red O (C) and Alizarin Red (D) staining. Scale bar: 200 μm (bottom), 50 μm (up); E: Representative images of induced pADMSCs differentiation into chondrocytes using Alcian Blue. Scale bar: 200 μm (bottom), 25 μm (up). (F) The flow cytometry of surface marker molecules about pADMSCs; (a) the negative marker CD34; (b) the positive marker CD105; (c) the negative CD11b (leukocyte marker); (d) the positive marker CD44; Blue trace: the isotype control IgG2a negative (excluding non-specific binding of Fc segment). (G–H) The EVs derived from pADMSCs were observed using TEM and NTA. (I) WB was used to detect the specific surface proteins CD63, CD81 and Calnexin. (J) Uptake of pADMSC-EVs by Fb. EVs were fluorescently labeled with PKH26 (red); Cytoskeleton was stained with phalloidin (green); Cell nuclei were stained with DAPI (blue); Scale bar: 10 μm. (K, L) Fluorescence image and intensity of the sustained EVs labeled with PKH26 in hEECs. (M) Quantification of EVs released from GelCT assessed via CD63 positive ELISA. Subsequently, PKH26-labeled EVs were encapsulated in GelCT to evaluate their retention and release behavior. As shown in Fig. 1J–L, the EVs were efficiently internalized by human dermal fibroblasts and endometrial epithelial cells (hEECs) within 24 h. Compared with free EVs, the EVs-GelCT group maintained higher intracellular fluorescence signals at 48 and 72 h, indicating enhanced EV retention and sustained release. Furthermore, ELISA analysis demonstrated that EVs were continuously released from GelCT over 8 days, confirming the sustained-release capacity of the hydrogel system (Fig. 1M). At 4°C, the GelCT hydrogel was in a fluid state, whereas at 37°C, it transformed into a gel state (Fig. 2A). Biocompatibility is a key parameter for evaluating the safety of biomaterials. To determine the safety of the hydrogel, we conducted cell scratch and live/dead staining experiments. In the scratch assay, the wound closure rates of the GelCT group (66.72% at 12 h and 28.73% at 24 h) were comparable to those of the control group, with no statistically significant difference (P > 0.05), indicating no adverse effects on cell migration (Fig. 2B). The live/dead cell staining results showed that the cell survival rates of both groups were higher than 90%, and the cells also demonstrated > 90% viability with typical MSC morphology (Fig. 2C). Simultaneously, we confirmed that EVs had no effect on the surface network structure, chemical composition, antibacterial effect, rheological and mechanical properties of GelCT (Fig. 2D–J and S2B–C). These experiments confirmed that GelCT can be used as a sustained-release carrier for EVs. The characterization, biocompatibility, and antibacterial properties of GelCT. (A) Morphological appearance of GelCT at different temperatures GelCT remained in a flowable state at 4°C and transitioned into a gel state at 37°C. (B) Representative images and quantitative analysis of the wound injury rate in the control and GelCT groups at 12 and 24 h. (C) Representative live/dead images and quantification of cell viability in the control and GelCT medium after 24, 48, and 72 h of culture. The data are presented as mean ± SD, ns means not significant, n ≥ 3. (D) Morphological characterization of GelCT by SEM and quantitative analysis of pore size distribution (n=3). (E) The FTIR spectra of GelCT and EVs-GelCT. Scale bar = 10 μm. (F–G) Antimicrobial effects and quantitative results of different groups against E. coli and S. aureus. (H–I) The biofilm formation ability and quantitative analysis. ns means non-significant. ****P < 0.0001. (J) Storage and loss modulus of GelCT measured over the indicated frequency range. (K) In vivo biosafety evaluation of EVs-GelCT. H&E staining of heart, liver, spleen, lung, and kidney in the control and GelCT groups. Scale bar = 200 μm. To assess the safety of EVs-GelCT therapy, peripheral blood was collected from pigs at post-wounding day 45 (PWD 45) for the analysis of hematological parameters and serum biochemistry. The intrauterine administration of EVs-GelCT did not induce abnormalities in porcine blood cell counts or hematocrit levels compared to the sham group (Table S1). Similarly, key biochemical indicators such as aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine (Cre), and blood urea nitrogen (BUN) remained within the normal range. The analysis showed no significant differences in the serum levels of ALT, AST, Cre, and BUN between the groups (Table S2). To comprehensively evaluate the biosafety of