Single-cell transcriptomics-guided dynamic hydrogel delivery of artemisia argyi-derived EVs relieves ER stress and promotes diabetic wound regeneration

Abstract

Diabetic wounds are characterized by persistent inflammation, impaired angiogenesis, and delayed tissue regeneration, yet effective strategies to regulate the pathological immune microenvironment remain limited. Here, integrated single-cell transcriptomics and multi-omics analyses revealed significant enrichment of endoplasmic reticulum stress (ERS)-related signaling in pro-inflammatory macrophages within diabetic wounds, suggesting a potential mechanism driving chronic inflammation. To target this process, we developed an injectable dynamic hydrogel (GPOK-2@EVs) incorporating Artemisia argyi-derived extracellular vesicles (EVs) for localized immunomodulatory therapy. The hydrogel was formed through dual dynamic crosslinking between phenylboronic acid-modified gelatin and oxidized konjac glucomannan, providing self-healing capability, injectability, tissue adhesion, and pH/ROS-responsive degradation for sustained EV release. In vitro experiments demonstrated that GPOK-2@EVs exhibited excellent biocompatibility and effectively reprogrammed macrophages toward an anti-inflammatory phenotype, accompanied by reduced inflammatory cytokine secretion, alleviated ERS activation, decreased ROS accumulation, and improved mitochondrial integrity. GPOK-2@EVs significantly enhanced endothelial angiogenic activities, including cytoskeletal remodeling, tube formation, and migration. In a diabetic mouse wound model, GPOK-2@EVs accelerated wound closure, promoted granulation tissue formation and collagen deposition, enhanced vascularization, and suppressed inflammatory and ERS responses. These findings demonstrate that EV-loaded dynamic hydrogels can modulate the inflammatory microenvironment and promote diabetic wound regeneration, providing a promising biomaterial strategy for chronic wound therapy.

Graphical Abstract

GPOK-2@EVs hydrogel promotes wound healing through ROS-responsive EV release, macrophage modulation, inflammatory cytokine reduction, and enhanced angiogenesis.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12951-026-04759-7.

Highlights

1. ER stress is identified as a key driver of pro-inflammatory macrophage activation in diabetic wounds.

2. The GPOK-2@EVs self-healing hydrogel combines microenvironment-adaptive delivery and macrophage-reprogramming capacities.

3. GPOK-2@EVs accelerate diabetic wound healing by alleviating ER stress and promoting angiogenesis.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12951-026-04759-7.

Introduction

With the global aging population and the rising prevalence of diabetes, the incidence of diabetic wounds is increasing annually [1]. A hallmark complication of diabetes is the development of chronic, non-healing wounds, characterized by persistent hyperglycemia, bacterial infection, and sustained inflammation, which collectively impede wound closure [2]. Despite the application of various therapeutic strategies in clinical management, their efficacy remains limited, particularly when the immune response in diabetic patients cannot be effectively modulated. This leads to significant patient suffering and imposes a substantial economic burden [3–5].

Macrophages serve as central immune regulators in wound repair, with their phenotypic dynamic switching dictating the healing trajectory. Activated macrophages are broadly categorized into pro-inflammatory M1 and anti-inflammatory/pro-reparative M2 subsets [6, 7]. In normal healing, macrophages rapidly transition from the M1 phenotype in the early inflammatory phase to the M2 phenotype during inflammation resolution, thereby promoting tissue proliferation and remodeling [8]. However, in diabetic wounds, the persistent microenvironment of hyperglycemia, hypoxia, and oxidative stress disrupts macrophage polarization. This results in the excessive activation of M1 macrophages and a failure to switch effectively to the M2 phenotype, creating a vicious cycle of chronic inflammation. Consequently, targeting the M1/M2 polarization balance to reverse this pathological inflammatory microenvironment has emerged as a critical strategy for accelerating diabetic wound repair [9–12]. The advent of single-cell RNA sequencing (scRNA-seq) provides a powerful tool for dissecting cellular heterogeneity and fate transitions, enabling comprehensive profiling of gene expression and tracking of dynamic cellular states at the single-cell level [13]. This technology has been widely applied in tumor immunology and chronic inflammatory diseases. Utilizing scRNA-seq, we systematically analyzed the transcriptomic dynamics of macrophages during diabetic wound healing and identified unique molecular signatures in M1 macrophages within diabetic wounds. Compared to non-diabetic wounds, these cells exhibited persistent enrichment of endoplasmic reticulum stress (ERS) related genes and pro-inflammatory signaling pathways, suggesting the existence of specific upstream regulatory mechanisms. This finding offers new insights for targeting the pro-inflammatory characteristics of M1 macrophages to accelerate inflammation resolution.

ERS may act as a key upstream mechanism regulating macrophage polarization, playing a pivotal role in the impaired healing of diabetic wounds [14]. Metabolic stress induced by hyperglycemia disrupts ER function, leading to the accumulation of unfolded proteins and sustained ERS, which triggers the UPR [15–18]. While moderate ERS can restore protein homeostasis via the UPR and exert cytoprotective effects, severe or prolonged ERS induces apoptosis, disrupts the wound microenvironment, and hinders healing [19]. Notably, ERS signaling pathways are potentially linked to the regulation of immune cell function, influencing cytokine secretion and the maintenance of immune phenotypes [20, 21]. Targeting ERS signaling pathways thus represents an innovative strategy to reshape macrophage polarization and improve diabetic wound healing.

In the field of immunomodulation for diabetic wounds, developing safe and effective therapeutic strategies remains a core challenge. Traditional immune interventions often rely on single-target blockade, such as anti-inflammatory antibodies or synthetic small-molecule inhibitors. Although these can alleviate inflammation in the short term, they struggle to fundamentally reverse the pathological polarization state of macrophages [22–24]. Synthetic nanodelivery systems, while achieving targeted drug delivery, face issues such as poor biodegradability, potential cytotoxicity, and rapid immune clearance [25–27]. Animal-derived EVs, despite their natural delivery advantages, are limited by restricted sources, significant batch-to-batch variability, and risks of pathogen contamination, hindering clinical translation [28, 29]. In this context, plant-derived nanovesicles have emerged as a promising biocompatible delivery platform. They not only circumvent the safety concerns associated with animal-derived materials but also offer advantages such as scalability, high storage stability, and the endogenous loading of diverse bioactive molecules. Plants have accumulated rich secondary metabolites through evolution, which often possess multi-target synergistic regulatory capabilities, acting simultaneously on inflammation, oxidative stress, and cellular metabolism [30–32]. Plant-derived small molecules and vesicle preparations have garnered significant attention for their multi-target synergistic effects, low toxicity, and excellent biocompatibility. As natural nanocarriers with diameters of 30–150 nm, plant-derived EVs inherit the low immunogenicity and efficient intercellular communication capabilities of EVs while overcoming bottlenecks such as high extraction costs, potential viral risks, and ethical controversies associated with animal sources [33, 34]. They offer unique advantages, including ease of scalable acquisition, stable physicochemical properties, and richness in endogenous bioactive components (e.g., flavonoids, terpenoids, and specific miRNAs), enabling natural modulation of the immune microenvironment [35–37]. Artemisia argyi, a traditional Chinese herb with a long history, is renowned for its exceptional anti-inflammatory, antibacterial, and pro-healing properties [38, 39]. Extracts from A. argyi have also shown broad application prospects in anti-aging, improving insulin resistance, and neuroprotection [40, 41]. However, whether A. argyi-derived extracellular vesicles (EVs) contain key active components capable of regulating ERS, and the specific molecular mechanisms by which they promote healing in diabetic wounds by reshaping macrophage polarization, remain to be thoroughly explored.

Recently, the emergence of network pharmacology has enabled researchers to explore the complex relationships between plant components and gene regulation, revealing how phytochemicals interact with multiple targets to jointly regulate cellular processes and pathological states [42, 43]. By integrating network pharmacology with scRNA-seq, we discovered that these EVs are enriched with various active components related to ERS regulation, capable of significantly suppressing ERS levels in macrophages. Specifically, EVs can downregulate the expression of ERS markers such as GRP78, CHOP, p-PERK, and p-IRE1α, thereby blocking ERS-mediated activation of inflammatory signaling, inhibiting M1 macrophage polarization, and promoting the switch to the M2 phenotype. However, free EVs suffer from limitations at the wound site, including short retention times and susceptibility to enzymatic degradation, which restrict their therapeutic efficacy [44, 45]. Hydrogels, as three-dimensional polymeric networks, offer excellent biocompatibility, injectability, and sustained-release characteristics. They provide a stable platform for loading and delivering EVs, extending their local retention time at the wound site and enabling controlled release [46–49].

Based on the aforementioned background, this study constructs a dynamic cross-linked hydrogel system loaded with Artemisia argyi-derived EVs (GPOK-2@EVs) to promote diabetic wound healing. Artemisia argyi-derived EVs were isolated via differential centrifugation coupled with filtration and thoroughly characterized. The GPOK hydrogel was engineered by conjugating boronic acid to gelatin and oxidizing konjac glucomannan, forming a dual-dynamic network via Schiff base and boronate ester bonds. This system exhibits exceptional self-healing, injectability, and tissue adhesion. Notably, it demonstrates pH/ROS dual-responsive degradability, enabling accelerated degradation and controlled EV release specifically within the diabetic wound microenvironment. Subsequent in vivo studies in a diabetic murine full-thickness wound model were conducted to elucidate its therapeutic efficacy in tissue regeneration and immunomodulation. Figure 1 schematically illustrates the overall design of the GPOK-2@EVs hydrogel system and its proposed therapeutic mechanism in diabetic wound healing, including ERS alleviation, macrophage reprogramming, enhanced angiogenesis, and accelerated tissue regeneration. This work offers a novel strategy for developing smart plant EV-based dressings for chronic wound management.

Fig. 1.

Schematic illustration of GPOK-2@EVs-mediated ers relief and macrophage reprogramming for diabetic wound healing

Materials and methods

Processing of single-cell transcriptomic data and functional enrichment analysis

In this study, single-cell transcriptomic datasets were retrieved from publicly available Gene Expression Omnibus (GEO) resources. Control samples were obtained from foot skin tissues of non-diabetic individuals in dataset GSE165816, whereas chronic diabetic foot ulcer (DFU) samples were derived from dataset GSE245703. All raw gene expression matrices were processed uniformly in the R environment using the Seurat package. Cells with fewer than 300 or more than 6000 detected genes, total UMI counts below 500 or above 7000, or mitochondrial gene proportions greater than 15% were excluded to remove low-quality cells, apoptotic cells, and potential doublets. These thresholds were determined based on standard practices in single-cell RNA-seq analysis and the distribution characteristics of the datasets.

Following quality control, each dataset was normalized separately, and highly variable genes were identified using Seurat. The datasets were then merged, scaled, and subjected to principal component analysis (PCA). To minimize batch effects caused by different GEO sources, batch correction was performed using the Harmony algorithm, with dataset origin set as the batch variable. The Harmony-corrected embeddings were used for downstream nearest-neighbor graph construction, clustering, and UMAP visualization. A shared nearest-neighbor graph was constructed using the top 25 Harmony-corrected dimensions, followed by clustering with the FindNeighbors and FindClusters functions. Cell-type annotation was performed based on canonical marker genes reported in the literature and manually curated to ensure accuracy.

