Engineering human PEG10-based nanoparticles for RNA self-packaging, delivery and cancer therapy

Abstract

PEG10 protein was recently uncovered to self-assemble and self-package its own mRNA into nanoparticles, but the particles require expensive transfection for production and have yet to be explored for cancer therapy. Here we develop a human PEG10-based nanoparticles (PBNPs) platform for cargo RNA self-packaging and delivery for cancer therapy. We design a process to improve the PBNPs production for 11.3-fold while reducing the cost. The PBNPs self-package mRNA of at least 7336 nucleotides and remain stable for 7 months. We engineer the PBNPs surface and tremendously improve mRNA delivery efficiencies to various cancer cells, particularly colon cancer cells (≈71%). We further reprogram the PBNPs to deliver an immunotherapeutic mRNA cocktail to colon cancer cells, which elicits T cell responses in vitro. In vivo co-administration of the engineered PBNPs and chemodrug in female mice synergizes immune responses and promotes anti-cancer efficacy, implicating the potential of PBNPs as an RNA delivery vehicle for immunotherapy.

mRNA cancer therapies are limited by costly production, inefficient RNA packaging, and poor tumour targeted delivery of non-viral vectors. The authors engineer PEG10 based nanoparticles enabling efficient RNA selfpackaging, enhanced cancer targeting ( ~ 71% in colon tumours), and cytokine delivery that boosts anti-tumour immunity and improves therapeutic efficacy.

Introduction

The success of COVID-19 vaccines, recent regulatory approvals and hundreds of ongoing clinical trials¹ have marked the emergence of mRNA therapeutics for vaccines, diseases and cancer therapies^2,3. However, cost-effective mRNA production, efficient packaging, and targeted delivery to specific cells while minimizing side effects remain major challenges^3,4. These challenges are particularly critical in cancer therapy, where precise delivery and localized immune activation are essential to achieve therapeutic efficacy while limiting systemic toxicity⁴.

To date, non-viral vectors such as lipid nanoparticles (LNPs)^1,5–8, membrane-derived extracellular vesicles (EVs)^9–11 and virus-like particles (VLPs)^12–15 have been extensively exploited for mRNA delivery. Despite their promise, each platform has inherent limitations, including complex manufacturing processes, limited cargo control, suboptimal targeting and safety concerns. Recently, a novel humanized PEG10-based RNA delivery strategy was developed¹⁶. PEG10 is an endogenous retrotransposon-derived protein expressed in mammals and is homologous to retroviral Gag capsid proteins¹⁶. PEG10 proteins self-assemble into nanoparticles and preferentially bind to the 5’ and 3’ untranslated regions (UTRs) of peg10 mRNA, hence self-packaging its own mRNA¹⁶. The mRNA cargo of PEG10 particles was reprogrammed by flanking the gene of interest with peg10 UTRs for functional transfer of the mRNA cargo to neuroblastoma (N2a) cells. The mRNA delivery requires ectopic expression of a fusogenic protein, such as vesicular stomatitis virus G protein (VSVG). By transfecting 3 plasmids encoding PEG10, peg10 UTR-flanking mRNA and VSVG into the producer HEK293T cells, both mouse and human PEG10 particles were produced to package, secrete, and deliver mRNA cargos. Human PEG10 particles were harnessed to deliver Cas9 mRNA for CRISPR-mediated gene disruption in the mouse N2a cells, thus underscoring their potential as a novel mRNA delivery platform¹⁶.

PEG10-based particles possess features that are particularly attractive for cancer therapy, including their human origin, potential for reduced immunogenicity, and intrinsic RNA packaging capability^17,18. However, two key challenges remain. First, production currently relies on co-transfection of three plasmids, which is costly and difficult to scale. Second, their application in cancer therapy, particularly for delivering immunostimulatory mRNAs, has yet to be explored.

To address these challenges, we first developed a baculovirus (BV)-based system for efficient production of PEG10-based nanoparticles (PBNPs). BV is a non-pathogenic insect virus with a favorable biosafety profile (BSL-1) and is widely used in vaccine manufacturing¹⁹ (e.g., COVID-19 VLP vaccines²⁰). Importantly, BV can transduce various mammalian cells at efficiencies exceeding 95% without the need of expensive transfection reagents²¹. By driving the transgene with a promoter active in mammalian cells, BV has been harnessed as a gene delivery vector for tissue regeneration^22,23, pain mamagement²⁴ and gene editing^25–27. Here we exploited BV vectors to co-express PEG10, peg10 UTR-flanked mRNA and VSVG in HEK293T cells, achieving an 11.3-fold improvement in PBNPs production with reduced cost compared to plasmid transfection. PBNPs produced by BV can package large RNA (at least 7336 nt) and remain stable at 4 °C for 7 months. By engineering the surface protein of PBNPs, we tremendously improved the mRNA delivery efficiency to different cancer cells, reaching ≈ 71% in colon carcinoma cells. We further reprogrammed PBNPs to deliver a cytokine mRNA cocktail (Il12 and OX40L) to colon cancer cells and demonstrated that the transfected cells elicit immune responses in vitro. In vivo, co-administration of engineered PBNPs with oxaliplatin enhanced anti-tumor responses in a mouse tumor model. Altogether, this study establishes an efficient PBNPs production platform and demonstrates their potential for RNA self-packaging, targeted delivery, and cancer immunotherapy.

Results

Production and characterization of PBNPs produced by the BV system

To produce PBNPs more robustly and cost-effectively, we constructed two BV donor plasmids: (i) pBac-PEG10-VSVG expressing human PEG10 and VSVG, and (ii) pBac-cEGFP expressing the cargo egfp mRNA flanked by 5’ and 3’ peg10 UTRs (Fig. 1a). These plasmids were used to generate two corresponding BV vectors (Bac-PEG10-VSVG and Bac-cEGFP, Fig. 1a). After determining the doses of BV and plasmid that conferred similar delivery efficiencies (Supplementary Fig. 1a), we co-transduced HEK293T cells with Bac-PEG10-VSVG/Bac-cEGFP or co-transfected cells with the two plasmids, followed by supernatant harvest and ultracentrifugation purification (Fig. 1b). The functional titers (transfecting unit per milliliter, TU/ml) were determined by transfecting HEK293T cells with diluted PBNPs (Supplementary Fig. 1b, c). Compared with plasmid co-transfection, BV co-transduction exhibited an 11.3-fold improvement in PBNPs functional titer (Fig. 1c) with lower cost (Supplementary Fig. 2), indicating the advantage of BV for PBNPs production.

