Bacterial vesicles from intratumoral L. salivarius enhance PD-1 blockade via FPR1-mediated macrophage polarization in gastric cancer

Summary

The immunomodulatory function of the gastric microbiota in cancer is poorly understood, partly due to the stomach’s acidic environment and limited microbial colonization. Here, by analyzing 68 paired human gastric cancer (GC) samples, we identify Ligilactobacillus salivarius as a commensal bacterium depleted in tumors but enriched in immune checkpoint blockade (ICB) responders. Oral administration of L. salivarius enhances anti-PD-1 efficacy in multiple GC mouse models by promoting pro-inflammatory macrophage activation. Mechanistically, bacterial extracellular vesicles (bEVs) derived from L. salivarius deliver 2,3-bisphosphoglycerate-dependent phosphoglycerate mutase (2,3-BdpM) to tumors, where it activates formyl peptide receptor 1 (FPR1) on macrophages, triggering mitogen-activated protein kinase (MAPK) and nuclear factor κB (NF-κB) signaling. Moreover, 2,3-BdpM augments the cytotoxic activity of chimeric antigen receptor (CAR)-Claudin18.2⁺ macrophages in an FPR1-dependent manner. These findings describe a microbial-macrophage axis that enhances GC immunotherapy and highlights the translational potential of orally deliverable microbial adjuvants.

Graphical abstract

Highlights

• •Intratumoral reduction of L. salivarius correlates with GC immunotherapy efficacy
• •bEVs derived from L. salivarius enhance immunotherapy efficacy in GC mouse models
• •2,3-BdpM in bEV triggers pro-inflammatory macrophage remodeling via FPR1
• •The cytotoxicity of CAR-Claudin18.2⁺ macrophages was amplified with 2,3-BdpM alone

Yu et al. identify Ligilactobacillus salivarius as a gastric commensal enriched in immunotherapy responders. Oral administration enhances anti-PD-1 efficacy by delivering 2,3-bisphosphoglycerate-dependent phosphoglycerate mutase (2,3-BdpM) via bacterial extracellular vesicles (bEVs) to activate pro-inflammatory macrophages through formyl peptide receptor 1 (FPR1), revealing a microbial-macrophage axis that potentiates gastric cancer immunotherapy.

Introduction

Gastric cancer (GC) is characterized by high intratumoral heterogeneity and aggressive biological behavior, leading to late-stage diagnosis and frequent recurrence in most patients.^1,2,3,4,5 Although therapeutic strategies have evolved substantially, clinical outcomes remain unsatisfactory for many GC patients, largely due to the complex and poorly defined interactions within the tumor microenvironment (TME). Recent clinical trials and preclinical models have demonstrated that immune checkpoint blockade (ICB) confers significant benefit in a subset of GC patients, particularly those with highly immunogenic TMEs enriched in CD8⁺ T cells, Th1 helper cells, pro-inflammatory macrophages, and natural killer (NK) cells.^6,7,8,9

Historically, the acidic environment of the stomach was considered inhospitable to microbial colonization. However, advances in high-throughput sequencing have uncovered a distinct and diverse gastric microbiota, partially overlapping with microbial ecosystems in other anatomical sites but exhibiting unique functional features.^10,11,12 While certain species, such as Streptococcus anginosus and Fusobacterium nucleatum, have been implicated in promoting gastric tumorigenesis via direct interactions with epithelial cells, the immunological roles of gastric microbes remain poorly understood.^13,14 Emerging evidence suggests that intratumoral bacteria are active modulators of antitumor immunity. These microbes can enhance immune surveillance by promoting antigen presentation and immune activation, yet they may also drive immune suppression by inducing T cell dysfunction and shaping an immunosuppressive TME.^15,16,17 Deciphering host-microbiota interactions in the stomach may reveal mechanisms of immune evasion and identify potential immunoregulatory strategies for GC.

The immune system has evolved to detect and counteract a wide array of microbial threats, rendering microorganisms some of the most potent natural inducers of immune responses.^18,19 Recent studies have highlighted that bacterial outer membrane vesicles (bacterial extracellular vesicles [bEVs]), released by both Gram-negative and Gram-positive bacteria, can elicit robust and durable antitumor immune responses, primarily through interferon (IFN)-γ- and CXCL10-mediated pathways, without causing overt toxicity.²⁰ bEVs are enriched in outer membrane components such as lipopolysaccharides, proteins, and lipids and encapsulate periplasmic contents including enzymes, polysaccharides, and nucleic acids.^21,22 Given their continuous release during host colonization and abundance of microbe-associated molecular patterns (MAMPs), bEVs are thought to serve as natural delivery systems that convey immunostimulatory signals to host receptors, thereby activating innate and adaptive immune programs.^23,24 However, the role of intratumoral bacterial bEVs in modulating antitumor immunity in GC remains unexplored.

To address this gap, we used a rigorously controlled spontaneous GC mouse model with precise gastric content sampling and 16S rRNA sequencing to identify tumor-associated microbes. Analysis of paired tumor and adjacent normal tissues from patients with or without ICB therapy revealed Ligilactobacillus salivarius as positively associated with immunotherapy response. Functional studies in GC organoid-immune co-cultures, genetically engineered mouse, and syngeneic models showed that L. salivarius and its secreted bEVs accumulate in gastric tumors and markedly enhance ICB efficacy. Mechanistically, bEVs deliver 2,3-bisphosphoglycerate-dependent phosphoglycerate mutase (2,3-BdpM), which at low concentrations activates macrophage formyl peptide receptor 1 (FPR1), triggering mitogen-activated protein kinase (MAPK) and nuclear factor κB (NF-κB) signaling and promoting pro-inflammatory, antitumor macrophage polarization. Together, these findings identify L. salivarius as a modulator of GC immunotherapy and reveal a bEV-mediated mechanism of macrophage reprogramming, highlighting tumor-resident microbiota as a promising target in GC immunotherapy.

Results

The abundance of L. salivarius is reduced in GC tissues

To investigate microbiome alterations associated with GC, we employed a previously established murine model in which GC is induced by Helicobacter pylori SS1 (HP.SS1) infection in combination with N-methyl-N-nitrosourea (MNU).²⁵ To minimize potential contamination from oral and esophageal microbes during human gastric fluid collection via endoscopy, we developed a refined in situ sampling technique in mice. Specifically, both the proximal and distal ends of the stomach were ligated, followed by slow saline perfusion to wash the gastric lumen. This approach allowed for the collection of gastric fluid under controlled conditions, effectively excluding microbial interference from other regions of the gastrointestinal tract. Using 16S rRNA gene sequencing, we profiled the gastric microbiota of age- and gender-matched tumor-bearing and non-tumor-bearing mice (Figure 1A). Among the identified taxa, we observed a significant reduction in the abundance of the genus Ligilactobacillus in GC mice, whereas the closely related genera Lactobacillus and Limosilactobacillus showed no marked differences between groups (Figures 1B and S1A). Notably, several Ligilactobacillus species, including L. salivarius, Ligilactobacillus ruminis, and Ligilactobacillus animalis, have been previously reported to stably colonize the gastrointestinal tract in both humans and animals. To identify species with stronger colonization potential in human gastric tissues, we performed quantitative reverse-transcription polymerase chain reaction (RT-qPCR) and fluorescence in situ hybridization (FISH) on tumor samples from GC patients. L. salivarius was found to robustly colonize the gastric mucosa (Figures 1C and 1D), while L. ruminis showed minimal presence and L. animalis was virtually undetectable (Figure S1B).

Figure 1.

L. salivarius is deficient in the GC tissues of patients and enhances immunotherapy in ATPM-GC mice

(A) Schematic workflow. Four GC mouse models received L. salivarius (BNCC367991) combined with anti-PD-1 antibody following antibiotic (ABX) pretreatment. Experimental groups included control, L. salivarius, anti-PD-1, and L. salivarius + anti-PD-1.

(B) 16S rRNA analysis of differential genera in gastric fluids. Ca, HP.SS1 + MNU-induced GC mice; N, age-matched normal mice.

(C–H) FISH (C, E, G) and RT-qPCR (D, F, H) detection of L. salivarius in GC and matched normal tissues (>8 cm from tumor, n = 36) and patient GC tissues without (n = 46) or with immunotherapy (n = 22; responders n = 17, non-responders n = 5). L. salivarius, red; DAPI, blue. Scale bars, 50 μm.

(I) Representative pre- and post-computed tomography (CT) images of immunotherapy responder and non-responder.

(J and K) Subcutaneous MFC tumor-bearing mice treated with L. salivarius and/or anti-PD-1; tumor volume and weight measured on day 28 (n = 8/group).

(L–N) Subcutaneous HM GC tumor-bearing mice treated similarly; tumor volume and weight (n = 8/group) and survival (>2,000 mm³ endpoint, n = 10/group).

(O–R) Flow cytometry of CD8⁺, CD4⁺ T cells, IFN-γ⁺/GZMB⁺ subsets, CD86⁺/CD206⁺ macrophages, and DCs in HM GC subcutaneous tumors (n = 5/group).

(S–U) In vivo bioluminescence, stomach weight, and morphology in ATPM-GC mice (n = 4/group).

(V and W) H&E, CK-7, and Ki-67 staining in GC tissues; scale bars, 100 or 250 μm, n = 4/group.

(X–AA) Flow cytometric analysis of T cells (X), DCs (Y), and macrophage subsets (Z), as well as IFN-γ⁺ and GZMB⁺ T cells (AA) in ATPM-GC tissues (n = 4/group).

(AB and AC) Tumor volume and weight in subcutaneous ATPM GC mice after treatment (n = 8/group).

Data are mean ± SEM. Statistical comparisons: t test or one-way ANOVA with Tukey’s post hoc test. Significance: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

To further explore the clinical relevance of L. salivarius, we analyzed paired tumor and adjacent normal tissues (>8 cm from the tumor margin) from 68 GC patients. FISH and RT-qPCR analysis of 36 paired samples from patients who had not received immunotherapy revealed a consistent decrease in L. salivarius abundance in tumor tissues (Figures 1C and 1D). Intriguingly, when stratified by treatment status, patients who had received immunotherapy (n = 22) exhibited significantly higher L. salivarius levels compared to those who had not (n = 46) (Figures 1E and 1F). Among immunotherapy recipients, responders (n = 17), defined by iRECIST criteria,²⁶ showed markedly higher L. salivarius abundance in tumor tissues compared to non-responders (n = 5) (Figures 1G–1I). These findings suggest that L. salivarius may be inversely associated with GC progression and could potentially modulate the efficacy of cancer immunotherapy.

L. salivarius enhances immunotherapy efficacy in GC mice

To investigate whether L. salivarius exerts antitumor effects or enhances immunotherapy efficacy in GC, we employed multiple mouse models, including the genetically engineered autochthonous GC mouse model (ATPM: Atp4b-iCre^+/−, Tff1^flox/flox, Trp53^flox/flox, Myc^LSL/−) (Figure S1C), subcutaneous tumors derived from ATPM or H. pylori (SS1 strain)/MNU-induced primary GC cells (HM GC cells), and the MFC murine GC model (Figure 1A). The ATPM GC mouse model recapitulates key GC mutations, including Tff1 deletion, Trp53 loss, and Myc amplification, specifically in gastric epithelial cells. Histopathological analysis confirmed the development of gastric adenocarcinoma based on H&E staining and CK-7 immunohistochemistry. Due to the long latency and limited throughput of this model (n = 4–5/group), we established subcutaneous tumors using primary GC cells derived from ATPM tumors to increase experimental consistency and sample size (n = 8/group). Mice pretreated with a 2-week antibiotic cocktail (ABX) were then administered L. salivarius, with or without anti-PD-1 monoclonal antibody therapy (Figures S1D and S1E). In both MFC- and HM-derived subcutaneous tumor models, L. salivarius monotherapy moderately suppressed tumor growth, whereas its combination with anti-PD-1 markedly enhanced antitumor efficacy (Figures 1J–1N and S1F–S1K); notably, L. salivarius BNCC367991 exhibited stronger antitumor activity than BNCC194724 (Figure S1I). Similarly, in the ATPM orthotopic model and its corresponding subcutaneous tumor model, bioluminescent imaging revealed that L. salivarius alone did not elicit a marked antitumor effect in situ (Figures 1S–1W, S2A, and S2B), but substantially inhibited tumor proliferation in the subcutaneous setting (Figures 1AB, 1AC, and S2E-S2H). Notably, both models demonstrated that L. salivarius markedly potentiated the therapeutic efficacy of anti-PD-1 treatment (Figures 1S–1W, 1AB, and 1AC). Collectively, these findings suggest that while L. salivarius exhibits modest intrinsic antitumor activity, its primary benefit lies in its capacity to significantly augment the response to ICB in GC.

