Non-pathogenic E. coli displaying decoy-resistant IL18 mutein boosts anti-tumor and CAR NK cell responses

Abstract

The tumor microenvironment can inhibit the efficacy of cancer therapies through mechanisms such as poor trafficking and exhaustion of immune cells. Here, to address this challenge, we exploited the safety, tumor tropism and ease of genetic manipulation of non-pathogenic Escherichia coli (E. coli) to deliver key immune-activating cytokines to tumors via surface display on the outer membrane of E. coli K-12 DH5α. Non-pathogenic E. coli expressing murine decoy-resistant IL18 mutein (DR18) induced robust CD8⁺ T and natural killer (NK) cell-dependent immune responses and suppressed tumor progression in immune-competent colorectal carcinoma and melanoma mouse models. E. coli K-12 DH5α engineered to display human DR18 potently activated mesothelin-targeting chimeric antigen receptor (CAR) NK cells and enhance their trafficking into tumors, which extended survival in an NK cell treatment-resistant mesothelioma xenograft model by enhancing TNF signaling and upregulating NK activation markers. Our live bacteria-based immunotherapeutic system safely and effectively induces potent anti-tumor responses in treatment-resistant solid tumors, motivating further evaluation of this approach in the clinic.

The development of recent immunotherapeutic approaches, including the use of immune checkpoint blockade, chimeric antigen receptor (CAR) T cells, CAR natural killer (NK) cells and tumor-infiltrating lymphocytes, has substantially helped to improve the treatment of patients with advanced malignancies^1–4. However, most advanced and metastatic malignancies remain incurable and, therefore, represent a major unmet need^5,6. Lack of tumor-specific targets in most cancer types, poor tumor trafficking of the immune effector cells, including CAR T and NK cells, and their dysfunction and exhaustion in the highly immune-suppressive tumor microenvironment (TME) are still some of the major hurdles to further improving immunotherapeutic approaches to cancer^2,3.

Recently, many genera of bacteria (Escherichia, Salmonella, etc.) have been reported to preferentially colonize hypoxic tumor tissues, generating renewed interest in potentially harnessing live bacteria to deliver novel payloads to modulate the TME⁷. Additionally, several early-phase clinical trials demonstrated the feasibility and safety of using engineered bacteria, although with very limited efficacy^8–10. Modern synthetic biology and genetic modification tools have allowed the engineering of bacteria to deliver payloads, including various immune-modulating molecules^11–15, toxins^12–14,16 and CAR T cell-stimulating tags¹⁷, or to guide their spatiotemporal delivery to the TME^18,19. However, most of these approaches depend on the secretion and/or self-lysis of the bacteria to release the therapeutic payload, which often limits their half-life and efficacy^12,15.

Although bacterial surface display technologies have found wide applications in antibody–antigen screening and biocatalysis^20,21, the use of this technology to display immune-activating cytokines for cancer immunotherapy has not been previously explored. We hypothesized that the expression of these molecules on the bacterial outer membrane would enhance their ability to effectively induce potent immune responses through surface anchoring and clustering, and the use of non-pathogenic facultative anaerobe bacteria would mediate their safe, efficient and preferential delivery into the TME. The Escherichia coli K-12 DH5α, originally isolated from native gut microbiota in humans, is non-pathogenic with no major toxins reported (BSL-1 strain)²². Furthermore, these bacteria are sensitive to the commonly used antibiotics (pan-sensitive), thus providing an additional layer of safety for their potential clinical applications²³.

In the present study, we developed an immunotherapy approach for the preferential TME delivery and enhanced exposure and retention of key immune-activating cytokines by assembling them on the outer membrane of tumor-tropic and non-pathogenic bacteria E. coli K-12 DH5α. We successfully expressed murine and human IL15, IL18, decoy-resistant IL18 mutein (DR18) and IL21 on the outer membrane of E. coli K-12 DH5α with multiple bacterial scaffolds. The bacteria displaying murine DR18 induced the most potent CD8⁺ T and NK cell-dependent immune responses in tumor-bearing immune-competent syngeneic mouse models. The engineered bacteria demonstrated tumor tropism, and the abscopal and recall responses suggested epitope spreading and induction of immunologic memory. The E. coli K-12 DH5α displaying human DR18 combined with human CAR NK cells in tumor-bearing xenograft mice demonstrated its ability to enhance CAR NK cell tumor trafficking, thus further broadening its translational potential (Supplementary Fig. 1). Moreover, gene expression analysis provided clues to the underlying mechanisms mediating their CAR NK cell activation. In summary, we describe here the development of a bacterial therapy platform by successfully engineering non-pathogenic E. coli K-12 DH5a to display potent cytokines for enhanced immunotherapy.

Results

Non-pathogenic E. coli K-12 DH5α engineered to display cytokines

We hypothesized that engineering non-pathogenic facultative anaerobe E. coli K-12 DH5α would allow us to display highly activating cytokines on its outer membrane and, thereby, deliver them preferentially to the TME and trigger a potent immune response. IL12, IL15, IL18 (and its muteins) and IL21 are some of the most potent immune-activating cytokines, demonstrating promising activity alone and/or in combination with other immunotherapeutic approaches in pre-clinical tumor models and patients with advanced malignances²⁴. Additionally, IL18 muteins, such as DR18, were developed for decreased binding affinity to the IL18 binding protein, thus promoting its interaction with the IL18 receptor²⁵. Despite the promising efficacy of these cytokines, short half-life, lower effective concentrations in the TME and systemic toxicity are some of the major limitations hampering their therapeutic application²⁴.

To overcome these major hurdles, we engineered E. coli K-12 DH5α to surface display these major cytokines. For this, we chose five widely used bacterial scaffolds: lipoprotein fused with E. coli outer membrane protein A (Lpp–OmpA); C terminal of IgA proteinase (C-IgAP); N terminal of eaeA (Neae); YiaT protein with the N terminal 1–232 amino acids (YiaT232); and YiaT protein with the N terminal 1–181 amino acids (YiaT181)^26–28. We engineered these scaffolds to present eight different cytokines: murine interleukin-15 (mIL15), murine decoy-resistant interleukin-18-CS2 (mDR18), murine interleukin-21 (mIL21), human interleukin-15 (hIL15), human interleukin-18 (hIL18), human decoy-resistant interleukin-18–6-12 (hIL18–6-12), human decoy-resistant interleukin-18–6-29 (hIL18–6-29) and human interleukin-21 (hIL21). We cloned these cytokine–scaffold fusion constructs into pLygo, pDS861 or pDSG232 plasmids using rhamnose or tetracycline-inducible promoters (P_rha or P_tet). E. coli K-12 DH5α was then transformed with these plasmids, and the expression of the cytokines was assessed by flow cytometry after staining the bacteria with the corresponding antibodies (Fig. 1a and Supplementary Fig. 2). Human and murine IL15, IL18 and DR18 were all expressed regardless of the scaffold used, whereas human and murine IL21 could be displayed only by Neae (Fig. 1b). We were unable to display IL12 (data not shown), most likely due to its post-translation modifications requiring a eukaryotic cell system. Based on these results, we decided to focus on bacteria displaying murine and human IL15, IL18 and DR18 for further evaluation.

Fig. 1 |. Murine DR18 displayed by bacterial scaffold Lpp–OmpA induces potent anti-tumor responses in an immune-competent syngeneic mouse model (MC38).

a, Schematic figure showing the surface display of cytokines in E. coli by their transformation with the cytokine–scaffold coding plasmids and the surface expression assessed by flow cytometry. b, Heatmap showing the expression levels of eight cytokines (mIL15; mIL18-CS2, decoy-resistant murine interleukin-18; mIL21; hIL15; hIL18; two types of human decoy-resistant IL18, hIL18–6-12 and hIL18–6-29; and hIL21) displayed by five different bacterial scaffolds: C-IgAP, Neae, Lpp–OmpA (OmpA), YiaT181 and YiaT232. The display levels are represented by median fluorescence intensity (MFI) measured by flow cytometry. c, Mean tumor volumes as assessed on day 15 after tumor inoculation in different groups of mice treated with PBS, mIL18-CS2 (mDR18, 0.4 mg kg⁻¹), mIL15 (0.4 mg kg⁻¹) or bacteria displaying mDR18, mIL15 or scaffold only by C-IgAP, Lpp–OmpA, YiaT181, YiaT232 or Neae (10⁹ CFU) i.t. three times on days 7, 10 and 14 (n = 6 each group). d, As mentioned in Methods, mice were treated with PBS, mDR18 (mIL18-CS2, 4 mg kg⁻¹), OmpA (0.25 × 10⁹ CFU) or OmpA–mDR18 (0.25 × 10⁹ CFU). e,f, Mean tumor growth (e) and Kaplan–Meier survival (f) curves for mice bearing MC38 tumors after treatment with OmpA–mDR18 (0.25 × 10⁹ CFU, n = 10), OmpA (0.25 × 10⁹ CFU, n = 10), mDR18 (4 mg kg⁻¹, n = 8) and PBS (n = 10). g, As mentioned in Methods, mice were treated with PBS, OmpA (0.5 × 10⁹ CFU) or OmpA–mDR18 (0.5 × 10⁹ CFU). h,i, Mean tumor growth (h) and Kaplan–Meier survival (i) curves for mice bearing B16F10 tumors after treatment with OmpA–mDR18 (0.5 × 10⁹ CFU, n = 10), OmpA (0.5 × 10⁹ CFU, n = 10) and PBS (n = 10). Two-way ANOVA test for tumor growth curve (e,h) and Mantel–Cox test for survival curve (f,i). Data represent mean ± s.d. (c,e,h).

To assess and compare their ability to induce an immune response, we first screened these leading candidates (murine IL15, IL18 and DR18 with different bacterial scaffolds) in an immune-competent mouse model: C57BL/6 bearing the syngeneic colon cancer cell line MC38 (murine colorectal carcinoma cell line). In total, 0.5 million MC38 cells were implanted into both the flanks (subcutaneously (s.c.)) of each mouse; and starting from day 7, when the tumor volume reached 50–100 mm³, 1 billion colony-forming units (CFU) of bacteria were injected intratumorally (i.t.) into each tumor on days 7, 10 and 14. The bacteria with murine DR18 displayed by OmpA (OmpA–mDR18) demonstrated the most effective tumor control and were, therefore, chosen for further assessment (Fig. 1c and Supplementary Figs. 3a,b and 4a–e).

