Engineered BCG selectively triggers trained immunity in tumor-associated macrophages and sensitizes glioblastoma to radiotherapy in mice

Abstract

Radiotherapy plays a crucial role in antitumor immunity in glioblastoma, yet its efficacy is often limited, resulting in tumor recurrence. Here, we engineer a macrophage membrane-camouflaged Bacillus Calmette-Guérin (BCG) via bioorthogonal chemistry to enhance radiotherapy against glioblastoma. This engineered BCG penetrates the blood-brain barrier, targets tumors, and alleviates hypoxia through intrinsic catalase activity, exerting antitumor effects in both murine orthotopic glioblastoma and humanized mouse models. Notably, it also initiates trained immunity in tumor-associated macrophages. Depletion and adoptive transfer of tumor-associated macrophages demonstrate that trained immunity promotes inflammatory cytokine production, reactive oxygen species release, phagocytosis and the recruitment of CD8⁺ T cells, ultimately amplifying immune responses to radiotherapy. Moreover, immune checkpoint blockade further augments the antitumor efficacy of engineered BCG combined with radiotherapy. Here, we show that trained immunity in tumor-associated macrophages is a promising strategy to sensitize glioblastoma to radiotherapy and improve treatment outcomes.

Radiotherapy (RT) is standard-of-care in cancer management; however, RT efficacy remains limited. Here, the authors test whether membrane camouflaged BCG bacteria (MBCG) enhance response to RT in preclinical models of glioblastoma. MBCG efficiently targets tumor tissues, induces trained immunity in tumor-associated macrophages, and enhances the RT-induced anti-tumor responses.

Introduction

Radiotherapy (RT) remains a mainstay of cancer treatment, yet its clinical efficacy is often compromised by the immunosuppressive tumor microenvironment (TME) and hypoxia-induced radioresistance¹. RT can induce immunogenic cell death and activate systemic antitumor immune responses through the cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) signaling pathway, thereby recruiting and priming CD4⁺ and CD8⁺ T cells^2–4. Unfortunately, these antitumor effects are frequently insufficient to eradicate tumors. Hypoxia and the accumulation of immunosuppressive cells, including myeloid-derived suppressor cells and M2-type macrophages, along with elevated immune checkpoint expression, severely impair RT-mediated immune activation⁵. Notably, clinical and preclinical studies have demonstrated that restoring tumor oxygenation and alleviating TME immunosuppression can substantially enhance the therapeutic efficacy of RT⁶. Despite its widespread use in glioblastoma (GBM) treatment, the efficacy of RT remains limited by the immunologically “cold” TME, highlighting the urgent need for strategies that simultaneously alleviate hypoxia and enhance RT-induced antitumor immunity.

Trained immunity, a de facto type of innate immune memory, is characterized by the heightened responsiveness of innate immune cells upon secondary stimulation^7,8. Mechanistically, initial stimuli such as β-glucan or microbial components reprogram innate immune cells through metabolic and epigenetic modifications, including increased H3K4me3 at proinflammatory gene promoters, thereby enabling robust cytokine production upon re-exposure^9,10. In addition, trained immunity can enhance macrophage phagocytosis and amplify subsequent innate and adaptive immune responses against pathogens or tumors^11–13. Our previous study revealed that inducing trained immunity in tumor-associated macrophages (TAM) sensitized murine lung cancer to immune checkpoint blockade¹⁴. However, whether targeting trained immunity in TAMs can be harnessed to potentiate RT-induced antitumor immune responses in GBM remains unknown.

Bacillus Calmette-Guérin (BCG), a live attenuated strain of Mycobacterium bovis and an approved immunotherapeutic agent for non-muscle-invasive bladder cancer, is a well-established inducer of trained immunity^15–17. Recent studies indicate that BCG induces trained immunity by engaging nucleotide-binding oligomerization domain-containing protein 2 (NOD2), activating the Akt-mTOR signaling pathway and triggering epigenetic and metabolic reprogramming, which enhances innate immune responses to secondary stimuli^18–20. However, whether BCG-induced trained immunity can be harnessed to amplify RT-mediated antitumor effects, particularly within the immunosuppressive context of GBM, remains unexplored.

Despite its immunostimulatory properties, the clinical application of naked BCG for GBM faces substantial challenges, including poor tumor targeting, limited blood-brain barrier (BBB) penetration, and systemic immunogenicity. Biomimetic delivery strategies based on cell membranes hold promise in improving biocompatibility, prolonging circulation, and enhancing tumor targeting^21–23. In particular, macrophage membranes have been utilized to coat several types of nanoparticles, including poly(lactic-co-glycolic acid) (PLGA) and black phosphorus, for the targeted therapy of murine orthotopic GBM^24–26. Nevertheless, maintaining membrane stability remains a major hurdle for clinical translation, as membranes are prone to shedding from the core structure.

In this work, we engineer a macrophage membrane-camouflaged BCG (MBCG) using bioorthogonal chemistry to generate a stable core-shell biologic for efficient delivery and immune modulation in GBM (Fig. 1). We find that MBCG induces trained immunity in TAMs and enhances the response of GBM to RT in both murine orthotopic and humanized models, with additional benefit in combination with immune checkpoint blockade. Collectively, these findings identify trained immunity in TAMs as a promising approach for radiosensitization in GBM. More broadly, this work underscores the therapeutic potential of innate immune reprogramming in cancer treatment.

Fig. 1. Schematic illustration of the preparation of macrophage-membrane camouflaged BCG (MBCG) and its mechanism for sensitizing glioblastoma to radiotherapy.

A bioorthogonal reaction mediates the preparation of macrophage-membrane encapsulated BCG (MBCG). After systemic administration, MBCG penetrates the blood-brain barrier (BBB) and targets tumor-associated macrophages (TAM) to initiate trained immunity, thereby reprogramming tumor microenvironment to sensitize glioblastoma radiotherapy. N3-DAA, 3-Azido-D-alanine hydrochloride. DSPE-PEG-DBCO, 1,2-distearoyl-sn-glycero-3-phosphoethanolamine-poly(ethylene glycol)-dibenzocyclooctyne. MDP, muramyl dipeptide. DAMPs, damage associated molecular patterns. NOD2, Nucleotide binding oligomerization domain-containing 2. cGAS, Cyclic GMP-AMP Synthase. cGAMP, Cyclic GMP-AMP. STING, stimulator of interferon genes.

Results

Preparation and characterization of MBCG

MBCG was prepared via a bioorthogonal reaction between DBCO and azide (N₃), to improve the biological stability and tumor-targeting capacity of BCG (Fig. 2a). The successful introduction of azide groups on BCG cell walls was confirmed by incubation with DBCO-Cy5, followed by confocal laser scanning microscopy (CLSM) imaging. As shown in Fig. 2b, the azide-modified BCG (BCG_N3) labeled with DiO clearly colocalized with DBCO-Cy5, in contrast to unmodified BCG. Furthermore, the macrophage membranes pre-modified with DSPE-PEG-DBCO exhibited greater colocalization with the BCG_N3 to form MBCG than membranes directly encapsulated onto BCG (M@BCG) under ultrasound conditions (Fig. 2c and Supplementary Fig. 1a), indicating that the bioorthogonal reaction facilitates more stable membrane encapsulation. To assess the efficiency and reproducibility of membrane coating, we performed flow cytometry to quantify the bioorthogonal conjugation efficiency (Supplementary Fig. 1b–e). The conjugation efficiency reached up to ~91.3% and remained consistent across three independent experimental batches. Dynamic light scattering (DLS) revealed that the average diameter of the MBCG was approximately 1.27 μm, which was slightly larger than that of naked BCG_N3 (Fig. 2d, e). Furthermore, the click chemistry-mediated membrane coating conferred improved stability. As shown in Fig. 2f and Supplementary Fig. 1f–i, the diameter of the MBCG remained stable (~1.2 μm) over 72 h in both phosphate-buffered saline (PBS) and 10% fetal bovine serum (FBS), whereas M@BCG progressively increased in size, possibly due to membrane detachment from the bacterial surface. Additionally, the freeze-dried formulation of MBCG exhibited excellent redispersibility, maintaining a particle diameter comparable to that of freshly prepared MBCG (Fig. 2g, h). Notably, the bioorthogonal conjugation process did not compromise BCG viability or its proliferative capacity upon reconstitution, as confirmed by in vitro culture experiments (Fig. 2i and Supplementary Fig. 1j). Finally, transmission electron microscopy (TEM) and scanning electron microscopy (SEM) images revealed altered surface features, exhibiting a roughened exterior compared with naked BCG (Fig. 2j, k).

Fig. 2. Preparation and characterization of MBCG.

a Schematic illustration of MBCG preparation via bioorthogonal click chemistry between azide-functionalized BCG (BCG_N3) and DBCO-modified macrophage membranes derived from RAW 264.7 cells. b Representative confocal laser scanning microscopy (CLSM) images of DiO-labeled BCG and BCG_N3 after incubation with DBCO-Cy5 for 6 h at 37 °C. DiO, green; DBCO-Cy5, magenta. Scale bars, 1 µm. c Representative CLSM images of macrophage membrane-coated BCG prepared by different methods. M@BCG indicates BCG coated with macrophage membranes by ultrasonication alone, whereas MBCG was prepared by bioorthogonal coupling between BCG_N3 and DBCO-modified macrophage membranes. DiO-labeled BCG, green; DiD-labeled macrophage membranes, magenta. Scale bars, 1 μm. d, e Size distribution of BCG_N3 (d) and MBCG (e) in PBS measured by dynamic light scattering (DLS). f Time-dependent size changes of M@BCG and MBCG in 10% fetal bovine serum (FBS) at 37 °C over 72 h, measured by DLS (n = 4 biologically independent preparations per group). Data are presented as mean ± SD. Statistical significance was assessed using two-way repeated-measures ANOVA with Geisser-Greenhouse correction, followed by Sidak’s multiple-comparisons test. Adjusted P value for the comparison at 72 h is indicated. g Photographs of lyophilized MBCG and reconstituted MBCG in PBS. Scale bars, 1 cm. h Size distribution of reconstituted MBCG in PBS measured by DLS. i Growth curves of lyophilized BCG and MBCG after reconstitution in 7H9 broth medium at 37 °C, monitored by OD₆₀₀ over 15 days (n = 3 biologically independent cultures per group). j Representative transmission electron microscopy (TEM) images of BCG and MBCG. Scale bars, 0.5 µm. k Representative scanning electron microscopy (SEM) images of BCG and MBCG showing surface morphology. Scale bars, 0.5 μm. CLSM images (b, c), DLS profiles (d, e, h), TEM (j), and SEM (k) images are representative of three biologically independent preparations. Source data are provided as a Source Data file.

