Converting TCR-based chimeric antigen receptor STAR into dual-specific targeting receptor for cancer immunotherapy

Abstract

Chimeric antigen receptor (CAR) T cell therapy has achieved great success in treating hematopoietic malignancies; however, post-therapy relapse remains a challenge. Traditionally, multi-specific CAR engineering requires precise arrangement of single-chain variable fragments (scFvs), which can lead to aggregation issues when assembled linearly. In this study, we developed a novel chimeric receptor, the dual-targeting synthetic TCR and antigen receptor (D-STAR). D-STAR exhibited structural advantages, activating T cells and inducing effector functions in response to single antigen stimulation while mediating robust killing against various malignant B cells. In mouse models, D-STAR demonstrated superior antitumor efficacy compared to single- and dual-targeting CAR-T cells. To enhance its effectiveness, we integrated the OX40 costimulatory cytoplasmic domain with flexible linkers, boosting T cell proliferation and fitness under higher tumor burdens in vivo. This study illustrates the superior structural capacity and antitumor potency of D-STAR T cells.

Graphical abstract

Lin and colleagues developed a dual-target chimeric receptor with structural advantages for both solid and liquid tumors, showing robust activity against malignant B cells and surpassing second-generation CARs in a tumor-escape mouse model. This work offers a promising therapeutic approach for relapsed/refractory tumors and potential optimizations in cell therapy.

Introduction

Since the late 1960s, adoptive cell therapy (ACT) has rapidly developed, showcasing significant therapeutic and curative potential for treating malignancies.¹ This approach includes various cell types, such as cytokine-induced killer (CIK) cells, tumor-infiltrating lymphocytes (TILs), chimeric antigen receptor T (CAR-T) cells, and T cell receptor engineered T (TCR-T) cells. Among these, CAR-T cell therapy has garnered considerable attention and achieved remarkable progress in cancer therapy research.²

CAR-T cell therapy has achieved significant success in treating CD19-positive and BCMA-positive malignancies.^3,4 Since 2017, 12 CAR-T cell therapy products have been approved by the US Food and Drug Administration (FDA).^5,6 Despite the encouraging results in treating malignant lymphoma, post-therapy relapse occurs in 30%–60% patients following CD19-CAR-T cell infusion.⁷ There are two patterns of post-CAR-T cell relapse: antigen-positive relapse, caused by poor persistence of CAR-T cells,⁸ and antigen-negative relapse. In the latter case, tumors can evade CAR-mediated recognition and clearance when CD19 is absent or truncated, despite CAR-T cell persistence.⁹ One of solutions to overcome this issue is to engineer multi-specific CAR-T cells. However, current strategies face various obstacles, such as the proper arrangement and positioning of each single-chain variable fragment (scFv), which can affect the effector function of CAR-T cells.^10,11 Additionally, CAR-T cell-related toxicity is complex, multi-factorial, and potentially lethal, including cytokine release syndrome and neurotoxicity.^12,13 In contrast, TILs or TCR-T, which rely on natural T cell signaling, have reported lower adverse events rate on clinical trials.^14,15 Unlike TCRs, CARs trigger T cell response through the intracellular domain CD3ζ and activation domains derived from costimulatory receptors.^16,17,18 On the other hand, TCR activation depends on the changes in the arrangement of its CD3 subunits, which are stabilized with the TCRαβ chain.¹⁹

Previously, we developed a multi-chain chimeric receptor, termed synthetic TCR and antigen receptor (STAR),²⁰ by fusing antibody variable regions with the constant regions of TCRαβ chain. STAR combines antigen specificity with enhanced signaling capability inherited from TCR. This innovative design has demonstrated functional advantages over CAR-T cells in solid tumor models, including faster tumor regression, improved T cell infiltration, and a reduced exhaustion phenotype. Based on promising preclinical evidence, CD19 STAR T cell therapy has completed an investigator-initiated trial (IIT) clinical trial (NCT03953599), with results demonstrating technical feasibility, clinical safety, and efficacy for patients with refractory and relapsed (R/R) B cell acute lymphoblastic leukemia (B-ALL).²¹

In this study, we unveiled the engineering process of the double-chain STAR receptor and utilized this structure to further develop a dual-targeting STAR, termed as dual-targeting STAR (D-STAR). D-STAR receptor leverages natural TCR signaling pathways and incorporates the variable regions of antibodies to recognize two different tumor antigens. D-STAR receptor possess a structural advantage in its compatibility, making it easily engineered for multi-targeting. D-STAR T cells can be activated by either antigen and exhibit potent cytotoxicity against highly heterogeneous malignant B cells. Furthermore, in a tumor-escape mouse model, D-STAR T cell outperformed single- and dual-target second-generation CAR-T cells in tumor control and extended survival. By incorporating the OX40 costimulatory domain, D-STAR T cells exhibited enhanced proliferation, leading to improved performance under higher tumor burdens in vivo. These features suggest that D-STAR T cells hold promising potential for engineered-T cell therapy in treating malignancies.

Results

Engineering process of the double-chain STAR receptor

In the previous study, single-targeting STAR receptor was designed with variable regions of immunoglobulin heavy and light chains (VH and VL) fused to murine-mutant TCR-Cα and TCR-Cβ. This structure could pair with endogenous CD3 module and initiate downstream signaling in human T cells.²¹ We speculated whether transduced single-chain receptor featuring human wild-type (WT) TCR-Cα or -β could similarly pair with endogenous TCRβ/α to initiate downstream TCR signaling (Figure 1A). Contrary to expectations, surface expression analysis using FMC63⁺ and red fluorescence protein (RFP) reporter showed that these receptors failed to localize on the human T cell membrane (Figure 1B). Subsequent analysis suggested that an imbalance in TCR subunit assembly within the endoplasmic reticulum (ER) may lead to degradation.^22,23

Figure 1.

