A living therapeutic platform for localized in situ modulation of macrophage pyroptosis ameliorates GVHD while preserving GVL

Abstract

Background

Graft-versus-host disease (GVHD) is a major cause of morbidity and mortality following allogeneic hematopoietic stem cell transplantation, with the gastrointestinal tract being a primary target organ. Dysregulated innate immune activation, particularly within intestinal macrophages, is increasingly recognized as a key contributor to GVHD pathogenesis; however, strategies for achieving localized and controllable immunomodulation remain limited.

Methods

Single-cell transcriptomic analyses were performed to characterize macrophage-associated inflammatory programs in intestinal tissues during GVHD. A programmable living therapeutic platform was engineered using Escherichia coli Nissle 1917 to secrete outer membrane vesicles (OMVs) encapsulating BigLEN, a peptide inhibitor of NLRP3 inflammasome activation. Following oral administration, OMV-mediated delivery, inflammasome-associated signaling, and therapeutic efficacy were evaluated using in vitro macrophage assays and multiple murine GVHD models, including a graft-versus-leukemia (GVL) setting.

Results

Single-cell transcriptomic analysis of intestinal biopsies revealed a marked enrichment of inflammasome and pyroptosis-associated transcriptional programs in macrophages during GVHD. To locally modulate this inflammatory axis, we engineered Escherichia coli Nissle 1917 to secrete outer membrane vesicles (OMVs) encapsulating BigLEN, a peptide inhibitor of NLRP3 inflammasome activation. Following oral administration, engineered OMVs enabled effective delivery of BigLEN to lamina propria macrophages, resulting in attenuation of inflammasome activation and associated pyroptosis-related inflammatory responses. In multiple murine GVHD models, this in situ immunomodulatory strategy significantly alleviated intestinal pathology, reduced systemic inflammatory cytokine production, and improved survival. Importantly, anti-leukemia activity was preserved in a graft-versus-leukemia (GVL) setting.

Conclusion

This study demonstrates the feasibility of a programmable, OMV-based living therapeutic platform for localized control of macrophage inflammasome activity in the gastrointestinal tract. This strategy achieves potent, local control of GVHD without overt impairment of graft-versus-leukemia (GVL) activity, and presents a novel targeted paradigm for treating gastrointestinal inflammatory disorders.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12951-026-04285-6.

Introduction

Allogeneic hematopoietic stem cell transplantation (allo-HSCT) is a curative therapy for many hematologic malignancies but is frequently complicated by acute graft-versus-host disease (GVHD) [1]. Among target organs, the gastrointestinal tract is particularly vulnerable, where epithelial injury and uncontrolled inflammation drive morbidity and mortality [2, 3]. Current GVHD therapies rely largely on systemic immunosuppression, which although effective in dampening inflammation [4, 5], often compromises protective immune surveillance and the graft-versus-leukemia (GVL) effect [6]. This therapeutic trade-off underscores the need for strategies that locally suppress GVHD-associated inflammation while preserving systemic anti-tumor immunity.

While donor T cells remain central mediators of GVHD, increasing evidence indicates that innate immune cells critically shape the inflammatory microenvironment that sustains T-cell–mediated tissue damage [7, 8]. In the gut, macrophages rapidly respond to conditioning-induced injury and microbial translocation, amplifying inflammation through the production of pro-inflammatory cytokines [9]. Recent studies and single-cell transcriptomic analyses of GVHD patient biopsies have highlighted macrophage pyroptosis as a prominent feature of intestinal GVHD, implicating the NLRP3 inflammasome as a key upstream regulator of this process [10].

Pyroptosis is a highly inflammatory form of programmed cell death mediated by gasdermin pore formation and accompanied by the release of interleukin-1β (IL-1β) and interleukin-18 (IL-18) [11]. Activation of the NLRP3 inflammasome, composed of NLRP3, ASC, and caspase-1, is a principal driver of macrophage pyroptosis [12, 13]. Although NLRP3 represents an attractive therapeutic target, systemic inhibition raises concerns regarding off-target toxicity and impaired host defense in the post-transplant setting [14–16].

To achieve functional inhibition of NLRP3-associated inflammasome activation, we sought to identify a highly specific NLRP3-inhibitory peptide that could serve as a molecular payload for localized immunomodulation. Through structure-guided in silico screening of a structurally diverse in-house peptide library, we identified BigLEN, a 16-amino-acid peptide derived from the endogenous neuropeptide precursor proSAAS (PCSK1N), as a candidate capable of binding the conserved NACHT domain of NLRP3. Although BigLEN has been previously characterized in neuroendocrine contexts, its role in innate immune regulation has not been reported. Our computational, biophysical, and functional analyses support a previously unrecognized NLRP3-inhibitory activity of BigLEN. Importantly, BigLEN was not envisioned as a stand-alone therapeutic agent, as systemically administered peptides face major barriers to achieving effective bioactive concentrations within the inflamed intestinal immune niche, including rapid clearance, limited tissue accumulation, and inefficient uptake by lamina propria macrophages [17, 18]. These considerations led us to propose that the therapeutic efficacy of NLRP3 inhibition in intestinal GVHD critically depends on the development of a gut-restricted delivery strategy capable of concentrating the inhibitory payload at the site of disease. To this end, we engineered the probiotic Escherichia coli Nissle 1917 (EcN) [19, 20] as a living delivery platform capable of producing and deploying BigLEN in situ [21]. By harnessing EcN-derived outer membrane vesicles (OMVs), we developed an inducible system termed EcN^Smart that enables preferential delivery of the NLRP3-inhibitory payload within the intestinal immune microenvironment, including macrophages.

Using murine models of allo-HSCT, we demonstrate that EcN^Smart-derived OMVs could cross the intestinal barrier, inhibit macrophage NLRP3 inflammasome activation and pyroptosis in vivo, and markedly attenuate intestinal GVHD while preserving the GVL effect (Fig. 1). Together, these findings establish a delivery-dependent paradigm for localized innate immune modulation in GVHD and highlight the potential of engineered probiotics as precision therapeutics for inflammatory diseases of the gastrointestinal tract.

Fig. 1.

Schematic of the engineered probiotic strategy for treating intestinal GVHD. Orally administered, engineered Escherichia coli Nissle 1917 produces outer membrane vesicles (OMVs) displaying the NLRP3-inhibitory peptide BigLEN in the gut. These OMVs enable delivery of the displayed peptide to intestinal tissues and are uptake by lamina propria macrophages. Inside macrophages, BigLEN binds to the NLRP3 NACHT domain, inhibiting inflammasome assembly and subsequent pyroptosis. By attenuating macrophage pyroptosis and the associated cytokine storm, the treatment reduces the activation and infiltration of pathogenic T‑cell subsets (such as Th1 and Th17 cells) and promotes a shift in macrophage polarization toward an anti‑inflammatory phenotype. Ultimately, this locally restricted immunomodulation alleviates intestinal GVHD

Integrated single-cell profiling and in vivo validation implicate NLRP3-mediated macrophage pyroptosis in intestinal GVHD