Fig. 1. Production and characterization of PBNPs produced by BV.

a Plasmids and BVs for PBNPs production. pBac-PEG10-VSVG and pBac-cEGFP are plasmids encoding PEG10/VSVG and egfp mRNA cargo flanked by 5’ and 3’ peg10 UTR, respectively. The plasmids were used to generate corresponding recombinant BV by Bac-to-Bac^TM system. b Workflow of PBNPs production and purification by BV transduction. For each production batch, HEK293T cells were cultured in ten 15-cm dishes (≈ 2 × 10⁷ cells/dish) and co-transduced with Bac-PEG10-VSVG/Bac-cEGFP at MOI 50/50, or co-transfected with the two plasmids (250,000 ng each per 10 dishes). PBNPs released into supernatant were harvested after 48 h, filtered and concentrated by ultracentrifugation. The PBNPs were further purified by ultracentrifugation. c Functional titers of PBNPs produced by BV transduction and plasmid transfection. The functional titers, expressed as transfecting unit (TU) per milliliter (TU/mL), were determined by PBNPs transfection of HEK293T cells, followed by flow cytometry analysis of GFP⁺ cells (see Supplementary Fig. 1). d RNA extracted from PBNPs as analyzed by denaturing gel electrophoresis. e DNA electrophoresis of cDNA and the template plasmid pBac-cEGFP. The RNA was subjected to RT-PCR using primer pairs spanning different regions of the cargo egfp. pBac-cEGFP was subjected to PCR using the same primers. f Relative abundance of protein-coding transcripts as analyzed by RNA-seq. The transcripts per million (TPM) of each protein-coding transcript were normalized to the total protein-coding TPM. Only the top 6 transcripts are shown. RNA-seq analysis was performed using a single biological sample. g Western blot analysis of producer cell lysates and purified PBNPs. GP64 is a BV envelope protein. GAPDH served as a control. h Representative TEM images of PBNPs. For panel c, data are presented as mean±SD from three independent production experiments. Source data are provided as a Source Data file.

To characterize the PBNPs produced by BV, we analyzed PBNPs RNA by gel electrophoresis and found a predominant band with a length corresponding to egfp and flanking peg10 UTRs (Fig. 1d). PCR using primer pairs spanning different regions of cDNA and the template plasmids (Fig. 1e), together with Sanger sequencing (Supplementary Fig. 3a), confirmed that egfp was flanked by the 5’ and 3’ peg10 UTRs. qRT-PCR further showed comparable copy numbers of egfp, 5’ UTR and 3’ UTR on the same mRNA (Supplementary Fig. 3b, c), indicating intact packaging of full-length transcripts.

RNA sequencing (Supplementary Data 1) revealed that egfp mRNA is the most abundant protein-coding transcript (36.1%) in the PBNPs, followed by peg10 mRNA (18.9%), whereas other protein-coding transcripts (e.g., RPL41, RPS27, TMSB10, and RPS2) were detected at substantially lower levels (Fig. 1f). In addition, host-derived small RNAs, including pseudogene transcripts, small cytoplasmic RNA (scRNA) and ribosomal RNA fragments, were detected (Supplementary Fig. 4a). These RNAs represent highly expressed cellular species^28–30 and are therefore likely incorporated through passive background encapsulation, consistent with observations in extracellular vesicles^31–33. Despite this background, transcript ranking analysis showed that egfp mRNA is the third most abundant transcript overall, exceeded only by two small cytoplasmic non-coding transcripts (RNA5SP202, ≈ 120 nt; RN7SL1 ≈ 300 nt) (Supplementary Fig. 4b). These findings indicate that RNA packaging by PBNPs is selective and enriched, with preferential incorporation of the engineered peg10 UTR–flanked cargo transcript.

Biochemical characterization further confirmed that PBNPs consist of PEG10 and VSVG without detectable BV envelope protein GP64 (Fig. 1g), and contained only trace amounts of infectious BV (Supplementary Fig. 5). PBNPs exhibited a narrow size distribution and negative surface charge (Supplementary Fig. 6). Transmission electron microscopy revealed spherical particles with retrovirus-like morphology and a diameter of ≈ 100 nm (Fig. 1h).

Collectively, these data demonstrate that BV-mediated production substantially enhances PBNPs yield compared to plasmid transfection, and that the resulting particles self-assemble from PEG10/VSVG while preferentially packaging peg10 UTR–flanked cargo mRNA. Furthermore, complementary analyses (functional titer, cargo RNA quantification, and physicochemical characterization) indicate that PBNPs retain structural integrity and functional activity during storage at 4 °C for up to 7 months (Supplementary Fig. 7).

Packaging capacity of PBNPs

BV has a large cloning capacity³⁴ that allowed us to increase the cargo size in the BV vector. To explore the packaging capacity of PBNPs, we constructed more BV vectors to carry cargoes with lengths ranging from ≈ 1720 to 7336 nt (Fig. 2a). The PBNPs produced from these BV vectors had similar genome titers (mRNA copy numbers/μl, Fig. 2b) and functional titers (Fig. 2c). When the PBNPs encoding EGFP and mCherry were used to transfect HEK293T cells, the transfection efficiencies fell in the range of ≈ 9–21% (Fig. 2d-e). Importantly, the transfection efficiencies did not significantly (p > 0.05) vary with the cargo lengths, indicating that the PBNPs self-package mRNA cargoes of at least 7336 nt without compromising the functional titers.

Fig. 2. Packaging capacity of PBNPs.

a BVs encoding cargoes with different lengths. Bac-cEM encoded EGFP and mCherry reporter proteins linked by a P2A sequence with a cargo length of 2536 bp. Bac-EMLuc encoded an additional luciferase with a total cargo length of 4264 bp. Bac-cEMdLB-pH encoded EGFP, mCherry and a dCas12a-p65-HSF1 fusion protein, with a total length of 7336 bp. The PBNPs were produced by co-transducing HEK293T cells with Bac-PEG10-VSVG (MOI 50) and one of the following (MOI 50): Bac-cEGFP, Bac-cEM, Bac-EMLuc or Bac-cEMdLB-pH. b Genome titers (mRNA copy number) of PBNPs self-packaging cargoes of different lengths as determined by qRT-PCR (using primers for 3’ UTR). c Functional titers of PBNPs self-packaging cargoes of different lengths. d Fluorescence micrographs of HEK293T cells transfected with different PBNPs at 10 copies/cell. e Flow cytometry of EGFP⁺ and mCherry⁺ cells. Data are presented as mean±SD from three independent experiments. Statistical significance was determined using ordinary one-way ANOVA. ns, not significant. Source data are provided as a Source Data file.

Engineering PBNPs to improve RNA delivery into different cancer cells

EGFR and PD-L1 are two common cancer surface proteins³⁵. To enhance PBNP-mediated transfection efficiency of cancer cells, we engineered targeting ligands by fusing scFvs specific for EGFR or PD-L1 to the N-terminus of VSVG (Supplementary Fig. 8a), which corresponds to the ectodomain exposed on the particle surface³⁶. Alternatively, we swapped VSVG with the intracellular envelope protein SYNA¹⁶. For initial screening, these PBNPs were produced by plasmid transfection. Wild-type PBNPs displaying VSVG exhibited low transfection efficiencies (from< 5% to 15%) across a panel of cancer cells (Supplementary Fig. 8b, c). In contrast, pseudotyping with alternative envelope proteins yielded variable results. Among them, PBNPs displaying the EGFR-targeting fusion protein (hereafter termed eVSVG) showed the most consistent improvement, achieving up to 29% transfection efficiency in CT26 colon carcinoma cells. Based on this result, eVSVG was selected for subsequent BV-mediated production of engineered PBNPs (ePBNPs).

