Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal solid tumour with a 5-year survival rate of less than 10%.1 2 Conventional chemotherapy fails to significantly improve life expectancy and immune checkpoint blockade therapy is not effective against PDAC.3 Anti-claudin-18 isoform 2 (CLDN18.2)/anti-cluster of differentiation 3 (CD3) immunoglobulin G (IgG) format bispecific T-cell engagers (CLDN18.2/CD3 BiTEs), which redirect T cells to recognise and effectively eliminate tumour cells, are increasingly used for cancer treatment.4 CLDN18.2/CD3 BiTEs have shown encouraging preliminary results for treatment of PDAC, but additional strategies are needed to achieve durable responses.

Various constructs of BiTEs targeting CD3 are broadly classified into two categories based on the presence/absence of Fc domains, which contribute to maintain stability, simplify the purification process and extended the half-life of bispecific Abs (BsAbs).5 However, interactions of the Fc domains and related receptors (FcγRs) of immune effector cells, such as natural killer (NK) cells, monocytes and macrophages, induce Ab-dependent cell-mediated cytotoxicity/phagocytosis (ADCC/ADCP) of T cells,6 7 which limits the efficacy of BiTEs. FcγRs are functionally classified as activating (human FcγRI/CD64, FcγRIIA/CD32A, and FcγRIII/CD16; mouse FcγRI, FcγRIII, and FcγRIV) or inhibitory (human and mouse FcγRIIB).8 Although functionally diverse, most previous studies of FcγRs have focused on leucocytes.9 10 Thus, the roles of FcγRs in melanoma and lymphoma cells have been recently investigated.11 12 FcγRIIb expressed by smooth muscle cells contributes to vascular remodeling13 and FcγRIII/CD16+ fibroblasts was shown to foster a trastuzumab-refractory microenvironment in breast cancer.14 However, the clinical significance, function and underlying mechanisms of FcγRs in non-leucocyte cells in PDAC remain elusive. Thus, further investigations are warranted to assess the effects of Fc-FcγR on the therapeutic efficacy of CLDN18.2 BiTEs.

The importance of CD8+ T cell stemness in cancer immunotherapy has been established.15–20 Antigen-experienced stem-like CD8+ T cells express both the high mobility group-box transcription factor T cell factor 1 (TCF-1) and the checkpoint receptor programmed cell death protein 1 (PD-1) and possess self-renewal capacity in response to viral or tumour antigens. PD-1+TCF-1+ CD8+ T cells act as a reservoir that continually produces TCF-1- effector T cells exhibiting cytotoxic functions. Enhancing proliferation of stem-like tumour-reactive CD8+ T cells could effectively improve the effects of BiTEs.15 21 However, the intrinsic regulators of T cell stemness in the tumour and the ability of CLDN18.2 BiTEs to regulate proliferation and maintenance of stem-like CD8+ T cells remain unclear.

The aim of the present study was to clarify the mechanisms limiting the efficacy of CLDN18.2 BiTEs and the roles of FcγRI (CD64) in desmoplasia and the cyclic guanosine-adenosine monophosphate synthase (cGAS)/stimulator of interferon genes (STING) pathway in maintenance of CD8+ T cell stemness. Based on these findings, a regimen of a STING agonist and vilanterol to target CD64+ cancer-associated fibroblasts (CAFs) was developed to enhance the therapeutic efficacy of CLDN18.2 BiTEs against PDAC.

ResultsCLDN18.2 BiTEs induced target-dependent tumour cytotoxicity and T cell expansion and activation in vitro

CLDN18.2/CD3 BiTEs were generated by immunising Veloc immune mice separately for CLDN18.2 and CD3 BsAbs.22 23 Flow cytometry showed that the resulting hinge-stabilised, effector minimised, IgG4 isotype CLDN18.2/CD3 BiTEs (figure 1A) specifically bound to human primary patient-derived xenograft (PDX)1–3# cells expressing Claudin18.2, but not Claudin18.2-knockout (KO) cancer cells (figure 1B), and murine cell lines expressing Claudin18.2, indicating conservation of the Claudin18.2 sequence in human and mouse cells (figure 1C). Moreover, the CLDN18.2/CD3 BiTEs bound to human embryonic kidney 293T expressing CLDN18.1, but not CLDN18.2 (figure 1D and online supplemental figure S1A), indicating differentiation of the CLDN18 isoforms. Additionally, the BiTEs specifically bound to CD3 of human CD4+ and CD8+ T cells (figure 1E) and humanised hCD3e (online supplemental figure S1B).