Rational development of STAR receptor

(A) Schematic design of the single-chain STAR receptor paired with endogenous TCRα or β chain. (B) Surface expression of the single-chain STAR receptor on Jurkat cells, transduced with various receptor structures. Representative flow plots showing the mean fluorescence intensity (MFI) of the STAR receptor on the T cell membrane. (C) Alanine mutations at ubiquitination sites in human WT-TCRα transmembrane and intracellular regions, referred to as hu-mut-TCRA. The left flow plots illustrate human TCRα/β expression in Jurat and JC5 cells. The right flow plots show the MFI of the STAR receptor on the Jurkat or JC5 cell membrane. (D) Surface expression of the STAR receptor, designed as either a single chain or co-transduced with myc-hu-WT-TCRB/his-hu-WT-TCRA at a multiplicity of infection (MOI) of 10, ensuring equivalent expression levels of TCRα and β constant regions on Jurkat cells. Numbers in the panel represent surface expression efficiency. (E) Gene schematics of the double-chain STAR receptor, designed with human WT TCRα/β, human mut TCRα/β, and murine-mutant TCRα/β constant regions. (F) Surface expression of double-chain STAR receptors on human primary T cells and the activation level following sorting of STAR⁺ T cells incubated with Raji cells for 6 h at an E:T ratio of 1:1. This experiment was repeated twice with different donors.

To address this issue, we introduced alanine substitutions at ubiquitination sites on the TCRα chain to prevent degradation in ER, termed hu-mut-TCRA variant (Figure 1C). Compared with hu-WT-TCRA, single-chain STAR receptor with hu-mut-TCRA (FMC-mut-TCRA) yielded slightly enhanced surface expression on human T cell membrane. This indicated that STAR receptor may share a similar assembly pattern with TCR (Figure 1C). We also transduced FMC63-mut-TCRA into JC5 cells, which have knocked out the endogenous TCRα/β chain, as detected by fluorescence-activated cell sorting (FACS) (Figure 1C). The results confirmed the necessity of hu-mut-TCRA pairing with endogenous WT TCRβ for proper membrane localization. However, these modifications did not completely meet our expectation.

Given the structural and spatial similarities between STAR and TCR, we hypothesized that the imbalance between TCRα and β chain expression could induce degradation. To investigate this, we co-infected Jurkat cells with the single-chain STAR receptor and myc/his-human WT TCRβ/α at the same multiplicity of infection (MOI). Notably, the structure of myc/his-human WT TCRβ/α excluded the effect of variable region on TCRα/β chain, as shown in Figure 1D. Compared to previous structure, the co-transduced single-chain STAR receptor showed increased surface display efficiency on human T cells (Figure 1D).

In summary, we found that the single-chain STAR receptor with human WT TCR-Cα/β failed to pair with endogenous TCR due to imbalanced expression, which led to ER-dependent degradation. Based on these results, we generated a double-chain STAR receptor containing either human WT/mut TCR-Cα and β or humanized murine-mutant TCR-Cα and β as previously reported (Figure 1E). The results demonstrated that the double-chain STAR receptor with humanized murine-mutant TCR-Cα and -β showed better surface display and activation efficiency on human T cell when incubated with Raji cell for 6 h. Moreover, we observed that the linearly arranged anti-CD19 (FMC63) scFv did not affect surface display or T cell activation, which enlighten the potential application of a dual-targeting STAR receptor, featuring scFvs that target different tumor antigens assembled on both TCR-Cα and β (Figure 1F).

D-STAR receptor features structural properties suited for engineering dual-targeting T cells

Despite the success of CAR-T cell therapy in treating hematopoietic tumors, its effectiveness can be compromised by high rates of post-therapy relapse. Engineering CARs with multiple targets is one promising strategy, but the arrangement and positioning of each scFv in the CAR’s extracellular domain are critical for optimal antitumor efficacy.^24,25 Initially, we developed several dual-targeting 4-1BBζ CARs with scFvs positioned differently (Figure S1A). Consistent with previous reports, structural variations in dual-targeting CARs can affect their effector function against tumor cells (Figure S1B). Additionally, dual-CARs exhibited autoactivation and early T cell exhaustion before antigen stimulation (Figure S1C). Given these challenges and previous successes with double-chain STAR receptors, we decided to address these issues by remodeling the traditional CAR structure.

Based on the above results, we engineered a D-STAR with two different scFvs separately arranged on the humanized murine-mutant TCRα and β constant regions (Figure 2A). To investigate whether the position of the scFvs affects D-STAR T cell function, we switched the positions, termed CD19-CD20-STAR and CD20-CD19-STAR (Figure 2A). We then assessed the effector function of D-STARs transduced into human primary T cells. The results demonstrated that changing the position of the scFvs on D-STAR did not impact T cell performance regarding tumor lysis ability (Figure 2B) or cytokine production toward CD19^KO-Raji cells and CD20^KO-Raji cells (Figure 2C).

Figure 2.

Advantages of the D-STAR structure

(A) Design and gene schematics of D-STAR, featuring swapped position of anti-CD19 and anti-CD20-scFvs. (B) Cytotoxicity of D-STARs against CD19^KO or CD20^KO Raji cells at various E:T ratios. Data shown as means ± SEM are technical duplicates from one of three independent experiments. p values were determined using two-way ANOVA with Tukey’s correction for multiple comparisons. (C) Effector cytokine production by human primary T cells transduced with D-STARs, targeting CD19^KO or CD20^KO Raji cells after a 24-h coculture at an E:T ratio of 1:1. Data shown as means ± SEM are technical duplicates from one of three independent experiments. p values were determined using one-way ANOVA with Tukey’s correction for multiple comparisons. (D) Design of D-STAR structures incorporating 2C6 or rituximab, which have lower affinity compared to ofatumumab-derived anti-CD20 scFv. (E) Percentage of mixed tumor cells composed of CD19^KO and CD20^KO Raji cells and D-STAR T cells after a 24-h coculture at an E:T ratio of 1:3. This experiment was repeated twice with human primary T cells derived from different donors. (F) CD107a expression on D-STAR T cells incubated with Raji cells at various time points and an E:T ratio of 1:1. Data are from one of three independent experiments. (G) Intracellular staining of granzyme B in D-STAR T cells after an 8-h coculture with CD19^KO or CD20 ^KO Raji cells at an E:T ratio of 1:5. This experiment was repeated twice with different donor-derived human primary T cells. ∗p < 0.05; ns, not significant.

