GPC3 and CD133-targeted peptide dual modification enhances the therapeutic effect of doxorubicin carried by OMVs on hepatocellular carcinoma

Abstract

Hepatocellular carcinoma (HCC) has a high mortality rate because of the limitations of conventional chemotherapies (e.g., doxorubicin, DOX), including poor tumor targeting, systemic toxicity, and chemoresistance. Bacterial outer membrane vesicles (OMVs) are promising drug carriers but have limitations like nonspecific phagocytosis and weak targeting. In this study, a dual-targeted nanosystem ((GPC3 + CD133)^T-OMVs@DOX) was developed using GPC3/CD133 peptide-modified OMVs loaded with DOX to enhance HCC-specific targeting, optimize drug release, and reduce off-target toxicity. Compared with single-targeted or unmodified OMVs, dual-targeted (GPC3 + CD133)^T-OMVs significantly increased binding/internalization in GPC3/CD133-coexpressing Huh-7 cells (p < 0.05). DOX loading yielded high LE (∼81%), with a pH-responsive release of 40% at pH 7.4 vs. 80% at pH 5.0 over 48 h. Sparstolonin B (SsnB), a TLR2/4 inhibitor, reduced macrophage phagocytosis by ∼60%, prolonging the duration of DOX-loaded OMVs in circulation. In vivo, the dual-targeted system inhibited tumor growth by 75%, outperforming free DOX (40%) and single-targeted OMVs-DOX (55%). Mice treated with the OMVs maintained a stable weight, whereas those given free DOX showed ∼15% weight loss. Histology revealed minimal organ damage in the dual-targeted group vs. severe cardiomyocyte injury with free DOX. In conclusion, the dual-targeted OMV system enhances tumor specificity via cooperative GPC3/CD133 recognition, optimizes pH-responsive release, and reduces nonspecific clearance by modulating macrophages. These features improve antitumor efficacy and mitigate toxicity, positioning OMV-based nanocarriers as promising platforms for precision HCC therapy.

Graphical abstract

1. Introduction

Hepatocellular carcinoma (HCC) ranks among the sixth most lethal malignancies globally (Llovet et al., 2021; Vogel et al., 2022). Given that the early symptoms of liver cancer are often subtle, the majority of patients are diagnosed at the middle to late stages, at which point they have missed the opportunity for surgical resection (Brown et al., 2023; Llovet et al., 2021; Vogel et al., 2022). Traditional radiotherapy and chemotherapy have limited efficacy. Although targeted therapy and immunotherapy have achieved certain advancements, challenges such as drug resistance and low response rates persist (Brown et al., 2023; Hwang et al., 2025; Sacks et al., 2018). Therefore, the development of novel, highly effective, and low-toxicity treatment strategies is urgently needed.

Nanotechnology-driven drug delivery systems exhibit diverse functional advantages, including enhancing drug solubility/stability, regulating controlled release, prolonging the duration of systemic circulation, traversing biological tissue barriers, improving drug accumulation in the target tissue, and facilitating cellular internalization and intracellular transport (Wang et al., 2024a; Mitchell et al., 2021; Nie et al., 2020). Bacterial outer membrane vesicles (OMVs) are bilayer spherical lipid structures with a diameter of 10–250 nm produced by the outer membrane blistering of gram-negative bacteria (Toyofuku et al., 2019; Toyofuku et al., 2023). OMVs contain peribacterial proteins, while the outer membrane is composed mainly of phospholipids and lipopolysaccharides (LPS). OMVs can participate in a series of biological processes, such as interbacterial communication, nutrient acquisition, and horizontal transmission of antibiotic resistance genes. On the other hand, OMVs can cross the mucosal barrier into the blood circulation and mediate bacteria–host interactions, including immunomodulation, promotion of pathogen adhesion, and invasion of host cells (Cheng et al., 2025; Xia et al., 2025). OMVs are emerging as “new stars” in the field of nanodrug carriers, with several studies utilizing them as drug carriers to treat diseases (Liang et al., 2021).

Compared to eukaryotic cell-derived vesicles, OMVs offer potential advantages such as easier large-scale production and more straightforward engineered modification (Ho et al., 2024; Liu et al., 2022a). Moreover, intrinsic adjuvant properties can modulate the tumor immune microenvironment (Mehanny et al., 2020; Feng et al., 2022). Li et al. reported that OMVs can promote the polarization of macrophages from the M2 subtype to the M1 subtype and induce pyroptosis to enhance antitumor immunity. The photosensitizer Chlorin e6 (Ce6) and the chemotherapy drug doxorubicin (DOX) were encapsulated in OMVs and combined with photodynamic therapy, chemotherapy, and immunotherapy to eradicate triple-negative breast cancer tumors in mice (Li et al., 2023b).

However, the clinical translation of OMV-based therapeutics is hindered by several recognized challenges. Foremost, their intrinsic immunogenicity, largely attributed to components such as LPS, poses risks of systemic toxicity and inflammatory responses, thereby constraining the therapeutic window (Chen et al., 2025a). Furthermore, scalable, reproducible, and cost-effective production remains elusive; OMV yield, size, and composition are often heterogeneous, varying with bacterial strain and culture conditions (Torres-Vanegas et al., 2024). Perhaps most critically for drug delivery applications, native OMVs lack inherent tumor-targeting specificity, necessitating the development of engineering strategies to confer this functionality.

Several previous studies have attempted to modify OMVs to enhance the targeting of bacterial vesicles(Dong et al., 2024; Cui et al., 2023). Drug delivery systems (DDSs) modified with a single targeting moiety are prone to limitations such as suboptimal therapeutic efficiency and the development of acquired resistance. This shortcoming is primarily attributed to tumor heterogeneity—where cancer cells often exhibit heterogeneous expression of target antigens—or dynamic downregulation of the targeted receptor on tumor cell surfaces, which reduces DDS binding and internalization (Wang et al., 2024b). Additionally, prolonged exposure to single-targeted therapy may trigger compensatory signaling pathways or activate alternative survival mechanisms in residual cancer cells, further diminishing treatment efficacy and promoting resistance phenotypes. In this study, we developed a dual-targeted nanodelivery system using glypican-3 (GPC3) and CD133 peptide-modified OMVs loaded with DOX (GPC3 + CD133)ᵀ-OMVs@DOX to increase tumor specificity. Moreover, sparstolonin B (SsnB), a selective inhibitor of TLR 2 and TLR 4, was used to inhibit the nonspecific phagocytosis of OMVs by macrophages. As a result, (GPC3 + CD133)ᵀ-OMVs@DOX exhibited good targeting ability and antitumor properties.

2. Material and methods

2.1. Cell lines and culture conditions

Huh-7 cells, a highly differentiated hepatocellular carcinoma cell line, and RAW264.7 cells, a mouse macrophage line, were purchased from Pricella (Wuhan, China) and cultured with Dulbecco's modified Eagle's medium (DMEM; Servicebio, China) supplemented with 10% fetal bovine serum (FBS; Bioexplorer, USA) and 1% penicillin–streptomycin (Solarbio, China) under standard culture conditions (37 °C, 5% CO₂). Cell lines were identified using STR analysis and tested negative for mycoplasma contamination.

2.2. Construction of engineered bacteria

The hypervesiculating Escherichia coli Nissle 1917 strain (ΔE), which was established by deleting the nlpl gene, was purchased from BioSci (Hangzhou, China). The plasmid pET21a-OmpA-Spy Catcher-HA Tag was designed and synthesized by Gene Create (Wuhan, China), and the sequence is shown in Table S1. A total of 1 μL of purified plasmid was added to 100 μL of competent cells, mixed gently, and then stored on ice for 20 min. Next, the mixture was transferred to an electroporation cuvette, and electroporation was performed at 2.5 kV, 25 μF, and 200–400 Ω. Afterward, the cells were mixed with 1 mL of prewarmed Luria–Bertani (LB) medium (37 °C) and recovered by shaking at 200 rpm for 1 h at 37 °C. Finally, the transformants were selected on LB agar plates supplemented with ampicillin (AMP).