EVs-GelCT, heart, liver, spleen, lung, and kidney tissues from different treatment groups were subjected to H&E staining to detect pathological and structural changes. The results revealed no abnormalities in organ architecture, providing preliminary evidence for the non-toxicity of this therapeutic approach (Fig. 2K). EVs-GelCT restores endometrial architecture, promotes angiogenesis, and attenuates fibrosis in a porcine IUA model To determine the therapeutic efficacy of EVs-GelCT in IUA, we applied the formulation to a double-injury IUA model induced by mechanical curettage followed by 95% ethanol exposure (Fig. 3A–C and S1). The results showed that on PWD 60, the IUA group exhibited more filamentous or flaky adhesions within the uterine cavity, with increased thickness and density (Fig. 3D). Some areas presented dense fibrotic changes, appearing whitish and pale yellow, with increased, tortuous, dilated, or clustered blood vessels. The groups treated with pADMSC-EVs or GelCT alone showed milder conditions. In contrast, the EVs-GelCT group displayed a pale pink endometrial mucosa, smooth surface, and regular folds, suggesting preliminary evidence of favorable repair outcomes (Fig. 3D). H&E staining was used to assess uterine morphology and to measure endometrial thickness and gland number (Fig. 3E). The average endometrial thickness in the sham group was 3.87 mm, which decreased to 1.34 mm in the IUA group. Comparatively, the EVs-GelCT group exhibited a thickness of 4.35 mm, which is significantly higher than the IUA group (Fig. 3F). The gland structures were intact and continuously distributed in the sham group. Compared to the IUA group, the EVs-GelCT group had significantly more glands (332 vs. 120, P < 0.0001) with relatively intact structures (Fig. 3E and G). Masson's trichrome staining showed that the degree of fibrosis in the IUA group was significantly increased (Fig. 3H). In the other groups, the degree of fibrosis was reduced to a certain extent, and the improvement was most significant in the EVs-GelCT group (Fig. 3H–I). Immunohistochemical staining revealed that the expression of the fibrotic marker α-SMA, TGF-β were significantly reduced in the EVs-GelCT group (Fig. 4C and D), which was further confirmed by Sirius red staining (Fig. 4E). Subsequently, we detected the expression of the CD31 protein and found that there were a large number of CD31-positive signals in the EVs-GelCT group, whereas the IUA sections exhibited minimal CD31 signal. Compared with the IUA group, the CD31 signals in the other control groups were partially restored (Fig. 4F–G and S2D). To further investigate whether EVs-GelCT treatment contributes to the restoration of endometrial receptivity, we evaluated the expression of HOXA10 and ITGAV, two critical molecular markers. The results revealed that EVs-GelCT treatment significantly restored the expression levels of both HOXA10 and ITGAV in the regenerated endometrium, suggesting enhanced endometrial receptivity and the establishment of a more favorable microenvironment for embryo implantation (Fig. 4H–J and S2E). CD81 immunohistochemical staining further demonstrated that EVs-GelCT significantly enhanced EV retention in injured uterine tissues compared with free EVs (Fig. 4K–L). The assessment of endometrial repair via Sirius red staining and immunofluorescence at PWD60 treatment. (A, B) Immunohistochemical staining and quantitative analysis of α-SMA expression. (C, D) Immunohistochemical staining and quantitative analysis of TGF-β expression. (E) Representative images of Sirius red staining of uterine tissues from the indicated groups. Scale bar = 200 μm. (F, G) Immunofluorescence staining and quantitative analysis of CD31 expression. Scale bar = 100 μm. (H-J) Immunofluorescence staining and quantitative analysis of HOXA10 and ITGAV expression Scale bar = 200 μm. (K–L) Immunofluorescence staining and quantitative analysis of CD81 expression. Data are presented as mean ± SD. n = 5, *p < 0.05, **p < 0.01, and ***p < 0.001. Collectively, these data demonstrate that the intrauterine administration of EVs-GelCT can significantly increase endometrial thickness and glandular density while suppressing fibrosis. ScRNA-seq reveals that EVs–GelCT cross-linking markedly downregulates fibroblasts in IUA uterine tissue We used scRNA-seq to investigate the effect of EVs-GelCT on wound cellular composition at PWD60 (Fig. 5A). After quality control, 42,914 high-quality cells were retained, including 20,039 cells from the IUA group and 22,875 cells