BBB penetration and tumor-targeting capacity of the MBCG

Although macrophage membranes have been widely used to coat nanoparticles for the treatment of murine orthotopic glioma, it remained unclear whether MBCG, with a diameter greater than 1000 nm, could cross the BBB and target tumor tissues. We therefore first established an in vitro BBB model to evaluate the ability of MBCG to penetrate the BBB (Fig. 3a). Compared with naked BCG, flow cytometric analysis showed that MBCG penetrated the bEnd.3 monolayer in the upper chamber more efficiently and was more readily endocytosed by TAMs seeded in the lower chamber (Fig. 3b, c). In addition, the membrane modification enhanced macrophage uptake of BCG. As shown in Fig. 3d and Supplementary Fig. 2a–c, TAMs exhibited significantly enhanced uptake of MBCG compared to naked BCG, as reflected by higher endocytosis rates and fluorescence intensity over time. We next evaluated the in vivo BBB penetration and tumor-targeting capacity of MBCG in a murine orthotopic GBM model. The tumor-bearing mice were randomly divided into two groups and intravenously injected with DiR-labeled naked BCG or MBCG. Biodistribution was then monitored at the indicated time points using the IVIS imaging system. As shown in Fig. 3e, f, fluorescence was observed in the brains of the MBCG-treated mice at 6 h post injection and gradually increased until 12 h, whereas no signal was detected in the brains of mice treated with naked BCG. More importantly, ex vivo imaging of major organs demonstrated that DiR-labeled MBCG clearly colocalized with the tumor tissues, as indicated by bioluminescence (Fig. 3g, h). The presence of MBCG in both tumor and brain tissues was further validated by ex vivo colony- forming units (CFU) assays (Supplementary Fig. 2d, e). Notably, MBCG was also distributed in the liver, spleen, and lungs, likely owing to physiological uptake by the mononuclear phagocyte system and transient pulmonary retention (Fig. 3g). We further assessed the biocompatibility of MBCG by measuring proinflammatory cytokine levels in blood and normal brain tissues. Although serum IL-6 and TNF levels increased at 24 h after injection, they returned to baseline by 48 h. In addition, qPCR analysis and H&E staining revealed no evident cytotoxicity in normal brain regions following MBCG administration (Supplementary Fig. 2f–l).

Fig. 3. MBCG penetrates the BBB, targets tumor tissues, and is preferentially endocytosed by TAMs.

a Schematic illustration of an in vitro blood-brain barrier (BBB) model. bEnd.3 endothelial cells were seeded in the upper chamber and tumor-associated macrophages (TAM) in the lower chamber. b, c Flow cytometric analysis of TAM uptake in the lower chamber after addition of DiO-labeled BCG or MBCG (2 × 10⁴ CFU/mL) to the upper chamber and incubating for 12 h at 37 °C. Representative histogram (b) and quantification of endocytosis as the percentage of DiO⁺ TAMs (c). d Time-course analysis of TAM uptake following direct exposure to DiO-labeled BCG or MBCG (2 × 10⁵ CFU/mL). Uptake from 0–12 h was quantified as the percentage of DiO⁺ TAMs by flow cytometry (n = 3 biologically independent experiments). e, f In vivo imaging system (IVIS) analysis of orthotopic GL261-luc tumor bearing mice after intravenous injection of DiR-labeled BCG or MBCG (1 × 10⁶ CFU in 100 µL PBS; DiR, 10 µM). Representative IVIS images (e) and quantification of region-of-interest (ROI) signals over 12 h (f) (n = 6 mice per group). g Representative ex vivo DiR fluorescence imaging of tumors and major organs at 12 h post injection (n = 6 mice per group). h Quantification of DiR fluorescence and tumor bioluminescence signals in tumor tissues (n = 6 mice per group). i Representative flow cytometric gating strategy for identifying DiR⁺ BCG within tumor-infiltrating leukocytes and TAMs (corresponding to j). Debris was excluded by FSC/SSC gating; leukocytes were defined as CD45⁺ cells, and TAMs as CD45⁺ CD11b⁺ F4/80⁺ cells. j Quantification of DiR⁺ cells within leukocyte and TAM populations at 12 h post injection (n = 6 mice per group). Data are presented as mean ± SD. Statistical significance was assessed using one-way ANOVA with Tukey’s multiple-comparisons test (c), two-way repeated-measures ANOVA with Geisser-Greenhouse correction followed by Sidak’s multiple-comparisons test (f; adjusted P value at 12 h is shown), and two-sided unpaired Welch’s t-test (h). P values are indicated. Flow cytometry plots (b, d, i) and IVIS images (e, g) are representative of three biologically independent experiments. Source data are provided as a Source Data file.

The intratumoral cellular distribution of MBCG was further analyzed by flow cytometry. As shown in Fig. 3i, j, nearly all MBCG were distributed in CD45⁺ leukocytes, indicating preferential uptake by immune cells. Among these cells, approximately 21.2% of DiR⁺ CD45⁺ CD11b⁺ F4/80⁺ cells were identified as TAMs, suggesting that TAMs represent a major cellular target of MBCG. In addition to TAMs, MBCG was also internalized by other immune subsets within the tumor microenvironment. Flow cytometric analysis showed that approximately 2.8% of CD11c⁺ dendritic cells (DC) and 11.5% of CD11b⁺ Ly6G⁺ neutrophils were DiR⁺ (Supplementary Fig. 3a, b). Moreover, MBCG treatment increased the proportion of mature DCs (CD11c⁺ MHCII⁺) and activated neutrophils (CD11b⁺ Ly6G⁺ CD62L⁺) in tumor tissues (Supplementary Fig. 3c–e).

MBCG mediates the amelioration of the hypoxic TME in glioblastoma

As a mycobacterium, BCG naturally expresses catalase, which may contribute to ameliorating the hypoxic TME, thereby enhancing the antitumor efficacy of RT. The hypoxic TME of GBM was first assessed using data from TCGA and GTEx. As shown in Fig. 4a, hypoxia-inducible factor-1α (HIF-1α), a hallmark of the hypoxic TME, was highly expressed in gliomas, including low-grade glioma (LGG) and GBM. In addition, the TISIDB and CIBERSORT analysis showed that the expression of pro-inflammatory genes, as well as tumor-infiltrating immune cells, including CD8⁺ T and NK cells, were negatively correlated with HIF-1α expression, suggesting that hypoxia may be associated with reduced antitumor immune activation (Fig. 4b, c). We therefore investigated the effect of MBCG on H₂O₂ decomposition by assessing its catalytic activity (Fig. 4d). A reduction in H₂O₂ was detected in both H₂O₂ solutions and murine GBM tissues after incubation with MBCG, indicating its catalytic activity toward H₂O₂ (Fig. 4e, f and Supplementary Fig. 4). Application of an O₂-sensitive fluorescence probe confirmed that MBCG could convert H₂O₂ into O₂, whereas heat-inactivated MBCG (iMBCG) did not release O₂ (Fig. 4g). These in vitro measurements of H₂O₂ degradation and O₂ generation indicated that MBCG had the potential to ameliorate the hypoxic TME in vivo.

Fig. 4. MBCG alleviates hypoxia by decomposing intratumoral H₂O₂.

a RNA-seq analysis of HIF-1α expression in GBM (n = 166), LGG (n = 523), and normal brain tissues (n = 1136) using TCGA and GTEx datasets. b Spearman correlation coefficients (ρ) between HIF-1α and inflammation-related genes in glioma, stratified by grade (anti-inflammatory, red; pro-inflammatory, blue). c Spearman correlation between HIF-1α expression and tumor-infiltrating immune cell abundance inferred using CIBERSORT. P values are provided in the Source Data file. d Schematic illustration of catalase-mediated decomposition of H₂O₂ into O₂ by MBCG. e In vitro degradation of H₂O₂ (100 μM) by MBCG (5 × 10⁵ CFU/mL) at 25 °C; PBS and heat-inactivated MBCG (iMBCG) served as controls (n = 3 biologically independent experiments). Adjusted P value (iMBCG vs. MBCG) at 20 min is shown. f Intratumoral H₂O₂ levels in GL261 tumors 24 h after intravenous administration of PBS, iMBCG or MBCG (1 × 10⁶ CFU in 100 µL PBS) (n = 6 mice per group). g O₂ generation detected with an O₂-sensitive probe during reaction of MBCG with H₂O₂ (100 μM); representative images (insets) and ROI quantification (n = 3 biologically independent experiments). Scale bars, 2 mm. h Experimental timeline for in vivo hypoxia evaluation. i, j Photoacoustic imaging of tumor oxygen saturation (sO₂) 24 h after intravenous administration of PBS, iMBCG, or MBCG (1 × 10⁶ CFU in 100 μL PBS). Dashed circles indicate tumors (i; black, O₂-free; blue, O₂ present; red, oxygen-rich). Quantification is shown in (j) (n = 6 mice per group). Scale bars, 4 mm. k HIF-1α immunofluorescence staining in tumors 72 h after intravenous administration of PBS, iMBCG, or MBCG (1 × 10⁶ CFU in 100 μL). Blue, DAPI; red, HIF-1α; scale bars, 100 μm. Data are presented as mean ± SD. Statistical analyses: Welch’s one-way ANOVA with Games-Howell test (a), two-sided Spearman correlation (b, c), two-way repeated-measures ANOVA with Geisser-Greenhouse correction and Sidak’s test (e), and ordinary one-way ANOVA with Tukey’s test (f, g, j). P values are indicated. Images in (g, i, and k) are representative. All in vivo experiments were independently repeated three times with similar results. Source data are provided as a Source Data file.

Accordingly, a murine orthotopic GBM model was used to investigate the in vivo catalytic activity of MBCG toward H₂O₂ within tumor tissues (Fig. 4h). Compared with those in the PBS and iMBCG groups, the orthotopic tumors in the mice intravenously injected with MBCG presented significantly attenuated hypoxia after 24 h of treatment (Fig. 4i, j), suggesting improved oxygenation within tumors. In parallel experiments, the expression of HIF-1α in tumor slices collected from mice after the injection of different BCG formulations for 72 h was examined by immunofluorescence staining. The results demonstrated that MBCG markedly decreased the expression of HIF-1α compared with that in the control and iMBCG groups (Fig. 4k). Thus, MBCG exhibited potential to alleviate tumor hypoxia by catalytically converting H₂O₂ into O₂. This capability may be leveraged to enhance the therapeutic efficacy of RT.

MBCG induces trained immunity in TAMs in vitro and in vivo

MBCG penetrated the BBB, targeted tumor tissues, and was endocytosed by TAMs. Consequently, trained immunity in TAMs isolated from murine GBM tissues might be initiated by MBCG to magnify RT-induced immune activation (Fig. 5a). Transcriptomic profiling of TAMs treated with MBCG compared with untreated controls revealed significant activation of inflammatory and innate immune signaling pathways, including the NOD-like and MAPK pathways, as shown by GO and KEGG enrichment analyses, together with a gene expression heatmap (Fig. 5b–d). Western blot analysis confirmed elevated phosphorylation of AKT, TAK1, and p38 in MBCG-treated TAMs compared with PBS or BCG controls (Supplementary Fig. 5a), supporting the activation of key nodes associated with trained immunity. Accordingly, the in vitro training effect of MBCG on TAMs collected from subcutaneous murine glioblastoma tissues was evaluated. The results indicated that MBCG training significantly increased IL-6 and TNF expression as well as ROS production following LPS restimulation, whereas the NOD2 inhibitor GSK717 reversed these effects (Fig. 5e–g). Notably, MBCG-treated TAMs exhibited increased phagocytic ability toward tumor cells, which was dependent on the NOD2 signaling pathway as determined by CLSM and flow cytometric assays (Fig. 5h–j). These findings indicate that MBCG induces trained immunity in TAMs.

