CAR T cells expressing a bacterial virulence factor trigger potent bystander antitumour responses in

Chimeric antigen receptor T cells (CAR T cells) are effective against haematologic malignancies. However, in solid tumours, their potency is hampered by local immunosuppression and by the heterogeneous expression of the antigen that the CAR targets. Here we show that CAR T cells expressing a pluripotent pro-inflammatory neutrophil-activating protein (NAP) from Helicobacter pyloritrigger endogenous bystander T-cell responses against solid cancers. In mice with subcutaneous murine pancreatic ductal adenocarcinomas, neuroblastomas or colon carcinomas, CAR(NAP) T cells led to slower tumour growth and higher survival rates than conventional mouse CAR T cells, regardless of target antigen, tumour type and host haplotype. In tumours with heterogeneous antigen expression, NAP secretion induced the formation of an immunologically ‘hot’ microenvironment that supported dendritic cell maturation and bystander responses, as indicated by epitope spreading and infiltration of cytotoxic CD8⁺ T cells targeting tumour-associated antigens other than the CAR-targeted antigen. CAR T cells armed with NAP neither increased off-tumour toxicity nor hampered the efficacy of CAR T cells, and hence may have advantageous translational potential.

Therapies with adoptive transfer of autologous T cells engineered ex vivo with a chimeric antigen receptor (CAR) directed against CD19 are efficacious and approved for refractory and relapse B-cell leukaemia and lymphoma, while B-cell maturation antigen-directed CAR T-cell treatment is approved for multiple myeloma^1,2,3,4. In solid tumours, a particular tumour-associated antigen that CAR T cells could target is often heterogeneously expressed, which impairs the efficiency of CAR T-cell therapy⁵. Furthermore, the immunosuppressive microenvironment in solid tumours obstructs CAR T-cell efficacy^6,7. Therefore, combating CAR target antigen heterogeneity and reducing immunosuppression is of utmost importance in improving CAR T-cell therapy of solid tumours.

Various approaches to reinforce the efficacy of CAR T cells against solid tumours have been evaluated, including engineering CAR T cells to produce endogenous enzymes, cytokines, chemokines that potentiate either CAR T cells or other host immune cells, or both^{8,9,10,11,12,13,14,15}. Many exogenous molecules of bacterial origin have strong immunomodulatory properties. One such protein is the neutrophil-activating protein (NAP) from Helicobacter pylori, which attracts innate immune cells, induces maturation and T helper type-1 (Th1) polarization of dendritic cells (DCs), and creates a local pro-inflammatory milieu with enhanced interleukin (IL)-12 production^16,17,18. We investigate and report that arming of CAR T cells with NAP can improve their activity in various solid tumour models.

We use retroviral vectors (RV) (Fig. 1a) to engineer murine T cells. After validating inducible NAP expression (Supplementary Fig. 1), we evaluated the efficacy of CAR(NAP) T cells to target endogenously expressed murine CD19 and disialoganglioside (GD2) in two immunocompetent, syngeneic mouse models of cancer, A20 (lymphoma) and NXS2 (neuroblastoma) (Fig. 1b,c). Conventional CAR T cells and CAR(NAP) T cells exhibited similar cytotoxicity against both cell lines in vitro (Extended Data Figs. 1a,b and 2a–c), confirming that insertion of the NAP transgene did not compromise the cytotoxic capacity of CAR T cells. When evaluated in vivo, CAR(NAP) T cells controlled tumour growth and significantly prolonged the survival of tumour-bearing mice in both models (Fig. 1d–g, and Extended Data Figs. 1c–e and 2d–f). Approximately 75% of the mice became tumour-free when treated with CAR(NAP) T cells, while conventional CAR T-cell treatment only cured 30% of the mice with A20 lymphoma (Fig. 1d,e). Mice cured from A20 tumours were protected against rechallenge with the same tumour cell line (Fig. 1e), indicating the establishment of immunological memory. Decreased B-cell counts were observed in the blood of mice after treatment with CD19 targeting CAR(NAP) T cells but not in mice treated with conventional CAR T cells (Extended Data Fig. 1f), which is also a sign of potency of CAR(NAP) T cells. Of note, this enhanced efficacy of CAR(NAP) T cells was neither dependent on the nature of the target antigen, nor the tumour types or host haplotypes, as superior therapeutic efficacy was also observed for CAR(NAP) T cells in various tumour models, including NXS2-mCD19 neuroblastoma (Extended Data Fig. 3), Panc02-mCD19 pancreatic adenocarcinoma (Extended Data Fig. 4) and CT26-hPSCA colon carcinoma (Extended Data Fig. 5). These data confirm that CAR(NAP) T cells perform better than CAR T cells in vivo despite exhibiting similar potency in vitro.