from the EVs-GelCT group (Fig. S3, Table S3). After dimensionality reduction and clustering, ten distinct cell types were annotated: B cells (Bc, MS41A1high/CD79Ahigh/CD79Bhigh), epithelial cells (Etl, EDN1high/PECAM1high/EMCNhigh), endothelial cells (Edt, EPCAMhigh/KRT8high/SAL1high), fibroblasts (Frb, COL1A1high/COL1A2high/DCNhigh), lymphocytes (Lec, LYVE1high/CCL21high/FABP5high), macrophages (Mrp, CD74high/C1QAhigh/C1QBhigh), MSCs (MSC, HMGB2high/H2AC4high/PCLAFhigh), neutrophils (Ntp, LYZhigh/S100A8high/S100A12high), smooth muscle cells (SMC, ACAT2high/ITH4high/TPM2high), and T cells (Tc, CD3Ehigh/CD3Dhigh/TRAT1high). These were visualized using Uniform Manifold Approximation and Projection (UMAP) plots (Fig. 5B–D). Cell cluster identities were validated through highly expressed signature genes and canonical markers (Fig. 5B–D). EVs-GelCT modulates cellular composition and tissue repair-associated responses in full-thickness injured porcine uterine tissue. (A) Schematic illustration of the scRNA-seq workflow. (B) UMAP dimensional reduction representing cells categorized into ten main clusters with each cell color-coded based on its cell type. (C) Table along with bar plots showing cell populations of each cluster presented by percentages of total sequenced cells or percentages of specific cell type across treatments. The red indicates cell populations of interest. Pale green shows populations enriched in the group IUA, green shows populations enriched in the group EVs-GelCT. (D) Dot plot showing the expression of canonical marker genes for each identified cell type. Dot size represents the proportion of cells expressing each gene, and dot color represents the expression level. Bubble chart of total cells showing expression of canonical markers of each cell type. ((E) UMAP visualization of cell type-associated marker genes. All data presented as mean ± SD. Source data are provided as a source data file. Fibroblasts dominated the wound cell population (Fig. 5B and C), averaging 31.7% of the total number of cells. The highest proportion was observed in the IUA group (45.6%), whereas the EVs-GelCT treatment reduced this to 17.8%, indicating an effective attenuation of endometrial fibrosis, consistent with the tissue-repair observations described above (Fig. 5C). Macrophages accounted for an average of 5.91% of the total cells, and the proportions in the two groups were almost the same (5.23% in the IUA group vs. 6.01% in the EVs-GelCT group) (Fig. 5C). Meanwhile, the feature plots of the key functional genes were generated to show the functional characteristics of each cluster, which served as a reference for the final annotation. Among them, the markers of different major cell categories, including CD79B, PECAM1, EPCAM, COL3A1, LYVE1, CD68, HMGB2, CSF3R, ACTA2, and CD3E, further demonstrated the distribution of each cell population (Fig. 5E). Overall, these results demonstrate that EVs-GelCT treatment remodels the uterine microenvironment in porcine samples, with the most significant difference being Frb abundance. EVs-GelCT attenuates fibrosis through suppression of F2 (ECM) and F3 (inflammatory) fibroblast populations Next, we investigated the regulatory effect of EVs-GelCT crosslinking on the phenotype of Frb. Based on the functional characterization of differentially expressed genes (DEGs), the Frb population (COL1A11high/COL1A2high/DCNhigh) could be divided into four distinct functional subpopulations: tissue-reparative Frb (F1), ECM–reparative Frb (F2), inflammation-associated Frb (F3), and myofibroblasts (F4) (Fig. 6A and B and S4A–B, Table S4). These subtypes exhibited significant differences between the EVs-GelCT and IUA groups, and the specific analysis is as follows. EVs-GelCT attenuates fibrosis by suppressing ECM-associated F2-Frb and inflammation-related F3-Frb populations. (A) UMAP visualization of Frb cells from IUA and EVs-GelCT-treated groups, showing four distinct subpopulations with cells colored according to their assigned clusters. (B) Table and bar plots showing the proportions of Frb in each cluster, presented as percentages of total sequenced cells and percentages of specific cell populations across treatments. Color indicates the enriched treatment: pale green (IUA), green (EVs-GelCT). (C) Dot plot showing enriched GO biological processes of highly expressed genes in each Frb subpopulation. Dot size represents the proportion of cells expressing the corresponding genes, and dot color represents the GO enrichment p-value. (D) UMAP visualization