2.3. Isolation of SpC-OMVs

Engineered bacteria were scraped from the screening plate and cultured in LB medium supplemented with AMP (100 μg/mL) by shaking at 200 rpm and 37 °C until the OD 600 reached 0.8. The supernatants were harvested by centrifugation (5000 ×g, 15 min, 4 °C) and filtered through a 0.45 μm polyvinylidene fluoride filter to remove bacterial debris. The clarified supernatant was concentrated using a 100 kDa tangential flow filtration (TFF) membrane cartridge (Sartorius, Germany), followed by centrifugation (12,000 ×g, 30 min, 4 °C) and sterile filtration (0.22 μm, Millipore, USA). Finally, Spy Catcher-displaying OMVs (SpC-OMVs) were isolated via ultracentrifugation at 150,000 ×g at 4 °C for 3 h twice. The SpC-OMVs were quantified by measuring the total protein content using a bicinchoninic acid (BCA) assay and stored at −80 °C.

2.4. Preparation of (GPC3 + CD133)^T-OMVs

CD133/GPC3-targeting peptides (the sequences are shown in Table S2) were designed and synthesized by Wuhan TianDe Biotechnology Co., Ltd. GPC3-targeting peptide-Spy Tag003 (GPC3^T-SpT) and CD133-targeting peptide-Spy Tag003 (CD133^T-SpT) were added to SpC-OMVs suspensions and coincubated at 4 °C for 12 h on a rotary shaker. Unbound targeted peptides were removed by ultrafiltration.

2.5. SpC-OMVs and (GPC3 + CD133)^T-OMVs characterization

The morphology of OMVs was observed by transmission electron microscopy (TEM; Hitachi, Japan). The particle size distribution and Zeta potential were analyzed using dynamic light scattering (Zetasizer Nano ZSP, Malvern, UK). OmpA-SpC-HA tag fusion proteins displayed on SpC-OMVs were characterized using western blotting analysis.

2.6. Endotoxin quantification

Endotoxin levels in purified OMVs preparations were measured using a Limulus Amebocyte Lysate (LAL) chromogenic endpoint assay kit (Beyotime, China) according to the manufacturer's instructions. Briefly, OMVs were appropriately diluted with endotoxin-free water and mixed with LAL reagent in a 96-well plate. After incubation at 37 °C for 10 min, the chromogenic substrate was added, and the reaction was stopped with stop solution. The absorbance at 405 nm was measured using an EnSight™ Multimode Microplate Reader (PerkinElmer, USA). A standard curve was generated using E. coli O111:B4 endotoxin standard, and the endotoxin activity in samples was calculated and expressed as endotoxin units (EU) per microgram of OMV protein.

2.7. Western Blotting

To perform the protein analysis, total protein was extracted from OMVs or cells using RIPA buffer (Beyotime, China) supplemented with a protease inhibitor cocktail. Following extraction, the protein concentration was determined by a BCA assay, after which the samples were mixed with loading buffer (Beyotime, China) and stored at −80 °C. The proteins were subsequently separated by 5–10% SDS–PAGE and then electrophoretically transferred to PVDF membranes. After blocking with 5% skim milk for 1 h, the membranes were incubated first with primary antibodies and subsequently with HRP-conjugated secondary antibodies for 2 h each. Finally, the protein bands were visualized using a chemiluminescence gel imaging system (Bio-Rad, USA). To ensure equal loading, GAPDH was used as an internal control. The following antibodies were used: anti-HA tag (Bioswamp, 1:5000), anti-TLR2 (Huabio, 1:1000), anti-MyD88 (Abcam, 1:1000), anti-NF-κB (Abcam, 1:1000), anti-p-NF-κB (Abcam, 1:1000), anti-MAPK (Abcam, 1:1000), anti-p-MAPK (Abcam, 1:1000), and anti-GAPDH (Huabio, 1:5000).

2.8. OMVs internalization

OMVs were stained with PKH-67 (D0031, Solarbio, China) and Huh-7 cells (20,000 cells/well) or RAW264.7 cells (100,000 cells/well) were coincubated with 30 μg of labeled OMVs for 6 h, fixed with 4% polyformaldehyde (Servicebio, China), and stained with DAPI. Images were acquired by fluorescence microscopy (LSM 780; Zeiss, Germany) and analyzed with ImageJ software (National Institute of Health, USA).

2.9. Peptide competition assay

To validate the specificity of receptor-mediated uptake, a competition assay was performed. Huh-7 cells were pre-incubated with a 50-fold molar excess of unlabeled GPC3 and CD133 targeting peptides for 1 h at 37 °C. Subsequently, PKH67-labeled (GPC3 + CD133)ᵀ-OMVs were added directly into the medium and cells were further incubated for 6 h. After fixed with 4% polyformaldehyde, nuclei were stained with DAPI. Images were acquired using fluorescence microscopy (LSM 780; Zeiss, Germany) and analyzed with ImageJ software (National Institute of Health, USA).

2.10. Drug loading and release kinetics

DOX (S17092, Shanghai Yuanye, China) was loaded into OMVs using a sonication method. Briefly, different amounts of DOX were added to 1 mL of OMVs suspension (containing 500 μg of OMVs protein), and the mixture was sonicated (20% amplitude, 10 s on/5 s off for 60 cycles) followed by incubation at 37 °C for 1 h to restore membrane integrity. Free DOX was removed by ultracentrifugation (150,000 ×g, 3 h, 4 °C). The amount of encapsulated DOX was quantified by measuring fluorescence intensity (Ex/Em: 478/596 nm) using an EnSight™ Multimode Microplate Reader (PerkinElmer, USA).To correct for potential background fluorescence from the vesicle components, blank OMVs were measured in parallel, and their signal was subtracted from that of the OMVs@DOX samples. Loading efficiency (%) = (Mass of DOX loaded in OMVs / Total mass of DOX used in the loading process) × 100%. Drug content (μg/μg OMVs) = Mass of DOX loaded in OMVs / Mass of the drug-loaded OMVs. The mass of OMVs is based on the total protein content as determined by BCA assay.

To perform the DOX release studies, OMVs@DOX suspension (250 μL) was loaded into a dialysis bag (MWCO 14 kDa, Bkmamlab, China) and immersed in 20 mL of release medium (PBS, pH 5.0 or 7.4). The system was incubated at 37 °C under gentle shaking (100 rpm). The released DOX content was measured and calculated fluorometrically (Ex/Em: 478/596 nm) against a DOX standard curve. Drug release (%) = (DOX content in PBS/total DOX content in OMVs) × 100%.

2.11. Cytotoxicity assay

The cytotoxicity of DOX, GPC3^T-OMVs@DOX, CD133^T-OMVs@DOX, and (GPC3 + CD133)^T-OMVs@DOX to Huh-7 cells was evaluated using a cell counting kit (CCK-8) assay. In brief, Huh-7 cells were seeded into 96-well plates (3000 cells/well) and exposed to different concentrations of DOX or other drugs (containing the same amount of DOX) for 24 h. Next, the culture medium was replaced with CCK-8 reagent-containing medium, and the cells were incubated for 2 h, after which the absorbance at 450 nm was measured using a microplate reader (Perkin Elmer, UK).