Figure 1

Anti-claudin-18 isoform 2 (CLDN18.2)/anti-cluster of differentiation 3 (CD3) bispecific antibody induces target-dependent tumour cytotoxicity and T cell expansion and activation in vitro. (A) The structure of CLDN18.2 bispecific T-cell engagers (BiTEs). (B–D) The specific binding to human (B) and murine (C) CLDN18.2 but not CLDN18.1 (D) in indicated cell lines. (E) Specific binding to CD3 on human CD4+ and CD8+ T cells. (F–M) Specific cell killing assay of indicated 2D primary patient-derived xenograft (PDX) cell lines and 3D organoids incubated with serial dilutions of CLDN18.2 BiTEs in the presence of human T cells. Cellular and organoid viability was evaluated by CellTiter-Glo assay. (N–O) The real-time killing assay of indicated cell lines incubated with CLDN18.2 BiTEs in the presence of human T cells. (P–Q) The immunological synapse-induced BiTEs were evaluated by flow cytometry. (R–S) The effects of BiTEs on the function and proliferation of CD4+ T cells (R) and CD8+ T cells (S) by flow cytometry. All experiments were repeated three times independently. Statistical analysis in N–O was carried out using repeated measure two-way analysis of variance, in P–Q was carried out using two-tailed Mann-Whitney U test, in R–S was carried out using paired Student’s t-test. *P<0.05; **p<0.01; ***p<0.001; ****p<0.001 and n.s., non-significant.

An in vitro killing assay with a T cell/cancer cell coculture system was used to evaluate the cytotoxicity of the CLDN18.2/CD3 BiTEs against tumour cells. As shown in figure 1F–H, BiTEs demonstrated strong cytotoxicity against human PDX cell lines, but not after KO of CLDN18.2 (figure 1I). Similar results were observed in organoids (figure 1J–M). Additionally, a time-lapse monitoring system verified that the BiTEs effectively induced formation of immunological synapses between the tumour and CD8+ T cells, apoptosis of target tumour cells via secretion of granzyme B, and proliferation of T cells (figure 1N–S).

CLDN18.2 BiTEs effectively inhibited early tumour growth, but late-stage efficacy was significantly diminished in humanised mouse models

The promising in vitro data led to further in vivo investigations of CLDN18.2 BiTEs activities in humanised mouse models. As shown in figure 2 and online supplemental figures S2–S4, CLDN18.2 BiTEs (0.1 mg/kg) significantly delayed tumour progression, reduced tumour burden (figure 2A–D), exhibited minor cytotoxicity (figures 4), decreased proportion of Ki67+ tumour cells (figure 2F and online supplemental figure S5A) and increased apoptosis of tumour cells (figure 2G and online supplemental figure S5B) and in CD34+ humanised mice, consistent with a human PBMC-transferred mice model (figures 4 and online supplemental figure S5C,D). Besides, in the orthotopic KPC humanised hCD3e mice model, CLDN18.2 BiTEs effectively decreased tumour growth and the proportion of Ki67+ tumour cells, while promoting apoptosis of tumour cells, which significantly prolonged survival of mice, demonstrating that CLDN18.2 BiTEs inhibited growth of PDAC cells (figure 2O–V and online supplemental figure S5E,F). However, CLDN18.2 BiTEs had no effect on growth of CLDN18.2-negative PDX tumours and KPC-Cldn18.2-KO GEMMs (figure 2W–Y and online supplemental figure S6A–D).