We analyzed single-cell RNA sequencing data from the GSE database [22], comparing intestinal tissues from patients with graft-versus-host disease (GVHD) to healthy adjacent colon tissues obtained from colorectal cancer resections. Cell clustering was performed based on established lineage-specific markers. UMAP dimensionality reduction identified seven major cell populations: B cells, cycling cells, epithelial cells, mast cells, myeloid mononuclear cells, plasma cells, and T cells (Fig. 2A–B). Samples were categorized according to clinical status into healthy controls and disease groups, with the latter further stratified into mild and severe GVHD based on clinical severity. The relative proportions of each cell population across groups were quantified (Fig. 2C). Compared to healthy controls, the HSCT group exhibited a significant increase in the proportions of T cells, macrophages, mast cells, and epithelial cells (Figure S1A), with the most notable elevation observed in T cells. Given the central role of dysregulated T cell activation in GVHD pathogenesis [23], we further performed T cell subset analysis. Based on established T cell markers (Figure S1C). T cells were classified into subsets including CD4⁺ ANXA1⁺, CD4⁺ CD40LG⁺, CD4⁺ IL7R⁺, CD4⁺ CCR7⁺, CD4⁺ FOXP3⁺ (regulatory T cells, Treg), CD4⁺ MKi67⁺, CD8⁺ GZMB⁺, CD8⁺ ITGAE⁺, and CD8⁺ LAG3⁺ (Figure S1D). As GVHD involves complex intercellular crosstalk, we conducted cell-cell communication analysis. Results revealed strong interactions between CD8⁺ GZMB⁺ T cells and macrophages (Figure. S1F), suggesting coordinated involvement of cytotoxic T cells and macrophages in GVHD progression. To investigate transcriptional changes in macrophages during GVHD, we compared differentially expressed genes between control and disease groups. Gene set variation analysis (GSVA) indicated significant enrichment of pyroptosis-related pathways with advancing GVHD (Fig. 2D). Further enrichment analysis showed that NLRP3, a key regulator of pyroptosis, was markedly upregulated in GVHD macrophages (Figure S1B), suggesting enhanced inflammasome activation and engagement of pyroptosis-associated inflammatory programs in GVHD macrophages. UMAP visualization (Fig. 2E) and violin plots confirmed NLRP3-specific overexpression in macrophages, with expression levels significantly higher in mild and severe GVHD groups compared to healthy controls, and further elevated in severe versus mild GVHD (Figure S1E). Moreover, violin plots demonstrated upregulated expression of multiple pyroptosis markers—including NLRP3, IL‑1β and GBP2—in disease groups, correlating positively with GVHD severity (Fig. 2F). To validate these findings in vivo, we established an allogeneic hematopoietic stem cell transplantation model in Balb/c mice using bone marrow and splenocytes from C57BL/6 donors, with syngeneic Balb/c transplantation as control. The survival period of mice in the allogeneic hematopoietic stem cell transplantation group was significantly shorter compared to the control group (Figure SG). Colon tissues were harvested on day 14 post-transplantation and subjected to HE staining, immunohistochemistry (IHC), and immunofluorescence (IF) analysis. HE staining revealed pronounced inflammatory infiltration, mucosal edema, and architectural disruption in GVHD colon tissues compared to controls (Fig. 2G). IHC staining confirmed high NLRP3 expression in intestinal tissues of GVHD mice (Fig. 2H). IF staining further demonstrated an increased proportion of macrophages, elevated NLRP3 expression, and enhanced levels of the pyroptosis executioner protein GSDMD in GVHD colon tissues (Figure. 1I–J). Relative fluorescence data were provided in Supplementary Figure S1H–I. Collectively, these results indicate that macrophage infiltration and activation of pyroptosis-related signaling pathways are associated with intestinal GVHD, corroborating the human single-cell sequencing data. Our findings suggest that macrophage pyroptosis contributes to GVHD progression and highlight NLRP3 as a potential target for localized therapeutic intervention.

Fig. 2.

Single-cell transcriptomic and histopathological profiling reveals enrichment of NLRP3 and pyroptosis-associated signatures in intestinal macrophages during GVHD. (A) UMAP visualization of single‑cell RNA sequencing data from intestinal tissues of GVHD patients and healthy controls, showing seven major cell populations. (B) Established lineage-specific markers of these seven major cell populations. (C) The relative proportions of each cell population across different groups. (D) GSVA indicating significant enrichment of pyroptosis‑related pathways in GVHD samples compared to controls. (E) UMAP projection highlighting NLRP3 expression, which is specifically upregulated in the macrophage cluster under GVHD conditions. (F) Violin plots showing elevated expression levels of pyroptosis‑associated genes (NLRP3, IL‑1β, GBP2) in intestinal macrophages from GVHD patients, correlating with disease severity. (G) H&E staining of colon sections from syngeneic and allogeneic transplanted mice, revealing pronounced inflammatory infiltration and mucosal disruption in GVHD mice. Colon tissues were collected on day 14 post-transplantation. (H) IHC staining confirming increased NLRP3 protein expression in intestinal tissues of GVHD mice. (I-J) IF staining of colon sections showing enhanced infiltration of F4/80⁺ macrophages, elevated NLRP3 expression, and increased levels of the pyroptosis executioner protein GSDMD in GVHD mice compared to controls. Scale bars: 200 μm (H&E, IHC) and 50 μm (IF). Data are representative of at least three independent experiments. Colon tissues were collected on day 14 post-transplantation

Identification and validation of BigLEN as a novel NLRP3-inhibitory peptide attenuating pyroptosis

Based on the aforementioned findings, we posited that macrophage pyroptosis serves as a critical effector mechanism exacerbating GVHD, with NLRP3 identified as its key regulatory molecule. Consequently, screening for potent NLRP3 inhibitors may present a novel therapeutic strategy for mitigating GVHD. The screening pipeline is depicted in Fig. 3A. Initially, the high-resolution crystal structure of the NLRP3 NACHT domain, which exhibits high conservation between humans and mice, was retrieved from the PDB database. Structural analysis of NLRP3 in complex with known inhibitors revealed a conserved hydrophobic binding pocket within the NACHT domain (Fig. 3B). Detailed interaction profiling demonstrated that the inhibitor forms hydrogen bonds with residues Arg351, Gln624, Glu629, Arg578, and Ala228, with corresponding binding energies of -4.7, -2.0, -2.5, -8.7, and − 0.7 kcal/mol, respectively. Utilizing these key interactions, a pharmacophore model comprising four features (F1: hydrogen bond acceptor; F2: hydrogen bond acceptor; F3: hydrogen bond donor; F4: hydrogen bond acceptor) was constructed (Fig. 3C) to facilitate virtual screening for novel NLRP3 inhibitors. Primary pharmacophore-based screening of an in-house library of 1,041 peptides yielded 196 candidate hits. These candidates were subsequently subjected to molecular docking against the NACHT domain. The top 10% of compounds, ranked by binding score, were selected for further validation. Among these, the peptide BigLEN which is a 16 amino acids polypeptide demonstrated pronounced inhibitory activity. Further molecular docking analysis (score: -14.48 kcal/mol) indicated that BigLEN occupies the hydrophobic pocket of the NACHT domain (Fig. 3D) and establishes hydrogen bonds with key residues His367, Arg578, Glu369, and Glu629, with binding energies of -1.1, -0.8, -13.76, and − 0.6 kcal/mol, respectively (Fig. 3E). These data suggest a potential mechanism of action for BigLEN as an NLRP3 inhibitor.

Fig. 3.

Identification and functional validation of BigLEN as a novel NLRP3‑inhibitory peptide. (A) Schematic workflow for the virtual screening of NLRP3‑inhibitory peptides. (B) Structural analysis of the NLRP3 NACHT domain in complex with a known inhibitor, highlighting the conserved hydrophobic binding pocket. (C) Pharmacophore model constructed based on key inhibitor‑protein interactions, featuring four critical chemical features (hydrogen bond acceptors and donors) for virtual screening. (D) Molecular docking pose of the candidate peptide BigLEN within the NACHT domain hydrophobic pocket. (E) Detailed interaction diagram between BigLEN and key residues (His367, Arg578, Glu369, Glu629) of the NLRP3 NACHT domain, with corresponding binding energies. (F) Surface plasmon resonance (SPR) sensograms showing concentration‑dependent binding of BigLEN to immobilized NLRP3 protein. The equilibrium dissociation constant (KD) was calculated to be 1.749 × 10⁻⁷ M. (G-H) Flow cytometry analysis of cell death in mouse bone marrow‑derived macrophages (BMDMs). Cells were primed with LPS and stimulated with nigericin in the presence or absence of BigLEN (10 or 50 µM). (I) Representative Western blot images of key pyroptosis‑related proteins and ASC in BMDMs in different groups. (J) Lactate dehydrogenase (LDH) release assay of supernatant from BMDMs following LPS/nigericin stimulation in the presence or absence of BigLEN (10 or 50 µM), supernatants were collected 12 h after BigLEN treatment. (K) Representative TEM image of BMDMs in different groups. Scale bar = 1 μm. Data are representative of three independent experiments. Statistical significance was determined by one‑way ANOVA with Tukey’s post‑hoc test (*p < 0.05). Quantitative data are presented as mean ± SEM