a, Retroviral constructs used for mouse T-cell engineering. b,c, Treatment schedules for CAR-engineered T cells of subcutaneous murine A20 lymphoma that endogenously express murine CD19 (b), and murine NXS2 neuroblastoma that endogenously express GD2 (c). d, e, A20 tumour size (mean) until the first mouse had to be killed (d) and mouse survival (Kaplan–Meier curve) after treatment (e); arrows in e indicate rechallenge of cured mice with A20 tumour cells. f, g, NXS2 tumour size (mean) until the first mouse had to be killed (f) and mouse survival (Kaplan–Meier curve) after treatment (g). The A20 experiment was performed once, and the NXS2 experiment was performed twice and all data points were pooled. Tumour sizes between treatment groups were compared using two-way ANOVA, and survival curves were compared using the log-rank test. Error bars represent s.e.m. (NS, no statistical significance; *P < 0.05, ***P < 0.001, ****P < 0.0001). Precise P values are reported in Supplementary Table 3.

Source data

Full size image

We also investigated whether the potential immunogenicity of NAP can counteract the performance of CAR(NAP) T cells. Encouragingly, CAR(NAP) T-cell persistence was not affected, as CAR T cells were detected in the blood of approximately half of the tumour-free mice in both treatment groups (Supplementary Fig. 2). Importantly, the therapeutic efficacy of CAR(NAP) T cells was not diminished in mice with pre-existing anti-NAP antibodies (Supplementary Fig. 3). Additionally, repeated treatment with CAR(NAP) T cells did not result in elevated toxicity compared with conventional CAR T cells when assessing systemic cytokine release (Supplementary Fig. 4) or body weight (Supplementary Fig. 5). In conclusion, arming CAR T cells with NAP therefore constitutes a viable strategy for targeting various solid tumours.

Antigen heterogeneity is a major hurdle for CAR T-cell therapy of solid tumour¹⁹. To evaluate whether CAR(NAP) T cells can overcome antigen heterogeneity, we established a heterogeneous tumour model by subcutaneously injecting a mixture (1:1 ratio) of wild-type (antigen-negative) NXS2 cells and antigen-positive NXS2-mCD19 cells (Fig. 2a). CAR(NAP) T-cell treatment significantly reduced the growth of tumours with heterogeneous CAR-target antigen expression and prolonged mouse survival, while conventional CAR T-cell treatment failed to yield any therapeutic benefit compared with mock T-cell treatment (Fig. 2b,c and Supplementary Fig. 6). Furthermore, mice cured by CAR(NAP) T-cell treatment were protected against rechallenge with wild-type NXS2 cells, indicating establishment of protective bystander immunological memory (Fig. 2c). Tumour-infiltrating CD8⁺ T cells isolated from CAR(NAP) T-cell-treated mice exhibited an antigen-experienced memory-like (CD62L⁺, CD44⁺) phenotype and were degranulated (CD107a⁺) to a higher degree than those isolated from CAR T-cell-treated mice (Fig. 2d–f). Importantly, splenocytes isolated from CAR(NAP) T-cell-treated mice, but not those from CAR T-cell-treated mice, responded to wild-type (CAR-targeted antigen-negative) NXS2 tumour cell lysate (Fig. 2g and Supplementary Fig. 7), indicating systemic activation of an endogenous bystander T-cell response, probably explaining why cured mice were protected when rechallenged with wild-type NXS2 (Fig. 2c).