2.12. Flow cytometry

2.12.1. Analysis of peptide conjugation efficiency

SpC-OMVs were incubated with varying concentrations of GPC3^T-SpT (labeled with Rhodamine B) and CD133^T-SpT (labeled with FITC) peptides to allow conjugation. The mixture was then incubated with anti-HA-tag magnetic beads at 4 °C for 12 h with rotation to specifically capture the SpC-OMVs. After incubation, the beads were collected using a magnetic stand and washed three times with PBS. The bead-bound complexes were finally resuspended in a fixed volume (500 μL) of PBS for immediate analysis by flow cytometry using an Accuri C6 flow cytometer (BD Biosciences, San Jose, CA). The median fluorescence intensity (MFI) of the OMVs population was used as the primary quantitative metric to assess peptide conjugation efficiency, as it correlates with the average peptide density on the vesicle surface. Data were analyzed using FlowJo v10 software (FlowJo LLC, Ashland, OR). Appropriate controls (e.g., bare beads, single-stained samples) were included for accurate gating and compensation.

2.12.2. Induction and validation of M1 macrophage polarization

To establish an M1-polarized macrophage model for subsequent experiments. RAW264.7 cells were seeded in 6-well plates (5 × 10⁵ cells/well) and stimulated with different concentration of LPS (L2630, Sigma-Aldrich, USA) for 24 h. Following LPS stimulation, the RAW264.7 cells were harvested and subjected to fixation and permeabilization according to the manufacturer's protocol. Cells were stained with fluorescently labeled with APC-conjugated anti-CD86. Flow cytometric analysis was performed on an Accuri C6 flow cytometer (BD Biosciences, San Jose, CA), and the data were analyzed using FlowJo v10 software (FlowJo LLC, Ashland, OR).

2.13. Animal experiments

All the animal studies were approved by the Ethics Committee of Wuhan University. Male nude mice (4–6 weeks old) were purchased from Hunan SJA (Changsha, China) and raised under special pathogen-free (SPF) animal conditions. A total of 5 × 10⁶ Huh-7 cells were resuspended in 200 μl of PBS and injected subcutaneously into the left side of each mouse. The mice were kept until the tumors were visible (approximately 2 weeks). The tumor-bearing mice were then randomly allocated into 6 groups (n = 5): (1) SsnB + PBS; (2) SsnB + free DOX; (3) SsnB + blank OMVs; (4) SsnB + GPC3^T-OMVs@DOX; (5) SsnB + CD133^T-OMVs@DOX; and (6) SsnB + (GPC3 + CD133)^T-OMVs@DOX. SsnB (HY-116213, MedChemExpress, USA) was dissolved in DMSO and diluted with corn oil. The mice received intraperitoneal injections of SsnB 30 min before intravenous treatment (5 mg/kg DOX equivalent). The treatments were repeated every 3 days for 5 cycles. The tumor size and mouse body weight were monitored every 2 days (tumor volume = length × width²/2). At the experimental endpoint, the mice were euthanized for tissue collection for subsequent experiments. All subsequent analyses (tumor weighing, histopathology, immunohistochemistry) were performed by investigators blinded to the group assignments.The tumor inhibition rate (TIR) for a treated group was calculated as: [1 − (Tumor Weight ^treatment / Tumor Weight ^control)] × 100%, where the control refers to the tumor-bearing mice treated with SsnB and PBS.

2.14. In vivo imaging

To detect the metabolism of OMVs in vivo, different types of OMVs were labeled with DiR (UR21017, Umibio, China) or Cy7 (HY-D0825, MedChemExpress, USA). Nude mice (n = 3 each group) that carried tumors were pretreated with SsnB (24 mg/kg, intraperitoneal injection) or DMSO/oil (200 μl, intraperitoneal injection), labeled OMVs were subsequently administered via tail vein injection, and biodistribution was monitored using a NightOWL II imaging system (Berthold, Germany) at the designated time points. Fluorescence intensity was quantified as the average radiant intensity (p/s/cm²) over the region of interest using IndiGO™ software.

2.15. Hematoxylin and eosin (H&E) staining

The mouse tissues (liver, kidney, spleen, and heart) were processed for standard histopathological analysis. Briefly, the samples were fixed in 4% paraformaldehyde for 24–48 h, dehydrated in ethanol, embedded in paraffin, and sectioned at 4 μm thickness. The sections were then stained with hematoxylin and eosin (H&E). Finally, the slides were examined under a microscope (Olympus BX-40, Japan).

2.16. Biochemical analysis

The plasma was isolated (3000 rpm, 15 min) and stored at −20 °C. ALT/AST (liver function), and urea/creatinine (renal injury) levels were measured using a Siemens biochemical analyzer.

2.17. Immunohistochemical (IHC) staining

Paraffin-embedded tumor sections (4 μm) underwent antigen retrieval in Tris-EDTA buffer (pH 9.0). After blocking endogenous peroxidase and nonspecific sites, sections were incubated with primary antibodies against Ki-67 (1:500, Bioswamp), E-cadherin (1:200, Bioswamp), and cleaved caspase-3 (1:200, Bioswamp) at 4 °C overnight. Detection was performed using an HRP-polymer system (Bioswamp) with DAB visualization. Quantitative analysis (positive cell percentage for Ki-67, caspase-3; average optical density for E-cadherin) was performed with ImageJ software.

2.18. Statistical analysis

All the quantitative data are expressed as the mean ± SD. GraphPad Prism 7 (GraphPad Software) was used for both graphical representation and statistical analysis. All datasets satisfied the normality assumption (Shapiro-Wilk test, p > 0.05). Subsequently, intergroup differences were assessed by an unpaired Student's t-test (for two groups) or by one-way/two-way ANOVA (for multiple groups). Statistical significance is denoted using a dual-marker system for clarity: the hash symbol (#) indicates comparisons against the control group (or baseline group as defined in each experiment), while the asterisk (*) denotes comparisons between different treatment groups (^{###/⁎⁎⁎} p < 0.001, ^##/⁎⁎ p < 0.01, ^#/⁎ p < 0.05).

3. Results

3.1. Engineering and characterization of SpyCatcher-OMVs

We developed a high-yield SpyCatcher-OMVs (SpC-OMVs) system by expressing a OmpA-Spy Catcher-HA tag fusion protein (OmpA serves as a membrane-anchoring domain) in the hypervesiculating E. coli Nissle 1917 mutant (ΔE). A schematic of the OMVs extraction process is shown in Fig. 1A. Under TEM, the resulting OMVs and SpC-OMVs exhibited typical cup-shaped nanostructures with diameters less than 200 nm, showing no significant differences (Fig. 1B). Western blotting analysis using an anti-HA antibody confirmed successful expression and integration of the fusion protein exclusively in SpC-OMVs (Fig. 1C). DLS measurements revealed similar hydrodynamic diameters for both OMVs and SpC-OMVs (66.48 ± 0.27 nm vs. 67.25 ± 0.35 nm), demonstrating that genetic modification did not significantly affect vesicle size or distribution (Fig. 1D). Additionally, compared with the wild-type strain (116 ± 23.07 μg/500 mL supernatant), the ΔE mutant produced approximately five times more OMVs (509.3 ± 41 μg/500 mL supernatant), providing a robust foundation for scalable OMV-based drug delivery (Fig. S1).To ensure the reproducibility of our delivery platform, key characteristics—including vesicle size distribution, protein content, and endotoxin levels—were assessed across three independent batches of SpC-OMVs. The results demonstrated good batch-to-batch consistency (Table. S3), confirming the robustness of our production and isolation protocol.

Fig. 1.

Preparation and characterization of SpyCatcher-displayed OMVs (SpC-OMVs).

(A) Schematic diagram of the extraction process for OMVs.