To experimentally validate the binding between BigLEN and NLRP3, surface plasmon resonance (SPR) assays were performed. The results indicated rapid association and dissociation kinetics, with an equilibrium dissociation constant (KD) of 1.749 × 10⁻⁷ M, confirming a high-affinity interaction between BigLEN and NLRP3 protein (Fig. 3F). We next investigated the ability of BigLEN to counteract pyroptosis induced by LPS and nigericin in mouse bone marrow-derived macrophages (BMDMs). Cells were primed with 0.5 µg/ml LPS for 4 h prior to nigericin (10 µM) challenge. Flow cytometry analysis revealed that BigLEN (10 and 50 µM) significantly suppressed cell death in a dose-dependent manner (Fig. 3G-H). Consistent with this, western blot analysis showed that BigLEN treatment attenuated the expression of key pyroptosis-executing molecules downstream of NLRP3 activation, including cleaved Caspase-1, GSDMD-N, mature IL-18, and cleaved IL-1β, in LPS/nigericin-stimulated BMDMs (Fig. 3I). ASC oligomerization is an essential step for NLRP3 activation [24], we found that BigLEN could suppress LPS/nigericin-induced ASC oligomerization in a dose-dependent manner (Fig. 3I), suggesting that BigLEN acts upstream of ASC oligomerization to suppress NLRP3 activation. Furthermore, measurement of lactate dehydrogenase (LDH) release in supernatant indicated a significant increase in the LPS/nigericin group compared to the PBS control, which was markedly reduced by BigLEN treatment (Fig. 3J). The inhibitory effect of BigLEN on pyroptosis was further corroborated by transmission electron microscopy (TEM), which showed preservation of cellular integrity in BigLEN-treated cells compared to the characteristic morphological changes of pyroptosis in controls. In summary, we have successfully identified and validated BigLEN as a high-affinity NLRP3-binding peptide capable of effectively inhibiting NLRP3-mediated pyroptosis in macrophage in vitro.

Limited in vivo efficacy of systemically administered BigLEN reveals a delivery bottleneck

Although BigLEN potently inhibited NLRP3 inflammasome activation and macrophage pyroptosis in vitro, we next evaluated its therapeutic efficacy following systemic administration in vivo. Intravenous administration of BigLEN did not result in a statistically significant improvement in overall survival compared with control-treated GVHD mice (Figure S2A).

Despite the lack of survival benefit, BigLEN-treated mice exhibited a measurable attenuation of disease severity at early time points. Clinical GVHD scores were significantly reduced on day 7 and day 14 after transplantation in mice receiving intravenous BigLEN compared with untreated GVHD controls (Figure S2B), indicating partial in vivo biological activity. However, this improvement was transient and insufficient to translate into durable disease control or survival benefit. These findings suggest that systemic administration may be suboptimal for achieving sustained modulation of intestinal immune responses.

Development of an engineered E. Coli nissle 1917 for localized immunomodulatory payload delivery

The limited therapeutic efficacy observed following intravenous administration of free BigLEN indicated that effective inhibition of NLRP3-driven macrophage pyroptosis in intestinal GVHD is constrained by delivery rather than by intrinsic inhibitory activity. We therefore sought to develop a gut-targeted delivery platform capable of concentrating the NLRP3-inhibitory payload within the intestinal immune microenvironment. Given its established safety profile and ability to colonize the gastrointestinal tract, the probiotic Escherichia coli Nissle 1917 (EcN) was selected as the delivery chassis. To enhance vesicle-mediated cargo delivery while minimizing inflammatory risk, we engineered EcN by deleting genes involved in outer membrane integrity and lipopolysaccharide biosynthesis. Preliminary evidence suggests that deletion of genes involved in outer membrane integrity, such as tolA and tolR, can increase the secretion of bacterial outer membrane vesicles (OMVs) [25, 26]. Furthermore, knocking out msbB reduces Lipopolysaccharide (LPS) synthesis, thereby potentially attenuating bacterial toxicity [27]. Using CRISPR/Cas9 technology, we generated knockout mutants of the lpxM gene (also known as msbB), which encodes a Lipid A biosynthesis myristoyltransferase, as well as the tolR and tolA genes, which encode components of the Tol-Pal system (Fig. 4A). Successful gene deletions were confirmed by DNA gel electrophoresis (Fig. 4B-C). The genetically modified strain was designated EcN-e. Growth curve analysis monitoring OD600 over time indicated that EcN-e exhibited a slightly slower proliferation rate compared to the wild-type EcN (Fig. 4D). We then compared OMV production between the strains. Quantitative analysis using a BCA protein assay revealed a significant increase in OMV yield from EcN-e compared to EcN (Fig. 4E). Transmission electron microscopy (TEM) and subsequent nanoparticle tracking analysis confirmed that the genetic modifications did not alter the characteristic morphology or size distribution of the OMVs (Fig. 4F-G). We next assessed the potential toxicity attenuation in EcN-e. Quantification of LPS content demonstrated a significant reduction in EcN-e compared to the parental strain (Fig. 4H). To evaluate in vivo safety, mice were orally gavaged with EcN-e (1 × 10⁹ CFU in 200 µL PBS) every 7 days for 40 days. No mouse mortality or overt signs of distress were observed throughout this period, indicating no acute toxicity (Fig. 4I). Complete blood count analysis showed that chronic administration of EcN-e did not significantly alter key hematological parameters, including white blood cell count, red blood cell count, hemoglobin concentration, and platelet count (Fig. 4J, Figure S3A-D). Finally, we evaluated the in vitro biocompatibility of the purified OMVs. Incubation of macrophages with OMVs derived from EcN-e did not induce significant apoptosis (Fig. 4K, L). In summary, we successfully constructed an engineered EcN strain (EcN-e) characterized by enhanced OMV production and a favorable safety profile, both in vitro and in vivo. This engineered bacterium provides a robust and safe platform for the subsequent targeted delivery of therapeutic agents, such as the NLRP3-inhibitory peptide identified earlier.

Fig. 4.

Engineering of an E. coli Nissle 1917 for enhanced and safer outer membrane vesicle production. (A) Schematic of the genetic engineering strategy for the probiotic Escherichia coli Nissle 1917 (EcN). Key genes involved in lipopolysaccharide (LPS) biosynthesis (msbB) and outer membrane integrity (tolR, tolA) were deleted using CRISPR/Cas9 to generate the engineered strain EcN-e, aiming to reduce endotoxicity and increase OMV secretion. (B-C) Gel electrophoresis confirmation of successful gene deletions. PCR products show the expected size shifts for the msbB (B) and tolRA (C) loci in the knockout strain (EcN-e) compared to the wild-type (WT) EcN. (D) Growth curves of WT EcN and EcN-e monitored by optical density at 600 nm (OD600). (E) Relative quantification of OMV production. Protein content of purified OMVs from equal numbers of bacteria was measured using a BCA assay. (F) Representative transmission electron microscopy (TEM) images of OMVs isolated from EcN and EcN-e. Scale bar = 100 nm. (G) Nanoparticle tracking analysis (NTA) of OMVs, confirming a size distribution profile typical of bacterial OMVs (peak around 80 nm). (H) Relative quantification of LPS content in bacterial lysates. (I) Survival of mice in each group (n = 6/group). In vivo safety assessment, mice were orally administered EcN/EcN-e (1 × 10⁹ CFU) every three days for 40 days. (J) Representative image of complete blood count analysis after oral administration of PBS/EcN/EcN-e. (K-L) In vitro biocompatibility of purified OMVs. Macrophages were incubated with OMVs derived from PBS/EcN/EcN-e, and apoptosis was assessed via flow cytometry. Data are representative of at least three independent experiments. Statistical significance was determined by unpaired two-tailed Student’s t test (*p < 0.05) and one‑way ANOVA with Tukey’s post‑hoc test (*p < 0.05). Quantitative data are presented as mean ± SEM