Furthermore, we found that the affinity of FMC63 antibody is 10-fold lower than that of the CD20 antibody ofatumumab (OFA)²⁶ but similar to two other CD20 antibodies, rituximab and 2C6 (Figure 2D). Therefore, we substituted the anti-CD20 scFv derived from ofatumumab with 2C6 or rituximab in the D-STAR structure, termed FMC-2C6-STAR and FMC-RTX-STAR (Figure 2D). We then evaluated whether the affinity of the scFvs could affect D-STAR effector function. D-STARs with the lower affinity of anti-CD20 scFv showed comparable CD107a expression toward Raji cells and tumor lysis ability against a mixture of antigen-loss tumor cells (Figures 2E and 2F). For CD19^KO-Raji cells, which were recognized solely through CD20, we found that FMC-RTX-STAR T cells secreted slightly lower granzyme B compare to others. For CD20^KO-Raji cells, which were recognized by FMC63, all of the group showed equipotent granzyme B secretion, as expected (Figure 2G).

In addition to hematopoietic tumors, we also engineered D-STARs to target solid tumor. Firstly, we assessed the expression of EGFR, ROR1, and PDL1 on various solid tumor and leukemia cell lines (Figure S2A). The results showed that ROR1 is more widely expressed among these cell lines compared to EGFR and PDL1. We then engineered D-STARs targeting either EGFR/ROR1 or EGFR/PDL1, and both of them could display on human T cells (Figure S2B). Compared to EGFR-STAR and EGFR-PDL1-STAR, EGFR-ROR1-STAR demonstrated potential against solid tumors and hematopoietic malignancies, as shown by increased granzyme B secretion and tumor lysis ability, coupled with reduced cytokine secretion (Figures S2C–S2E). These results suggest that dual targeting is readily achievable with the STAR construct.

Additionally, we tested the antigen specificity of D-STAR. First, we incubated matched STAR (CD19-CD20-STAR) and unmatched STAR (EGFR-ROR1-STAR) with CD19^KO, CD20^KO Raji and parental Raji cell, which had confirmed antigen expression earlier (Figure S3A). As expected, only the CD19-CD20-STAR exhibited apparent cytotoxicity against these cell line (Figure S3B). Next, we developed a stable K562 cell line overexpressing either CD19 or CD20, as K562 cells do not naturally express either antigen (Figure S3C). Only the CD19 or CD20-K562, but not the WT K562, were killed by CD19-CD20-STAR T cells, indicated that there is no mispairing between the scFvs. Besides, none of the cell lines were killed by EGFR-ROR1-STAR T cells, demonstrating no aggregation between the two scFvs in the D-STAR structures and underscoring the specificity of STAR structure.

D-STAR T cells exhibited stronger tumor elimination than single-targeting STAR T cells in different lymphoma cells

Our previous work demonstrated that single-targeting STAR exhibited potent cytotoxicity in both hematopoietic and solid tumor models.^20,21 We then compared the D-STAR to the single-targeting STAR receptor (S-STAR) to assess whether D-STAR has the potential to exhibit similarly strong cytotoxicity.

Initially, we transduced Jurkat cells with D-STAR and S-STAR targeting CD19 or CD20 (Figure 3A) and confirmed that both S-STAR and D-STAR could specifically bind to their respective antigens. The binding ability to CD20 was consistently higher than to CD19 due to the difference in scFv affinity, as FMC63 has an approximately 10-fold lower affinity than ofatumumab (Figure 3B). Next, we constructed three types of 293T cell line overexpressing CD19/CD20 and incubated with S-STAR and D-STAR T cells. After stimulation by either CD19- or CD20-expressed 293 T cells, D-STAR could effectively trigger T cell activation, as measured by CD69 and CD25 expression. Upon CD19 antigen stimulation, both CD19 STAR and D-STAR showed similar CD69 and CD25 expression. When stimulated by CD20, D-STAR exhibited CD69 expression similar to CD20-STAR, while CD25 expression was lower in CD20-STAR. This difference may be due to longer incubation time when detecting CD25 expression (Figure 3C). When incubated with CD19^KO or CD20^KO Raji cells at various time points, D-STAR outperformed both CD19-STAR and CD20-STAR in terms of efficacy (Figure 3D), although CD20-STAR exhibited higher cytokine production than D-STAR upon activation by CD19^KO Raji cells (Figure 3E). We hypothesized that this phenomenon may due to the different spatial arrangement of the VH-VL regions derived from FMC or OFA antibodies. In S-STAR, these regions are separate, while, in D-STAR, they are aligned linearly. This structural difference could affect the tumor antigen recognition, although results might change with a different scFv, warranting further exploration.

Figure 3.

Effector functions of D-STAR T cells under single-antigen stimulation

(A) Design and gene schematics of S-STAR and D-STAR on lentiviral vectors, co-expressing a red fluorescence protein (RFP) reporter. (B) CD19 and CD20 protein-binding abilities of S-STAR and D-STAR. FSC-H indicating the forward scatter detector-height. Data are from one of three independent experiments. (C) Engineered T cells were incubated with antigen-expressing 293T cells and WT 293T cells at an E:T ratio of 2:1. Early T cell activation marker CD69 was detected after 6 h, and CD25 expression was detected after 9 h. Data shown are from one of three independent experiments. (D) Cytotoxicity of S-STAR and D-STAR against CD19^KO or CD20^KO Raji cells at an E:T ratio of 1:1 was assessed at various time points using luciferase assay; this experiment was repeated with human primary T cells derived from two different donors. p values were determined using two-way ANOVA with Tukey’s correction for multiple comparisons. (E) Cytokine concentrations released by S-STAR and D-STAR after a 24-h coculture at an E:T ratio of 1:1. Data shown as means ± SEM are technical duplicates from one of three independent experiments. p values were determined using one-way ANOVA with Tukey’s correction for multiple comparisons. ∗p < 0.05, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001; ns, not significant.

