One dose of PDGFB circular RNA may boost growth factor levels to speed up healing of diabetic wounds

The schematic diagram illustrates the using of LNP-circPDGFB to accelerate diabetic wound healing (Fig. 1A). The circular RNA encoding PDGFB (Supplementary Table S1) was synthesized through a group I intron-based autocatalytic cyclization strategy, as outlined in Supplementary Figure S1A. Capillary electrophoresis was employed to analyze the molecular weights of both linear and circular PDGFB RNA transcripts (Supplementary Figure S1B, C). Encapsulation was performed using microfluidic technology. Briefly, the ionizable lipid SM-102, cholesterol, DSPC, and DMG-PEG2000 dissolved in ethanol were mixed with mRNA dissolved in citric acid buffer, and the resultant nanoparticles encapsulating linRNA and circRNA were referred to as LNP-linPDGFB and LNP-circPDGFB, respectively (Fig. 1A).

LNP-linPDGFB and LNP-circPDGFB were characterised by dynamic light scattering (DLS), as shown in Fig. 1B-E and Supplementary Figure S1D. The results revealed that LNP-linPDGFB had a mean average hydrodynamic diameter of 100.9 nm with a narrow polydispersity index of 0.161 and a zeta potential of −0.6873 mV (Fig. 1B, C and Supplementary Figure S1D). LNP-circPDGFB had a mean average hydrodynamic diameter of 110.6 nm with a narrow polydispersity index of 0.065 and a zeta potential of −0.1335 mV (Fig. 1D, E and Supplementary Figure S1D). Encapsulation rate of LNP-linPDGFB and LNP-circPDGFB were evaluated by Quant-it RiboGreen RNA Assay and were 94.97% and 94.77%, respectively (Supplementary Figure S1D).

To validate the stability and prolonged expression of LNP-encapsulated circRNA, we first evaluated the sustained protein expression capacity of circRNA using GFP as a reporter gene. Following transfection of LNP- circGFP and LNP-linGFP (Supplementary Figure S2A) into 293 T cells, fluorescence imaging and quantitative analysis revealed significantly prolonged GFP expression mediated by circRNA (Fig. 2A). While fluorescence intensities were comparable between groups on day 1, the circRNA group demonstrated substantially higher signal than linear mRNA by day 3. Notably, strong circRNA-mediated fluorescence persisted through day 9, with detectable signal remaining at day 11 (Fig. 2B). To systematically compare the temporal expression dynamics of linear versus circular mRNA, we transfected HEK293T cells with constructs encoding either GFP or PDGFB in both linear and circular forms. Western blot analysis revealed low levels of protein expression from both linear and circular GFP mRNA at 24 h post-transfection. Notably, linear GFP expression peaked at 48 h and began to decline by 72 h, whereas circular GFP exhibited sustained upregulation from 24 to 72 h, resulting in significantly higher protein levels at later time points compared to the linear form (Fig. 2C). In the PDGFB expression system, both linear and circular mRNA reached their highest expression at 24 h, followed by a gradual decrease. However, the circular PDGFB showed higher initial expression than its linear counterpart. Although both declined over time, circular PDGFB maintained relatively high expression until 48 h, in contrast to the linear form, which decreased to very low levels. By 96 h, circular PDGFB remained detectable, whereas linear PDGFB expression was nearly undetectable as early as 72 h (Fig. 2D). Based on these temporal profiles, we selected 72 h as the optimal pretreatment duration for subsequent in vitro experiments. To quantitatively assess fluorescence signal dynamics, we transfected HEK293T cells with equal amounts of LNP-linFluc and LNP-circFluc (Supplementary Figure S2B) and periodically measured luciferase activity. Consistent with previous observations, both mRNA forms showed low luciferase activity on day 1. Circular Fluc activity continued to increase steadily, surpassing linear Fluc levels significantly by day 5 (Fig. 2E).

We next evaluated potential immunogenicity by measuring three proinflammatory cytokines (IFN-α, IL-6 and IL-1β) following transfection. Strikingly, compared to negative controls, neither linear nor circular RNA showed substantial immunogenicity (Fig. 2F-H). Moreover, circular RNA demonstrated comparable or even lower expression for some inflammatory cytokines relative to controls (Fig. 2H). In summary, our comprehensive analyses demonstrate that circRNA provides prolonged, stable expression with minimal immunogenicity compared to linear mRNA, while maintaining sustained high-level protein production over extended periods.