Engineering an arabinose-inducible E. Coli nissle strain for targeted gut delivery and pyroptosis inhibition

For the targeted and regulated delivery of the BigLEN peptide to the gastrointestinal tract, we selected the pBAD vector system—a reliable and tightly controlled system for recombinant protein expression in bacteria. This system is based on the araBAD operon, which regulates L-arabinose metabolism in E. coli. By placing the gene of interest downstream of the araBAD promoter in the pBAD vector, high-level expression is induced in the presence of L-arabinose, while expression is repressed in the presence of glucose.To visualize bacterial localization, we first engineered an E. coli Nissle 1917 (EcN) strain, designated EcN-e, to express luciferase fused to a hemagglutinin (HA) tag from human influenza virus. This fusion gene was placed under the control of the arabinose-inducible promoter, allowing expression to be induced by arabinose. As shown in Fig. 5A, after culture with arabinose and addition of the luciferase substrate, the bacteria emitted bright luminescence. Consistent with this, Western blot analysis (Fig. 5B) confirmed expression of the HA-tagged protein only upon arabinose induction. Mice were then oral treated with this luciferase-expressing engineered EcN (1 × 10⁹ CFU in 200 µL PBS). As shown in Fig. 5D, luminescence was primarily localized to the colon over time and remained detectable 24 h after administration. Quantified IVIS data of stomach, small intestine and colon tissues were shown in Figure S7B. Next, we constructed an engineered strain for BigLEN delivery. The BigLEN peptide was fused to the HA tag, and this fusion protein was linked to the C-terminus of ClyA protein to enable its display on the surface of bacterial outer membrane vesicles (OMVs). This construct was also cloned into the pBAD vector. The resulting strain was named EcN^Smart. Western blot analysis of bacterial outer membrane proteins confirmed successful expression of the HA-tagged fusion protein (Fig. 5B), indicating that EcN^Smart effectively displays the fusion peptide on OMV surfaces. Colon tissues collected 12 h after gavaging mice with EcN^Smart (along with arabinose induction) were subjected to immunofluorescence staining. Unlike systemically delivered free peptide, engineered OMVs readily accessed the intestinal tissue compartment. The results showed fluorescence signals from HA-tagged OMVs within colon tissue, with co-localization observed with macrophages (Fig. 5C). These data demonstrated that oral administration of the engineered probiotic enables targeted delivery to intestinal tissue and subsequent uptake by intestinal macrophages. To further evaluate the in vivo distribution of engineered OMVs, mice were orally gavaged with EcN^Smart, and major organs were harvested 24 h later for analysis. OMVs carried an HA-tag fused to the therapeutic peptide, enabling tissue detection by anti-HA immunofluorescence staining. Fluorescence imaging demonstrated predominant HA signal enrichment in the colon. Detectable signal was also observed in the spleen, whereas liver, lung, kidney, and heart showed minimal fluorescence levels comparable to background. These findings indicate preferential localization of orally administered EcN-derived OMVs to the gastrointestinal tract, with limited distribution to the spleen. Representative images and quantitative analyses are shown in Figure S5A–B. Also, to assess cellular uptake, DiI labeled OMVs were incubated with BMDMs and analyzed by confocal microscopy. Robust intracellular fluorescence signals were observed within macrophages, indicating efficient internalization of OMVs in vitro. Representative images are shown in Figure S7A.

Fig. 5.

Targeted delivery and pyroptosis inhibition by an arabinose‑inducible engineered E. coli Nissle strain. (A) Luminescence imaging of EcN/engineered EcN (Luc‑HA) under the control of the arabinose‑inducible araBAD promoter in the presence (+) or absence (-) of arabinose. (B) Representative Western blot images confirming arabinose‑dependent expression of the HA‑tagged fusion protein in the engineered EcN (Luc-HA) and OMV (ClyA-BigLEN-HA). (C) Representative immune fluorescence images of colon tissues from mice treated with EcN^Smart (BigLEN‑displaying strain) in the presence (+) or absence (-) of arabinose. HA (red) signals indicate the presence of engineered OMVs, F4/80 (green) marks macrophages, and DAPI (blue) stains nuclei. Scale bar = 50 μm. (D) In vivo bioluminescence imaging over time showing persistent luminescence signal in the gastrointestinal tract after oral administration of luciferase‑expressing EcN and arabinose. (E-F) Flow cytometry analysis of cell death in BMDMs. Cells were primed with LPS and stimulated with nigericin in the presence or absence of BigLEN‑displaying OMVs (50–200 µg/mL). (G) Representative TEM image of BMDMs in different groups. Scale bar = 1 μm. Data are representative of three independent experiments. (H) Representative Western blot images of key pyroptosis‑related proteins and ASC in BMDMs in different groups. (I) Lactate dehydrogenase (LDH) release assay of supernatant from BMDMs following LPS/nigericin stimulation in the presence or absence of BigLEN‑displaying OMVs (50–200 µg/mL). Statistical significance was determined by one‑way ANOVA with Tukey’s post‑hoc test (*p < 0.05). Quantitative data are presented as mean ± SEM. (J) Quantification of Western blot band intensity of different proteins. Densitometric analysis was performed using ImageJ and normalized to GAPDH. Data represent three independent experiments. Data are presented as mean ± SEM

We further evaluated in vitro whether OMVs secreted by EcN^Smart could inhibit pyroptosis induced by LPS and nigericin. BMDMs were primed with 0.5 µg/mL LPS for 4 h, followed by challenge with 10 µM nigericin to induce pyroptosis. Flow cytometry results showed that treatment with BigLEN-displaying OMVs (50 µg/mL and 200 µg/mL) significantly inhibited cell death in a dose-dependent manner (Fig. 5E–F). Transmission electron microscopy (TEM) similarly demonstrated that BigLEN-OMVs suppressed the pyroptotic morphological changes induced by LPS/nigericin (Fig. 5G). Western blot revealed that BigLEN-OMVs inhibited the expression or activation of key pyroptosis-related proteins induced by LPS/nigericin, including cleaved caspase-1, GSDMD-N, IL-18, and mature IL-1β (cleaved-IL-1β) (Fig. 5H). Also, we found that BigLEN-OMVs could suppress LPS/nigericin-induced ASC oligomerization in a dose-dependent manner. Additionally, we measured lactate dehydrogenase (LDH) release in the cell supernatant. Compared with the PBS control group, LDH levels were significantly increased in the LPS/nigericin-stimulated group, whereas BigLEN-OMV treatment effectively reduced LDH release (Fig. 5I). To quantitatively assess inflammasome-dependent cytokine release, IL-1β and IL-18 concentrations in macrophage culture supernatants were measured by ELISA. Consistent with the Western blot analyses, EcN^Smart treatment significantly reduced the secretion of both IL-1β and IL-18 following LPS/nigericin stimulation. Quantitative data are presented in Figure S10A-B. In summary, we successfully constructed the engineered E. coli strain EcN^Smart, which allows regulated delivery of a therapeutic peptide to the gut.

EcN^Smart treatment improves survival and suppresses pyroptosis in GVHD

To directly compare the therapeutic outcomes of systemic NLRP3 inhibition and gut-restricted delivery, we evaluated the effects of intravenous BigLEN, orally administered wild-type EcN, and EcN^Smart in a murine model of allogeneic HSCT induced GVHD. Treatment was initiated one week after transplantation, coinciding with neutrophil engraftment. Mice were treated with PBS/BigLEN/EcN/EcN^Smart every three days for 40 days.