Before further evaluation of the OmpA–mDR18 system, we increased its expression levels by optimizing the induction conditions and switching to a high-copy plasmid backbone with an alternate replication origin (Supplementary Fig. 5). All subsequent experiments were performed using these optimized OmpA–mDR18 bacteria. To further evaluate the ability of OmpA–mDR18 to both control the tumor growth and improve survival in these animals, we treated C57BL/6 mice bearing s.c. injected (unilaterally in flanks) MC38 cells with the following treatment groups: OmpA–mDR18 bacteria, OmpA (bacteria with empty scaffold), mDR18 (purified murine DR18) and PBS, all injected i.t. on days 7, 10 and 14 (Fig. 1d). The optimized OmpA–mDR18 bacteria in these mice demonstrated very effective anti-tumor responses even with a lower bacterial dose (0.25 billion CFU versus 1 billion CFU of the non-optimized bacteria tested previously; Fig. 1c). In these mice, tumors were non-palpable by approximately day 21 after tumor inoculation in most mice treated with OmpA–mDR18, with 50% of these mice remaining tumor free beyond day 100. OmpA–mDR18 bacteria were more potent than the purified mDR18 in these mice and demonstrated highly effective tumor control, with a median survival of 53 d versus 28.5 d in mice receiving purified mDR18 (P < 0.0001) (Fig. 1e,f). Furthermore, OmpA–mDR18 bacteria at 0.25 billion CFU dose used in these experiments were safe. At a higher dose (1 billion CFU), however, even though tumor control was more dramatic, most (~80%) of the mice demonstrated weight loss, hunched backs or slowed movements within days after the bacterial injections, potentially indicating induction of a hyper-inflammatory response at these doses.

To assess the broader applicability of the OmpA–mDR18 bacteria, we also evaluated them in a melanoma tumor model with C57BL/6 mice bearing B16F10 tumor cells. Because B16F10 is widely resistant to immunotherapy due to the downregulation of major histocompatibility complex class I (MHC-I) and lack of major histocompatibility complex class II (MHC-II) expression²⁹, we used five (two times weekly) injections in this model, with 0.5 billion CFU per dose (Fig. 1g). Again, we observed substantial tumor control and improved survival in the OmpA–mDR18 bacteria treatment group, with a median survival of 38 d in the OmpA–mDR18 group versus 21 d in the OmpA group (P < 0.0001). Approximately 30% of mice in the OmpA–mDR18 group stayed tumor free until 60 d (last follow-up) and, thus, were deemed ‘cured’ (Fig. 1h,i).

To address if the in vivo anti-tumor responses induced by the OmpA–mDR18 bacteria were associated with the induction of immunologic memory to the tumors, we rechallenged tumor-free animals from the above experiment with MC38 cells on the contralateral flank 2 months after the initial clearance of tumors. All mice in the rechallenged group remained tumor free for at least 60 d after their tumor inoculation, thus indicating the induction of a strong immunologic memory to these tumors (Extended Data Fig. 1a,b). We also evaluated recall responses in the B16F10 model in a similar manner, and, as before, all rechallenged mice remained tumor free by the study endpoint (60 d after rechallenging) (Extended Data Fig. 2a,b).

Furthermore, tocompare the anti-tumor efficacy of OmpA–mDR18 to the clinically approved immunotherapy, we subsequently treated four groups of mice bearing MC38 (50–100 mm³) with PBS, OmpA–mDR18 (0.25 × 10⁹ CFU, i.t.), anti-PD-1 (InVivoMAb anti-mouse PD-1, clone: RMP1–14, 8 mg kg⁻¹, intraperitoneally (i.p.)) or OmpA–mDR18 (0.25 × 10⁹ CFU, i.t.) + anti-PD-1 (8 mg kg⁻¹, i.p.) as described in Extended Data Fig. 3a. OmpA–mDR18 (i.t., 60% cure rate) was more effective than anti-PD-1 in mediating tumor control in these mice (P = 0.0005 for tumor growth curve, P = 0.03 for survival curve). Additionally, we saw very potent synergistic effect with anti-PD-1 (i.p.) + OmpA–mDR18 (i.t.), as 90% were tumor free as of the last follow-up in this treatment group (Extended Data Fig. 3b,c).

To address the safety concern that i.t. injection of bacteria would possibly lead to bacterial dissemination to other organs, a separate cohort of mice bearing MC38 (300–500 mm³) was injected with bacteria (OmpA and OmpA–mDR18) and then euthanized separately at 12 h, on day 3 or day 7 post injection, tumor was collected for cryosection slides, and other organs were homogenized for CFU assay. No bacteria were detected in liver, lung, spleen, bone marrow and blood at these timepoints (Supplementary Fig. 6). Within the tumor beds, the bacterial distribution highly depended on the injection site. Specifically, bacteria distributed evenly close to the injection site at 12 h after injection. However, at 72 h after injection, within the tumor tissue, bacteria localized primarily at the edge of the necrotic region closer to the injection site (Supplementary Fig. 7).

The strong recall responses seen in these immune-competent models implied the induction of an adaptive immune response and possibly epitope spreading typically mediated by T cells. To investigate this possibility, we evaluated abscopal responses induced by the engineered bacteria by first injecting MC38 tumor cells (s.c., 0.5 million cells on the left flank and 0.3 million cells on the right flank) of each animal. When the tumor volume of the right flank reached 50–100 mm³, 0.5 billion CFU bacteria were injected only into the left flank tumors on days 7, 11, 14, 17 and 21, and the tumor volumes were assessed (Fig. 2a). The mice treated with OmpA–mDR18 bacteria had improved tumor control compared to mice treated with non-mDR18 bacteria (OmpA) on both the treated (P < 0.0001, left flank) and the untreated (P < 0.0001, right flank) sides, supporting the induction of an abscopal effect and, thus, a strong systemic immune response (Fig. 2b,c).

Fig. 2 |. Systemically delivered tumor-homing OmpA–mDR18 E. coli induces an abscopal effect and is safe.

a, Mice were s.c. engrafted with 0.5 × 10⁶ and 0.3 × 10⁶ MC38 cells on the left and right flanks. Starting from day 7, tumors on the left side were treated with PBS (n = 8), OmpA–mDR18 (0.5 × 10⁹ CFU, n = 10) or OmpA (0.5 × 10⁹ CFU, n = 10) (i.t.) five times on days 7, 11, 14, 17 and 21. b,c, Mean tumor growth on the treated side (b) and untreated side (c) of mice. d, Mice engrafted with 10⁶ MC38 cells were i.v. treated with PBS (n = 9), OmpA (n = 10), OmpA–mDR18 (n = 10) or mDR18 (n = 9) on days 8, 11, 15 and 18. e,f, Mean tumor growth (e) and Kaplan–Meier survival (f) curves of mice. g, On day 8 after engraftment of 10⁶ MC38 cells, mice (n = 8 each group) were treated once (i.v.). Plasma was isolated from blood collected from the submandibular vein on days 11 and 15 for cytokine measurement. Tissues were collected on days 7, 10 and 14 after injection for biodistribution. h, Bacterial distribution in tumor, liver, kidney and spleen on days 7, 10 and 14 after injection. CFU g⁻¹ is bacterial concentration. i, The concentration of cytokines associated with CRS: IL1β, IL6, monocyte chemoattractant protein-1 (CCL2), IFNγ, IL2 and granulocyte macrophage colony-stimulating factor (GM-CSF) in the plasma (on days 3 and 7 after treatment unless otherwise mentioned) from mice treated with LPS (5 h after treatment), PBS, mDR18, OmpA and OmpA–mDR18. Mice were treated with PBS, 4 mg kg⁻¹ mDR18, 10⁹ CFU OmpA or OmpA–mDR18 (d–h). Two-way ANOVA test for growth curve (b,c,e), Mantel–Cox test for survival curve (f), one-way ANOVA test and two-sided unpaired t-test for comparison (i). Data represent mean ± s.d. (b,c,e,i).

In summary, we successfully displayed key activating cytokines on the outer membrane of E. coli, with mDR18 displayed by the OmpA scaffold (OmpA–mDR18) inducing very potent tumor control (and prolonging survival), recall and systemic responses in two immune-competent mouse models. Furthermore, the induction of immunological memory and abscopal effects underscores the high translation potential of this immunotherapeutic system.

Tumor enrichment, systemic efficacy and safety of OmpA–mDR18

To assess the feasibility, safety and potential efficacy of systemically delivered engineered bacteria, we used C57BL/6 mice bearing MC38 tumors as above but infused the bacteria through tail vein injections (Fig. 2d). The mice subsequently received intravenous (i.v.) injections of OmpA–mDR18-expressing E. coli (1 × 10⁹ CFU) or controls (Fig. 2d) on days 8, 11, 15 and 18. The mice received treatment after their tumor volume was larger than 100 mm³ (typically by day 8), as a previous study showed that systemically delivered E. coli efficiently homes into the tumors of this volume in mice³⁰. We saw better tumor control and survival benefit in mice that received OmpA–mDR18 versus OmpA-expressing bacteria (P = 0.0003), and neither OmpA nor mDR18 led to any tumor control on its own (Fig. 2e,f). Furthermore, despite systemic delivery of the live bacteria, none of the mice treated with bacteria in our study developed any obvious toxicity as assessed by weight loss, hunched posture or unexpected death (Supplementary Fig. 8a). Twenty percent of mice (2/10) treated with the purified recombinant cytokine (mDR18) were found to have liver failure and hydroperitoneum upon euthanization, suggesting considerable toxicity upon systemic delivery of this cytokine. Moreover, systemically (i.v.) delivered OmpA–mDR18 outperformed the treatment with anti-PD-1 therapy in controlling the growth of relatively large established tumors (100–200 mm³, P < 0.0001 for tumor growth curve and P = 0.0018 for survival curve) in another independent batch of mice bearing MC38. Additionally, systemic co-administration of OmpA–mDR18 (i.v.) and anti-PD-1 (i.p.) furtherly enhanced anti-tumor efficacy with 60% of mice remaining tumor free as of the last follow-up (Extended Data Fig. 4a–c). Furthermore, we tested the engineered bacterial cancer therapy in a well-established B16F10 melanoma lung metastasis model by injecting B16F10-luciferase tumor cells i.v. followed by PBS, anti-PD-1, OmpA–mDR18 and anti-PD-1 + OmpA–mDR18 weekly for four times in total. The mice were euthanized on day 30, and lungs were harvested and assessed for the surface tumor nodule formation. We found that the i.v. injection of OmpA–mDR18 was associated with a substantially reduced number of tumor nodules (metastases) in the lungs. In addition, we observed a potential synergism between the engineered bacteria and anti-PD-1 in reducing metastasis, as the least number of tumor nodules was observed in the mice treated with this combination (Extended Data Fig. 5). These promising data, particularly with the systemic delivery of the engineered bacteria, make a strong case for the evaluation of this combination in the clinic in the near future.