Figure 2

Anti-claudin-18 isoform 2 (CLDN18.2)/anti-cluster of differentiation 3 (CD3) bispecific antibody effectively inhibits growth of pancreatic ductal adenocarcinoma (PDAC) tumour in the early stage, but the anti-tumour efficacy was significantly diminished in the late stage. (A–V) Therapeutic effects of CLDN18.2 bispecific T-cell engagers (BiTEs) (0.1 mg/kg) on the CLDN18.2-positive patient-derived xenografts (PDXs) and KPC cell lines in three humanised mice models. CD34+HSC-transplanted humanised mice (A–G), human PBMC-transferred humanised mice (H–N) and hCD3e humanised mice (O–V). Schematic illustration for the design (A, H and O), representative tumour images (B, I and P), analysis of tumour growth curve (C, J and Q), tumour weight (D, K and R), body weight (E, Land S), Ki67+ tumour cells (F, M and T), TUNEL+ tumour cells (G, N and U) and survival curve (V). (W–Y) Therapeutic effects of BTEs on CLDN18.2-negative PDXs in CD34+ humanised mice. Representative tumour images (W), tumour growth curve (X) and tumour weight (Y). All experiments were repeated three times independently. Six mice per group in all experiments (n=6). Statistical analysis in C, E, J, L, Q, S and X was carried out using repeated measure two-way analysis of variance, in D, F, G, K, M, N, R, T, U and Y was carried out using unpaired Student’s t-test. Survival analysis in V was carried out using Kaplan-Meier survival curves with log-rank test. Representative histological images for F–G, M–N and T–U were shown in Fig. S**. *P<0.05; **p<0.01; ***p<0.001; ****p<0.001 and n.s., non-significant.

Figure 3

Anti-claudin-18 isoform 2 (CLDN18.2) bispecific T-cell engagers (BiTEs) could interact with CD64+ cancer-associated fibroblast (CAF) and enhance desmoplasia in pancreatic ductal adenocarcinoma (PDAC). (A–D) The desmoplasia of patient-derived xenograft (PDX) tumours in CD34+HSC-transferred humanised mice each group in day 0, day 12 and day 32. Masson staining (A), alpha smooth muscle actin (α-SMA) immunohistochemistry (IHC) staining (B), Collagen-I IHC staining (C) and tissue stiffness (D). (E–I) CLDN18.2 BiTEs induced desmoplasia of CAFs in vitro. Oil red staining (E–F), mRNA expression of α-SMA and Collagen-I (G), collagen contraction assay (H–I). (J–L) Fc fragment of BiTEs accounted for the prodesmoplasia of BiTEs. Schematic illustration of the Fc/Fab digestion assay (J). Oil red staining (K) and mRNA expression of α-SMA and Collagen-I (L). (M–O) The expression of different FCGRs was evaluated in primary CAFs by flow cytometry (FCM) (M–N), fresh PDAC tissues by sc-RNA sequence (O). (P) Representative of CD64+ CAFs in PDAC tissues by multiplexed IHC staining. (Q) Survival analysis of PDAC patients in our centre according to the expression of CD64+ CAF. (R–S) The correlation between percentage of CD64+ CAFs and tissue desmoplasia (R) and CD8+ T cell infiltration (S) in PDX tumours after CLDN18.2 BiTEs administration. (T–W) CLDN18.2 BiTEs induced desmoplasia (T–V) and decreased T cell infiltration (W) in CD64+ CAF-organoid coinjection models in hCD34+ humanised mice. All experiments were repeated three times independently. 6 replicates per group (n=6) in A–F, H–I and K, 6 mice per group (n=6) in T–W, 3 replicates per group in G and L (n=3), 15 fresh perspective PDAC specimens in M–N (n=15), 111 retrospective PDAC cases in Q (n=111), 21 PDXs in R–S (n=21). Statistical analysis in A–D, U–W was carried out using unpaired Student’s t-test, in E–L was carried out by paired Student’s t-test, in Q was performed using Kaplan-Meier survival curves with log-rank test, in R–S was carried out by Spearman’s rank correlation analysis. Representative histological images for A–C and U and dot plots for W were shown in Fig. S**. *P<0.05; **p<0.01; ***p<0.001; ****p<0.001 and n.s., non-significant.