a, Treatment schedule of the mixed tumour model with 50% of tumour cells expressing the mCD19 CAR target antigen. b,c, Tumour size (mean) until the first mouse had to be killed (b) and mouse survival (Kaplan–Meier curve) after treatment and rechallenge (arrow) of cured mice with wild-type NXS2 tumour cells (c). The experiment was performed twice and all data were pooled. Tumour sizes were compared using two-way ANOVA, and survival curves were compared using log-rank test. d–f, Characteristics of tumour-infiltrating CD8⁺ T cells on day 18. Representative plots (d) and the percentage of memory-like (CD44⁺CD62L⁺) CD8⁺ T cells (e), and CD107a expression (f) on CD8⁺ T cells in treatment groups. Groups were compared using t-test. MFI, mean fluorescence intensity. g, Analysis of epitope spread as a consequence of treatment, with IFN-γ levels from co-cultures of splenocytes from treated mice and autologous DCs loaded with wild-type NXS2 (CD19^–) tumour cell lysate. Values from co-cultures with unloaded DCs were seen as background and subtracted. Groups were compared using ANOVA with Bonferroni correction. h,i, Epitope spreading towards a non-targeted model antigen (OVA) for tumour-infiltrating CD8⁺ T cells after mCD19 CAR T-cell treatment of Panc02-mCD19-OVA tumours in C57BL/6 mice. Representative density plots (h) showing OVA(SIINFEKL)-specific CD8⁺ T cells and percentage (i) of these T cells in tumours from each group. Groups were compared using ANOVA with Bonferroni correction. j, IFN-γ production by splenocytes after mCD19 CAR T-cell treatment of NXS2-mCD19-OVA tumours in A/J mice upon rechallenge with different OVA peptides. In all panels, error bars represent s.e.m. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). Precise P values are reported in Supplementary Table 3.

Source data

Full size image

A key feature of the bystander T-cell response is epitope spreading, that is, activation of endogenous CD8⁺ T cells directed against antigens other than the antigen targeted by the CAR, which is vital for eradication of antigen-heterogeneous solid tumours¹⁹. We investigated epitope spreading using two tumour models: Panc02-mCD19-OVA and NXS2-mCD19-OVA, wherein ovalbumin (OVA) was introduced as a bystander antigen not targeted by the CAR T cells. Tumour infiltration of OVA-specific (SIINFEKL-directed) CD8⁺ T cells were significantly increased in mice treated with CAR(NAP) T cells compared with CAR T cells (Fig. 2h,i). Furthermore, treatment with CAR(NAP) T cells induced a broad bystander CD8⁺ T-cell response to all predicted H-2D^d-restricted OVA-derived peptides, with higher interferon (IFN)-γ response compared with conventional CAR T-cell treatment (Fig. 2j), indicating epitope spreading resulting in both a stronger and broader bystander immune response. We identified six subpopulations of tumour-infiltrating CD8⁺ T cells (Pop1–6, Extended Data Fig. 6a) on the basis of surface marker expression (Extended Data Fig. 6b,d). Memory-like (CD44⁺, CD62L⁺, CD127⁺), proliferating (Ki67⁺), activated and cytolytic (CD69⁺, CD107a⁺) T cells (Pop1) were significantly enriched after CAR(NAP) T-cell treatment (Extended Data Fig. 6c). In contrast, exhausted (PD1^hiTim3^hiLAG3^hi) cytolytic T cells (Pop2) were enriched after conventional CAR T-cell treatment (Extended Data Fig. 6c,e). These observations support the notion that CAR(NAP) T-cell treatment induces epitope spreading, with an endogenous bystander response of highly activated CD8⁺ T cells directed against various epitopes of tumour-associated antigens.