Because of the lack of side effects and potent tumor control with the systemic delivery of the engineered bacteria, we hypothesized that these bacteria were preferentially enriched in the TME due to their facultative anaerobic nature. To test this hypothesis, we injected OmpA–mDR18 bacteria i.v. through the tail vein in another batch of C57BL/6 mice bearing MC38 tumors. Upon euthanization on days 15, 18 and 22, various organs and tumors were harvested for the enumeration of CFU (Fig. 2g). In agreement with our hypothesis, we observed preferential enrichment of engineered bacteria in the tumors of the treated mice. Limited number of bacteria also colonized liver, lung and kidney on day 15 but were undetectable after day 18 (day 10 after bacterial injection). Bacterial colonization within tumor was undetectable after day 22 (day 14 after bacterial injection). No bacteria were detected in other organs, including central nervous system, lung, heart, blood, stomach, small intestine, colon, bone marrow and cecum (Fig. 2h and Supplementary Fig. 9). Furthermore, at 72 h after injection within the tumor tissue, the bacteria preferentially colonized the necrotic region relatively evenly (Supplementary Fig. 10). The lack of any toxicity was also consistent with the cytokine profiles from the treated mice (purified lipopolysaccharide (LPS)-treated mice as positive control) showing no major increase in the cytokines typically associated with the development of sepsis and cytokine release syndrome (CRS) in the mice after i.v. injection of the OmpA–mDR18 displaying E. coli^31,32 (Fig. 2g–i and Supplementary Figs. 11 and 12). However, we observed increased levels of some inflammation-relevant proteins in the mice treated with the purified cytokine (mDR18) but not with engineered bacteria, consistent with systemic toxicity seen in mice treated with this cytokine (Fig. 2i and Supplementary Figs. 11 and 12). The results from these experiments highlight the potential safety of the engineered bacteria even when delivered systemically in future clinical trials.

OmpA–mDR18 E. coli-induced anti-tumor responses are dependent on NK cells and CD8⁺ T cells

The potent anti-tumor responses along with the recall and abscopal effects induced with the engineered bacteria above in immune-competent animals suggested activation of the adaptive immune system in vivo. To investigate the potential impact of bacterial therapy on immune cells in the TME, we first performed a detailed profiling of immune cells within the tumors after bacterial injections. Specifically, on days 1 and 3 after a single i.t. injection (OmpA, OmpA–mDR18, mDR18 or PBS, in two separate cohorts of mice), the s.c. MC38 tumors were harvested after euthanization, and single cells were stained with a broad antibody panel for flow cytometry analysis (day 1 and day 3 after treatment) (Fig. 3a and Extended Data Fig. 6a). We observed an increase in the CD8⁺ T cells (P = 0.0067), NK cells (P = 0.007) and granulocytes (P = 0.0086) infiltrating the tumors after treatment with the OmpA–mDR18 bacteria versus controls (OmpA bacteria) (Fig. 3b). We also observed a decrease in the number of tumor-associated macrophages (TAMs; P = 0.0024, PBS versus OmpA–mDR18) and mononuclear myeloid-derived suppressor cells (M-MDSCs) (P = 0.0479, PBS versus OmpA–mDR18) in the mice treated with OmpA–mDR18 (and OmpA, to a lesser extent) bacteria, potentially indicating a broader TME modulation beyond the effects on T and NK cells (Fig. 3b and Extended Data Fig. 6b).

Fig. 3 |. Tumor control by OmpA–mDR18 is mediated by CD8⁺ T cells and NK cells in the TME.

a, C57BL/6 mice were s.c. engrafted with 0.5 × 10⁶ MC38 cells in the flanks. When tumor size reached 150–200 mm³ on day 10, mice (n = 3–5 each group) were i.t. treated with PBS, OmpA–mDR18 (0.5 × 10⁹ CFU), OmpA (0.5 × 10⁹ CFU) or mDR18 (4 mg kg⁻¹). The mice were euthanized, and the tumors were harvested for flow cytometric analysis of tumor-infiltrating immune cells (CD8⁺ T cells, CD4⁺ T cells, NK cells, T-like NK (NKT) cells, macrophages, monocytes, granulocytes and M-MDSCs) 3 d after the bacterial injection. b, The percentages of different cell types (as a proportion of live cells): CD8⁺ T cells (CD45⁺CD3⁺CD8⁺), CD4⁺ T cells (CD45⁺CD3⁺CD4⁺), NK cells (CD45⁺CD3⁻NK1.1⁺), NKT cells (CD45⁺CD3⁺NK1.1⁺), granulocytes (CD45⁺CD11b⁺Ly6C⁺Ly6G⁺), monocytes (CD45⁺CD11b⁺Ly6C^lowLy6G⁻), macrophages (CD45⁺CD11b⁺F4/80⁺) and M-MDSCs (CD45⁺CD11b⁺Ly6C^highLy6G⁻) in the TME. c, C57BL/6 mice (n = 10 each group) were s.c. engrafted with 0.5 × 10⁶ MC38 cells and, on day 6, i.p. treated with monoclonal antibodies anti-CD8α, anti-NK1.1, anti-ly6G or PBS and then i.t. treated with OmpA–DR18 (0.5 × 10⁹ CFU) or PBS on days 7, 10 and 13. d,e, Mean tumor growth (d) and Kaplan–Meier survival (e) curves for mice bearing MC38 treated with mAbs or PBS and bacteria or PBS. Two-sided unpaired t-test for percentage data (b), two-way ANOVA test for tumor growth curve (d) and Mantel–Cox test for survival curve (e). Data represent mean ± s.d. (b,d).

To confirm the potential contribution of T cells, NK cells and granulocytes to the anti-tumor responses induced by OmpA–mDR18 engineered bacteria, we depleted the relevant cell populations using anti-NK1.1 (NK cell depletion), CD8a (CD8⁺ T cell depletion) and anti-ly6G (granulocyte depletion) monoclonal antibodies (mAbs), respectively (Fig. 3c). In brief, the MC38 tumor cells were implanted into the flanks of C57BL/6 mice and then divided into the groups receiving PBS (control) or different depletion mAbs i.p. on days 6, 7, 10 and 13. In addition, 0.5 × 10⁹ CFU of OmpA–mDR18 bacteria were administered i.t. on days 7, 10 and 13 after tumor cell inoculation. In parallel, a separate group of mice was i.p. and i.t. treated with PBS with a similar schedule as mAbs and engineered bacteria as a negative control. The depletion of the specific cell types was confirmed by the peripheral blood mononuclear cells (PBMCs) collected from mice in different treatment groups on day 14 and evaluated by flow cytometry (Extended Data Fig. 6c). Consistent with the data above, we detected potent tumor control with approximately 60% of the mice deemed cured and still alive (as of the last follow-up) in the group treated with OmpA–mDR18 (i.t.) and PBS (i.p.). We observed no impact of granulocyte depletion (anti-ly6G), whereas the depletion of NK cells (P < 0.0001, median survival 28.5 d) and CD8⁺ T cells (P < 0.0001, median survival 16 d) led to a major decrease in tumor control, thereby underscoring the importance of these two cell types in mediating the anti-tumor responses seen with our engineered bacteria in these mice (Fig. 3d). Compared to the negative control group (PBS, i.t. and PBS, i.p.), the depletion of CD8⁺ T cells (anti-CD8a) resulted in almost complete abrogation, whereas the depletion of NK cells (anti-NK1.1) resulted in only partial loss of tumor control in these mice (Fig. 3d,e). We also performed similar experiments in the B16F10 model using similar mAbs and bacterial doses and schedules (Extended Data Fig. 7a). Again, we saw minimal to no impact of the granulocyte depletion, whereas the depletion of NK cells (P = 0.0064, median survival 27 d) and CD8⁺ T cells (P = 0.0025, median survival 20 d) both eliminated the treatment efficacy (Extended Data Fig. 7b,c).

To assess potential contribution of the humoral immune response in mediating the therapeutic efficacy by the bacterial therapy, we performed two sets of experiments. First, we investigated the proportion of plasma cells and B cells in the TME and inguinal lymph nodes (LNs) (tumor-draining LN) on day 7 after bacterial injections (i.t.) (Supplementary Fig. 13a). We saw a significant increase in the B cell percentages in the LN but not in the TME. We did not see any increase in the plasma cells in the LN, although very rare plasma cells were detected in the TME (Supplementary Fig. 13b). Next, to evaluate a possible contribution of the humoral immune responses to the tumor control seen with OmpA–mDR18, B cell depletion in these mice was performed using anti-CD19 and anti-B220 mAbs (i.p.). We did not observe any considerable difference in the tumor growth or survival between the B-cell-depletion versus no-depletion animals (P = 0.7734 for tumor growth curve and P = 0.3360 for survival curve) (Supplementary Fig. 13c–f).