Given the pivotal role of macrophages in regulating the DFU microenvironment and wound repair, subsequent analyses focused on this cell population. To investigate macrophage heterogeneity, macrophage subclusters were reanalyzed and annotated according to established phenotypic markers. Differential expression analysis between macrophages derived from healthy and DFU tissues was conducted using the FindMarkers function in Seurat. Differentially expressed genes (DEGs) were identified using the Wilcoxon rank-sum test with thresholds of |log2FoldChange|> 0.3 and p < 0.05. GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were subsequently conducted using clusterProfiler to elucidate functional differences among macrophage subtypes in inflammatory regulation, metabolic reprogramming, and immune responses.

Lipidomics and untargeted metabolomics of artemisia argyi-derived EVs

To characterize the lipid composition and metabolic profile of Artemisia argyi-derived EVs, lipidomics and untargeted metabolomics analyses were performed using mass spectrometry platforms. The experimental workflow included batch sample processing, organic solvent extraction of lipids and polar metabolites, liquid chromatography–mass spectrometry (LC–MS) acquisition, and downstream bioinformatic analysis.All samples were processed under identical conditions to minimize technical bias, and quality control (QC) samples were included to evaluate instrument stability and reproducibility. Untargeted metabolomics was conducted using an ultra-high-performance liquid chromatography–high-resolution mass spectrometry (UPLC-HRMS) system in both positive and negative ion modes to maximize metabolite coverage.Raw data were processed using Progenesis QI (v2.3), including peak detection, alignment correction, noise filtering, metabolite annotation, and calculation of relative abundances. Data were subsequently normalized to reduce batch effects. Multivariate statistical analysis, including PCA and differential significance testing, was employed to identify key metabolic features, thereby constructing a differential metabolic profile and providing a foundation for investigating the metabolic regulatory and potential pharmacological roles of Artemisia argyi-derived EVs.

Network pharmacology analysis of artemisia argyi-related targets and pathways

To systematically elucidate the molecular interaction network of bioactive components derived from Artemisia argyi EVs, a network pharmacology approach was employed. Candidate bioactive molecules with favorable drug-like properties were first screened based on gastrointestinal absorption, lipophilicity, molecular weight, and drug-likeness parameters using the SwissADME platform. Potential molecular targets of the selected compounds were then predicted using SwissTargetPrediction, and only targets with a probability score > 0.1 were retained for subsequent analysis. To identify shared targets between Artemisia argyi-derived EVs and M1 macrophage-associated DEGs, intersection analysis was performed using Venny, thereby identifying potential core mediators underlying the immunomodulatory effects of EVs. An integrated active component-target-pathway interaction network was constructed using Cytoscape (v3.10.1) to visualize the regulatory relationships between EV-derived bioactive molecules and pathological gene signatures. Functional annotation and pathway enrichment analyses of intersecting genes were performed using clusterProfiler (v3.14.3), with adjusted p < 0.05 considered statistically significant.

Molecular docking analysis was further performed using AutoDock Vina to evaluate the potential interaction between representative EV-derived bioactive components and key ERS-related candidate targets. The three-dimensional structure of GRP78 was obtained from the RCSB Protein Data Bank. Water molecules, co-crystallized ligands, and redundant ions were removed using PyMOL. The receptor protein was then prepared by adding polar hydrogens and assigning charges using AutoDock Tools. The three-dimensional structure of quercetin was downloaded from the PubChem database and energy-minimized using the MMFF94 force field. A docking grid box was centered on the predicted active pocket of GRP78, and the conformation with the lowest binding energy was selected as the optimal docking pose. Ligand–receptor interactions were visualized using PyMOL.

Materials

Gelatin (Gel) from porcine skin, 3-aminophenylboronic acid (3APBA), N-hydroxysuccinimide (NHS), 1-(3-dimethylaminopropyl)-3-ethylcarbodiimide hydrochloride (EDC), konjac glucomannan (KGM), sodium periodate (NaIO4), hydrochloric acid (HCl), and sodium hydroxide (NaOH) were purchased from Aladdin (Shanghai, China). All reagents were used as received without further purification.

Isolation and characterization of artemisia argyi-derived EVs

Fresh Artemisia argyi leaves were washed with deionized water, soaked in PBS, and homogenized. The homogenate was filtered through sterile gauze and sequentially centrifuged at 500 × g for 10 min, 2,000 × g for 20 min, 5,000 × g for 30 min, and 10,000 × g for 60 min at 4 °C to remove cells, debris, and large vesicles. The supernatant was filtered through a 0.22 μm membrane and ultracentrifuged at 100,000 × g for 90 min at 4 °C using an SW 32 Ti rotor (Beckman Coulter). The pellet was washed with PBS and ultracentrifuged again at 100,000 × g for 90 min. The final EV pellet was resuspended in 200 μL sterile PBS and stored at − 80 °C for subsequent experiments. For density-gradient validation, part of the EV suspension was loaded onto a discontinuous sucrose gradient (8%, 30%, 45%, and 60%, w/v) and ultracentrifuged at 150,000 × g for 2 h at 4 °C using an SW 41 Ti rotor (Beckman Coulter). The EV-enriched fraction at the 30%/45% interface was collected, washed with PBS, and concentrated by ultracentrifugation at 100,000 × g for 90 min. EVs prepared by differential ultracentrifugation and density-gradient centrifugation were compared by Western blotting for plant EV-associated markers and endotoxin detection. The hydrodynamic size distribution and zeta potential of EVs were measured using a Litesizer™ 500 analyzer, and EV protein concentration was determined using a BCA assay.

Synthesis and characterization of GP

Gelatin (4.0 g) was dissolved in 200 mL of solvent at 50 °C to obtain solution A. The pH of the solution was adjusted to 5.0 with 0.1 mol/L hydrochloric acid. 3.38 g of 3APBA, 0.65 g of EDC, and 0.39 g of NHS were each dissolved in 50 mL of reaction solvent to obtain solution B. The pH of this solution was adjusted to 5.0 with 0.1 mol/L HCl. Subsequently, solutions A and B were cooled to room temperature and then mixed. The mixture was continuously stirred magnetically at room temperature for 48 h. The reaction mixture was then dialyzed against deionized water for one week using a dialysis membrane with a molecular weight cut-off of 3500 Da, until no detectable 3APBA was found in the eluent. After freeze-drying for 3 days, GP was obtained. The physicochemical properties of the prepared GP were characterized by Fourier transform infrared (FTIR) spectroscopy (Nicolet 6700, Thermo Fisher Scientific, USA).

Synthesis and characterization of OKGM

KGM (2.0 g) was dispersed in 350 mL of deionized water and formed a transparent homogeneous system by magnetic stirring (1000 rpm) for 30 min. Subsequently, 2.0 g of sodium periodate (NaIO₄) was added to the system, and the reaction vessel was transferred to a light-protected environment. The reaction was carried out continuously by stirring in a constant-temperature water bath at 30 °C for 12 h. After the reaction was completed, 7 mL of ethylene glycol (EG) was added to the mixed solution at a uniform rate. The reaction was continued for 1 h under the same temperature and stirring conditions. Subsequently, the product was transferred to a dialysis bag with a molecular weight cut-off (MWCO) of 8000 Da. Dialysis purification was carried out under light-protected conditions at room temperature for 3 days. During this period, deionized water was replaced every 12 h to remove low-molecular-weight impurities. The final solution was freeze-dried at − 50 °C for 72 h to obtain white flocculent oxidized konjac glucomannan (OKGM). The physicochemical properties of the obtained OKGM were characterized by Fourier transform infrared (FTIR) spectroscopy (Nicolet 6700, Thermo Fisher Scientific, USA).

Preparation and characterization of the hydrogels

GP was dissolved in 1 mL of phosphate-buffered saline (PBS) to form 5, 7.5, and 10 wt% GP solutions. Subsequently, 100 mg of OKGM was dissolved in 1 mL of PBS to obtain a 10 wt% OKGM solution in another centrifuge tube. The GP and OKGM solutions were then mixed at a volume ratio of 1:1 and stirred thoroughly. Subsequently, 1 M sodium hydroxide (NaOH) was added to adjust the pH to 7.5–8.5 and form a GPOK hydrogel. The obtained GPOK hydrogels were designated as GPOK-1, GPOK-2, and GPOK-3 based on the different concentrations of GP. To prepare GPOK@EVs, 120 μg of EVs was uniformly dispersed in 1 mL of GP solution before the GP and OKGM solutions were mixed, and the pH was adjusted to obtain the GPOK@EVs hydrogel. The chemical structure and functional groups of the composite hydrogels were characterized by Fourier transform infrared (FTIR) spectroscopy. Morphological characterization of the hydrogel samples was conducted using scanning electron microscopy (SEM; ZEISS Ultra 55, Germany). To prepare for SEM analysis, lyophilized hydrogels were rapidly frozen by immersion in liquid nitrogen to preserve their microstructure.

Rheology analysis

The mechanical properties of the hydrogel were evaluated using a rotational rheometer (Physica MCR302, Anton Paar) at a strain amplitude of 1.0% over a frequency range of 0.1–10 Hz. To evaluate the injectability of the hydrogel, the dependence of viscosity (η) on shear rate was analyzed using a shear rate sweep mode. The step-strain test was conducted at a fixed frequency of 10 rad/s. The oscillation strain alternated between 3.0% (low strain) and 300% (high strain), and each strain level was maintained for 100 s to assess the self-healing properties of the hydrogel.

Swelling tests

Swelling rates (SRs) of the hydrogels were measured through an immersion method. Defined weights of the hydrogel (W₀) were immersed in PBS for set durations, following which the swollen hydrogels were retrieved and their wet weights (W_t) recorded. Swelling rates were calculated with the equation: SRs = (W_t − W₀)/W₀ × 100%.

Adhesion tests

The adhesion strength of the hydrogel was evaluated using an electromechanical universal testing machine (MCR 301, Anton Paar, Austria). The hydrogel solution was applied between two porcine skin substrates, with an overlapping area of 1 cm × 1 cm for each sample. After gelation, the samples were tested at a crosshead speed of 10 mm·min⁻¹.

Responsive degradability

To evaluate the in vitro degradation characteristics of GPOK-2@EVs, equal amounts of hydrogel were freeze-dried and weighed to record the initial mass. The samples were then immersed in different media, including PBS at pH 7.4, PBS at pH 5.5, PBS containing 1 mM H₂O₂ at pH 7.4, and PBS containing 1 mM H₂O₂ at pH 5.5. The pH 7.4 PBS group was used as the neutral physiological control, whereas the pH 5.5 PBS group was used as the constant acidic pH control. The pH 5.5 + H₂O₂ group was used to mimic the acidic and ROS-rich microenvironment of diabetic wounds. At predetermined time points, the samples were collected, freeze-dried, and weighed to calculate the remaining mass ratio.

In vitro release profile of EVs in hydrogel

To determine the EV release behavior of the GPOK@EVs hydrogels, the hydrogels were placed in different solutions (PBS and PBS containing 1 mM H₂O₂ at pH 5.5) and then incubated at 37 °C. At predetermined time points, the supernatant was collected and replaced with an equal volume of the corresponding fresh solution. The protein concentration in the collected supernatant was determined using the bicinchoninic acid (BCA) protein assay. The release amount of EV-associated protein was calculated based on the bovine serum albumin (BSA) standard curve. The cumulative release profile was subsequently plotted to analyze the EV release behavior of the hydrogels.