To investigate whether the observed therapeutic synergy was mediated through modulation of the tumor immune microenvironment, we conducted flow cytometric analyses. In the HM-derived subcutaneous tumor mouse model, L. salivarius, either alone or in combination with anti-PD-1, significantly increased the frequency of intratumoral CD8⁺ T cells and CD86⁺ F4/80⁺ pro-inflammatory macrophages, as well as IFN-γ expression in CD8⁺ T cells. While the proportions of dendritic cells (DCs) and CD4⁺ T cells remained largely unchanged, the combination therapy, compared to anti-PD-1 alone, further enhanced Granzyme B and IFN-γ expression in CD4⁺ T cells (Figures 1O–1R and S1L–S1N). Consistently, in the orthotopic ATPM tumor model, L. salivarius, alone or with anti-PD-1, markedly increased the infiltration of CD8⁺ T cells and CD86⁺ F4/80⁺ macrophages, accompanied by elevated Granzyme B and IFN-γ expression in both CD8⁺ and CD4⁺ T cells. Notably, although the DC population remained stable, L. salivarius significantly reduced the proportion of CD4⁺ T cells (Figures 1X–1AA, S2C, and S2D). These findings suggest that L. salivarius enhances the efficacy of anti-PD-1 therapy by promoting a cytotoxic immune response within the TME.

bEVs from L. salivarius modulate immune responses more potently than metabolites

These findings suggested that L. salivarius markedly remodels the intratumoral immune microenvironment (Figures 1O–1R and 1X-1AA). To delineate the underlying immunoregulatory mechanism, we selected two reference L. salivarius strains, BNCC367991 and BNCC194724, and treated peripheral blood mononuclear cells (PBMCs) with their culture supernatants. Flow cytometric analysis showed that both supernatants increased the frequency of CD8⁺ T cells and enhanced IFN-γ and Granzyme B production in CD8⁺ and CD4⁺ T cells (Figures S3A–S3E). Notably, the BNCC367991 supernatant elicited a substantially stronger response and, importantly, induced a pronounced expansion of pro-inflammatory CD86⁺ F4/80⁺ macrophages. These results prompted us to focus on the BNCC367991 strain for further mechanistic characterization of its active components.

Metabolomic profiling of the BNCC367991 supernatant revealed significant enrichment of citric acid, 2-hydroxycaproic acid, and phenyllactic acid (PLA) compared with MRS medium (Figure S3F). Exogenous treatment of PBMCs with these metabolites demonstrated that 2-hydroxycaproic acid decreased the proportion of CD8⁺ T cells, whereas PLA modestly enhanced IFN-γ and Granzyme B production without increasing CD8⁺ T cell frequency, indicating limited immunomodulatory capacity (Figures S3G–S3L). Given accumulating evidence that bEVs mediate long-range immune regulation,^23,24 together with our in vivo findings showing that oral administration of L. salivarius reshapes the tumor immune landscape, we investigated whether BNCC367991 exerts its effects through bEVs. Upon ultracentrifugation-based removal of bEVs, the bEV-containing supernatant displayed a markedly greater ability to increase CD8⁺ T cell frequencies, promote IFN-γ and Granzyme B production, and expand CD86⁺ F4/80⁺ pro-inflammatory macrophages compared to the bEV-depleted supernatant (Figures S3M–S3R). These data support the notion that bEVs constitute a key immunostimulatory component through which L. salivarius mediates its immune-enhancing effects.

bEVs inhibit GC cell proliferation and promote apoptosis via CXADR

Although the above results indicated that L. salivarius culture supernatant (LSCS) containing bEVs can markedly modulate immune responses, it remained unclear whether they exert direct effects on GC cells. To address this, we isolated bEVs from L. salivarius and confirmed their presence via ultracentrifugation and electron microscopy (Figures 2A and 2B). Treatment of GC cells and normal gastric epithelial cells (GES-1) with different concentrations of LSCS or bEVs revealed that 20 μg/mL LSCS and 2 μg/mL bEVs significantly inhibited proliferation of GC cells without affecting GES-1 cells (Figures S4A and S4B). Furthermore, we established GC organoids derived from H. pylori SS1 + MNU-induced syngeneic mouse tumors (Figures 2C and S4C). Treatment with LSCS (10 or 20 μg/mL) markedly reduced organoid size and volume, promoting collapse and shrinkage, indicative of growth suppression (Figures 2D and S4D). Treatment with 1 or 2 μg/mL bEVs recapitulated these antiproliferative effects (Figures 2E and S4E). Similar growth inhibition and induction of apoptosis were observed in human GC organoids and GC cell lines (AGS and HGC-27) following exposure to either LSCS or bEVs (Figures S4F–S4K).

Figure 2.

L. salivarius-derived bEVs suppress GC cell proliferation and promote apoptosis via CXADR

(A and B) Transmission electron microscopy (TEM) and nanoparticle tracking analysis (NTA) images of bEVs. Scale bars: 1 μm (TEM) and 100 nm (NTA).

(C) Experimental design: organoids were derived from HP.SS1+MNU GC mice and human GC tissues and treated with LSCS or bEVs to assess proliferation.

(D and E) Morphological and Live/Dead analysis of organoids treated with LSCS (10 or 20 μg/mL) or bEVs (1 or 2 μg/mL) at indicated time points. MRS, control; cEVs, heat-inactivated bEVs (56°C, 90 min). Calcein AM, green; PI, red. Scale bars, 100 μm.

(F) Bulk RNA sequencing (RNA-seq) analysis of differential gene expression in AGS and GES-1 cells after 36 h of bEV treatment (n = 3).

(G–J) Western blot analysis of NLRP1, CXADR, and F11R in AGS, HGC-27, and GES-1 cells following bEV treatment (n = 3).

(K and L) Western blot showing CXADR overexpression efficiency in AGS and HGC-27 cells transfected with shRNA (n = 3).

(M and N) Flow cytometry analysis of apoptosis in AGS and HGC-27 cells after bEV or CXADR overexpression (n = 3).

Data are mean ± SEM. Statistical analysis: one-way ANOVA with Tukey’s multiple comparisons test; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

To identify the molecular target through which bEVs suppress GC cell proliferation, we performed transcriptomic profiling of AGS and GES-1 cells following bEV treatment. Three receptor-related genes, Nod-like receptor protein 1 (NLRP1), coxsackievirus and adenovirus receptor (CXADR), and F11 receptor (F11R), were highly expressed in AGS but minimally expressed in GES-1 cells, and among these, CXADR and F11R exhibited further downregulation upon bEV exposure in AGS cells (Figure 2F). Protein-level validation in AGS, HGC-27, and GES-1 cells revealed that NLRP1 and CXADR were markedly elevated in GC cells compared with GES-1 (Figures 2G and 2H). Subsequent immunoblotting of bEV-treated GC cells demonstrated a marked suppression of CXADR and F11R, whereas NLRP1 expression remained unchanged (Figures 2I and 2J). These findings implicated CXADR as a putative functional mediator of the antiproliferative activity of bEVs. Indeed, CXADR overexpression substantially reversed bEV-induced apoptosis in GC cells, indicating that CXADR is required for the pro-apoptotic and growth-suppressive effects of bEVs (Figures 2K–2N).

bEVs enhance immune-mediated cytotoxicity and selectively target gastric tumors

Although bEVs were able to inhibit GC cell proliferation and bind to both GC cells and organoids, flow cytometry analysis revealed that LSCS-containing bEVs induced substantial immunomodulatory effects (Figures 2, 3A, 3B, and S3M–S3R). To investigate immune modulation, we established a co-culture system using GC organoids and splenocytes from the same ATPM-GC mouse (Figures 3C and 3D). To control for potential immunogenic effects of Matrigel, we compared splenocyte profiles after treatment with PBS, Matrigel alone, or Matrigel containing GC organoids. While Matrigel alone had no effect, GC organoids significantly reduced CD8⁺ T cell frequency, decreased IFN-γ production in both CD4⁺ and CD8⁺ T cells, and increased immunosuppressive CD206⁺ F4/80⁺ macrophages (Figures S5A–S5G). To separate direct cytotoxicity from immunomodulation, we reduced the concentrations of LSCS and bEVs to 1/10 of those used in tumor suppression assays. At these lower doses, organoid growth was unaffected; however, 1 μg/mL LSCS and 0.1 μg/mL bEVs significantly enhanced PBMC-mediated cytotoxicity against GC cells (Figures S4A, S4B, S5H, and S5I). Flow cytometry of co-cultured splenocytes revealed enhanced proportions of CD86⁺ F4/80⁺ macrophages, DCs, and CD8⁺ T cells, along with increased IFN-γ production in both CD4⁺ and CD8⁺ T cells, and a reduction in total CD4⁺ T cells (Figures 3E–3L). Moreover, splenocyte-mediated cytotoxicity against GC organoids was significantly enhanced (Figures 3M and S5J).

Figure 3.

bEVs potentiate splenocyte-mediated cytotoxicity against GC organoids and target gastric tumors

(A and B) Immunofluorescence detection of bEV adhesion to primary GC cells (A) and human GC organoids (B). Cells (DIO), green; bEVs, red; DAPI, blue. Scale bars: 10 μm in (A) and 50 μm in (B); n = 3/group.

(C and D) Splenocytes from ATPM-GC mice co-cultured with tumor organoids. Organoid proliferation and splenocyte colocalization (green, GC organoids; red, splenocytes) were assessed microscopically; scale bar, 25 μm.

(E–H) Flow cytometry analysis of CD4⁺ and CD8⁺ T cell proportions (E and F) and IFN-γ expression (G and H) after LSCS or bEV treatment (n = 3/group).

(I–L) Alterations in proportions of DCs, CD86⁺F4/80⁺, and CD206⁺F4/80⁺ macrophages (n = 3/group).

(M) GC organoid proliferation following LSCS or bEVs combined with splenocytes at 0, 12, and 50 h; scale bars, 50 μm.

(N–Q) In vivo imaging of ATPM-GC mice (N and O) and subcutaneous HM GC tumor-bearing mice (P and Q) after intraperitoneal injection of DIR-labeled bEVs or E. coli EVs at indicated time points (n = 3/group).

Data are mean ± SEM. Statistical significance determined by unpaired t test or one-way ANOVA with Tukey’s multiple comparisons: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

We next treated splenocytes from healthy mice with 0.1 μg/mL bEVs to further evaluate the immunomodulatory potential of L. salivarius bEVs. Flow cytometry analysis revealed a marked increase in CD8⁺ T cells and enhanced IFN-γ production in both CD8⁺ and CD4⁺ T cells, while DC populations remained unchanged (Figures S5K–S5N). Importantly, bEVs induced a shift in macrophages toward a pro-inflammatory, anti-tumor phenotype (Figure S5N). Together, these findings demonstrate that while high-dose L. salivarius-derived bEVs suppress GC cell proliferation directly, low doses primarily act by reprogramming the immune microenvironment to enhance antitumor immunity.

To elucidate how L. salivarius-derived bEVs enhance the cytotoxicity of splenic lymphocytes against GC cells, we investigated whether bEVs can selectively target GC cells using multiple models, including GC cell lines, patient-derived organoids, and both orthotopic and subcutaneous mouse models of GC (Figures 3A, 3B, and S5O). In vitro binding assays further demonstrated that L. salivarius bEVs preferentially associated with human GC cell lines (AGS and HGC-27) over normal gastric epithelial cells (GES-1) (Figures S5O and S5P). In ATPM-induced GC mice, intraperitoneally injected DiR-labeled bEVs preferentially accumulated in gastric tumors, as shown by in vivo imaging, with minimal distribution in healthy controls (Figures 3N and 3O). Similar tumor tropism was observed in a subcutaneous GC model established using primary tumor cells from H. pylori SS1 + MNU-induced GC mice, where bEVs derived from L. salivarius showed greater accumulation than E. coli bEVs, without observable toxicity on histopathological analysis (Figures 3P, 3Q, S5Q, and S5R). Together, these findings indicate that L. salivarius bEVs exhibit selective tropism for GC cells, which likely underpins their ability to facilitate lymphocyte-mediated tumor killing. However, the molecular basis of this tumor-targeting mechanism requires further investigation.

bEVs enhance immunotherapy efficacy in GC mice

To investigate whether L. salivarius-derived bEVs recapitulate the immunotherapeutic benefits of L. salivarius in GC, we employed multiple mouse models, including the ATPM transgenic model, subcutaneous tumors derived from ATPM or H. pylori (SS1 strain)/MNU-induced primary GC cells, and the MFC murine GC model (Figure 4A). Mice were administered L. salivarius bEVs alone or in combination with anti-PD-1 antibodies. In both MFC- and HM-derived subcutaneous tumor models, bEVs from L. salivarius (BNCC367991) moderately suppressed tumor growth and, together with anti-PD-1 antibodies, elicited a markedly enhanced antitumor effect, whereas bEVs from L. salivarius (BNCC194724) similarly potentiated anti-PD-1 efficacy but showed no significant tumor-suppressive activity when administered alone (Figures 4B–4F, S5S, S5T, and S6A–S6H). Consistently, in the ATPM orthotopic and corresponding subcutaneous tumor models, L. salivarius (BNCC367991) bEVs alone failed to induce a robust antitumor response (Figures 4L–4Q, S6J, and S6K). Although no significant reduction in tumor volume was observed in the ATPM subcutaneous model, tumor weight was notably decreased (Figures 4W, 4X, and S6N–S6Q). Importantly, both models demonstrated that L. salivarius bEVs markedly potentiated the efficacy of anti-PD-1 therapy (Figures 4L–4Q, 4W, and 4X). Across all tumor models, bEVs significantly enhanced anti-PD-1-mediated tumor control, as evidenced by reduced tumor burden and prolonged survival. These findings indicate that while L. salivarius bEVs exert limited intrinsic antitumor activity, their principal value lies in amplifying the therapeutic response to ICB in GC.