To study how CAR(NAP) T-cell treatment affects the tumour microenvironment, we investigated gene expression in tumour-infiltrating immune cells (CD45⁺) (Fig. 3a) using the NXS2-mCD19 model. Compared with mock T-cell treatment, both CAR(NAP) T cells and CAR T cells exhibited unique gene expression patterns (Fig. 3b,c). Gene set enrichment analysis for genes upregulated more than 2-fold in the CAR(NAP) T-cell treatment group compared with the CAR T-cell treatment group, revealed enrichment of Gene Ontology (GO) terms involved in leucocyte migration, proliferation, inflammatory response and regulation of response to an external stimulus (Fig. 3d,e). Cell type profiling from the gene expression analysis²⁰ indicated enhanced infiltration of neutrophils (Fig. 3f), macrophages (Fig. 3g) and cytotoxic natural killer (NK) cells (Fig. 3h) after CAR(NAP) T-cell treatment. Infiltration of neutrophils and M1 macrophages were also validated by flow cytometry (Fig. 3i,j). Flow cytometry analysis also revealed an increased infiltration of antigen-presenting DCs (CD103⁺CD11c⁺) after CAR(NAP) T-cell treatment (Fig. 3k). We also observed a similar immune cell infiltration profile in the A20 tumour model, where CAR(NAP) T-cell treatment led to higher infiltration of CD8⁺ T cells and Gr1⁺MPO⁺ neutrophils, and lower infiltration of FoxP3⁺CD4⁺ regulatory T cells (Extended Data Fig. 7). Elevated infiltration of activated immune cells, in turn, resulted in higher levels of Th1-related cytokines and chemokines (Fig. 3l), and higher infiltration of IFN-γ (Fig. 3m) and IL-2 expressing (Fig. 3n) T cells in the tumour tissues after CAR(NAP) T-cell treatment compared with CAR T-cell treatment. Of note, CAR-positive T-cell infiltration in the tumours was detected in a higher proportion of mice after CAR(NAP) T-cell treatment compared with CAR T-cell treatment (Fig. 3o), consistent with an increased proliferative capacity and less exhausted phenotype. Collectively, these observations suggest that treatment with CAR(NAP) T cells induce a more pronounced immunologically ‘hot’, Th1-type pro-inflammatory tumour microenvironment compared with conventional CAR T cells, which explains the superior therapeutic efficacy of CAR(NAP) T cells.

a–h, CD45⁺ cells were isolated from NXS2-mCD19 tumours after CAR T-cell treatment and gene expression was analysed using NanoString. a, Schematic illustration of the experimental set-up. b, Heat map of gene expression presented as normalized z-score of raw gene counts. c, Upregulated genes (unique and overlapping) after treatment with CAR(NAP) T cells (red) and CAR T cells (blue) compared with mock T-cell treatment. d, Gene Ontology (GO) term enrichment network of uniquely upregulated genes after treatment with CAR(NAP) T cells versus CAR T cells, generated by Metascape analysis. Values in parentheses, –log₁₀(P). e, Heat map of gene expression presented as normalized z-score of raw gene counts and their GO term pathways classification, colour-coded as in d. f–h, Abundance of tumour-infiltrating neutrophils (f), macrophages (g) and cytotoxic NK cells (h), determined from NanoString data. a.u., arbitrary units. i–k, Flow cytometry analysis of tumour-infiltrating immune cells in the NXS2-mCD19 tumour model. Gating strategies (left panels) and percentage (right panels) of tumour-infiltrating Ly6C⁺Ly6G⁺ neutrophils (i), Ly6C^lowMHC-II⁺ M1 tumour-associated macrophages (j) and CD103⁺CD11C⁺ antigen-presenting DCs (k) are shown. l, Cytokine and chemokine levels in tumour lysate (X, below detection level). m, Percentage of IFN-γ producing cells in the CD3⁺ tumour-infiltrating T-cell population. n, Percentage of IL-2 producing cells in the CD3⁺ tumour-infiltrating T-cell population. o, Proportion of mice with qPCR-detectable tumour-infiltrating CAR-engineered T cells (n = 8). Groups were compared using t-test (*P < 0.05, **P < 0.01). In all panels, error bars represent s.e.m. Precise P values are reported in Supplementary Table 3.