Figure 4

Vilanterol, a promising inhibitor for VAV2, effectively inhibited desmoplasia induced by CD64+ fibroblasts in anti-claudin-18 isoform 2 (CLDN18.2) bispecific T-cell engagers (BiTEs) and sensitised pancreatic ductal adenocarcinoma (PDAC) to CLDN18.2 BiTEs. (A) Gene set enrichment analysis (GSEA) analysis of CD64+ CAFs treated with CLDN18.2 BiTEs according to the RNA-seq. (B) Differentially expressed genes (DEGs) analysis of CD64+ cancer-associated fibroblasts (CAFs) treated with CLDN18.2 BiTEs according to the proteomics (left) and phosphoproteomics (right). (C) The potential signal pathway in CD64+ CAF induced by CLDN18.2 BiTEs according to the DEGs. (D) The SYK-VAV2-RhoA-MLC2-α-SMA/Collagen-I axis in primary CD64− and CD64+ CAFs treated with BiTEs was evaluated in vitro. (E) The CD64− and CD64+ CAFs isolated from BiTEs-treated patient-derived xenograft (PDXs) and then the SYK-VAV2-RhoA-MLC2-αSMA/Collagen-I axis was evaluated. (F) Deletion of SYK, VAV2, RhoA or myosin light chain 2 (MLC2) could abrogate the expression of alpha smooth muscle actin (α-SMA) and Collagen-I in CD64+ CAFs treated with BiTEs. (G–H) The subcellular distribution of MRTF-A in CD64− and CD64+ CAFs treated with BiTEs was evaluated by IF. (I–L) CD64+ CAFs with MRTF-A deletion were treated with BiTEus and then the expression of MRTF-A, α-SMA and collagen I in nuclear (Nuc) or cytoplasm (Cyto) fractions (I), percentage of oil red+ cells (J), HA concentration (K) and collagen contraction capacity (L) were determined. (M–O) The molecular docking for the Vilanterol and VAV2. (P–T) The effects of vilanterol to sensitise BiTEs treatment were evaluated in CD64+ CAFs-organoid coinjection Hcd34+ humanised mice model. Growth curve (P), tumour desmoplasia (Q–S) and infiltration of CD8+ T cells (T). All experiments were repeated three times independently. 6 replicates per group (n=6) in H, J–L, 6 mice per group (n=6) in P–T. Statistical analysis in H, J–L was carried out using paired Student’s t-test, in p was carried out using repeated measure two-way analysis of variance, in Q–T was performed using unpaired Student’s t-test. Representative histological images for Q–S and dot plots for T were shown in Fig. S**. *P<0.05; **p<0.01; ***p<0.001; ****p<0.001 and n.s., non-significant.

Growth curve analysis of CLDN18.2 BiTEs group showed that tumour grow slightly in the early stage (<20 days), while accelerated in the late stage (>20 days) (figure 2C, J and Q). However, CLDN18.2 BiTEs at 1 mg/kg was insufficient to control late-stage tumour growth (online supplemental figure S6E,F). Collectively, these results suggested that CLDN18.2 BiTEs effectively inhibited early tumour growth, but late-stage efficacy was significantly diminished.

CLDN18.2 BiTEs interact with CD64+ CAFs and induced desmoplasia in PDAC

An investigation of the mechanisms underlying the late-stage tumour-suppressive effects demonstrated that CLDN18.2 BiTEs moderately increased the proportion of CD8+ and CD4+ tumour-infiltrating lymphocytes (TILs) in two humanised mouse models (Online supplemental figure S7A–H), consistent with a previous report.24 No obvious difference of other immune populations (including NK cells, myeloid-derived suppressor cells, macrophages, B cells and dendritic cells) between control and BiTEs group was observed (online supplemental figure S8). However, the proportion of TILs was sharply reduced in the BiTEs group on day 32 as compared with day 12 (online supplemental figure S7B–D and S7F–H). Desmoplasia, as a salient feature of the tumour microenvironment, significantly decreases vascular perfusion, immune infiltration and drug delivery.25 Further analysis by Masson and alpha smooth muscle actin (α-SMA)/collagen I staining showed significantly increased fibrosis in the BiTEs group from days 12 to 32 (figure 3A–C and online supplemental figure S9A–C). Besides, CLDN18.2 BiTEs could notably increase the tissue stiffness in PDAC (figure 3D), suggesting that CLDN18.2 BiTEs exacerbated tumour fibrosis. Moreover, BiTEs significantly induced activation of primary quiescent CAFs isolated from fresh PDAC tumours, as determined by oil red staining, expression of α-SMA/collagen I, and collagen contraction capacity of CAFs (figure 3E–I), demonstrating that CLDN18.2 BiTEs directly induced desmoplasia and fibrosis in PDAC.