Figure 4.

bEVs enhance immunotherapy in GC mice

(A) Schematic of experimental design. Four GC mouse models received bEVs alone or combined with anti-PD-1 antibody. Groups: control, bEVs, anti-PD-1, bEVs + anti-PD-1.

(B–F) Tumor volume and weight (n = 8/group) and survival (n = 10/group) in subcutaneous MFC and HM GC tumor models.

(G–K, R–V) Flow cytometry analysis of immune cell populations in HM GC primary tumor-bearing and ATPM-GC tumors, including CD8⁺/CD4⁺ T cells, IFN-γ⁺ and GZMB⁺ T cells, DCs, and macrophage subsets (n = 4–5/group).

(L and M) In vivo bioluminescence imaging of ATPM-GC mice before and after treatment (n = 4–5/group); red crosses indicate tumor-related deaths, empty squares indicate anesthesia-related deaths.

(N and O) Stomach weight and gross morphology in ATPM-GC mice (n = 4/group).

(P and Q) H&E (100 μm) and immunohistochemical staining of CK-7 (250 μm) and Ki-67 (100 μm) in GC tissues (n = 4/group).

(W and X) Tumor volume and weight in ATPM GC primary tumor-bearing mice on day 28 (n = 8/group).

Data are mean ± SEM. Statistical significance: unpaired t test or one-way ANOVA with Tukey’s multiple comparisons; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

To determine whether the synergistic effect was mediated through modulation of the tumor immune microenvironment, we conducted flow cytometric analysis. In the HM-derived subcutaneous tumor model, L. salivarius bEVs, either alone or combined with anti-PD-1, significantly increased intratumoral CD8⁺ T cell infiltration, CD86⁺ F4/80⁺ pro-inflammatory macrophages, and IFN-γ production by CD8⁺ T cells. While the frequencies of DCs and CD4⁺ T cells remained largely unchanged, the combination therapy further enhanced Granzyme B and IFN-γ expression in CD4⁺ T cells compared to anti-PD-1 monotherapy (Figures 4G–4K and S6I). Similarly, in the orthotopic ATPM tumor model, L. salivarius, alone or with anti-PD-1, promoted robust infiltration of CD86⁺ F4/80⁺ macrophages. Although DC and CD4⁺ T cell levels were not significantly altered, combination treatment further upregulated Granzyme B and IFN-γ in CD8⁺ T cells relative to anti-PD-1 alone (Figures 4R–4V and S6L). These results indicate that L. salivarius-derived bEVs augment the efficacy of anti-PD-1 therapy by enhancing cytotoxic immune responses within the TME.

bEVs enhance CD8⁺ T cell-mediated cytotoxicity by promoting pro-inflammatory macrophage activation via FPR1

Given that L. salivarius-derived bEVs markedly promote macrophage polarization toward a proinflammatory phenotype and enhance IFN-γ secretion by CD8⁺ T cells, we next investigated whether this effect on CD8⁺ T cells was direct or mediated through macrophage activation. Bone marrow-derived macrophages (BMDMs) and CD8⁺ T cells sorted from murine spleens were isolated and used in co-culture assays. BMDMs were first stimulated with L. salivarius-derived bEVs and then co-cultured with CD8⁺ T cells. Flow cytometry revealed that bEVs robustly activated BMDMs, increasing the proportion of CD86⁺ iNOS⁺ F4/80⁺ cells (Figure 5A). Notably, bEVs alone had minimal impact on CD8⁺ T cell functionality (Figure S7A). However, CD8⁺ T cells co-cultured with bEV-treated BMDMs exhibited significantly elevated expression of Granzyme B and IFN-γ, indicating that macrophages are required for bEV-mediated T cell activation (Figure 5B). To assess whether these activated macrophages enhance CD8⁺ T cell-mediated cytotoxicity against GC cells, we established a three-cell co-culture system comprising primary GC cells, BMDMs, and CD8⁺ T cells derived from the same ATPM mice (Figure 5C). Consistent with earlier findings, bEVs had little effect on CD8⁺ T cell cytotoxicity in the absence of BMDMs. In contrast, the presence of bEV-primed BMDMs significantly enhanced CD8⁺ T cell-mediated killing of GC cells (Figures 5D–5F). The strongest cytotoxicity was observed when all three components—bEVs, BMDMs, and CD8⁺ T cells—were combined, highlighting a critical intermediary role for macrophages in this process.

Figure 5.

bEVs promote pro-inflammatory macrophage activation via FPR1, enhancing the cytotoxic function of CD8⁺ T cells

(A) BMDMs were treated with 0.1 μg/mL bEVs for 36 h. Macrophage polarization (CD86⁺ iNOS⁺ F4/80⁺/CD206⁺ Arg-1⁺ F4/80⁺) was analyzed by flow cytometry (n = 3/group).

(B) Sorted BMDMs were co-cultured with CD8⁺ T cells and treated with bEVs for 36 h. IFN-γ and Granzyme B expressions in CD8⁺ T cells were measured (n = 3/group).

(C–F) Co-culture system established from ATPM-GC mice: GC cells and CD8⁺ T cells in the lower chamber, BMDMs in the upper chamber. Cytotoxicity was assessed after 12 h (D), and GC cell apoptosis was measured after 16 h (E-F) (n = 3/group).

(G) Bulk RNA-seq of BMDMs following 36-h bEV treatment (n = 3/group).

(H and I) Fpr1/Fpr2 expression in BMDMs after bEV intervention, measured by flow cytometry (H) and western blot (I) (n = 3/group).

(J and K) Effects of bEVs combined with shFPR1 or shFPR2 on CD86⁺ and CD206⁺ macrophage ratios in BMDMs (n = 3/group).

(L) Co-localization of bEVs and FPR1 in BMDMs. FPR1, green; bEVs, red; DAPI, blue. Scale bar, 10 μm.

(M) FPR1 and CD68 co-localization in tumor tissues from 12 GC patients, stratified by immunotherapy. Scale bar, 50 μm.

(N) Immunofluorescence of FPR1 and F4/80 in ATPM-GC tissue treated with bEVs ± anti-PD-1. Scale bar, 15 μm (n = 4–5/group).

(O–Q) Co-culture of CD8⁺ T cells with BMDMs treated with bEVs or shFPR1, showing cytotoxicity (O) and IFN-γ and Granzyme B expression (P and Q) (n = 3/group).

(R–T) Western blot of P-STAT3, P-STAT1, JNK1/2/3, MEK1/2, ERK1/2, p38, p65, P-p65, P-IκB-α, and IκB-α in control and shFPR1 groups (n = 3/group).

Data are mean ± SEM. Statistical comparisons: unpaired t-tests or one-way ANOVA with Tukey’s multiple comparisons. Significance: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

To elucidate how bEVs reprogram macrophages toward a proinflammatory phenotype, we performed RNA sequencing on bEV-treated BMDMs (Figure S7B). Differential expression analysis revealed upregulation of proinflammatory genes such as Il1a, Il1b, Il12a, and Tnf, along with downregulation of immunosuppressive markers including Cd163, Cd209a, Irf4, and Csf1r, corroborating the flow cytometry data (Figure S7C). Notably, Fpr1 and Fpr2 were among the most upregulated genes at both the mRNA and protein levels (Figures 5G–5I). To determine their functional relevance, we silenced Fpr1 and Fpr2 using short hairpin RNA (shRNA) in BMDMs. Knockdown of either gene impaired the proinflammatory activation induced by bEVs, with Fpr1 knockdown exerting a more pronounced effect (Figures 5J, 5K, S7D, and S7E). Fluorescence imaging revealed co-localization of labeled bEVs with FPR1 on macrophages, suggesting direct interaction (Figure 5L). In vivo, bEV treatment increased the frequency of FPR1⁺ macrophages in GC tissues, an effect similarly observed in gastric tumors from immunotherapy-treated mice and patients (Figures 5M, 5N, and S7F). Analysis of the GSE183904 single-cell RNA sequencing dataset further showed that tumors enriched for FPR1⁺ macrophages exhibited higher proportions of GZMB⁺ CD8⁺ T cells (Figure S7G). Functionally, Fpr1 silencing in BMDMs abrogated the ability of bEV-primed macrophages to enhance Granzyme B and IFN-γ expression in CD8⁺ T cells and weakened their cytotoxicity against GC cells (Figures 5O–5Q). These data identify FPR1 as a key mediator of macrophage activation and downstream CD8⁺ T cell function. Moreover, transcriptomic analysis of macrophages treated with bEVs revealed significant enrichment of the MAPK and NF-κB signaling pathways based on Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis of differentially expressed genes and transcription factors. To investigate the role of FPR1 in mediating these effects, we knocked down Fpr1 in macrophages and examined key components of the JAK-STAT (phosphorylated STAT3 and STAT1), MAPK (JNK1/2/3, MEK1/2, ERK1/2, and p38), and NF-κB (p65, phosphorylated p65, IκB-α, and phosphorylated IκB-α) pathways by western blotting. Fpr1 knockdown markedly reduced the expression of MAPK pathway proteins and the phosphorylation of NF-κB pathway components, whereas proteins of the JAK-STAT pathway remained unaffected. These results indicate that bEVs promote pro-inflammatory macrophage activation predominantly through FPR1-dependent regulation of the canonical MAPKs and NF-κB signaling pathways (Figures 5R, 5S, and S7H–S7J).

Finally, to assess whether FPR1 is involved in the direct anti-tumor effects of bEVs on GC cells, we treated primary GC cells with high-dose bEVs (2 μg/mL) in the presence or absence of Fpr1 knockdown (Figures S7K and S7L). Although bEVs upregulated FPR1 expression in GC cells, silencing Fpr1 did not affect bEV-induced inhibition of proliferation or apoptosis induction (Figures S7M–S7P), indicating that the cytotoxic effects of bEVs on tumor cells are independent of FPR1 signaling in cancer cells.

2,3-BdpM in bEVs promotes pro-inflammatory macrophage activation by activating the FPR1-dependent MAPKs and NF-κB signaling pathways

To identify the key bEV component responsible for targeting the FPR1 receptor on macrophages, we performed a pull-down assay using glutathione S-transferase (GST)-tagged human FPR1 protein to capture interacting proteins from bEV lysates (Figure S8A). Subsequent liquid chromatography-mass spectrometry analysis identified 21 candidate bEV-binding proteins (Figure 6A). Subsequent in silico molecular docking was conducted using AlphaFold 3, assessing the interaction between each candidate and FPR1. Based on predicted TM (pTM) and interface pTM (ipTM) scores (>0.7), two high-affinity candidates were identified: 2,3-BdpM and a LysM domain-containing peptidoglycan-binding protein (LysM) (Figure S8B). We next recombinantly expressed both proteins and assessed their physical interaction with macrophage-derived FPR1 using co-immunoprecipitation. Among them, 2,3-BdpM showed stronger binding affinity to FPR1 than LysM (Figure 6B and S8C). Consistently, GST pull-down assays using purified recombinant proteins demonstrated a direct physical interaction between 2,3-BdpM and FPR1 in vitro (Figure 6C). Functional assays further demonstrated that 2,3-BdpM significantly upregulated Fpr1 expression in macrophages, whereas LysM had minimal effects (Figures 6D, 6E, and S8D).

Figure 6.

2,3-BdpM from bEVs promotes pro-inflammatory macrophage activation via the FPR1/MAPKs/NF-κB pathway and enhances immunotherapy efficacy in GC mice

(A) Protein analysis of bEVs by GST pull-down and mass spectrometry, with molecular docking of FPR1 using AlphaFold 3. ipTM and pTM values are shown.

(B and C) Interaction between FPR1 and GST-2,3-BdpM detected by co-immunoprecipitation (B) or GST pull-down (C).

(D and E) Western blot of FPR1 in BMDMs treated with 0.1 μg/mL GST-2,3-BdpM (n = 3/group).

(F) Flow cytometry of CD86⁺ iNOS⁺ F4/80⁺ and CD206⁺ Arg-1⁺ F4/80⁺ macrophages in BMDMs after GST-2,3-BdpM and/or shFPR1 treatment (n = 3/group).

(G–I) CD8⁺ T cell activation (IFN-γ, Granzyme B) and cytotoxicity against primary GC cells after co-culture with treated BMDMs (n = 3/group).

(J) Western blot of signaling proteins (P-STAT3, P-STAT1, JNK1/2/3, MEK1/2, ERK1/2, p38, P-p65, P-IκB-α, and IκB-α) in control, GST-2,3-BdpM, shFPR1, and combination groups (n = 3/group).

(K) Experimental design: HM GC primary tumor-bearing and ATPM-GC mice received GST-2,3-BdpM (25 μg/mouse) and/or anti-PD-1 antibody.

(L–N) Tumor weight and volume (n = 8/group) and survival on day 28 (n = 10/group).

(O–S) Flow cytometry of immune cells in subcutaneous tumors: macrophages (O), CD8⁺/CD4⁺ T cells (P), DCs (Q), and cytokine-positive T cells (R and S) (n = 5/group).

(T and U) In vivo bioluminescence imaging of ATPM-GC mice before and after treatment (n = 4/group).

(V and W) Stomach weight and gross morphology (n = 4/group).