Fig. 5. MBCG induces NOD2-dependent trained immunity in TAMs.

a Schematic illustration of MBCG-induced trained immunity and enhanced responses to radiotherapy treated tumor cells. b–d Transcriptomic profiling of TAMs treated with PBS or MBCG (2 × 10⁵ CFU/mL) for 12 h (n = 3 biologically independent samples per group). Gene Ontology enrichment analysis of differentially expressed genes (DEGs) (b), KEGG pathway enrichment analysis of DEGs (c), and heatmap of selected inflammatory and innate immune DEGs (row Z-scores; red, higher; blue, lower expression) (d). e, f IL-6 (e) and TNF (f) mRNA expression in TAMs subjected to a trained immunity protocol. TAMs were pretreated ± MBCG (2 × 10⁵ CFU/mL) and the NOD2 inhibitor GSK717 (5 μg/mL) for 24 h, rested for 48 h, then restimulated with LPS (100 ng/mL) for 12 h (n = 3 biologically independent experiments). g Reactive oxygen species production measured by DCFH-DA following the same protocol as in (e, f) (n = 3 biologically independent experiments). h–j Phagocytosis of GL261-GFP cells by TAMs pretreated as described in (e, f). TAMs were co-cultured with GL261-GFP cells for 12 h and analyzed by confocal microscopy (h) and flow cytometry (i), with quantification (j) (n = 3 biologically independent experiments). Images in (h) are representative of three biologically independent experiments. Magenta, DiD-labeled TAM membranes; green, GFP⁺ tumor cells; blue, DAPI; scale bars, 10 μm. k, l qPCR analysis of IFN-α (k) and IFN-β (l) expression in TAMs treated with or without MBCG and restimulated with RT-exposed GL261-conditioned medium for 12 h (n = 3 biologically independent experiments). m, n qPCR analysis of IL-6 (m) and TNF (n) expression in TAMs following the same restimulation protocol as in (k, l) (n = 3 biologically independent experiments). Data are presented as mean ± SD. Enrichment P values in (b, c) were calculated using two-sided Fisher’s exact test with multiple testing correction. Statistical significance in (e, f and j–n) was determined using ordinary one-way ANOVA with Tukey’s multiple-comparisons test. P values are indicated. Experiments were independently repeated three times with similar results. Source data are provided as a Source Data file.

To investigate whether trained immunity could reinforce the antitumor efficacy of RT against GBM, GL261 cells were subjected to various RT doses, all of which significantly inhibited cell growth (Supplementary Fig. 5b–f). Based on these results and previous studies employing hypofractionated RT to elicit immune responses within limited experimental windows^27–29, a dose of 4 Gy per fraction was selected for subsequent in vitro and in vivo experiments. Notably, the combination of MBCG with RT-conditioned GL261 culture medium resulted in enhanced activation of the STING pathway (Fig. 5k, l and Supplementary Fig. 5g), which may facilitate immune activation within the TME. Furthermore, the in vivo training effect of MBCG on TAMs was also evaluated in GL261 tumor-bearing mice. TAMs isolated from MBCG-treated mice exhibited increased responsiveness to restimulation by RT-exposed GL261 medium, as indicated by increased expression of IL-6 and TNF, which was reversed by co-administration of a NOD2 inhibitor (Fig. 5m, n). These results demonstrate that MBCG efficiently induces trained immunity in TAMs in vitro and in vivo, and augments RT-associated immune activation.

To further functionally validate whether MBCG-induced trained immunity enhances macrophage-mediated anti-tumor activity, we co-cultured MBCG-trained TAMs with GL261 cells with or without RT exposure. MBCG-trained TAMs exhibited significantly increased cytotoxicity and elevated reactive oxygen species (ROS) production in response to irradiated tumor cells (Supplementary Fig. 5h–m). Furthermore, the expression of proinflammatory cytokines, including IL-6 and TNF, was markedly upregulated (Supplementary Fig. 5n, o), consistent with a trained proinflammatory phenotype. These effects were abrogated by the NOD2 inhibitor GSK717, indicating that the MBCG-induced training effect was NOD2-dependent (Supplementary Fig. 5p, q). Collectively, these findings further support that MBCG-trained TAMs not only acquire molecular characteristics of trained immunity but also exhibit enhanced tumoricidal function and immune-stimulatory capacity.

The combined antitumor activity of MBCG and RT against glioblastoma

MBCG targeted tumor tissues to ameliorate the hypoxic TME and induced trained immunity in TAMs, which enhanced RT-associated immune activation. Therefore, a murine orthotopic GBM model was established to evaluate the combined antitumor activity of MBCG with RT (Fig. 6a). Tumor growth was monitored by bioluminescence imaging, and the bioluminescence signal intensity of the tumor tissues was quantified (Fig. 6b, c). The results demonstrated that three RT cycles significantly decreased tumor growth compared with that in the PBS and MBCG groups. In contrast, the combination of MBCG and RT produced the greatest reduction in tumor burden among the groups, indicating that MBCG effectively potentiated the antitumor activity of RT. Consistent with the results of tumor growth monitoring, hematoxylin and eosin (H&E) tumor pathology evaluations and the survival times of the mice also confirmed that the combination of MBCG with RT had a superior antitumor effect on murine GBM (Fig. 6d, e). In addition, the effects of these treatments on the body weight of tumor-bearing mice were monitored for 12 days, demonstrating that these treatments did not induce significant systemic toxicity (Supplementary Fig. 6a).

Fig. 6. In vivo antitumor efficacy of MBCG combined with radiotherapy (RT).

a Experimental design. C57BL/6 mice were intracranially inoculated with GL261-Luc cells (1 × 10⁵ cells per mouse) on day −5, treated intravenously with MBCG (1 × 10⁶ CFU in 100 μL PBS) on days 0, 6, and 12, and received focal brain irradiation (4 Gy per fraction) on days 3, 9, and 15. b, c Tumor growth monitored by bioluminescence imaging (b) and quantified by ROI analysis (c) (n = 5 mice per group). Adjusted P values at day 16 are shown (MBCG vs. RT; MBCG + RT vs. RT). d Representative H&E-stained brain sections from mice receiving the indicated treatments at day 16 (n = 5 mice per group). Images are representative of three biologically independent in vivo experiments. Scale bars, 1 mm. e Kaplan–Meier survival curves of mice receiving different treatments (n = 5 mice per group). P values for the indicated pairwise comparisons are shown. f–j qPCR analysis of IL-6, TNF, IFN-γ, IL-12β, CXCL9 mRNA expression in tumor tissues collected after the indicated treatments (n = 6 mice per group). k–o Flow cytometric analysis of tumor-infiltrating immune populations, including macrophages (k, CD45⁺ CD11b⁺ F4/80⁺), TNF⁺ M1-like macrophages (l, CD86⁺ CD45⁺ CD11b⁺ F4/80⁺ TNF⁺), CD8⁺ T cells (m, CD45⁺ CD3⁺ CD8⁺), IFN-γ⁺ CD8⁺ cytotoxic T cells (n, CD45⁺ CD3⁺ CD8⁺ IFN-γ⁺), and IFN-γ⁺ CD4⁺ T cells (o, CD45⁺ CD3⁺ CD4⁺ IFN-γ⁺) (n = 6 mice per group). Data are presented as mean ± SD. Statistical significance was assessed using two-way repeated-measures ANOVA with Geisser-Greenhouse correction, followed by Sidak’s multiple-comparisons test (c), log-rank (Mantel-Cox) test (e), and ordinary one-way ANOVA with Tukey’s multiple-comparisons test (f–o). For in vivo experiments, each data point represents an individual mouse, and experiments were independently repeated three times with similar results. Source data are provided as a Source Data file.

Unexpectedly, MBCG alone did not exhibit significant antitumor efficacy, which might be attributed to the fact that GBM is a typical “cold” tumor and that MBCG monotherapy was insufficient to effectively impede tumor growth. Further analysis of the tumor immune microenvironment was performed using qPCR and flow cytometric assay. As shown in Fig. 6f–j, combining MBCG with RT resulted in the highest expression levels of inflammatory factors including IL-6 and TNF. Moreover, immunomodulation-related factors such as IFN-γ, IL-12β and CXCL9 were also significantly induced by the combination of MBCG and RT. Accordingly, the distribution of immune cells within tumor tissues was analyzed by a flow cytometric assay (Fig. 6k–o, and Supplementary Fig. 6b). Although RT treatment alone did not significantly increase the proportion of TNF⁺ M1 macrophages, its combination with MBCG markedly elevated this population (Fig. 6k, l). In contrast, MBCG monotherapy did not significantly alter the frequency of IFN-γ⁺ CD8⁺ or CD4⁺ T cells compared with PBS, potentially explaining its limited antitumor efficacy. However, the combination of MBCG and RT significantly enhanced IFN-γ production in CD8⁺ and CD4⁺ T cells, indicating enhanced adaptive immune activation (Fig. 6m–o).

To further assess the generalizability and clinical relevance of MBCG, we employed an additional immunocompetent orthotopic GBM model using the poorly immunogenic CT2A-Luc cell line. Tumor-bearing mice were treated with MBCG in combination with either a preclinical hypofractionated RT regimen (4 Gy per fraction) or a clinically relevant schedule (2 Gy per fraction). The results demonstrated that MBCG combined with 2 Gy per fraction RT significantly suppressed tumor growth and prolonged survival compared to monotherapies, achieving therapeutic effects comparable to those of the higher RT dose (Supplementary Fig. 7a–e). Furthermore, flow cytometric analysis revealed increased infiltration of M1-like macrophages (CD45⁺ CD11b⁺ F4/80⁺ CD86⁺) and CD8⁺ T cells in tumors treated with MBCG plus 4 Gy per fraction RT (Supplementary Fig. 7f–h). Importantly, the combination therapy did not induce a significant increase in immunosuppressive cell populations compared to RT alone (Supplementary Fig. 7i–k). In addition, the enhanced antitumor efficacy and immune activation of MBCG in combination with RT were recapitulated in male mice bearing orthotopic CT2A-Luc tumors (Supplementary Fig. 8a–f). These findings highlight the reproducibility and translational potential of combining MBCG with RT for GBM treatment.