Cell culture

Mouse macrophage RAW264.7 cells were obtained from the National Collection of Authenticated Cell Cultures (NCACC, China). Mouse fibroblast L929 cells and human umbilical vein endothelial cells (HUVECs) were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). RAW264.7 macrophages, L929 fibroblasts, and HUVECs were cultured in Dulbecco’s Modified Eagle Medium (DMEM, Gibco, USA) supplemented with 10% fetal bovine serum (FBS, Gibco, USA) and 1% penicillin–streptomycin (100 U mL⁻¹ penicillin and 100 U mL⁻¹ streptomycin, Gibco, USA). Cells were maintained in a humidified incubator at 37 °C with 5% CO₂. For routine passaging, adherent cells were detached using 0.05% trypsin–EDTA solution or gently scraped depending on the cell type. Cells in the logarithmic growth phase were used for all subsequent experiments.

Macrophage treatment groups

RAW264.7 macrophages were stimulated with LPS to establish an inflammatory model. The Control group refers to macrophages treated with LPS alone, whereas the GPOK-2, EVs, and GPOK-2@EVs groups refer to LPS-stimulated macrophages receiving the corresponding treatments. For in vitro experiments, the free EV group was administered at an EV dose equivalent to the EV amount contained in the corresponding volume of GPOK-2@EVs hydrogel.

In vitro cytocompatibility assessment

The cytocompatibility of GPOK-2@EVs was evaluated using RAW 264.7 macrophages, HUVECs, and L929 fibroblasts. Cells were seeded into 96-well plates and cultured overnight to allow cell attachment. Subsequently, the culture medium was replaced with fresh medium containing different treatments, including Control, GPOK-2, EVs, and GPOK-2@EVs. After 24 h of incubation, cell viability was assessed using a CCK-8 assay according to the manufacturer’s instructions, and the absorbance was measured at 450 nm using a microplate reader. To further visualize cell survival, Live/Dead staining was performed. Cells were incubated with calcein-AM and propidium iodide staining solution, followed by observation under an inverted fluorescence microscope. Live cells exhibited green fluorescence, whereas dead cells showed red fluorescence, allowing evaluation of cytocompatibility under different treatments.

Hemolysis assay

The hemocompatibility of GPOK-2@EVs was evaluated using a standard hemolysis assay. Fresh mouse blood was collected and centrifuged to isolate red blood cells (RBCs), which were subsequently washed with PBS and diluted to obtain an erythrocyte suspension. The RBC suspension was incubated with PBS (negative control), GPOK-2@EVs, or Triton X-100 (positive control) at 37 °C for 2 h. After incubation, the samples were centrifuged and the absorbance of the supernatant was measured at 540 nm to quantify hemoglobin release. The hemolysis ratio was calculated relative to the positive control to assess the blood compatibility of the materials.

In vivo biocompatibility assessment of GPOK-2@EVs

To investigate the in vivo biocompatibility of GPOK-2@EVs, healthy C57BL/6 mice were randomly allocated into two groups. Mice in the control group received saline, while those in the treatment group were administered GPOK-2@EVs via subcutaneous injection in the dorsal region.During the observation period, the body weight of the mice was recorded to monitor their overall physiological condition. At the predetermined endpoint, blood samples were collected for routine hematological and biochemical analyses, including white blood cell count (WBC), hemoglobin (HGB), red blood cell count (RBC), neutrophil count (NEU), alanine aminotransferase (ALT), aspartate aminotransferase (AST), urea (UREA), and creatinine (CREA). Subsequently, the major organs, including the heart, liver, spleen, lung, and kidney, were harvested, fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned for hematoxylin and eosin (H&E) staining to examine potential histopathological alterations.

Immunofluorescence staining

Immunofluorescence staining was performed to evaluate macrophage polarization and endothelial cell responses. RAW264.7 macrophages and HUVECs were seeded on sterile coverslips, treated with Control, GPOK-2, EVs, or GPOK-2@EVs for 24 h, fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and blocked with 5% BSA. The samples were incubated overnight at 4 °C with primary antibodies against CD86 (ab239075, Abcam), CD206 (ab64693, Abcam), F-actin (ab130935, Abcam), and CD31 (ab222783, Abcam), followed by fluorescent secondary antibodies and DAPI counterstaining. Images were acquired using a confocal microscope and quantified with ImageJ software.

Western blotting

Total proteins from macrophages, HUVECs, or wound tissues were extracted using RIPA buffer supplemented with protease and phosphatase inhibitors. After centrifugation, protein concentrations were determined using a BCA protein assay kit (Thermo Fisher Scientific, Cat. No. 23225). Equal amounts of protein were separated by SDS-PAGE and transferred onto PVDF membranes. Membranes were blocked with 5% non-fat milk and incubated overnight at 4 °C with primary antibodies against iNOS (#95423, CST), Arg-1 (No. 66129–1-Ig, Proteintech), CD206 (ab64693, Abcam), CD86 (ab239075, Abcam), GRP78 (No. 11587–1-AP, Proteintech), PERK (No. 24390–1-AP, Proteintech), p-PERK (No. 82534–1-RR, Proteintech), eIF2α (No. 11170–1-AP, Proteintech), p-eIF2α (No. 87179–2-RR, Proteintech), ATF4 (No. 10835–1-AP, Proteintech), CHOP (No. 15204–1-AP, Proteintech), CD31 (ab222783, Abcam), VEGF (ab32152, Abcam), TET8 (CSB-PA212091XA01DOA, CUSABIO), HSP70 (ab2787, Abcam), PEN1 (CSB-PA875527XA01DOA, CUSABIO), and GAPDH (ab9485, Abcam). After incubation with HRP-conjugated secondary antibodies, protein bands were detected using ECL reagents and quantified with ImageJ software. GAPDH was used as the loading control.

Quantitative real-time PCR (qPCR)

Total RNA was extracted from cells using TRIzol reagent (Invitrogen, USA, Cat. No. 15596026) according to the manufacturer’s instructions. RNA concentration and purity were determined using a NanoDrop spectrophotometer (Thermo Fisher Scientific, USA). Complementary DNA (cDNA) was synthesized from 1 μg of total RNA using a PrimeScript RT reagent kit. qPCR was performed using SYBR Green Master Mix on a QuantStudio 6 real-time PCR system (Applied Biosystems, USA). Relative gene expression levels were calculated using the 2^ − ΔΔCt method and normalized to the housekeeping gene β-actin. All primer sequences used in this study are listed in Supplemental Resources-1.

Enzyme-linked immunosorbent assay

To determine the secretion levels of inflammatory cytokines, RAW 264.7 macrophages were seeded into 6-well plates at a density of 5 × 10⁴ cells per well and cultured for 24 h to allow cell attachment. The culture medium was then replaced with fresh medium containing different treatments, including Control, GPOK-2, EVs, and GPOK-2@EVs, and the cells were incubated under standard culture conditions for the indicated time. After treatment, the cell culture supernatants were collected and centrifuged to remove cell debris. The concentrations of inflammation-related cytokines, including IL-1β, IL-6, TNF-α, IL-10, TGF-β, and CCL2, were quantified using commercial ELISA kits according to the manufacturer’s protocols. Briefly, standards and samples were added to ELISA plates pre-coated with specific capture antibodies and incubated at 37 °C for the recommended duration. After washing to remove unbound substances, detection antibodies were added and incubated, followed by the addition of substrate solution to initiate the colorimetric reaction. The reaction was terminated using stop solution, and the optical density was measured at 450 nm with a microplate reader. Cytokine concentrations were calculated based on the standard curves generated from known standards.

Flow cytometry analysis

Flow cytometry was used to quantify macrophage polarization. After the indicated treatments, RAW264.7 macrophages were harvested, blocked with Fc receptor blocking buffer, and stained with fluorochrome-conjugated antibodies against F4/80 (No. 98236–1-RR, Proteintech), CD86 (No. 65165–1-Ig, Proteintech), and CD206 (No. 98031–1-RR, Proteintech) for 30 min at 4 °C in the dark. The proportions of M1 macrophages (F4/80⁺CD86⁺) and M2 macrophages (F4/80⁺CD206⁺) were analyzed using FlowJo software.

Intracellular reactive oxygen species (ROS) evaluation

Intracellular ROS levels were detected using the fluorescent probe DCFH-DA (HY-D0940, MCE). RAW 264.7 macrophages were seeded in culture plates and treated with Control, GPOK-2, EVs, or GPOK-2@EVs according to the experimental design. After treatment, cells were incubated with DCFH-DA working solution in serum-free medium at 37 °C for 30 min in the dark. Subsequently, the cells were washed three times with PBS to remove excess probe. Fluorescence images were obtained using a fluorescence microscope, and intracellular ROS levels were quantified by measuring the mean fluorescence intensity using ImageJ software.

ERS probe assay

RAW 264.7 macrophages from different treatment groups (Control, GPOK-2, EVs, and GPOK-2@EVs) were seeded onto confocal dishes and cultured overnight. After the indicated treatments, cells were incubated with ERS probe working solution prepared from an ERS detection probe kit (#9956, CST) at 37 °C for the recommended time in the dark. Fluorescence images representing different probe states were captured using a confocal laser scanning microscope. The relative ERS level was determined by calculating the ratio of green to red fluorescence intensity using ImageJ software.

Mitochondrial staining

To visualize mitochondrial morphology, RAW 264.7 macrophages were seeded onto confocal dishes and cultured overnight. After the designated treatments, cells were incubated with MitoTracker Red working solution (Solarbio, M9940) at 37 °C for approximately 20–30 min in the dark according to the manufacturer’s protocol. After staining, cells were gently washed with PBS to remove excess dye. Fluorescence images were acquired using a confocal laser scanning microscope, and mitochondrial morphology was analyzed using ImageJ software.

Transmission electron microscopy

For ultrastructural analysis of mitochondria, RAW 264.7 macrophages were collected after the indicated treatments and fixed in 2.5% glutaraldehyde at 4 °C overnight. The samples were then post-fixed with 1% osmium tetroxide, dehydrated through a graded ethanol series, and embedded in epoxy resin. Ultrathin sections were prepared and stained with uranyl acetate and lead citrate before observation using a transmission electron microscope. Mitochondrial morphology and structural alterations were analyzed, and the mean mitochondrial length was quantified using ImageJ software.

Scratch wound healing assay

HUVECs were seeded in 6-well plates at a density of 1 × 10^6 cells/well and cultured to near-complete confluence. A uniform linear scratch was created in each well using a sterile 200 μL pipette tip perpendicular to pre-marked reference lines. Detached cells were gently removed by washing with PBS. HUVECs were then incubated with conditioned media collected from macrophages subjected to different treatments, including Control, GPOK-2, EVs, and GPOK-2@EVs. Scratch closure was photographed at 0, 24, and 48 h using an inverted microscope (Nikon Eclipse Ti, Japan, or equivalent). The wound area at each time point was quantified using ImageJ software, and the relative scratch area was calculated to evaluate cell migratory capacity.

Transwell migration assay

Cell migration was further assessed using 8 μm pore-size Transwell inserts (Corning, USA). HUVECs were suspended in serum-free medium and seeded into the upper chambers at a density of 2 × 10^4 cells/well, while the lower chambers were filled with conditioned media collected from RAW264.7 macrophages treated with Control, GPOK-2, EVs, or GPOK-2@EVs. After incubation for 24 h at 37 °C, non-migrated cells on the upper surface of the membrane were removed using cotton swabs. Cells that migrated to the lower surface were fixed with 4% paraformaldehyde for 15 min, stained with 1% crystal violet for 30 min, photographed under an optical microscope, and quantified using ImageJ software.