of cell category-associated marker genes. (E) Pseudotime trajectory analysis of Frb subpopulations. (F–G) Volcano plots showing differential gene expression analysis of F2-Frb and F3-Frb populations. (H) Venn diagram showing overlapping downstream differentially expressed genes (DEGs) of F2-Frb and F3-Frb populations. (I) Quantitative analysis of CXCL10 immunohistochemical staining. Data are presented as mean ± SD. n = 3; **p < 0.01. (J) KEGG pathway enrichment analysis of DEGs in F2/F3-Frb populations between EVs-GelCT and IUA groups. F1 (RORB high/APCDD1 high/HAPLN1high) was prominent in both groups, accounting for an average of 21.3% of Frb. The gene expression pattern of F1 was markedly reduced after therapy (27.95% IUA vs. 14.68% EVs–GelCT), consistent with a transient ECM scaffold function that must be terminated to avert fibrosis (Fig. 6C and S4D). Compared with the IUA-Frb population, the gene expression levels of subpopulations F2 and F3 were significantly downregulated in EVs-GelCT (Fig. 6B). F2 was more common in IUA-Frb (12.26%) than in EVs-GelCT-Frb (2.08%). It had the ECM component marker MFAP5, vascular smooth muscle cell marker MGP, and cell proliferation marker SFRP4, which was consistent with the Frb subtype and related Gene Ontology (GO) terms for ECM tissue and structural repair. This subtype mainly responds to damage signals and blocks the expression of ECM-related genes, thereby inhibiting repair (Fig. 6C and S4D). The gene expression level of F3 was also significantly downregulated in EVs-GelCT-Frb (0.86%) compared to IUA-Frb (5.08%). This inflammation-associated Frb cluster expressed chemotaxis-related gene CXCL10, cell immune regulation-related genes GBP1 and HLA-DRA, and genes involved in the GO terms of signal transduction, inflammatory response, and transcription factor activity. This primarily drives the production of inflammatory factors, leading to fibrosis (Fig. 6C and S4D). Further tracking of the markers RORB, MGP, CXCL10, and A2M of different Frb subgroups showed the location of each group (Fig. 6D). In EVs-GelCT-Frb, F4 was up-regulated, specifically expressing genes related to wound contraction-closure and scar formation (fibrosis-related marker A2M, immune cell marker C7, and tumor stem cell marker ALDH1A1) (Fig. 6C and S4D). Based on the above data analysis, we first performed dimensionality reduction and clustering on Frb using Seurat to subdivide the subpopulations, and then conducted Monocle2 pseudotime analysis on the purified subpopulations (Fig. 6E and S4C). By tracking gene expression dynamics along the pseudotemporal axis, we categorized temporal DEGs from five cell types—MSCs, F1, F2, F3, and F4—into three phases– early-, mid-, and late-stage highly expressed genes (corresponding to green, red, and blue). These expression profiles revealed functional states across the inferred trajectories. The analysis demonstrated that EVs-GelCT could recruit MSCs to the tissue injury site and then induce their sequential differentiation into F2, F3, and F4, and finally into F1 (Fig. 6E and S4E). This differentiation trajectory clearly demonstrated that during the tissue repair process mediated by the EVs-GelCT complex, MSCs could be recruited and differentiated into different types of mature Frb. In summary, from the perspective of fibrosis, EVs-GelCT treatment significantly downregulates profibrotic subtypes F2/F3-Frb, which play pivotal roles in attenuating pathological fibrosis. To clarify the role in the anti-fibrosis process, we further presented the DEGs of the two subtypes, ECM–reparative F2–Frb and inflammation-associated F3-Frb, to explore their in-depth repair relationships. Compared with the IUA group, the EVs–GelCT group showed 1242 DEGs in the F2-Frb subtype (530 upregulated, 713 downregulated; Log2FoldChange cut-off = ±0.26) (Fig. 6F and Table S5). In the F3-Frb subtype of the EVs-GelCT group, 995 DEGs were identified (480 upregulated, 515 downregulated) (Fig. 6G and Table S6). Notably, the important marker CXCL10 was downregulated in both subtypes (Fig. 6H). Furthermore, the immunohistochemical staining results further showed that the expression of CXCL10 was down-regulated to a certain extent in the IUA wound tissue treated with EVs-GelCT (Fig. 6I and S5A). We also explored the communication relationships between F2/F3-Frb and other cells using the CellPhoneDB database25. The results indicated that, compared with the IUA group, EVs-GelCT treatment significantly promoted the crosstalk between MSCs and F2/F3-Frb