Consistent with earlier observations, intravenous administration of BigLEN did not result in a statistically significant improvement in overall survival compared with untreated GVHD mice (Fig. 6A). Similarly, wild-type EcN treatment alone failed to confer a survival advantage. In contrast, mice receiving EcN^Smart exhibited a marked and sustained improvement in survival. Tissues were harvested on day 28 post-transplantation for histological, biochemical, and immunological analyses. Histopathological analysis of colonic tissues revealed no significant attenuation of tissue damage in BigLEN- or EcN-treated mice, whereas EcN^Smart treatment substantially improved intestinal architecture and reduced inflammatory injury (Fig. 6B, Semi-quantitative score showed in Figure S4F). To assess whether EcN^Smart exerts effects beyond the gastrointestinal tract, additional histopathological analyses were performed on established GVHD target organs, including liver and lung. Semi-quantitative scoring of H&E-stained sections revealed no statistically significant differences in liver or lung pathology among treatment groups (Figure S8A).

Fig. 6.

Treatment with engineered EcN^Smart ameliorates intestinal GVHD and attenuates pyroptosis-associated inflammatory responses in vivo. (A) Survival curves of allogeneic HSCT recipient mice treated with free BigLEN peptide, wild type EcN, or EcN^Smart. Treatment began at one week post-transplantation (n = 10). (B) Representative H&E staining of colon tissues harvested on day 28 post-transplantation. Scale bar = 100 μm. (C) Representative immune fluorescence images of colon tissues from mice with different treatment. Scale bar = 50 μm (D) Clinical GVHD severity scores in mice with different treatment, scores recorded every 7 days. (E) Measurement of lactate dehydrogenase (LDH) activity in intestinal lavage fluid of mice with different treatment. (F) Representative Western blot images of key pyroptosis‑related proteins and ASC in colon tissue lysates from mice with different treatment. Statistical significance for survival was determined by the log-rank test. Clinical scores and other datas were analyzed by one-way ANOVA with Tukey’s post-hoc test. (*p < 0.05). Data are presented as mean ± SEM. (G) Quantification of Western blot band intensity of different proteins are shown. Densitometric analysis was performed using ImageJ and normalized to GAPDH. Data represent three independent experiments. Data are presented as mean ± SEM

Despite the absence of survival or histopathological benefit, both BigLEN and wild-type EcN–treated mice exhibited measurable reductions in pyroptosis-associated phenotype. Immunofluorescence staining demonstrated decreased expression of the pyroptosis executor GSDMD in colonic macrophages in both BigLEN and EcN groups compared with untreated GVHD mice, although the magnitude of reduction was significantly greater in the EcN^Smart group (Fig. 6C). Relative quantitative data of GSDMD IF signals from colon tissues of different groups are shown in Figure S4A. To quantitatively evaluate macrophage lytic cell death in vivo, intestinal lamina propria cells were analyzed by flow cytometry. Macrophages were gated as F4/80⁺ cells, and membrane-compromised cells were identified as 7-AAD⁺ within this population. EcN^Smart treatment significantly reduced the proportion of 7-AAD⁺ F4/80⁺ macrophages compared with free BigLEN or wild-type EcN, indicating attenuation of inflammasome-associated lytic cell death in the intestinal tissue. Quantitative data are shown in Figure S6A–B. Consistently, lactate dehydrogenase (LDH) levels in intestinal lavage fluid were modestly reduced in BigLEN and EcN-treated mice but were markedly suppressed following EcN^Smart administration (Fig. 6E). Western blot analysis of colonic tissue lysates further confirmed partial inhibition of ASC oligomerization and downstream pyroptosis-associated signaling in the BigLEN and EcN groups, whereas EcN^Smart treatment resulted in robust suppression of inflammasome activation in vivo (Fig. 6F). The dissociation between reduced pyroptosis markers and the absence of survival or histopathological improvement in the BigLEN and wild-type EcN groups suggests that partial or transient suppression of inflammasome activity is insufficient to interrupt the self-sustaining inflammatory cascade of established GVHD. Although systemic BigLEN administration and wild-type EcN treatment attenuated macrophage pyroptosis to a detectable extent, the magnitude, spatial distribution, and duration of this inhibition were likely inadequate to prevent downstream amplification of tissue damage and pathogenic T-cell responses.

Notably, wild-type EcN alone also conferred modest suppression of pyroptosis-associated readouts, consistent with prior reports that probiotic strains can exert baseline immunomodulatory effects through microbiota interaction, barrier reinforcement, or tonic regulation of innate immune signaling. However, these nonspecific effects remained subtherapeutic in the absence of targeted payload delivery.

EcN^Smart is associated with altered T-cell responses and macrophage polarization in intestinal GVHD

Given that inflammasome-activated macrophages are a major source of IL-1β and IL-18, cytokines implicated in the differentiation and expansion of pathogenic CD4⁺ T‑helper subsets, we assessed whether EcN^Smart treatment influences T‑cell infiltration and macrophage polarization in the intestinal tissues of mice with acute GVHD. Spleen and colon tissues were harvested on day 28 post-transplantation for flow cytometric analysis. Flow cytometric analysis revealed that the infiltration of CD4⁺IFN‑γ⁺ T cells and CD4⁺IL‑17⁺ T cells was significantly reduced in both the spleen and colon of EcN^Smart ‑treated mice compared to controls (Fig. 7A-D). To evaluate overall T cell activation status, CD69 expression on CD4⁺ and CD8⁺ splenic T cells was assessed. Flow cytometric analysis revealed that EcN^Smart treatment did not significantly alter CD69 expression compared with GVHD controls (Figure S9A-D). These findings indicate that EcN^Smart preferentially ameliorates intestinal inflammation while reducing pathogenic T cell polarization without detectable suppression of global T cell activation. Concurrently, the level of reactive oxygen species (ROS) in colon tissue was markedly decreased after EcN^Smart administration (Figure S4B-C). We further assessed the polarization state of intestinal macrophages by immunofluorescence staining. Colon tissue from EcN^Smart treated mice exhibited increased expression of the anti‑inflammatory marker eNOS and decreased expression of the pro‑inflammatory marker iNOS in macrophages, indicating a shift toward an anti‑inflammatory phenotype (Fig. 7E-F). Quantitative data of IF images were provided in Figure S4D-E. These results demonstrated that EcN^Smart treatment was associated with a shift in macrophage inflammatory marker expression toward a less pro-inflammatory state, concomitant with the amelioration of GVHD.

Fig. 7.

EcN^Smart is associated with altered T-cell responses and macrophage polarization in intestinal GVHD. (A-B) Flow cytometric analysis of CD4⁺IFN‑γ⁺ (Th1) and CD4⁺IL‑17⁺ (Th17) T cells in the spleen (A) and colon (B) of mice at day 28 post‑transplantation. (C-D) Quantitative data of flow cytometric analysis in different groups (n = 5). (E-F) Immunofluorescence staining of colon sections for macrophage polarization markers. Representative images show macrophages (F4/80⁺, pink) co‑stained with the pro‑inflammatory marker iNOS (green, left panel) or the anti‑inflammatory marker eNOS (yellow, right panel). Nuclei are stained with DAPI (blue). Scale bar = 50 μm. Statistical significance was determined by one‑way ANOVA with Tukey’s post‑hoc test (*p < 0.05). Data are presented as mean ± SEM. Additional quantitative data are provided in Figure S4D-E

EcN^Smart treatment preserves anti-leukemia activity while reducing GVHD-associated mortality

Preserving the graft-versus-leukemia (GVL) effect while treating graft-versus-host disease (GVHD) is crucial for preventing relapse in malignant hematologic disorders. To evaluate whether EcN^Smart treatment affects the GVL effect, we established a model using A20 murine B‑cell lymphoma cells stably expressing luciferase. All recipient mice were co‑transplanted with donor bone marrow and luciferase‑transfected A20 cells. Mice in BMT groups received spleen T cells along with BM cells. Mice in EcN^Smart group orally treated with EcN^Smart at 1 week post transplantation (1 × 10⁹ CFU in 200 µL PBS, once every three days for 40 days). Leukemia burden was assessed by bioluminescence imaging on days 7, 14, and 21 post-transplantation, Mice were monitored for survival up to 40 days post-transplantation. Survival analysis revealed that mice treated with EcN^Smart exhibited significantly prolonged survival, whereas mice transplanted with BM and A20 cells dead from leukemia progression and mice transplanted with BMT and A20 cells predominantly succumbed to GVHD (Fig. 8A). Correspondingly, clinical scoring showed that EcN^Smart treated mice had significantly lower scores at day 28 compared to untreated controls (Fig. 8B). In vivo bioluminescence imaging revealed rapid tumor progression in non‑transplanted mice. While all transplanted mice showed substantial control of tumor growth, those untreated with EcN^Smart experienced high mortality due to GVHD. (Fig. 8C). In summary, EcN^Smart treatment preserves GVL from GVHD by maintaining potent anti‑tumor activity while significantly reducing GVHD‑related mortality.