As VSVG naturally forms trimers on the particle surface^37,38, we next investigated whether the stoichiometry of eVSVG and VSVG influences incorporation and function. To this end, we constructed 4 BVs for co-transduction of HEK293T cells (Fig. 3a). Bac-PEG10 (MOI 50) and Bac-cEGFP (MOI 50) expressed PEG10 and cargo mRNA. Bac-VSVG and Bac-eVSVG expressed VSVG and eVSVG, respectively. The total MOI for Bac-eVSVG/Bac-VSVG was fixed at 50, but the MOI percentage (%) of Bac-eVSVG was varied (Fig. 3a). Western blot analysis of concentrated particles confirmed the incorporation of eVSVG together with VSVG into ePBNPs at Bac-eVSVG ratios of 33% and 67% (Fig. 3b). Functional titers were highest at 0% and 33% eVSVG but decreased at 67% and 100% (Fig. 3c). Consistently, transfection of CT26 cells showed that ePBNPs produced with 33% Bac-eVSVG achieved markedly enhanced efficiency (71%), significantly (p < 0.05) outperforming wild-type PBNPs (wtPBNPs) and those produced with 67% or 100% Bac-eVSVG (Fig. 3d). Immunogold TEM further confirmed surface presentation of eVSVG on ePBNPs (Fig. 3e). Based on these results, the 33% Bac-eVSVG condition was selected for subsequent ePBNPs production.

Fig. 3. Engineering PBNPs to improve RNA delivery into different cancer cells.

a BVs for producing wild-type PBNPs (wtPBNPs) and engineered PBNPs (ePBNPs). Bac-PEG10 encoded PEG10. Bac-cEGFP encoded EGFP. Bac-VSVG encoded wild-type VSVG. Bac-eVSVG encoded engineered VSVG (eVSVG) consisting of the heavy chain (V_H) and light chain (V_L) of the EGFR-specific scFV, which was fused to the N-terminus of VSVG by a GS linker. His₆, a His₆ tag. SP, signal peptide. HEK293T cells were transduced with Bac-PEG10 (MOI 50), Bac-cEGFP (MOI 50) and Bac-eVSVG/Bac-VSVG (total MOI 50 with varying MOI% of Bac-eVSVG). When the MOI% for Bac-eVSVG was 0%, the resultant PBNPs were wtPBNP. b Western blot analysis of eVSVG/VSVG in the PBNPs. eVSVG contained a His₆ tag that enabled the detection of eVSVG using the anti-His₆ antibody. c Functional titers of PBNPs produced from different MOI%. d EGFP⁺ CT26 cells using PBNPs produced from different MOI% as determined by flow cytometry. e Immunogold TEM image of PBNPs. Arrows indicate the gold particle-labeled His₆. f Fluorescence micrographs of HEK293T and various cancer cells transfected by PBNPs (10 copies egfp mRNA/cell). g Transfection efficiencies as determined by flow cytometry. CT26, colon carcinoma; U-87, glioblastoma; MC38, colon adenocarcinoma; Hepa1-6, hepatoma; Huh7, hepatocellular carcinoma; Hep3B, hepatocellular carcinoma; 4T1, breast cancer. Data are presented as mean ± SD from three independent experiments. Statistical significance in panels c and d was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test, with 33% of eVSVG as the reference group. Statistical significance in panel g was determined using two-way ANOVA followed by Šídák’s multiple comparisons test. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns, not significant. Source data are provided as a Source Data file.

We next evaluated expression kinetics in CT26 cells. ePBNPs outperformed wtPBNPs, achieving 71% and 57% EGFP⁺ cells at 24 h and 48 h post-transfection, respectively. (Supplementary Fig. 9). The ePBNPs produced by BV also transfected various cancer cells at higher efficiencies than the wtPBNPs produced by BV transduction (Fig. 3f, g) or by plasmid transfection (Supplementary Fig. 8). Notably, the ePBNPs also transfected glioma cancer cells U-87 and three liver cancer cells (Hepa1-6, Huh7 and Hep3B) at efficiencies >60% (Fig. 3f, g).

ePBNPs self-packaging Il12 and OX40L mRNA stimulated T cell responses

mRNA cocktails encoding multiple cytokines show promising antitumor effects by enhancing the cellular immunities³⁹. We further swapped the EGFP cargo with well-known immunostimulatory cytokines IL-12⁴⁰ and OX40L⁴¹, constructed Bac-cOI (Fig. 4a) and produced the ePBNP.OI self-packaging the Il12 and OX40L mRNA (Fig. 4a). We determined that 10 cargo copies/cell conferred reasonable CT26 cell viability and most robust IL-12 expression (Supplementary Fig. 10). The wild-type PBNP.OI (wtPBNP.OI) was similarly produced as the control.

Fig. 4. ePBNPs self-packaging Il12 and OX40L mRNA stimulated T cell response.

a BVs for producing ePBNP.OI. Bac-cOI encoded IL-12 and OX40L, which were flanked by 5’ and 3’ peg10 UTR. ePBNP.OI was produced by transducing HEK293T cells with Bac-PEG10-VSVG (MOI 33.3), Bac-eVSVG (MOI 16.7) and Bac-cOI (MOI 50). wtPBNP.OI was produced by transducing HEK293T cells with Bac-PEG10-VSVG (MOI 50) and Bac-cOI (MOI 50) as a control. ePBNP carrying egfp was produced as another control. b Co-culture of splenocytes with CT26 cells that were mock-transfected or transfected with ePBNP.OI, wtPBNP.OI or ePBNPs (expressing EGFP). The splenocytes were stimulated with Dynabeads™ Mouse T-Activator CD3/CD28 for T-Cell Expansion and Activation. c IL-12 and OX40L expression kinetics after co-culture as determined by ELISA. d IFN-γ and TNF-α expression kinetics after co-culture as determined by ELISA. The naïve splenocytes without co-culture/Dynabeads™ treatment served as the control. e Splenocytes proliferation after 3 days of co-culture. f Percentage of activated CD4⁺ and CD8⁺ T cells in the splenocyte population. The cells were subjected to triple immunostaining for T cells (CD3/CD69/CD4 or CD3/CD69/CD8). Data are presented as mean ± SD from three independent experiments. Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test, with ePBNP.OI as the reference group. **p < 0.01; ***p < 0.001; ****p < 0.0001. Source data are provided as a Source Data file.