Angiogenesis is a key factor in the healing of diabetic wounds [1, 3]. To evaluate the therapeutic potential of LNP-circPDGFB in diabetic wound healing, we systematically examined its effects on key endothelial cell functions. Human umbilical vein endothelial cells (HUVECs) were subjected to a 72-h pretreatment with five experimental conditions: LNP-linFluc, LNP-circFluc, PDGF-BB protein, LNP-linPDGFB and LNP-circPDGFB. Functional assessments were conducted using EdU incorporation for proliferation, Transwell assays for migration, and Matrigel-based tube formation for angiogenic capacity. EdU staining revealed robust proliferative activation in groups treated with PDGF-BB protein, LNP-linPDGFB and LNP-circPDGFB, with LNP-circPDGFB exhibiting the most potent effect (Fig. 3A-B). Transwell migration assays demonstrated significantly enhanced cell motility in both LNP-linPDGFB and LNP-circPDGFB groups compared to controls, with the circular mRNA formulation yielding superior migratory outcomes (Fig. 3C-D). Angiogenesis experiment results and quantitative analyses of tube formation further confirmed the strong pro-angiogenic phenotype induced by LNP-circPDGFB, as evidenced by significant increases in total tube length and branch point number relative to all other groups (Fig. 3E-G). Our collective results indicate that LNP-circPDGFB effectively promotes endothelial cell proliferation, migration, and angiogenesis, underscoring its promising therapeutic utility in facilitating vascularization during diabetic wound repair.

Fibroblast proliferation, migration, and collagen synthesis play critical roles in wound healing [21, 22]. We next evaluated PDGFB mRNA impact on mouse embryonic fibroblast cell line NIH3T3. EdU and CCK-8 proliferation assays revealed that compared to LNP-linFluc and LNP-circFluc control groups, treatments with PDGF-BB protein, LNP-linPDGFB and LNP-circPDGFB for 72 h all significantly enhanced NIH3T3 cell proliferation, with LNP-circPDGFB demonstrating the most pronounced pro-proliferative effects (Fig. 4A-C). Transwell migration assays further confirmed that LNP-circPDGFB treatment significantly increased cellular migration rates compared to controls (Fig. 4D-E). These results align with our HUVEC findings, demonstrating that LNP-circPDGFB not only enhances vascular endothelial cell functions but also significantly promotes fibroblast proliferation and migration. By simultaneously augmenting these two critical repair cell populations, our study presents a novel therapeutic strategy for improving diabetic wound healing.

Having established the potent in vitro effects of LNP-circPDGFB on both endothelial cells and fibroblasts - the two pivotal cell types involved in wound repair - we next sought to evaluate its in situ therapeutic performance in a physiologically relevant context. To dynamically monitor protein expression kinetics in vivo, we encapsulated firefly luciferase-encoding circular RNA (Fluc-circRNA) or linear mRNA (Fluc-linRNA) into optimized lipid nanoparticles (LNPs) and applied them topically to full-thickness dorsal wounds in C57 mice. Longitudinal bioluminescent imaging (Fig. 5A-B) revealed that while linear mRNA showed higher initial expression but rapid decline by day 3, circRNA-mediated expression remained robust throughout the first week and persisted at detectable levels until day 11, demonstrating superior pharmacokinetics ideal for wound healing applications.

We next evaluated the in situ therapeutic performance of LNP-circPDGFB in a diabetic wound healing model using extended comparative controls. In addition to LNP-linFluc and LNP-circFluc, diabetic mice were treated under identical dosing conditions with clinically relevant and mechanistically distinct comparators, including LNP-linVEGF, LNP-circVEGF, recombinant human PDGF-BB protein, Regranex^®, LNP-linPDGFB, and LNP-circPDGFB. Macroscopic wound closure analysis demonstrated that a single topical administration of LNP-circPDGFB resulted in the most rapid and sustained wound healing throughout the observation period (Fig. 5C–E). Although LNP-circVEGF significantly promoted wound closure relative to control groups, its overall efficacy was inferior to that of LNP-circPDGFB, while Regranex^® showed a moderate therapeutic effect that was less pronounced than either circular RNA–based LNP formulation. In contrast, linear RNA formulations, reporter RNA controls, and untreated wounds exhibited comparatively limited or delayed healing responses. Quantitative analysis of wound area trajectories further confirmed a clear efficacy hierarchy, with LNP-circPDGFB achieving the fastest wound closure, followed by LNP-circVEGF and Regranex^®, whereas linear RNA formulations and control groups showed minimal therapeutic benefit (Fig. 5D–E). Collectively, these results indicate that LNP-circPDGFB enables superior and durable in situ wound repair compared with both clinically approved growth factor therapy and other LNP-based RNA formulations.