(X and Y) H&E and immunohistochemical staining of CK-7 and Ki-67 (scale bars, 100 μm; n = 4/group).

(Z–AE) Flow cytometry of immune populations in ATPM-GC tissues (n = 4/group).

Data are mean ± SEM. Statistical analysis: unpaired t test or one-way ANOVA with Tukey’s multiple comparisons test. Significance: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Given that 2,3-BdpM belongs to the dPGM family, we next examined whether its mammalian homolog exerts similar immunomodulatory activities. Phosphoglycerate mutase 1 (PGAM1), the mammalian enzyme within the same family, was used to exclude potential effects from endogenous host dPGM.^27,28 Exogenous PGAM1 treatment of macrophages neither upregulated FPR1 expression nor promoted pro-inflammatory polarization (Figures S8E and S8F), indicating that the immunostimulatory function of 2,3-BdpM is bacteria specific. To further establish the bacterial specificity of 2,3-BdpM, PCR analysis of the gpmA locus confirmed that this gene is present exclusively in bacterial genomes (Figure S8G). Furthermore, proteomic analysis of bacterial culture supernatants depleted of bEVs and of isolated bEVs revealed that 2,3-BdpM is present at high abundance exclusively within bEVs (Tables S1 and S2). Remarkably, macrophages responded to 2,3-BdpM at concentrations as low as 0.1 μg/mL, undergoing robust polarization toward a pro-inflammatory state (Figures S8H and S8I). Notably, the capacity of 2,3-BdpM to drive pro-inflammatory macrophage activation was completely abolished by shRNA-mediated knockdown of Fpr1 (Figure 6F and S8J). Moreover, macrophages treated with 2,3-BdpM promoted CD8⁺ T cell activation, as evidenced by increased Granzyme B and IFN-γ expression and enhanced cytotoxicity against GC cells (Figures 6G–6I). These effects were significantly attenuated when Fpr1 was silenced in BMDMs. Collectively, these findings identify 2,3-BdpM as a critical bEV-derived protein that activates pro-inflammatory macrophages via the FPR1 receptor.

To further determine whether 2,3-BdpM drives pro-inflammatory macrophage polarization through FPR1, we performed follow-up experiments and assessed relevant signaling pathways by western blotting. Treatment with 2,3-BdpM markedly upregulated key components of the MAPK pathway (JNK1/2/3, MEK1/2, ERK1/2, and p38) and NF-κB pathway (phosphorylated p65 and phosphorylated IκB-α). Importantly, this effect was largely abrogated upon Fpr1 knockdown, indicating that 2,3-BdpM acts through FPR1 to activate these pathways (Figures 6J, S8K, and S8L). In contrast, no changes were observed in the JAK-STAT pathway (phosphorylated STAT3 and STAT1), confirming that 2,3-BdpM specifically engages the FPR1-dependent MAPKs and NF-κB signaling pathways to promote pro-inflammatory macrophage polarization (Figure 6J and S8L).

2,3-BdpM enhances immunotherapy response in GC mice

We next evaluated whether 2,3-BdpM exerts direct inhibitory effects on GC cells by treating primary GC cells derived from ATPM mice with 2 μg/mL 2,3-BdpM. Despite the relatively high concentration, no significant changes were observed in cell proliferation or apoptosis, indicating that the antitumor activity of 2,3-BdpM is not mediated by direct cytotoxicity (Figures S8M–S8P). Instead, these results suggest that 2,3-BdpM exerts its antitumor effects primarily through modulation of macrophage activation and subsequent enhancement of CD8⁺ T cell-mediated cytotoxicity against GC cells.

To investigate whether 2,3-BdpM enhances antitumor immunity in vivo, similar to bEVs, we tested its efficacy in two GC mouse models: the ATPM orthotopic model and a subcutaneous model established using primary tumor cells from H. pylori (SS1 strain) and MNU-induced mice (Figure 6K). Mice were treated with 2,3-BdpM alone or in combination with anti-PD-1 antibody. Tumor volume, weight, and histological analyses revealed that 2,3-BdpM significantly suppressed tumor progression and potentiated the therapeutic efficacy of anti-PD-1 treatment (Figures 6L–6N, 6T–6W, and S9A–S9C). These findings were supported by Ki-67 immunohistochemistry, which confirmed reduced proliferative activity at the cellular level (Figures 6X, 6Y, S9B–S9D, and S9H). Flow cytometric analysis of tumor-infiltrating lymphocytes showed that treatment with 2,3-BdpM alone or in combination with anti-PD-1 significantly increased the proportion of CD86⁺ F4/80⁺ pro-inflammatory macrophages and CD8⁺ T cells within the TME (Figures 6O, 6AD, S9E, and S9I). Notably, the combination treatment further enhanced CD8⁺ T cell effector function, as evidenced by elevated expression of Granzyme B and IFN-γ (Figures 6P–6S, 6Z–6AC, S9E, and S9I). These results indicate that 2,3-BdpM from L. salivarius-derived bEVs enhances the efficacy of anti-PD-1 therapy by promoting cytotoxic immune responses in the TME.

2,3-BdpM enhances the cytotoxicity of CAR-Claudin 18.2⁺ macrophages against GC via FPR1 activation

To investigate the in vivo mechanism by which 2,3-BdpM enhances anti-tumor immunity via FPR1-mediated activation of pro-inflammatory macrophages, we engineered chimeric antigen receptor macrophages (CAR-Claudin18.2⁺ M) using an optimized CAR plasmid and lentiviral transduction system (Figure 7A). This was achieved through targeted modifications to the intracellular and extracellular domains of the CAR construct (Figure 7B). Subsequent treatment of CAR-Claudin18.2⁺ M with 2,3-BdpM or Fpr1-shRNA lentivirus revealed that 2,3-BdpM significantly promoted pro-inflammatory activation of these cells, an effect notably attenuated by Fpr1 knockdown (Figures 7C–7E). Co-culture experiments with primary GC cells or GC organoids from ATPM-GC mice further demonstrated that the enhanced cytotoxicity and apoptosis-inducing capacity of 2,3-BdpM-treated CAR-Claudin18.2⁺ M was similarly dependent on FPR1 signaling (Figures 7F–7L and S10A).

Figure 7.

The cytotoxic activity of CAR-Claudin18.2⁺ M against GC cells is enhanced using 2,3-BdpM through FPR1

(A) Schematic of CAR-Claudin18.2 in murine macrophages.

(B) CAR-Claudin18.2 lentivirus infection efficiency in BMDMs (n = 3/group).

(C and D) CD86⁺ F4/80⁺ and CD206⁺ F4/80⁺ macrophage ratios in CAR-M cells after 2,3-BdpM (1 μg/mL) and/or shFPR1 treatment (n = 3/group).

(E) Immunofluorescence detection of Claudin18.2 in primary GC cells. Claudin18.2, green; DAPI, blue. Scale bar, 100 μm.

(F) Cytotoxicity of CAR-M cells against primary GC cells at different effector/target ratios.

(G) Cytotoxicity of 2,3-BdpM and/or shFPR1-treated CAR-M cells (E:T = 1:1) (n = 3/group).

(H and I) Flow cytometry of apoptosis induced by CAR-M cells (4 h) or treated CAR-M cells (8 h) in primary GC cells (n = 3/group).

(J and K) Co-culture of treated CAR-M cells with ATPM GC organoids; organoid growth observed at 16, 24, and 30 h. Scale bar, 50 μm.

(L) Immunofluorescence imaging of CAR-M (DIO, green) cytotoxicity on GC organoids (DIL, red). Scale bar, 50 μm.

(M–P) NOD-SCID mice bearing primary GC cells were treated with preconditioned CAR-M cells. Body weight and in vivo fluorescence monitored (n = 5/group).

Data are mean ± SEM. Statistical comparisons: unpaired t-tests or one-way ANOVA with Tukey’s multiple comparisons. ∗p < 0.05; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

To evaluate therapeutic efficacy and safety in vivo, primary GC cells from ATPM-GC mice were subcutaneously implanted into NOD/SCID mice (Figure S10B), followed by adoptive transfer of CAR-Claudin18.2⁺ M pretreated with 2,3-BdpM or Fpr1 knockdown (Figure 7M). Compared to controls or monotherapy groups, 2,3-BdpM-treated CAR-Claudin18.2⁺ M markedly suppressed tumor growth without causing observable toxicity in major organs (Figures 7N–7P and S10C–S10E). Together, these findings support the clinical translational potential of combining 2,3-BdpM with CAR-Claudin18.2⁺ macrophage therapy for the treatment of GC.

Discussion

Recent studies implicate the microbiota in tumor progression and therapeutic response. In GC, intragastric bacteria such as Streptococcus anginosus and Fusobacterium nucleatum promote tumor growth, yet how intratumoral microbes shape the immune microenvironment and influence immunotherapy remains unclear.^13,14,29 Using a spontaneous GC model with contamination-controlled gastric sampling, together with human GC specimens and multiple murine models, we identified a significant depletion of Ligilactobacillus salivarius in tumor tissues. We further demonstrate that L. salivarius enhances ICB efficacy in GC. Mechanistically, L. salivarius does not act directly on CD8⁺ T cells but delivers bEVs to tumors, reprogramming tumor-associated macrophages (TAMs) toward a pro-inflammatory, antitumor phenotype, thereby augmenting CD8⁺ T cell cytotoxicity. We identify the bEV-associated protein 2,3-BdpM as the key mediator, which engages FPR1 on macrophages to activate MAPK and NF-κB signaling and drive macrophage reprogramming, ultimately enhancing antitumor immunity.

Previous studies show that Streptococcus anginosus promotes GC progression via the TMPC-ANXA2-MAPK axis and Fusobacterium nucleatum facilitates tumor growth and immunosuppression through the interleukin (IL)-17/NF-κB/RelB pathway.^13,14 However, how intratumoral microbes modulate the immune microenvironment to affect immunotherapy response in GC remains unclear. Using murine GC models and human specimens, we observed a consistent depletion of L. salivarius in tumors, which was restored in patients receiving ICB therapy. L. salivarius, a probiotic with strong epithelial adherence and anti-inflammatory, antioxidant, and metabolic regulatory properties, can colonize the stomach and inhibit H. pylori-induced IL-8 production.^30,31,32,33 Its culture supernatants also enhance Th1 cytokine production and inhibit colon cancer cell proliferation.^31,34,35 Despite these immunomodulatory effects, the role of L. salivarius in tumor immunity, particularly within the TME, remains largely unexplored. Here, we found that gastric L. salivarius promotes pro-inflammatory macrophage polarization and enhances CD8⁺ T cell function, without affecting CD4⁺ T cells. Importantly, exogenous L. salivarius administration significantly improved ICB efficacy in multiple GC mouse models, highlighting its potential as an adjuvant probiotic in cancer immunotherapy.

The interplay between the microbiota and host immunity is increasingly recognized in cancer, yet how intratumoral bacteria modulate tumor progression and antitumor immunity remains unclear. Bacteria can influence the TME via antigen presentation, epitope mimicry, or production of metabolites such as short-chain fatty acids, bile acids, and inosine that regulate NK and regulatory T cell activity.^34,35 Metabolomic analysis of L. salivarius supernatants revealed enrichment of citric acid, 2-hydroxycaproic acid, and PLA. While citric acid and 2-hydroxycaproic acid showed no effect on human PBMCs, PLA enhanced IFN-γ and Granzyme B production by CD8⁺ T cells without changing their frequency. PLA, a phenolic compound produced by Lactobacillus, has broad antimicrobial activity and can improve immune traits in vivo, yet its overall immunomodulatory effect is modest.^36,37,38 bEVs are emerging as long-range immune modulators.^39,40,41 We show that L. salivarius produces bEVs that accumulate in gastric tumors and induce immune modulation. Unlike bEVs from other bacteria, L. salivarius bEVs do not directly activate CD8⁺ T cells but enhance antitumor immunity indirectly via proinflammatory activation of tumor-associated macrophages. High bEV concentrations can suppress GC cell proliferation, but at lower concentrations, their immunomodulatory effect remains strong. Compared with live bacteria, bEVs are more stable, safer, and do not require colonization, making them promising immunotherapeutics.

bEVs are enriched in metabolic enzymes, small molecules, and MAMPs, which can modulate tumor cell metabolism and proliferation and activate host innate immunity via pattern recognition receptors such as Toll-like receptors (TLRs) and NOD-like receptors (NLRs), triggering proinflammatory responses, cytokine release, and cell death.⁴² Importantly, we show that L. salivarius-derived bEVs directly suppress GC cell proliferation through CXADR. CXADR was initially identified as a high-affinity receptor for coxsackievirus B and other enteric viral subtypes, and elevated CXADR expression has been documented in several cancer types.^43,44,45 Consistent with these observations, we found that bEV-mediated suppression of CXADR in GC cells leads to marked induction of apoptosis. Notably, CXADR expression was higher in GC cell lines than in non-malignant GES-1 cells, suggesting it may serve as a preferential target for bEV-tumor cell interactions, although the molecular basis of this selectivity remains to be determined.