We next transfected CT26 cells with the ePBNP.OI or wtPBNP.OI and co-cultured the cells with mouse splenocytes (Fig. 4b). As negative controls, we also mock-transfected cells (Mock group) or transfected cells with ePBNPs expressing EGFP (ePBNP group). As expected, ePBNP and Mock groups barely conferred IL-12 and OX40L expression, while ePBNP.OI-transfected CT26 cells simultaneously expressed IL-12 and OX40L at levels higher than those of the wtPBNP.OI group at 24 h (Fig. 4c). Concurrently, the co-cultured splenocytes in the ePBNP.OI group secreted more IFN-γ and TNF-α than those in the wtPBNP.OI, ePBNP, and Mock groups (Fig. 4d). Meanwhile, the ePBNP.OI group induced more apparent splenocyte proliferation (Fig. 4e) as well as CD4⁺ and CD8⁺ T cell activation (Fig. 4f) than the wtPBNP.OI, ePBNP and Mock groups. These data proved that ePBNP.OI transfection of CT26 cells confers sufficient IL-12 and OX40L expression to stimulate T cell responses in a paracrine manner.

ePBNPs self-packaging Il12 and OX40L mRNA improved immunotherapy effects in vivo

We next evaluated the therapeutic efficacy of ePBNP.OI in combination with the chemotherapeutic drug oxaliplatin (OXA) in the CT26 tumor model. When tumors reached ≈ 50 mm³ (day 0), we intratumorally injected PBS (PBS group), OXA (OXA group), ePBNP.OI (1 × 10⁷ cargo copies/dose, ePBNP.OI group) or OXA combined with ePBNP.OI (Combo group, n = 6 for all groups, Fig. 5a).

Fig. 5. ePBNPs self-packaging Il12 and OX40L mRNA improved immunotherapy effects in vivo.

a Scheme of animal experiments. After the tumor size reached 50 mm³ (day 0), we intratumorally injected PBS (PBS group), OXA (3 mg/kg, OXA group), ePBNP.OI (1 × 10⁷ cargo copies/dose, ePBNP.OI group) or OXA combined with ePBNP.OI (Combo group). n = 6 for all groups. b Tumor volume for each group. CR, complete response (tumor volume < 100 mm³). c Survival rate. d TNF-α levels. e INF-γ levels. TNF-α and INF-γ levels in the sera were analyzed by ELISA at 24 h after the complete therapy regime. Data are presented as mean ± SD from three independent experiments. Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test, with Combo as the reference group. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001. Source data are provided as a Source Data file.

Both PBS and OXA groups failed to repress tumor growth while ePBNP.OI group moderately delayed the tumor progression and achieved complete remission (CR, tumor size < 100 mm³) in 1 out of 6 mice (Fig. 5b). By priming the immune responses with OXA first and subsequent immunostimulation by ePBNP.OI, the Combo group resulted in markedly enhanced tumor control, with tumor growth suppressed in 5 out of 6 mice and CR observed in 2 mice. Consistently, the Combo group achieved 100% survival at day 30, compared to 17% and 33% survival in the OXA and ePBNP.OI groups, respectively (Fig. 5c).

Mechanistically, ELISA analysis revealed that the Combo group induced higher levels of TNF-α and IFN-γ than the ePBNP.OI, OXA and PBS groups (Fig. 5d, e), indicating enhanced immune activation. In contrast, ePBNPs carrying egfp did not elicit appreciable increases in these cytokines (Supplementary Fig. 11). Furthermore, the Combo group was associated with increased infiltration of CD4⁺ T cells (Supplementary Fig. 12), suggesting changes in the tumor microenvironment. Biodistribution analysis by qRT-PCR showed substantially lower Il12 and OX40L mRNA levels in major organs (heart, lung, kidney, spleen and liver) than in the tumor (Supplementary Fig. 13). Blood chemistry and histological analyses did not reveal significant changes (p > 0.05) in hepatic (AST, ALT) or renal (BUN) markers, and no apparent pathological abnormalities were observed in major organs compared with untreated mice (Supplementary Fig. 14). Together, these results indicate that combining ePBNP.OI with OXA improves anti-tumor responses and immune activation, without evidence of overt toxicity under the conditions tested.

Discussion

Humanized PEG10 system is a promising mRNA delivery vehicle¹⁸. Here, we addressed two major issues that hamper the advancement of PEG10-based RNA delivery to cancer therapy: production and cancer targeting. To overcome production limitations, we developed a BV-based system to replace plasmid transfection. Using this approach, PBNPs yield increased by 11.3-fold with reduced cost, while maintaining particle composition and morphology (Fig. 1 and Supplementary Figs. 1, 2) comparable to those produced by triple plasmid transfection¹⁶. RNA profiling revealed that peg10 UTR–flanked cargo mRNA is preferentially enriched, although not exclusively packaged (Fig. 1f and Supplementary Fig. 4).

PBNPs have a packaging capacity of at least 7336 nt (Fig. 2), hence enabling potential delivery of a large mRNA transcript encoding single or multiple proteins. Furthermore, the large cloning capacity of BV (at least 38 kb³⁴) enables the generation of an all-in-one BV that efficiently transduces suspended mammalian cells, such as CHO cells, for PBNPs production (Supplementary Fig. 15), rendering BV amenable to large-scale production using the existing recombinant protein production facilities. These data collectively justified the use of BV, in lieu of plasmid transfection, for PBNPs production. In addition to PEG10, other mammalian retrotransposon Gag-like proteins such as Arc^42–44 and PNMA2⁴⁵ were recently discovered to self-assemble into particles and engineered to package RNA for delivery into mammalian cells. Our BV system may also be adapted to produce Arc- and PNMA2-based particles for RNA delivery.

Notably, recent studies uncovered regulators that inhibit the excessive production of PEG10 particles. PEG10 is selectively targeted by the proteasome shuttle factor ubiquilin 2 (UBQLN2)⁴⁶ and ubiquitin E3 ligase (UBE3A)⁴⁷ for degradation, hence decreasing the particle abundance. PEG10 particle secretion is also antagonized by the Gag-like RTL8 in humans through competitive incorporation into the particles⁴⁸. As such, screening and engineering of producer cells with minimal perturbation from these negative regulators may improve PBNPs production. Further enhancement of PBNPs production may be achieved by optimizing the producer cells for suspension culture and BV transduction.

Importantly, we found that wtPBNPs produced by plasmid transfection exhibited low delivery efficiencies ( < 5–15%) across multiple cancer cell types (Supplementary Fig. 8), consistent with the finding that VSVG-pseudotyped PEG10 particles are 4 ∼ 5 times less efficient than lentivirus for RNA delivery¹⁶. This limited efficiency may explain why wtPBNPs were not explored for cancer therapy. Although a recent study engineered PEG10 Gag domain with liver cancer-specific neoantigens and chemically coupled a CpG oligonucleotide to the particle surface, it neither engineered the fusogen nor exploited the particles for mRNA delivery⁴⁹.