MBCG enhances the antitumor immunity of RT in a manner dependent on TAM training

To determine the contribution of immune cells to the combined antitumor effects of MBCG and RT, CD8⁺ T cells, NK cells and TAMs were depleted in a murine orthotopic GBM model, and a NOD2 inhibitor (GSK717) was used to reverse the induction of trained immunity (Fig. 7a). Flow cytometric analysis confirmed effective depletion of TAMs, CD8⁺ T cells, and NK cells (Fig. 7b–d). Tumor growth was examined by bioluminescence imaging, and the bioluminescence intensity of the tumor tissues was quantified as indicated in Fig. 7e and Supplementary Fig. 9a. The depletion of CD8⁺ T and NK cells reversed the augmented antitumor effect of MBCG combined with RT, indicating that these two cell types contribute to the tumor-killing effect of RT-induced immune activation^30,31. As anticipated, macrophage elimination and NOD2 inhibition also markedly reduced the antitumor efficacy of this combination therapy, suggesting that trained immunity by TAMs plays an important role in the combined antitumoral activity of MBCG with RT. Notably, the bioluminescence intensity in the NOD2 inhibition group was even higher than that in the PBS group, indicating that NOD2 activation might play a pivotal role in inhibiting GBM development.

Fig. 7. MBCG sensitizes murine glioblastoma to radiotherapy through trained immunity of TAMs.

a Experimental timeline for immune cell depletion or NOD2 inhibition before MBCG and RT treatment in orthotopic GL261-Luc tumor-bearing mice. Macrophages were depleted by intravenous clodronate liposomes (200 μL, 5 mg/mL). CD8⁺ T cells and NK cells were depleted using anti-CD8 or anti-NK1.1 antibodies (100 μg per mouse). b–d Flow cytometric assessment of depletion efficiency for macrophages (b, CD45⁺ CD11b⁺ F4/80⁺), CD8⁺ T cells (c, CD45⁺ CD3⁺ CD8⁺), and NK cells (d, CD45⁺ CD3⁻ NK1.1⁺) (n = 6 mice per group). e Representative bioluminescence imaging of GL261-Luc tumors from mice receiving the indicated treatments (n = 5 mice per group). f Experimental timeline for evaluating the antitumor effect of adoptive transfer of MBCG-trained macrophages followed by RT (4 Gy per fraction). g, h Bioluminescence imaging (g) and ROI quantification (h) of tumor burden after adoptive transfer with or without RT (n = 5 mice per group). Adjusted P values at day 9 are shown. i–m qPCR analysis of IL-6, TNF, IFN-γ, IL-12β, and CXCL9 mRNA expression in tumors collected on day 10 (n = 6 mice per group). n–r Flow cytometric analysis of the immune cell distribution in tumor tissue with different treatments: macrophages (n, CD45⁺ CD11b⁺ F4/80⁺), CD8⁺ T cells (o, CD45⁺ CD3⁺ CD8⁺), CD4⁺ T cells (p, CD45⁺ CD3⁺ CD4⁺), CD8⁺/CD4⁺ T cell ratio (q), and NK cells (r, CD45⁺ CD3^- NK1.1⁺) (n = 6 mice per group). Mice were intracranially inoculated with GL261-Luc cells (1 × 10⁵) on day −5, treated with MBCG (1 × 10⁶ CFU in 100 µL PBS) on days 0 and 6, and subjected to RT (4 Gy per fraction) on days 3 and 9. Data are presented as mean ± SD. Statistical analyses: two-sided unpaired Welch’s t-test (b–d), two-way repeated-measures ANOVA with Geisser-Greenhouse correction followed by Sidak’s test (h), Welch’s one-way ANOVA with Games-Howell test (i–o, q), and ordinary one-way ANOVA with Tukey’s test (p, r). Adjusted P values are indicated. For in vivo experiments, each data point represents one mouse, and experiments were independently repeated three times with similar results. Source data are provided as a Source Data file.

To further explore whether the augmented antitumor effects of MBCG combined with RT depended on TAM trained immunity, the experiments involving macrophage depletion and adoptive transfer of MBCG-trained TAMs were performed (Fig. 7f). The results showed that macrophage depletion accelerated tumor growth, and a single RT treatment or adoptive transfer of in vivo MBCG-trained TAMs resulted in moderate antitumor activity in macrophage-depleted mice (Fig. 7g, h). However, tumor growth in macrophage-depleted mice was significantly inhibited when the mice received adoptive transfer of MBCG-trained TAMs followed by RT treatment in vivo. Furthermore, the immune microenvironment of tumor tissues was analyzed by determining the expression of cytokines and the extent of immune cell infiltration. As shown in Fig. 7i–m, intratumoral expression of multiple immune-activating cytokines and chemokines, including IL-6, TNF, IFN-γ, IL-12β, and CXCL9, was markedly reduced following macrophage depletion. Radiotherapy (RT) alone failed to restore the decreased expression induced by TAM elimination. In contrast, adoptive transfer of MBCG-trained TAMs significantly increased the expression of these immune-activating mediators, an effect that was further potentiated when combined with RT.

In addition, flow cytometric assay for immune cells within tumor tissues was performed, and the results revealed that TAM elimination significantly decreased the infiltration of CD8⁺ T and NK cells. However, this effect was reversed through the adoptive transfer of MBCG-trained TAMs (Fig. 7n–r and Supplementary Fig. 9b). Together, these results demonstrate that MBCG boosts the antitumor immunity of RT through its ability to induce trained immunity in TAMs, which plays a crucial role in RT-mediated immune activation, thereby facilitating immune cell infiltration and augmenting RT efficacy in GBM treatment.

Immune checkpoint inhibition further improves the combined antitumor efficacy of MBCG and RT in vivo

The combination of MBCG and RT exhibited significant antitumor effects in a murine glioblastoma model. The therapeutic efficacy of this combination was further evaluated in a humanized LN229 orthotopic mouse model (Fig. 8a). The combination of MBCG with RT did not exert antitumor effects in NOD/SCID/IL2rγNull (NSG) mice alone, but significantly suppressed tumor growth in mice receiving adoptive transfer of human peripheral blood mononuclear cells (PBMC). To characterize the immune composition of PBMCs, flow cytometric analysis prior to transfer confirmed that approximately 27.2% of CD45⁺ cells were CD3⁺ T cells (Supplementary Fig. 10a, b). Moreover, flow cytometric analysis of tumor-infiltrating immune cells in PBMC-engrafted mice confirmed the presence of human CD45⁺ leukocytes within the tumor microenvironment, including CD3⁺ T cells comprising both CD8⁺ and CD4⁺ subsets (Supplementary Fig. 10c, d). These data showed that tumor growth in LN229 tumor-bearing mice receiving adoptive transfer of PBMCs was significantly inhibited by the combination treatment of MBCG with RT, while RT monotherapy only moderately inhibited tumor growth (Fig. 8b–d). H&E staining of brain tissues also revealed that combining MBCG with RT produced the greatest tumor suppression in the mice that received PBMCs (Fig. 8e). Furthermore, the qPCR results demonstrated that the combination treatment of MBCG with RT significantly increased the expression of human IL-6 and IFN-γ compared with MBCG or RT monotherapy (Fig. 8f, g).

Fig. 8. MBCG enhances radiotherapy efficacy in humanized GBM models and synergizes with immune checkpoint blockade.

a Experimental timeline for evaluating MBCG plus radiotherapy (RT) in a humanized orthotopic glioblastoma model. NSG mice were intracranially inoculated with LN229-Luc cells (1 × 10⁵ cells per mouse) on day −5 and received intravenous adoptive transfer of 1 × 10⁵ human PBMCs on days −1, 4, and 8. MBCG (1 × 10⁶ CFU in 100 μL PBS) was administered on days 0, 5, and 9, and RT (4 Gy per fraction) on days 3, 7, and 12. b–d Representative bioluminescence imaging of LN229-Luc tumors (b), individual mouse ROI quantification (c), and group-level ROI analysis from day 0 to day 12 (d, n = 5 mice per group). Tumor imaging was performed every 4 days. Adjusted P values at day 12 are shown. e Representative H&E-stained brain sections collected on day 13. Images are representative of three biologically independent in vivo experiments. Scale bars, 1 mm. f, g qPCR analysis of human IL-6 (f) and IFN-γ (g) transcripts in tumor tissues collected from humanized GBM-bearing mice after the indicated treatments (n = 6 mice per group). h Experimental timeline for triple therapy with MBCG, RT and immune checkpoint inhibitors (anti-PD1 and anti-CD47, 100 μg per mouse, i.v.) in orthotopic GL261-Luc tumor-bearing mice. i–k Representative bioluminescence imaging of GL261-Luc tumors (i), individual mouse ROI quantification from day 0 to 30 (j), and group-level ROI analysis from day 0 to day 12 (k, n = 5 mice per group). Tumor imaging was performed every 3 days. Adjusted P values at day 12 are shown. l Kaplan–Meier survival curves of GL261-Luc tumor-bearing mice receiving the indicated treatments (n = 8 mice per group). Data are presented as mean ± SD. Statistical analyses: Two-way repeated-measures ANOVA with Geisser-Greenhouse correction followed by Sidak’s multiple-comparisons test (d, k), ordinary one-way ANOVA with Tukey’s multiple-comparisons test (f), Welch’s one-way ANOVA with Games-Howell test (g), or log-rank (Mantel-Cox) test (l). For in vivo experiments, each data point represents one mouse, and experiments were independently repeated three times with similar results. Source data are provided as a Source Data file.

The enhanced antitumor effect was primarily attributed to the amelioration of the hypoxic TME and the immune augmentation associated with trained immunity in TAMs. However, the superior antitumor efficacy of combining MBCG with RT was accompanied by increased expression of immune checkpoints, including PD-1, PD-L1, and CD47, as shown in Supplementary Fig. 11a–c, which might impair the combined antitumor activity. To further substantiate these observations, we performed qRT-PCR analysis and immunohistochemical staining of tumor tissues. Quantitative PCR analysis confirmed the significant upregulation of PD-1, CTLA-4, CD274, and CD47 mRNA levels in tumor tissues following MBCG combined with RT treatment (Supplementary Fig. 11d–g), and the IHC results also revealed consistently increased expression of PD-L1 and CD47 in the MBCG + RT group compared with other treatment groups (Supplementary Fig. 11h). These findings collectively support the notion that combination therapy induces adaptive immune resistance through checkpoint molecule upregulation, thereby providing a mechanistic rationale for incorporating immune checkpoint inhibitors (ICI). Consequently, ICIs targeting CD47 and PD-1 may cooperate with MBCG and RT to further hinder the development of GBM. A murine GBM model was used to evaluate the enhanced therapeutic effect of ICIs combined with MBCG and RT (Fig. 8h). Tumor growth in vivo was monitored by bioluminescence imaging and quantification. As shown in Fig. 8i, j, k, ICI monotherapy resulted in only moderate antitumor activity, which might be attributed to the low expression of CD47 and PD-L1 in the “cold” TME. In contrast, ICIs combined with MBCG and RT strongly inhibited tumor growth compared with the other groups. Additionally, the bioluminescence intensity in the mice that received triple combination therapy on days 20 to 30 almost completely disappeared. Notably, seven mice in the PBS group died of the disease between 17 and 26 days, whereas no mice in the triple combination therapy group died, even after more than 40 days (Fig. 8l). These results indicated that MBCG could serve as a sensitizer to augment the antitumor activity of RT in both murine and human GBM, and that this effect could be further potentiated by ICIs, highlighting its potential for clinical application in tumor immunotherapy.