Digestion with pepsin to determine whether the Fc or Fab fragment of BiTEs was responsible for activation of CAFs found that BiTEs without the Fc fragment failed to induce fibrosis both in vitro and in vivo (figure 3J–L, online supplemental figure S10A–E). Interaction with the IgG Fc-FCG receptor (FCGR) is required for optimal efficacy of various Abs.26 Flow cytometry of the expression pattern of human FCGRs on CAFs found that only CD64 (FCGRI) was abundantly expressed on CAFs, while signalling of Abs specific for CD16 (FCGRIII), CD32b/c (FCGRIIb/c), and CD32a (FCGRIIa) was minimal (figure 3M,N). Consistently, scRNA-seq from our center and multiplex immunohistochemical staining confirmed a high proportion of CD64+ CAFs in PDAC (figure 3O and P). Similarly, the Fc fragment of CLDN18.2 BiTEs demonstrated strong affinity for mouse FCGR1, but not FCGR2, FCGR3 and FCGR4 (online supplemental figure S11A,B). These results indicate that although the LALA mutation was introduced to the Fc region to avoid ADCC and ADCP,27 the Fc fragment of BiTEs still interacted with FCGR1. Moreover, the proportion of CD64+ CAFs was significantly positively correlated with poor prognosis and tissue fibrosis, while negatively correlated with infiltration of CD8+ T cells (figure 3Q–S) in PDAC. Furthermore, BiTEs failed to induce the desmoplasia in CD64-deleted human and murine CAFs in vitro (online supplemental figure S12A–H). Consistently, in vivo, BiTEs effectively increased expression of α-SMA and the elastic modulus, and decreased the proportion of CD8+ T cells in the CD64+ CAF-tumour coinjected group (figure 3T–W). Moreover, we generated a fibroblast-specific conditional Fcgr1/CD64 KO model by crossing Fcgr1/CD64 flox/flox mice with Acta2-Cre mice (online supplemental figure S12I) and we also found that CLDN18.2 BiTEs failed to induce the collagen deposition in CD64-KO mice compared with wild type mice (online supplemental figure S12K). Moreover, the orthotopically-implanted KPC organoids model was performed to validate the prodesmoplasia of CLDN18.2 BiTEs and consistent results were also observed (online supplemental figure S13). These results suggest that the interaction of CLDN18.2 BiTEs with CD64+ CAFs enhanced desmoplasia in PDAC.

Besides, one responder and one non-responder from a clinical trial about the CLDN18.2 BiTEs (NCT05164458) had low and high proportions of CD64+ CAF, respectively (online supplemental figure S14A–C), suggesting that PDAC patients with high tumour cell expression of CLDN18.2 and low stromal cell expressions of CD64 could benefit from CLDN18.2 BiTEs therapy. However, a larger cohort is needed to confirm the correlation between CD64+ CAFs and responsiveness to BiTEs.

Vilanterol, a promising VAV2 inhibitor, effectively inhibited desmoplasia induced by CLDN18.2 BiTEs and improved treatment efficacy

RNA-seq analysis was conducted to elucidate the molecular mechanism underlying activation and fibrosis of CAFs found that CLDN18.2 BiTEs induced abnormal activation of the Immunoglobulin G (IgG)-Receptor for the constant fragment of immunoglobulin G (FCGR)-Rac family small GTPase 1 (RAC) and extracellualr matrix (ECM) production pathways (figure 4A). Comprehensive proteome and phosphorproteome profiling showed that CLDN18.2 BiTEs sharply increased expression of α-SMA, collagen I, p-SYK (Tyr525), p-VAV2 (Y172), and RhoA-GTP, p-MLC2 (Ser19) (figure 4B). Interaction of Fc with FcgR activated SYK, which subsequently phosphorylated the VAV guanine nucleotide exchange factor (VAV) (figure 4C).28–30 Tyrosine-phosphorylated VAV catalyses GDP/GTP exchange on Rac-1, which further activates myosin light chain 2 (MLC2) and induces actin polymerisation and ECM production in fibroblasts.31 Consistently, p-SYK (Tyr525), p-VAV2 (Y172), RhoA-GTP and p-MLC2 (Ser19) were markedly increased in BiTEs-treated CD64+ CAFs, corresponding with enrichment of α-SMA and collagen I (figure 4D). Besides, p-SYK (Tyr525), p-VAV2 (Y172), RhoA-GTP and p-MLC2 (Ser19) were markedly elevated in CD64+ CAFs isolated from BiTEs-treated PDX tumours (figure 4E). More importantly, silencing SYK, VAV2, RhoA and MLC2 reduced activation of related downstream mediators and levels of a-SMA/collagen I (figure 4F). Additionally, MLC2 phosphorylation induces actin polymerisation, which recruits G-actin to F-actin stress fibres with subsequent release of MRTF-A from G-actin to the nucleus to regulate genes encoding various cytoskeletal and cell adhesion components, including α-SMA, calponin 1 and collagen 1.32–34 Immunofluorescence staining showed increased nuclear translocation of MRTF-A in CD64+ fibroblasts treated with BiTEs (figure 4G,H). MRTF-A depletion abrogated fibrosis of CD64+ CAFs induced by BiTEs (figure 4I–L). Taken together, these data suggest that CLDN18.2 BiTEs activate the CD64-SYK-VAV2-RhoA-ROCK-MLC2-MRTF-A pathway in CD64+ fibroblasts to mediate desmoplasia.