To address this cancer-targeting limitation, we engineered the PBNPs surface and uncovered that EGFR scFV fusion to VSVG conferred the most pronounced improvement in mRNA delivery into various cancer cells, reaching 29% in CT26 cells (Supplementary Fig. 8). Switching the production system to BV further improved the ePBNPs transfection efficiency to 71% (Fig. 3d). Moreover, ePBNPs transfected glioma cancer cells and three liver cancer cells at efficiencies > 60% (Fig. 3g, h), implicating their potential for targeting diverse tumor types.

We further found that the stoichiometry ratio of Bac-eVSVG and Bac-VSVG influenced ePBNPs function (Fig. 3). Adjusting the ratio to 33% enhanced transfection efficiency, whereas complete replacement of VSVG reduced both particle titer and activity. These results suggest that a balanced fusogen composition is required for optimal particle performance. Immunogold TEM further confirmed the surface localization of eVSVG on ePBNPs, supporting its functional display despite the absence of high-resolution structural characterization.

Leveraging the modularity of ePBNPs, we engineered ePBNP.OI to self-package Il12 and OX40L mRNAs for delivery into CT26 cells, leading to enhanced T cell responses (Fig. 4). In combination with OXA, ePBNP.OI elicited stronger immune activation, resulting in improved tumor control, survival and anti-tumor immunity (Fig. 5 and Supplementary Fig. 12). Although mRNA expression is transient, IL-12 promotes Th1 polarization, stimulates IFN-γ production and activates cytotoxic lymphocytes^50–52, while OX40L provides co-stimulatory signals that support T cell expansion and persistence⁵³. Therefore, the IL-12/OX40L cocktail can substantiate T cell activation and persistence, enabling amplification of anti-tumor immunity. In addition, the treatment regimen involved four doses of ePBNP.OI (Fig. 5a). Such repeated mRNA delivery can reinforce T cell priming and activation within the tumor microenvironment, while also promoting cumulative remodeling of the immune milieu, resulting in sustained antitumor efficacy^39–41. Consistent with this mechanism, the combination treatment increased TNF-α and IFN-γ levels, and enhanced CD4⁺ T cell infiltration (Fig. 5d, e and Supplementary Fig. 12). Immune activation was attributable to the encoded cytokines, as ePBNPs carrying egfp did not induce appreciable TNF-α or IFN-γ responses (Supplementary Fig. 11). These findings support that repeated delivery of immunostimulatory mRNAs via ePBNPs can induce a self-reinforcing anti-tumor immune response that persists beyond transient expression of the delivered mRNA. Biodistribution analysis further showed preferential accumulation of Il12 and OX40L mRNAs in tumors with minimal off-target distribution, and no evident acute toxicity was found (Supplementary Figs. 13, 14). Our results collectively support the potential of ePBNPs for self-packaging an immunostimulatory mRNA cocktail, targeted delivery and cancer immunotherapy.

In comparison with other non-viral RNA delivery systems, ePBNPs offer several advantages. Current LNPs platforms involve complex and costly workflows, including DNA template production in E. coli, in vitro transcription, mRNA purification and packaging into 4 expensive lipids in the microfluidic devices⁶. In addition, ionizable lipid in the LNPs can trigger immune responses⁵⁴, and LNPs tend to accumulate in the liver⁵⁵. Although next-generation LNPs have been developed through lipid engineering⁵ or surface modification⁵⁶ to improve targeting and safety, surface decoration typically necessitates chemical conjugation of targeting molecules to the lipid, making the production process even more complicated and costly. Furthermore, a substantial fraction of LNPs may not efficiently encapsulate mRNA⁵⁷. In contrast, ePBNPs can be generated through a streamlined process involving molecular cloning and BV-mediated transduction. BV stock is readily produced to high titers simply by infecting insect cells²¹. All subsequent ePBNPs assembly and RNA-self-packaging processes occur spontaneously in the transduced mammalian cells without the need of further modification. Moreover, ePBNPs remain stable at 4 °C for up to 7 months as supported by preserved functional activity, RNA cargo integrity, and physicochemical properties (Supplementary Fig. 7), highlighting a potential advantage over less stable RNA delivery systems such as LNPs.

Alternatively, EVs are naturally secreted for cell-to-cell communication, induce low immune responses and are exploited for RNA delivery⁹. However, EVs are an impure mixture of particles that carry proteins, RNA and DNA¹⁰, making specific packaging of cargo mRNA difficult^9,11. Other limitations of EVs include poor cell targetability and low cargo encapsulation efficiency^10,58. Compared with EVs, PBNPs are superior thanks to their ability to enrich the peg10 UTR–flanked mRNA in the particles. Although various engineered VLPs are also developed to carry RNA for genome editing¹², base editing¹³, prime editing¹⁴ and even T cell engineering¹⁵, the production of these VLPs requires transfection of multiple plasmids encoding essential components, which is expensive and less efficient than viral transduction. Furthermore, these VLPs are typically produced by plasmids encoding Gag proteins derived from murine leukemia virus or human immunodeficiency virus^13,14, which may induce virus-associated immunogenicity. In contrast, ePBNPs can be produced readily by BV and are composed of PEG10 protein naturally expressed in humans^16,59, which may reduce immunogenicity¹⁸. Although the viral VSVG may still elicit immune responses, this may be circumvented by future engineering or replacement of VSVG with other human envelope proteins¹⁸.

To advance ePBNPs towards clinical applications, several challenges remain. PEG10 is involved in diverse physiological and pathological processes, including placental⁶⁰ and brain neuron development^16,47, cancer progression^61,62 and neurodegeneration⁴⁷. Therefore, long-term studies are required to assess whether ePBNPs administration may perturb endogenous PEG10 functions or promote tumorigenesis. Manufacturing processes must also be optimized for scalability. Production is currently dependent on adherent HEK293T cultures, which are not readily scalable. Although PBNPs can be produced in suspension CHO cells (Supplementary Fig. 15), the impact of producer cell type on particle quality, RNA cargo composition, and delivery efficiency remains to be systematically evaluated. Addressing these factors will be important for establishing scalable manufacturing of ePBNPs. In addition, current purification by ultracentrifugation is labor-intensive and not amenable to large-scale applications, highlighting the need to develop chromatography-based approaches to improve yield, throughput and reproducibility.

In summary, this study addresses key limitations of PEG10-based RNA delivery by establishing a production strategy and engineering ePBNPs for enhanced cancer cell targeting and immunotherapy. These data support the potential of ePBNPs as a platform for RNA-based cancer immunotherapy.

Methods

Ethics statement

Animal experiments were performed in compliance with the Guide for the Care and Use of Laboratory Animals (National Science and Technology Council, Taiwan), with the approval of the National Tsing Hua University Institutional Animal Care and Use Committee (IACUC protocol number NTHU-11203H008).