(B) Representative TEM images of OMVs (left) and SpC-OMVs (right). Scale bar: 200 nm.

(C) Western blotting analysis of OmpA-SpC-HA Tag fusion protein expression in bacterial membrane and purified OMVs.

(D) Hydrodynamic diameter distribution of OMVs and SpC-OMVs measured by DLS.

3.2. Synthesis and in vitro targeting validation of dual-targeted OMVs

To increase the specificity of our delivery platform for hepatocellular carcinoma (HCC), we functionalized SpC-OMVs with two distinct targeting peptides. GPC3-targeting peptide-Spy Tag003 (GPC3^T-SpT) and CD133-targeting peptide-Spy Tag003 (CD133^T-SpT) were designed to bind GPC3 and CD133, respectively, both of which are highly expressed on HCC cells. These peptides were site-specifically and covalently conjugated to SpC-OMVs via spontaneous isopeptide bond formation, yielding dual-targeted (GPC3 + CD133)ᵀ-OMVs (Fig. 2A). To determine the optimal conjugation efficiency, we incubated SpC-OMVs (500 μg/mL) with varying concentrations of the peptides (GPC3^T-SpT conjugated to Rhodamine B and CD133^T-SpT conjugated to FITC). Following enrichment of OMVs using anti-HA magnetic beads and removal of unbound peptides, quantification by median fluorescence intensity (MFI) revealed a concentration-dependent increase in signal, which reached a plateau at a concentration of 100 μg of each peptide per mL of OMVs suspension (Fig. 2B). Increasing the peptide concentration beyond 100 μg/mL did not yield a statistically significant increase in MFI. This ratio was subsequently adopted for all further experiments. DLS analysis confirmed that the conjugation process did not alter the physicochemical properties of the OMVs, as (GPC3 + CD133)ᵀ-OMVs maintained a monodisperse distribution with a hydrodynamic diameter of 70.97 ± 0.29 nm (Fig. 2C).

Fig. 2.

Functionalization and targeting validation of dual-targeted OMVs.

(A) Schematic illustration of the conjugation between SpyCatcher-OMVs (SpC-OMVs) and SpyTag-fused targeting peptides (GPC3T-SpT and CD133T-SpT).

(B) Flow cytometry analysis of peptide conjugation efficiency. SpC-OMVs were incubated with increasing concentrations of GPC3T-SpT (labeled with Rhodamine B) and CD133T-SpT (labeled with FITC).

(C) Hydrodynamic diameter of dual-targeted (GPC3 + CD133)ᵀ-OMVs after peptide conjugation, as determined by DLS.

(D) Cellular uptake of PKH67-labeled (green) OMVs by Huh-7 cells after 6 h incubation. Dual-targeted (GPC3 + CD133)ᵀ-OMVs show enhanced targeting and uptake compared to non-targeted or single-targeted OMVs. Scale bar: 100 μm.

(E) Peptide competition assay validating the receptor-specificity of uptake. Confocal microscopy shows that pre-incubation with excess free targeting peptides significantly reduces the internalization of fluorescent (GPC3 + CD133)ᵀ-OMVs by Huh-7 cells. Scale bar: 100 μm.

Data are presented as mean ± SD (n = 3). Statistical significance is denoted as follows: *p < 0.05, **p < 0.01, ***p < 0.001. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

We next evaluated the targeting efficacy of the engineered vesicles in Huh-7 cells. Incubation with PKH67-labeled OMVs demonstrated that (GPC3 + CD133)ᵀ-OMVs achieved markedly greater cellular uptake than single-targeted OMVs (GPC3ᵀ-OMVs or CD133ᵀ-OMVs) or non-targeted SpC-OMVs, as quantified by fluorescence microscopy (Fig. 2D). The two single-targeted versions showed comparable but lower targeting efficiency.To quantify whether this enhancement represented a mere additive effect or a cooperative gain, we compared the observed uptake of the dual-targeted formulation with the arithmetic sum of the two single-targeted versions. After subtracting non-specific background, the mean fluorescence intensity of (GPC3 + CD133)ᵀ-OMVs significantly exceeded the additive sum (p < 0.01), indicating a cooperative advantage.

Finally, to confirm that the enhanced uptake was specifically mediated by peptide-receptor interaction, we performed a competition assay. Pre-incubation of Huh-7 cells with an excess of free GPC3 and CD133 targeting peptides prior to the addition of PKH67-labeled (GPC3 + CD133)ᵀ-OMVs resulted in a marked reduction of perinuclear fluorescent signal compared to the untreated control, as observed by confocal microscopy (Fig. 2E). Quantitative analysis of the fluorescence intensity per cell corroborated this observation: pre-incubation with the free peptides significantly inhibited OMVs uptake (p < 0.001). Notably, the fluorescence intensity in the competition group was reduced to a level that showed no statistically significant difference from that of the non-targeted SpC-OMVs.

Together, these results demonstrate that the dual-targeting strategy not only enhances cellular association but does so through specific ligand-receptor interactions, thereby significantly improving the binding specificity of OMVs to HCC cells.

3.3. Preparation and in vitro antitumor efficacy of doxorubicin-loaded dual-targeted OMVs

Doxorubicin (DOX) was efficiently encapsulated into (GPC3 + CD133)ᵀ-OMVs using an ultrasonic coincubation method (Fig. 3A). The drug content and loading efficiency were evaluated at various initial DOX concentrations. As shown in Fig. 3B, the loading efficiency significantly increased when the DOX concentration increased from 200 μg/mL to 400 μg/mL, reaching 81.65 ± 1.30%. A further increase in concentration only marginally improved the loading efficiency. Therefore, 400 μg/mL was selected as the optimal concentration, yielding a final drug loading concentration of 1.10 ± 0.01 μg of DOX per μg of OMVs. The release behavior of DOX from the vesicles was assessed under different pH conditions (Fig. 3C). While initial burst release was observed in both neutral (pH 7.4, mimicking physiological conditions) and acidic (pH 5.0, mimicking the lysosomal environment) buffers, markedly enhanced and sustained release (∼80% at 48 h) occurred at acidic pH compared with that at neutral pH (∼40%). After drug loading, the diameter of the OMVs increased to 78.20 ± 2.84 nm, indicating successful drug encapsulation (Fig. 3D). Drug uptake and antitumor effects were subsequently evaluated in Huh-7 cells. Fluorescence microscopy revealed stronger intracellular red fluorescence (DOX) in cells treated with (GPC3 + CD133)ᵀ-OMVs@DOX than in those treated with an equivalent dose of free DOX, demonstrating enhanced cellular internalization mediated by the dual-targeted vesicles (Fig. 3E). Consistent with these findings, the results of the CCK-8 assays revealed that (GPC3 + CD133)ᵀ-OMVs@DOX exhibited significantly enhanced cytotoxicity, with an IC₅₀ of 5.88 ± 2.01 μM, which was considerably lower than that of free DOX (18.62 ± 1.38 μM) (Fig. 3F). Moreover, at the same DOX concentration (18.62 μM), the dual-targeted formulation induced superior tumor cell killing compared with that induced by either single-targeted OMVs (GPC3ᵀ-OMVs@DOX or CD133ᵀ-OMVs@DOX), while blank OMVs showed no apparent cytotoxicity (Fig. 3G). In summary, the encapsulation of DOX into (GPC3 + CD133)ᵀ-OMVs promotes drug internalization and significantly enhances its antitumor efficacy in vitro, resulting in it outperforming both free drug and single-targeted formulations.

Fig. 3.

Doxorubicin loading, release profile, and in vitro antitumor efficacy of dual-targeted OMVs.