subtypes. This may imply that EVs may play an anti-fibrosis role by recruiting MSCs, which undergo directional chemotactic migration to the damaged tissue, colonizing the environment and surviving in it (Fig. S4B–D). Meanwhile, the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis showed that these DEGs modulate the TGF-β/SMAD pathway (Fig. 6J), which is a core fibrotic signaling cascade targeted by EVs-GelCT. MiRNA-seq reveals that EVs attenuate fibrosis via miR-423-5p-mediated suppression of Smad3 in the TGF-β/smad pathway Initially, WB was employed to assess the protein expression levels of α-SMA, COL1A1, and CXCL10 in EECs treated with EVs-GelCT, aiming to evaluate the effects on fibrosis improvement. The results showed that while the fibrosis markers were significantly increased in EECs treated with TGF-β1, they were significantly reduced after a 48 h treatment with EVs-GelCT (Fig. 7A and B). The CCK-8 assay and scratch test confirmed that EVs-GelCT could significantly promote the proliferation and migration of EECs (Fig. 7C–E). These results confirm that EVs-GelCT can alleviate fibrosis in EECs and promote their proliferation and migration in vitro. EVs-GelCT attenuate TGF-β1-induced fibrosis and promote proliferation and migration of EECs. (A, B) Expression levels of fibrosis-associated proteins in EECs were assessed by WB. (C) EEC proliferation was evaluated using the CCK-8 assay. (D, E) EEC migration was assessed by a wound healing assay. (F) The top 30 miRNAs identified in pADMSC-EVs. (G) Relative expression profiles of the top 30 miRNAs from the miRNA microarray dataset in the GEO database (GSE160634). (H) KEGG pathway analysis performed using DIANA-miRPath v3.0 based on the top 30 miRNAs identified in pADMSC-EVs. (I) miRNA–mRNA interaction network. miRNAs and mRNAs are represented by red and green nodes, respectively. (J) Protein–protein interaction (PPI) network. Node size represents the degree value. (K, L) WB analysis of the phosphorylation levels of SMAD2 and SMAD3. The miRNAs in EVs-GelCT were profiled using miRNA-seq (Fig. 7F). The IUA miRNA expression profile was obtained from the microarray dataset GSE160634 in the GEO database. Notably, among the top 30 miRNAs in pADMSC-EVs, 25 were downregulated, 4 were upregulated, and 1 showed no change in expression in the IUA samples (Fig. 7G). To clarify the functional characteristics of the top 30 miRNAs in pADMSC-EVs, the KEGG pathway analysis revealed their involvement in regulating the TGF-β/Smad pathway (Fig. 7H). By using TargetScan, GO analysis, and PPI network analysis to predict the target genes and related functions of the obtained miRNAs, their potential roles were determined, and the interaction maps of the miRNA–mRNA network and PPI network were drawn. The results showed that the biological processes (BP), cellular components (CC) and molecular functions (MF) related to TGF-β/Smad were enriched (Fig. S6). Notably, miRNAs could directly target key genes in TGF-β/Smad, such as TGFBR1, TGFBR2, and Smad2 (Fig. 7I). In the PPI network analysis of these miRNAs, Smad2 and Smad3 had the highest node degrees and served as hub genes (Fig. 7J). The WB experiment further confirmed these results. After treatment with TGF-β1, while phospho-Smad2/3 were significantly increased, their expression levels were significantly decreased after treatment with EVs-GelCT (Fig. 7K and L). The TargetScan database analysis predicted 76 miRNAs targeting Smad2 and 60 miRNAs targeting Smad3. Among them, only miR-423-5p was one of the top 30 miRNAs in EVs and significantly down-regulated in IUA (Fig. 8A), suggesting its key role in the regulation of the TGF-β/Smad pathway. The significantly increased expression level of miR-423-5p after treatment with EVs-GelCT indicated that it could be effectively delivered to EECs through EVs (Fig. 8B). In addition, transfection with miR-423-5p mimics could also effectively alleviate TGF-β1-induced fibrosis, which was confirmed by the protein expression of α-SMA, COL1A1, and CXCL10 (Fig. 8C and D). Conversely, miR-423-5p inhibition aggravated fibrotic responses and partially diminished the anti-fibrotic effects of EVs-GelCT, further suggesting that miR-423-5p is an important functional mediator contributing to the therapeutic effects of EVs-GelCT (Fig. S7A–B). The effects of miR-423-5p on the proliferation and migration of EECs showed that transfection with miR-423-5p mimics did not significantly affect the proliferation and migration of EECs. In contrast, miR-423-5p inhibition