Preparation of plasmids and recombinant BV

To construct recombinant BVs for PBNPs production, the human peg10 gene driven by CMV promoter (CMV-PEG10) and the VSVG gene driven by CMV promoter (CMV-VSVG) were PCR-amplified from pCMV-PEG10 (#174859, Addgene) and pMD2.G (#12259, Addgene), respectively. CMV-PEG10 and CMV-VSVG fragments were subcloned together into pFastBac^TM Dual (Thermo Fisher) by Gibson Assembly, yielding a BV donor plasmid pBac-PEG10-VSVG. The fragment encoding the 5’ UTR and the first 500 bp of 3’ UTR of peg10 was PCR-amplified from pCMV-Hs.cargoCas9 (#174864, Addgene) and subcloned into pFastBac^TM Dual. The egfp fragment was subsequently PCR-amplified from pEGFP-N1 and inserted between the 5’ and 3’ UTRs, yielding pBac-cEGFP (cargo length 1720 bp). The donor plasmids pBac-PEG10-VSVG and pBac-cEGFP were used to generate the corresponding BV vectors Bac-PEG10-VSVG and Bac-cEGFP, respectively, following the instructions of the Bac-to-Bac^TM system (ThermoFisher).

To generate BV with a larger cargo size, the mCherry gene was amplified from pJZC78 (#62339, Addgene) and fused with the egfp fragment with an intervening P2A peptide by NEBuilder^® HiFi DNA Assembly (New England Biolabs). The fusion egfp-P2A-mCherry was inserted into pBac-cEGFP to replace egfp, yielding pBac-cEM (cargo length 2536 bp). Similarly, the firefly luciferase (Luc) gene was amplified from the TA cloning vector (with the identical sequences as pCMV-FLuc (#170575, Addgene) and fused with a P2A sequence, followed by insertion into pBac-cEM to yield pBac-cEMLuc (cargo length 4264 bp). Similarly, we amplified a fusion gene consisting of dLbCas12 and p65-HSF1 from pEF1a-dLbCpf1-A1.1²². The dLbCas12-p65-HSF1 fusion gene was fused with a P2A sequence and inserted into pBac-cEM at the 3’ end of EGFP-P2A-mCherry to form pBac-cEMdLb-pH (cargo length 7336 bp). In addition, the CMV-PEG10 fragment and CMV-VSVG fragment were separately subcloned into pFastBac^TM Dual to generate pBac-PEG10 and pBac-VSVG, respectively. The EGFR-specific scFv flanked by a 5’ signal peptide (SP) and His₆ tag and a 3’ GS linker was chemically synthesized (Genomics, Taiwan) and inserted into pBac-VSVG for fusion to the N-terminus of VSVG, yielding pBac-eVSVG. To construct the plasmid encoding OX40L and IL-12, DNA fragment of OX40L were chemically synthesized and Il12 was PCR-amplified from pmIL12-N1 (#123139, Addgene), were fused with intervening P2A and cloned into pmIL12-N1 to generate pmOX40L-P2A-IL12 (OI) using Gibson assembly. The OI fragment was cloned into pBac-cEGFP to replace egfp, yielding pBac-cOI. All these pBac donor plasmids were used to generate the corresponding BV using the Bac-to-Bac^TM system.

All BV vectors were amplified by infecting insect cells (Sf-9) with titers typically ranging from 10⁸-10⁹ pfu/ml. BV vectors were titrated by the endpoint dilution method²¹. The multiplicity of infection (MOI) was defined as pfu/cell.

Cell culture

Sf-9 cells (Gibco, 11496015) were routinely cultured in SF900 II serum-free medium. HEK293T cells (ATCC, CRL-3216) were routinely cultured in DMEM medium containing 10% fetal bovine serum (FBS). CT26 (mouse colon carcinoma (ATCC, CRL-2638)) and 4T1 (breast cancer (ATCC, CRL-2539)) cells were cultured in the RPMI 1640 medium supplied with 10% FBS. U87 (glioblastoma (ATCC, HTB-14)), MC38 (colon adenocarcinoma (abm, T8291)), Hepa1-6 (hepatoma (ATCC, CRL-1830)), Huh7 (hepatocellular carcinoma (abm, T8973)) and Hep3B (hepatocellular carcinoma (ATCC, HB-8064)) cells were cultured in DMEM containing 10% FBS.

PBNPs production by plasmid transfection and BV transduction

For each PBNPs production batch, HEK293Tproducer cells were cultured in ten 15 cm dishes (≈ 2.0 × 10⁷ cells/dish). For production by plasmid transfection, pBac-PEG10-VSVG (25,000 ng/dish) and pBac-cEGFP (25,000 ng/dish) were mixed with Lipofectamine^TM 3000 (Thermo Fisher) in 3 ml Opti-MEM^TM medium (Thermo Fisher) and added to HEK293T cells in 17 ml DMEM (20 ml/dish). The cells were cultured for 48 h for PBNPs secretion.

Alternatively, PBNPs were produced by BV co-transduction of HEK293T cells as described²¹ with minor modifications. Depending on the MOI and BV titer, a certain amount of BV solution was mixed with NaHCO₃-free DMEM at a volumetric ratio of 1:4 and added to the HEK293T cells (≈ 2.0 × 10⁷ cells/dish), followed by gentle mixing on a rocking plate (10 times/min) at room temperature. After 6 h, the mixed BV solution was decanted, and 20 ml/dish DMEM medium containing 3 mM sodium butyrate was added. After 18 h incubation, the cells were washed and then cultured in DMEM without sodium butyrate for 24 h for PBNPs secretion.

PBNPs concentration and purification

The supernatants (≈200 ml from 10 dishes per batch) were harvested (2000 × g, 10 min) and filtered through 0.45 μm. For each ultracentrifugation tube, approximately 30–32 ml of samples were loaded to the top of 3 mL sucrose (20%) cushion and ultracentrifuged (120,000 × g for 2 h) at 4 °C. The pellets from each tube were resuspended into 350 μl phosphate-buffered saline (PBS) and stored at 4 °C.

Further purification was performed following the protocols reported previously¹⁶. Specifically, 5 batches of resuspended PBNPs were subjected to density gradient ultracentrifugation using different concentrations of OptiPrep^TM Density Gradient Medium (iodixanol solution, Sigma-Aldrich). The iodixanol solution was diluted with PBS-MK buffer (PBS containing 1 mM MgCl₂ and 2.5 mM KCl) to concentrations of 40%, 20%, 10%, 5% and loaded to the tubes layer by layer. After loading the concentrated PBNPs to the top, the tubes were ultracentrifuged (200,000×g for 3 h) at 4 °C. Fractions in the gradient layers were collected (1 ml/fraction), desalted with centrifugal filters and subjected to Western blot to determine which fractions contained PEG10. The fractions containing PEG10 were ultracentrifuged again (120,000 × g for 2 h) at 4 °C and resuspended in PBS (pH 7.4).