We found that CAFs exhibited the high expression level of VAV2 (online supplemental figure S15A,B) and VAV2 is dispensable for ADCC by NK cells35 and ADCP by macrophages.32 So, to determine whether inhibition of VAV2 in CAFs could improve the efficacy of BiTEs, VAV2 was silenced in CD64+ fibroblasts, which substantially suppressed the proportion of oil red+ CAFs and collagen contraction capacity (online supplemental figure S15C,D). By contrast, VAV2 knockdown in NK cells and macrophages had no effect on ADCC and ADCP (online supplemental figure S15E–H). Consistently, VAV2-targeting AAVs markedly improved BiTEs efficacy, suppressed tumour desmoplasia, and increased CD8+ T cell infiltration in the CD64+ CAF-organoid coinjection model (online supplemental figure S16). Collectively, these data suggest that VAV2 is a potential therapeutic target to enhance the efficacy of CLDN18.2 BiTEs.

Considering the clinical translational significance of VAV2, high-throughput virtual screening was performed to target Y172 of p-VAV2 (online supplemental figure S17A–D, table S1). Considering the inhibition efficacy of Y172 of p-VAV2, vilanterol was selected for further studies (figure 4M–O and online supplemental figure S18A, table S2). Vilanterol significantly inhibited the SYK-VAV2-RhoA-GTP-MLC2-SMA/collagen-I axis and desmoplasia in CD64+ CAFs (online supplemental figure 18B). Mechanically, the effects of vilanterol were almost abrogated by mutating Y172 of VAV2 (online supplemental figure S18C). An in vivo assay demonstrated that vilanterol effectively enhanced BiTEs efficacy, suppressed tumour desmoplasia and increased CD8+ T cell infiltration (figure 4P–T and online supplemental figure S19A–C). Collectively, these results indicate that vilanterol is a potential novel inhibitor of VAV2 to prevent desmoplasia and enhance the efficacy of CLDN18.2 BiTEs.

The proportion of stem-like CD8+ T cells was sharply decreased in the CLDN18.2 BiTEs group and further hampered the therapeutic effects of CLDN18.2 BiTEs

Considering that CLDN18.2 BiTEs promote a T cell-mediated anti-tumour immune response, in vivo CD4 or CD8 deletion assays were performed to determine the essential immune phenotype associated with BiTEs (figure 5A,B). CD8 deletion completely and CD4 deletion partially abolished the anti-tumour effect of BiTEs, suggesting a contributory role of CD8+ T cells (figure 5C). Considering the key role of CD8+ T cells in the killing effects of CLDN18.2 BiTEs in vivo, tumour-infiltrated CD8+ T cell subtypes of the BiTEs group were profiled at different stages by flow cytometry (figure 5D–J). The proportions of stem-like resource PD1+ TCF-1+ CD8+ T cells, progeny PD-1+TCF-1− CD8+ T cells, central memory-like subset CD45RO+CD62L+ CD8+ T cells, effector CD45RA− CD62L− CD8+ T cells and Ki67+/Granzyme B+/IFN-α+/Perforin+ CD8+ T cells were significantly reduced on day 32 as compared with day 12 (figure 5K–O). Moreover, the CD8+ TILs-organoid coculture assay were performed (figure 5P). The CD8+ TILs isolated from BiTEs group in day 12 exhibited strong anti-tumour effects, while TILs in day 32 showed impaired anti-tumour capacity (figure 5Q–R). Stem-like CD8+ T cells mediate responses of adoptive cell immunotherapy against human cancers.33 Thus, CD8+ T cells were activated with anti-CD3/CD28 beads in the presence of interleukin-15 (IL-15) and IL-7 to induce production of stem cell-like T cells before transfer (figure 5S), which enriched PD-1+TCF-1+ stem-like CD8+ T cells (figure 5T). Activated CD8+ T cell significantly delayed tumour growth (figure 5U), while depletion of TCF1 by specific shRNA in adoptively transferred activated CD8+ T cells impaired anti-tumour effects, similar to the untreated group (figure 5V–X and online supplemental figure S20A–B). These results show that decreasing the proportion of late-stage stem-like tumour-reactive CD8+ T cells hampered the therapeutic effects of CLDN18.2 BiTEs.