Given the heterogeneous antigen expression among these malignant B cells (Figure S4), we evaluated whether D-STAR T cell could eliminate various types of malignant B cells. After incubating S-STAR or D-STAR T cells with various malignant B cells for 24 h, D-STAR T cells exhibited stronger cytotoxicity and comparable effector cytokines secretion against multiple malignant B cell lines compared to S-STAR T cells (Figures 4A and 4C). Modulated coculture time confirmed that D-STAR demonstrated tumor lysis capabilities equivalent to CD20-STAR T cells against Raji cells but outperformed CD19-STAR T cells (Figure 4B).

Figure 4.

D-STAR T cells effectively lyse malignant B cell with variable CD19 and CD20 expression

(A) Cytotoxicity of S-STAR and D-STAR against various malignant B cell lines, measured by LDH release in the supernatant. Data are presented as means ± SEM, representing technical duplicates from one of three independent experiments. p values were determined using one-way ANOVA with Tukey’s correction for multiple comparisons. (B) Tumor lysis capability of S-STAR and D-STAR, transduced into donor-derived primary T cells against Raji cells at an E:T ratio of 1:1 after 6-, 12-, and 24-h coculture, as determined by luciferase assay. Data are shown as means ± SEM, representing technical duplicates from one of three independent experiments. p values were determined using two-way ANOVA with Tukey’s correction for multiple comparisons. (C) Concentration of effector cytokines produced by S-STAR and D-STAR against various malignant B cells after 24-h coculture at an E:T ratio of 1:1. Data are shown as means ± SEM, representing technical duplicates from one of three independent experiments. p values were determined using one-way ANOVA with Tukey’s correction for multiple comparisons. (D) Intracellular staining for granzyme B and perforin in S-STAR and D-STAR T cells after 8-h coculture with Raji and Nalm6 cells at an E:T ratio of 1:3. Data are from one of two experiments using two different donors. (E) Expression of apoptosis-related proteins detected after 3- and 6-h coculture with Raji or Nalm6 cells at E:T ratios of 1:5 or 1:10. Data are from one of two experiments using two different donors. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001; ns, not significant.

We further assessed S-STAR and D-STAR T cell function using Raji cells (Burkitt’s lymphoma) and Nalm6 cells (acute lymphoblastic leukemia). Consistent with tumor lysis, intracellular staining of granzyme B suggested that D-STAR T cells outperformed both CD19-STAR and CD20-STAR T cells, while D-STAR secreted perforin similar to CD20-STAR when incubated with Raji but higher in terms of Nalm6, which showed undetectable CD20 expression (Figure 4D). Traditionally, granzyme B could directly induce tumor cell death, leading us to assess cleaved-caspase 3, a marker of cell death. We found that D-STAR T cells began inducing cell death in Raji and Nalm6 cells after 3 h, with increased levels of cleaved-caspase 3 correlating with higher tumor cell number or prolonged incubation times (Figure 4E). When comparing CD19-STAR, CD20-STAR, and D-STAR at different E:T ratios with Raji cells, D-STAR induced stronger cell death than CD19-STAR and performed similarly to CD20-STAR (Figure S6C).

In summary, these results indicate that, under stimulation by one or two antigens, D-STAR T cells can activate and exert superior effector functions, efficiently killing various malignant B cells.

Compared to second-generation CARs, D-STAR T cells mediate robust antitumor efficacy

Given the promising results of dual-targeting CAR-T cell therapies in tumor elimination, we proceeded to compare D-STAR T cells with second-generation CD19-BBz-CAR and CD19-CD20-BBz-CAR in term of cytotoxicity, cytokine release, and antitumor efficacy. The CD19-BBz-CAR is identical to the structure of tisagenlecleucel (CTL019),²⁷ while the CD19-CD20-BBz-CAR is based on the published tandem CD19/CD20 CAR (Figure 5A).²⁵

Figure 5.

D-STAR T cells exhibit robust antitumor efficacy compared to single-target and dual-target CAR-T cells both in vitro and in vivo

(A) Gene schematics of single-target, dual-target second-generation CAR and D-STAR on lentiviral vectors. (B) Intracellular staining for perforin, granzyme A, and granzyme B in T cells after an 8-h coculture with Raji cells at an E:T ratio of 1:5. Data are from one of two experiments using two different donors. (C) Expression of apoptosis and pyroptosis-related proteins, detected after 3- and 6-h coculture with Raji cell at an E:T ratio of 1:10. Data are from one of three independent experiments using different donors. (D) Concentration of effector cytokines and tumor lysis observed after a 24-h coculture with CD19^KO, CD20^KO, and WT Raji cells at an E:T ratio of 1:1. Data shown as means ± SEM are technical duplicates from one of three independent experiments. p values were determined using one-way ANOVA with Tukey’s correction for multiple comparisons. (E) Bioluminescence images of mixed CD19^KO, CD20^KO, and WT Raji cell tumors. Black arrowheads indicate mice re-challenged with mixed tumor cells across all groups on day 19. (F) Progression of Raji tumors evaluated by bioluminescence imaging, with n = 5 mice per group. Differences are compared between the CD19-CD20-BBz CAR and CD19-CD20-STAR groups at each time point. (G) Survival curve up to 90 days of each group. This experiment was repeated twice; p values determined by Kaplan-Meier analysis. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001; ns, not significant.