To further evaluate tissue regeneration and angiogenesis following different therapeutic interventions, wound tissues from mice treated with PDGF-BB protein, Regranex^®, LNP-linPDGFB, LNP-circPDGFB, LNP-linVEGF, LNP-circVEGF, and control formulations were subjected to systematic histological and immunofluorescence analyses (Fig. 6A). Histological examination at day 15 post-injury revealed marked differences in tissue repair outcomes among the treatment groups (Fig. 6B). Among these, wounds treated with LNP-circPDGFB exhibited the most pronounced regenerative features, including a thicker neo-epidermal layer and more abundant collagen deposition (Fig. 6B–D). In comparison, PDGF-BB protein, Regranex^®, LNP-linPDGFB, and LNP-circVEGF also promoted tissue regeneration to varying extents, as evidenced by reduced wound width, increased epidermal thickness, and enhanced collagen deposition. Notably, across these parameters, LNP-circVEGF showed overall better effects than its linear RNA counterpart and Regranex^®, but remained inferior to the LNP-circPDGFB treatment.

Quantitative analysis of the Collagen I and CD31-positive area proportion revealed significant differences among the treatment groups in promoting wound Collagen I expression and angiogenesis (Fig. 6E-G). Compared with the control, LNP-linFluc, and LNP-circFluc groups, all treatments associated with functional growth factors significantly increased the proportion of CD31-positive micro vessels. Among them, the LNP-linVEGF and Regranex^® groups exhibited a moderate pro-angiogenic effect, whereas the LNP-circVEGF group further significantly increased the CD31-positive proportion, with a pro-angiogenic capacity comparable to that of the LNP-circPDGFB group. Consistent with enhanced tissue remodeling, α-SMA staining revealed increased accumulation of myofibroblasts in PDGFB-related treatment groups, with the most prominent effect observed in the LNP-circPDGFB group, suggesting enhanced wound contraction and extracellular matrix remodeling (Fig. 6E, H). Notably, these comprehensive therapeutic benefits were achieved without eliciting detectable immune responses or toxicity (Supplementary Fig. 3). Overall, these histological and immunofluorescence analyses indicate a clear efficacy hierarchy in tissue regeneration and remodeling, with LNP-circPDGFB producing the most pronounced improvements, followed by LNP-circVEGF, while linear RNA formulations exhibited comparatively weaker effects. The observed effects can be attributed to the unique advantages of our platform: the nuclease-resistant circular structure enables sustained PDGFB expression, while LNP delivery maintains local bioavailability. This dual approach overcomes the transient growth factor availability that limits current treatments, simultaneously addressing angiogenesis, fibroblast activation, and ECM reorganization - the key pathophysiological deficits in diabetic wounds.

To further understand the mechanisms of therapeutic effects of LNP-circPDGFB on diabetic wound, we conducted single-cell analysis on the wound tissue samples treated by LNP-circPDGFB or LNP-circFluc, respectively. UMAP clustering revealed 18 distinct clusters, representing 14 unique cellular types (Fig. 7A-B and Supplementary Figure S4A-C). We analyzed the proportions of these cell types and observed increased proportions for keratinocytes, fibroblasts and endothelial cells in PDGFB group (Fig. 7C). We next explored the number of DEGs between PDGFB and FLUC cell types. The results showed that fibroblasts have a predominance of upregulated DEGs, suggesting enhanced cellular activity (Fig. 7D).