Fig. 8.

EcN^Smart treatment preserves the GVL effect. (A) Survival analysis of recipient mice treated with BM cells+A20 cells/BM cells + T cells + A20 cells/ BM cells + T cells + A20 cells + EcN^Smart (n = 10). (B) Clinical GVHD scores at 7, 14, 21, 28 days post‑transplantation in different groups. (C) Representative in vivo bioluminescence imaging of leukemia burden. Statistical significance for survival was determined by the log‑rank test. Clinical scores were analyzed by one‑way ANOVA with Tukey’s post‑hoc test (*p < 0.05). Data are presented as mean ± SEM. Abbreviation: NS, not significant

Discussion

The quest to disentangle graft-versus-host disease (GVHD) from the graft-versus-leukemia (GVL) effect represents one of the most pressing challenges in transplantation immunology. Systemic immunosuppression, the longstanding cornerstone of GVHD management, inevitably blunts anti-tumor immunity, creating a perilous therapeutic trade-off [28]. Our study explores a resolution to this dilemma by redirecting therapeutic intervention to the anatomical epicenter of disease—the gastrointestinal tract. We demonstrate that an orally administered, engineered probiotic can function as a localized bioreactor, producing vesicles that locally attenuate NLRP3 inflammasome activation and associated pyroptosis-related inflammatory responses in intestinal macrophages. This spatially confined strategy achieved robust protection against intestinal GVHD while faithfully preserving systemic GVL responses, offering a paradigm shift from broad immunosuppression toward precise, tissue-targeted immunomodulation.

Our findings resonate with yet critically extend the growing interest in NLRP3 as a therapeutic target. Small-molecule NLRP3 inhibitors such as MCC950 have shown efficacy across sterile inflammatory models [29, 30]. However, systemic application of small-molecule NLRP3 inhibitors in the context of HSCT raises legitimate concerns, as pan-inhibition of NLRP3 might impair antimicrobial defense and potentially affect lymphocyte function, thereby muddying the desired separation from GVL. In contrast, our peptide-based inhibitor, BigLEN, was delivered via a bacterial vesicle system engineered for gut restriction. This approach capitalizes on the innate tropism of outer membrane vesicles (OMVs) for mucosal immune cells, ensuring high local concentrations of the inhibitor precisely where inflammation rages, while minimizing systemic exposure. This logistical advantage underscores a key principle: in mucosal diseases, where drug acts can be as important as what it targets. Our findings highlight that efficient local delivery represents a critical determinant of in vivo efficacy in this context.

This spatial precision also distinguishes our platform from other emerging biological therapies for SR-aGVHD. While JAK inhibitors like ruxolitinib have improved response rates, they exert broad effects on cytokine signaling across multiple cell types and compartments [31]. Similarly, infused cellular therapies (e.g., MSCs, regulatory T cells) require complex manufacturing and exhibit variable engraftment and potency, with mechanisms often involving pleiotropic, contact-dependent suppression [32, 33]. Our engineered EcN system represents a convergence of synthetic biology and immunotherapy: a well-defined, reproducible single strain programmed to execute a specific, inducible action at the disease site. Compared to the ecological complexity of fecal microbiota transplantation, our approach offers a mechanistic clarity and a superior safety profile, utilizing a chassis with a decades-long history of safe human use, further refined by targeted detoxification.

The mechanistic elegance of this system lies in its sequential action: Arabinose-induced production of BigLEN-displaying OMVs, lies in its coordinated action, wherein inducible OMV production enables local delivery and uptake of the inhibitory payload by lamina propria macrophages. Once inside, BigLEN acts as a molecular wedge at the NLRP3 NACHT domain, preventing the crucial oligomerization step. This upstream blockade effectively uncouples the inflammatory signal from its devastating executive phase, halting both pyroptosis and the maturation of IL-1β/IL-18. IL-18 plays a complex and context-dependent role in intestinal physiology. Under homeostatic conditions, basal IL-18 contributes to epithelial barrier maintenance and antimicrobial defense. However, during acute GVHD, excessive inflammasome activation can drive pathological overproduction of IL-18, promoting epithelial injury and amplifying inflammatory cascades. In our model, EcN^Smart reduced IL-18 concentration in the context of active GVHD-associated inflammation. Importantly, this reduction was accompanied by improved epithelial histology and reduced macrophage pyroptosis, without evidence of worsened barrier damage. Beyond local tissue protection, EcN^Smart treatment reduced splenic Th1 and Th17 frequencies without significantly altering CD69 expression on CD4⁺ or CD8⁺ T cells. These data argue against generalized T cell hypoactivation. Instead, we interpret the reduction in pathogenic Th1/Th17 populations as a secondary consequence of diminished gut-driven inflammatory amplification. Intestinal inflammasome-derived cytokines, including IL-1β and IL-18, are known to stabilize and reinforce pathogenic T cell polarization. By interrupting this upstream inflammatory loop, EcN^Smart indirectly attenuates systemic immune dysregulation while preserving overall T cell activation capacity. The preserved graft-versus-leukemia (GVL) effect further supports the absence of broad immunosuppression.

Despite these promising results, we must thoughtfully consider the limitations and translational hurdles of our platform. First, while macrophages were the primary target, the potential influence of OMV-BigLEN on other intestinal sentinel cells, such as dendritic cells or the epithelium itself, remains an open question for future research. Second, our models convincingly show efficacy in prophylaxis and early intervention; however, the therapeutic window in established, severe fibrotic GVHD may be narrower and necessitates dedicated study. Third, although our genetic modifications (ΔmsbB, ΔtolRA) aimed at enhancing safety, the long-term dynamics of engineered bacterial colonization in the context of a disrupted post-transplant microbiome, and the remote risk of horizontal gene transfer, require thorough investigation in advanced preclinical models. Finally, the transition to patients will demand optimization of the arabinose induction regimen and rigorous GMP-compliant manufacturing to ensure consistent OMV biogenesis and peptide loading.

Looking forward, the modularity of our platform is one of its most compelling features. The EcN-OMV chassis can be repurposed as a versatile delivery vehicle for other therapeutic payloads—be they peptides, nanobodies, or RNA-based agents—targeting different pathways in GVHD or other inflammatory bowel diseases. This plug-and-play potential could significantly accelerate the development of an entire class of targeted, orally administered biotherapeutics.

In conclusion, we have developed a synthetic biology-driven strategy that successfully localizes potent immunomodulation to the gut. By inhibiting NLRP3-dependent pyroptosis specifically within intestinal macrophages, we achieve effective attenuation of GVHD without overt compromise of graft-versus-leukemia activity. This work not only presents a promising new candidate for SR-aGVHD but also validates a generalizable platform for spatial precision medicine in mucosal immunity, paving the way for therapies that are as precise in their delivery as they are in their target.

Methods

Animals and cells

Female Balb/C and C57BL/6 (6–8 weeks old) were obtained from Animal centers from Fourth Military Medical University. All animal experiments were approved by the Animal Experimentation and Ethics Committee of Fourth Military Medical University (Approval NO.250751). Mice were housed in a room at 22–24 °C with a 12 h light/dark cycle and at 40–60% humidity. BMDM were isolated from mouse bone marrow. A20 were donated by Dr. Zhang from Department of Hematology of Tangdu Hospital. BMDM were cultured in complete media containing DMEM with 10% fetal bovine serum and 1% penicillin/streptomycin (Hyclone). A20 cells were cultured in complete media containing RPMI 1640 with 10% fetal bovine serum and 1% penicillin/streptomycin (Hyclone).