In vitro angiogenesis assay

The angiogenic ability of HUVECs was evaluated using a tube formation assay. Briefly, 50 μL of Matrigel (Corning, USA, Cat. No. 356234) was added to each well of a 24-well plate and allowed to solidify at 37 °C for 40 min. HUVECs were then seeded onto the Matrigel-coated wells at a density of 2 × 10^4 cells/well and treated with Control, GPOK-2, EVs, or GPOK-2@EVs. After incubation for 5 h at 37 °C, tube-like structures were analyzed using ImageJ software, and the number of branch points and relative tube length were quantified to assess angiogenic activity.

In vivo diabetic wound model and treatment

Male C57BL/6 mice (8 weeks old, 18–20 g) were used to establish the diabetic wound model. All animal experiments were conducted in accordance with the guidelines approved by the Ethics Committee of Peking University People’s Hospital (Approval No. 2026PHB185). After fasting for 10 h, diabetes was induced by intraperitoneal injection of streptozotocin (STZ, 50 mg/kg) dissolved in citrate buffer (pH 4.5) once daily for five consecutive days [50]. Fasting blood glucose levels were monitored using tail-vein blood samples, and mice with fasting blood glucose levels > 11.1 mmol/L in two consecutive measurements were defined as diabetic mice. Full-thickness skin wounds were created after successful diabetes induction. For wound creation, mice were anesthetized and the dorsal hair was carefully removed. A 10 mm diameter full-thickness circular skin wound was generated on the dorsum using a sterile biopsy punch. The mice were then randomly assigned to four treatment groups: Control, GPOK-2, EVs, and GPOK-2@EVs. The corresponding formulations were applied directly to the wound area immediately after wound creation according to the experimental design. In an additional positive-control experiment, INTRASITE™ Gel was applied to diabetic wounds under the same experimental schedule as a clinically relevant commercial hydrogel control. EVs were administered at 1.2 mg/kg according to the body weight of each animal. Digital images of the wounds were recorded at days 1, 3, 7, 10, and 14 after treatment. Wound areas were quantified using ImageJ software, and the wound closure rate was calculated relative to the initial wound size. On day 14, the mice were euthanized, and wound tissues were harvested for histological staining, immunohistochemical analysis, and protein detection.

Histological evaluation

Wound tissues were fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned into 4 μm slices. H&E and Masson’s trichrome staining were performed to evaluate re-epithelialization, granulation tissue formation, and collagen deposition. For IHC, sections were incubated with antibodies against iNOS (ab3523, Abcam), Arg-1 (ab215843, Abcam), CD31 (ab281583, Abcam), CHOP (ab317378, Abcam), and GRP78 (ab21685, Abcam), followed by HRP-conjugated secondary antibodies and DAB visualization. For tissue immunofluorescence, sections were stained with anti-PGP9.5 antibody (ab27053, Abcam) and fluorescent secondary antibody. Images were acquired by light or fluorescence microscopy and quantified using ImageJ software.

Statistical analysis

Data are presented as the mean ± SD from at least three independent experiments or biological replicates. Before applying parametric tests, normality and homogeneity of variance were assessed using the Shapiro–Wilk test and Levene’s test, respectively. For two-group comparisons, a two-tailed Student’s t-test was used when data met the assumptions of normality and equal variance; otherwise, the Mann–Whitney U test was applied. For multiple-group comparisons, one-way ANOVA followed by Tukey’s post hoc test was performed when these assumptions were satisfied; otherwise, the Kruskal–Wallis test followed by Dunn’s multiple-comparison test was used. Statistical significance was defined as P < 0.05. Graphical analyses were carried out with GraphPad Prism 8.0 (GraphPad Software, USA).

Results and discussion

Single-cell and multi-omics analyses identify ERS as a pathological feature of pro-inflammatory macrophages in diabetic wounds

After quality control, Harmony-based batch-effect correction, and dimensionality reduction, unsupervised clustering identified 20 transcriptionally distinct cell clusters (Fig. 2A), and the overall distribution of cells from control and DFU samples is shown in Figure S1A. Based on canonical marker gene expression patterns, these clusters were annotated into nine major cell populations, including smooth muscle cells (TAGLN⁺), differentiated keratinocytes (KRT1⁺), fibroblasts (DCN⁺), basal keratinocytes (KRT5⁺), vascular endothelial cells (ACKR1⁺), NK/T cells (CCL5⁺), macrophages (CD163⁺), plasma cells (MZB1⁺), and B cells (CD79A⁺) (Fig. 2B–C). Similar cell-type distributions were observed when control and DFU samples were visualized separately (Figure S1B). Comparative analysis of cellular composition revealed an increased relative proportion of macrophages in DFU tissues (Fig. 2D), which was further supported by sample-level quantification (Figure S1C). These observations indicate that the immune microenvironment of DFU undergoes substantial remodeling, consistent with previous reports that excessive or unresolved inflammation contributes to chronic diabetic wound pathology [51, 52]. Given the central role of macrophages in regulating inflammation and tissue repair, we next examined macrophage heterogeneity in DFU tissues. Subclustering analysis identified two principal macrophage phenotypes corresponding to classical pro-inflammatory M1 macrophages and anti-inflammatory or pro-repair M2 macrophages (Fig. 2E). The distribution of these subsets across individual samples is shown in Figure S1D. Expression patterns of canonical polarization markers further confirmed the reliability of this classification, with M1 macrophages expressing CXCL8, IL1B, and CD86, whereas M2 macrophages preferentially expressed APOE, CD163, and MRC1 (Figure S1E–F). Notably, DFU samples displayed a relative enrichment of M1 macrophages and an imbalance between M1 and M2 subsets (Figure S1G), suggesting persistent pro-inflammatory activation within the diabetic wound microenvironment. Such dysregulated macrophage polarization has been widely recognized as a key driver of chronic inflammation and impaired healing in diabetic wounds [53, 54].

Fig. 2.

Single-cell analysis identifies ER stress signatures in M1 macrophages in DFU. (A) UMAP plot showing 20 transcriptionally distinct cell clusters. (B) UMAP plot of major annotated cell types based on canonical marker genes. (C) Dot plot of representative marker genes across annotated cell types; dot size denotes the percentage of expressing cells and color indicates average expression. (D) Relative proportions of major cell types in control and DFU samples. (E) UMAP visualization of macrophage subsets, highlighting M1 and M2 populations. (F) Volcano plot of DEGs in M1 macrophages between DFU and controls (|log2FC|> 0.3, P < 0.05). (G) GO enrichment analysis of upregulated genes in DFU M1 macrophages versus controls. (H) KEGG pathway enrichment analysis of upregulated genes in DFU M1 macrophages versus controls. (I) Metabolite-class composition of EVs. (J) Lipid-class composition of EVs. (K) Integrated network linking Artemisia argyi-derived EV bioactive components to predicted targets and DFU-associated gene sets, illustrating putative mechanisms. (L) Venn diagram showing overlap between predicted EV target genes and genes upregulated in DFU M1 macrophages. (M) KEGG pathway enrichment analysis of the overlapping gene set. (N) GO enrichment analysis of the overlapping gene set

To further characterize the molecular features of DFU-associated M1 macrophages, differential gene expression analysis was performed. The volcano plot revealed substantial transcriptional reprogramming between M1 macrophages derived from DFU tissues and those from healthy controls (Fig. 2F). GO enrichment analysis of the upregulated genes demonstrated significant enrichment in biological processes related to response to ERS, IRE1-mediated UPR, and ER-to-Golgi transport vesicle membrane (Fig. 2G). KEGG pathway analysis also highlighted signaling networks associated with cellular stress adaptation and protein processing in the ER (Fig. 2H). These results indicate that DFU-associated M1 macrophages are characterized not only by pro-inflammatory gene signatures but also by pronounced activation of ERS related programs. Persistent ERS has been shown to amplify inflammatory signaling, disrupt macrophage plasticity, and aggravate tissue injury under chronic metabolic conditions [55]. Therefore, targeting ERS-associated inflammatory amplification may represent a potential strategy to restore immune balance and promote wound repair.

To explore whether Artemisia argyi-derived EVs could modulate this pathological process, we next performed metabolomic and lipidomic analyses of EV cargo. Metabolite classification revealed that these EVs were enriched in multiple classes of bioactive compounds, including flavonoids, terpenoids, amino acids, and alkaloids (Fig. 2I), many of which have been reported to possess antioxidant and anti-inflammatory activities [56, 57]. Lipidomic profiling further identified abundant lipid species such as ceramides (Cer), triacylglycerols (TG), phosphatidylcholines (PC), and diacylglycerols (DG) (Fig. 2J), suggesting potential roles in membrane dynamics and cellular stress signaling [58]. To systematically predict potential molecular targets associated with these EV-derived components, a compound–target interaction network was constructed (Fig. 2K). Intersection analysis between predicted EV targets and genes upregulated in DFU-associated M1 macrophages identified a subset of overlapping genes (Fig. 2L), indicating that EV components may directly influence macrophage inflammatory programs. KEGG pathway analysis further supported the involvement of stress-responsive signaling networks associated with ER homeostasis and UPR programs (Fig. 2M). Functional enrichment of these overlapping targets revealed strong association with ERS related biological processes, including response to ERS, ER UPR, IRE1-mediated UPR, and PERK-mediated UPR (Fig. 2N). Combined with the GO enrichment results, these data pointed to ERS related signaling, particularly the PERK–eIF2α-associated axis, as a potential regulatory mechanism underlying the effects of Artemisia argyi-derived EVs in macrophages. To further identify the upstream target linking EV-derived bioactive components to ERS regulation, molecular docking analysis was performed. Quercetin, a representative active component identified in Artemisia argyi-derived EVs, showed stable binding to the active pocket of GRP78 through multiple hydrogen bonds with key residues, including Lys81, Tyr65, Gln83, and Asn87, suggesting that GRP78 may serve as a key upstream target of EV-mediated ERS regulation (Figure S2A). To functionally validate this target, GRP78 was knocked down in LPS-treated macrophages. In si-NC cells, EV treatment reduced ERS-related proteins and shifted macrophage markers from iNOS toward Arg-1. However, GRP78 knockdown attenuated the regulatory effects of EVs on ERS-related proteins and macrophage polarization markers, further supporting the involvement of GRP78 in EV-mediated ERS inhibition and macrophage reprogramming (Figure S2B).

Based on these bioinformatics predictions, we next investigated whether EV treatment could modulate ERS signaling at the protein level. To further validate the involvement of the PERK–eIF2α pathway, macrophages were stimulated with LPS and then treated with EVs in the presence or absence of the PERK inhibitor GSK2606414. Western blot analysis showed that LPS markedly activated ERS-related signaling, whereas EV treatment attenuated the expression of key ERS markers, including phosphorylated PERK (p-PERK), phosphorylated eIF2α (p-eIF2α), ATF4, GRP78, and CHOP (Figure S2C). Moreover, the addition of GSK2606414 further altered the activation status of this pathway, supporting the involvement of PERK-mediated ERS signaling in the regulatory effects of EVs. Given the established role of ERS signaling in promoting inflammatory macrophage activation and chronic tissue damage [59], suppression of this pathway may contribute to the resolution of persistent inflammation in diabetic wounds. Collectively, the integration of single-cell transcriptomics, metabolomics, lipidomics, and network pharmacology indicates that ERS represents a central molecular feature of DFU-associated macrophages. Artemisia argyi-derived EVs appear capable of modulating this ERS–inflammation axis, thereby providing a mechanistic basis for their potential application in the treatment of chronic diabetic wounds. However, the therapeutic application of EVs requires an efficient delivery strategy to ensure their stability and sustained activity within the wound microenvironment. Therefore, we next developed a hydrogel-based delivery system to encapsulate and locally deliver EVs for diabetic wound repair.