Figure 5

The stem-like CD8+ T cells were significantly decreased in the late-stage of anti-claudin-18 isoform 2 (CLDN18.2) bispecific T-cell engagers (BiTEs) group, which hampered the therapeutic effects of CLDN18.2 BiTEs. (A–C) In vivo CD4 or CD8 deletion assay in BiTEs-treated Hcd34+ humanised mice. Schematic illustration (A), representative CD4/CD8 dot plots in each group (B) and growth curve (C). (D–O) The percentage of different subtypes of CD8+ T cells in Hcd34+ mice treated with CLDN18.2 BiTEs or not in day 0, day 12 and day 32 was evaluated by flow cytometry. (PR) CD8+ tumour-infiltrating lymphocytes (TILs) isolated from patient-derived xenograft (PDX) tumours in days 0, 12 and 32 were cocultured with organoid in vitro and then the killing effects were evaluated by Caspase3/7 probe assay (P). Representative apoptotic organoid images (Q) and statistical analysis (R). (S–U) CD8+ TILs isolated from BiTEs group in day 32 were pretreated with IL-7/interleukin-15 (IL-15) to enrich TCF-1+PD-1+ stem like T cells and then transferred into the PDX mice. Schematic illustration (S), percentage of subtypes of T cells in CD8+ TILs treated with IL-7/IL-15 (T) and tumour growth curve (U). (V–X) CD8+ TILs isolated from BiTEs group in day 32 were first transfected with shRNAs for TCF-1, then pretreated with IL-7/IL-15, and subsequently transferred into the PDX mice. Schematic illustration (V), percentage of subtypes of T cells in CD8+ TILs treated with IL-7/IL-15 (W) and tumour growth curve (X). All experiments were repeated three times independently. 6 mice per group (n=6) in C, D–O, U, X, 6 replicates per group (n=6) in Q–R, 3 replicates per group (n=3) in T, W. Statistical analysis in C, U and X was carried out using repeated measure two-way analysis of variance, in D–O was carried out using unpaired Student’s t-test, in R, T and W was carried out using paired Student’s t-test. *P<0.05, **p<0.01; ***p<0.001; ****p<0.001 and n.s., non-significant.

Decreased STING activity reduced the proportion of late-stage stem-like tumour-reactive CD8+ T cells in the CLDN18.2 BiTEs group

RNA-seq was performed to clarify the early and late-stage roles of CD8+ T cells in subcutaneous tumours of mice following BiTEs administration. GSEA analysis showed increased activation of multiple immunity-related pathways in CD8+ T cells on day 12, but decreased activity on day 32 (figure 6A). Differentially expressed genes were related to migration and adhesion, cytokines and cytokine receptors, effector genes, inhibitory receptors and some transcription factors (figure 6B). Importantly, late-stage cGAS/STING signalling and type I interferon (IFN) response pathways were significantly downregulated in TILs (figure 6A). Western blot analysis indicated that expression of p-STING, p-TBK1, p-IRF3, p-cGas and p-p65 in addition to several stemness-related genes was decreased in TILs on day 32 versus day 12, consistent with stem-like CD8+ T cells (figure 6C). The T cell-specific DNA-binding protein, T cell factor 1 (Tcf1, encoded by Tcf7), is the key transcription factor of the canonical Wnt signalling pathway, which plays a crucial role in T cell fate specification by initiating a T cell gene programme downstream of Notch signalling.34 Accordingly, Tcf7 germline knockout mice display severely impaired T cell development and stemness.36 Chromatin immunoprecipitation and the luciferase assay of CD8+ T cells showed that IRF3 and p65 bound to the promoter of TCF-7, which increased transcription on day 12 versus day 32 (online supplemental figures S21A–F, 6D,E). Meanwhile, binding of the transcription activation markers H3K27ac and H3K9ac to the TCF-7 promoter was increased on day 12 versus day 32, while binding of the transcription suppression markers H3K27me3 and K3K27me was decreased (online supplemental figure S21G–J). STING/cGAS signalling is activated by enrichment of cytosolic DNA.37 The genomic marker TERT, but not the mitochondrial DNA marker DLOOP1, exhibited dynamic changes similar to TCF-1 in CD8+ TILs (figure 6F,G).