The data from these mechanistic studies potentially point toward OmpA–mDR18 E. coli mediating substantial TME modulation and impacting multiple immune cells, with CD8⁺ T cells and NK cells being the major effector cells responsible for inducing potent anti-tumor responses by these engineered bacteria.

E. coli displaying human DR18 enhances tumor trafficking and therapeutic efficacy of CAR NK cells

To assess if this bacterial engineering approach would also work with human cytokines, we replaced the murine DR18 with a human decoy-resistant IL18 (hDR18) that was previously developed²⁵. Based on our murine data, this could work in combination with both CAR T and NK cells; however, we chose CAR NK cells to avoid the possibility of exacerbating the risk of developing side effects, such as severe cytokine release commonly seen with CAR T cells^33–35. We first evaluated the expression of wild-type (WT) and two IL18 muteins (hIL18–6-12 and hIL18–6-29, both decoy resistant) in combination with different bacterial scaffolds using flow cytometry and assessed their activity upon co-culture with the HEK-Blue IL18 reporter cells (Fig. 4a). IL18–6-12 mutein (hereafter referred to as hDR18) displayed by C-IgAP, YiaT232 or YiaT181 scaffolds had the highest IL18 activity (Fig. 4b). The optimized hDR18-displaying E. coli (YiaT232–hDR18 and YiaT181–hDR18) demonstrated high IL18 activity, with multiplicity of infection (MOI) of 10 equivalent to approximately 1,500 pg ml⁻¹ soluble purified hDR18 (Supplementary Fig. 14a–c). Next, we assessed the ability of hDR18-displaying E. coli to enhance the cytotoxicity of the mesothelin (MSLN)-targeting CAR NK cells generated from the peripheral blood of normal healthy volunteers using our optimized baboon lentivirus system³⁶. We chose MSLN-CAR as proof-of-principle NK cell-based therapy as MSLN is highly expressed by several solid tumors, including ovarian cancer, pancreatic cancer and mesothelioma³⁷. Activation of MSLN-CAR NK cells by engineered hDR18 bacteria (with different scaffolds) enhanced their cytotoxicity against MSLN⁺ cell lines (NCI-H226 and NCI-H259). Bacteria displaying hDR18 with the YiaT232 scaffold were the most potent in enhancing the NK cell CAR cytotoxicity (P < 0.0001 toward H226, P = 0.0012 toward H2591) and were, therefore, chosen for further evaluation (Fig. 4c–e).

Fig. 4 |. Bacteria displaying hDR18 enhance anti-tumor responses of MSLN-CAR NK cells.

a, Schematic figure showing the use of HEK-Blue IL18 reporter cells to screen multiple bacterial scaffolds expressing hDR18 for activation. b, The activity (assessed using HEK-Blue in a) of WT human IL18 and two variants of hDR18 (6–12 and 6–29) displayed by five bacterial scaffolds—Neae, C-IgAP, OmpA, YiaT232 and YiaT181—with the MOI of E. coli displaying cytokines and HEK-Blue IL-18 reporter at 10. Human IL18 activity is represented by OD₆₂₀. c, Schematic figure for in vitro co-culture killing assay with MSLN-CAR NK and tumor cells. In brief, MSLN-CAR NK cells were primed by bacteria or other control groups overnight and then co-cultured with the tumor cells for 4 h. Viability of the tumor cells was assessed by Zombie NIR viability dye. d,e, Viability of H226 (d) and H2591 (e) cell lines after co-culture with MSLN-CAR NK cells primed by hDR18-displaying bacteria (YiaT232–IL18–6-12, YiaT232–hDR18; YiaT181–IL18–6-12, YiaT181–hDR18; C-IgAP–IL18–6-12, C-IgAP–hDR18) or control groups for overnight. The MOI of E. coli displaying cytokines and MSLN-CAR NK cells was 1,000. E (effector NK cells):T (target tumor cell lines: H2591 or H226) = 1:1. Two-sided unpaired t-test. Data are representative of three independent experiments with MSLN-CAR NK cells from three independent donors. Data represent mean ± s.d. (b,d,e).

Based on the above promising in vitro results, we hypothesized that the combination of the bacteria with MSLN-CAR NK cells would also lead to enhanced tumor control in vivo in a xenograft mouse model. Furthermore, based on the increased NK cell numbers that we saw in the tumors from immune-competent mice upon their treatment with the engineered bacteria (Fig. 3b), we also hypothesized that this approach would help home adoptively transferred CAR NK cells into tumors and, thus, potentially overcome one of the major challenges in adoptive cell therapies, namely their poor trafficking particularly in the solid tumors. To assess this, we used NSG mice bearing highly resistant NCI-H226 (H226, mesothelioma) tumor cells s.c. and i.t. injected the engineered bacteria (Fig. 5a). In total, 5 × 10⁶ H226 cells were s.c. injected (flanks), and, on day 30 after tumor inoculation when the tumor size reached 80–200 mm³, the mice received 3–5 × 10⁶ CAR NK cells i.v. (tail vein injection). YiaT232–hDR18 (1 × 10⁹ CFU) or purified hDR18 (4 mg kg⁻¹) versus PBS control was i.t. administrated on days 30, 37 and 44 after tumor inoculation. None of the mice had any apparent toxicity from the bacteria despite being highly immune compromised. Notably, the treatment with YiaT232–hDR18 markedly enhanced tumor control (P < 0.0001) and prolonged the survival (median survival 112 d) in these mice, outcompeting the cohort treated with the purified cytokine (hDR18, median survival 80.5 d) (Fig. 5b,c). For the further assessment of the efficacy of our bacteria-NK therapy, we next compared NSG mice bearing H226 treated with YiaT232–hDR18 (i.t.) + MSLN-CAR NK (i.v.) to combination chemotherapy with pemetrexed (100 mg kg⁻¹, i.p.) and cisplatin (5 mg kg⁻¹, i.p.), an FDA-approved therapy for the patients with mesothelioma (Supplementary Fig. 15a). Although both the chemotherapy and bacteria-NK therapy considerably delayed the tumor growth (P < 0.0001), the bacteria-NK therapy outperformed chemotherapy in these mice (P = 0.0023) (Supplementary Fig. 15).

Fig. 5 |. Bacteria displaying hDR18 enhance the proliferation, tumor trafficking and efficacy of MSLN-CAR NK cells in vivo, leading to improved tumor control.

a, NSG mice were s.c. engrafted with 5 × 10⁶ H226 cells. Starting on day 30, mice (n = 10 each group) were treated with PBS, YiaT232 (10⁹ CFU), YiaT232–hDR18 (10⁹ CFU) or purified hDR18 (4 mg kg⁻¹) (i.t.) three times (on days 30, 37 and 44). In total, 3–5 million MSLN-CAR NK cells were i.v. administrated except for mice in the tumor-only groups. All mice were i.p. injected with 75 kU of human recombinant IL2 every other day to support the survival of human NK cells in vivo. b,c, Mean tumor growth (b) and Kaplan–Meier survival (c) curves for the tumor-bearing mice after treatment. Mean tumor growth and survival curves are the combination of two independent experiments, with n = 10 mice per group. d, NSG mice were s.c. engrafted with 5 × 10⁶ H226 cells on day 40, and mice (n = 10 each group) were treated with PBS, YiaT232 (10⁹ CFU), YiaT232–hDR18 (10⁹ CFU) or purified hDR18 (4 mg kg⁻¹) (i.t.). In total, 5 million MSLN-CAR NK cells were i.v. administrated except in the mice from tumor-only groups. All mice were i.p. injected with 75 kU of human recombinant IL2 every other day. On day 47, mice were euthanized, and organs (tumor, liver, lung, spleen and bone marrow) of all mice were collected for analysis by flow cytometry. e, The percentage of human CD45⁺ cells in livers, tumors, spleens, lungs and bone marrow after treatment. Two-way ANOVA test for tumor growth curve (b), Mantel-Cox test for survival curve (c) and two-sided unpaired t-test for NK percentage (e). Data represent mean ± s.d. (b,e).

To further investigate the biodistribution and tumor infiltration of MSLN-NK cells in these mice, a separate cohort of NSG mice was treated in the same fashion with similar tumors and treatments (as in Fig. 5), except that they received only one injection of the engineered bacteria, hDR18 or PBS control and then, a week later, were euthanized for further analysis (Fig. 5d and Extended Data Fig. 8a). Treatment with YiaT232–hDR18 led to an obvious increase in adoptively transferred MSLN-CAR NK cells and Ki67⁺ CAR NK cells into the tumors (P = 0.0067 for MSLN-CAR NK cells, P = 0.0048 for Ki67⁺ CAR NK cells), with the purified hDR18 having only a limited impact (P = 0.0856) (Fig. 5e). There was also an increase in the CAR NK cells and Ki67⁺ CAR NK cells in the livers and lungs with YiaT232–hDR18 bacteria and hDR18, suggesting that this treatment could boost the proliferation of CAR NK cells in vivo (Fig. 5e). Although i.t. injection of bacteria in immune-competent mice did not result in bacterial dissemination to other organs (Supplementary Fig. 6), bacteria were detected in the livers and lungs other than in the tumors, possibly due to the attenuated immune system of the NSG mice, which also potentially explains the proliferation of NK cells in these organs (Extended Data Fig. 8b).

These data support the assumption that E. coli expressing hDR18 YiaT232 scaffold (YiaT232–hDR18) is effective in enhancing anti-tumor responses of the CAR NK cells by their direct activation and increasing their tumor infiltration, thus making this combinatorial treatment approach attractive for potential translation in future clinical trials.