(A) Schematic illustration of the preparation of (GPC3 + CD133)ᵀ-OMVs@DOX.

(B) Loading efficiency (LE) and drug content (DC) of (GPC3 + CD133)ᵀ-OMVs@DOX at different initial DOX concentrations.

(C) In vitro DOX release profiles from (GPC3 + CD133)ᵀ-OMVs@DOX in PBS at pH 7.4 and pH 5.0.

(D) Hydrodynamic diameter of (GPC3 + CD133)ᵀ-OMVs after DOX loading, as measured by DLS.

(E) Cellular uptake of free DOX (red) and PKH67-labeled (green) (GPC3 + CD133)ᵀ-OMVs@DOX by Huh-7 cells after 6 h incubation. Scale bar: 100 μm.

(F) Cell viability of Huh-7 cells treated with various concentrations of free DOX or (GPC3 + CD133)ᵀ-OMVs@DOX for 24 h, as determined by CCK-8 assay.

(G) Cell viability of Huh-7 cells treated with different formulations at a concentration of 18.62 μM (IC50 of free DOX) for 24 h.

Data are presented as mean ± SD (n = 6). Statistical significance is denoted as follows: *p < 0.05, **p < 0.01, ***p < 0.001. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.4. Sparstolonin B prolongs the duration of OMVs in circulation by suppressing macrophage phagocytosis

To improve the in vivo stability and delivery efficiency of OMVs, we investigated whether sparstolonin B (SsnB), a specific Toll-like receptor 2 (TLR2) antagonist, could inhibit macrophage-mediated clearance of OMVs. We first established an M1-polarized macrophage model by stimulating RAW264.7 cells with LPS. From a tested range (500 ng/mL to 3 μg/mL), 3 μg/mL was selected as the optimal concentration that effectively induced polarization while maintaining cell viability. Cells were treated with 3 μg/mL LPS for 24 h to establish the M1 phenotype. Successful polarization was confirmed by the significant upregulation of the M1 marker CD86 via flow cytometry (Fig. S2), assessment of the M2 marker CD206 confirmed no induction under these conditions, validating the specificity of the M1 phenotype (data not shown). These activated macrophages were then pretreated with different concentrations of SsnB before they were incubated with PKH67-labeled OMVs. As shown in Fig. 4A, pretreatment SsnB significantly reduced the uptake of OMVs, as indicated by the decrease in green fluorescence intensity. No significant difference was observed between the NC group and the DMSO group, ruling out solvent-related effects. Furthermore, SsnB pretreatment markedly suppressed the production of proinflammatory cytokines, including IL-6 and TNF-α (Fig. 4B), and reduced the phosphorylation of key proteins in the MAPK and NF-κB pathways (Fig. 4C), indicating the inhibition of TLR2/MyD88 downstream signaling. Finally, to evaluate the effect of SsnB on OMVs pharmacokinetics in vivo, mice were pretreated with SsnB followed by intravenous injection of DiR-labeled OMVs. In vivo imaging revealed that SsnB pretreatment significantly reduced OMV accumulation in the liver at all time points examined (Fig. 4E), suggesting that SsnB effectively inhibited systemic macrophage phagocytosis, thereby prolonging the circulation time of OMVs. Taken together, these results demonstrate that SsnB increases the circulation time and delivery efficiency of OMVs by inhibiting TLR2/MyD88/MAPK/NF-κB-mediated macrophage phagocytosis.

Fig. 4.

SsnB inhibits macrophage phagocytosis of OMVs by antagonizing the TLR2 signaling pathway in vitro.

(A) Phagocytosis of PKH67-labeled (green) OMVs by LPS-activated RAW264.7 cells pretreated with different concentrations of SsnB or vehicle control (DMSO).

(B) ELISA analysis of IL-6 and TNF-α levels in cell culture supernatants after treatment.

(C) Western blotting analysis of key proteins in the TLR2/MyD88/MAPK/NF-κB pathway in RAW264.7 cells following SsnB treatment. # # #p < 0.001, # # p < 0.01, and # p < 0.05 vs. control group.

(D) In vivo fluorescence imaging showing biodistribution of DiR-labeled OMVs post-injection in mice pretreated with SsnB or oil/DMSO.

Data are presented as mean ± SD (n = 3). Statistical significance is denoted as follows: #, compared to the control group; *, compared between treatment groups (#/* p < 0.05, ##/** p < 0.01, ###/*** p < 0.001,). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.5. In vivo antitumor efficacy and biodistribution of dual-targeted OMVs@DOX

The in vivo antitumor performance of the engineered OMVs was systematically evaluated in a murine hepatoma model. Tumor-bearing mice were randomly assigned to different treatment groups, as schematically illustrated in Fig. 5A. Importantly, all treatment regimens described below were well tolerated throughout the study period, with no signs of acute systemic toxicity observed. A detailed and quantitative analysis of biosafety is provided in Section 3.6. In vivo fluorescence imaging revealed markedly reduced accumulation of the pretreatment with SsnB combined with the dual-targeted formulation (GPC3 + CD133)ᵀ-OMVs@DOX in nontargeted organs compared with that in other groups (Fig. 5B). This observation was further confirmed by ex vivo imaging of dissected organs, which revealed particularly increased in tumor uptake and diminished in liver (Fig. 5C), indicating enhanced tumor-selective targeting and reduced macrophage phagocytosis. Consistent with this enhanced biodistribution, the SsnB + (GPC3 + CD133)ᵀ-OMVs@DOX group demonstrated the most potent suppression of tumor growth. Furthermore, the antitumor activity of the single-targeted OMVs (GPC3ᵀ-OMVs@DOX or CD133ᵀ-OMVs@DOX) was superior to that of free DOX or the PBS control, although their efficacy was lower than that of the dual-targeted formulation (Fig. 5D, E). The final tumor weights further corroborated the enhanced inhibitory effect of the combination treatment (Fig. 5F). Immunohistochemical analysis of tumor tissues revealed a significant reduction in Ki-67-positive cells in the (GPC3 + CD133)ᵀ-OMVs@DOX group (Fig. 5G), indicating suppressed tumor cell proliferation. Additional IHC data demonstrating enhanced E-cadherin expression and caspase-3 activation are presented in Fig. S3. In conclusion, these in vivo results demonstrate that (GPC3 + CD133) ᵀ-OMVs@DOX, especially when combined with macrophage phagocytosis inhibition via SsnB, achieves significantly enhanced antitumor efficacy by improving tumor-specific accumulation and drug delivery, outperforming both free drug and single-targeted OMVs therapies.

Fig. 5.

In vivo antitumor efficacy and biodistribution of SsnB-pretreated and dual-targeted OMVs@DOX in a murine hepatoma model.

(A) Schematic illustration of the experimental timeline and treatment groups (n = 5). Mice received SsnB pretreatment followed by intravenous injection of: 1) PBS; 2) Blank OMVs; 3) Free DOX; 4) GPC3ᵀ-OMVs@DOX; 5) CD133ᵀ-OMVs@DOX; 6) (GPC3 + CD133)ᵀ-OMVs@DOX.

(B) In vivo fluorescence imaging showing the biodistribution of Cy7.7 -labeled OMVs in tumor-bearing mice post-injection.

(C) Ex vivo fluorescence imaging of dissected major organs and tumors at 24 h post-injection. # # #p < 0.001, # # p < 0.01, and # p < 0.05 vs. Oil/DMSO+SpC-OMVs group.

(D) Tumor volume growth curves of mice during the treatment period.

(E) Body weight curves of mice during the treatment period.

(F) Representative photographs and weights of excised tumors at the endpoint of the study.

(G) Immunohistochemical staining of Ki-67 in tumor tissues, showing proliferation index across treatment groups. Scale bar: 50 μm.