Full size image

Finally, to test the applicability of human CAR(NAP) T cells, we performed an in vitro proof-of-concept validation in freshly isolated human peripheral blood mononuclear cells (PBMCs) (Fig. 4a). NAP was expressed by human CAR(NAP) T cells upon target-antigen recognition (Fig. 4b,c), and human CAR(NAP) T cells and conventional CAR T cells killed human cancer cells equally well in vitro (Fig. 4d), even though transduction efficacy was higher for CAR T cells than for CAR(NAP) T cells (Supplementary Fig. 8a). Phenotypic characterization of CAR(NAP) T cells and CAR T cells upon transduction revealed similar CD4 and CD8 (Supplementary Fig. 8b), memory and effector T-cell composition (Supplementary Fig. 8c), and activation/exhaustion phenotype (Supplementary Fig. 8d). However, upon target cell recognition in vitro, gene expression analysis (Fig. 4e) revealed a distinct gene expression pattern for CAR(NAP) T cells compared with conventional CAR T cells, with GO terms enriched for adaptive immune response, cellular response to IFN-γ, regulation of cytokine production, type-I IFN signalling, NK cell-mediated immunity and granulocyte migration (Fig. 4f,g). In addition, the supernatants from co-cultures of CAR(NAP) T cells and tumour cells expressing the antigen targeted by the CAR were rich in potent chemo-attractants that induced migration and activation of DCs (Fig. 4h and Extended Data Fig. 8), neutrophils (Extended Data Fig. 9) and monocytes (Extended Data Fig. 10), confirming the findings of the gene set enrichment analysis (Fig. 4f,g).

a, Self-inactivating (ΔU3) lentiviral constructs used for human T-cell engineering. b,c, Representative histogram (b) and MFI (c) of NAP expression in CAR-positive T cells upon target cell recognition. Dashed line, MFI of isotype staining. d, Relative viability of target cells after exposure to CAR-engineered T cells. e–g, Gene expression in CAR T cells and CAR(NAP) T cells after exposure to target cells. e, Schamatic illustration of the experimental setting. f, GO term enrichment network of the uniquely upregulated genes in CAR(NAP) T cells according to Metascape analysis of the NanoString gene expression data. Numbers in brackets are –log₁₀(P) values. g, Heat map of genes uniquely upregulated in CAR(NAP) T cells presented as normalized z-score of raw gene counts. h, Percentage of DCs migrating towards supernatant collected from a 5:1 co-culture of engineered T cells and Daudi target cells. i–k, Potency of CAR(NAP) T cells assisted by autologous DCs, presented as relative viability of target cells (i), IFN-γ secretion (j) and CD107a expression (k). l–n, CAR-engineered human T cells were co-cultured with autologous immature DCs and Daudi target cells (5:1:1) and assayed for T-cell polarization (l), and production of IL-12 (m) and other cytokines and chemokines (n). All experiments were performed three times and data were pooled. Error bars represent s.e.m. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). Precise P values are reported in Supplementary Table 3.

Full size image

In the presence of autologous DCs, CAR(NAP) T cells exhibited higher cytotoxicity than conventional CAR T cells (Fig. 4i), with enhanced IFN-γ secretion (Fig. 4j) and degranulation (Fig. 4k). Furthermore, gene expression analysis revealed that genes representing pathways involved in cytokine-mediated signalling were upregulated (Supplementary Fig. 9), and a Th1 transcriptional profile was apparent (Fig. 4l). The levels of IL-12 (Fig. 4m) and those of various pro-inflammatory cytokines and chemokines (Fig. 4n) were increased in supernatants from the co-culture of CAR(NAP) T cells, DCs and tumour cells, verifying the observed gene expression profile. Taken together, these observations indicate that the cytotoxic and immunostimulatory capacity of human CAR(NAP) T cells are superior to those of conventional human CAR T cells, which is consistent with our findings in the various mouse models.

The above observations align with published data that NAP, provided as a recombinant protein²¹ or delivered as a therapeutic transgene^16,17, promotes DC maturation and Th1 polarization, and recruits and activates monocytes and neutrophils¹⁸. Accordingly, we propose that the enhanced therapeutic efficacy of CAR(NAP) T cells is associated with the ability of secreted NAP to attract and activate innate immune cells to kill tumour cells, and to activate DCs to facilitate the induction of epitope spreading, leading to CAR-target-independent CD8⁺ T-cell bystander immunity (Fig. 5).