Chronic hyperglycemia has been shown to impair fibroblast migration, alter mechanotransduction, and disrupt the balance between matrix deposition and remodeling, leading to immature granulation tissue and structurally compromised repair. The upregulation of genes involved in collagen synthesis, matrix assembly, and cytoskeletal dynamics observed in our scRNA-seq data suggests that sustained PDGFB expression may help overcome these intrinsic cellular deficits, promoting a reparative fibroblast phenotype capable of coordinated migration and matrix remodeling. We next compared the differences between PDGFB and FLUC sample of fibroblasts and endothelial cells. We identified upregulated DEGs directly associated with PDGF signaling pathway (Fos, Junb, Egr1), genes involved in extracellular matrix (ECM) synthesis and remodeling (Col1a2, Col5a2, Col7a1, Loxl2, Bgn, Mmp4, Serpinh1 Timp2), and genes related to cell migration or cytoskeleton reorganization (Vim, Tpm1, Tpm2, Cfl2, Cd44, Enah, Rock1). The expression of these genes was significantly upregulated in fibroblasts from PDGFB samples (Fig. 7E, left panel). GO and GSEA analysis of the upregulated genes in fibroblasts highlighted significant enrichment in pathways related to PDGF signaling, PI3K–Akt signaling, angiogenesis, growth factor signaling, and ECM organization in PDGFB samples (Fig. 7E, right panel, Supplementary Figure S5). The enrichment in PDGF-related signaling pathways suggests activation of mechanisms involved in fibroblast proliferation and migration. Notably, PI3K–Akt signaling is a key downstream pathway of PDGF and is frequently dysregulated in diabetic fibroblasts, contributing to impaired migration, survival, and matrix remodeling [23, 24]. Therefore, the observed enrichment of ECM-related and cytoskeleton-associated pathways may reflect a partial restoration of fibroblast reparative functions that are typically compromised under diabetic conditions. The enrichment in angiogenesis-related pathways further implies a potential role of these fibroblasts in supporting vascular development, which is often insufficient in chronic diabetic wounds. Collectively, these findings suggest that LNP-circPDGFB may promote fibroblasts toward a more reparative phenotype under diabetic wound conditions.

Analysis of endothelial cells revealed upregulation of genes associated with key biological functions (Fig. 7F, left panel). Notably, genes such as Itgav and Fgfr2 were enriched, implicating activation of the PI3K-Akt pathway. Additionally, MAPK pathway genes, including Dusp1, Dusp6, Junb, and Fos, also showed increased expression. Angiogenesis related genes, such as Egfl7, Angptl4, Col15a1, and Serpine1, were also upregulated, indicating potential promotion of blood vessel formation. These findings suggest that the identified genes in endothelial cells may play a significant role in disease-related processes. Enrichment analysis of the upregulated genes in endothelial cells revealed significant associations with pathways related to PI3K–Akt and MAPK signaling, cellular migration, vascular biology, and ECM dynamics (Fig. 7F, right panel). These pathways are closely linked to endothelial survival, activation, and angiogenic responses. Under diabetic conditions, PI3K–Akt signaling is frequently impaired, contributing to endothelial dysfunction and non-productive angiogenesis despite the presence of pro-angiogenic stimuli [1, 3, 4]. In this context, the enrichment of PI3K–Akt- and MAPK-associated pathways may indicate reactivation of endothelial programs required for productive angiogenesis rather than dysfunctional vascular responses. The involvement of vascular biology and ECM organization pathways further suggests improved coordination between endothelial cells and the surrounding matrix, which is critical for stable vessel formation during diabetic wound repair.

Citrate buffer (Cat: C1013-100 ml) was purchased from Solarbio (Beijing, China). Streptozotocin (STZ) (Cat: HY-13753-100 mg) was purchased from MedChemExpress (New Jersey, USA). Ionizable lipid (SM-102), Cholesterol, 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC), and 1,2-dimyristoyl-rac-glycero-3-methoxypolyethylene glycol-2000 (DMG-PEG2000) were purchased from Genscript (Guangzhou, China). Ready-to-use tribromoethanol anesthetic (Cat: M2920) was purchased from Nanjing Aibei Biotechnology Co.Ltd. (Nanjing, China). The Dual-Luciferase Reporter Assay Kit (Cat: DL101-01) was purchased from Vazyme Biotech Co.Ltd. (Nanjing, China). The EdU Detection Kit (Cat: C10310-1/2/3) was purchased from RiboBio Co.Ltd. (Guangzhou, China). Anti-fade mounting medium with DAPI (Cat: P0131-25 ml) was purchased from Beyotime (Shanghai, China). Matrigel (Cat:356234) was purchased from Corning Inc. (New York, USA). Transwell plates (Cat:3422) were purchased from Corning Inc. (New York, USA). HiScript IV ALL-in-One Ultra RT SuperMix (Cat: R433-01) for qPCR was purchased from Vazyme Biotech Co.Ltd. (Nanjing, China).