Data are presented as mean ± SD (n = 5).Statistical significance is denoted as follows: #, compared to the control group (Oil/DMSO + SpC-OMVs); *, compared between treatment groups (#/* p < 0.05, ##/** p < 0.01, ###/*** p < 0.001,).

3.6. Evaluation of OMVs toxicity

To fully evaluate the therapeutic potential of our OMV platform by complementing the antitumor efficacy assessment, we performed a comprehensive analysis of its systemic safety profile. Hepatic and renal functions were further evaluated using serum biomarkers (Fig. 6A, B). Assessment of hepatic function revealed that serum ALT levels were elevated in the free DOX group, indicating potential drug-induced liver injury, whereas all OMV-treated groups maintained ALT levels comparable to those in the control group (Fig. 6A). Reflecting the well-documented cardiotoxicity of DOX, the free DOX group also exhibited a significant increase in AST, a marker indicative of cardiac and hepatic damage. In contrast, the AST level in the dual-targeted (GPC3 + CD133)ᵀ-OMVs@DOX group showed no significant difference from the control group (Fig. 6A). Renal function markers, including creatinine and urea, remained within normal ranges across all groups, confirming the absence of nephrotoxicity (Fig. 6B).Histological analysis of major organs (liver, spleen, kidney, and heart) via hematoxylin and eosin (H&E) staining provided definitive evidence of safety (Fig. 6C). OMVs- and OMVs@DOX-treated mice displayed an intact tissue architecture with no signs of inflammation, necrosis, or apoptosis in any organ. In stark contrast, myocardial sections from the free DOX-treated group exhibited characteristic signs of cardiotoxicity, including reduced cardiomyocyte volume, nuclear shrinkage, and interstitial edema, suggesting substantial myocardial damage. These findings corroborate the biochemical and physiological data, confirming that OMVs mitigate the off-target toxicity of encapsulated drugs.

Fig. 6.

Biosafety evaluation of SsnB-pretreated and dual-targeted OMVs@DOX in tumor-bearing mice.

(A) Serum levels of ALT and AST as markers of liver function.

(B) Serum levels of creatinine (Cr) and Urea as markers of kidney function.

(C) Representative hematoxylin and eosin (H&E)-stained sections of major organs (liver, spleen, kidneys, and heart) collected at the end of the treatment period. Scale bars: 100 μm. The H-E staining of the organs from other groups could be seen in Fig. S4. Arrows indicate atrophied myocardial cells.

Data are presented as mean ± SD (n = 5). Statistical significance is denoted as follows: *p < 0.05, **p < 0.01, ***p < 0.001.

Collectively, these results demonstrate that OMVs-based formulations exhibit excellent in vivo biocompatibility. OMVs elicit minimal systemic toxicity, preserve normal hematological and organ function, and effectively reduce the cardiotoxicity associated with DOX when used as a delivery platform.

4. Discussion

Chemotherapy continues to serve as a critical non-first-line therapeutic option for hepatocellular carcinoma (Li et al., 2023a; Chen et al., 2024). Chemotherapy drugs, such as doxorubicin (DOX), often have poor tumor targeting and severe systemic toxicity, and acquired drug resistance can occur; these limitations significantly compromise their clinical efficacy (Lilienberg et al., 2017; Liu et al., 2022b). Targeted drug delivery systems, such as biological membrane vesicle carriers, have emerged as promising strategies for enhancing efficacy while reducing off-target toxicity (Wang et al., 2024a; Mitchell et al., 2021). Glypican-3 (GPC3) is a heparan sulfate proteoglycans aberrantly overexpressed in >70% of HCCs but nearly absent in normal hepatocytes. It promotes tumor progression via Wnt/β-catenin and Hedgehog signaling, and its expression is correlated with poor prognosis (Tehrani et al., 2025). CD133, a pentaspan transmembrane glycoprotein, is expressed on cancer stem cells (CSCs) and is considered responsible for the self-renewal ability, chemoresistance, and tumorigenicity of HCC (Wang et al., 2025; Song et al., 2019). Many studies, including those on targeted drug delivery, specific diagnosis, and CAR-T-cell therapy, have utilized GPC3 as a target in liver cancer (Makkouk et al., 2021; Deng et al., 2022). In contrast, few studies on CAR-T-cell therapy for cancer have used CD133 as a target (You et al., 2025; Zhai et al., 2025). In this study, we developed a dual-targeted nanodelivery system based on bacterial outer membrane vesicles (OMVs) modified with GPC3- and CD133-targeting peptides to enhance therapeutic efficacy. Ideally, targeting GPC3 could potently engage and eliminate these bulk tumor populations. Concurrently, targeting CD133 would specifically address the refractory characteristics and stemness of tumors. The cooperative interplay of these two targeting strategies is anticipated to improve antitumor activity.

OMVs have emerged as a promising class of biomimetic nanocarriers because of their unique structural and biological properties (Wang et al., 2024a; Mitchell et al., 2021; Toyofuku et al., 2019; Toyofuku et al., 2023; Cheng et al., 2025; Xia et al., 2025). In our study, we first validated the successful isolation and characterization of OMVs (Fig. 1). Transmission electron microscopy (TEM) confirmed their cup-shaped morphology, dynamic light scattering (DLS) revealed a uniform size distribution (∼70 nm), and western blotting verified the expression of the OmpA-SpC-HA tag fusion protein, confirming its purity and identity. The production process proved reproducible, with consistent vesicle size, protein yield, and low endotoxin levels across batches (Table. S3). This consistency in critical quality attributes supports the reliability of the platform. Furthermore, the low endotoxin profile associated with the probiotic ECN strain is a favorable feature that may help mitigate a common safety concern for bacterially derived therapeutic vesicles.These results align with previous reports that OMVs can be efficiently isolated and retain their native structural integrity, supporting their suitability as drug delivery platforms (Cheng et al., 2021).

As previously stated, GPC3 and CD133 are well-characterized biomarkers for HCC. By comodifying OMVs with GPC3 and CD133-targeting peptides (Fig. 2A), we aimed to achieve dual-specific recognition of both bulk tumor cells and CSCs, which are often resistant to conventional chemotherapy. Flow cytometry and confocal microscopy analyses (Fig. 2B & D) confirmed that compared with single-targeted (GPC3ᵀ-OMVs or CD133ᵀ-OMVs) or unmodified OMVs, dual-targeted OMVs ((GPC3 + CD133)ᵀ-OMVs) exhibited significantly higher binding and internalization rates in Huh-7 cells. This enhanced uptake was receptor-specific, as it was effectively blocked by pre-incubation with free targeting peptides (Fig. 2E). Quantitative analysis indicated that the dual-targeting effect surpassed a merely additive outcome, suggesting a cooperative interaction between the two targeting pathways. While this comparison at an optimized concentration supports a cooperative mechanism, a more rigorous analysis—such as employing the Bliss independence model across a gradient of ligand densities—would provide deeper insight into the nature of this interaction in future studies (Wang et al., 2024b).

Additionally, the lipid bilayer structure of OMVs facilitates efficient encapsulation of both hydrophilic (e.g., DOX) and hydrophobic drugs via passive diffusion or membrane engineering (Chai et al., 2025; Al-Dossary et al., 2022). In this work, we leveraged these properties to load DOX into OMVs, achieving a high drug loading efficiency (LE) of ∼81% (Fig. 3B), which is significantly greater than that of conventional liposomes (∼50–70%) (Toyofuku et al., 2019). This superior LE may be attributed to the negatively charged surface of OMVs (zeta potential ∼ − 15 mV), which electrostatically interact with positively charged DOX (pKa ∼8.3), enhancing drug retention within the vesicle (Table S4) (Feng et al., 2024). Besides, the moderate and consistent increase in hydrodynamic diameter post-loading, accompanied by a preserved monodisperse size distribution (Fig. 3D), suggests that the sonication process achieved effective drug encapsulation without causing substantial vesicle fragmentation or aggregation—an indication that core vesicle integrity was maintained. While these physicochemical and functional data support the robustness of our loading method, direct biophysical assessment of membrane integrity (e.g., under sonication stress) remains a subject for future mechanistic study.