GVHD model

One week prior to transplantation, recipient Balb/C mice were provided with antibiotic-supplemented drinking water for infection prophylaxis. On the day of transplantation, the mice received a lethal dose of 8 Gy total body irradiation via a linear accelerator and were allowed to rest for 4–6 h. Donor mice were euthanized by cervical dislocation, sterilized by immersion in 75% ethanol for 10 min, and dissected under a biological safety cabinet to isolate femurs, tibiae, and spleens. Bone marrow (BM) cells were flushed from bones using RPMI 1640 medium, and splenocytes were prepared by gentle mechanical dissociation followed by filtration through a 70 μm strainer. After cell counting and concentration adjustment, recipients were transplanted via tail vein injection with a mixture of 5 × 10⁶ BM cells and 7.5 × 10⁶ splenocytes. Clinical GVHD was assessed using a semi-quantitative scoring system as previously described, evaluating five parameters: weight loss, posture, activity, fur texture, and skin integrity. Each parameter was scored from 0 to 2 (0 = normal; 1 = moderate change; 2 = severe change). The cumulative clinical score (maximum score = 10) was recorded for each mouse. Animals were scored every 7 days post-transplantation by investigators.

Therapeutic administration in the acute GVHD model

Mice were treated with PBS/BigLEN/EcN/EcN^Smart in different groups. BigLEN was administered via intravenous injection at a dose of 3 mg/kg in sterile PBS, EcN and EcN^Smart were administered by oral gavage at 1 × 10⁹ CFU per mouse per dose in 200 µL sterile PBS. All treatment begins on day 7 post-transplantation, and continued every 3 days until day 40. To induce BigLEN expression in EcN^Smart in vivo, L-arabinose was administered via drinking water at a concentration of 5 mg/mL beginning on day 7 post-transplantation and maintained throughout the treatment period. Arabinose-containing water was replaced every 2–3 days. For short-term induction experiments, L-arabinose was administered by oral gavage at 10 mg/mL in 200 µL per mouse.

BMDM isolation

The mice were euthanized and their femurs and tibia were separated aseptically, and then the surrounding muscles and tissues were thoroughly cleaned. Bone marrow was subsequently flushed from the cavities using a cold culture medium, and the resulting cell suspension was gently dissociated and passed through a strainer to create a single-cell mixture. Then, the cell pellet underwent RBC lysis, washing, and counting. Cells were then cultured with M-CSF at 37 °C and 5% CO₂; medium was refreshed at day 3, yielding adherent mature macrophages by days 6–7.

Plasmid construction

The Luc-HA fusion gene was cloned into the pBAD vector at the NcoI and EcoRI restriction sites. The ClyA–BigLEN-HA were cloned into the pBAD vector at the NcoI and HindIII restriction sites. The sequences of these plasmids are listed in Supplementary file. The Nissle 1917, which had been transformed with the expression plasmids, pBAD-Luc-HA, pBAD- ClyA–BigLEN-HA, was grown at 37 °C in LB medium with shaking at 220 rpm. Ampicillin (50 µg/ml) was added when needed.

In vitro and in vivo bioluminescence assay

EcN transformed with Luc-HA plasmids were cultured with or without 2 mg/ml.

Ara solution. Then engineered EcN and natural EcN were suspended in PBS and 200 µl 5 mg/ml d-luciferin potassium solution. Female Balb/C mice were administered 1 × 10⁹ E. coli and administered 200 µL of 10 mg/mL L-arabinose by oral gavage. After 12 h, the mice were euthanized, 10 mg contents included faeces from the stomach, intestine, cecum and colon and were collected into 1.5 ml tubes and suspended in 500 µl 5 mg/ml d-luciferin potassium solution. The samples were visualized by using the IVIS^® Lumina II in vivo imaging system (PerkinElmer, Thermo Fisher, US). After 3 h, 6 h, 12 h, 24 h, the mice were euthanized, the stomach and the whole lower digestive tract tissue (from d uodenum to colon) was isolated. Two millilitres of 5 mg/ml d-luciferin potassium solution was infected into the digestive tract. And the signal were detected immediately within 30 s.

OMVs preparation and characterization

EcN were cultured and Ara (2 mg/ml) was added to cultured medium when the optical density at 600 nm (OD600) reached 0.6 for protein expression. EcN were incubated overnight at 16 °C and shaken at 160 rpm. The supernatant was collected and filtered through a 0.45 μm polyvinylidene fluoride filter. Then, OMVs were collected by ultracentrifugation at 4,000×g for 15 min at 4 °C, and the supernatant was collected and ultracentrifuged at 150,000×g for 2 h at 4 °C. The pellet was finally resuspended in 200 µl PBS. To remove contaminating proteins and bacteria, the pellet was washed with PBS and the isolation procedure was repeated another time. The isolated OMVs were resuspended in PBS and stored at -80℃. The morphology of OMVs was characterized by electron microscope (JEM-2000 EX TEM, JEOL Ltd., Tokyo, Japan). For size distribution analysis, isolated OMVs were uniformly diluted to 1 µg /mL (at protein concentration), followed by analysis on the dynamic light scattering (DLS; Zetasizer Nano ZS90).

scRNA-Seq Data acquisition and analysis

scRNA-seq data of intestinal samples of GVHD patients were derived from GEO data base (GEO: GSE234357) and analyzed using the R package. The analysis workflow was implemented with the Seurat package (v3.1.5) in R. Briefly, gene expression counts were normalized, and the top 2000 variable genes were identified for downstream analysis. Dimensionality reduction was achieved via PCA, with the optimal number of components statistically determined by the JackStraw function. These components were then used for graph-based cell clustering and non-linear visualization with UMAP. For trajectory inference, microglial cells were subjected to pseudotime analysis using Monocle 2 (v14.0), where cell ordering was performed with the DDR Tree reduction method and the order Cells function. Genes differentially expressed between identified clusters (defined as |log2FC| ≥ 1 and adj. p-val < 0.05) were finally subjected to functional annotation using GO and KEGG databases, along with core pathway analysis via Ingenuity Pathway Analysis (IPA, QIAGEN).

Virtual screening

Human NACHT Domain and inhibitor (PDB: 7ALV) high-resolution complex protein structure were available form the Protein Data Bank (PDB) (http://www.rcsb.org). Protein structure processing and interaction energy analysis were performed using the Molecular Operating Environment (MOE) software. Based on the analysis of interactions between the protein and the small-molecule inhibitor, a ligand-based pharmacophore model was constructed. Virtual screening was then carried out against a self-built library containing 1,041 short polypeptides (12–20 amino acids). This library was curated to maximize structural diversity and physicochemical compatibility with intracellular delivery systems. Compared with small molecules, peptide candidates were selected based on their potential for enhanced interface complementarity with protein targets and their compatibility with recombinant expression in E. coli, which facilitates scalable production and integration into engineered outer membrane vesicle (OMV)-based delivery platforms. Molecular docking was performed between the protein and peptides that matched the pharmacophore features; the resulting complexes were ranked according to docking scores. Binding free energies were calculated using the docking scoring function in MOE, with lower values indicating stronger binding affinity. The top 10% of hit peptides were selected for subsequent experimental validation.

Surface plasmon resonance (SPR)

Surface plasmon resonance (SPR) analysis was performed on a Biacore T200 instrument (GE Healthcare, USA) to measure the binding affinity between the NACHT protein (HY-P701027, MCE, USA) and BigLEN (501036-69-7, MCE, USA). NACHT was diluted in 10 mM acetate buffer (pH 4.0) and immobilized on a CM5 dextran chip (Cytiva, USA) via amine-coupling chemistry. The BigLEN was serially diluted in 1× HBS-EP+ buffer to concentrations of 3200, 1600, 800, 400, 200, and 100 nM. These dilutions were injected over the immobilized NACHT surface at a flow rate of 30 µL/min. The running buffer for all procedures was 1× HBS-EP⁺, and the resulting response data were characterized using the Kinetic Analysis Wizard fit model to calculate KD values.