Linear and circular mRNA vectors were constructed by Genscript (Guangzhou, China). The genes of firefly luciferase (Fluc), green fluorescent protein (GFP) and PDGFB (Gene ID: 18591) were taken as inserts. For construction of linear mRNA vectors, mRNA was produced in vitro by T7 High Yield RNA Synthesis Kit on a linearized DNA template, which encoding 5′-UTR sequence, Pdgfb open reading frame, 3′-UTR sequence, and poly-A tail sequence. UTP was replaced with 1-methylpseudouridine (m1Ψ) to produce mRNA containing modified nucleosides. The initial linear mRNA was capped and purified by the Genscript Capping System. Using linear DNA as a template, the Fluc-mRNA and GFP-mRNA were prepared in the same method. All mRNA was analyzed by agarose gel and cryopreserved at −20 °C.

For construction of circular mRNA vectors, the empty circRNA backbone with 5′ homology arm sequence, 3′ intron sequence, linker-1 sequence, IRES sequence, linker-2 sequence, 5′ intron sequence and 3′ homology arm sequence were designed in the authors’ laboratory and synthesized by Genscript (Guangzhou, China). The coding sequences (CDS) of PDGFB, Fluc and GFP were PCR amplified and inserted into the empty circRNA backbone via Gibson assembly, generating circRNA expression plasmids for in vitro transcription (IVT). PDGFB circRNA was synthesized by Genscript. Using T7 High Yield RNA Synthesis Kit, PDGFB circRNA precursor was synthesized via IVT from linearized circRNA plasmid template. RNA product was mixed with appropriate DNase I for 30 min to digest the initial DNA template and the product was column purified with Monarch RNA Cleanup Kit. Subsequently, the reaction mixture was added with GTP at a concentration of 2 mM, and then the mixture was incubated at 55 °C for 15 min to catalyze the cyclization of circRNA. To enrich circular RNA products, the sample was digested with RNase R at 37 °C for 15–30 min, followed by additional column purification. The same procedure was applied to synthesize Fluc and GFP circRNAs.

Lipid nanoparticles encapsulating either linear or circular RNA were prepared using a microfluidic-based method. In brief, an RNA aqueous solution at pH 4.0 is quickly combined with a lipid mixture dissolved in ethanol. The lipid mixture consists of an ionizable cationic lipid, DSPC, DMG-PEG2000 and cholesterol. The ratios for the lipid mixture are as follows: SM-102: DSPC: cholesterol: DMG-PEG2000 = 50:10:38.5:1.5. The resulting LNP suspension was then dialyzed against PBS and stored at −80 °C at a final concentration of 0.1 mg/mL for further use. The encapsulation efficiency of linear and circular mRNA was determined using the Quant-iT RiboGreen RNA Assay Kit (Invitrogen, R11490↗). Specifically, TE buffer and 2% TE-Triton buffer, which could break the lipid layer of the lipid nanoparticles, were used respectively to treat LNP samples in order to determine the amount of dissociating circRNA and total circRNA, and the fluorescence intensity was measured by microplate reader after adding Ribogreen reagent. The encapsulation efficiency was determined by the following formula: Encapsulation efficiency (%) = [(total circRNA-free circRNA)/total circRNA] × 100%. Particle size, polydispersity index (PDI) and Zeta potential of LNP-circRNA or LNP-linRNA was measured by dynamic light scattering (DLS) (Malvern, Zetasizer Nano ZS, Herefordshire, UK).