High concentrations of bEVs inhibited GC cell proliferation, whereas a 10-fold lower dose promoted macrophage polarization toward a pro-inflammatory phenotype. Transcriptomic analysis showed no significant changes in TLR or NLR expression, indicating that bEV-induced immune activation is independent of canonical pattern recognition receptor signaling. Instead, we identified FPR1 as a key mediator, activating MAPK and NF-κB pathways and driving pro-inflammatory, antitumor macrophage polarization. FPR1 is broadly expressed in human and mouse tissues, including immune cells (neutrophils, monocytes, and macrophages) as well as cells in the nervous vascular, endocrine, hepatic, and epithelial systems.⁴⁶ FPR1, a G protein-coupled receptor, along with FPR2, is known to stimulate intracellular signaling cascades and activate the NADPH oxidase complex, promoting reactive oxygen species generation.^47,48,49 Although FPR1 is classically known to recognize N-formylated peptides and to function in antimicrobial defense, its role in bEV-mediated antitumor immunity has not been reported.^50,51,52 2,3-BdpM, a bacterial enzyme encoded by gpmA, directly interacts with FPR1, as shown by GST pull-down assays using purified proteins. While 2,3-BdpM does not directly suppress tumor cell proliferation, it induces FPR1 activation, promotes inflammatory macrophage polarization, and enhances CD8⁺ T cell-mediated cytotoxicity against GC cells. Proteomic analyses further revealed that 2,3-BdpM is selectively enriched in L. salivarius-derived bEVs, identifying bEVs as the delivery vehicle. Together, these findings expand the ligand repertoire of FPR1 beyond peptides and describe a bEV-FPR1 axis that shapes antitumor immunity, with potential implications for GC immunotherapy.

CAR-T cell therapies have shown remarkable success in hematologic malignancies, yet their efficacy against solid tumors remains limited because the dense extracellular matrix and physical barriers, including collagen and glycosaminoglycans, impede T cell infiltration and persistence. In contrast, macrophages possess intrinsic chemotactic and tissue-penetrating capabilities, enabling them to access the tumor core more effectively.^53,54,55 In this study, we demonstrate that 2,3-BdpM enhances the proinflammatory activation of CAR-engineered macrophages and boosts their cytotoxicity against GC cells. These findings support a combinatorial strategy integrating CAR-macrophage therapy with immune-potentiating agents, offering a promising alternative for GC treatment.

In summary, we identify an intratumoral microbe-immune axis in GC. L. salivarius-derived bEVs selectively target tumor cells and reshape the immune microenvironment. The bEV-carried enzyme 2,3-BdpM activates FPR1 on macrophages, triggering MAPK and NF-κB signaling and promoting a pro-inflammatory, antitumor phenotype. Combining 2,3-BdpM-bEVs with ICB yields synergistic antitumor effects, supporting bEV-based combination immunotherapies. L. salivarius may also serve as a predictive biomarker for ICB responsiveness.

Limitations of the study

We acknowledge several limitations in this study. It should be recognized that although the in vivo dosage of bEVs was informed by previous reports, the physiological concentration of bEVs derived from L. salivarius in the gastric milieu remains unclear and warrants further investigation. In addition, the molecular determinants underlying the preferential association of bEVs with GC cells have not yet been delineated and the precise interaction targets require additional mechanistic exploration.

Resource availability

Lead contact

Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Yanfeng Hu (banby@smu.edu.cn).

Materials availability

All unique/stable reagents generated in this study are available from the lead contact with a completed materials transfer agreement.

Data and code availability

• •All data reported in this paper will be shared by the lead contact upon request. The 16S rRNA and transcriptomic data have been deposited in CNGBdb (CNP0007169) with sample IDs sub070112, sub070150, and sub080744. The single-cell RNA-seq dataset (GSE183904) was obtained from a publicly available repository.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.
• •No custom computer code was generated or used in this study.

Acknowledgments

This study received financial support from the National Natural Science Foundation of China (82272062); the Guangdong Basic and Applied Basic Research Foundation (2022A1515220014); the Key Areas Research and Development Programs of Guangdong Province (2023B1111050009); the Guangdong Provincial Key Laboratory of Precision Medicine for Gastrointestinal Cancer (2020B121201004); the Key Clinical Technique of Guangzhou (2023P-ZD01); and the Development Project of the Gastrointestinal Oncology Research Institute of Nanfang Hospital (J1010171102). The illustrations in the graphical abstract and Figures 1A, 2C, 3C, 4A, 5C, 6K, 7J, and 7M were designed using BioRender.com.

Author contributions

X.Y. and Y.H. conceptualized the study. Y.R., J.L., J.J., and L.W. conducted the experiments. X.M., Y.W., L.X., Y.X., H.H., and C.Z. processed the data. X.Y. and Y.R. drafted the manuscript. All authors contributed to the revision and refinement of the manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
FITC CD45 Clone S18009F	Biolegend	Cat# 157214; RRID:AB_2894427
PE-Cy7 CD45 Clone S18009F	Biolegend	Cat# 157206; RRID:AB_2860726
BV421 CD3 Clone 17A2	Biolegend	Cat# 100228; RRID:AB_2562553
APC-Cy7 CD3 Clone 17A2	Biolegend	Cat# 100222; RRID:AB_2242784
PE CD4 Clone GK1.5	Biolegend	Cat# 100408; RRID:AB_312693
FITC CD8 Clone 53-6.7	Biolegend	Cat# 100706; RRID:AB_312745
APC CD8 Clone 53-6.7	Biolegend	Cat# 100712; RRID:AB_312751
FITC CD11c Clone N418	Biolegend	Cat# 117306; RRID:AB_313775
PerCP/Cy5.5 CD11c Clone N418	Biolegend	Cat# 117328; RRID:AB_2129641
BV421 F4/80 Clone BM8	Biolegend	Cat# 123132; RRID:AB_11203717
APC CD86 Clone A17199A	Biolegend	Cat# 159216; RRID:AB_3106041
FITC CD206 Clone C068C2	Biolegend	Cat# 141703; RRID:AB_10900988
PE CD206 Clone C068C2	Biolegend	Cat# 141706; RRID:AB_10895754
PE Arginase 1 Clone W21047I	Biolegend	Cat# 165803; RRID:AB_3068116
PE-Cy7 iNOS Clone W16030C	Biolegend	Cat# 696813; RRID:AB_3106153
PE-Cy7 IFN-γ Clone XMG1.2	Biolegend	Cat# 505826; RRID:AB_2295770
BV421 IFN-γ Clone XMG1.2	Biolegend	Cat# 505829; RRID:AB_10897937
APC IFN-γ Clone XMG1.2	Biolegend	Cat# 505810; RRID:AB_315404
BV421 Granzyme B Clone GB11	Biolegend	Cat# 515410; RRID:AB_3106190
APC Granzyme B Clone QA16A02	Biolegend	Cat# 372204; RRID:AB_2687028
Purified CD16/32 Clone 93	Biolegend	Cat# 101302; RRID: AB_312801
Zombie R718 Fixable Viability Kit	Biolegend	Cat# 423116
Ki-67 Recombinant Rabbit Monoclonal Antibody	Thermo Fisher Scientific	Cat# MA5-14520; RRID:AB_10979488
CK7 Mouse mAb	Immunoway	Cat# YM3054
Anti-Cytokeratin 7 Rabbit mAb	Abcam	Cat# ab199718; RRID: AB_3076227
CD68 Monoclonal Antibody	Thermo Fisher Scientific	Cat# 14-0688-82; RRID:AB_11151139
F4/80 Monoclonal Antibody	Thermo Fisher Scientific	Cat# 14-4801-82; RRID:AB_467558
NLRP1 Rabbit pAb	Immunoway	Cat# YN6054
F11R Rabbit pAb	Immunoway	Cat# YT5479
CXADR Rabbit pAb	Immunoway	Cat# YT0631
FPR1 Rabbit pAb	Immunoway	Cat# YT1768
FPR2 Rabbit pAb	Immunoway	Cat# YN2518
Phospho-STAT3 (Y705) Antibody	Abmart	Cat# T56566; RRID:AB_2936393
Phospho-STAT1 (Tyr701) Antibody	Abmart	Cat# TP56498; RRID:AB_3076705
JNK1/2/3 Antibody	Abmart	Cat# T40073; RRID:AB_2937020
MEK1/2 Antibody	Abmart	Cat# T55168; RRID:AB_3713056
ERK1/2 Antibody	Abmart	Cat# T40071; RRID:AB_2936996
p38 MAPK Antibody	Abmart	Cat# T55600; RRID:AB_2936971
NF-κB p65 Antibody	Abmart	Cat# T55034; RRID:AB_2937049
Phospho-NF- kappaB p65 (Ser536) Antibody	Abmart	Cat# TP56372; RRID:AB_2936854
IKB alpha Antibody	Abmart	Cat# T55026; RRID:AB_2937048
Phospho-IKB alpha (Ser32/Ser36) Antibody	Abmart	Cat# TA2002; RRID:AB_3712518
Alpha Tubulin Monoclonal antibody	Proteintech	Cat# 66031-1-Ig; RRID:AB_11042766
GST Tag Monoclonal antibody	Proteintech	Cat# 66001-2-Ig; RRID:AB_2881488
His Tag (PT0481R) PT Rabbit mAb	Immunoway	Cat# YM8314; RRID:AB_3718184
Mouse IgG Antibody	Sigma-Aldrich	Cat# 12–371; RRID:AB_145840
HRP-conjugated Goat Anti-Rabbit IgG	Fdbio Science	Cat# FD0115
HRP-conjugated Goat Anti-Mouse IgG	Fdbio Science	Cat# FD0114
IPKine HRP, Goat Anti-Mouse IgG LCS	Abbkine	Cat# A25012; RRID:AB_2737290
IPKine HRP, Goat Anti-Mouse IgG LCS	Abbkine	Cat# A25222; RRID:AB_2922982
Dynabeads Mouse T-Activator CD3/CD28 beads	Thermo Fisher Scientific	Cat# 11453D
Anti-rabbit IgG (H + L), F(ab')2 Fragment (Alexa Fluor 555 Conjugate)	Cell Signaling Technology	Cat# 4413; RRID:AB_10694110
Anti-mouse IgG (H + L), F(ab')2 Fragment (Alexa Fluor 488 Conjugate)	Cell Signaling Technology	Cat# 4408; RRID:AB_10694704
Anti-rat IgG (H + L), (Alexa Fluor 488 Conjugate)	Cell Signaling Technology	Cat# 4416; RRID:AB_10693769
Anti-Mouse PD-1 Antibody	MCE	Cat# HY-P99144; RRID: AB_3718703
Deposited data
16s rRNA-sequencing data	This paper	CNGBdb: CNP0007169; sub070150
RNA-sequencing data	This paper	CNGBdb: CNP0007169; sub070112 and sub080744
Single-cell RNA-sequencing data	Kumar et al.56	GEO: GSE183904; DOI: https://doi.org/10.1158/2159-8290.CD-21-0683
Experimental models: Cell lines
AGS	Procell	Cat# CL-0022
HGC-27	Procell	Cat# CL-0107
GES-1	Procell	Cat# CL-0563
HEK 293T	Procell	Cat# CL-0005
Experimental models: bacteria strains
H. pylori SS1	Biosci Biotechnology	Cat# F1006
Ligilactobacillus salivarius	Beijing Beina Chuanglian Biotechnology Institute	Cat# BNCC367991; BNCC194724
Experimental models: Mouse strains
NOD-SCID mice	Zhuhai Baishitong Biotechnology	N/A
Atp4b-iCre mice (T054674)	GemPharmatech	Strain# T054674; RRID: IMSR_GPT:T054674
Tff1-flox mice (T016018)	GemPharmatech	Strain# T016018; RRID: IMSR_GPT:T016018
Trp53-flox mice (#008462)	Jackson Laboratory	Strain# 008462; RRID:IMSR_JAX:008462
Myc-LSL mice (#035557)	Jackson Laboratory	Strain# 035557; RRID:IMSR_JAX:035557
Chemicals, peptides, recombinant proteins and others
L.salivarius-specific probes labeled with CY5	IEMed	N/A
L.salivarius-specific probes labeled with CY3	IEMed	N/A
Columbia blood agar plates	HuanKai	Cat# 024070
MRS broth	Solarbio	Cat# M8540
N-methyl-N-nitrosourea (MNU)	Meryer	Cat# M75416
Ampicillin	Macklin	Cat# 69-53-4
Vancomycin	Macklin	Cat# 1404-90-6
Neomycin sulfate	Macklin	Cat# 1405-10-3
Metronidazole	Macklin	Cat# 443-48-1
Citric acid	MCE	Cat# HY-N1428
Phenyllactic acid	Sigma-Aldrich	Cat# 103-82-2
2-Hydroxycaproic acid	Sigma-Aldrich	Cat# 6064-63-7
FBS	Procell	Cat# 164210
1% penicillin-streptomycin	NCM Biotech	Cat# C100C5
Gastric Cancer Organoid Culture Medium	Novoprotein	Cat# OCMHC07
Gastric Cancer Organoid Culture Medium	Absin	Cat# abs9534
Collagenase Type IV	Gibio	Cat# 17104019
Collagenase Type II	Gibio	Cat# 17101015
Percoll cell separation solution	Biosharp	Cat# 65455-52-9
Typsin-EDTA	Gibio	Cat# 25200072
DAPI	Beyotime	Cat# C1002
D-luciferin potassium salt	GlpBio	Cat# GC43496
Citrate-EDTA Antigen Retrieval Solution	Biosharp	Cat# BL637A
Endogenous Peroxidase Blocking Buffer	Beyotime	Cat# P0100A
DIO	Beyotime	Cat# C1038
DIL	Solarbio	Cat# D8700
DIR	MCE	Cat# HY-D1048
Triton X-100	Solarbio	Cat# T8200
Anti-fade mounting medium	Solarbio	Cat# S2110
Goat serum	Bioss	Cat# C-0005
Collagenase Type IV	Yeasen	Cat# 40510ES60
Collagenase Type II	Yeasen	Cat# 40508ES60
DNase I	Solarbio	Cat# D8071
50 kDa molecular weight cutoff	Merck Millipore	Cat# GE28-9322-36
Dispase	Gibco	Cat# 17105041
Matrigel	Corning	Cat# 356231
Ficoll-Paque	Solarbio	Cat# P8860
RBC lysis buffer	Solarbio	Cat# R1011
Recombinant Mouse M-CSF	Novoprotein	Cat# CB34
Lipofectamine RNAiMAX	Thermo Fisher Scientific	Cat# 13778075
puromycin	Solarbio	Cat# P8230
RIPA Lysis Buffer	CWBIO	Cat# CW2334S
PVDF membrane	Merck Millipore	Cat# ISEQ85R
Recombinant human GST-tagged FPR1 protein	Abcam	Cat# ab158465
Recombinant human PGAM1 protein	MCE	Cat# HY-P701962
Dithiothreitol	Amresco	Cat# M109
Ammonium bicarbonate	Sigma-Aldrich	Cat# A6141
Iodoacetamide	Amresco	Cat# M216
Trypsin	Promega	Cat# V5280
BL21(DE3) competent E. coli	Thermo Fisher Scientific	Cat# C600003
LB medium	Biosharp	Cat# BL1056A
Isopropyl-β-D-thiogalactopyranoside	Solarbio	Cat# I8070
Lipo2000 transfection reagent	Biosharp	Cat# BL623B
Lentivirus concentrator	ECOtop	Cat# ES-7011
Polybrene	Solarbio	Cat# H8761
Critical commercial assays
Fluorescence in Situ Hybridization Kit for RNA	Beyotime	Cat# R0306S
All-In-One DNA/RNA Mini-Preps Kit	Sangon Biotech	Cat# B618203
2× Hieff Canace AdvanceFast PCR Master Mix	Yeasen	Cat# 10163ES08
Dead Cell Removal Kit	Miltenyi Biotec	Cat# 130-090-101
EasySep Mouse CD45 Positive Selection Kit	Stemcell	Cat# 18945
BD Cytofix/Cytoperm Fixation/Permeabilization Kit	BD Biosciences	Cat# 554714
Dako REAL EnVision Detection System	Dako	Cat# K5007; RRID:AB_2888627
BeyoBCA Protein Assay Kit	Beyotime	Cat# P0398S
BeyoClick EdU Cell Proliferation Kit with Alexa Fluor 555	Beyotime	Cat# C0075S
Cell Counting Kit-8 (CCK-8)	Dojindo	Cat# CK04
Annexin V-FITC/PI Apoptosis Kit	Elascience	Cat# E-CK-A211
Annexin V-APC/PI Apoptosis Kit	Elascience	Cat# E-CK-A217
EasySep Mouse Naive CD8+ T cell Isolation Kit	Stemcell	Cat# 19858
CytoTox 96 Non-Radioactive Cytotoxicity Assay Kit	Promega	Cat# G1780
NEBNext Ultra RNA Library Prep Kit	New England Biolabs	Cat# E7770L
GST Protein Pull-Down Kit	Elabscience	Cat# EA-IP-K008
GST-tagged Protein Purification Kit	Elabscience	Cat# EA-TP-K004
His-tagged Protein Purification Kit	Elabscience	Cat# EA-TP-K005
Oligonucleotides and recombinant DNA
Table S3	N/A	N/A
L.salivarius-specific probe sequene	IEMed	N/A
L.ruminis-specific probe sequene	IEMed	N/A
L.animalis-specific probe sequene	IEMed	N/A
L.salivarius-specific primer sequene	IGE	N/A
gpmA primer sequene	IGE	N/A
Mouse shRNA FPR1	IGE	N/A
Mouse shRNA FPR2	IGE	N/A
Mouse CAR-Claudin18.2 plasmid	iCarTab	N/A
pMD2.G	Sangon Biotech	Cat# A338961; RRID:Addgene_12259
psPAX2	Sangon Biotech	Cat# A338972; RRID:Addgene_12260
pGEX-4T1 vector	IGE	N/A
GST-2,3-BdpM plasmid	IGE	N/A
GST-2,3-Lysm plasmid	IGE	N/A
Software and algorithms
GraphPad Prism	GraphPad	RRID:SCR_002798
FlowJo (v10)	TreeStar	RRID:SCR_008520
Photoshop	Adobe	RRID:SCR_014199
Illustrator	Adobe	RRID:SCR_010279
ImageJ	National Institutes of Health	RRID:SCR_003070