A critical requirement for effective nanodrug delivery systems is the ability to release therapeutic agents selectively at the tumor site, minimizing off-target toxicity (Chai et al., 2025, Al-Dossary et al., 2022). DOX exerts its cytotoxic effect by intercalating into DNA and inhibiting topoisomerase II, but its efficacy is highly dependent on its intracellular accumulation. We therefore evaluated the drug release kinetics of OMVs-DOX under different pH conditions (pH 7.4, physiological; pH 5.0, endosomal/lysosomal compartments) (Fig. 3C). Notably, OMVs@DOX exhibited a pH-dependent release profile: only ∼40% of the DOX was released at a pH of 7.4 after 48 h, whereas ∼80% was released at a pH of 5.0. This pH sensitivity is is consistent with mechanisms reported for analogous vesicular systems, where protonation of membrane components and subsequent bilayer destabilization under acidic conditions can accelerate drug diffusion (Chen et al., 2020).

The tumor microenvironment (TME) is characterized by an acidic pH (6.0–6.8) because of increased glycolysis and poor vascularization, whereas endosomal compartments (pH 5.0–6.0) mediate the intracellular trafficking of nanocarriers (Sun et al., 2025; Liu et al., 2025). The pH-responsive release of OMVs@DOX thus ensures that most of the drug is retained in the bloodstream during circulation, reducing systemic exposure, and is rapidly released upon internalization into tumor cells or penetration into the TME. This is supported by our in vitro cytotoxicity assays, in which OMVs-DOX showed comparable IC₅₀ values to those of free DOX at 48 h but significantly greater cytotoxicity at earlier time points (6–12 h) when cells were exposed to acidic conditions (data not shown).Together, these functional data underscore the utility of the pH-responsive release. Direct biophysical investigation of the OMV membrane under varying pH conditions would provide more definitive mechanistic insight in future studies. These results highlight the potential of OMVs-DOX to enhance tumor-specific drug delivery while reducing off-target effects on healthy tissues.

One major challenge in nanodrug delivery is the rapid clearance of nanocarriers by the RES, primarily via nonspecific phagocytosis by macrophages in the liver and spleen (Szöllosi et al., 2023). OMVs are rich in PAMPs and exhibit a potent ability to activate immunity, effectively engaging TLRs, including TLR2, TLR4, and TLR5, alongside other innate immune pathways (Chen et al., 2025b; Russo et al., 2018; Hayashi et al., 2001; Kuzmich et al., 2017; Liang et al., 2015). We previously reported that OMVs can be recognized by TLR2 and TLR4 on macrophages, triggering proinflammatory responses and cellular uptake (Liang et al., 2015; Zhang et al., 2024). To mitigate this, we pretreated RAW 264.7 macrophages with SsnB, a selective inhibitor of TLR2/4 signaling, and observed an ∼60% reduction in OMV internalization (Fig. 4A–B). These findings suggest that TLR2/4 signaling is a key mediator of OMVs uptake by macrophages.

Importantly, in vivo biodistribution studies (Fig. 4D) likely revealed that SsnB pretreatment increased the tumor accumulation of OMVs@DOX by ∼30%, as reduced RES clearance would prolong circulation time and enhance passive targeting via the enhanced permeability and retention (EPR) effect. These findings align with recent studies demonstrating that modulating macrophage–TLR interactions can improve the pharmacokinetics of nanocarriers (Deng et al., 2022). By targeting both active (GPC3/CD133) and passive (EPR) mechanisms while reducing nonspecific clearance, (GPC3 + CD133)ᵀ-OMVs achieve a “dual-enhanced” tumor targeting strategy, which is critical for maximizing therapeutic efficacy (Nie et al., 2020).

The ultimate goal of targeted drug delivery is to translate preclinical findings into clinical applications. Our in vivo experiments (Fig. 5) demonstrated that (GPC3 + CD133)ᵀ-OMVs@DOX significantly inhibited tumor growth in Huh-7 xenograft models, with a tumor inhibition rate (TIR) of ∼75%, compared with ∼40% for free DOX and ∼ 60% for single-targeted OMVs@DOX. This superior efficacy may be attributed to three key factors: (1) enhanced tumor targeting via dual-specific peptide recognition, (2) pH-responsive drug release maximizing intracellular DOX accumulation, and (3) reduced RES clearance prolonging circulation time. The stable body weight of mice treated with (GPC3 + CD133)ᵀ-OMVs@DOX, in contrast to the significant weight loss induced by free DOX (∼15%), provides preliminary evidence for its reduced acute systemic toxicity (Fig. 5E). While this is a robust indicator, future studies incorporating hematological analysis will further substantiate the biocompatibility profile.

The improved pharmacokinetics, attributed to SsnB-mediated suppression of macrophage clearance, was a key factor in this efficacy. However, the role of SsnB as a TLR2/4 antagonist may have extended beyond merely prolonging circulation. In the tumor microenvironment, constitutive TLR signaling can promote a pro-tumorigenic state by driving NF-κB and MAPK pathways, leading to the production of cytokines (e.g., TNF-α, IL-6) that foster inflammation and cell survival (Tang et al., 2018). Intriguingly, beyond its anti-phagocytic effect, SsnB has been shown to directly inhibit cancer cell proliferation and induce apoptosis in models such as colorectal cancer by suppressing these same pathways (Çirçirli et al., 2025). It is plausible that in our study, SsnB pretreatment contributed to the therapeutic outcome not only by enhancing OMV delivery but also by simultaneously dampening detrimental TLR-driven inflammation within tumors. A critical caveat is that our use of immunocompromised nude mice precluded the evaluation of SsnB's potential interactions with the adaptive immune system or with the intrinsic immunostimulatory properties of OMVs. Therefore, dissecting the net immunomodulatory effect of combining SsnB with an OMV-based platform remains an essential direction for future research in immunocompetent models.