TNF signaling via NFκB as the major pathway mediating CAR NK cell activation with YiaT232–hDR18

To better understand potential mechanism(s) mediating enhanced CAR NK cell anti-tumor responses, we performed RNA sequencing (RNA-seq) of MSLN-CAR NK cells, co-cultured with the engineered bacteria. MSLN-CAR NK cells were primed for 3 h with bacteria expressing hDR18 (YiaT232–hDR18, MOI = 1,000), bacteria expressing empty YiaT232 scaffold (YiaT232, MOI = 1,000), soluble purified hDR18 protein (no bacteria, 100 ng ml⁻¹) or PBS control. We focused our analysis on the group of differentially expressed genes, with adjusted P values (P_adj) < 0.05, in the YiaT232–hDR18 bacteria versus PBS (named Y18-PBS) comparison groups (Extended Data Fig. 9a). Principal component analysis (PCA) suggested that the first principal was dominated by the effect of the engineered bacteria on NK cells, and the second principal was dominated by the healthy donor variability (Fig. 6a). In the Y18-PBS group, we further studied the genes that were differentially expressed between YiaT232–hDR18 and YiaT232 (colored green in Fig. 6b and detailed in Fig. 6c). The interaction network for the upregulated genes is described in Fig. 6d, color-coded by the main gene sets that were found to have significant overlaps (Molecular Signatures Database (MSigDB), Investigate tool). These sets included ‘TNFα signaling via NFκB’ (green, P_adj = 2.63 × 10^–23), ‘Cytokine signaling’ (light blue, P_adj = 9.61 × 10^–6) and ‘GPCR signaling’ (dark blue, P_adj = 2.48 × 10^–4). In addition, gene set enrichment analysis (GSEA) also identified ‘TNFα signaling via NFκB’ as one of the most enriched pathways (P_adj = 0.006) (Fig. 6e and Extended Data Fig. 9b). To confirm this, we evaluated the activation of NK cells within PBMCs, primed for 12 h using the above conditions, and then analyzed by flow cytometry. We also found upregulation of the NK activation markers CD25 and CD69 (Supplementary Fig. 16a,b) as well as TNF, interferon-γ (IFNγ) and CD107a (Supplementary Figs. 17 and 18). With these data, we conclude that YiaT232–hDR18 bacteria can efficiently activate NK cells (with or without MSLN-CAR), leading to increased killing efficacy.

Fig. 6 |. Gene expression changes in the MSLN-CAR NK cells induced by priming with E. coli hDR18.

Messenger RNA transcript levels were measured in MSLN-CAR NK cells 3 h after treatment with YiaT232–hDR18, E. coli (MOI = 1,000), YiaT232, E. coli (MOI = 1,000), hDR18 (no bacteria, 100 ng ml⁻¹) and PBS (a–e). a, PCA of genes that were differentially expressed among YiaT232–hDR18, E. coli and PBS (P_adj < 0.05, n = 2,923). Samples were labeled with the different treatment groups (top) and different donors (bottom). b, Volcano plot showing gene expression of YiaT232–hDR18, E. coli and YiaT232–E. coli-treated MSLN-CAR NK cells. Highlighted in green is a subgroup of genes mentioned in a with log₂FC > 0.5 (or log₂FC < −0.5) and P < 0.01 comparing YiaT232–hDR18, E. coli and YiaT232, E. coli. Highlighted by name are genes with corresponding P_adj < 0.05 (n = 43 upregulated genes; n = 6 downregulated genes). The dashed line depicts a P value of 0.05. P values were calculated using DESeq2. c, Normalized expression levels (z-scores) of the genes highlighted in b. d, STRING network analysis for the upregulated genes highlighted in b (n = 43). Genes were labeled according to the significantly enriched MSigDB gene sets: ‘TNFα signaling via NFκB’, ‘Cytokine signaling’ and ‘GPCR signaling’. e, GSEA enrichment plot for the KEGG ‘TNFα signaling via NFκB’ in YiaT232–hDR18, E. coli and YiaT232, E. coli groups.

Discussion

Here we describe the development of an immunotherapeutic approach using engineering non-pathogenic, tumor-tropic and pan-sensitive E. coli K-12 DH5α to surface display key activating cytokines and demonstrate the safety and promising efficacy of this approach in otherwise hard-to-treat tumors. E. coli K-12 DH5α displaying DR18 demonstrated tumor tropism, inducing dramatic CD8⁺ T and NK cell-dependent tumor controls, curing many tumor-bearing mice and outcompeting and synergizing with the anti-PD1 treatment. The recall and abscopal effects suggest induction of immunologic memory and epitope spreading with this approach. The enhanced tumor trafficking and promising efficacy of the engineered bacteria when combined with systemically delivered CAR NK cells in NSG mice bearing subcutaneous tumors demonstrate the potential of this approach to serve as a ‘tumor GPS’ for the systemically delivered CAR treatments in solid malignancies. We also demonstrate the safety of the engineered bacteria in immune-competent mice and in combination with CAR NK cells in highly immune-deficient NSG mice. These findings highlight the feasibility, safety and potential efficacy of this immunotherapy approach, making a strong case for its evaluation in the clinic in patients with otherwise poor prognoses.

The major novelty of our study is the application of surface display of immune-activating cytokines on the bacterial outer membrane, enabling this system to induce potent anti-tumor responses in contrast to the previous bacterial therapy studies using secretion and/or cell lysis-mediated delivery of biologics to the TME^12–15,17. The surface display of biologics in live bacteria leads to substantially enhanced potency due not only to the tumor tropism of the bacteria but also to its increased effective concentration due to the two-dimensional versus three-dimensional mobility^38–40 and prolonged half-life in the TME due to its continued biosynthesis by the live bacteria, E. coli. However, these studies have had limited efficacy in inducing strong and persistent immune responses, unless additional treatments, such as immune checkpoint blockers, were co-administered³⁰. We used E. coli strain K-12 DH5α here for several reasons: it is a facultative anaerobe that facilitates tumor tropism; it is pan-sensitive to multiple commonly used antibiotics; it lacks horizontal gene transfer machinery; and its genetic modification has included knocking out its recombinant system, which makes this strain harder to obtain any mutations for possible resistance to antibiotics^{22,23,41–44}. Furthermore, it also lacks pks gene islands, present in E. coli Nissle 1917 and known to produce colibactin, a genotoxin associated with colorectal cancer⁴⁵.

A previous study demonstrated superior efficacy of the soluble DR18 over regular IL18 in inducing anti-tumor repsonses²⁵. We saw that the bacteria displaying DR18 were not only better tolerated but also vastly superior to the soluble DR18, inducing a potent immune response leading to a cure in a substantial proportion of tumor-bearing animals. The surface displaying of this cytokine on the bacterial outer membrane may lead to its enhanced effective concentration and half-life in close proximity to the effector immune cells in TME along with the potential ‘adjuvant’ function of the bacteria. The differential efficacy of bacterial scaffold proteins for murine versus human DR18 is most likely related to the steric restrictions of their spatial display affecting ligand–receptor interactions (Supplementary Fig. 19). Furthermore, the rechallenge experiments and abscopal effects suggested induction of immunological memory and epitope spreading leading to systemic immune responses and, thus, further increasing the application of our approach to patients with advanced metastatic tumors.

The immune profiling of the tumors after the bacterial therapy demonstrated the ability of these bacteria to modulate otherwise highly immune-suppressive TME, particularly in solid tumors such as melanoma and colorectal carcinoma models. Although we saw increased infiltration of granulocytes into the tumors, our depletion studies demonstrated the anti-tumor tumor responses to be mediated primarily by CD8⁺ T and NK cells, possibly along with a decrease in TAM frequency in the TME. Bacteria have been known to reduce TAMs through pyroptosis and TNF-mediated killing of macrophages and, thus, contribute to their decreased frequency in the tumors observed after our bacterial therapy⁴⁶. The increased homing of the adoptively transferred CAR NK cells and enhanced tumor control seen in xenograft mice bearing otherwise very resistant mesothelioma tumor cells with the bacteria displaying human DR18 further demonstrate the potential application of this approach in combination with CAR therapy. Our RNA-seq data suggest that the increased cytotoxicity of these cells is mainly regulated by ‘TNFα signaling via NFκB’, ‘Cytokine signaling’ and ‘GPCR signaling’ pathways. The NFκB pathway is well known for its crucial role in regulating the proliferation and activation of effector NK and T cells⁴⁷. Indeed, examination of the core genes of the GSEA for the pathway ‘TNFα signaling via NFκB’ identifies key activation genes, such as TNFSF9 (41BB ligand), TNF, CXCL10 (an inflammatory chemokine), CD69 (activation signaling), IL15RA (receptor for IL15), TNFRSF9 (41BB), STAT5A (transcription factor), IFNGR2 (IFNγ receptor 2), MYC (transcription factor) and more. Similarly, CXCL10 is well known to induce NK cell trafficking into the tumors⁴⁸, thus potentially explaining the increased trafficking of the CAR NK cells seen in our study.

Although we saw the induction of potent immune responses that were associated with modulation of the key immune cells with the bacterial therapy, it was reassuring to see no major side effects and relatively low levels of cytokines, including interleukin-1β (IL1β), interleukin-2 (IL2) and interleukin-6 (IL6), known to mediate CRS^31,32,34,35. Similarly, the bacteria did not cause an overwhelming systemic infection even in immune-compromised NSG mice, supporting the potential safety of this approach even in patients with otherwise impaired immunity. The pan-sensitive nature of the E. coli K-12 DH5α to multiple commonly used antibiotics adds an additional layer of safety^23,44.

In summary, we developed a live bacterial therapy-based immunotherapy approach using E. coli K-12 DH5α bacteria displaying key cytokines. The dramatic anti-tumor responses along with the safety seen with these bacteria in otherwise hard-to-treat tumors make this approach quite promising and build a strong case for their evaluation alone or in combination with other immunotherapeutic approaches, including CAR T/NK cells and/or immune checkpoint blockade, in the clinic.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41587-024-02418-6.