To further validate the robustness and translational relevance of this strategy, we performed a delayed intervention experiment in which treatments were initiated on day 10 after tumor implantation, representing a more advanced disease stage. The combination of MBCG and RT continued to significantly inhibit tumor growth compared with control and monotherapy groups, and this effect was further enhanced by the addition of ICIs (Supplementary Fig. 11i, j). In addition, combining MBCG with either dose of radiotherapy significantly improved tumor control in the CT2A model lacking luciferase expression, as assessed by T2-weighted MRI (Supplementary Fig. 12a–c), compared with monotherapies or PBS, thereby excluding potential reporter-associated immunogenic effects. Together, these findings indicate that MBCG in combination with radiotherapy retains therapeutic efficacy across advanced stages of disease and across distinct glioma models, underscoring its potential therapeutic relevance in glioblastoma.

Discussion

Emerging evidence indicates that radiotherapy plays a pivotal role in activating systemic immune responses against tumors, in addition to its effects of directly killing tumor cells³². However, effective strategies to augment RT-triggered antitumor immune responses are currently lacking. In addition, a high radiation dose might impair immune activation and cause potential adverse effects, including structural damage to the brain, tumor recurrence and radioresistance, thereby leading to a poor prognosis of GBM²⁹. To address this challenge, an increasing number of strategies have been explored in combination with RT to amplify the antitumor immune response. For instance, ICIs such as PD-1, PD-L1, and CTLA-4 blockade have been widely explored in combination with RT. This combination can promote immune cell infiltration and activate cytotoxic CD8⁺ T cells, leading to a synergistic antitumor effect^33–35. Nevertheless, a major limitation is that only a minority of patients exhibit a durable response to ICIs in clinical trials. This limited efficacy may reflect the persistence of an immunologically “cold” TME that cannot be sufficiently reprogrammed by RT alone, thereby contributing to nonresponse to ICI therapy³⁶. Therefore, the development of efficient approaches that can be combined with RT to trigger an antitumor immune response and increase the responsiveness of GBM to ICIs remains an unmet need.

Trained immunity involves long-term epigenetic and metabolic reprogramming of innate immune cells in response to exogenous or endogenous stimuli. This rewiring can potentiate immune responses in these cells upon restimulation, thereby providing the host with long-term protective effects against pathogens or tumors¹¹. Consequently, agents that induce trained immunity might serve as radiosensitizers to boost RT-mediated immune activation for efficient tumor therapy. Microorganisms have emerged as promising tools in cancer therapy, and could be designed to target specific tumor markers, deliver therapeutic agents, or reprogram the TME³⁷. In this study, we engineered BCG with macrophage-membrane camouflage to improve tumor access in GBM and to induce trained immunity in TAMs, thereby augmenting RT-triggered antitumor immune responses and improving responsiveness to ICIs. BCG, which was originally developed as a vaccine for the prevention of tuberculosis, has been found to have potent nonspecific immunostimulatory effects in bladder cancer. One of the key mechanisms underlying the antitumoral activity of BCG might be the induction of trained immunity in innate immune cells such as monocytes, macrophages, and natural killer cells¹⁷. For example, BCG-trained macrophages undergo epigenetic and metabolic rewiring, resulting in a stronger ability to produce proinflammatory cytokines such as IL-6 and TNF, which are essential for recruiting and activating other immune cells within tumor tissues. Moreover, after BCG training, macrophages exhibit increased phagocytosis and tumor-killing capabilities, which could directly contribute to the elimination of tumor cells³⁸. Therefore, BCG might serve as an ideal inducer of trained immunity to synergize with RT to intervene in GBM development.

TAMs represent the largest component of infiltrating immune cells within the TME. These macrophages can suppress or promote tumor progression and are often associated with resistance to RT and ICIs, reflecting their complex roles in regulating the TME^39,40. Therefore, targeting TAMs may represent a promising avenue to potentiate the efficacy of immune activation mediated by RT. In addition, the hypoxic TME impairs the tumor-killing effect of RT and promotes the polarization of TAMs into immunosuppressive macrophages^41,42. BCG, in addition to inducing trained immunity, can convert hydrogen peroxide to oxygen via its intrinsic catalase activity, which could be utilized to ameliorate the hypoxic TME. Furthermore, BCG also exhibits an excellent safety profile in clinical practice and serves as an immunoadjuvant in bladder cancer treatment^15,16.

On the basis of the favorable characteristics of BCG, we devised a concise therapeutic strategy that leverages the engineered BCG-mediated TAM-targeted trained immunity and hypoxic TME alleviation to boost RT-induced antitumor immune responses for GBM treatment. While our initial experiments employed the GL261 model, we acknowledge its relatively high immunogenicity, and limited translational relevance to human GBM⁴³. To address this, we incorporated the CT2A model as a complementary immunocompetent system. CT2A tumors exhibit low immunogenicity and pronounced resistance to immunotherapy, while lacking IDH mutations, features that closely mirror the immunosuppressive microenvironment of glioblastoma. In addition, transcriptomic profiling indicates that CT2A aligns with the mesenchymal subtype of human GBM and was characterized by deficient antigen presentation, supporting its value as a clinically relevant model for evaluating immune-based therapeutic strategies^44–46. Furthermore, as sex-based differences in myeloid responses have been reported in GBM⁴⁷, we repeated key in vivo experiments in male mice. Similar therapeutic outcomes were observed, underscoring the robustness of our findings and addressing the relevance of sex differences in GBM immune responses.

Our findings demonstrated that the introduction of macrophage membranes via a bioorthogonal reaction endowed BCG with several advantageous properties, including a stable biomimetic “core-shell” structure, efficient BBB penetration, and TAM-targeting ability. These characteristics enhanced systemic antitumor immune responses to RT by efficiently inducing trained immunity in TAMs and alleviating the hypoxic TME. Further investigations showed that the ability of MBCG to increase RT-triggered antitumor immunity against GBM was reversed by TAM depletion or NOD2 inhibition. More importantly, this effect was restored by adoptive transfer of MBCG-trained TAMs, indicating that the enhanced antitumor activity of RT was dependent on the trained immunity of TAMs. Moreover, we systematically evaluated the therapeutic efficacy of MBCG combined with RT and ICIs, including αPD-1 and αCD47, for GBM treatment. This triple combination markedly enhanced antitumor efficacy without inducing discernible systemic toxicity, highlighting TAM trained immunity as a viable strategy to potentiate both RT and ICI responses. Notably, flow cytometry, qPCR, and IHC analyses revealed that MBCG combined with RT upregulated immune checkpoints such as PD-L1 and CD47 in the tumors, which attenuated immune efficacy but also supported the rationale for ICI coadministration. Despite pronounced macrophage and CD8⁺ T cell infiltration, no behavioral signs of elevated intracranial pressure, such as lethargy, ataxia, or delayed responsiveness, were observed. Survival monitoring beyond 40 days revealed no neurological abnormalities unrelated to the tumor burden. Although neuroinflammation has been reported with CAR-T and ICI-based therapies, macrophage-driven responses induced by MBCG are rarely linked to acute intracranial hypertension. Similarly, immune infiltration in glioma typically progresses gradually and seldom causes symptomatic pressure elevation, although careful monitoring is warranted in future translational applications.

While our study demonstrated the efficacy and immunomodulatory potential of MBCG in combination with RT, the intravenous administration of BCG warrants consideration for clinical translation. Unlike traditional intradermal or intravesical delivery⁴⁸, intravenous injection was employed to achieve systemic exposure and facilitate BBB penetration, which is critical for targeting intracranial GBM lesions. Although systemic BCG administration carries a known but rare risk of BCGitis⁴⁹, no signs of infection or toxicity were observed in our model. Future applications should explore safety-optimized formulations, including attenuated strains or brain-targeting ligands, to enhance therapeutic precision and minimize off-target effects.

In conclusion, targeting the trained immunity of TAMs by MBCG represents a novel and promising strategy to augment the RT-mediated antitumor immune response against GBM. The trained immunity of TAMs can increase phagocytosis of tumor cells, the release of ROS, and the expression of cytokines, thereby recruiting and activating antitumor immune cells. The underlying mechanisms likely involve metabolic and epigenetic rewiring of TAMs, as well as the activation of various proinflammatory signaling pathways. Future research is needed to elucidate these mechanisms and explore the clinical potential of MBCG-induced trained immunity for TAMs in combination with RT for GBM management.

Methods

Ethics statement

All animal experiments, including all surgical procedures, were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of China Pharmaceutical University (permit nos. 202409109 and 202510050). All procedures were performed in compliance with institutional guidelines and all relevant national regulations for the care and use of laboratory animals.

Human peripheral blood mononuclear cells were obtained from healthy adult volunteers. All procedures involving human participants were reviewed and approved by the Institutional Ethics Committee of Bayannur Hospital (Bayannur, China; approval no. BYYY-2024017). Written informed consent was obtained from all donors prior to sample collection, in accordance with the principles of the Declaration of Helsinki.

Materials

Dulbecco’s modified Eagle medium (DMEM, Cat. BC-M-005), Roswell Park Memorial Institute (RPMI) 1640 medium (Cat. BC-M-016), fetal bovine serum (FBS, Cat. BC-SE-FBS08), phosphate-buffered saline (Cat. BC-BPBS-01), and trypsin (Cat. BC-CE-005) were provided by Nanjing SenBeiJia Biological Technology Co., Ltd, China. Penicillin/streptomycin (Cat. C0222), 1,1′-Dioctadecyl-3,3,3′,3′-Tetramethylindodicarbocyanine,4-Chlorobenzenesulfonate Salt (DiD, Cat. C1995S), red blood cell lysis buffer (Cat. C3702), dyes of 4′,6-diamidino-2-phenylindole (DAPI, Cat. C1002), 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindotricarbocyanine iodide (DiR, Cat. Y239280), 3,3′-dioctadecyloxacarbocyanine perchlorate (DiO, Cat. C1038), D-luciferin potassium salt (Cat. Y023738) and LPS (Cat. ST1470) were obtained from Shanghai Beyotime Biotechnology Co., Ltd, China. In Vivo anti-mouse PD-1 (CD279) mAb was purchased from Selleck (Cat. A2122), USA. In Vivo anti-mouse CD47 mAb was obtained from Bioxcell (Cat. BE0283), USA. DSPE-PEG-DBCO was purchased from Xi’an Ruixi Biological Technology Co., Ltd. (Xi’an, China, Cat. R-DP-039-2k). 3-azido-D-alanine hydrochloride (Cat. HY-151642A) and GSK717 (Cat. HY-136555) were obtained from MedChemExpress, China. 2-amino-2-(hydroxymethyl)propane-1,3-diol (Tris) was obtained from Tianjin Ruijinte Chemicals Co., Ltd. (Cas. 77-86-1). Magnesium chloride anhydrous (MgCl₂, Cat. C01905294) was provided by Nanjing Reagent, China. Deoxyribonuclease I (DNase I, Cat. D8076) was obtained from Solarbio, China. Sucrose (Cas. 57-50-1) was provided by Xilong Scientific, China.