PBNPs transfection of cells

For PBNPs transfection, cells were seeded prior to the day of transfection. The volume of PBNPs required was calculated according to the following formula: V = (N × D)/T, where V = volume of required PBNPs, N = number of cells at time of transfection, D = cargo copies per cell, T = genome titer (cargo mRNA copies/μL). The required PBNPs were mixed with DMEM medium containing 6 μg/ml DEAE-dextran and added to the cells, followed by 5 h incubation at 37 °C. After transfection, the supernatant was removed and replaced by fresh medium. The cells were cultured for another 48 h at 37 °C and imaged with a fluorescence microscope (Eclipse Ti2, Nikon). Alternatively, the cells were analyzed by a flow cytometer (Guava^® easyCyte™, Luminex) to determine the percentage of fluorescing cells (GFP⁺ or mCherry⁺) analysis. The data were analyzed using guavaSoft™ (3.3) (Luminex).

Determination of PBNPs functional titer and genome titer

The functional titer of PBNPs was determined as described⁶³. HEK293T cells were seeded to 24-well plates (2 × 10⁵ cells/well) and cultured in DMEM medium for 1 day. The PBNPs samples were diluted to different concentrations using the DMEM medium containing 6 μg/ml DEAE-dextran and added to the cells (200 μl/well), followed by 5 h incubation at 37 °C. After transfection, the supernatant was removed and replaced by 500 μl DMEM medium. The cells were cultured for another 24 h at 37 °C, washed with 1 ml PBS and trypsinized for flow cytometry analysis. The functional titers were determined using a formula as described⁶³ and are expressed as transfecting units (TU)/ml (Supplementary Fig. 1).

The genome titers of PBNPs were determined by qRT-PCR as described^14,15 and are expressed as mRNA copy numbers/μl. Briefly, PBNPs resuspended in PBS were first treated with the DNA and RNA endonuclease (Micrococcal Nuclease, M0247S, New England Biolabs) for 2 h at 37 °C, followed by the addition of 2 μl RiboLock RNA inhibitor (40 μ/μl, Thermo Fisher). The RNAs in the PBNPs were extracted by QIAamp Viral RNA Mini Kit (Qiagen) and converted to cDNA using High-Capacity cDNA Reverse Transcription kit (Thermo Fisher). Cargo cDNA and 6 diluted standard plasmids containing the same gene cassette (1 × 10³-1 × 10⁸ copies/μl) were subjected to qPCR using the SyGreen^® Blue Mix (qPCRBIO) on LightCycler^®96 (Roche) with primers specific to the mRNA (the cargo gene, 5’ peg10 UTR or 3’ peg10 UTR) (Supplementary Data 2). qPCR was performed with the following program setting: 95 °C for the first 3 min, 45 cycles of 10 s at 95 °C, 5 s at 58 °C and 10 s at 72 °C, 10 s. Calibration curves were established from the standard plasmids data and were used to calculate the genome titers of PBNPs. The data were analyzed using LightCycler^® 96 SW 1.1 software (Roche). The primer sequences used for genomic titer in this study are listed in Supplementary Data 2.

Characterization of PBNPs

After nuclease treatment, total RNA was extracted using the QIAamp Viral RNA kit for denaturing RNA gel electrophoresis. The RNA was also reverse transcribed to cDNA using the High-Capacity cDNA Reverse Transcription kit. The cDNA of the cargo egfp (cEGFP) was PCR-amplified using different primer sets specific for 5’ UTR, egfp and 3’ UTR. pBac-cEGFP was used as a control plasmid for the same PCR. The amplicons were analyzed by DNA gel electrophoresis. All primer sequences used for DNA electrophoresis in this study are listed in Supplementary Data 2.

For RNA sequencing, the purified PBNPs were first treated with the nuclease, and the RNAs in the PBNPs were extracted by Direct-zol RNA Microprep Kit (Zymo Research, R2061) and treated with DNase I (Zymo Research, E1011). The purified RNA was used for the preparation of the sequencing library by TruSeq Stranded mRNA Library Prep Kit (Illumina). Briefly, mRNA was purified from total RNA (1 μg) by oligo(dT)-coupled magnetic beads and fragmented into small pieces under elevated temperature. The first-strand cDNA was synthesized using reverse transcriptase and random primers. After the generation of double-strand cDNA and adenylation on 3’ ends of DNA fragments, the adapters were ligated. The products were enriched with PCR and purified with the AMPure XP system (Beckman Colter, Beverly, USA). The libraries were qualified by the Qsep400 System (Bioptic Inc.) and quantified by Qubit 2.0 Fluorometer (Thermo Fisher). The qualified libraries were then sequenced on an Illumina NovaSeq platform with 150 bp paired-end reads (Genomics BioSci & Tech, Taiwan).

RNA sequencing data were processed using a standard pipeline. Raw reads were quality filtered and trimmed using fastp, with quality assessed by FastQC. Clean reads were aligned to the reference genome using HISAT2 and processed with SAMtools. Gene expression was quantified using featureCounts and normalized as transcripts per million (TPM) to account for gene length and sequencing depth, enabling comparison of relative transcript abundance across samples (Supplementary Data 1).

The PBNPs samples were subjected to Bradford Protein Assay for protein concentrations and then analyzed by 12% SDS-PAGE gel and Western blot following standard procedures⁶⁴. The primary antibodies included anti-PEG10 (rabbit, 1:1000, Cell Signaling), anti-VSVG (rabbit, 1:1000, Abcam), anti-GP64 (mouse, 1:1000, Abcam), anti-His₆ (mouse, 1:1000, Abcam) and anti-GAPDH (rabbit, 1:2000, Abcam). The secondary antibodies included goat anti-rabbit or anti-mouse IgG (1:10000, Abcam) conjugated with HRP. The membranes were developed with Clarity Western ECL substrate (Bio-Rad) and visualized with LightCyclerHR scanner (Syngene).

The morphology of PBNPs samples was imaged by transmission electron microscopy (TEM) as described¹⁶ with minor modifications. The PBNPs samples were adsorbed onto Formvar/carbon-coated copper grids (TED Pella Inc.), washed 3 times by deionized water and negatively stained with 2% phosphotungstic acid (Sigma). After drying at 80 °C overnight, the copper grids were observed by TEM (Talos F200C, Thermo Fisher). The presence of eVSVG on ePBNPs was verified by immunogold TEM as described¹⁶, using anti-His₆ Mab as the primary antibody.

The zeta potential and size distribution of PBNPs were detected by dynamic light scattering (DLS) following the manufacturer’s instructions (Zetasizer, Malvern Panalytical).