Experimental model and study participant details

Mice

C57BL/6J and NOD-SCID mice (15–25 g) were purchased from Zhuhai Baishitong Biotechnology (Zhuhai, China). 615 mice were obtained from Shulaibao Biotechnology (Chengdu, China). Atp4b-iCre (T054674) and Tff1-flox (T016018) mice, both on a C57BL/6 genetic background, were sourced from GemPharmatech Co., Ltd. (Jiangsu, China). Trp53-flox mice (#008462), on a C57BL/6 background, and Myc-LSL mice (#035557), on an FVB/N congenic background, were acquired from The Jackson Laboratory. All mice were maintained in a specific pathogen-free environment, and the experimental procedure was authorized by the Animal Ethics Committee of Nanfang Hospital, Southern Medical University (IACUC-LAC-20240423-002). Experiments were conducted using gender-matched mice, aged between 6 and 12 weeks, comprising both males and females. All mice were kept in filter cages containing corn cob bedding, with access to standard chow and water available ad libitum. A maximum of five mice were housed per cage. Cages were cleaned weekly in a laminar flow hood, and the mice were kept on a 12-h light-dark cycle. The use of mice in these experiments adhered to all ethical regulations related to animal research.

Human tissues

In this study, patients with GC from Nanfang Hospital, Southern Medical University (Guangzhou, China) were enrolled as a validation cohort to verify bacterial strains identified by 16S rRNA sequencing using FISH. Paired tumor and adjacent non-tumorous tissues were collected intraoperatively from 68 GC patients. Tumor samples were obtained from macroscopically confirmed lesions, while non-tumorous tissues were harvested from sites located at least 8 cm from the tumor margin, following established protocols. Among the 68 patients, 46 had not received immunotherapy, and 22 had been treated with immunotherapy prior to surgery. Of the latter group, 17 were classified as responders and 5 as non-responders, based on the iRECIST (immune Response Evaluation Criteria in Solid Tumors) guidelines. All patient samples were collected with informed consent and in accordance with the Declaration of Helsinki. The study was approved by the Ethics Committee of Nanfang Hospital (NFEC-2025-194). Detailed clinical information is provided in Table S4.

Bacteria and culture

H. pylori SS1 (HP.SS1, F1006) was obtained from Biosci Biotechnology Co., Ltd. (Hangzhou, China), while L. salivarius strains BNCC367991 and BNCC194724 were purchased from the Beijing Beina Chuanglian Biotechnology Institute (Beijing, China). HP.SS1 was cultured under microaerophilic conditions on Columbia blood agar plates, while L. salivarius was revived and expanded anaerobically in MRS broth (Solarbio, China). All cultures were maintained at 37°C, and bacterial density was monitored using a hemocytometer. Bacterial suspensions were then diluted in PBS to a final concentration of 2 × 10⁹ CFU/mL. For L. salivarius intervention in the GC mouse model, approximately 100 μL of the bacterial suspension was administered per mouse by oral gavage, repeated over multiple dosing sessions.

Tumor models and treatments

In the GC mouse model induced by MNU+HP.SS1, C57BL/6 mice (aged 6–8 weeks, weighing approximately 20 g) were utilized for model development. HP.SS1 was administered via oral gavage (0.2 mL of ∼10⁸ CFU/mouse) three times weekly for 2 weeks. Subsequently, MNU was provided in tap water (300 ppm) ad libitum for 6 weeks over a 12-week span, freshly prepared thrice weekly in light-protected bottles. Additionally, mice received oral gavage of 300 ppm MNU (Meryer, China) in 5% hypertonic saline (0.5 mL/mouse) thrice weekly for 6 weeks. A fasting-feeding-fasting regimen was applied every alternate day during MNU exposure for 12 weeks. After induction, mice were maintained under standard conditions until 52 weeks of age to establish the MNU+HP.SS1-induced GC model.

To establish the triple-gene GC mouse model, Atp4b-iCre, Tff1-flox, and Trp53-flox mice were intercrossed to generate Atp4b-iCre^+/−, Tff1^flox/-, and Trp53^flox/- (ATP) mice. In parallel, Tff1-flox, Trp53-flox, and Myc-LSL mice were crossed to obtain Tff1^flox/flox, Trp53^flox/flox, and Myc^LSL/LSL (TPM) mice. Subsequently, ATP and TPM mice were interbred to generate Atp4b-iCre^+/−, Tff1^flox/flox, Trp53^flox/flox, and Myc^LSL/- (ATPM) mice. The breeding process involved multiple generations over a three-year period. Considering that Myc amplification was coupled with luciferase (Luc), the successful establishment of the triple-genotype GC mouse model was confirmed at 6 weeks of age using in vivo bioluminescence imaging and pathological staining.

To evaluate therapeutic efficacy, we employed multiple murine GC models: (1) the ATPM spontaneous GC model, (2) subcutaneous implantation of MFC cells, (3) subcutaneous implantation of primary tumor cells derived from ATPM mice, and (4) subcutaneous implantation of primary tumor cells from Helicobacter pylori SS1 + MNU-induced GC mice. Mice were randomized into four treatment groups: Control, L. salivarius, anti-PD-1 antibody, and combination therapy (L. salivarius + anti-PD-1). To establish an antibiotic-treated (ABX) GC model, mice received a cocktail of antibiotics (Macklin, China)-ampicillin (1.86 mg), vancomycin (0.96 mg), neomycin sulfate (1.86 mg), and metronidazole (1.86 mg)-in 250 μL of sterile water, administered via oral gavage three times per week for two weeks. Concurrently, mice were given free access to drinking water containing 1 g/L of each antibiotic for two weeks. Following ABX treatment, 100 μL of L. salivarius suspension (2 × 10⁸ CFU) was administered via oral gavage on three occasions. Anti-PD-1 antibody (MCE, USA) was intraperitoneally injected weekly (100 μg/mouse) for three consecutive weeks. Mice were euthanized, and tissues were collected at week 4 post-intervention. In parallel, the same GC models were used to assess the effects of L. salivarius-derived bacterial extracellular vesicles (bEVs) in combination with anti-PD-1. Mice were divided into four groups: Control, bEVs, anti-PD-1, and bEVs + anti-PD-1. bEVs (25 μg/mouse) were administered intraperitoneally once weekly for two weeks, and anti-PD-1 was given as above. Similarly, to evaluate the combined effect of 2,3-BdpM and anti-PD-1, mice were assigned to four groups: Control, 2,3-BdpM, anti-PD-1, and 2,3-BdpM + anti-PD-1. 2,3-BdpM (25 μg/mouse) was administered intraperitoneally once weekly for three weeks, along with anti-PD-1 antibody following the same regimen.

In the subcutaneous tumor model of GC mice treated with CAR-macrophages (CAR-Ms), primary GC cells isolated from the triple-gene GC mouse were implanted subcutaneously (5 × 10⁶ cells/mouse) into 6-week-old NOD-SCID mice. After 10 days of tumor establishment, CAR-Ms were treated with 1 μg/mL 2,3-BdpM or Fpr1 knockdown viruses for 48 h (groups: UTD, NC,2,3-BdpM, shFPR1, and 2,3-BdpM + shFPR1), followed by intraperitoneal injection (5 × 10⁶ cells per mouse). Tumor progression was monitored using bioluminescence imaging three times over 2 weeks. To evaluate the potential toxicity of CAR-Ms therapy, we conducted blood biochemistry, hematological analyses, and histopathological evaluation (H&E staining) of the key organs.

Mice were euthanized in accordance with institutional ethical guidelines when any of the following humane endpoints were reached: tumor volume exceeded 2000 mm³, maximum tumor diameter surpassed 2.0 cm, body weight loss exceeded 20% of baseline, or signs of severe physical distress were observed. For survival analysis of subcutaneous tumor-bearing mice, reaching a tumor volume >2000 mm³ or a diameter >2.0 cm was considered as the event of death.