Cells lines and bacterial culture

RAW 264.7 (Cat. SCSP-5036) cell line was obtained from the Shanghai Cell Bank, China. GL261 (Cat. M8-0302) was purchased from OriCell®, Cyagen Biosciences Inc., Suzhou, China. L929 cell line (Cat. SRC-1023) was obtained from Saier Biotech. CT2A and CT2A-luc (luciferase-expressing) cell lines were obtained as a kind gift from ImmVira Co., Ltd. (Shenzhen, China). LN229 cell line was a kind gift from Prof. Can Zhang (China Pharmaceutical University, China). Primary tumor-associated macrophages (TAM) used in this study were isolated from GL261-bearing female C57BL/6 mice. Cell lines were cultured in complete DMEM supplemented with 10% FBS and 1% penicillin/streptomycin under humidified conditions equilibrated with 5% CO₂ at 37 °C. Mycobacterium bovis BCG (Bacillus Calmette-Guérin) was prepared by spreading M. bovis BCG from a vaccine stock vial (BCG for Therapeutic Use, Chengdu Biological Products, Sinopharm Approval No. S20123007) in Middlebrook 7H9 broth (Cat. LA7220, Solarbio, China,) supplemented with Middlebrook oleic albumin dextrose catalase enrichment liquid(OADC, Cat. LA9560, Solarbio, China) (10% v/v) and Tween 80 (0.05% v/v, Cat. C07805103, Nanjing Reagent, China) incubated at 37 °C. Bacterial growth density was monitored by measuring the OD₆₀₀ value using a microplate reader (Infinite F50, Tecan). Colony-forming units were determined by serial dilution of cultures on Middlebrook 7H10 agar (Cat. LA7230, Solarbio, China) plates and incubation at 37 °C for 4 weeks.

Animals

C57BL/6 mice (male/female, 6-week-old) and NOD/SCID/IL2rγNull (NSG) mice (female, 6-week-old) were provided by Hangzhou Ziyuan Laboratory Animal Science and Technology Co., Ltd, China (laboratory animal production license numbers: SCXK[Zhe] 2019-0004 and SCXK[Zhe] 2024-0004, Hangzhou, China). Female C57BL/6 mice were used for all primary experiments. Selected key experiments were independently replicated in male C57BL/6 mice, with consistent results (Supplementary Fig. 8). Male and female animals were not co-housed or mixed within experimental groups. Sex was therefore not included as a biological variable in statistical analyses. Animals were randomly assigned to experimental groups and housed in separate cages according to treatment groups. Experimental and control animals were not co-housed, and animals were not bred separately for experimental allocation. Equal numbers of mice were maintained per cage within each experiment. All mice were maintained in a specific pathogen-free (SPF) facility under controlled environmental conditions (21 ± 1 °C; 40–70% humidity; 12-h light/dark cycle) with ad libitum access to food and water.

All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of China Pharmaceutical University and conducted in accordance with institutional and national guidelines, with adherence to IACUC-approved humane endpoints and maximum tumor burden limits. For subcutaneous models, the maximum permissible tumor diameter (20 mm) was not exceeded. As orthotopic glioblastoma tumors were intracranial, tumor size could not be measured by calipers. Animals were monitored at least once daily for neurological symptoms and general health status. Humane endpoints included >20% body weight loss, severe neurological deficits (e.g., ataxia or impaired mobility), inability to access food or water, persistent lethargy, or other signs of distress. Mice exhibiting these signs in the presence of tumor burden and a body condition score ≤2 were euthanized. Euthanasia was performed by CO₂ inhalation followed by cervical dislocation to ensure death.

Cell membrane collection and DBCO-modified membrane preparation

The macrophage membrane was prepared according to the methods reported^26,50. Briefly, the RAW 264.7 cells were harvested and resuspended in ice-cold cell lysis buffer (containing 0.01 M Tris, 0.001 M MgCl₂, 2 mM PMSF, 10 μg/mL DNase, and 0.35 M sucrose; pH 7.4) at 4 °C for 30 min. The cells were destroyed in an ice bath using an ultrasonic instrument at 150 W for 3 min, followed by centrifugation (1500 × g) at 4 °C for 10 min. The supernatant was collected and vortexed again followed by centrifugation (9000 × g) at 4 °C for 20 min. After discarding the supernatant, the precipitate was washed three times with the lysis buffer and then dialyzed to remove the buffer salts. Finally, the purified macrophage membrane was stored as a freeze-dried precipitate. The DBCO-modified cell membrane was prepared by incubating the cell membrane (final concentration: 1 mg/mL) and DSPE-PEG-DBCO (final concentration: 1 mg/mL) at 37 °C for 1 h on a shaker at 200 rpm. Subsequently, dialysis was conducted for 2 h to remove free DSPE-PEG-DBCO.

Preparation of BCG_N3

Azide-functionalized Mycobacterium bovis BCG (BCG_N3) was generated using an azide-bearing D-alanine analog. Briefly, BCG was cultured in the presence of 3-azido-D-alanine hydrochloride at a final concentration of 100 μM at 37 °C with shaking at 200 rpm. After 48 h of incubation, the bacterial culture reached an OD₆₀₀ of approximately 0.6, corresponding to a density of ~1.0 × 10⁷ CFU/mL. Bacteria were then harvested by centrifugation at 6000 × g for 5 min, washed thoroughly with PBS, and resuspended in 1 mL PBS to remove excess precursor. The resulting bacteria, hereafter referred to as BCG_N3, possessed surface-accessible azide functionalities and were used directly for subsequent bioorthogonal strain-promoted azide-alkyne cycloaddition (SPAAC) with DBCO-modified membranes.

Preparation of MBCG

For the preparation of MBCG, the pre-prepared M-DBCO (final concentration, 3 mg/mL) was added to the BCG_N3 (final concentration, 1 × 10⁷ CFU/mL). The mixture was diluted fivefold with PBS and then incubated at 37 °C for 6 h with gentle shaking (200 rpm) to prepare the MBCG via the bioorthogonal reaction between DBCO and azides (N3). Finally, the mixture was washed with PBS and centrifuged at 6000 × g for 5 min to collect the MBCG. The obtained MBCG was resuspended in PBS and stored at 4 °C.

Characterization of BCG_N3 and MBCG

The generation of azide on the surface of BCG was confirmed by immunofluorescence assay. Briefly, the bacteria of natural BCG and pre-prepared BCG_N3 was stained with DiO (10 μM) and incubated with Cy5-coupled DBCO (DBCO-Cy5; Cat. BCD-32-1mg, Confluore, China) for 6 h at 37 °C on a shaker at 200 rpm (the final concentration of DBCO-Cy5: 50 µM), respectively. A dilution of the bacteria as mentioned above was spread on a cover slip, added with an anti-fluorescence quencher and detected by confocal laser scanning microscopy (CLSM, Olympus FV3000, Japan) to investigate the colocalization of BCG_N3 and DBCO-Cy5 depending on the successful introduction of N3 on the surface of BCG. Further, the fluorescence colocalization between DiO-stained BCG_N3 and DiD-labeled DBCO-modified or non-modified membrane was also detected by CLSM to explore the coating efficiency of cell membrane. The particle size distribution of BCG, BCG_N3, and MBCG was examined by DLS (Anton Paar Litesizer 500). The samples of BCG and MBCG were fixed with glutaraldehyde and then imaged by transmission electron microscope (TEM, JEM 2100) and scanning electron microscope (SEM, HITACHI SU8010), respectively.

Assessment of membrane conjugation via bioorthogonal reaction and BCG viability

To evaluate membrane conjugation efficiency, DiO-labeled BCG and DiD-labeled macrophage membranes were incubated under identical ultrasonication conditions to generate MBCG and M@BCG, respectively. After washing to remove unbound membranes, samples were analyzed using flow cytometry (CytoFLEX Beckman Coulter, USA). The BCG population was gated based on forward and side scatter (FSC/SSC), and the proportion of DiO⁺ DiD⁺ double-positive events was quantified to indicate membrane coating efficiency. Bioorthogonal conjugation efficiency was further calculated by comparing the MFI of DiD in different groups, using the following formula:

$$\mathrm{Bioorthogonal}\mathrm{conjugation}\mathrm{efficiency}\left(\%\right)=\left(1-\frac{{\mathrm{MFI}}_{\mathrm{Control}}}{{\mathrm{MFI}}_{\mathrm{Azide}+\mathrm{DBCO}}}\right)\times 100\%$$

Where MFI_Azide+DBCO represents the mean fluorescence intensity (MFI) of BCG particles after incubation with DiD-labeled, DBCO-modified membranes, MFI_Control refers to the background signal from BCG incubated with DiD-labeled, unmodified membranes (lacking DBCO). Membrane retention stability was assessed by analyzing the DiO⁺ DiD⁺ population at 48 h post-preparation in PBS. For viability analysis, BCG, MBCG, and heat-inactivated BCG (iBCG) were stained with propidium iodide (PI, 1 μg/mL; Cat. ST511, Beyotime, China) for 15 min at room temperature and analyzed by flow cytometry. PI-positive cells were defined as non-viable.

Extraction and preparation of mouse TAMs

Tumor tissues collected from euthanized GL261-bearing female C57BL/6 mice were cut into small pieces, and digested in collagenase type IV (Cat. BS165-1g, BioSharp) for 1 h at 37 °C. Single cell suspensions were obtained through a 70-μm basket filter (Cat. 352350, BD Biosciences). The obtained cells were incubated with red blood cell lysis buffer for 5 min at room temperature and then washed with PBS. The resuspended cells were mixed with 40% Percoll solution (Cat. P8370, Solarbio, China) and then carefully overlaid onto 80% Percoll solution. After centrifugation for 30 min at 750 × g, the cells were collected from the interlayer and then cultured in DMEM supplemented with 10% FBS, 20% L929 supernatant, and 1% penicillin-streptomycin solution. Four hours later, the culture medium was replaced with fresh medium to obtain macrophages.

Preparation of human PBMCs

PBMCs from blood donated by volunteers were isolated using a Human Whole Blood Mononuclear Cell Separation Solution (Cat. P9011, Solarbio, China) for density gradient separation. All the volunteers provided written informed consent. The fresh blood was diluted with PBS in a 1:1 ratio. Then, the diluted blood was carefully layered on Human Whole Blood Mononuclear Cell Separation Solution at a 1:1 ratio. The cells were obtained by centrifugation at 250 × g for 30 min, then the PBMC interphase was carefully collected and washed with PBS.