Preparation and characterization of hydrogels

Before constructing the EV-integrated hydrogel system, the physicochemical characteristics of the isolated Artemisia argyi-derived EVs were first examined to confirm their structural integrity and colloidal stability. Transmission electron microscopy (TEM) showed that the isolated EVs displayed a typical spherical to cup-shaped morphology with clear boundaries and without obvious aggregation or impurity contamination (Fig. 3A). Nanoparticle size analysis further demonstrated that the EV population was mainly distributed within the nanoscale range of approximately 50–200 nm and exhibited a relatively narrow unimodal distribution (Fig. 3B), indicating good homogeneity. In addition, zeta potential measurement revealed a negative surface charge under physiological conditions (Fig. 3C), suggesting that the vesicles possessed favorable colloidal stability in aqueous environments. These results confirmed the successful isolation of structurally intact EVs suitable for subsequent hydrogel loading. Western blot analysis further showed positive expression of the plant EV-associated markers TET8, HSP70, and PEN1 in the isolated vesicles. Comparable marker expression was observed between EVs prepared by differential centrifugation and density-gradient centrifugation (Figure S3A). In addition, the Limulus amebocyte lysate assay showed very low endotoxin levels in both EV preparations, with no significant difference between the two methods. (Figure S3B-C).

Fig. 3.

Preparation and characterization of hydrogels. (A) TEM image of EVs. (B) The particle size distribution of EVs. (C) The Zeta potential of EVs. (D) Schematic illustration of the preparation of GP and OKGM hydrogels. (E) FTIR spectra of Gel, 3APBA, and GP. (F) FTIR spectra of KGM and OKGM. (G) FTIR spectra of OKGM, GP, and GPOK. (H) Representative photographic images of GPOK hydrogels. (I) Frequency‑sweep rheological curves of GPOK hydrogels with different component ratios. (J) Quantitative analysis of the storage modulus (G′) of GPOK hydrogels with different component ratios at 1 Hz. (K–N) Strain‑sweep rheological curves of GPOK‑1 (K), GPOK‑2 (L), GPOK‑3 (M), and GPOK‑2@EVs (N). (O) Quantitative analysis of the fracture strain of hydrogels with different component ratios. (P) SEM images of GPOK‑2 and GPOK‑2@EVs. (Q) The swelling curve of GPOK hydrogel in PBS. Data are presented as mean ± SD. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

To achieve sustained delivery of EVs at the wound site, a dynamically cross-linked GPOK hydrogel was designed by first preparing two functional precursors, GP and OKGM, followed by in situ network formation. As illustrated in Fig. 3D, gelatin was modified with 3-aminophenylboronic acid (3APBA) through EDC/NHS-mediated amidation to yield GP, while konjac glucomannan (KGM) was oxidized by NaIO₄ to introduce aldehyde groups and generate OKGM. Subsequent mixing of GP and OKGM enabled the simultaneous formation of Schiff base bonds and borate ester linkages, thereby constructing a dual dynamic cross-linked hydrogel network. Such a design was expected to endow the hydrogel with both structural stability and dynamic responsiveness, which are desirable features for wound dressing applications. The successful synthesis of GP and OKGM, as well as the formation of the final GPOK hydrogel, was verified by Fourier-transform infrared (FTIR) spectroscopy. In the spectrum of GP, a characteristic B–O vibration peak appeared at approximately 1340 cm⁻¹, confirming the successful conjugation of 3APBA onto the gelatin backbone (Fig. 3E). For OKGM, the appearance of a distinct C = O stretching peak at around 1730 cm⁻¹, compared with native KGM, indicated successful oxidation and aldehyde generation (Fig. 3F). After hydrogel formation, the FTIR spectrum of GPOK simultaneously exhibited the characteristic signals derived from both GP and OKGM, together with an enhanced C = N peak near 1630 cm⁻¹ (Fig. 3G), supporting the formation of Schiff base linkages within the network. These spectroscopic results collectively demonstrated the successful establishment of the designed dual-crosslinked hydrogel structure. The macroscopic appearance of the precursor solutions and the formed hydrogel is shown in Fig. 3H. Upon mixing GP and OKGM, a stable hydrogel was rapidly formed and retained its shape upon vial inversion, confirming efficient gelation. To further evaluate the mechanical behavior of the hydrogels, rheological characterization was carried out. Frequency sweep analysis showed that, across the tested frequency range, the storage modulus (G′) of all formulations remained consistently higher than the loss modulus (G″) (Fig. 3I), indicating dominant elastic behavior and the successful formation of stable gel networks. Quantitative comparison of G′ at 1 Hz demonstrated that the mechanical strength of the hydrogels could be modulated by changing the component ratio (Fig. 3J). Among the tested formulations, GPOK-2 exhibited the highest storage modulus, whereas GPOK-1 showed the lowest. Notably, the incorporation of EVs into GPOK-2 did not significantly alter the storage modulus, indicating that EV loading had minimal impact on the bulk mechanical properties of the hydrogel network.

To assess the deformation resistance of the hydrogels, strain sweep rheological tests were performed for GPOK-1, GPOK-2, GPOK-3, and GPOK-2@EVs (Fig. 3K–N). Both GPOK-1 and GPOK-3 showed gel network collapse when the strain reached approximately 95.8%, as evidenced by the crossover of G′ and G″ (Fig. 3K,M). In contrast, GPOK-2 and GPOK-2@EVs tolerated substantially higher strain before network failure, with crossover occurring at around 191% strain (Fig. 3L,N). The quantitative analysis of fracture strain further confirmed that GPOK-2 and GPOK-2@EVs possessed markedly improved anti-deformation capability compared with the other formulations (Fig. 3O). These results indicate that the GPOK-2 formulation achieved a more favorable balance between stiffness and flexibility, which is particularly important for wound dressings expected to withstand dynamic mechanical disturbance in vivo. The internal microstructure of the optimized hydrogel was further examined by scanning electron microscopy (SEM). As shown in Fig. 3P, GPOK-2 exhibited a uniform and interconnected porous architecture, which is beneficial for nutrient transport and cargo loading. After EV incorporation, GPOK-2@EVs largely preserved this porous network, although the pore walls appeared slightly rougher, suggesting successful EV integration within the matrix. To further confirm EV encapsulation within the hydrogel matrix, PKH26-labeled EVs were incorporated into GPOK-2 and visualized by three-dimensional fluorescence imaging. As shown in Figure S4, red fluorescence signals were distributed throughout the reconstructed hydrogel volume, indicating successful loading of EVs within the three-dimensional network. The relatively uniform fluorescence distribution and absence of obvious large aggregates suggest that EVs were well dispersed during the in situ gelation process. This spatially homogeneous EV loading is favorable for local retention and sustained release, further supporting the suitability of GPOK-2 as an EV delivery platform for diabetic wound repair. The swelling behavior of the GPOK hydrogels was further evaluated (Fig. 3Q). The swelling ratios of all hydrogel groups increased rapidly during the initial stage and subsequently reached a plateau, indicating that their hydrophilic porous networks could effectively absorb water. Among the formulations, GPOK-1 exhibited the highest swelling ratio, which may be attributed to its relatively loose network structure and low crosslinking density. As the GP concentration increased, the swelling ratio gradually decreased. This decrease may result from the formation of a denser polymer network and enhanced chain entanglement, which restricted water diffusion and matrix expansion. GPOK-2 displayed a moderate swelling ratio, reflecting an optimal balance between water absorption capacity and structural stability. This property is advantageous for wound dressing applications, as it enables the hydrogel to absorb wound exudate while maintaining matrix integrity. Moreover, EV incorporation had minimal influence on the swelling behavior of GPOK-2, and GPOK-2@EVs retained a stable swelling profile.

Overall, these results demonstrate that Artemisia argyi-derived EVs were successfully isolated with appropriate nanoscale features and surface stability, and that the GPOK hydrogels were successfully fabricated through dual dynamic crosslinking. Among the tested formulations, GPOK-2 showed the most suitable mechanical profile, combining relatively high elasticity with enhanced anti-deformation performance. Importantly, EV incorporation did not compromise the structural or rheological characteristics of the hydrogel, providing a solid material basis for subsequent evaluation of its adhesive behavior, stimulus-responsive degradation, controlled EV release, and biological wound-healing activity.

Adhesion properties and responsive release performance of hydrogels

For wound dressings used for skin repair, appropriate mechanical properties are fundamental to their application, whereas excellent self-healing capacity, injectability, tissue adhesion, stimuli-responsive degradability, and controlled-release capability are the core criteria for evaluating their clinical translation potential. The dynamic covalently cross-linked network composed of Schiff base bonds and borate ester bonds endows the GPOK hydrogel with a rich dynamic response. The self-healing performance of the hydrogel determines its ability to resist structural damage caused by skin deformation and to maintain a stable physical barrier over the wound. This property was evaluated through alternating strain sweep tests, and the corresponding results are shown in Fig. 4A. At a fixed angular frequency of 10 rad/s, the oscillatory strain alternated between 3.0% (low strain) and 300% (high strain), with each strain level maintained for 100 s. At low strain, the storage modulus (G′) of the GPOK-2@EVs hydrogel was significantly higher than the loss modulus (G″), indicating a stable elastic solid-like state. When a high strain of 300% was applied, G′ and G″ instantly reversed and the moduli dropped sharply, indicating disruption of the dynamic cross-linked network and transformation of the hydrogel into a viscous-flow state. Notably, when the strain was reduced back to 3.0%, the modulus of the hydrogel returned to its initial value within a few seconds, and this rapid recovery behavior remained essentially unchanged over multiple continuous strain cycles. Macroscopic self-healing experiments further visually verified this property (Fig. 4B). Two cut hydrogel blocks, placed in contact at room temperature without any external stimulation, achieved self-healing after a short period of juxtaposition. This excellent self-healing property enables the hydrogel to continuously maintain the barrier function of the dressing after application to the wound surface, thereby avoiding secondary damage to the fragile wound tissue.

Fig. 4.

Adhesion and stimuli‑responsive release performance of hydrogels. (A) Alternating strain‑amplitude curve of GPOK‑2@EVs. (B) Representative image of the self‑healing behavior of the GPOK‑2@EVs hydrogel. (C) Shear‑thinning curve of the GPOK‑2@EVs hydrogel. (D) Representative images of the injectable behavior of the GPOK‑2 hydrogel. (E) Schematic illustration of the burst‑pressure test of the hydrogel. (F) Quantitative analysis of the burst pressure of hydrogels with different component ratios. (G) Lap‑shear adhesion curves of pig skin with different component ratios of hydrogels. (H) Quantitative analysis of the adhesion strength of hydrogels with different component ratios to pig skin. (I) In vitro degradation curves of GPOK-2@EVs hydrogels under different conditions, including PBS at pH 7.4, PBS at pH 5.5, PBS containing 1 mM H₂O₂ at pH 7.4, and PBS containing 1 mM H₂O₂ at pH 5.5. (J) Representative images of the degradation of the GPOK‑2@EVs hydrogel in PBS at pH 5.5 and in PBS containing 1 mM H₂O₂. (K) Cumulative EV release curve of GPOK-2@EVs in a simulated diabetic wound microenvironment, namely PBS containing 1 mM H₂O₂ at pH 5.5. Data are presented as mean ± SD. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

The injectability of hydrogels is a key advantage for adapting to irregular wounds and is closely related to their shear-thinning behavior. As shown by the results of the shear-thinning experiment (Fig. 4C), the viscosity of the GPOK-2@EVs hydrogel decreased significantly with increasing shear rate, exhibiting typical shear-thinning characteristics. This phenomenon is attributed to the reversible breakage of dynamic cross-linking bonds within the hydrogel network under high shear forces, which reduces the viscosity of the system. Macroscopic injection tests further confirmed the excellent injectability of the hydrogel (Fig. 4D). The GPOK hydrogel could be smoothly extruded through a standard syringe needle while maintaining its integrity without obvious diffusion.