Twenty-four hours prior to transfection, HEK293T cells were passaged into 12-well plates to achieve 30% confluency. The cells were transfected with either LNP-linGFP or LNP-circGFP (0.5 µg/well; stock concentration: 0.1 µg/µL). Fresh complete medium (DMEM supplemented with 10% fetal bovine serum) was added before transfection. Three replicate wells were set up for each group, with an untreated group serving as the blank control. After 6 h, the medium was replaced. GFP fluorescence was monitored every other day (on days 1, 3, 5, 7, 9, 11, 13, 15 and 17) using an inverted fluorescence microscope to assess mRNA expression and LNP delivery efficiency. The medium was refreshed daily to maintain cell viability.

Western blot was performed as previously described [32]. 293 T cells transfected with LNP-GFP circRNA, LNP-GFP linRNA, LNP-linPDGFB or LNP-circPDGFB were collected and lysed with 1× loading buffer. Proteins were separated by SDS-PAGE (90 V for stacking gel, 120 V for separating gel) and transferred to PVDF membranes (350 mA, 70 min). Membranes were blocked with skim milk for 1 h, incubated with primary(GFP 1:1000; Cell SignalingTechnology 2956 S、PDGFB 1:50;Santa Cruz sc-36580、β-tubulin (1:5000; BeijingRay Antibody Biotech, RM2003) and HRP-conjugated secondary antibodies, and visualized using the Bio-Rad ChemiDoc Imaging System after ECL substrate application.

293 T cells (1 × 10⁴/well) were cultured in 24-well plates until reaching 30% confluence. Cells were treated with 2 µL of LNP-Fluc circRNA or LNP-Fluc linRNA for 24 h. After removing the medium, cells were washed twice with PBS and lysed with 1× Cell Lysis Buffer for 5 min at room temperature. Lysates were centrifuged at 12,000 × g for 2 min, and supernatants were collected. For detection, 20 µL of lysate was mixed with 100 µL Luciferase Substrate (Dual Luciferase Reporter Assay Kit; DL101Vazyme), and Firefly luciferase activity was immediately measured using a microplate reader. Relative fluorescence intensity was assessed at 72 h and 120 h using the same method.

Real-time quantitative PCR was performed as previously described [33]. Total RNA was extracted using TRIzol reagent (Vazyme Biotech, R411-01). cDNA was synthesized from 1 µg RNA using the HiScript IV All-in-One Ultra RT SuperMix for qPCR (Vazyme Biotech, R433-01) according to the manufacturer’s protocol. Gene-specific primer pairs (designed using NCBI tools) are listed in Supplementary Table S2. Quantitative real-time PCR was performed using SYBR Green Master (Vazyme Biotech, Q711-03) according to the manufacturer’s protocol. All reactions were performed in technical triplicates, with β-actin/GAPDH as the internal control.

For proliferation assessment, HUVEC and NIH3T3 cells were seeded in 96-well plates (1,000 cells/well) and cultured until 30% confluence. The five treatment groups were incubated with cells for 72 h, with untreated cells serving as controls. Cell viability was measured using Cell Counting Kit-8 (CCK-8, BeyoTime C0037) according to the manufacturer’s protocol. Briefly, 10 µL of CCK-8 reagent (diluted 1:1000 in DMEM) was added to each well, followed by 2-h incubation at 37 °C. Absorbance at 450 nm was measured using a microplate reader. For EDU, twenty-four hours prior to transfection, both cell types were seeded into 48-well plates (2,000 cells/well). The same five treatment groups as before were applied: LNP-linFluc, LNP-circFluc, PDGF-BB protein, LNP-linPDGFB, and LNP-circPDGFB. After 72 h of pre-treatment, cell proliferation was assessed according to the EdU assay kit instructions (RiboBio, C10310-1/2/3).

Cell migration capacity was evaluated using Transwell chambers (Corning 3422, USA). NIH3T3 and HUVEC cells were seeded into 24-well plates 24 h before transfection to achieve 30% confluency. After 24 h, the cells were transfected with LNP-linFluc, LNP-circFluc, PDGF-BB protein, LNP-linPDGFB or LNP-circPDGFB (7 µg/well). Following 72 h of pretreatment, the cells were trypsinized, counted, and resuspended in 100 µL of serum-free DMEM medium. The cell suspension was then transferred to the upper chamber of a Transwell insert placed in a 24-well plate at a density of 60,000 cells/well. The lower chamber contained 700 µL of DMEM medium supplemented with 20% FBS. After incubation for 24 h (37 °C, 5% CO₂), the upper chamber was removed, washed three times with PBS, fixed with 4% paraformaldehyde for 2 h and stained overnight with 0.1% crystal violet (Solarbio, C8470). After staining, non-migrated cells on the upper surface were gently removed with a cotton swab. The migrated cells were then observed under a microscope and images were captured.