CD19-BBz-CAR secreted higher levels of perforin, granzyme A, and granzyme B compared to CD19-CD20-BBz-CAR and D-STAR after stimulation with Raji cells (Figure 5B). As reported, granzyme B can induce the cleavage of caspase 3, which further leads to GSDME cleavage and induces cell pyroptosis.^28,29 Notably, higher N-GSDME expression was observed at 6 h with CD19-CD20-BBz CAR and with CD19-CD20-STAR, compared to CD19-BBz CAR-T cells. Activated caspase-3 was detectable at 0 h with CD19-BBz CAR-T cells but was absent in CD19-CD20-BBz CAR and CD19-CD20-STAR T cells, potentially explaining the GSDME cleavage. Interestingly, both CD19-BBz CAR and CD19-CD20-STAR T cells showed higher expression of activated caspase-3 at 6 h compared with CD19-CD20-BBz CAR-T cells, which is consistent with granzyme B secretion (Figure 5C). In terms of cytotoxicity against WT Raji and CD19^KO-Raji, D-STAR, compared to CD19-CD20-BBz-CAR, showed no prominent difference in cytotoxicity but exhibited better killing ability than CD20^KO-Raji (Figure 5D). When comparing CD19-BBz-CAR to D-STAR T cells, both showed similar cytotoxicity against CD20^KO-Raji and WT Raji cells but CD19-BBz-CAR-T cells secreted higher effector cytokines, including interleukin (IL)-2 and tumor necrosis factor (TNF)-α (Figure 5D). This suggests that D-STAR T cells can mediate robust effector function with lower cytokine production.

To evaluate the antitumor efficacy of D-STAR in a mouse model, we inoculated CD19^KO-Raji, CD20^KO-Raji, and WT Raji cells in a 1:1:8 ratio into immunodeficient NCG mice to mimic tumor escape. Engineered T cells were then administrated intravenously on day 6 following tumor inoculation (Figure 5E). D-STAR T cells induced long-term tumor regression, exhibiting a significant delay in tumor growth within the first 20 days when compared to CD19-BBz-CAR and CD19-CD20-BBz-CAR (Figures 5E and 5F). Subsequently, we re-challenged the mice with a mix of target cells to assess whether the existing D-STAR T cell could continue to exert cytotoxicity on antigen-loss tumor cells. The results demonstrated that D-STAR T cells were capable of controlling tumor progression and prolonging survival in this tumor-escape mouse model (Figure 5G). Additionally, we repeated this experiment in vivo with an increased ratio of antigen-loss Raji cells, using a 1:1:2 ratio of CD19^KO-Raji, CD20^KO-Raji, and WT Raji cells (Figure S5A). D-STAR T cells still showed superior tumor control compared to CD19-BBz-CAR and CD19-CD20-BBz-CAR, together with prolonged survival of the mice (Figures S5B–S5D).

TanCAR7, a CD19/CD20 dual-targeting CAR (D-CAR), successfully finished a phase I clinical trial for patients with relapse/refractory B cell lymphoma yielding impressive outcomes.¹¹ Consequently, we compared D-STAR with TanCAR7. Since TanCAR7 uses a different anti-CD20 scFv (Leu16), we engineered the Leu16-FMC-STAR using the same scFvs for a directly comparison with Leu16-FMC-CAR (TanCAR7) (Figure S6A). While the functions of Leu16-FMC-STAR and Leu16-FMC-CAR were comparable, the OFA-FMC-STAR showed stronger antitumor cytotoxicity than OFA-FMC-CAR (Figure S6B). Additionally, we observed that D-STAR induced greater cell death than Leu16-FMC-CAR, as indicated by the expression of N-GSDEME and C-C3 (Figure S6C). These results suggest that D-STAR enhances T cell-mediated tumor lysis more efficiently than TanCAR7. We also examined cytokine production when D-STAR or TanCAR7 eliminated tumor cells. The results revealed that D-STAR T cells secreted lower levels of effector cytokines, such as IL-2, interferon (IFN)-γ, TNF-α, and granulocyte-macrophage colony-stimulating factor (GM-CSF), compared to TanCAR7, indicating that D-STAR T cells can mediate robust effector functions with reduced cytokine production, consistent with the activation through TCR (Figure S6D).

D-STAR assembled with OX40 costimulatory domain could enhance T cell fitness

Previously, we modified CD19-STAR by incorporating OX40 costimulatory domain.²¹ However, since STAR needs to pair with endogenous CD3 molecules, the modification to the intracellular domain needs to consider spatial limitations. Meanwhile, the cytoplastic domain of OX40 is positively charged and hydrophilic, which may appeal to negatively charged T cell membrane, as predicted (Figure 6A). To ensure optimal exposure of the binding site on the cytoplastic domain for TRAF2, which initiates downstream signaling of the OX40 costimulatory domain,^30,31 we designed different lengths of flexible linkers to increase the distance between TCRα chains and the OX40 cytoplasmic domain. This configuration is referred to as D-STAR-linkers-OX40 (Figure 6A).

Figure 6.