Firm and long-lasting tissue adhesion is essential for wound dressings to remain fixed on the dynamic skin surface, seal the wound, prevent invasion by external bacteria, and maintain a moist healing microenvironment. First, the sealing ability of the hydrogel on tissue was evaluated by a burst-pressure test (Fig. 4E), and the quantitative results are shown in Fig. 4F. The burst pressure of the GPOK hydrogels first increased and then decreased with increasing GP concentration. Among them, the burst pressures of GPOK-2 and GPOK-2@EVs were significantly higher than those of GPOK-1 and GPOK-3, indicating that these hydrogels can withstand higher physiological pressures and maintain stable sealing performance on the wound surface. Meanwhile, lap-shear tests were conducted on fresh pig skin to quantitatively evaluate the adhesion strength between the hydrogel and the skin tissue (Fig. 4G–H). Consistent with the burst-pressure results, the adhesion strength of GPOK-2 to pig skin was the highest and was significantly greater than that of the GPOK-1 and GPOK-3 groups. Notably, there was no significant difference in adhesion strength between GPOK-2 and GPOK-2@EVs, confirming that the loading of EVs does not impair the tissue-adhesion performance of the hydrogel. The excellent tissue adhesion of the GPOK hydrogel is mainly attributed to multiple interfacial interactions between the hydrogel and the skin tissue. The abundant amino and phenylboronic acid groups on the GP molecular chains can form hydrogen bonds and dynamic boronate ester bonds with hydroxyl groups and glycoproteins on the surface of skin tissue, whereas the residual aldehyde groups on OKGM can form Schiff base bonds with amino groups on tissue proteins. Thus, firm adhesion to the skin surface is achieved. The sealing ability of GPOK hydrogels depends not only on the density of adhesive functional groups but also on the balance between interfacial adhesion and stress dissipation. In GPOK-1, the relatively low GP concentration provides insufficient amino and phenylboronic acid groups for constructing an effective dual-crosslinked network with OKGM, resulting in a loosely crosslinked structure with limited cohesive strength and fewer tissue-interactive sites, thereby reducing resistance to pressure-induced rupture. In contrast, GPOK-2 contains an intermediate GP concentration, which provides a more favorable stoichiometric balance between GP and OKGM and enables the formation of a uniform and interconnected dynamic network through Schiff base and borate ester bonds. This optimized network possesses sufficient stiffness to maintain structural integrity while retaining adequate chain mobility for deformation and mechanical stress dissipation through reversible bond breakage and reformation. Compared with GPOK-1 and GPOK-3, GPOK-2 exhibited the highest adhesion strength and markedly improved fracture strain. Therefore, GPOK-2 can effectively withstand pressure loading without premature adhesive failure or interfacial detachment. However, further increasing the GP concentration in GPOK-3 may induce localized excessive polymer-chain entanglement or overcrosslinking. Such a network may restrict chain rearrangement, reduce deformability, lower the fracture strain of GPOK-3 relative to GPOK-2, and generate stress-concentration regions under pressure, thereby weakening the overall sealing performance. Collectively, GPOK-2 achieved the most favorable structural balance between adhesive performance and mechanical adaptability, thereby exhibiting the highest burst pressure among the tested formulations.

The degradation behavior of wound dressings needs to be closely matched to the dynamic healing process of skin wounds. Since diabetic wounds are characterized by an acidic and ROS-rich inflammatory microenvironment, we further evaluated the degradation behavior of GPOK-2@EVs under single and combined stimuli (Fig. 4I and 4J). In neutral PBS at pH 7.4, the hydrogel exhibited slow degradation and maintained structural stability for a relatively long period, indicating good stability under physiological conditions. In contrast, degradation was accelerated in acidic PBS at pH 5.5 and in PBS containing 1 mM H₂O₂, confirming the independent pH- and ROS-responsive degradation behavior of the hydrogel. Notably, when the hydrogel was incubated in PBS containing 1 mM H₂O₂ at pH 5.5, which mimics the acidic and oxidative microenvironment of diabetic wounds, the degradation rate was further increased compared with either single-stimulus condition. This accelerated degradation can be attributed to the simultaneous disruption of the dual dynamic network: Schiff base bonds undergo acid-promoted hydrolysis, whereas boronate ester bonds are cleaved under ROS stimulation. Therefore, the combined acidic/ROS condition induces more efficient network disintegration than either stimulus alone. These results confirm the pH/ROS dual-responsive degradation behavior of GPOK-2@EVs and support its capacity for on-demand degradation and EV release in the diabetic wound microenvironment. KGM is a natural β-1,4-linked heteropolysaccharide mainly composed of D-mannose and D-glucose residues [60]. Periodate oxidation introduces aldehyde groups into OKGM and partially opens the sugar rings, increasing its susceptibility to hydrolytic and oxidative degradation [61, 62]. In diabetic wounds, the acidic and ROS-rich microenvironment may promote hydrogel network disintegration through Schiff base hydrolysis and boronate ester cleavage, generating soluble OKGM-derived fragments that can be gradually cleared through wound exudate, local cellular uptake/phagocytic clearance, and systemic elimination.

On the basis of the pH/ROS dual-responsive degradation performance of the hydrogels, we further investigated the in vitro release behavior of EVs from GPOK-2@EVs hydrogels in a simulated diabetic wound microenvironment (PBS containing 1 mM H₂O₂ at pH 5.5), and the cumulative release curves are shown in Fig. 4K. In the simulated diabetic wound microenvironment, EVs were rapidly released within 72 h, and the release rate was significantly higher than that in the PBS environment, confirming the responsive release performance of the GPOK hydrogel for EVs. The pH/ROS-responsive degradation profile of GPOK-2@EVs may better match the temporal requirements of diabetic wound repair. During the early inflammatory stage, diabetic wounds are characterized by an acidic microenvironment and elevated oxidative stress, which require rapid therapeutic intervention [63]. Accordingly, the accelerated degradation of GPOK-2@EVs under the combined acidic/ROS condition enables early EV release within approximately 72 h. As the wound microenvironment gradually improves and the inflammatory burden decreases, reduced acidic and oxidative stimuli may slow further network dissociation, allowing the residual hydrogel matrix to maintain local coverage and a moist environment during subsequent granulation tissue formation, angiogenesis, and re-epithelialization.

In summary, the GPOK composite hydrogel exhibits excellent self-healing performance, injectability, and strong adhesion to skin tissue, as well as pH/ROS dual-responsive degradation and controlled EV-release performance. These functional characteristics fully meet the practical requirements of skin wound repair, especially under the complex inflammatory microenvironment of chronic diabetic wounds, providing a reliable functional carrier for the local and sustained delivery of EVs and laying a solid foundation for subsequent evaluation of its biological efficacy in skin wound repair.

Biocompatibility and biosafety evaluation of GPOK-2@EVs

Ideal dressings must not only facilitate tissue regeneration but also exhibit negligible cytotoxicity, hemocompatibility, and systemic toxicity to avoid exacerbating the already compromised diabetic microenvironment [64]. Previous studies have highlighted that while natural polymers generally possess favorable safety profiles, chemical modifications and high concentrations of nanovesicles can occasionally induce unforeseen cellular stress or immune reactions [65]. Therefore, we systematically evaluated the cytocompatibility, hemocompatibility, and systemic biosafety of the GPOK-2@EVs hydrogel through a series of rigorous in vitro and in vivo assays. The cytocompatibility of the hydrogel was first examined using RAW264.7 macrophages and L929 fibroblasts, two commonly used cell models for evaluating wound-healing materials [66]. Live/dead staining showed that the majority of cells remained viable after treatment with Control, GPOK-2, EVs, and GPOK-2@EVs (Fig. 5A,B). Strong green fluorescence corresponding to live cells was observed in all groups, while only a few red fluorescent dead cells appeared. Quantitative analysis further confirmed that cell viability remained above 90% with no significant differences among groups (Fig. 5C). In addition, CCK-8 assays demonstrated that cell viability was not significantly affected by hydrogel or EV treatment (Fig. 5D), indicating that both the hydrogel matrix and EV cargo possess favorable cytocompatibility. Hemocompatibility was further evaluated using a hemolysis assay. As shown in Fig. 5F, PBS (negative control) and GPOK-2@EVs showed minimal hemolysis, whereas Triton X-100 (positive control) resulted in complete hemolysis. Quantitative analysis demonstrated that the hemolysis ratio of GPOK-2@EVs was extremely low and comparable to that of PBS (Fig. 5G). This excellent hemocompatibility can be attributed to the hydrophilic nature of the hydrogel network and the negative surface charge of the plant-derived EVs, which minimize nonspecific interactions with the RBC membrane [67].

Fig. 5.

Biocompatibility and hemocompatibility evaluation of GPOK-2@EVs in vitro and in vivo. (A) Live/dead staining of RAW264.7 macrophages treated with Control, GPOK-2, EVs, and GPOK-2@EVs. Live cells are stained green and dead cells are stained red. Scale bar = 100 μm. (B) Live/dead staining of L929 fibroblasts under different treatments to evaluate cytocompatibility. Scale bar = 500 μm. (C) Quantitative analysis of cell viability of RAW264.7 and L929 cells after treatment with Control, GPOK-2, EVs, and GPOK-2@EVs (n = 3). (D) CCK-8 assay evaluating the viability of RAW264.7 and L929 cells after different treatments (n = 3). (E) Body weight changes of mice during the treatment period in Control and GPOK-2@EVs groups (n = 3). (F) Representative photographs of hemolysis assay tubes for PBS (negative control), GPOK-2@EVs, and Triton X-100 (positive control). (G) Quantitative analysis of hemolysis ratios in PBS, GPOK-2@EVs, and Triton X-100 groups (n = 3). (H) Hematological and blood biochemical parameters, including WBC, HGB, RBC, NEU, ALT, AST, UREA, and CREA, used to assess systemic biosafety (n = 3). (I) Representative H&E staining images of major organs (heart, lung, kidney, liver, and spleen) from the Control and GPOK-2@EVs groups, showing no obvious histopathological abnormalities. Scale bar = 100 μm. Data are presented as mean ± SD. ns, not significant

To further evaluate systemic biosafety in vivo, mice treated with GPOK-2@EVs were monitored for physiological and biochemical changes. No significant difference in body weight was observed between the control and treatment groups during the experimental period (Fig. 5E). Moreover, hematological parameters (WBC, HGB, RBC, and NEU) and biochemical indicators related to liver and kidney function (ALT, AST, UREA, and CREA) remained within normal ranges (Fig. 5H). Histological examination of major organs further revealed normal tissue structures without obvious inflammatory infiltration or pathological damage in the GPOK-2@EVs group (Fig. 5I). Collectively, these results demonstrate that GPOK-2@EVs hydrogel exhibits excellent cytocompatibility, hemocompatibility, and systemic biosafety.