CAR(NAP) T cells are CAR T cells armed with NAP, an immune activator derived from H. pylori. Within a solid tumour, heterogeneity in antigen expression makes it difficult to eradicate all tumour cells with conventional CAR T cells. CAR(NAP) T cells secrete NAP during killing of tumour cells (Steps 1 and 2). NAP works as a chemoattractant, recruiting innate immune cells, such as neutrophils and NK cells, and activates them to directly kill tumour cells independent of whether they express the CAR target or not (Steps 3 and 4). NAP also recruits monocytes and DCs, and activates them to secrete pro-inflammatory cytokines that directly counteract the hostile tumour microenvironment and strengthen the function of CAR(NAP) T cells. Immature DCs can phagocytose cell debris from dying tumour cells (Step 5). During this process, the DCs mature and migrate to a draining lymph node (Step 6) where they degrade the tumour cell debris and present peptides, including tumour-associated antigens, on the human leukocyte antigen (HLA) molecules to naive T cells. T cells with matching T-cell receptors (TCRs) recognize tumour-associated antigen epitopes and become activated (Step 7). The newly activated T cells recirculate back to the tumour site and kill the tumour cells through TCR-mediated recognition, which is independent of the antigen targeted by the CAR (Step 8). Therefore, this technology can counteract heterogeneity in antigen expression within solid tumours.

Full size image

CAR T cells have been armed with several host-derived endogenous molecules (IL-12¹², IL-18^8,9,10, Flt3L¹¹, IL-7/CCL19¹³, Heparinase¹⁴ and CD40L¹⁵) that engage host immune cells to counteract antigen heterogeneity in solid tumours. These factors primarily work by engaging antigen-presenting cells^{8,9,10,11,12,13}. Here we demonstrate that expression of a bacteria-derived virulence factor by CAR T cells can mediate bystander immunity with epitope spreading, leading to eradicating antigen-heterogeneous tumours. Our results suggest that NAP engages a broader range of immune effectors, for example, DCs, monocytes, NK cells and neutrophils to potentiate stronger antitumour immunity. NAP has been shown to substantially enhance the immunogenicity of poor immunogens²². This can be favourable as our data suggest that the immunogenicity of NAP can reverse the suppressive immune landscape in solid tumours. On the other hand, NAP immunogenicity can also be unfavourable, as immune response against NAP may hamper the persistence of the CAR(NAP) T cells, similar to what has been observed for T cells expressing other foreign proteins²³, or cause adverse effects such as anaphylaxis²⁴. However, we show that persistence and efficacy of CAR(NAP) T cells were not altered in mice pre-immunized to have anti-NAP antibodies. It is even possible that pre-existing anti-NAP immunity can neutralize potential NAP-related systemic toxicity. However, the actual situation needs to be evaluated in a careful dose-escalation clinical study with CAR(NAP) T cells.

We used an inducible nuclear factor of activated T-cells (NFAT) promoter²⁵ to control NAP expression as proof of concept. The same promoter was used in a γ-retroviral vector to control IL-12 expression in a clinical trial evaluating IL-12-armed tumour-infiltrating lymphocytes²⁶. Toxicities observed in patients in that trial were mainly attributed to basal level of systemic IL-12 expression²⁶. Therefore, other tightly regulated promoters may be worth further investigation^27,28. Whether NAP expressed from CAR(NAP) T cells would cause toxicity in a similar fashion as IL-12 in cancer patients is unknown. However, analysis of systemic cytokine levels in blood and body weight of mice during treatment did not show any CAR(NAP) T-cell-treatment-related toxicity. Furthermore, to the best of our knowledge, there are no reports describing NAP-related toxicity in patients with H. pylori-induced gastric inflammation, which further supports a safe profile of using NAP as a therapeutic transgene in CAR T cells.

Arming CAR T cells with a potent bacterial virulence factor is a unique concept in CAR T-cell design when compared with using host-derived factors ^{8,9,10,11,12,13,14}. There may be other pathogen-derived factors that are as effective, therefore our study opens up possibilities for evaluating intricate multifaceted pathogen-derived immune activators as ammunition for CAR T cells. We demonstrate that arming CAR T cell with NAP improves the treatment of solid tumours with heterogenous antigen expression. This approach may be used to engineer CAR T cells regardless of the CAR target antigen and tumour types, thus expanding translatability to various human cancers.