Armored D-STAR T cells with OX40 costimulatory domain enhance T cell proliferation

(A) Prediction and design of D-STAR modified with OX40 costimulatory domain, separated by different length of flexible linkers. (B) OX40 signaling events of D-STAR and D-STAR-linkers-OX40 T cells upon stimulation with Raji cells. Transduced T cells were sorted for RFP⁺ followed by coculture with Raji cell at an E:T ratio of 2:1 for 8 h. Tubulin and PCNA were used as the loading control. Representative results from one of two independent experiments. (C) T cell proliferation of D-STAR and D-STAR-linkers-OX40 upon Raji cell stimulation, as detected by Edu assay. Data are from one of two experiments using two different donors. (D) T cell proliferation under repeated stimulation with irradiated Raji cells for 5 days. Representative flow plots show CFSE dilution as a measure of cell proliferation on days 0 and 5. All cells are CFSE⁺ at day 0, shown as the gray population. Representative data are from one of two donors. (E) Bioluminescence images showing progression of Raji tumors; black arrowheads indicate mice re-challenged with CD19^KO-Raji cells. (F) Raji tumor progression as evaluated by bioluminescence imaging. Data shown as means ± SEM; p values were determined using two-way ANOVA with Tukey’s correction for multiple comparisons and compared on day 33. (G) Circulating RFP⁺ T cells in the bloodstream. Data shown as means ± SEM; p values were determined using two-way ANOVA with Tukey’s correction for multiple comparisons and comparing CD19-CD20-STAR with CD19-CD20-STAR-(G4S)3-OX40 on day 10 and CD19-CD20-BBz-CAR with CD19-CD20-STAR-(G4S)3-OX40 on day 25. (H) CD8⁺ T cell populations of each group. n = 5 mice per group. p values were determined using two-way ANOVA with Tukey’s correction for multiple comparisons and comparing CD19-CD20-BBz-CAR with CD19-CD20-STAR-(G4S)3-OX40 on day 13. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗∗p < 0.0001.

Firstly, we performed experiments to assess OX40 signaling in D-STAR-linker-OX40 T cells to verify the functionality of this domain. After 8-h stimulation, both D-STAR and D-STAR-linker-OX40 T cells exhibited equivalent levels of phosphorylated p100 in the cytoplasm. However, D-STAR-(G4S)3-OX40 and D-STAR-(G4S)7-OX40 T cells showed more RelB translocation into nucleus, indicating enhanced activation of non-canonical nuclear factor κB (NF-κB) signaling pathways (Figure 6B). Next, we assessed the impact of OX40 costimulatory domain on T cell function. Firstly, we incubated both D-STAR and D-STAR-linker-OX40 T cells with Raji cell for 8 h along with 5-ethynyl-2′-deoxyuridine (Edu) (Figure 6C). The results demonstrated that D-STAR-(G4S)3-OX40 T cells incorporated more Edu, indicating greater numbers of actively proliferating T cells. Another assay for detecting T cell proliferation is carboxyfluorescein succinimidyl ester (CFSE), a fluorescent cell-labeling dye used to track cell division and proliferation over extended periods. Compared to D-STAR and the other D-STAR-linkers-OX40 constructs, D-STAR-(G4S)3-OX40 exhibited more CFSE⁻ T cells, indicating an enhanced proliferative ability under prolonged stimulation (Figures 6D, S7A, and S7B). We hypothesized that this might be related to the cell cycles, so we measured ki-67 expression. The results showed that, compared to D-STAR, both D-STAR-(G4S)3-OX40 and D-STAR-(G4S)7-OX40 expressed higher Ki-67, as shown in Figure S7C. This difference may not be related to receptor expression on T cell membrane, as the mean fluorescence intensity (MFI) of FMC63⁺ expression in engineered T cells from two different donors indicated that the expression of D-STAR is similar or even higher compared to other constructs (Figure S7E). However, we observed a decreased activation level in D-STAR-(G4S)7-OX40 T cells after a 12-h incubation with tumor cells (Figure S7D). This reduced activation level over time may negatively impact T cell function.

Based on the results from in vitro experiments, we selected D-STAR-(G4S)3-OX40 to evaluate the antitumor efficacy in an intraperitoneal mouse model, which represents a more chronic condition with a higher tumor burden compared to the intravenous mouse model. In this scenario, we inoculated 2 million Raji cells intraperitoneally to immunodeficient NCG mice. Engineered T cells were then administrated intravenously on day 6 post inoculation. Both CD19-CD20-BBz CAR and CD19-CD20-STAR were unable to control tumor growth, as shown in Figure 6E. However, D-STAR-(G4S)3-OX40 T cells could induce long-term tumor regression, together with increased circulated T cells compared to D-STAR without OX40 (Figures 6E–6G). For CD19-CD20-BBz CAR, the higher tumor load allowed for continuous stimulation by tumor cells, enabling T cell proliferation to be maintained despite the decreasing trend. Meanwhile, we monitored the dynamic percentage of CD8⁺ T cells in circulated T cells. There was no significant difference between the percentage of CD8⁺ T cells in D-STAR and D-STAR-(G4S)3-OX40 T cells, but CD19-CD20-BBz CAR exhibited a higher CD8⁺ T cell population 13 days post T cell infusion compared to D-STAR-(G4S)3-OX40 T cells (Figure 6H).

In conclusion, our findings suggest that modifying D-STAR-OX40 with flexible linkers can enhance T cell proliferation under prolonged stimulation and further maintain T cell fitness in conditions with a higher tumor burden.

Discussion

Post-therapy relapses due to antigen negativity or mutation resistance and life-threatening T cell-related toxicity are two major limitations on CAR-T cell therapy.³² Engineering multi-specific CAR-T cells needs to consider the position and arrangement of different scFvs, which lacks a systemic solution.^33,34 Therefore, we decided to improve the traditional structure of CARs and to engineer a novel chimeric receptor with dual targeting to address the issue.