Methods

Cell culture

B16F10 (CRL-6475) was purchased from the American Type Culture Collection (ATCC). HEK-Blu IL18 cells were purchased from InvivoGen (hkb-hmil18-av). H226 (CRL-5826) and H2591 (CRL-5939) were requested from David Barbie’s laboratory at Dana-Farber Cancer Institute, originally purchased from ATCC. MC38 (ENH204-FP) was requested from Darrell Irvine’s laboratory at the Koch Institute, originally purchased from Kerafast. HEK-Blue IL18 cells, MC38, B16F10 and B16F10-luciferase-GFP (B16F10-luc; B16F10 transfected by lentivirus to express luciferase and GFP, sorted by GFP positive) were maintained in complete DMEM (Corning, 10–013CV) supplemented with 10% FBS (Corning, 35–016-CV) and 100 U ml⁻¹ penicillin–streptomycin (P/S; Corning, 30–002-CI). For HEK-Blue IL18 cells, 100 μg ml⁻¹ Normocin (InvivoGen, ant-nr-05) and 1× HEK Blue selection (InvivoGen, hb-sel) were added to maintain the cell lines. H226 and H2591 were cultured in RPMI 1640 (Corning, 10–041-CV) supplemented with 10% FBS, 100 U ml⁻¹ P/S and 1× NEAA. All cell lines mentioned were maintained at 37 °C in a humidified incubator with 5% CO₂. Cells in passages 2–10 were used for the experiments.

E. coli surface display and protein structure prediction

For the surface display of cytokines, we used the following plasmids: pLyGo-Ec-7 (Addgene, 163135), pLyGo-Ec-8 (Addgene, 163136), pDSG323 (Addgene, 115594) and pDS861 (from Quintara Biosciences) with YiaT232 or YiaT181. DNA sequences for the encoding cytokines including 3× GGGGS linker between scaffolds and the cytokines with DYKDDDDK-tag (FLAG-tag) in the N terminus and Myc-tag in the C terminus were inserted between two SapI sites for pLyGo-Ec-7 and pLyGo-Ec-8, between SpeI and PstI sites for pDSG323 and between NotI and BamHI sites for pDS861 by NEBuilder HiFI DNA Assembly Master Mix (New England Biolabs (NEB), M5520AVIAL). For the optimization of the surface display of Lpp–OmpA–mDR18, the high-copy plasmid pDS861-dsRed was digested with XbaI and NotI. The E. coli rhaSR-PrhaBAD-inducible promoter system containing regulator proteins (RhaR and RhaS), a rhamnose-responsive promoter (P_rhaBAD) and a ribosome binding site (RBS), along with Lpp–OmpA–mDR18, were amplified from pLyGo-Ec-8-mDR18. The DNA parts mentioned above were assembled with digested pDS861-dsRed backbone by NEBuilder HiFi DNA Assembly Master Mix. All plasmids were validated by Sanger sequencing before the next steps. The DNA fragments for cytokines with an N-terminal Myc tag and a C-terminal FLAG-tag were synthesized by Twist Bioscience, or 3× GGGGS linkers were introduced by primers.

For bacterial induction, E. coli K-12 DH5α with the corresponding plasmid was inoculated in fresh LB medium (Thermo Fisher Scientific, BP97235) with 50 μg ml⁻¹ kanamycin (Sigma-Aldrich, 25389–94-0). After overnight culture in a shaker (37 °C, 250 r.p.m.), bacterial suspensions were diluted by 10-fold in the fresh LB with 50 μg ml⁻¹ kanamycin and 10 mM L-rhamnose (Sigma-Aldrich, 10030–85-0) for derivatives of pLyGo-Ec-7, pLyGo-Ec-8 and pDS861 and in the fresh LB with 50 μg ml⁻¹ kanamycin and 100 ng ml⁻¹ anhydrotetracycline for modified plasmids based on pDSG323. After 48-h induction in a shaker (25 °C, 250 r.p.m.), bacteria were collected for the next step. For the optimized induction of hDR18 displayed by YiaT232, YiaT181 and C-IgAP, overnight bacterial culture was diluted 10 times in the fresh LB containing 50 μg ml⁻¹ kanamycin and 10 mM L-rhamnose monohydrate, and proteins were induced at 25 °C, 250 r.p.m., for 72–96 h. Bacteria were collected after induction for the next step. For bacterial surface display verification, 20 μl of bacterial suspension was collected, washed once with PBS and then incubated with anti-DYKDDDDK Tag Antibody (BioLegend, 637315) in FACS buffer (2% FBS, 0.1% w/v sodium azide, 2 mM EDTA dissolved in DPBS) for 20 min at room temperature, washed two times and then suspended in PBS for flow cytometry.

The structures of various fusion proteins were predicted by ColabFold. The structure images were analyzed by Visual Molecular Dynamics⁴⁹.

Protein expression and purification

Murine DR18, murine IL15 and human DR18 were cloned into pSH200 vector (a generous gift from Xiling Shen at Duke University) containing a 6×Histidine tag (His-tag) via the Gibson assembly after digestion with BamHI and XbaI. Plasmids were verified by Sanger sequencing before expression. The cytokines were expressed and purified as previously described^50–52. Validated protein was aliquoted and kept at −80 °C in the buffer of 3× PBS with 10% glycerol and 1 mM dithiothreitol (Thermo Fisher Scientific, BP172–5) until further use.

Information on primers for cloning and sequencing is included in Supplementary Table 1.

Animal experiments

Six-to-eight-week-old female C57BL/6J (The Jackson Laboratory ( JAX), 000664 or Charles River Laboratories (CRL), C57BL/6NCrl) mice and 4-week-old female NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG, JAX, 005557) mice were purchased and maintained in the animal facility at Northeastern University (NEU). Mice were housed in a specific pathogen-free facility and fed normal chow and water ad libitum under standard animal facility conditions (12-h light/dark cycle, temperature of 22 °C, relative humidity of 40–70%). All animal studies and procedures were performed following federal, state and local guidelines for institutional animal care and approved by the Institutional Animal Care and Use Committee at NEU.

For the scaffold screening experiment, 0.5 million MC38 cells in 30 μl of sterile PBS were s.c. injected into both flanks of C57BL/6 mice. On day 7, when tumor volume reached 50–100 mm³, mice were i.t. injected with PBS, purified protein or engineered bacteria in 20 μl of sterile PBS twice weekly for a total of three injections into tumors on both flanks.

For the i.t. injection experiments (Fig. 1c,d,g), 0.5 million MC38 or B16F10 cells in 30 μl of sterile PBS were s.c. injected into the flank of C57BL/6 mice. On day 7, when tumors reached 40–70 mm³ for B16F10 and 50–100 mm³ for MC38, mice were i.t. injected with PBS, purified protein or engineered bacteria in 20 μl of sterile PBS twice weekly for a total of three injections in MC38 and five injections in B16F10. In another cohort comparing engineered bacteria and anti-PD-1, 0.5 million MC38 cells in 30 μl of sterile PBS were s.c. injected into the flank of C57BL/6 mice. On day 7, mice with MC38 (50–100 mm³) were injected with sterile PBS (i.t., 20 μl), engineered bacteria (i.t. in 20 μl of sterile PBS), anti-PD-1 (i.p. in 100 μl of sterile PBS) or a combination of engineered bacteria (i.t. in 20 μl of sterile PBS) and anti-PD-1 (i.p. in 100 μl of sterile PBS) twice weekly for a total of three injections.

For the abscopal effect experiments (Fig. 2a), 0.5 million MC38 cells and 0.3 million MC38 cells in 30 μl of sterile PBS were s.c. injected into both flanks of C57BL/6 mice (0.5 million on the left flank and 0.3 million on the right flank). On day 7, when tumors on the left flank reached 50–100 mm³, mice were i.t. injected with PBS or engineered bacteria in 20 μl of sterile PBS twice weekly for a total of five injections.

For the systemic delivery experiments (Fig. 2d), 1 million MC38 cells in 30 μl of sterile PBS were s.c. injected into the flanks of C57BL/6 mice. On day 8, when the flank tumors reached 120–200 mm³, mice were i.v. injected with PBS or engineered bacteria in 100 μl of sterile PBS two times weekly for four injections in total. In another cohort comparing systemic efficacy of engineered bacteria and anti-PD-1, 1 million MC38 cells in 30 μl of sterile PBS were s.c. injected into the flanks of C57BL/6 mice. On day 8, mice bearing MC38 (100–200 mm³) were injected with sterile PBS (i.v., 100 μl), engineered bacteria (i.v. in 100 μl of sterile PBS), anti-PD-1 (i.p. in 100 μl of sterile PBS) or a combination of engineered bacteria (i.v. in 100 μl of sterile PBS) and anti-PD-1 (i.p. in 100 μl of sterile PBS) two times weekly for four injections in total.

For the lung metastasis model (Extended Data Fig. 5a), 0.5 million B16F10-luc cells in 100 μl of sterile PBS were i.v. injected in C57BL/6 mice. On day 7 after injection, mice were injected with sterile PBS (i.v., 100 μl), engineered bacteria (i.v. in 100 μl of sterile PBS), anti-PD-1 (i.p. in 100 μl of sterile PBS) or a combination of engineered bacteria (i.v. in 100 μl of sterile PBS) and anti-PD-1 (i.p. in 100 μl of sterile PBS) weekly for a total of four injections. On day 30, mice were euthanized, and lungs were harvested for assessment of tumor nodules.

InVivoMAb anti-mouse PD-1 (clone: RMP1–14) was used in experiments mentioned above.

For the in vivo experiments involving MSLN-CAR NK cells, 5 million H226 cells in 50 μl of sterile PBS were s.c. injected into the flanks of NSG mice. On day 30, when the flank tumors reached 80–200 mm³, 3–5 million MSLN-CAR NK cells (in 100 μl of sterile PBS) generated from three normal healthy donors were i.v. injected (tail vein). PBS, engineered bacteria and purified protein (hDR18) in 20 μl of sterile PBS were i.t. injected every week for a total of three injections. In the separate batch of mice comparing efficacy of chemotherapy and bacteria-NK therapy, 5 million H226 cells in 50 μl of sterile PBS were s.c. injected into the flanks of NSG mice. After 30 d, NSG mice with H226 (80–120 mm³) were injected with sterile PBS (i.t., 100 μl), engineered bacteria (i.t. in 100 μl of sterile PBS) or cisplatin (i.p., Thermo Fisher Scientific, 22–515-0, 5 mg kg⁻¹, in 100 μl of sterile PBS) + pemetrexed (i.p., Selleck Chemicals, S5971, 100 mg kg⁻¹, in 100 μl of sterile PBS) weekly for three injections. Mice treated with bacteria were also injected with 5 million MSLN-CAR NK cells (i.v. in 100 μl of sterile PBS) at day 30. To support human CAR NK cells, the mice received 75 kU of human recombinant IL2 (Miltenyi Biotec, 130–097-748) i.p. every other day. Tumor volume was measured by calipers and calculated by 0.5 × length × width² for 2–3 times per week.