GPOK-2@EVs reprograms macrophage polarization toward an anti-inflammatory phenotype

Our preceding single-cell transcriptomic and multi-omics analyses identified marked macrophage accumulation, an M1/M2 imbalance, and enrichment of endoplasmic reticulum stress-related signaling in pro-inflammatory macrophages in diabetic wounds. These findings suggest that macrophage dysregulation is a central event in the diabetic wound microenvironment. Because persistent M1 polarization sustains inflammatory signaling and oxidative stress, thereby impairing tissue repair and angiogenesis [8, 68, 69], we next investigated whether GPOK-2@EVs could modulate macrophage polarization and restore immune homeostasis. Immunofluorescence staining was first used to assess macrophage polarization. Macrophages in the Control and GPOK-2 groups exhibited relatively strong CD86 fluorescence, whereas EVs and GPOK-2@EVs treatment markedly reduced CD86 expression, which was further supported by quantitative analysis (Fig. 6A,C). In contrast, the M2 marker CD206 displayed the opposite trend: EVs and GPOK-2@EVs markedly increased CD206 fluorescence intensity compared with the Control and GPOK-2 groups, as confirmed by quantitative analysis (Fig. 6B,D). These data indicate that EV-containing treatments shift macrophages away from a pro-inflammatory phenotype toward a reparative state, with hydrogel-assisted delivery producing the strongest effect.

Fig. 6.

GPOK-2@EVs effectively reprograms macrophage polarization toward an anti-inflammatory M2 phenotype and attenuates pro-inflammatory cytokine release. (A) Immunofluorescence staining of the M1 macrophage marker CD86 in RAW264.7 cells treated with Control, GPOK-2, EVs, and GPOK-2@EVs. Nuclei were stained with DAPI. Scale bar = 50 μm. (B) Immunofluorescence staining of the M2 macrophage marker CD206 in RAW264.7 cells under different treatments. Nuclei were stained with DAPI. Scale bar = 50 μm. (C) Quantitative analysis of CD86 fluorescence intensity in macrophages (n = 3). (D) Quantitative analysis of CD206 fluorescence intensity in macrophages (n = 3). (E) Flow cytometry analysis of macrophage polarization using F4/80 together with CD86 and CD206 under different treatments. (F) Quantification of the proportions of CD86⁺ and CD206⁺ macrophages determined by flow cytometry (n = 3). (G) Western blot analysis of macrophage polarization-related proteins, including CD86 and CD206. GAPDH was used as the loading control. (H) Quantitative analysis of protein expression levels based on the Western blot results (n = 3). (I) RT–qPCR analysis of macrophage polarization-related genes, including CD206, IL-10, TGF-β, TNF-α, IL-1β, and iNOS (n = 3). (J) ELISA analysis of inflammatory cytokines secreted by macrophages, including TNF-α, IL-1β, IL-6, IL-10, TGF-β, and IL-4 (n = 3). Data are presented as mean ± SD. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

To further quantify this phenotypic shift, flow cytometry was performed using F4/80 together with CD86 and CD206 to identify M1- and M2-like macrophages, respectively. As shown in Fig. 6E,F, EVs and GPOK-2@EVs significantly decreased the proportion of CD86⁺ M1 macrophages while increasing the proportion of CD206⁺ M2 macrophages among F4/80⁺ cells. Such a shift is highly desirable in diabetic wound therapy, as increasing the M2/M1 ratio has been directly correlated with accelerated closure rates and improved collagen deposition in previous studies [70, 71]. Consistent with these observations, Western blot analysis showed that EVs and GPOK-2@EVs reduced the expression of CD86 while increasing the levels of CD206 (Fig. 6G,H). RT-qPCR further demonstrated increased expression of M2-associated markers and anti-inflammatory cytokines, including CD206, IL-10, and TGF-β, together with decreased transcription of pro-inflammatory genes such as TNF-α, IL-1β, and iNOS (Fig. 6I). ELISA analysis showed a similar pattern at the secretory level, with reduced release of TNF-α, IL-1β, and IL-6 and increased production of IL-10, TGF-β, and IL-4 following EV-containing treatments (Fig. 6J). Because chronic diabetic wounds are characterized by sustained inflammatory signaling and excessive cytokine production that hinder tissue repair [72, 73], these findings support the view that GPOK-2@EVs helps re-establish immune balance within the wound microenvironment. Collectively, GPOK-2@EVs effectively reprogrammed macrophages by suppressing M1-associated inflammatory activation while promoting M2-associated anti-inflammatory responses. This immunomodulatory effect is highly relevant to chronic diabetic wounds, where unresolved inflammation remains a major barrier to tissue regeneration [74, 75]. By rebalancing macrophage polarization and limiting inflammatory cytokine production, the GPOK-2@EVs hydrogel may create a more favorable microenvironment for subsequent wound repair.

GPOK-2@EVs alleviates ERS and oxidative stress in macrophages

In diabetic microenvironments, persistent ERS can chronically activate pro-inflammatory signaling and, through its close crosstalk with mitochondria, induce mitochondrial fragmentation, membrane potential loss, and excessive ROS generation, thereby aggravating local tissue injury [76]. Targeting the ERS–mitochondria–ROS axis may therefore represent an effective strategy to restore immune balance in chronic wounds [19, 77]. ERS probe staining was used to visualize the intracellular ERS state. As shown in Fig. 7A, the Control and GPOK-2 groups exhibited strong red fluorescence, indicating accumulation of aggregated misfolded proteins, whereas EV and GPOK-2@EVs treatments shifted the fluorescence toward the green monomeric state, suggesting alleviation of ERS. Mitochondrial dynamics, governed by the balance between fission and fusion, are critical for cellular health; excessive fission leads to mitochondrial fragmentation, rendering the organelles prone to ROS leakage and apoptosis [78]. MitoTracker staining showed that EV and GPOK-2@EVs treatments restored a more elongated mitochondrial morphology compared with the fragmented mitochondria observed in control macrophages (Fig. 7B). In parallel, ROS staining demonstrated a marked reduction in intracellular ROS accumulation in EV-treated cells (Fig. 7C), and quantitative analysis confirmed a significant decrease in ROS fluorescence intensity (Fig. 7D). Consistent with the observed improvement in mitochondrial morphology, quantitative analysis of JC-1 staining further indicated that EV-containing treatments restored mitochondrial membrane potential in macrophages (Figure S5).

Fig. 7.

GPOK-2@EVs attenuate ERS and oxidative stress in macrophages. (A) ERS probe staining of RAW264.7 macrophages treated with Control, GPOK-2, EVs, and GPOK-2@EVs. Green fluorescence represents the monomeric form of the probe, whereas red fluorescence indicates probe aggregation associated with ER stress. Nuclei were stained with DAPI. Scale bar = 10 μm. (B) MitoTracker staining showing mitochondrial morphology in macrophages under different treatments. Nuclei were stained with DAPI. Scale bar = 10 μm. (C) Intracellular reactive oxygen species (ROS) levels detected using the DCFH-DA fluorescent probe. Scale bar = 50 μm. (D) Quantitative analysis of intracellular ROS fluorescence intensity (n = 3). (E) Transmission electron microscopy (TEM) images showing mitochondrial ultrastructure in macrophages under different treatments. Scale bar = 250 nm. (F) Quantitative analysis of mean mitochondrial length (n = 3). (G) Western blot analysis of ER stress–related proteins, including GRP78, PERK, phosphorylated PERK (p-PERK), eIF2α, phosphorylated eIF2α (p-eIF2α), ATF4, and CHOP. GAPDH served as the loading control. (H) Quantitative analysis of protein expression levels based on the Western blot results (n = 3). Data are presented as mean ± SD. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Transmission electron microscopy further verified the protective effect of EV treatment on mitochondrial ultrastructure. As shown in Fig. 7E, mitochondria in the Control and GPOK-2 groups displayed swelling and disrupted cristae, whereas EV- and GPOK-2@EVs-treated cells preserved relatively intact mitochondrial ultrastructure. Quantitative analysis also showed a significant increase in mean mitochondrial length following EV treatment (Fig. 7F), indicating reduced mitochondrial fragmentation. To further validate these observations at the protein level, Western blot analysis was performed. Macrophages treated with EVs or GPOK-2@EVs exhibited markedly reduced expression of ERS related proteins, including GRP78, p-PERK, p-eIF2α, ATF4, and CHOP, whereas total PERK and eIF2α levels remained relatively stable (Fig. 7G,H). GRP78 and CHOP are classical indicators of ERS activation and sustained UPR signaling [76, 79, 80], suggesting that EV-containing treatments effectively suppress ERS signaling in macrophages. To further test whether EVs remained effective under forced ERS activation, macrophages were treated with tunicamycin. Tunicamycin markedly increased ERS-related proteins and iNOS expression while reducing Arg-1 expression (Figure S6A). Under this condition, EVs, especially GPOK-2@EVs, reduced ERS-related protein expression and shifted macrophage markers from iNOS toward Arg-1, suggesting preserved immunomodulatory activity under forced ER stress conditions (Figure S6B,C). Collectively, these results demonstrate that GPOK-2@EVs attenuates ERS signaling, reduces oxidative stress, and preserves mitochondrial integrity in macrophages.

GPOK-2@EVs enhances endothelial angiogenic activities in vitro

Angiogenesis is a critical step in the wound-healing process, particularly in diabetic wounds where impaired vascularization significantly delays tissue regeneration [81, 82]. Since macrophage-mediated inflammatory resolution can promote endothelial activation and vascular remodeling [83–85], we next investigated whether GPOK-2@EVs could directly enhance endothelial cell angiogenic activities. Cell viability was first evaluated to confirm the cytocompatibility of the treatments toward endothelial cells. Live/dead staining revealed that the majority of cells remained viable under all treatment conditions, with abundant green fluorescence and minimal red fluorescence observed in each group (Fig. 8A). Quantitative analysis of cell viability at 0 h, 24 h, and 48 h further confirmed that none of the treatments induced cytotoxic effects (Fig. 8B), indicating that the hydrogel formulations and EV cargo are well tolerated by endothelial cells. To examine the pro-angiogenic potential of the treatments, the expression of angiogenesis-related markers was evaluated. VEGF is a central regulator of endothelial proliferation and angiogenic signaling, whereas CD31 is widely used as a marker of endothelial differentiation and vascular structures [86]. Quantitative RT–qPCR analysis showed that EVs and GPOK-2@EVs significantly upregulated the transcription of angiogenesis-related genes in endothelial cells (Figure S7). Consistent with these transcriptional changes, Western blot analysis further demonstrated that EV and GPOK-2@EVs treatments markedly increased the protein expression levels of VEGF and CD31 (Fig. 8C), which was further confirmed by quantitative analysis (Fig. 8D). These findings indicate that EV-containing treatments can stimulate pro-angiogenic signaling pathways in endothelial cells.

Fig. 8.