In vitro BBB model

A 24-well Transwell system (insert pore size, 3.0 μm) was utilized to establish an in vitro BBB model. Primary tumor-associated macrophages (1 × 10⁴ cells per well) were seeded in the lower chamber. bEnd.3 cells (1 × 10⁴ cells per insert) were seeded onto the upper inserts and cultured in DMEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin until a confluent monolayer was formed. DiO-labeled BCG_N3 or MBCG (final concentration of BCG: 2 × 10⁴ CFU/mL) was then added to the upper chamber. After incubation for 12 h, TAMs were collected from the lower chamber, and the uptake of BCG_N3 or MBCG was examined by CLSM and flow cytometry.

In vivo biodistribution of MBCG

After the establishment of orthotopic GBM model by the orthotopic injection of 7 μL of GL261-Luc cells (1 × 10⁵ cells per mouse) into female C57BL/6 mice, the mice were anesthetized with isoflurane and received an intravenous injection of MBCG after 5 days following the construction of mice model. The tumor growth and in vivo MBCG biodistribution were analyzed by a near-infrared fluorescence imaging system (IVIS® Lumina™, USA). Briefly, tumor-bearing mice were randomly assigned to two groups (n = 6 mice per group), and each group received a single tail vein injection of 100 μL PBS containing either DiR-labeled BCG or MBCG at an equivalent BCG dose of 1 × 10⁶ CFU per mouse. The fluorescence images were acquired at different time intervals (0, 2, 4, 6, 8, 10, and 12 h). At the end of the experiment, the mice were euthanized and their main organs including heart, liver, spleen, lung, kidney, and brain were harvested for ex vivo fluorescence imaging.

In vitro cellular uptake assay

The TAMs were seeded in a 12-well plate and incubated at 37 °C for 24 h. Then, PBS, MBCG, and MBCG + GSK717 were added to the cells, respectively, and incubated at 37 °C for 24 h and then replaced with fresh culture medium for 48 h (final concentration of BCG is 2 × 10⁵ CFU/mL, and final concentration of GSK717 is 5 μg/mL). Subsequently, TAMs were co-cultured with GL261-GFP tumor cells at 37 °C for 12 h. Then, the cells were washed twice with PBS followed by fixing with 4% of paraformaldehyde. The cell nucleus were stained with DAPI (5 μg/mL) for 10 min and the cell membrane was stained with DiD (10 μM) for 30 min, both at room temperature, and then observed using CLSM. In a parallel experiment, the ability of cellular uptake indicated by the fluorescence was further quantified by flow cytometry.

In vitro cellular ROS release examination

The TAMs (1.0 × 10⁵ cells/well) were seeded in a 12-well plate and incubated at 37 °C for 24 h. The cells were added with PBS, MBCG, and MBCG + GSK717, respectively, and incubated at 37 °C for 24 h (final concentration of BCG is 2 × 10⁵ CFU/mL; final concentration of GSK717 is 5 μg/mL), and then replaced with fresh culture medium resting for 48 h. The rested cells were restimulated with LPS (100 ng/mL) for 12 h and added with DCFH-DA (Cat. S1105S, Beyotime, China.) for 30 min, followed by Flow cytometric analysis.

Catalyzing H₂O₂ by MBCG in vitro and in vivo

The capacity of MBCG to catalyze hydrogen peroxide (H₂O₂) in vitro was detected by Micro Hydrogen Peroxide (H₂O₂) Assay Kit from Solarbio (Cat. BC3595, China). Briefly, at room temperature (25 °C), PBS, iMBCG (heat-inactivated MBCG), and MBCG (final concentration of BCG is 5 × 10⁵ CFU/mL) were, respectively, added to 100 μM of H₂O₂ solution at different time points (0, 5, 10, 15, 20 min), and the concentration of H₂O₂ was subsequently determined by Micro Hydrogen Peroxide assay. The capacity of MBCG to catalyze H₂O₂ in vivo was evaluated using an established orthotopic GBM model in female C57BL/6 mice. Mice were administered with PBS, iMBCG or MBCG via tail vein injection, respectively. After 24 h, the tumor tissues were excised and grinded, and subjected to H₂O₂ detection.

In vivo reversing hypoxia activity of MBCG

Following establishment of the orthotopic GBM model in female C57BL/6 mice, the mice were randomly divided into three groups: PBS, iMBCG, and MBCG. The treatments were started when the murine GBM model was successfully established as confirmed by the IVIS system. Representative photoacoustic images of tumor sites were obtained 24 h after intravenous injection of PBS, iMBCG, and MBCG (final BCG-equivalent dose of 1 × 10⁶ CFU in 100 µL PBS). Mice were euthanized 72 h after injection of PBS, iMBCG or MBCG, and tumor tissues were excised and analyzed by immunofluorescence for HIF-1α.

In vivo irradiation

All in vivo irradiation was performed using a Varian X-series clinical linear accelerator (Varian Medical Systems) delivering 6 MV X-rays at a dose rate of 400 cGy/min. Mice were anesthetized with isoflurane and positioned on a custom platform. Laser alignment combined with a fixed radiation window was employed to consistently target the cranial region without the use of CT-guided imaging. The irradiation field was precisely marked at the level of the skull. To minimize off-target exposure, the remainder of the mouse body was protected with lead-equivalent rubber shielding. Fractionation schedules, including 2 Gy or 4 Gy per fraction, were applied in this study.

In vitro irradiation

Cell cultures were irradiated using the same Varian X-series clinical linear accelerator under identical beam settings (6 MV X-rays, 400 cGy/min). Cells were placed in tissue culture plates without lids at room temperature during irradiation to ensure uniform dose delivery. Single radiation doses (e.g., 4 Gy or 6 Gy) were administered according to the experimental design specified in the figure legends. Irradiated cells were immediately returned to the incubator following treatment for further experimental analysis.

Western blot

Western blot assay was used to determine protein expression. The cells were collected and lysed in NP40 lysis buffer (Cat. P0013F, Beyotime, China) for 10 min on ice. Protein concentrations were quantified using a BCA protein assay kit (Elabscience, China, Cat. E-BC-K318-M). Then, equal amounts of protein were denatured, separated by SDS-PAGE, and transferred onto polyvinylidene fluoride (PVDF) membranes (Millipore, USA). The membranes were blocked with 5% skim milk powder for 1 h and then incubated with primary antibodies at 4 °C for 12 h. The antibodies used in this study were as follows: rabbit pAKT (Abways CY6569, 1:1000), rabbit pTAK1 (Abways CY9059, 1:1000), rabbit pIκBα (Beyotime AF5851, 1:1000), rabbit pP65 (Beyotime AN371, 1:1000), rabbit pP38 (Abways CY6391, 1:1000), rabbit cGAS (Cell Signaling 31659T, 1:1000), rabbit pSTING (Cell Signaling 72971T, 1:1000), rabbit STING (Cell Signaling 50494T, 1:1000), rabbit pTBK1 (Cell Signaling 5483T, 1:1000), rabbit TBK1 (Cell Signaling 3504 T, 1:1000), rabbit GAPDH (Abways AB0037, 1:1000). After incubation with the HRP-conjugated anti-rabbit IgG (Abways AB0101, 1:50,000), the protein blots were treated with chemiluminescent substrates (Vazyme, China) and then visualized using a chemiluminescent (ECL) detection apparatus (Clinx, China) to show the protein levels.

Flow cytometry and gating strategy

Single cells were incubated with mouse Fc receptor blocking reagent (Absin, China, Cat. Abs9477) for 20 min at 4 °C, washed and stained with fluorochrome-conjugated antibodies (Supplementary Table 2) for 30 min at 4 °C. Fluorescence data were collected using Cytoflex and analyzed employing a FlowJo software (version 10.8.1). After gating out debris using forward scatter (FSC) and side scatter (SSC), single cells were selected by FSC-H vs. FSC-A. Live cells were identified by excluding dead cells using a viability dye. CD45⁺ leukocytes were then gated to define immune cells. Immune checkpoint expression within tumor tissues was assessed in CD45⁺ immune cells by staining for and identifying CD47⁺, PD-L1⁺, and PD-1⁺ cell populations. Within the CD45⁺ population, CD11b⁺ F4/80⁺ macrophages, CD11b⁺ Gr1⁺ MDSCs, CD11c⁺ MHC-Ⅱ⁺ DCs, CD11b⁺ Ly6G⁺ CD62L⁺ neutrophils, CD3⁻ NK1.1⁺ NK cells, and CD3⁺ T cells were identified. M1-like macrophages were defined as CD86⁺ TNF⁺ within the macrophage population. CD3⁺ T cells were further classified into CD4⁺ and CD8⁺ subsets. To assess functional activation, intracellular IFN-γ staining was performed on CD8⁺ T cells, whereas CD4⁺ T cells were evaluated based on CD25 staining and intracellular Foxp3 staining. FMO and isotype controls were applied to define positive populations.

Quantification of cytokine gene and protein expression

To assess cytokine expression at both the transcriptional and protein levels, quantitative real-time PCR (qPCR) and enzyme-linked immunosorbent assay (ELISA) were performed. For qPCR, tumor-bearing mice were euthanized, and brains were exposed under sterile conditions. Tumor tissues were carefully dissected under a stereomicroscope to distinguish them from adjacent normal brain regions. Tumor and control brain tissues were snap frozen in liquid nitrogen and stored at −80 °C. Approximately 10 mg of frozen tissue was pulverized in liquid nitrogen using a pre-chilled mortar and pestle. Total RNA was extracted using TRIzol reagent (Cat. R411-01/02, Vazyme, China), and RNA concentration was determined by spectrophotometry. cDNA was synthesized from 500 ng of total RNA using the SuperScript VILO cDNA Synthesis Kit (Thermo Fisher Scientific) according to the manufacturer’s protocol. qPCR was performed on an Applied Biosystems 7500 Fast Real-Time PCR System using Power SYBR Green PCR Master Mix (Cat. AG11701, Accurate Biology, China). Each sample was run in triplicate. Gene expression levels were normalized to β-actin and analyzed using the 2^−ΔΔCt method. Primer sequences are provided in Supplementary Table 1. For ELISA, the peripheral blood of mice was collected via retro-orbital venous plexus puncture under manual restraint. Approximately 400 μL of blood was collected into EDTA-coated tubes and centrifuged at 1000 × g for 15 min at 4 °C to obtain plasma. The supernatant was carefully collected and stored on ice until analysis. The plasma levels of TNF (Cat. EK282) and IL-6 (Cat. EK2153) were measured using commercial ELISA kits (Multisciences, China), following the manufacturer’s instructions. Absorbance was measured at 450 nm using a microplate reader (Infinite F50, Tecan). Cytokine concentrations were calculated based on standard curves generated from recombinant cytokine standards using GraphPad Prism software (version 10.1.2). Each sample was measured in duplicate.