In this study, we generated a double-chain STAR receptor and converted it into the dual-targeting STAR, termed D-STAR. From in vitro experiments, we demonstrated that the position and choice of scFvs on D-CAR structure could affect its effector function, while D-STAR with switched scFv position or choice of scFv shows no obvious difference, indicating that D-STAR structure was more universal for multi-specificity. This phenomenon is also related to the activation threshold between D-CAR and D-STAR. In addition, we found it is necessary to pair the constant domains of TCRα and TCRβ when engineering STAR receptor. Otherwise, the overexpressed TCR-Cα or TCR-Cβ constant region would be degraded by ER-associated pathway. This degradation could be reversed by mutated ubiquitinated amino acid. In our previous works, a systematical comparison among TCR, CAR, and STAR demonstrated that STAR could trigger TCR-like signaling processes and transcriptional programs, which are distinct from CAR signaling.²⁰ It is known that TCR maintains high antigen sensitivity and forms stable immune synapses upon activation, whereas it was found that CAR cannot form well-organized immune synapses, which may result in its increased activation threshold.³⁵ Although we have not detected the immune synapse of D-STAR yet, based on previous data and structural property, we hypothesized that STAR may form a stable immune synapse similar to TCR to maintain higher sensitivity and lower activation threshold.³⁶

The original D-STAR receptor only contains the first signal of T cell activation and lacks the second and third signals, which are known to have important effects on the proliferation and differentiation of T cells. Therefore, we considered armoring an additional motif on D-STAR to enhance its versatile function to encounter the complex tumor microenvironment. OX40 costimulatory receptor-modified D-STAR could promote T cell proliferation upon short-term and long-term antigen stimulation and improve T cell performance in mouse model with more tumor load. Although this signaling component appears to successfully recapitulate signal 2 of T cell activation, it is still unclear whether it is subject to the same regulation as native TCR and costimulatory receptors. A major limitation of D-STAR is that the turnover of D-STAR chains might be controlled by TCR-like feedback regulation mediated by endocytosis, recycling, and degradation. Receptor degradation of T cells serves as a protective strategy to avoid continuous activation leading to exhaustion.^37,38 Since TCR complex can be endocytosed within 30 min, we will examine whether D-STAR can be quickly endocytosed upon antigen stimulation. We wonder whether this process could affect the function of D-STAR T cells. Further experiments will explore whether modulating D-STAR endocytosis could enhance its effector function.

Besides STAR, several studies also engineered TCR-based CARs; for example, antibody-TCR (abTCR),³⁹ T cell antigen coupler (TAC),⁴⁰ and T cell receptor fusion constructs (TRuC).^41,42 In our study, STAR receptor fuses scFv with Cα and Cβ, which is different from Cγ and Cδ used by abTCR, CD3ε used by TRuC, and antigen-binding domains used by TAC. Based on these properties, STAR structure is more similar to the natural TCR and easier to design as dual targeting on its double chain.

In conclusion, our work involves engineering a novel double-chain chimeric receptor with structural advantage targeting two different antigens, applicable to both solid and liquid tumors. The D-STAR, targeting CD19 and CD20, demonstrates robust effector function against various malignant B cells and outperforms second-generation CARs in a tumor-escape mouse model. Additionally, we modified the D-STAR structure with OX40 cytoplasmic domain. Introducing a medium-length flexible linker between the TCRα chains and the OX40 cytoplasmic domain enhances T cell proliferation and antitumor efficacy in high-tumor-burden scenarios compared to D-STAR without OX40. This study offers a promising therapeutic approach against relapsed/refractory tumors and explores optimizations on cell therapy.

Materials and Methods

Study design

In this study, we designed a novel double-chain chimeric receptor named D-STAR, which had antigen-binding ability derived from antibody. We examined D-STAR’s structural features, activation pattern, cytotoxicity against tumor cell line, and antitumor efficacy in mouse model and conducted systematic comparisons with second-generation CARs and STAR T cells. Moreover, we enhanced D-STAR with an additional costimulatory motif to improve T cell fitness under high tumor burden. The statistical tests used and the number of experimental replicates are specified within each figure legend. We utilized female NCG mice aged 6 to 8 weeks for in vivo studies. Prior to T cell treatment, mice were classified based on their bioluminescence readings to ensure an even distribution of the average tumor load among all groups. The experiments were conducted unblinded, and no outliers were removed from the data.

Mouse model

Female NCG mice aged 6 to 8 weeks were acquired from GemPharmatech and housed in a pathogen-free environment at Tsinghua University Animal Facility. Mice were kept at 12:12 light/dark cycles with unrestricted food and water availability. All mouse experiments were conducted according to Institutional Animal Care and Use Committee (IACUC)-approved protocols. For the intraperitoneal tumor model, mice were inoculated with equal numbers of tumor cells combined with Matrigel (BD Biosciences, #356234). For the intravenous tumor model, Raji-luc cells were administered by tail vein injection. T cells (cell numbers varied for different experiments and are noted in figures or figure legends) were infused intravenously via tail vein at indicated time points. Tumor progression was monitored via bioluminescence emission using a Lumina II instrument (CALIPER) following D-luciferin (YEASEN) injection. Blood samples were collected from orbital sinus without any randomization or blinding procedure.

Vector construction

We used a pHAGE backbone to construct the vector for transducing T cell lines and primary T cells via lentivirus. The vector contained an EF-1α promoter and an IRES-linker RFP reporter. The scFvs sequences of FMC63, OFA, rituximab, 2C6, and Leu-16 were previously described. The TCR constant regions, costimulatory domains, and additional sequence were cloned from primary T cell cDNA or synthesized after codon optimization. TCR constant region employing a murine-mutant sequence was previously reported. CD3ζ cytoplasmic domain (UniProtKB-P20963, amino acids 52–164), 4-1BB cytoplasmic domain (UniProtKB-Q07011, amino acids 214–255), and OX40 cytoplasmic domain (UniProtKB-P43489, amino acids 236–277). Gene fragments were assembled by seamless cloning kit (Clone Smarter, #C5891-50) into vector backbones together with a Kozak sequence for optimal translation.

Cell line and culture

Lenti-293T cells were purchased from Takara Biomedical for lentivirus packaging. Jurkat E6.1 and various human cancer cell lines were purchased from ATCC and cultured in specified media supplemented with FBS. CD19^KO and CD20^KO Raji cells were engineered via CRISPR-Cas9. JC5 cells were derived from Jurkat E6.1 cells by knocking out TCRα/β chains with a CRISPR-Cas9 system. The Raji, CD19^KO-Raji, CD20^KO-Raji, and Nalm6 were engineered to express firefly luciferase GFP. CD19-K562 and CD20-K562 were generated by lentivirally transducing K562 cells with CD19 or CD20. Regular mycoplasma testing confirmed the absence of contamination.