HUVEC cells were seeded into 24-well plates for pretreatment under the same conditions and groups as used in the Transwell migration assay. First, 30 µL of Matrigel (Corning, 356243) was added to each well of a 24-well plate and evenly spread to cover the bottom, followed by incubation at 37 °C in a CO₂ incubator for 60 min to allow gel polymerization. After 72 h of pretreatment with the five mRNA groups, the cells were trypsinized, counted and seeded onto the polymerized Matrigel at a density of 2 × 10⁴ cells per well. The cells were then incubated for 12 h (37 °C, 5% CO₂), and tube formation was documented under a microscope.

Male C57BL/6 mice (20–25 g) were anesthetized with isoflurane and received 9-mm full-thickness dorsal wounds created using sterile biopsy punches. The animals were divided into two treatment groups receiving single topical applications of: LNP-linFluc and LNP-circFluc (10 µg mRNA in 20 µL solution per wound, the mRNA dose, administration route, and key time points were primarily determined based on previously published studies that systematically evaluated the expression kinetics, duration of biological activity, and safety of LNP-mediated mRNA or delivery in diabetic wound models [26, 34]. After 60-min immobilization to ensure formulation absorption, bioluminescence imaging was performed following intraperitoneal injection of D-luciferin (15 mg/ml, MB1834, MeilunBio) using an AniView100 (PHOTON) with standardized acquisition parameters which mice were maintained under isoflurane anesthesia. The imaging protocol enabled quantitative comparison of temporal expression patterns between linear and circular RNA constructs, with photon flux (photons/sec/cm²/sr) quantified using Living Image software. Experimental rigor was maintained through: uniform wound dimensions (9 ± 0.2 mm) and exclusion of specimens with dosing irregularities. Three independent replicates were performed per timepoint, with system calibration conducted before each imaging session to ensure measurement consistency. This longitudinal approach provided direct visualization of the enhanced pharmacokinetics conferred by circRNA’s nuclease resistance in the wound microenvironment.

All animal studies were approved by the Institutional Animal Care and Use Committee of South of China Agricultural University. Mice were housed in specific pathogen-free conditions at 20–26℃, with 30–70% humidity. Animals were monitored regularly throughout the study and that predefined humane endpoints were applied to minimize suffering, in accordance with institutional animal care guidelines. For the in vivo experiments, mice were age-matched and randomly assigned to experimental groups, using at least n = 3 mice per genotype or treatment. C57BL/6J (WT) mice were purchased from Guangzhou Yancheng Biological Technology Co. Ltd (Guangzhou, China). Diabetes was induced in C57BL/6 male mice (20–25 g) via 2 weeks of high-fat/high-sugar diet followed by intraperitoneal injection of streptozotocin (STZ, HY-13753-50 mg MedChemExperss) dissolved in citrate buffer (PH = 4.5 C1013-100 ml Solarbio). Mice with fasting blood glucose > 11.1 mmol/L (measured via tail vein blood using Sinocare glucometer) 1-week post-STZ were considered diabetic. Diabetic mice were randomly divided into 9 groups (n = 3 per group): LNP-linFluc, LNP-circFluc, PDGF-BB protein, Regranex, LNP-linVEGF, LNP-circVEGF, LNP-linPDGFB, LNP-circPDGFB treatment groups, and an untreated control group. All procedures were performed under anesthesia induced by tribromoethanol (M2920, Nanjing Aibei Biotechnology). After removing dorsal hair, full-thickness skin wounds (6 mm in diameter) were created using a sterile biopsy punch, and sew a ring with an inner circle of 9 mm along the edge of the wound and stick it on. The nine treatment groups received topical application of their respective solutions, while the control group remained untreated. Anesthesia was maintained for 60 min to facilitate absorption. Wound healing progress was recorded every other day post-treatment using a ruler for size calibration. Quantitative analysis was performed using ImageJ software. On day 15 after surgery, the mice were euthanized, and wound specimens were collected for paraffin embedding and histological examination.