In vivo combined antitumoral activity of MBCG with RT

The in vivo antitumor efficacy of MBCG combined with RT was evaluated in two independent orthotopic murine GBM models: GL261-Luc and CT2A-Luc. For the GL261-Luc orthotopic glioma model in female C57BL/6 mice, the tumor-bearing mice were randomly divided into four groups: PBS, RT, MBCG, MBCG + RT. Each group of tumor-bearing mice were injected with 100 μL of different formulations through the tail vein (final BCG-equivalent dose of 1 × 10⁶ CFU per mouse, 4 Gy per fraction for RT). The mice were subjected to MBCG injection on day 0, 6, and 12, and RT on day 3, 9, and 15, respectively. The tumoral growth was monitored by the bioluminescence examination on day 0, 4, 8, 12, and 16. At the end of experiments (on day 16), six of the mice were euthanized and tumor tissues were collected for immune cells analysis by flow cytometry and cytokines/chemokines examination by qPCR, and another five of the mice were euthanized for the H&E staining. The survival time of residual mice in each group were recorded. In a parallel validation study, the CT2A-Luc model was used to further assess the combined therapeutic effect of MBCG and RT in a more immunologically relevant setting. Mice were orthotopically implanted with CT2A-Luc cells and randomly divided into six groups: PBS, MBCG, RT (2 Gy/1 fraction), RT (4 Gy/1 fraction), MBCG + 2 Gy/1 fraction, and MBCG + 4 Gy/1 fraction. MBCG and RT were administered on the same schedules as in the GL261 study. Tumor burden was evaluated by bioluminescence imaging (IVIS, PerkinElmer), and regions of interest (ROI) were quantified on day 12. At the endpoint, tumor-derived cells from six mice per group were analyzed by flow cytometry to quantify tumor-infiltrating M1 macrophages (CD45⁺ CD11b⁺ F4/80⁺ CD86⁺) and CD8⁺ T cells (CD45⁺ CD8⁺ CD3⁺). Survival was monitored in the remaining mice.

Immune cells depletion and verification

After the establishment of orthotopic GBM model in female C57BL/6 mice, the mice were randomly divided into four groups: PBS, macrophage depletion (M_del), CD8⁺ T cells depletion (CD8⁺ T_del), and NK cells depletion (NK_del). Briefly, macrophages depletion (M_del) was achieved through the intravenous administration of clodronate liposomes (5 mg/mL, 200 μL per mouse; Cat. 40337ES05, YEASEN, Shanghai, China). The efficiency of macrophage depletion was evaluated using CD11b and F4/80 antibodies. CD8⁺ T cells depletion (CD8⁺ T_del) was achieved by intravenous administration of an anti-CD8 monoclonal antibody (Cat. BE0061, BioXCell, China) at a dosage of 100 μg per mouse. Depletion efficiency was confirmed using anti-CD8-APC monoclonal antibodies. NK cells depletion (NK_del) was carried out by the intravenous injection of 100 µg anti-NK1.1 monoclonal antibody (Cat. BE0036, BioXCell) per mouse. The NK cell depletion was assessed using anti-NK1.1 monoclonal antibodies.

Combined antitumoral evaluation of MBCG and RT after immune cells depletion and NOD2 inhibition

After the establishment of orthotopic GBM model in female C57BL/6 mice, the mice were randomly divided into six groups: PBS, MBCG + RT, MBCG + RT in combination with GSK717, macrophage depletion (M_del) in combination with MBCG + RT, CD8⁺ T cells depletion (CD8⁺ T_del) in combination with MBCG + RT, and NK cells depletion (NK_del) in combination with MBCG + RT. MBCG was administered at an equivalent BCG dose of 1 × 10⁶ CFU in 100 μL PBS (RT, 4 Gy per fraction). The tumoral growth was monitored by the bioluminescence examination at day 0, 3, 6, and 9. Tumor-bearing mice in each group received tail vein injections of the indicated MBCG formulations at an equivalent BCG dose of 1 × 10⁶ CFU in 100 μL PBS, and radiotherapy was administered at 4 Gy per fraction. The mice were subjected to MBCG injection on day 0 and 6, and RT on day 3 and 9, respectively. The depletion of immune cells, including CD8⁺ T cells, NK cells, and macrophages, as well as NOD2 pathway inhibition, was achieved in the mouse model through two injections of antibodies or chemical agents via the tail vein on the day −1 and 5, respectively. The NOD2 inhibitor GSK717 was delivered by MBCG and intravenously administered at a dosage of 5 mg/kg.

In vivo macrophage depletion and adoptive transfer of TAMs

After establishment of the orthotopic glioblastoma (GBM) model in female C57BL/6 mice, mice were randomly assigned to five groups: PBS, macrophage depletion (M_del), M_del + RT, M_del + adoptive transfer (TAMs, 1 × 10⁵ cells per mouse), M_del + adoptive transfer + RT. For adoptive transfer experiments, TAMs were pretreated as indicated  with MBCG at a final BCG-equivalent dose of 1 × 10⁶ CFU per mouse. Radiotherapy was delivered at 4 Gy per fraction. The mice were subjected to macrophage elimination on day −1 and 5, adoptive transfer of MBCG-trained TAMs on day 0 and 6, RT on day 3 and 9, respectively. The tumoral growth was monitored by the bioluminescence examination on day 0, 3, 6, and 9. At the end of experiments (on day 10), the tumor tissues were collected for immune cells analysis by flow cytometry and cytokines/chemokines examination by qPCR.

In vivo anti-tumor performance of MBCG combined with RT in humanized NSG mice

Female NSG mice were orthotopically implanted with human glioma LN229-Luc cells (1 × 10⁵ cells in 7 μL per mouse) on day −5. The LN229-Luc tumor bearing mice were randomly assigned into five groups: PBS, MBCG + RT, adoptive transfer + RT, adoptive transfer + MBCG, adoptive transfer + MBCG + RT. Adoptive transfer of human PBMCs (1 × 10⁵ cells per mouse) was performed on days −1, 4, and 8. MBCG was administered intravenously at an equivalent BCG dose of 1 × 10⁶ CFU in 100 μL PBS per mouse on days 0, 5, and 9, and radiotherapy was administered at 4 Gy per fraction on days 3, 7, and 12. Tumor growth was monitored by bioluminescence imaging on days 0, 4, 8, and 12. At the experimental endpoint (day 13), mice were euthanized, and brain tumor tissues were harvested for hematoxylin and eosin (H&E) staining and quantitative PCR (qPCR) analysis.

In vivo combined therapy of MBCG, RT, and ICIs

After establishment of the orthotopic GBM model in female C57BL/6 mice, mice were randomly divided into five groups: PBS, ICIs, ICIs + RT, MBCG + RT, MBCG + RT + ICIs. Mice received intravenous injections of αCD47 and αPD-1 antibodies (ICIs, 100 μg per mouse) or MBCG at an equivalent BCG dose of 1 × 10⁶ CFU in 100 μL PBS per mouse on days 0, 6, 12, and 16. Radiotherapy was administered at 4 Gy per fraction on days 3, 9, and 15. Tumor growth was monitored by bioluminescence imaging of the brain on days 0, 3, 6, 9, 12, 20, and 30. Survival was assessed over a 50-day observation period, and the survival time of the remaining mice in each group was recorded.

RNA-seq data analysis for MBCG-trained TAMs

The heterotopic GBM-bearing model was established in female C57BL/6 mice by subcutaneous injection of GL261 cells (5 × 10⁶ cells per mouse) into the left flank. When the tumors reached about 600 mm³, the TAMs were separated and incubated with or without MBCG for 12 h (final concentration of BCG is 2 × 10⁵ CFU/mL). Then, the TAMs were washed with PBS and added with TRIzol. Total RNA was extracted by chloroform separation, isopropanol precipitation, and 75% ethanol. Ribosomal RNA removal, cDNA library construction, and paired-end sequencing were performed using the NEBNext® UltraTM RNA Library Prep Kit (Illumina ®, USA) from Genechem Biotechnology Co., Ltd., China. Gene expression values were generated by log2 transformation of FPKM values after adding a pseudocount of 1. The number of reads mapped to each gene was counted using FeatureCounts software (version v1.5.0-p3). Paired differential gene expression analyses were performed in R (version 3.5.0) using the DESeq2 package (version 1.20.0) based on the read counts generated by FeatureCounts, with a p-value threshold of 0.05. Volcano plots of the resulting differential genes were generated using the ggplot2 package (version 3.3.6) in R. Gene Ontology (GO) and KEGG enrichment analysis of the differential genes were conducted by the clusterProfiler package in R.

Statistical and reproducibility

Statistical analyses were performed using GraphPad Prism (version 10.1.2). Data are presented as mean ± SD. Statistical significance was assessed using unpaired two-sided t-test (Student’s or Welch’s, as appropriate), two-sided one-way ANOVA, or two-sided two-way repeated-measures ANOVA, as appropriate. For repeated-measures analyses, Geisser-Greenhouse correction was applied when necessary. Post hoc multiple-comparisons tests (including Tukey’s or Sidak’s tests) were used as specified in the figure legends. Survival curves were analyzed using the log-rank (Mantel-Cox) test. Correlation analyses were performed using two-sided Spearman’s rank correlation coefficient, as indicated. Exact P values are provided directly in the figures whenever calculable. For comparisons yielding P values below the numerical precision limit of GraphPad Prism, values are reported as P < 0.0001. For data where direct labeling was not feasible, the corresponding P values are included in the Source Data file. A P value < 0.05 was considered statistically significant. Sample sizes were not predetermined using formal statistical power calculations. For in vivo experiments, group sizes of 5–8 mice per condition were selected based on prior studies with similar tumor models and established practice in the field. The exact sample size for each experiment is specified in the corresponding figure legends. All in vivo and in vitro experiments were performed with at least three independent biological replicates.

Reporting summary

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

Author contributions

K.R. Z.Y.Y., and L.L. contributed equally to this work. K.R., Z.Y.Y., D.H.Y., L.Z.J., and H.S.Q. conceived and designed the experiments. K.R., Z.Y.Y., L.L., Z.Y.X., N.Y.M., N.Y.S., and G.D.W. performed the experiments. L.L., Y.X., T.Y., Z.Y.C., L.Q., and D.H.Y. performed the data analysis and generated figures. K.R., Z.Y.Y., L.L., and H.S.Q. wrote the manuscript and K.R., D.H.Y., L.Z.J., and H.S.Q. supervised the study.

Peer review

Peer review information

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

Data availability

The RNA sequencing data generated in this study have been deposited in the Genome Sequence Archive (GSA) of the National Genomics Data Center (NGDC) under accession number CRA040567 (BioProject: PRJCA060899), and can be accessed directly at https://ngdc.cncb.ac.cn/gsa/browse/CRA040567. Public transcriptomic data supporting the findings of this study are available from The Cancer Genome Atlas (TCGA; https://www.cancer.gov/tcga) and the Genotype-Tissue Expression (GTEx) project (https://www.gtexportal.org/home/). Publicly accessible platforms used for analysis included UCSC Xena, TISIDB, and TIMER3. Source data underlying all quantitative figures, including numerical datasets and uncropped western blot images, have been deposited in Figshare (10.6084/m9.figshare.31848097). All other data supporting the findings of this study are available within the Article and its Supplementary Information. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72067-7.