GPOK-2@EVs promote endothelial cell angiogenic activities in vitro. (A) Live/dead staining of HUVECs treated with Control, GPOK-2, EVs, and GPOK-2@EVs. Green fluorescence indicates live cells and red fluorescence indicates dead cells. Scale bar = 500 μm. (B) Quantitative analysis of HUVEC viability at 0 h, 24 h, and 48 h after different treatments (n = 3). (C) Western blot analysis of angiogenesis-related proteins VEGF and CD31 in HUVECs. GAPDH was used as the loading control. (D) Quantitative analysis of VEGF and CD31 protein expression levels based on the Western blot results (n = 3). (E) Immunofluorescence staining of F-actin in HUVECs. Nuclei were stained with DAPI. Scale bar = 50 μm. (F) Immunofluorescence staining of CD31 in HUVECs. Nuclei were stained with DAPI. Scale bar = 50 μm. (G) Quantitative analysis of F-actin fluorescence intensity (n = 3). (H) Quantitative analysis of CD31 fluorescence intensity (n = 3). (I) Representative images of tube formation assay of HUVECs cultured on Matrigel. Scale bar = 200 μm. (J) Quantitative analysis of the number of branch points in the tube formation assay (n = 3). (K) Quantitative analysis of relative tube length in the tube formation assay (n = 3). (L) Transwell migration assay showing migrated HUVECs stained with crystal violet. Scale bar = 100 μm. (M) Quantitative analysis of endothelial cell migration rates (n = 3). (N) Representative images of scratch wound-healing assay at 0 h, 24 h, and 48 h. Scale bar = 500 μm. (O) Quantitative analysis of relative scratch area in the wound-healing assay (n = 3). Data are presented as mean ± SD. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

F-actin immunofluorescence staining was used to evaluate endothelial cytoskeletal organization. As shown in Fig. 8E, EVs and GPOK-2@EVs enhanced F-actin fluorescence intensity and promoted a more organized cytoskeletal structure compared with the control group, which was confirmed by quantitative analysis (Fig. 8G). Similarly, CD31 immunofluorescence was also increased in EV-containing groups (Fig. 8F, H). The angiogenic activity of endothelial cells was further assessed by tube formation assay. As shown in Fig. 8I, EVs and GPOK-2@EVs markedly promoted the formation of more organized capillary-like networks compared with the Control and GPOK-2 groups. Quantitative analysis further confirmed significant increases in both branch points and relative tube length (Fig. 8J, K), indicating enhanced endothelial angiogenic capacity. Endothelial migration is another key step in vascular remodeling during tissue repair [87]. Transwell assays showed that EVs and GPOK-2@EVs significantly promoted endothelial cell migration, which was confirmed by quantitative analysis (Fig. 8L, M). Consistently, scratch assays showed faster wound closure in EV-containing groups at both 24 and 48 h (Fig. 8N, O). Collectively, these results indicate that GPOK-2@EVs enhances endothelial angiogenic behaviors, including cytoskeletal remodeling, tube formation, and migration. Together with its effects on macrophage reprogramming and stress attenuation, these findings suggest that GPOK-2@EVs may facilitate vascular regeneration and microenvironment remodeling in diabetic wounds.

GPOK-2@EVs accelerates diabetic wound healing and modulates inflammation and ERS in vivo

Encouraged by the favorable in vitro immunomodulatory and pro-angiogenic effects, we next evaluated the therapeutic efficacy of GPOK-2@EVs in an STZ-induced diabetic wound model (Fig. 9A). To further assess EV retention in vivo, fluorescently labeled GPOK-2@EVs were applied to diabetic wounds. In vivo fluorescence imaging showed that EV-associated signals persisted at the wound site during the 14-day observation period (Figure S8), suggesting prolonged local retention after hydrogel delivery. Representative photographs showed that GPOK-2@EVs markedly accelerated wound closure compared with the Control, GPOK-2, and EV groups throughout the 14-day observation period (Fig. 9B). Quantitative analysis confirmed a significantly smaller residual wound area in the GPOK-2@EVs group, particularly during the middle and late stages of healing (Fig. 9C, D). As a positive control, INTRASITE™ Gel also promoted wound closure over 14 days. GPOK-2@EVs showed a comparable healing effect to this commercial hydrogel, supporting its favorable therapeutic efficacy (Figure S9A, B). These results indicate that hydrogel-assisted EV delivery provides superior therapeutic efficacy over free EVs, likely by improving local retention and sustained bioactivity in the diabetic wound microenvironment, where excessive inflammation and poor vascular support usually compromise the activity of therapeutic agents [88]. Histological analysis further verified the improved healing quality. H&E staining showed more complete re-epithelialization and thicker granulation tissue in the GPOK-2@EVs group (Fig. 9E), while Masson’s trichrome staining revealed denser and more organized collagen deposition (Fig. 9F). Quantitative analysis confirmed significant increases in granulation tissue thickness, granulation tissue area, and collagen deposition after GPOK-2@EVs treatment (Fig. 9G–I). We further evaluated neural growth in regenerated wound tissues by PGP9.5 immunofluorescence staining. GPOK-2@EVs showed higher PGP9.5-positive signals than the other groups, suggesting favorable neural growth in the regenerated wound tissue (Figure S10A, B). Because re-epithelialization, granulation tissue formation, and extracellular matrix remodeling are key indicators of functional wound repair rather than simple surface closure [89, 90], these findings suggest that GPOK-2@EVs not only accelerates wound closure but also improves the quality of tissue reconstruction.

Fig. 9.

GPOK-2@EVs accelerate diabetic wound healing and modulate inflammation and ERS in vivo. (A) Schematic illustration of the streptozotocin (STZ)-induced diabetic mouse wound model and treatment timeline. Diabetes was induced by STZ injection, followed by full-thickness dorsal skin wound creation and hydrogel treatment. Wound healing progression was monitored at days 1, 3, 7, 10, and 14. (B) Representative photographs of wound healing in the Control, GPOK-2, EVs, and GPOK-2@EVs groups at different time points after treatment. (C) Quantitative analysis of wound area during the healing process among different treatment groups (n = 3). (D) Representative wound area distribution maps illustrating wound contraction and closure dynamics in different treatment groups at different time points. (E) Hematoxylin and eosin (H&E) staining of wound tissues collected on day 14, showing re-epithelialization and granulation tissue formation. Scale bar = 500 μm. (F) Masson’s trichrome staining showing collagen deposition and extracellular matrix remodeling in wound tissues. Scale bar = 50 μm. (G) Quantitative analysis of granulation tissue thickness in wound sections (n = 3). (H) Quantitative analysis of granulation tissue area in wound sections (n = 3). (I) Quantitative analysis of collagen deposition based on Masson staining (n = 3). Data are presented as mean ± SD. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

To explore the underlying mechanism in vivo, we examined inflammatory status, angiogenesis, and ERS in wound tissues. Immunohistochemical staining showed that GPOK-2@EVs reduced the pro-inflammatory marker iNOS and increased the M2-associated marker Arg-1 (Fig. 10A), which was confirmed by quantitative analysis (Fig. 10B, C). At the same time, CD31 expression was markedly elevated in the GPOK-2@EVs group (Fig. 10A, D), indicating enhanced neovascularization. Since persistent M1-dominant inflammation is a major barrier to diabetic wound healing, while M2 polarization and vascular regeneration are closely associated with tissue restoration [91], these findings suggest that GPOK-2@EVs promotes repair by shifting the wound microenvironment toward a more reparative and pro-angiogenic state. Notably, GPOK-2@EVs also suppressed ERS in vivo. CHOP and GRP78 staining was markedly reduced in the GPOK-2@EVs group (Fig. 10A, E and F), and quantitative RT–PCR further showed decreased expression of ERS-related genes, including GRP78, ATF4, and CHOP, in wound tissues (Figure S11). Western blot analysis provided additional support, showing increased Arg-1 and CD31 expression together with reduced iNOS, p-PERK, p-eIF2α, GRP78, ATF4, and CHOP levels in the GPOK-2@EVs group (Fig. 10G). These findings suggest that the therapeutic benefit of GPOK-2@EVs is associated with coordinated regulation of macrophage polarization, angiogenesis, and ERS signaling. Compared with many previously reported EV-loaded hydrogels that mainly emphasize anti-inflammatory or pro-angiogenic effects [88, 92], our results additionally highlight suppression of ERS as an important component of the repair process, suggesting a more direct intervention in the stress-associated pathological microenvironment of diabetic wounds. Taken together, GPOK-2@EVs effectively accelerates diabetic wound healing by improving tissue regeneration, reducing inflammatory activation, enhancing vascularization, and attenuating ERS in wound tissues. This supports the potential of GPOK-2@EVs as a stress-responsive immunomodulatory hydrogel system for diabetic wound repair.

Fig. 10.

GPOK-2@EVs modulates macrophage polarization, angiogenesis, and ERS signaling in diabetic wounds. (A) Representative immunohistochemical staining images of inflammation, angiogenesis, and ERS-related markers in wound tissues from different treatment groups. Scale bar = 50 μm. (B–F) Quantitative analysis of the immunohistochemical staining intensity of iNOS, Arg-1, CD31, CHOP, and GRP78, respectively (n = 3). (G) Western blot analysis of proteins associated with macrophage polarization, angiogenesis, and ERS signaling in wound tissues. GAPDH served as the loading control. Data are presented as mean ± SD. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Although silicone splints were not used in this mouse wound model, the therapeutic effects of GPOK-2@EVs were supported by consistent histological, immunohistochemical, and molecular improvements; future studies using splinted excisional wound models are warranted to further validate these findings. From a translational perspective, scalable production of Artemisia argyi-derived EVs remains an important challenge. Future clinical application will require standardized plant sourcing, extraction, purification, storage, and batch-to-batch quality control, as these factors may affect EV yield, cargo composition, and biological activity. GMP-compatible workflows and defined quality attributes, including purity, endotoxin level, sterility, stability, and potency, should be established before clinical translation.

Conclusion

In conclusion, this work identifies ERS as an important mechanism contributing to macrophage dysfunction and persistent inflammation in diabetic wounds. Integrated single-cell transcriptomics and multi-omics analyses revealed significant enrichment of ERS related signaling in pro-inflammatory macrophages, providing mechanistic insight into the chronic inflammatory microenvironment of diabetic wounds. Guided by this finding, we constructed a dynamic injectable hydrogel (GPOK-2@EVs) incorporating Artemisia argyi-derived EVs for localized and sustained delivery. The hydrogel exhibited favorable self-healing capability, injectability, tissue adhesion, and pH/ROS-responsive degradation, making it well suited for the diabetic wound microenvironment. In vitro, GPOK-2@EVs effectively promoted macrophage reprogramming toward an anti-inflammatory phenotype, alleviated ERS and oxidative stress, and enhanced endothelial angiogenic activity. In vivo, it significantly accelerated diabetic wound closure, improved granulation tissue formation and collagen deposition, increased vascularization, and reduced inflammatory and ERS responses. Overall, this study establishes a promising plant EV-based hydrogel platform for diabetic wound therapy and highlights the potential of stress-responsive immunomodulatory biomaterials in chronic wound repair.

Author contributions

Dazhuang Miao: Conceptualization, Methodology, Investigation, Data curation, Writing – original draft. Xiaopeng Suo: Methodology, Investigation, Data curation, Validation, Writing – original draft. Xinqian Geng: Investigation, Data curation, Formal analysis, Validation, Writing – original draft. Ying Yang: Investigation, Data curation, Formal analysis, Validation, Writing – original draft. Gaoming Wang: Methodology, Formal analysis, Validation. Yang Zhou: Investigation, Data curation, Validation. Yunhe Jia: Supervision, Writing – review & editing, Project administration. Shuofei Yang: Supervision, Writing – review & editing, Resources. Heping Li: Supervision, Writing – review & editing, Funding acquisition. Gaohua Li: Conceptualization, Supervision, Project administration, Resources, Funding acquisition, Writing – review & editing.

Funding

Research Fund of Beijing Medical AwardFoundation, YXJL-2020-1223-0320, Yunnan Province Pilot Project on Clinical Collaboration of Integrated Chinese and Western Medicine for Major and Difficult Diseases, 202402100327, Basic Training Special Program of Peking University People’s Hospital, RDJP2025-12,Medical Innovative Technology Fund, CSJSYS-2023-09, National Natural Science Foundation of China, 82370497

Data availability

Data will be made available on request.

Declarations

Competing interest

The authors declare that they have no competing financial interests or personal relationships that may have influenced the work reported in this paper. The authors declare no competing interests.