Immunohistochemistry, immunofluorescence, imaging and processing

Tumors at different timepoints after treatment were harvested and incubated in 3.7% paraformaldehyde (PFA) in 1× PBS for 2 h at room temperature. The samples were washed in PBS for 10 min at room temperature three times, after which samples were immersed in PBS with 30% sucrose at 4 °C overnight and then quickly washed with PBS and dried in a paper towel, embedded in OCT in tissue cassettes and frozen at −80 °C. Cryosection was done by iHisto. The frozen slides were washed with PBS for three times and incubated in PBS with 2 mg ml⁻¹ BSA (Sigma-Aldrich, 05470) and 0.5% Triton X-100 (Sigma-Aldrich, T8787 blocking buffer) for 30 min at room temperature. Then, sections were stained by FITC-anti-E. coli pAb (Abcam, ab30522, 200:1) and Alexa Fluor 647 anti-DYKDDDDK (200:1) overnight at 4 °C, followed by three washes of PBS in 0.1% Tween 20 and three washes of PBS. The slides were fixed in PBS with 1% PFA for 5 min and washed twice in PBS, after which they were mounted by EverBrite Mounting Medium with DAPI (Biotium, 23002) for 24 h at room temperature and then stored at 4 °C until requisition. Samples were acquired by a Zeiss LSM 800 confocal microscope. Samples were excited at 495 nm, detected in the range of 510–620 nm for FITC with the power as 0.26% and detector gain as 623 V, excited at 653 nm, detected in the range of 656–700 nm for Alexa Fluor 647 with the power as 1.00% and detector gain as 712 V and excited at 353 nm, detected in the range of 400–510 nm for DAPI with the power as 0.95% and detector gain as 850 V. For acquisition and processing, we used ZEN Blue 2.6 software, and pseudo color was applied.

Staining and FACS analysis

All antibodies were diluted 1:50. The live/dead dyes were diluted 1:1,000–1:300. The information on antibodies and live/dead dye is included in Supplementary Table 1.

Tumor-bearing mice were euthanized, and tumors, lungs and livers were sliced and digested by digestion buffer (RPMI-10 with 5% FBS, 10 mM HEPES, 100 μg ml⁻¹ P/S, 20 μg ml⁻¹ DNase I and 1 mg ml⁻¹ Collagenase IV) for 1 h at 37 °C with agitation, followed by treatment with ammonium–chloride–potassium (Gibco, A1049201) buffer for red blood cell (RBD) lysis and then filtered through a 70-μm strainer to remove debris. Spleen and bone marrow from rear limbs were mechanically dissociated and then filtered through a 70-μm strainer and lysed by ammonium–chloride–potassium buffer. Single-cell suspensions filtered through 70-μm strainers without RBD were washed with PBS three times, blocked by Fc blocker (BD Biosciences, clone 2.4G2) at 4 °C for 10 min, washed with PBS three times, stained with live/dead dyes in PBS for 20 min at room temperature in the dark, washed with FACS buffer (PBS, 5 mM EDTA, 2% FBS, 0.1% NaN₃ sodium azide) three times and then stained with antibody mix for 30 min at room temperature in the dark, washed with PBS three times, fixed with 4% PFA in PBS for 15 min at room temperature in the dark and washed with PBS three times before running on a flow cytometer. Stained samples were analyzed using a Sony ID7000 and FlowJo software. Gating strategies are shown in Supplementary Fig. 20.

Luminex analysis

Blood samples were collected from the submandibular vein at two timepoints (3 d and 7 d) using blood collection tubes with lithium heparin (Greiner Bio-One, 450479). Heparinized plasma was obtained by centrifugation at 4,500g for 15 min at 4 °C and sent to Eve Technology for Luminex analysis for measurement of cytokines and chemokines.

In vivo cell depletions

Antibodies against CD8α (clone 2.43, Bio X Cell, 8 mg kg⁻¹ i.p. twice weekly), NK1.1 (clone PK136, Bio X Cell, 8 mg kg⁻¹ i.p. twice weekly), Ly6g (clone 1A8, Bio X Cell, 8 mg kg⁻¹ i.p. twice weekly), CD19 (clone 1D3, Bio X Cell, 8 mg kg⁻¹ i.p. twice weekly) and B220 (clone RA3.3A1/6.1 (TIB-146), Bio X Cell, 8 mg kg⁻¹ i.p. twice weekly) were used to deplete CD8⁺ T cells, NK cells, granulocytes and B cells, respectively, in vivo. Blood samples were collected from the submandibular vein, harvested in EDTA-coated tubes (BD Biosciences, 02–657-32) and lysed using LCK buffer to remove RBD. The cells were then stained and analyzed by flow cytometer as previously described for verification of cell type removal⁵³.

NK isolation and transduction

NK cells were isolated from healthy donor leukapheresis collars (Crimson Core, T0197) via Ficoll-Paque density gradient using a RosetteSep Human NK Cell Enrichment Kit (STEMCELL Technologies, 15065). Purified NK cells were cultured in NK MACS medium (Miltenyi Biotec) supplemented with 5% human serum and 500 U ml⁻¹ human IL2 (Miltenyi Biotec, 130–114-429) and maintained at 37 °C in a humidified incubator with 5% CO₂.

MSLN-CAR gene construct was designed with the following components: signal peptide (CD8), MSLN scFv derived from the YP218 antibody clone⁵⁴and transmembrane domains (CD8), followed by intracellular signaling domains, 4–1BB and CD3ξ. The CAR construct was linked to eGFP by P2A self-cleaving peptide, and it allowed the assessment of transduction efficiency. Lentiviral supernatants were produced using our baboon lentivirus system as reported previously³⁶. Forty-eight hours after transduction, MSLN-CAR NK cells were stained for violet live/dead stain, and the transduction rates were evaluated by flow cytometry, gating on live, GFP⁺ cells (Supplementary Fig. 21).

In vitro co-culture assays

For HEK-Blue IL-18 cells, 54,000 cells and bacteria with different MOI were added to each well of a 96-well plate in 200 μl of DMEM supplemented with 10% FBS and 100 U ml⁻¹ P/S. After 20-h incubation, 20 μl of media from each well was mixed with 180 μl of Quanti-Blue (InvivoGen, rep-qbs) for 10−20 min at 37 °C, and optical density at 620 nm (OD₆₂₀) was measured by a plate reader.

For NK cell killing assays, 15,000–50,000 effector NK cells (MSLN-CAR NK) labeled with CFSE dye were co-cultured with bacteria with MOI of 1,000 overnight (12–16 h) and then co-cultured with 50,000 target tumor cells (H226, Raji, OVCAR8, K562 or H2591) for 4–6 h. Afterwards, the cells were treated with a detaching buffer (PBS, 2 mM EDTA) and harvested for staining. In brief, cells washed with PBS three times were stained with Zombie NIR Fixable Viability Dye for 15 min at room temperature in the dark. Afterwards, samples were washed three times with the FACS buffer before running on a flow cytometer.

For marker analysis, human PBMCs were treated with the bacteria at MOI of 1,000 for 12 h and then collected for antibody staining. In brief, samples were stained with live/dead dye green and subsequently stained with the antibody cocktails with PBS or FACS washes in between. For the NK cell staining, 50,000 primary human NK cells were treated by bacteria with MOI of 1,000 for 12 h and co-cultured with 50,000 target H226 for 6 h. GolgiPlug and GolgiStop (Brefeldin A and Monensin, 554715 and 555028, BD Biosciences) were added 1 h after the starting point of co-culture. After 6-h co-culture, cells were collected and stained with live/dead dye violet and then stained with antibodies against surface markers before fixation with three washes of PBS or FACS buffer in between. Afterwards, the cells were fixed with BD Cytofix/Cytoperm Buffer (BD Biosciences, 554714) for 30 min on ice and stained with intracellular antibodies in BD Perm/Wash Buffer (BD Biosciences, 554723) for 30 min on ice with three washes of BD Perm/Wash Buffer or FACS buffer in between. Cells were acquired using a Sony ID7000.

RNA-seq and analysis

In total, 1–3 million MSLN-CAR NK cells from four different donors (two donors per biological replicate) were co-cultured with bacteria YiaT232 (MOI of 1,000), bacteria YiaT232–hDR18 (MOI of 1,000), purified hDR18 (100 ng ml⁻¹) or PBS for 3 h, after which cells were harvested for RNA extraction and RNA-seq. The FASTQ files were processed using STAR alignment⁵⁵ for the first run and HISAT2 (ref. 56) for the second run, followed by DESeq2 (ref. 57) analysis. P values and adjusted P values were determined by DESeq2, and pathway analysis was performed using GSEA^58,59 for pre-ranked lists (based on log₂ fold changes (FCs)) and MSigDB^58,59 focusing on genes with P_adj ≤ 0.05 in the comparison of NK cells cultured with YiaT232–hDR18 E. coli versus PBS, with P < 0.01 and log₂FC > 0.5 (or log₂FC < −0.5) in the comparison of NK cells cultured with YiaT232–hDR18 E. coli versus YiaT232. MATLAB 2013b was used for analysis.

Statistics and reproducibility

Data were analyzed by Student’s t-test (two-sided), one-way ANOVA, two-way ANOVA, Kaplan–Meier methods or log-rank test by GraphPad Prism version 10.1.1. No statistical methods were applied to determine sample size. Mice were randomized into different groups after tumor inoculation. Mice with signs of sickness (hunched back, severe ulceration and more than 15% body weight loss) were immediately euthanized and excluded from the analysis. The investigators were not blinded to allocation during experiments and outcome assessments. RNA-seq data were evaluated using P values and adjusted P values produced by DESeq2.