impaired the proliferative and migratory capacity of EECs, further supporting the functional role of miR-423-5p in endometrial repair (Fig. 8E–G). Therefore, EVs could exert therapeutic effects by delivering miR-423-5p to EECs. EVs–GelCT inhibit SMAD3 phosphorylation by delivering miR-423-5p, thereby suppressing the TGF-β/Smad pathway and attenuating fibrosis. (A) Venn diagram showing the candidate miRNAs targeting Smad2 and Smad3 identified in EVs–GelCT. (B) Expression level of miR-423-5p in EECs after treatment with EVs–GelCT. (C, D) Effects of miR-423-5p overexpression on the expression of fibrosis-related proteins in EECs. (E-G) Effects of miR-423-5p mimics and inhibitor transfection on the migration and proliferation of EECs. (H) Dual-luciferase reporter assays were performed in 293T cells co-transfected with miR-423-5p mimics and wild-type (WT) or mutant (MUT) Smad3 3′-UTR reporter vectors. After transfection of miR-423-5p mimics and WT or MUT vectors into 293T, dual-luciferase reporter assays were performed. (I) WB was performed to detect p-SMAD3 expression in EECs. The dual-luciferase reporter gene assay showed that miR-423-5p could bind to the 3′-UTR of Smad3 (Fig. 8H). Moreover, we also detected the expression level of p-Smad3 after transfection with miR-423-5p mimics. As shown in Fig. 8I, compared with the TGF-β1 group, miR-423-5p significantly reduced the protein expression of p-Smad3. Conversely, miR-423-5p inhibition increased Smad3 phosphorylation, further supporting the negative regulation of TGF-β/Smad signaling by miR-423-5p through targeting Smad3 (Fig. S7C–D). To further validate the regulatory effect of miR-423-5p on the TGF-β/SMAD3 signaling pathway in vivo, functional validation experiments were performed in a rat IUA model. qRT-PCR analysis showed that miR-423-5p overexpression significantly reduced the expression levels of fibrosis-related genes, including α-SMA and COL1A1, whereas inhibition of miR-423-5p resulted in increased expression of these fibrotic markers (Fig. S8A). Immunofluorescence staining further demonstrated that miR-423-5p agomir treatment markedly decreased p-SMAD3 expression without affecting total SMAD3 levels, whereas miR-423-5p antagomir treatment significantly increased p-SMAD3 expression (Fig. S8B). These findings indicate that miR-423-5p alleviates endometrial fibrosis through inhibition of SMAD3 phosphorylation. These results indicate that pADMSC-EVs released from EVs–GelCT inhibit Smad3 phosphorylation by delivering miR-423-5p, thereby disrupting the TGF-β/Smad pathway and reducing fibrosis. Discussion In this study, EVs-GelCT significantly improved the endometrial structure in a porcine IUA model, increasing endometrial thickness and the number of glands. In-depth mechanistic studies revealed that miR-423-5p released by EVs-GelCT specifically targeted and inhibited the phosphorylation of Smad3, effectively blocking the TGF-β/Smad fibrotic signaling pathway. This innovative finding provides a novel strategy for the treatment of IUA. Currently, standard treatments for IUA, such as hysteroscopic adhesiolysis, often face the challenge of high recurrence rates [12]. In recent years, treatment strategies based on hydrogels or EVs have attracted extensive attention in the scientific community due to their great potential in tissue regeneration and immune regulation [[24], [25], [26], [27]]. However, the underlying mechanisms of reversing fibrosis during endometrial reconstruction remain unclear. In response to this critical challenge, EVs, which are enriched with various bioactive molecules (such as miRNAs and proteins), have emerged as a promising therapeutic strategy due to their ability to regulate cell proliferation, immune responses, and fibrotic processes [13,14,28]. However, the rapid clearance, short-term retention at the target site, and limited delivery efficiency of EVs in vivo significantly restrict their therapeutic efficacy. To overcome these limitations, hydrogel-based EV delivery systems have attracted increasing attention. Multifunctional hydrogels not only enable sustained EV release at the injury site, thereby improving EV stability and bioavailability, but also enhance EV-mediated biological functions by establishing a favorable local microenvironment [26,27]. Despite these advances, the underlying mechanisms by which hydrogel-mediated sustained EV delivery platforms contribute to the reversal of fibrosis in IUA remain insufficiently understood, and existing studies have largely been limited to small