Cell lines culture

Human GC cell lines (AGS and HGC-27) and the normal gastric epithelial cell line GES-1 were cultured in DMEM supplemented with 10% fetal bovine serum (FBS; Procell, China) and 1% penicillin-streptomycin (M&C Gene Technology, China) at 37°C in a humidified incubator with 5% CO₂. Cell lines were obtained from Procell (authenticated) and were verified to be mycoplasma-free by PCR during culture. Cells were passaged upon reaching confluence. Cells were passaged upon reaching confluence. To establish primary GC cells from mice, tumor tissues were harvested from ATPM transgenic mice and Helicobacter pylori SS1 + MNU-induced GC mice. Tumors were minced into ∼1 mm³ fragments and digested with type II and IV collagenases (Gibco, USA). Following filtration, isolated cells were cultured and passaged for five generations. The identity of primary GC cells derived from ATPM mice was confirmed via bioluminescence imaging due to the overexpression of a Myc-luciferase fusion gene. For subcutaneous tumor model establishment, primary GC cells and murine forestomach carcinoma (MFC) cells were digested with trypsin-EDTA, washed twice with sterile PBS, and resuspended at 5 × 10⁶ cells/mL. A 100 μL cell suspension was then injected subcutaneously into the axillary region of each mouse. For xenograft models, primary tumor cells were similarly implanted into NOD-SCID mice.

Organoid culture and treatment

Tumor tissues from MNU+HP.SS1 GC mice, ATPM GC mice, and patients with GC were washed with sterile pre-chilled PBS, to remove blood and debris. The tissues were cut into small pieces (approximately 1–2 mm³) and incubated at 37°C for 45 min with gentle agitation in an enzymatic solution comprising 1 mg/mL Collagenase IV and 0.5 mg/mL Dispase (both from Gibco, USA). After digestion, cells were filtered, washed, and resuspended in GC organoid culture medium (Absin, China; Novoprotein, China) supplemented with 10% Matrigel (Corning, USA). Aliquots of 25 μL were plated into pre-warmed 24-well plates and allowed to solidify. Plates were incubated at 37°C in an atmosphere containing 5% CO₂, with the culture medium replaced every 2–3 days. Organoids typically formed within 5–7 days. Mature organoids were passaged mechanically. The identity and proliferative status of GC organoids were confirmed using H&E and immunofluorescence staining for Ki-67 and CK-7.

In the organoid experiments using either mouse-derived or human-derived models with L. salivarius medium or bEVs intervention, the intervention concentration of L. salivarius culture supernatant (pH adjusted to 7.4) was 10 or 20 μg/mL, while the intervention concentration of bEVs was 1 or 2 μg/mL. After the intervention, organoid volume changes were observed via microscopy on days 1, 3, and 5 to assess proliferation. For the co-culture experiment of splenocytes and ATPM GC mouse-derived organoids, the splenocytes concentration was 2 × 10⁴, with L. salivarius culture supernatant (LSCS) intervention at 1 μg/mL and bEVs intervention at 0.1 μg/mL. After a 36-h intervention, organoid volume changes were observed microscopically, and the proportion and functional changes of immune cells in splenocytes were detected using flow cytometry. The following groups: PBS, LSCS, bEVs, splenocytes, splenocytes + MRS, splenocytes + LSCS, splenocytes + cEVs, and splenocytes + bEVs. For the CAR-macrophage intervention mouse-derived organoid experiment, 2 × 10⁴ CAR-Ms were treated with 0.1 μg/mL of 2,3-BdpM or Fpr1 knockdown virus for 48 h (groups: UTD, NC, 2,3-BdpM, shFPR1, and 2,3-BdpM + shFPR1), followed by co-culture with organoids. The proliferation of organoids was observed at 16, 24, and 30 h.

Method details

16S rRNA sequencing and analysis

Gastric juice was collected from both HP.SS1+MNU-induced in situ GC mice and age-matched (55-week-old) control mice. Under gas anesthesia, the pylorus and esophagus were ligated. Small incisions were made at the lower esophagus and upper pylorus to insert soft tubing from scalp needles, which were sutured in place to prevent leakage. Using an anesthetic infusion pump, sterile saline was slowly infused through the upper gastric tube, while gastric juice was simultaneously collected through the lower tube. Collection was performed over 1 h for each mouse, yielding approximately 2 mL of gastric juice. The bacterial composition of the gastric juice was analyzed via 16S rRNA gene sequencing, following methods detailed in our previous publications, with technical support provided by LC-Bio Technology Co., Ltd (Hangzhou, China).²⁵

Reverse transcription-quantitative polymerase chain reaction (RT-qPCR)

Total DNA from gastric carcinoma tissues and their matched normal counterparts, along with total RNA isolated from L.salivarius, was extracted using the All-In-One DNA/RNA Mini-Preps Kit (Sangon). qPCR analysis was carried out with SYBR Green Real-Time PCR Master Mix (Yeasen) on an Applied Biosystems QuantStudio 5 system, with all reactions performed in triplicate. Primer sequences used for the qPCR assays are provided in Table S3.

RNA FISH assay

FISH was performed on paraffin-embedded gastric tissues (normal and cancerous) from humans and mice using L. salivarius-specific probes labeled with CY5 or CY3 (IEMed, China). Tissue sections were incubated overnight at 56°C in a hybridization buffer (comprising 0.9 M NaCl, 20 mM Tris-HCl, 0.1% SDS, and 10% formamide) with fluorescent probes. Following hybridization, the sections were washed twice, counterstained with DAPI (Beyotime, China), and then mounted with an antifade reagent. Imaging was conducted using an automated inverted microscope (Nikon Ti2-E).

In vivo imaging

Luciferase activity at the Myc amplification locus in the GC mouse model was visualized using an in vivo small animal imaging system to track tumor progression. Mice were intraperitoneally injected with 75 mg/kg D-luciferin potassium salt (GlpBio, USA). Anesthesia was induced with isoflurane, and the mice were then imaged using the Newton 7.0 system (Vilber Bio Imaging) for fluorescence intensity analysis.

For bEVs labeled with the DIR dye (Beyotime, China), 1 mg/kg was administered intraperitoneally, and in vivo imaging was conducted at 30, 60, and 120 min post-injection at 780 nm. Key organs, such as the stomach, were harvested for imaging to assess the preferential accumulation of bEVs in GC tissue.

Metabolomics analysis

L. salivarius culture supernatants (100 μL) were precipitated with pre-chilled methanol (−80°C), vortexed, incubated at low temperature, centrifuged, and lyophilized for storage at −20°C. Pooled QC samples were prepared to monitor instrument performance, and blanks were used for background correction. Untargeted metabolomics was performed using LC-MS/MS on a Q Exactive HF-X (Thermo, USA) coupled to an Ultimate 3000 UHPLC with an ACQUITY UPLC HSS T3 column (2.1 mm × 100 mm, 1.8 μm, Waters). Data were acquired in positive and negative ion modes with full-scan and DDA MS/MS. Gradient elution used 0.1% formic acid in water (A) and acetonitrile (B) over 15 min at 250 μL/min. Raw data were processed in Compound Discoverer for peak extraction, alignment, quantification, and molecular formula prediction, with metabolite identification via mzCloud, MZvault, and Chemspider; blanks were used to remove background ions. PCA visualized overall data distribution, while OPLS-DA achieved group separation and calculated VIP scores. Paired Student’s t test identified significant metabolites (VIP > 1, p < 0.05), which were annotated with KEGG for pathway and enrichment analyses.

Immunohistochemistry and immunofluorescence

Paraffin-embedded tumor tissues from GC mice were subjected to Ki-67 and CK-7 immunohistochemical staining to assess tumor cell proliferation and confirm the adenocarcinoma phenotype. Tissue sections were first deparaffinized with xylene, then rehydrated using a graded ethanol series and rinsed in PBS. Antigen retrieval was carried out with a citrate-EDTA solution (Biosharp, China) under microwave heating, followed by cooling. Endogenous peroxidase activity was blocked by incubating the sections in a peroxidase-blocking buffer (Beyotime, China) for 10 min. Afterward, sections were incubated with 10% goat serum for 30 min to block non-specific binding. Primary antibodies against Ki-67 (1:200, Thermo Fisher Scientific, USA) and CK-7 (1:200, Immunoway/Abcam, USA) were applied overnight at 4°C. After incubation, sections were treated with HRP-conjugated secondary antibodies (Dako, Denmark). Immunoreactivity was detected using DAB, and nuclei were counterstained with hematoxylin. Finally, the sections were dehydrated, mounted, and analyzed under a light microscope (Leica RM2245). Digital images were captured with a pathology scanner (Teksqray SQS-40R), and positive staining areas were quantified using Image-Pro Plus 6.0 software.

Immunofluorescence staining was conducted on human and murine GC tissues, organoids, and cell cultures, including cells labeled with membrane dyes DIO (Beyotime, China) or DIL (Solarbio, China). Paraffin-embedded tissue sections were subjected to a deparaffinization process and rehydrated with xylene and sequential ethanol solutions, followed by antigen retrieval via microwave heating. Cells were fixed using paraformaldehyde, washed with PBS, and permeabilized with 0.25% Triton X-100 in PBS for 10 min at room temperature. After blocking with goat serum for 30 min, primary antibodies were applied overnight at 4°C. The antibodies used included CK-7 (1:200, Immunoway, USA), Ki-67 (1:500, Thermo Fisher Scientific), FPR1 (1:200, Immunoway, USA), F4/80 (1:200, Thermo Fisher Scientific), and CD68 (1:1000, Thermo Fisher Scientific). After washing with PBS, the samples were incubated with fluorescence-conjugated secondary antibodies (Alexa Fluor 488 or 555; Cell Signaling Technology, USA) for 1 h at room temperature. Following additional PBS washes, nuclei were stained with DAPI and mounted using an anti-fade medium (Solarbio, China). Fluorescent and confocal images were obtained using a fluorescence microscope (Nikon Ti2-E) or a confocal laser scanning microscope (Nikon AXNIS-Elements 5.4). Positive cell ratios and co-localization areas were quantitatively analyzed using Image-Pro Plus 6.0 software.

Flow cytometry

Immune cells were isolated from spleens of tumor-bearing mice, in vitro cultured macrophages, sorted CD8⁺ T cells, and tumor tissues. Subcutaneous and orthotopic gastric tumor tissues were minced into 1–2 mm³ fragments and enzymatically digested in RPMI 1640 containing 1 mg/mL collagenase II, 1 mg/mL collagenase IV (Yeasen, China), and 200 μg/mL DNase I (Solarbio, China) at 37°C for 60 min on a shaker. The digested mixture was passed through a 100-μm cell strainer to remove debris. Cells were pelleted (450 × g, 5 min), washed with PBS, and subjected to density gradient centrifugation (80%/40% Percoll, Biosharp) to remove red blood cells and debris. Dead cells were depleted using a Dead Cell Removal Kit (Miltenyi Biotec). For orthotopic tumors, CD45⁺ immune cells were enriched using the EasySep Mouse CD45 Positive Selection Kit (Stemcell Technologies) due to low yield and cellular complexity. Cells were resuspended in FACS buffer for downstream staining. For subcutaneous tumors, CD45⁺ enrichment were not performed. Cells were then resuspended in FACS buffer and stained with surface antibodies against CD45, CD3, CD4, CD8, CD11c, F4/80, CD86, and CD206 (BioLegend, USA) for 30 min at room temperature. For intracellular cytokine staining, surface-stained cells were fixed and permeabilized using a commercial kit (BD Biosciences, USA), then stained for iNOS, Arg-1, IFN-γ and Granzyme B (GZMB; BioLegend, USA). Flow cytometry was performed on a CytoFLEX instrument (Beckman, BF15167), and data were analyzed using FlowJo software (v10).

Bacterial outer membrane vesicle isolation

The purification of bEVs was performed following previously described protocols with slight adjustments.^57,58 Briefly, L. salivarius was cultivated anaerobically in MRS broth at 37°C. During the logarithmic growth phase, culture supernatants were collected and subjected to centrifugation at 10,000 × g for 30 min at 4°C to eliminate bacterial cells. The supernatant was subsequently passed through a 0.22 μm membrane filter to remove any remaining cells and debris. The clarified fluid was concentrated using a 50 kDa molecular weight cutoff centrifugal filter device (Merck Millipore, USA) at 10,000 × g for 50 min. Following concentration, the samples were transferred to ultracentrifuge tubes and centrifuged at 130,000 × g for 2 h to pellet the bEVs. The resulting vesicle pellet was resuspended in 1 mL PBS, assessed for protein content, and stored at −80°C for subsequent experiments. Protein quantification of the bEVs was performed utilizing the BeyoBCA Protein Assay Kit (Beyotime, China) in accordance with the manufacturer’s protocol.

Transmission electron microscopy and nanoparticle tracking analysis

For ultrastructural visualization, 10 μL of L. salivarius-derived bEVs suspension was applied onto carbon-coated copper grids and left to stand for 1 min. The excess solution was carefully blotted away using filter paper. Subsequently, the grids were negatively stained with 10 μL of 2% uranyl acetate for 1 min. After air drying at room temperature, samples were examined with a transmission electron microscope (Hitachi HT-7700) operating at 80 kV.

For particle size distribution analysis, bEV samples were adjusted to a volume of 30 μL through dilution. Instrument calibration was performed using standardized particles to ensure measurement accuracy. To prevent clogging of the injection needle, serial dilutions were prepared. The size distribution and particle concentrations were then assessed using a NanoFCM N30E flow nanoanalyzer.