Reporting summary

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

Extended Data

Extended Data Fig. 1 |. Treatment with the engineered bacteria induces immunological memory leading to enhanced recall responses (MC38).

a, b, C57BL/6 mice (n = 5) cured from MC38 upon treatment with OmpA-mDR18 and naïve C57BL/6 (gender and age-matched with the cured mice) were subcutaneously (s.c.) engrafted with 0.5 × 10⁶ MC38 cells and then monitored for tumor growth and survival. Mean tumor growth (a) and Kaplan–Meier survival curves (b) for mice injected with 0.5 × 10⁶ MC38 cells on Day 0. Two-way ANOVA test for tumor growth curve (a) and Mantel-Cox test for survival curve (b). Data are representative of one independent experiment, with n = 5 mice per group (a, b). Data represent means ± SD (a).

Extended Data Fig. 2 |. Treatment with the engineered bacteria induces immunological memory leading to enhanced recall responses (B16F10).

a, b, C57BL/6 mice (n = 3) cured from B16F10 upon treatment with OmpA-mDR18 and naïve C57BL/6 (gender and age-matched with the cured mice) were subcutaneously (s.c.) engrafted with 0.5 × 10⁶ B16F10 cells and then monitored for tumor growth and survival. Mean tumor growth (a) and Kaplan–Meier survival curves (b) for mice injected with 0.5 × 10⁶ B16F10 cells on Day 0. Two-way ANOVA test for tumor growth curve (a) and Mantel-Cox test for survival curve (b). Data are representative of one independent experiment, with n = 3 mice per group (a, b). Data represent means ± SD (a).

Extended Data Fig. 3 |. Co-administration of OmpA-mDR18 and anti-PD-1 increases anti-tumor responses in an immune-competent syngeneic mouse model (MC38).

a, C57BL/6 mice were subcutaneously (s.c.) engrafted with 0.5 × 106 MC38 cells, starting on day 7 (tumor size reaches 50 – 100 mm3), mice were treated with PBS (i.t., n = 5), OmpA-mDR18 (i.t., 0.25 ×109 CFU, n = 5), anti-mouse PD-1 (anti-PD-1; i.p., 8 mg/kg, n = 10) or OmpA-mDR18 (i.t., 0.25 ×109 CFU) + anti-PD-1 (i.p., 8 mg/kg, n = 10) three times (days 7, 10, 14). b, c, Mean tumor growth (b) and Kaplan-Meier survival curves (c) for mice bearing MC38 tumors after treatment. Two-way ANOVA test for tumor growth curve (b) and Mantel-Cox test for survival curve (c). Data are representative of two independent experiments. Data represent means ± SD (b).

Extended Data Fig. 4 |. Systemic co-administration of OmpA-mDR18 and anti-PD-1 increases anti-tumor efficacy in immune-competent mice bearing MC38.

a, C57BL/6 mice were subcutaneously (s.c.) engrafted with 10⁶ MC38 cells, starting on day 8 (tumor size reaches 100 – 200 mm³), mice were treated with PBS (i.v., n = 5), OmpA-mDR18 (i.v., 10⁹ CFU, n = 5), anti-PD-1 (i.p., 8 mg/kg, n = 8) or OmpA-mDR18 (i.v., 10⁹ CFU) + anti-PD-1 (i.p., 8 mg/kg, n = 10) four times (days 8, 11, 15, 18). b, c, Mean tumor growth (b) and Kaplan-Meier survival curves (c) for mice bearing MC38 tumors after treatment. Two-way ANOVA test for the tumor growth curves (b) and Mantel-Cox test for the survival curves (c). Data are representative of two independent experiments. Data represent means ± SD (b).

Extended Data Fig. 5 |. Efficacy of OmpA-mDR18 and anti-PD-1 in the lung metastasis model.

a, 0.5 × 106 B16F10-luciferase-GFP cells were injected into each C57BL/6 mouse by tail vein injection. Starting on day 7, mice were treated with PBS (i.v., n = 5), OmpA-mDR18 (i.v., 109 CFU, n = 8), anti-PD-1 (i.p., 8 mg/kg, n = 6) or OmpA-mDR18 (i.v., 109 CFU) + anti-PD-1 (i.p., 8 mg/kg, n = 8) weekly for four times (on days 7, 14, 21 and 28) in total. Mice were sacrificed on day 30 and lungs were harvested for further analysis. b, Number of tumor nodules on lung surface. c, pictures of lungs from different treatment groups. Two-sided unpaired t-test for tumor nodes (b). Data were combined from two independent experiments. Data represent means ± SD (b).

Extended Data Fig. 6 |. OmpA-mDR18 rapidly promotes tumor-infiltrating NK and CD8 + T cells in MC38 model.

a, C57BL/6 mice were subcutaneously (s.c.) engrafted with 0.5 × 10⁶ MC38 cells. When tumor size reached 150 – 200 mm³ on day 10, mice were treated with PBS (n = 5), OmpA-mDR18 (0.5 × 10⁹ CFU, n = 5), OmpA (0.5 × 10⁹ CFU, n = 5) or mDR18 (4 mg/kg, n = 5) intratumorally (i.t.). Tumor tissues were harvested for further analysis for tumor-infiltrating immune cells (CD8⁺ T cells, CD4⁺ T cells, NK cells, NKT cells, macrophages, monocytes, granulocytes, and mononuclear myeloid-derived suppressor cells, M-MDSCs) 1 day after treatment by flow cytometer. b, The percentage of key cell types (as a proportion of live cells): CD8⁺ T cells (CD45⁺CD3⁺CD8⁺), CD4⁺ T cells (CD45⁺CD3⁺CD4⁺), NK cells (CD45⁺CD3⁻NK1.1⁺), T like NK cells (NKT, CD45⁺CD3⁺NK1.1⁺), granulocytes (CD45⁺CD11b⁺Ly6C⁺ Ly6G⁺), monocytes (CD4 5⁺CD11b⁺Ly6C^lowLy6G⁻), macrophages (CD45⁺CD11b⁺F4/80⁺) and M-MDSCs (CD 45⁺CD11b⁺Ly6C^highLy6G⁻) in tumor microenvironment were shown in the bar graph. c, Summary of the data showing the percentage of CD8⁺ T cells, NK cells, and granulocytes as a proportion of CD45⁺ cells blood cells in mice bearing MC38 on day 1 after being treated with bacteria (i.t.) or bacteria (i.t.) plus monoclonal antibody (i.p.). Two-sided unpaired t-test (b, c). Data represent means ± SD.

Extended Data Fig. 7 |. Tumor control by OmpA-mDR18 in B16F10 is mediated by CD8 + T cells and Natural Killer (NK) cells in the tumor microenvironment (TME).

a, C57BL/6 mice were subcutaneously (s.c.) engrafted with 0.5 × 10⁶ B16F10 cells. On day 6, mice were treated with monoclonal antibodies anti-CD8α, anti-NK1.1, anti-ly6G, and PBS intraperitoneally (i.p.). When tumor size reached 40 – 70 mm³ on day 7, mice were treated with OmpA-DR18 (0.5 × 10⁹ CFU) or PBS intratumorally (i.t.) and monoclonal antibody (mAb) intraperitoneally (i.p.) on day 7, day 10, day 14, day 17 and day 21. b, c, Mean tumor growth (b) and Kaplan–Meier survival curves (c) for mice bearing B16F10 treated with monoclonal antibodies or PBS, and bacteria or PBS. Two-way ANOVA test for tumor growth curve (b) and Mantel-Cox test for survival curve (c). Data are representative of one independent experiment, with n = 5 mice per group. Data represent means ± SD (b).

Extended Data Fig. 8 |. Engineered bacteria disseminate to key organs in the immune compromised NSG mice and are associated with increased proliferation of the adoptively transferred CAR-NK cells.

a, NSG mice were subcutaneously (s.c.) engrafted with 5 × 10⁶ H226 cells, on day 40, mice (n = 5 each group) were treated with PBS, YiaT232 (10⁹ CFU), YiaT232-hDR18 (10⁹ CFU) and purified hDR18 (4 mg/kg) (i.t.). 5 million MSLN-CAR NK were administrated intravenously (i.v.). All mice were injected with 75kU human recombinant IL2 every other day intraperitoneally (i.p.). On Day 47, mice were sacrificed and organs (tumor, liver, lung, spleen, and bone marrow) of all mice were collected for analysis by flow cytometry and CFU. b, The concentration of bacteria (CFU/g) in tumors, livers, spleens, lungs, and bone marrow post-treatment. c, Percentage of Ki67⁺ cells gated on human CD45⁺ cells in livers, tumors, spleens, lungs, and bone marrow. Two-sided unpaired t-test for percentage of Ki67⁺ cells (c). Data represent means ± SD (b, c).

Extended Data Fig. 9 |. Gene expression changes in the MSLN-CAR NK cells induced by priming with E. coli-hDR18.

a, Log2 fold changes for differentially expressed genes (YiaT232-hDR18, E. coli vs PBS, adjusted p-value < 0.05, n = 2923). Fold changes were calculated for YiaT232-E. coli, YiaT232-hDR18, E. coli and hDR18 compared with PBS (paired). Data is row normalized. P-values were calculated using deseq2. b, GSEA enrichment FDR (false discovery rates) values for the comparisons YiaT232-hDR18, E. coli vs PBS, YiaT232-E. coli and hDR18. The left panel represents down-regulated gene sets, and the right panel represents upregulated gene sets. The dashed line depicts 5% FDR. Highlighted in red are gene sets that passed 5% FDR in all comparisons. P-values were calculated using GSEA.

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1038/s41587-024-02418-6.

Data availability

All data generated during this study are available within the paper. The bulk RNA-seq data were deposited to the Gene Expression Omnibus under accession number GSE275391 (ref. 60). Datasets from the MSigDB were used in this study. Source data are provided with this paper.