animal models, with validation in large animal models remaining scarce [29]. Therefore, in this study, we employed a porcine IUA model to evaluate an EVs-GelCT hydrogel delivery system and demonstrated that this platform promotes the reversal of IUA-associated fibrosis and functional endometrial regeneration by enabling sustained EV release and remodeling the fibrotic microenvironment. We found that EVs-GelCT can achieve a precise balance between endometrial repair and anti-fibrosis during the IUA repair cycle by regulating fibroblast subpopulations. F1, as the terminally differentiated subpopulation of fibroblasts, plays a crucial role in the middle and late stages of IUA repair. Relying on its mature differentiation characteristics, it constructs a temporary ECM, providing a structural scaffold for tissue repair and inhibiting pathological fibrosis [30]. The EVs-GelCT regulates the directional differentiation of MSCs into F1 and moderately controls its expression, promoting the transition of repair from structural construction to mature remodeling and ensuring the normal reconstruction of the endometrium. F2 and F3 are the early intermediate subpopulations of MSC differentiation. In the early stage of IUA repair, they are involved in the initial repair of the ECM, the conduction of inflammatory signals, and the regulation of immune cell chemotaxis. However, if their expression remains high in the middle and late stages, F2 will lead to excessive deposition of the ECM, and F3 will induce chronic inflammation. The two subpopulations act synergistically to exacerbate fibrosis [19,31]. The EVs-GelCT precisely inhibits the abnormal high expression of F2 and F3 in the middle and late stages, blocks the fibrotic cascade reaction, and promotes ECM remodeling and inflammation regression. This result is consistent with previous research findings, indicating that MSC-EVs can regulate fibroblast phenotypes and reduce fibrosis [24]. Mechanistically, miR-423-5p carried by the EVs-GelCT targets and inhibits the phosphorylation of Smad3, blocking the core fibrotic TGF-β/Smad pathway. Although miR-423-5p has been proven to have a protective effect against fibrosis in kidney injury and other conditions [32], this study is the first to clarify its anti-fibrotic function in IUA through the EV-released miR-423-5p, which regulates the sequential differentiation of MSCs into fibroblast subpopulations and the stage-specific expression of each subpopulation. However, given the complex biological composition of EVs, other candidate miRNAs may also participate in the therapeutic effects of EVs through independent or synergistic mechanisms. The specific functions and potential interactions of these miRNAs are currently being explored. The CXCL10 gene was found among the downregulated DEGs in the F2 and F3 subsets. As previously reported, CXCL10, a C-X-C motif chemokine, plays a crucial role in the fibrotic process of IUA. When cells are damaged, CXCL10 is released by epithelial cells and can bind to CXCR3, guiding cells to migrate to the fibrotic site [33]. Subsequently, these cells release pro-fibrotic factors, such as TGF-β, thus initiating a fibrotic cascade reaction, such as the activation of the TGF-β/Smad signaling pathway, which leads to excessive ECM deposition and fibrosis formation [34]. Notably, DPP4 truncates two terminal amino acids at the N-terminus of CXCL10, generating an inactivated form of CXCL10 [35]. Although this antagonist form of CXCL10 can still bind to the CXCR3 receptor, it may not induce IUA-related fibrosis and, thus, promote wound repair. This discovery provides new insights into the potential application of CXCL10 in fibrosis treatment, and its underlying mechanisms are still under further investigation. Our study innovatively used a large-animal model, the Rongchang pig. Compared with small-animal models, large animals have significant advantages in anatomical structure, pathological processes, and clinical translational relevance [21]. Currently, only one study has reported the in situ repair capacity of a complex of stem cell–self-crosslinking hyaluronic acid hydrogel in a non-human primate (rhesus monkey) IUA model [29]. This provides more reliable pre-clinical evidence for subsequent clinical translation. Additionally, combined with scRNA-seq technology, we clearly demonstrated that the complex specifically inhibits the pro-fibrotic F2 and F3 fibroblast subsets. In the existing literature on hydrogel-based treatments for IUA, only two studies used