Flow cytometry

The following antibodies were used: Biotin-anti-FMC63 scFv (bioswan, #500014), Streptavidin-BV421 (BioLegend, #405225), Streptavidin-APC (eBioscience, #17-4317-82), CD69-FITC (BioLegend, #310903), mTCRβ-APC (BioLegend,#109212), human TCRα/β-BV421 (BioLegend, #306721), Granzyme B-BV421 (BioLegend, #372217), CD107a-BV421 (BioLegend, #328625), recombinant human CD19 protein (Abcam, # ab234966), recombinant human CD20 protein (Abcam, #ab158047), CD25-APC (BioLegend, #302610), human CD8-percp-cy5.5 (BioLegend, #344710), human CD3-PE-cy7 (BioLegend, #344816), Perforin-Percp-cy5.5 (BioLegend, #353313), Granzyme A-PE/Cyanine7 (BioLegend, #507221), CD19-PE-cy7 (BioLegend, #302216), CD20-APC (BioLegend,#302310), PD1-PerCP/Cy5.5 (BioLegend, #329914), TIM3-BV421 (BioLegend, #345007), LAG3-AF647 (BD Biosciences, #565717), EGFR-EF660 (eBioscience, #50-9509-41), PDL1 (CD274)-APC (BioLegend, #329707), ROR1-PE (BD Biosciences,#564474), and ki-67-APC (BioLegend,#350514). Transduced Jurkat E6.1 cells were incubated with his-tag-labeled antigen protein (10 μg/mL) on ice for 30 min, followed by staining with fluorescence-conjugated anti-His-tag antibody (BioLegend, #652513).

Western blot

Cells were collected in lysis buffer (50 mM HEPES [pH 7.4], 150 mM NaCl, 1% NP-40, and 1 mM EDTA) containing 1 mM sodium orthovanadate, 1 mM sodium fluoride, 1 mM phenylmethylsulfonyl fluoride (PMSF), and a protease inhibitor mixture (Roche). Primary antibodies included β-tubulin (EASY BIO, #BE0025-100), c-caspase-3 (Cell Signaling Technology, #9664s), caspase-3(Cell Signaling Technology, #9662s), N-GSDME (Abcam, ab215191), GSDMB (Abcam, ab215729), Phospho-NF-κB2 p100 (Ser866/870) (CST, #4810), RelB (CST, #4954), and PCNA (Santa Cruz, #sc-56).

Enzyme-linked immunosorbent assay

The cytokine response of T cells was determined by cytokines released into culture medium. The cytokine concentrations were measured using the following enzyme-linked immunosorbent assay kits per the manufacturer’s instructions: human IL-2 (Invitrogen, #88-7025-88), human IFN-γ (Invitrogen, #88-7316-88), and human TNF-α (Invitrogen, #88-7346-88), human GM-CSF (Invitrogen, #88-8337-88).

Cytotoxicity assays

The cytotoxicity of T cells was determined by luciferase reporter and lactate dehydrogenase (LDH)-based assays (CytoTox 96 Non-Radioactive Cytotoxicity Assay, Promega). The transduced T cells and tumor target cells were cocultured in replicates at the indicated effector:target ratio.

Proliferation assay

This study involves two kinds of proliferation assay, namely Edu (Beyotime, #C0071S) and CFSE (Invitrogen, #C34554). Edu is a thymidine analog that integrates into newly synthesized DNA during the DNA-replication phase of the cell cycle, reflecting T cell proliferation during the initial activation stage. A higher incorporation of Edu reflects a greater number of actively proliferating T cells. Transduced T cells were stimulated with irradiated tumor cells in the presence of Edu. After 8-h incubation, cells were isolated and analyzed by flow cytometry with a 488-nm excitation source. Another assay is CFSE, a fluorescent cell-labeling dye used to track cell division and proliferation over extended periods. Cells were stained with CFSE first to ensure all cells were CFSE⁺. As T cells proliferate during coculture with irradiated Raji cells for 5 or 7 days, the CFSE staining becomes diluted. These cells were then harvested and analyzed by flow cytometry with a 488-nm excitation source.

Statistical analysis

Statistical analysis was performed using GraphPad Prism 6 software. No statistical methods were used to predetermine sample size. Comparisons between two groups used two-tailed parametric or nonparametric t tests, while multiple group analyses utilized one-way analysis of variance (ANOVA) or two-way ANOVA, and Tukey’s multiple comparison test. Survival data were analyzed using the log rank test and plotted with the Kaplan-Meier method. Statistical tests applied are mentioned in each figure legend.

Data availability

All data associated with this study are provided in the manuscript or supplemental information. The DNA plasmids and cell lines described here are available from X.L. under a material-transfer agreement with the Tsinghua University.

Acknowledgments

The flow cytometry was carried out at Core Facility of Institute for Immunology in Tsinghua University. The animal experiments were carried out at Laboratory Animal Resources Center in Tsinghua University. The authors acknowledge the help of W.D. Han for providing TanCAR7 plasmid. This work was supported by the National Natural Science Foundation of China (82293664, 82341212, and 31930039), annual funds from Tsinghua-Peking Center for Life Sciences, and annual funds from Changping Laboratory.

Author contributions

X.L., J.W., and L.Y. conceived and designed the project. L.Y., Z.Z., and J.W. carried out experimental work and interpreted data. L.Y., Z.Z., H.Y., D.H., and Y.L. developed the methodology. L.Y. wrote the manuscript. J.W. and X.L. revised the manuscript and supervised the project. All authors read and approved the manuscript.

Declaration of interests

The authors declare no competing interests.