Western blot

Take cellular protein as an example, the total protein was extracted using RIPA lysis buffer according to the instructions. The lysis solution was centrifugated at 12,000×g for 20 min at 4 °C. Then the supernatant was collected into protease-free tube. The protein concentration was quantified using BCA protein assay kit (Beyotime, Beijing). Protein lysates (40 µg per sample) were resolved by SDS–PAGE (ACE biotechnology) and transferred to a PVDF membrane (MerckMillipor). After blocking in 5% non-fat milk, the PVDF membrane was incubated with primary antibodies followed by the specific secondary antibodies. The protein expression was detected using ECL hypersensitive chemiluminescent substrate for imaging.

LDH detection

The LDH release assay was performed by first collecting cell culture supernatant, which was centrifuged to remove debris. An aliquot of the clear supernatant was then incubated with a reaction mixture containing lactate, NAD⁺, and a tetrazolium salt. During the incubation period, LDH catalyzes the reduction of NAD⁺ to NADH, which subsequently reduces the tetrazolium salt to a colored formazan product. The reaction was stopped with a weak acid, and the absorbance of the formazan dye was measured at 490 nm. The LDH activity was quantified by comparing the sample absorbance to the control group cells.

Safety evaluation of the EcN oral administration

For the safety assessment, mice were randomly divided into three groups: the EcN-treated group, the EcN-e treated group. Both group received oral gavage of 1 × 10^9 CFU of live Escherichia coli suspended in 200 µL phosphate-buffered saline (PBS) every 7 days for 40 days. The control group received an equivalent volume of sterile PBS. Body weight and general health status were monitored daily. At the designated endpoint (24 h after the final gavage), mice were humanely euthanized. Blood was collected via cardiac puncture into EDTA-coated tubes for hematological analysis. Major organs, including the heart, liver, spleen, lungs, kidneys, and sections of the gastrointestinal tract (duodenum, jejunum, ileum, and colon), were immediately harvested.

H&E staining

Harvested organs were fixed in 4% paraformaldehyde for 24–48 h at room temperature, followed by standard dehydration through a graded ethanol series, clearing in xylene, and embedding in paraffin. Tissue sections were cut at a thickness of 5 μm using a rotary microtome [Leica RM2255]. For histological evaluation, sections were deparaffinized, rehydrated, and stained with Hematoxylin and Eosin (H&E) according to standard protocols. Briefly, sections were stained with Harris’s hematoxylin for nuclear staining, differentiated, blued, and then counterstained with eosin for cytoplasmic staining. After dehydration and clearing, the slides were mounted with a neutral resin.

Immunofluorescence staining

Colons were harvested, flushed with cold PBS, and immediately embedded in Optimal Cutting Temperature (OCT) compound. Tissues were snap-frozen in liquid nitrogen and stored at -80 °C until sectioning. Cryosections (10 μm thickness) were prepared using a cryostat (Leica CM1950), fixed in 4% paraformaldehyde for 15 min at room temperature, and permeabilized with 0.3% Triton X-100 in PBS for 10 min. After blocking with 5% normal donkey serum in PBS for 1 h at room temperature, sections were incubated overnight at 4 °C with primary antibodies diluted in blocking buffer. Primary antibodies used included: anti-F4/80 (for macrophages, Cell Signaling Technology, clone D2S9R, 1:200), anti-GSDMD (for pyrototic cells, Abmart, 1:200), anti-iNOS (for M1 macrophages, Abmart, 1:250), and anti-eNOS (for M2 macrophages, Abmart, 1:200). After washing, sections were incubated with appropriate fluorophore-conjugated secondary antibodies (Alexa Fluor 488, 594, or 647, Invitrogen, 1:500) for 1 h at room temperature in the dark. Nuclei were counterstained with Hoechst for 5 min. Finally, sections were mounted with anti-fade mounting medium (Beyotime). Fluorescent images were captured using a confocal laser scanning microscope (Olympus, FV3000) and analyzed with ImageJ (NIH). Quantification of positive cells or fluorescence intensity was performed on at least three randomly selected fields per section.

Flow cytometry

Dissected colon tissues were washed, stripped of fat and Peyer’s patches, and opened longitudinally. Tissues were cut into small pieces and incubated in pre-warmed digestion buffer containing HBSS with 5% FBS, 1 mM DTT, 5 mM EDTA, and 1 mg/mL Collagenase D (Roche) at 37 °C with gentle shaking for 30–45 min. The cell suspension was filtered through a 70-µm cell strainer, washed, and resuspended in FACS buffer (PBS with 2% FBS). Red blood cells were lysed using ACK lysis buffer. For surface staining, approximately 1 × 10^6 cells were incubated with anti-mouse CD16/CD32 (Fc Block, BD Biosciences) for 10 min on ice, followed by staining with fluorochrome-conjugated antibodies against surface markers for 30 min in the dark at 4 °C. For intracellular cytokine staining, cells were treated with fixation/permeabilization using the Cytofix/Cytoperm kit (BD Biosciences). Data were acquired on CytoFLEX (Beckman). Gating strategies were based on forward/side scatter to exclude debris and doublets, followed by sequential gating for specific cell populations. The antibody panel included: PerCP-Cy5.5 anti-CD4, FITC anti- IFN-γ, PE anti-IL17 and APC anti-F4/80, all from BioLegend.

Transmission electron microscopy (TEM)

To assess pyroptotic morphology, isolated primary macrophages were processed for TEM. Cells were pelleted and fixed similarly. After primary fixation, samples were washed three times with 0.1 M phosphate buffer and post-fixed with 1% osmium tetroxide in the same buffer for 1.5 h at 4 °C. Samples were then dehydrated through a graded ethanol series (50%, 70%, 90%, 100%), followed by infiltration with propylene oxide and embedding in Epon 812 epoxy resin. Ultrathin Sect.  (70 nm thickness) were cut using an ultramicrotome (Leica EM UC7), collected on copper grids, and stained with 2% uranyl acetate and Reynold’s lead citrate. Grids were examined under a transmission electron microscope (Hitachi HT7800) operated at 80 kV. Images were captured using a digital camera. Pyroptotic cells were identified by classical morphological features, including plasma membrane rupture, large pores, cellular swelling, and release of cytoplasmic contents while maintaining intact organelles initially.

Experimental timeline

Recipient mice received lethal irradiation on day 0 followed by bone marrow and splenocyte transplantation. For early mechanistic validation studies, tissues were harvested on day 14 post-transplantation. For therapeutic efficacy studies, treatment was initiated on day 7 post-transplantation and tissues were collected on day 28 unless otherwise specified. Mice were monitored for survival up to 40 days.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Zhuo Wan, Zhenyu Xu and Xuanting Liang designed the experiments, performed major investigations, analyzed data, and wrote the original manuscript; Yueru Ji and Jiayue Xi carried out experimental validation and performed formal analysis; Zhuoyue Shi and Yufei Hou provided critical resources, conducted data curation and bioinformatic processing; Xueqian Yan , Li Liuand Nian Li assisted in experimental work and visualization; Chaojun Song supervised the project; Li Fan and Weiwei Qin jointly supervised the research, reviewed and edited the manuscript, and secured financial support.

Funding

This study was supported by the Special Support Program for High Level Talents in Shaanxi Province, and the Shaanxi Provincial Health Scientific Research Innovation Platform Project (2025PT-07) to Weiwei Qin, “Clinical medicine + Pharmacy” research center research fund of the Air Force Medical University (Grants No. LHJJ2023-YX05) to Li Fan, and Xinfei Project of the Air Force Medical University to Zhuo Wan.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This study was designed and monitored in accordance with sponsor procedures and in compliance with the principles of the Declaration of Helsinki and approved by the Medical Ethics Committee of Tangdu Hospital, the Fourth Military Medical University of the People’s Liberation Army.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.