The biological toxicity of OMVs@DOX complexes warrants comprehensive evaluation, encompassing two key components: the inherent toxicity of OMVs and the well-documented cytotoxicity of DOX. First, high-dose OMVs (>200 μg/mouse) derived from pathogenic or nonprobiotic bacterial strains may induce inflammatory responses, hemolysis and organ damage, including hepatic/renal injury and splenomegaly, potentially because of their LPS content or transport of bioactive molecules (Zhuang et al., 2021; Noh et al., 2023; Li et al., 2020; Qing et al., 2020). However, in this study, we utilized OMVs derived from E. coli Nissle 1917 (ECN), a probiotic strain recognized for its low immunogenicity and high biocompatibility (Grozdanov et al., 2002; Gelfat et al., 2022). ECN-derived OMVs (ECN-OMVs) have been shown to exhibit minimal systemic toxicity in previous in vivo investigations, with reduced proinflammatory cytokine secretion and negligible histopathological changes in major organs (Li et al., 2024). Notably, the use of nude mice in our model further mitigated potential immune-mediated toxicity, as these animals lack functional adaptive immunity, which is a key driver of OMV-associated inflammatory reactions. Second, DOX toxicity remains a critical concern because of its well-characterized off-target effects, particularly cardiotoxicity and myelosuppression, which arise from its intercalation into DNA and inhibition of topoisomerase II (Mattioli et al., 2023; Xie et al., 2025; Alavi and Varma, 2020). However, the tumor-targeting ability of ECN-OMVs may alleviate systemic DOX exposure. By leveraging the natural tropism of ECN-OMVs for tumor microenvironments (e.g., via enhanced permeability and retention effects or specific receptor binding), our complex is hypothesized to concentrate DOX in tumor tissues, thereby reducing accumulation in vital organs and minimizing cardiotoxicity compared with free DOX (Li et al., 2019). Preliminary histopathological analyses of heart and kidney tissues in our nude mouse model further support this notion, as no significant DOX-induced tissue damage was detected, which is consistent with the enhanced targeting efficiency of OMV-based delivery systems. Consistent with these findings, histological analysis of major organs (liver, spleen, kidney, and heart) revealed minimal damage in the (GPC3 + CD133)ᵀ-OMVs@DOX group, whereas free DOX induced severe cardiomyocyte vacuolation and nuclear pyknosis (Fig. 6C). The favorable biocompatibility of our OMVs formulation is multifaceted. First, the encapsulation by OMVs prevents direct interaction of DOX with healthy tissues (You et al., 2025). Second, the intrinsic properties of the vesicle itself contribute to its safety profile. The endotoxin activity of our OMVs was quantified at 1.2 ± 0.3 EU/μg protein, a level consistent with expectations for vesicles derived from the probiotic ECN.This defined and relatively low endotoxin burden is a key factor in mitigating uncontrolled systemic inflammation, thereby supporting the safety of repeated systemic administration observed in our study and informing its translational feasibility.

While our study provides compelling evidence for the utility of dual-targeted OMVs in HCC therapy, several limitations warrant consideration. First, our targeting validation was primarily conducted in the Huh-7 cell line, which was selected for its robust co-expression of GPC3 and CD133 and served as an ideal model for the initial proof-of-concept. Given the known heterogeneity of HCC, future work should extend this evaluation to a broader panel of HCC cell lines with varying GPC3 and CD133 expression profiles to more comprehensively assess the platform's applicability across different molecular subtypes.

Second, the in vivo model used (Huh-7 xenografts in immunocompromised mice) does not fully recapitulate the complexity of the human TME, including the presence of a functional immune system, tumor stroma, and interstitial pressure (Cejas et al., 2024; Liu et al., 2023; Chiorazzi et al., 2023). Future studies should evaluate (GPC3 + CD133)ᵀ-OMVs@DOX in immunocompetent HCC models (e.g., transgenic mice or patient-derived xenografts) to better assess its immunomodulatory effects and therapeutic potential in a more physiologically relevant setting (Liu et al., 2023, Chiorazzi et al., 2023).

Third, while our acute toxicity studies revealed no significant organ damage, the long-term safety and immunogenicity profile of repeated OMV administration remains a critical translational consideration. This is because OMVs, as carriers of bacterial components like LPS, are potent innate immune activators. We acknowledge the recognized challenge that chronic exposure could potentially lead to altered immune responses (Effah et al., 2024; Gerritzen et al., 2017; Guo et al., 2025). Importantly, this inherent immunostimulatory property is a double-edged sword; it underpins the clinical success of licensed OMV-based vaccines (e.g., meningococcal vaccines), demonstrating that safety and immunogenicity can be balanced in a controlled manner. Consequently, dedicated repeat-dose studies in immunocompetent models are essential. Future work must investigate the pharmacokinetics of OMVs at multiple doses and evaluate their biodistribution in nontargeted tissues (e.g., the brain and muscle) to ensure safety. Furthermore, leveraging established engineering strategies—such as genetic detoxification of LPS or surface engineering to fine-tune immune recognition—will be key to systematically optimizing the long-term therapeutic window of this platform.

Finally, the scalability of OMVs production needs optimization. While OMVs are biologically advantageous, their isolation via ultracentrifugation is time-consuming and results in low yields compared with those of synthetic nanocarriers (Gerritzen et al., 2017, Guo et al., 2025). The development of scalable purification methods (e.g., tangential flow filtration) and genetic engineering of bacteria to overexpress targeting peptides on OMVs could increase their clinical translatability.

5. Conclusion

In summary, we developed a dual-targeted nanodelivery system using GPC3 and CD133 peptide-modified OMVs to deliver DOX for HCC therapy. Our results demonstrate that this system enhances tumor-specific uptake via combinatorial biomarker recognition, optimizes drug release kinetics in response to the acidic TME, reduces nonspecific phagocytosis by modulating TLR signaling, and significantly improves in vivo antitumor efficacy while minimizing systemic toxicity. These findings highlight the potential of OMVs as versatile and biocompatible platforms for targeted cancer therapy and provide a foundation for further preclinical and clinical development of dual-targeted OMV-based nanodrugs for HCC treatment.

The following are the supplementary data related to this article.

Fig. S1.

Yield comparison between ECN and ΔE.

(A) Comparison of OMV production yields between wild-type EcN and the hypervesiculating ΔE mutant.

Data are presented as mean ± SD (n=3). Statistical significance is denoted as follows: *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. S2.

Validation of M1 polarization in RAW264.7 cells induced by LPS.

(A) Flow cytometry analysis of CD86-positive cells (M1 macrophage marker) after 24 h stimulation with different concentrations of LPS.

(B) Representative bright-field images of RAW264.7 cells under M0 (unstimulated) and M1 (LPS-stimulated) conditions. Scale bar: 100 μm.

Data are presented as mean ± SD (n=3). Statistical significance is denoted as follows: *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. S3.

Immunohistochemical analysis of tumor tissues.

(A) Representative images of E-cadherin and cleaved caspase-3 staining in tumor sections from different treatment groups. E-cadherin expression was quantified by analyzing average optical density (AOD) using ImageJ software. Caspase-3 activation was assessed by calculating the percentage of positive cells. Scale bars: 50 μm.

Data are present as mean ± SD (n = 5). Statistical significance is denoted as follows: #, compared to the control group (SsnB + PBS); *, compared between treatment groups (#/* p < 0.05, ##/** p < 0.01, ###/*** p < 0.001,).

Fig. S4.

Histopathological evaluation of major organs from additional treatment groups.

(A) Representative H&E-stained sections of liver, spleen, kidneys, and heart from mice treated with SsnB+blank OMVs, SsnB+GPC3ᵀ-OMVs@DOX, and SsnB+CD133ᵀ-OMVs@DOX group. Scale bars: 100 μm.

Supplementary tables

Ethics approval and informed consent

This study was approved by the Ethics Committee of the Animal Experiment Center of Wuhan University. All methods were performed in accordance with the Institutional Review Board guidelines and regulations.

Availability of data and materials

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

CRediT authorship contribution statement

Zuo Mou: Writing – original draft, Methodology, Investigation, Formal analysis, Data curation. Yuefeng Zhang: Writing – review & editing, Writing – original draft, Resources, Project administration, Conceptualization. Xiaodian He: Investigation, Data curation. Mingze Zhang: Visualization, Formal analysis. Xiaoqin He: Writing – review & editing, Resources, Funding acquisition. Wei Wang: Writing – review & editing, Supervision, Data curation. Zehao Liu: Writing – review & editing, Visualization, Validation. Xinxin Xiong: Writing – review & editing, Methodology, Formal analysis. Peng Ma: Supervision, Resources, Project administration. Kaihuan Yu: Supervision, Resources, Project administration.

Funding

This project was financially supported by the National Natural Science Foundation of China (No. 82001940).

Declaration of competing interest

The authors report no conflicts of interest in this work.

Data availability

Data will be made available on request.