CCK-8 cell viability assay

Cell viability and proliferation of AGS, HGC-27, and GES-1 cells were determined using the Cell Counting Kit-8 (CCK-8, Dojindo). Cells were seeded in 96-well plates at a density of 5 × 10³ cells per well and treated with LSCS, bEVs, or GST-2,3-BdpM for the indicated times. Subsequently, 10 μL of CCK-8 solution was added to each well, followed by incubation at 37°C for 1 h. Absorbance at 450 nm was measured using a microplate reader. Cell proliferation inhibition was calculated based on relative absorbance compared with the control group.

EdU cell proliferation assay

Cell proliferation was evaluated using the BeyoClick EdU Cell Proliferation Kit with Alexa Fluor 555 (Beyotime, China) according to the manufacturer’s instructions. AGS and HGC-27 cells cultured in confocal dishes were incubated with 10 μM EdU working solution at 37°C under 5% CO₂ for 4 h. Following incubation, cells were fixed with 4% paraformaldehyde, permeabilized with 0.3% Triton X-100, and subsequently stained with Hoechst 33342 to visualize nuclei. Fluorescence images were captured with a fluorescence microscope, and the percentage of EdU-positive cells was quantified using Image-Pro Plus 6.0 software.

Apoptosis analysis

Apoptosis was evaluated using Annexin V-FITC or APC apoptosis detection kits (Elascience, China). Cells were harvested, washed twice, and then stained with Annexin V and PI following the manufacturer’s instructions. The stained cells were analyzed using a CytoFLEX flow cytometer.

Extraction of splenocytes and isolation of naive CD8⁺ T cells

Spleen tissue from ATPM mice was collected, homogenized, and passed through a 70 μm filter to prepare a single-cell suspension. Ficoll-Paque (Solarbio, China) was then added to the suspension, and the mixture underwent centrifugation at 400 × g for 20 min. After centrifugation, the splenocyte layer was carefully collected, washed, and resuspended to yield purified splenocytes.

Naive CD8⁺ T cells were isolated using the EasySep Mouse Naive CD8⁺ T cell Isolation Kit (Stemcell, Canada) according to the manufacturer’s instructions. The isolation process involved extracting splenocytes and isolating the naive CD8⁺ T cells. A total of 5×10⁴ CD8⁺ T cells were seeded in a 96-well plate and cultured in T cell medium (Lonza, Switzerland). These cells were then activated with 5 μg/mL of anti-CD3/anti-CD28 antibodies (Thermo Fisher Scientific) for 2 days. Following stimulation, the cells were used for co-culture and cytotoxicity assays.

BMDM isolation

Bone marrow cells were extracted from the femurs and tibias of C57BL/6 or ATPM mice. Red blood cells were removed using a lysis buffer (Biosharp, China), and the cell suspension was filtered through a 70 μm strainer, followed by washing. The resulting cell suspension was seeded into BMDM medium supplemented with 10% FBS and 30 ng/mL M-CSF (Novoprotein, China). Cells were cultured for 7 days to induce macrophage differentiation. The resulting BMDMs were harvested for subsequent co-culture and functional assays.

Cytotoxicity assay

Cytotoxicity assays were conducted to assess the tumor-killing ability of CD8⁺ T cells and macrophages co-cultured with primary ATPM GC cells, following the protocol provided by the CytoTox 96 Non-Radioactive Cytotoxicity Assay Kit (Promega, USA). Briefly, the following five experimental conditions were established in triplicate: (A) cell-free control wells containing only culture medium; (B) target cell solvent control wells with untreated ATPM tumor cells and vehicle equivalent to that used in experimental wells; (C) maximum LDH release control was achieved by lysing target cells with 10 μL of 10× lysis buffer, followed by a 45-min incubation before the addition of the detection reagent; (D) experimental wells containing co-cultures of isolated CD8⁺ T cells or BMDMs from ATPM mice with ATPM GC cells at an effector-to-target ratio of 1:5, (E) and effector cell control wells with untreated CD8⁺ T cells or macrophages alone in culture medium. After 12 h of co-culture, supernatants were collected, and absorbance was recorded at 490 nm. Cytotoxicity was then calculated using the formula: Cytotoxicity (%) = [(D - A) - (B - A) - (E − A)]/[(C - A) - (B - A)] × 100.

For CAR-M experiments, CAR-engineered macrophages were cultured alongside ATPM tumor cells in a 1:1 ratio for 8 h under otherwise identical plating and detection conditions.

RNA-sequencing and analysis

To assess the functional impact of bEVs on GC cells and macrophage programming, AGS and GES-1 cells, as well as bone marrow cells from ATPM mice, were treated with bEVs for 48 h, followed by transcriptomic profiling. Total RNA was extracted and assessed for quality and integrity using standard procedures. Libraries were generated from at least 1 μg of high-quality RNA utilizing the NEBNext Ultra RNA Library Prep Kit, with steps including poly(A)+ mRNA enrichment, RNA fragmentation, and subsequent cDNA synthesis. After adaptor ligation and size selection (∼250–300 bp), the libraries underwent PCR amplification, quality assessment, and were subsequently sequenced using the Illumina NovaSeq platform with 150 bp paired-end reads. After sequencing, raw reads were processed to eliminate adaptor contamination and low-quality reads. The resulting clean reads were then aligned to the mouse reference genome using HISAT2 (v2.0.5), and gene expression levels were calculated as FPKM values with FeatureCounts (v1.6.0). Differential expression analysis was performed using either DESeq2 or edgeR, contingent on the number of biological replicates. Genes exhibiting a false discovery rate (FDR) below 0.05 were considered significantly differentially expressed. To uncover critical biological processes, functional enrichment was analyzed via Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses.

shRNA knockdown

BMDMs were serum-starved prior to transfection using a serum-free medium. Transfection was then performed with Lipofectamine RNAiMAX (Thermo Fisher Scientific), delivering either a non-specific control plasmid or plasmids (IGE, China) targeting CXADR, Fpr1, or Fpr2. Transfection was performed for 36 h, followed by a medium change. After an additional 72 h, cells were subjected to puromycin (Solarbio, China) selection for 72 h to isolate GC cells with CXADR overexpression or BMDMs with Fpr1 or Fpr2 knockdown, which were subsequently used for functional analyses.

Western blot

Cells were lysed using RIPA buffer supplemented with protease and phosphatase inhibitors (CWBIO, China). After centrifugation at 12,000 × g for 20 min, the supernatant was harvested. Protein concentrations were quantified with a bicinchoninic acid (BCA) assay kit (Beyotime, China). Equal amounts of protein were separated by electrophoresis on a 10% SDS-polyacrylamide gel and subsequently transferred onto polyvinylidene difluoride (PVDF) membranes (Merck Millipore, USA). Membranes were blocked and then incubated overnight at 4°C with primary antibodies. After washing, membranes were incubated with HRP-conjugated secondary antibodies (Fdbio Science, China) for 1 h at room temperature. Signals were visualized using an Alliance Q9 imaging system, and band intensities were quantified with ImageJ software.

Proteomic analysis

The protein solutions obtained from the GST pull-down assay, bEVs, and L. salivarius culture supernatant were reduced at 56°C for 30 min in a solution containing 5 mM dithiothreitol (DTT; Amresco, USA) and 50 mM ammonium bicarbonate (Sigma-Aldrich, USA). Subsequently, alkylation was performed in the dark with 15 mM iodoacetamide (Amresco, USA) for 30 min. Samples were enzymatically digested with trypsin (Promega, USA) at 37°C overnight. The resulting peptides were purified by desalting and then reconstituted in 0.1% formic acid for subsequent analysis using a RIGOL L-3000 high-performance liquid chromatography (HPLC) system. Mass spectrometry was conducted on a Q Exactive HF-X instrument, equipped with a Nanospray Flex ion source and a C18 analytical column (QL-HPLC-100∗15). Data acquisition was performed in data-dependent acquisition (DDA) mode. The raw mass spectrometry files were processed with MaxQuant software, using the L.salivarius protein database for identification. Trypsin was designated as the proteolytic enzyme, with carbamidomethylation of cysteine residues set as a fixed modification, and oxidation of methionine along with N-terminal acetylation defined as variable modifications. The precursor ion mass tolerance was maintained within 15 ppm.

Molecular docking

Proteins potentially interacting with FPR1 were identified via GST pull-down assays coupled with the mass spectrometry analysis of bEVs. The amino acid sequences of candidate target proteins were retrieved and saved in FASTA format for downstream structural modeling. Protein structure prediction and docking with FPR1 were performed using the AlphaFold3 server (https://alphafoldserver.com/welcome) under default settings. All structural models used in this study were automatically ranked by AlphaFold3, with the top-ranked conformations selected based on confidence scores (ipTM and pTM values), reflecting the predicted accuracy of both intra- and inter-chain interactions. To further evaluate the biological relevance of the predicted complexes, structural visualization and analysis were conducted using PyMOL and Protein Viewer (https://marketplace.visualstudio.com/items?itemName=Arian Jamasb.protein-viewer).

Purification of GST- or His-tagged proteins

The gpmA, O2U05_09325 and Fpr1 genes were cloned into the pGEX-4T1 vector and transformed into Escherichia coli (Thermo Fisher, USA). The resulting E. coli strains expressing GST-tagged gpmA or O2U05_09325, or His-tagged Fpr1, were cultured in LB medium. When the optical density (OD₆₀₀) reached approximately 0.6, protein expression was induced by adding 0.4 mM isopropyl-β-D-thiogalactopyranoside (IPTG) (Solarbio, China), and cultures were subsequently incubated overnight at 25°C. Cells were harvested by centrifugation and lysed in PBL buffer containing 10 mg/mL lysozyme and a protease inhibitor cocktail. The lysate was treated with DNase I (Solarbio, China) and subjected to sonication, followed by incubation on ice for 1 h. After centrifuging at 10,000 × g for 20 min at 4°C, the supernatant was collected. GST- or His-tagged fusion proteins were purified using GST or His affinity columns (Elabscience, China) and eluted with 30 mM reduced glutathione in PBS or 50 mM imidazole, respectively. Protein concentrations of the purified GST-2,3-BdpM, GST-Lysm, and His-FPR1 were determined using a BCA protein assay.

GST pull-down assay

Recombinant human GST-tagged FPR1 protein was purchased from Abcam. bEVs derived from L. salivarius, as well as GST-2,3-BdpM and His-FPR1 proteins, were isolated as described above. GST pull-down assays were performed using a GST Protein Pull-Down Kit (Elabscience, China) according to the manufacturer’s instructions. Briefly, bEV lysates were prepared by incubating the samples with lysis buffer containing protease inhibitors and DNase I (Solarbio, China) on ice for 1 h. The lysates were then centrifuged at 10,000 × g for 20 min at 4°C to remove cellular debris. Meanwhile, Glutathione 4FF beads were washed with equilibration buffer prior to use. Diluted GST-FPR1 or GST-2,3-BdpM proteins were incubated with the pre-equilibrated beads overnight at 4°C to allow immobilization. Following extensive washing to remove unbound proteins, bEV lysates or His-FPR1 protein were added to the beads and incubated at room temperature for 3 h to facilitate protein–protein interactions. The bound complexes were eluted with 10 mM glutathione elution buffer and subsequently analyzed by silver staining, mass spectrometry, and Western blotting.

Co-immunoprecipitation

The induced BMDMs (2×10⁷) were suspended in 200 μL of PBS. Further, 20 μg of GST-2,3-BdpM or GST-Lysm protein was added and incubated at 37°C for 4 h. After treatment, the BMDMs were washed with PBS and lysed using RIPA lysis buffer (CWBIO, China). Following centrifugation at 12,000 rpm for 20 min at 4°C, the supernatant was collected and incubated overnight with 5 μg of anti-GST antibody (Proteintech, China) (normal mouse IgG antibody as a control), followed by incubation with Protein A/G agarose beads (Proteintech, China) for 4 h. The released proteins were analyzed using Western blot, to detect the bound FPR1 protein.

Lentiviral packaging and generation of CAR-Claudin18.2⁺ macrophages

Fpr1 knockdown plasmids were obtained from IGE (Guangzhou, China), and the CAR-Claudin18.2 plasmid from iCarTab (Suzhou, China). The viral packaging plasmids, pMD2.G and psPAX2 (Sangon Biotech, China), were co-transfected with the target gene plasmids into HEK 293T cells using the Lipo2000 reagent (Biosharp, China). After 48 h of culture, the supernatants were harvested, filtered, and concentrated 40-fold using a lentivirus concentrator (ECOtop, China).

Differentiated BMDMs were plated in 6-well plates. the medium was replaced with antibiotic-free medium, and lentivirus was added at a multiplicity of infection (MOI) of 100, along with 10 μg/mL polybrene (Solarbio, China). The cells were incubated for 36 h, and infection was repeated once. Following 72 h of culture, the cells underwent puromycin selection for another 72 h. Overexpression efficiency was assessed using flow cytometry.

Quantification and statistical analysis

Sample sizes were not estimated using statistical power calculations in advance. Each experiment was carried out independently at least three times to confirm reproducibility. Randomization and blinding procedures were not employed. Data analysis was performed using GraphPad Prism version 9.0. For comparisons between two groups, either the Wilcoxon rank-sum test or Student’s t test was utilized based on data distribution and experimental design. For comparisons involving more than two groups, one-way ANOVA or the Kruskal-Wallis test was applied as appropriate. Statistical significance was defined as p < 0.05 (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001), while results with p ≥ 0.05 were considered not significant (ns).

Published: February 17, 2026