Co-culture of transfected CT26 cells with splenocytes

The CT26 cells were seeded to the lower chamber (2×10⁵ cells/well) of 24-well transwell plates (Corning) and mock-transfected or transfected with the PBNPs at 10 cargo copies/cell. After 12 h, the cells were washed with PBS. In parallel, the splenocytes were harvested from the spleens of 8-week-old BALB/c mice using the gentleMACS™ Dissociator (Miltenyi Biotech), cultured in RPMI 1640 medium containing 10% FBS and seeded to the upper chamber of transwell plates (2 × 10⁵ cells/well). The splenocyte-containing chambers were transferred to the transwell plates seeded with the transfected CT26 cells and were stimulated with Dynabeads™ Mouse T-Activator CD3/CD28 for T-Cell Expansion and Activation (Thermo Fisher).

ELISA (Enzyme-linked immunosorbent assay)

The co-culture supernatants were collected and centrifuged at 300 × g for 10 min to remove cell debris. The concentrations of mouse IL-12, INF-γ and TNF-α were measured using the kits for each cytokine: ELISA MAX™ Deluxe Set (Biolegend) for Mouse IL-12 (p70), IFN-γ or TNF-α. OX40L was measured by Mouse TNFSF4 (OX40L) ELISA Kit (Invitrogen).

T cell response

The co-cultured splenocytes were washed with PBS and placed onto the microscope (Nikon ECLIPSE Ti2) to observe the T cell proliferation. Images were captured at 3 days post-coculture. The cells were also dissociated from the upper chambers and immunostained with 3 antibodies (Biolegend): (1) FITC anti-CD3 (mouse, 1:1000, #100203), (2) PE anti-CD4 (mouse, 1:1000, #100407) or PerCP/Cyanine5.5 anti-CD8a (mouse, 1:1000, #100733) and (3) FITC anti-CD69 (mouse, 1:1000, #104505). Following staining, the cells were analyzed using a flow cytometer (Guava^® easyCyte™, CYTEK). The gating was performed as follows: FSC-Area and SSC-Area were applied to identify splenocytes, and singlet cells were further gated on FSC-Height and FSC-Area (Supplementary Fig. 16). The data were analyzed using guavaSoft™ (3.3) (Luminex).

Tumor model and animal experiments

Six-week-old female BALB/c mice (BioLASCO) were anesthetized with isoflurane (Abbott) delivered in oxygen (2 L/min) using an induction chamber (1–5% isoflurane) followed by maintenance anesthesia via a nose cone (2–4% isoflurane). Adequate anesthesia was confirmed by loss of the pedal withdrawal reflex before procedures were performed. Mouse CT26 cells were resuspended in Opti-MEM^TM (5 × 10⁵ cells/ml) and subcutaneously injected (200 μL) into the back of the thigh. The width (W) and length (L) of the tumors were measured with a digital caliper (Mitutoyo), and the volume (V) was calculated by the formula: V (mm³) = L × W²× (π/6). When the tumor volume reached ≈ 50 mm³ (day 0), the mice were randomly divided into 4 groups for intratumoral injections (n = 6 for all groups). The PBS group received four doses of PBS (100 μl/dose) at days 1, 3, 5 and 7. The OXA group received one dose of oxaliplatin (3 mg/kg mice) at day 1. The ePBNP.OI group received four doses of the engineered PBNPs (ePBNP.OI) encoding IL-12 and OX40L (1 × 10⁷ cargo copies/dose) at days 1, 3, 5 and 7. The Combo group received one dose of oxaliplatin at day 1 to prime the immune response and subsequently received four doses of ePBNP.OI (1 × 10⁷ cargo copies/dose) at days 3, 5, 7 and 9. The mice were considered dead and sacrificed for ethical issues when the tumor size exceeded 20 mm in any direction or the tumor volume exceeded 1500 mm³. Complete response (CR) was defined as the disappearance of measurable tumor mass ( < 100 mm³) for at least one time point, according to the IACUC standard on solid tumors in mice. The sera were taken from the mice at 24 h after complete therapy and analyzed by ELISA for TNF-α and IFN-γ.

To determine biodistribution, tumor-bearing mice received a single intratumoral injection of ePBNP.OI and were euthanized 24 h post-injection. Approximately 50–100 mg of each major tissue, including the heart, lung, liver, kidney, spleen and tumor, was harvested and snap-frozen in liquid nitrogen. Frozen tissues were homogenized using disposable pestles, and total RNA was extracted with TRIzol™ (Invitrogen). cDNA was synthesized from 500 ng of total RNA using the High-Capacity cDNA Reverse Transcription Kit. Subsequent qPCR was performed using diluted cDNA (10 ng/μl), and 8 diluted standard plasmids containing OX40L and IL-12 cassettes (10¹-10⁸ copies/μl) were subjected to qPCR on LightCycler^®96 with the target-specific primers (Supplementary Data. 2). Calibration curves plotted according to serial diluted standards and measured Ct values were used to quantify the mRNA copy number in each tissue. The data were analyzed using LightCycler^® 96 SW 1.1 software (Roche). The primer sequences used for qPCR in this study are listed in Supplementary Data 2.

To assess the toxicity, serum samples were collected from healthy mice (control group) and from the Combo group 24 h after the endpoint. Levels of liver and kidney function markers, including aspartate aminotransferase (AST), alanine aminotransferase (ALT), and blood urea nitrogen (BUN), were quantified using a clinical chemistry analyzer (Hitachi). All measurements were performed at the National Center for Biomodels (Taiwan). In parallel, major organs (liver, kidneys, heart, lungs, and spleen) were excised, fixed in 4% paraformaldehyde, embedded in paraffin, sectioned, and stained with hematoxylin and eosin (H&E) for histopathological analysis.

Statistics & reproducibility

No statistical method was used to predetermine sample size. No data was excluded from the analysis. Mice were randomly assigned to experimental groups where applicable. The investigators were not blinded to allocation during experiments and outcome assessment. Statistical analyses were performed using GraphPad Prism (9.5.0). Data are presented as mean±SD from at least three independent experiments unless otherwise indicated. Detailed statistical information, including statistical tests, exact adjusted p-values, and sample sizes, is provided in the corresponding figure legends and Source Data files.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

P.Y.C. and T.L.T. conceptualized the project, designed and performed experiments and wrote the paper. V.A.T., Q.T.D, N.T.K.N., and P.H.C. performed experiments. Y.C.H. conceptualized and supervised the project, designed experiments, acquired funding and wrote the paper.

Peer review

Peer review information

Nature Communications thanks João Conde and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

The authors further acknowledge the financial support from the National Science and Technology Council (NSTC 114-2223-E-007-013, 113-2223-E-007-010, 112-2223-E-007-002, 112-2314-B-007-004-MY3) and National Health Research Institutes (NHRI-EX115-11329EI, EX114-11329EI, EX113-11329EI), Taiwan.

Data availability

All data supporting the results of this study are available within the paper, its Supplementary Information, and the Source Data files. Next-generation sequencing data generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE331505 and are currently private pending public release. Source data are provided in this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74352-x.