scRNA-seq, but neither delved into the mechanisms of reversing fibrosis [36,37]. However, this study has a few limitations. For example, the complex biological factors in EVs (including other miRNAs, proteins, etc.) may act synergistically, and their specific delivery mechanisms and intracellular kinetics require further in-depth research [16]. In the future, engineering modifications to optimize the targeting of EVs or to explore their combined application with other biomaterials are expected to further enhance the therapeutic effect [19,36]. From a translational perspective, EVs-GelCT may potentially be applied locally into the uterine cavity following hysteroscopic adhesiolysis. The injectable and thermosensitive properties of GelCT enable in situ gelation, which may enhance EV retention and provide sustained therapeutic effects in the injured endometrium [7]. However, the optimal clinical dosage and dose–response relationship of this platform remain to be established. Although the EV dosage used in the porcine model (0.032 mg/kg) demonstrated therapeutic efficacy, further studies involving dose optimization, pharmacokinetic characterization, biodistribution analysis, and comprehensive safety evaluation are required to define the appropriate therapeutic window and support future clinical translation [38]. In conclusion, this study clearly demonstrates that the EVs-GelCT complex can effectively reverse fibrosis in porcine IUA caused by mechanical injury, with significant repair effects. Specifically, after the EVs-GelCT complex is applied to the porcine uterine cavity, it slowly releases EVs onto the endometrial surface. The miRN-423-5p carried by the EVs targets and binds to the 3’ -UTR region of Smad3, inhibiting its phosphorylation, thereby blocking the formation of the Smad2/3-Smad4 complex. This further reduces the expression levels of α-SMA, TGF-β, CXCL10, and collagen fibers, ultimately reducing the activation of the F2 and F3 fibroblast subsets and alleviating fibrosis (Fig. 9). This strategy provides a promising preclinical foundation for the future development of EV-based biomaterial therapies for IUA. Schematic illustration of the therapeutic effects of EVs-GelCT on IUA. EVs-GelCT treatment promotes angiogenesis, restores endometrial thickness, and attenuates fibrosis, thereby facilitating endometrial repair. CRediT authorship contribution statement Xiao Xu: Writing – review & editing, Writing – original draft, Methodology, Data curation. Xiaomin Wang: Writing – review & editing, Resources. Quan Tian: Validation, Data curation. Junlin Lv: Writing – review & editing, Investigation. Nailong Pan: Conceptualization. Zhen Shang: Validation. Yiwei Xu: Resources. Xiaotong Li: Investigation. Xiaoyan Li: Investigation. Teng Wang: Validation. Yanjun Liu: Resources. Liang Zhang: Validation. Xiaoying Kong: Validation. Xiaodan Hao: Writing – review & editing, Supervision, Conceptualization. Dan Han: Writing – review & editing. Wenhua Xu: Writing – review & editing, Funding acquisition. Funding This work was supported by Shandong Taishan scholars special expert project (tstp20240824); Shandong Provincial Major Basic Research Project (ZR2026ZD35); Project of Marine Pilot Laboratory of Ministry of Science and Technology (10-02); National Key Research and Development Program (2022YEF0132500); Qingdao Natural Science Foundation (24-4-4-zrjj-150-jch); National Natural Science Foundation of China(82302653); National Natural Science Foundation of China (Grant No. 32101137); National Natural Science Foundation of China (81770900); Horizontal Major Projects in Shandong Province (RH2200000157); Key Projects of Qingdao Science and Technology Department (20-3-4-43-nsh); the Postdoctoral Fellowship Program (Grade C) of China Postdoctoral Science Foundation (GZC20240771). Thanks to the software of Adobe Illustrator and Chemdraw for drawing the figures. Declaration of competing interests 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. Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.mtbio.2026.103574. Contributor Information Xiaodan Hao, Email: haoxiaodan@qdu.edu.cn. Dan Han, Email: handan_happy123@163.com. Wenhua Xu, Email: qd.wh@qdu.edu.cn. Supplementary data The following is the Supplementary data to this article: Multimedia component 1 Data availability No data was used for the research described in the article. References Associated Data Supplementary Materials Multimedia component 1 Data Availability Statement No data was used for the research described in the article.