On day 15, wound tissues (1 × 1 cm squares) were fixed in 4% paraformaldehyde for 24 h, paraffin-embedded, and sectioned (5 μm thickness) using hematoxylin and eosin (H&E) staining, Masson’s trichrome staining (performed according to the kit instructions), and immunofluorescence (IF) staining to assess tissue formation, collagen deposition, capillary density, immune cell infiltration, and inflammatory cytokine expression. For immunofluorescence staining, antigen retrieval was performed on tissue sections using antigen retrieval buffer under microwave treatment. The sections were then blocked with ready-to-use goat serum at 37 °C for 30 min. The primary antibodies used were: α-SMA (ab124964, 1:4000; Abcam), CD31 (AF3628, 1:2000; R&D Systems; stock concentration 0.2 mg/ml), Collagen I (ab270993, 1:2000; Abcam), CD3 (100206, 1:2000; BioLegend), CD20 (PA516701↗, 1:600; Thermo), CD68 (ab955, 1:50; Abcam), TNFα (26405-1-AP, 1:5000; Proteintech), IL-6 (A0286, 1:5000; ABclonal), and IL-17 (506904, 1:2000; BioLegend). The primary antibodies were incubated at 4 °C overnight. The corresponding secondary antibodies were: Alexa Fluor 647 donkey anti-goat IgG (ab150131, 1:667; Abcam), Alexa Fluor 488 goat anti-mouse IgG (ab150113, 1:667; Abcam), Alexa Fluor 488 goat anti-rabbit IgG (ab150077, 1:667; Abcam), and Alexa Fluor 555 donkey anti-mouse IgG (ab150110, 1:667; Abcam). Finally, the sections were mounted with an anti-fade mounting medium containing DAPI (P0131-25 ml; Beyotime) and imaged using a confocal microscope. Quantitative analysis of α-SMA, CD31, Collagen I, IL-6, IL-17, and TNF-α expression was performed using ImageJ software.

The sequencing reads were assessed by quality metrics, and raw reads were mapped to the mouse reference genome (mm10) to generate gene expression matrices via CellRanger pipeline [35] (10× Genomics, v8.0.0). The gene expression matrix was processed using Seurat package [36] (version 4.3.0). Low-quality cells were filtered out if they had fewer than 500 UMI counts, fewer than 200 detected genes, more than 6000 detected genes, or mitochondrial read percentages exceeding 5% [37, 38]. Furthermore, to avoid the effects of doublets, potential doublets were predicted using the DoubletFinder package [39] (v2.0.3) and excluded.

Gene expression was normalized, scaled by a factor of 10,000, and then log-transformed. The top 2000 highly variable genes (HVGs) were detected with the FindVariableFeatures function and used for subsequent analysis. Principal component analysis (PCA) was performed on those HVGs with the PCA function. Batch effects were corrected using the canonical correlation analysis (CCA) method in Seurat package, based on the top 30 PCA components. For visualization, unsupervised Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction and cell clustering were performed using the top 30 principal components via the RunUMAP, FindNeighbors and FindClusters functions.

To identify differentially expressed genes (DEGs) between groups, we used the FindMarkers function with test.use = “wilcox”. Genes with |log2FoldChange| > 0.25 and P value < 0.05 were considered DEGs. Gene Ontology (GO) [40] and Gene Set Enrichment Analysis (GSEA) [41] pathway enrichment analyses were performed using the clusterProfiler package [42] (v4.0.5) with an adjusted P-value threshold of 0.05. P-values were adjusted using the Bonferroni method.

Statistical analyses were performed using GraphPad Prism 8.0 software (GraphPad Software, La Jolla, CA, USA). All quantitative data are presented as mean ± standard error of the mean (SEM) from at least three independent experiments. Prior to parametric statistical analyses, data distributions were assessed for normality to confirm that the assumptions underlying these tests were met. Comparisons among multiple groups were conducted using one-way analysis of variance (ANOVA), while comparisons between two groups were performed using Student’s t-test. A P value < 0.05 was considered statistically significant.